CONTENTS PREFACE ......................................................................................................................xiii INTRODUCTION The spread of the Indo-Europeans.............................................................................1 General linguistics and Indo-European reconstruction ......................................... 7 On Russenorsk............................................................................................................21 The origin of the Goths............................................................................................. 27 C.C. Uhlenbeck on Indo-European, Uralic and Caucasian................................... 31 An outline of Proto-Indo-European ........................................................................37 Schleicher’s fable ........................................................................................................ 47 INDO-EUROPEAN PHONOLOGY *H2o and *oH2 .............................................................................................................. 51 Proto-Indo-European glottalic stops: The comparative evidence ........................53 Proto-Indo-European tones?.................................................................................... 67 An Indo-European substratum in Slavic?................................................................73 INDO-EUROPEAN MORPHOSYNTAX 1st sg. middle *-H2 .......................................................................................................81 Proto-Indo-European verbal syntax.........................................................................91 GREEK Greek numerals and Proto-Indo-European glottalic consonants.......................105 The Aeolic optative....................................................................................................111 The Greek 3rd pl. endings........................................................................................ 117 INDO-IRANIAN Glottalic consonants in Sindhi and Proto-Indo-European ................................. 121 Archaic ablaut patterns in the Vedic verb.............................................................. 125 Accent and ablaut in the Vedic verb........................................................................131 The origin of the Indo-Iranian desiderative.......................................................... 139 viii Contents TOCHARIAN On the development of Proto-Indo-European final syllables in Tocharian ...... 143 The Tocharian word for ‘woman’............................................................................149 The fate of the sigmatic aorist in Tocharian ...........................................................151 A note on the Tocharian dual.................................................................................. 155 The Tocharian imperfect ......................................................................................... 159 GERMANIC PHONOLOGY Vestjysk stød, Icelandic preaspiration, and PIE glottalic stops ........................... 165 Proto-Germanic obstruents ....................................................................................169 Kluge’s law and the rise of Proto-Germanic geminates........................................ 175 Labials, velars and labiovelars in Germanic ..........................................................179 Preaspiration or preglottalization? ......................................................................... 185 Germanic *ē1 and *ē2.................................................................................................189 Proto-Germanic obstruents and the comparative method ................................. 193 English bottom, German Boden, and the chronology of sound shifts................197 GERMANIC VERB CLASSES The Germanic first class of weak verbs..................................................................201 The Germanic third class of weak verbs............................................................... 205 The Germanic seventh class of strong verbs ........................................................ 209 The Germanic fifth class of strong verbs ............................................................... 211 The Germanic sixth class of strong verbs .............................................................. 215 The Germanic fourth class of weak verbs..............................................................219 Old Norse taka, Gothic tekan, Greek τεταγών...................................................... 221 GERMANIC VERBAL INFLEXION The Germanic weak preterit ...................................................................................227 The Proto-Germanic pluperfect .............................................................................235 GERMANIC NOMINAL INFLEXION The inflexion of the Indo-European ā-stems in Germanic................................ 239 The inflexion of the Germanic n-stems .................................................................343 GERMAN Old High German umlaut ...................................................................................... 247 The High German consonant shift........................................................................ 249 The origin of the Franconian tone accents............................................................255 Contents ix ENGLISH The origin of the Old English dialects ...................................................................259 How old is the English glottal stop? ...................................................................... 265 The origin of the Old English dialects revisited................................................... 269 Anglo-Frisian ............................................................................................................275 SCANDINAVIAN The Old Norse i-umlaut...........................................................................................285 On breaking.............................................................................................................. 289 Glottalization, preaspiration and gemination in English and Scandinavian.... 293 Early Runic consonants and the origin of the younger futhark......................... 299 Björketorp and Stentoften .......................................................................................305 The origin of the vestjysk stød ................................................................................ 313 Vestjysk stød again ................................................................................................... 317 ALBANIAN Proto-Indo-European *s in Albanian..................................................................... 319 Proto-Indo-European *j in Albanian .....................................................................325 Reflexes of Indo-European consonants in Albanian........................................... 329 ARMENIAN Armenian ewł ‘oil’..................................................................................................... 333 BALTO-SLAVIC The Baltic word for ‘in’............................................................................................. 335 All’s well that ends well ............................................................................................337 Balto-Slavic accentuation revisited......................................................................... 341 Lithuanian žinóti ‘to know’......................................................................................359 ITALO-CELTIC More on the chronology of Celtic sound changes ................................................361 ANATOLIAN Initial laryngeals in Anatolian.................................................................................365 Hittite ammuk ‘me’ .................................................................................................. 369 Hittite hi-verbs and the Indo-European perfect ...................................................373 Stative and middle in Hittite and Indo-European ................................................383 x Contents INDO-URALIC Eight Indo-Uralic verbs?..........................................................................................387 The Indo-Uralic verb ............................................................................................... 391 Nivkh as a Uralo-Siberian language ...................................................................... 405 Indo-Uralic consonant gradation ..........................................................................409 Indo-Uralic and Altaic ............................................................................................. 415 Indo-Uralic and Altaic revisited .............................................................................419 APPENDIX A parasitological view of non-constructible sets ................................................. 429 The origin and nature of the linguistic parasite....................................................435 REFERENCES .............................................................................................................. 439 INDEX ......................................................................................................................... 493 PREFACE This book contains most of what I have written about Germanic, IndoEuropean and Indo-Uralic. It is complementary to the earlier volumes on Armenian (K194), Celtic (K239) and Baltic (K263, see the references under Kortlandt). Together they represent the bulk of my scholarly output with the exception of studies on Slavic and general linguistics. The red thread which runs through my work is a quest for relative chronology of linguistic developments. A methodological advantage of this approach is that it offers the possibility of considering the compatibility of different solutions before assessing their correctness, which enhances the motivation to integrate different views. Moreover, the probability of a reconstruction can be judged against the background of the transitions which it implies for the linguistic system as a whole. This is of special importance when the gap between the results of internal reconstruction and the comparative evidence is huge, as it is in the case of Slavic accentuation or Indo-Uralic. Another point which I would like to emphasize here is that my reconstructions are always bottom-up, never top-down. Thus, my reconstruction of preglottalized stops in Proto-Germanic is based on preglottalization in English and Danish, preaspiration in Scandinavian, affrication in High German, and various types of gemination in all West and North Germanic languages. It is independent of any reconstruction of IndoEuropean. Similarly, my reconstruction of glottalic stops in Proto-IndoEuropean is based on glottalization in Indo-Iranian, Armenian, Balto-Slavic and Germanic and indirect evidence from Greek and Latin, not on typological or other general considerations. The same holds for my reconstruction of morphological systems in their chronological perspective. The method of “forward reconstruction” may be useful as a heuristic device but can easily put one on the wrong track and lead to circular reasoning. It follows that the chapters on Germanic can be read without reference to the Indo-European background and that the Indo-Uralic part of the book can be left out of consideration if one does not want to look beyond Proto-Indo-European. The initial chapters of the book may serve as an introduction to the background and methodology of my reconstructions. One point which deserves special attention is the necessity to limit the number of possibilities. It is easy to posit a distinction between palatovelars, plain velars and labiovelars for ProtoIndo-European. However, such a reconstruction does not explain why the plain velars are largely in complementary distribution with the other series (cf. Meillet 1894, Steensland 1973), nor why we find many dozens of examples of alternation xii Preface between different series (e.g. Čekman 1974). These distributional properties of the system are explained by a reductionist approach. Similarly, it is easy to posit five short and five long vowels in the proto-language to cover any possible correspondence between the daughter languages. However, this way of reconstructing yields an enormous amount of complementary distribution in combination with the Indo-European laryngeals. If one takes the laryngeal theory seriously, it follows that *i and *u were resonants with vocalic and consonantal variants on a par with *r, *l, *n, *m, that the colored variants of *e before and after the laryngeals were distinct from later *a and earlier *o, and that the Proto-Indo-European long vowels originated as phonetic variants of short *e and *o (cf. Wackernagel 1896: 66-68). In a similar vein, it is pointless to reconstruct a large number of grammatical categories for the proto-language without considering the lack of evidence for the implied possibilities in the daughter languages. A final note on the transcription: following Oettinger (1979) I write h for ḫ but š, not s in Hittite. For the Proto-Indo-European laryngeals I avoid the symbol *h, which suggests the wrong phonetics. In my earlier work I used to write *H1, *H2, *H3; I now prefer *Ȥ, *ȥ, *ȥw, which represent the most probable phonetic realization (at least for the non-Anatolian branch of Indo-European), or *q1, *q2, *q3, which suggest any back articulation from post-velar to laryngeal. In view of the different contexts, I have decided against a uniform transcription for all chapters. From a structural point of view, one might expect a larger number of laryngeals on the basis of the number of velar obstruents, but there is no comparative evidence to support their reconstruction. For Old Norse I have most often followed Heusler’s spelling (1967). In the course of the past decades I have profited much from scholarly discussions with Robert Beekes, Cornelis Ruijgh, Alexander Lubotsky, Jörundur Hilmarsson, Dirk Boutkan, Rick Derksen, Leonid Kulikov, Michiel de Vaan, Martine Robbeets, Alwin Kloekhorst, Tijmen Pronk, Guus Kroonen, Michaël Peyrot, Lucien van Beek and other colleagues at various stages of my work. It goes without saying that they cannot be held responsible for any mistakes in the following pages. Leiden, April 1st, 2010 THE SPREAD OF THE INDO-EUROPEANS The publication of Mallory’s book (1989) has rendered much of what I had to say in the present contribution superfluous. The author presents a carefully argued and very well written account of a balanced view on almost every aspect of the problem. Against this background, I shall limit myself to a few points which have not received sufficient attention in the discussion. First of all, the relation between archaeology and linguistics is a precarious and asymmetrical one (cf. already Schmitt 1974). Mallory’s lucid discussion of the problem (1989: 164-168) should be required reading for anybody who ventures into this realm of shadows. It is a methodologically legitimate activity to look for archaeological traces of a linguistic group, but the converse does not hold. Speculations about the linguistic affinity of a prehistoric culture are futile because it is reasonable to assume that the vast majority of prehistoric linguistic groups have vanished without leaving a trace. Thus, it is certainly attractive to assign the ancestors of the speakers of Proto-Tocharian to the Afanasievo culture (cf. Mallory 1989: 62 and 225), but we must never forget that the very existence of the Tocharian texts which have survived is a purely accidental fact of history, due to a number of factors which happened to concur thousands of years after the eastward migrations of the Indo-Europeans. It is not merely possible, but very probable that many groups of Indo-Europeans migrated eastward before the ancestors of the Indo-Iranians, and that the distinguishing feature of the Tocharians is merely the preservation of their historical records. If the differences between East and West Tocharian lead us to date ProtoTocharian to the second half of the first millennium BC, this still leaves a gap of two or three millennia after the purported arrival of the Indo-Europeans in the area. Many things may have happened in the meantime. The real argument for an early eastward migration of the ancestors of the Tocharians is the remarkably archaic character of the attested languages (see Penney 1989 for a point of particular importance). It has often been argued that Tocharian has special connections with the western Indo-European languages. In my view, this is the result of a methodological bias in our way of reconstructing Proto-Indo-European. As Mayrhofer has noted (1983), the history of reconstruction can be described as a gradual shift away from the languages on which the reconstruction is primarily based. The similarities which link Tocharian to the western Indo-European languages reflect precious archaisms which were obscured by more recent developments affecting the dialectal area from which Greek and Indo-Iranian were to evolve. The bias is strengthened by the presence of later parallel innovations in the latter two 2 Introduction branches, e.g. in the development of the middle voice (cf. K044: 130 and K239: 151-157). Similarly, Mallory’s inconclusiveness about the westward Indo-European migrations (1989: 257) appears to result from a search for archaeological evidence beyond what can be motivated from a linguistic point of view. If we follow the traditional opinion and assign the ancestors of the speakers of Celtic and Germanic to the La Tène and Jastorf cultures, respectively, this again leaves us with a gap of two millennia after the Corded Ware horizon to which the ancestors of the western Indo-Europeans may have belonged. Here again, we can be sure that a lot of things happened in the meantime, and it is most probable that many linguistic groups were irretrievably lost. This leads me to the second point I want to make. There seems to be a general tendency to date proto-languages farther back in time than is warranted by the linguistic evidence. When we reconstruct Proto-Romance, we arrive at a linguistic stage which is approximately two centuries later than the language of Caesar and Cicero (cf. Agard 1984: 47-60 for the phonological differences). When we start from the extralinguistic evidence and identify the origins of Romance with the beginnings of Rome, we arrive at the eighth century BC, which is almost a millennium too early. The point is that we must identify the formation of Romance with the imperfect learning of Latin by a large number of people during the expansion of the Roman empire. Similarly, we may identify the formative period of Proto-Indo-European with the earliest expansions of the Indo-Europeans. The issue involved here is partly terminological. Elsewhere I have presented a relative chronology of 22 stages for the phonological developments which characterize the formation of Old Irish (K035; K239: 6-17 and 140f.). All of these developments are posterior to the Ogam inscriptions, which lack the characteristic features of the Old Irish language. If we use the term “Primitive Irish” for the period before the apocope (my stage 15) and the term “Archaic Irish” for the period between the apocope and the syncope (my stage 19), we may wonder about the applicability of the term “Irish” to the Ogam inscriptions; it may be more appropriate to speak of the variety of Insular Celtic spoken by the ancestors of the Irish. In any case, no reconstruction of Proto-Irish on the basis of Old Irish and later materials comes close to anything resembling the language of the Ogam inscriptions. Since the latter can hardly be older than the beginning of the Christian era and the syncope may be dated to the sixth century, it will be clear that I have little confidence in a theory which relegates Proto-Indo-European to the fifth or sixth millennium BC. The radical changes which embody the formation of Irish in the first half of the first millennium AD are probably due to imperfect learning by speakers of an unknown substrate language which was lost forever. The spread of the Indo-Europeans 3 Perhaps the best example of a disintegrating proto-language is furnished by the Slavic material. Apart from the rise of x, all the major developments which differentiate Slavic from its Baltic prototype are usually dated to the first millennium AD (e.g. Shevelov 1964, K058). The earliest dialectal divergences within Slavic which have survived into historical times can hardly be older than the fourth century, and the last shared innovations of the entire group, such as the rise of the neo-acute tone, may be dated to the ninth century. The modern dialectal situation is essentially the same as it was in the twelfth century. When we reconstruct Proto-Slavic, the result can largely be identified with the language of the ninth century, apart from the dialectal differentiation which started half a millennium earlier, apparently in connection with the earliest expansion of the Slavic territory. It is reasonable to assume that many dialects arose and disappeared at earlier stages, but it is not obvious that the term “Slavic” is appropriate before the expansions of the first millennium AD. This brings me to the third point I want to make here. If a proto-language can be dated to the period of its expansion, the mechanism of this process must be examined in detail. It comprises two phases, each of which has its own dynamics. First, a number of people have to move from their original homeland to a new territory. Second, a larger number of people must find it expedient to adopt the language of the intruders. Both developments are determined by specific social and economic circumstances. Population movements are determined by three factors. Firstly, there must be a reason to leave one’s homeland. This factor has rightly been stressed by Anthony, who observes that people living along the boundary between the poorer lowland steppe and the richer upland forest “risked periodic exposure to severe stress, for small variations in precipitation, temperature, population density, or deforestation rate would dramatically alter the local distribution of critical resources in these fragile borderland communities” (1986: 292). This periodic exposure to severe stress prompted expansion when the opportunity presented itself. Secondly, there must be a place where life seems to be better in order to make the journey worth while. This is a reason to expect migrations toward rather than away from more developed areas such as Assyria in the third and second millennia BC. Thirdly, the cost of the journey must not be prohibitive. It is now generally recognized that the domestication of the horse played a crucial part in reducing the cost of physical mobility. The expansion of Indo-European presupposes not only the migrations of Indo-Europeans, but also the adoption of Indo-European languages by local populations. With respect to this issue Mallory refers to Barth’s work (1981) in a discussion which is really too short. The complexity of the problem is illustrated by the following passage, which I cannot refrain from quoting at some length (Mallory 1989: 260f.): 4 Introduction “Barth examined the linguistic relations between the Pathans and Baluchi on the Afghan-Pakistan border. The Pathans were the more numerous, the wealthier, better armed, and even possessed a better military reputation. Nevertheless, it is the Baluchi who have been making the sustained linguistic assimilation of the Pathans. The Baluchi social structure is hierarchic and encourages vertical relationships between local leaders and clients. The various bands offer opportunities for social advancement within these hierarchies, and displaced Pathans in a frontier situation are attracted individually and in groups to join Baluchi communities. On the other hand, the more egalitarian society of the Pathans was ill-suited to absorb foreigners who could only enter it either in roles despised by the Pathans or by undertaking a more complicated process to being admitted as an equal in Pathan society. The nub of the issue here is not weapons, wealth or population size but the social permeability of the competing social organizations. As numerous historical instances testify, pastoral societies throughout the Eurasian steppe are typified by remarkable abilities to absorb disparate ethno-linguistic groups. Indo-European military institutions may have encouraged membership from local groups in the form of clientship which offered local populations greater advantages and social mobility.” This must have been the decisive force in the spread of the Indo-European languages. Starting from the linguistic evidence and trying to fit the pieces into a coherent whole, we arrive at the following picture. The best candidate for the original Indo-European homeland is the territory of the Sredny Stog culture in the eastern Ukraine. The attested languages reflect a number of waves of migration to the east, north of the Caspian Sea (Tocharian, Indo-Iranian), to the south, west of the Black Sea (Anatolian, Greek, Armenian, Albanian), and to the west, south of the Baltic Sea (Italo-Celtic, Germanic). As Mallory notes, there may have been a fourth, abortive wave of migration to the southeast, west of the Caspian Sea, which is not reflected in the linguistic records, perhaps because the Indo-Europeans were assimilated to the local population at an early stage. The earlier migrations yielded the peripheral languages (Tocharian, Anatolian, ItaloCeltic), which did not take part in the late Indo-European innovations of the central dialects (Indo-Iranian, Greek, Germanic, Balto-Slavic, etc.). Some innovations affected only a part of the central dialects, such as the assibilation of the palatovelars (which did not reach Greek and Germanic). Other developments had a more local character. An interconsonantal laryngeal voiced the following stop in North Iranian (Avestan, Sogdian) dugdar- ‘daughter’, but not in its Persian and Indic cognates. This must have been a very early development. It appears that Phrygian was rather closely related to Greek (cf. Lubotsky 1988b), Thracian to Armenian (cf. K101), and Venetic to Italic. The position of Illyrian remains unclear. The spread of the Indo-Europeans 5 The Indo-Europeans who remained after the migrations became speakers of Balto-Slavic. If the speakers of the other satem languages can be assigned to the Yamnaya horizon and the western Indo-Europeans to the Corded Ware horizon, it is attractive to assign the ancestors of the Balts and the Slavs to the Middle Dnieper culture. If the origin of this culture “is to be sought in the Sredny Stog, Yamnaya and Late Tripolye cultures” and this phase is “followed by a middle period where the classic Corded Ware amphorae and beakers appear” (Mallory 1989: 248), the course of events corresponds nicely with the development of a satem language which was drawn into the western Indo-European sphere of influence. The disintegration of Balto-Slavic is closely parallel to that of IndoEuropean: the Slavs migrated to the west, the south, and the east, the Latvians to the north, and the Prussians were assimilated to the Germans. The deceptively archaic character of the Lithuanian language may be compared to the calm eye of a cyclone. The resulting picture can be summarized as follows. Eastward migrations: 1 Tocharian 2a Indic 2b South Iranian 2c North Iranian (3 East Slavic) Southward migrations: 1 Anatolian 2a Greek 2b Phrygian 2c Armenian 2d Thracian 2e Daco-Albanian (3 South Slavic) Westward migrations: 1a Italic 1b Venetic 1c Celtic 2 Germanic (3 West Slavic) Once again it must be emphasized that many linguistic groups may have vanished without leaving any historical record. We must now examine how the view developed here can be related to Gimbutas’ theory of two homelands and three waves of migration into the Balkans. The main objection which can be raised against Gimbutas’ scheme (e.g. 6 Introduction 1985: 198) is that it starts from the archaeological evidence and looks for a linguistic interpretation. As a consequence, the scheme does not fit the linguistic evidence very well. It seems to me that we arrive at a much better representation if we start from the linguistic side and try to find an archaeological corroboration. The natural solution then is to link Gimbutas’ first wave (4400-4200 BC) to the ancestors of the Anatolians, her second wave (3400-3200 BC) to the ancestors of the Greeks and the Phrygians, and her third wave (3000-2800 BC) to the ancestors of the Armenians and the Thracians. If this identification is correct, the satemization process can be dated to the last centuries of the fourth millennium. It is possible that the speakers of Italo-Celtic must be assigned to the Globular Amphora culture, and that Germanic grew out of a later component of the Corded Ware horizon. Since the beginnings of the Yamnaya, Globular Amphora, Corded Ware, and Afanasievo cultures can all be dated between 3600 and 3000 BC, I am inclined to date Proto-Indo-European to the middle of the fourth millennium, and to recognize Proto-Indo-Hittite as a language which may have been spoken a millennium earlier. If we can identify Indo-Hittite and Indo-European with the beginning and the end of the Sredny Stog culture, respectively, it will be clear that the linguistic evidence from our family does not lead us beyond Gimbutas’ secondary homeland and that the Khvalynsk culture on the middle Volga and the Maykop culture in the northern Caucasus cannot be identified with the Indo-Europeans. Any proposal which goes beyond the Sredny Stog culture must start from the possible affinities of Indo-European with other language families. It is usually recognized that the best candidate in this respect is the Uralic language family, while further connections with the Altaic languages and perhaps even Dravidian are possible. The hypothesis that Indo-European is genetically related to a Caucasian language family or to Afro-Asiatic seems much less probable to me. What we do have to take into account is the typological similarity of ProtoIndo-European to the North-West Caucasian languages. If this similarity can be attributed to areal factors, we may think of Indo-European as a branch of UraloAltaic which was transformed under the influence of a Caucasian substratum. It now appears that this view is actually supported by the archaeological evidence. If it is correct, we may locate the earliest ancestors of the speakers of ProtoIndo-European north of the Caspian Sea in the seventh millennium (cf. Mallory 1989: 192f.). This is essentially in agreement with Gimbutas’ theory (cf. also K112). GENERAL LINGUISTICS AND INDO-EUROPEAN RECONSTRUCTION I Denmark has always been a superpower in linguistics. There is no need to list all famous scholars who worked in this country and left their imprint on the history of linguistics, but there are two names which I want to mention here, viz. Otto Jespersen and Holger Pedersen. The point is that we have a lesson to learn from these two great scholars in connection with the relation between general linguistics and Indo-European reconstruction. Otto Jespersen was not only a great phonetician, but is regarded by some as the founder and by others as the forerunner of modern syntax. His Philosophy of Grammar is a classic for linguists of very different theoretical persuasions. The point I want to emphasize here is that Jespersen was very well informed about the comparative linguistics of his time, and was therefore in a very good position to hold strong views about what his colleagues were and were not doing. Holger Pedersen was perhaps the greatest comparative linguist of all time. But he also had a keen sense of scholarly atmosphere, as is evident from his history of 19th century linguistics. One of the characteristic features of his work is the insistence on comparison with what is actually observed in living languages, and on the role of naturalness in what is expected of linguistic development. He simply was a very good general linguist. The fast-growing body of scholarly literature in the field of linguistics and the concomitant rise of specialization have led to a regrettable disintegration of the community of linguists. This is not to say that things were in all respects better in the past. Scholars were not always very nice to each other in former days, as can easily be gleaned from older issues of linguistic journals. There are many more jobs around nowadays. Yet I think that the discipline of linguistics has suffered from a fragmentation which could and should have been avoided. It is clear that nobody can read more than a very small percentage of the total scholarly output in linguistics nowadays. This raises a fundamental question: how to choose what to read? The answer is simple: there is no general way to choose, because you never can tell where to find the unexpected clue. One can only try and look. It is therefore most important to have a general idea of what colleagues are doing elsewhere in the field. A fair assessment of what general and comparative linguistics have to offer each other can only be reached if there is some consensus about the goals of the linguistic enterprise. The comparative linguist is in search of a picture which mirrors as closely as possible a historical reality, whereas the general linguist is primarily concerned with predicting the unknown. It is far from obvious that 8 Introduction the former’s reconstructions should conform to the latter’s predictions. In the following I shall give a few examples of how these two lines of inquiry do not run parallel. Perhaps the most common objection against a proposed reconstruction which is raised time and again on general grounds, is that a linguistic form is impossible because it does not conform to typological expectations. The classic example is Brugmann’s reconstruction of nasalis sonans in 1876, e.g. in the first syllable of *kmtóm ‘hundred’. Brugmann published his article in a journal of which Curtius had made him co-editor before going on a journey. When the latter read the article after his return, he became so enraged that he dissolved the journal and started a new one, without Brugmann (cf. Pedersen 1962: 293). The new reconstruction has now been part of the communis opinio for over a century. The case of the nasalis sonans is particularly instructive because the new theory soon gained general acceptance. The same cannot be said of the hypothesis that the Indo-European proto-language had no more than a single vowel. It is therefore important to compare the two cases in order to establish the reason for the different treatment. Note that I am not primarily concerned with the correctness of the reconstructions but with their reception by the scholarly community. If we can find out what motivates our colleagues to agree or to disagree, it may be possible to save a lot of time when trying to convince them. There are two types of objection against the reconstruction of a single vowel for Proto-Indo-European. On the one hand, it is claimed that not all of the material can be explained from such a reconstruction. On the other hand, it is argued that there can be no such thing as a language with no more than one vowel. Both arguments have their counterparts in the reconstruction of the Proto-Indo-European syllabic resonants. In the case of the nasalis sonans, there were two factors which rendered the new reconstruction more palatable. While the concept of syllabic nasal was an innovation, the syllabic liquids l and r were familiar from Czech and Sanskrit. The new theory did not therefore affect the idea of syllabicity as a vocalic property but only its distribution. Moreover, the class of possible reconstructed forms was not greatly affected because Brugmann recognized, beside the zero grade vocalism of the syllabic resonants, a reduced grade vocalism which could be invoked for those instances where others might see counter-evidence. It can be argued that the real victory of the Sonantentheorie was eventually achieved by the elimination of the reduced grade. That was a development which took much longer than the acceptance of the nasalis sonans. The reconstruction of a single Proto-Indo-European vowel is a natural consequence of the laryngeal theory. It differs from the Sonantentheorie in two respects. Firstly, it strongly reduces the class of possible reconstructed forms. As General linguistics and Indo-European reconstruction 9 a result of this much higher predictive power, it much more easily generates a class of counter-examples. Secondly, the way out which the reduced grade offered in the case of the Sonantentheorie is blocked by the fact we are now dealing with the vowel system itself. What is remarkable here is not that the new reconstruction of the vowel system met with a lot of opposition, which is only natural, but that it found any acceptance at all. The far-reaching consequences inherent in the new reconstruction of the vowel system render the impact of the typological argument all the more important. It has been claimed that languages with less than two vowels are unattested or even impossible. This objection has been countered by the observation that there is a consensus among specialists of North-West Caucasian languages about the existence of minimal vowel systems, matched by extremely large consonant inventories, in that area. This shows the weakness of the typological approach: it causes a bias toward what is regular, normal, or frequent in the languages of the world and thereby renders the reconstruction of deviant patterns impossible (cf. Kuipers 1968: 78f.). The range of animal species living today would not allow us to reconstruct a dinosaur. The typological argument against the reconstruction of a minimal vowel system for Proto-Indo-European is particularly regrettable because typological evidence could actually be used to support such a reconstruction. According to what probably is the majority view, the original homeland of the IndoEuropeans must be situated in the South Russian steppe. The non-IndoEuropean language family which is closest to that area is precisely the NorthWest Caucasian. If we start from the assumption that the Proto-Indo-European sound system resembled that of its neighbours, with which it may have formed a Sprachbund, the North-West Caucasian system is as close as we can get from a typological point of view. Moreover, we know that the area around Majkop, which is Circassian territory, was a cultural center in the formative years of the Indo-European proto-language. It is therefore easily conceivable that the IndoEuropean sound system originated as a result of strong Caucasian influence. In fact, the typological argument is not only weak and ambiguous, but can even be harmful. It has long been recognized that cognate languages tend to develop along similar lines after the dissolution of their ancestor. The Romance languages of today resemble each other much more closely than any of them resembles Latin. As a consequence, the history of Indo-European reconstruction shows a gradual shift away from the principal languages (cf. Mayrhofer 1983). If Bopp’s Indo-European resembled Sanskrit, and Brugmann’s Indo-European resembled Sanskrit no more than Greek, and Cowgill’s Indo-European resembled Sanskrit and Greek no more than Hittite, it is to be expected that future reconstructions will diverge more widely from our traditional image of what an Indo-European language should look like, and thus move farther away from our typological expectations. 10 Introduction What has just been said must not be taken as a plea against the use of typological evidence. On the contrary, I think that typological considerations are most useful as a heuristic device. They must never take the place of the evidence, however. In practice, the typological argument has too often served as a rationalization of traditional prejudice. Curtius’ reaction to Brugmann’s nasalis sonans is a case in point. I shall give two more examples of such unwarranted generalization about possible sound systems. In Bella Coola, a Salish language, there are words consisting entirely of voiceless consonants, e.g. t’xɷt ‘stone’. When Boas reported about this hitherto unknown phenomenon, his article is said to have been rejected by the editor of a journal because everyone knows that it is impossible to have words without vowels (cf. Hockett 1955: 57). On the basis of my own fieldwork I can testify to the existence of the same word structure in Heiltsuk, an unrelated, Wakashan language, which is also spoken on the Canadian Pacific coast, e.g. qqs ‘eye’. Here again, it appears that the typological argument can indeed be harmful. According to Jakobson’s Kindersprache, Aphasie und allgemeine Lautgesetze, there can be no language without nasal consonants. The fundamental oppositions vowel – consonant and oral – nasal must be present everywhere: “sie sind die einzigen die nirgends fehlen dürfen” (Jakobson 1941: 34). Unfortunately, the existence of consonant systems without nasals has been reliably reported for Quileute and for Duwamish and Snoqualmie, languages which are spoken in the state of Washington and which belong to two different language families (cf. Hockett 1955: 119). This counter-evidence subsequently led Jakobson to replace his “universals” by “near-universals”, without, incidentally, mentioning the languages which forced his theoretical retreat (Jakobson 1962: 526). Here again, typological reasoning had an adverse effect on the progress of linguistics. Since the universal character of the opposition between oral and nasal consonants has been disproved, we may wonder if the same can be done for the opposition between consonants and vowels. This has actually been achieved by Pulleyblank in his analysis of Mandarin Chinese, which is not a minor language. Pulleyblank treats all vowels as syllabic variants of glides with which they alternate (1984: 57). Since the vowels are derived by rules of syllabification, all morpheme structures consist of consonants only. In comparison with this analysis, the reconstruction of a single vowel for Proto-Indo-European looks rather conservative. This raises the question whether our reluctance to admit certain possibilities may be a consequence of the tools we have been accustomed to use. In particular, is it possible that our conception of vowels and consonants is conditioned by our use of the Latin alphabet? Here it may be appropriate to have a look at the Japanese syllabary, which offers an instructive parallel. Unlike the well-known Semitic and Indic scripts, the Japanese syllabary does not offer the possibility to denote a consonant without a following vowel. General linguistics and Indo-European reconstruction 11 Consequently, it is impossible to describe the language in terms of stems ending in a consonant followed by suffixes beginning with a vowel. Thus, the paradigm of the verb ‘to speak’, indicative hanasu, infinitive hanasi, imperative hanase, subjunctive hanasoo, negative hanasanai, cannot be described as a consonantal stem hanas- followed by a variety of suffixes, but only as an alternating stem hanasu, -si, -se, -so-, -sa-, which may be followed by other syllabic elements. This is indeed what is done in traditional Japanese grammar, where the verb belongs to the so-called godan-katuyoo, or five-step conjugation. The problem of notation as an obstacle to progress is not limited to linguistics, as any mathematician can testify. Consider the multiplication of 19 by 44. Accustomed as we are to the system of Arabic figures, we immediately see that 20 times 44 is 880, and when you subtract 44 you get 836, which is the correct answer. But now suppose that you live in Rome, two thousand years ago, then you have to multiply XIX by XLIIII in order to arrive at DCCC’VI. There can be little doubt that this is a more cumbersome procedure. Against this background, we may wonder if the difficulty of analyzing Japanese verbs with a consonantal stem in terms of the syllabary has a parallel in languages with an alphabetic script. In fact, it is not difficult to find examples. Take the English noun house and the verb to house. The latter is derived from the former by voicing the final consonant. It would therefore be appropriate to write the voicing feature as a suffix, if the writing system allowed us to do so. Conversely, the noun use is derived from the verb to use by devoicing the final consonant. Note that the direction of the derivation is different here: we can have a house without housing someone, but we cannot do any housing without a house; however, the use of something presupposes somebody using it, whereas we may use something without invoking the abstract notion of ‘use’. While a traditional analysis must treat these instances in terms of stem alternation, a generative analysis may postulate an underlying suffix. Such a suffix does not necessarily correspond to the suffix which a comparative linguist would reconstruct. The example of house and use brings us to the problem of markedness. It is not always evident which of the two members of a pair must be considered the marked one. Since this may have consequences for linguistic reconstruction, it will be appropriate to look into the origin of the term. The concept of markedness as applied to the meaning of morphological categories can be traced back to Jakobson’s essay Zur Struktur des russischen Verbums, which appeared in 1932. Jakobson’s original example accompanying the introduction of the term is the Russian pair of words telenok ‘calf ’ – telka ‘heifer’ (1932: 75). When he reprinted the article in his Selected Writings, Jakobson replaced these words by osel ‘donkey’ – oslica ‘she-ass’, without, incidentally, drawing the reader’s attention to the fact that he changed his prime example (1971: 4). There are two remarkable things about this. Firstly, it turns out that the example was not very 12 Introduction well chosen. It thus appears that the concept lacks the clarity which should render it applicable in an unambiguous way. Secondly, the ill-chosen example was evidently of no consequence to the theory because it was tacitly replaced by a different one. This does not inspire great confidence in the usefulness of the proposal. Since the Urheber apparently had a hard time making up his mind about the applicability of his theory to his own language, one can hardly blame others for avoiding the concept of markedness as a tool of analysis. To summarize our findings thus far, it appears that there is good reason to be ambivalent about the usefulness of general considerations in linguistic reconstruction. As a heuristic device, a theoretical framework can certainly be helpful, but the negative potential of aprioristic considerations must not be underestimated. Since theory can easily embody the reflection of rationalized prejudice, it is important that comparative work be carried out inductively, as Holger Pedersen knew a long time ago. The accumulated experience of comparative linguistics offers a sound basis for a general theory of language change, which is part of a general theory of language. II We may now examine the hypothesis that the traditional voiced stops of the Indo-European proto-language were actually glottalic. Before the advent of the laryngeal theory, it was generally assumed that the proto-language had the same four series of stops as we find in Sanskrit, e.g. the dental series *t, *th, *d, *dh. When it turned out that the voiceless aspirate was rare and must in a number of cases be derived from an earlier sequence of *t plus a laryngeal consonant, the inductive generalization that no more than three series can be reconstructed for the proto-language left scholars with a typologically anomalous consonant system: *t, *d, *dh. There are two ways out of this intuitively unsatisfactory situation. On the one hand, one may return to the traditional reconstruction of four series of obstruents, in spite of the fact that there is insufficient evidence for the existence of original voiceless aspirates. This possibility does not offer an explanation for the peculiar asymmetry in the attested material. On the other hand, one can try to reinterpret the three series of *t, *d, *dh in such a way as to bring the reconstructed system into agreement with typological expectations. This research strategy invites scholars to look for additional evidence, which might change our views of the proto-language in a more radical way. The first to pursue the latter possibility in print was Holger Pedersen, at the age of 84. Pedersen argued that there are no reliable Indo-European etymologies which point to an initial voiced labial stop *b- (1951: 10-16). Since the voiceless labial stop p- is easily lost in a number of languages, he suggested that ProtoIndo-European *b was originally voiceless and weak, while the traditional voiced aspirate *bh may have developed from a voiceless aspirate. He compared General linguistics and Indo-European reconstruction 13 the interchange of voiced and voiceless stops with the West Armenian consonant shift. The point to be noted here is the primacy of the empirical evidence. Typological considerations only served as a heuristic device for developing an explanatory hypothesis. Pedersen’s article inspired Martinet to propose two years later in a footnote that the Proto-Indo-European voiced stops could be derived from an earlier glottalic series without a labial member (1953: 70). He compared the absence of the labial with the same phenomenon in Proto-Semitic, for which he reconstructed a glottalic series as the origin of the so-called emphatic stops. Here again, typological considerations served as a heuristic device. The problem was posed by the unexpected absence of empirical evidence for the reconstruction of a labial stop. A few years later, Andreev proposed an Indo-European proto-language without distinctive voicedness (1957: 7). He reconstructed voiceless fortes, voiceless lenes, and voiceless aspirates, corresponding to traditional t, d, dh, and suggested that this system is apparently preserved in Hittite. He introduced the incompatibility of fortes and aspirates in the root structure, which he (like Meillet) explained by an assimilation rule, into the discussion of the consonant system. His reinterpretation of the consonant shifts in the separate branches anticipates an argumentation which was put forward much later by the proponents of the glottalic theory. A proposal which looks like an integrated view of the hypotheses put forward by Pedersen, Martinet and Andreev is Swadesh’s theory that ProtoIndo-European and its neighbours had simple, glottalic, and aspirated stops, and that the difference between voiced and voiceless articulation was a matter of local variation (1971: 127). Since this theory was published posthumously, its origin is difficult to determine. Swadesh remarks that the traditional IndoEuropean voiced stops are equivalent to the glottalic series of other language families with respect to sound symbolism (1971: 219). Twenty years after the publication of Martinet’s suggestion that we may have to reconstruct glottalic stops for Proto-Indo-European, Gamkrelidze and Ivanov proposed the same (1972: 16), again on the basis of Pedersen’s reasoning. Their proposal became much more widely known, probably because it was put forward time and again in different places. They explained the absence of roots with two glottalic stops by a dissimilation rule (1973: 153). They also reformulated Grassmann’s Hauchdissimilationsgesetz as a Proto-Indo-European rule of allophonic variation (1980: 30-32). Here the primacy of the empirical evidence has been lost: the glottalic theory is not used to explain Grassmann’s law, but Grassmann’s law is adapted in order to serve as evidence for the glottalic theory. It seems to me that Latin fīdō ‘I trust’ < *bheidh- suffices to show that the argument cannot be used. 14 Introduction Around the same time, a similar proposal was put forward by Hopper, who adduced not only the absence of *b and the root structure constraints, but also the absence of glottalic stops from inflectional affixes (1973: 157). Here again, theoretical considerations evidently provided an obstacle to observation of the material, as is clear from the comparison of Latin quod with Old High German hwaz ‘what’, on the basis of which we have to reconstruct a Proto-IndoEuropean neuter ending *-d. On the basis of the proposals by Pedersen and Andreev, Rasmussen derived traditional *t, *d, *dh from earlier *T, *t, *d, where the first represents any emphatic stop, however phonetically realized: glottalized, pharyngealized, or just stronger (1974: 11). The same reconstruction is implied in Illič-Svityč’s Nostratic dictionary (1971: 147). The problem with this hypothesis is that there is no reason to assume an emphatic or otherwise strong character for a glottalic series. There are many varieties of glottalization, some of them weak, others strong. The relatively weak character of glottalization in Georgian and Armenian is evident from the fact that we often find glottalic rather than aspirated stops in loanwords from Russian. This suggests that we have aspirated fortes and glottalic lenes in these languages. In Avar, a North-East Caucasian language, there is an opposition between tense and lax voiceless consonants which is independent of the opposition between plain and glottalic stops and affricates, e.g. k, k:, k’, k:’. Moreover, there is also an opposition between geminate and single tense consonants, so that we have e.g. xɷásel ‘winter’ vs. t’ás:a ‘from above’ vs. xɷás:s:ab ‘special’ (cf. Ebeling 1966: 63). Thus, it appears that unwarranted generalization on the basis of theoretical considerations can easily interfere with observation of the facts and lead one astray in linguistic reconstruction. This can block scholarly progress for many years. Haudricourt reports (1975: 267) that as early as 1948 he arrived at the conclusion that the traditional voiced stops of the Indo-European protolanguage were in fact glottalic and that the original pronunciation has been preserved in East Armenian. His argumentation was based on the types of phonetic development attested in the Far East. The negative attitude of Bloch and Kuryłowicz toward his view apparently kept him from publication. If Haudricourt, Pedersen, Martinet, Andreev and Swadesh had met at a conference in the late ’forties, the glottalic theory might have become popular a generation earlier than it actually did. I conclude that the typological argument has too often been invoked as a constraint on linguistic reconstruction rather than as a device to broaden the horizon of possibilities. As a result, our reconstructions tend to have a strong bias toward the average language type known to the investigator. The more deviant the structure of the proto-language actually was, the stronger the bias and the larger the difference between reality and reconstruction we should expect. We must therefore first and foremost pay attention to the comparative General linguistics and Indo-European reconstruction 15 evidence, which remains the ultimate basis for choosing between alternative options in linguistic reconstruction. It is remarkable that the comparative evidence has largely been left out of consideration in the discussion of the glottalic theory. III Glottalization is found in five out of the ten surviving branches of IndoEuropean, viz. Indic, Iranian, Armenian, Baltic, and Germanic. This is not the place to reconsider the comparative value of the evidence in the separate branches, which is very uneven (cf. K075). My point is methodological: can we establish the circumstances under which certain facts are admitted as evidence for a reconstruction? The answer to this question is far from obvious. There are two varieties of stød in Danish. As a rule, standard Danish stød appears in monosyllabic words which have pitch accent 1 in Swedish and Norwegian. Though the distribution of the stød has partly been obscured by analogical developments, it seems clear that it developed from a falling tone movement. I shall leave the standard Danish stød out of consideration in the following. The so-called vestjysk stød is an entirely different phenomenon because it is characteristic of originally polysyllabic words, which have accent 2 in Swedish and Norwegian. It cannot possibly be connected with the Jylland apocope because it is also found in the northeastern part of vestfynsk dialects, where the apocope did not take place. While the vestjysk stød is clearly linked to a following plosive which represents an earlier voiceless stop, it does not represent original gemination because it distinguishes e.g. the verbs dampe [dam’b] ‘to steam’, kante [kan’d] ‘to border’ from the nouns damp [damb] ‘steam’, kant [kand] ‘edge’, which never had a geminate (cf. Ejskjær 1990: 64). As the glottalization in the infinitive vente [ven’d] ‘to wait’ is absent from the imperative vent [vend] ‘wait!’ (Ejskjær 1990: 65), it looks like a feature of the following stop which was lost in word-final position. This leads us to consider the possibility that it may reflect some kind of Proto-Germanic glottalization. In his monograph on the vestjysk stød, Ringgaard concludes that “the v-stød is only found immediately before the plosives p, t, k, and that it is found wherever these stand in an original medial position, following a voiced sound in a stressed syllable. The exceptions to this are certain types of loan-words from a later period” (1960: 195). He dates the rise of the vestjysk stød to the 12th century because it is characteristic of “all then existing medial plosives” (1960: 199). The view that the vestjysk stød is a spontaneous innovation of the westernmost dialects of Danish, which Jespersen had in fact proposed almost half a century earlier already (1913: 23), can hardly be called an explanation. Moreover, it does not account for the vestjysk stød in the isolated pocket of dialects on the island 16 Introduction of Fyn, which suggests that it is a retention rather than an innovation. The hypothesis of a local origin also neglects the parallel development of preaspiration in Icelandic. Preaspiration is not only found in Icelandic, but also in Faroese, Norwegian, and the Gaelic dialects of Scotland. Phonetically, the preceding vowel is cut short and continued as a whisper, while a preceding resonant is partly or wholly unvoiced. The distribution of preaspiration in Icelandic is the same as in the Norwegian dialect of Jæren (cf. Oftedal 1947). We can therefore conclude that it is “an example of a feature taken to Iceland by the original settlers” (Chapman 1962: 85). Marstrander has argued that the preaspiration in Scottish Gaelic is due to a Norse substratum (1932: 298). He advances the hypothesis that the Norwegian preaspirated stops represent a retention of the clusters hp, ht, hk, which developed into geminates elsewhere (1932: 302). This theory implies three developments, viz. ht > tt in East Norse, tt > ht in West Norse, and t > ht in West Norse in those positions where the preaspirated stop does not reflect a cluster, e.g. Icelandic epli ‘apple’, vatn ‘water’, mikla ‘to increase’, hjálpa ‘to help’, verk ‘work’. Here the preaspirated plosives correspond to the traditional voiced stops of the Indo-European proto-language. Both the vestjysk stød and the preaspiration receive a natural explanation if we assume that early Proto-Germanic possessed a series of preglottalized voiced stops ’b, ’d, ’g (cf. K075: 196, K102: 8). Devoicing yielded a series of late ProtoGermanic sequences ’p, ’t, ’k, the glottal element of which was lost under various conditions. Weakening of the glottal stop in West Norse yielded preaspiration while its assimilation to the following plosive gave rise to a series of geminates in East Norse, with the exception of Danish, where the sequences were subject to lenition and the glottal stop was preserved in the vestjysk dialects. It is difficult to escape the impression that the reluctance of earlier investigators to take the vestjysk stød and the Icelandic preaspiration seriously as comparative evidence in the reconstruction of Proto-Germanic deprived them of an insight which could have changed our view of Proto-Indo-European. What was the cause of their restraint? What kept them from regarding preglottalization and preaspiration as evidence on a par with other features? Was it the Latin alphabet which constrained their thinking? Apart from the straightforward explanation of the vestjysk stød and the Icelandic preaspiration, the reconstruction of Proto-Germanic preglottalized stops has the advantage of accounting in a principled way for the existence of several layers of gemination, which can now be viewed as retentions rather than innovations (cf. K102: 7). Firstly, it is possible that the unexplained gemination in Swedish, e.g. in vecka ‘week’, droppe ‘drop’, skepp ‘ship’, reflects a dialect which escaped an early loss of the glottal stop, in contrast with Old Norse vika, dropi, skip, Old English wice, dropa, scip. Secondly, mp, nt, nk yielded pp, tt, kk in the General linguistics and Indo-European reconstruction 17 larger part of Scandinavia. This development becomes understandable if we assume that the nasal consonant was devoiced by the preaspiration of the following plosive and subsequently lost its nasal feature. Thirdly, *k was geminated before *j and *w, e.g. in Old Norse bekkr ‘brook’, røkkr ‘dark’. Similarly, *t was geminated before *j in a limited area, e.g. Swedish sätta ‘to set’. (West Germanic geminated all consonants except r before *j and is therefore inconclusive.) Fourthly, the stops p, t, k were geminated before l and r in West Germanic, e.g. English apple, bitter, cf. Gothic baitrs. The same development is found sporadically in Scandinavia, which suggests that we are dealing with the loss of an archaic feature rather than with an innovation. Here again, the geminate may have originated from the assimilation of a glottal stop to the following plosive. In fact, the evidence for Proto-Germanic preglottalized stops is not limited to Scandinavian, but can also be found in English and German. It is common knowledge that standard English inserts a glottal stop before a tautosyllabic voiceless plosive, e.g. sto’p, tha’t, kno’ck, wa’tch, also lea’p, soa’k, hel’p, pin’ch (cf. Brown 1977: 27). There is no reason to assume that this is a recent phenomenon. The High German sound shift yielded affricates and geminated fricatives, e.g. Old High German pfad ‘path’, werpfan ‘to throw’, offan ‘open’, zunga ‘tongue’, salz ‘salt’, wazzar ‘water’, kind, chind ‘child’, trinkan, trinchan ‘to drink’, zeihhan ‘token’. These reflexes suggest a complex articulation for the Proto-Germanic voiceless plosives from which they developed. The origin of the gemination is unexplained in the traditional doctrine. If we start from the assumption that the Proto-Germanic plosives were preceded by a glottal stop which is preserved in the vestjysk stød and the English glottalization, the High German sound shift can be explained as a lenition of the plosives to fricatives with a concomitant klusilspring of the preceding glottal stop. Note that the High German sound shift has a perfect analogue in the English dialect of Liverpool, where we find e.g. [kx] in can’t, back (Hughes and Trudgill 1987: 66), which again remains unexplained in the traditional doctrine. Thus, it appears that there is a whole range of phenomena which receive a natural explanation when we assume that glottalization is ancient in Germanic. The methodological question is: why have scholars been reluctant to identify the vestjysk stød with the English glottalization, which according to Ringgaard gives the same auditory impression and apparently has the same articulation, as a historical reality which may have been inherited from the proto-language? Is there an implicit assumption that unwritten features must not be ancient? Is this the same factor which made Curtius reject Brugmann’s nasalis sonans, in spite of the fact that we have a syllabic nasal in standard English words such as button and in the standard German infinitive ending of most verbs, e.g. leiten ‘to lead’, where both examples end in [tn]? Is it all the result of our Latin upbringing, 18 Introduction which Jespersen blamed for our lack of insight into the grammar of modern English? IV It will be clear from what has been said that I am not particularly impressed by the contribution of theoretical reasoning to historical linguistics. Both Jespersen and Pedersen emphasized time and again that linguistics is an inductive enterprise, and I agree whole-heartedly. This does not mean that the comparative linguist can disregard what is going on in general linguistics, however. It rather means that we must look at those branches of linguistics which deal with language change in progress. Language is the interface between society and the individual, and sociolinguistics is the area of research where we can expect results which may be of immediate relevance to linguistic reconstruction. Rapid linguistic change in bilingual communities of nomadic traders and ethnically mixed groups offers a test-case for historical linguistics. There is no reason to assume that the sociolinguistic conditions of prehistoric linguistic development were very different from what can be observed today among comparable groups. The remarkable spread of the Indo-European languages was determined by specific social and economic circumstances. It presupposes that a number of people moved from their original homeland to a new territory. As is now generally recognized, the domestication of the horse played a crucial role in the increase of physical mobility. However, the Indo-European expansions required not only the migration of Indo-Europeans, but also the adoption of IndoEuropean languages by local populations. This implies that a large number of people must have found it expedient to adopt the language of the intruders. As Mallory has pointed out, “pastoral societies throughout the Eurasian steppe are typified by remarkable abilities to absorb disparate ethno-linguistic groups. Indo-European military institutions may have encouraged membership from local groups in the form of clientship which offered local populations greater advantages and social mobility” (1989: 261). This must have been the decisive factor in the spread of the Indo-European languages. When we look at language interference in bilingual communities, it appears that there is a marked difference in the ease of linguistic borrowing between grammar and lexicon, between bound and free morphemes, and between verbs and nouns. As a result, the older strata of a language are better preserved in the grammatical system than in the lexical stock, better in morphology than in phonology or syntax, better in verb stems and pronouns than in nouns and numerals. The wide attestation of the Indo-European numerals must be attributed to the development of trade which accompanied the increased mobility of the Indo-Europeans at the time of their expansions. Numerals do General linguistics and Indo-European reconstruction 19 not belong to the basic vocabulary of a neolithic culture, as is clear from their absence in Proto-Uralic and from the spread of Chinese numerals throughout East Asia. The inequality between different parts of the language in linguistic borrowing is of particular importance when we are dealing with distant affinity. In a beautiful and convincing article which appeared a number of years ago (1988), Michael Fortescue has demonstrated on the basis of case suffixes, pronouns and verbal morphology that Eskimo and Aleut are genetically related to Yukagir, which is most probably related to the Uralic language family. His reconstructions support the possibility that Tungus and Japanese also belong to the same language stock. It is clear that such affinity could never be demonstrated by the mere comparison of words. In a study of the earliest contacts between the Indo-European and Uralic language families (1986), Rédei lists 64 words which were supposedly borrowed from Indo-European into Uralic at an early date. The material is divided into three groups: 7 Indo-European words which are attested in both Finno-Ugric and Samoyedic, 18 Indo-European or Indo-Iranian words which are attested in Finno-Ugric but not in Samoyedic, and 39 Indo-Iranian words which are not found either in Ugric or in Samoyedic. Now it turns out that the number of verbs in the oldest material is too large to support the hypothesis that they were borrowed: verbs constitute 43% of the first group, 28% of the second group, and 5% of the third group. This is strong evidence for the thesis that the oldest layer was in fact inherited from an Indo-Uralic proto-language. Though the material is very small, the case for an original genetic relationship is particularly strong because we are dealing with basic verbs, meaning ‘to give’, ‘to wash’, ‘to bring’, ‘to drive’, ‘to do’, ‘to lead’, ‘to take’ (cf. K112). Moreover, it is difficult to see how Proto-Indo-European words could have been borrowed into Proto-Uralic if the Indo-Europeans lived in the South Russian steppe when the ancestors of the Finno-Ugrians and the Samoyeds lived on the eastern side of the Ural mountains. The earliest contacts between Indo-European and Uralic languages must probably be identified with the eastward expansion of the Indo-Iranians and the simultaneous spread of the Finno-Ugrians to the southwest. Thus, it appears that we do not need a large number of obvious cognates, which cannot be expected in the case of distant linguistic affinity, in order to establish genetic relationship between languages. What we need to find are morphological correspondences and a few common items of basic vocabulary because these are the elements which are least likely to be borrowed. We can then try to match the linguistic evidence with what can be gathered from anthropological and archaeological sources. In my view, the last decade has brought decisive proof of genetic relationship between the whole range of languages from Indo-European to Eskimo. The next step should comprise an establishment of chronological layers in the material and a specification of the 20 Introduction connections with the Altaic language family. The role of general linguistics in this enterprise is to provide an idea of what can be expected in linguistic development, not by theoretical reasoning but by inspection of what actually happens in situations of language contact. Language is a social phenomenon, and linguistic change must be examined in its social and historical context. NOTE This is the revised text of a paper read at the Institute of general and applied linguistics, University of Copenhagen, on December 2, 1993. ON RUSSENORSK The concept of mixed language has recently gained some popularity, to my mind for no good reason. It is unclear how a mixed language can be distinguished from the product of extensive borrowing or relexification. I therefore think that the concept only serves to provoke muddled thinking about linguistic contact and language change. Note e.g. that Munske adduces German as an example “because the author is a professor of German linguistics and because the phenomenon of language mixing can be explained better in relation to a language on which a large amount of research has been done than, for example, in relation to pidgin and creole languages” (1986: 81). Peter Bakker and Maarten Mous claim that “extreme borrowing never exceeds roughly 45% of the lexicon, whereas in some of the mixed languages discussed the proportion of ‘foreign’ lexical elements is closer to or over 90%, and this figure is the same whether one counts types or tokens. There do not seem to be languages with a proportion of borrowed items between 45% and 90%, so that there is no continuum between languages with heavy borrowing and mixed languages” and that “in the mixed languages most of the core vocabulary tends to be foreign” (1994: 5). When we look at the short text in Bakker’s prime example of a mixed language, Michif (1994: 28-30), we find the following distribution of French and Cree items: – French elements: un vieux ‘an old’, un matin ‘a morning’, ses pièges ‘his traps’, une tempête ‘a storm’, pas moyen ‘no way’ (2×), son shack ‘his cabin’ (2×), le vieux ‘the old’, d’un gros arbre ‘at a big tree’, une bonne place ‘a good place’, le loup de bois ‘the timber wolf ’, le loup ‘the wolf ’ (4×), sa bouche ‘his mouth’, son bras ‘his arm’, dans la queue ‘by the tail’, par la queue ‘by the tail’. – Cree stems: ‘trap’ (2×), ‘wake up’, ‘be sick’ (2×), ‘want-go-see’, ‘leave’, ‘be busy’, ‘bad-weather’, ‘find’ (2×), ‘be lost’, ‘walk’ (2×), ‘play out’, ‘sit upright’, ‘die’ (2×), ‘think of ’, ‘see’, ‘run’, ‘wait for’, ‘look at’, ‘come-thus-run’, ‘sit’, ‘open’ (2×), ‘wanttake’, ‘push forward’, ‘take from’, ‘pull inside out’, ‘run back go home’. – Cree words: ‘and’ (3×), ‘still’, ‘this’ (4×), ‘meantime’, ‘there’ (2×), ‘here’, ‘it is said’, ‘that’ (2×), ‘where’, ‘again’. – Cree affixes: 50 instances (12 prefixes and 38 suffixes). It is clear that Bakker’s prime example does not fulfil his own criteria of mixed languages and that we are simply dealing with heavy borrowing of French nouns into Cree. French articles and possessives are treated as prefixes which were borrowed together with the head noun. 22 Introduction In the following I shall discuss another example of a mixed language, viz. Russenorsk, a pidgin which was spoken by Norwegian and Russian fishermen along the Arctic coast of Norway in the 19th century. Our major source of information on this language is Broch & Jahr 1984; other important contributions are Lunden 1978 and Peterson 1980. Note that Russenorsk was a ‘seasonal’ language, not used continuously throughout the year but only during part of the summer fishing season (Lunden 1978: 213). It cannot therefore be compared with a natural language but illustrates the process of language mixing in a situation where creolization never had a chance. Following Olaf Broch, Lunden lists three salient features of Russenorsk which are frequently quoted in the literature: (1) the use of moja and tvoja as 1st and 2nd sg. personal pronouns, (2) the use of the preposition po/på as a general marker of oblique relations, (3) the development of a verbal marker -om, e.g. kapitan paa kajuta slipom ‘the captain is asleep in his cabin’. Since these features serve as a kind of shibboleth, they may reflect “a case of speaking “the way foreigners speak”, rather than speaking “the other man’s language””, as Lunden puts it (1978: 215). It follows that we must distinguish between a person speaking his own language, a person simplifying his own language in order to make himself understood, and a person trying to use his interlocutor’s language. We should therefore expect to find six components of Russenorsk, depending on the nationality of the speaker and his three types of linguistic performance. The effort which goes into “speaking the way foreigners speak” beside “speaking the other man’s language” accounts for the fact that both partners may believe that they are actually speaking each other’s language. I shall now test this hypothesis by analyzing a dialogue recorded by A.W.S. Brun, cited and translated by Peterson (1980: 253f.), and reproduced by Broch & Jahr (1984: 130f.). The speakers are abbreviated as Nor[wegian] and Rus[sian]; unambiguous Norwegian and Russian words are indexed by subscript N and R while German, Dutch, English and ambiguous words remain unmarked (in accordance with Peterson 1980); the orthography follows Broch & Jahr (1984) except for æ and aa, which are replaced by e and o. Nor: Kjøp|N IN seika, treska|R, tiksa oN balduska? “Are you buying pollack, cod, haddock, and halibut?” The speaker uses his own language, except for the fact that the fish names are apparently language-independent. Rus: DaR, daR – moja|R kopom|R altsamma|N, davai|R po skip|N kom|N. “Yes, yes – I’ll buy all of it, come on board.” The speaker adapts his own language moja kopom and switches to Norwegian altsamma, po skip kom. On Russenorsk 23 Nor: Spasiba|R! har|N IN mokka|R, har|N IN groppa|R? “Thank you! Do you have flour, do you have grain?” The speaker uses his own language, except for the word of thanks and for the names of the commodities he wants to purchase. Rus: DaR, daR! Davai|R po skip|N kom|N, brat|R, po tjei|R driki|N. “Yes, yes! Come on board, brother, drink some tea.” The speaker clearly tries to speak Norwegian, though the interjections da, davai, brat are Russian. Nor: Blagdaru|R pokorna|R! Kok|R tvoja|R betalom|N for|N seika? “I humbly thank you! What are you paying for pollack?” The speaker now tries to adopt the simplified language of his interlocutor, tvoja betalom echoing moja kopom, but the main verb and focus of the message is still Norwegian betalom for seika. Rus: Pet|R pudof|R seika 1 pud|R moki|R. “Five poods of pollack for one pood of flour.” This is Russian. Nor: Kor|N iN tykje|N eN deN laga|N? IN moN gjer|N deN billiar|N! “How the hell is that figured out? You have to make it cheaper!” This is pure Norwegian. Rus: Kak|R sprek? Moje|R niet|R forsto|N. “What did you say? I don’t understand.” Again, the speaker adapts his own language and uses verbs from the language of his interlocutor. Nor: Dorgo|R, dorglo|R Rusmain|N – prosjai|R! “Expensive, expensive, Russian – goodbye!” The speaker uses Russian words without any sentence structure. Rus: Nietsjevo|R! sjetiri|R – gall|N! “Okay! four – and a half!” This is Russian, except for the focus of the message, which is in Norwegian. Nor: Davai|R fir|N – nietsjevo|R verrigod|N. “Make it four, okay, good.” Apart from the interjections davai and nietsjevo, this is Norwegian. Rus: Njet|R, brat|R! Kuda|R moja|R selom|N desjevli|R? Grot djur|N mokka|R po Rusleien|N dein|N orN. “No, brother! Where can I sell it cheaper? Flour is very expensive in Russia this year.” The speaker adapts his own language and substitutes the Norwegian verb, then switches to Norwegian in the second sentence. Nor: Tvoja|R niet|R sainferdi|N sprek. “You’re not telling the truth.” The speaker imitates the simplified language of his interlocutor but the focus of the message is in Norwegian. Rus: Jes, grot sainferdi|N, moja|R niet|R lugom|N, djur|N mokka|R. “Yes, it’s very true, I’m not lying, flour is expensive.” The speaker tries to answer in Norwegian, adapts his own language and uses Norwegian words in the three foci of the message. 24 Introduction Nor: Kak|R tvoja|R kopom|R – davai|R fir|N pud|R; kak|R tvoja|R niet|R kopom|R – soN prosjai|R! “If you want to buy – four poods; if you don’t want to buy – then, goodbye!” The speaker imitates the simplified language of his interlocutor, but the focus of the message fir is still Norwegian. Rus: NoN, nietsjevo|R brat|R, davai|R kladi|R po dek|N. “Well, okay brother, put the fish on the deck.” This is essentially a Russian sentence. When we evaluate the evidence, it is clear that there is a substantial difference in linguistic behavior between the two parties of the dialogue. The Norwegian uses his own language; when he adopts simplified Russian expressions from his interlocutor, the focus of the message always remains Norwegian. The Russian on the other hand simplifies his own language for the sake of his interlocutor and switches to Norwegian all the time, the only exceptions being his first offer pet pudof seika 1 pud moki and his final consent nietsjevo brat, davai kladi po dek. There is no mixed language here but a dialogue between a Russian speaking foreigners’ talk and limited Norwegian and a Norwegian speaking his own language and imitating the Russian’s foreigners’ talk. The focus of the messages is always in Norwegian, whether the speaker is Norwegian or Russian. It follows that Russenorsk is a variant of Norwegian with an admixture of Russian foreigners’ talk. While the concept of mixed language seems to have originated from underanalysis of linguistic data, the putative grammar of Russenorsk appears to result from linguistic overanalysis. The alleged nominal suffixes -a and -i (Broch & Jahr 1984: 43f., 63) are simply the Russian sg. and pl. endings which were borrowed as part of the names of the merchandise. When a Norwegian asks in Russian foreigners’ talk: Nogoli dag tvoja reisa po Archangel otsuda|? “How many days did you travel from/to Archangel to/from here?” (nogoli dag < mnogo li dag, R. otsuda ‘from here’), the Russian says in Norwegian: Ja po madam Klerck tri daga lige ne. “I lay three days at Mrs. Klerck’s.”, with the regular ending -a after the Russian numeral tri (Broch & Jahr 1984: 113, 118). The adjectival -a is the Russian feminine and unstressed neuter ending which was borrowed as part of the adjectives. The verbal ending -om represents the Scandinavian hortative ending -om, not only because the preceding verb stem is usually Germanic and because over 50% of the instances are introduced by davai or værsego (Broch & Jahr 1984: 47), but especially because it is pronounced [um], as is clear from the manuscripts, and cannot therefore be of Slavic origin. There are only four Russian verbs in -om, viz. kopom ‘buy’, robotom ‘work’, smotrom ‘see’, kralom ‘steal’; the isolated form podjom ‘let’s go’ does not count because it is not attested in a sentence. Russian verb forms are usually On Russenorsk 25 imperatives or infinitives; the form vros ‘(you) lie’ is not 2nd sg. but uninflected: moja njet vros (lygom) ‘I do not lie’. Thus, there is no trace of Russian grammar in the language. I conclude that Russenorsk is a variant of Norwegian with an admixture of Russian foreigners’ talk and elements from the native language of the speaker. The concept of mixed language is misleading because there is a fundamental asymmetry between the two parties in the dialogue, both of whom essentially speak Norwegian. There is no Russian element in the grammar, which is Norwegian, though not limited by the standard language but full of pragmatic variation, especially topicalization. The attested material illustrates the regular mechanism of language change through imperfect learning. THE ORIGIN OF THE GOTHS Witold Mańczak has argued that Gothic is closer to Upper German than to Middle German, closer to High German than to Low German, closer to German than to Scandinavian, closer to Danish than to Swedish, and that the original homeland of the Goths must therefore be located in the southernmost part of the Germanic territories, not in Scandinavia (1982, 1984, 1987a, 1987b, 1992). I think that his argument is correct and that it is time to abandon Iordanes’ classic view that the Goths came from Scandinavia. We must therefore reconsider the grounds for adopting the latter position and the reasons why it always has remained popular. The reconstruction of Gothic history and the historical value of Iordanes’ Getica have been analyzed in detail by Peter Heather (1991: 3-67). As he points out about this prime literary source (p. 5): “Two features have made it central to modern historical reconstructions. First, it covers the entire sweep of Gothic history. [...] Second, there is a Gothic origin to some of the Getica’s material, which makes it unique among surviving sources.” Iordanes’ work draws heavily on the lost Gothic histories of Ablabius and Cassiodorus, who “would seem to have been in the employ of Gothic dynasts and had to produce Gothic histories of a kind that their employers wished to hear” (Heather 1991: 67). As to the origin of the Goths and their neighbours, the Gothic migrations and the great kings of the past, oral history is the most likely source of the stories. This material must therefore be handled with particular care: “Oral history is not unalterable, but reflects current social configurations; as these change, so must collective memory” (Heather 1991: 62). It appears that Iordanes knew of several alternative accounts of early Gothic history, and Heather concludes (1991: 66): “There was thus more than one version of Gothic origins current in the sixth century. Jordanes, as we have seen, made his choice because he found written confirmation of it, but this is hardly authoritative: the Scandinavian origin of the Goths would seem to have been one sixth-century guess among several. It is also striking that Jordanes’ variants all contained islands: Scandinavia, Britain, ‘or some other island’. In one strand of Graeco-Roman ethnographic and geographic tradition, Britain, Thule, and Scandinavia are all mysterious northern islands rather than geographical localities. ‘Britain’ and ‘Scandinavia’ may well represent interpretative deductions on the part of whoever it was that recorded the myths. The myths themselves perhaps referred only to an unnamed, mysterious island, which the recorder had then to identify. The Scandinavian origin-tale would thus be similar to much else in the Getica, depending upon a complex mixture of material from Gothic oral and GraecoRoman literary sources.” 28 Introduction If we are to maintain continuity between the Baltic Gutones of the 1st and 2nd centuries and the Pontic Goths of the 3rd and 4th centuries, this only reflects the tradition of the ruling clans (cf. Wolfram 1979: 6-7). The historical evidence suggests that the Scandinavian Goths came from the south across the Baltic Sea rather than the other way round (cf. Hachmann 1970: 454-457 and 465). The Lithuanian name Gudai ‘Byelorussians’ < *-dh- has nothing to do with the Goths < *-t- but must be derived from Prussian gudde ‘woods’, like the Polish place names Gdańsk and Gdynia (cf. Fraenkel 1950: 64). There is no archaeological evidence for a large-scale migration of Goths from the Baltic to the Black Sea (cf. Heather 1991: 6 and Hachmann 1970: 467). In fact, there are several reasons why such a migration is highly unlikely. First of all, there is a clear discontinuity between the Przeworsk culture in Poland and the Černjahov culture in the Ukraine which are identified with the Goths before and after the migration, respectively (see the map of Green 1998: xiv). The only reason to assume that the Goths followed the rivers Bug or San and Dniestr is that “the terrain did not offer many alternatives between a common starting-point and a shared goal” (Green 1998: 166). Secondly, the territory between these two areas north of the Carpathian mountains is precisely the homeland of the Slavs, who do not appear to have stirred before the arrival of the Huns in the fourth century. This can hardly be reconciled with a major migration of Goths through their territory. Thirdly, the periodic exposure to severe stress in the fragile borderland communities of the steppe prompted westward population movements toward areas of more stable climatic conditions. An eastward migration of Goths from the richer upland forest into the poorer lowland steppe was both unmotivated and difficult to realize against the natural forces to be encountered. Fourthly, the expected direction of a migration is toward more developed areas where life seems to be better, which in the present context means toward the nearest border of the Roman Empire. We would therefore expect the Goths to move to the south through the Moravian Gate toward the Danube, as did the Slavs a few centuries later. Fifthly, there is little reason to assume that the Goths behaved differently from the Burgundians, the Vandals, the Marcomanns and the Langobards, all of whom crossed the upper Danube at some stage. It therefore seems probable to me that the historical Goths followed the course of the Danube downstream and entered the Ukraine from the southwest. The Gepids may have lagged behind on this journey, which accounts for Iordanes’ etymology of their name (cf. Heather 1991: 5). Putting the pieces together, I think that the most likely chain of events is the following. The Gutones, like their East Germanic brethren, moved south toward Italy and the riches of the Roman Empire until they reached the river Danube. They may have adopted the speech of Alemannic tribes which had arrived there from the west, where these had been in close contact with the Romans for a longer period of time. It is possible that Gothic ethnogenesis actually took place The origin of the Goths 29 in Lower Austria when East Germanic tribes from the north met with West Germanic tribes from the west and, having been prevented from entering the Roman Empire in large numbers, joined forces in their quest for a place to cross the lower Danube. This scenario is well-motivated in terms of pressures and attractions. It renders the southern origin of the Gothic language compatible with the northern origin of the name. The ‘Gothicization’ of large numbers of non-Goths was not brought about by “the predominance of ‘true Goths’” (Heather 1991: 327) but by the absence of major linguistic differences between the Germanic tribes of the 2nd century. It is only to be expected that the most prestigious Germanic dialect was spoken close to the border of the Roman Empire and largely taken over by the newcomers. The Gothic majority did not exist at the outset but came into being as a result of the process of assimilation as the groups adapted to one another. The scenario outlined here has the additional advantage of accounting for a number of peculiar characteristics of the Gothic language in comparison with its closest relatives. Gothic phonology resembles that of Latin and Romance more than that of the other Germanic languages (cf. K102: 8-9 and K138: 54). Though Gothic is more archaic than its sisters, its morphology appears to have been regularized to a large extent. The Latin suffix -ārius was evidently productive in Gothic bokareis ‘scribe’, laisareis ‘teacher’, liuþareis ‘singer’, motareis ‘toll-taker’, sokareis ‘disputer’. The Gothic words siponeis ‘disciple’, kelikn ‘tower’, alew ‘oil’, lukarn ‘lamp’ were probably borrowed from the Celts in Moravia (cf. Green 1998: 156-158), which explains their limited distribution in Germanic. The word for ‘vinegar’ is of particular interest because it has seven different variants in Germanic (cf. Wollmann 1990: 526-542): 1. Gothic aket, akeit; 2. Swiss German (Wallis) achiss; 3. Old High German ezzih; 4. Middle Low German etik; 5. Middle Dutch edic; 6. Old English eced, Old Saxon ekid; 7. Icelandic edic, Swedish ättika, which were apparently borrowed from Low German. It is clear that the Gothic word came from Alemannic in the 1st century before viticulture spread to the Palatinate and the middle Rhine in the 2nd century (cf. Wollmann 1990: 540). The words Kreks ‘Greek’ and dat.pl. marikreitum ‘pearls’ also betray the influence of an Upper German dialect without voiced obstruents (cf. K102: 9). Furthermore, Greek words usually appear in their Latin form in Gothic (cf. especially Jellinek 1926: 179-183 and 188-194), which points to a western origin of the Goths, e.g. aipistula ‘letter’ (but aipistaule ‘Pauline epistle’), drakma 30 Introduction ‘drachma’, paurpura ‘purple’, gen.sg. sinapis ‘mustard’, dat.pl. Rumonim ‘Romans’, Saurim ‘Syrians’, also aikklesjo ‘congregation’, aiwaggeljo ‘gospel’, aiwaggelista ‘evangelist’, diabulus ‘devil’ (but diabaulus in St. John), Marja ‘Mary’ (but Maria in St. Luke), and Iesus Xristus. It seems to me that gen.pl. skaurpjono ‘scorpions’ almost suffices to show that the Goths entered the Balkans from the west, not from the north. Most important is that Greek o-stems are inflected as u-stems in Gothic, e.g. Iudaius ‘Jew’, gen.sg. -aus, dat.pl. -um, acc.pl. -uns, but as i-stems in nom.pl. Iudaieis, gen.pl. -e (Jellinek 1926: 108), which can only be explained by Latin transmission. Other pieces of evidence are cultural loans such as aurali ‘napkin’ and kubitus ‘reclining (company) at a meal’ and loan translations, e.g. armahairtei ‘mercy’, which were taken from Latin orarium, cubitus, misericordia, not from their Greek equivalents. A final point to be noted is that Baltic loanwords from Gothic were transmitted through Slavic (cf. Stender-Petersen 1927: 134 and Green 1998: 172-174), which suggests that the Balts never had direct contact with the Goths but were separated from them by the Slavs. C.C. UHLENBECK ON INDO-EUROPEAN, URALIC AND CAUCASIAN In his early years, C.C. Uhlenbeck was particularly interested in the problem of the Indo-European homeland (1895, 1897). He rejected Herman Hirt’s theory (1892) that the words for ‘birch’, ‘willow’, ‘spruce’, ‘oak’, ‘beech’ and ‘eel’ point to Lithuania and its immediate surroundings and returned to Otto Schrader’s view (1883, 1890) that the original homeland must rather be sought in southern Russia and may have included some of the later Germanic and Iranian territories. It is clear that the Mediterranean region and the area around the North Sea can safely be excluded because the arrival of the Indo-Europeans was comparatively recent here, as it was in Iran and the Indian subcontinent. It is difficult to be more specific within the limits of central and eastern Europe and central Asia. Uhlenbeck was impressed by the lexical correspondences between Indo-European and Semitic which had been adduced in favor of an eastern homeland but pointed out that borrowings from Semitic may have reached the Indo-Europeans through an intermediary. He agrees that the Indo-European words for trees and animals point to a moderate climate but questions the possibility of a more specific localization as well as the concept of homeland itself. Uhlenbeck identifies the Slavic word for ‘dog’ pьsъ with the Indo-European word for ‘livestock’ *pekʄu and its original meaning as ‘domestic animal’. Unlike Hirt (1895), he recognizes that the Indo-Europeans were pastoralists before they became agriculturalists, as is clear from the absence of common words for ‘plough’, ‘field’, ‘grain’ and suchlike. While Armenian shares many agricultural terms with the languages of Europe, these are absent from Indo-Iranian. The common Indo-European vocabulary reflects a stage of development when weaponry was made of stone, wood, bones and hides (cf. Schrader 1890: 320-346). It includes words for ‘cart’ (ὄχος), ‘wheel’ (κύκλος), ‘axle’ (ἄξων), ‘yoke’ (ζυγόν), ‘carpenter’ (τέκτων), ‘(wooden) house’ (δόμος), ‘vessel’ (ναῦς), ‘to plait’ (πλέκω), ‘to weave’ (ὑφαίνω), ‘to spin’ (νέω), ‘garment’ (εἷμα) and ‘to clothe’ (ἕννῡμι). The population of Denmark and Scandinavia did not speak an IndoEuropean language before the advent of pastoralism, which puts these countries beyond the original homeland. The introduction of the cart was evidently more recent than the domestication of cattle, which were used as draught animals. All Indo-European languages have the same words for ‘o’x, ‘sheep’, ‘goat’, ‘horse’ and ‘swine’, but not for tame birds, which had not yet been domesticated, nor for ‘donkey’, which came from the south. The ancient character of cattle-breeding is corroborated by the words for ‘livestock’, ‘to herd’, ‘herd’ (πῶυ), ‘herdsman’ (ποιμήν), ‘cow’ (βοῦς), ‘to milk’ (ἀμέλγω), ‘butter’ (ἔλπος) and ‘bull’. The word 32 Introduction steer (ταῦρος, Latin taurus) is probably an early Indo-European borrowing from Semitic denoting ‘wild bull’, as is indicated by Lithuanian taũras ‘aurochs’ and Old Prussian tauris ‘wisent’, while ox is the old word for ‘domesticated bull’. Cattle were the most important animals in prehistoric pastoralism, as is clear from such Vedic expressions as ‘desire of cows’ for ‘struggle’ and ‘cattle-master of horses’ for ‘lord’ and from the peculiar Greek word ἱπποβουκόλος ‘horsecowherd’ for ‘horse-herd’. Other domesticated animals included sheep, as is also clear from the words for ‘wool’ and ‘lamb’, and goats and horses, but not donkeys, chicken, ducks and geese, which were domesticated more recently (cf. Schrader 1890: 390). While the Indo-European vocabulary contains an abundance of words reflecting a pastoral society, there is no common agricultural terminology. The Indo-Iranians evidently belonged to a different cultural unity when the languages of Europe, including Armenian, developed their agricultural terminology (cf. now Kuz’mina 2007). The Latin word grānum originally meant ‘grain of corn’, as is also clear from its Germanic, Baltic and Slavic cognates, while hordeum ‘barley’ may have designated a wild variety. The Sanskrit cognate of the word for ‘field’ ájras means ‘plain’ while the word vápati means both ‘throws’ and ‘sows’. The European words for ‘seed’ (Latin sēmen), ‘to mow’ (ἀμάω), ‘to mill’ (Latin molere), ‘to plough’ (ἀρόω) and ‘plough’ (ἄροτρον) are absent from Indo-Iranian (cf. Schrader 1890: 410). These etyma can now be identified with Hittite šai- ‘to throw’, ānš- ‘to wipe’, malla- ‘to grind’ and harra‘to crush’, respectively (cf. Kloekhorst 2008). The word for ‘wine’ has an IndoEuropean etymology (cf. Beekes 1987) but was limited to the Mediterranean countries (Italy, Greece, Asia Minor), from where it spread to northern and eastern Europe and to the Middle East. The original Indo-European word probably denoted ‘vine’ rather than ‘wine’ because this is the meaning of Basque ayen, aihen (cf. also Beekes 2010: 1059). Though metallurgy was unknown to the Indo-Europeans, they had words for gold, which is cognate with yellow and has been preserved in Germanic, Baltic, Slavic and Indo-Iranian, ‘silver’ (Latin argentum), cognate with ‘bright’ and preserved in Celtic, Italic, Greek, Armenian and Iranian, ‘copper’ (Latin raudus ‘piece of brass’), which is cognate with red and has been preserved in Italic, Germanic (where it may also be represented by lead), Slavic and IndoIranian, and ore, which became the word for ‘brass, bronze’ in Italic (Latin aes), Germanic and Indo-Iranian. There were no common Indo-European terms for ‘iron’, ‘lead’ and ‘tin’. The original words for ‘gold’ and ‘silver’ were apparently replaced with the advent of metallurgy, the first in Italy (Latin aurum), from where the new term spread to Baltic, Celtic and eventually Basque, and the second in Spain (Celtiberian silapur, Basque zillar), from where the new term spread to Germanic, Baltic and Slavic. The latter word cannot be separated from Berber aẓref ‘silver’ and aẓarif ‘alum’, where a- is a nominal prefix. Here the word C.C. Uhlenbeck on Indo-European, Uralic and Caucasian 33 for ‘alum’ is evidently a borrowing from Punic and can be derived from the Semitic root ṣrp, Akkadian ṣarāpu ‘to refine (metals by firing)’, Hebrew ṣārap ‘to smelt (metal)’ (see Boutkan & Kossmann 2001 for references). I conclude that Uhlenbeck was well ahead of his time in his discussion with Hirt and Schrader. He recognized that it is necessary to distinguish between two components of Indo-European language and culture, an older common inheritance which reflects a pastoral society and a later European complex with a common agricultural vocabulary, both of them dating from before the introduction of metallurgy. It is interesting that before the end of the 19th century he had already reached the position which has now become dominant among Indo-Europeanist scholars and is supported by the archaeological evidence (cf. Mallory 1989). The major point which he did not see is the crucial role of the domesticated horse in the Indo-European expansions (but see below). The tentative localization of the Indo-European homeland was logically followed by the question if the proto-language could be related to other language families. Uhlenbeck remarked that the identity between the nominative and the accusative in the neuter, both singular and plural, points to an original absolutive case (“Passivus”) which was identical with the bare stem (except in the o-stems), whereas the subject of transitive verbs was in an ergative case (“Aktivus”), marked by a suffixed *-s which he identified with the demonstrative pronoun *so (1901). His student Nicolaas van Wijk argued that the nominal genitive singular in *-s was identical with the original ergative (1902). Among other things, he adduced such constructions as Latin miseret mē ‘I feel pity’, pudet mē ‘I am ashamed’, where the logical subject is in the genitive case. Uhlenbeck claimed that the Indo-European proto-language was characterized by polysynthesis, suffixation and infixation and drew attention to its typological similarity to unrelated languages such as Basque, Dakota and Greenlandic. He also observed that the Indo-European mediopassive voice is reminiscent of the verbal construction with an incorporated dative and an object in the absolutive case which is found in Basque and North American languages. In a later study (1904), he adduced the strong resemblance between Basque and Indo-European nominal composition as an example of typological similarity between unrelated languages (“Sprachen zwischen welchen man selbst keine entfernte Verwandtschaft nachzuweisen vermag”, cf. also Uhlenbeck 1913). While he considered a common origin of Eskimo and Aleut with the Uralic languages probable (1905a), he rejected the possibility that Basque is related to Uralic and Altaic and suggested that it might rather be of Afro-Asiatic provenance (1905b). After a detailed examination of the available evidence, Uhlenbeck concluded that a genetic relationship between Basque and Caucasian languages cannot be established (1923), but later he changed his opinion and 34 Introduction considered the latter to be highly probable (“de onmiskenbare verwantschap met het Kaukasisch”, 1946: 17), regarding the Afro-Asiatic elements as borrowings. Holger Pedersen has listed numerous examples of Russian impersonal sentences with an inanimate agent in the instrumental case, e.g. tečeniem ego poneslo nazad ‘the current carried him back’, vetrom sneslo kryšu ‘the wind blew off the roof ’, and similar constructions in Iranian, Celtic and Germanic (1907: 134-140). He argues that this sentence type is older than the rise of grammatical gender in Indo-European and compares it with the ergative construction in North Caucasian languages, where the subject of a transitive verb is in the instrumental or in the genitive, as it is in Tibetan and Eskimo, and in the Armenian l-preterit, where it is in the genitive. For the Indo-European protolanguage he proposes that the subject of an intransitive verb and the object of a transitive verb were in the absolutive (unmarked) case while the subject of a transitive verb was in the genitive when the agent was animate but in the instrumental when it was inanimate (1907: 152). After the differentiation between the ergative and the genitive, the former came to be used for the subject of intransitive verbs in the case of animate individuals, as distinct from collectives and inanimates. This three-way distinction was subsequently grammaticalized as masculine, feminine and neuter gender. In a later study, Pedersen argues that a comparison of the Indo-European and Uralic grammatical systems leaves no doubt about their genetic relationship (“Es liegt hier eine Summe von Übereinstimmungen vor, die den Zufall ausschliesst”, 1933b: 309) and proposes that athematic presents (τίθημι ‘I put’) were originally transitive while thematic presents (φέρω ‘I carry’) and perfects (οἶδα ‘I know’) were originally intransitive, which is reminiscent of Hungarian várom ‘I wait for him’ versus várok ‘I wait’. In the meantime, Uhlenbeck’s doubts about the reality of an Indo-European proto-language had grown. He now defined the proto-language as the group of dialects spoken by the original community of Indo-European conquerors. Since the conquests of non-Indo-European territories took place at different times, the language of the later conquerors was no longer identical with the original protolanguage. While the separate branches of Indo-European arose when the language of the invaders was adopted by local populations speaking quite different substratum languages, Uhlenbeck claims that Proto-Indo-European itself already consists of two unrelated groups of elements, which he calls A and B (1933, 1934a, 1937b). Here A contains pronouns, verbal roots and derivational suffixes whereas B contains isolated words which are not related to verbal roots, such as numerals, some kinship terms, and many names of body parts, animals and trees. Uhlenbeck compares A with Uralic and Altaic and attributes irregular features such as heteroclitic inflection and grammatical gender to B, for which one might think of Caucasian languages. The relation between Indo-European and Uralic can be extended to Eskimo (cf. Uhlenbeck 1905a, 1906, 1907, 1934b, C.C. Uhlenbeck on Indo-European, Uralic and Caucasian 35 1937a, 1941, 1942). This is in accordance with the view that the Indo-Europeans arrived in southern Russia from the Asian steppes east of the Caspian Sea, where they allegedly led a nomadic life with horses, chariots and large herds of cattle. Since such words as kinship terms and names of body parts are usually regarded as belonging to the basic vocabulary of a language, Uhlenbeck rejected the terms “genetic relationship” for A and “borrowing” for B and presented Proto-Indo-European as a “mixed language”, an idea which had first been put forward by Sigmund Feist (1910). This is an unfortunate notion which is based on underanalysis of the data and can easily lead to muddled thinking about linguistic contact and language change (cf. K197). The point is that the two components A and B have an entirely different status. The situation is reminiscent of Michif, which has been adduced as a prime example of a mixed language. Here we find numerous French nominal stems which were borrowed together with their determiners, e.g. le loup ‘the wolf ’, sa bouche ‘his mouth’, son bras ‘his arm’, while the verbal stems and grammatical elements are purely Cree (cf. K197: 123). Uhlenbeck himself adduces the Sanskrit influence on Indonesian, the French influence on English and the Romance influence on Basque as parallels (1941: 204f.). He appears to have realized his mistake because he later returned to an analysis in terms of genetic relationship and borrowing (1946). The two major findings which Uhlenbeck has contributed to IndoEuropean linguistics are the reconstructed ergative (which was established independently by Pedersen (1907: 157), who provided the comparative evidence) and the twofold origin of the vocabulary. Both discoveries have been slow in their acceptance by the scholarly community. André Vaillant has identified the Indo-European ergative in *-s as an original ablative, the animate accusative in *-m as a lative (casus directivus), and the neuter pronominal ending *-t with the instrumental ending in Hittite and the ablative ending of the o-stems in the other Indo-European languages (1936). The ending *-m was originally limited to animate individuals, like the preposition a in Spanish, e.g. veo a Pedro ‘I see Peter’ (see Pottier 1968 for details). Robert Beekes has shown that the entire paradigm of the o-stems was built on an ergative case form in *-os (1985). Blissfully ignorant of the data and unaware of the comparative evidence, Alan Rumsey has argued on typological grounds that there cannot have been a ProtoIndo-European ergative because this case is absent from the neuter paradigm (1987a, 1987b). Since his objection was effectively answered by Pedersen a hundred years ago (1907), there is no reason to return to the matter here. It illustrates how a tool which in itself is useful becomes harmful in the hands of the unskilled (cf. also K130). The idea of a genetic relationship between Indo-European and Uralic has become fairly well accepted among specialists (e.g. Collinder 1965, 1974). Gimbutas’s theory that the Indo-Europeans moved from a primary homeland north of the Caspian Sea to a secondary homeland north of the Black Sea (e.g. 36 Introduction 1985) is fully in agreement with the view that their language developed from an Indo-Uralic proto-system which was modified under the influence of a North Caucasian substratum, perhaps in the sixth millennium BC (cf. Mallory 1989: 192f., K111, K203). Johannes Knobloch has suggested that the thematic vowel *-e/o- in the Indo-European verbal inflection represents an earlier object marker (1953). I have argued that the thematic present and the perfect originally had a dative subject, reflecting an earlier intransitive construction with an indirect object (K049). For Proto-Indo-Uralic we can reconstruct a genitive in *-n, which is reflected in the oblique stem form of the Indo-European heteroclitics, a lative-accusative in *-m, a dative-locative in *-i, an ablative-instrumental in *-t, which is reflected as both -t and -s in Indo-European, plural markers *-t and *-i, dual *-ki, personal pronouns *mi ‘I’, *ti ‘thou’, *me ‘we’, *te ‘you’ and corresponding verbal endings, reflexive *u, demonstratives, participles, derivational suffixes of nouns and verbs, negative *n- and interrogative *k- (cf. K203). The rise of the ergative construction, grammatical gender and adjectival agreement can be attributed to North Caucasian influence and may have proceeded as indicated by Pedersen (1907). It is important to note that the accusative is of Indo-Uralic origin and therefore older than the ergative. This explains the peculiar construction of Russian vetrom sneslo kryšu ‘the wind blew off the roof ’, where the inanimate agent is in the instrumental and the object is in the accusative. While the Indo-Uralic component of the lexicon (Uhlenbeck’s A) has been a focus of research in the past, the identification of the non-IndoUralic component (Uhlenbeck’s B) remains a task for the future. In view of the large number of consonants and the minimal vowel system of Proto-IndoEuropean, the northern Caucasus seems to be the obvious place to look (cf. Starostin 2007). AN OUTLINE OF PROTO-INDO-EUROPEAN Indo-European is a branch of Indo-Uralic which was radically transformed under the influence of a North Caucasian substratum when its speakers moved from the area north of the Caspian Sea to the area north of the Black Sea (cf. K247). As a result, Indo-European developed a minimal vowel system combined with a large consonant inventory including glottalized stops, also grammatical gender and adjectival agreement, an ergative construction which was lost again but has left its traces in the grammatical system, especially in the nominal inflection, a construction with a dative subject which was partly preserved in the historical languages and is reflected in the verbal morphology and syntax, where it gave rise to new categories, and a heterogeneous lexicon. The Indo-Uralic elements of Indo-European include pronouns, case endings, verbal endings, participles and derivational suffixes. In the following I shall give an overview of the grammar of Proto-Indo-European as it may have been spoken around 4000 BC in the eastern Ukraine, shortly after the ancestors of the Anatolians left for the Balkans (for more recent developments I refer to Beekes 1995). This stage preceded the common innovations of the non-Anatolian languages such as *mer- ‘to die’ < ‘to disappear’ , *tu << *ti ‘thou’, *seȥ- ‘to satiate’ < ‘to stuff ’, *dhugʄȥtēr << *dhuegʄȥtr ‘daughter’, *ȥerȥw- ‘to plough’ < ‘to crush’, *meȤ ‘don’t!’ < ‘say no!’, *Ȥekʄuos << *Ȥekʄu ‘horse’ (cf. Kloekhorst 2008: 8-10). It also preceded the rise of the subjunctive and the optative and dialectal Indo-European developments such as the rise of distinctive voicedness (not shared by Tocharian), the creation of a thematic middle voice (cf. K239: 151-157), and the satemization of the palatovelars (cf. K263: 43). The lexicon included words for ‘cart’, ‘wheel’, ‘axle’, ‘yoke’, ‘carpenter’, ‘house’, ‘vessel’, ‘to plait’, ‘to weave’, ‘to spin’, ‘to clothe’, ‘o’x, ‘sheep’, ‘goat’, ‘horse’, ‘swine’, ‘cow’, ‘dog’, ‘to herd’, ‘to milk’, ‘butter’, ‘wool’, ‘lamb’, ‘gold’, ‘silver’, ‘copper’, ‘ore’, but not for ‘donkey’, ‘cat’, ‘chicken’, ‘duck’, ‘field’, ‘to sow’, ‘to mow’, ‘to mill’, ‘to plough’, ‘iron’, ‘lead’, ‘tin’. There was no agricultural or metallurgical vocabulary at this stage. PHONOLOGY Proto-Indo-European had two vowels: *e [æ] and *o [Ȝ], which had long variants *ē and *ō in monosyllabic word forms and before word-final resonants (cf. Wackernagel 1896: 66-68). At a later stage, *e was colored by a contiguous *ȥ or *ȥw to *a or *o, respectively (cf. K194: 39-44, 54-56, 75-78 and K202, Lubotsky 1989, 1990). Even more recently, *o was colored by a contiguous *ȥ to *a in Greek (cf. K034). The vowel *a is widespread in borrowings from European substratum languages, e.g. Latin albus ‘white’, Greek ἀλφός, Hittite alpa- ‘cloud’. PIE *e may represent any Indo-Uralic non-final vowel under the stress, e.g. 38 Introduction *uegʄh- ‘carry’ < *wiqi-, *uedh- ‘lead’ < *weta-, *ȥegʄ- ‘drive’ < *qaja-, *mesg- ‘plunge’ < *mośki-, cf. Finnish vie- ‘take’, vetä- ‘pull’, aja- ‘drive’, Estonian mõske- ‘wash’. PIE *o has a twofold origin: it developed phonetically from unstressed *u and *e and was introduced by analogy in stressed syllables (cf. K203: 221, K213: 165). Proto-Indo-European had six resonants with syllabic and consonantal allophones: *i, *u, *r, *l, *m, *n. There were twelve stops, one fricative *s, and three laryngeal consonants *Ȥ, *ȥ, *ȥw. The distinction between the laryngeals was neutralized before and after *o (cf. K194, K202). The stops were the following: labials dentals palatovelars labiovelars fortis glottalic lenis *p [p:] *t [t:] *kʄ [kʄ:] *kw [kw:] *b [p’] *d [t’] *gʄ [kʄ’] *gw [kw’] *bh [p] *dh [t] *gʄh [kʄ] *gwh [kw] Word-initial *b- had already become *p-, e.g. Vedic píbati ‘drinks’, Old Irish ibid, Armenian ǩmpem ‘I drink’ (with a nasal infix, cf. K194: 80), Luwian pappaš- ‘to swallow’ (Kloekhorst 2008: 628) with analogical fortis *-p- and Latin bibō with restoration of initial *b-. A similar rule may account for the absence of PIE roots with two glottalic stops such as *degʄ- or *gweid- because the fortes were almost as frequent as the lenes and the glottalics together. The opposition between palatovelars and labiovelars was neutralized after *u and *s and the palatovelars were depalatalized before *r, *s and laryngeal consonants (cf. Meillet 1894, Steensland 1973, Villanueva 2009), e.g. Luwian k- < *kʄ- in karš- ‘cut’ < *krs-, kiš‘comb’ < *ks-, kattawatnalli- ‘plaintiff ’ < *kȥet- (cf. Kloekhorst 2008) and similarly in Vedic cyávate ‘moves’ < *kȤieu-, Greek σεύομαι, Prussian etskī- ‘rise’ < *kȤiei-, Latin cieō (cf. K263: 176) and in Vedic kṣáyati ‘rules’ < *tkȤei-, Avestan xš-, as opposed to Vedic kṣéti ‘dwells’ < *tkʄei-, Avestan š- (cf. Beekes 2010: 789, 791). It has been observed that PIE fortis and lenis stops could not co-occur in the same root, so that roots of the type *teubh- or *bheut- are excluded. It follows that the distinction between fortes and lenes was a prosodic feature of the root as a whole, which may be called “strong” if it contained a fortis and “weak” if it contained a lenis stop. This system can be explained in a straightforward way from an earlier system with distinctive high and low tones. Lubotsky has shown that there is a highly peculiar correlation between Indo-European root structure and accentuation (1988a: 170), which again points to an earlier level tone system. In any case, the PIE prosodic system was very close to the system attested in Vedic Sanskrit. I have proposed that the PIE distinction between fortis and lenis stops resulted from a consonant gradation which originated from an IndoUralic stress pattern that gave rise to strong and weak syllables (K213). It is probable that the whole inventory of PIE stops and laryngeal consonants can be An outline of Proto-Indo-European 39 derived from the five Indo-Uralic stops *p, *t, *c, *k, *q with palatalization, labialization and uvularization under the influence of contiguous vowels (cf. K203: 220). Note that Proto-Uralic *q (=*x in Sammallahti 1988) is strongly reminiscent of the Indo-European laryngeals, being lost before a vowel and vocalized before a consonant in Samoyedic and lengthening a preceding vowel before a consonant in Finno-Ugric. NOMINAL MORPHOLOGY There were four major types of nominal paradigm in Proto-Indo-European: static, proterodynamic, hysterodynamic and thematic. In the singular, the proterodynamic paradigm had radical stress in the nom. and acc. forms and suffixal stress in the loc. and abl. forms whereas the hysterodynamic paradigm had radical stress in the nominative, suffixal stress in the acc. and loc. forms, and desinential stress in the ablative, which later adopted the function of genitive in these paradigms. A comparative analysis of the non-Anatolian languages leads to the following reconstruction (cf. K263: 104). Here Rsd stands for radical stress, rSd for suffixal stress, and rsD for desinential stress; the accentuation of the inst.sg. forms was probably identical with that of the loc.sg. forms at an earlier stage. The examples are: Vedic sūnús ‘son’, Old Irish ainm ‘name’, Greek θυγάτηρ ‘daughter’, Lithuanian piemuõ ‘shepherd’, and Old Norse oxe ‘o’x. nom.sg. acc.sg. gen.sg. loc.sg. dat.sg. inst.sg. sūnús sūnúm sūnós sūnáu sūnáve sūnúnā Rsd *-s Rsd *-m rSd *-s rS *-ø rSd *-i Rsd *-Ȥ ainm ainm anmae ainm Rs *-ø Rs *-ø rSd *-s rS *-ø rSd *-i Rsd *-Ȥ nom.pl. acc.pl. gen.pl. loc.pl. dat.pl. inst.pl. sūnávas sūnn sūnnām sūnúṣu sūnúbhyas sūnúbhis rSd *-es Rsd *-ns rsD *-om rsD *-su rsD *-mus rsD *-bhi anman anman anman rSd *-ȥ rSd *-ȥ rsD *-om rsD *-su rsD *-mus rsD *-bhi nom.sg. acc.sg. gen.sg. loc.sg. dat.sg. inst.sg. θυγάτηρ θυγατέρα θυγατρός θυγατρί piemuõ píemenį piemeñs piemenyjè píemeniui píemeniu anmanaib oxe oxa oxa oxa Rs *-ø rSd *-m rsD *-os rSd *-i rsD *-ei rsD *-eȤ 40 nom.pl. acc.pl. gen.pl. loc.pl. dat.pl. inst.pl. Introduction θυγατέρες θυγατέρας θυγατρῶν θυγατράσι píemenys píemenis piemenũʛ piemenysè piemenìms piemenimìs yxn yxn yxna yxnom rSd *-es rSd *-ns rsD *-om rsD *-su rsD *-mus rsD *-bhi It appears that these case endings largely originated after the split between Anatolian and the other Indo-European languages and that their common ancestor had no genitive, no dative, and no distinct oblique plural endings. The original situation has partly been preserved in Hittite, which has no number distinction in the case endings of the genitive (original ablative, which also replaced the locative in the plural) -aš < *-os, the instrumental -t, and the new ablative -z < *-ti (which is the instrumental with an added locative marker). Kloekhorst has shown that the acc.pl. ending -uš reflects *-(o)ms (2008: 929), which became *-(o)ns in the non-Anatolian languages. It appears that the plural marker *-s was added to the case marker *-m here, as in the Indo-Iranian and Armenian inst.pl. ending *-bhis. The nom.pl. ending -eš represents *-eies (cf. Kloekhorst 2008: 249). The Proto-Indo-European thematic paradigm was probably uninflected except for the accusative in *-om because the Hittite replacement of the ending *-os by all. -a < *-o (cf. Kloekhorst 2008: 161), loc. -i, inst. -it, abl. -az < *-oti, nom.pl. -eš < *-eies is incompatible with the addition of pronominal endings to the thematic vowel in dat.sg. *-oȤei, loc.sg. *-oȤi, abl.sg. *-oȤed and the extensions in nom.pl. *-oses, later *-oȤes, inst.pl. *-oȤois in the other branches of Indo-European. Either of these sets of developments would render the other superfluous and incomprehensible. I propose to derive the inst.sg. ending *-Ȥ from *-d [t’] < *-t after the full grade suffix *-en- in the n-stems because this consonant was phonetically lost word-finally after an obstruent but preserved after a vowel. The nominative had four different endings in Proto-Indo-European: *-s, *-d, *-i and zero. As Pedersen argued a long time ago (1907: 152), “bei transitiven verben stand das objekt in der grundform, das subjekt aber im genitiv [i.e. my ablative], wenn wirklich von einer thätigkeit desselben die rede sein konnte, also wenn es der name eines lebenden wesens war; dagegen stand es im instrumentalis, wenn es ein unpersönlicher begriff war.” Thus, *-s and *-d < *-t represent the endings of the ergative of animate and inanimate nouns, respectively, while the zero ending continues the original absolutive case. When the ergative in *-os was reanalyzed as a sigmatic nominative, it gave rise to an accusative in *-om which was subsequently generalized as an absolutive form of inanimate nouns, supplying a singulative to a collective formation in *-ȥ, and to an uninflected predicative nominal (which later adopted the function of a genitive plural, cf. K030: 294f.). This development was anterior to the split An outline of Proto-Indo-European 41 between Anatolian and the other branches of Indo-European but more recent than the rise of the lengthened grade before word-final resonants. The ending *-i is found primarily in pronominal plurals, e.g. demonstrative *toi, anaphoric *Ȥei, interrogative *kwei, also present 3rd pl. *-nti, which represents the original nom.pl. form of the nt-participle, and Latin quae, haec, Prussian fem.sg. quai, stai, where the feminine gender continues the earlier collective formation in *-ȥ, perhaps also the Hittite neuter pl. ending -i. At an earlier stage, the ending *-i had been added to the Indo-Uralic plural suffix *-t-, yielding the PIE nom.pl. ending *-es (cf. K203: 222). In Uralic we find e.g. Finnish talot ‘houses’, obl. taloi-, where *-i originally marked the dependent status of the noun (cf. Collinder 1960: 237, Janhunen 1982: 29f.). While the PIE endings nom. < abl. *-s, nom. < inst. *-d < *-t, acc. *-m, loc. *-i, and nom.pl. *-es and *-i have impeccable Indo-Uralic etymologies, this does not hold for the genitive, the dative, and the oblique plural endings. Genitival and adjectival relationships were apparently expressed by simple juxtaposition and partial agreement. Other syntactic and semantic relationships were expressed by a large number of particles. Pronouns never developed an animate ergative or an inanimate accusative and had not yet developed other oblique case forms in Proto-Indo-European, so that we can only reconstruct animate nom. *so, *Ȥe, *kwe, acc. *tom, *im, *kwim, inanimate abs. *to, *i, *kwi, erg. *tod, *id, *kwid < *-t. After the split between Anatolian and the other Indo-European languages, full paradigms were created by the addition of case endings in the former and by composition with the word for ‘one’ *si (cf. Kloekhorst 2008: 750), obl. *sm- in the latter. New adjectival paradigms originated from the thematicization of pronominal and adverbial forms, e.g. *kwo-, *io-, Hittite a- < *Ȥo-, kā- ‘this’ < *kʄo- from *kʄi ‘here’, apā- ‘that’ < *Ȥobho- from *Ȥe-bhi ‘at him’, cf. Vedic inst.pl. ebhís. For the personal pronouns, which probably used the accusative (Indo-Uralic allative) as a general oblique case form, I refer to my earlier treatment (K223, cf. also Kloekhorst 2008: 111-116). The creation of genitive, dative and oblique plural endings belongs to the separate histories of Anatolian and the other branches of Indo-European. After the rise of the thematic accusative ending *-om, Anatolian created a new oblique case in *-o, evidently on the analogy of the endingless locative forms of the consonant stems, to replace the allative function of the accusative. As *-s and *-d < *-t had become animate and inanimate nominative endings, respectively, and the former adopted the function of genitive in the consonant stems, the ablative ending was replaced by *-ti in Anatolian and by *-d in pronominal paradigms in the other languages, which then generalized *-Ȥ in the instrumental case. It follows from this scenario that the common development of final *-d < *-t, e.g. in Latin quod, Old High German hwaz, was not shared by Anatolian. The early loss of word-final *-t after an obstruent in the non-Anatolian languages explains the removal of the root-final obstruent in Greek ἔσβη ‘(the fire) went out’ < 42 Introduction *gwēs(t) and the rise of the k-perfect in Greek and Latin (cf. K239: 155). The nonAnatolian languages also created a full grade dat.sg. ending *-ei and a full grade inst.sg. ending *-eȤ, probably after the reanalysis of abl.sg. *-d as *-ed in the pronoun (cf. K223). Anatolian created a pronominal genitive in *-el which is reminiscent of Greek φίλος ‘friend(ly)’ < ‘own’ < *bhi-l- ‘belonging to the inner circle’. The other languages created a pronominal gen.sg. form by composition: *kwe-so, *Ȥe-so, *to-si with addition of *-o from *-so, then loc.sg. *Ȥesmi, *tosmi beside *Ȥei, *toi, feminine *Ȥesieȥ-, *tosieȥ-, etc. In the plural we have new endings in acc. *tons < *toms, inst. *tois, later *toȤois from the thematic paradigm, abl. *toios (cf. K194: 50), gen. *toisom (cf. K030), dat. *toimus (cf. K194: 49) and loc. *toisu. Since the endings *-mus and *-su are not found in the singular, they probably originated from distributive usage. Comparing these forms with Russian vsem po odnomu ‘one each to all’ and po-vsjudu ‘everywhere’, respectively, I would regard *-s as a plural marker (vs-), *-u as a distributive suffix (po), and *-m- as a reflex of the word for ‘one’ (odn-). The suffix *-u may be compared with Greek πάνυ ‘altogether’, Vedic u ‘also’, Hittite hūmant- ‘every, each’. The inst. ending *-bhi was still an independent particle at the stage under consideration. VERBAL MORPHOLOGY As I have treated the prehistory of the Indo-European verb in some detail elsewhere (K203, K241, K244), I can be brief here. There were six different sets of verbal endings (thematic and athematic present and aorist, perfect and stative) which originally corresponded with different types of syntactic construction. When the ergative became a nominative case, the formal distinction between transitive and intransitive verbs disappeared, but the construction of the thematic present and the perfect with a logical subject in the dative (or locative) was preserved, except in Anatolian. This gave rise to an expansion of the transitive middle paradigm, where the subject and the indirect object were identical. The Proto-Indo-European verb had an indicative, an injunctive, an imperative, a participle in *-nt-, verbal adjectives in *-lo-, *-mo-, *-no-, *-to-, and verbal nouns in *-i-, *-u-, *-m-, *-n-, *-s-, *-t- (cf. Beekes 1995: 249-251). The optative may originally have been a derived present in view of the 1st sg. ending -m < *-mi in Tocharian, both A and B. Derived verb stems were formed by reduplication and/or suffixation. The PIE stem-forming suffixes are: present *-(e)i-, *-(e)m-, *-(e)s-, *-n(e)-, *-dh(e)-, *-ske-, *-ie-, perhaps *-i(e)Ȥ(optative), aorist *-s-, *-eȤ-, *-eȥ- (cf. K239: 71f., 134f., 152f.). While the aorists may represent original nominal formations (cf. Greek χρή ‘must’ and K263: 57, 187), I have proposed to identify the derived presents as (Indo-Uralic?) compounds with the roots of Indo-European ‘to go’, ‘to take’, ‘to be’, ‘to lead’, ‘to put’, ‘to try’, perhaps also *ieȤ- ‘let’ (K241 in fine). An outline of Proto-Indo-European 43 Elsewhere I have argued that the Hittite hi-flexion comprises original perfects, new perfects created on the basis of derived presents, and transitive zero grade thematic formations corresponding to the Vedic 6th class presents (cf. K241; there were no full grade 1st and 10th class presents at this stage). The original athematic i-presents are reflected in Latin capiō ‘take’, Old Irish gaibid, Gothic hafjan, and the Balto-Slavic i-presents. Slavic verbs in -ěti (Lith. -ėti) with an i-present continue four different formations: o-grade perfects, zero grade i-presents, e-grade statives, and verbs denoting processes which originally had a thematic present, e.g. gorěti ‘to burn’, bъděti ‘to be awake’, sěděti ‘to sit’, svьtěti sę ‘to shine’ (cf. K121, K228). The second type corresponds to the Vedic root-stressed 4th class presents and the third type to Gothic sitan and Old Irish saidid (cf. K105: 7f., K239: 135). While the derivation of Hittite hi-verbs from reduplicated and nasal presents belongs to the Anatolian developments, the creation of derived perfects from athematic i-presents evidently dates back to the common Indo-European proto-language, being reflected in Vedic 4th class middle presents such as búdhyate. After the loss of the ergative construction, the stressed suffix *-ie- which is still found in Vedic syáti ‘binds’ could easily spread as a suitable device to derive new presents, primarily of transitive verbs. The introduction of full grade thematic stems gave subsequently rise to new imperfective presents, e.g. dáyate ‘distributes’ beside dyáti ‘cuts’ (cf. Kulikov 2000: 277f.). The new suffix *-eie- then spread to o-grade perfects, giving rise to the 10th class causative presents (cf. K241). In Hittite, the ie- and ske-presents adopted the mi-flexion in prehistoric times. The statives in -āri resulted from an Anatolian inovation which preceded the merger of the perfect with the transitive thematic flexion (cf. K244). POSTSCRIPT Melchert and Oettinger have recently (2009) published an alternative reconstruction of the PIE case endings which may be compared with my reconstruction presented above. Here I shall point out the differences between the two analyses. 1. On the basis of the Anatolian evidence Melchert and Oettinger start from a special dat.pl. ending *-os beside general abl. *-ti and inst. *-H1, for which I shall write *-Ȥ, and adverbial forms in *-ti, *-m, *-(a)d, *-is and *-bhi. Allowing for a separate development of the Proto-Anatolian case system, I start from abl. *-(o)s, inst. *-t, loc. *-i and acc. (Indo-Uralic allative) *-m(e) beside a separate particle *bhi ‘near’. Moreover, I derive *-s from Indo-Uralic *-t-i. 2. Melchert and Oettinger do not take the endings nom. *-(o)s, acc. *-m, nom.acc. *-d and loc. *-i into account in their analysis. In my view, these endings have the same Indo-Uralic origin as the oblique case endings *-os, *-m, 44 Introduction *-d, *-ti and *-Ȥ. The nom.sg. endings *-s and *-d represent the ergative case endings of animates (ablative) and inanimates (instrumental), respectively (cf. already Pedersen 1907: 152), and the acc.sg. ending *-m represents the IndoUralic allative. 3. Melchert and Oettinger think that the ending *-d, which they identify with Latin ad ‘to’, replaced the original ablative ending *-ti first in the pronoun and later by addition of *ad to the thematic vowel in the noun. This is highly improbable for semantic (allative replacing ablative?), formal (*-d in the pronoun but *-ad after the thematic vowel?), structural (with thematic but not with athematic nouns?) and chronological (Univerbierung first in the pronoun and later in the noun?) reasons. In my view, the ending *-d developed phonetically from *-t in word-final position and the ending *-ti was created as a new local case marker when abl. *-s and inst. *-t had adopted the function of ergative case endings. The development of *-ti into a regular case ending never took place in the non-Anatolian languages. The comparison with Armenian and Tocharian is spurious (cf. K194: 49). 4. Melchert and Oettinger propose that the gen.sg. and dat.pl. forms replaced the original ablative in the non-Anatolian languages. This is highly improbable because the generally attested development is a replacement of grammatical by local case forms, not the reverse. In my view, the gen.sg. and dat.pl. endings *-os represent the original ablative. 5. The dat.pl. ending of the non-Anatolian languages was *-mus, not *-mos or *-bhios, as is clear from Old Lithuanian *-mus, Slavic *-mъ (cf. Vermeer 1991), u-infection in Old High German tagum and Old Norse dNjgom ‘days’ (cf. Van Helten 1891: 460-462), and the Armenian evidence (cf. K194: 49). The ending was replaced by the new ablative endings *-bhos in Italo-Celtic and *-bhios in Indo-Iranian which had replaced the earlier abl.pl. ending *-ios (which is still found in the Armenian pronoun, cf. K194: 50). 6. The instrumental ending *-t was preserved in Proto-Anatolian (cf. Kloekhorst 2008: 232, 799) and became *-d [t’] in the non-Anatolian languages, where it was preserved after vowels and liquids and lost after obstruents. I have proposed to derive the ending *-Ȥ from *-d after the full grade suffix *-en- in the n-stems (see above). There is no evidence for an ending *-Ȥ in Anatolian. There is no reason to identify the verbal suffix *-eȤ with the non-Anatolian instrumental ending *-eȤ. The Old Hittite clause conjunctive particle ta represents *to, not *toȤ (cf. Kloekhorst 2008: 801). 7. Melchert and Oettinger assume a thematic ablative ending *-oȤad for which there is no evidence. The Lithuanian gen.sg. ending -o represents unstressed *-ōd < *-oȤed (cf. K263: 6, 184, 192), with addition of pronominal *-ed to the An outline of Proto-Indo-European 45 thematic vowel. The creation of oblique case forms of thematic nouns was an independent development in the Anatolian and the non-Anatolian languages. 8. Melchert and Oettinger posit inst.pl. endings *-is, *-mis, *-oiis for which there is no evidence whatsoever. They were clearly influenced by the scrabble methodology still practiced in some quarters of the Indo-European scholarly community (cf. K239: 145f.). The thematic inst.pl. ending *-ōis represents *-oȤois, with *-ois from the oblique plural form *tois found in gen.pl. *toisom, as is clear from Vedic -aiḥ, Greek -οῖς, Lith. -aĩs and Slavic (Slovene, Posavian, Czech, Slovincian) long -ȳ. The particle *bhi replaced the inst.pl. ending in the athematic paradigm, where it had been lost phonetically after obstruents. The addition of final *-s after *-bhi on the analogy of dat.pl. *-mus and abl.pl. *-ios was probably an independent development of Insular Celtic and Indo-Iranian because it apparently did not take place in Greek and Gaulish and may have been recent in Armenian and especially in Balto-Slavic, where the ending *-miȤs adopted its laryngeal either from inst.sg. *-Ȥ or from acc.pl. *-Ȥns. Note that the singular had thematic *-oȤ and athematic *-mi in Balto-Slavic, e.g. Lith. vilkù ‘wolf ’, sūnumì ‘son’, OCS vьčera ‘yesterday’, synъmь. SCHLEICHER’S FABLE The first to compose a text in Proto-Indo-European was August Schleicher (1868). Emendations of his text were published by Hirt and Arntz (1939), Lehmann and Zgusta (1979), and others. Though these efforts must not be taken too seriously, they can serve as an illustration of the linguistic systems reconstructed by their authors. As my chief concern has always been the relative chronology of linguistic developments, it may be appropriate to present my version of Schleicher’s fable for a number of successive reconstructed systems, viz. Proto-Indo-European, non-Anatolian Indo-European, Classic IndoEuropean (excluding Tocharian), Central Indo-European (excluding ItaloCeltic), satǩm Indo-European (excluding Greek and Germanic), Balto-Slavic, East Baltic, and modern Lithuanian. As I have stuck as much as possible to Schleicher’s and Hirt’s versions, many of the choices made in the following are far from compelling. It must be emphasized that the purpose of this exercise is purely to illustrate the consequences of a systematic approach and to exemplify the cohesive nature of the reconstructions. For the reconstruction of Proto-Indo-European I refer to my outline (K269). The absence of lengthening in Vedic áviḥ ‘sheep’ shows that the coloring of *e to *o after *ȥw was more recent than Brugmann’s law in Indo-Iranian (cf. Lubotsky 1990). Since the distinction between the three laryngeals was preserved in Greek (Central IE) and Armenian (satǩm IE), it was not lost before Balto-Slavic times, when they merged with the glottalic feature of the PIE “unaspirated voiced” stops into a glottal stop (Winter’s law, cf. K263: 44f.). The rise of the semivowels /j/ and /w/ resulted from the loss of the laryngeals in initial position. The relative pronoun *io- cannot be reconstructed for Anatolian and Tocharian, where the uninflected form *i may have survived at an early stage (cf. K263: 145). The word for ‘wool’ is a ȥ-stem in Latin (Classic IE) and other languages, perhaps a collective of a Ȥn-derivative, cf. Arm. gełmn, which appears as Hittite huliya- beside hulana-, pointing to an original form *ȥueli (cf. Kloekhorst 2008: 357f., Beekes 2010: 858). If the reconstruction of a zero grade root vowel in Old Irish ní ‘is not’, Slavic ně is correct (cf. K239: 126), the vowel of the monosyllable *neȤst was regularly lengthened in Proto-Indo-European (cf. Wackernagel 1896: 66-68) and the following laryngeal was lost in the satǩm languages (cf. K263: 52-57). The final *-t was lost after an obstruent in nonAnatolian Indo-European (cf. K269). 48 Introduction PROTO-INDO-EUROPEAN NON-ANATOLIAN INDO-EUROPEAN ȥweuis i ȥueli nēȤst Ȥekʄ:ums uēit:, t:o kw’rȥeum uokʄom ukʄent:m, t:o mkʄ’eȥm porom, t:o tkʄmenm ȤoȤkʄ:u prent:m. ȥweuis i ȥuelȤn nēȤs Ȥekʄ:uns Ȥe uēit’, t:om kw’rȥeum uokʄom uekʄont:m, t:om mkʄ’eȥm porom, t:om tkʄmenm ȤoȤkʄ:u peront:m. Ȥe uēuk: ȥweuis Ȥekʄ:umus, ȥetkʄo Ȥme kʄ:ērt’ ȥnerm uit’ent:i Ȥekʄ:uns ȥekʄ’ont:m. Ȥe ueuk:nt’ Ȥikʄ:ues, kʄ:luti ȥwuei, ȥetkʄo nsme kʄ:ērt’ uit’ent:i, ȥnēr p:ot:is ȥwuiom ȥulȤenm sue kwermom uesti kw:rneut:i, ȥwuimus kw:e ȥuelȤn neȤsti. t:ot’ kʄ:ekʄ:luus ȥweuis pleȥnom Ȥe pēuk’. uēuk:t ȥoeuis Ȥikʄ:uos, ȥetkʄo Ȥme kʄ:ērt: ȥnerm uit’ent:i Ȥekʄ:ums ȥkʄ’ent:m. ueuk:nt: Ȥikʄ:ues, kʄ:luti ȥwue, ȥetkʄo nsme kʄ:ērt: uit’ent:i, ȥnēr p:ot:is ȥwuiom ȥueli sue kwermom uesti kw:rneut:i, ȥwuēi kw:e ȥueli neȤsti. t:o kʄ:ekʄ:luus ȥweuis pleȥnom pēuk’t. CLASSIC INDO-EUROPEAN CENTRAL INDO-EUROPEAN ȥweuis ioi ȥulȤneȥ nēȤs Ȥekʄuns Ȥe uēi’d, tom ’gwrȥeum uogʄom uegʄontm, tom m’gʄeȥm borom, tom dgʄmenm ȤoȤkʄu berontm. Ȥe uēuk ȥweuis Ȥekʄumus, ȥedgʄo Ȥmoi kʄēr’d ȥnerm ui’denti Ȥekʄuns ȥe’gʄontm. Ȥe ueukn’d Ȥikʄues, kʄludi ȥwuei, ȥedgʄo nsmi kʄēr’d ui’denti, ȥnēr potis ȥwuiom ȥulȤneȥm subi gwermom uesti kwrneuti, ȥwuimus kwe ȥulȤneȥ neȤsti. to’d kʄekʄluus ȥweuis pleȥnom Ȥe bēu’g. ȥweuis ioi ȥulȤneȥ nēȤs Ȥekʄuns Ȥe uēi’d, tom ’gwrȥum uogʄom uegʄontm, tom m’gʄām borom, tom gʄmenm ȤoȤkʄu berontm. Ȥe uēuk ȥweuis Ȥekʄumus, ȥedgʄoi Ȥmoi kʄēr’d ȥnerm ui’denti Ȥekʄuns ȥe’gʄontm. Ȥe ueukn’d Ȥikʄues, kʄludi ȥwuei, ȥedgʄoi nsmi kʄēr’d ui’denti, ȥnēr potis ȥweuiom ȥulȤnām subi gwermom uesti kwrneuti, ȥweuimus kwe ȥulȤneȥ neȤsti. to’d kʄekʄluus ȥweuis pleȥnom Ȥe bēu’g. The PIE word for ‘horse’ is attested in Hittite ekku- < *Ȥekʄu- (cf. Kloekhorst 2008: 237-239), of which Greek ἵππος < *Ȥikʄuos represents the original ergative case, with reduction of pretonic *e to *i before a consonant cluster (cf. K065: 222 and K069). The u-stem must have been preserved in Balto-Slavic times in view of Lith. ašvà because the cluster *kʄw was depalatalized before a back vowel in this branch of Indo-European (cf. K263: 43). The acc.pl. ending *-ms was preserved in Anatolian (cf. Kloekhorst 2008: 929). The thematic aorist, e.g. Vedic ávidat ‘found’, probably represents an earlier root aorist, cf. ¡raik ‘left’ beside Greek ἔλιπε, Latin -līquit, with regular lengthening in the monosyllabic injunctive forms (cf. K188: 14 and K239: 155). The word-final fortis stop became glottalic in the non-Anatolian languages (cf. K269). There is evidence for the augment *Ȥe in Tocharian (cf. K149) and for its partial replacement by the enclitic particle *tu in Balto-Slavic (cf. K263: 285). The aorist suffix *-eȤ became *-ē word-finally in Balto-Slavic (cf. K239: 84 and K263: 187). Schleicher’s fable 49 SATƏM INDO-EUROPEAN BALTO-SLAVIC ȥweuiṣ ioi ȥulȤneȥ nēs Ȥekʄuns Ȥe uēi’d, tom ’gwrȥum uogʄom uegʄontm, tom me’gʄām borom, tom gʄmenm ȤoȤkʄu berontm. Ȥe uēuk ȥweuiṣ Ȥekʄumuṣ, ȥedgʄoi Ȥmoi kʄēr’d ȥnerm ui’denti Ȥekʄuns ȥe’gʄontm. Ȥe ueukn’d Ȥekʄues, kʄludi ȥweuei, ȥedgʄoi nsmi kʄēr’d ui’denti, ȥnēr potiṣ ȥweuiom ȥulȤnām subi gwermom uesti kwrneuti, ȥweuimuṣ kwe ȥulȤneȥ neȤsti. to’d kʄekʄluuṣ ȥweuiṣ pleȥnom Ȥe bēu’g. ’owiṣ joi wi’lna’ nēs ’ećuns wi’dē, i’nun tingun woȢɴun welkontin, i’nun welīn krowun, i’nun Ȣɴmōnin burȢɴu neśontin. EAST BALTIC LITHUANIAN avìs kurí’ netùri vì’lnās règē źì’rgNj’s, v¦’ną smàgų’s ràtNj’s vèlkantį, v¦’ną dìdį kràvą, v¦’ną źmṓnį grḗʖtai nèśantį. avìs kurì netùri vìlnos pamãtė árklius, víeną sunkiùs ratùs tráukiantį, víeną dìdelį króvinį, víeną žmõgų greĩtai nẽšantį. avìs pasãkė arkliáms, mán šìrdį skaũda mãtant výrą gẽnantį árklius. arkliaĩ pasãkė, klausýk aviẽ, mùms šìrdį skaũda mãtant, výras viẽšpats iš aviẽs vìlnos sáu dãro šìltą drabùžį, õ ãvys vìlnos netùri. išgirɶdusi taĩ avìs pab°go į lýgumą. avìs sàkē źi’rgàmus, màni sá’pā śi’rdìs règinti ví’rą gènantį źì’rgNj’s. źì’rgai sàkē, klàusi’ avḗʖ, mùmus sá’pā śi’rdìs règinti, ví’ras patìs iź avḗʖs vì’lnās sèvi dàrā śìltą drṓbį, ā àvies vì’lnās netùri. gìrdusi’ tà avìs bé’gā į lí’gumą. tēr tu ’owiṣ ’ećumuṣ, bolei mini ćēr wi’run wi’denti ’ećuns genontin. tren tu ’ećues, ćludi ’owei, bolei i’nmuṣ ćēr wi’denti, wi’ros potiṣ iȢɴ ’owiun wi’lnās subi teplān drōbin kurneuti, ’ā ’owimuṣ wi’lna’ ne’sti. to ćlu’ṣuṣ ’owiṣ pla’nun be’ga’. Pronominal *to, like *i, may have been uninflected at an early stage (cf. K269). The stem *in- developed an acute in Balto-Slavic times (cf. Derksen 2003). Word-final *-om became *-un in Balto-Slavic (cf. K263: 43f.). The ablaut in u-stem adjectives appears to have been preserved in Italo-Celtic and lost in Central Indo-European. The thematic nt-participle was an innovation of the non-Anatolian languages. Word-final *-eȥm was probably preserved in ItaloCeltic and became *-ām in Central Indo-European (cf. K263: 23 and 132f.), e.g. Greek ἄγᾱν ‘(too) much’. Clusters of dental plus velar stops lost their initial element before a consonant in Central Indo-European or earlier (cf. K268). Greek ὠκύς ‘fast’ may be related to the word for ‘horse’. The Vedic 3rd sg. subjunctive vócati beside vócāti ‘speak’ points to an original athematic reduplicated aorist injunctive with a long vowel in the monosyllabic 3rd sg. form (cf. K188: 9f.). For Balto-Slavic I reconstruct *tēr, cf. Hittite ter-, Lith. tarɶti, Prussian acc.sg. tārin ‘voice’ (=Latin vōcem). The dat.pl. 50 Introduction ending *-mus is an innovation of the non-Anatolian languages, where it replaced the general oblique plural ending *-os which is still found in Anatolian (cf. Kloekhorst 2008: 214). For Greek ἄχνυμαι ‘grieve’, ἄχθομαι ‘be grieved’, Hittite hatk- ‘tighten’ I reconstruct an original stative ‘be tight’ with reduction of the cluster *dhgʄh before the nasal suffix in Greek. If Old Irish -ágathar ‘fears’ is related, which seems probable, the conditioned reduction of the cluster can be dated to the Classic Indo-European period. The primary ending *-oi dates from the Central Indo-European period (cf. K044). The oblique case forms *Ȥme ‘me’, *tue ‘thee’, *nsme ‘us’, *usme ‘you’ may have been uninflected before they adopted the dat.loc. ending *-i in Classic Indo-European times (cf. K223 and K269). The word for ‘heart’ had a long vowel in the monosyllabic nom.acc. form, where the final stop is preserved in Armenian sirt and lost in Balto-Slavic, e.g. Prussian seyr. The participial dat.loc. form in *-enti is coreferential with the oblique pronoun and developed into the Lithuanian gerund. The original nom.pl. ending of the consonant stems was *-es (cf. K269). The vocative was marked by apocope before the particle *e and later restoration of the suffix in the consonant stems (cf. Beekes 1985: 99-108), e.g. Lith. aviẽ < *owe-i. The gen.pl. form in *-om developed from an uninflected predicative nominal (cf. K030 and K269). The creation of the reflexive dat.sg. form *subhi beside the general oblique case form *sue can be dated to the Classic IndoEuropean period. The PIE locative form in *-ēi may have been used for both singular and plural referents before the creation of separate case endings in the non-Anatolian languages (cf. K269). For Greek ἔφυγε ‘fled’, Latin fūgit I again reconstruct a root aorist (cf. K239: 155). Translation of the text (cf. Lehmann & Zgusta 1979: 456, 462): “A sheep that had no wool saw horses, one pulling a heavy cart, one carrying a big load, and one quickly carrying a man. The sheep said to the horses: My heart pains me seeing a man driving horses. The horses said: Listen sheep, our hearts pain us when we see that a man, the master, makes the wool of sheep into a warm garment for himself, and the sheep have no wool. Having heard this, the sheep fled into the plain.” *H2o AND *oH2 1. Since the early days of laryngeal theory there has been disagreement about the question whether a PIE *H2 merged with *H1 and *H3 in the neighbourhood of *o (e.g. Saussure, Kuryłowicz, Cowgill, Beekes) or colored a contiguous *o to *a (e.g. Benveniste, Kuryłowicz, Lejeune, Lindeman). In his review of Beekes’ dissertation (1971), Ruijgh lists the following arguments in favor of the latter view: (1) Perfect πέπηγα < πέπᾱγα, λέληθα < λέλᾱθα, cf. ἔρρωγα, πέποιθα. These forms can hardly be explained otherwise than by assuming that -ā- developed phonetically from *-oH2-· They provide a model for the analogical introduction of the present tense vocalism in the perfect: “il est facile de rendre compte de formes telles que κέκευθα (κεύθω ‘cacher’) à partir du type λέλᾱθα (λ³θω ‘être caché’); noter la parenté sémantique entre les deux verbes” (p. 190). (2) Compounds ἱππᾱμολγός < *-ο-Η2molgó-, στρατᾱγός < *-ο-Η2ogo-. This phonetic development created the model for the analogical introduction of the long vowel in such forms as ἐπημοιβός < ἐπᾱμοιβός next to phonetically regular ἐπᾰμοιβός. (3) 1st sg. ending -μαι next to 2nd sg. -σοι and 3rd sg. -τοι. I think that we have to start from *-(m)ai, *-(s)toi, *-(t)oi. The problem is that *-ai does not necessarily continue *-H2oi (cf. now K044). (4) Expected o-grade in ἀγός, ἄνεμος, ἀριθμός, φήμη < φ³μᾱ, cf. τροφός, πότμος, πορθμός, λόχμη. As for ἄνεμος, Latin animus, I am inclined to disagree both with Ruijgh’s *H2onH1mos and with Beekes’ *H2enH1mos and to posit *H2ṇH1emos, cf. κάλαμος < *kḷH2emos and νηνεμίη < *ṇH2nH1em-. Armenian hołm is probably of non-Indo-European origin. In ἀριθμός we may assume zero grade, as Ruijgh remarks himself, cf. OHG, ΟN, OIr. rīm. The o-grade in ἀγός and φήμη is a reasonable reconstruction. 2. Beekes has returned to the matter in a separate article (1972), where he regards the following cases as certain: (1) βωμός, cf. ἔβην. (2) φωνή, cf. φημί. (3) ποιμήν, πῶυ, cf. Latin pāscō. (4) θῶξαι, cf. θήγω. (5) πτώσσω, cf. πτήσσω. (6) ὄγκος, cf. ἀγκάλη, ἀγκών. (7) ὄκρις, cf. ἄκρος. 52 Indo-European phonology (8) οἰωνός, cf. Latin avis. (9) οὖς, cf. Latin auris. (10) Skt. ¡yu, cf. αἰεί. Το these we may add ὄγμος (cf. ἄγω, Skt. ájma- ‘way’), ὄναρ (PIE *H2onṛ, *H2ner-), ἀγωγή, ἀκωκή (cf. ἐδωδή, ὀπωπή < *ΗCοΗC-). In view of the examples where *H2 colors a neighbouring *o to *a, Ruijgh suggests: “Il vaut donc mieux expliquer les cas isolés de mots tels que βω-μός, φω-νή, ὄγ-μος comme résultat d’une apophonie plus récente, qui n’a pas réussi à pénétrer plus profondément dans le système de la morphologie grecque”. Here I agree with Beekes: “It is not probable that isolated cases are due to a recent reshaping” and “for some good cases such an analogical secondary ablaut cannot possibly be taken into consideration” (p. 120). Beekes’ examples appear to belong to an older layer. 3. Since both points of view rest upon considerable evidence, neither can be refuted: they must be integrated into a single consistent theory. I agree with Ruijgh that *H2 colored a contiguous *o to *a in Greek. However, I agree with Beekes that the relevant instances do not date back to the Indo-European protolanguage. The simplest assumption is that the opposition between the laryngeals was neutralized in the neighbourhood of PIE *o, where they merged into *H1, and that *H2 was restored in certain productive categories in Proto-Greek. Thus, we have ἀγός < *H2ogós, φήμη < *bhoH2meH2, βέβηκα < βέβᾱκα < *-gwοH2- (cf. δέδοικα) on the analogy of ἄγω < *Η2eg-, φημί < *bheH2-, βίβημι < *-gweH2- next to ὄγμος < *H1og- < *H2og-, ἀγωγή, φωνή, βωμός < *-οH1- < *-οH2-. The analogical development must be dated before the loss of the laryngeals because the latter eliminated the motivation for it. Semantically, the restoration of *H2 in φήμη ‘saying’ and the preservation of the old ablaut in φωνή ‘voice’ is quite acceptable. The timbre neutralization of the laryngeals in the neighbourhood of *o has its analogue in Shuswap, which offers the closest typological parallel to the PIE laryngeals. In this language, all consonants which are members of pairs exhibiting the rounding correlation are rounded before and after the rounded vowel (cf. Kuipers 1974: 22). As in Shuswap, the opposition between *H3e- and *H3o- was apparently not neutralized in Proto-Indo-European: initial *H2 and *H3 were preserved as h- before *e but lost before *o in Armenian, while *H1 was always lost before a vowel, e.g. hot < *H3ed- (original s-stem, cf. Latin odor), hoviw < *H3eui-peH2- (cf. Latin ovis, pāstor), but orb, orjik‘, orɷk‘, ost (cf. ὀρφανός, ὄρχις, ὄρρος, ὄζος, OHG ars, ast). There is a zero grade in oskr < *H3stuer- (cf. ὀστέον). For ὠμός, Arm. hum, Skt. āmáḥ I would suggest PIE *H2eH3mós, Latin amārus < *H2Ḥ3m-, Old Irish om < *H2H3e/om-. PROTO-INDO-EUROPEAN GLOTTALIC STOPS: THE COMPARATIVE EVIDENCE 1. THE TYPOLOGICAL ARGUMENT According to the traditional view, the Indo-European proto-language possessed four series of stops, which correspond to Old Indic t, th, d, dh. The laryngeal theory has made clear that th does not reflect a unitary phoneme of the protolanguage but a sequence of t plus laryngeal. Many scholars have seen that this leaves the reconstructed proto-language with a typologically improbable system of obstruents. Some have concluded that we have to return to the traditional reconstruction, even if the available evidence offers insufficient support for this view. Others have tried to reinterpret the triad *t, *d, *dh in such a way that the reconstructed system becomes more in accordance with typological expectations. 1.1. Holger Pedersen has argued that there are no reliable etymologies which point to PIE initial *b- (1951: 10-16). Since the voiceless labial stop *p- is easily lost in several languages (e.g. Celtic, Armenian, Japanese), Pedersen suggests that PIE *b was probably voiceless and weak, while *bh may have developed from a voiceless aspirate. He compares the interchange of voiced and voiceless stops with the West Armenian consonant shift. 1.2. Referring to Pedersen’s view, André Martinet suggests in a footnote that the PIE unaspirated voiced stops can be derived from a glottalic series (1953: 70). The absence of the labial can be compared with the same phenomenon in ProtoSemitic, for which he reconstructs a glottalic series from which the emphatic obstruents are derived. 1.3. In a neglected article, N.D. Andreev proposes a reconstruction of ProtoIndo-European without voiced obstruents: he reconstructs voiceless fortes, voiceless lenes, and voiceless aspirates, corresponding to traditional *t, *d, *dh (1957: 7). He explains the incompatibility of fortes and aspirates in the root by an assimilation rule. 1.4. Morris Swadesh has suggested that Proto-Indo-European and its neighbours had simple, glottalic, and aspirated stops, and that the difference between voiced and voiceless articulation was a matter of local variation (1971: 127). Since his book was published posthumously, the origin of his view is hard to determine. He remarks that the simple voiced stops of Indo-European are equivalent to the glottalic set in other language families with regard to ancient symbolism (p. 219). 54 Indo-European phonology 1.5. Gamkrelidze and Ivanov have proposed on the basis of Pedersen’s reasoning that the PIE unaspirated voiced stops were glottalic (1972: 16). This interpretation allows them to explain the absence of roots with two glottalic stops by a dissimilation rule (1973: 153). They also reformulate Grassmann’s law as a PIE rule of allophonic variation (1980: 30-32). This seems to be at variance with the Latin evidence, e.g. fīdō ‘I trust’ < *bheidh-. 1.6. A similar proposal was put forward by Paul Hopper, who pointed not only to the absence of *b and the root structure constraints, but also to the absence of the glottalic stops from inflectional affixes (1973: 157). His view is repeated several times in later articles. 1.7. Jens Rasmussen has proposed to derive traditional *t, *d, *dh from earlier *T, *t, *d, “T being a cover-symbol for any emphatic stop however phonetically realized (glottalized, pharyngealized, or just stronger)” (1974: 11). The same reconstruction is implied in Illič-Svityč’s Nostratic dictionary (1971: 147). It is based on the false assumption that glottalic or emphatic stops are stronger than others. 1.8. As early as 1948 André-Georges Haudricourt reached the conclusion that the PIE unaspirated voiced stops were glottalic and that the original pronunciation was preserved in East Armenian (1975: 267). His argumentation was based on the types of phonetic development attested in the Far East. The negative attitude of Jules Bloch and Jerzy Kuryłowicz toward his view apparently kept him from publishing it. 1.9. In my own exposition I have reinterpreted *t, *d, *dh as fortis, glottalic lenis, and aspirated lenis, respectively (K032: 107). The rephonemicization of either the aspirated or the glottalic stops (or both) as voiced provoked a number of consonantal mutations (or mergers) in the separate branches of Indo-European. In Indo-Iranian, Balto-Slavic, Celtic and Albanian, both *d and *dh became phonemically voiced. Greek has voiced *d and voiceless *dh, while the converse holds for Germanic. Italic, where *d is voiced, and East Armenian, where it is voiceless, have both voiced and voiceless reflexes of *dh. No rise of phonemically voiced stops took place in Anatolian and Tocharian. 1.10. George Dunkel has rightly pointed out the circularity of the typological argument (1981: 566). If our reconstructions are tailored to typological expectations, they acquire a bias toward the average language type. The more aberrant the structure of the proto-language is, the stronger the bias and the larger the difference between the real and the reconstructed proto-language becomes. In my view, the discussion suffers from an unfortunate lack of distinction between theory and method. Typological considerations are an extremely useful Proto-Indo-European glottalic stops 55 heuristic device. They can never take the place of the evidence, however. It is therefore remarkable that so little attention has been paid to the comparative evidence, which is abundantly present for those who are ready to see it. 2. BALTIC Latvian preserves the glottalic feature of the unaspirated voiced stops as a glottalic tone on the preceding vowel in originally pretonic syllables, e.g. pêʚds ‘footstep’, nuôgs ‘naked’, Vedic padám, nagnás (K025). The glottalic tone represents the merger of the glottalic feature with the reflex of the PIE laryngeals. Under the stress, it is preserved in the Žemaitian dialects of Lithuanian (Zinkevičius 1966: 34). The usual view that the glottalic tone is of secondary origin cannot be correct because it does not explain the rise of the glottalization. More probably, the Proto-Baltic acute was a glottal stop which was lost under rising and falling tone movements that originated in the separate languages (K025: 324-328). It is widely believed that the Proto-Baltic circumflex, like its Greek counterpart, resulted from early contractions while all other ancient long vowels are acute. This view is incorrect. As I have argued elsewhere, the acute is the phonetic reflex of the PIE laryngeals and the glottalic feature of the unaspirated voiced stops, whereas all other ancient long vowels are circumflex (K064). The latter include the following categories: (1) Long vowels from contractions, e.g. Lith. gen.sg. algõs ‘salary’, cf. Greek ἀλφῆς. (2) Lengthened grade vowels in the nom.sg. form of stems in a resonant, e.g. Lith. akmuõ ‘stone’, duktÄ ‘daughter’, Greek ἄκμων, θυγάτηρ. (3) Long vowel preterits, e.g. Lith. Ämė ‘took’, bÄrė ‘strewed’, lÄkė ‘flew’. The acute of g°rė ‘drank’ reflects the root-final laryngeal, not the long root vowel. (4) Lith. 3rd person future forms, e.g. duõs ‘will give’, kalbÄs ‘will speak’. Since the long vowel is not shortened in polysyllabic stems, the metatony must be older than Leskien’s law (K014: 86). Assuming that the PIE laryngeals were lost after lenghtened grade vowels, I connect the metatony in this category with the lengthened grade in the 2nd and 3rd sg. active forms of the Vedic sigmatic aorist injunctive (K065). (5) Lengthened grade vowels in original root nouns, e.g. Lith. gėlà ‘pain’, žolÄ ‘grass’, mėsà ‘meat’, cf. Slavic žalь, Prussian acc.sg. sālin, Vedic m¡s. (6) Latvian nom.sg. sÅls ‘salt’ and gùovs ‘cow’, cf. Greek ἅλς, βοῦς, Vedic gáus. Here again, I assume that the laryngeal was lost after a lengthened grade vowel. The expected acute reflex of the laryngeal is found in Lith. sólymas ‘brine’ (Būga 1959: 584). (7) Lith. nom.sg. -Ä. In my view, this ending originated from the loss of the laryngeal after a lengthened grade vowel in the nom.sg. form of the root noun 56 Indo-European phonology which is represented in arklìdė ‘stable’, avìdė, ‘sheepfold’, alùdė ‘pub’, pelùdė ‘chaff store’, also žvaigždÄ ‘star’, Vedic -dh¡, Latin -dēs, and the Greek passive aorist sufix -θη-. Thus, I conclude that the PIE laryngeals merged into a glottal stop, which merged with the glottalic feature of the unaspirated voiced obstruents. The Proto-Baltic circumflex is simply the absence of a glottal stop. 3. SLAVIC Baltic and Slavic shared the merger of the PIE laryngeals with the glottalic feature of the unaspirated voiced stops into a glottal stop. Slavic subsequently lost the glottal stop with compensatory lengthening of a contiguous vowel in pretonic and post-posttonic syllables (K014: 11). Under the stress and in the first posttonic syllable the loss of the glottal stop yielded the rise of the new timbre distinctions. As a result, the presence versus absence of the glottal stop is reflected as a difference between short and long reflexes of the “long” vowels. The difference is usually preserved in Serbo-Croatian and has left traces in the other languages. Thus, the short ě, a of SCr. jÇbuka ‘apple’, jÈdēm ‘I eat’, pÇdnēm ‘I fall’, sjÈdnēm ‘I sit down’, pòbjegnēm ‘I flee’, jÇgnje ‘lamb’, jÇgoda ‘strawberry’ reflects the glottalic feature of the following unaspirated voiced obstruent. Long vowels without a glottal stop were not shortened in Slavic, except under special conditions. Thus, lengthened grade vowels are generally long in Serbo-Croatian, e.g. in the following instances: (1) The word žÈrāv ‘crane’, Czech žeráv, reflects an original nom.sg. form *gʄerōu, cf. Latin grūs (Vaillant 1958: 172). The long vowel is in agreement with the circumflex of Lith. akmuõ ‘stone’ and Latvian âbuõls ‘apple’. (2) Sigmatic aorist: 1st sg. dònijeh next to donèsoh ‘brought’, and the isolated infinitive rËjet (Dubrovnik) next to rèći ‘to say’ (Vaillant 1966: 60). Similarly 1st sg. mrËjeh, ùmrijeh ‘died’, klÌh, zàklēh ‘swore’. (3) The tonal alternation between dÇh ‘I gave’ and dÍ ‘he gave’ is the same as between Lith. dúosiu and duõs ‘will give’. I think that it reflects the loss of the laryngeal after a lengthened grade vowel in the aorist injunctive. (4) Original root nouns, e.g. rËječ ‘word’, Tocharian B reki, cf. Vedic v¡k, Latin vōx, Prussian acc.sg. tārin. Other examples: žÍr ‘live coals’, žára ‘nettles’, pΞār ‘fire’, Ïgār ‘fallow’, gÍr ‘soot’, čÍr ‘magic’ (Czech čár and čára), nÈmār ‘negligence’, sÍm, sáma, sámo ‘alone’. There is additional evidence for the view that the acute was a glottal stop in the fact that it blocked the progressive accent shift (K014: 14). This constraint has a significant parallel in Avar, where “stress shifted to the second syllable from the first non-pharyngealized one” (Dybo et al. 1978: 19). It seems to me that the Proto-Indo-European glottalic stops 57 Balto-Slavic evidence suffices to shift the burden of proof onto the adversaries of the glottalic theory. 4. ARMENIAN Shortly after the turn of the century, Pedersen challenged the obtaining views on the Armenian consonant system with the hypothesis that the voiced stops of the classical language were aspirated (1906: 336-342). This view was accepted by Vogt (1938: 327), who discussed the matter in detail in a separate study (1958), and later by Allen (1951: 134) and Benveniste (1959). Garibjan went a step further and surmised that the voiced stops in the western dialects which correspond to the voiceless stops of the classical language constitute an archaism (1959). Agajan has demonstrated that this view cannot be upheld (1960). The very sources from which Vogt and Benveniste drew their inspiration (Adjarian 1909, Allen 1950) permit entirely different conclusions, which are apparently supported by the newly discovered southern dialects (Garibjan 1958). The following analysis will be based on three principles: (1) A reconstruction of the Common Armenian consonant system on the basis of the modern dialects must logically precede a comparison with material from other Indo-European languages. (2) If the consonant systems of two related dialects differ in more than a single feature, the historical connection between them involves at least two distinct developments. (3) If a single uninterrupted central area differs from the peripheral areas with respect to a specific feature, it is probable that the central dialect has innovated. In order to simplify the discussion, I number the modern dialects in such a way that the first digit reflects the correspondence with classical t and the second digit the correspondence with classical d, both in word-initial position, and that a minimum difference between numbers reflects a minimum difference between dialects in terms of features. In the following list I give, next to the number of each dialect, the reflex of classical t, d, th, the corresponding number in the classifications of Vogt (1958) and Garibjan (1959), and a typical representative. 11 12 13 21 22 23 20 t t= t t d d= d d d t d dh t d dh th = th th th th th th th th Vogt 4 3 1 – 5a 2 5b Garibjan 7 6 2 4 3 1 5 examples: Van Agulis Erevan Sasun Trabzon Sivas Malatia 58 Indo-European phonology The dialects 11, 22, 20 have apparently come into existence as a result of the neutralization of a phonological opposition. According to Vogt, “il s’agit évidemment d’une simplification secondaire des systèmes centraux” (1958: 148), i.e. 13 and 23. Should 11 indeed be derived from 13? Since these dialects differ in two features, we have to assume an intermediate stage. If the voicedness was lost earlier than the aspiration, the reflexes of classical d and th must have merged, which is not the case in 11. If the aspiration was lost earlier than the voicedness, the intermediate stage was identical to the system of dialect 12. But there is no reason why 12 should be derived from 13; both 11 and 13 may actually have to be derived from 12. Geographically, the area which 11 and 12 occupy together forms a semi-circle around the central dialects: Arčeš – Van – Xoy (11) – Agulis – Meghri (12) – Karabagh – Kanaker (11) – Lori – Tiflis – Artvin (12). This situation suggests that 13 must be derived from 12. The dialects 22 and 20 may indeed be derived from 23, but either of them can also be derived from 21, a dialect which Vogt does not take into account because it is not covered in Adjarian’s monograph. Moreover, 22 may be derived from 12 in the same way as 23 can be derived from 13. A choice can only be made by taking into account the geographical distribution of the dialects. Since the position of Trabzon (22) with respect to the Artvin-Tiflis area (12) is the same as that of Little Armenia (23) with respect to Central Armenia (13), it is reasonable to assume that the historical relationship between the Trabzon dialect and its eastern neighbour is the same as that between 23 and 13. The suggestion that 13 must be derived from 12 and the impossibility of deriving 12 from 22 then involve the consequence that the dialects of Trabzon and Little Armenia must be derived from 12 and 13, respectively. If this is correct, the semi-circle discussed above can be extended to Trabzon. Other parts of 22 may have different historical connections. Thus, the isolated dialect of Maraš (22) must probably be derived from the contiguous dialect of Hadjin-Zeytun (21). The Malatia-Urfa area (20) is situated within the semi-circle Sasun – Dersim – Hadjin – Beylan – Svedia (21) and must therefore be derived from the latter. Which distinctive features can be reconstructed for the oldest stage of the apparently archaic dialect 12? I think that the answer is provided by Allen's phonetic analysis of an East Armenian dialect (1950). The unaspirated voiceless plosives are glottalic (“ejective”) in this dialect, whereas the voiced stops of the classical language are voiceless in initial position. Thus, it is a transitional dialect between 12 and 11, having lost the voicedness of d while retaining the opposition between d and t. Actually, the opposition between voiced and voiceless initial stops was restored by the introduction of the loan words beg and boy (Allen 1950: 202). The term “potential voiced aspirates” which Allen applies to the reflexes of classical d etc. has given rise to misunderstanding on the part of Benveniste, who inferred the existence of voiced aspirates from the description (1959: 50). In fact, voicing and aspiration are mutually exclusive, the “potential Proto-Indo-European glottalic stops 59 voiced aspirates” being voiced and unaspirated if preceded by a nasal, and lightly aspirated and voiceless in final position after -r-. These are precisely the positions where almost all Armenian dialects show unaspirated voiced and aspirated voiceless plosives, respectively (Pisowicz 1976a: 61-62). In initial position, the “potential voiced aspirates” are voiceless. “They are distinguished from the ejectives by having pulmonic as opposed to glottalic plosion, and from the aspirates by the absence of voiceless breath on release. The most notable feature differentiating them from the ejectives, however, is to be found in a following vowel, which is articulated with markedly stronger breath-force and on a lower pitch than is general in other but comparable contexts” (Allen 1950: 200). The transfer of the distinctive feature to the following vowel is carried through completely in a part of the dialects 11 (Pisowicz 1976b: 215-216). In the original system, the glottalic articulation of the “ejectives” was apparently distinctive. Indeed, 19th century loan words from Russian showed aspirated plosives in Armenian, e.g. p‘eč‘, manet‘ from peč', moneta (Pisowicz 1976a: 18). Thus, I do not share the usual view that the glottalic articulation of the unaspirated voiceless stops in the Tiflis-Artvin area is due to a Caucasian substratum. It is more probable that the latter simply favored the preservation of a feature which was already present. The newly discovered dialects 21 seem to corroborate the antiquity of the glottalic articulation. According to Garibjan (1959: 85-86) and Pisowicz (1976a: 78), these dialects must be derived from the western system 23. Since 21 and 23 differ in two features, we have to assume an intermediate stage. If the voicedness was lost earlier than the aspiration, the intermediate stage was identical to 20. If the aspiration was lost earlier than the voicedness, the intermediate stage was identical to 22. In either case two series would have merged, which is not the case in 21. I conclude that 21 must be derived from an eastern dialect. Since the geographical position of the southern dialects (21) with respect to the Van area (11) is the same as that of Little Armenia (23) with respect to Central Armenia (13), it is reasonable to assume that the historical relationship between 21 and 11 is the same as that between 23 and 13. We come to the conclusion that the southern dialects developed from the system of their eastern neighbour at a time when the latter had not yet lost the distinction between the original (glottalic) unaspirated voiceless stops and the ones that originated from the devoicing of the classical voiced stops. Indeed, the dialect of Šatax (between Van and Sasun) distinguishes between two series of unaspirated voiceless stops (Pisowicz 1976a: 66) and is in this respect transitional between 11 and 21. We can now connect the two semi-circles discussed above: together they constitute an uninterrupted line of dialects which are archaic with respect to the encircled areas. 60 Indo-European phonology We have now established the following relative chronology: (1) Rise of aspiration in voiced stops (12 > 13). (2) Devoicing of unaspirated voiced stops (12 > 11). (3) Voicing of glottalic stops (11 > 21, 12 > 22, 13 > 23). (4) Elimination of unaspirated voiceless stops (21 > 22, 21 > 20). The absolute chronology can only be established on the basis of loan words. Since loans from Arabic are subject to shifting while loans from Turkic are not (Agajan 1960: 44), we have to date (3) between the 7th and the 10th centuries. It is possible that (1) was in progress during the classical period, as Djahukian suggests (1967: 76). Thus far I have limited the discussion to word-initial plosives because it is the position of maximal contrast in the Armenian dialects. The same type of analysis can be applied to other positions. The generalization of voiced stops after nasals and voiceless aspirates after prefinal -r- are probably early developments because they have affected the large majority of dialects. Intervocalically, the following types are found: 10a 11a 12a 20a 21a 23a t t t= t d d d d th = t d th = t dh th th th th th th th examples: Karabagh, Lori Van Agulis Malatia, Trabzon, Erevan, Tiflis Sasun Sivas The areas 12a and 23a are small islands within 10a and 20a, respectively, while 11a and 21a are considerably smaller than 11 and 21. A large part of the eastern dialects have the western (voiced) reflex of intervocalic t etc. Almost all northern dialects have been subject to the aspiration of intervocalic d etc. Following the principles which have been put forward above, one arrives at the same reconstruction and the same relative chronology as have been established for initial plosives, plus one additional development: (5) Devoicing of voiced aspirates (13a > 10a, 23a > 20a). The reconstructed Common Armenian obstruent system now appears as follows: aspirated voiced voiceless th plain d glottalic t Proto-Indo-European glottalic stops 61 The first Armenian consonant shift is seen to consist in the loss of the aspiration of PIE *dh, the rephonemicization of lenis PIE *d as voiceless, and the weakening of the occlusion of PIE *t. 5. VEDIC AND AVESTAN The Balto-Slavic evidence points to a series of preglottalized voiced obstruents as the earliest reconstructible reflex of the PIE glottalic stops. Alexander Lubotsky has adduced Indo-Iranian evidence in support of this interpretation (1981: 137). The root of Greek πήγνῡμι ‘fasten’ is represented in the Rgveda as follows: 3rd sg. p¡paje ‘stiffened’, pajrá- ‘firm’, pakṣá- ‘wing’, pakṣín- ‘bird’, pákṣas- ‘side’, p¡jas- ‘frame’, pājasyà- ‘flank’. The distribution of short and long root vowel can be explained by the assumption that the laryngeal was lost before a glottalic obstruent when the latter was followed by another consonant. Lubotsky adduces fourteen roots in laryngeal plus glottalic stop with short root vocalism in Old Indic, five of which have Avestan correspondences with a short root vowel. Thus, Lubotsky’s law must be dated to the Indo-Iranian period. The development is understandable if a sequence of laryngeal plus glottalic stop was realized as a glottal stop plus preglottalized voiced obstruent. 6. SINDHI AND PANJABI The hypothesis that the unaspirated voiced stops of Old Indic were preglottalized is confirmed by immediate comparative evidence from Sindhi, which has preserved the glottalic articulation. This language has a threefold opposition between voiced stops: the unconditioned reflexes of the d and dh series are glottalic and aspirated, respectively, while dissimilation of the dh series before aspirates of recent origin has given rise to a plain voiced series, e.g. ’gāhu ‘bait’ < grāsa-, gāhu ‘fodder’ < ghāsa-. It can be shown that the glottalic stops are at least older than the loss of the PIE laryngeals (K038: 17-18). There is no reason to disclaim their PIE origin. The glottalic articulation cannot be attributed to external influence because the neighbouring languages do not present anything comparable. The Panjabi material also requires the former existence of preglottalized voiced obstruents at a recent stage. In this language, the voiced aspirates have become voiceless and unaspirated, yielding a low tone on the following vowel, e.g. kòṛā ‘horse’, Hindi ghoṛā. Since the voiceless aspirates have been preserved as a separate category, the dh series was not phonemically aspirated at the time of the devoicing (Haudricourt 1975: 271). It follows that the glottalic stops were preserved at that stage. Moreover, the d series did not lower the tone of a following vowel. This also points to the preservation of the glottalic feature. 62 Indo-European phonology 7. GREEK The usual reconstruction of the PIE word for ‘100’ *kʄṃtóm does not account for the initial vowel of Greek ἑκατόν. The comparison of this vowel with the word for ‘one’ *sem- is at variance with the indeclinability and the syntactic behavior of ἑκατόν. The original character of the indeclinability is evident both from the preservation of the final nasal in composition and from the absence of a plural form. If the latter had existed at an earlier stage, it would hardly have been replaced with a derivative formation. Thus, I think that the Greek form and its syntax are more archaic than is generally assumed. The initial vowel of ἑκατόν can be explained if we start from PIE *dkʄṃtóm and assume that the buccal features of the initial consonant were lost while its glottalic feature merged with the reflex of the PIE laryngeal *H1 to yield e-, with subsequent adoption of the aspiration from the word for ‘one’ (K060: 98). This explanation, which presupposes that *d was a glottalic obstruent at the time under consideration, has the additional advantage of accounting for the long vowel in the decades, e.g. πεντήκοντα ‘50’ < *penkwe-dkʄomt-. It also accounts for the coexistence of the southern form εἴκοσι < *ἐϝÕκοσι and the northern form ϝÕκατι ‘20’, both of which can be derived from PIE *dwidkʄṃti if we assume that partial dissimilation of the initial consonant yielded *H1- whereas total dissimilation yielded zero. This again presupposes that the glottalic articulation of *d had been preserved at the time of dissimilation. When did *d merge with *H1 in ἑκατόν, εἴκοσι and πεντήκοντα? I think that τρι³κοντα ‘30’ < *triH2-dkʄomt- and West Greek τετρώκοντα ‘40’ < *kwetwṛdkʄomt- allow us to date the merger of *d with the laryngeals to a stage which was posterior to the rise of colored epenthetic vowels, but anterior to the eventual loss of the laryngeals. Since the development of colored epenthetic vowels is specifically Greek, it follows that the PIE glottalic stops were preserved up to a comparatively recent stage. 8. LATIN In Latin we find a long root vowel in āctus ‘driven’, lēctus ‘gathered’, where the velar stop belongs to the PIE glottalic series, and a short root vowel in factus ‘made’, vectus ‘carried’, where it belongs to the fortis or aspirated series. The Balto-Slavic and Indo-Iranian developments show that the glottalic feature characterized the initial part of the obstruent, whereas the aspiration characterized its final part in view of Bartholomae’s law. When voicedness became phonemic, the aspiration in the cluster *-ght- was simply lost, but the glottalic articulation in *-gt- was preserved as a feature of the preceding vowel. As in the Greek and Balto-Slavic instances cited above, it merged with the reflex of *H1. Proto-Indo-European glottalic stops 63 The same development accounts for the long vowel in the decades, e.g. vīgintī ‘20’, quadrāgintā ‘40’, septuāgintā ‘70’. The voicing of *k to g in these words must be attributed to the preceding nasal in *septṃdkont- ‘70’ and *newṇdkont- ‘90’ (Thurneysen 1883: 313). This voicing rule was Italo-Celtic, as is clear from Old Irish sechtmogo ‘70’, as opposed to cethorcho ‘40’. The quantitative difference between Latin vīgintī and Old Irish fiche ‘20’, like the difference between Latin rēctus ‘straight’ and Old Irish recht ‘law’, shows that the loss of the glottalic articulation was posterior to the disintegration of Italo-Celtic. The glottalic feature was lost after a consonant, e.g. in the nasal presents stringō ‘tighten’, pingō ‘paint’ < *-g-, which merged with fingō ‘touch‘, mingō ‘urinate’ < *-gh-. This accounts for the absence of a long vowel in strictus, pictus, which adopted the vocalism of fictus, mictus on the analogy of the present tense. The glottalic feature was apparently absorbed by the preceding laryngeal in lassus ‘tired’ < *lḤ1dtos, just as the laryngeal was absorbed by the following glottalic obstruent in Vedic pajrá- ‘firm’. The initial syllable of sedeō ‘sit’ was prefixed to the form *sdtos in -sessus. Thus, the Latin evidence supports the reconstruction of a series of preglottalized obstruents. 9. GERMANIC In his monograph on the West Jutland stød, K. Ringgaard concludes “that the v-stød is only found immediately before the plosives p, t, k, and that it is found wherever these stand in an original medial position, following a voiced sound in a stressed syllable. The exceptions to this are certain types of loan-words from a later period” (1960: 10, 195). “The v-stød is a complete occlusion of the vocal cords, combined with the diaphragm’s movement of inhalation” (p. 199), i.e. a full-fledged glottal stop. Ringgaard dates the rise of the West Jutland stød to the 12th century because it is characteristic of “all then existing medial plosives”. His view that it is a spontaneous innovation of the westernmost dialects of Danish can hardly be called an explanation. Moreover, it does not account for the parallel development of preaspiration in Icelandic. Preaspiration is found not only in Icelandic, but also in Faroese, Norwegian, and the Gaelic dialects of Scotland. Phonetically, the preceding vowel is cut short and continued as a whisper; a preceding resonant (m, n, l, r) is partly or wholly unvoiced. The distribution of preaspiration in Icelandic is the same as in the Norwegian dialect of Jæren (Oftedal 1947). We can therefore conclude that it is “an example of a feature taken to Iceland by the original settlers” (Chapman 1962: 85). Carl Marstrander has argued that the preaspiration in Scottish Gaelic is due to a Norse substratum (1932: 298). He advances the hypothesis that the Norwegian preaspirated stops represent a retention of the clusters hp, ht, hk, which developed into geminates elsewhere (p. 302). This theory implies three 64 Indo-European phonology developments: tt < ht in East Norse, ht < tt in West Norse, and ht < t in West Norse in those positions where the preaspirated stop does not reflect a cluster, e.g. Icelandic epli ‘apple’, vatn ‘water’, mikla ‘increase’, hjálpa ‘help’, verk ‘work’. Here the preaspirated stop appears to be the phonetic reflex of a PIE unaspirated voiced obstruent. Both the West Jutland stød and the preaspiration receive a natural explanation if we assume that Early Proto-Germanic, like Proto-Balto-Slavic and Proto-Indo-Iranian, possessed a series of preglottalized voiced stops. Devoicing yielded a series of Late Proto-Germanic sequences Ȥp, Ȥt, Ȥk, the glottal stop of which was lost under various conditions. Weakening of the glottal stop in West Norse yielded preaspiration, while its assimilation to the following obstruent gave rise to a series of geminates in East Norse, with the exception of Danish, where the sequences were subject to lenition and the glottal stop was preserved in the western dialects. I wonder if Swedish vecka ‘week’, droppe ‘drop’, skepp ‘ship’ reflect a dialect that escaped the earliest loss of the glottal stop. One may wonder if preglottalization had been preserved in word-initial position in Late Proto-Germanic. There is positive evidence for this in the West Jutland stød of fattig < fátǿkr ‘poor’ < ‘few-taking’, sytten ‘seventeen’. Apart from the straightforward explanation of the West Jutland stød and the Icelandic preaspiration, the theory advanced here has the advantage of accounting in a principled way for the existence of several layers of gemination, which can now be viewed as retentions rather than innovations: (1) mp, nt, nk yielded pp, tt, kk in the larger part of Scandinavia. The nasal consonant was apparently devoiced by the glottal feature which preceded the stop, and subsequently lost its nasalization. (2) k yielded kk before j and w. Similarly, t yielded tt before j in a limited area, e.g. Swedish sätta ‘set’. The development cannot easily be identified with the change of g into gg before j because the latter involves the transformation of a fricative into a stop. (3) p, t, k yielded pp, tt, kk before r and l in West Germanic. The same development is found sporadically in Scandinavia. Here again, the geminate may have originated from the assimilation of a glottal stop to the following buccal stop. It is possible that the theory put forward here has certain consequences for the interpretation of the West Germanic material. Firstly, the High German sound shift may have resulted from a lenition of the buccal stops with concomitant oralization of the preceding glottal stop. If this is correct, the glottalization must have been preserved at the time of the shift. Secondly, the absence of aspirated stops from Dutch and Frisian may be due to an early loss of preglottalization in this area. Thirdly, the English glottal stop may be much more ancient than is Proto-Indo-European glottalic stops 65 commonly assumed. It appears that these possibilities merit further consideration. 10. CONCLUSION The Balto-Slavic, Indo-Iranian, Italo-Celtic and Germanic evidence points to the former existence of a series of preglottalized voiced obstruents. This reconstruction is supported by evidence from Greek and Armenian. In the modern languages, glottalization has been preserved in Latvian and Sindhi, and in dialects of Lithuanian, Armenian, and Danish. The preglottalized voiced obstruents may have originated from a common innovation of all IndoEuropean dialects except Anatolian and Tocharian, where voicedness never became a phonemically relevant feature. We may now reconsider the typological argumentation cited in the introduction. It will be seen that my own view is closest to the positions held by Andreev and Swadesh: the proto-language had simple fortes, glottalic lenes, and aspirated lenes stops, corresponding to traditional *t, *d, *dh. As in modern Icelandic and in the southern dialects of East Armenian, all stops were usually voiceless. This fits in with the absence of a voiced counterpart to the PIE fricative *s. The typological reinterpretation of the obstruent system was based on the poor attestation of the labial stop *b, the incompatibility of two glottalic stops in the root, and the incompatibility of fortes and aspirates in the root. The latter constraint has a notable parallel in Austronesian (Bradshaw 1979). The absence of initial *b- can hardly be explained by the loss of an earlier *p-, as Pedersen maintained (1951: 12). I think that the correct solution was indicated by Thurneysen (1908): *b- lost its glottalic feature and merged with *p-, cf. especially Vedic píbati ‘drinks’, Old Irish ibid, Armenian ǩmpem (with secondary nasal infix), Latin bibō (with restored reduplication). Medial *-b- was preserved, e.g. in Latvian âbuõls ‘apple’ < *abōl-, Lith. obelìs < *abel-, Russian jábloko < *abl-, where the ablaut guarantees the Indo-European origin of the word. It has been conjectured that Germanic and Armenian preserve the original voicelessness of the PIE glottalic stops. I think that this is incorrect. In the case of Germanic, the fact that the glottal feature precedes the buccal stop suggests that the latter was voiced at an earlier stage. In the case of Armenian, its close relationship with Greek, Albanian, Balto-Slavic and Indo-Iranian suggests that the voicelessness is secondary. Thus, I think that the original system was not preserved outside Anatolian and Tocharian. PROTO-INDO-EUROPEAN TONES? 1. Classical Indo-Europeanists assumed the existence of a distinction between acute and circumflex long vowels in the proto-language on the basis of evidence from Indo-Iranian, Greek, Balto-Slavic and Germanic (e.g. Hirt 1929). It will be argued that the attested differences arose in the separate branches after the dissolution of the proto-language. 1.1. The Indo-Iranian evidence for a circumflex tone is based on the disyllabic reading of certain long vowels in Vedic and Gathic, e.g. Vedic bh¡s-, bháas‘light’, préṣṭha-, práyiṣṭha- ‘dearest’, acc.sg. pánthām, pánthaam ‘path’, Gathic acc.sg. paθaam ‘path’, mazdaam ‘God’, gen.pl. -aam, subj. -aa-. The disyllabic reading is regular in Gathic and sporadic in Vedic. It can be shown that in Gathic it occurs whenever we expect an intervocalic laryngeal (cf. Monna 1978: 97-103). It follows that intervocalically the laryngeals yielded a glottal stop in Indo-Iranian and that this glottal stop was preserved as a hiatus in the separate languages. This eliminates the evidence for tonal distinctions in Indo-Iranian and, consequently, the Indo-Iranian evidence for a Proto-Indo-European circumflex. 1.2. The Greek evidence for tonal distinctions is limited to the word-final syllable, where a long vowel or diphthong may be either rising = acute or falling = circumflex (cf. Risch 1975: 472). The same distinction on unstressed final diphthongs is reflected as the converse tonal distinction on a long vowel or diphthong in the preceding syllable. Since tonal distinctions on unstressed syllables are exceptional in a language with free stress, this suggests that the Greek tones are due to a secondary development. If we eliminate the cases where the circumflex is the result of recent contractions or analogical patterning (cf. Bally 1945: 42), we find that endings are circumflex whenever they were disyllabic at an earlier stage whereas original long vowels are acute. The following instances are of particular interest: 1) Acute ending in nom.pl. οἶκοι < *-oy ‘houses’ versus circumflex ending in loc.sg. οἴκοι < *-o-i ‘at home’, cf. the dat.sg. ending of the consonant stems -i. 2) Circumflex 3rd sg. present ending in κελεύει < *-e-i ‘orders’, where -i was taken from the athematic flexion as a tense marker and added to the original thematic present ending -e. 3) Circumflex 3rd sg. optative ending in κελεύοι < *-o-īt (cf. K097: 237). The distinction between acute and circumflex was lost in non-final syllables, cf. 1st pl. κελεύοιμεν, 2nd pl. δύναισθε ‘be able’, ὄναισθε ‘have profit’. The generalization of the root vowel before *-ī- yielded the accentuation of 1st pl. διδοῖμεν ‘give’, 68 Indo-European phonology ἱσταῖμεν ‘set up’, τιθεῖμεν ‘put’, 2nd pl. ἱσταῖσθε ‘stand’, where the place of the stress points to earlier -οϊ-, -αϊ-, -εϊ- (cf. ibidem). I conclude that Greek offers no evidence for an inherited distinction between acute and circumflex tones. The attested distinction arose after the loss of the PIE laryngeals. 1.3. The tonal distinctions of Baltic and Slavic are extremely important for the reconstruction of Proto-Indo-European. They do not, however, represent a tonal distinction of the proto-language. As I have argued in detail elsewhere (K064), the acute tone of Baltic and Slavic resulted from the phonetic merger of the PIE laryngeals with the glottalic feature of the unaspirated voiced stops into a glottal stop. It has been preserved unchanged in originally pretonic syllables in Latvian, e.g. pêʚds ‘footstep’, nuôgs ‘naked’, Vedic padám, nagnás, and under the stress in the Žemaitian dialects of Lithuanian. The original circumflex of Baltic and Slavic was simply the absence of a glottal stop. It is found in the following categories: 1) Long vowels from contractions, e.g. Lith. gen.sg. algõs ‘salary’, Gr. ἀλφῆς. 2) Lengthened grade vowels in the nom.sg. form of stems in a resonant, e.g. Lith. akmuõ ‘stone’, duktÄ ‘daughter’, Latvian âbuõls ‘apple’, Serbo-Croatian žÈrāv ‘crane’, Czech žeráv. 3) Sigmatic aorist, e.g. SCr. 1st sg. dònijeh next to donèsoh ‘brought’, ùmrijeh ‘died’, zàklēh ‘swore’. 4) Long vowel preterit, e.g. Lith. Ämė ‘took’, bÄrė ‘strewed’, lÄkė ‘flew’. The acute of g°rė ‘drank’ reflects the root-final laryngeal, not the lengthened grade root vowel. 5) Lith. 3rd person future forms, e.g. duõs ‘will give’, kalbÄs ‘will speak’. Since the long vowel is not shortened in polysyllabic stems, the metatony must be older than Leskien’s law. Assuming that the PIE laryngeals were lost after lengthened grade vowels, I connect the metatony in this category with the lengthened grade in the 2nd and 3rd sg. active forms of the Vedic sigmatic aorist injunctive (cf. K065). 6) The tonal alternation between SCr. 1st sg. dÇh and 3rd sg. dÍ ‘gave’ is the same as between Lith. dúosiu and duõs ‘will give’. I think that it reflects the loss of the laryngeal after a lengthened grade vowel in the aorist injunctive. 7) Lengthened grade vowels in original root nouns, e.g. Lith. gėlà ‘pain’, žolÄ ‘grass’, mėsà ‘meat’, SCr. rËječ ‘word’, čÍr ‘magic’, sÍm ‘alone’. 8) Latvian nom. sg. sÅls ‘salt’ and gùovs ‘cow’, cf. Latin sāl, Vedic gáus. Here again, I assume that the laryngeal was lost after a lengthened grade vowel. The expected acute reflex of the laryngeal is found in Lith. sólymas ‘brine’. 9) Lith. nom.sg. -Ä. In my view, this ending originated from the loss of the laryngeal after a lengthened grade vowel in the nom.sg. form of the root noun Proto-Indo-European tones? which is represented in arklìdė ‘stable’, avìdė ‘sheepfold’, alùdė ‘pub’, pelùdė ‘chaff store’, also žvaigždÄ ‘star’, Vedic -dh¡, Latin -dēs. Thus, the Balto-Slavic circumflex is the normal reflex of any long vowel of nonlaryngeal and non-glottalic origin. It provides no evidence for a Proto-IndoEuropean tonal distinction. 1.4. The Germanic evidence for tonal distinctions is limited to originally final syllables, where original long vowels and diphthongs have a short and a long reflex. In my view, the long reflex is regular before original word-final obstruents and analogical in the gen.pl. ending, while the short reflex is regular word-finally and before word-final nasals. The following categories are of major importance: 1) Gothic undaro ‘under’ < *-ōt, cf. Vedic adhar¡t ‘below’, and other adverbs in -o, e.g. Go. þiubjo ‘secretly’, þaþro ‘thence’, galeiko ‘similarly’, OHG gilīhho, OE gelīce, as opposed to the word-final reflex of the same vowel in Go. giba ‘gift’, ON giNjf, OE giefu, etc. 2) Go. hidre ‘hither’, OE hider, cf. Latin citrō. This category apparently adopted the original ablative ending. The instrumental in *-ē was preserved in dat.sg. daga ‘day’, imma ‘to him’, hindana ‘(from) behind’, OE hindan. 3) Go. 3rd sg. weak preterit nasida ‘saved’ reflects the original final vowel *-ē, not *-ēt, and can be identified with the Lithuanian preterit in -ė, as I have argued elsewhere (K071). The other languages adjusted the preterit endings of the weak verbs to those of the verb ‘to do’. 4) The Go. gen.pl. endings -o and -e must be derived from Proto-Germanic *-ōan and *-eian, as I have argued in detail elsewhere (K030: 291-293). 5) The nom.sg. ending of the n-stems is phonetically regular in Go. guma ‘man’, OE tunge ‘tongue’, ēage ‘eye’, OHG zunga, ouga, and analogical in Go. tuggo, augo, ON tunga, auga, OE guma, OHG gomo (ibidem). The ending of ON gumi was taken from the ion-stems (cf. Lid 1952). 6) Since final *-ai yielded -a in Go. bairada ‘he is borne’, bairanda ‘they are borne’, faura ‘before’, cf. Gr. παραί ‘beside’, the preservation of the diphthong in the optative bairai ‘he may bear’ must be attributed to the original final *-t. The pronominal nom.pl. ending of the adjective, e.g. blindai ‘blind’, was taken from the monosyllabic demonstrative pronoun þai at a recent stage. Long final diphthongs had been preserved in Proto-Germanic, e.g. Go. ahtau ‘eight’, dat.sg. gibai ‘gift’, anstai ‘favor’. I conclude that the Germanic material is fully explained in terms of segmental features and offers no evidence for an inherited tonal distinction. 2. Though the distinction between acute and circumflex long vowels cannot be attributed to Proto-Indo-European, it does not follow that the proto-language 70 Indo-European phonology had no distinctive tone. There are several indications that point to the existence of a tonal distinction at a stage in the development of Proto-Indo-European. (On the morphological nature of the tone cf. the literature cited under 2.3 below.) 2.1. The Old Indic accentual system is tonal in the sense that the feature which distinguishes stressed from unstressed syllables is inherent in the syllabic nucleus and independent of contiguous syllables. As a consequence, we find such sequences as RV 1.1.6 távét tát satyám on the one hand and 10.75.5 imám me gaṅge yamune sarasvati śútudri on the other. In principle, the number of high tones in a phrase may vary between one and the number of syllables, though it is never larger than the number of independent morphemes. The Indian grammarians leave no doubt about the tonal character of the accent (cf. Allen 1953: 87-91). Like its Japanese counterpart, which is to some extent comparable from a typological point of view, the Vedic accentual system can be derived from a system with a two-way level tone distinction on each syllable (cf. McCawley 1978: 301). The attested system may have developed through the loss of high tones following a high tone in the same syntactic unit and subsequent grammaticalization of the resulting tone alternations. 2.2. The accentual system of Greek can be derived from that of Vedic, but not conversely. It is characterized by a number of innovations, among which are the following: 1) Limitation of the place of the stress to the last three syllables of the word, and to the last two syllables if the final syllable contains a long vowel or diphthong. This implies a shift of the stress from the antepenult to the penult before a long ultima. 2) Wheeler’s law: retraction of the stress from a short ultima to a short penult after a long antepenult, e.g. ποικίλος ‘spotted’, βουκόλος ‘cowherd’. These stress shifts have in common that a long syllable attracts the stress not onto itself, but onto the intervening syllable. They are best explained by the assumption that unstressed long vowels or diphthongs received a rising or falling tone movement in the neighbourhood of a high tone vowel, even if there was an intervening syllable, which in that case adopted the high tone of its environment. The assumption of a tonal assimilation which yielded (subphonemic) rising and falling tone movements on unstressed long vowels and diphthongs also provides a basis for understanding two more accentual developments in Attic: 3) Stressed long vowels and diphthongs in the penultimate syllable are falling before short and rising before long final syllables, e.g. gen.sg. σωτῆρος ‘saviour’, Proto-Indo-European tones? gen.pl. σωτήρων. This is understandable if the short ultima was low whereas the long ultima was falling. 4) Vendryes’ law: a short antepenult receives the stress from a long penult before a short ultima, e.g. ἔγωγε ‘I at least’, cf. ἐγώ ‘I’. The stressed vowel had become falling as a result of the previous development and lost the stress to a preceding short syllable, which had apparently become high as a result of tonal assimilation, but not to a preceding long syllable, which had only become rising. The tonal assimilation surmised for Greek was probably posterior to and perhaps evoked by the loss of the PIE laryngeals, which gave rise to a large number of new long vowels. It presupposes the earlier existence of a level tone system. 2.3. It has recently been argued that the accentuation of Baltic and Slavic derivatives can be explained on the basis of inherent accentual properties of the constituent morphemes (cf. Dybo 1968, 1973, Garde 1976). If the separate morphemes are assigned “high” or “low” tone according to their accentual properties, the Balto-Slavic place of the stress can be identified with the leftmost high tone of a word form. The stress patterns can therefore be explained if we assume that late Proto-Indo-European possessed a tonal distinction between “high” and “low” morphemes. 2.4. The reconstructed Indo-European proto-language has a remarkable root structure constraint: a PIE root may not contain a voiced aspirate and a voiceless stop simultaneously, as in **bheut- or **teubh-, unless it is preceded by *s-, as in *steigh- ‘mount’ (e.g. Meillet 1937: 174). If we assume progressive voice assimilation after initial *s-, the distinction between voiceless stops and voiced aspirates was apparently a prosodic feature of the root as a whole. It can therefore be compared with the proposed tone distinction. If this is a meaningful comparison, we expect a correlation between voiceless stops and high tone on the one hand, and between voiced aspirates and low tone on the other. It follows that the incompatibility of voiceless stops and voiced aspirates in a PIE root is a corollary of the incompatibility of high and low tone on the same root. This hypothesis has been investigated by Alexander Lubotsky (1988a). The actual state of affairs is complicated by the influence of the PIE laryngeals on the tonal patterns, as Lubotsky points out. The PIE obstruent harmony has a typological parallel in the Austronesian language Jabêm, where obstruent voicing is actually correlated with a level tone distinction (cf. Bradshaw 1979). 2.5. It has long been recognized that the traditional reconstruction of the PIE obstruent system is typologically deviant in view of the double marking of the voiced aspirates. It must be noted that both voicing and aspiration are often accompanied by a low tone on the following vowel. We must therefore consider 72 Indo-European phonology the possibility that voicing, aspiration and low tone all originated from a single phonological feature, e.g. “lax”. Though it is generally assumed that “consonant types affect tone but tone does not affect consonant types” (Hyman 1975: 229), this does not account for the rise of distinctive tone in syllables which do not contain obstruents. It is therefore probable that the proposed PIE tones were older than the distinction between voiceless stops and voiced aspirates. The latter distinction may have arisen at the same time as zero grade vocalism, which points to a period of strong dynamic stress. It is remarkable that distinctive tone was apparently not eliminated at that stage but survived into the attested languages. The theory advanced here now suggests an explanation for the alternation of voiceless stops and voiced aspirates in root enlargements, where the original distribution may have depended on the accentual properties of the root. 3. It will be clear that the indications of a PIE level tone system listed above require further investigation. The point I want to stress here is their compatibility and mutual reinforcement. Indeed, it seems difficult to account in a principled way for the peculiarities of the Vedic accentual system, the Greek accent shifts, the accentuation of Baltic and Slavic derivatives, the PIE root structure constraint, and the markedness of the voiced aspirates in relation to the voiceless stops without the assumption of a PIE level tone system. The proposed tonal distinction has nothing to do with the traditional difference between acute and circumflex, which arose in the separate languages after the dissolution of Proto-Indo-European. AN INDO-EUROPEAN SUBSTRATUM IN SLAVIC? I Georg Holzer has argued for the prior existence of an unknown Indo-European substratum language from which Baltic and Slavic borrowed a collection of etyma with a distinctive consonant shift (1989). In this hypothetical language, which he calls “Temematic”, (1) the Indo-European voiceless plosives *p, *t, *k became voiced b, d, g whereas (2) the Indo-European voiced aspirates *bh, *dh, *gh became voiceless p, t, k. Since Winter’s law did not operate in this hypothetical language, original *p, *t, *k merged with Indo-European *b, *d, *g, which remained distinct from *bh, *dh, *gh. Moreover, (3) the original syllabic liquids *ṛ, *ḷ yielded ro, lo while (4) long vowels were shortened before resonants. Holzer adduces 45 etyma in his main text and 17 additional examples in an appendix. Since I have little to add to his full discussion of the material, I shall limit myself here to listing the 45 etyma with their new derivations (some of which were already proposed by Machek) and adding the main alternative etymologies (cf. also Brozović 1989, Loma 1990, Moszyński 1992). I shall follow his practice of omitting asterisks before obvious Slavic proto-forms. For the details I refer to Holzer’s book. I. “ACKERBAU” 1. borzda ‘Furche’ < *borgʄ-dā- < *porkʄ-tā-, Latin porca, OHG furuh; better than Gr. φάρος ‘Furche’. 2. proso ‘Hirse’ < *proso- < *bhṛso-, Latin far ‘Dinkel, Spelt’, ON barr ‘Gerste’, OCS brašьno ‘Mehl’; better than *per- ‘schlagen’. 3. bъrъ ‘Hirse’ < *buro- < *pūro-, Gr. πῡρός ‘Weizen’, OE fyrs ‘Quecke’, Lith. pūraĩ, RCS puro; better than *bher- ‘hervorstehen’. 4. zobь ‘Hafer, Futter’ < *gʄobi- < *kʄopi-, Lith. šãpas ‘Halm’, ON hafri; better than zobъ ‘Kropf, Schnabel’. 5. zona ‘taube Körner’ < *gʄonā- < *kʄonā-, Gr. κεν(εϝ)ός ‘leer, eitel’, Arm. sin; better than znobiti ‘frieren lassen’. 6. loboda ‘Gänsefuß, Melde’ < *lo-bodā- < *ḷ-podā-, Latin olor ‘Schwan’, pēs ‘Fuß’; better than *albho- ‘weiß’. 7. smьrdъ ‘Bauer’ < *kʄmir-do- < *gʄhmer-to-, Av. zamar ‘in der Erde’, Vedic jmán, OCS zemlja; better than smьrd- ‘stinken’. 8. sębrъ ‘Bauer’ < *kʄem-ro- < *gʄhēm-ro-, same root; better than ORu. sěmьja ‘Gesinde’, Lith. šeimà. 74 Indo-European phonology 9. těsto ‘Teig’ < *toikʄ-to- < *dhoigʄh-dho-, Gr. τοῖχος ‘Mauer’, Latin fingō ‘forme’, Gothic daigs ‘Teig’, ORu. děža ‘Teigmulde’; better than Arm. t‘rem ‘knete Teig’, OHG theismo, deismo ‘Sauerteig’. II. “VIEHZUCHT” 10. krot- ‘gezahmt’ (in derivatives) < *kroto- < *ghṛdho-, Gothic garda ‘Hürde, Viehhof ’, Lith. garɶdas ‘Pferch’, OCS ograda; better than Gr. κροτέω ‘schlage’, κράτος ‘Starke’. 11. sverěpъ ‘ungezahmt’ < *kʄwer- < *gʄhwēr-, Gr. θήρ ‘wildes Tier’, Latin ferus, Lith. žvėrìs, OCS zvěrь; better than Lith. svarùs ‘schwer’, Latin sērius, sevērus. 12. za-tvoriti ‘schließen’ < *tworo- < *dhworo-, Gr. θύρᾱ ‘Tür’, Latin forēs, Gothic daur, OCS dvьri; better than tvoriti ‘schaffen, bauen’, Lith. tvérti ‘formen, schaffen’. 13. pNjto ‘Fessel, Strick’ < *ponto- < *bhondho-, Vedic bandhá- ‘Binden, Band’, Gr. πεῖσμα ‘Seil, Strick’, Gothic bandi; better than pęti < *penH- ‘spannen’. 14. zvon- ‘*Hund’ (in derivatives) < *gʄwon- < *kʄwon-, Vedic ś(u)ván-, Gr. κύων, Lith. šuõ, Gothic hunds; better than zvonъ ‘Ton, Schall, Glocke’. 15. těxa, těšiti ‘saugen’ < *toi-sā- < *dhoi-(sā-), Vedic dháyati ‘saugt’, Gr. θῆσθαι ‘melken’, OCS doiti ‘saugen’; better than Lith. taisýti ‘bereiten’, tiesà ‘Wahrheit’, tiesùs ‘gerade’, teisùs ‘(ge)recht, aufrichtig’. 16. toliti ‘Durst stillen’ < *tol- < *dhōl-, Vedic dhārú- ‘saugend’, Gr. θῆλυς ‘nährend’, Latin fīlius ‘Sohn’, Latv. dêls; better than OIr. tuilid ‘schlaft’, OHG stilli ‘still’. 17. telę ‘Kalb’ < *tel- < *dhel-, same root; better than Latin tollō ‘hebe empor’, OHG dolēn ‘ertragen’. 18. drěv- ‘alt’ (in derivatives) < *drewo- < *trewo-, Av. θraošti- ‘Reife, Vollendung, Ende’, OHG trowwen ‘pubescere, crescere’; better than Gothic triggws ‘treu’, Lith. drtas ‘stark, kraftig’. 19. bolna ‘Fell, Haut’ < *bol-nā- < *pol-nā-, Gr. acc.pl. πέλλᾱς ‘Häute’, Latin pellis ‘Fell’, OHG fel; better than Gr. φολίς ‘Reptilschuppe’, φελλός ‘Kork’. 20. golěnь ‘Unterschenkel’ < *gol- < *kol-, Gr. κωλήν ‘Hüftknochen’, Latin calx ‘Ferse’, OCS kolěno ‘Knie’; better than golъ ‘nackt’. 21. bedro ‘Oberschenkel’ < *bed-ro- < *ped-ro-, Vedic pad- ‘Fuß’, Gr. πούς, Latin pēs, Gothic fotus; better than Latin femen, femur ‘Oberschenkel’. 22. ědro ‘Busen, Eingeweide’ < *ēdro- < *ētro-, Gr. ἦτρον ‘Bauch, Unterleib’, OE ǃdre ‘Ader’, OIr. inathar ‘Eingeweide’; better than *oid- ‘schwellen’. 23. zNjbrъ ‘Wisent, *hornlos’ < *gʄom-ro- < *kʄom-ro-, Vedic śáma- ‘hornlos’, Gr. κεμάς ‘junger Hirsch’, ON hind ‘Hirschkuh’, Lith. (Žem.) šmùlas ‘hornlos’; better than zNjbъ ‘Zahn’. 24. bedr- (in derivatives) ‘Feder, Flügel’ < *bedro- < *petro-, Vedic pátra-, Latin penna; better than bodNj ‘steche’, bedro ‘Oberschenkel’, bъdrъ ‘wach’. An Indo-European substratum in Slavic? III. “IMKEREI” 25. svepetъ ‘Honigwabe’ < *swep- < *(s)webh-, Gr. ὑφή ‘Weben’, Latin vespa ‘Wespe’, Lith. vapsà, OHG weban; better than Latin sapa ‘Saft’, OHG saf, Vedic vápati ‘streut’, Lith. siti. 26. trNjtъ ‘Drohne’ < *tron- < *dhrōn-, Gr. θρῶναξ ‘Drohne’, OHG treno; better than *ter- ‘reiben, bohren’. IV. “BEVORRATUNG” 27. prokъ ‘Vorrat, Rest’ < *proko- < *bhṛgho-, Gothic bairgan ‘bergen’, Ru. beregú ‘hüte, bewahre’; better than Gr. πρόκα ‘sofort’, Latin procul ‘ferne, weit’. 28. gъrnъ ‘Topf, Kessel’ < *gwir-no- < *kwer-no-, MIr. cern ‘Schüssel’, ON hverna ‘Kochgeschirr’, Vedic carú- ‘Kessel, Topf ’; better than gъrnъ ‘Ofen’. V. “GESELLSCHAFT” 29. svobodь ‘frei’ < *swo-bodi- < *swo-poti-, Vedic svá-pati- ‘sein eigener Herr’, Latin sui potens ‘unabhangig’; better than *swo-bh(w)o-, Gothic sibja ‘Sippe’. 30. slobodь ‘frei, konnend’ < *slo-bodi- < *sḷ-poti-, ON salr ‘Saal, Zimmer’, OCS selo ‘Feld, Acker, Ort’, Lith. salà ‘Dorf ’, Vedic páti- ‘Herr’; better than Gothic silba ‘selbst’, OIr. selb ‘Besitz’. 31. sirъ ‘verwaist’ < *kʄei-ro- < *gʄhei-(ro-), Vedic hīyate ‘wird verlassen’, Gr. χήρᾱ ‘Witwe’, Latin hērēs ‘Erbe’, Gothic gaidw ‘Mangel’; better than Av. saē- ‘verwaist’. 32. trNjtъ ‘Wache, Schar’ < *trunk-to- < *dhrungh-dho-, OIr. drong ‘Schar’, OHG truht, OCS družina ‘Gemeinschaft’; better than German dringen, drängen. 33. gojь ‘Ruhe, Friede’ < *gwojo- < *kwojo-, Av. šāiti- ‘Freude’, Latin quiēs ‘Ruhe’, OCS pokoi; better than gojiti ‘pflegen, mästen, heilen’, žiti ‘leben’, Vedic gáya‘Haus, Hof ’. 34. iz-gojь ‘Freigekaufter’ < *-gwojo- < *-kwojo-, Vedic cáyate ‘racht, straft’, Gr. τίνω ‘büße, bezahle’, OCS cěna ‘Wert, Preis’; better than gojь ‘Friede’, gojiti ‘ernähren’. 35. mьsta ‘Rache, Strafe’ < *mistā- < *misdhā-, Av. mižda- ‘Lohn’, Gr. μισθός, Gothic mizdo, OCS mьzda; better than *mei- ‘wechseln, tauschen’. 36. pъtati ‘achtgeben, fragen’ < *putā- < *bhudhā-, Vedic bódhati ‘wacht, merkt’, Gr. πεύθομαι ‘(er)frage’, OCS bъděti ‘wachen’; better than Latin putāre ‘rechnen, vermuten, meinen’, Slavic pъvati ‘hoffen’. 37. ne-pьtja ‘Vorwand’ < *-pit-jā- < *-bhidh-jā-, Gr. πείθω ‘überrede’, Latin fīdō ‘vertraue, glaube’, Gothic bidjan ‘bitten’, OCS běditi; better than pъvati ‘hoffen’, pъtati ‘fragen’. 76 Indo-European phonology VI. “ZIMMEREI” 38. dolga ‘Brett, Fußboden’ < *dol-gā- < *tol-kā-, Vedic tala- ‘Fläche, Ebene’, German Diele, OCS tьlo ‘Boden’, OPr. talus ‘Fußboden’; better than Gr. δέλτος ‘Schreibtafel’, German Zelge ‘abgeschnittener Zweig’. 39. porkъ ‘Turm, Katapult’ < *porko- < *bhorgho-, Gothic baurgs ‘Stadt, Turm’, OHG burg, Ru. beregú ‘hüte, bewahre’; better than *per- ‘schlagen’ oder ‘fliegen’. 40. tvьrdъ ‘eingeschlossen’ < *twir-do- < *dhwer-to-, Gr. θύρᾱ ‘Tür’, Latin forēs, OCS dvьri; better than tvoriti ‘schaffen’. 41. tъkъ ‘zusammenpassend’ <*tuko- < *dhugho-, Gr. τεύχω ‘mache zurecht’, OE dugan ‘nützen, passen’, Lith. daũg ‘viel’; better than Gr. τύκος ‘Hammer’. VII. “SONSTIGES” 42. gvězda ‘Stern’ < *gʄwoid-dā- < *kʄwoit-tā-, Vedic śveta- ‘weiß’, Gothic ŵeits, OCS cvětъ ‘Blume, Blüte’, světъ ‘Licht, Welt’; better than Gr. φοῖβος ‘leuchtend, strahlend’, φαικός ‘klar, leuchtend’. 43. pojetь ‘singt’ < *paje- < *bhāje-, Latin fōr ‘spreche, sage, besinge’, OE bōian ‘prahlen’, Slavic bajati ‘erzahlen, besprechen’, Gr. φημί ‘sage’; better than Gr. παιάν ‘Heil- und Lobgesang’, Gothic faian ‘tadeln’. 44. ędro ‘Kern, Hode’ < *endro- < *entro-, Vedic antrá- ‘Eingeweide’, Gr. ἔντερα, OCS jętro ‘Leber’; better than Gr. ἁδρός ‘voll, ausgewachsen’. 45. golNjbь ‘Taube’ < *golumbo- < *kolumbo-, Gr. κόλυμβος, Latin columba; better than Lith. guÞbė ‘Schwan’. While some of these etymologies may turn out to be false, it is improbable that all of them will eventually be rejected. We must therefore reckon with the definite possibility of a “Temematic” substratum in Slavic, and perhaps in Baltic. Apart from the interchange of the tenues and the mediae aspiratae as a result of (1) and (2) discussed above and the absence of Winter’s law, Holzer has proposed the following developments for his hypothetical substratum language (1989: 13): (3) The syllabic liquids *ṛ, *ḷ became ro, lo. There are only five examples in Holzer’s 45 etyma: 2. 6. 10. 27. 30. proso < *bhṛso-, Latin far, loboda < *ḷ-podā-, Latin olor; krot- < *ghṛdho-, Gothic garda; prokъ < *bhṛgho-, Gothic bairgan; slobodь < *sḷ-poti-, ON salr. None of these instances has an obvious zero grade. The assumption of an original zero grade in the compounds loboda and sloboda is no more than a theoretical construct. There is a zero grade in Vedic gṛhá- ‘Haus’, which is clearly An Indo-European substratum in Slavic? cognate with Gothic gards ‘Haus’, but the absence of zero grade in the immediate neighbourhood of the Slavic territory renders the idea rather arbitrary. The zero grade in Lith. birɶginti ‘sparen’ can easily be secondary. I therefore think that the alleged development is no more than a possibility. (4) Long vowels were shortened before resonants. In fact, there is only a single root where an original long vowel was NOT shortened: 22. ědro < *ētro-, Gr. ἦτρον. There is only one other etymon with an initial vowel, and this is an instance where the Slavic word has not preserved a vowel length contrast: 44. ędro < *entro-, Gr. ἔντερα. Thus, we may just as well say that there is no trace of the original distinction between long and short vowels in the material. (5) Raising of *e to *i before tautosyllabic r. There are three examples in Holzer’s 45 etyma: 7. smьrdъ < *gʄhmer-to-, Av. zamar; 28. gъrnъ < *kwer-no-, ON hverna; 40. tvьrdъ < *dhwer-to-, Gr. θύρᾱ. Since the number of examples is very small and raising of e to i before r is not a natural development, it is preferable to assume a morphological reduced grade in these derivatives. Note that the other derivatives usually have o- or zero grade (cf. Holzer 1989: 170-174). (6) Diphthongs became acute before single consonant plus vowel. This rule is reminiscent of the one that assigns acute tone to Slavic borrowings from Germanic (cf. K014: 70). It suggests that the “Temematic” language may have had an expiratory accent on the initial syllable. In his tentative identification of the substratum with the language of the Cimmerians, Holzer proposes two additional developments (1989: 179): (7) Unrounding of *o, *ō to *a, *ā, and (8) Raising of *e to *i before tautosyllabic nasals. These hypothetical changes fit the earliest Slavic developments after the end of the Balto-Slavic period very nicely (cf. K036: 264f. and K066: 46f.). We may therefore wonder if the “Temematic” substratum provoked the earliest developments of Slavic as a separate language, before the Scythian expansions took place. 78 Indo-European phonology II The next question is: what was the position of the proposed substratum language between the surrounding Indo-European dialects? We find limited voicing of the tenues in Germanic (Verner’s law) and total devoicing of the aspirates in Greek. These are precisely the two branches of Indo-European which share most lexical isoglosses with the newly proposed language (cf. Holzer 1989: 165). It is therefore reasonable to locate the new language between Germanic in the north and Greek in the south. The unaspirated voiced (or glottalic) plosives were devoiced in Phrygian (cf. Lubotsky 1998: 42022), Armenian, and Thracian, but not in Greek, Albanian, and Dacian. As Phrygian was close to Greek (cf. Lubotsky 1988b) and Thraco-Armenian to DacoAlbanian (cf. K101), it is probable that the new language was originally spoken to the west of the dialect continuum which connected Slavic with Albanian and Armenian. (A possible borrowing into Albanian is mjet ‘Mittel, Trennwand’ < *meti- < *medhi-, cf. Demiraj 1997: 274f.) This places the speakers of “Temematic” in the western Ukraine, which is precisely the most probable homeland of the Slavs, at the time when the ancestors of the Greeks moved into the Balkan peninsula. We must therefore consider the possibility that their language, like Anatolian and perhaps Greek, had not yet developed a distinction of voicedness in the plosives at that time. In my earliest discussion of the PIE system of obstruents (K032), I reconstructed plain fortes *T: (traditional *t etc.), glottalic lenes *T’ (traditional *d etc.), and aspirated lenes *T‘ (traditional *dh etc.), and assumed that glottalization was lost at an early stage in Indic, Greek and Italic while aspiration was lost at an early stage in Balto-Slavic, Armenian and Germanic, yielding a complementary series of voiced lenes in opposition to the original voiceless fortes. Both glottalization and aspiration were lost yielding voiced lenes in Iranian, Albanian and Celtic, which occupied an intermediate position between the dialects just mentioned, while the original absence of voiced obstruents was preserved in Anatolian and Tocharian. I have now changed my view in two respects: (1) It has turned out that glottalization was preserved not only in BaltoSlavic and Armenian, but also in Indo-Iranian, Greek, Italic and Germanic (cf. K075: 192-197). It is therefore possible that the absence of distinctively voiced lenes was maintained much longer in Greek and Italic, which have voiceless reflexes of the aspirates. This opens the possibility of reconstructing a GraecoPhrygian proto-language with voiceless reflexes of both the glottalics (as in Phrygian) and the aspirates (as in Greek). Thus, we arrive at an outer ring of Indo-European dialects where the original system of obstruents was preserved, comprising Tocharian in the East, Anatolian and Graeco-Phrygian in the south, and Italic in the west, and a central area where voicedness arose as a distinctive An Indo-European substratum in Slavic? feature in the aspirates, containing the satem languages and their western neighbours. (2) The evidence for aspiration in the aspirates is limited to Indic and Greek, which have very different types of aspiration and do not even share it with Iranian and Phrygian. The Italic fricatives can easily have developed from a lenition of voiceless plosives without an intermediate stage of aspirated stops. The Greek aspiration may be the result of delayed voice onset after the voiceless lenes. It now appears that Indo-Iranian shared the rise of distinctive voicedness with the central Indo-European dialects and that the rise of voiced aspirates in Indic was a secondary development, as it was in the central dialects of Armenian (cf. K075: 188-191). It follows that we must reconstruct fortes *T:, glottalics *T’, and lenes *T for Proto-Indo-European, with neither voicedness nor aspiration as distinctive features. The phonetic distinction between fortes *T: and lenes *T was probably a matter of consonantal length, as is essentially the case in modern Germanic languages, where voicedness and aspiration are concomitant features (cf. Goblirsch 1994: 11 and passim). We may now reconsider the position of “Temematic” among its IndoEuropean neighbours. The absence of voicedness in the lenes (“aspirates”) suggests that the language was originally close to Graeco-Phrygian. The subsequent loss of glottalization and rise of voiced obstruents was a development shared by Daco-Albanian. The distinctive merger of the fortes with the new voiced obstruents resulted from a weakening process which can be dated before this development because the merger of fortes with glottalics and subsequent rise of voicedness is a more natural chain of events than the unmotivated rise of voicedness in the fortes at a stage when the glottalics had become voiced and the lenes were voiceless. In the latter chronology, one would rather expect both fortes and lenes to remain voiceless, as in Greek, or the lenes to become voiced, as in Daco-Albanian. The weakening of the fortes has a counterpart in Germanic, where the details are quite different because we find voicing after unstressed vowels (Verner’s law) and frication elsewhere (Grimm’s law) and rise of new fortes from the original glottalics. These developments were probably more recent. Farther to the west we find lenition of the original fortes yielding voicing in British and frication in Irish. Though it is difficult to see a historical connection here, it is remarkable that there is no trace of lenition in Balto-Slavic languages. One therefore wonders if the various kinds of lenition were induced by a non-Indo-European substratum language spoken on the western border of the Indo-European homeland when successive waves of migration (Italic, Celtic, Germanic, “Temematic”) passed through their territory. In any case, this hypothesis is compatible with the idea that after the migration of the Graeco-Phrygians into the Balkan peninsula, the speakers of “Temematic” moved from the southwestern part of the Indo-European homeland into the territory which was abandoned by the ancestors of Germanic 80 Indo-European phonology speakers when these moved westwards. Though it is difficult to prove the former existence of a “Temematic” language in any strict sense of the word, it is important to note that it fits into our picture of the original Indo-European dialects very nicely. 1ST SG. MIDDLE *-H2 1. The athematic secondary indicative 1st sg. middle ending is -i in IndoIranian, e.g. Vedic náṃśi, Gathic aojī, Old Persian adaršiy. I think that this ending continues PIE *-H2 (cf. 1979a: 67). The 1st sg. middle forms akri (RV 10.159.4 = 10.174.4) < *e-kwrḤ2 and ajani (RV 8.6.10) < *e-gʄṇH1Ḥ2 have nothing to do with the 3rd sg. passive aorists akāri < *e-kwori and ajani < *e-gʄonH1i. The latter are probably uninflected neuter i-stems which were incorporated into the verbal system. 2. In spite of the fact that the corresponding thematic ending is *-ai in IndoIranian, Meillet suggested already that the 1st sg. middle ending -i may be an alternant of the long vowel in the Greek secondary ending -μᾱν (1964: 234). If this is correct, the Indo-Iranian thematic ending must be due to analogical reshaping. 3. Conversely, Petersen assumed that the athematic secondary ending -i was due to analogy on the basis of the 1st pl. forms: impf. ¡si : pres. ¡se = impf. ¡smahi : pres. ¡smahe (1936: 162). This is an unnatural type of analogic change. The 1st sg. form is generally reshaped on the basis of other sg. forms, not on the basis of the 1st pl. form. Besides, it remains unclear why the proposed analogy affected neither the optative ending -a nor the thematic endings, where both the model and the motivation for analogic change should be the same. Moreover, the 1st pl. ending -mahe is of analogical origin itself. A subsequent analogic development would be expected to replace the secondary endings -i and -mahi with **-a and **-maha on the basis of 3rd sg. -(t)a, primary -(t)e, cf. also the subjunctive 1st pl. ending -mahai on the basis of 1st sg. -ai. 4. Following Meillet, Ruipérez assumes that the 1st sg. ending -i continues PIE *-ǩ and concludes that the Indo-Iranian thematic ending *-ai is analogical (1952: 23). In his opinion, this *ǩ yielded *a before and after nonsyllabic *i in the primary ending *-ai and the optative ending *-īya. The latter development can no longer be maintained, especially since Hoffmann’s discussion of the athematic optative (1968: 5). A vocalized laryngeal never yielded anything different from i in Indo-Iranian. We must therefore assume that both the primary ending *-ai and the optative ending -a are of analogical origin. 5. Kuryłowicz shares Petersen’s view that the ending -i is analogical (1964: 59). He states that it replaced earlier *-a < *-H2o without specifying the motivation for the replacement or giving evidence for the reconstruction of the earlier ending. Conversely, earlier -i could easily have been replaced by **-a in Indo- 82 Indo-European morphosyntax Iranian on the basis of 3rd sg. -(t)a. Ruipérez pointed out already that the Hittite 1st sg. ending -ha may have taken its vowel from the other persons (1952: 24). 6. Cowgill has given a detailed account of his views on the 1st sg. middle ending in Indo-Iranian (1968). Like Kuryłowicz, he starts from a PIE ending *-H2o. He explicitly rejects Ruipérez’s suggestion that Hittite -ha represents PIE *-H2 plus analogical -a from the other persons because in that case *H2 “would be the only Indo-European person marker that functioned equally in both voices, without the need of a specific voice marker in the mediopassive” (1968: 26). I conclude that we have to separate the laryngeal of the perfect and the middle from the one in the thematic present ending *-oH and that it cannot be regarded as a person marker. Elsewhere I have identified the laryngeal in the 1st sg. endings of the perfect and the middle with the one in 2nd sg. *-tH2e/o and 1st pl. *-medhH2 and the laryngeal of the thematic 1st sg. ending *-oH with the one in 2nd sg. *-eH1i (acute tone in Lithuanian) and 2nd pl. *-etH1e (aspiration in Indo-Iranian); if there ever was a laryngeal in the other 1st and 2nd person endings of these paradigms *-m(H2)e, *-(H2)e, *-dh(H2)ue, *-om(H1)om, it was lost phonetically in the available material (1979a: 68). 7. The reconstruction *-H2o brings Cowgill into major difficulties because *H2 appears not to have affected the timbre of a neighbouring *o, so that the Greek and Tocharian 1st sg. middle endings -μαι, -μᾱν, A -mār, B -mar, -mai cannot be derived from *-H2o-. The postulate of an ablauting voice marker *e is unmotivated and offers no explanation, as Cowgill points out. His suggestion that *-o- was analogically replaced with *-a- lacks a motivation if *o was characteristic of the middle: the converse development would be expected. In fact, there is no evidence at all for o-vocalism in the 1st sg. middle ending of the proto-language. 8. Cowgill sees the motivation for the introduction of -i in the athematic paradigm in the confusion of 1st sg. *-H2o and 3rd sg. *-o after the loss of postconsonantal laryngeals. It is unclear why the same confusion continued to be tolerated in the primary ending and in the perfect, where 1st sg. *-H2e and 3rd sg. *-e merged at the same stage. In the perfect, the homonymy was even extended to the middle endings and, in classical Sanskrit, to those cases where the 1st and 3rd sg. forms had not merged phonetically as a result of Brugmann’s law. 9. The reconstruction *-H2o forces Cowgill to assume a substantial amount of remodelling in cases where I see phonetically regular forms (1968: 30f.): “akri (X 159,4 = 174,4), probably not to be read akuri, must be analogic, while āvṛṇi (X 33,4) can be explained as regular only by reading āvṛṇī < *āvṛniyi ← *āvṛniya < *ēwḷnḤAo”, where I reconstruct *e-kwrḤ2 and *e-uʢḷnH1Ḥ2. Also, “huvé, hinvé, vṛṇe, tasthe, etc. ought to be *hūve, *hinuve, *vṛṇiye, *tasthiye if they were 1st sg. middle *-H2 83 faithful continuations of *gʄhuHAoy, *gʄhinuAoy, *wḷnḤAoy, *stestẠAoy”, whereas in my opinion these forms reflect disyllabic *gʄhuH1Ḥ2, *gʄhinuH2, *uʢḷnH1Ḥ2, *stestH2Ḥ2 with primary -e for secondary -i < *-Ḥ2. “Similarly Gatha Avestan has dadē (Y. 28,4) and vǩrǩnē (Y. 46,3)”, where I read /dadai/ and /vṛnai/ with primary *-ai for secondary *-i in disyllabic *dhedhH1Ḥ2 and *uʢḷnH1Ḥ2. Thus, I disagree with Cowgill’s view that “it is not likely that the difference between 1st sg. tatane RV. VII 29,3 and 3rd sg. tatne RV. X 130,2 is due to the laryngeal originally present in the ending of the former”: these forms reflect PIE *tetonH2e and *tetone with substitution of the Indo-Iranian primary middle ending *-ai for *-a and introduction of zero grade into the root. 10. In his article on the origin of the Sanskrit passive aorist, Insler adopts Petersen’s view that the 1st sg. middle ending -i was created on the analogy of the 1st pl. ending (1968: 323). He regards this 1st sg. middle ending as the only possible origin of the 3rd sg. passive ending -i and reconstructs a complicated chain of analogic developments to account for the different root vocalism and accentuation. Here I shall not enter upon a discussion of his theoretical constructs, which are not strictly relevant to the subject of the present study, but I want to draw attention to the distribution of the 3rd person middle endings which Insler establishes and which can be summarized in Table I (cf. 1968: 327). Table I: Vedic endings 3rd sg. 3rd pl. deponent roots: present imperfect aorist -e, -te -a(t), -ta -a(t), -ta, -i -re, -ate -ra(n), -ata -ra(n), -ata transitive roots: transitive present transitive imperfect/aorist passive present passive aorist -te -ta -e -i -ate -ata -re -ra(n) It must be noted that the 3rd pl. passive ending -ran is limited to three roots: adṛśran ‘have been seen’, ayujran ‘have been yoked’, asṛgran ‘have been discharged’. The endings -a and -ra were apparently confined to deponent roots at an earlier stage. In Avestan, the form in -i is limited to the passive of transitive roots: vācī ‘was spoken’, srāvī ‘was heard/tried’, jaini ‘was slain’, ǩrǩnāvī ‘was allotted’ (Insler 1968: 320). This was probably the earlier distribution. In the root aorist of the Sanskrit deponents we find 3rd sg. -ta after a root ending in a short vocalic element versus -i, replacing earlier -a, after a root ending in a consonant (including laryngeal), e.g. amṛta, apādi, ajani. The older ending was preserved 84 Indo-European morphosyntax in the form -at in ādat ‘took’, akhyat ‘looked’, and ahuvat ‘called’. Unlike Insler, I think that this formal distribution is secondary and that we must assume a semantic opposition for the proto-language: PIE 3rd sg. *-o and 3rd pl. *-ro in deponents versus 3rd sg. *-to and 3rd pl. *-ntro in transitive middles. 11. Watkins’ view is the exact opposite of Insler’s: he assumes that the 3rd sg. passive aorist ending -i replaced earlier *-Ho in the 1st sg. middle aorist and imperfect, the older ending being preserved in the optative (1969: 138f.). Here again, both the model and the motivation for the analogic change lack sufficient justification. The 1st sg. middle and 3rd sg. passive forms did not belong to the same paradigm and differed in ablaut and accentuation. It remains unclear how and why the 1st sg. ending -i appeared in the imperfect and the transitive middle aorist, where the 3rd sg. ending was -a or -ta. In order to account for the ablaut difference Watkins changes avri /avuri/ into **avari (RV 4.55.5) and assumes that the zero grade in RV ayuji (5.46.1), akri (2×), avṛṇi is analogical. Thus, the only example in the Rgveda which can have served as a model is ajani (8.6.10), which is ambiguous for the determination of the original ablaut grade. 12. In his excellent article on the Vedic passive aorist in -i, Migron shows that this form is impersonal in the sense that it serves any person and number without generally specifying the agent, “not because the agent is unknown, but because it is either unimportant or too well-known to require mention” (1975: 299). He demonstrates that it really is a passive perfect, “i.e. that its aspectual rôle is to focus the hearer’s attention on the moment at (or since) which the ‘Einwirkung’ has been accomplished, has become a fact of some consequence to him”, e. g. víśvaṃ jīváṃ támaso nír amoci ‘Every living thing has been released from darkness’ (RV 10.107.1). It follows from his observations that the connection between the “passive aorist” and the causative and the ya-passive is even closer than was hitherto assumed. I think that the latter were simply derivatives from a deverbative noun of the type *kwori, which could itself be used predicatively in the sense of a passive perfect, e.g. ‘This is a construction’ = ‘This has been constructed’, cf. English revolutionize as a factitive of revolution. There are remnants of this type in other languages, e.g. Slavic bolь ‘sick’, navь ‘dead’, factitive naviti (cf. Vaillant 1974: 23), and Gothic muns ‘thought’ (with zero grade taken from the verb munan). I think that it is also the origin of the Germanic weak preterit, e.g. Gothic 2nd sg. nasides ‘(you) saved’ < *nosi dhēs, cf. also Old Irish -suidigedar < *sodi sagītro. This type of nouns must not be confused with those i-stems which have lengthened grade in the root and continue suffixless deverbative nouns in Slavic, e.g. rěčь ‘speech’ (Vedic v¡k), tvarь ‘creation’ (cf. Vaillant 1974: 28). 13. Elsewhere I have presented my reconstruction of the PIE verbal endings, a part of which is reproduced in Table II (cf. 1979a: 67). 1st sg. middle *-H2 85 Table II: Proto-Indo-European endings 1st sg. 2nd sg. 3rd sg. 3rd pl. secondary active -m -s -t -nt transitive middle -mH2 -stHo -to -ntro intransitive middle -H2 -tHo -o -ro The 1st sg. transitive middle ending *-mH2 yielded *-ā after a consonant and *-mi after a vowel in Indo-Iranian. Thus, it was eliminated because it merged with the 1st sg. active ending of the thematic and athematic present, respectively. The corresponding intransitive middle ending survived as -i after a consonant. This ending was extended to the thematic flexion (subjunctive), where the 1st sg. active and intransitive middle endings had merged phonetically into *-ā. The 2nd sg. middle endings *-stHo and *-tHo merged after an obstruent. They also merged with the 3rd sg. transitive middle ending *-to if that obstruent was an aspirate. This was the motivation for remodelling these endings on the basis of the corresponding active forms. The 3rd pl. transitive middle ending *-ntro lost its *r on the analogy of the active endings. 14. In the optative, which had intransitive middle endings, the 1st sg. ending yielded Indo-Iranian *-aiyi < *-aiHi < *-oâH1Ḥ2 in the thematic flexion and *-ī < *-īH < *-iH1H2 in the athematic flexion, cf. Vedic nom.acc.du. devã < *daivīH < *deiuʢiH2H1, which cannot be analogical. The 3rd sg. ending yielded *-aiya and *-iya, respectively. The latter form was regularized to *-īya and substituted for the seemingly endingless 1st sg. form. Finally, the thematic 1st sg. ending was replaced with *-aiya. 15. When the 3rd person endings *-a and *-ra received additional clarity by the addition of the secondary active endings -t and -n, this created a problem in the optative, where the 3rd pl. active ending was *-iHat < *-iH1nt. The resulting ambiguity led to the substitution of 3rd sg. middle -ta for *-a and the creation of a 3rd pl. active ending *-iHṛ on the basis of the corresponding middle form. The ending *-ṛ for earlier *-at < *-nt also spread to the athematic aorist and the reduplicated imperfect. 16. The primary middle endings were created in Indo-Iranian on the basis of the secondary endings: *-ai : -a = *-tai : -ta = -ti : -t. The 2nd sg. ending *-sai was apparently modelled after the primary 3rd sg. ending *-tai, not after the secondary 2nd sg. ending *-(s)tha. The thematic secondary 1st sg. ending *-ai came to be used as a primary ending both in the thematic and in the athematic flexion. It was replaced with *-āi in the subjunctive in order to differentiate this mood from the indicative. I think that the Vedic 3rd sg. and pl. subjunctive 86 Indo-European morphosyntax endings -ate and -anta replace earlier *-a and *-ara, respectively. On the coexistence of primary and secondary endings in the Indo-Iranian subjunctive paradigm see Beekes 1981. 17. In Greek, the 1st sg. transitive middle ending *-mH2 yielded *-mā after a consonant and *-ma after a vowel. The first variant was generalized and received an additional *-m for the sake of clarity. The corresponding intransitive ending *-H2 yielded *-a after a consonant and lengthening after the thematic vowel. It was eliminated because it merged phonetically with the perfect and thematic present endings, respectively. The 2nd sg. intransitive middle ending *-tHo merged with the 3rd sg. transitive middle ending *-to, and after dental obstruents also with the 2nd sg. transitive middle ending *-stHo. This was the motivation for the replacement of the 2nd sg. endings with *-so, which was created on the analogy of the active endings. The 3rd person intransitive middle endings were lost, and the *r in the 3rd pl. transitive middle ending *-ntro was eliminated on the basis of the active endings. 18. As in Indo-Iranian, the primary middle endings were created on the basis of the secondary endings in Greek: -μαι : *-μα = -σοι : -σο = -τοι : -το = -τι : *-τ, etc. This development cannot have been a shared innovation of Greek and IndoIranian because it depends crucially on the different vocalization of the syllabic resonants. Thus, the difference between the presence of a nasal in Greek -μαι and its absence in the corresponding Sanskrit ending -e reflects ultimately the different vocalization of the syllabic nasal before *H in these languages, just as the difference between Greek -α- and Sanskrit -i reflects the different vocalization of the syllabic laryngeal. Similarly, the difference between the Sanskrit 2nd sg. ending -thās on the one hand and Iranian *-sa and Greek -σο on the other reflects the different development of PIE *tt, which yielded -tt- in Sanskrit and -st- in Iranian and Greek. The s in Sanskrit -se betrays the more recent origin of this ending. The dialectal Greek rise of -σαι and -ται on the basis of 1st sg. -μαι can be compared with the generalization of -ai in the Vedic subjunctive. 19. In Italic and Celtic, the 3rd pl. transitive middle ending *-ntro lost its *r on the analogy of the active endings. The 2nd sg. transitive middle ending was remodelled in the same way: *-so : -s = *-to : -t. On the other hand, *nt was introduced as a 3rd pl. marker into the intransitive middle ending, which became *-ntro. At this stage, the final *-ro of this ending was reinterpreted as a voice marker and spread to the other intransitive middle endings: 1st sg. *-ōro (thematic ending), 2nd sg. *-toro, 3rd sg. *-oro. Analogy created another 3rd sg. form: *-tro : *-to = *-ntro : *-nto. The addition of *-ro to the 3rd sg. and pl. transitive middle endings yielded passive forms of transitive verbs in *-toro and *-ntoro. Thus, we arrive at the verbal system which is presented in Table III. 1st sg. middle *-H2 87 Table III: Italo-Celtic endings 1st sg. 2nd sg. 3rd sg. 3rd pl. secondary active -m -s -t -nt transitive middle -ma -so -to -nto passive -toro -ntoro intransitive middle -a, -ōro -to, -toro -o, -oro, -tro -ntro The 2nd sg. transitive middle ending has been preserved in Latin, the 3rd sg. ending in Venetic doto, donasto ‘gave’ and in the Old Irish imperfect and imperative endings, and the 3rd pl. ending also in the latter paradigms (cf. K239: 45-49). The passive and intransitive middle (deponent) endings have all been preserved in Old Irish with the exception of 1st sg. *-a and 3rd sg. *-o, which was replaced with *-to in the deponent imperative. The final *-o of 1st sg. *-ōro and 2nd sg. *-toro explains the absence of palatalization in the absolute forms (cf. K035: 49). The palatalization in -ther originated from the syncope of a preceding front vowel. The passive endings are also found in Latin. From the intransitive middle paradigm, the 1st sg. ending *-ōro is preserved in Latin -ōr and the 3rd person endings are attested in Umbrian ferar ‘feratur’, Oscan sakarater ‘sacratur’, Marrucinian ferenter ‘feruntur’ (cf. K239: 156f.). It is beyond doubt that the r spread from the 3rd pl. ending because it is absent from the Latin 2nd sg. and pl. forms and from the entire deponent imperative paradigm in Old Irish with the exception of the 3rd pl. form (cf. already Pedersen 1938: 105). 20. The only attested 1st sg. middle form in Germanic is Old Norse heite ‘am called’, which belongs to the same paradigm as Old English 3rd sg. hātte and Gothic 2nd sg. haitaza, 3rd sg. haitada, 3rd pl. haitanda. These forms point to a Proto-Germanic middle present set of endings 1st sg. *-ai, 2nd sg. *-asai, 3rd sg. *-adai, 3rd pl. *-andai, which was apparently created on the basis of a paradigm *-a, *-asa, *-ada, *-anda by the addition of *-i from the athematic primary active endings. The vocalism of the 3rd sg. ending betrays the PIE intransitive middle ending *-o, which was apparently extended with *-to. The new ending *-oto served as a model for the creation of 2nd sg. *-oso and 3rd pl. *-onto, and for the optative endings *-oiso, *-oito, *-ointo, Gothic -aizau, -aidau, -aindau, and the imperative 3rd person endings *-otōu and *-ontōu, Gothic -adau, -andau, which can be compared with Vedic -tām, -ntām, Greek -τω, -ντω, all from the PIE transitive middle endings with lengthening of the final vowel before the added particle *u. Latin -tō is apparently a merger of 2nd sg. active *-tōd (Vedic -tāt) and 3rd sg. transitive middle *-tō. Greek created a new set of 3rd person middle imperatives on the basis of the 2nd pl. ending (cf. Chantraine 1967: 271f.). The fact that the Germanic 1st sg. form was not remodelled in the same way suggests that it did not end in *-o at that stage. Thus, it offers additional support for the 88 Indo-European morphosyntax reconstruction *-H2. If the latter yielded *-a, it follows that *a and *o had not yet merged at the stage under consideration. 21. The Tocharian primary endings 1st sg. A -mār, B -mar, 2nd sg. A -tār, B -tar, 3rd sg. AB -tär, 3rd pl. AB -ntär suggest a Proto-Tocharian paradigm *-mar, *-tar, *-tṛ, *-ntṛ. The secondary endings 1st sg. A -e, B -mai, 2nd sg. A -te, B -tai, 3rd sg. A -t, B -te, 3rd pl. A -nt, B -nte point to Proto-Tocharian *-ai, *-tai, *-to, *-nto. The terms “primary” and “secondary” have given rise to misunderstanding among Indo-Europeanists. The Tocharian primary active endings (which are found in the present, subjunctive, optative, and B imperfect) represent the PIE primary and secondary active endings: 1st sg. AB -m < *-mi, B -u < *-ō, 2nd sg. AB -t < *tu (cf. Pedersen 1944: 5), which was added to the zero ending from PIE *-(e)s, 3rd sg. A -ṣ and B -ṃ are pronominal clitics which were added to the zero ending (which has been preserved in the B imperfect and optative) from PIE primary *-e and secondary *-(e)t, 3rd pl. A -ñc < *-nti, B -ṃ < *-nt. The Tocharian secondary active endings (which are found in the preterit and usually in the A imperfect) correspond to the PIE perfect endings: 1st sg. A -ā, -wā, B -wa replace *-a < *-H2e, 2nd sg. A -ṣt, B -sta replace *-ta < *-tH2e, 3rd sg. AB zero from PIE *-e, 3rd pl. AB -r = PIE *-(ē)r, B -re reflects *-ro. Thus, the endings were originally differentiated according to the voice (active or perfect) of the verb form. In the same way, the primary and secondary middle endings continue the PIE transitive and intransitive middle endings, respectively. 22. Unlike Italic and Celtic, Tocharian extended final *-ro to the 3rd sg. ending *-to of the transitive middle paradigm: *-tro : *-t = *-ntro : *-nt. The 3rd person intransitive middle endings *-o and *-ro were subsequently replaced with *-to and *-nto after the model of the active and transitive middle endings. The 2nd sg. ending *-(s)to, which lost its *s after an obstruent (cf. in this connection Melchert 1977), adopted the vowel of the 1st sg. endings *-ma < *-mH2 and *-a < *-H2. Tocharian appears to have developed real primary and secondary endings in the intransitive middle (mediopassive) paradigm: *-toi : *-to = *-ti : *-t, etc. Thus, we arrive at the system of verbal endings which is presented in Table IV. Table IV: Proto-Tocharian endings 1st sg. 2nd sg. 3rd sg. 3rd pl. secondary active -m -s -t -nt transitive middle -ma -ta -tro -ntro secondary mediopassive -a -ta -to -nto primary mediopassive -ai -tai -toi -ntoi At this stage, the perfect and mediopassive endings spread to the active and transitive middle aorist, respectively. The merger of the perfect with the aorist 1st sg. middle *-H2 89 yielded a preterit paradigm with perfect endings, as in Latin. The middle preterit received the secondary 3rd person mediopassive endings *-to and *-nto, but the primary 1st and 2nd sg. mediopassive endings *-ai and *-tai in order to avoid homonymy with the active preterit (perfect) endings *-a < *-H2e and *-ta < *-tH2e. This distribution is also reminiscent of Latin. I assume that the 1st sg. perfect ending *-ai in Latin served for differentiation from the deponent ending *-a, which was subsequently replaced with -ōr. Similarly, the Slavic ending -ě < *-ai in vědě ‘I know’ indicates the previous existence of a middle ending *-o < *-H2o (with added *-o from the other persons). The introduction of the final *-o from *-ro into the 3rd pl. perfect ending *-ēr in Latin and Tocharian was of course anterior to the introduction of *nt into the 3rd pl. intransitive middle ending. It did not affect the corresponding zero grade ending *-ṛ, which is preserved in the Tocharian s-preterit, because the perfect and intransitive middle endings would otherwise have become homonymous. The 3rd person transitive middle endings *-tro and *-ntro lost their *-o, which became a marker of the preterit, and their *r was introduced into the 1st and 2nd sg. endings of the paradigm. Finally, the mediopassive adopted the transitive middle flexion. The new tense markers *-o and *-r were extended to the 1st pl. ending *-meta < PIE *-medhH2, which was replaced with *-meto (A -mät, B -mte) in the preterit and with *-metr (AB -mtär) in the present system. I think that the 2nd pl. ending of the middle preterit, which is A -c and B -t, is the phonetic reflex of PIE *-dhue, cf. Α sparcwatär, B sporttotär ‘turns’ from the Proto-Tocharian active stem *spartwe- with the thematic 3rd sg. middle ending *-otṛ. The final -r of the corresponding present ending A -cär, B -tär was apparently added in the separate Tocharian dialects, cf. also the active ending B -cer next to A -c < *-te. The expected *-tt < *-twe in the B middle ending seems to have been extended to the 1st pl. endings -mtte, -mttär (but cf. Peyrot 2008: 155-157). The 1st pl. active ending A -mäs, B -m reflects PIE *-me(s) with a pronominal clitic in the former dialect (cf. Pedersen 1941: 143). The 2nd pl. ending AB -s of the active preterit represents a clitic which was taken from the imperative and added to the zero ending that corresponds to the Sanskrit perfect ending -a. 23. In Anatolian, the spread of *nt in 3rd pl. middle forms led to the coexistence of the endings *-nto, *-ntro, and *-ntoro. The introduction of final *-ro into the 3rd sg. forms in *-o and *-to gave rise to the endings *-oro and *-toro. The 1st sg. middle ending *-H2 received a final *-o from the other members of the paradigm. Eventually *-ro became an optional clitic in all persons and was remodelled to -ri in the present and -ru in the imperative on the basis of the corresponding active forms. The vocalization of the 1st sg. middle ending *-H2 after a consonant yielded *-a, to which the regular ending *-ho could be added. Restoration of the laryngeal yielded an ending *-ha, which could be extended to 90 Indo-European morphosyntax *-haho. Thus, the assumption that the PIE 1st sg. middle ending was *-H2 rather than *-Η2ο explains the origin of the surprising Hittite ending -hahari. 24. I conclude that the Indo-Iranian 1st sg. middle ending -i is the phonetic reflex of PIE *-H2 and that the correctness of this reconstruction is supported by evidence from Greek, Germanic, Tocharian and Hittite. The primary middle endings originated as a result of parallel developments in the separate languages, as is clear from the formal and functional incongruities. The motivation for these parallel developments lies in the absence of a distinction between primary and secondary endings in deponent paradigms, the presence of a clear present tense marker in the athematic active flexion, and the asymmetrical status of *r in the PIE 3rd pl. endings. Since the early substitution of *-nto for the transitive middle ending *-ntro is common to Indo-Iranian, Greek, Italic, Celtic and Germanic, it may be a dialectal Indo-European development which was not shared by Tocharian and Anatolian. On the other hand, the spread of *r in the transitive middle paradigm and its elimination from the intransitive middle paradigm in Proto-Tocharian is also found in Armenian. Thus, I think that Armenian, like Irish, preserves the transitive middle flexion in the imperfect and the middle imperative and the mediopassive flexion in the middle aorist: 2nd sg. -r < *-ro, 3rd sg. -(w)r < *-tro, 2nd pl. -ruk‘ < *-ro-, aorist 3rd sg. -w < *-to, 3rd pl. -n (without loss of the preceding vowel) < *-nto, 2nd pl. -ǐik‘ < *-dhue- (cf. K194: 34-38). The active and mediopassive endings merged in the present tense as a result of the apocope. The prohibitive imperative in -r < *ra belongs to the present system and cannot be connected with the middle aorist imperative ending -r. Phrygian seems to have shared the Armenian development: αδδακετορ = αδδακετ ‘afficit’. PROTO-INDO-EUROPEAN VERBAL SYNTAX I In 1901 C.C. Uhlenbeck concluded from the identity of the nominative and the accusative of the neuter in the Indo-European languages that the differentiation of these cases is secondary. For an early period of the proto-language he assumes the existence of an agentive case in *-s, which expressed the subject of transitive verbs, and a general case in *-m (after o-stems) or zero (in other flexion classes), which expressed the object of transitive verbs and the subject of passive and intransitive verbs. The sigmatic nominative developed from the original agentive case, while the accusative in *-m and the asigmatic nominative continue the general case. Uhlenbeck follows Bopp in the identification of the ending *-s with the PIE demonstrative pronoun *so. A few years later Holger Pedersen presented a much more elaborate view of PIE verbal syntax (1907: 148-157). His exposition seems to have fallen into total oblivion. In the handbook on ergativity (1979), K.H. Schmidt does not even mention Pedersen’s article, which is for several reasons one of the most remarkable publications in the history of linguistics. As Pedersen’s view has not lost any of its significance since it was written a hundred years ago, the following rather extensive quotation (1907: 152f.) seems to be justified: “In einer vorhistorischen periode haben, wie ich vermuthe, die folgenden regeln gegolten: bei intransitiven verben stand das subjekt in der (u. a. auch als objekt fungirenden) grundform (bei o-stämmen die form auf -om, bei den -ā-, -n-, -r-stämmen die historische nominativform); bei transitiven verben stand das objekt in der grundform, das subjekt aber im genitiv, wenn wirklich von einer thätigkeit desselben die rede sein konnte, also wenn es der name eines lebenden wesens war; dagegen stand es im instrumentalis, wenn es ein unpersönlicher begriff war. Die beiden sätze: “der bruder tödtet das thier” und “der baum tödtet das thier” wurden also als “des bruders thiertödten” und “durch den baum thiertödten” ausgedrückt. Dabei ist der subjektsgenitiv natürlich als possessiver genitiv aufzufassen [...]. Allmählich differenzirt sich jedoch der subjektsgenitiv (der casus activus) von dem genitiv in seinen sonstigen Verwendungen [...]. Nachdem sich in dieser weise ein selbständiger casus activus entwickelt hatte, konnte dieser casus seine gebrauchssphäre erweitern, sodass er auch bei intransitiven verben als subjekt fungirte; eine zeitlang wird er in dieser funktion mit der grundform regellos abgewechselt haben, bis schliesslich bei den o-stämmen die grundform auf die nicht-subjektivische Verwendung beschränkt und dadurch zum accusativ gestempelt wurde. Die endung -m wurde dann als accusativendung auf die übrigen stammklassen übertragen; so trat 92 Indo-European morphosyntax beispielsweise eine form *ekʄuā-m ‘die stute’ (acc.) an stelle des älteren *ekʄuā, das nur noch als nominativ bewahrt blieb, in dieser verwendung aber den casus activus ganz verdrängte”. Concerning the original function of the ending *-m Pedersen remarks (1907: 156): “Ich dachte damals auch an die arabische nunation, die beim determinirten substantiv fehlt (farasun ‘ein pferd’, al farasu ‘das pferd’), und ich will jetzt diese vermuthung nicht verheimlichen. Falls das idg. -s des genitivs (und des casus activus) ursprünglich ein artikel war (was nicht ausgeschlossen ist, da eine verwendung des artikels beim genitiv, während es beim regens fehlte, mit mehreren lebendigen sprachen parallel sein würde), so wäre die indogermanische regel für das vorkommen des beweglichen -m mit dem Arabischen parallel”. The next major step in the reconstruction of PIE verbal syntax was taken by Holger Pedersen in another article which modern investigators have ignored (1933b: 311-315). The title is not mentioned in Collinder’s survey (1974), for example. Pedersen bases himself on the assumption that there were three series of personal endings in the Indo-European proto-language: 1. the “normal” endings, which are best preserved in the athematic flexion, 2. the perfect endings, which are also found in the thematic present, and 3. the middle endings. He puts forward the hypothesis that the perfect endings belonged originally to the flexion of intransitive verbs, and the “normal” endings to the flexion of transitive verbs. The distinction between these two sets of personal endings thus corresponds to the difference of verbal government between intransitive verbs, where the subject was in the nominative, and transitive verbs, where it was in an oblique case. Pedersen points to the identity of the “intransitive” 1st sg. ending *-ō with the ending of the nominative pronoun ἐγώ, and to the identity of the “transitive” endings 1st sg. *-m and 3rd sg. *-t with the oblique pronominal stems *me- and *to-. He also points to the possibility of identifying 3rd pl. “intransitive” *-r and “transitive” *-nt with the formative suffix of nom. ὕδωρ, obl. ὕδατ- < *-nt-. In an article which has received more attention than Pedersen’s studies, André Vaillant presented three arguments in favor of the hypothesis that the nom.sg. ending *-s is an ancient ergative ending, which he identifies with an original ablative ending (1936). First, there is a morphological opposition between animate and inanimate in Indo-European, which is reflected in nom. πολύς, acc. πολύν, neuter πολύ. Following Meillet (1931a), Vaillant assumes that the rise of the feminine gender is a recent development, which did not reach the Anatolian languages. Second, there is a suppletive nominative in pronominal paradigms, which is reminiscent of the suppletive ergative in Chechen. Third, Proto-Indo-European verbal syntax 93 there are two types of verbal flexion, which correspond to the Hittite conjugations in -hi and -mi. Vaillant assumes that the Hittite flexion in -hi corresponds to the Indo-European perfect, which is originally intransitive, while the flexion in -mi originated from the addition of pronominal elements to a verbal noun in -t-: 1st sg. 2nd sg. 3rd sg. 1st pl. 2nd pl. 3rd pl. *gwhenmi < *gwhent-m-i *gwhensi < *gwhent-t-i *gwhent-i *gwhenmes < *gwhent-m-es *gwhentes < *gwhent-w-es *gwhnont-i (participle) The final -i may be the vestige of a copula. In his monograph on Hittite (1938), Holger Pedersen repeated some of the considerations from his 1933 article. This account is again disregarded by K.H. Schmidt (1979). The cardinal point of Pedersen’s theory is the existence of a relation between the two types of verbal flexion in Hittite (-hi and -mi) and the two types of nominative ending (with and without -s). The sigmatic nominative expressed the subject of transitive verbs, which correspond to the Hittite flexion in -mi, whereas the asigmatic nominative expressed the object of transitive verbs and the subject of intransitive verbs. The Indo-European perfect, which corresponds to the Hittite flexion in -hi, was originally intransitive. The original distribution of -hi and -mi has been obscured and cannot be recovered. I think that the principal flaw in the conception of Pedersen and Vaillant is the insufficient distinction between flexion types. The identification of the intransitive perfect with the thematic flexion, which is predominantly transitive at the earliest reconstructible stage, cannot be substantiated. The same unwarranted assumption, among others, is made by Watkins (1969: 107-112). Similarly, we have to make a strict distinction between transitive and intransitive middle paradigms. II The status of the thematic flexion in the PIE verbal system has been the subject of much controversy. According to Meillet, the thematic type was originally limited to suffixed stems, e.g. in -ske- and -ne-, and to the subjunctive of athematic stems (1931b: 202). Vaillant assumed a twofold origin of the thematic present: on the one hand the sixth class of Sanskrit (tudáti) corresponds to the thematic flexion in -mi of Hittite (waššezzi, lukezzi, -škezzi), and on the other the paradigm of φέρω, -εις, -ει can be identified with the Hittite flexion in -hi of denominative stems in a laryngeal, e.g. newahhi (1937). These theories must now be reconsidered in the light of the Hittite evidence, which has gained much 94 Indo-European morphosyntax wider accessibility thanks to the publication of Norbert Oettinger’s monograph (1979). It follows from Oettinger’s analysis that the flexion in -mi is found with athematic stems, simple thematic stems, and derived stems in -ske- and -ie-, whereas the flexion in -hi is characteristic of old perfects, causatives and iteratives, denominative stems in -ahh-, and derived stems in -ie- after a rootfinal laryngeal. In the course of the historical development, the flexion in -hi is gradually eliminated. Stems in -ahh- generally belong to the flexion in -mi after the Old Hittite period. In my opinion, the principal step toward a solution of the problem of the thematic flexion was made in 1953 by Johannes Knobloch, who identified the thematic vowel with an object marker. His article does not seem to have evoked any response in the literature, probably because he limited himself to a typological comparison with Circassian and did not adduce any historical evidence in support of his view. Against Pedersen’s identification of the flexion in -mi with the transitive conjugation Knobloch objects that the distribution of Hittite -mi and -hi does not correspond to a distinction between transitive and intransitive verbs, e.g. ešmi ‘I am’, pāimi ‘I go’ versus dāhhi ‘I take’, pihhi ‘I give’. Athematic root verbs in -mi are particularly often intransitive in the IndoEuropean languages. Referring to an article by Jakolev, Knobloch cites the following Kabardian examples as an illustration of the three types of syntactic construction which are found with Circassian verbs: ś’ále-r mà-Ǐe ‘le garçon crie’ ś’ále-r txǩɳλǩ-m yó-Ǐe ‘le garçon lit (dans) le livre’ ś’ále-m txǩɳλǩ-r ye-Ǐ ‘le garçon lit le livre (en entier)’ In the first example, the subject is in the absolute case. Knobloch compares this intransitive construction with the Indo-European type with a verb in *-mi. In the second example, the subject is in the absolute and the (indirect) object in the relative case. The verb has a zero subject prefix and an indirect object marker yó-. This is the construction which Knobloch compares with the Indo-European thematic flexion, the thematic vowel corresponding to the object prefix. In the third example, the subject is in the relative and the (direct) object in the absolute case, while the verb has a zero object marker and an actor prefix ye-. Knobloch compares this transitive construction with the Indo-European perfect, where the thematic vowel is absent. Thus, he arrives at the following reconstruction of the Indo-European verb phrase: Proto-Indo-European verbal syntax 95 – construction of the ergative type: – objective flexion: -o-H- (thematic present) – athematic flexion: -H- (perfect) – construction of the nominative type: – objective flexion: -o-m (thematic aorist and imperfect) – athematic flexion: -m(i) (present and aorist) Knobloch adds that the thematic vowel of nominal o-stems can also be regarded as a petrified object marker. For a more detailed and accurate description of the Circassian case system I refer to Kuipers 1962. The main objection which can be raised against Knobloch’s reconstruction is that the Indo-European perfect was undoubtedly intransitive at the earliest reconstructible stage, so that the hypothesis that it was construed with an ergative is highly unnatural. Moreover, the conjecture that the thematic present was construed with an ergative while the thematic aorist was construed with a nominative or absolutive case runs counter to the expected state of affairs. It seems preferable to return to Pedersen’s suggestion that the flexion in *-Hcorresponds to an intransitive type of construction whereas the ergative case correlated with the endings 1st sg. *-m, 2nd sg. *-s, 3rd sg. *-t. III As I have indicated elsewhere (K033: 67f.), I think that we have to assume six Proto-Indo-European classes of verbal stems, which were characterized by the following sets of endings: imperfective perfective dynamic subjective 1st sg. 2nd sg. 3rd sg. 3rd pl. -mi -si -ti -nti -m -s -t -nt dynamic objective 1st sg. 2nd sg. 3rd sg. 3rd pl. -oH -eH1i -e -o -om -es -et -ont 1st sg. 2nd sg. 3rd sg. 3rd pl. -H2 -tHo -o -ro -H2e -tH2e -e -r static These sets of endings correspond to the historically attested athematic present and aorist, thematic present and aorist, stative (intransitive middle) and perfect. 96 Indo-European morphosyntax The opposition between the laryngeals was neutralized in the neighbourhood of PIE *o (cf. K034: 128). The six types of paradigm were interconnected by a network of derivative, not flexional relations. For the origins of the middle paradigms I refer to the exposition which I have given elsewhere (K044, K203). The stative and the perfect were inherently intransitive, while the objective flexion was transitive and the subjective flexion could be either. The distinction between subjective and objective flexion is characteristic of the Uralic languages. In Hungarian, for example, the verb várni ‘to wait’ has the following paradigms: subjective objective present 1st sg. 2nd sg. 3rd sg. várok vársz vár várom várod várja preterit 1st sg. 2nd sg. 3rd sg. vártam vártál várt vártam vártad várta The objective flexion is used with a definite direct object, e.g. várom a fiamat ‘I am waiting for my son’. The subjective flexion is used if there is no definite direct object, e.g. várok valakit ‘I am waiting for somebody’. The objective personal endings are identical with the possessive suffixes, cf. karom ‘my arm’, karod ‘thine arm’, karja ‘his arm’. It is usually assumed that the opposition between subjective and objective conjugation cannot be traced back to Proto-Uralic (e.g. Collinder 1960: 244, Hajdú 1975: 101). There are two weighty arguments against this view. Firstly, the objective conjugation is common to Hungarian, Ob-Ugric, Samoyed, and Mordvin, and traces have been claimed for Saami and, less convincingly, for Cheremis and the Permian languages (see Tauli 1966: 171f. for references). It is improbable that the rise of the objective flexion is an independent development in all branches of Uralic. Secondly, there is to a certain extent formal agreement between the languages. In Yurak, which is in several respects the most archaic of the Uralic languages, the endings of 1st sg. subj. madāu ‘I cut’ and obj. madādm ‘id.’, where -d- is an infixed object pronoun and -m is the subject marker, correspond formally to those of Hung. váro-k and várom, respectively, and the 2nd sg. obj. endings of Yurak madān ‘you cut’, Yenisei motadd-o ‘id.’, Selkup ńoand ‘you hunted’, and Hung. várod all point to Proto-Uralic *-nt (cf. Pedersen 1933: 323f.), cf. also 2nd sg. subj. Selkup ńoal, Hung. vártál. The main argument against the hypothesis that the verbal system of the Uralic languages is very ancient is the widespread identity of the flexional endings with the personal or possessive pronouns. This argument is inconclusive because the identity may be the result of analogic remodelling. Proto-Indo-European verbal syntax 97 Thus, the identity of the Polish 1st pl. ending -my with the personal pronoun my is comparatively recent but cannot be used as evidence for a recent origin of the flexional system. The ending -my replaces Old Polish -m, which had merged with the athematic 1st sg. ending as a result of the loss of final jers. There is evidence that the Uralic endings were subject to a similar type of restructuring at various stages in the development of the separate languages. In Hungarian, the objective personal endings are identical with the possessive suffixes, as opposed to the subjective personal endings. In Yurak, which is representative for the Samoyed languages, the subjective endings are identical with the possessive suffixes after singular nouns, while the objective endings for dual and plural objects are identical with the possessive suffixes after dual and plural nouns, in contradistinction to the different objective endings for singular objects. This distribution is undoubtedly secondary. The infixed object pronoun which is present in 1st sg. obj. Yurak madādm, Yenisei motaro’, is also present in 3rd sg. subj. madāda, motara, Selkup ńoed, but absent from 3rd sg. obj. madā, mota, ńoe-k. This reversal of the 3rd sg. subj. and obj. endings must be attributed to the influence of the possessive suffixes. In Selkup, the 1st sg. subj. and obj. suffixes have also interchanged places, e.g. subj. ńoap < *-m, obj. ńoa-k ‘I hunted’, cf. Yurak subj. madāu, obj. madādm, Hung. subj. váro-k, obj. várom. The hypothesis that there was a distinction between subjective and objective flexion in Proto-Indo-European cannot be proven in any strict sense of the word, but it offers an explanation for at least three sets of data in the oldest material of the historically attested languages: the distribution of the Hittite thematic flexion, the origin of the sixth class of Sanskrit (tudáti), and the rise of the subjunctive. It may also offer an explanation for the distribution of the thematic aorist in Greek, which will not be discussed here. In the course of the historical development, the Hittite flexion in -hi is gradually replaced with the flexion in -mi. It is probable that this development had started before the earliest texts already, so that we can assume that some of the verbs which belong to the flexion in -mi in the oldest material had been transferred to that class at an earlier stage. The thematic 3rd sg. ending *-e was identical with the perfect 3rd sg. ending *-e at the very outset, while the 1st sg. endings *-oH and *-H2e were sufficiently alike to induce analogic leveling. As a result of the loss of *H1, the 2nd sg. ending *-eH1i merged with the 3rd sg. ending when the latter took the characteristic *-i from the athematic flexion: it was therefore predisposed to replacement with a more distinctive ending. In view of all this, it is remarkable that the thematic present did not entirely merge with the perfect. I think that the reason must be sought in the addition of *-i from the athematic present to the perfect endings at a stage when the thematic present was still a distinct inflexional type. The transfer of causatives and iteratives to the flexion of the perfect can be understood if we assume that the final vowel of 3rd sg. *-eie was dropped before the loss of intervocalic *i, so that the ending 98 Indo-European morphosyntax merged with the corresponding perfect ending at a stage which was posterior to the addition of *-i to the perfect endings but anterior to the loss of the thematic present flexion. This explanation is far more probable than the complicated mechanism which Oettinger suggests (1979: 452ff.). The remaining thematic presents were subsequently transferred to the flexion in -mi, perhaps under the influence of the secondary endings. The transfer was late in the case of the denominatives in -ahh- < *-eH2e-, which in Old Hittite belong to the flexion in -hi. Thus, the expected reflex of the PIE thematic flexion in Hittite can be found in the simple flexion in -ami, the derived flexions in -škami and -(j)ami, the causatives and iteratives in -ahhi, and the denominatives in -ahhahhi (e.g. happinahhahhi, Oettinger 1979: 41). All of the simple verbs are transitive with the exception of papre- ‘unrein sein’ (Oettinger 1979: 282ff.). The inherited verbs in -ske- are transitive, and so are the causatives and iteratives and the verbs in -ahh-. The numerous verbs in -je- constitute a heterogeneous class, the nondenominatives among them being almost exclusively transitive. Oettinger’s view that šišzi < *sisd-ti represents PIE *sisde- ‘sit’ with analogical athematic flexion (1979: 216) must be rejected because no such verb existed in the proto-language. The intransitive meaning of Skt. sãdati ‘sits’ and tíṣṭhati ‘stands’ is the result of a secondary development, as Thieme has demonstrated (1929: 55), cf. ἵζω, ἵστημι. The sixth class of Sanskrit (tudáti) has punctual meaning in Vedic, except in the case of originally athematic verbs which were transferred to the thematic flexion (cf. Renou 1925: 310). The verbs of this class are characteristically accompanied by an implicit or explicit definite object. In addition to the examples which Renou adduces (l.c.), the following instances can serve for illustration (the translation is from Geldner 1951). 1.67.7-8 yá īṃ cikéta gúhā bhávantam ¡ yáḥ sas¡da dh¡rām ṛtásya, ví yé cṛtánty ṛt¡ sápanta ¡d íd vásūni prá vavācāsmai “Wer ihn entdeckt hat, da er sich versteckt hielt, wer zum Strom der Wahrheit gelangt ist – jedem der (den Strom der Wahrheit) entbindet, die Wahrheit pflegend, – dem hat (Agni) darnach Gutes verheissen.” 3.29.14 prá saptáhotā sanak¡d arocata mātúr upásthe yád áśocad dhani, ná ní miṣati suráṇo divédive yád ásurasya jaṭhárād ájāyata “Von sieben Opferpriestern umgeben erstrahlte er seit alters, wenn er im Schosse der Mutter, an ihrem Euter erglühte. Nicht schliesst der Erfreuliche Tag für Tag die Augen, nachdem er aus dem Leibe des Asura geboren wurde.” (cf. Latin micāre ‘to twinkle’) 5.30.13 supeśásaṃ m¡va sṛjanty ástaṃ gávāṃ sahásrai ruśámāso agne “Reich geschmückt entlassen mich die Ruśama’s mit Tausenden von Kühen nach Hause, o Agni.” Proto-Indo-European verbal syntax 99 5.53.6 ¡ yáṃ náraḥ sud¡navo dadāśúṣe diváḥ kóśam ácucyavuḥ, ví parjányaṃ sṛjanti ródasī “Wenn die gabenschönen Herren für den Opferspender des Himmels Eimer heraufgezogen haben, so lassen sie den Parjanya (Regen) über beide Welten sich ergiessen.” 6.36.3 táṃ sadhrãcīr ūtáyo vṛɴṣṇyāni páuṃsyāni niyútaḥ saścur índram, samudráṃ ná síndhava uktháśuṣmā uruvyácasaṃ gíra ¡ viśanti “Den Indra begleiten vereint die Hilfen, die Bullenkräfte, die Manneskräfte, die Gaben. Wie die Ströme in das Meer, so gehen die Lobreden, durch Loblieder verstärkt in den Geräumigen ein.” In all of these instances, it is the object rather than the subject which experiences a change of state as a result of the action. Renou regards the verbs of the sixth class as originally modal forms and compares them with the subjunctive, which he considers to be the starting-point for the formation of numerous thematic indicatives (1925: 315). The Vedic subjunctive is a thematically inflected stem: “le seul trait qui caractérise le subjonctif est la voyelle thématique” (Renou 1932: 14). As Renou points out, the original meaning of this form is best preserved in those cases where the athematic stem does not constitute an indicative paradigm: “pour rendre compte des notions liées au système thématique, il faut tabler sur les formations autonomes, non sur celles qui ont adhéré à un système particulier de présent et d’aoriste” (1932: 15). And here we find that “une forme telle que karati, que rien ne rattache à un thème spécial, possède une valeur trouble, mi-réelle mi-modale, et telle qu’il serait vain de restituer un karati indicatif à côté d’un karati subjonctif ” (ibidem). The best example is precisely the stem kara-, which is attested 75 times in the Ṛgveda: “en majorité subjonctif, mais subjonctif indéterminé, éventuel, plutôt que modal,” without regard to the presence of either primary or secondary endings. Compare the following examples: 2.35.1 ap¡ṃ nápād āśuhémā kuvít sá supéśasas karati jóṣiṣad dhí “Gewiss wird Apām Napāt, der Rossetreiber, (meine Lobrede) zieren, denn er soll seine Freude daran haben,” “peut-être Apām Napāt, animateur de coursiers, rendra-t-il (mes chants) richement ornés?” 7.88.1 prá śundhyúvaṃ váruṇāya préṣṭhāṃ matíṃ vasiṣṭha mīḷhúṣe bharasva, yá īm arv¡ñcaṃ kárate yájatraṃ sahásrāmaghaṃ vṛɴṣaṇaṃ bṛhántam “Vasiṣṭha! Bring ein sauberes, recht angenehmes Gedicht dem belohnenden Varuṇa dar, der den verehrungswürdigen, tausend Gaben bringenden grossen Bullen herwärts lenken soll,” “présente à Varuṇa la prière la mieux aimée, qui amène (qui amènera) le taureau.” 6.18.14 ánu tv¡highne ádha deva dev¡ mádan víśve kavítamaṃ kavīn¡m káro yátra várivo bādhit¡ya divé jánāya tanvè gṛṇānáḥ “Da jubelten alle Götter dir, o 100 Indo-European morphosyntax Gott, dem Weisesten der Weisen im Drachenkampf zu, in dem du gepriesen dem bedrängten Himmel, dem Volke, dir selbst einen Ausweg schufest,” “alors, ô dieu, les dieux se réjouirent à ton sujet, ô tueur du Dragon, quand à l’opprimé tu procuras le libre espace.” 5.31.11 śraś cid ráthaṃ páritakmyāyāṃ prvaṃ karad úparaṃ jūjuv¡ṃsam, bhárac cakrám étaśaḥ sáṃ riṇāti puró dádhat saniṣyati krátuṃ naḥ “Auch den Wagen der Sonne, der vorausgeeilt war, brachte er im entscheidenden Augenblick ins Hintertreffen. Etaśa trug das Rad davon; er stellt es her. Wenn er (ihn) an die Spitze bringt, wird er unsere Absicht erreichen.” In connection with the last two examples Hoffmann remarks: “Auch an der zweiten Stelle VI 18.14, wo káraḥ allgemein präterital übersetzt wird, braucht durchaus kein präteritaler Tatbestand vorzuliegen: ‘Da jubeln (mádan, Inj. Präs.) dir alle Götter beim Drachenkampfe zu, in dem (yátra) du Weite dem bedrängten Himmel, dem Volke, dir selbst schaffen wirst (káraḥ)’. Wenn man aber an einer präteritalen Situation festhalten will, so lässt sich auch rechtfertigen: ‘Da haben dir die Götter beim Drachenkampfe zugejubelt, in dem du ... schaffen solltest (d.h. damit du dabei schaffest)’ . [...] Geldner übersetzt karat auch V 31.11 präterital, doch schon die auf karat folgenden Verbformen (bhárat, sáṃ riṇāti, saniṣyati) machen das unwahrscheinlich” (1967a: 5537). Renou concludes from the Vedic facts that the subjunctive was originally an independent formation, characterized by the mere presence of the thematic vowel, with a semi-modal value which could develop either into the historical subjunctive or into the inexpressive and aspectually indeterminate indicative of the first present class: “Le subjonctif prévaudra dans la mesure où le verbe conserve un présent ou un aoriste athématique qui soutient ce mode; l’indicatif, dans la mesure où le thème en -a- est senti comme isolé et indépendant” (1932: 29). He remarks that “dans bhárati, en regard de bíbharti, l’évolution est à son terme, l’incorporation du thème bhara- au système indicatif est totale et rien ne décèle immédiatement l’origine modale. Mais on observera que bíbharti fournit le présent expressif ‘tenir en mains, soutenir, maintenir’, et aussi ‘porter dans son sein’; bhárati ‘apporter ou emporter, procurer, offrir’ et au moyen ‘recevoir’, implique une participation de la volonté du sujet [...]. Au point de vue des désinences, dans le Rv., on observera que, si bíbharti possède uniquement la série primaire et la voix active, bhárati reçoit aussi la série secondaire et le moyen, avec une répartition des finales qui rappelle celle du subjonctif. Quant à bhárti, la forme fait corps avec bíbharti pour le sens; exceptionnellement rare, elle ne saurait appuyer l’origine thématique normale de bhárati” (1932: 23f.). On the endings of the subjunctive see now Beekes 1981. The facts which have been adduced here can be understood if we start from the assumption that the thematic vowel was originally an object marker. Consider the following Bulgarian examples: Proto-Indo-European verbal syntax 101 spj-a ‘I sleep’ spi mi se ‘I am sleepy’ In the first example the stem is followed by the 1st sg. ending -a. In the second it is followed by the zero 3rd sg. ending, the enclitic 1st sg. dative pronoun, and the reflexive particle. The structure of these forms is immediately comparable with that of Skt. ádmi ‘I eat’, where -mi is the 1st sg. subject marker, and Gr. ἔδομαι ‘I will eat’, where the root is followed by the thematic vowel -o-, the 1st sg. marker -m-, and the middle voice marker -ai. While the Bulgarian case shows how the subjunctive can have originated from a type of objective flexion, the nonvolitional variant which underlies Skt. bhárati is found in Polish. In this language, where the translation of the above examples is śpię and chce mi się spać (same syntactic construction with 3rd sg. chce ‘wants’ and inf. spać ‘to sleep’), the “objective” construction is found in such instances as spało mi się bardzo smacznie, which is practically equivalent to spałem bardzo smacznie ‘I slept very soundly’. The position of Russian appears to be intermediate in this respect, e.g. mne ne spitsja ‘I cannot sleep’ (Polish nie mogę zasnąć), but mne xočetsja spat' ‘I am sleepy’ (Bulg. spi mi se). As the Slavic parallel demonstrates, the fact that the stem is intransitive is no obstacle to the derivation of a modal category from an objective construction. The Slavic examples adduced here contain a reflexive particle. As Thieme observed (1929: 53), there is a correlation between thematic flexion and middle voice, as opposed to an athematic active paradigm, in the oldest Indo-European material. “II est certain qu’il y a plus généralement une tendance vers la voix moyenne dans la plupart des systèmes thématiques; à cet égard le contraste hánti jíghnate, sácate síṣakti est saisissant [...]. Le moyen est aussi rare dans les présents et aoristes radicaux, dans le présent redoublé athématique, qu’il abonde dans les présents thématiques” (Renou 1932: 211). IV Now we return to the syntax of the PIE finite verb. It has long been recognized that there is a striking resemblance between the verbal systems of Georgian and classical Greek. It may therefore be profitable to have a look at the syntax of the Georgian verb. The following examples are characteristic of literary Georgian (cf. Boeder 1979: 437): txa modi-s ‘the goat comes’ (present) txa movid-a ‘the goat came’ (aorist) txa mosul-a ‘the goat has apparently come’ (perfect) txa č’am-s venax-s ‘the goat eats the vine’ (present) txa-m šeč’am-a venax-i ‘the goat ate the vine’ (aorist) txa-s šeuč’ami-a venax-i ‘the goat has apparently eaten the vine’ (perfect) 102 Indo-European morphosyntax The subject of a regular intransitive verb is always in the nominative, which ends in zero after a vowel and -i after a consonant. With transitive verbs, the case forms depend on the tense system. In the present tense, the subject is in the nominative and the object is in the dative, which ends in -s. In the aorist, the subject is in the ergative, which ends in -m after a vowel and -ma after a consonant, and the object in the nominative. In the perfect, the subject is in the dative and the object in the nominative. There is a class of intransitive verbs which have an ergative subject in the aorist, e.g. c’q’al-i duγ-s ‘the water boils’ (present) c’q’al-ma iduγ-a ‘the water boiled’ (aorist) is t’iri-s ‘he weeps’ (present) man it’ir-a ‘he wept’ (aorist) There are transitive verbs which have a dative object in the aorist. There is a class of verbs which have a dative subject and a nominative object in the present (indirect or inverted verbs), e.g. deda-s uq’var-s švil-i ‘the mother loves the child’. The verbal syntax of Georgian is more archaic than that of the related Megrelian and Laz languages. In Megrelian, the use of the ergative case was generalized in the aorist, irrespective of transitivity or intransitivity of the verb. In Laz, the use of the ergative case with transitive verbs was generalized, irrespective of tense. A particularly instructive survey of the historical development of verbal syntax in Georgian, including the dialects, can be found in Boeder’s contribution to the handbook on ergativity (1979). For Proto-Indo-European we can assume that the subject was in the absolutive (asigmatic nominative) case in the stative and the perfect because these categories were intransitive. The original derivative relationship between a transitive present and an intransitive perfect has been preserved in πείθω ‘I persuade’, πέποιθα ‘I trust’, ῥήγνῡμι ‘I break (tr.)’, ἔρρωγα ‘I am broken’. If the agent was mentioned with the perfect, it was probably in the dative if it was animate and in the instrumental if it was inanimate. If this is correct, the original syntactic construction is preserved in τοῦτό μοι πέπρακται, where the verb has received an analogical middle ending, while the syntax of πέπρᾱχα τοῦτο was taken from the present tense. The original perfect πέπρᾱγα has preserved the intransitive meaning. In the thematic flexion, which always had two arguments, the thematic vowel referred to an object in the absolutive case. I stick to Pedersen’s view that the secondary endings *-m, *-s, *-t etc. referred to a subject in the ergative (sigmatic nominative) case. For the thematic present I assume that the subject was originally in the dative if it was animate and in the instrumental if it was inanimate. Thus, the syntactic construction of the thematic present was the same as that of Bulg. spi mi se or rather Eng. me dreamed a strange dream. The substitution of I for me in modern English had its analogue in late Proto-Indo- Proto-Indo-European verbal syntax 103 European, cf. also German mir träumt and ich träume. The fact that Bulgarians are discouraged from saying az mi se spi, where az is the nominative of the 1st sg. pronoun, testifies to the same development happening right now in that language. The hypothesis that the subject of a thematic present was originally in the dative accounts for the correlation between middle presents and active aorists in a number of instances, e.g. δέρκομαι ‘I see’, ἔρχομαι ‘I go, come’ (often with ὁδόν ‘road, journey’: the original meaning was perhaps ‘to cover a distance’, cf. Skt. ṛccháti ‘reaches’), aor. ἔδρακον, ἤλυθον. The PIE transitive middle expressed the identity of the subject with the indirect object: it can be compared with the subjective version in Georgian, e.g. me vimzadeb sadils ‘I prepare myself a dinner’ (cf. neutral version me vamzadeb sadils ‘I prepare a dinner’, intransitive middle vemzadebi ‘I prepare myself). Like its Georgian counterpart, the PIE transitive middle had probably the same syntactic construction as the corresponding active forms. The presence of a dative subject in the thematic present prompted the spread of the transitive middle endings, which correlated with the identity between the subject and the indirect object. In its turn, the spread of the transitive middle endings facilitated the substitution of the ergative (sigmatic nominative) case for the dative with thematic presents in late ProtoIndo-European. With athematic presents and aorists, the subject was probably in the ergative if the verb was transitive and in the absolutive if the verb was intransitive. The apparent contradiction between Pedersen’s view that the endings *-m, *-s, *-t etc. referred to the ergative subject of a transitive verb and Knobloch’s observation that intransitive verbs belong as a rule to the flexion in -mi has a remarkable counterpart in Samoyed, where intransitive verbs characteristically receive the endings of the objective flexion (cf. Castrén 1854: 207), e.g. Yurak adm ‘I am’. Indeed, I think that this is a major argument in favor of the Indo-Uralic hypothesis. As is commonly assumed, the accusative developed from a directive case (e.g. Haudry 1977: 155). The same development is attested at a later stage in Romance, e.g. Spanish la madre quiere a su niño, Rumanian mama iubeşte pe pruncul său ‘the mother loves her child’, cf. Latin ad ‘to’, prae ‘for’. The substitution of the accusative for the absolutive was probably early in the case of effective verbs, individualized objects, and emphasis (cf. Pottier 1968). The generalization of -m in the neuter nom.acc.sg. ending of the o-stems can perhaps be attributed to the semantics of this category, which supplied an expression for individual members to a collective in -ā. It is recalled that the neuter does not function as a subject of transitive verbs in Hittite. GREEK NUMERALS AND PROTO-INDO-EUROPEAN GLOTTALIC CONSONANTS Gamkrelidze and Ivanov have suggested on typological grounds that the reconstructed voiced occlusives of the Indo-European proto-language were actually glottalic (1973). A similar argumentation was put forward earlier by Holger Pedersen (1951: 14). Elsewhere I have argued that this hypothesis is supported by immediate comparative evidence from Latvian (K025), Armenian (K031), and Sindhi (K038), and by indirect evidence from Balto-Slavic (Winter’s law, cf. Winter 1978), Latin (Lachmann’s law, cf. K086), and Indo-Iranian (Bartholomae’s law, cf. K032). Additional evidence from Indo-Iranian has been adduced by Alexander Lubotsky (1981). Elsewhere I have argued that the new theory provides a possible explanation for the rise of preaspiration in Icelandic and the so-called vestjysk stød in Danish (cf. K072). Here I intend to show that it offers an explanation for several problems in connection with the formation of the numerals in Greek. The PIE word for ‘100’ is usually reconstructed as *kʄṃtóm. This reconstruction does not account for the initial vowel of ἑκατόν. The initial vowel is usually derived from *sem- (e.g. Risch 1962: 132) or *sṃ- (e.g. Chantraine 1967: 150) if its origin is not simply called “unknown” (e.g. Beekes 1969: 53) or left out of consideration altogether (e.g. Szemerényi 1960). These explanations meet with several difficulties. There is no support for the ad hoc hypothesis that the final nasal of *sem was dissimilated before *kṃtóm. The alleged substitution of ἑκατόν for *ἁκατόν < *sṃ- cannot be compared with the replacement of ἅτερος with ἕτερος because the latter is limited to a part of the dialects. Above all, the assumption of a composite form ‘one hundred’ is at variance with the indeclinability and the syntactic behavior of ἑκατόν. The original character of the indeclinability is evident both from the preservation of the final nasal in composition and from the impossibility of using *κατά after τρία etc. If such a construction had been possible at an earlier stage, it would hardly have been replaced with a derivative formation in -io-. Thus, I think that the Greek form and its syntax are more archaic than is generally assumed. The hypothesis that the unaspirated voiced stops of the Indo-European proto-language were glottalic offers a straightforward explanation for the initial vowel of ἑκατόν. If we start from *dkʄṃtóm, we can assume that the buccal features of the initial consonant were lost while its glottalic feature merged with the reflex of the PIE laryngeal *H1 and yielded *e-: *ἐκατόν. The aspiration was apparently taken from ἕν. Thus, I agree with Frisk’s prudent variety of the above hypotheses that initial ἑ- “muß irgendwie mit ἕν ‘eins’ oder idg. *sṃ- (gr. ἁ-) zusammenhängen” (1973: 475). The ending of *dkʄṃtóm can be identified with 106 Greek the gen.pl. ending *-om, not because the numeral represents an original gen.pl. form (this is Szemerényi’s view, o.c. 140, which Risch has effectively rebutted, o.c. 135) but because the gen.pl. must be derived from an uninflected PIE form (cf. K030: 294f.). The original meaning of *dkʄṃtóm was “pertaining to the category of portions of reality which carry the feature ‘consisting of 10 members’”. As Laroche remarked on the Hittite genitive in -an (1965: 40), the ending is characteristic of “êtres ou catégories allant par groupes”. The explanation put forward here has the additional advantage of simultaneously accounting for the long vowel in the decades, a traditional analysis of which leads to the following reconstruction (cf. Szemerényi, o.c. 24): ‘20’ ‘30’ ‘40’ ‘50’ ‘60’ ‘70’ ‘80’ ‘90’ *wīkṇtī *trīkont*kwetwṝkont*penkwēkont*swekskont*septṃɷkont*oktōkont*newṇɷkont- εἴκοσι τρι³κοντα τετταράκοντα πεντήκοντα ἑξήκοντα ἑβδομήκοντα ὀγδοήκοντα ἐνενήκοντα vīgintī trīgintā quadrāgintā quīnquāgintā sexāgintā septuāgintā octōgintā nōnāgintā I now reconstruct *d instead of the vowel length in the proto-forms, e.g. *penkwedkʄomt. This view has been put forward a number of times in the earlier literature, as Szemerényi points out (o.c. 136). It has never been explained, however, that *d merged with the laryngeals in this environment. Szemerenyi’s reference to a chronological difference begs the question because he does not discuss the origin of the Indo-European lengthened grade. I agree with Szemerényi that originally there was no final vowel in ‘30’ through ‘90’, especially because Indic -śat would otherwise be very difficult to explain. The assumption that *d was also present in *swekʄsdkʄomt, where it lost its glottalic feature and was assimilated to *t before *kʄ, provides an explanation for the rise of the Indo-Iranian suffix -ti- in the higher decades if it is correct that the final *kʄ of the cluster was lost in this branch of Indo-European. Greek eliminated the cluster by introducing -ē- from πεντήκοντα into ἑξήκοντα. It can be objected against the theory advanced here that the long vowel is absent from ἑπτακόσιοι etc. The objection does not hold because the latter formation is limited to Greek and must apparently be dated to the period after the reanalysis of *ἐκατόν as ἑ-κατό-ν. The Indo-European proto-language had no names for the hundreds. In the case of *septṃdkʄomt, the hypothesis that *d yields the same reflex as *H1 accounts for the difference between -ē- in ἑβδομήκοντα and -ā- in the corresponding Latin form. It may also account for the irregular voicing in Greek. The alleged development of PIE *septmós to *ἑβδαμός rather than *ἑπταμός lacks parallels and can hardly be correct. More probably, the voicing Greek numerals and Proto-Indo-European glottalic consonants 107 arose in *ἑβδμήκοντα after the development of the syllabic nasal at a stage which was posterior to Sievers’ law, cf. θνητός < *dh(w)ṇH2tós, στρωτός < *stṛH3tós. (This is in agreement with Szemerényi’s view, o.c. 8). It then spread to ἕβδομος, from there to ὄγδοϝος, which replaced *ὀκτοϝός < *H3kʄtH3wós, and finally to *ὀγδϝήκοντα, which replaced *ὀκτώκοντα < *H3ekʄtoHdkʄomt-. (On the neutralization of the timbre opposition between the laryngeals in the neighbourhood of PIE *o see K034. Vedic aśītí- developed by dissimilation from *HaśtḤti-, which reflects *H3ekʄtHdkʄomt with medial zero grade from the ordinal *HaśtHa- < *H3ekʄtHo-.) The introduction of the medial vowel from the ordinals into ἑβδομήκοντα and ὀγδοήκοντα is probably late. The form *H1newṇdkʄomt yielded *ἐνεϝνήκοντα, in which *w was apparently lost at an early stage because of the aberrant syllable structure. At a later stage, *w was assimilated or became u before a following resonant (cf. Lejeune 1972: 181f.). The usual derivation from *enwen- cannot be correct because that would yield **εἰνενήκοντα in Ionic, cf. εἴνατος < ἔνϝατος. The theory advanced here also provides an explanation for the coexistence of εἴκοσι < *ἐϝÕκοσι and ϝÕκατι, both of which represent PIE *dwidkʄṃti. Partial dissimilation of the initial consonant yielded *H1widkṇti, from which the southern form must be derived, whereas total dissimilation yielded the northern form. Unlike Szemerényi (o.c. 24), I assume that Indo-Iranian and Greek faithfully reflect PIE short -i, which was lengthened in Latin and British by the addition of the dual ending *-H1. The long vowel of τρι³κοντα cannot represent an original plural ending -ā, as is often assumed, because it is absent from τρία < *triH2. I think that the form continues *triaHkonta, which developed from PIE *triH2dkʄomt under the influence of τρία. Indeed, this is the form where in my view the final vowel of -κοντα originated on the analogy of *dwidkʄṃti and from where it spread to the higher decades. The difference between *dwi- and *triH2- has been preserved in Old Irish fiche < *wikent- versus trícho < *trīkont- and in Tocharian B ikäṃ < *wikṇt versus täryāka < *triaka, with Proto-Tocharian *a as the phonetic reflex of PIE *H2. The ending -a < *-H2 is the regular Proto-Tocharian plural ending, which replaces *-ont in the decades. The short vowel which is reflected in Breton tregont is doubtless of analogical origin. British ü- in Old Welsh uceint and Breton ugent also points to a short vowel: it represents *wi- before a syllable with a front vowel, cf. Welsh ucher ‘evening’, Latin vesper. The development of üfrom *wi- is not due to umlaut but to the preservation of the palatal feature in this environment. The initial part *dwi- is apparently the PIE neuter form *dwoi of the root *du- ‘2’ with zero grade in composition, and *triH2- is similarly the neuter form with zero grade of the root *tri ‘3’. While *d merged with *H1 in πεντήκοντα, ἑβδομήκοντα and ἐνενήκοντα, it apparently merged with *H3 in τετρώκοντα, which was preserved in West Greek and regularized to *τετϝαράκοντα elsewhere. I think that τετρώκοντα developed 108 Greek regularly from *kwetwṛdkʄomt and that the rounding of the medial vowel represents the lost *w, as is the case with the rounding of the epenthetic vowel in τρυφάλεια < *kwtwṛ-. The combined evidence of τρι³κοντα, τετρώκοντα and πεντήκοντα allows us to date the merger of *d with the laryngeals to a stage which was posterior to the rise of colored epenthetic vowels, but anterior to the eventual loss of the laryngeals (which had merged as a result of the previous development): *triHaHkont, *kwetwroHkont, *penkweHkont. Since the development of colored epenthetic vowels is specifically Greek, it follows that the PIE glottalic consonants were preserved up to a stage which was posterior to the separation from the other languages. This result is in accordance with what has been demonstrated earlier for Indo-Iranian (cf. Lubotsky, o.c.), Armenian, Balto-Slavic, and Germanic (cf. K032: 110-114). I think that the same holds for Albanian, Italic and Celtic. The Albanian material is difficult to interpret, as it usually is. The initial consonant of -zet ‘20’ must be derived from *gʄw- because it requires the simultaneous presence of labial, palatal, and velar articulation (cf. K022: 249; the rule was first established by Pedersen 1900b: 338). It probably originated from assimilation in *dwigʄṇti < *dwidkʄṃti, with *gʄ combining the glottalic feature of *d with the palatovelar articulation of *kʄ. A similar explanation could be put forward for the voiced stop in Latin vīgintī if the voicing were not absent from the ordinal vīcēsimus and from Old Irish fiche. Moreover, Lachmann’s law suggests that a glottalic consonant dissolved into a sequence of a laryngal and a voiceless buccal part, the former of which merged with the reflex of the PIE laryngeals, when it was preceded by a vowel and followed by a voiceless stop (cf. K032: 117). One therefore expects *dkʄ to develop into *Hk, not into *g. However, it is probable that the cluster became voiced after a nasal (cf. Thurneysen 1883: 313). The neutralization of the opposition between glottalic and aspirated stops in the position after a nasal accounts for the absence of Lachmann’s law in strictus and pictus, which adopted the short vowel of fictus and mictus, cf. stringō, pingō < *-g- and fingō, mingō < *-gh-. The glottalic feature was apparently absorbed by the preceding laryngeal in lassus < *lH1dtos, just as the laryngeal was absorbed by the following glottalic obstruent in Vedic pajrá- ‘firm’ < *peH2gʄro- (Lubotsky, o.c.). The initial syllable of sedeō was prefixed to *sdtos in -sessus, where the glottalic feature had been lost at an early stage, cf. the zero grade in nīdus < *nisdos. In my view, *septṃdkʄomt and *H1newṇdkʄomt developed into *septṃHgont and *newṇHgont, which subsequently yielded *septmāgont and *newnāgont in Italo-Celtic. Szemerényi’s view that the introduction of zero grade in Latin -gintā was anterior to the elimination of the long resonants (o.c. 169) cannot be correct because the latter development was Italo-Celtic and the former was not (cf. K046: 14). Note that Old Irish sechtmogo can represent either *septmāgont or *septmākont, with medial o for a under the influence of the preceding labial (cf. Greek numerals and Proto-Indo-European glottalic consonants 109 Thurneysen 1946: 50). The former reconstruction is more probable because -ach- seems to have resisted the voicing of voiceless fricatives between unstressed vowels (ibidem, 82). The form nócha ‘90’ was apparently modelled after trícho ‘30’, while cethorcho developed regularly from *kwetwrākont. For coíca ‘50’, Modern Irish caogad, I assume metathesis of *kōgexo to *kōxego and voicing of *xʄ after the syncope (ibidem, 80). The rounded vowel of the initial syllable is due to the original labiovelar environment, as it is in guidid < *gwhedhand gonaid < *gwhen- (cf. Cowgill 1980). Together with *kwetwrākont, these forms are the source of the final vowel in Latin -gintā, which was introduced on the analogy of vīgintī. Thus, the ultimate origin of the difference between Greek -a and Latin -ā is the different vocalization of the laryngeals in the two languages. After the disintegration of Italo-Celtic, the influence of ‘70’ and ‘90’ first affected *swekskont, which was preserved in Irish sesca, and *oktōkont, where Latin preserved -ō- and Celtic adopted *-mā-. The preservation of the difference between the short vowel of Old Irish fiche < *dwidkʄṃt- and the long vowel of trícho < *triH2dkʄomt- suggests that the *d was simply lost after a vowel in Celtic. If the length in Latin vīgintī is correctly attributed to the glottalic feature of the lost *d, it shows that the elimination of the glottalic obstruents was posterior to the disintegration of Italo-Celtic. Note that the difference between fiche and vīgintī corresponds with the difference between Old Irish recht ‘law’ and Latin rēctus, where the long vowel originated from Lachmann’s law. THE AEOLIC OPTATIVE 1. Despite considerable effort which has been spent on a variety of possible solutions to the problem (cf. especially Thomas 1957 and Forbes 1958, with a survey of the earlier literature), the origin of the so-called Aeolic optative has not been clarified: “Le problème reste posé” (Chantraine 1967: 266).1 I think that the absence of a convincing solution is the consequence of an imperfect understanding of the original, Proto-Indo-European state of affairs. In the following I intend to discuss a few points which, though relevant to the problem, have not received sufficient attention and to present an alternative solution. 2. Proto-Indo-European verbal paradigms could have either fixed or mobile stress. When the stress was fixed, as in the sigmatic aorist and the thematic flexion, the optative suffix was *-iH1, followed by the personal endings with no vowel intervening: 1st sg. 2nd sg. 3rd sg. 1st pl. 2nd pl. 3rd pl. -siH1m -siH1s -siH1t -siH1me -siH1te -siH1nt -oiH1m -oiH1s -oiH1t -oiH1me -oiH1te -oiH1nt Outside these two categories, I find no trace of an original paradigm with fixed stress in Greek. When the stress was mobile, the optative suffix was *-ieH1- in the singular and *-iH1- in the plural of the active voice, and *-iH1- in the middle voice. The stress was on the ending in the 1st and 2nd pl. forms of the mobile paradigms, and evidently also in the sg. forms of the middle voice, but not in the 3rd pl. forms, where a number of indications point to original root stress. First of all, the 3rd pl. active ending of the Vedic optative is -ur, not -an. The ending -ur is found in root presents with fixed stress, e.g. inj. takṣur of tákṣati ‘they fashion’, in reduplicated imperfects, e.g. ádadhur of dádhati ‘they put’, in the sigmatic aorist, which has -sur, and in root aorists of roots in a laryngeal, e.g. ádhur ‘they put’, i.e. in all those athematic forms where the stress is either on the root or on a preceding syllable. It follows that -ur replaces earlier *-at from syllabic *-nt. Secondly, the agreement between Latin velint, Gothic wileina and Old Church Slavic velętъ, which are all related to English will, suggests that the optative paradigm from which these forms are derived had an e-grade in the root. The oldest paradigm of the Slavic compound stem do-vьlje- ‘suffice’, which has a reduced grade in the root and is evidently based on the 3rd sg. form in *-ieH1t, has an irregular 3rd pl. form dovьlętъ, which must be derived from 112 Greek *-i(H1)nt. The same alternation is found in xošte- < *-tye- ‘want’, 3rd pl. xotętъ. It points to an original paradigm *ulieH1t, *ueliH1nt. Thirdly, the Vedic optative of the type dheyām ‘I may put’ requires an explanation. This form cannot have replaced *dheya(m) < *dheH1iH1m or *dhāyām < *dheH1ieH1m because neither of these forms is attested in the material while both are supported by other paradigms and would not therefore easily be lost, cf. gaméyam, gamés beside gamy¡s of gam- ‘go’. This suggests that the full grade of the root *dheH1- and the full grade of the suffix *-ieH1- were taken from different forms of the same paradigm, which means that the stress alternated between the root and the suffix. Since the suffix had full grade in the singular, the obvious source of the full grade root vowel is the 3rd pl. form dheyur, which is the expected reflex of *dheH1iH1nt. On the basis of these considerations I arrive at the following reconstruction of PIE paradigms for the present optative of the root *H1ei- ‘go’ and the aorist optative of the root *dheH1- ‘put’: 1st sg. 2nd sg. 3rd sg. 1st pl. 2nd pl. 3rd pl. H1iieH1m H1iieH1s H1iieH1t H1iiH1me H1iiH1te H1eiiH1nt dhH1ieH1m dhH1ieH1s dhH1ieH1t dhH1iH1me dhH1iH1te dheH1iH1nt In the middle voice, which will not be discussed here, I also assume full grade of the root in the 3rd pl. form and zero grade elsewhere (cf. K065). 3. What is the expected development of the reconstructed paradigms in Greek? This question hinges on the development of the laryngeals. In the position after a vowel and before a consonant, the laryngeals were apparently lost at an early stage with compensatory lengthening of the preceding vowel, cf. especially μείς < *mēns < *meH1ns ‘month’, where the accent points to an original monosyllable, and similarly acc.pl. -³ς, -άνς < *-āns < *-eH2ns, acc.sg. -³ν < *-ām < *-eH2m, also acc.sg. -ῡν < *-uHm, acc.pl. -ῡς < *-uHns, but acc.sg. -yan on the analogy of nom.sg. -ya < *-iH2, where the vocalization of the word-final laryngeal is regular.2 Thus, I assume that the phonetic reflex of 1st sg. *-siH1m and 3rd pl. *-siH1nt should be *-sīn. In the thematic flexion, *-oiH1- yielded *-oiy- (or *-oyy-) with vocalization of the following nasal in Arc. ἐξελαύνοια ‘I may drive out’ and Hom. ἑποίατο ‘they may follow’. Before the nonsyllabic consonant of the endings *-s, *-t, *-me, *-te, the sequence *-oiy- was evidently reduced to *-oi- (or *-oy-).3 It is clear from the 3rd sg. form in -οι that the assimilation of the laryngeal to the preceding semivowel was anterior to the loss of final *-t because the laryngeal would otherwise have been vocalized to yield -ε, as it was in ὄσσε < *H3ekwiH1 ‘(pair of) The Aeolic Optative 113 eyes’. However, the circumflex ending of κελεύοι ‘he may order’, where the accent was not retracted to the initial syllable, points to a disyllabic ending *-oyi or *-oī, suggesting that the assimilation of the laryngeal was anterior to the rise of a distinction between *i and *y. Thus, we arrive at the following relative chronology: (1) assimilation of a laryngeal before a final consonant (cluster) to a preceding (semi)vowel, (2) vocalization of the syllabic nasals and loss of final *-t, (3) reduction of *-oii- to *-oi- before a consonant, (4) rise of an opposition between *i and *y, (5) loss of the laryngeals in antevocalic and intervocalic position. The laryngeals of *H1i- and *dhH1- were lost after the vocalization of the following *i, cf. especially ἔπιον < *H1e-pH3iom ‘I drank’.4 In intervocalic position, the laryngeals were retained longer than elsewhere, as is clear from the circumflex tone which reflects the original disyllabic character of the resulting long vowels and diphthongs. This leads us to the following reconstruction of Proto-Greek paradigms: 1st sg. 2nd sg. 3rd sg. 1st pl. 2nd pl. 3rd pl. -sīn -sīs -sī -sīme -sīte -sīn -oiya -ois -oī -oime -oite -oiya(n) iyēn iyēs iyē iīme iīte eyīn thiēn thiēs thiē thīme thīte theīn The disyllabic character of *theīn is still preserved in τιθεῖεν < *titheī-en ‘they may put’, where the accent was not retracted to the initial syllable, cf. δύναιο, δύναισθε ‘you may be able’, which replace earlier *dunīso, *dunīsthe. 4. The 3rd pl. ending *-sīn, which was homophonous with 1st sg. *-sīn, was now replaced by *-seīn on the analogy of *-theīn, the ending of which was also found in the passive aorist and in the paradigm of ἵημι ‘let go’. This is the origin of the Aeolic optative. The ending *-seīn was subsequently replaced by *-seiyan on the analogy of the thematic ending *-oiyan. This replacement accounts for the retraction of the accent in λύσειαν ‘they may loosen’ in accordance with the limitation law, as compared with τιθεῖεν. The ending *-seiyan then gave rise to the 3rd sg. ending *-seiye on the analogy of the indicative, cf. ἔλυσε, ἔλυσαν ‘he, they loosened’, also 2nd sg. *-seiyas. In the 1st and 2nd pl. forms, however, the model of ἐλύσαμεν, ἐλύσατε yielded λύσαιμεν, λύσαιτε on the analogy of the thematic endings. The latter analogy did not work in the 3rd sg. form, where the indicative ending was -ε. Thus, the distribution of -αι- and -ει- is ultimately based on the spread of *-eifrom the 3rd pl. form on the one hand and the absence of -α- from the 3rd sg. indicative form on the other. 114 Greek In the thematic flexion, the isolated 1st sg. ending *-ya was replaced by the usual athematic ending *-mi, e.g. λύοιμι ‘I may loosen’, which then gave rise to the analogical form λύσαιμι. The substitution of -εν for *-an in the 3rd pl. ending must have taken place at a time when *-en < * -ent had not yet been replaced by -ον, -αν or -σαν in the indicative, as in Hom. ζεύγνυον, ζεύγνυσαν ‘they yoked’, ἤιον ‘they went’, Skt. ¡yan < *-H1ient. In the original paradigm with mobile stress, the full grade of the 3rd pl. form spread to the other forms of the paradigm, e.g. τιθείης, τιθεῖτε < *titheiyēs, *titheīte ‘you may put’. This development is analogous to the rise of Skt. dheyām. The 3rd pl. ending *-īn was replaced by *-iyen on the basis of the indicative paradigm, e.g. τιθεῖεν < *titheiyen, also Delphi περιιειεν < *-i-eiyen ‘they may go round’, Hom. ἰείη < *i-eiyē ‘he may go’. Similarly, εἰδείη, εἰδεῖεν < *weideiyē, *weideiyen ‘he, they may know’ represent *uidieH1t (Skt. vidy¡t), *ueidiH1nt (with original full grade in the root) plus *-eī- from *theīn and 3rd pl. -εν < *-ent. The Cretan forms δικακσιε ‘he may judge’, κοσμησιε ‘he may arrange’ (Dreros), ϝερκσιεν ‘they may perform’, διαλυσιαν ‘they may dissolve’ (Gortyn) are apparently built on the zero grade of the suffix *-s-iH1-. It is highly improbable that the singular forms represent *-yē- because there is no trace of the full grade suffix in the sigmatic aorist, which had fixed stress from the outset. These forms rather represent a variety of the Aeolic optative with generalization of the zero grade *-ī- instead of the 3rd pl. vocalism *-eī-. The endings -σιε, -σιαν suggest that we have to reconstruct a real Aeolic optative (-σειε, -σειαν) with *-ī- not yet replaced by -αι- in the 1st and 2nd person forms and subsequently generalized throughout the paradigm. The form ϝερκσιεν adopted -εν from the other optative paradigms, and the eventual substitution of -αι- for *-ī- is clear from the forms ϝερκσαι ‘he may perform’, ρηκσαιεν ‘they may break’ (Gortyn). It appears that Cretan lagged behind in a development of the optative which was the same as in the other dialects. NOTES Cf. Chantraine’s footnote “On mesurera dans ces articles l’extrême complication de toutes les solutions proposées.” Ri’xs conception of “-ī/iâ- umgebildet [...] zu -eââa/e- mit den Ind.-Ausgangen und Dissimilation -iâ- > -eâ(â)-, nur als Variante in der 2. 3. Sg. 3. Pl.” (1976: 233) stretches the imagination and does not explain the distribution of the e-grade. Cf. also Risch (1982: 32829): “Nicht eindeutig geklärt ist noch immer die Herkunft des sog. ‘äolischen’ Optativs, z. B. δείξειας, -ειε, -ειαν.” 1 Professor Ruijgh draws my attention to γλωχÕς ‘point’, which is based on the original acc.sg. form *glōkhīn of γλῶσσα < *-iH2 ‘tongue’. 2 The Aeolic Optative 115 The e-grade of δέατο ‘he seemed’ suggests that this form represents an original stative *deiH2-o, cf. κεῖται ‘he lies’, κρέμαται ‘he hangs’, 3rd pl. *deiH2ento (replacing earlier *-ro), cf. Hittite kitta(ri) ‘he lies’, kiyanta(ri) ‘they lie’. The 3rd pl. ending *-ento was regular in the middle root aorist, e.g. Skt. kranta ‘they made’, ranta ‘they went’ (cf. K065: 220), also θέντο < *dhH1ento ‘they put’. 3 Cf. also *ἵᾱ < *sH2ieH2 ‘strap’ (Ruijgh 1967: 205) and ὑγιής < *-gwH3iēs ‘healthy’, βίος < *gwwiH3os < *gwH3iuos ‘life’, ἐβίων < *-gwwiH3eH1m < *-gwH3iueH1m ‘I lived’ (cf. SCr. žívjeti, OPr. giwīt), βέομαι < *gwweiH3omH2- ‘I will live’ where the *-wapparently inhibited the palatalization of the preceding labiovelar. The word ζωός < *gwyōwos ‘alive’ and its derivatives contain a secondary full grade which replaces the original zero grade of *gwīwos < *gwH3iuos, Skt. jīvás, where the Balto-Slavic and Celtic evidence shows that the laryngeal preceded the *i, e.g. Latvian dzîvs (with broken tone reflecting preservation of final stress), Welsh byw (with a short root vowel), cf. K014: 76-82. The verb λοέω < *lowesō adopted the root vowel of λόω < *lowō ‘I wash’, which replaces athematic *loumi < *leH3umi, where the phonetic loss of *H3 before *u in such forms as 3rd pl. *lH3uenti led to confusion with the paradigm of λύω ‘I loosen’. The root vowel of λόω cannot represent a vocalized laryngeal because in that case there would be no motivation for the rise of the secondary full grade which is attested in Myc. re-wo-to-ro-ko-wo, metathesized in Hom. λοετροχόος ‘bath-pourer’. The rise of *lewo- may be due to the influence of the quasi-synonymous root χεϝ- ‘pour’, as Professor Ruijgh suggests to me. Note that ταναός < *tṇH2euos and Πλάταια < *pḷtH2euiH2 are no counterexamples to the loss of a laryngeal before a vocalized semivowel, cf. also Breton tanao < *tanawos ‘thin’. 4 THE GREEK 3RD PL. ENDINGS The Vedic 3rd pl. active ending is -ur instead of -an in the following instances: (1) root presents with fixed stress, e.g. inj. takṣur of tákṣati ‘they fashion’; (2) reduplicated imperfects, e.g. ádadhur of dádhati ‘they put’; (3) sigmatic aorist -sur; (4) root aorists of roots in a laryngeal, e.g. ádhur ‘they put’; (5) optative -yur. Besides, the ending -ur is found in the perfect. Elsewhere (K065, cf. also K097) I have argued that -ur replaced earlier *-at from syllabic *-nt in those athematic forms where the stress was either on the root or on a preceding syllable. It follows that inj. dhúr adopted the vocalism of ind. ádhur, whereas the converse substitution took place in ind. ávran ‘they covered’, inj. vrán. This view is supported by the apophonic difference between 3rd pl. indicative and injunctive forms in the middle root aorist: ákrata ‘they made’, ¡rata ‘they went’ versus kranta, ranta (cf. Meillet 1920: 203, 205). It also provides an explanation for the remarkable 3rd pl. middle subjunctive ending -anta, which must originally have been the inj. ending corresponding to ind. -ata from syllabic *-nto. This ending was evidently reinterpreted as a subjunctive because it differed from the other inj. endings by the presence of an initial vowel. What is the expected distribution of full and zero grade 3rd pl. endings in Greek? In the thematic flexion we expect -ον, -οντι (-ουσι), -οντο, -οντοι (-ονται). Outside the thematic flexion, the sigmatic aorist, and the original stative, I find no evidence for an original paradigm with fixed stress in Greek. I therefore expect zero grade 3rd pl. endings in the sigmatic aorist, reduplicated and augmented forms, the optative, and the original stative, and e-grade endings in unreduplicated athematic presents and augmentless root aorist forms. I claim that the actual distribution is closer to this expectation than is usually assumed. In the following, page numbers will refer to the discussion of the 3rd pl. forms by Risch (1982). The ending -εντι (-ενσι) is attested in Myc. e-e-si /ehensi/ ‘they are’, contracted in Ionic εἰσί and West Greek ἐντί (324), also in Myc. ki-ti-je-si /ktiyensi/ ‘they cultivate’, Vedic sánti, kṣiyánti. Risch puts the middle verbs κει‘lie’, ἡσ- ‘sit’ and ϝεσ- ‘wear’ in the same category, which is inappropriate because these represent original statives with a root-stressed 3rd pl. form, e.g. κέαται, εἵατο, which are in perfect agreement with the corresponding Vedic forms. The original zero grade ending -ατι (-ασι) from syllabic *-nti is well preserved in the perfect, e.g. Hom. πεφύκασι ‘they have grown’. Elsewhere we find -αν and -αντι (-ᾱσι) for the zero grade active endings. 118 Greek Turning now to the reduplicated presents and the root aorists, we find the zero grade 3rd pl. endings in Attic τιθέᾱσι, διδόᾱσι, Arcadian συνέθεαν, Cyprian ka-te-ti-ya-ne /katethiyan/, Boeotian and Locrian ἀνέθεαν, also Arc. imp. ὐνθεάντω beside καθέντω, Elean opt. ἐπιθεῖαν, γνοῖαν (325). These forms have not received an adequate explanation. Risch’s comment deserves full quotation (327): “Die Tatsache, daß der Typus ἔθεαν, ἔδοαν sowohl im Arkadischen als auch im Kyprischen gilt, spricht m.E. eindeutig dafür, daß er wenigstens bei dieser Dialektgruppe alt ist. Auch das Böotische ist ein Dialekt, der neben verschiedenen Neuerungen doch manches Alte bewahrt hat, so z.B. die Endung -φι in ἐπιπατρόφιον, pronominale Formen mit -νί in προτηνί ‘früher’, Patronymika auf -ιος. Also darf ἔθεαν auch im äolischen Bereich eventuell als alt betrachtet werden. Dafür spricht auch das Zeugnis des homerischen Gebrauchs. Bekanntlich haben wir hier bei den langvokalischen Aoristen sowohl ἔσταν als auch ἔστησαν, sowohl ἔφανεν als auch ἐφάνησαν usw., wobei die längere Form typisch ionisch-attisch ist und die kürzere vermutlich dem Äolischen zugewiesen werden kann. Ebenso ἔφαν und ἔφασαν und bei -νῡμι ἐζεύγνυον und ἐζεύγνυσαν. Um so überraschender ist, daß nur ἔθεσαν, ἔδοσαν, ἐδίδοσαν, nie ἔθεν, ἔδον, ἔδιδον vorkommen. Der Schluß liegt nahe, daß auch das vorhomerische Äolisch sie nicht kannte, sondern vermutlich ἔθεαν, ἔδοαν (oder *ἔδοεν?) hatte, die dann ohne weiteres durch metrisch gleichwertiges ἔθεσαν, ἔδοσαν ersetzt werden konnten.” In view of this, the West Greek forms τίθεντι, ἔτιθεν, ἔθεν, δίδοντι, ἔδιδον, ἔδον can hardly be ancient. They are easily explained as the result of a secondary development, while their replacement by forms in -αντι (-ᾱσι) and -αν in the other dialects can hardly be motivated. Risch thinks that the Attic forms τιθέᾱσι, διδόᾱσι, ἱᾶσι, ἱστᾶσι are recent (similarly Chantraine 1973: 471 and Rix 1976: 252) because we find Myc. di-do-si /didonsi/, -i-je-si /-hiyensi/, and Hom. τιθεῖσι, ἱεῖσι, διδοῦσι, ἱστᾶσι, where he attributes the accent to the influence of ἑστᾶσι, τεθνᾶσι from -άᾱσι (329). The problem is that the perfect provides a small and none too obvious basis for such an analogical development. Moreover, the ending -ατι (-ασι) from syllabic *-nti must have been introduced into the perfect from the reduplicated present. It seems much more plausible that the zero grade ending was never eliminated from the reduplicated present in Attic and was replaced by the regular athematic ending -εντι (-εισι) in Ionic. Thus, we may surmise that τιθεῖσι, διδοῦσι represent earlier *τιθέενσι, *διδόενσι (West Greek τίθεντι may actually represent *τιθέεντι). Since neither the replacement of -ανσι by -ενσι, nor the converse replacement can easily be motivated, we may conjecture that the two endings were alternative solutions for the elimination of an irregular ending. In this way I arrive at the tentative reconstruction of the 3rd pl. forms *τιθέᾰσι, *διδόᾰσι for Attic-Ionic. The Greek 3rd pl. endings 119 The situation in Aeolic and Arcado-Cyprian is different. Though the evidence of these dialects points to the aorist forms ἔθεαν, ἔδοαν, it appears that the present forms were replaced by τίθεντι, δίδοντι, as they were in West Greek (where the substitution may have taken place independently at a more recent stage). First, Myc. di-do-si and -i-je-si seem to reflect /didonsi/ and /hiyensi/. Second, Risch has called attention to Arc. present ποίενσι beside imperfect παρεκάλεαν (329). This can hardly be a recent distribution. Third, there is evidence for *τίθεισι beside τιθεῖσι in the Hom. imperfect (329): “Es ist vielleicht auch nicht zufällig, daß bei Homer zwar keine Aoriste vom Typus ἔθεν, wohl aber einige Imperfekta dieser Art, nämlich ἵεν Μ 33, ξύνιεν Α 273, μέθιεν φ 377 (stets vor der Hephthemimeres-Zäsur), dazu als Lesung Aristarchs πρότιθεν α 112 (cod. προτίθεντο) neben sonstigen τίθεσαν, δίδοσαν, ἵστασαν bezeugt sind.” These may be Aeolic forms. Fourth, the remarkable spread of the secondary ending -εν in Thessalian requires an explanation. The raising of a to e before and after i in Thess. -ει, διέ for -αι, διά does not suffice to explain the rise of the ending -αεν for -αν, e.g. ὀνεθείκαεν beside ὀνέθεικαν. The spread of -εν suggests the earlier replacement of *ἐτίθεαν by ἔτιθεν, which presupposes the replacement of *τιθέατι by τίθεντι, as in West Greek. Putting the evidence together, I tentatively reconstruct the following distribution: 3rd pl. West Greek Aeolic Arcado-Cyprian Attic-Ionic present τίθεντι τίθεντι τίθενσι τιθέᾰσι imperfect ἔτιθεν ἔτιθεν ἐτίθεαν ἐτίθεαν aorist ἔθεν ἔθεαν ἔθεαν ἔθεαν This scheme does not necessarily reflect a single chronological layer. In particular, the Aeolic state of affairs may be a younger development of the one posited for Arcado-Cyprian. The West Greek simplification may have taken place at any stage; the Elean optative forms in -αν, e.g. ἐπιθεῖαν, suggest that it was a recent development. In any case, the Attic-Ionic distribution is evidently archaic and the cleavage between this dialect and Arcado-Cyprian must be ancient. We now come to the discrepancy between ἔβαν and ἔσταν on the one hand, and ἔθεαν and ἔδοαν on the other. There can be little doubt that the disyllabic forms are secondary. If they replace earlier ἔθεν and ἔδον, it is hard to see a motivation for the introduction of the new ending. Such a motivation is even more difficult to find in the case of *τιθέᾰσι and *διδόᾰσι if these replace earlier τίθεντι and δίδοντι. Ι therefore think that we have to start from *ἔθαν, *ἔδαν, *τίθατι, *δίδατι, where the introduction of the root vowel is a natural development. It follows that after a consonant the laryngeals were lost without a 120 Greek trace before a syllabic nasal, which was regularly vocalized to a, in spite of the fact that word-initial laryngeals were vocalized before a tautosyllabic nasal. This brings us to a reconsideration of the nt-participle. Since Beekes’ discussion of Latin iens (1985: 67-71) we have to start from nom. *dheH1nts, acc. *dhH1entm, gen. *dhH1ntos, which now yields Greek θείς, θέντα, *θατός. The oblique stem may be reflected in Arc. ἀπυδόας, Elean ἀνταποδιδῶσσα from *-διδόασσα (325). The original participle of ἔτλην is preserved in τάλᾱς ‘wretched’, which represents *telH2nts, *tlH2(e)nt-, while the 3rd pl. form ἔτλαν evidently replaces *ἔταλαν. Similarly, I think that ἔγνον replaces *ἔγαναν from *-gnH3nt and cannot be used as evidence for a root aorist with fixed stress. After a consonant, the nom.sg. ending *-as from zero grade *-nts was replaced by *-ōn(t), e.g. ἑκών ‘willing’, evidently because polysyllabic consonant stems usually had an asigmatic nominative in Greek. Thus, the ending of ἐών, ἰών, βαλών (not -είς) corroborates the reconstruction *H1esnts, *H1eints, *gwelH1nts, as opposed to δούς, γνούς from *deH3nts, *gneH3nts, cf. also δράκων beside δρακείς reflecting *derknts, *drkentm. As I pointed out above, the Vedic evidence leads us to expect e-grade 3rd pl. endings in augmentless and unreduplicated forms, as opposed to zero grade endings in the forms which have just been discussed. This is actually what we find in the imperative θέντων, δόντων; the -a- of Arc. ὐνθεάντω beside καθέντω must have been taken from the indicative. The e-grade ending is also found in the optative θεῖεν, as opposed to -σειαν in the sigmatic aorist (cf. K097). There seem to be traces of the original distribution in the middle voice as well, cf. Hom. ξύμβληντο for *-βάλεντο, πλῆντο for *πάλεντο and *πάλαντο, but βεβλήαται and βεβλήατο for *βεβάλατο from *-gwlH1nto. The 3rd sg. form ἐπρίατο ‘bought’ for *ἔπρῑτο must be based on augmentless πρίαντο from *kwriH2ento, similarly δίενται ‘they hasten’ from *diH1ento, also ἔπιον ‘they drank’ for *ἔπῑν on the basis of *pH3ient, and κέονται ‘they lie’. The e-grade endings were largely replaced by the thematic o-grade endings outside the optative, where the model for this substitution was lacking. Thus, we find -o- for e-grade 3rd pl. endings in τανύουσι, ζεύγνυον, ὤμνυον, ὄρνυον, καταείνυον, τανύοντο beside the usual zero grade in δεικνύᾱσι, ζεύγνυσαν, δαίνυνται, ῥήγνυντο. The accent of δαμνᾶσι, ῥηγνῦσι may reflect earlier *-νάενσι, *-νύενσι, as in the case of τιθεῖσι. It appears that Greek preserved the original PIE alternations more faithfully than is generally assumed. GLOTTALIC CONSONANTS IN SINDHI AND PROTO-INDO-EUROPEAN 1. In 1973 Gamkrelidze and Ivanov suggested on typological grounds that the reconstructed voiced occlusives of the Indo-European proto-language were actually glottalic. Elsewhere I have argued that this hypothesis is supported by comparative evidence from Baltic (K025) and Armenian (K031) and offers an explanation for the lengthening in Latin āctus, lēctus (Lachmann’s law) and the devoicing in Sanskrit yuktáḥ < *-gt-, vásuttiḥ < *-dHt-, as opposed to dugdháḥ < *-ght-, vásudhitiḥ < *-dhHt- (K032). I have not discussed the Sindhi material, which will be the subject of the present article. 2. Apart from the usual obstruents which are found in the other Indo-Aryan languages (voiced and voiceless, aspirated and unaspirated), Sindhi possesses a series of voiced implosive stops, which will be denoted by ’b, ’ḍ, ’j, ’g in this article (on the phonetics see Nihalani 1974). There is no dental implosive stop. The phonemic status of the implosives is beyond doubt, cf. bakhu ‘sacrifice of a goat’, ’bakhu ‘embrace’, ḍiṭhō ‘bold’, ’ḍiṭhō ‘seen’ jaü ‘barley’, ’jaü ‘lac’, gacu ‘mortar’, ’gacu ‘much’. These examples are taken from Turner 1924, who gives a few dozens of minimal pairs; his article serves as a basis for the following analysis. The topic of this article is their historical origin. 3. As Turner points out, “a detailed examination of the words in which they occur shows that initially ’g ’j ’ḍ ’b correspond to initial g-, j- (dy-), d-, b- (dv-) in Sanskrit, and intervocalically to consonant groups in Sanskrit that in Prakrit became -gg- -jj- -ḍḍ- (-dd-) -bb- (-vv-), while the simple voiced stops in Sindhī, g j d b, are the result of special conditions and in most cases (where not occurring in loan-words from other languages) are descendants of Primitive Indian sounds other than g j d b. The simple voiced dental d, except in the group nd, is found only in loan-words. That is to say, except for certain specific conditions, Sindhī has shifted Primitive Indian g-, j-, d-, b- to ’g ’j ’ḍ ’b” (1924: 305 = 1975: 196). 4. The rise of the glottalic articulation which is characteristic of the Sindhi implosives can hardly be attributed to external influence because the neighbouring Dravidian (Brahui), Iranian (Baloči), and Indo-Aryan languages do not present anything comparable, with the exception of the westernmost dialect of the closely related Lahnda language. The southern (Kacchi) and eastern (Thareli) dialects of Sindhi do not have implosives. On the other hand, it is difficult to perceive an internal motivation for the rise of the glottalic consonants, which are not found elsewhere in the Indo-Aryan linguistic area. Since the Sanskrit reflexes of the sounds for which Proto-Indo-European glottalic articulation has been assumed are reflected as implosives in Sindhi, e.g. 122 Indo-Iranian ’ḍiaṇu ‘to give’ (Indo-European *dō-), we must consider the possibility that the latter language has preserved an archaism which was lost elsewhere. 5. As Turner points out, “the simple voiced stops – g j ḍ b – can result from the disaspiration of the corresponding aspirated voiced stops – gh jh ḍh bh. A voiced aspirate when followed by an aspirate or by h in the same word lost its aspiration and became the corresponding simple voiced stop without glottal closure. It makes no difference whether h represents a Sanskrit intervocalic sibilant or has been inserted simply to avoid hiatus” (1924: 311f. = 1975: 202). Examples: bathī ‘quiver’ (Skt. bhástrā), ḍiṭhō ‘bold’ < *ḍhiṭhō (Skt. dhṛṣṭá-), gāhu ‘fodder’ (Skt. ghāsá-), cf. ’gāhu ‘bait’ < grāsa-. It follows that the rise of the implosives was anterior to the dissimilation of voiced aspirates before another aspirate or h. 6. A surd stop preceded by a nasal becomes the corresponding voiced stop, which remains without glottal closure. The original voiced stops preceded by a nasal are completely assimilated to the preceding nasal. In the group stop + r, the r is lost, except when the stop is a dental. In the latter case, the dental becomes a cerebral and the r (except in South Sindhi) remains. Even when the resultant cerebral is voiced, there is no glottal closure, e.g. ḍrākha ‘a small grape’ (drākṣā), āṇḍrō ‘entrails’ (āntrá-), caṇḍru ‘moon’ (candrá-). Thus, where the voiced stop is preceded by a nasal or followed by r there is no glottal closure. These clusters do not offer a basis for a relative chronology because it cannot, at this point, be decided whether the glottal closure arose in ’gāhu ‘bait’ after the loss of r in grāsa-, or was lost secondarily in ḍrākha. 7. Initial vy- became w- in Sindhi, e.g. wāghu ‘tiger’ (Skt. vyāghrá-), but intervocalic -vy- became -b- without glottal closure, e.g. sibaṇu ‘to sew’ (sãvyati), katabu ‘business’ (kártavya-). It follows that the rise of the intervocalic implosive -’b- from -br-, -rb-, and -dv- was anterior to the development of -vy- into a bilabial stop. The rise of ’b from initial and intervocalic dv and the rise of ’j from intervocalic dy do not offer a basis for a relative chronology because the glottal closure may have been a feature of the dental obstruent before the sound change already, e.g. ’ba ‘two’, ’bījō ‘second’ (Skt. dv¡, dvitãya-), u’baṭaṇu ‘perfumed flour to rub the body with’ (udvartana-), a’ju ‘today’ (adyá). 8. Sindhi, like Kashmiri and Sinhalese, keeps Skt. j and y apart. The former became ’j, the latter j, e.g. jō, jā ‘who’ (yáḥ, y¡), jō ‘because’ (yátaḥ), jaṇyō ‘sacred cord’ (yajñopavītá-), jāhu ‘coitus’ (yābha-). It follows that the rise of the implosive ’j was anterior to the development of initial y into a stop. In the same way, Skt. -īya- (Pali -īya- or -iyya-) is reflected in Sindhi as -ija-, while Skt. -idyais reflected as Sindhi -i’ja-, e.g. ’ḍijaṇu ‘to be given’ (dīyáte, Pali diyyati, Prakrit dijjaï), chi’jaṇu ‘to be broken’ (chidyate). Cf. also bhāṇējō ‘sister’s son’ (bhāgineyaka-, Pali bhāgineyya-), māṭrējō ‘belonging to a stepmother’ (*mātreya-, Pali matteyya-), sēja ‘couch’ (śayyā, Prakrit sejjā), as opposed to Glottalic consonants in Sindhi and Proto-Indo-European 123 bha’jaṇu ‘to be broken’ (bhajyáte), upa’jaṇu ‘to be produced’ (utpadyate), wē’ju ‘doctor’ (vaidya-). Thus, I assume that -j- is the regular reflex of intervocalic -yyin Sindhi. 9. The treatment of -ry- and -rv- differs from that of -yy- and -vy- in the presence versus absence of glottal closure, e.g. kā’ju ‘ceremony, work’ (kāryà-), ca’baṇu ‘to chew’ (carvati), sēja ‘couch’ (śayyā), katabu ‘business’ (kártavya-). We have to conclude that r was not assimilated to a following y or v, but developed into glottal closure. I think that -ry- and -rv- became -dy- and -dv- at an early stage and subsequently shared the regular development of these clusters in Sindhi. They do not offer a basis for a relative chronology because the glottal feature may or may not have been present in the original clusters -dy- and -dv- at the time when the obstruent replaced the alveolar or retroflex flap in -ry- and -rv-. 10. The facts which have been discussed so far do not inhibit the identification of Sindhi and Proto-Indo-European glottalization. In particular, Sindhi may have preserved the glottal closure which has been surmised as a feature of the PIE voiced occlusives except for certain specific environments. There are a number of instances where the identification is impossible, however. First of all, the unaspirated voiced stops which came into being as a result of Grassmann’s law are reflected as implosives in Sindhi, e.g. ’badhō ‘bound’ (baddhá-), ’ḍahī ‘curds’ (dádhi), ’jaṅgha ‘leg’ (jáṅghā), ’ginhaṇu ‘to buy’ (gṛhn¡ti). If the identification of Sindhi and PIE glottalization is to be upheld, we must assume that the initial consonant of these words became glottalic as a result of the aspiration dissimilation. This is no serious objection because there was no series of unaspirated voiced stops without glottal closure at the time of the dissimilation if PIE glottalization had not yet been lost at that stage. 11. Besides, there is an implosive in the words pē’ju ‘drinking’, pē’ja, pē’jī ‘ricewater’, pē’ju, pē’jō ‘watering land after sowing’ (Pali peyya-). The implosive cannot continue -yy-, which yielded -j- (see above). I think that the glottal closure is the reflex of the laryngeal in the Indo-European root *poH(i)- ‘drink’. The preservation of the laryngeal before the semivowel in this word family is a consequence of its particular root structure. The development of the laryngeal into glottal closure gave rise to a marginal phoneme -’y- in pre-Sindhi, which was to merge with -’j- from -dy- and -jy-. A similar merger of laryngeals with the glottal closure of the PIE voiced occlusives took place in Baltic and Slavic (cf. K025). 12. The hypothesis that the laryngeal was preserved in the position before a semivowel up to a stage which was posterior to the rise of the Sindhi implosives is supported by the absence of glottal closure in the words jīu ‘living being’ (jīvá), jiarō ‘alive’ (jīvalá-), jiaṇu ‘to live’ (jãvati). Turner attributes the absence of 124 Indo-Iranian glottal closure in these words to the following -ī- (1924: 309 = 1975: 199). This is unsatisfactory because there is an initial implosive in ’jibha ‘tongue’ (jihv¡) and it is not clear why the length of the following vowel should have caused the loss of the initial glottal closure. I assume that the latter was lost by dissimilation before the glottal closure which had developed from the laryngeal of Indo-European *gwiH(u)- ‘live’, which had been preserved before the semivowel and was subsequently lost. Several conjectures can be made in connection with jīrō ‘cummin-seed’ (jīraka-), none of which seems to be demonstrable. 13. Moreover, there is an implosive in su’jaṇu ‘to be heard’ (śrūyáte, Pali suyyati), su’jaṇu ‘to be swollen’ (śūyate), u’jaṇu ‘to be woven’ (ūyate). Here again I assume earlier -’y-, where the glottal closure continues a laryngeal which was original in seṭ verbs and spread to aniṭ verbs in the passive. In nijaṇu ‘to be carried away’ (nīyáte) etc. the glottal closure was apparently lost between -i- and -y- at an early stage. 14. One may wonder if it is probable a priori that an archaic feature which was lost elsewhere in the Indo-Aryan linguistic area should be preserved in Sindhi. It must be remembered that Sindhi belongs to the more conservative dialects, e.g. in the preservation of the distinction between Skt. j and y, which it shares with Kashmiri and Sinhalese, and in the preservation of r after a dental obstruent. Since there are indubitable traces of early dialectal diversity in the Rgveda (cf. Emeneau 1966), the possibility that Sindhi has preserved an archaism which was lost elsewhere cannot be rejected a priori. The original Indian script does not provide the means of distinguishing between glottalic and plain voiced obstruents, which became phonemically relevant only as a result of the Sindhi dissimilation of voiced aspirates before another aspirate or h. 15. In any case, the Sindhi material cannot be adduced in support of “the general diachronic hypothesis that at least one source of injectives might be a sound shift from voiced plain to voiced implosive stops” (Greenberg 1970: 134). The comparative evidence points the other way. ARCHAIC ABLAUT PATTERNS IN THE VEDIC VERB 1. The 1st sg. active form of the Vedic sigmatic aorist injunctive does not take vṛddhi. This is a remarkable archaism which has not been sufficiently appreciated. 2. In his article on the “proterodynamic” root present, Insler calls attention to the fact that “the system of proterodynamic present inflection reflected in Vedic forms is nearly identical to the oldest system of Vedic sigmatic aorist inflection” (1972: 56). “It is only when we compare the act. indic.-inj. of proterodynamic root presents that the complete parallelism breaks down” (Insler 1972: 57). The active forms of the sigmatic aorist have lengthened grade vocalism throughout the whole paradigm and do not show the expected alternation between lengthened grade in the singular and full grade in the plural which is found in tāṣṭi, tákṣati. We must therefore ask the question: “which paradigm seems to continue the original ablaut relationship?” (Insler 1972: 58). 3. The obvious explanation is that the active paradigm of the sigmatic aorist “has participated in the same sort of leveling of vocalism observed in act. root aorists of the type ákar, ákarma, ákarta” (Insler 1972: 58). Insler rejects this view because the lengthened grade vocalism was extended to the 3rd pl. form of the sigmatic aorist, whereas the corresponding form of the root aorist maintains the original zero grade, e.g. ákran. The argument does not hold because the ending of the root aorist was -an < *-ent, whereas the sigmatic form ended in *-sat < *-snt. The ending *-at was replaced with -ur, as it was in the injunctive takṣur and in the reduplicated imperfect. The retention of the ablaut contrast in the paradigm of tāṣṭi and the extension of the lengthened grade to the 3rd pl. form of the sigmatic aorist fit “the general tendencies of the Vedic verb system to characterize act. athematic present inflection by ablaut differences, but to mark act. athematic aorist inflection by the predominant absence of any alternating vocalism” (Insler 1972: 61). 4. Lengthened grade vocalism was generalized in the active paradigm of the sigmatic aorist indicative, but not in the injunctive, which betrays the original distribution of the ablaut grades. It is noteworthy that the original distribution was already indicated by Wackernagel in his Old Indic grammar (1896: 68): the lengthened grade spread from the monosyllabic 2nd and 3rd sg. forms to the rest of the paradigm. The archaic character of this distribution is supported by the Balto-Slavic evidence (cf. K014: 84-86 and K064: 114-117). It is also clear from the Vedic material. 126 Indo-Iranian 5. The 1st sg. indicative has lengthened grade in RV ajaiṣam, aprākṣam, abhārṣam, ayāṃsam, aspārṣam, ahārṣam, akāniṣam, akāriṣam, acāriṣam, asāniṣam, and ambiguous vocalism in ayāsam. It has full grade in akramiṣam and in the analogic forms akramīm and aśaṃsiṣam. The 1st sg. injunctive has full grade in VS TS TB jeṣam, TS KS TB JB yoṣam, and RV stoṣam, vadhīm, and lengthened grade in the analogic form rāviṣam (ru- ‘break’). 6. Following Hoffmann, Narten interprets jeṣam and 1st pl. RV jeṣma as precative forms (1964: 120). The reason for this interpretation is evidently the absence of lengthened grade (cf. Hoffmann 1967a: 254). The functional evidence for the interpretation as precative (Hoffmann 1967b: 32f.) or subjunctive (Insler 1975b: 1526) is very weak, while the formal objections against it are prohibitive. It is therefore preferable to retain the traditional view that these forms are what they look like: full grade injunctive forms, which were interchangeable with the corresponding subjunctive in certain contexts and which could be interpreted as precative when the latter category became common. 7. Narten assumes that the injunctive forms yoṣam and stoṣam took their vocalism from the subjunctive (1964: 213, 277). The model for this analogic development is lacking, however, because the subjunctive ending was -āni, not -am. Hoffmann attributes the alleged substitution of the injunctive ending -am for the earlier subjunctive ending -ā to the influence of the 2nd sg. imperative: “Das Bestreben, den Konjunktivausgang -ā von dem durch Auslautsdehnung gleichlautend gewordenen Imperativausgang zu sondern, hat das Ausweichen zu -am, wodurch die 1. Person deutlich gekennzeichnet wurde, gefördert” (1967a: 248). I find such influence highly improbable. The use of the 1st sg. injunctive for the subjunctive must be explained from the meaning of the forms. Note that Standard British English offers an exact parallel in the use of ‘I shall’ where other persons ‘will’. During my stay in Dublin, Dr Patrick Sims-Williams told me that when an Irish friend asked him in front of an open door: “Will I go first?”, the only reasonable answer to him would be: “I don’t know”. Compare in this connection RV 7.86.2 kad¡ nv àntár váruṇe bhuvāni ... kad¡ mṛḷīkáṃ sumánā abhí khyam ‘When will I be inside Varuna? When shall I, cheerful, perceive his mercy?’ Also 10.27.1 ásat sú me jaritaḥ s¡bhivegó, yát sunvaté yájamānāya śíkṣam ‘That will be my excitement, singer, that I shall be helpful to the pressing sacrificer’. In 10.28.5 kath¡ ta etád ahám ¡ ciketam ‘How shall I understand this (word) of yours?’ the substitution of the subjunctive for the injunctive would yield a quite different shade of meaning: it would shift the responsibility from the singer to Indra. 8. The indicative has lengthened grade in RV 3rd du. asvārṣṭām, 1st pl. ajaiṣma, abhaiṣma, atāriṣma, 2nd pl. achānta, 3rd pl. achāntsur, abhaiṣur, atāriṣur, apāviṣur, amādiṣur, arāṇiṣur, arāviṣur, avādiṣur, asāviṣur, and Archaic ablaut patterns in the Vedic verb 127 ambiguous vocalism in ayāsur, arājiṣur, and āviṣur. It has full grade in 3rd du. amanthiṣṭām, 1st pl. agrabhīṣma, 3rd pl. atakṣiṣur, adhanviṣur, anartiṣur, amandiṣur, all of which have a root in a double consonant (cf. gṛbhītá- < *gṛbhH-ita-). It has zero grade in amatsur, anindiṣur, and ākṣiṣur (naś- ‘attain’). 9. The injunctive has full grade in Rgveda 2nd du. aviṣṭam, kramiṣṭam, gamiṣṭam, caniṣṭam, cayiṣṭam, mardhiṣṭam, yodhiṣṭam, vadhiṣṭam, śnathiṣṭam, 3rd du. aviṣṭām, 1st pl. jeṣma, śramiṣma, 2nd pl. aviṣṭa(na), grabhīṣṭa, raṇiṣṭana, vadhiṣṭa(na), śnathiṣṭana, zero grade in hiṃsiṣṭa, and ambiguous vocalism in 3rd pl. dhāsur, hāsur. It has lengthened grade in 2nd du. yauṣṭam (ĀpMB yoṣṭam), tāriṣṭam, 2nd pl. naiṣṭa (ĀpŚS yoṣṭa), 3rd pl. yauṣur, jāriṣur, and in the analogic form 2nd du. yāviṣṭam (yu- ‘unite’). Note that the difference between 1st pl. śramiṣma and atāriṣma parallels the one between jeṣma and ajaiṣma. 10. One may wonder if the ablaut difference between the indicative and the injunctive is also found in the asigmatic aorist. It has long been noticed that the 3rd pl. middle indicative forms ákrata and ārata correspond to the injunctive forms kranta and ranta (Meillet 1920: 203, 205). The archaic character of this distribution is supported by the isolated 3rd pl. injunctive forms naśan and naśanta, which correspond to indicative ākṣiṣur (for āśur replacing *āśat) and āśata. Hoffmann’s conjecture that the initial n- of the injunctive is of secondary origin (1957: 124f.) does not explain why it is limited to the 3rd pl. forms, cf. 3rd sg. middle aṣṭa. As in the case of the sigmatic aorist, it is probable that the vocalic alternation was eliminated in the indicative paradigm. This must have occurred at a much earlier stage, however, because it affected the form which was to yield ásthur. The full grade injunctive ending -anta survived in the paradigm of the subjunctive, which shared the thematic vowel. There is a trace of the original distribution in Homer τάνυται, τανύοντο. 11. As I indicated above (section 3), the 3rd pl. ending -ur replaced earlier *-at < *-nt, not -an < *-ent. Since the optative ends in -yur, the original form must have had zero grade both in the suffix and in the ending. This suggests that it had full grade in the root. 12. Hoffmann has argued that the root aorist optative had fixed stress on the root (1968). His proposal offers a straightforward explanation for 3rd pl. Latin velint, Gothic wileina, and OCS velętъ, but not for the remarkable alternation which the latter language shows between 2nd pl. xoštete, dovьljete and 3rd pl. xotętъ, dovьlętъ. It appears that the 3rd pl. form differed from the other persons in the original paradigm. This enables us to remove the unlikely assumption that the root aorist differed from the root present in the accentuation of the optative. 128 Indo-Iranian 13. Insler connects the type dheyām with the type gaméyam, the two being in complementary distribution (1975b: 15). His explanation falters on two points. First, it requires the previous existence of both *dheyam and *dhāyām, of which the attested form represents a blending. It is highly improbable that neither of the earlier forms would have survived because both were supported by other paradigms, while the alleged blending created a new type. Second, the motivation for the spread of the new vocalism to the 3rd person forms is very weak. The long chain of analogic changes which Insler’s theory requires is too complicated to be credible. 14. Thus, I arrive at the following reconstruction of the Proto-Indo-European active root optative: 1st 2nd 3rd singular dhH1ieH1m dhH1ieH1s dhH1ieH1t plural dhH1iH1me dhH1iH1te dheH1iH1nt After Sievers’ law and the loss of tautosyllabic laryngeals this paradigm turned into the following: 1st 2nd 3rd singular dhiyām dhiyās dhiyāt plural dhīma dhīta dhaȤiȤat The generalization of *dhaȤ- and the substitution of -ur for *-at yielded 1st sg. dheyām, 3rd pl. dheyur. 15. The isolated 1st pl. middle optative form naśīmahi (3×) next to aśīmahi (5×) suggests that this paradigm also contained a form with full grade in the root. Since the initial n- is lacking elsewhere in the middle optative and indicative paradigms, it was probably taken from the unattested 3rd pl. optative form. 16. The accentual mobility in the paradigm of the optative is reminiscent of the one in the reduplicated present, where 3rd pl. bíbhrati and dádhati have both initial stress and zero grade in the root and in the ending. Thus, I reconstruct PIE *dhedhH1nti ‘they put’. It follows that the 3rd pl. form does not have the same origin as the other forms of the paradigm. 17. The reduplicating syllable da- of dádhāmi replaces earlier di-, which is preserved in τίθημι and in the desiderative present dídhiṣāmi. It is difficult to agree with Leumann’s view that da- was taken from the perfect (1952: 27) because the motivation for such an analogic development was very weak. More probably, the paradigm of the present contained a form with da- from the very outset. This must have been the 3rd pl. form. In my view, PIE *dhi- was simply Archaic ablaut patterns in the Vedic verb 129 the pretonic (zero grade) variant of *dhe- before a double consonant, cf. πίτνημι, ἵππος, πίσυρες beside πέσσυρες and πέτταρες, Czech čtvrtý < *čьtvьrtyj ‘fourth’, OCS šьlъ < *šьdlъ ‘went’. 18. The 3rd pl. forms yánti, kranta, dheyur and dádhati have in common that the initial syllable contains a full grade vowel. They have the same vocalism as the participles yánt-, kránt-, dádhat-. It is therefore probable that the form in -nti represents the original nominative plural form of the participle. The plural ending -i is also found in the Proto-Indo-European pronominal inflection: nom. *to-i, gen. *to-i-s-om, dat. *to-i-mus, abl. *to-i-os, inst. *to-i-bhi, loc. *to-i-su. It follows from this point of view that the secondary ending *-nt was created on the analogy of the singular forms, where the primary -i had a different origin. In my view, the plural ending -i is of Indo-Uralic origin. It can be identified with the Finnic and Northern Samoyed oblique plural suffix -i-, e.g. Finnish talo ‘house’, pl. talot, taloi-. It is also found as a plural object marker in the Northern Samoyed objective conjugation, e.g. Yurak mada-i-n ‘I (did) cut (more than two things)’, cf. Finnish pala-n ‘I burn (intr.)’, Lappish puolám < *palak-mi. ACCENT AND ABLAUT IN THE VEDIC VERB Most scholars nowadays reconstruct a static root present with an alternation between lengthened grade in the active singular and full grade in the active plural and in the middle. I am unhappy about this traditional methodology of loosely postulating long vowels for the proto-language. What we need is a powerful theory which explains why clear instances of original lengthened grade are so very few and restrains our reconstructions accordingly. Such a theory has been available for over a hundred years now: it was put forward by Wackernagel in his Old Indic grammar (1896: 66-68). The crucial element of his theory which is relevant in the present context is that he assumed lengthening in monosyllabic word forms, such as the 2nd and 3rd sg. active forms of the sigmatic aorist injunctive. Since the sigmatic aorist is the prototypical static paradigm in the verbal inflection, it offers the possibility of testing the relative merits of the two theories, Wackernagel’s lengthening in monosyllabic word forms versus a static paradigm with lengthened grade in the singular and full grade in the plural. As I have pointed out elsewhere (K065), the evidence substantiates Wackernagel’s view and forces us to reject the alternative because we find full, not lengthened grade in the 1st sg. form, e.g. Vedic jeṣam ‘conquer’, stoṣam ‘praise’. The only 1st sg. active form with lengthened grade in the sigmatic aorist injunctive is rāviṣam of the root ru- ‘hurt’, which is clearly analogical. It is therefore reasonable to assume that originally the static present also had lengthened grade in the 2nd and 3rd sg. active forms of the injunctive and full grade elsewhere. Narten assumes that the injunctive forms yoṣam and stoṣam took their vocalism from the subjunctive (1964: 213, 277). The model for this analogic development is lacking, however, because the subjunctive ending was -āni, not -am. The use of the 1st sg. injunctive for the subjunctive must be explained from the meaning of the forms. Note that standard British English offers an exact parallel in the use of ‘I shall’ where other persons ‘will’. Similarly, 8.74.15 dediśam ‘I shall point out’ must be identified as an injunctive (cf. Hoffmann 1967a: 253281), not a subjunctive (thus Schaefer 1994: 42f. “will ich [...] hinweisen”), and the same holds for yoṣam and stoṣam. The injunctive presents the event as a fact without specifying its time frame. As a result, the listener has to supply a time frame in which the event is part of reality, and is driven by the context to choose the most obvious possibility. The subjunctive, however, presents the will to achieve a situation as part of reality, and thereby suggests that its accomplishment may be beyond the subject’s control. The “Spezialfall” of the “Nebeneinander von Injunktiv und Konjunktiv in der 1. Person Singularis” (Hoffmann 1967a: 247) is a result of the fact that the first person can take full responsibility for his own actions, cf. also 2.18.3 hárī nú kaṃ rátha índrasya 132 Indo-Iranian yojam āyái sūkténa vácasā návena “die Falben schirre ich nun an Indras Wagen [‘now indeed shall I harness the steeds to Indra’s chariot’] mit wohlgesprochener neuer Rede, auf daß er komme” beside 1.82.1-5 yójā nv ìndra te hárī “ich will dir nun deine Falben anschirren, Indra” followed by 82.6 yunájmi ‘I harness’ (Hoffmann 1967a: 253). If Wackernagel’s theory is correct, as I think it is, we also expect lengthened grade in the 2nd and 3rd sg. active forms of the root aorist injunctive. Perhaps the clearest piece of evidence for this original distribution is the long vowel in 3rd sg. *gwēmt ‘came’, Latin vēnit, Gothic qēm-, Toch. B śem, which can hardly be explained otherwise. Another instance is Greek σβη- < *sgwēst ‘(the fire) went out’ (cf. Ruijgh 1998: 226). A third example may be Old Irish ro-mídair ‘he judged’ < *mēd- of midithir ‘judges’ < *med-, Gothic mēt- ‘measured’, Greek μηδ‘be disposed, inclined’ beside μεδ- ‘be observant, attentive’. This raises the question of why in Indo-Iranian the long vowel was generalized in the sigmatic aorist indicative, which had fixed stress, and eliminated in the root aorist, which had mobile stress. The reason must be sought in the difference between static and dynamic paradigms. The problem will be taken up below. We first have to establish the nature of the static present, which is not a frequent type of inflection. I subscribe to Alexander Lubotsky’s unpublished theory that it must be derived from a reduplicated formation (cf. already Rix apud Harðarson 1993: 2912). A clear instance is Vedic 3rd sg. t¡ṣṭi, 3rd pl. tákṣati ‘fashion’, which cannot be separated from Greek τέκτων ‘carpenter’ < *tetk-. Another example is Vedic 3rd sg. d¡ṣṭi ‘makes offering’ beside dákṣate ‘is able’ < *dedk- (cf. Lubotsky 1994: 204). These verbs may have provided a model for kāś‘appear’ beside cáṣṭe ‘sees’ < *kwetś- < *kwekwk-, then śās- ‘order’ < *ke(k)Hsbeside aorist śās-, śiṣ- < *keHs-, *kHs-, and 3rd sg. m¡rṣṭi ‘wipes’ < *me(m)rg-, stáuti ‘praises’ < *ste(st)u-, with lengthening of the vowel replacing the lost consonants. The original formation can be identified with Greek τίκτω ‘engender’. It may be useful to have a look at the place of this formation in the original verbal system. Following a line of thought developed by Pedersen (1921: 25f.) and Kuiper (1934: 212), I reconstruct a hysterodynamic s-present, 3rd sg. *tresti, 3rd pl. *trsenti, beside a static s-subjunctive (Indo-Iranian aorist injunctive), 3rd sg. *tērst, 3rd pl. *tersnt, the coexistence of which is perhaps best preserved in Tocharian (cf. already K064: 1173), where we find B täs- < *dhH1es- beside A tās< *dhH1s- in the present and B tes-, A cas- < *dhēH1s in the preterit of the verb tā< *dheH1- ‘put’. If the reduplicated formations followed a similar pattern, we may reconstruct a hysterodynamic reduplicated present, Vedic 3rd sg. vívakti ‘speaks’ < *wiwekwti, weak stem *wiwkw-, but with retracted stress in 3rd pl. *wewkwnti, cf. 3rd sg. síṣakti, 3rd pl. sáścati ‘accompany’ < *sisekwti, *seskwnti, beside a static reduplicated subjunctive (Indo-Iranian aorist injunctive), 3rd sg. *wēwkwt, 3rd pl. *wewkwnt, Vedic ávocat ‘he spoke’, subjunctive vócati beside vócāti. This Accent and ablaut in the Vedic verb 133 reconstruction actually explains the long *-ē- in the reduplication syllable of original reduplicated aorists, as opposed to original presents and perfects, in Tocharian (cf. K149: 173). The original accentuation of the hysterodynamic reduplicated present is preserved in Vedic 3rd sg. juhóti, 1st pl. juhumás, 3rd pl. júhvati ‘sacrifice’. For the reduplication syllable cf. also 3rd sg. yuyóti ‘separates’, aorist injunctive yūyot for *yāyut < *yēywt (?), also 3rd sg. ápīpatat for *-pāp(t) < *pēpt beside ápaptat, 3rd pl. ápaptan ‘flew’ for *-paptat < *peptnt, and jáhāti ‘leaves’ beside jíhīte ‘goes forth’, further 2nd sg. vavákṣi beside 3rd sg. vivaṣṭi ‘desires’, imperative rirīhi beside subjunctive rárate ‘give’, also jígāti ‘goes’, jágat ‘going, world’, like Greek τίκτω ‘engender’, τέκτων ‘carpenter’. There is no reason to assume two types of reduplicated present which as a result of partial adaptation under mutual influence gave rise to four different combinations of accent and ablaut in Vedic (thus e.g. Harðarson 1993: 3014) because this assumption does not explain the coexistence of the two types of reduplication within a single paradigm. The rise of the static reduplicated present may have been provoked by the raising of pretonic -e- to -i- in the reduplication syllable (cf. K065: 222). If the historical background of the reduplicated formation proposed here is correct, we should expect full grade reduplication and zero grade root vocalism throughout the paradigm of the Vedic intensive. It follows that full grade root vocalism in the paradigm of the intensive is always the result of analogy after the hysterodynamic flexion types. Thus, I think that 1st sg. dediśam ‘point out’ is the regular injunctive form and that e.g. 2nd sg. dardar ‘split’, 3rd sg. adardhar ‘held’ are analogical for *dardur, *adardhṛt, which were anomalous forms. On the other hand, 3rd pl. forms in -at < *-nt could easily give rise to a thematic injunctive paradigm with 3rd sg. -at and 3rd pl. -an, e.g. davidyutat or -an ‘flashed’ (cf. Thieme 1929: 12f., Hoffmann 1967a: 200f., Schaefer 1994: 41). Jamison asks the question “why the intensive was not thematized throughout: it is the restriction, the relative rarity of these thematic forms that is curious, not their existence” (1983: 48). The answer is precisely that the subjunctive had zero grade root vocalism in the intensive so that the thematic paradigm existed already with a different function. Since the intensive was the only athematic present without full grade vocalism in the predesinential syllable of the active singular forms, the analogical introduction of a full grade root vowel is only to be expected, e.g. dávidyot for *-dyut beside davidyutat or -an. Interestingly, the two instances which Jamison adduces as clear examples of thematized injunctives, as opposed to subjunctives, are precisely the 3rd pl. forms carkiran ‘commemorate’ and pāpatan ‘fly’ (1983: 47), where -an may have replaced -at < *-nt (cf. also Schaefer 1994: 41f. and Lubotsky 1997: 561), like -anta replacing -ata in the middle. This replacement must have been earlier than the general substitution of -ur for *-at < *-nt in Vedic (cf. K044: 1292). 134 Indo-Iranian As I have argued elsewhere (K065), the 3rd pl. form occupies a special position in the paradigm. This is clear not only from the alternating vowel in the reduplication syllable of 3rd sg. síṣakti, 3rd pl. sáścati ‘accompany’ and jígāti ‘goes’, jágat ‘going, world’, but also from the alternating vocalism in the active and middle root aorist and in the paradigm of the optative. As Meillet noticed a long time ago (1920: 202-205), the 3rd pl. middle indicative forms ákrata ‘made’, ārata ‘went’ correspond to the injunctive forms kranta, ranta. The archaic character of this distribution is supported by the isolated 3rd pl. injunctive forms naśan and naśanta ‘attain’, which correspond to indicative ākṣiṣur (for āśur replacing *āśat) and āśata. Hoffmann’s conjecture that the initial n- of the injunctive is of secondary origin (1957: 124f.) does not explain why it is limited to the 3rd pl. forms, cf. 3rd sg. middle aṣṭa. In the active root aorist we find 3rd pl. ásthur ‘stood’, ávran ‘covered’ beside the corresponding injunctive forms sthúr, vrán, which suggest an original alternation between double zero grade in the indicative and a full grade ending in the injunctive. This distribution must be old because the double zero grade is supported by comparative evidence from Greek, where 3rd pl. ἔθεαν ‘put’ replaces earlier *ἔθαν, with loss of the laryngeal and vocalization of the nasal (cf. K098: 67), and from Germanic, where original *dunþ is reflected in the Old English preterit sg. dyde, pl. dydon ‘did’ (cf. K109: 102). In the optative, the alternation between a full grade suffix in the singular and double zero grade in the suffix and the ending in the 3rd pl. form is best preserved in the Old Church Slavic je-presents xošte- ‘want’ and dovьlje- ‘satisfy’, which have *-iHnt in 3rd pl. xotętъ and dovьlętъ, corresponding to Latin velint, Gothic wileina (cf. K065: 221). As in the case of the reduplicated presents, there is no reason to assume different flexion types in the optative, an assumption which does not explain the coexistence of the two types within a single paradigm. Instead we must reconstruct an original alternation between suffixal stress in the active singular, desinential stress in the active 1st and 2nd pl. forms and in the middle, and root stress in the active and middle 3rd pl. forms. This reconstruction actually offers an explanation for the Vedic isolated 1st pl. form naśīmahi (3×) beside aśīmahi (5×) ‘attain’, which suggests that this paradigm also contained a form with full grade in the root. Since the initial n- is lacking elsewhere in the middle optative and indicative paradigms, it was probably taken from the unattested 3rd pl. middle optative form. The reconstruction of a triple accent and ablaut alternation advocated here also accounts for the root aorist optative type exemplified by 1st sg. Vedic dheyām, Greek θείην ‘put’ (cf. Harðarson 1993: 126-142 for a survey of the scholarly literature). In his elaborate treatment, Insler connects the type dheyām with the type gaméyam, the two being in complementary distribution (1975b: 15). His explanation falters on two points. First, it requires the previous existence of both *dheyam and *dhāyām, of which the attested form represents a blending. It is highly improbable that neither of the earlier forms would have survived Accent and ablaut in the Vedic verb 135 because both were supported by other paradigms, while the alleged blending created a new type. Second, the motivation for the spread of the new vocalism to the third person forms is very weak. The long chain of analogic changes which Insler’s theory requires is too complicated to be credible. Thus, I think that the paradigm of dheyām was based on the 3rd pl. form *dhaȤiȤat, which had full grade in the root and double zero grade in the suffix and the ending, because the zero grade of the root was reduced to dh- before the optative suffix -yā-, -ī- in the other persons. Similarly, Greek introduced the stem vowel from 3rd pl. *theīn into the other persons, where the zero grade of the root had been reduced to th- before -iē-, -ī- (cf. K097: 238). The disyllabic character of *theīn is still preserved in τιθεῖεν < *titheī-en ‘they may put’, where the accent was not retracted to the initial syllable, unlike δύναιο, δύναισθε ‘you may be able’, which replace earlier *dunīso, *dunīsthe. The 3rd pl. ending of the sigmatic aorist optative *-sīn < *-sīnt, which had become homophonous with 1st sg. *-sīn < *-sīm, was replaced by *-seīn on the analogy of *theīn, the ending of which was also found in the passive aorist and in the paradigm of ἵημι ‘let go’. This is the origin of the so-called Aeolic optative. The correctness of these reconstructions is corroborated by the Old High German preterit subjunctive (Indo-European optative) of weak verbs. The difference between Alemannic nāmi ‘took’ and suohtī ‘sought’ (Notker nâme versus suohtî), which cannot be explained as a secondary development, shows that the two paradigms represent different formations. While nāmi can be compared with wili ‘wants’ (Notker wile) and derived from *-īt, the weak ending -tī must be compared with Vedic 1st sg. dheyām, 3rd pl. dheyur, Greek θείην, θεῖεν, and derived from *dheīt (cf. K109: 105). It provides conclusive evidence for the compound origin of the Germanic weak preterit. The peculiar accentuation of the 3rd pl. forms such as Vedic tákṣati ‘fashion’, sáścati ‘accompany’, júhvati ‘sacrifice’, *jágati ‘go’, ákrata ‘made’, ārata ‘went’, naśan, naśanta, āśata, ākṣiṣur ‘attain(ed)’, ásthur ‘stood’, ávran ‘covered’, dheyur ‘put’ points to a different origin from the other forms of the verbal paradigm. In fact, the accentual alternation in 1st sg. juhómi, 3rd sg. juhóti, 1st pl. juhumás, 3rd pl. júhvati ‘sacrifice’ suggests that these forms have three distinct origins: the singular looks like a regular verbal paradigm, with suffixed endings which may go back to clitics, while the 1st pl. form resembles a derivative, perhaps a compound, and the 3rd pl. form has the appearance of a participle. As I pointed out earlier (K065: 222), I think that the form in *-nti represents the original nom.pl. form of the participle, with the Indo-Uralic plural ending *-i which is also found in the Proto-Indo-European pronominal inflection, e.g. *toi ‘they, these’, gen. *toisom, etc. Since Beekes’s discussion of Latin iens, eunt- ‘going’ (1985: 67-71), we have to start from a reconstructed paradigm with nom.sg. *H1eints, acc.sg. *H1ientm, gen.sg. *H1intos, in which Vedic 3rd pl. yánti < *H1ienti may have been the original nom.pl. form of the participle. If the present 136 Indo-Iranian indicative *trsenti and the aorist injunctive *tersnt originally belonged to the same paradigm, the latter form looks like the original neuter of the participle. This is indeed the expected form if the agent of a transitive verb in the aorist was in the ergative case (cf. K049). Thus, I tentatively reconstruct present indicative *toi trsenti beside aorist injunctive *tois tersnt, where *tois is the original ergative from which the genitive *tois-om and the instrumental *to-ois were derived. A typological parallel is offered by the dialectal Russian (plu)perfect, e.g. v jix košalí bal'šýje nabíto (býli) sénom ‘they have/had filled the big bags with hay’ (cf. Honselaar 1998: 303), literally: ‘at them [gen.pl.] bags [nom.pl.] big [nom.pl.] filled [neuter past participle] (were [pl.]) hay [inst.sg.]’. It appears that the participial form was cliticized after the augment in Vedic ásthur ‘stood’ for *ásthat < *H1e-stH2nt, also ákrata ‘made’ < *H1e-krnt-, and after the reduplication, e.g. neuter dádhat ‘putting’ < *dhedhH1nt, 3rd pl. dádhati < *dhedhH1nti. In this view, forms like naśan, naśanta ‘attain’, ávran ‘covered’ adopted the full grade ending on the analogy of the primary (i.e. nom.pl.) form in *-enti, whereas the static paradigm is ultimately based on the secondary (i.e. neuter) form with zero grade *-nt exemplified in *tersnt and *dhedhH1nt. The model for the creation of the full grade secondary endings *-ent, *-ento beside primary *-enti was of course provided by 3rd sg. *-t, *-to beside *-ti, while the zero grade primary ending of dádhati ‘they put’ may have been the original nom.pl. ending of the reduplicated participle. The remaining question is why the lengthened grade was eliminated from the Indo-Iranian root aorist, e.g. Vedic 3rd sg. ágan ‘went’ < *-gwemt, cf. Latin vēnit, Gothic qēm-, Toch. B śem ‘came’ < *gwēm-. As in the case of the s-present and the s-aorist, I think that we have to start from a reduplicated present indicative, 3rd sg. *wiwekwti, 3rd pl. *wewkwnti, beside a reduplicated aorist injunctive, 3rd sg. *wēwkwt, 3rd pl. *wewkwnt, cf. Vedic vívakti ‘speaks’, ávocat ‘spoke’. The meaning of this formation must have been iterative or intensive (cf. Bybee et al. 1994: 166-174 on the semantic development of reduplicated formations). When lengthened grade superseded reduplication in the active singular of the static present, first in TeK-roots such as t¡ṣṭi ‘fashions’, d¡ṣṭi ‘makes offering’, then analogically in m¡rṣṭi ‘wipes’, stáuti ‘praises’, the long vowel became characteristic of this type of derived present and thereby anomalous in the paradigm of the root aorist, where it was limited to the 2nd and 3rd sg. injunctive forms and could easily be eliminated, cf. 3rd sg. imperfect akrāmat beside aorist akramīt ‘strode’. The original lengthened grade may have been preserved in 2nd and 3rd sg. akrān ‘cried’, asyān ‘moved’, āraik ‘left’, acait ‘perceived’, aśvait ‘brightened’, adyaut ‘shone’, which are isolated in the paradigm of the root aorist and could be reanalyzed as sigmatic aorist forms. The hypothesis that these forms are independent analogical creations (Narten 1964: 18) does not explain their isolated character in the oldest texts. I therefore think that they may be relics from the stage when lengthened grade had not yet Accent and ablaut in the Vedic verb 137 adopted the function of reduplication in the static present, which provoked its elimination from the root aorist. THE ORIGIN OF THE INDO-IRANIAN DESIDERATIVE In his detailed account of the Indo-Iranian desiderative (2006), François Heenen offers a new functional analysis in order to establish the invariant meaning of the category and to determine the semantic roles of the morphological components in this meaning. A typical example is jíghāṃsati ‘he wishes to slay’ in JB 1.193 indro vai vṛtram ajighāṃsat “Indra wished to slay Vṛtra”. The desiderative is characteristically a (i) thematically inflected (ii) sigmatic derivation from a (iii) verbal root with (iv) zero grade vocalism and (v) initial reduplication with (vi) i-vocalism and (vii) initial accentuation. Heenen concludes that the semantic value of the reduplication is the image of a “développement progressif graduel. Le sujet exécute des efforts répétés qui convergent vers un résultat” (p. 68) and that the suffix -sa- has “une origine commune à celui du subj. aor.” (p. 72), combining perfective aspect -s- with a non-factive mode of expression -a- (p. 70). The desiderative itself denotes neither the successful accomplishment of the basic action, nor the modal value of a subjunctive. Heenen suggests that the reduplication was the original carrier of the categorial meaning and that the suffx -sa- was added later on the analogy of the aorist subjunctive. This does not explain how the distinction between “actions répétées convergentes des thèmes dés. redoublés avant l’addition du suffixe -sa-, et actions répétées des prés. redoublés du type de bíbhar-” developed (p. 73). In a presentation at Leiden University (28 October 2005) Heenen took a different view, viz. that the desiderative must be derived from the aorist subjunctive and that the reduplication appeared at a later stage. In the following I intend to show that the comparative evidence points to a third possibility, viz. that the perfective -s- was the oldest component and the thematic -a- the most recent addition and that the meaning of the desiderative reflects this chronology. Elsewhere I have argued that all subjunctive and strong future formations of Old Irish can be derived from a single athematic paradigm with a stem in *-sand secondary endings (K239: 71). This unitary inflection combined with six types of stem: 140 Indo-Iranian (a) (b) (c) (d) (e) (f) unreduplicated, e-grade, aniṭ, e.g. fo-ló ‘supports’ < *leugs-, unreduplicated, e-grade, seṭ, e.g. -genathar ‘is born’ < *genas-, reduplicated, e-grade, aniṭ, e.g. fo-cicherr ‘will throw’ < *kikerds-, reduplicated, e-grade, seṭ, e.g. -gignethar ‘will be born’ < *gigenas-, reduplicated, zero grade, aniṭ, e.g. fo-lil ‘will support’ < *lilugs-, reduplicated, zero grade, seṭ, e.g. -géna ‘will wound’ < *gignās-. As a rule, the unreduplicated forms are subjunctives and the reduplicated forms are futures, but there is a class of basic verbs (‘lie’, ‘sit’, ‘run’, ‘flee’, ‘arise’, ‘protect’) where the unreduplicated forms are used both for the subjunctive and for the future (cf. Thurneysen 1946: 410f.). It is therefore probable that all of these formations represent an earlier athematic subjunctive paradigm with secondary endings. Moreover, there is a seventh type without reduplication but with a zero grade root and a full grade suffix *-es-, viz. subjunctive -bé < *bes- < *bhwest ‘be’, future -bia < *biyas- < *bhwiHes- (cf. K239: 73), cf. Latin fu- ‘be’, fi- ‘become’. It appears that the reduplication was introduced in order to differentiate the future from the subjunctive. From an Indo-European point of view, the full grade forms can be compared with the sigmatic aorist injunctive and the zero grade forms with the injunctive of the s-presents (cf. Pedersen 1921, Kuiper 1934). I have proposed to reconstruct a hysterodynamic present 3rd sg. *tresti, 3rd pl. *trsenti beside a static aorist 3rd sg. *tērst, 3rd pl. *tersnt (e.g. K188: 9). The system is perhaps best preserved in Tocharian, where we find B täs- < *dhH1es- beside A tās- < *dhH1sin the present and B tes-, A cas- < *dhēH1s- in the preterit of the verb tā- < *dheH1- ‘put’. For the reduplicated formations I reconstruct a hysterodynamic present, e.g. Vedic 3rd sg. vívakti ‘speaks’ < *wiwekwti, weak stem *wiwkw-, but with retracted accent in 3rd pl. *wewkwṇti, cf. 3rd sg. síṣakti < *sisekwti, 3rd pl. sáścati < *seskwṇti ‘accompany’, beside a static aorist 3rd sg. *wēwkwt, 3rd pl. *wewkwṇt, reflected in Vedic ávocat ‘he spoke’, subjunctive vócati beside vócāti. This reconstruction explains the long *-ē- in the reduplication syllable of original reduplicated aorists, as opposed to original presents and perfects, in Tocharian (cf. K149: 173). The original accentuation of the hysterodynamic reduplicated present is preserved in Vedic 3rd sg. juhóti, 1st pl. juhumás, 3rd pl. júhvati ‘sacrifice’. In the same vein I now reconstruct a reduplicated s-present sg. *titrésmi, -si, -ti, pl. *titṛsmés, -tH1é, *tétrsṇti, which explains the combination of accented i-vocalism in the reduplication syllable and zero grade in the root of the Indo-Iranian desiderative. It follows that the thematic vowel is the most recent component of the formation. The unreduplicated athematic s-present *bhHues-, *bhHus- is also attested in Balto-Slavic and Italic, e.g. Lithuanian bùs ‘will be’, Old Church Slavic byšęšteje ‘the future’ (which is an athematic nt-participle), Oscan fust, Umbrian fust, 3rd pl. furent ‘will be’, Oscan fusíd, Latin foret, 3rd pl. forent ‘were’, with full grade The origin of the Indo-Iranian desiderative 141 suffix *-es- in Umbrian ferest ‘will bring’, Oscan pertemest ‘will prevent’, also Umbrian menes ‘you will come’ reflecting *gwmes- with zero grade in the root. This formation is clearly identical with the Old Irish subjunctive -bé < *bhwest. The Balto-Slavic s-subjunctive became a future in East Baltic, e.g. Lith. bùs ‘will be’, but assumed the function of an imperative in Old Prussian, e.g. teīks ‘put!’, where the addition of thematic present endings yielded a future paradigm, e.g. postāsei ‘you will become’. The reduplicated s-present is attested in Italic, where it adopted the function of a future perfect, e.g. Latin ēmerō ‘I will have bought’, Oscan fifikus ‘you will have decided’, fefacust ‘he will have done’, Umbrian fefure < *fifusent ‘they will have been’, dirsust < *didust ‘he will have given’ (cf. K239: 72, 152). The data presented here invite a consideration of the following scenario. We have to start from a hysterodynamic s-present with zero grade root vocalism and a static s-aorist with e-vocalism in the root. Both the present and the aorist injunctive could turn into a non-factive subjunctive. The s-present became a future in Lith. bùs and in Oscan and Umbrian fust ‘will be’, 3rd pl. furent, but not in Old Irish -bé ‘be’ and in Latin foret, 3rd pl. forent ‘were’, which adopted the productive suffix *-ē-. These forms show that the future meaning was comparatively recent here, as it was in Latin erō. The non-factive meaning of the s-present could easily be combined with the aspectual meaning of the reduplicated present, which denoted repetitive activity (cf. Kulikov 2005: 442-444, with references to Ul’janov 1903 and Elizarenkova 1982), e.g. Vedic píbati ‘drinks’ (series of sips), jíghrati ‘smells’ (series of sniffs), jígāti ‘goes’ (series of steps) in RV 10.73.3 ṛṣv¡ te p¡dā prá yáj jígāsi “your feet are high when you (Indra) are treading”. A clear example of the difference between telic bhárati ‘brings’ (Russ. nesët) and atelic bíbharti ‘carries’ (Russ. nósit) is RV 10.30.13 yád ¡po ádṛśram ... ghṛtám páyāṃsi bíbhratīr mádhūni ... índrāya sómaṃ súṣutam bhárantīḥ “when the waters, which carry (bíbhratīr) ghee, milk and honey, which bring (bhárantīḥ) the well-pressed soma-sap to Indra, became visible ...”. The reduplicated s-present with secondary endings denoted repetitive activity aimed at an accomplishment which does not happen. This is Heenen’s “développement progressif graduel. Le sujet exécute des efforts répétés qui convergent vers un résultat” (2006: 68) which is implied by the perfective -s- but not accomplished because the action is not telic. The meaning may be compared with that of the secondary imperfective verb pridumyvat' in Russian Levin slušal i pridumyval i ne mog pridumat' čto skazat' (Lev Tolstoj, Anna Karenina) “Levin listened and tried to think of something to say but could not”, covering effort and attempt but not necessarily desire (except by implication). The expected meaning of the reduplicated s-present is conative, not desiderative. The last step in the development of the Indo-Iranian desiderative is the thematicization of the reduplicated sigmatic formation. Elsewhere I have argued that the Vedic subjunctive presents the will to achieve a situation as part of 142 Indo-Iranian reality, thereby suggesting that its accomplishment may be beyond the subject’s control (K188: 8). This modal value differs from that of the injunctive, which simply presents the event as a fact without specifying its time frame. It follows that the thematic subjunctive of a reduplicated s-present depicts the will to achieve a situation of repetitive activity aimed at the accomplishment of a goal which may be beyond the subject’s control. When this mode of expression lost its non-factive (subjunctive) character, it became a desiderative present, stating the will to achieve the desired situation as a fact. The conflation of the situation desired (-a-) and the situation aimed at (-s-) into a single goal of action (-sa-) enabled the creation of a full paradigm of the desiderative, e.g. vívāsati ‘desires to win’, imperfect āvívāsat, subjunctive āvívāsāt, optative ¡vivāset, imperative 2nd sg. ¡vivāsa, participle āvívāsant- (cf. Heenen 2006: 223-226). There never was an Indo-European desiderative formation with a reduplicated thematic present of the type *titrse/o-. ON THE DEVELOPMENT OF PROTO-INDO-EUROPEAN FINAL SYLLABLES IN TOCHARIAN The fate of PIE final syllables in Tocharian has largely been clarified by Pedersen (1941, 1944) and Lane (1976). In the following I intend to discuss a few details against the background of Jörundur Hilmarsson’s dissertation (1986). As far as I see, there is little evidence for a difference between final and nonfinal syllables in the development of the Proto-Tocharian vowel system. I subscribe to the view summarized by Beekes (1985: 208), which is reproduced here with the slight modification that I write ē, ā, ō for his eh1, aH, oH. pre-Toch. e, i, u a, ǩ o ē/_CC ē/_CV ā/_CC ā/_CV ō/_CC ō/_CV iH/_C iH/_V uH/_C uH/_V ei ai oi eu au ou Proto-Toch. ǩ a e e a a o o u ya (ǩ)y wa (ǩ)w (ǩ)i ai ei (ǩ)u au eu West Toch. a/ä ā/a e e ā/a ā/a o o u yā/ya (i)y wā/wa (u)w i ai ai u au e͡u > au East Toch. ä ā a a ā ā a a u yā y wāɻ w i e e u o o Examples of the long vowel developments: *uēntos ‘wind’, A want, B yente, *mē ‘not’, AB mā, *bhrātēr ‘brother’, A pracar, B procer, *uōstu ‘house’, A waṣt, B ost (Beekes apud Hilmarsson 1986: 11), see further below. Proto-Tocharian consonants: p, ṕ, t, ť, ts, ś, s, š, k, kw, m, ḿ, n, ń, l, ľ, r, (ŕ), y, w, ẃ, e.g. *ṕǩńś ‘five’, A päń, B piś, *ṕǩkwǩl ‘year’, A p͡ukäl, B pikul, *sǩkw < Skt. sukha- ‘happiness’, A suk, B sakw, *ẃǩšǩye ‘night’, A wṣe, B yṣiye, *sporťẃotr ‘turns’, A sparcwatär, B sporttotär, caus. *spartwǩšš, A spārtwäṣ, B ṣparttaṣṣäṃ. It appears that final *-m, *-n, *-s, *-t were lost at an early stage without leaving a trace, and that a preceding vowel developed in accordance with the 144 Tocharian rules given above (cf. Lane 1976: 151 on *-om). The thesis which I would like to defend here is that this loss of final single consonants was anterior to the proposed long vowel shift in open syllables, while final *-nt, *-nts had become *-n at the time of the shift. This final *-n was lost after long vowels at a later stage. Final *-ns was preserved in Proto-Tocharian and developed into A -s, B -ṃ. Eventually, final vowels were apocopated in East Tocharian. The hypothesis that *-ns was preserved in Proto-Tocharian and yielded A -s, B -ṃ was rejected by Pedersen and Lane for different reasons. Pedersen called the derivation of A -s from *-ns “entschieden unrichtig. In einer Sprache, wo das -s der Endung -nts geschwunden ist [...], kann dass -s in der Gruppe -ns nicht geblieben sein” (1941: 81). Lane objected to the derivation of B -ṃ from *-ns: “je n’y crois guère pour ma part; ceux des dialectes grecs qui, diversement, altèrent *-ns le font tous aux dépens de la même consonne, la nasale” (1976: 153). The latter objection had already been raised by Pedersen: “In der Gruppe -ns pflegt das -s der stärkere Laut zu sein, der sich auf Kosten des -n- behauptet” (1941: 77), but he did not regard it as decisive: “Trotz allen Bedenken wird man also den Obl. Pl. auf den ieur. Akk. Pl. zurückführen” (1941: 78). I think that the typological argument cannot be maintained. Sanskrit preserves a reflex of the final *-s in the length of acc.pl. -ān < *-ons but not in nom.sg. -an < *-onts. Armenian preserves the final spirant unchanged in acc.pl. -s < *-ns but not after a vowel. Old High German has a reflex -ā < *-ōns which is distinct from both -a < *-ōn and -o < *-ōs. Since a comparison of the Tocharian languages points to the ending *-ns, I think that strong evidence is required for an eventual rejection of this reconstruction. The early loss of final *-s after an obstruent in the Tocharian branch of Indo-European partly accounts for the rise of the root subjunctive corresponding to s-presents and s-preterites (cf. K068: 181).1 Thus, I derive A wäl, gen. lānt, B walo, gen. lānte ‘king’ from Proto-Tocharian *wǩlo, *lante, earlier *wǩlōn, *wlante from *uelonts, *ulǩntos, cf. Old Irish flaith ‘lordship’ < *ulǩti- of the root *uelH-. The nom.sg. ending *-ōn of the nt-stems gave rise to doublets such as B obl.sg. klyomont beside klyomoṃ of klyomo ‘noble’ (cf. Stumpf 1990: 102, Peyrot 2008: 119). Turning now to the development of long vowels in final syllables, we find the expected reflex of *ē in the palatalizing ā-preterite, e.g. A klyoṣ, B klyauṣa ‘heard’ < *klēusēt. The reflex of *-ā and *-ām is found in the flexion of the original H2-stems, which has preserved two types of apophonic alternation. The two most common types of nouns with accentual mobility in Indo-European are proterodynamic neuters, where the accent alternates between the root and the stem formative, and hysterodynamic masculines and feminines, where the accent alternates between the suffix and the case ending, e.g. Skt. d¡ru ‘wood’, gen. drós, Greek πατήρ ‘father’, gen. πατρός (cf. Beekes 1985: 167-171). Pedersen already remarked that the accusative adopted the stem form of the weak cases in Tocharian (1944: 37), as it did in the plural in Sanskrit. Thus, we may expect that On the development of Proto-Indo-European final syllables in Tocharian 145 the ideal reflex of a proterodynamic H2-stem has full grade in the root, zero grade in the nom.sg. ending, full grade of the stem formative in the other case forms, and neuter plural endings (class II), and that the ideal reflex of a hysterodynamic H2-stem has zero grade in the root, full grade in the nom.sg. ending, zero grade of the stem formative in the other case forms, and no neuter plural endings (class VI). This is exactly what we find. The proterodynamic type is represented by B śana ‘wife’ < *śǩna < *gwenH2, obl. śano < *-ām, gen. śnoy < *-ās plus *-eis, pl. śnona < *-ās plus *-nΗ2, A obl.pl. śnās < *-āns, and B lāntsa ‘queen’ < *wlantya < *-iH2, obl. -o, gen. -oy, pl. lantsona, A nom.pl. läntsañ < *-oń < *-ās plus *-nes, also B ṣarya ‘beloved’, obl. -o. The hysterodynamic type is represented by B kantwo ‘tongue’ < *kǩntwo < *dnghuā, obl. kantwa < *-uH2-m, A nom.pl. käntwāñ < *-uH2-nes, B perl.pl. kantwaṃtsa for *käntwāntsa < *-uH2ns plus a secondary case ending which may be derived from PIE inst. *-H1, and B kātso ‘stomach’ < *kuHtiā (Hilmarsson 1986: 236), obl. -a, nom.pl. katsāñ, also tāno ‘grain’ < *dhH3nā, obl. -a, nom.pl. tanāñ, and kāswo ‘rash’, māskwo ‘hindrance’, tsāro ‘monastery’, all obl. -a. The pronominal adjective A ālak, B alyek, allek ‘other’ < *alios plus -k, nom.pl. A ālyek, B alyaik < *alioi plus -k, shows the reflex of the proterodynamic type in the feminine: A ālyāk, B alyāk, obl. A ālyäkyāṃ (with -ä- for -a-), B alyok, allok, pl. A ālkont, B alloṅk, alloṅkna, pointing to Proto-Tocharian nom.sg. *aľľak, obl.sg. *aľľok, nom.pl. *aľľo(na)k < *aliH2, *aliām, *aliās.2 The phonetic reflex of final *-ō is found in B 1st sg. -u, for which no alternative explanation is satisfactory. The same development is found in the past participle AB -u < *-wu < *-wōs, where A apocopated the vowel and vocalized the preceding semivowel while B preserved the vowel and lost the semivowel, and also in the monosyllables AB ku ‘dog’ < *kuō, A wu ‘two’ < *duō, and in A oktuk ‘eighty’, which reflects the original final vowel of okät ‘eight’ < *ekwtu < *oktō. The corresponding B word okt lost its final *-u on the analogy of the word for ‘seven’, which conversely adopted the labiovelar of the word for ‘eight’, cf. A ṣpät < *šǩpt < *septm, B ṣukt < *šǩkwt. The medial vowel of A oktats, B oktatse ‘eightfold’ and A āknats, B aknātsa ‘ignorant’ must be derived from a vocalized laryngeal, not from a full grade vowel.3 In the decades, A ṣäptuk ‘seventy’ adopted the suffix of oktuk while the converse development is found in B ṣuktaṅka, oktaṅka.4 The phonetic reflex of *ō in a closed syllable is found in the AB verbal nouns in -or < *-uōr and in the B obl.sg. and nom.pl. forms in -oṣ, -ont, -oñc, which must be derived from *-uōsm, *-uōses, *-uōntm, *-uōntes, with early loss of the final *-ǩ which arose from these endings.5 The o-grade was preserved in B -weṣ < *-uoses and the e-grade in A -unt < *-uentm, cf. obl.sg. B śaumoṃ ‘man’ < *-ōnm but A śomäṃ < *-enm. I also assume *ō in B po ‘all’ < *pōn < *pōnts < *peH2onts, nom.pl. masc. poñc < *pōntes, fem. ponta < *pōnta, where the circumflex of Greek πᾶς, πᾶν points to a contraction in the masc. and neuter nom.sg. forms.6 It follows from the position taken here that A ytār, B 146 Tocharian ytārye ‘road’ cannot be derived from “I.-E. *H1itōr” (Hilmarsson 1986: 44), which has little to commend itself anyhow. We must rather start from a zero grade suffix *-tr and assume analogical introduction of the vowel from A ysār, B yasar ‘blood’ < *yǩsar, where it represents a vocalized laryngeal, cf. Hittite ešhar. The same analogy is found in A wmār ‘jewel’, cf. B wamer < *wǩmer. The word AB kos ‘how much’ may be derived from *kwō-su, certainly not from *kwo-suō(s) (Hilmarsson 1986: 46), which should yield AB -u, as in the past participle. The original nom.sg. ending of the n-stems *-ō(n) is preserved in the word for ‘dog’, AB ku, where it cannot be analogical. The usual ending B -o must ultimately be derived from *-ōnts (with analogical lengthened grade), which may have yielded *-ōn at an early stage in the development of Proto-Tocharian. It was evidently introduced into the flexion of the n-stems at a time when the final *-n was still preserved, cf. the converse substitution in Greek φέρων ‘carrying’. Thus, it shows the expected reflex of *ō in a closed syllable. When the final *-n was lost, the ending merged with the reflex of *-eH2. In spite of Hilmarsson’s efforts to prove the contrary, I still think that obl.sg. A koṃ, B kweṃ ‘dog’ represent Proto-Tocharian *kwen < *kwonm, like A por, B pūwar ‘fire’ < *pwar < *puHr (cf. Winter 1965: 192).7 The preservation of -w- in A śtwar, B śtwer ‘four’ < *śǩtwer < *kwetuores is probably due to the intervocalic position of the cluster. If A kaṃ, B kene ‘melody’ is correctly derived from *gʄhuʢonos (Hilmarsson 1986: 158), the *-w- must have been lost at an early stage. A sas, B ṣe ‘one’ point to Proto-Tocharian *šens < *sēms, with A s- for *šassimilated to -s, and with analogical elimination of B *-ṃ, which was preserved in the feminine A säṃ, B sana < *sǩna < *sm (neuter?) plus *-H2. The nom.sg. ending -e represents *-ēn, where the final nasal must have been restored at a time which was anterior to the long vowel shift in open syllables, cf. the contrast between Latin liēn ‘spleen’ and homō ‘man’. I find no evidence for original ē-stems in Tocharian. NOTES Final *-s was regularly lost in *seks, A ṣäk ‘si’x, but analogically restored in B ṣkas, evidently on the basis of the ordinal *seksos, A ṣkaṣt, B ṣkaste. 1 Hilmarsson’s view that the theory outlined here makes the o-vocalism of A *śoṃ ‘continent’ and *mok ‘greatness’ incomprehensible (1986: 350) must be based on a misunderstanding. Even if his derivation of these asterisked words from *ścono < *styānā and *moko < *mākā is correct, their failure to delabialize in East Tocharian is a dialectal problem which should not affect the reconstruction of the Proto-Tocharian forms. Moreover, Hilmarsson’s “difficulties fitting B karyor, A kuryar” ‘commerce’ < *kwǩryor into his system because “the finals B -or and A -ar seem irreconcilable” (1986: 16) are a consequence of his unwillingness to adopt the position taken here. His 2 On the development of Proto-Indo-European final syllables in Tocharian 147 assumption of three different types of Proto-Tocharian *ō in posttonic syllables is not warranted. The vocalized laryngeal is also found in Skt. aśītí- ‘eighty’ and probably in Greek ὄγδο(ϝ)ος ‘eighth’ (K060: 99, 103). This obviously does not imply a direct derivational connection between A oktats and Skt. aśītí- (Hilmarsson 1986: 349). It simply means that we have to start from zero grade before the suffix. 3 Hilmarsson points out correctly that the -u of *oktō was preserved in dialectal times whereas the final nasal of *septm was apocopated in pre-dialectal times already (1986: 159). This does not affect the reconstructed analogical developments in any significant way, however. 4 5 See now Peyrot 2009, however. For the possibility of deriving Greek πᾶς from *peH2onts with restored *H2 see K034. The flexion type must evidently be compared with Skt. bháran < *-onts, bhárat- ‘bearing’, whereas Greek ἱστ³ς ‘setting’ < *-stH2ents seems to agree with Skt. dviṣán < *-ents, dviṣat- ‘hating’. The root of Toch. B po and Greek πᾶς was apparently ousted in the other Indo-European languages because of its homonymy with *peH2- ‘protect’. This is no reason to question the IndoEuropean origin of the word, cf. Greek λᾶας ‘stone’, which is doubtless IndoEuropean (cf. Beekes 1985: 15-17) though it has no obvious cognates in other languages. Hilmarsson’s derivation of B po from *H1pōnts (1986: 214) is impossible because there is no prothetic vowel in Greek. 6 I agree with Schindler (1967: 243) that Winter’s derivation of A por and B pūwar from different nominatives is “ein Weg, den man bei so nahe verwandten Sprachen nach Möglichkeit vermeiden sollte”. We have to start from PIE *peH2ur, gen. *pH2uens, later *pH2unos > *puH2nos, then pre-Toch. *puHr > *pwar. There is no evidence for a collective in *-ōr in Tocharian. 7 THE TOCHARIAN WORD FOR ‘WOMAN’ Consider the following paradigms: ‘wife’ A nom.sg. śäṃ obl.sg. śäṃ nom.pl. śnu obl.pl. śnās B śana śano śnona śnona ‘woman’ A k͡uli k͡ule k͡ulewāñ k͡ulewās B klyiye, klīye klaiñ, klaiṃ, klai klaina klaina I agree with Klaus Schmidt that the word for ‘woman’ must be derived from the word for ‘wife’ with dissimilation of the radical nasal to -l- before a suffixal nasal (1980: 410). For a similar dissimilation cf. Lith. glinda ‘nit’ < *gninda. I disagree with Schmidt’s view that the two words represent different stem forms of the same paradigm, to be compared with Arm. kin < *gwena, pl. kanay- < *gwnai-, Greek γυνή, gen. γυναικός. These forms must probably be derived from an original proterodynamic neuter paradigm, PIE nom.acc. *gwenH2, gen.abl. *gwneH2s, dat.loc. *gwneH2i (cf. Beekes 1985: 167). The Toch. word for ‘woman’ is rather a derivative of this root. For the derivation cf. Slovene žéna ‘wife’, žéʛnski ‘female (adj.)’, žéʛnska ‘woman’, and the English noun female. Thus, we may look for an adjectival suffix which yields the proper output when added to the root *gwnH2-. It is clear that the suffix must have contained an *n for the dissimilation of the radical *n to take place. This is not the only requirement, however. Hilmarsson observes that B klyiye shows “the main characteristics of Class VI,2: nom.sg. -iye and nom.pl. with -ai-. This fact might prompt one to assume that this word belonged to Class VI,2 at an earlier stage and was later transferred to Class II,1 because of its meaning” (1986: 218, also 333). I think that this is quite correct, but one may wonder if it is compatible with his conclusion that -iye- is “an intruder in Class VI,2, it seems to be more at home in Class VI,1” because “-iye in its natural element (Class VI,1) had a palatalizing effect, but when transferred to a foreign flexional category (Class VI,2) -iye had no such effect” (1986: 229). Thus, we have to account for the palatalizing effect of -iye in the word for ‘woman’. Hilmarsson has convincingly argued that the ending -(iy)e must be derived from *-(iy)ēn (1986: 231-236). The problem with the word for ‘woman’ is that there is no obvious reason for the loss of the root-final laryngeal of *gwnH2before this ending. On the one hand, *H2 is not lost before *i, cf. especially A sne, B snai ‘without’ < PIE *snHi, Latin sine. On the other hand, *H2 colors a following *e to *a, so that palatalization cannot take place. Thus, we must look 150 Tocharian for a suffix before which *H2 is lost. The only possibility I can think of is *-H1en-, *-H1n- (cf. Beekes 1985: 53), which fits perfectly. Assuming that *Η2 was lost before *H1, I arrive at the following reconstruction: stage I *gwnΗ2-Η1ēn *gwnΗ2-Η1n- stage II *kwlH1ēn *kwlH1n- stage III *kwlye *kwlan- stage IV klyiye klain- For the substitution of -iye, -ain- for *-e, *-an- I refer to Hilmarsson (l.c.) and for the vocalization of *H1 as -a- to Klingenschmitt (1975: 161-162). POSTSCRIPT It appears that B klīye is older than klyiye (cf. Hilmarsson 1996: 157, Peyrot 2008: 109). This suggests the possibility that *H1 blocked the palatalization by the following front vowel at stage III. THE FATE OF THE SIGMATIC AORIST IN TOCHARIAN 1. Krause and Thomas write (1960: 247): “In den aktiven Formen tritt das -s- nur in der 3. Sg. auf, z.B. B Sg. 1 prekwa, 2 prekasta (mit Endung -s(a), 3 preksa, Pl. 3 prekär [A prakwā, prakäṣt, prakäs, prakär]. Im Medium führt das Wtoch. das -s- durch das ganze Paradigma durch, z.B. ersamai, -tai, -te, -nte. Im Otoch. dagegen findet sich eine doppelte Flexion im Medium: Einerseits wird auch hier das -s- durchgeführt, z.B. rise, risāte, risāt, risānt; andrerseits erscheinen s-lose Formen wie tamät, tamänt; nakät, nakänt; pakät, pakänt; lyokät, lyokänt; tsakät, tsakänt. Nur bei Wz. yām- zeigen sich beide Bildungen: yāmtse, *yāmtsāte, yāmtsāt, yāmtsānt neben yāmwe, yāmte (nicht bezeugt in den übrigen Personen).” The only active paradigm with generalized -s- is ‘gave’: B wsāwa, wsāsta, wasa (wsā-ne), wasam, wsare (wsär-ñi), A wsā, wäs (wsā-ṃ), wsär, part. wawu (cf. already Pedersen 1941: 186). We must look for an explanation of this distribution. 2. As a rule, final obstruents are lost in Tocharian, so that the expected reflex of 3rd sg. *prēkst (Vedic áprāṭ ‘asked’) is B *prek, A *prak, and the sigmatic ending must be analogical, cf. especially A ṣäk ‘si’x < *seks, B ṣkas with restoration of -s on the basis of the ordinal ṣkaste. The phonetic loss of *s accounts for the rise of an asigmatic paradigm on the basis of Proto-Tocharian 3rd sg. *prek. 3. If sigmatic and asigmatic forms existed side by side at some stage, the main question is: why was the sigmatic ending restored in the 3rd sg. form and ousted in the rest of the active paradigm? This question must be viewed in relation to the corresponding subjunctive. As Lane observed, there is a pattern: “s-present, athematic subjunctive originally only active; e-subjunctive, only middle, s-preterit” (1959: 165). If the asigmatic forms in the s-preterite arose from the phonetic loss of *s, the root subjunctive is best derived from the sigmatic aorist injunctive, a derivation which moreover explains the absence of an s-subjunctive. 4. This brings us to a reconsideration of the root vocalism. Elsewhere I have argued that the sigmatic aorist injunctive had lengthened grade in the 2nd and 3rd sg. forms and e-grade in the rest of the paradigm, and that this distribution is still reflected in the Vedic material, where lengthened grade was already generalized in the indicative (K065). Since apart from the palatalization *ē in closed syllables merged with *o while *e plus resonant merged with the corresponding zero grade reflex in Proto-Tocharian (cf. K095: 80), the root 152 Tocharian vocalism of the perfect and the sigmatic aorist merged outside the 1st sg. form, which was easily subject to analogical leveling. As a consequence, palatalization and its absence became associated with transitive versus intransitive paradigms (cf. Winter 1980a). 5. When the subjunctive and the preterite adopted the endings of the present and the perfect, respectively, the sigmatic ending served to disambiguate the 3rd sg. form *prek. This presupposes the earlier coexistence of sigmatic and asigmatic paradigms. The sigmatic paradigm of AB wäs- ‘gave’, also A cas-, B tes‘put’, and the asigmatic middle paradigm of A nak- ‘perished’, pak- ‘cooked’, tsak‘burned’ suggest a correlation between vocalic stems and sigmatic forms on the one hand, and between consonantal stems and asigmatic forms on the other. This is understandable if original *s was lost in interconsonantal position, as was the case in the verb stems in -tk- (cf. Melchert 1977). 6. Thus, I propose the following scenario. The loss of word-final and interconsonantal *s yielded a mixed paradigm which gave rise to the root subjunctive on the one hand and to sigmatic and asigmatic preterites on the other. The generalized reflex of lengthened grade in the active preterite is strongly reminiscent of the Vedic aorist indicative. In Tocharian, it even spread to the middle paradigm, e.g. A lyokät ‘was illuminated’, tamät ‘was born’, nakät, pakät, tsakät, which point to *lēuksto, etc. The initial palatalization was partly redistributed according to transitiveness. The original vowel alternation was largely eliminated. 7. It will be clear from the foregoing that I strongly disagree with both Adams (1988) and Jasanoff (1988). While the latter’s rash comparison with Hittite does not account for the multifarious sigmatic formations which are attested in the Tocharian material, the former’s reliance on the initial palatalization for the identification of a preterite as an original aorist or perfect is at variance with the productive character of this feature. I think that it is hardly possible to identify an original perfect at all (cf. already Pedersen 1941: 183) because the root aorist adopted the endings of the perfect at an early stage. Apart from the endings and the reduplicated participle, the only clear trace of the perfect appears to be the mobile stress in suffixless preterites with an original o-grade root vowel, as opposed to the fixed initial stress in the s-preterite. 8. The e-subjunctive, like the e-present, appears to be ultimately based on the stative 3rd sg. form in *-o, which was extended by the regular middle endings. The s-present is a thematically inflected sigmatic formation. It seems probable to me that it represents a thematicization of an original present in 1st sg. *-esmi, 1st pl. *-smes (cf. Pedersen 1921: 26). If this is correct, Tocharian has preserved an archaic feature in the correlation between s-present, root subjunctive (sigmatic aorist injunctive), and s-preterit. The original vowel alternation in the suffix has The fate of the sigmatic aorist in Tocharian 153 left a trace in the present of AB tā- ‘put’, e.g. 3rd pl. A tāseñc < *dhH1s-, B taseṃ < *dhH1es-, with a different generalization in the two languages, both of which have tās- in the subjunctive and täs- in the middle preterite. This suggests that the athematic s-present was still preserved in Proto-Tocharian. POSTSCRIPT Michaël Peyrot points out to me that B taseṃ is an archaic form and may therefore represent *tāseṃ (with zero grade root vocalism). A NOTE ON THE TOCHARIAN DUAL 1. It appears that the PIE word for ‘two’ was *duo (cf. Cowgill 1985). I reconstruct the nom.du. ending as *-H1(e) (cf. Oettinger 1988) for animates and *-i for inanimates. The fundamental character of the animacy distinction must be recognized since Weitenberg’s basic article (1987: 225 and passim). Other endings are analogical, e.g. *duoH1 ‘two (persons)’, *duoi ‘two (things)’, Skt. akṣã < *-i-H1 ‘both eyes’, Greek ὄσσε < *-i-H1 or *-i-e, with addition of the regular ending of the consonant stems at a prehistoric stage. It is usually assumed that the neuter ending was *-iH1 (e.g. Oettinger 1988: 355). While the *-H1 can easily be analogical in this ending, there are several indications that the earlier ending was *-i. First of all, PIE *duidkmti ‘twenty’ contains a dual ending, as opposed to *triH2dkomt ‘thirty’ (cf. K060: 100), e.g. Skt. viṃśatí, triṃśát, Greek ϝÕκατι, τρι³κοντα. The short -i cannot be analogical in these forms because there is no model. Secondly, Vedic ákṣi ‘eye’ may represent the original dual form, which was enlarged to akṣã, AV ákṣiṇī. The short -i of ákṣi is difficult to explain otherwise. Thirdly, I think that the nom.du. ending *-i was preserved in Gathic. In his Gathic grammar, Beekes lists eight instances of short i for expected long ī (1988: 42f.). If we disregard the YH forms, which have a shortened vowel in an inner syllable of a polysyllabic word followed by the enclitic -cā ‘and’, the five remaining instances are the following (with translation from Insler 1975a): (30.11b) xvīticā (ǩɷnǩitī) ‘both a way of easy access (and one with no access)’ (31.4b) ašicā (ārmaitī) ‘also reward (and piety)’ (32.10b) ašibyā ‘with (his) eyes’ (53.5a) kainibyō ‘to (these) girls’ (53.8c) šyeitibyō (vīžibyō) ‘for the (peacefully) dwelling (settlements)’ The first three instances are dual forms while the last two represent analogical shortening before the oblique plural endings, as in the following cases: (49.4b) (xvāiš) hizubīš ‘with (their own) tongues’ (43.7e) (aibī θwāhū gaēθāhū) tanušicā (read tanušucā) ‘(among thy creatures) and (thine) own’ Though the evidence is not conclusive, the prominence of dual forms among the instances of unexpected short i and the absence of neuter dual forms with long ī suggest that there was a short neuter ending *-i in Gathic. 2. Turning to the Tocharian languages, I think that an explanation which starts from PIE *duo and the endings *-H1(e) and *-i is preferable to possible alternatives. It follows that I find myself unable to adopt the interpretation 156 Tocharian which Hilmarsson has proposed in his careful analysis of the material (1989). In this note I shall not go into a detailed discussion but limit myself to a few basic points. As Pedersen pointed out already (1941: 75), “eine Flexionsform, die in B -ne, im A -n geben sollte, müsste den ieur. Vokal o enthalten haben. Für eine solche Flexionsform gibt es aber nirgends eine Stütze. Wir müssen also in -ne eine enklitische Partikel sehen, so unklar es auch ist, wie sie dazu gekommen ist, sich mit Dualformen zu verbinden”. I think that this particle can be identified with the deictic particle *an < *Η2en (cf. K063: 320f.) followed by the numeral *dwo in the same way as Lith. mùdu ‘we two’, jùdu ‘you two’, fem. mùdvi, jùdvi, OE wit, git. The same particle is found in the 3rd sg. suffixed pronoun A -ṃ, B -ne. The genitive forms in A -nis, B -naisäñ, -naisi can be compared with Arcadian -οιυν (cf. Hilmarsson 1989: 61f.) and Lith. locative mùdviese, jùdviese. The expected reflex of the PIE endings *-H1e and *-i is palatalizing *-ä in Proto-Tocharian, and this is indeed the regular ending before the dual particle, e.g. A aśäṃ, B eś(a)ne ‘both eyes’ < *okwi-ndwo, which therefore reflects an older stage than Greek ὄσσε (which has an analogical -e). I think that the expected reflex of *-iH1 would be *-yā in view of AB tās- < *dhH1s- and A -māṃ, B -mane < *-mH1nos (cf. Klingenschmitt 1975: 161f.). The objection that we find -i- in the optative does not hold because this mood was thematicized, as is clear from the 3rd pl. ending B -yeṃ, -yentär, so that the formative -i- may represent *-iH1e-. I agree with Ringe (apud Hilmarsson 1989: 10) that the dual ending *-ä can easily be analogical in the thematic flexion. The expected reflex of the original thematic ending is found in A wu < *dwoH1, we < *dwoi ‘two’, B tai < *toi ‘the two’. On A ti- for *te- in tiṃ ‘these two’, tim ‘those two’ see K063: 319ff. The neuter dual ending *-i palatalized the preceding *w in B kenī(ne) < *gonwi(-ndwo) ‘both knees’, but not in A kanweṃ (which adopted -e- from we < *dwoi). The resulting West Tocharian neuter dual ending -i < *-wi then spread to pwāri(ne) ‘two fires’ because original *-i was lost without a trace after the consonant r, which was unique in resisting palatalization, and further to the Buddhist vocabulary (cf. Hilmarsson 1989: 59). After the loss of final *-ä, the West Tocharian ending -i < *-wi was the only nonzero dual ending outside the pronouns, and I think that this is why it replaced *-u and *-ai in B wi ‘two’, where the absence of palatalization shows that it cannot have been original, cf. A wiki, B ikaṃ ‘twenty’ < *dwi-. If Greek ἀμφί ‘on both sides’ represents *H2nt-bhi (Jasanoff 1976), Toch. A āmpi, B āntpi, antapi ‘both’ reflect *H2nt-bhi-H1e, with similar addition of the regular dual ending as in Greek ἄμφω, Latin ambō < *H2nt-bhoH1. More precisely, the initial vowel and the preservation of -t(ä)- show that *H2nt- was replaced by the reflex of acc.sg. *H2entm at some stage. My colleague Robert Beekes suggests to me that a derivation from *H2en- ‘that’ rather than *H2ent‘face’ is preferable from a semantic point of view. I disagree, not only because A note on the Tocharian dual 157 this leaves the -t- unexplained but also because a development from ‘two face to face, opposite to each other’ actually seems more appropriate to me than from ‘the two beyond, on the other side’, cf. also Greek ἄντα, ἀντίος. The form A āmpuk may reflect *H2nt-bhi-kw(e), with -uk < *-äkw as in yuk, B yakwe ‘horse’, etc. 3. Pedersen’s derivation of B ikäṃ ‘twenty’ < *wiknt (1941: 253) cannot “be safely discarded” (Hilmarsson 1989: 121) because short *i probably caused palatalization in the same way as the other front vowels, as is especially clear from the 3rd pl. ending A -ñc < *-nti. The corresponding B ending -ṃ represents the secondary ending *-nt (cf. K033: 66 on the A zero ending beside -ñc). The sequence *wi gave rise to the B dual ending -i, as was pointed out above. The absence of palatalization in B wase ‘poison’ does not warrant the assumption of a new sound law but rather points to an early borrowing from Indo-Iranian (Skt. viṣám), as is indeed suggested by its specialized meaning and lack of phonetic congruity with Greek ᾿ῑός and Latin vīrus. Thus, I think that PIE *duidkmti lost its *-i on the analogy of *dekmt and *triH2dkomt and developed into A *wikäṃ, B ikäṃ. Though the ending of A wiki looks like the regular nom.pl. ending AB -i < *-eies of the masculines which did not take the nasal suffix -ñ < *-nes, it seems probable to me that it originated as a copy from the particle -pi in such instances as A wiki ṣapi ‘twenty-one’, cf. B ikäṃ ṣe. In any case this -i must be of recent origin because it did not palatalize the preceding consonant. It cannot be a dual ending because there is no such ending in East Tocharian. THE TOCHARIAN IMPERFECT Krause and Thomas list the following Common Tocharian imperfects (1960: 217): 1sg. 2sg. 3sg. 1pl. 2pl. 3pl. ‘to go’ TB yaim yait yai, yey yeyem yaicer, yeycer yeyeṃ, yeṃ TA yem yet yeṣ *yemäs *yec yeñc ‘to be’ TB ṣaim, ṣeym ṣait ṣai, ṣey ṣeyem ṣaicer, ṣeycer ṣeyeṃ, ṣeṃ TA ṣem ṣet ṣeṣ ṣemäs *ṣec ṣeñc These paradigms are generally derived from the PIE optative (e.g. Pedersen 1941: 206, Lane 1953: 279, Pinault 1989: 128, Klingenschmitt 1994: 406, Winter 1994: 294). Other verbs show different formations (cf. Krause & Thomas 1960: 218-221): TB i-imperfects of present stem classes I-IV and VII-XII, e.g. act. 3pl. priyeṃ to pär- ‘to carry’ (thematic present), 1sg. klyauṣim to klyaus- ‘to hear’ (thematic present), mid. 3pl. kraupiyentär to kraup- ‘to collect’ (thematic present), act. 3sg. aiṣṣi to ai- ‘to give’ (sk-present), 3sg. weṣṣi to we- ‘to say’ (sk-present), 3sg. kälpāṣṣi to kälp- ‘to obtain’ (sk-present), 3pl. lkāṣyeṃ to läk- ‘to see’ (sk-present); TB oy-imperfects of present stem classes V and VI, e.g. mid. 3pl. lkoyentär to läk- ‘to see’ (ā-present), act. 3pl. kärsanoyeṃ to kärs- ‘to know’ (nā-present), 1sg. tarkanoym to tärk- ‘to release’ (nā-present); TA ā-imperfects of present stem classes I-XII, e.g. act. 3sg. keñā to ken- ‘to call’ (thematic present, preterit kāk), 3sg. klyoṣā to klyos- ‘to hear’ (thematic present, preterit klyoṣ), 3sg. eṣā, 3pl. eṣār to e- ‘to give’ (s-present, preterit wäs), 3sg. karyā to kar- ‘to laugh’ (ye-present), 3sg. kātäñśā to kātk- ‘to stand up’ (nā-present, preterit kātäk), mid. 3sg. kropñāt to krop- ‘to collect’ (nā-present, preterit kropat); TA ā-imperfects of subjunctive stems, which are very few and hard to analyze; in fact, these forms are probably based on unattested present formations (cf. Lane 1953: 281); TA imperfects derived from the root, which is the root of the present stem in the case of suppletive paradigms. Krause and Thomas mention the following instances (1960: 221, cf. Pedersen 1941: 174): 160 Tocharian (1) act. 3pl. śārsar to kärs- ‘to know’ (nā-present, preterit 3sg. śärs, 3pl. krasar, mid. 3sg. kärsāt), mid. 3sg. śālpat to kälp- ‘to obtain’ (nā-present, preterit 3sg. kälpāt), act. 3pl. cārkar, mid. 3sg. cārkat to tärk- ‘to release’ (nā-present, preterit 3sg. cärk, 3pl. tarkar), mid. 3sg. pārat, 3pl. pārant to pär- ‘to carry’ (thematic present, preterit 3sg. kāmat), act. 3sg. lyāk, 3pl. lyākar to läk- ‘to see’ (ā-present, preterit 3sg. pälkāt), mid. 3pl. śākant to tsäk- ‘to pull out’ (nā-present, preterit 3pl. tsakar, mid. 3sg. tskāt); (2) act. 3sg. craṅkäs, 3pl. craṅkär to träṅk- ‘to say’ (athematic present, preterit 3sg. we, weñā-, 3pl. weñār), act. 3pl. śepär to tsip- ‘to dance’ (athematic present). It is generally assumed that the TA ā-imperfect must be connected with the stative ē-verbs of other Indo-European languages (e.g. Pedersen 1941: 179, Lane 1953: 285), with *-ē- yielding Tocharian -ā- in open syllables (cf. K095: 80). This explains the palatalization of the preceding consonant. The same formation is found in the preterit, e.g. 3sg. TA klyoṣ, TB klyauṣa ‘heard’. While Pedersen does not doubt that these preterits are “dem Ursprunge nach Imperfekta” (1941: 190), Lane thinks that all TA imperfects except ye- ‘go’ and ṣe- ‘be’ are “in origin identical with preterit formations” which are “all derived from IE perfects and aorists” (1953: 278). The latter view cannot be correct because the development of an imperfect into an aorist is commonplace but “der umgekehrte Vorgang, Übergang eines gewöhnlichen Aorists zu imperfektischer Funktion, kommt nicht vor” (Pedersen 1941: 175). In fact, Pedersen argues that the ā-imperfect was preserved beside the i-imperfect in TB 3sg. yamaṣṣa, 3pl. yamaṣṣare, mid. 3sg. yamaṣṣate ‘made’ (1941: 181), with durative meaning, as distinct from the frequentative i-imperfect yamaṣṣi, yamaṣṣitär. Similarly, the TA preterit weñābeside we ‘said’ may be compared with Russian ipf. govoríl “seine Worte waren” beside pf. skazál “er sprach die Worte aus”, and the TA preterit klyoṣ beside the imperfect klyoṣā ‘heard’ with Russian slýšal, “wo wir den Vorgang perfektiv auffassen” (Pedersen 1941: 175). It follows that Tocharian A is more archaic than Tocharian B in the formation of the imperfect. Interestingly, the TA 3sg. preterit endings -ā and zero reflect the original distinction between imperfect and preterit. The former ending is found with ā-imperfects and with preterits of classes IV and V, which are derived from skand n-stem formations, whereas the latter ending is limited to root imperfects, which will be discussed below, and preterits of classes I-III and VI, which reflect perfects and aorists. The difference between imperfect klyoṣā and preterit klyoṣ ‘heard’ is strongly reminiscent of the one between Old Church Slavic imperfect slyšaaše and aorist slyša, with loss of the final vowel in Tocharian A. Elsewhere I have derived the Slavic imperfect from a nominal form in *-ē- followed by the perfect of the verb ‘to be’ (K071). In a similar vein, we may attribute the preservation of the TA imperfect ending -ā to the former presence of a verbal The Tocharian imperfect 161 clitic. It seems probable to me that this clitic has left a trace in the 1sg. ending -wā, perhaps from *bhwē-, which serves to disambiguate the ending from 3sg. -ā. Thus, I think that the TB i-imperfect is at least partly the result of an innovation. As a rule, the i-imperfect is derived from the present stem in the same way as the Tocharian optative is derived from the subjunctive stem. The oy-imperfect is the i-imperfect of stems in a laryngeal. It presupposes an earlier formation in *-o < *-ā-. The former existence of an oy-optative in Tocharian A is indicated by the lack of palatalization in 3sg. tākiṣ ‘were’, kärsiṣ ‘knew’ (cf. Pedersen 1941: 203). The original formation in *-ā- is reflected in TA yā ‘went’ beside i- ‘to go’, TB iyaṃ ‘goes about’, imperfect īyoy, cf. Skt. yā- beside i- ‘to go’, Lith. jóti ‘to ride’. It thus appears that the i-imperfect developed from the iterative use of a present optative and replaced an earlier imperfect which was partly preserved in Tocharian A. In fact, the modal use of the TB present optative has been preserved in a number of instances (cf. Pedersen 1941: 206). The TB i-subjunctive must also be derived from an earlier optative (cf. Lane 1959: 166), e.g. 1sg. wṣiyau ‘(will) stay’, cf. wsaṣṣäṃ ‘stays’. The TB “intensive preterit” is formally a preterit of an i-subjunctive, e.g. 3sg. wṣīya ‘stayed’ or ‘would stay’. Against this background, it is improbable that the Common Tocharian imperfects of the verbs ‘to go’ and ‘to be’ represent original optatives. Such a derivation is also difficult from a morphological point of view because the vocalism of TA ye-, ṣe-, TB yai-, yey-, ṣai-, ṣey- requires *-oi-, which is unattested in Tocharian. The expected reflexes of the PIE optatives of *ei- and *es- in Tocharian would have yielded *iyā-, *ṣyā-, and undoubtedly have joined the ā-flexion and not have been replaced by a unique formation in -ey- on the basis of an unknown model. It is much more probable that the attested paradigms are what one expects them to be, viz. original imperfects. What are the expected reflexes of the original imperfects? Consider the following paradigms: 1sg. 2sg. 3sg. 1pl. 2pl. 3pl. Vedic ¡yam áis áit áima áita ¡yan Indo-European *ēim *ēis *ēit *ēime *ēite *ēient Proto-Tocharian *yēy *yēy *yēy *yēym *yēyc *yēyn << *yāyn This is a perfect reconstruction of the Proto-Tocharian paradigm for the verb ‘to go’. In the case of the verb ‘to be’, the paradigm was evidently remodeled by substituting ṣey- for sg. *yēs, *yē, pl. *yēṣ- < *ēs-. It follows that we can reconstruct the augment in Tocharian, as we can in Germanic (cf. K135: 138) and Celtic (cf. K239: 109f.). 162 Tocharian The derivation of TA ye- and ṣe-, TB yey- and ṣey- from PIE augmented imperfects has led me to propose a new explanation for the origin of the TA root imperfects śārs- of kärs- ‘to know’, śālp- of kälp- ‘to obtain’, cārk- of tärk- ‘to release’, pār- of pär- ‘to carry’, lyāk- of läk- ‘to see’, śāk- of tsäk- ‘to pull out’, craṅkof träṅk- ‘to say’, śep- of tsip- ‘to dance’, all of which seem to reflect a palatalizing long root vowel *-ē-. The development is phonetically regular in the last five verbs (cf. K095: 80) and may be either analogical or conditioned by the following tautosyllabic resonant in the first three. The last two verbs, where *-ēdid not shift to -ā-, joined the flexion of the sigmatic aorist. The long root vowel could have arisen on the analogy of *yēy- and *ṣēy- at an early stage of ProtoTocharian, analogous to the rise of long vowel preterits in Latin and Germanic on the basis of such perfects as *ēd-, *ēm-, *ēp- (cf. K120: 104). However, these forms evidently represent original imperfects which became root aorists by a differentiation between present and aorist stems (cf. K239: 155). Thus, TA lyāk ‘saw’ and TA pārat ‘carried’ reflect an earlier stage than Latin lēgī and Gothic bērum (cf. also Pedersen 1941: 176, Lane 1953: 282). The theory advanced here also provides an explanation for the discrepancy between the causative (class II) preterits of Tocharian A and B, which must be derived from the PIE reduplicated aorist (cf. Pedersen 1941: 187, Krause & Thomas 1960: 244), e.g. TA 3sg. cacäl to täl- ‘to raise’, 2sg. śaśrāṣt to tsär- ‘to split’, 3sg. lyalyäm, lyalymā- to läm- ‘to set’, which point to *tētl-, *dēdr-, *lēlm-, also 3sg. kakäl to käl- ‘to endure’ with restoration of the initial consonant (cf. Winter 1980b: 561), but TB 3sg. cāla < *tēl-, 3sg. tsyāra with recent tsy-, mid. 3sg. lyāmate < *lēm- with recent stress on the initial syllable. The latter forms are clearly innovations and may have been modeled after the root imperfects when these had become preterits in Tocharian B. The reduplicated formation survived in TB 3sg. yaika < *wēwik-, TA wawik ‘removed’, and in the participle, e.g. TB ñeñmu < *nēnm-, TA nanmu to näm- ‘to bow’, also TB śceścamu, TA śaśmu < *stēstm- ‘put’, cf. with a restored initial consonant TB kekamu, TA kakmu < *gēgm- ‘come’, as opposed to TB papeku < *pāpeku < *pēpokwōs, TA pakku < *pēpk- ‘boiled’. It appears that the reduplication vowel is TB -e-, TA -a- in original aorists and TB -a-, TA -ā- in original perfects, cf. TB paspārttau, causative peṣpirttu < *(s)pēsp(e)rtwōs ‘turned’, but TA sāspärtwu, causative sāspärtwṣu with analogical vocalism. It follows that TB paspārttau may represent *-spertw- < *-sportw-, with the same lowering as in the TA imperfects śārs- ‘knew’ and cārk- ‘released’. When the imperfects became preterits, they formed participles with the reduplication vowel of the aorist, not the perfect, cf. TB lyelyku to 3sg. lyāka ‘saw’, keklyauṣu to klyauṣa ‘heard’, weweñu to weña < *woknē ‘said’, TA kaklyuṣu, wewñu. As this reduplication is irregular from a synchronic point of view (cf. Krause & Thomas 1960: 238, 242), it follows that aorist and perfect were separate categories when these participles were created. The imperfective aorists which developed from earlier imperfects may have The Tocharian imperfect 163 been preserved as a distinct category in Proto-Tocharian, as they were in Slavic, e.g. OCS slyša, TA klyoṣ ‘heard’. When the root imperfects were ousted by i-imperfects in Tocharian B, they resembled the reduplicated preterits formally by the palatalization of the initial consonant and semantically by their opposition to the simple preterits. As Pedersen has pointed out, the meaning of the Indo-European reduplicated aorist is not only causative but also that of “einer als abgeschlossen vorgestellten Wiederholung”, i.e. an imperfective aorist, e.g. Greek (ἐ)πέπληγον “brachten durch wiederholtes Schlagen zustande” (1941: 177). The former root imperfect therefore provided a suitable model for an elimination of the reduplication in the preterit, e.g. TB cāla ‘raised’, śārśa ‘let know’, TA cacäl, śaśärs. The original root imperfect was preserved as a preterit in TB lyāka, lyakā- ‘saw’ (not ‘let see’), lyawā- ‘rubbed’. I agree with Lane that “the long-vowel preterits and imperfects in question have in origin nothing to do with the reduplicated forms” but disagree with his view of “a developing category of imperfects” in Tocharian A (1953: 282f.). The accentuation of the causative preterit in Tocharian B shows that it is a recent formation. VESTJYSK STØD, ICELANDIC PREASPIRATION, AND PROTO-INDO-EUROPEAN GLOTTALIC STOPS 1. In his monograph on the vestjysk stød, Kristian Ringgaard concludes “that the v-stød is only found immediately before the plosives p, t, k, and that it is found wherever these stand in an original medial position, following a voiced sound in a stressed syllable. The exceptions to this are certain types of loanwords from a later period” (Ringgaard 1960: 10, 195). “The v-stød is a complete occlusion of the vocal chords, combined with the diaphragm’s movement of inhalation, equalizing the difference in pressure, and caused by an attempt to pronounce pure and unaspirated fortes plosives when medial” (Ringgaard 1960: 199), i.e., a full-fledged glottal stop. Ringgaard dates the rise of the vestjysk stød to the 12th century because it is characteristic of “all then existing medial plosives”. The view that the vestjysk stød originated from a phonetic development of fortes plosives in medial position was already put forward by Anders Pedersen (1912: 42), who compared the development with the rise of preaspiration in Icelandic. It is an elaboration of Axel Kock’s hypothesis that the vestjysk stød represents “en ljudaffektion, som inträtt vid tenues i vissa ställningar” (Kock 1891: 368 fn., similarly 1901: 26 fn.). Accepting the thesis that the rise of the vestjysk stød has nothing to do with accentuation or apocope, we are faced with three conflicting theories on its origin: T1: Ȥt < ht (A. Pedersen 1912: 42). This view is accepted by H. Pedersen (1942: 119). T2: Ȥt < t (Jespersen 1913: 23). This view is accepted by Ringgaard (1960: 108). T3: Ȥt < tt (Skautrup 1928: 45). This view is explicitly rejected by Hansen (1943: 135). The difference between these three proposals must not be exaggerated. All of them start from the conviction that the origin of the glottal stop must be sought in a series of fortes plosives. The disagreement concerns the phonetic probability of the proposed developments. Ringgaard unjustly repeats Jespersen’s objection to Pedersen’s view: both the rise of glottalization from preaspiration and the rise of preaspiration from a preceding glottal stop are well attested, e.g. in Burmish languages (cf. Bradley 1979: 127-131). Ringgaard’s own view that the vestjysk stød is a spontaneous innovation of the western dialects can hardly be called an explanation. Moreover, it does not account for the parallel development of preaspiration in Icelandic. 166 Germanic phonology 2. Preaspiration is found not only in Icelandic, but also in Faroese, West Norwegian, and the Gaelic dialects of Scotland. Phonetically, the preceding vowel is cut short and continued as a whisper; a preceding resonant (m, n, l, r) is partly or wholly unvoiced. The distribution of preaspiration in Icelandic has been clarified in Einar Haugen’s lucid analysis of the phonemic system (1941, cf. also 1958). It turns out that /b, d, g/ differ from /p, t, k/ in the absence of preaspiration, except in initial position, where they are distinctively unaspirated. The opposition is neutralized after fricatives and after long vowels, where preaspiration is lost. All obstruents are usually voiceless. The same distribution is found in the Norwegian dialect of Jæren (Oftedal 1947). Preaspiration is also attested in Hallingdal, northern Gudbrandsdal, Trøndelag, and even Herjedal. We can conclude that it must have been common to a much larger area and that it is “an example of a feature taken to Iceland by the original settlers” (Chapman 1962: 85). Carl Marstrander has argued that the preaspiration in Scottish Gaelic is due to a Norse substratum (1932: 298). He advances the hypothesis that the Norwegian preaspirated stops represent a retention of the clusters hp, ht, hk, which developed into geminates elsewhere (302). He adduces Middle English haht, saht, slahter < ON hǽtta ‘danger’, sátt, sǽtt ‘agreement’, sláttr ‘mowing’ as evidence for the view that the cluster ht had not yet merged with the geminate tt in the 10th century. His theory implies three developments: D1: tt < ht in East Norse; D2: ht < tt in West Norse; D3: ht < t in West Norse in those positions where the preaspirated stop does not reflect a cluster, e.g. Icelandic epli ‘apple’, vopn ‘weapon’, opna ‘open’, gutl ‘dabbling’, vatn ‘water’, batna ‘improve’, mikla ‘increase’, þukla ‘touch’, teikn ‘token’, líkna ‘show mercy’, hjálpa ‘help’, verpa ‘throw’, elta ‘pursue’, erta ‘tease’, fólk ‘people’, verk ‘work’. Here the preaspirated stop appears to be the phonetic reflex of a Proto-Indo-European unaspirated voiced plosive. 3. Elsewhere I have argued that the reconstructed voiced plosives of the IndoEuropean proto-language were actually glottalic and that the glottalization has been preserved in Balto-Slavic, Armenian, and Indo-Iranian, and has left traces in Greek and Latin (K032, K060). Both the vestjysk stød and the Icelandic preaspiration receive a natural explanation if we assume that Early ProtoGermanic, like Proto-Balto-Slavic and Proto-Indo-Iranian, possessed a series of preglottalized voiced stops. Devoicing yielded a series of Late Proto- Germanic sequences Ȥp, Ȥt, Ȥk. Syllable-final glottal stop was lost. Subsequently, weakening of the glottal stop in West Norse yielded preaspiration, while its assimilation to the following plosive gave rise to a series of geminates in East Norse, with the exception of Danish, where the plosives were subject to lenition and the glottal stop was preserved in the westernmost dialects. I wonder if Swedish vecka Vestjysk stød, Icelandic preaspiration, etc. 167 ‘week’, droppe ‘drop’, skepp ‘ship’ reflect a dialect that escaped the earliest loss of the glottal stop. One may wonder if preglottalization had been preserved in word-initial position in Late Proto-Germanic. There is positive evidence for this idea in the vestjysk stød of fattig < fátǿkr ‘poor’ < ‘few-taking’ and sytten ‘seventeen’. Apart from the straightforward explanation of the vestjysk stød and the Icelandic preaspiration, the theory advanced here has the advantage of accounting in a principled way for the existence of several layers of gemination, which can now be viewed as retentions rather than innovations: G1: mp, nt, nk yielded pp, tt, kk in the larger part of Scandinavia. The nasal consonant was apparently devoiced by the laryngeal feature which preceded the plosive, and subsequently lost its nasalization. G2: k yielded kk before *j and *w. Similarly, t yielded tt before *j in a limited area, e.g. Swedish sätta ‘set’. The development cannot easily be identified with the change of g into gg before *j because the latter involves the transformation of a fricative into a plosive. Modern Icelandic /b, d, g/ are lengthened after a short vowel before n and l, e.g. rigna ‘rain’, sigla ‘sail‘, with half-long g. West Germanic geminated all consonants except r before *j. G3: p, t, k yielded pp, tt, kk before r and l in West Germanic. The same development is found sporadically in Scandinavia. Here again, the geminate may have originated from the assimilation of a glottal stop to the following plosive. 4. It is possible that the theory put forward here has certain consequences for the interpretation of the West Germanic material. Firstly, the High German sound shift may have resulted from a lenition of the plosives with concomitant oralization of the preceding glottal stop. If this is correct, the glottalization must have been preserved at the time of the shift. Secondly, the absence of aspirated stops from Dutch and Frisian may be due to an early loss of preglottalization in this area. Thirdly, the English glo’ttal sto’p may be much more ancient than is commonly assumed. It appears that these possibilities merit further consideration. The evidence for the preservation of the Proto-Indo-European glottalic obstruents in Proto-Germanic supports the hypothesis that there also was a glottal stop of laryngeal origin. Oralization of the latter yielded k before *w in the following instances (cf. Austin 1946, 1958; Lehmann 1965: 216): 168 Germanic phonology – OE haccian ‘hack’ < *kaHw- next to hēawan ‘hew’ with laryngeal metathesis. – OE leccan ‘moisten’ < *laHw-. – OE naca ‘boat’ < *naHw-. – OE cwic ‘alive’ < *gwiHw- < *gwHiw-. – ON skeika ‘swerve’ < *skaiHw- < *skaHiw-. – OHG speichaltra ‘spit’ < *speiHw- < *speHiw- next to spīwan with further laryngeal metathesis. – OE spic ‘fat’ < *spiHw- < *spHiw-. – OE staca ‘haystack’ < *staHw-. – OE tācor ‘brother-in-law’ < *daiHw- < *daHiw-. There is no evidence for a similar development before *j, where a laryngeal lengthened the preceding vowel, e.g. OHG tāju ‘suck’. The rise of -ug- from antevocalic *-uH- must be explained as a secondary development (cf. Winter 1965: 198). The intervocalic sequences *-wH- and *-jH- yielded *-ww- and *-jj- in Proto-Germanic. PROTO-GERMANIC OBSTRUENTS Ten years ago I published a new reconstruction of the Proto-Indo-European system of obstruents, arguing for the absence of an original distinction between voiced and voiceless phonemes in the proto-language (K032). According to the conception outlined there, the historically voiced plosives were earlier aspirated and glottalized lenes, and the rise of an opposition between voiced and voiceless obstruents was a common innovation of all branches except Anatolian and Tocharian, resulting from the loss of aspiration in the north and glottalization in the south of the Indo-European language area. In subsequent years I argued that there is evidence for later preservation of the original glottalized plosives not only in Balto-Slavic and Armenian, as I had done earlier, but also in IndoIranian, Greek, Italo-Celtic, and Germanic (e.g. K075). This obviously has consequences for the reconstruction of the Proto-Germanic system of obstruents, which will be discussed in the present contribution. As I pointed out in my article on the PIE obstruents (K032: 111f.), there is no evidence that the original aspirated plosives ever yielded fricatives in ProtoGermanic times (cf. already Meillet 1908: 89f.). The major indications which are relevant in this connection are the following: (1) A comparative analysis of the Scandinavian evidence points to original plosives in word-initial position (cf. Einarsson 1941: 43ff.). (2) In Old English, initial /g/ was a plosive before consonants and back vowels (Moulton 1954: 24), so that the rise of a fricative before front vowels must have been posterior to the palatalization. (3) There is no Old Saxon spelling evidence for a fricative pronunciation of initial /g/, which can be established on the basis of the Middle Low German and modern dialectal evidence only (Moulton 1954: 32; Dr Norbert Voorwinden informs me that Iosef alliterates with God and good in Old Saxon, which renders this point immaterial). (4) In Old Norse, the preterit suffix of weak verbs with a stem ending in l or n is a dental plosive if the preceding syllable is heavy, but a dental fricative if the preceding syllable is light, e.g. deilda ‘divided’, kenda ‘taught’ (inf. kenna), valða ‘chose’, vanða ‘accustomed’. The simplest way to account for the difference is the assumption that an intervocalic plosive *d became a fricative at the stage between the first and the second syncope. Thus, the syncope in *dailidō yielded *däildō, the intervocalic development of the dental plosive in *walidō then yielded *waliðō, the syncope in the latter form yielded *walðō, and the shortening of final vowels eventually yielded the attested forms deilda and valða. 170 Germanic phonology If this reconstruction of the relative chronology is correct, it follows that the intervocalic dental plosive was preserved at the time of the earlier syncope. (5) Intervocalic *ð was lost before r when the intervening vowel was syncopated in Proto-Norse, e.g. fiórir ‘four’, hvárir ‘which’, Gothic fidur-, ŵaþarai. Since the cluster was not simplified in veðr ‘weather’ < *uedhrom, Russ. vëdro, the dental plosive in this word had apparently not yet become a fricative at the time of the younger syncope. Thus, the rise of a fricative in veðr was evidently posterior to the loss of the fricative with compensatory lengthening in fiórir and hvárir, which was in its turn posterior to the syncope in *fiður- and *hwaðar-. (6) Old Norse batt ‘bound’, helt ‘held’, Gothic haihald point to preservation of the plosive before the apocopated ending *-e. The final fricative of Gothic faifalþ ‘folded’ was assimilated to the preceding resonant in ON fell ‘covered the head’, as it was in fimm ‘five’; the forms hell and felt are analogical. If the obstruent *d had been a fricative at the time of the apocope, it would have yielded -þ in Gothic and have been assimilated to the preceding resonant in Old Norse. It follows that the fricative of Gothic anabauþ ‘ordered’, ON bauð from *bhoudhe is an innovation while the corresponding dental plosive of West Germanic must be an archaism. Since the Proto-Germanic voiced obstruents have a twofold origin, their reconstruction as plosives poses a chronological problem with respect to Verner’s law. There are two possibilities: (1) Verner’s law was posterior to the rise of the voiceless fricatives f, þ, x which resulted from Grimm’s law. This is the usual view. It implies a development *t > *þ > *ð > d > ð, e.g. in English father, OE fæder, ON faðir, Greek πατήρ. Since the dental obstruent was preserved in ON dat.sg. feðr, Greek πατρί, it must have been a plosive at the stage discussed under (5) above (cf. K032: 113). The objection that the dental obstruent may have been restored in this form does not hold in view of the word leðr ‘leather’, Old Irish lethar, where such a restoration is impossible. (2) Verner’s law was anterior to the rise of voiceless fricatives which resulted from Grimm’s law. This is Vennemann’s view (1984). I think that it is correct. In my earlier discussion I stuck to the traditional view, mainly because of ON enn ‘still’, which cannot be separated from endr ‘again’, OHG enti ‘earlier’ (K032: 113). The assimilation in ON enn points to a fricative, as in fimm and fell. The antinomy is resolved if we derive enn from acc.sg. *anþu, Greek ἄντα, and assume that the umlaut was analogical, while OHG enti can be identified with Greek ἀντίος. There is no evidence for a root-stressed loc.sg. form *anþi beside *andi (Greek ἀντί) and *unþē in Germanic, in spite of Lühr’s effort to prove the contrary (1979). If we identify ON enn with Skt. ánti, the plosive of the common 171 Germanic prefix and- remains unexplained. (For the stress of *unþē cf. Gathic inst.sg. ašī < *ṛɴtī ‘reward’ and Beekes 1985: 197. Perhaps we do have to assume a root-stressed paradigm on the strength of enn and *unþē and a derivative to account for endr and OHG enti. If this is correct, it remains unclear why the prefix and- adopted the plosive of the derivative.) The hypothesis that Verner’s law was anterior to Grimm’s law has several advantages beyond the points which have been mentioned thus far. It provides a simple chronology for the Runic forms fadir (Strö), faþir (Rök), ON faðir (Lehmann 1986: 101). The rise of the younger futhark was evidently conditioned not by the voicing of the fricatives f, þ, x, but by the loss of occlusion in the plosives b, d, g. Moreover, the total merger of the preterit formative *-dh(ē)- with the participial formative *-t(o)- is best explained by an early phonological merger of the two obstruents, followed by a long period of analogical adjustments. There can be no doubt that the analogical processes operated in both directions. On the one hand, the suffix of OE gehæfd ‘had’, gesægd ‘said’, gehygd ‘mind’, Gothic gahugds was taken from the dental preterit. The original consonant was preserved in OE hæft ‘captive’, Gothic andahafts ‘answer’. On the other hand, the perfect presents (“praeterito-praesentia”) created a preterit on the basis of the participle, e.g. OE cūðe ‘could’, cūð ‘known’. Rückumlaut preterits are also a creation on the basis of the participle, e.g. OE worhte ‘worked’, þūhte ‘seemed’. They replace earlier perfects (strong preterits), as is clear from the root vowel of warhte beside worhte, also þōhte beside þūhte, with secondary transfer to the paradigm of þencan ‘think’ on the analogy of sōhte, sēcan ‘seek’. The original perfect was preserved in brēac, brūcan ‘use’, which adopted the regular ending of the strong verbs in the participle brocen, cf. Gothic bruhta ‘used’ and OE coren ‘chosen’ replacing earlier cost. The ablaut was most widely preserved in the case of Gothic brahta, briggan ‘bring’, OE brōhte, bringan, OHG brāhta, bringan. The hypothesis that Verner’s law was anterior to the rise of the voiceless fricatives is compatible with the early rise of Proto-Germanic voiced plosives as a result of the dialectal Indo-European loss of aspiration. Both developments are at variance with the traditional reconstruction of PIE simple voiced stops yielding Proto-Germanic simple voiceless stops which were subject to a number of gemination processes in various Germanic language areas and shifted to affricates in Old High German. We must therefore reconsider the phonetic character of the Proto-Germanic voiceless plosives. The usual reconstruction of simple voiceless plosives for Proto-Germanic does not account for the multifarious reflexes which we find in the daughter languages. The following developments are of major importance: (1) Preaspiration in Icelandic, e.g. in epli ‘apple’, opna ‘open’, vatn ‘water’, batna ‘improve’, mikla ‘increase’, teikn ‘token’, verpa ‘throw’, elta ‘pursue’, verk ‘work’. 172 Germanic phonology These examples show that the preaspirated stops do not reflect clusters but directly represent the voiceless plosives of Proto-Germanic. Since the same reflexes are found in the Norwegian dialect of Jæren (cf. Oftedal 1947), preaspiration is an inherited feature in these words. (2) Preglottalization in the western dialects of Danish: the so-called vestjysk stød (cf. Ringgaard 1960). The classic view that it represents “en ljudaffektion, som inträtt vid tenues i vissa ställningar” (Kock 1891: 368 fn.) does not explain the rise of the glottal stop. (3) Gemination in Swedish, e.g. in vecka ‘week’, droppe ‘drop’, skepp ‘ship’, cf. ON vika, dropi, skip, OE wice, dropa, scip, Finnish viikko. This gemination is unexplained. (4) Assimilation of mp, nt, nk to pp, tt, kk in the larger part of Scandinavia. The nasal consonant was apparently devoiced by the preaspiration of the following plosive and then lost its nasal feature. (5) Gemination of k before j and w, e.g. ON lykkja ‘coil’, bekkr ‘brook’, nNjkkvi ‘boat’, røkkr ‘dark’. Similarly, gemination of t before j in a limited area, e.g. Swedish sätta ‘set’. (West Germanic geminated all consonants except r before j and is therefore inconclusive.) (6) Gemination of p, t, k before r and l in West Germanic. The same development is found sporadically in Scandinavia; this suggests that we have to do with the loss of an archaic feature (such as preaspiration) rather than with an innovation. In Icelandic, preaspiration is lost before r and preserved before l (Haugen 1941: 101). (7) Standard English inserts a glottal stop before a tautosyllabic voiceless plosive, e.g. lea’p, hel’p (Brown 1977: 27). There is no reason to assume that this is a recent phenomenon. (8) The High German sound shift yielded affricates and geminated fricatives, e.g. OHG pfad ‘path’, werpfan ‘throw’, zunga ‘tongue’, salz ‘salt’, kind, chind ‘child’, trinkan, trinchan ‘drink’, offan ‘open’, wazzar ‘water’, zeihhan ‘token’. These reflexes suggest a complex articulation for the Proto-Germanic voiceless plosives from which they developed. In the traditional theory, the origin of the gemination is unexplained. Note that the High German sound shift has a perfect analogue in the English dialect of Liverpool, where we find e.g. kx in can’t, back (Hughes and Trudgill 1987: 66). The developments listed here receive a natural explanation if we start from the system of obstruents which must be reconstructed for Balto-Slavic. In this branch of Indo-European, the unaspirated lenes of the proto-language are reflected as voiced plosives preceded by a glottal stop, e.g. Latvian pêʚds ‘footstep’, 173 nuôgs ‘naked’ from *pe’d-, *no’g-, cf. OE fōt, nacod. (The circumflex accent denotes a glottal catch.) When the voiceless plosives were lenited to fricatives in Germanic, voicedness was lost as a distinctive feature. This is my reformulation of Grimm’s law. The Proto-Germanic system of obstruents, which lacked voiced phonemes, has been preserved largely unchanged in modern Icelandic (and in the Norwegian dialect of Jæren), except for the fact that relaxation of the glottal stop yielded preaspiration (cf. Haugen 1941). The original glottal stop was preserved in the western dialects of Danish in spite of the general lenition of obstruents characteristic of this language. It was assimilated to the following plosive in Swedish vecka, droppe, skepp, sätta. It devoiced and subsequently eliminated a preceding nasal consonant in the larger part of Scandinavia, e.g. ON drekka ‘drink’. It was assimilated before kj and kw in ON lykkja, nNjkkvi, and before p, t, k plus resonant in West Germanic. It was preserved before a tautosyllabic plosive in modern English. It was oralized and after a vowel assimilated when the following plosive was lenited to a fricative as a result of the High German sound shift, a development which was probably arrested by the loss of the glottal stop in the Low German area. I find no evidence for preservation of the glottalic feature in Gothic. Unlike the other Germanic languages, Gothic appears to have developed a distinction between voiced and voiceless phonemes at an early stage, probably under the influence of its non-Germanic neighbours. The new voiced plosives developed fricative allophones, which still remained strictly distinct from the voiceless fricatives in Wulfila’s days, as Roberge has shown (1983). There is a trace of the Proto-Germanic absence of voiced obstruents in the Gothic words Kreks ‘Greek’ and dat.pl. marikreitum ‘pearls’, which were apparently borrowed from Latin Graecus, margarīta at a stage when no voiced plosives were available. As a summary, the following schematic representation illustrates my view of how the system of obstruents developed from Proto-Indo-European to Gothic: I. Proto-Indo-European glottalized fortis lenis plain t: t’ aspirated th II. Dialectical Indo-European (Proto-Balto-Slavic) glottalized plain voiceless t voiced ’d d III. Proto-Germanic (Proto-Norse, Proto-English, Proto-German) plosive fricative fortis ’t lenis t þ 174 Germanic phonology IV. Gothic voiceless voiced plosive t fricative þ d Verner’s law must be dated to stage II. It yielded a voiced variant of the PIE fricative *s, which became phonemic when the stress was fixed on the initial syllable of the word. The rise of r from *z may be viewed as a consequence of the loss of voicedness as a distinctive feature which Grimm’s law entailed in the theory advocated here. It appears that the rephonemicization of voicedness in Gothic forestalled the rhotacism. I am inclined to date the fixation of the stress before Grimm’s law in Gothic and after Grimm’s law in the other Germanic languages. This chronological difference explains a number of seemingly independent characteristics of Gothic: (1) the preponderance of barytone forms, (2) the elimination of Verner alternations, and (3) the absence of rhotacism. KLUGE’S LAW AND THE RISE OF PROTO-GERMANIC GEMINATES According to Friedrich Kluge (1884: 174) “trat nach Vollzug des zweiten lautverschiebungsaktes, d.h. nach der Wirkung des Verner’schen gesetzes, die angleichung der betonten n-suffixe an die vorausgehenden tönenden laute ein. Nachdem sich so das gebiet der tönenden verschlusslaute erweitert hat, tritt der letzte Verschiebungsakt ein, wodurch alle medien, einfache wie geminierte zu [tenues] werden.” Though Kluge was not the first to propose the assimilation of suffixal *n to a preceding consonant in order to explain the rise of ProtoGermanic *kk, *tt, *pp, his presentation is rightly acknowledged as the canonical view in the later discussion of the problem. It therefore seems appropriate to call his rule “Kluge’s law” (thus already Kauffmann 1887). The rise of the voiceless geminates belongs to the most debated issues in Germanic linguistics. It has received a comprehensive treatment in Rosemarie Lühr’s Habilitationsschrift (1988), an impressive piece of work in the best German tradition (with 44 pages of bibliography and 2493 footnotes; I have noticed very few printing errors, mainly subscript dots missing in Indic forms, pp. 199, 200, 312). After a detailed discussion of earlier treatments in terms of “expressive gemination” and “expressive nasalization” in the first two chapters of her book, the author concludes (188): “Es steht nichts im Wege, die Gemination oder den Konsonantenwechsel in den meisten der sogenannten ‘Gefühlswörter’ auf die gleiche Weise zu deuten wie im Falle von Wörtern neutralen Gefühlswertes.” I agree whole-heartedly; it follows that Fagan’s theory (1989) must be rejected. In the central chapter of her book (189-215), Lühr presents her own theory, which basically conforms to Kluge’s. She lists the following crucial arguments (191): “1) Die Bedeutung der meisten Nomina mit Doppeltenuis oder Konsonantenwechsel läßt keine expressive, lautnachahmende oder Intensität beziehungsweise Iteration ausdrückende Lautgebung vermuten. 2) Die Doppelobstruenten treten vor allem in n-Stämmen auf, was in der Flexion dieser Stämme begründet ist. 3) n-Stämme mit *ll < *l-n, *nn < *n-n verhalten sich morphologisch wie die n-Stämme mit Doppeltenuis. 4) Der Umstand, daß ein und dasselbe Wort Doppeltenuis aufweist und als u-Stamm flektiert, ist bei der Annahme einer nicht lautgesetzlichen Entstehung der Doppeltenues nicht erklärbar.” These arguments decide the issue. In the last two chapters, Lühr adduces lists of nouns and verbs and analyzes the material. She derives the intensive and iterative verbs with voiceless geminates from factitives on the basis of deverbal 176 Germanic phonology adjectives in *-na- < *-no-, e.g. German bücken ‘stoop’, Skt. bhugná- ‘bent’, not from nasal presents. This looks very convincing (but see now Kroonen 2009: 47). Comparing Lühr’s final text with her preliminary statement (1980), I find myself in agreement with her earlier rather than later analysis of the phonetic development involved. First of all, Kluge limited the assimilation rule to stressed nasal suffixes in view of such instances as Gothic aþn ‘year’, auhns ‘oven’.1 According to Lühr’s more recent view (1988: 192) “erscheint es ratsam, den Akzent bei der Beschreibung der n-Gemination außer Betracht zu lassen, auch wenn sich mit Hilfe des Akzentes eine Reihe von Gegenbeispielen leichter erklären ließe.” I certainly disagree. Secondly, Lühr reformulates Kluge’s assimilation rule as gemination before *n and subsequent loss of the nasal. There is no material evidence for this view, which is based on an aprioristic theory of syllable structure.2 I think that Kluge’s formulation quoted above is superior to any modification which has been proposed since. The relative chronology of Kluge’s law poses a problem which has not been solved yet. On the one hand, the rise of the new geminates was posterior to Verner’s law because it affected the voiced reflexes of the ΡIΕ voiceless plosives in the same way as the original aspirates (thus already Paul 1880: 133 fn., which Lühr does not mention). On the other hand, the devoicing of the geminates suggests that it was anterior to Grimm’s law, or at least to the “Medienverschiebung”, as Kluge pointed out already. The logical conclusion is that Verner’s law preceded Grimm’s law, a chronology for which there is other evidence as well (cf. K102: 5f.). Those who refuse to accept this conclusion can choose between two alternatives. Either they may assume that the voiced fricatives which resulted from Verner’s law assimilated a following *n and became plosives (Kluge 1884: 175) or became plosives before *n which was then assimilated (Lühr 1980: 259) before devoicing. Or they may assume that “zwischen der urgermanischen und der hochdeutschen lautverschiebung ausser der durch Verner aufgeklärten noch eine weitere Verschiebung liegt, durch welche die lange verschlusslenis zur verschlussfortis verschoben wird” (Paul 1880: 133 fn.), “wobei die verdoppelten stimmhaften Reibelaute in der Gemination zunächst zu Verschlußlauten und dann wie im Oberdeutschen zu stimmlosen Verschlußlauten wurden” (Lühr 1988: 196), which means that it was “eine von der germanischen Lautverschiebung unabhängige spätere Erscheinung”. Both solutions are unsatisfactory. The problem can be resolved if we start from the reconstruction of the Proto-Germanic system of obstruents which I have presented earlier (K102). In my view, the ΡIΕ aspirates lost their aspiration in dialectal Indo-European times already. They remained distinct from the “mediae” (unaspirated lenes) because the latter were preglottalized, as is clear from their reflexes in Balto-Slavic. After Verner’s law had “das gebiet der tönenden verschlusslaute erweitert” (as Kluge 177 put it), the remaining voiceless plosives were lenited to fricatives and voicedness was lost as a distinctive feature. This is Grimm’s law. The resulting system of obstruents is preserved largely unchanged in modern Icelandic, except for the development of preglottalization into preaspiration, e.g. in epli ‘apple’, mikla ‘increase’, verk ‘work’. We can now date Kluge’s law between Verner’s law and Grimm’s law: this enables us to explain the attested reflexes from well-known developments without any additional assumptions. The range of Kluge’s law must not be underestimated. Lühr writes (1988: 250): “Nun erscheinen auch bei dem Wort Schiff Lautungen, die auf ein *pp weisen; diese sind in den etymologischen Wörterbüchern im allgemeinen nicht erwähnt”, and she cites a number of forms on the basis of which “ist für das Althochdeutsche neben *skipa- ein *skippa- n. vorauszusetzen. [...] Demgegenüber weisen alle übrigen germanischen Sprachen allein auf eine Vorform mit einem urgerm. *p.” The geminate is actually corroborated by Swedish skepp ‘ship’, which relates to ON skip as Swedish droppe ‘drop’, vecka ‘week’ relate to ON dropi, vika and can safely be taken to be the reflex of *skibna< *skipnó-, cf. Lith. skiẽpas ‘graft’. It would be interesting to reconsider the whole problem of East Norse gemination against this background. Similarly, the rootfinal consonant of the word for ‘deep’ was probably taken from *dubna- < *dhubhnó-, Russ. dno ‘bottom’, cf. Lith. dubùs ‘deep, hollow’, with a short vowel pointing to *bh, not *b, also MHG topf ‘pot’ < ‘vas profundum’ (Lühr 1988: 232f. and 349f.).3 One can only hope that Kluge’s law will now be recognized as an integral part of our knowledge in the standard treatments of Germanic historical linguistics. NOTES Cf. already Sievers 1878: 149 fn., which Lühr does not mention in her book, and Osthoff 1882: 300 fn. The formulation that “nicht sicher ist, ob ein im Frühurgermanischen auf ein *-n- folgender Akzent die n-Gemination verhindert hat” (Lühr 1988: 331) is of course a slip of the pen. 1 On the detrimental influence of preconceived theoretical ideas on the investigation of the material see my discussion (K104) of Vennemann’s application of this theory to Armenian data. 2 Unfortunately, the usefulness of Lühr’s book is reduced by the limited character of the index. Thus, the page numbers given here can only be found through the words Topf and tupfen, respectively, not through tief, OE doppettan, or the Celtic, Baltic or Slavic cognates cited in the text. 3 LABIALS, VELARS AND LABIOVELARS IN GERMANIC The eminent Indologist F. B. J. Kuiper has emphasized the necessity to distinguish between at least three different layers of pre-Germanic substratum (1995): (A1) “European” words with a-vocalism and frequent voiced aspirates, e.g. *bhabh- ‘bean’, *bhardh- ‘beard’, *bhar(e)s- ‘barley’,*ghasdh- ‘rod’, perhaps *sam(a)dh‘sand’. (A2) Germanic words with a-, i-, u-vocalism, initial clusters kn-, kl-, and prenasalization, which is reflected as variation between -b-, -mb-, -mp-, -pp-, e.g. ON klífa ‘climb’, klif, kleppr, kleif, ‘cliff ’, OE climban, clif. (A3) “Old European” words with frequent a-vocalism, few consonants and lots of suffixes, reflected in the European hydronymy. Theo Vennemann has presented considerable evidence for the view that this substratum language is cognate with Basque (1994a). Kuiper goes on to discuss two Proto-Germanic words which may belong to layer A1, viz. *babnō ‘bean’, which he assumes to have replaced *babō < *bhabhā reflected in Latin faba and Old Prussian babo,1 and *babmaz, Gothic bagms ‘tree’, OE bēam, OHG boum. Here he assumes dissimilation to *baunō and *baumaz in West Germanic and dissimilation of *babmaz to *bagmaz in East and North Germanic, cf. Old Swedish bagn ‘trunk’, and concludes that the source language must have had two different root morphemes*bhabh- ‘bean’ and *bhabh- ‘tree’. The latter conclusion seems highly improbable to me. It is much more plausible that we are dealing with two derivatives of a single root *bhabh-, which may have carried the verbal meaning ‘grow’. This leads me to a reconsideration of the possibility that *bhabh- represents the PIE root *bheH3u- in an unknown substratum language which may have branched off at an early stage.2 I have long suspected the existence of a western Indo-European substratum language where the PIE laryngeals are reflected as velar obstruents to a larger extent than they are in Anatolian. It seems appropriate to give vent to my speculations here. As I have indicated elsewhere (K112), I think that the PIE laryngeals developed from Indo-Uralic velar plosives, e.g. PIE*dheH1- ‘put’ < PIU *deka-, Finnish teke- ‘do, make’, and PIE *deH3- ‘give, take’ < PIU *tagu-, Finnish tuo‘bring’, southern Lappish duokĕ- ‘sell’. For Armenian I have argued that PIE *H1eand *Ho- yielded e- and o- whereas *H2e- and *H3e- yielded ha- and ho-, with voiced h-, respectively (K061). If PIE *e yielded a while *H2 and *H3 merged with *gh and *gwh in the western Indo-European substratum language, the expected 180 Germanic phonology correspondences fit Kuiper’s category A1 perfectly. We may then conclude that *bhabh- < *bhagwhw- < *bheH3w- is a natural development. Another example where I suspect an obstruent reflecting a PIE laryngeal is the Gothic name of the s-rune sugil < *suH2l, with laryngeal metathesis in the zero grade of *seH2ul ‘sun’, Latin sōl, cf. Gothic sauil < *sāwel-, ON sól. The alternative hypothesis that Gothic sugil is a dialectal variant of sauil meets with several difficulties. Firstly, there is no evidence for this type of dialectal variation elsewhere in the language. Secondly, there is no evidence for an intrusive -g- in Gothic, as there is in West Germanic, e.g. OE nigon < *newun, geoguð < *juwundō, Gothic niun ‘nine’, junda ‘youth’. Thirdly, it is unclear why a form from an obscure dialect should be used to designate the s-rune. It is more probable that the name represents an archaic symbol. If the name of the s-rune represents a borrowing from a West Germanic dialect (Old English?), these objections are invalid. An example to which my colleague Robert Beekes has drawn my attention is the word for ‘bridge’, OE brycg, OHG brucka, if this is cognate with the word for ‘eyebrow’, OE brū < *H3bhruH2-, Greek ὀφρῦς. Here again, the semantic shift precludes the possibility of simple dialectal variation. Dr Peter Schrijver has drawn my attention to OE sugu ‘sow’, which may be another case in point if it represents the PIE word for ‘pig’, OE sū. However, it seems to me that the velar may have been taken from the verb sūgan ‘suck’ if it was not simply intrusive before vocalic case endings. An example with an initial laryngeal which is reflected as a velar obstruent is Gothic gazds ‘sting’, OHG gart ‘goad’ < *ghasdh- < *H3esd-, Latin hasta ‘pole, spear’, Middle Irish gat ‘withe’. The regular development is reflected in Gothic asts ‘branch’, OHG ast, Greek ὄζος. A possible instance with a voiceless reflex of an initial laryngeal is Slavic kostь ‘bone’, Greek ὀστέον (for alternative proposals see Hamp 1985 and Rousseau 1990), cf. especially the a-vocalism of Welsh asgwrn ‘bone’ beside Latin os. However, the meaning is not favorable for borrowing and Latin costa ‘rib’ rather suggests that we are dealing with two different roots here. A better example is Slavic koza ‘goat’ < *kagʄh- < *H2egʄ-, Vedic aj¡, cf. Lith. ožỹs ‘he-goat’ < *H2egʄ-, Vedic ajás (the long vowel in the Lithuanian word is a result of Winter’s law, according to which an original short vowel became long and acute before an unaspirated voiced plosive in Balto-Slavic). The same root is recognized in Gothic hakuls, OHG hahhul ‘cloak’. Another possible instance with an initial laryngeal is OHG habaro ‘oats’ if this word is cognate with Latin avēna, Lith. avižà, Slavic ovьsъ. It is tempting to derive the Balto-Slavic words from *avikʄdh- < *H3ewi-H1d- ‘fodder’ with different simplification of the final cluster, despite Pedersen’s reluctant attitude.3 However, the Germanic word may not be related because Finnish kakra and Middle Irish coirce, Welsh ceirch suggest that a different root may be involved. 181 A final example with a voiceless reflex of an initial laryngeal and a-vocalism is Latin cabō, caballus ‘horse’, Slavic kobyla < *kabh- < *H1ekw-, Finnish hepo. I find it difficult to separate these words. The possibility of deriving Latin faba ‘bean’ and caballus ‘horse’ from*bheH3w- and *H1ekw- leads me to reconsider the labialization of labiovelars in Germanic. As far as I can see, there has been little progress here since Streitberg’s unsurpassed Proto-Germanic grammar appeared in 1896.4 As a rule, PIE labiovelars were preserved in initial and intervocalic positions in ProtoGermanic but were delabialized before rounded vowels at an early stage. The development of the voiced aspirate *gwh in initial position is unclear: “labialization to b- is improbable, delabialization to g- is unwarranted, assumed development to w- rests on disputable etymologies” (Poloméɴ 1987: 309). Yet I think that the number of good etymologies where a PIE labiovelar is reflected as a labial obstruent in Germanic is too large to be ignored, cf. especially Gothic wulfs ‘wolf ’ < *wḷkwos, fimf ‘five’ < *penkwe, ainlib ‘eleven’, twalib ‘twelve’ < *-likwa, bilaif ‘remained’ < *-loikwe, laiba ‘remnant’< *loikwā, ON lifr ‘liver’ << *yekwṛ, Gothic wairpan ‘throw’ < *wergw-. While the initial labial of Gothic fidwor ‘four’ can easily have been taken from the word for ‘five’, the alleged assimilation of the labiovelar to the initial labial in fimf seems rather problematical to me. It is improbable that the labial feature spread phonetically through the non-labialized vowel. In cases of phonetic influence at a distance, we usually find either regressive assimilation resulting from mental anticipation, as in Latin quīnque ‘five’, or progressive dissimilation, as in Spanish árbol ‘tree’ or in the Germanic instances which Kuiper adduces (1995: 83). I therefore regard the word for ‘five’ as a prime example for the phonetic development of labials from labiovelars in Germanic. In view of the delabialization of labiovelars before rounded vowels in ProtoGermanic (cf. Streitberg 1974: 110f., 134), it is clear that the rise of the labial obstruent in the word for ‘wolf ’ must have been very early. It follows that the delabialization in ON ylgr ‘she-wolf ’ must have been even earlier because it is hardly probable that an alternation between labial and velar obstruent within a single paradigm would have been preserved for a very long period of time. As the labial obstruent of OHG wulpa, OE wylf ‘she-wolf ’ can easily have been taken from the word for ‘wolf ’ before the consonant shift, I arrive at the following relative chronology: 182 Germanic phonology (1) (2) (3) (4) (5) (6) (7) *wḷkwjā- > *wḷkjā-, *wḷkwos > *wulkwos, *wulkwos > *wulpos, *wulkjā- > *wulgjā- (Verner’s law), merger of *a, *ā with *o, *ō, *wulgjō- >> *wulbjō- in West Germanic, *wulpaz > *wulfaz (Grimm’s law).5 On the basis of Gothic wulfs, fimf, wairpan I now assume that the development of the labiovelars into labial obstruents was regular after resonants, and also before resonants in view of ON lifr ‘liver’, ofn ‘oven’, with later delabialization in Gothic auhns, Swedish ugn. The phonetic motivation for this development was provided by the dissolution of the intervocalic labiovelars into heterosyllabic sequences of plain velar plus *w, which is especially clear in the case of Gothic naqaþs ‘naked’, where ON nøkkviðr and OHG nackot show gemination of the velar before *w, and in Gothic riqis ‘darkness’, ON røkkr, and Gothic aqizi ‘a’x, OHG ackus. Thus, I regard the labialization of the labiovelars as a form of cluster simplification. In the case of the root *likw- we may assume the former existence of a nasal present *linkw-, to be compared with Latin linquō ‘leave’. The main objection to the explanation proposed here is the existence of the verbs Gothic siggwan ‘sing’, sigqan ‘sink’, and stigqan ‘thrust’. The latter verb appears to belong to Kuiper’s prenasalizing category A2, cf. ON stinga ‘thrust’, OHG stehhan, stehhen, sticken ‘prick’, stehho, stecko ‘stick’ (but cf. Kroonen 2009: 48f.). The same may be suspected for the other two verbs. The connection of Gothic sigqan with Greek ἑάφθη and Armenian ankanim ‘fall’ is no more than a possibility and the connection of Gothic siggwan with Greek ὀμφή ‘voice, oracle’ is not compelling. No conclusions can be based on Gothic arŵazna ‘arrow’. It is therefore possible that the labiovelars in these words belong to a layer which is younger than the labialization of the PIE labiovelars. NOTES The accentuation of Russian bob ‘bean’ and its cognates in other Slavic languages points to an original root-stressed neuter, which was evidently created as a singulative formation. Note that Old Prussian babo is translated as Bonen in the Elbing vocabulary. 1 The reconstruction of the root as *bheH3u- rather than *bheuH2- is based on the Vedic imperative bodhí ‘be!’, the Old Irish preterit -bá, -boí, -bo ‘was’, Welsh bu, and other evidence (cf. K239: 77-79 and 125). 2 “Hat man denn je Schafe und Böcke mit Hafer gefüttert? Oder is der Name etwa ein Schimpfwort, von solchen Leuten ausgegangen, denen der Haferbrei 3 183 nicht mundete?” (Pedersen 1895: 42). Wild oats may certainly have been part of the livestock’s diet. “Die Bedingungen für den Eintritt dieser Verschiebung der Artikulationsstelle sind bis jetzt noch nicht ermittelt” (Streitberg 1974: 111). 4 On the relative chronology of Verner’s law and Grimm’s law cf. K102. Note that OHG lebara ‘liver’, not **-ur-, shows that the stem extension of this word arose prior to my stage (2). 5 PREASPIRATION OR PREGLOTTALIZATION? In an earlier study (K102) I argued that we have to reconstruct preglottalized stops for Proto-Germanic on the basis of the following reflexes: (1) Preaspiration in Icelandic, e.g. in epli ‘apple’, opna ‘open’, vatn ‘water’, batna ‘improve’, mikla ‘increase’, teikn ‘token’, verpa ‘throw’, elta ‘pursue’, verk ‘work’. These examples show that the preaspirated stops do not reflect clusters but directly represent the voiceless plosives of Proto-Germanic. Since the same reflexes are found in the Norwegian dialect of Jæren (cf. Oftedal 1947), preaspiration is an inherited feature in these words. (2) Preglottalization in the western dialects of Danish: the so-called vestjysk stød (cf. Ringgaard 1960). The classic view that it represents “en ljudaffektion, som inträtt vid tenues i vissa ställningar” (Kock 1891: 368 fn.) does not explain the rise of the glottal stop. (3) Gemination in Swedish, e.g. in vecka ‘week’, droppe ‘drop’, skepp ‘ship’, cf. ON vika, dropi, skip, OE wice, dropa, scip, Finnish viikko. This gemination is unexplained. (4) Assimilation of mp, nt, nk to pp, tt, kk in the larger part of Scandinavia. The nasal consonant was apparently devoiced by the preaspiration of the following plosive and then lost its nasal feature. (5) Gemination of k before j and w, e.g. ON lykkja ‘coil’, bekkr ‘brook’, nNjkkvi ‘boat’, røkkr ‘dark’. Similarly, gemination of t before j in a limited area, e.g. Swedish sätta ‘set’. (West Germanic geminated all consonants except r before j and is therefore inconclusive.) (6) Gemination of p, t, k before r and l in West Germanic. The same development is found sporadically in Scandinavia; this suggests that we have to do with the loss of an archaic feature (such as preaspiration) rather than with an innovation. In Icelandic, preaspiration is lost before r and preserved before l (cf. Haugen 1941: 101). (7) Standard English inserts a glottal stop before a tautosyllabic voiceless plosive, e.g. lea’p, hel’p (Brown 1977: 27). There is no reason to assume that this is a recent phenomenon (cf. K142). (8) The High German sound shift yielded affricates and geminated fricatives, e.g. OHG pfad ‘path’, werpfan ‘throw’, zunga ‘tongue’, salz ‘salt’, kind, chind ‘child’, trinkan, trinchan ‘drink’, offan ‘open’, wazzar ‘water’, zeihhan ‘token’. These reflexes suggest a complex articulation for the Proto-Germanic voiceless plosives from which they developed. In the traditional theory, the origin of the 186 Germanic phonology gemination is unexplained (but see K138). Note that the High German sound shift has a perfect analogue in the English dialect of Liverpool, where we find e.g. kx in can’t, back (Hughes & Trudgill 1987: 66). The reconstruction of glottalization for Proto-Germanic has been challenged by Goblirsch, who claims that this feature “has not been directly attested in Germanic” (1999: 117), disregarding the vestjysk stød and the English glottal stop which I have adduced as direct evidence. He claims that the “main argument in favor of the glottalic theory is a typological one” (1999: 118), disregarding both the comparative evidence which I have adduced in the course of the last 25 years (see especially K075) and the argumentation against the use of typological considerations in linguistic reconstruction (cf. K130). It is simply not true that “there is a nearly complete lack of direct evidence in Germanic and the other branches of Indo-European” (Goblirsch 1999: 119). What can I do but refer to my earlier publications? Goblirsch returns to the traditional reconstruction of voiced rather than glottalized stops for Proto-Indo-European and posits “spirants, phonologically undifferentiated as to voice” for the traditional Indo-European voiced aspirates (1999: 120), disregarding the comparative evidence from Germanic (cf. K102) and other languages and assuming independent rise of occlusion in Celtic, Germanic, Baltic, Slavic, Greek, Iranian, Albanian, Armenian, and Indic (1999: 121). He claims that his reconstruction is “the simplest system possible. There is every advantage to a simple protosystem” (1999: 122) and thereby unwittingly offers a perfect exemplification of the thesis that “the negative potential of aprioristic considerations must not be underestimated. Since theory can easily embody the reflection of rationalized prejudice, it is important that comparative work be carried out inductively” (K130: 97). Perhaps the most characteristic feature of the Germanic languages in comparison with their Romance and Slavic neighbours is the incomplete voicing in the obstruents. This feature is most striking in the peripheral dialects, especially in Icelandic and Upper German, but also in Danish. I find it very difficult to assume that these dialects have innovated and that the ProtoGermanic system resembled that of Spanish or Greek more than that of the attested Germanic languages. If we take the evidence of the peripheral dialects at face value, we must reconstruct a series of voiceless fricatives, a series of unaspirated voiceless plosives, and a series of voiceless obstruents with a complex articulation which is reflected as (pre)aspiration in the north and (af)frication in the south. The latter series remained distinct from the original geminates (cf. especially Petersen et al. 1998: 27 on Faroese, K142: 177 on northern English, and K138: 56 on southern German). Thus, I think that the alleged strengthening of initial obstruents in North Bavarian prōǩd ‘breit’, tum ‘dumm’, tōx ‘Tag’, kēm ‘geben’ and Middle Bavarian pām ‘Baum’, taitš ‘deutsch’, tō 187 ‘Tag’, krāw ‘grau’ (cf. Goblirsch 1994: 33) in fact reflects an archaism. The West Germanic gemination of consonants before *j gave rise to a sixfold distinction in the obstruents without introducing voicing as a distinctive feature (cf. K138: 55). The question now is: what was the feature which is reflected as preaspiration in Icelandic, preglottalization in the western dialects of Danish, gemination in Swedish vecka, droppe, skepp, assimilation of mp, nt, nk to pp, tt, kk in the larger part of Scandinavia, gemination of k in ON bekkr, røkkr and of t in Swedish sätta, gemination of p, t, k before r and l in West Germanic and sporadically in Scandinavia, preglottalization before a tautosyllabic voiceless plosive in English, and affrication in High German and in the English dialect of Liverpool? It cannot have been gemination, which remained a separate distinctive feature, e.g. in North Tyrol (Imst) prukkǩ ‘Brücke’, lNjxxǩ ‘lachen’, deŋkkxǩ ‘denken’ (cf. Goblirsch 1994: 35). But it must have been a feature that could easily give rise to gemination under various conditions in the separate languages. As the phonetic difference between gemination and preglottalization is small, it can easily have been the latter, which may have been preserved in English and western Danish and by lenition have developed into preaspiration in West Scandinavian. Thus, I reconstruct for Proto-Germanic the preglottalization which is actually attested in standard British English and which offers by far the simplest explanation for the reflexes in the other Germanic languages. Note that this reconstruction of Proto-Germanic glottalization is wholly independent of any theories one may have on its Indo-European origins. In a recent article (1997), Page argues that preaspiration was present in the Common Scandinavian period and that the vestjysk stød developed from this preaspiration (thus already Pedersen 1912: 42). This view was already rejected by Jespersen, who called it an example of “papirfonetik” (1913: 24). Ringgaard agrees with Jespersen and regards the vestjysk stød as a spontaneous innovation of the westernmost dialects of Danish (1960: 108). This does not explain the origin of the glottalization and dissociates it from the same phenomenon in English and from the rise of preaspiration elsewhere in Scandinavia. Page suggests that in comparison with the Scandinavian dialects with preaspiration, “dialects with stød are more centrally located, a pattern associated with innovations” (1997: 185). On the contrary, I would maintain that the vestjysk dialects are peripheral in relation to other varieties of Danish, which in their turn were peripheral to the dialects of Norway and Sweden at the time of the Viking expeditions. It therefore seems much more probable to me that preglottalization is ancient in Germanic and that the West Scandinavian preaspiration developed from it by lenition. GERMANIC *ē1 AND *ē2 It appears that most scholars nowadays reconstruct North-West Germanic *ē1 and *ē2 as [ā] and [ē], respectively. Elsewhere I have argued for a reconstruction as [ǃ] and [ea] instead (K084, K115, K124, K180). Patrick Stiles has now emphatically endorsed the majority view (2004). Since he is evidently unaware of my publications, it may not be superfluous to return to the subject here. I shall try to avoid repeating what I have said earlier. There is evidence for a fronted pronunciation of *ē1 in Gothic, Old English, Old Frisian, Old Saxon, and the Dutch dialects of West Flanders, Zealand, Holland and Utrecht (cf. K084: 440), in early West Germanic names such as Suebi and Segimerus, and in Finnish loanwords from North Germanic such as miekka and niehla, Old Icelandic mækir ‘sword’ (early Runic acc.sg. makija), nál ‘needle’ (cf. Grønvik 1998: 88). The retraction of *ē1 to ā evidently spread from the High German area to the north and apparently took place independently in Scandinavia when new long vowels arose. In my view, the crucial factor is precisely the Proto-Germanic asymmetry in the low vowels between long front ǃ and short back a which could be resolved either by fronting a to æ, as in Anglo-Frisian, or by backing ǃ to ā, as in High German and Scandinavian. If ǃ had been retracted to ā in West Germanic already, the Anglo-Frisian fronting of short a to æ would be entirely unmotivated. Moreover, we would have to reconstruct a totally unmotivated sequence of alternating backing and fronting developments under different conditions (cf. K180: 46). If we want to avoid the assumption that fronted æ was again retracted to a at various stages, it follows that the Anglo-Frisian fronting of the short vowel was blocked by a following l, r, h plus consonant and in open syllables by a back vowel in the following syllable (cf. already Heuser 1903: 1). Since we do not find palatalization before *ai and *au in Frisian, it is natural to assume that *ai had been monophthongized to ā before the Anglo-Frisian fronting of a to æ and that *au had remained unchanged. The Anglo-Frisian palatalization then affected k and g before front vowels. After the earlier, “Saxon” migration which resulted in the conquest of Kent and Sussex in the fifth century, the fronting of a to æ affected the remaining instances of a in closed syllables, and also *au with a before tautosyllabic u, in the dialect of the new settlers. This “Saxon” second fronting was followed by breaking and second palatalization, e.g. in eald, cēapian, Old Frisian ald ‘old’, kāpia ‘buy’. In fact, the first stage of breaking can be identified with the “Saxon” fronting because the conditions were largely identical: it appears that the process of breaking began as incomplete fronting of a before tautosyllabic l, r, h and u and subsequently affected e and i. After the “Anglian” migration which can be connected with the subjugation of the 190 Germanic phonology Humber estuary starting around the middle of the sixth century, these developments spread to the northern territories, leaving traces only of the earlier situation. In the meantime, Anglian shared the development of Frisian on the continent, in particular the raising of ǃ to ē, which had been preceded by the Anglo-Frisian retraction of ǃ to ā before w. The Kentish raising of ǃ to ē was probably a local development, perhaps under the influence of a second invasion in Kent in the sixth century. After the “Anglian” migration, Frisian fronted ā < *ai to ǃ unless it was followed by a back vowel in the following syllable and monophthongized *au to ā. The distinction between ē < *ǃ and ǃ < *ai is still preserved in modern dialects (cf. Campbell 1939: 1011). The Anglo-Frisian and second English palatalizations preceded i-mutation because the umlauted vowels did not palatalize k and g but phonemicized the opposition between palatals and velars, so that Old Frisian shows palatalization before e < *æ and ē < *ǃ but not before e < *ä or ǃ < *ai, e.g. tsetel < *katilaz ‘kettle’ and tziake < *kǃkōn ‘jaw’ versus kenna < *kannjan ‘make known’ and kēi, kāi < *kaijō ‘key’. The Proto-Germanic adverbs ‘here’, ‘there’ and ‘where’ can be reconstructed on the basis of Gothic her, þar, ŵar and Old Norse hér, þar, hvar as *hiar, *þar, *hwar and compared with Lithuanian šia-, ta-, ka- from Indo-European *ki-o-, *to-, *kwo-. This explains the difference between the long vowel in ‘here’ and the short vowel in ‘there’ and ‘where’. In West Germanic, the latter was lengthened so as to yield High German and Saxon -ār, Frisian and Anglian -ēr, and West Saxon -ǃr. Stiles now reconstructs Proto-West-Germanic *þār and *hwār and concludes that *ē1 had been retracted to *ā at the time of the lengthening (2004: 389). In fact, things are more complicated. While the vowel of Old English and Old Saxon hēr and of Old High German hear and hier can be identified with *ē2 < *ea, with lowering of *i to e before a in *hiar (cf. K124: 16), the high vowel in Old Frisian and Old Saxon hīr suggests that the lowering was forestalled by the raising of *ia to ie in the Frisian and Low German dialectal areas, so that *ē2 never arose in this word. This view is supported by the fact that the Saxon variant hēr is frequently spelled hier. As I have pointed out earlier (K124: 17), I think that original *ea was preserved not only in High German, but also in Scandinavian at the time when breaking arose, and that the monophthongization of *ea to ē took place under the pressure of the rise of ǃ from umlauted *ā here. In Old English, *ea yielded ēo in fēoll ‘fell’, hēold ‘held’, spēonn ‘clasped’, gēong ‘went’ (cf. K115: 99). Thus, we arrive at the following scenario. After *ā merged with *ō in ProtoGermanic, symmetry could be restored by lowering *ē1 to ā, and this is what happened in High German and Scandinavian. This development was forestalled by the monophthongization of *ai to ā in Anglo-Frisian, where short a was fronted to æ instead unless it was followed by a blocking environment. These Germanic *ē1 and *ē2 191 developments preceded the Anglo-Saxon migrations. The lengthening in *þar and *hwar could take place at any time during these developments because there was no other long vowel than *ē1 [ǃ, ā] corresponding to short *a [a, æ] in early West Germanic. The rise of *þǃr and *hwǃr in Anglo-Frisian evidently preceded breaking and i-mutation, which belong to the period after the early migrations. Symmetry in the vowel system was again restored in North and West Germanic by the monophthongization of *ea to *ē2 except in High German, where *ō was diphthongized instead. I conclude that there is no reason to assume a general retraction of *ē1 to ā in North-West Germanic and that Stiles pays insufficient attention to the chronological and motivational aspects of the developments. PROTO-GERMANIC OBSTRUENTS AND THE COMPARATIVE METHOD In his impressive new monograph (2005), Kurt Goblirsch rejects my reconstruction of Proto-Germanic preglottalized stops. It may therefore be useful to examine his treatment of the evidence which I have adduced for my reconstruction. I have claimed that the following processes are direct reflexes of ProtoGermanic preglottalization (K102: 6f., see also K138, K192, K211, K212): (1) Preaspiration in Icelandic, e.g. in epli ‘apple’, opna ‘open’, vatn ‘water’, where the aspiration developed from the lenition of a glottal stop. (2) Preglottalization in the western dialects of Danish: the so-called vestjysk stød (cf. Ringgaard 1960), which is in my view an archaism. (3) Gemination in Swedish, e.g. in vecka ‘week’, droppe ‘drop’, skepp ‘ship’, ON vika, dropi, skip, OE wice, dropa, scip, which developed from assimilation of a glottal stop to the following plosive. (4) Assimilation of mp, nt, nk to pp, tt, kk in the larger part of Scandinavia. The nasal consonant was apparently devoiced by the preaspiration of the following plosive and then lost its nasal feature. (5) Gemination of k before j and w in ON lykkja ‘coil’, bekkr ‘brook’, røkkr ‘dark’, and of t before j in Swedish sätta ‘to set’ and other examples, from assimilation of a glottal stop to the following plosive. (6) Gemination of p, t, k before r and l in West Germanic. The same development is found sporadically in Scandinavia; this suggests that we have to do with the loss of an archaic feature rather than with an innovation. (7) Standard English inserts a glottal stop before a tautosyllabic voiceless plosive, e.g. lea’p, hel’p (cf. Brown 1977: 27), which is in my view an archaism. (8) The High German sound shift yielded affricates and geminated fricatives, e.g. OHG offan ‘open’, wazzar ‘water’, zeihhan ‘token’. The gemination originated from the oralization of a glottal stop and its assimilation to the following fricative. Goblirsch does not discuss the vestjysk stød and the English glottalization at all. He claims that the rise of distinctive preaspiration before p, t, k was a reaction to the loss of voicedness in b, d, g in order to maintain the distinction between voiced and voiceless geminates (p. 132). This does not explain the specific phonetic development. Moreover, there is reason to assume that all obstruents were voiceless in Scandinavian until the 12th century (cf. K212). 194 Germanic phonology Goblirsch discusses neither the spontaneous gemination in Swedish vecka etc. nor the conditioned gemination in ON lykkja, røkkr, Swedish sätta and similar instances. He attributes the West Germanic gemination of p, t, k before r and l to “Systemzwang” and to the relative length of the voiceless stops (p. 141), without discussing the origin of the length and the reason for further strengthening. He acknowledges but does not explain the rise of High German long fricatives from single p, t, k “die durch die hochdeutsche Spirantisierung geminiert wurden”, e.g. OHG offan ‘open’, ezzan ‘to eat’, sahha ‘matter’ (p. 154). What all of these lengthenings have in common is that they affected ProtoGermanic *p, *t, *k and that they remain unaccounted for in Goblirsch’s theory. Thus, it turns out that Goblirsch simply disregards the comparative evidence for preglottalized stops in Proto-Germanic. This is a consequence of his a priori assumption that, since there is in Germanic “heute in keiner Sprache und in keinem Dialekt eine Serie der glottalisierten Konsonanten” (p. 59), no such series can be reconstructed for Proto-Germanic, in spite of the fact that preglottalization is actually attested in English and in the western dialects of Danish (cf. K192: 8). This has grave consequences because he thereby eliminates glottalization as a possible source of the Germanic consonant shifts from his reconstructions. A strict comparative analysis of Proto-Germanic *b, *d, *g yields a reconstruction of word-initial plosives which have been preserved in Scandinavian, English and High German, and probably in Gothic and Low German as well (cf. Moulton 1954 and K102). While it is generally recognized that *d was always a plosive in West Germanic, I have argued that in Scandinavian an intervocalic plosive *d became a fricative between the earlier and the later syncope and a postvocalic *d before -r- after the younger syncope, and that a word-final plosive *d was preserved both in Gothic and in Scandinavian at the time when final obstruents were devoiced after the apocope (K102: 4). Goblirsch disregards my argumentation and posits voiced fricatives which became plosives in all Germanic languages under various conditions and even projects this reconstruction back into Indo-European, in spite of the comparative evidence of neighbouring branches of the language family. This is what Moulton called a “reductio ad absurdum” which “immediately destroys the whole comparative method of reconstruction, and with it the whole concept of a proto-language. We must apply the comparative method either strictly or not at all” (1954: 38). Moreover, Goblirsch assumes that Proto-Germanic *b, *d, *g became devoiced independently in High German, Danish (and neighbouring dialects of Swedish, Norwegian, Low German and Frisian) and Icelandic (and Faroese), and then word-initially in Swedish, Norwegian, English and Low German (p. 17). In a strict comparative analysis one would rather assume an archaism in High German, Danish and Icelandic, and substratum influence from Celtic, Romance, Slavic and Finnish in northern English, Low and Central 195 Franconian, northern Middle German and eastern Swedish, in spite of Goblirsch’s statement to the contrary (p. 79). The reason why scholars are tempted to assume voiced fricatives for ProtoGermanic is of course that this simplifies the formulation of Verner’s law, according to which voiceless fricatives became voiced under certain conditions. However, there are reasons to suppose that Verner’s law preceded Grimm’s law (cf. K102: 5f. and K119: 2f.), which Goblirsch does not discuss. If Verner’s law voiced *t to *d before the consonant shift, it follows that the traditional IndoEuropean *d and Proto-Germanic *t cannot have been simple voiced and voiceless stops. Since the reflexes of these consonants lengthened a preceding vowel under certain conditions in the same way as did the Proto-Indo-European laryngeals both in Latin (Lachmann’s law) and in Balto-Slavic (Winter’s law) and moreover left a trace in the glottalization which we find in the modern Baltic languages, the hypothesis of an original series of preglottalized stops is not farfetched. In Goblirsch’s theory, voiceless plosives first become aspirates, then affricates, and finally fricatives. Note that the first development regards the glottal articulation (delay of the voice onset time) whereas the second and the third concern the oral articulation (oralization of friction and loss of oral occlusion). It is presupposed that the glottal occlusion is released during the oral occlusion and lost together with the latter. In the case of preglottalization and preaspiration, the glottal occlusion is released and the glottis is opened wide before the oral occlusion takes place. When a preglottalized voiced stop becomes devoiced, the glottis is opened wide after the release of the glottal occlusion and remains so until the new voice onset. When the latter is delayed, the new voiceless stop becomes aspirated. Thus, the devoicing of a preglottalized voiced stop yields a preglottalized voiceless stop which may develop into a preaspirated voiceless stop and subsequently into a voiceless aspirated stop. This is what probably happened in the Germanic languages. An oral fricative may be shortened to an oral plosive, but it is improbable that this happened repeatedly on a large scale. I conclude that the Germanic consonant shifts started with a lenition of the original voiceless plosives *p, *t, *k to fricatives f, þ, x and the concomitant general devoicing of obstruents which yielded the characteristic absence of voicedness in High German, Danish and Icelandic, where the original preglottalization is reflected as consonantal strengthening, vestjysk stød and preaspiration, respectively. No such development took place in the neighbouring Indo-European languages (Baltic, Slavic, Italic, Celtic), where the inherited voicedness in the obstruents was never lost, except in Italic after the lenition of the simple voiced stops (i.e. the so-called voiced aspirates) to fricatives. Preglottalization was preserved in English and western Danish, lenited to preaspiration in West Norse, and oralized in East Norse and High German, 196 Germanic phonology except in word-initial position, where delay of the voice onset time yielded postaspiration. Further delay of the voice onset time and oralization of the friction yielded affricates and fricatives in High German. In Gothic, voicedness was restored under the influence of a Romance substratum which produced a system of obstruents resembling that of modern Spanish (cf. K102: 8). The reconstruction of Proto-Germanic preglottalized stops thus gives a complete, consistent and well-motivated picture of the multifarious consonant shifts in the Germanic languages. ENGLISH bottom, GERMAN Boden, AND THE CHRONOLOGY OF SOUND SHIFTS Guus Kroonen has convincingly shown (2006) that English bottom and German Boden represent an original formation in *-men- which can be identified with Greek πυθμήν ‘bottom’, with Indo-European loss of *-m- in the zero grade suffix *-mn- after a labial consonant in the root. This explains the alternation between Germanic *d in OFr. bodem from *budmēn and *tt in OE botem from gen.sg. *buttas < *budnós. It does not explain *þ in OS dat.sg. bothme beside bodme, ME bothme beside botme and bodme, OHG bodam. In view of Mercian oexen, Northumbrian exen, OFr. ixen, ON yxn, øxn < *uksnes ‘oxen’, the expected paradigm of Proto-Germanic *budmēn is the following (cf. Beekes 1995: 176, Boutkan 1995: 278, K219: 3): nom.sg. acc.sg. gen.sg. dat.sg. inst.sg. loc.sg. *budmēn *budmenun *buttas *buttei *buttē *budmeni nom.pl. acc.pl. gen.pl. dat.pl. *buttes *buttuns *buttan *budumus It now appears that *-m- was restored in all languages, but not with the same result. While the geminate was preserved so as to yield preglottalization in OE botm, botem and ON botn, it was simplified and lenited by Grimm’s law in OHG bodam and OS dat.sg. bothme < *boþma-. Similarly, OFr. wetma and OHG wedamo ‘dowry’ show preservation of the geminate in the northwest and lenition by Grimm’s law in the southeast in the paradigm of Proto-Germanic *wetmēn. While High German generalized the lenited obstruent in bodam, bodem, Boden and Scandinavian generalized the fortis stop in botn, the original alternation was apparently preserved in OE bodan beside botem and in OS dat.sg. bodme beside bothme. It follows that the simplification of the geminate before the restored -mpreceded Grimm’s law in the German dialectal area but not in Anglo-Frisian and Scandinavian. Elsewhere I have claimed that the Germanic fixation of the stress on the initial syllable of the word preceded Grimm’s law in Gothic but not in North-West Germanic (K102: 9, K138: 54). These two datings support each other if Gothic originated in the southernmost part of the Germanic territories, as I have argued earlier (K198). Thus, we arrive at the following relative chronology (with a reconstruction of traditional *t, *d, *dh, cf. K138 with references): 198 Germanic phonology (1) Proto-Indo-European t:, t’, t. (2) Dialectal Indo-European t, ’d, d. (3) Verner’s law: non-initial t > d before the stress. (4) Kluge’s law: medial tn, ’dn, dn > tt before the stress. (5a) Retraction of the stress in Gothic. (5b) -ttm- > -tm- in High and Low German and simplification of most geminates in Gothic. (6) Grimm’s law: t > þ, ’d > ’t (> t in Gothic), d > t (not in Gothic). (7a) Retraction of the stress in North-West Germanic. (7b) Simplification of geminates under certain conditions in North and West Germanic. (8) West Germanic raising of e to i before i and j. (9) West Germanic loss of j before i. (10) West Germanic gemination before j: þj > þþ, ’tj > ’tt, tj > tt. (11) High German consonant shift: ’t > tz > zz, ’tt > ttz > tz. (12) Restoration of voicedness in northern English, Low and Central Franconian, northern Middle German and eastern Swedish under the influence of neighbouring Celtic, Romance, Slavic and Finnic languages (cf. K230). The fricative *þ is found not only in the German territories but also in English texts of northern provenance, e.g. bythne ‘keel’, ME bothem beside OE botm, bytme (cf. Campbell 1959: 171, Brunner 1965: 163). It appears that the AngloFrisian preglottalized reflex of *ttm was replaced by the German fricative reflex *þm between the earlier, “Saxon” migration of the 5th century and the later, “Anglian” migration of the 6th century AD (cf. K084: 440-442, K180: 45). This development can now be added to the list of innovations which reached the continental homeland of the Anglo-Saxons from the south between the two waves of migration (ibidem). In Old Frisian, we find the original word wetma ‘dowry’ beside the German loan withume ‘consecrated ground near the church’ (Boutkan & Siebinga 2005: 453) < ‘endowment’. The recent date of Grimm’s law invalidates Vennemann’s objection to the glottalic theory that “for Semitic loanwords with voiced plosives [...] we have to reckon with Germanic T or D reflexes depending on the time of borrowing” (2006: 137) because traditional Indo-European *d was a preglottalized voiced stop from dialectal Indo-European times almost until the dissolution of ProtoGermanic. The argument can only be used against the outdated view that the Germanic voiceless stops directly continue the voiceless ejectives of the IndoEuropean proto-language, which is still supported by Perridon (2008: 416). It appears that Perridon has given up on the comparative method: “we need no longer look for a common origin for all the cases of preaspiration in [the Nordic] languages. Preaspiration may have emerged, and possibly have English bottom, German Boden, and the chronology of sound shifts 199 disappeared again, in different areas at different points in time” (2008: 425). As I have argued elsewhere (K211, K212, K230), the hypothesis that preaspiration originated from a lenition of preglottalized stops in the 7th century at a time when all obstruents were voiceless accounts for a wide range of different phenomena. Perridon’s view that preglottalization “seems to be closely connected with the loss of the distinction between long and short consonants” (2008: 426) is clearly mistaken in view of the vestjysk stød in such instances as hjæl’b ‘to help’ < *hjæl’pe, hen’d ‘to fetch’ < *hen’te, skar’b ‘sharp (pl.)’ < *skar’pe, kjøv’d ‘bought’ < *kjø’pte, bruw’d ‘used’ < *bru’kte. After the late Proto-Germanic devoicing of the original preglottalized voiced stops, further delay of the voice onset time yielded post-aspiration in word-initial position so as to maintain the distinctive glottal articulation when the lenes were also devoiced. THE GERMANIC FIRST CLASS OF WEAK VERBS 1. The first weak class of Germanic continues two PIE present stem formations, viz. *-ie/o- and *-eie/o-, e.g. 3rd sg. 3rd pl. *bugjeþi *bugjanþi ‘buy’ *wurkjeþi *wurkjanþi ‘work’ *nasejeþi *nasejanþi ‘save’ *kausejeþi *kausejanþi ‘probe’ The question to be answered in the following is: when did these two types merge? 2. The two types were redistributed according to the length of the stem in the attested languages, e.g. Gothic: 3rd sg. 3rd pl. bugjiþ bugjand waurkeiþ waurkjand nasjiþ nasjand kauseiþ kausjand It is reasonable to assume that bugjiþ, kauseiþ, and the 3rd pl. forms are phonetically regular. This suggests that *e was raised to i, intervocalic *j was lost, and the high front vowel became nonsyllabic before a. Conversely, *j became vocalic in waurkeiþ, lengthening the thematic vowel. This must have been a phonetic development because it is also found in the nominal inflection. It eliminated the difference between the two types of present stem formation after long verb stems. In the case of short verb stems, the elimination of the difference must have been due to an analogical development because the form nasjiþ can hardly be phonetically regular, as will be argued in the following. 3. The 2nd sg. imperative ending -ei, which is uniform after long and short verb stems, can only be derived from PIE *-eie because *-ie would yield -i, as is clear from the vocative of the io-stems, e.g. hairdi ‘shepherd’, and related forms. Consequently, verbs with a present stem in *-ie/o- must have adopted the imperative in *-eie at a stage when the latter still represented a distinct flexion type. This poses a problem with respect to the chronology and the motivation of the required analogical development. 4. Chronologically, the simplest assumption is that verbs of the type of bugjan and waurkjan adopted the imperative of nasjan and kausjan after the loss of final *-e, which eliminated the thematic vowel from the imperative. As a result of this phonetic development, the imperative in *-i, *-ei became somewhat isolated from the indicative paradigm and therefore more easily subject to analogical pressure. When intervocalic *j was subsequently lost and the two types of present stem merged before the low thematic vowel a, there was little reason to keep the two types distinct in the imperative. 202 Germanic verbal inflexion 5. The motivation for generalizing the long ending -ei rather than the short ending *-i was simply the relative frequency of the two types. Krause lists 12 primary verbs versus 42 causatives, intensives and iteratives, beside 46 denominatives (1953: 225-227). The latter may belong to either type, cf. Greek ποιμαίνω ‘herd’ of ποιμήν, φιλέω ‘love’ of φίλος, and may have been subject to redistribution in Proto-Germanic times. The generalization of -ei may have been facilitated by the vocalization of *j in waurkeis, waurkeiþ, which induced the transition to the other type. 6. Just as *-eie yielded -ei, PIE *-eies yielded -eis in nom.pl. naweis of naus ‘corpse’. This makes the conclusion that nasjiþ replaces an earlier form *naseiþ practically inevitable. This analogical development again poses a problem of chronology and motivation. It is clear that it must have been posterior to the generalization of the long imperative ending -ei, which would otherwise be a quite incomprehensible development. 7. After the neutralization of the opposition between /i/ and /j/ in waurkeiþ and nasjand, the flexion of nasjan differed from that of bugjan only in those forms where long verb stems automatically took -ei- alternating with -ja-. The flexion of bugjan was of greater regularity because it always had a stem-final -jbefore the thematic vowel. Moreover, the same flexion was found in strong verbs with a present stem in -j-, e.g. bidjan ‘pray’, hafjan ‘raise’. Krause lists 17 verbs of this type (4 primary weak verbs, 4 denominatives with a vocalic stem, and 7 strong verbs with a short stem, to which *sitjan and *ligjan must be added on the evidence of the North and West Germanic languages) versus 9 causatives and intensives, beside 6 denominatives with a consonantal short stem (1953: 219-220 and 225-227). It appears that the substitution of nasjiþ for *naseiþ was determined both by the tendency toward regularity of the paradigm and by the relative frequency of the types. The substitution may or may not have taken place in the West Germanic languages, which lost *j before i. 8. Thus, I claim that the phonemic distinction between -ji- (from *-je-) and -ei- (from *-ei- and *-ī-) was never lost after short stems. There is no reason to assume a *-j- at any stage in the development of the word marei ‘sea’, which represents the neuter *mari plus the PIE suffix *-H1n-. Similarly, verbal nouns such as naseins ‘salvation’ do not replace earlier *nasjins from *-jen-, as is sometimes assumed, but represent a lost neuter *nasi plus the suffix *-H1n-. The form *nasi, which is also found as the first member of a compound in the preterit, can be identified with the Vedic passive aorist in -i, as I have indicated elsewhere (K044: 1281). There is no evidence for a sequence -ji- which is older than the raising of *e to i in unstressed syllables. 9. The neutralization of the opposition between /i/ and /j/ took place not only before a, but also before other back vowels, cf. especially nom.pl. sunjus of sunus The Germanic first class of weak verbs 203 ‘son’, PIE *-eues. Since j after a long stem was not excluded before back vowels, it was bound to spread to the position before front vowels. This is what happened in gen.sg. andbahtjis ‘service’, gawairþjis ‘peace’ next to andbahteis, gawairþeis, also reikjis ‘power’, arbjis ‘heritage’, fiskjins ‘fisher’, bandjins ‘prisoner’, and similarly in the adjective. 10. In conclusion, I assume the following chain of developments: (1) (2) (3) (4) (5) (6) loss of final *-e, raising of *e to i and loss of intervocalic *j, vocalization of *j after a long stem before front vowels, desyllabification of *i before back vowels, generalization of the long imperative ending -ei, analogical introduction of stem-final j before front vowels. The last development finally eliminated the difference between the flexion types of bugjan and nasjan. Though the West Germanic evidence for the developments (5) and (6) is inconclusive as a result of subsequent changes, the difference between Old High German 2nd sg. neris ‘save’ and subj. nāmīs ‘took’ seems to reflect the one between Gothic nasjis and nemeis quite faithfully. THE GERMANIC THIRD CLASS OF WEAK VERBS 1. More than half a century after Flasdieck’s comprehensive study (1935), consensus has not been reached on the origin of the third weak class of Germanic (cf. Bammesberger 1987). It appears that the majority of scholars nowadays agree on the reconstruction of an ai/ja-paradigm of the Old English type for Proto-Germanic (cf. especially Dishington 1978). This reconstruction leaves several questions unanswered. Firstly, we find evidence for an ai/a-paradigm in Old High German, Old Norse and Gothic. Secondly, the expected gemination of palatals is lacking in ON segia ‘say’, þegia ‘be silent’, Old Swedish sighia, þighia. Thirdly, the Proto-Indo-European origin of the ai/ja-paradigm remains to be clarified. 2. Dishington has convincingly argued that the ai/a-paradigm of Old Norse and the corresponding forms in Old High German arose under the influence of the thematic inflection (1978). He does not discuss the Gothic paradigm, which cannot easily be dismissed because of its early attestation and in view of the generally archaic character of this language. I shall return to the Gothic evidence below. 3. There is evidence for an ai/ēja-paradigm in Old English, e.g. lifgan ‘live’, feccan ‘fetch’ (Flasdieck 1935: 106). Cowgill has shown that this is the result of a recent innovation which did not even reach Old Saxon and Old Frisian (1959: 14). Since the lack of gemination in ON segia and þegia requires the earlier presence of a vowel before *-ja- (cf. already Mahlow 1879: 24), Dishington proposes a similar restructuring for Old Norse (1978: 312). This seems improbable to me, not only because the Old English innovation is rather specific, but especially because it was apparently evoked by a similar restructuring of the ō-inflection to an ō/ōja-paradigm in Old English, Old Frisian and Old Saxon, a development for which there is no evidence in Old Norse. It seems better to connect the lack of gemination in segia and þegia with the same phenomenon in the causatives and intensives vekia ‘wake up’, þekia ‘thatch’, rekia ‘stretch’, Old Swedish vrækia ‘drive’. 4. In an earlier article (K088) I argued that the first weak class of Germanic comprises two inflection types, a je/ja-paradigm and an eje/eja-paradigm, and that the distinction between these two was preserved after short verb stems up to a recent stage. The lack of gemination in ON vekia, þekia, rekia, Old Swedish vrækia suggests that the gemination of velars before *j was anterior to the loss of the distinction between the two inflection types, so that the operation of the rule was limited to the je/ja-paradigm, e.g. ON hyggia ‘think’, byggia ‘marry’. The gemination in leggia ‘lay’ can easily be analogical, either on the basis of liggia 206 Germanic verbal inflexion ‘lie’, or under the influence of hyggia and byggia. It now follows from the lack of gemination in segia and þegia that the ai/ja-paradigm which was reconstructed on the basis of the West Germanic evidence must be derived from an ai/ija- or ai/eja-paradigm in order to accommodate the North Germanic material. 5. We now turn to the prehistory of the paradigm which has been reconstructed for North and West Germanic. I agree with Jasanoff (1978: 60-67) that all theories which operate with athematic forms or with an apophonic alternation in the suffix must be rejected. Incidentally, the reconstruction of an ēi/ī/ĭ-paradigm for Balto-Slavic is incompatible with the accentual evidence from that branch of Indo-European. Furthermore, I agree with Dishington (1978) and Bammesberger (1987) that Jasanoff ’s own theory, which is basically a modification of Collitz’s (1891: 51-2), cannot be maintained. We must therefore look for an invariable thematic suffix which yields the correct output in a phonetically regular way. If the final part of the ai/ija- or ai/eja-suffix represents the thematic vowel, the initial part must be derived from a vowel which was low before -i- and front before -a-, i.e. from a low front vowel *æ, which leads to the reconstruction of an æi/æja-paradigm. Since West Saxon and Old West Low Franconian ǃ is the most archaic reflex of Indo-European *ē in Germanic (cf. K084: 440, with ref.), we arrive at the reconstruction of a pre-Germanic ēje/ējo-paradigm on the basis of the internal evidence from North and West Germanic alone. Jasanoff ’s objection that *-ēje- should give *-ē- because *-ājeyielded *-ō- (1978: 62) does not hold if *ē was low while *ā and *e were mid vowels at the time of contraction. Similarly, his conjecture that *-ējo- should give *-ē- because *-ājo- yielded *-ō- is unfounded because *ē was a front vowel while *ā and *o were not. His assertion that the first vowel of Gothic þahta, OHG dāhta ‘thought’ was nasalized at the stage under consideration is spurious. The transfer of Gothic þeihan, OE þēon, OHG dīhan ‘thrive’ from the third to the first strong class rather suggests that this type of long vowel was denasalized at an early stage. We must therefore consider the possibility that Gothic haba, habam, haband ‘have’ represent -ā, -ām, -ānd (cf. already Brugmann 1904: 527). 6. What is the expected reflex of an ēje/ējo-paradigm in Gothic? This question must be viewed in connection with the development of the i- and u-diphthongs in the Germanic languages. The Proto-Indo-European loc.sg. forms in *-ēi and *-ēu are reflected in the Germanic dat.sg. endings of the i- and u-stems. We find a low reflex in Gothic anstai ‘favor’, sunau ‘son’, Runic winai ‘friend’ (AD 300), faþai ‘master’ (France, 6th century), Hakuþo (AD 450, cf. Antonsen 1975: 34, 77, 55), OE suna, OS suno, and a high reflex in Runic Kunimu(n)diu (AD 500), magiu ‘son’ (7th century, ibidem: 79, 85), OHG suniu, ensti, meri ‘sea’, OS ensti, wini, ON brúþe ‘bride’. It appears that the long diphthongs were originally low and were raised around the end of the 5th century. If the earlier Norse syncope can be dated to the 7th century, this fits in with the development of the The Germanic third class of weak verbs 207 æi/æja-paradigm advocated here. Interestingly, the gen.sg. ending of the i-stems *-ois which is reflected in Gothic anstais and ON brúþar appears to have merged with the reflex of *-ēi in OHG and OS ensti while remaining distinct from the nom.pl. ending *-oi and the dat.sg. ending *-ōi, OHG blinte ‘blind’, tage ‘day’, OS blinde, dage. This suggests a development of *-ois > *-aiz > *-aij > *-æi with subsequent raising after its merger with *-æi < *-ǃi < *-ēi, whereas *-āi and *-ōi were shortened to *-ai and later monophthongized to *-ē. The dat.sg. ending of OE suna and OS suno shows that *-ǃu merged with *au rather than *eu in these dialects, perhaps because it was shortened slightly earlier here than in OHG and ON. 7. Now we turn to the development of intervocalic *j in Gothic. Elsewhere I have argued that the gen.pl. ending -e represents Proto-Indo-European *-eiom (K030: 291-2 and K055: 171-2). The lowering of *ei to *ē before *-an < *-om, as compared with its unconditioned raising to *ī, has a parallel in the low reflex of Proto-Germanic *ǃ from Indo-European *ē in saian ‘sow’ and waian ‘blow’, as compared with its unconditioned reflex *ē. The absence of raising suggests that intervocalic *j was assimilated to the preceding vowel. This development can be dated after the loss of final *-e (cf. K088: 30). After the raising of *e to i, the expected reflex of an ēje/ējo-paradigm is an æi/æa-paradigm in early Gothic. I think that the long vowel in the suffix was now shortened to *a before the raising of *ǃ to *ē, as in ŵamma ‘to whom’ beside ŵammeh ‘to everyone’. This development can be compared with the reduction of *ōw to au before vowels in sauil ‘sun’, taui ‘deed’, staua ‘judge, judgment’, stauida ‘judged’, cf. OE stōw ‘place’, stōwian ‘restrain’. The preservation of -j- in Gothic stojan < *stōwjan shows that the loss of *w was more recent than the elimination of intervocalic *j. It seems probable to me that the root vowel of saian, waian, taui, staua was short. This hypothesis accounts not only for the automatic lowering in the latter words, but also for the absence of raising in the former. It has the additional advantage of accounting for the inflection of haban, which now turns out to be an ai/ā-paradigm of recent origin. While the ō-inflection is apparently of ProtoGermanic date, the development of the third weak class belongs largely to the separate languages. Such forms as waiwoun ‘they blew’ and the calque armaio ‘mercy’ are recent and must be considered analogical in any theory. 8. The ēje/ējo-paradigm which has been reconstructed for pre-Germanic is not of Proto-Indo-European origin. It represents a je/jo-derivative of a preterital ē-stem, to be compared with Slavic iměti, iměje- rather than Lith. tur°ti, tùri ‘have’. One may wonder if Germanic has preserved any traces of the original present tense inflection which was ousted by the ēje/ējo-paradigm. On the basis of the Balto-Slavic evidence we can reconstruct an ei/i-paradigm, which is still preserved in Prussian turei, turri ‘has, have’ (cf. K087). I think that this 208 Germanic verbal inflexion inflection can actually be recovered in Germanic, but not in the third class of weak verbs. 9. It has often been observed that the absence of the verbs for ‘sit’ and ‘lie’ from the third weak class is remarkable. Gothic sitan and ligan belong to the simple thematic inflection, whereas their North and West Germanic cognates have a je/jo-paradigm. Though it is usually assumed that the Gothic inflection is secondary, the motivation for such a morphological transfer of two extremely common verbs is quite unclear. It seems much more probable to me that these verbs represent an original i-inflection which was brought into line with the i/a-paradigm in Gothic and the i/ja-paradigm in the North and the West. The inflection which can be reconstructed for Proto-Germanic is the same as we find in Lith. sėd°ti, s°di ‘sit’. The Gothic development has a perfect analogue in Old Irish saidid, -said ‘sits’, laigid ‘lies’, 3rd pl. sedait, -legat (cf. Thurneysen 1946: 354). 10. The reconstruction of a Proto-Germanic i-inflection also offers an explanation for OHG stān, stēn ‘stand’, gān, gēn ‘go’, OE gān, 3rd sg. gǃþ, pl. gāþ. Mahlow already reconstructed *staji-, *staja- (1879: 138), which should yield an ai/ā-paradigm in view of Gothic aiz ‘brass’, maiza ‘greater’, OE ār, māra, OHG ēr, mēro. There are two problems with this reconstruction. On the one hand, the vowel of OHG stēn and gēn is not the unconditioned ei of stein ‘stone’ but the lowered variant ē which is found word-finally in wē ‘woe’ and sē ‘behold’, as if it were followed by a hiatus. The reflex ei in Otfrid 3rd sg. steit, geit represents -ē-i-, cf. duit ‘does’ from -uo-i-, similarly OE gǃþ from -ā-i-. On the other hand, the vowel of OE gān, gāþ is not the retracted variant ō which could be expected as a reflex of *ā before a nasal but the unconditioned reflex ā of *ai, as in stān ‘stone’. These problems disappear if we start from an athematic i-inflection *stai-, which could be restructured either as *sta-i/a-, which is reflected in OHG stān, or as *stai-i/a-, which is suggested by OHG stēn. As to the etymology of the verb ‘to go’, I am inclined to return to Kluge’s derivation of the word from gaplus *eimi, in spite of Streitberg’s objection that the meaning is durative (1896: 319). If this is correct, we can reconstruct a disyllabic stem *ga-i- with an athematic paradigm for Proto-Germanic, and the OE reflex gān is regular. The difference between durative *stai-, *seti-, *legi-, preterite *-ē-, and inchoative *stand-, *sent-, *leng-, preterite 3rd sg. *stōþ, *sat(e), *lag(e) was apparently lost in Proto-Germanic times. The same can be assumed for *(ga-)i-, *(ga-)eaj(e), which yielded OE gā-, *ee, OHG gē-, *ee. The nasal present of Gothic gaggan, OE OHG gangan may have been created on the analogy of *stai-, *stōþ, *stōdun, standan (OHG stantan), *gai-, *gegai, *gegijun if the incorporation of ga- was sufficiently early. THE GERMANIC SEVENTH CLASS OF STRONG VERBS 1. The Proto-Germanic reduplicated preterit has a threefold reflex in the attested languages (cf. Bech 1969: 3): (1) reduplicated preterits, e.g. Gothic haihait ‘called’, lailaik ‘leaped’, OE heht ‘called’, leolc ‘played’, also reord ‘advised’; (2) r-preterits, e.g. ON sera ‘sowed’, grera ‘grew’, snera ‘turned’, OHG steroz ‘struck’, pl. biruun ‘dwelt’, OE leort ‘let’, also reord ‘advised’; (3) ē-preterits, e.g. ON hét ‘called’, fell ‘fell’, hlióp ‘leaped’, OE hēt, fēoll, hlēop. Bech derives the latter two types from ez-infixation before the root vowel and syncope. This accounts nicely for such forms as ON snera, OHG steroz, OE leort, but less satisfactorily for the ē-preterits. The main drawback of Bech’s theory is the narrow basis for the analogical development of the ez-infix, which he derives from *sezō ‘sowed’ and *sezalt ‘salted’ (1969: 21). It is hardly possible that these two verbs provided the model for a pervasive restructuring of the reduplicated preterit. 2. Robert Fulk has now demonstrated in a conclusive way that the ē-preterit must be derived from e-infixation before the root vowel of the present stem (1987). The apex of his explanation is the disyllabic interpretation of OHG geang ‘went’, feang ‘seized’, feal ‘fell’. The following remarks are intended to supplement Fulk’s theory. 3. As in the case of Bech’s proposal, the model for the analogical development of e-infixation is not obvious. Fulk adduces ON auka ‘increase’, ausa ‘pour’, OE eaden and OS ōdan ‘granted’, OHG erien ‘plough’, and Gothic (af-aikan ‘deny’ and) us-alþan ‘grow old’ (1987: 162), for which we can reconstruct the preterits *eauk, *eaus, *eaud, *ear (OHG ier), *eaik, *ealþ. It is not evident that these forms sufficed to generate a wholesale restructuring of the reduplicated preterit. We may therefore have to look for a more powerful model, which can only have been supplied by a high frequency verb. 4. In his discussion of OE ēode ‘went’, Cowgill assumes a development of 3rd sg. *eaje to *eae, then contracted to *eō, which yielded West Germanic *eu, then *euda (1960: 494, 499). This theory meets with a number of difficulties. First of all, the loss of intervocalic *j can hardly have preceded the loss of final *-e because the 2nd sg. imperative ending *-eje yielded Gothic -ei /-ī/, not **-ē (cf. K088: 28 and K105: 6), so that we would expect *eaj instead of *eae. Secondly, the alleged contraction of *-ae to *-ō is not very probable in view of Gothic haba /-ā/ ‘I have’ from *-ējō (K105: 5). Thirdly, it is not evident that a form *eō should 210 Germanic verbal inflexion escape remodeling to *eaj. These difficulties can be removed if we identify the stem vowel of OE ēode with that of fēoll and hlēop. 5. If *eaje yielded *eaj and 3rd pl. *eijun was remodeled to *eajun, the loss of intervocalic *j in the latter form yielded *eaun, which developed into OE *ēon, cf. rēon ‘rowed’ from *reōun. The stem form ēo- then served as a basis for the creation of a weak preterit while rēon was replaced by rēowon on the basis of other forms in the paradigm of rōwan. The difference between ēode on the one hand and sēow ‘sowed’, mēow ‘mowed’, blēow ‘blew’, cnēow ‘knew’ on the other shows that the 3rd sg. form ‘went’ cannot have been *eō at an earlier stage. 6. It follows from Fulk’s theory that disyllabic *ee, *ea, *eo yielded OE ē, ēo, ēo in lēt ‘let’, hēt ‘called’, fēoll ‘fell’, hēold ‘held’, spēonn ‘clasped’, gēong ‘went’, hlēop ‘leaped’, hrēop ‘shouted’. The rounding in the reflex of *ea must evidently be attributed to the influence of the following consonants. The unrounded reflex is found in fēng ‘seized’, hēng ‘hung’, which must be derived from disyllabic *feah, *heah. Fulk assumes “either a chronological or a morphological discrepancy in the formation of the etyma of ΟΕ gēong, on the one hand, and fēng and hēng on the other: the latter may have been formed at a later date” (1987: 172). This view is unsatisfactory because these verbs would be the only exceptions to the rule of e-insertion, and a phonetic explanation is therefore more probable. I think that *eah yielded *eeh instead of *eoh, which implies that the disyllabic pronunciation was preserved at the stage when posttonic long vowels were shortened and that the latter development preceded the retraction of *ā to ō in stressed syllables. 7. Fulk’s theory does not account for the r-preterits in a satisfactory way. As Bech observed, the geographical distribution of the forms suggests that they represent an older process. The original formation is attested in ON sera ‘I sowed’, rera ‘I rowed’, also snera ‘I turned’ if the nasal resonant in *seznōw was subject to metathesis, and OE reord ‘advised’ < *rerōd-, cf. beoft ‘beat’ < *bebaut-. It appears from ON grera ‘I grew’ that the reduplication C1eC1C2- was replaced by C1C2eC2-, cf. Gothic gaigrot /gegrōt/ ‘wept’. This substitution, which is reminiscent of Latin stetī ‘I stood’ and spopondī ‘I vowed’, accounts for OE speoft ‘spat’ and OHG screrot ‘cut’. While ON gnera ‘I rubbed’ was evidently built on the analogy of snera and grera, dissimilation of l–l to l–r yielded OE leort ‘let’ and OHG pleruzzun ‘they sacrificed’. Since the vowel of OE leolc ‘played’ cannot be the result of either breaking or umlaut before -lc, we have to assume a form *lelōk- on the analogy of *lelōt-. The furthest extension of the r-preterit is found in Old High German, where it is attested in biruun ‘they dwelt’ (cf. ON snera and gnera), steroz ‘struck’ (cf. screrot and pleruzzun), and even scrirun ‘they cried’, which does not belong to the seventh class of strong verbs. THE GERMANIC FIFTH CLASS OF STRONG VERBS 1. As a rule, reduplication and qualitative ablaut are in complementary distribution in the Germanic preterit. There are two classes of exceptions to this. On the one hand, both reduplication and qualitative ablaut are found e.g. in Gothic lailot ‘let’, saiso ‘sowed’, inf. letan, saian. This is obviously an archaism. It follows that no conclusions can be based on the absence of reduplication in verbs with qualitative ablaut for the reconstruction of the original state of affairs. On the other hand, neither reduplication nor qualitative ablaut is found e.g. in for ‘fared’, stoþ ‘stood’, inf. faran, standan. This is a heterogeneous class. The qualitative ablaut in the strong preterit is usually accompanied by a quantitative ablaut distinction between sg. and pl. forms. The elimination of the latter alternation in waiwoun ‘they blew’, lailoun ‘they despised’ is evidently recent (cf. K105: 7). The sg. form fret ‘devoured’ will be discussed below. While the use of either reduplication or ablaut as a preterit marker has a clear motivation, the remarkable preservation of a quantitative alternation between sg. and pl. forms in ablauting preterits suggests the existence of a complex productive pattern for an earlier stage of the language. I cannot therefore agree with Bammesberger “daß der Langvokal -ē- in der schwachen Präteritalalternante das Resultat einer analogischen Neuerung ist, wobei die Alternation -a- : -ō- in der VI. Klasse als Vorbild wirkte” (1986a: 56). Such a development would undoubtedly have replaced nam ‘took’, gaf ‘gave’ by **nem, **gef. The expected direction of analogical change is actually clear from the perfect presents (“Präteriopräsentien”), where ablaut was not used as a preterit marker. Here we find generalization of the full grade with preservation of the Verner alternation in aih ‘has’, aigun ‘they have’, also full grade in magum ‘we can’. It follows that the long vowel of gebum, gebun ‘we, they gave’ can hardly be of analogical origin. 2. The largely complementary distribution of reduplication and ablaut gives rise to the question of how it originated because both seem to reflect the PIE perfect. While ablaut is found with present stems of the type CeRC-, CeR-, CeC-, and CēC-, reduplication is found with present stems of the type CaRC-, aRC-, CēC-, and CōC-. This distribution suggests that the separation between ablauting and reduplicating preterits was triggered by the merger of earlier *a and *o, which obliterated the ablaut alternations in verbs with an original *a in the present stem. As a result, ablaut became redundant in reduplicating preterits and reduplication could be abolished in ablauting preterits. We may now wonder what happened to verbs with a present stem of the type eR-, eC-, aR-, aC-, CaR-, CaC-. Elsewhere I have argued that *eaj- ‘went’ 212 Germanic verbal inflexion provided the model for restructuring the reduplicating preterits in North and West Germanic (K115). It is probable that this formation itself was fairly recent and replaced an earlier suppletive aorist (perhaps *lud-, which may have been ousted because of its homonymy with the preterit of *leud- ‘grow’). Anyway, it is clear that reduplication was fairly recent in *eaik- ‘denied’, *eauk- ‘added’, *eaus‘poured’, *eaud- ‘granted’ because the initial sequence of vowels would hardly survive a longer period of time. This brings us to the verb ‘to eat’. It is highly improbable that Gothic fret, ON át, OE ǃt, OHG āz reflect *eet- (Flasdieck 1936: 335), firstly because there was no motivation to create this form rather than *eat-, and secondly because *eet- would undoubtedly have yielded the same vowel as we find in ON lét ‘let’, hét ‘called’, OE lēt, hēt, OHG liaz, hiaz. It is also highly unlikely that *ēt replaced a well-motivated earlier form *eat, as Cowgill maintained (1960: 492f.), especially in view of the MHG substitution of az for āz. I therefore think that we have to start from a paradigm *ōte, *ētun ‘he, they ate’, with substitution of *ētfor *ōt- on the analogy of *ōk- ‘traveled’, *ōl- ‘nourished’, *ōn- ‘breathed’, ON ók, ól, Gothic uzon ‘expired’, where the vowel length distinguished the preterit from the present stem *ak-, *al-, *an-. It follows that the long vowel of *ōk-, *ōl-, *ōn- must be relatively old. The motivation for the replacement of *ōt- by *ēt- arose from the merger of earlier *ā and *ō, which eliminated the ablaut distinction between sg. and pl. forms in the preterit of verbs with initial a- and provoked the merger of this type with the root aorist of the verb ‘to stand’, Gothic stoþ. The ancient character of the initial long vowel in the perfect is supported by the perfect present og ‘I fear’, unagands ‘fearless’, OIr. -ágathar ‘fears’, cf. Vedic bibhéti with reduplication from the perfect, OHG bibēn ‘tremble’. The separation of og from agis, OE ege ‘fear’ (Cowgill 1960: 489) is arbitrary. It follows that *ear-, OHG ier- ‘plowed’ is a recent formation. This is indeed what could be expected because the verb has a je-present in Celtic, Germanic, Baltic and Slavic, which suggests that the root had aorist meaning in the northern Indo-European languages. I therefore think that *ear- replaced an earlier aorist *ar- by prefixation of *e-. Since reduplication was more productive in the High German area than elsewhere at an early stage (cf. K115), there is no evidence for the view that the form *ear ever existed in the other languages, which may simply have replaced the aorist *ar- by the weak preterit. 3. Thus, I think that the ē-vocalism in the plural forms of the strong preterit reflects the reduplication syllable of eC-verbs, as in Gothic etun ‘they ate’. It is clear that the type could never have spread if this were the only verb with initial *e-, as it is in the oldest Germanic languages. Note that in German “fast alle starken verba des germanischen, die vocalisch anlauteten, untergegangen sind: aikan, akan, alan, alþan, anan, arjan, audan, aukan, ausan; agan, îgan und The Germanic fifth class of strong verbs 213 innan sind nur noch in den präsentisch gewordenen og, aig und an vorhanden. Einzige ausnahme das unentbehrliche etan” (Behaghel 1924: 129). We must therefore ask ourselves what happened to the original eC-verbs in Germanic. It is important to realize that the elimination of verbs with an initial vowel in Germanic is closely connected with the existence of a productive ablaut pattern in the strong preterit. If the root vowel served to distinguish the preterit from the present stem, the lexical meaning was carried by the root-final consonant, which was itself subject to phonetic variation under the influence of a following consonant or the place of the stress, cf. Gothic þarf, þaurbum ‘I, we need’, OE geneah, genugon ‘it, they suffice(s)’. Since the ablaut pattern remained productive in CeC-verbs, the annexation of an initial consonant could solve the problem. I think that this is indeed what happened. The process may be compared with the generalization of triconsonantal roots in the Semitic verb, where a similar problem existed. It has long been recognized that it is difficult to separate *nem- ‘take’ from Latin em- and its cognates in Celtic, Baltic and Slavic. The root can easily be explained from a reanalysis of the compound *gan-em- as *ga-nem-, which is attested all over the Germanic area. The same reanalysis is found in Latvian ņemt beside jemt ‘to take’ and in Slavic. Note that OE nōm, nōmon ‘took’ beside more recent nam, nāmon may actually reflect *-ōm, *-ēmun. The usual connection of *geb- ‘give’ with Latin habeō ‘hold’ and OIr. gaibid ‘takes’ must be rejected because the vocalism is incompatible. The verb can be derived from the Germanic prefix ga- plus the root of Hittite epzi ‘seizes’, Latin apīscor ‘reach’, coēpī ‘have begun’. Note that the meaning ‘reach’ offers a suitable basis for the semantic development. Interestingly, the consistent spelling of Gothic gaf ‘gave’, as opposed to grob ‘dug’ and gadob ‘was fitting’ (cf. Roberge 1983: 129), supports the derivation proposed here. I therefore reconstruct sg. *ōf-, pl. *eb-. A third example is the verb ‘to eat’, where we find OHG gezzan beside ezzan in Otfrid and Notker (cf. Seebold 1970: 179), also modern German gegessen, Dutch gegeten ‘eaten’. A fourth example is MHG gan, gunnen ‘grant’ and -bunnen (with -b- from ab- instead of ga-), also gunst ‘favor’, cf. OHG an, unnun, unst, abunst ‘grudge’. These instances exemplify the general tendency to eliminate initial vowels in ablauting paradigms. A fifth example may be the preterit of the verb ‘to be’. It is not immediately clear why the root aorist *bū- should have been replaced by the preterit of *wes‘stay’. Elsewhere I have argued for an original perfect *ōs-, *ēs- which is reflected in Indo-Iranian, Greek, Slavic and Celtic (K071). I therefore wonder if we have to start from a Germanic preterit *ubōs-, *ubēz- ‘was, were’. For the prefix cf. Greek ὕπειμι ‘subsist’, Latin subsum ‘am present’, beside OIr. foïd ‘spends the night’ from *upo-wes-. This would explain the alternation in *was-, *wēz-. Moreover, it seems possible that Anglian (e)arð, (e)aron ‘art, are’ and Old 214 Germanic verbal inflexion Swedish aru represent a back-formation from *ōs- on the analogy of *ōk-, *ōl-, *ōn-. A sixth example may be the verbs ‘to suffice’ and ‘to bring’. The ablaut of OE geneah, genugon points to an earlier alternation *ga-nah, *gan-ung-, which in its turn suggests a paradigm *bra-nah-, *bra-ng- from *pro-n(e/o)k-. The preterit evidently replaces an earlier root aorist, cf. Vedic ¡naṭ ‘attained’. The present stem *breng- was probably created to replace the earlier suppletive present *ber-, cf. Gothic atbairan ‘to bring’. Greek προφέρω ‘bring forward’, Latin (prō)ferō. Note that the Old Irish cognates con-ic ‘can’ and r-ic ‘reaches’ also suggest secondary ablaut on the basis of a zero grade formation. The high frequency verbs adduced here provide a suitable basis for the spread of ē-vocalism in the plural forms of the strong preterit. The Balto-Slavic evidence suggests that *sat-, *sēt- and *lag-, *lēg- replaced earlier aorists *set- ‘sat down’ and *leg- ‘lay down’ belonging to presents with a nasal infix, cf. Gothic standan, stoþ ‘stand, stood’, Old Prussian sindats ‘sitting’. These aorists were eliminated because they were nearly homonymous with the stative presents *seti- ‘sit’, *legi- ‘lie’ (cf. K105: 7f.). There is no reason to assume the existence of reduplicated preterits **sesat-, **lelag- at any stage in the development of Germanic. The incorporation of original aorists into the system of strong verb classes supported the elimination of reduplication in the ablauting preterits. THE GERMANIC SIXTH CLASS OF STRONG VERBS 1. In an earlier article (K120), I argued that Gothic nam, nemun ‘took’, gaf, gebun ‘gave’, -et, etun ‘ate’ represent original *ōme, *ēmunþ, *ōfe, *ēbunþ, *ōte, *ētunþ. Similarly, Gothic -on ‘breathed’, ON ól ‘nourished’, ók ‘traveled’ continue reduplicated preterits *ōn-, *ōl-, *ōk-, where the long vowel arose from the merger of earlier *ō- in the singular and *ā- in the plural. This development eliminated the apophonic distinction between sg. and pl. forms in these preterits and provoked the merger of the type with the root aorist Gothic stoþ, stoþun, OE stōd, stōdon ‘stood’. This root aorist correlates with a d-present with nasal infix stand-. Another verb which apparently combines a root aorist with a d-present is Gothic -hlaþan, OE hladan ‘to load’, cf. Lith. klóti, Slavic klad-. Thus, the sixth class of strong verbs seems to have originated as a distinct category from the merger of the reduplicated preterits of IE *an-, *al-, *ak- with the root aorists of IE *stā- and *klā-. (For the root aorist of IE *dhē- cf. K109: 102.) Since the resulting class lacked the usual apophonic difference between sg. and pl. forms in the preterit, it provided a convenient model for analogical extension to verbs with a present stem of the types CaR- and CaC-. This raises a number of questions. 2. First of all, we may wonder why *ar- ‘plow’ did not join the sixth class of strong verbs but developed a seventh class reduplicated preterit *ear- in Old High German and joined the first class of weak verbs in the other North and West Germanic languages. As I pointed out already (K120: 103), I think that *ear- replaced an earlier root aorist *ar- at a recent stage because the je-present of this verb in Celtic, Germanic, Baltic and Slavic suggests that the root had aorist meaning in the northern IE languages. We must therefore ask ourselves if the vocalism in the preterit of the other Germanic je-presents points to an earlier aorist or to a perfect formation. In Gothic, six out of the seven verbs with a je-present and a strong preterit belong to the sixth class: bidjan ‘to pray’, hafjan ‘to lift’, hlahjan ‘to laugh’, fraþjan ‘to understand’, skaþjan ‘to harm’, ga-skapjan ‘to create’, wahsjan ‘to grow’. It appears that strong verbs with a root in a velar stop and a je-present created a weak preterit on the basis of the past participle in *-htas (K109: 107): bugjan ‘to buy’, waurkjan ‘to work’, brukjan ‘to use’, þugkjan ‘to believe’, preterit bauhta, waurhta, bruhta, þuhta. The earlier strong preterit is still evident from OE brēac ‘used’ and warhte beside worhte ‘worked’, also þōhte, Gothic þahta ‘thought’, which gave rise to a secondary present *þankeje- in Proto-Germanic times already. The new formation gave way to a regular first weak class preterit in the case of sokjan ‘to seek’ and faurhtjan ‘to fear’: sokida ‘sought’, faurhtidedun ‘they feared’, OE sōhte, OHG. suohta, forahtun. The vocalism of OE brēac, warhte, 216 Germanic verbal inflexion þōhte points to an earlier perfect formation while sōhte is ambiguous because it represents the IE root *sāg-. Elsewhere I have suggested that Gothic brahta, OE brōhte ‘brought’ represents an original root aorist which adopted the pattern of a perfect formation (K120: 106). From an Indo-European point of view, Gothic hafjan, bugjan, waurkjan, þugkjan, faurhtjan and þaursjan ‘to thirst’ (cf. Schmid 1963: 97) belong to the flexion type of Latin capiō ‘I seize’ with a zero grade root vowel and a ya-present in Indo-Iranian (cf. also K239: 134-136). The same probably holds for wahsjan ‘to grow’, hazjan ‘to praise’ and *-safjan ‘to perceive’.1 The verbs bugjan and sokjan can hardly be separated from biugan ‘to bend’ and sakan ‘to quarrel’, respectively. If sokjan is a secondary formation on the basis of the preterit stem *sōk-, it is possible that sakan had a je-present at an earlier stage and that we may compare the relation between the two verbs with that between þagkjan and þugkjan. This is in accordance with the fact that Old Irish saigid ‘seeks’ has a je-present which may be identified with the flexion of Latin capiō (Thurneysen 1946: 354, cf. K239: 134-137).2 I conclude that there is no evidence for an original aorist among the sixth class preterits of je-presents while all of them may represent earlier perfects. The vocalism of *hōf and *sōk is ambiguous and the other verbs have an o-grade vowel. Since *ar- ‘plow’ did not develop a sixth class preterit, it is probable that *an-, *al-, *ak- joined the sixth class at a recent stage and never served as a model for other verbs. It follows that we may regard the sixth class preterit as the regular outcome of the root aorist of IE roots in *-ā-, which may have served as a model long before the merger of *ā and *ō. 3. Chr. Stang has proposed for verbs with an o-grade present such as Lith. bárti ‘to scold’, kálti ‘to forge’, málti ‘to grind’, kàsti ‘to dig’, OCS brati ‘to fight’, klati ‘to chop’, bosti ‘to stab’, Latin fodiō ‘I dig’, “dass in diesen und anderen Verben mit den Bedeutungen ‘stechen, graben, schlagen, mahlen, zermalmen’, wo neben e-stufigen auch o-stufige Formen vorkommen, die letzteren auf einen alten intensiven Präsenstypus mit o-stufiger Wurzelsilbe zurückgehen. Ich verweise auf skt. jaṅghanti, (Präs. Part.) ghánighnat, die wegen der Erhaltung von gh gegenüber h in hánti einmal o-Stufe gehabt haben dürften” (1966: 333, cf. already 1942: 41f.). Here belong Gothic graban ‘to dig’, slahan ‘to strike’, malan ‘to grind’, ga-draban ‘to hew out’, ga-dragan ‘to heap up’, skaban ‘to shave’, swaran ‘to swear’, þwahan ‘to wash’, also faran ‘to travel’, cf. Dutch doorsteken ‘to pierce, to cross’, oversteken ‘to cross’. Gothic ga-daban ‘to be suitable’ (cf. Schrijver 1991: 102) may also belong here, cf. German treffend ‘apt, apposite’. It is clear that not all of these verbs are equally ancient. Since they constitute the bulk of the sixth class of strong verbs, we may regard this class as the reflex of the IE intensives. This raises the question of why the intensives followed the pattern of the root aorist in *-ā- rather than the perfect ubiquitous in the formation of new The Germanic sixth class of strong verbs 217 preterits in the other strong verb classes. Since the intensives were a reduplicated o-grade formation, the perfect was not suitable as a clear model for an analogical preterit. As the original preterit of the intensives was an athematic imperfect, the root aorist was a much more likely candidate from the outset. Thus, I think that the imperfect took the place of an aorist at an early stage and came under the influence of the pattern which is reflected in OE hladan ‘to load’, hlōdon ‘they loaded’ after the merger of short *a and *o. This model was not available to present stems of the type CaRC-, which joined the seventh class of strong verbs, e.g. Gothic stautan ‘to strike’, skaidan ‘to separate’, blandan ‘to mi’x. Apart from the o-grade thematic presents and zero grade je-presents, there are a few o-grade je-presents, viz. Gothic fraþjan, skaþjan, hlahjan. These are probably old denominatives, cf. fraþi ‘understanding’, skaþis ‘harm’, also Russ. xóxot ‘laughter’, je-present xoxóčet ‘laughs’. They adopted the strong preterit as if they were primary verbs. Besides, there are three o-grade je-presents of roots in *-w-, viz. ON deyja ‘to die’, geyja ‘to bark’, kleyja ‘to scratch’, preterit dó, gó, kló. These may represent earlier intensives, cf. Lith. káuti ‘to strike’, where the rootfinal laryngeal yielded gemination in ON hNjggva, OE hēawan, which therefore joined the seventh class of strong verbs. NOTES For wahsjan cf. Avestan uxšyeiti ‘grows’. I think that the full vowel was generalized on the basis of the o-grade perfect stem. Though it is conceivable that the long vowel of the preterit wohs was taken from the reduplication syllable of *wōwahs-, this is improbable for chronological reasons. The je-present was replaced by the regular thematic flexion in North and West Germanic for differentiation from the causative *wahseje-, cf. ΟΝ vaxa vs. vexa, Avestan vaxšaiti ‘causes to grow’. It seems to me that hazjan must be identified with Indo-Iranian *śās- ‘instruct’. On *-safjan see Seebold 1970: 383, Schrijver 1991: 93. 1 Cf. again wahsjan, ΟΝ vaxa versus vexa. Differently swaran ‘to swear’, ON sverja, which belongs to the next category, cf. Latin susurrō ‘I whisper’. 2 THE GERMANIC FOURTH CLASS OF WEAK VERBS The formation of Gothic fullno-, 3rd pl. fullnand ‘become full’ cannot be separated from that of Vedic pṛṇ¡-, 3rd pl. pṛṇánti ‘fill’. It has long been recognized that the two paradigms cannot be identified, however. From a formal point of view, the expected stem form in Germanic should be *fullō-, with loss of the nasal and gemination of the preceding consonant. The attested verb is a denominative formation, derived from fulls ‘full’ < *plHnos. From a semantic point of view, the verbs of the fourth weak class are intransitives, which is contrary to what can be expected on the basis of the older Indo-European languages. This requires an explanation. It is probable that the nasal was always assimilated to a preceding consonant before the stress, cf. especially OHG storrēn ‘stand out’ beside stornēn ‘be rigid’, where the nasal was apparently restored, similarly sterro beside sterno ‘star’, ferro ‘far’ beside firni ‘old’, and swimman ‘swim’, Old Irish do-seinn ‘pursues’. As it is improbable that the nasal suffix was generalized on the basis of vocalic roots only, the fourth class of weak verbs must have originated from an extension of the suffix to stressed roots before the nasal assimilation (Kluge’s law, cf. K119). When we disregard the denominative verbs, the Gothic evidence for the fourth weak class is limited to the following material: af-lifnan ‘be left’, dis-skritnan ‘become torn’, us-gutnan ‘be poured out’, fra-lusnan ‘be lost’, ga-luknan ‘be shut up’, and-bundnan ‘become unbound’, ga-þaursnan ‘become dry’, us-bruknan ‘be broken’, af-taurnan ‘be torn away’, ga-skaidnan ‘become parted’, ga-waknan ‘awake’. It thus appears that the formation represents a nasal present derived from a middle root aorist with zero grade vocalism. This brings us to the problem of the aorist in Germanic. Elsewhere (K109) I have argued that the endings of the Germanic weak preterit reflect the root aorist *dē- < *dhē- and that in order to explain OE dyde ‘did’ we must assume that 3rd pl. *dēdunþ replaced earlier *dunþ. The latter form is important because it has zero grade vocalism in the stressed syllable. It suggests a paradigm with fixed stress, unlike Gothic stoþ, OE stōdon ‘he, they stood’, which reflect a more recent paradigm with mobile stress. In fact, the previous existence of a 3rd pl. form *stunþ provides the motivation for the creation of the form *stōdunþ. The Verner alternation in Gothic stoþ, OE stōdon suggests an early date for this analogical development. The paradigm of *dē-, 3rd pl. *dunþ is strongly reminiscent of the Vedic root aorist ádhāt, ádhur ‘he, they put’. Elsewhere (K065) I have argued that the 3rd pl. form originally had zero grade vocalism in the indicative and full grade vocalism in the injunctive, which would yield Proto-Germanic *e-dunþ, *e-stunþ versus *denþ, *stanþ. The zero grade vocalism of the reconstructed form *dunþ, 220 Germanic verbal inflexion which is reflected in OE dyde, now suggests that we have to start from an augmented root aorist *e-dē- in Germanic. When the augment was lost, the stress was probably fixed on the initial syllable of the paradigm. There are three weak preterits which directly reflect root aorists in Germanic (cf. K109: 108), viz. *kunþē- ‘knew’, *unþē- ‘granted’ and *wulþē‘ruled’, which are attested in Gothic kunþa, OE cūðe, ūðe, ON olla. These forms may be compared with Greek ἔγνω ‘perceived’, ὄνητο (for ὤνατο) ‘profited’, Vedic ávṛta ‘chose’. Here again, we may assume that the stress was fixed on the initial syllable when the augment was lost at an early stage. The historical paradigm was evidently built on the zero grade form of the root, e.g. *kununþ ‘they knew’, *wulþa ‘he chose’. Thus, it appears that the fourth class of weak verbs originated from an extension of the athematic nasal present suffix to the zero grade stem form of the middle root aorist which had received fixed stress on the initial syllable when the augment was lost. This yields the following relative chronology: (1) loss of the augment and establishment of fixed stress on the initial syllable in the root aorist, (2) derivation of nasal presents from middle root aorists with initial stress, (3) loss of fixed stress in the root aorist and expansion of the perfect, (4) Verner’s voicing of obstruents before the stress (cf. K102), (5) Kluge’s nasal assimilation before the stress (cf. K119), (6) Grimm’s rise of new fricatives and fixation of the stress on the initial syllable (cf. K102). It follows that I should have written *dunt instead of *dunþ etc. above. As this is not the place to discuss the Proto-Germanic consonant system, I have chosen to use a more traditional transcription. OLD NORSE taka, GOTHIC tekan, GREEK τεταγών In his analysis of this etymon, Karl-Heinz Mottausch lists the following Germanic verbs (1993: 152): Gothic tekan ‘touch’ < *tǃkan-; Old Norse taka and Middle Dutch taken ‘take’; Middle Low German tacken ‘touch’ < *takkō-; Old Frisian tetsia ‘appropriate’ < *takjan; Old High German zascōn ‘rob’ < *tak-skō-; Old English þaccian ‘pat’, Old Saxon thakolon ‘stroke’. The Gothic reduplicated preterit taitok can be identified with Latin tetigī < *tetagai ‘touched’ and Greek τεταγών ‘having seized’. The problem is that neither the initial t- nor the vowel -ē- of Gothic tekan is compatible with the regular sound laws. While the initial t- beside þ- seems to be a Germanic problem, the long vowel is also found in the Tocharian B present ceś- ‘touch’ < *tēk-. Mottausch derives the long vowel from *-ēH2- and reconstructs a static root present with an alternation between sg. *-tēH2g- and pl. *teH2g- (1993: 154f.). I am very unhappy about this traditional methodology of loosely postulating long vowels for the proto-language. What we need is a powerful theory which explains why clear instances of original lengthened grade are so very few and restrains our reconstructions accordingly. Such a theory has been available for over a hundred years now: it was put forward by Wackernagel in his Old Indic grammar (1896: 66-68). The crucial element of his theory which is relevant in the present context is that he assumed lengthening in monosyllabic word forms, such as the 2nd and 3rd sg. forms of the sigmatic aorist injunctive. Since the sigmatic aorist is the prototypical static paradigm in the verbal inflection, it offers the possibility of testing the relative merits of the two theories, Wackernagel’s lengthening in monosyllabic word forms versus a static paradigm with lengthened grade in the singular and full grade in the plural. As I have pointed out elsewhere (K065), the evidence substantiates Wackernagel’s view and forces us to reject the alternative because we find full, not lengthened grade in the 1st sg. form, e.g. Vedic jeṣam ‘conquer’, stoṣam ‘praise’. It is therefore reasonable to assume that originally the static present also had lengthened grade in the 2nd and 3rd sg. forms of the injunctive and full grade elsewhere. We now have to establish the nature of the static present, which is not a frequent type of inflection. I subscribe to Alexander Lubotsky’s unpublished theory that it must be derived from a reduplicated formation. A clear example is Vedic 3rd sg. t¡ṣṭi, 3rd pl. tákṣati ‘fashion’, which cannot be separated from 222 Germanic verbal inflexion Greek τέκτων ‘carpenter’ < *tetk-. Another example is Vedic 3rd sg. d¡ṣṭi ‘makes offering’ beside dákṣate ‘is able’ < *dedk- (cf. Lubotsky 1994: 204). I therefore think that Gothic tekan and Tocharian B ceś- represent a reduplicated paradigm which can be identified with Greek τεταγών. It may be useful to have a look at the place of this formation in the original verbal system. Following a line of thought developed by Pedersen (1921: 25f.) and Kuiper (1934: 212), I reconstruct a hysterodynamic s-present, 3rd sg. *tresti, 3rd pl. *trsenti, beside a static s-subjunctive (injunctive), 3rd sg. *tērst, 3rd pl. *tersnt, the coexistence of which is perhaps best preserved in Tocharian (cf. already K064: 1173), where we find B täs- < *dhH1es- beside A tās- < *dhH1s- in the present and B tes-, A cas- < *dhēH1s in the preterit of the verb tā- < *dheH1- ‘put’. If the reduplicated formations followed a similar pattern, we may reconstruct a hysterodynamic reduplicated present, 3rd sg. *wiwekwti, weak stem *wiwkw-, but with retracted stress in 3rd pl. *wewkwnti, cf. Vedic 3rd sg. síṣakti, 3rd pl. sáścati ‘accompany’ < *sisekwti, *seskwnti, beside a static reduplicated subjunctive (injunctive), 3rd sg. *wēwkwt, 3rd pl. *wewkwnt. This reconstruction actually explains the long *-ē- in the reduplication syllable of original reduplicated aorists, as opposed to original presents and perfects, in Tocharian (cf. K149: 173, and B tätt- ‘put’ < *dhidhH1-, *dhedhH1-). The rise of the static reduplicated paradigm may have been provoked by the raising of pretonic -e- to -i- in the reduplication syllable, which was probably earlier than Wackernagel’s lengthening. While the Tocharian B present ceś- < *tēke, subjunctive tek- < *tēk (with restored t-), and preterit teks- < *tēks (idem) all have original long *-ē-, short *-eis preserved in the optative taśi < *te(t)kī. The loss of the laryngeal in the root -(t)k- < *tΗ2g- shows that the lengthened grade arose by analogy with the paradigm of biconsonantal roots. The long -ē- of Gothic tekan cannot simply be derived from the same source because it is an isolated formation, and there is no indication that lengthened grade ever spread in the verb in Germanic. We therefore have to start from the stem *tetH2g- which is found in Greek and Latin, with a short vowel in the reduplication syllable. As interconsonantal laryngeals in medial syllables were lost in Germanic, the stem became *tetg-, assimilated to *tedg-, with restoration of the reduplication *dedg- (cf. already Mottausch 1993: 158). The latter development has a perfect analogue in Latin bibit beside Vedic píbati and Old Irish ibid ‘drinks’ < *pibe-. This solves the problem of the initial tbeside þ- in Old English þaccian ‘pat’ < *tH2g-, but not the problem of the long -ē- of Gothic tekan ‘touch’, which requires a reconsideration of the ProtoGermanic consonantal system. In a number of articles I have argued that the Indo-European unaspirated voiced plosives *b, *d, *g, *gw were actually preglottalized [’b], [’d], [’g], [’gw], and that the preglottalization was preserved after their devoicing to [’p], [’t], [’k], [’kw] in Germanic (cf. K138 and K142). The preglottalization has partly been Old Norse taka, Gothic tekan, Greek τεταγών 223 preserved in modern British English and western Danish and is reflected as preaspiration in West Norse and as gemination in all of North and West Germanic under various conditions. Theories which derive the western Danish glottalization from preaspiration or gemination, the West Norse preaspiration from gemination and the East Norse gemination from preaspiration, or the rise of glottalization, preaspiration and several layers of gemination from spontaneous local developments are all inadequate (cf. K072). In addition to the arguments which I have adduced elsewhere, I may now add an irrefutable piece of evidence from Faroese, which has preserved preaspiration before intervocalic plosives after originally short vowels, e.g. eta [e:hta] ‘eat’, opin [o:hpin] ‘open’ (cf. Petersen et al. 1998: 27; I am indebted to Dr Höskuldur Þráinsson for access to this volume). This preaspiration cannot possibly be derived from gemination because it remains distinct from the preaspiration before original intervocalic geminates, where a preceding short vowel was not lengthened. It must therefore be identified with the “illogical” gemination in Northumbrian, e.g. eatta (Lindisfarne Gospels, cf. Campbell 1959: 27, Brunner 1965: 189), and in Old High German, e.g. ezzan, offan. The Proto-Germanic preglottalization can be identified with the one in other Indo-European languages, e.g. Latvian êst ‘eat’ < *e’d-. Against this background we have to assume that *tetg- [tet’g-] was assimilated to *tedg- [te’d’g-], with restored reduplication *dedg- [’de’d’g-], and then simplified by the loss of dental articulation in the cluster *-dg-, which yielded a sequence of glottal stop [Ȥ] plus *-g- [’g]. If *H1 was still a glottal stop at the time, we can now write the stem as *deH1g-, which regularly developed into *dēg- [’de:’g-], later *tēk- [’te:’k-]. This explanation of the long -ē- in Gothic tekan is the same as has been proposed for the long -ē- in Greek πεντήκοντα ‘50’, ἑβδομήκοντα ‘70’, ἐνενήκοντα ‘90’ < *-H1kont- < *-dkomt- (cf. K060: 98f., Lubotsky 1994: 203). It has the advantage of accounting for both the initial t- and the long -ē- of Gothic tekan on the basis of a single property of the stem formation, viz. a reduplication which is also attested in Greek and Latin. Thus, we arrive at a Proto-Germanic aorist *tēk-, which gave rise to a Gothic 7th class verb tekan, a Scandinavian 6th class verb taka, and a variety of weak verb formations in West Germanic. Mottausch thinks that the Scandinavian verb is a “Neubildung” for *táka on the basis of the preterit tōk, which he derives from *tetōk- by loss of the reduplication syllable (1993: 159). This seems improbable to me for a number of reasons. First of all, the converse development is attested in the case of láta ‘put’, which has a secondary preterit lót beside lét on the basis of the unstressed variant lata, as Mottausch indicates himself. This verb belongs to the 7th class in all branches of Germanic. Secondly, the restructuring of the reduplicated preterit in North and West Germanic is based on the present stem, e.g. lēt < *leǃt, not **liót, cf. hlióp ‘leaped’ < *hleaup, Gothic lailot ‘let’, aiauk ‘added’. Thirdly, the 7th class was still productive in late Proto-Germanic, 224 Germanic verbal inflexion whereas the 6th class, which essentially reflects the root aorist (cf. K122), was not. This is especially clear from the verb arjan ‘plough’, which is a perfect candidate for a 6th class preterit but has a 7th class preterit in Old High German and a weak preterit elsewhere (cf. K120: 103). Fourthly, the West Germanic formations remain to be explained. After the merger of *ā with *ō in late Proto-Germanic, there must have been considerable pressure to change *tēk- into *tōk- so as to conform to the 6th class pattern, especially if there was a zero grade form with -a- in the paradigm. Gothic evidently created a (new) present tense by adding the regular present endings to the aorist stem *tēk- and substituted a newly formed perfect stem *tetōk- for the aorist stem in the preterit. There is no reason to assume that such forms ever existed in North and West Germanic, where they would probably have been preserved as a 7th class paradigm. Instead, we find tak- in Old Norse and Middle Dutch, with gemination in Middle Low German, with palatalization in Old Frisian, and with a sk-suffix in Old High German. This suggests either that there was no present tense of this verb at all in Proto-Germanic or that it was a suppletive formation. In either case the aorist stem *tēk- must have survived long enough to give rise to the derivative formations. The initial consonant of Old English þaccian ‘pat’ < *þakkō- < *tagnā- (cf. K119), of which Old Saxon thakolon ‘stroke’ is evidently a derivative, shows that this is an ancient formation. It can be identified with Middle Low German tacken ‘touch’ < *takkō- if the latter took its initial t- from the aorist stem *tēk-. Though this is predominantly a denominal type of inflection, it is perfectly possible that we are dealing here with an original nasal present which joined the regular type. Note that Middle Low German has preserved the original meaning attested in Gothic, Latin and Tocharian. There is no reason to assume that it is an intensive formation (thus Mottausch 1993: 152), especially because we cannot reconstruct another present tense for Proto-Germanic. Thus, I think that *þakkō- < *tagnā- and *tēk- < *tetg- originally belonged to the same paradigm, which can be directly compared with Latin tangō, tetigī ‘touch’. The elimination of the suppletive paradigm was a gradual process. While Gothic simply added present endings to the aorist stem *tēk-, this device was apparently no longer available when the process reached North and West Germanic. When Scandinavian substituted tak- for *þakkō-, the suppletive paradigm was still preserved in West Germanic. Old Frisian created a j-present *takj-, either on the analogy of 6th class verbs like Gothic hafjan ‘raise’ or as an intensive formation. The suppletive paradigm must still have existed at the time when Middle Low German adopted the initial t- of the preterit in tacken ‘touch’ < *takkō-. The Middle Dutch verb taken may be a borrowing because it belongs to the weak inflection and has the same meaning as Middle English taken and Old Norse taka (but cf. Kroonen 2009: 48). Old High German has only preserved a derivative of a deverbal noun in zascōn ‘rob’ < ‘make a grab’, cf. Old Norse taka, Gothic tekan, Greek τεταγών 225 forscōn ‘ask’ < ‘put a question’, eiscōn ‘demand’ < ‘lay a claim’. It thus appears that the original paradigm was retained longer in the Ingvaeonic and Low German territories than elsewhere. THE GERMANIC WEAK PRETERIT The main difficulty with the Germanic weak preterit is that one cannot endeavor an explanation of its origin without taking into account almost every aspect of the historical phonology and morphology of the Germanic languages. In the following I intend to show how a number of problems receive a natural explanation in a unified treatment on the basis of earlier studies. The theory presented here is not revolutionary but aims at integrating earlier findings into a coherent whole. There is no reason to give a detailed account of the scholarly literature, which is easily accessible (cf. Tops 1974, Bammesberger 1986a). The best starting-point for the discussion is perhaps the following quotation from Ball (1968: 186f.), to which I wholly subscribe: “It is surely a remarkable fact that the stem and dental of any and every weak verb are the same in the preterite and past participle. This immediately suggests either a common origin or that one is derived from the other. Now, the -to-participle is an IE formation while the weak preterite is Germanic, and, since a common origin seems out of the question, if they are related at all the dental preterite must be derived from the past participle. This hypothesis would avoid all the difficulties produced by Gothic wissa, brūhta, etc., which have been discussed above: it would, in fact, at once solve the problems both of the origin of the dental and of the form of the stem in the preterite-presents and class I preterites without medial vowel. And I have argued above that the class III preterites like OE hæfde can only be accounted for on the assumption that the weak preterite was introduced into this class at a far later date in Germanic.” As Ball recognizes, the “really serious problem is, of course, to account for the endings” (187): if the weak preterit “was a Germanic innovation, we might expect it to adopt a ready-made set of endings, such as those of the strong preterite” (183). This is where the verb ‘to do’ enters the picture: I agree with Ball that it “has always been the main strength of the composition theory that it provided a fairly satisfactory explanation of the endings” (183). The verb ‘to do’ has three different preterit stems in Germanic: *dud- in OE dyde, *ded- in OS deda, OHG teta, and *dēd- in OS dādun, OHG tātun. While *de- is evidently the reduplication syllable, the root forms *dē- and *du- must be derived from the root aorist, cf. Vedic ádhāt, ádhur ‘he, they put’. The coexistence of a perfect stem *dedō- and an aorist stem *dē- is corroborated by the 2nd sg. endings OHG -ōs, OS -os and Gothic -ēs, OE OS -es. Thus, I think that OS dedos and -des represent the perfect and the aorist of the verb ‘to do’, respectively. There is another root aorist which has survived into Germanic, viz. *stōþ, Vedic ásthāt ‘he stood’, which gave rise to a 3rd pl. form *stōdun(þ), cf. Gothic 228 Germanic verbal inflexion stōþ, OE stōdon. Similarly, the 3rd sg. form *dēþ gave rise to a plural form *dēdun(þ), OHG tātun, Gothic -dēdun. I thereby arrive at the following reconstruction of the Proto-Germanic root aorist of the verb ‘to do’: 1st sg. 2nd sg. 3rd sg. 1st pl. 2nd pl. 3rd pl. PGmc. *dēn *dēs *dēþ *dēdume *dēdude *dēdunþ Gothic -da -dēs -da -dēdum -dēduþ -dēdun ON -þa -þer -þe -þom -þoþ -þo OE -de -des -de OS -da -des -da -don -dun In order to account for OE dyde we must assume that *dēdunþ replaced earlier *dunþ at a stage which was more recent than the introduction of *du- into the optative (subjunctive). The perfect (strong preterit) of the verb ‘to do’ can be reconstructed as follows: 1st sg. 2nd sg. 3rd sg. 1st pl. 2nd pl. 3rd pl. PGmc. *dedōa *dedōþa *dedōe *dedume *dedude *dedunþ OS deda dedos deda dedun OHG teta -tōs teta -tum, -tōm -tut, -tōt -tun, -tōn The formation can be compared with Gothic saisō, ON sera ‘I sowed’. After the loss of final *-a, *-e, the 2nd sg. ending *-þ was evidently replaced by *-s on the analogy of the aorist (weak preterit), cf. Gothic saisōst, with added -t. This *-s spread to the other strong preterits in West Germanic on the analogy of the weak endings *-dæ, *-dǃs, a development which must have occurred at a stage when the Verner alternation of final *-s was still productive. On the form dedun cf. Lühr 1984: 39f. and 49f. We may now wonder if the development of the endings is in accordance with the Germanic Auslautgesetze. Elsewhere I have proposed the following rules for the phonetic development of final syllables in Germanic (K055: 172, K084: 437): The Germanic weak preterit PGmc. *-ō *-ōn *-ōns *-ōs *-ōt *-ōa(n) Gothic -a -a -os -os -o -o ON zero zero -ar -ar -a -a 229 OE -(u) -e -e -a -a -a OS -(u) -a -a -o -o -o OHG -(u) -a -ā -o -o -o Originally stressed *-ōs developed into OE -as, OS -os, OHG -ōs. The unstressed gen.sg. ending *-ōs was replaced by the acc.sg. ending *-ōn in West Germanic in order to eliminate the homophony with the gen.pl. ending *-ōan which resulted from the loss of *-s and *-an. The difference between *-ō and *-ōt is paralleled by the difference between Gothic -a < *-ai in the middle and -ai < *-ait in the optative (subjunctive). I do not share the usual view that the ON acc.sg. ending was replaced by the nom.sg. ending in giNjf ‘gift’ < *gebō, *gebōn because I fail to see the motivation for such a replacement, the two case forms being distinct in the other flexion classes of this language. The fem. acc.sg. form of the adjective spaka ‘wise’ has a pronominal ending. Like the introduction of the pronominal ending in the neuter form spakt, this is an innovation of Old Norse. The nonzero nom.sg. ending of ON hane ‘rooster’ was taken from the ion-stems (cf. Lid 1952). The reconstructed gen.pl. ending *-ōan was evidently a ProtoGermanic innovation (cf. K030). I see no evidence for tonal distinctions in Proto-Germanic. Here I add the expected reflexes of the corresponding front vowel endings: PGmc. *-ē *-ēn *-ēs *-ēt Gothic -a -a -ēs -ē ON zero zero -er -e OE -e -e -e(s) -e OS -a -a -e(s) -e OHG -a -a -e, -ēs -e Apart from the expected zero endings in Old Norse, the attested singular forms of the weak preterit appear to reflect a Proto-Germanic paradigm *-dēn, *-dēs, *-dē, without final *-þ in the 3rd sg. form. This is strongly reminiscent of the Balto-Slavic ē-preterit, which has a nominal origin (cf. K071: 256) and therefore suggests a derivation of the Germanic weak preterit from compounds with the PIE root noun *dhē- (cf. K064: 120), but it is more probable that the final *-þ was eliminated on the analogy of the strong preterit in view of the Gothic paradigm 3rd sg. -da, 3rd pl. -dēdun, which is otherwise difficult to explain. It appears that OS deda and OHG teta adopted the endings of the weak preterit. The Alemannic plural endings -tōm, -tōt, -tōn presuppose an earlier 3rd sg. form *tetō. 230 Germanic verbal inflexion Hollifield has argued that *ē always yielded *ā in North and West Germanic (1980). Though this may be correct for *-ē and *-ēn, the evidence is unfavorable in the case of *-ēi, *-ēu, *-ēr and *-ēs and inconclusive in the case of *-ēt. Moreover, I maintain that Proto-Germanic *ǃ was preserved in Ingvaeonic stressed syllables (K084: 440). Elsewhere I have argued for the following reflexes of long final diphthongs (K105: 5f.): PGmc. *-ēi *-ōi *-ēu *-ōu Gothic -ai -ai -au -au Runic -ai, -ē -ai, -ē -ō, -iu -ō ON -e -e -e -a OE -i -e -a -a OS -i -e -o -o OHG -i -e -iu, -e -o It appears that *-ēi and *-ōi remained distinct in OS and OHG, e.g. dat.sg. ensti ‘favor’ versus dage, tage ‘day’, and that *-ēu and *-ōu remained distinct in ON dat.sg. syne ‘son’ (Runic magiu) versus átta ‘eight’ and OHG suniu, sune versus ahto. The high reflex -i of *-ēi in OS and OHG and the fronted reflex -iu, -e of *-ēu in ON and OHG suggest that *ē was a front vowel when the long final diphthong was shortened to *-ei, *-eu, while the merger with the corresponding back vowel diphthongs in the other languages suggests that *ē was a low vowel at the time of the shortening, which was apparently early in OE and late in OHG. It follows that we must reconstruct *-ǃi and *-ǃu for North and West Germanic. There is no reason to assume different apophonic grades in these Germanic endings. In the case of Runic swestar ‘sister’ I assume preservation of PIE *-ōr and later replacement by the reflex of *-ēr in ON syster on the analogy of faþer, móþer, dótter. If PIE *-ēr had yielded *-ār, the rise of ON -er would be incomprehensible. Final *-ēs is found in the 2nd sg. ending of the weak preterit and in OHG 1st pl. -mēs, which can be compared with the corresponding long vowel ending in Lithuanian. The ON nonzero endings 1st sg. -a and 3rd sg. -e have not yet been explained. The attested older Runic endings are the following (cf. Antonsen 1975): 1st sg. -ō: Vetteland Stone (Norway, 350 AD), Einang stone (Norway, 350-400 AD), Gallehus gold horn 2 (Jutland, 400 AD), Rö stone (Bohuslän, 400 AD), Tune stone (Norway, 400 AD), Kjølevik stone (Norway, 450 AD), Ellestad stone (Östergötland, 550-600 AD). 1st sg. -aa: Etelhem clasp (Gotland, 500 AD). The Germanic weak preterit 231 3rd sg. -ai: Nøvling clasp (Jutland, 200 AD), Vimose chape (Fyn, 250-300 AD), Darum bracteate 3 (Jutland, 450-550 AD). 3rd sg. -ē: Garbølle wooden box (Sjælland, 400 AD), Halskov bracteate (Sjælland, 450-550 AD), Tjurkö bracteate 1 (Blekinge, 500 AD), By stone (Norway, 500-550 AD), Gummarp stone (Blekinge, 600-650 AD). These endings appear to reflect 1st sg. *-au or *-ōu, 3rd sg. *-ai or *-ōi, as if the optative endings had been added to the aorist or perfect endings of the Gothic forms. Here the OE paradigm of dyde comes to mind: it suggests that the aorist indicative adopted the endings of the optative when the perfect became the regular preterit of strong verbs. This leads us to an examination of the optative (subjunctive) endings in Germanic. The difference between Alemannic nāmi ‘took’ and suohtī ‘sought’ (Notker nâme versus suohtî) cannot be explained as a secondary development: it shows that the two paradigms represent different formations. While nāmi can be compared with wili ‘wants’ (Notker wile) and derived from *-īt, the weak form must be compared with Vedic 1st sg. dheyām, 3rd pl. dheyur, Gr. θείην, θεῖεν, and derived from *dheīt (cf. K065: 221). It provides conclusive evidence for the compound origin of the weak preterit. The Old English forms suggest an early substitution of *duī- for *deī- in the simple verb, and later replacement by *dudī-, which yielded dyde. This was evidently the subjunctive of the regular preterit *dedō-, *dedu- in Proto-Germanic times. It now appears that North Germanic disambiguated the weak indicative ending *-da by adding 1st sg. *-u, 3rd sg. *-i from the subjunctive *-diu, *-dii, which supplied a convenient model for disambiguation. Thus, I reconstruct Proto-Norse 1st sg. *-dau, 3rd sg. *-dai for the weak preterit indicative. It is clear that these endings cannot account for the West Germanic material. Following Collitz, Hollifield tries to demonstrate 1st sg. -a versus 3rd sg. -e for the Monacensis ms. of the OS Heliand, but this distribution is not supported by the evidence: the ratio of -a to -e in the first (I), middle (II-III), and final (IV-VI) part of the ms. is as follows (Hollifield 1980: 157): -a : -e 1st sg. 3rd sg. I 2:1 101:46 II-III 2:2 63:114 IV-VI 0:5 18:215 total 4:8 182:375 It follows that we have to start from a single homophonous ending *-æ which was first written -a and later -e (and twice -æ in the final part). The fronted character of this ending, as opposed to the regular endings of the ō-stems (Hollifield 1980: 152f.), may reflect the original timbre of Proto-Germanic *ǃ. It appears to differ from the even more fronted reflex of the dat.sg. ending *-ōi of the a-stems (Hollifield 1980: 156): 232 Germanic verbal inflexion -a : -e dat.sg. I 105:82 II-III 52:295 IV-VI 14:324 total 171:701 It must be investigated whether the differences can be attributed to the preceding consonant (cf. Lühr 1984: 75). In view of the general agreement between OS and OHG I assume that the expected strong preterit form *dedo, like *teto, adopted the weak ending. It has been proposed that the weak preterit represents the imperfect rather than the aorist of the verb ‘to do’ (e.g. Bech 1963, Lühr 1984). This hypothesis explains neither the absence of reduplication in Gothic -da, nor the long vowel of 3rd pl. -dēdun, OHG tātun. The derivation of these forms from a root aorist, to be compared with Gothic stōþ, OE stōdon, has the additional advantage of offering an explanation for OE dyde, as was pointed out above. It is highly improbable that the present stem *dedhē- survived beside aorist *dhē- and perfect *dedhō- when reduplication became characteristic of the strong preterit. It think that the attested present stem represents a thematic derivative *dōje- of the perfect and that the 1st sg. ending -m is secondary in this paradigm, cf. already early OHG 2nd sg. tōis, 3rd sg. tōit, tuoit, part. tōenti (Braune & Eggers 1975: 304). Now we turn to the problem of the stem form before the dental suffix. If the weak preterit must be derived from the past participle in the formations without a connecting vowel while the endings represent the root aorist of the verb ‘to do’, the origin of the weak preterit must be sought in compounds which are reflected as weak verbs with a connecting vowel. The correctness of this hypothesis is nicely corroborated by the existence of a class of verbs where the connecting vowel cannot have been introduced from the present tense. The Germanic first class of weak verbs originated from a merger of earlier je-presents, e.g. *bugje‘buy’, *wurkje- ‘work’, and eje-presents, e.g. *naseje- ‘save’, *kauseje- ‘probe’, as a result of Sievers’ law and raising of *e before *j. Elsewhere I have argued that the distinction between these two formations was preserved with short stems in Proto-Germanic (K088). In Gothic we find seven je-presents with a strong preterit (bidjan, hafjan, hlahjan, fraþjan, skaþjan, gaskapjan, wahsjan), five je-presents with a preterit in -ta (bugjan, waurkjan, brūkjan, þugkjan, kaupatjan), and eight primary je-presents with a preterit in -ida (hazjan, taujan, siujan, sōkjan, hrōpjan, wōpjan, þaursjan, faurhtjan). The connecting vowel was spreading in this language, as is clear from sōkida ‘sought’ (OE sōhte, OHG suohta), faurhtidēdun ‘they feared’ (OHG forahtun, cf. Krause 1953: 212), part. kaupatidai beside kaupastēdun ‘they buffeted’. It appears that brūhta ‘used’ and waurhta ‘worked’ replace earlier strong preterits in view of OE brēac and warhte beside worhte (cf. Bammesberger 1986a: 80), where the apophonic alternation cannot otherwise be explained; similarly Gothic brāhta ‘brought’ and þāhta ‘thought’ beside þūhta ‘seemed’, which gave rise to a secondary present *þankeje- The Germanic weak preterit 233 in Proto-Germanic times. Thus, I think that all strong verbs with a root in k or g and a je-present created a weak preterit on the basis of the past participle, which must have ended in -htas at that stage. It follows that the original first class of weak verbs had an alternation between *-eje- in the present and *-i- in the preterit, e.g. *nasejeþi, *nasidē(þ) ‘he saves, saved’. Elsewhere I have identified the stem *nasi- with the Indo-Iranian passive aorist as a neuter verbal noun ‘salvation’ (K044: 127f.), of which Gothic naseins is a derivative *nosi-H1n-i(K088: 29). The Germanic third class of weak verbs remains to be discussed. It has convincingly been argued that neither Gothic habaida, ON hafþa, -e, OHG habēta, nor OE hæfde, OS habda, Alemannic hapta ‘had’ can represent the original preterit of this class. As I have indicated elsewhere (K105: 7), I think that the inherited preterit was *habē-, without an intervening dental. This formation was replaced by *habdē- in West Germanic and by the present stem followed by the dental suffix elsewhere, and later also in Old High German. The formation of *kunþē- ‘knew’, *unþē- ‘granted’, *wulþē- ‘ruled’ is peculiar because we expect *d in the past participle. It seems to have originated from a root aorist 3rd pl. *kunþ (replacing *knunþ or *kununþ), cf. *dunþ above, and 3rd sg. middle *wulþa, Vedic ávṛta ‘he chose’. These forms reflect the original relation between nasal presents (*kunn-, *unn-, Vedic vṛṇ-) and root aorists, as opposed to underived presents with reduplicated perfects. The stem form *wulþ- cannot be identified with ON vald- ‘rule’ because the latter is identical with Lith. valdýti ‘to rule’, which has PIE *dh. It must rather be compared with Slavic velěti ‘to command’, which is a derivative of PIE *uel- ‘want’. The absence of a connecting vowel in Gothic wilda and the zero grade in OE wolde, OHG wolta suggest that these forms replace an earlier preterit *wulþ-, which apparently survived in ON olla ‘I ruled’. THE PROTO-GERMANIC PLUPERFECT The Germanic perfect presents (Präteritopräsentien) form a past tense by adding the endings of the weak preterit to the stem of the past participle, e.g. Gothic wissa ‘knew’. This is a recent formation (cf. K109). We may therefore ask ourselves if we can reconstruct the earlier formation which was ousted by the weak preterit. We may also try to recover the motivation for the replacement. There was no pluperfect in Proto-Indo-European. In Greek we find a derivative stem (ϝ)ειδη- ‘knew’, seemingly with the same suffix as in Slavic vědě-. If such a formation had existed in Germanic, it would hardly have been replaced by *wissē-, cf. Gothic witaida ‘observed’, which corresponds to Latin vidē-. The Vedic pluperfect can be defined as a perfect stem with secondary (aorist, imperfect) endings, e.g. ávedam ‘I knew’ (cf. especially Thieme 1929). This formation, which is occasionally found in Greek, may also have existed in Germanic. Indeed, I think that the hypothesis of a former pluperfect with secondary endings offers an explanation for a number of unclear points in Germanic historical morphology. The interpretation of Gothic (ni) ogs (þus) ‘(do not) fear’ as an injunctive (Hirt, Meid), subjunctive (J. Schmidt, Bammesberger), or irregular optative (Scherer, Hiersche) is not supported by independent evidence and must therefore be rejected (see Bammesberger 1986b for references and discussion). The form evidently represents a perfect stem with a secondary ending. It is difficult to separate OHG ni curi ‘noli’, Tatian ni curet beside ni curīt ‘nolite’ semantically from Gothic ogs and formally from the West Germanic strong preterit indicative, e.g. OHG 2nd sg. zugi, 2nd pl. zugut ‘drew’. The derivation of (ni) curi from an aorist optative (cf. Bammesberger 1986b: 676) is not supported by additional evidence and does not explain the plural form in -et, which can hardly be analogical. Moreover, the regular optative form ni churīs in the 2nd Reichenauer Glossar suggests that the plural form curīt replaced curet, not the other way round. Thus, the forms curi and curet appear to reflect the perfect stem with secondary ending *-es, *-ete. The usual view that the West Germanic 2nd sg. strong preterit ending -i was taken from the aorist (indicative, injunctive or optative) cannot be correct because both the model and the motivation for such a replacement are lacking. Apart from *dē- < *dhē- in the weak preterit and *stō- < *stā- in the sixth class strong preterit, which are not suitable as a model, it is difficult to find traces of the aorist in Germanic. It is highly improbable that an isolated 2nd sg. aorist ending replaced the regular perfect ending in a limited area without leaving a trace in the more archaic dialects. 236 Germanic verbal inflexion If the 2nd sg. ending -i < *-es belonged to a fully inflected paradigm, we can reconstruct 1st sg. *-om and 3rd sg. *-et, both of which yielded a zero ending in the attested Germanic languages. The merger of these endings with those of the perfect makes clear why the category disappeared. What remains to be discussed is the mechanism which produced the distribution of the 2nd sg. endings *-es and *-ta which is actually attested in the material. East and North Germanic preserved *-ta both in the perfect presents and in the strong preterit with the exception of stems in a long vowel, which adopted the weak endings in Scandinavian, e.g. 2nd sg. serer, 3rd sg. sere ‘sowed’. This type has a 2nd sg. ending -st in Gothic saisost. Conversely, the ending -t was added to the athematic 2nd sg. form of the verb ‘to be’ in Scandinavian est, later ert. Thus, we find spread of *-es after a vowel in the preterit and spread of *-ta after a consonant in the present tense. The West Germanic elimination of *-ta from the strong preterit and the addition of this ending in the athematic 2nd sg. forms, e.g. OE eart, bist, dēst, gǃst, wilt, can be viewed as a continuation of the same development. On the basis of what has been said I claim that we can reconstruct a thematic preterit of a perfect with secondary endings *-om, *-es, *-et, *-ete, at least for the perfect presents, where it was ousted by the newly formed weak preterit at a recent stage. The previous existence of this thematic formation explains the generalization of -u- as a tense marker in the plural endings -um, -ud, -un from *-me, *-te, *-nt in the perfect. It is hard to determine to what extent (if at all) the thematic formation supplied a real pluperfect to the strong preterit. This brings us to the question of the model and the motivation for the creation of the pluperfect. While most simple verbs probably had at least an aorist or a perfect at an early stage, derived verbs only had an imperfect beside the present tense. This holds for the causatives and iteratives (1st class weak verbs), denominatives (1st and 2nd class weak verbs), and intensives (6th and 7th class strong verbs with an original o-grade root vowel, cf. K122). While the former categories had a thematic imperfect which was eventually replaced by the weak preterit, the intensives were an athematic reduplicated formation, cf. Vedic jáṅghanti ‘strikes’, ádardar ‘pierced’. We may therefore wonder if Gothic lailaik ‘leaped’, OHG steroz ‘struck’, feang ‘seized’ directly continue an athematic imperfect. I think that this is not the case. The remarkable fact about the development of the Indo-European intensives in Germanic is not that they may form a 7th class strong preterit, which clearly represents the perfect, but that they do so only if the root structure prevented the formation of a 6th class strong preterit, which originated from the root aorist of roots in a long vowel (cf. K122). This is particularly striking in view of the fact that the root *ar- ‘plow’, which probably had aorist meaning in the northern Indo-European languages because it has a je-present in Celtic, The Proto-Germanic pluperfect 237 Germanic, Baltic and Slavic, did not join the 6th class of strong verbs but developed a 7th class strong preterit in Old High German iar- and a weak preterit in the other languages. We would therefore expect to find Gothic *saislah, *slahta or *slahida instead of sloh ‘struck’. It follows that the athematic imperfect of the intensives remained distinct from both the root aorist and the perfect. The solution to this anomaly seems to be that the athematic imperfect, like the athematic present, became thematicized at an early stage and did not therefore merge with either the root aorist or the perfect. A trace of the original athematic inflection may be preserved in Gothic reiran ‘to tremble’, OE rārian, OHG rēren ‘to roar’, where the preterit Gothic inreiraida ‘quaked’ suggests an original present stem *reiroi-, similarly gageigaidedjau ‘I might gain’. These verbs may have escaped early thematicization because their inflection was supported by the present stem *stai- ‘stand’ (cf. K105: 8). The remarkable fact that intensives with a root of the type CaR- or CaC-, but not CaRC-, created a long vowel preterit on the analogy of the root *stācannot be explained on the basis of a present stem *stistā-, as in Greek ἵστημι ‘I set up’, or *stand-, as in Gothic. I think that the latter stem form must be derived from an athematic imperative of a nasal present *standi, cf. Greek ἐσθίω ‘I eat’ from *ed- plus the Indo-European athematic imperative ending *-dhi, as in Vedic addhí ‘eat!’. It follows that we have to assume an inchoative nasal present *stanbeside the root aorist *stā-, which may be compared with Gothic fulln- ‘become full’ beside the preterit fullnō- which apparently ousted the root aorist attested in Greek πλῆτο. While the replacement of a root aorist by an imperfect in the case of fullnōis a natural development, the creation of an aorist beside a regular imperfect on the anomalous pattern of *stan-, *stā- is much more difficult to understand. If the original athematic imperfect *pulnāt became an aorist when the present tense was thematicized on the basis of 3rd pl. *pulnanti, it must have been the reduplication which prevented the same thing happening to the intensives. This renders the presence of a semantic distinction between aorist and perfect at that stage highly probable. The creation of a new aorist beside the imperfect in these intensives and inchoatives also demonstrates the presence of a semantic distinction between aorist and imperfect at the time of the thematicization. This is nicely corroborated by the perfect present Gothic kann ‘know’, which is evidently built on a thematicized imperfect *kunna-, cf. Vedic jānánti ‘they know’, ájānan ‘they knew’, Lith. žinóti ‘to know’. The athematic imperfect which turned aorist *kunnā- was replaced by *kunnē- in Gothic -kunnaida, obviously because the verb had durative meaning. The latter formation is found in competition with the original root aorist in ufkunnaida beside ufkunþa ‘recognized’. The creation 238 Germanic verbal inflexion of an athematic aorist *nasidē- ‘saved’ beside the thematic imperfect *naseje/a(cf. K109: 107) can probably be dated to the same stage of development. The thematic imperfect *kunna- was eventually ousted by the aorist kunþa ‘knew’. If the same development can be demonstrated for a perfect preterit rather than a perfect present, this vindicates the hypothesis of a real pluperfect. I think that conclusive evidence is provided by the verbs Gothic briggan ‘to bring’, brukjan ‘to use’, waurkjan ‘to work’, preterit brahta, bruhta, waurhta, OE brōhte, brēac, and warhte beside worhte. Since these forms represent analogical weak preterits derived from strong preterit stems, their original function can hardly have been anything else than that of supplying a pluperfect to a strong preterit. As in the case of the past tense of perfect presents, I think that they replaced a thematic formation. The thematic imperfect of intensives with a root of the type CaRC- was integrated into the perfect system, e.g. Gothic lailaik ‘leaped’, OHG steroz ‘struck’, feang ‘seized’, evidently because it was a reduplicated formation. This incidentally explains the absence of quantitative ablaut in the 7th class strong preterit in Gothic. We can now assume that these perfects were created in a similar way as kann was on the basis of *kunna-. The paradigm reconstructed above for OHG ni curi ‘noli’, ni curet ‘nolite’ fits into the picture rather nicely. When the pluperfect was lost as a tense, these forms survived in a modal function, like the Old Spanish pluperfect indicative in -ra which has become a past subjunctive in the modern language. The theory developed here provides an explanation for the remarkable redundancy which characterizes the personal endings in the oldest Germanic material, e.g. Gothic -a, -is, -iþ, -am, -iþ, -and after a present stem versus -Ø, -t, -Ø, -um, -uþ, -un after a preterit stem versus -au, -s, -Ø, -ma, -þ, -na after an optative stem. This awkward system becomes understandable if it resulted from the loss of an imperfect and a pluperfect which were formed from the present and preterit stems by the addition of a set of secondary thematic endings which regularly developed into -Ø, -s, -Ø, -am, -iþ, *-an. Note that the function of the thematic vowel was particularly unfortunate because it distinguished on the one hand the present and imperfect from the weak preterit (aorist) and on the other the pluperfect from the strong preterit (perfect). THE INFLEXION OF THE INDO-EUROPEAN ā-STEMS IN GERMANIC At the Eichstätt conference on the languages of prehistoric Europe (1999), the gentle Dirk Boutkan was unpleasantly surprised by an unexpected frontal attack launched by his then Leiden colleague Peter Schrijver. When the conference volume was published (2003), Dirk had passed away at the age of 37 and was therefore no longer in a position to answer his critic. Against this background I feel the need to clarify my own position in the debate. In the following I shall limit myself to the inflexion of the Indo-European ā-stems in Germanic. It is important to note that the principal differences of opinion between Boutkan and Schrijver are hidden in the footnotes to the latter’s diagrammatic presentation of the data (2003: 197), where Boutkan’s reconstructions *-ōs, *-ōns (1995: 163) appear as *-ōz, *-ōnz. Boutkan’s paradigm of the ā-stems is the following (1995: 222-224, abbreviations: P[roto-]G[er]m[anic], O[ld] R[unic], Ic[elandic], E[nglish], S[axon], H[igh] G[erman], N[ominative], A[ccusative], G[enitive], D[ative], I[nstrumental], s[ingular], p[lural]). Ns As Gs Ds Is Np Ap Gp Dp PGm *-ō *-ōn *-ōs *-ōi *-ō *-ōs *-ōns *-ōan *-ōmus Gothic -a -a -os -ai Ns As Gs Ds Is Np Ap Gp Dp PGm *-ō *-ōn *-ōs *-ōi *-ō *-ōs *-ōns *-ōan *-ōmus OE -Ø, -u -e -a, -e -e -os -os -o -om -a, -e -e, -a -a -um OR -u -o OIc -Ø -Ø -ar -u -oz -oz, -Az -o -Ø -ar -ar -a -om OS -Ø, -a/e -a/e -a/e -a/e -u -a/e, -o -a/e -o -un OHG -Ø, -a -a -a -u -o, -ā -ā -ōno -ōm This distribution of the endings differs from the one advocated by Schrijver in the following respects. 240 Germanic nominal inflexion The phonetically regular Ns ending was -a in Gothic and -u in North and West Germanic, with u-umlaut and loss of the ending in Scandinavian and with loss of the ending after long stems in West Germanic. The same ending is found in the Is form and in the 1s form of the present indicative. While Ns -u was preserved in Old English, where the Is form was lost, it was replaced by the As ending in OS -a/e and OHG -a, where Is -u was generalized as Ds ending. “A model for this development was provided by the o-stems, where the Ns and As had coalesced as a result of regular phonetic development” (Boutkan 1995: 225). Schrijver does not even mention this point. The As ending differed from the Ns ending by the final nasal, which evidently gave rise to a nasal vowel at an early stage and thereby inhibited the vowel from raising to -u in North and West Germanic, yielding OR -o. Like Gothic -a, OR -o was short, as is clear from the fact that it was subsequently lost with u-umlaut, as was Ns -u in Scandinavian. In West Germanic, the nasal vowel was delabialized and yielded OE -e, OS -a/e, OHG -a. This development may be compared with the lowering, delabialization and fronting in French masc. un ‘one’, as opposed to fem. une, where the high rounded oral vowel was preserved: “Close vowels such as [i], [y] and [u] are less liable to nasalization than more open vowels, for the reason that, as they involve raising the tongue almost to the maximum height possible without causing friction and so producing a consonant, the air passage through the mouth is already restricted and there is less ‘room for manoeuvre’ for the lowering of the velum in anticipation of a following nasal consonant.” (Price 1979: 8724). The view that the As ending was replaced by the Ns zero ending in Scandinavian cannot be correct because there is no motivation for this replacement, as the Ns and As forms are distinct in the other inflexional paradigms of this language. The fem. As form of the adjective spaka ‘wise’ has a pronominal ending. Like the introduction of the pronominal ending in the neuter form spakt, this is a Scandinavian innovation. Schrijver’s suggestion that the As zero ending was taken from the recessive u- or consonant stems (2003: 19711) is for the same reasons quite unacceptable, also because in the latter paradigms Ns -R < *-s was replaced by u-umlaut on the analogy of the ā-stems, e.g. mNjrk ‘mark’, Gs and NAp merkr. His statement that Boutkan does not discuss the problem is simply not true (cf. 1995: 139). It appears that West Germanic substituted the As for the Gs ending in order to avoid the merger with the Gp ending which resulted from the phonetic developments (cf. Boutkan 1995: 227). Like Schrijver, I agree with Dahl (1938: 141-143) that the Gs ending OE -a does not reflect original *-ōs but is the result of a secondary development of -e, which also affected the Ds and As forms. Schrijver raises three objections against the view that the As ending spread to the Gs in order to avoid homonymy with the Gp form (2003: 200): “In this way one homonymy is traded for another, but it is not made clear why the new homonymy would be more tolerable than the old one.” It is obvious that a The inflexion of the Indo-European ā-stems in Germanic 241 merger of the Gs and Gp forms always obliterates an important semantic distinction whereas the difference between Gs and As forms is usually no more than a superfluous syntactic index. “More to the point, OE had no difficulty with the homonymy Gsg. – Gpl. in the u-stems, where -a is the ending of the GDsg. and NAGpl. but not of the Asg.” This is simply not true: when Gs *-aus (OHG -oo) and Gp *-ewan (OHG -eo) merged into -a in Old English, the former was replaced by the corresponding ending of the o- and ā-stems (cf. Boutkan 1995: 256). “Furthermore, all stem classes preserve a formal difference between the Asg. and Gsg. if one was inherited.” This again is not true, as is clear from As for Gs hond ‘hand’, without -a, and guman ‘man’, not -en. Thus, Schrijver’s objections do not hold water. I have discussed the inflexion of the n-stems elsewhere (K219, cf. Boutkan 1995: 278-282). While Np *-ōs and Ap *-ōns merged in East and North Germanic when the nasal was lost, the two endings remained distinct in West Germanic because, as in the singular, the nasal vowel was delabialized, yielding Ap forms in OE -e, OS -a/e, OHG -ā (with length preserved before the lost consonant), contrasting with Np forms in OE -a, OS OHG -o. The Ap ending now replaced the Np ending in West Germanic, probably because the latter had become homophonous with the Gp ending. This development “did not reach the Kentish and West Saxon dialects, which had apparently left the continent by this time, and did not affect the pronominal flexion, where the homophony never arose” (K084: 438). Conversely, the Np ending replaced the Ap ending in West Saxon and Kentish in order to remove the homophony with the As ending which arose when the long vowel corresponding to OHG -ā was shortened. In German, the original Np ending -o was preserved in the peripheral Alemannic dialects. This explanation accounts in a straightforward way for the peculiar distribution of the attested forms. Schrijver’s supposition that Alemannic -o is a recent development (2003: 207) is at variance with the fact that it is limited to the earliest stage of the language. His view that the distinction between the Np and Ap forms of ā-stem nouns and adjectives was lost in Germanic, later restored in West Saxon on the basis of the demonstrative pronoun, then lost again in West Saxon by generalization of the nominal ending in the pronoun and the pronominal ending in the noun (2003: 208f.) is totally arbitrary, indeed spurious. I conclude that Schrijver’s attack on Boutkan was unnecessarily offensive and does not constitute a useful contribution to our knowledge of the Germanic Auslautgesetze. THE INFLEXION OF THE GERMANIC n-STEMS Elsewhere I have expressed my disagreement with the usual view that the Old Norse acc.sg. ending of the ō-stems was replaced by the nom.sg. ending, e.g. giNjf ‘gift’ < *gebō, *gebōn (K030: 298, K055: 172). I fail to see the motivation for such a replacement because the nom.sg. and acc.sg. forms are distinct in the other flexion classes of this language. The fem. acc.sg. form of the adjective spaka has a pronominal ending. Like the introduction of the pronominal ending in the neuter form spakt, this is an innovation of Old Norse. It follows that the expected nom.sg. ending of the n-stems is zero in Scandinavian. This explains the introduction of the front vowel from the jan-stems in hane ‘rooster’ (cf. Lid 1952). The inflexion of the Germanic n-stems has largely been clarified by Lane (1963) and Boutkan (1995), which enables me to refer to these authors for most points of discussion. Suffice it to emphasize that I start from nom.sg. *-ōn for all genders, masc. acc.sg. *-en- versus nom.acc.pl. *-on-, and neuter nom.acc.pl. *-ōn- < Indo-European *-ōn versus fem. *-ōn- < *-ā-n-. The alternation between sg. *-en- and pl. *-on- is also found in Armenian and must be ancient (cf. Meillet 1936: 79). This yields the following Proto-Germanic paradigms: nom.sg. acc.sg. gen.sg. dat.sg. nom.pl. acc.pl. gen.pl. dat.pl. masculine -ōn -enun -enas -eni -anes -anuns -nan -mus feminine -ōn -ōnun -ōnas -ōni -ōnes -ōnuns -ōnan -ōmus neuter -ōn -ōn -enas -eni -ōnō -ōnō -nan -mus As Boutkan points out (1995: 280), *-an- was raised to *-un- before the acc.pl. ending *-uns in North and West Germanic (Van Helten’s law). This *-un- spread to the nom.pl. and acc.sg. forms in the German area, where it appears as -on in the north (Saxon and Franconian) and -un in the south (Boutkan 1995: 285). The latter development did not reach the Ingveonic dialects, where the nom.pl. form in -an replaced the acc.pl. form, as in the case of the ō-stem ending -a replacing -e in the Old English dialects which left the continent at an early time (cf. K084: 438). In Scandinavian and Ingveonic, *-an- had evidently replaced *-en- in the singular at an early stage after Van Helten’s law. This leads to a reconstruction of the following paradigms after the apocope: 244 Germanic nominal inflexion acc.sg. gen.sg. dat.sg. nom.pl. acc.pl. dat.pl. Scandinavian -an -an -an -an -an -umR English -an -an -an -an -an -um Upper G. -un -en -in -un -un -um In the feminines, Boutkan assumes raising of final *-ōn to *-ūn after the apocope both in North Germanic and in the German dialectal area (1995: 284-290). This development, which may now properly be called Boutkan’s law, did not affect the Ingveonic dialects, where *-ōn had been shortened to -an at an early stage (ibidem). Thus, we arrive at the following paradigms: acc.sg. gen.sg. dat.sg. nom.pl. acc.pl. dat.pl. Scandinavian -ūn -ūn -ūn -ūn -ūn -ōmR English -an -an -an -an -an -um High G. -ūn -ūn -ūn -ūn -ūn -ōm As I think that the nom.sg. ending of Gothic guma ‘man’, OE tunge, OHG zunga ‘tongue’, also neuter OE ēage, OHG ouga ‘eye’ represents the phonetic reflex of Proto-Germanic *-ōn, we now turn to Go. tuggo, augo, ON tunga, auga (Runic -o), OE guma and OHG gomo, which must have taken their final vowel from the acc.sg. and nom.pl. forms after the delabialization of original *-ōn. This is unproblematic in the case of the East and North Germanic etyma because there was no suffixal variant *-an- in their paradigms, so that *-ōn could easily be restored on the basis of the medial variant *-ōn-. In the case of West Germanic it apparently requires a stage after *-an- had replaced *-en- in Ingveonic but before Van Helten’s *-un- spread from the acc.pl. to the nom.pl. and acc.sg. forms in the German dialectal area. On the basis of the Old Saxon evidence (cf. Boutkan 1995: 152-162), we may assume that at that stage the suffix *-an- was in fact [ån] while original *-ōn had developed into [æ], which was now replaced by [å] in West Germanic so as to yield OE guma, OHG gomo. This replacement can be viewed as an extension of the substitution of *-an- for *-en- in the acc.sg. form. The feminines could not be affected by a similar development in West Germanic because there was no final *-ō corresponding to the suffix *-ōn- after the raising of original *-ō to -u. When new *-ō originated from the loss of final consonants, earlier *-ōn- had apparently been shortened to -an or raised to -ūn already. The final question is about the chronology of the delabialization of original *-ōn. The Runic evidence for the acc.sg. ending *-ōn is limited to Einang (and The inflexion of the Germanic n-stems 245 perhaps Noleby and Eikeland) runo (cf. Nielsen 2000: 85, 152) and possibly Reistad wraita and Björketorp uþArAbA (Syrett 1994: 157-159). The interpretation of runo as acc.pl. (Syrett 1994: 119-132) is not convincing. The Runic evidence for the nom.sg. ending *-ōn is abundant. As Syrett has made perfectly clear (1994: 143-146), we find both -a and -o in the masculines and -o in the feminines (cf. also Grønvik 1998: 127 and Nielsen 2000: 154, 282). Masculine gender seems to be certain for Illerup wagnijo (also Vimose) and niþijo, probable for Udby lamo, Himlingøje hariso, Strårup leþro, Førde aluko, and possible for Berga fino and Vimose talijo. The ending -o appears to have petered out in the 5th century while -a is attested from the beginning of the Runic tradition, e.g. Illerup swarta, Vimose harja, and becomes general in the bracteates. If -o and -a represent consecutive stages of a nasal vowel [ån] which developed from original *-ōn, we may conclude that the delabialization was under way at the time of our earliest records and was completed in the 5th century. The same development affected the acc.pl. ending *-ōns in West Germanic, where *-s was assimilated to the preceding nasal, but not in North Germanic, where both this ending and the analogical nom.sg. ending *-ōn of the feminine n-stems became denasalized at an early stage (cf. Boutkan 1995: 142). The delabialization may be compared with the centralization of nasal vowels in French, cf. un bon vin blanc, where the vowels are much closer together than their oral counterparts are in une bonne vigne basse because lowering the velum for the articulation of the nasal vowel leaves little room for moving the tongue around. If Proto-Germanic *-ōn developed into the nasal vowel of bon and then into that of blanc [ån] while original *-ō was raised to -u, the spellings -o and -a are quite appropriate. The nasal vowel was sufficiently rounded to cause u-umlaut in Scandinavian before it was eventually lost in the same way as -u, yielding homophony in nom.acc.sg. giNjf. In the n-stems, the u-umlaut was evidently eliminated with the introduction of the front vowel from the jan-stems, e.g. hane. In the oblique cases, Stenstad igijon and Rosseland agilamudon (both 5th century) represent the suffix *-ōn- after the apocope but before the raising of -ōn to -ūn (Boutkan’s law). In the masculines, Kalleby þrawijan and Tune halaiban show that *-an- had already replaced *-en- in the 4th century, which supplies a terminus ante quem for Van Helten’s law. Tune arbijano exemplifies the generalized suffix *-an- in the gen.pl. form as well as the new post-apocope ending *-ō. A final example of original *-ōn is provided by the ending of Kjølevik minino and probably Strøm hino, the origin of which unfortunately remains unclear (cf. Boutkan 1995: 297-300). OLD HIGH GERMAN UMLAUT The paradigm of OHG anst ‘favor’, GDsg. ensti, NApl. ensti, Gpl. ensteo, Dpl. enstim shows in a straightforward way that *a was umlauted to e before a preserved i, but not before a lost *i. This is supported by the paradigm of man ‘man’, GDsg. man, NApl. man, where the root vowel was not umlauted before the lost endings. It suggests that the (phonemic) umlaut of *a to e was later than the loss of short endings after a long root syllable. It has been proposed that the absence of umlaut in NAsg. anst is due to influence from the consonant stems (Antonsen 1970). This cannot be correct because the two paradigms had no case endings in common before the apocope. The consonant stems had the same endings as the u-stems in the Asg., Apl. and Dpl. forms, whereas the u-stems merged with the i-stems in the Npl. and Gpl. forms, e.g. suni, suneo ‘sons’, cf. Gothic sunjus, suniwe. Thus, the influence of the i-stems upon the consonant stems is an indirect one, while the converse influence is recent: “Formen des G.D.Sg. ohne Endung (die mhd. sehr häufig sind) kommen im Ahd. nur äußerst selten vor” (Braune 1975: 201). Though the conditions under which *a was umlauted to e before an original *i or *j in the following syllable are rather well known (e.g. Braune 1975: 27f.), the development before original *e requires some discussion (cf. Cercignani 1979, with references). It seems to me that OHG beret ‘you bear’ is the unambiguous phonetic reflex of Proto-Indo-European *bherete. While the Monsee ending -it and the Alemannic ending -at are easily explained as analogical on the basis of the 3sg. and 3pl. endings, respectively (e.g. Braune 1975: 260), no such explanation is possible for the ending -et in the strong verbs because there was no model. It follows that the distinction between 3sg. -it < *-eti and 2pl. -et < *-ete is ancient. This is in accordance with the absence of umlaut in Old Icelandic falleþ ‘you fall’. There is an additional piece of evidence for the distinction between *-eti and *-ete in the first class of weak verbs. The absence of gemination in OHG zelit ‘tells’ and its presence in zellet ‘you tell’ show that *j was lost before *i but not before *e at a stage after the raising in *-eti. This yields the following relative chronology: (1) (2) (3) (4) (5) (6) raising of *e to i before *i and *j, loss of *j before *i, gemination before *j, raising of *e to i before *u, loss of final short vowels, umlaut of *a to e before i and *j. 248 German Stages (2)-(3) are West Germanic while stages (4)-(6) are specifically German. Dirk Boutkan has pointed out to me that such OHG doublets as gewi, gouwi ‘county’ suggest the converse chronology of (3) and (5): “Wörter mit w vor dem j wie geuui, G.Sg. geuues ‘Gau’, heuui ‘Heu’, gistreuui ‘Streu’, haben unumgelautete Nebenformen: gouwi, houwi, gistrouwi, meist in obliquen Kasus. Das Nebeneinander richtet sich lautgesetzlich danach, ob ein i oder ein Verdoppelung bewirkendes j folgte: gawi-, gawwj-” (Braune 1975: 190). I rather think that the apocopated NAsg. form *gawwj yielded *gawi before the early umlaut, which was apparently blocked by the cluster *ww in the same way as it was in garwen ‘to prepare’, scatwen ‘to shadow’. Voyles’ “alternative theory” (1991) is actually a return to pre-structuralist methodology. It follows from the above that the original paradigm of OHG hano ‘rooster’ can be reconstructed as Gsg. hanen < *-enos, Dsg. henin < *-eni, Asg. hanun < *-enum, Npl. *hanan < *-ones, Apl. hanun < *-onuns. For the u-infection in the Asg. form cf. Dpl. tagum ‘days’ < *-omus (K062: 106). The idea that this form had a different ablaut grade than the GDsg. forms seems highly improbable to me. It follows that the u-infection in Asg. hanun < *-enum was earlier than the West Germanic raising of *-ō to *-ū which led to the raising of *e to i in NApl. lembir ‘lambs’ < *-ezu < *-esā, cf. (4)-(6) above. This chronology seems to account for the material in an adequate way. The absence of umlaut in the subjunctive zalti ‘told’ < *zalteī (cf. K109: 105) is due to the analogical influence of the indicative. The new umlaut which arose at the end of the Old High German period yielded a different reflex than the earlier umlaut of *a to e, e.g. mähte ‘powers’, wähset ‘grows’, earlier mahti, wahsit. THE HIGH GERMAN CONSONANT SHIFT 1. Elsewhere (K075, K102, K119) I have put forward a new theory on the ProtoGermanic consonant shifts. In the following I intend to show how the High German consonant shift fits into the picture. The relevant dialect material has recently become more easily accessible through Goblirsch’s lucid and wellinformed treatise (1994), to which the reader may be referred for further details. 2. It may be useful to recapitulate the reconstructions which I have proposed for the successive stages of development from Proto-Indo-European to ProtoGermanic. For the argumentation I refer to the articles mentioned above. (1) Proto-Indo-European: plain fortes p:, t:, k:, kw: aspirated lenes ph, th, kh, kwh glottalic lenes p’, t’, k’, kw’ At this stage, all plosives were usually voiceless, as they are in modern Icelandic and the southern dialects of eastern Armenian. This is in agreement with the absence of a voiced counterpart to the PIE fricative *s. Initial *p’- now lost its glottalic feature and merged with its fortis counterpart *p:-, e.g. Vedic píbati ‘drinks’, Old Irish ibid (but Latin bibit with restored reduplication). We may also assume dissimilation of C’...C’ to C:...C’ and assimilation of C:...Ch and Ch...C: to Ch...Ch in roots at this stage. (2) Dialectal Indo-European: plain voiceless p, t, k, kw plain voiced b, d, g, gw glottalic voiced ’b, ’d, ’g, ’gw Apart from the development of the palatovelars, this is the system which can be reconstructed for Proto-Balto-Slavic, where the glottalic feature merged with the reflex of the PIE laryngeals into a glottal stop, e.g. Latvian pêʚds ‘footstep’ < *pe’d-, nuôgs ‘naked’ < *no’g-, cf. OE fōt, nacod. This is the starting-point for the Germanic developments. (3) Verner’s law: non-initial voiceless obstruents became voiced unless they were immediately preceded by the stress. As a result, the suffix of the participle in *-tó- practically merged with the formative element of the weak preterite *-dē-. Verner’s law did not change the system of obstruents but limited the distribution of the voiceless plosives and yielded a voiced variant *z of the phoneme *s. 250 German (4) Kluge’s law: the initial *n of a stressed suffix was assimilated to a preceding consonant. This development yielded a series of geminates which were in complementary distribution with the voiceless plosives. There is no distinction between plain and glottalic geminates. (5) Retraction of the stress in Gothic (cf. K102: 94). As a result, *z became a phoneme in this language, while the geminates were all but eliminated. (6) Grimm’s law: the voiceless plosives were lenited to fricatives and voicedness was lost as a distinctive feature. In Gothic, the glottalic feature was lost and the distinction of voicedness was restored, probably under the influence of its nonGermanic neighbours. (7) Retraction of the stress in North-West-Germanic yielded the following system of obstruents: plain fricatives f, θ, s, x, xw plain plosives p, t, k, kw glottalic plosives ’p, ’t, ’k, ’kw geminates pp, tt, ss, kk, kkw The voiced fricative *z became r, the labiovelars were eliminated, and the glottalization yielded a variety of reflexes (cf. K075, K102). 3. In Scandinavia we find a threefold reflex of the glottalization. Weakening of the glottal occlusion in West Norse yielded preaspiration, e.g. Icelandic epli ‘apple’, vatn ‘water’, mikla ‘increase’, hjálpa ‘help’, verk ‘work’. Weakening of the geminates also yielded preaspirated stops in this area. Conversely, assimilation of the glottal occlusion to the following plosive gave rise to new geminates in East Norse, e.g. Swedish vecka ‘week’, droppe ‘drop’, skepp ‘ship’, cf. ON vika, dropi, skip, OE wice, dropa, scip. In Danish, however, the geminates merged with the preglottalized stops, which were subject to lenition with preservation of the glottal occlusion in the western dialects, giving rise to the so-called vestjysk stød, e.g. hjæl’b ‘hjælpe’. It appears that preservation of the glottalization in the separate languages accounts for the existence of several layers of gemination, which can now be viewed as retentions rather than innovations, e.g. ON bekkr ‘brook’ < *-kj-, røkkr ‘dark’ < *-kw-, Swedish sätta ‘set’ < *-tj-, English apple < *-pl-, bitter < *-tr-, cf. Gothic baitrs. 4. The evidence for Proto-Germanic preglottalized stops is not limited to Scandinavian but can also be found in English and German. It is common knowledge that Standard English inserts a glottal stop before a tautosyllabic voiceless plosive, e.g. lea’p, hel’p. There is no reason to assume that this is a recent phenomenon. The High German consonant shift yielded affricates and geminated fricatives, e.g. OHG pfad ‘path’, werpfan ‘throw’, offan ‘open’, zunga ‘tongue’, salz ‘salt’, wazzar ‘water’, kind, chind ‘child’, trinkan, trinchan ‘drink’, The High German consonant shift 251 zeihhan ‘token’. These reflexes suggest a complex articulation for the ProtoGermanic voiceless plosives from which they developed. The origin of the gemination is unexplained in the traditional doctrine. If we start from the assumption that the Proto-Germanic plosives were preceded by a glottal occlusion which is preserved in the vestjysk stød and the English glottalization, the High German consonant shift can be explained as a lenition of the plosives to fricatives with a concomitant klusilspring of the preceding glottal stop. Note that the High German consonant shift has a perfect analogue in the English dialect of Liverpool, where we find e.g. [kx] in can, back, which also remains unexplained in the traditional doctrine. 5. The number of geminates was increased substantially by the West Germanic gemination of all consonants except r before a following *j, which was a comparatively recent development (cf. K123): (8) raising of *e to i before *i and *j, (9) loss of *j before *i, (10) gemination before *j. The absence of gemination in OHG zelit ‘tells’ < *-ljeti and its presence in zellet ‘you tell’ < *-ljete show that *j was lost before *i but not before *e at a stage after the raising in *-eti. The gemination before *j gave rise to the following system of obstruents: simple fricatives f, θ, s, x geminated fricatives ff, θθ, ss, xx simple plosives p, t, k geminated plosives pp, tt, kk simple glottalics ’p, ’t, ’k geminated glottalics ’pp, ’tt, ’kk The new geminated plosives did not merge with the Proto-Germanic geminates, which merged with the geminated glottalics instead, probably because they were longer. The glottalization was sometimes removed by analogy, e.g. in the words for ‘raven’ and ‘roe’ (cf. Lühr 1988: 332). Examples of geminated obstruents: OE skeþþan ‘injure’, cyssan ‘kiss’, hliehhan ‘laugh’, hebban ‘raise’, biddan ‘pray’, lecgan ‘lay’, scieppan ‘create’, settan ‘set’, leccan ‘moisten’. As Goblirsch has shown (1994: 11 and passim), the distinction between the plain and glottalic plosives, or lenes and fortes, is generally reflected as consonantal length in the modern languages, while voicedness and aspiration are concomitant features. Air pressure and muscular tension do not offer a reliable criterion because in Danish, lenis stops “have a greater tension than the fortes, which sounds like a terminological paradox” and in Icelandic, “unaspirated fortis and lenis stops were found to be tense compared to the aspirated fortes. In the studies on these two languages, 252 German the lack of firm closure associated with the aspirated stops was considered the decisive factor” (Goblirsch 1994: 9). In my view, a lack of firm closure in the latter part of the glottalic plosives was the origin of the High German consonant shift. 6. In High German, the glottalic plosives were lenited to fricatives with concomitant oralization of the glottal occlusion: (11) High German consonant shift. The new dental fricative, which I shall write z, remained distinct from the earlier alveolar fricative s. After a vowel, simple pf, tz, kx became ff, zz, xx, e.g. OHG offan ‘open’, wazzar ‘water’, zeihhan ‘token’. The interdental fricatives þ, þþ were subsequently shortened to t, tt. The full range of geminates is preserved e.g. in the North Tyrol dialect of Imst (cf. Goblirsch 1994: 35): rNjppǩ ‘Rabe’, hittǩ ‘Hütte’, prukkǩ ‘Brücke’, kxouffǩ ‘kaufen’, hNjassǩ ‘heißen’, lNjxxǩ ‘lachen’, huppfǩ ‘hüpfen’, sittsǩ ‘sitzen’, deŋkkxǩ ‘denken’. This is undoubtedly an archaism because the same is found in the isolated pockets of South Bavarian speech in Italy and former Yugoslavia (cf. Goblirsch ibidem and map 5 on p. 43). When “p, k became b, g” as a result of the High German lenition, this was not “a partial reversal of the second consonant shift” (Goblirsch 1994: 36), but of Grimm’s law, when distinctive voicing was lost. Similarly, the South Alemannic “strongly geminating” dialects have preserved an archaism. The geographical distribution of the “final strengthening” in Bavarian and Alemannic (Goblirsch 1994: 43, 45) shows that this is a retention, not an innovation, and must be compared with the same phenomenon in Icelandic (e.g. Haugen 1941: 101). The same holds for the alleged strengthening of initial obstruents in the North Bavarian dialects of Bohemia and Egerland, e.g. prōǩd ‘breit’, tum ‘dumm’, tōx ‘Tag’, kēm ‘geben’, and the Middle Bavarian dialects of Burgenland, e.g. pām ‘Baum’, taitš ‘deutsch’, tō ‘Tag’, krāw ‘grau’ (cf. Goblirsch 1994: 33). These peripheral dialects have preserved a precious archaism (cf. also Goblirsch 1994: 78f.). 7. The position taken here differs from Vennemann’s (cf. especially 1988, 1991, 1994b) in several respects. It is closer to the traditional doctrine in dating the High German consonant shift unambiguously after the West Germanic gemination, which I regard as its impulse. While I agree with Vennemann’s “repression theory” (1994b: 280f.), according to which the High German system of obstruents was gradually eliminated from north to south as a consequence of Franconian rule, I do not share his assumption of an early split between High German and the rest of the Germanic language family. If the High German affrication had preceded the West Germanic gemination, we would expect the Proto-Germanic geminates to have merged either with the simple fortes, as they in fact may have in Gothic, or with the geminated lenes, and the geminated fortes with the geminated fricatives. In my view, OHG helpfan differs from The High German consonant shift 253 Vestjysk hjæl’b in the oralization of the glottal occlusion and the frication rather than voicing of the obstruent which is found in English hel’p. THE ORIGIN OF THE FRANCONIAN TONE ACCENTS Elsewhere I have argued for the Proto-Germanic existence of a series of preglottalized voiceless stops which were preserved in English and yielded preaspiration in West Norse, gemination in East Norse, vestjysk stød in Danish, and affrication in High German (K102, K138, K211). In the remaining German territory, the glottalization may have affected the preceding vocalic segment in different ways. “In various Low German dialects, a length distinction on old and new long vowels arose in disyllables, depending on the phonation of the intervocalic consonant. [...] An example of this is the minimal pair ik riet /rit/ ‘I tear’ versus ik ried /ri:t/ ‘I ride’. [...] It is furthermore assumed that this length distinction is sometimes realized as an intonational opposition” (de Vaan 1999: 38). This development resembles the West Norse preaspiration and the East Norse gemination in the fact that the preceding vowel is cut short by the original preglottalized stop but differs from these because the vowel is not continued as a whisper and the following consonant is not geminated. Thus, the distinction between “long” and “overlong” vowels may have arisen from the loss of glottalization which shortened the preceding long vowel without either leaving a devoiced segment (as in West Norse) or lengthening the following consonant (as in East Norse). There is no reason to assume that voicedness played an independent role here. Björn Köhnlein’s M.A. thesis (2005) has led me to a reconsideration of the possibility that the rise of the Franconian tone accents can also be attributed to the loss of glottalization. The relevant data are the following. In Central Franconian, there is a distinctive opposition between a falling tone 1 and a stretched tone 2 that seems to be reversed in a strip of land along the southeastern border, which is formed by the “thick bundle of isoglosses separating Central Franconian from Rhine Franconian, the most characteristic one being the isogloss between the pronoun dat ‘that’ to the northwest and das to the southeast” (de Vaan 1999: 41). Phonetically, the Franconian tones strongly resemble the Latvian falling (`) and stretched (~) tones and the Lithuanian acute (1) and circumflex (2) tones, respectively. Since it has been argued that the Lithuanian acute (1) and the Latvian stretched (~) tone arose from a loss of glottalization (e.g. K025) while the distinction between the Franconian tones is to a large extent determined by the earlier presence or absence of a following preglottalized stop, it may be useful to consider the possibility of a similar origin. 256 German The distribution of the tone accents in the larger northwestern (A) and the smaller southeastern (B) parts of the Central Franconian area is as follows (cf. de Vaan 1999: 26-27 and Köhnlein 2005: 14-16): I. Non-high long vowels and diphthongs are falling in A and stretched in B. II. High long vowels and diphthongs, lengthened short vowels, and short vowels with tautosyllabic resonants are stretched in A and falling in B when they are followed by an original final consonant or non-final preglottalized stop. III. Elsewhere these vowels and sequences are falling in A and stretched in B, except lengthened short vowels, which are falling in both A and B. Thus, it appears that glottalization was lost after non-high long vowels and diphthongs at an early stage, after which it yielded a stretched tone in A, as in Latvian, and a falling tone in B, as in Lithuanian. The falling tone in A and the stretched tone in B were evidently the unconditioned, unmarked (“spontaneous”) reflexes before the lengthening of short vowels in open syllables and the apocope blurred the picture and rendered the distribution of the tones opaque. We must now ask: how did the bifurcation of glottalization into a stretched tone in the northwest and a falling tone in the southeast come about? It is important to understand the phonetic influence of glottalization on word melody. When the glottal closure is formed, the vocal cords are tightened so that the pitch of the sound goes up. The flow of air is then interrupted and subsequently continued at a lower pitch when the glottal closure is released. When a following voiceless consonant is short, the rise of the pitch may be more prominent than its fall, but when it is long, the fall of the pitch may be more prominent than its rise. As a result, we expect a falling tone near the Rhine Franconian area, where the glottal stop was oralized and lengthened the following fricative, but a stretched tone in the northwest, where the rise of the pitch before the short interruption of the air flow prevented the pitch from falling below the level it would reach without the glottal closure. This is indeed what we find. Thus, the bifurcation of the glottalization into a stretched tone in A and a falling tone in B is explained by the relative prominence of the rise of pitch before and the fall of pitch after the glottal closure. The word melody was then transferred to the initial syllable, yielding a tonemic opposition on long nuclei. The transfer was clearly anterior to the apocope because original monosyllables adopted the new, marked tone, as if followed by a word-initial glottal stop. It also preceded the lengthening of short vowels in open syllables in B because these never developed a stretched tone. The rise of the stretched tone in A and the falling tone in B may to some extent be compared with the rise of preaspiration in West Norse, where glottalization cut short the preceding vowel, and gemination in East Norse, The origin of the Franconian tone accents 257 where it lengthened the following consonant. The big difference is that the rise of a devoiced segment in West Norse and of an oral occlusion in East Norse eliminated the pitch effects and thereby prevented the rise of a distinct tone pattern. There was no such long voiceless segment in the Low German dialects which developed a distinction between “long” and “overlong” vowels, which “is sometimes realized as an intonational opposition” (de Vaan 1999: 38). THE ORIGIN OF THE OLD ENGLISH DIALECTS 1. It has been argued that the Old English dialects either reflect old tribal divisions or developed after the Anglo-Saxon emigration. I think that neither view is correct. In the following I intend to show that the early divergences between West Saxon and Kentish on the one hand and Anglian (Mercian and Northumbrian) on the other are the result of a chronological difference between two waves of migration from the same dialectal area in northern Germany. 2. Hans Nielsen (1981a: 251-252) lists thirteen pre-invasion correspondences between Anglian and continental Germanic languages and four correspondences of Kentish and West Saxon. I shall briefly review the material. 2.1. The usual nom.acc.pl. ending of the ō-stems is -a in West Saxon and Kentish and -e in Anglian. Since Kern (1906) has shown that the former represents the original nominative and the latter the original accusative ending, they can be derived from Proto-Germanic *-ōs and *-ōns, respectively. Though it is usually assumed that the nasal was lost in the latter ending (e.g. Hollifield 1980: 43), there is no evidence for this view (cf. in this connection Beekes 1982: 55). There is no reason to postulate tonal distinctions for Proto-Germanic. As I have indicated elsewhere (K055: 172), I assume the following developments of ProtoGermanic final syllables: PGmc. *-ō *-ōn *-ōns *-ōs *-ōt *-ōa(n) Gothic -a -a -os -os -o -o ON -Ø -Ø -ar -ar -a -a OE -(u) -e -e -a -a -a OS -(u) -a -a -o -o -o OHG -(u) -a -ā -o -o -o Apart from the compensatory lengthening in Old High German, ProtoGermanic *-ōns merged with *-ōs in the north and the east, and with *-ōn in the west. This divergence must evidently be connected with the different chronology of the rise of nasal vowels on the one hand and the loss of *-s on the other. The acc.sg. ending of the ō-stems (PGmc. *-ōn) replaced the gen.sg. ending (PGmc. *-ōs) in West Germanic in order to remove the homophony with the gen.pl. ending (PGmc. *-ōan, cf. K030: 293). For the same reason the acc.pl. ending (PGmc. *-ōns) replaced the nom.pl. ending (PGmc. *-ōs), but this development did not reach the Kentish and West Saxon dialects, which had apparently left the continent by this time, and did not affect the pronominal flexion, where the homophony never arose. Conversely, the nom.pl. ending 260 English replaced the acc.pl. ending in West Saxon and Kentish in order to remove the homophony with the acc.sg. ending. The latter development has nothing to do with the merger of nom.pl. *-ōs and acc.pl. *-ōns in Old Norse and Gothic, which resulted from the phonetic loss of the nasal. 2.2. The nom.pl. form Mercian oexen, Northumbrian exen, Old Frisian ixen, West Norse yxn, øxn preserves the zero grade of the proto-form *uksnes ‘oxen’, which was lost elsewhere. 2.3. Mercian and Northumbrian share with Old Norse the preservation of the zero grade gen.sg. ending -ur, -or in the word for ‘father’. This ending was original in the words for ‘mother’ and ‘brother’, from where it spread to the word for ‘father’ in North and West Germanic. Unlike the other languages, Old English preserves the difference between acc.sg. fæder on the one hand, and mōdor and brōþor on the other. 2.4. The Kentish hapax lǃresta can be identified with Old Frisian lērest ‘least’. This unique correspondence between Kentish and a continental language is insignificant. 2.5. Kentish and West Saxon share with Old Frisian and Old Saxon the use of the dative for the accusative of the 1st and 2nd sg. personal pronouns. This is apparently a common North Sea Germanic innovation which can be dated to the period before the migrations. On the other hand, the dialect from which Old High German evolved differentiated the accusative from the dative of the 1st and 2nd pl. personal pronouns by the addition of *-ik. The latter innovation spread to the Anglian dialects of Old English, leaving traces in Old Saxon and Old Low Franconian, but not in West Saxon or Kentish, which had apparently left the continent at the time already. The long accusative forms are evidently stylistically marked in Old English and disappear after the oldest records. 2.6. The Northumbrian 1st pl. possessive pronoun ūsa, which has been preserved in modern dialects between York and Stafford (cf. Orton et al. 1978, map M75), corresponds to Old Saxon ūsa, Old Frisian ūse, and Old Low Franconian unsa, representing an innovation in comparison with Gothic unsar, Old High German unserēr, Old Franconian unsēr, and the usual Old English forms ūser, ūre. The innovation can evidently be dated to the period of the last migrations. It did not affect the 2nd pl. form ēower, Old High German iuwerēr, Old Franconian iuwēr, cf. Old Saxon iuwa, Old Frisian iuwe, Old Low Franconian iuuwa. 2.7. The Northumbrian acc.sg. form of the masc. demonstrative pronoun þene beside þone can be identified with Old Frisian thene, Old Saxon thena beside thana, Middle Dutch dien beside Old Low Franconian thana, Old High German den. The introduction of e-vocalism in the acc.sg. form evidently spread from The origin of the Old English dialects 261 the southern dialects to the north and reached the pre-English dialects at the time of the last migrations. It also affected the interrogative pronoun in Old High German (hwenan, wen) and Old Saxon (hwena), but not in Old English (hwone). 2.8. Anglian has preserved the reduplicated preterits heht ‘called’, leolc ‘played’, leort ‘let’, reord ‘advised’, cf. Gothic haihait, lailaik, lailōt, rairōþ. 2.9. The r-forms in the present tense of the verb ‘to be’ are more widespread in Anglian than in West Saxon and Kentish. The -r- probably spread from the 1st pl. form *erum, ON erom, OHG birum, to 2nd pl. ON eroþ, OHG birut, and subsequently to 3rd pl. ON ero, OE (e)aron, -un, and 2nd sg. OE eart, (e)arþ. It was eventually generalized in Scandinavian. The West Saxon 2nd sg. form eart must have originated before the substitution of 3rd pl. sind(on) for the 1st and 2nd pl. forms. Since the latter development was a shared innovation of Old English, Old Frisian and Old Saxon, it must be dated to the period before the migrations. The Anglian plural form (e)aron must therefore be regarded as an archaism representing an earlier common innovation. 2.10. West Saxon preserves the original coexistence of 1st sg. eom and bēo ‘am’. The latter form adopted the -m of the former in Anglian bēom, bīom, Old Saxon bium. This was apparently a shared innovation dating from the time between the early and the later migrations. Conversely, the former paradigm adopted the b- of the latter in Old High German bim, bist, birum, birut, Old Frisian bim, Middle Dutch bem. 2.11. The rare Northumbrian 2nd sg. form of the verb ‘to be’ is can hardly be identified with Gothic is. It is probably the 3rd sg. form used for the 2nd sg. 2.12. The Anglian plural form dedon beside dydon ‘did’ can be compared with Old Frisian deden, Old Saxon dedun or dādun, Old High German tātun, Gothic -dedun. It represents an archaism in comparison with OE dydon. 2.13. Reflexes of *waljan in the paradigm of the verb ‘will’ are frequent in Old High German, but occur also in Old Saxon, Old Frisian, and Anglian. In the preterit, the stem is limited to Anglian walde and Old Saxon walda. It is evidently an innovation which reached pre-English from the south at the time of the later migrations. 2.14. The reflex of Proto-Indo-European *ē is ē in Anglian, Kentish, Old Frisian, and Gothic, ǃ in West Saxon, and ā in Old High German, Old Saxon, and Old Norse. There are also instances of ē in Old Saxon, e.g. bērun, lēsun. Van Wijk (1911) has established that the original reflex of PIE *ē is ǃ in West Flanders, Zealand, South Holland, Utrecht, and the southern part of North Holland, whereas it is ē in the remainder of North Holland and in Old Frisian. The reflex ā invaded the Low Franconian area from the south, while the entire coast from 262 English Flanders to Ostfriesland preserved the fronted reflex until the eleventh century (cf. Gysseling 1962: 7-8). There can be little doubt that Dutch and West Saxon ǃ is an archaism. The retraction of this vowel to ā in Old High German, Old Saxon, and Old Norse must be viewed in connection with the rise of ē2, and its raising to ē in Anglian, Kentish, Old Frisian, and Gothic with the monophthongization of ai. The twofold reflex in Old Saxon is matched by a twofold reflex of ai in the same area. We can therefore date the raising of ǃ to ē in Anglian, Kentish and Old Frisian to the period of the migrations. 3. A reconsideration of the correspondences between the Old English dialects and the continental Germanic languages shows that the early divergences between Anglian and West Saxon can be explained from a chronological difference between two stages of a single continental dialect. In comparison with Anglian, West Saxon has preserved two structural archaisms: the nom.pl. ending of the ō-stems -a, and the reflex ǃ of PIE *ē. On the other hand, Anglian has preserved five accidental irregularities: the umlauted nom.pl. form of ‘oxen’, the zero grade gen.sg. ending of ‘father’, the reduplicated preterits, and the plural forms (e)aron ‘are’ and dedon ‘did’. Three of these retentions are also found in Old Norse, one in Gothic, and one in Old Saxon and Old High German. Besides, Anglian differs from West Saxon as a result of seven innovations shared with continental West Germanic languages: the substitution of the acc.pl. ending of the ō-stems -e for the nom.pl. ending, the creation of a distinct accusative of the 1st and 2nd sg. personal pronouns, the creation of the 1st pl. possessive pronoun ūsa, the introduction of e-vocalism in the acc.sg. form of the masc. demonstrative pronoun, the creation of 1st sg. bēom ‘am’, the spread of *waljan to the paradigm of the verb ‘will’, and the raising of ǃ to ē. These developments, all of which have at least left traces in Old Saxon, can be dated to the period after the early migrations. 4. According to the explanation put forward here, we must distinguish between an earlier, “Saxon”, and a later, “Anglian” migration. One may wonder if there is any historical evidence for this view. Nielsen states that the Saxons lived in present-day Holstein according to Ptolemy (2nd century) and appear to have been in control of the whole region between the Elbe and the Weser from the middle of the third century (Nielsen 1981a: 265). They reached the Netherlands in the fourth century. The Angles can hardly be separated from the present-day district of Angeln in eastern Schleswig. I would suggest that “Anglian” refers to the original Saxons of Angeln, more or less as the French word allemand refers to the original Germans of Alemannia. As Nielsen (1981a: 271) points out, Bede does not always observe the distinction between Angles and Saxons, and the eventual preference for the term “Anglian” is probably due to its distinctiveness from the continental Saxons. The origin of the Old English dialects 263 The traditional designation for the Germanic invaders in Celtic sources is “Saxons”. This name was evidently established at the first stage of the invasion, which can be identified with the period from the time of Vortigern (around 450) until the battle of Mount Badon (about 500, cf. Jackson 1953: 199). There followed almost half a century of peace, the “Saxons” having settled in Kent and Sussex. The territory of Essex and Middlesex was largely uninhabited at that time. In the north, “the great gateway by which the Angles penetrated into the north Midlands and Yorkshire was the estuary of the Humber” (Jackson 1953: 207). Though in the Yorkshire Wolds and at York itself “archaeological finds seem to indicate a more or less unbroken continuity of occupation between the late Roman and pagan Saxon periods” (Jackson 1953: 212), there is no historical evidence for a kingdom of Deira before the second half of the sixth century. During the latter period Deira must have gained considerable strength in view of the spectacular expansion after the battle of Catterick about 600 which is described in the Gododdin. It seems that the battle of Catterick can be viewed as the northern equivalent of the battle of Mount Badon, except for the fact that it was won by the other side. Thus, I suggest that the “Saxon” invasion yielded the conquest of Kent and Sussex in the fifth century, whereas the “Anglian” invasion can be connected with the subjugation of the north which started around the middle of the sixth century. There is no linguistic evidence for a different continental homeland, especially because the shared innovations of Anglian and Old Saxon point to geographical contiguity after the early migrations. HOW OLD IS THE ENGLISH GLOTTAL STOP? The discussion of the English glottal stop by Christophersen (1952) and O’Connor (1952) motivated Anatoly Liberman to reconsider the problem of its origin (1972). The main difficulty is that “neither in RP, nor in the dialects can the glottal stop ever differentiate meaning,” so that “we must trace whether the glottal stop has not yet achieved relevancy or already lost it” (Liberman 1972: 50, 51). In his later work, Liberman definitely opts for the second possibility (1982: 237). Here I intend to make clear why I agree with this view in spite of the fact that I do not subscribe to his accentological theory. In his survey of the earlier literature, Andrésen comes to the conclusion that “there is strong evidence in favor of the view that about 1860 the phenomenon of pre-glottalization existed only in a few dialects in Western Scotland” (1968: 24). However, it is clear from his examples that this statement refers to the glottal stop which replaces [p, t, k], not to the concomitant glottal closure which accompanies these sounds in more widespread varieties of English. As Andrésen points out himself, it is “not impossible that pre-glottalization started as a characteristic feature of a certain class dialect, viz. the dialect of the working classes in the big industrial areas” (1968: 29). The spread of the replacing glottal stop and its social stigma adequately account for the “increasing space given by phoneticians from about 1920 onwards to the treatment of the glottal stop” (Andrésen 1968: 34), but this does not indicate that the “reinforcing” glottal closure of [’p], [’t], [’k] “is actually on the increase among educated people” (ibidem). In fact, there is evidence to the contrary. Collins and Mees have recently advanced our knowledge of the matter by listening “to a number of pre-1930 audio recordings, together with two recordings of later date, to hunt for what evidence, if any, could be found of glottalisation in the speech of people who had been born in the latter half of the nineteenth century” (1994: 75). They were impressed by “the general pervasiveness of glottalisation in the material we have at our disposal. Far from having to search for odd examples, as we thought might be the case at the outset of our investigation, we have found glottalisation in the speech of all our subjects, even in formal delivery” (1994: 78). They conclude that glottalization was well-established in upper-class English speech by the latter half of the nineteenth century and suggest that “this would imply that glottalisation was even more widespread in the standard language than our observations would indicate” (1994: 79). This is in accordance with the view that earlier glottalization went unnoticed because it was not distinctive. We may therefore have a look at the comparative evidence. 266 English The comparison of the English glottal stop with the Danish stød is commonplace. There are two varieties of stød in Danish. The standard Danish stød appears in monosyllables which have an acute pitch accent in Swedish and Norwegian. Though its distribution has partly been obscured by analogical levelings, it seems clear that it developed from a falling tone movement. The socalled vestjysk stød of the western dialects is an entirely different phenomenon because it is characteristic of original polysyllables which have a circumflex accent in Swedish and Norwegian. It cannot be connected with the Jylland apocope because it is also found in the northeastern part of vestfynsk dialects, where the apocope did not take place. In his monograph on the vestjysk stød, Ringgaard concludes that “the v-stød is only found immediately before the plosives p, t, k, and that it is found wherever these stand in an original medial position, following a voiced sound in a stressed syllable. The exceptions to this are certain types of loan-words from a later period” (1960: 195). He dates the rise of the vestjysk stød to the twelfth century because it is characteristic of “all then existing medial plosives” (1960: 199). The view that the vestjysk stød is a spontaneous innovation of the westernmost dialects of Danish, which Jespersen had in fact already proposed almost half a century earlier (1913: 23), can hardly be called an explanation. Moreover, it does not account for the vestjysk stød in the isolated pocket of dialects on the island of Fyn, which suggests that it is a retention rather than an innovation. The hypothesis of a local origin also neglects the parallel development of preaspiration in Icelandic and of the glottal stop in English. Both the vestjysk stød and the preaspiration receive a natural explanation if we assume that early Proto-Germanic possessed a series of preglottalized voiced stops ’b, ’d, ’g (cf. K075: 196, K102: 8). Devoicing yielded a series of late ProtoGermanic sequences ’p, ’t, ’k, the glottal stop of which was lost under various conditions. Weakening of the glottal stop in West Norse yielded preaspiration, while its assimilation to the following plosive gave rise to a series of geminates in East Norse, with the exception of Danish, where the sequences were subject to lenition and the glottal stop was preserved in the vestjysk dialects. Apart from the straightforward explanation of the vestjysk stød and the Icelandic preaspiration, the reconstruction of Proto-Germanic preglottalized stops has the advantage of accounting in a principled way for the existence of several layers of gemination, which can now be viewed as retentions rather than innovations (cf. K102: 7). Firstly, it is possible that the unexplained gemination in Swedish, e.g. in vecka ‘week’, droppe ‘drop’, skepp ‘ship’, reflects a dialect which escaped an early loss of the glottal stop, in contrast with Old Norse vika, dropi, skip, Old English wice, dropa, scip. Secondly, mp, nt, nk yielded pp, tt, kk in the larger part of Scandinavia. This development becomes understandable if we assume that the nasal consonant was devoiced by the preaspiration of the following plosive and subsequently lost its nasal feature. Thirdly, *k was geminated before *j and *w, How old is the English glottal stop? 267 e.g. in Old Norse bekkr ‘brook’, røkkr ‘dark’. Similarly, *t was geminated before *j in a limited area, e.g. Swedish sätta ‘to set’. (West Germanic geminated all consonants except r before *j and is therefore inconclusive.) Fourthly, the stops p, t, k were geminated before l and r in West Germanic, e.g. English apple, bitter, cf. Gothic baitrs. The same development is found sporadically in Scandinavia, which suggests that we are dealing with the loss of an archaic feature rather than with an innovation. Here again, the geminate may have originated from the assimilation of a glottal stop to the following plosive. Thus, I propose that the English glottal stop directly continues the glottalic feature of the Proto-Germanic preglottalized stops. The gemination in Old High German offan ‘open’, wazzar ‘water’, zeihhan ‘token’ also suggests a complex articulation for the Proto-Germanic voiceless plosives from which they developed. The origin of the gemination is unexplained in the traditional doctrine. If we start from the assumption that the Proto-Germanic plosives were preceded by a glottal stop which is preserved in the vestjysk stød and the English glottalization, the High German sound shift can be explained as a lenition of the plosives to fricatives with a concomitant klusilspring of the preceding glottal stop. Note that the High German sound shift has a perfect analogue in the English dialect of Liverpool, where we find e.g. [kx] in can’t, back (Hughes and Trudgill 1987: 66), which again remains unexplained in the traditional doctrine. If the English glottal stop was inherited from Proto-Germanic, we may look for traces in the form of unexpected gemination in medieval sources, to be compared with the gemination in Swedish vecka, droppe, skepp and Old High German offan, wazzar, zeihhan. Such unexplained gemination is indeed found in the Northumbrian Lindisfarne Gospels and Rushworth glosses, e.g. gætt, scipp, fætt, eatta, brecca (cf. Brunner 1965: 189, further Luick 1964: 400 and 886). Here pp, tt, cc may represent earlier [’p], [’t], [’k]. Hofmann writes about this gemination: “Da die Vokaldehnung im Nordenglischen erst für das 12. Jahrhundert angesetzt wird [...], müßte es sich um eine andersartige, echte Geminierung einzelner Formen handeln, die die distinktive Funktion der Konsonantenlänge wohl noch nicht aufhob und von der Vokaldehnung unabhängig war” (1989: 208). This is a perfect analogue of Swedish vecka, droppe, skepp, which can be explained by the existence of preglottalized stops in the Middle Ages. The hypothesis that the geminates represent a purely graphical phenomenon does not explain the short vowel reflexes in later northern English. It now turns out that the English language offers perhaps the most straightforward evidence for the theory that the unaspirated voiced stops of the Indo-European proto-language were actually glottalized. It follows that the preglottalization which can now be assumed for Old English fōt ‘foot’, nacod ‘naked’ can be identified with the glottal stop which is attested in Latvian pêʚds ‘footstep’, nuôgs ‘naked’. The antiquity of the English glottal stop is corroborated by glottalization in Danish, Latvian, Lithuanian, Armenian and Sindhi, and 268 English supported by indirect evidence from Indo-Iranian, Greek, Latin and Slavic (cf. K075). This exemplifies once more the importance of re-examining time and again the primary data in linguistic reconstruction. THE ORIGIN OF THE OLD ENGLISH DIALECTS REVISITED Did the Old English dialects first diverge in Britain or on the continent? In an earlier study (K084) I argued that neither view is correct and that the early divergences between West Saxon and Kentish on the one hand and Anglian on the other are the result of a chronological difference between two waves of migration from the same dialectal area in northern Germany. I argued that West Saxon has preserved two structural archaisms, viz. the nom.pl. ending of the ō-stems -a and the reflex ǃ of PIE *ē, whereas Anglian has retained five accidental irregularities which are also found in Old Norse, Gothic or Old High German. Besides, Anglian differs from West Saxon as a result of seven innovations shared with continental West Germanic languages: the substitution of the acc.pl. ending of the ō-stems -e for the nom.pl. ending, the creation of a distinct accusative of the 1st and 2nd sg. personal pronouns, the creation of the 1st pl. possessive pronoun ūsa, the introduction of e-vocalism in the acc.sg. form of the masc. demonstrative pronoun, the creation of 1st sg. bēom ‘am’, the spread of *waljan to the paradigm of the verb ‘will’, and the raising of ǃ to ē. I therefore distinguished between an earlier, “Saxon” invasion which resulted in the conquest of Kent and Sussex in the fifth century and a later, “Anglian” invasion which can be connected with the subjugation of the north starting around the middle of the sixth century. The shared innovations of Anglian and Old Saxon point to geographical contiguity after the early, “Saxon” migration. Reconsidering the relative chronology of Anglo-Frisian sound changes, Robert Fulk arrives at the following conclusion for the Northumbrian dialect of Old English (1998: 153): 1. Backing and nasalization of West Gmc. a, ā before a nasal consonant. 2. Loss of n before a spirant, resulting in lengthening and nasalization of the preceding vowel. 3. Fronting of West Gmc. a, ā to æ, ǃ, including a in the diphthongs ai and au. 4. Palatalization (but not yet phonemicization of palatals). 5. Retraction of æ, ǃ to a, ā due to the influence of neighbouring consonants. 6. Non-Saxon (and Frisian) ǃ > ē. 7. Restoration of a before a back vowel of the following syllable; at this time æu was retracted to au in Old Frisian. 8. Breaking; in West Saxon, palatal diphthongization follows. 9. i-mutation, followed by syncope; Old Frisian breaking follows. 10. Phonemicization of palatals and assibilation, followed by second fronting in part of West Mercia. 11. Smoothing and back mutation. 270 English In this chronology, English and Frisian begin to diverge at stage 5 and tend to diverge widely at stage 7. The main difficulty with Fulk’s chronology is the unmotivated character of the sound changes: we find backing at stage 1, fronting at stage 3, backing at stage 5, fronting at stage 6, backing at stage 7, fronting at stage 9, and backing at stage 11. What was the driving force behind these alternating developments? Following Krupatkin’s observation that “every time the initial shifts in the field of the long vowels raised similar transformations in the field of the short vowels” (1970: 63), we may look for structural pressure as a determinant factor. In my view, the basic element is the Proto-Germanic asymmetry in the low vowels between long front ǃ and short back a, which could be resolved either by fronting a to æ, as in Anglo-Frisian, or by backing ǃ to ā, as in the other languages (except Gothic, where ǃ was raised to ē at an early stage). If ǃ had been retracted to ā in West Germanic already, the Anglo-Frisian fronting would be entirely unmotivated. Moreover, Caesar refers to the Swabians as Suēbi, not **Suābi, which shows that we must reconstruct a front vowel for an early stage of Old High German. I therefore think that West Saxon ǃ is an archaism and that the early retraction of ǃ to ā did not reach Anglo-Frisian. Hans Nielsen lists three reasons for the assumption that ǃ was first retracted to ā and then fronted to ǃ in Anglo-Frisian (1981b: 52f.). First of all, “the development of Gmc. *-ēn, -ēm > OE/OFris. -ōn, -ōm could hardly have taken place except by way of *-ān, -ām”. Secondly, “the borrowing of Latin strāta as strāzza in OHG and strāta in OS and as strēt(e) in Angl./Kt./OFris. and strǃt in WS suggests that the forbears of OE/OFris. had an open vowel, which was subsequently fronted”. And thirdly, “the expansion of ē1 to ā was a direct consequence of the appearance of ē2 in the long/tense subsystem of late Gmc. (NG/WG)”. I think that none of these three arguments holds water. First of all, it must be noted that the retraction of Proto-Germanic *-ēn to *-ān is not only shared by Old Saxon and Old High German but also matched by a West Germanic delabialization of *-ōn and *-ōns to *-ān, *-āns (cf. K109: 103). This centralization before a tautosyllabic nasal is typologically similar to the development of nasal vowels in French, e.g. main, plein, bien, fin, un, brun, all with a nasalized central vowel in the modern language. It follows that no conclusions can be based on this new *ā < *ǃ, *ō before nasals, which evidently was an early West Germanic development. Secondly, the borrowing of Latin strāta as West Germanic *strǃta only shows that there was no *ā in the receiving language at the time and that *ǃ was closer than *ō, which is unremarkable. Note that the final -a was identified with the delabialized acc.sg. ending *-ōn. And thirdly, the rise of new ē was probably recent because it represents earlier *ea in Scandinavian and is preserved as ea in early Old High German (cf. K124). In fact, the diphthongization of ō to uo in Old High German is best explained by the hypothesis that ea > ia > ie was never The origin of the Old English dialects revisited 271 monophthongized in the southern dialects of West Germanic. The spelling ea is typical of Alemannic, as is the spelling ua for ō, as opposed to ie, uo in Franconian and ie beside ea and oa in Bavarian (cf. Rauch 1967: 37f., 25, 90). Now we turn to the Anglo-Frisian palatalization. Fulk distinguishes between an early palatalization (stage 4) and a later phonemicization of palatals and assibilation (stage 10). This is an unfortunate split, not only because the late phonemicization of palatals effectively obliterates the explanatory value of the early palatalization, but also because it implies that the fronting of velar consonants was reversed by the restoration of a following æ to a. When we look at other languages (Celtic, Slavic, Indic), we usually perceive a rising tide of palatalization, which first affects certain positions and then spreads to other environments (e.g. Greene 1974, K035). The similarities and differences between the conditions for palatalization in English and Frisian rather suggest that we have to distinguish between an early Anglo-Frisian development and a later Old English innovation. Such a chronological split is strongly criticized by Hogg, who claims that “the various types of palatalization are prime candidates for simultaneous application” (1979: 108). On the contrary, it yields a much more natural chain of events than the alternating developments of fronting and backing listed above. Thus, I would start from a vowel system with long front *ǃ and short back *a, a general tendency to retract ǃ to ā, and a local tendency to front a to æ. If we want to avoid the assumption that fronted æ was again retracted to a, it follows that the Anglo-Frisian fronting of the short vowel was blocked by a following l, r, h plus consonant and in open syllables by a back vowel in the following syllable. (Dirk Boutkan has pointed out to me that the same view was already put forward by Heuser 1903: 1.) Since we do not find palatalization before *ai and *au in Frisian, it is natural to assume that *ai had been monophthongized to ā before the Anglo-Frisian fronting of a to æ and that *au had remained unchanged. The Anglo-Frisian palatalization then affected k and g before front vowels. After the “Saxon” migration to Britain, the fronting of a to æ affected the remaining instances of a in closed syllables, and also *au with a before tautosyllabic u, in the dialect of the settlers. This “Saxon” second fronting was followed by breaking and second palatalization, e.g. in eald, cēapian, OFr. ald ‘old’, kāpia ‘buy’. In fact, the first stage of breaking can be identified with the “Saxon” fronting because the conditions were largely identical: it appears that the process of breaking began as incomplete fronting of a before tautosyllabic l, r, h and u and subsequently affected e and i. After the “Anglian” migration, these developments spread to the north, leaving traces only of the earlier situation. In the meantime, Anglian shared the development of Frisian on the continent, in particular the raising of ǃ to ē, which had been preceded by the Anglo-Frisian retraction of ǃ to ā before w (cf. Fulk 1998: 141). The Kentish raising of ǃ to ē was probably a local development, perhaps under the influence 272 English of a second invasion in Kent in the sixth century. After the “Anglian” migration, Frisian fronted ā (from *ai) to ǃ unless it was followed by a back vowel in the following syllable and monophthongized *au to ā. The distinction between ē < *ǃ and ǃ < *ai is still preserved in modern dialects (cf. Campbell 1939: 1011). The Anglo-Frisian and second English palatalizations preceded umlaut (i-mutation) because the umlauted vowels did not palatalize k and g but phonemicized the opposition between palatals and velars, so that Old Frisian shows palatalization before e < *æ and ē < *ǃ but not before e < *ä or ǃ < *ai, e.g. tsetel < *katilaz ‘kettle’ and tziake < *kǃkōn ‘jaw’ versus kenna < *kannjan ‘make known’ and kēi, kāi < *kaijō ‘key’ (cf. already van Haeringen 1920: 31f.). The main difference between the conventional wisdom that a was fronted to æ and then retracted to a before a back vowel in the following syllable and my view that these developments never took place concerns the interpretation of the form slēan ‘strike’, which serves as the hackneyed example to demonstrate fronting and breaking in *slahan (e.g. Hogg 1979: 92, Fulk 1998: 150). It seems to me that insufficient attention has been paid to the paradigm of this word. If slēan were the phonetic reflex of *slahan as a result of fronting and breaking, it would be quite impossible to account for Northumbrian ēa < *ahō ‘water’, where restoration of a before the back vowel in the following syllable should have prevented breaking. In fact, Mercian ēo- and North. (Bede) -ēu and the preservation of the contrast between the reflexes ēo < *ehō and ēa < *eha in the Vespasian Psalter (cf. Campbell 1959: 103) show that we are not dealing with breaking but with contraction here. It appears that the loss of intervocalic *-hbefore rounded vowels was sufficiently early for the resulting diphthong *au to undergo the “Saxon” fronting to *æu. The verb slēan relates to faran as sēon < *sehan to beran. It has long been recognized that strong verbs of the sixth class have a strong tendency to restore the root vowel a in West Saxon (e.g. Campbell 1959: 62). This is already an indication that the vocalism of slēan cannot simply be attributed to generalization of breaking. We must rather assume that a was restored in the imperative far and the subjunctive (optative) fare because this limited the front vowel to the 2nd and 3rd sg. forms which had an umlauted vowel in other verbs and ask why the same development did not take place in the paradigm of *slahan. The parallelism between slēan and sēon suggests that their vocalism must be attributed to the early loss of *-h- before a rounded vowel and contraction in the 1st sg. and 3rd pl. forms *seu, *slæu, *seoþ, *slæoþ, which eventually developed into sēo, slēa, sēoþ, slēaþ. When breaking yielded 2nd sg. *seohist, *slæohist, 3rd sg. *seohiþ, *slæohiþ, imp. seoh, *slæoh, subj. (opt.) *seohe, *slæohe, the stage was set for generalization of the broken vowel in the infinitives *seohan, *slæohan. The original distribution of front æ and back a in the root is actually preserved in Old Frisian, where we find 3rd sg. present ind. sleith < The origin of the Old English dialects revisited 273 *slæhiþ, subj. sle < *slæhe, past participle slein, infinitive sla < *slahan, gerund slande (cf. Boutkan 1996: 147). Thus, I regard the “Saxon” dialect of English as a variety of Ingvaeonic which generalized Anglo-Frisian fronting and palatalization and developed early breaking. In a similar vein, we may regard Mercian second fronting as a generalization of “Saxon” fronting after umlaut (cf. Fulk 1998: 149) under the influence of the “Anglian” raising of ǃ to ē, and the same holds for Kentish raising of æ to e after umlaut. While Old English breaking supplied short counterparts to the u-diphthongs, the Old Frisian breaking of e yielded a short diphthong *eu which was raised to iu when original *eu and *iu developed into iā and iū, respectively, e.g. siucht < *seuxþ < *sexþ < *sexiþ ‘sees’ (with restored root vowel, cf. Boutkan 1998a: 82). The restoration of the root vowel -e- in the strong verbs was probably an Anglo-Frisian development which was obliterated by “Saxon” umlaut (cf. Campbell 1959: 76, 300f., Fulk 1998: 149). All these developments seem to corroborate Krupatkin’s view quoted above that changes in the short vowel system were adaptive to changes in the long vowel system. ANGLO-FRISIAN Patrick Stiles has argued against the reconstruction of an Anglo-Frisian protolanguage “because it is not possible to construct the exclusive common relative chronology that is necessary in order to be able to establish a node on a family tree” (1995: 211). His starting-point are the following shared features of Old English and Old Frisian (1995: 177f.): 1. Front reflex of *ē1 in OE dǃd, dēd, OF dēd(e) ‘deed’, OHG tāt. 2. Fronting of *a in OE dæg, OF dei ‘day’, OHG tag. 3. Rounding of *ē1 before a nasal in OE, OF mōna ‘moon’, OHG māno. 4. Rounding of *a before a nasal in OE, OF lond ‘land’, OHG lant. 5. Loss of nasal with compensatory lengthening before a homorganic spirant in OE, OF tōþ, tōth ‘tooth’, gōs ‘goose’, fīf ‘five’, ūs ‘us’, OHG zan(d), gans, fimf, uns. 6. Rounded reflex of * in OE, OF brōhte ‘brought’, OHG brāhta. 7. Breaking in OE reoht, OF riucht ‘right’, OHG reht. 8. Reflexes of unaccented vowels in OE, OF mōna ‘moon’, OHG māno and OE, OF sunne ‘sun’, OHG sunna. 9. Palatalization of velar consonants in E cheese, F ts(j)iis, G Käse, also OE dæg, OF dei ‘day’, OHG tag. 10. Uniform plural ending in verbal paradigms, e.g. OE berað, OF berath, OHG berumēs, birit, berant ‘we, you, they carry’. It seems to me that the last point, which is fully shared by Old Saxon, is quite irrelevant to the relative chronology to be established and must be kept apart. It is well-known that breaking and palatalization took place under different conditions in English and Frisian. I have argued that the West Saxon front reflex ǃ of *ē1 is an archaism, not an innovation (K084, K180, K225). Reconsidering the relative chronology of Anglo-Frisian sound changes, Robert Fulk arrived at the following conclusion for the Northumbrian dialect of Old English (1998: 153): 1. Backing and nasalization of West Gmc. a, ā before a nasal consonant. 2. Loss of n before a spirant, resulting in lengthening and nasalization of the preceding vowel. 3. Fronting of West Gmc. a, ā to æ, ǃ, including a in the diphthongs ai and au. 4. Palatalization (but not yet phonemicization of palatals). 5. Retraction of æ, ǃ to a, ā due to the influence of neighbouring consonants. 6. Non-Saxon (and Frisian) ǃ > ē. 7. Restoration of a before a back vowel of the following syllable; at this time æu was retracted to au in Old Frisian. 276 English 8. Breaking; in West Saxon, palatal diphthongization follows. 9. i-mutation, followed by syncope; Old Frisian breaking follows. 10. Phonemicization of palatals and assibilation, followed by second fronting in part of West Mercia. 11. Smoothing and back mutation. In this chronology, English and Frisian begin to diverge at stage 5 and tend to diverge widely at stage 7. The main difficulty with Fulk’s chronology is the unmotivated character of the sound changes: we find backing at stage 1, fronting at stage 3, backing at stage 5, fronting at stage 6, backing at stage 7, fronting at stage 9, and backing at stage 11. What was the driving force behind these alternating developments? Following Krupatkin’s observation that “every time the initial shifts in the field of the long vowels raised similar transformations in the field of the short vowels” (1970: 63), we may look for structural pressure as a determinant factor. In my view, the basic element is the Proto-Germanic asymmetry in the low vowels between long front ǃ and short back a, which could be resolved either by fronting a to æ, as in Anglo-Frisian, or by backing ǃ to ā, as in the other languages (except Gothic, where ǃ was raised to ē at an early stage). If ǃ had been retracted to ā in West Germanic already, the Anglo-Frisian fronting would be entirely unmotivated. Moreover, Caesar refers to the Swabians as Suēbi, not **Suābi, which shows that we must reconstruct a front vowel for an early stage of Old High German. I therefore think that West Saxon ǃ is an archaism and that the early retraction of ǃ to ā did not reach Anglo-Frisian. Fulk distinguishes between an early palatalization (stage 4) and a later phonemicization of palatals and assibilation (stage 10). This is an unfortunate split, not only because the late phonemicization of palatals effectively obliterates the explanatory value of the early palatalization, but also because it implies that the fronting of velar consonants was reversed by the restoration of a following æ to a. When we look at other languages (Celtic, Slavic, Indic), we usually perceive a rising tide of palatalization, which first affects certain positions and then spreads to other environments (e.g. Greene 1974, K035). The similarities and differences between the conditions for palatalization in English and Frisian rather suggest that we have to distinguish between an early Anglo-Frisian development and a later Old English innovation. Such a chronological split is strongly criticized by Hogg, who claims that “the various types of palatalization are prime candidates for simultaneous application” (1979: 108). On the contrary, it yields a much more natural chain of events than the alternating developments of fronting and backing listed above. Thus, I would start from a vowel system with long front *ǃ and short back *a, a general tendency to retract ǃ to ā, and a local tendency to front a to æ. If we want to avoid the assumption that fronted æ was again retracted to a, it Anglo-Frisian 277 follows that the Anglo-Frisian fronting of the short vowel was blocked by a following l, r, h plus consonant and in open syllables by a back vowel in the following syllable (cf. already Heuser 1903: 1). Since we do not find palatalization before *ai and *au in Frisian, it is natural to assume that *ai had been monophthongized to ā before the Anglo-Frisian fronting of a to æ and that *au had remained unchanged. The Anglo-Frisian palatalization then affected k and g before front vowels. After the “Saxon” migration to Britain, the fronting of a to æ affected the remaining instances of a in closed syllables, and also *au with a before tautosyllabic u, in the dialect of the settlers. This “Saxon” second fronting was followed by breaking and second palatalization, e.g. in eald, cēapian, OF ald ‘old’, kāpia ‘buy’. In fact, the first stage of breaking can be identified with the “Saxon” fronting because the conditions were largely identical: it appears that the process of breaking began as incomplete fronting of a before tautosyllabic l, r, h and u and subsequently affected e and i. After the “Anglian” migration, these developments spread to the north, leaving traces only of the earlier situation. In the meantime, Anglian shared the development of Frisian on the continent, in particular the raising of ǃ to ē, which had been preceded by the Anglo-Frisian retraction of ǃ to ā before w (cf. Fulk 1998: 141). The Kentish raising of ǃ to ē was probably a local development, perhaps under the influence of a second invasion in Kent in the sixth century. After the “Anglian” migration, Frisian fronted ā (from *ai) to ǃ unless it was followed by a back vowel in the following syllable and monophthongized *au to ā. The distinction between ē < *ǃ and ǃ < *ai is still preserved in modern dialects (cf. Campbell 1939: 1011). The Anglo-Frisian and second English palatalizations preceded umlaut (i-mutation) because the umlauted vowels did not palatalize k and g but phonemicized the opposition between palatals and velars, so that Old Frisian shows palatalization before e < *æ and ē < *ǃ but not before e < *ä or ǃ < *ai, e.g. tsetel < *katilaz ‘kettle’ and tziake < *kǃkōn ‘jaw’ versus kenna < *kannjan ‘make known’ and kēi, kāi < *kaijō ‘key’ (cf. already van Haeringen 1920: 31f.). Krupatkin’s observation quoted above explains the fact that the “fractured reflexes of i and e have rounded second elements in OE, in early OE -u”, and possibly “the second element of the broken reflex of æ was also -u at first” (H.F. Nielsen 1984: 76). Thus, io, eo, ea were the short counterparts of īo, ēo, ēa from the time when these were still u-diphthongs. Nielsen can now explain the fact that the reflex of *a is not broken in Old Frisian: “Gmc. au was monophthongized to ā in OFris., and consequently there was no systematic pressure in terms of creating another short diphthong in the way that iu (io) was phonemicized as a short counterpart of Gmc. eu/iu” (1984: 77f.). In Scandinavian, “the diphthong arisen by breaking was ia both before a and before u”, and this diphthong “is found on the stones of Sparlösa and Rök; by u-mutation it passes into iNj in WN, as appears from scaldic rhymes and vowel harmony; at the further development into ö in Icelandic, bNjrn and biNjrn go 278 English together. In EN the development into iNj only takes place with lost u, a stage which perhaps is expressed in biaurn in the runic inscriptions; iNj is developed into io, which appears in the biurn of the runic inscriptions and in the biorn of the medieval MSS” (K.M. Nielsen 1961: 40f.). Since *e was the short counterpart of *ē2 in North and West Germanic, we must look into the origin of the latter. As I have pointed out elsewhere (K115, K124, K225), I think that original *ē2 developed from disyllabic *ea by monophthongization, first in the Low German area, probably under Romance influence, then in Anglo-Frisian and Scandinavian, and eventually in the large majority of High German dialects. The primary locus of disyllabic *ea are the seventh class strong preterits, where reduplication was replaced by e-infixation before the root vowel of the present stem in North and West Germanic (cf. Fulk 1987, K115), e.g. *heait ‘called’, *hleaup ‘leaped’, *feall ‘fell’, *heald ‘held’, *speann ‘clasped’, *geang ‘went’, *leǃt ‘let’, *hreōp ‘shouted’, *feh ‘seized’. The monophthongization yielded ēo in OE hlēop, fēoll, hēold, spēonn, gēong, hrēop, but ē in hēt, lēt, fēng, Northumbrian fēll and ēa, ǃ from *eā in oncnēaw ‘knew’, gesēaw ‘sowed’, oncnǃw, sǃwe (Campbell 1959: 1191) for WS cnēow, sēow. If we take Krupatkin’s observation seriously, we can now date the “Saxon” fronting and breaking after the migration to Britain in the 5th century but before the monophthongization of *ea to *ē2 and assume that the latter development reached the other Anglo-Frisian dialects from the Low German area before the “Anglian” migration in the 6th century. It may be useful to distinguish between a-breaking, which created a short counterpart to earlier *ea, and u-breaking, which gave rise to a short counterpart of original *eu. The first is found in West Saxon and Scandinavian while the second affected Anglian and Frisian. Breaking and palatalization were preceded by rounding of *ǃ, *a, * in OE, OF mōna ‘moon’, lond ‘land’, brōhte ‘brought’, tōþ, tōth ‘tooth’, gōs ‘goose’, and by loss of a nasal with compensatory lengthening before a fricative in the latter words, cf. OHG māno, lant, brāhta, zan(d), gans. These had already been preceded by the West Germanic centralization and unrounding of *ō in the endings *-ōn and *-ōns (cf. K084: 437 and K219: 5). It appears that the centralization before nasals first affected unaccented vowels in final syllables in West Germanic and then spread to stressed vowels in Anglo-Frisian, yielding a central vowel *ǩ, *ǩɷ, where phonetically *ǩ may have been [Ȝ] or [Ȝɶ]. When nasalization was lost in the endings *-ōn, *-ōns in West Germanic, these had been delabialized and became OHG -a, -ā, OS -a, subject to fronting in OE -e. In stressed syllables, *ǩ, *ǩɷ evidently yielded å, ō in Anglo-Frisian, where å denotes the vowel which is written as a or o before a nasal, e.g. in OE camb, comb, lamb, lomb, land, lond, lang, long, mann, monn, nama, noma, OF lang, long, man, mon, nama, noma. West Germanic stressed *e, *o before nasals are reflected as i, u, e.g. OE niman ‘take’, guma ‘man’, OF nima. These early Anglo-Frisian developments also affected some of the West Germanic dialects which were to Anglo-Frisian 279 develop into Old Saxon, which is not a homogeneous linguistic entity (cf. Klein 1990). On the basis of these considerations I now arrive at the following relative chronology: WG 0. Centralization of unaccented vowels before nasals in final syllables and subsequent loss of the nasalization in these endings. AF 1. Centralization of stressed vowels before nasals. AF 2. Loss of nasalization before f, þ, s, h. AF 3. Retraction of *ǩɷ to ō. AF 4. Monophthongization of *ai to ā. AF 5. Retraction of *ǃ to ā before w and before g plus back vowel. AF 6. Fronting of *a to æ, which was blocked by a following l, r, h plus consonant and in open syllables by a back vowel in the following syllable. AF 7. Palatalization of *k and *g before front vowels, also *g after front vowels. AF 8. Analogical restoration of the back vowel in 2sg. *farist, 3sg. *fariþ ‘go’, 2sg. *slahist, 3sg. *slahiþ ‘slay’, also OE sc(e)acan ‘shake’, sc(e)afan ‘shave’ (cf. Campbell 1959: 315f., K180: 49f.). WS 9. Breaking of *ǃ to *ǃa and of *ī to *īǩ before *h and raising of *ea to *ēǩ, which was simplified to *ē in *hēit, *lēæt, *fēah, *hlēup, *hrēop, *sēau, later hēt, lēt, fēng, hlēop, hrēop, sēow, but not in *fēǩll, *hēǩld, *spēǩnn, *gēǩng, where *ēǩ later (at stage E 13) developed into -ēo-. WS 10. Development of the West Germanic diphthongs *eu, *iu into *ēu, *īu, later ēo, īo. WS 11. Fronting and breaking of *a to short *æa before anteconsonantal l, r, h, u and simplification of *æau to *ǃu. WS 12. Breaking of *e to *eǩ and of *i to *iǩ before anteconsonantal l, r, h. AF 9. Monophthongization of *ea to ē, e.g. North. fēll, fēng, also *eā to ēa in cnēaw, sēaw. AF 10. Raising of *ǃ to ē (this development did not reach Insular North Frisian, cf. Hofmann 1964). A 11. Fronting of *a to æ before anteconsonantal r, h, u with breaking to *eǩ before r plus non-velar consonant and breaking of *e to *eǩ and of *i to *iǩ before r plus non-velar consonant, e.g. North. eorm, WS earm ‘arm’ (cf. Campbell 1959: 117). A 12. Development of the diphthongs *æu, *eu, *iu into *ǃu, *ēu, *īu and monophthongization to ǃ, ē, ī before velar consonants. K 11. Raising of *æ to e and of *eu to *iu and breaking of *e and *i to *iǩ before r plus consonant. K 12. Adjustment to developments WS 9-12. 280 English E 13. Merger of *ǃa, *ēǩ, *īǩ with *ǃu, *ēu, *īu into ēa, ēo, īo and development of *æa, *eǩ, *iǩ into ea, eo, io. E 14. Palatalization of *k and *g before front vowels. E 15. Palatal diphthongization. E 16. Umlaut (i-mutation). F 11. Fronting of *ā to ǃ, which was blocked by a back vowel in the following syllable. F 12. Monophthongization of *au to ā. F 13. Umlaut (i-mutation). F 14. Breaking of *e yielding iu (cf. Boutkan 1998a). In the chronology proposed here, West Saxon separated from Anglo-Frisian as a result of the “Saxon” migration to Britain in the 5th century and the break-up of Anglo-Frisian resulted from the “Anglian” migration in the 6th century. The stages AF 1-10 can appropriately be called Anglo-Frisian. Kentish sides with Anglian in this chronology. POSTSCRIPT It should be clear from the above that my chronology is an improvement on Fulk’s, just as Fulk’s is an improvement on Stiles’s. In order to elucidate the differences, it may be useful to compare Stiles’s chronology (PS 1-9) with Fulk’s (RF 1-11) and mine (WG 0, AF 1-10, WS 9-12, A 11-12, K 11-12, E 13-16, F 11-14), as will be shown here. Note that every development gave rise to at least one isogloss (between groups of speakers who were and who were not affected) and that the chronology only holds for the dialects which have actually come down to us in the historical record. PS 1-2 = RF 3 = AF 6 + WS 11 + A 11. In my chronology, ǃ was never retracted to ā in West Germanic but provided the asymmetry in the low vowels between long front ǃ and short back a which could be resolved either by fronting a to æ, as in Anglo-Frisian, or by backing ǃ to ā, as in the other languages. If ǃ had been retracted to ā in West Germanic already, the Anglo-Frisian fronting would be entirely unmotivated. I think that West Saxon ǃ is an archaism and that the early retraction of ǃ to ā did not reach Anglo-Frisian. If we want to avoid the assumption that fronted æ was again retracted to a, it follows that the AngloFrisian fronting of the short vowel was blocked by a following l, r, h plus consonant and in open syllables by a back vowel in the following syllable (AF 6). After the “Saxon” migration to Britain, the fronting of a to æ affected the remaining instances of a in closed syllables, and also *au with a before tautosyllabic u, in the dialect of the settlers (WS 11). This “Saxon” second fronting was followed by breaking and second palatalization, e.g. in eald, cēapian, OF ald ‘old’, kāpia ‘buy’. After the “Anglian” migration, these Anglo-Frisian 281 developments spread to the north, leaving traces only of the earlier situation (A 11). PS 3-6 = RF 1-2 = AF 1-3. Breaking and palatalization were preceded by rounding of *ǃ, *a, * in OE, OF mōna ‘moon’, lond ‘land’, brōhte ‘brought’, tōþ, tōth ‘tooth’, gōs ‘goose’, and by loss of a nasal with compensatory lengthening before a fricative in the latter words. These had already been preceded by the West Germanic centralization and unrounding of *ō in the endings *-ōn and *-ōns (WG 0). It appears that the centralization before nasals first affected unaccented vowels in final syllables in West Germanic and then spread to stressed vowels in Anglo-Frisian, yielding a central vowel *ǩ, *ǩɷ, where phonetically *ǩ may have been [Ȝ] or [Ȝɶ]. When nasalization was lost in the endings *-ōn, *-ōns in West Germanic, these had been delabialized and became OHG -a, -ā, OS -a, subject to fronting in OE -e. These early Anglo-Frisian developments also affected some of the West Germanic dialects which were to develop into Old Saxon, which is not a homogeneous linguistic entity. PS 7 = RF 8 = WS 9-12 + A 11-12 + K 11-12 + F 14. Following Krupatkin’s observation that “every time the initial shifts in the field of the long vowels raised similar transformations in the field of the short vowels” (1970: 63), we may look for structural pressure as a determinant factor. As I have pointed out elsewhere, I think that original *ē2 developed from disyllabic *ea by monophthongization. If we take Krupatkin’s observation seriously, we can now date the “Saxon” fronting and breaking after the migration to Britain in the 5th century but before the monophthongization of *ea to *ē2 and assume that the latter development reached the other Anglo-Frisian dialects from the Low German area before the “Anglian” migration in the 6th century. PS 8 = WG 0. I find it difficult to separate this development from the same in the other Germanic languages (cf. also K219). PS 9 = RF 4/10 = AF 7 + E 14. The similarities and differences between the conditions for palatalization in English and Frisian suggest that we have to distinguish between an early Anglo-Frisian development and a later Old English innovation. Since we do not find palatalization before *ai and *au in Frisian, it is natural to assume that *ai had been monophthongized to ā before the AngloFrisian fronting of a to æ and that *au had remained unchanged. The AngloFrisian palatalization then affected k and g before front vowels. After the “Saxon” migration to Britain, the fronting of a to æ affected the remaining instances of a in closed syllables, and also *au with a before tautosyllabic u, in the dialect of the settlers. This “Saxon” second fronting was followed by breaking and second palatalization, e.g. in eald, cēapian, OF ald ‘old’, kāpia ‘buy’. 282 English RF 6 = AF 10. Anglian shared the development of Frisian on the continent, in particular the raising of ǃ to ē, which had been preceded by the Anglo-Frisian retraction of ǃ to ā before w. After the “Anglian” migration, Frisian fronted ā (from *ai) to ǃ unless it was followed by a back vowel in the following syllable and monophthongized *au to ā. The distinction between ē < *ǃ and ǃ < *ai is still preserved in modern dialects. RF 9 = E 16 + F 13. After the analogical restoration of back vowels (AF 8) had introduced palatal consonants before back vowels and the second fronting (WS 11, A 11) had introduced new velar consonants before front vowels which were again palatalized (E 14), the umlaut (i-mutation) once again introduced new velar consonants before front vowels. H.F. Nielsen subscribes to Stiles’s chronology of the rounding before a nasal (PS 3-6 = RF 1-2 = AF 1-3) but not to his early fronting (PS 1-2 = RF 3 = AF 6) and proposes the following chronology for Old Frisian (2001: 521): HN 1. HN 2. HN 3. HN 4. HN 5. Monophthongization of *ai to ǃ (AF 4 + F 11) and of *au to ā (F 12). Fronting of long *ā (< *ē1) to ǃ and of short *a to æ (AF 6). Palatalization of *k and *g (AF 7). Umlaut = i-mutation (F 13). Breaking (F 14). The problem with this chronology is threefold. First, short *æ < *a and long *ǃ < *ē1 must be older than *ǃ < *ai because the former palatalize *k and *g and the latter does not, e.g. tsetel < *katilaz ‘kettle’ and tziake < *kǃkōn ‘jaw’ versus kēi ‘key’ (cf. van Haeringen 1920: 31f. and Campbell 1939: 107). It follows that the fronting of *ai (HN 1) must be more recent than the fronting of *ē1 and *a (HN 2) and the palatalization of *k and *g (HN 3). Second, the fronting of long *ā (< *ē1) to ǃ (HN 2) must be older than the rise of *ā < *au (HN 1) because the latter was not fronted until the i-mutation (HN 4). Third, in the modern dialects which preserve the distinction between ē < *ǃ < *ē1 and ǃ < *ai, the reflex of *ē1 merges with that of umlauted *ō and *ū and must therefore have been higher than the reflex of *ai, which merges with that of umlauted *au (cf. Campbell 1939: 1011). This in fact eliminates Nielsen’s chronology as a possibility. Indeed, it is precisely the front articulation of *ē1 versus the back articulation of *ai which characteristically distinguishes Anglo-Frisian from German (including Old Saxon), the latter displaying back articulation of *ē1 versus front articulation of *ai. Since the agreement between English and Frisian comprises not only the centralization of stressed vowels before nasals with loss of nasalization before fricatives and rounding but also the monophthongization of *ai to ā, the limited retraction of *ǃ to ā, the fronting of non-breaking *a to æ which was blocked by a back vowel in the following syllable, the palatalization of *k and *g by nonbroken front vowels and the analogical restoration of the back vowel in 2sg. Anglo-Frisian 283 *farist, 3sg. *fariþ ‘go’, 2sg. *slahist, 3sg. *slahiþ ‘slay’, there can be little doubt about a period of common development not shared by the German dialects. There is no reason to postulate a whole series of back and forth movements in order to unify different varieties of fronting, breaking, raising, palatalization and monophthongization in the separate languages into sweeping developments which affected all possible instances at a stroke. It is much more reasonable to assume a gradually widening scope of developments that affected progressively more instances under various conditions. According to the explanation put forward here, we must distinguish between an earlier, “Saxon”, and a later, “Anglian” migration. One may wonder if there is any historical evidence for this view. H.F. Nielsen states that the Saxons lived in present-day Holstein according to Ptolemy (2nd century) and “appear to have been in control of the whole region between the Elbe and the Weser from the middle of the third century” (1981a: 265). They reached the Netherlands in the fourth century. The Angles can hardly be separated from the present-day district of Angeln in eastern Schleswig. I would suggest that “Anglian” refers to the original Saxons of Angeln, more or less as the French word allemand refers to the original Germans of Alemannia. As Nielsen points out (1981a: 271), Bede does not always observe the distinction between Angles and Saxons, and the eventual preference for the term “Anglian” is probably due to its distinctiveness from the continental Saxons. The traditional designation for the Germanic invaders in Celtic sources is “Saxons”. This name was evidently established at the first stage of the invasion, which can be identified with the period from the time of Vortigern (around 450) until the battle of Mount Badon (about 500, cf. Jackson 1953: 199). There followed almost half a century of peace, the “Saxons” having settled in Kent and Sussex. The territory of Essex and Middlesex was largely uninhabited at that time. In the north, “the great gateway by which the Angles penetrated into the north Midlands and Yorkshire was the estuary of the Humber” (Jackson 1953: 207). Though in the Yorkshire Wolds and at York itself “archaeological finds seem to indicate a more or less unbroken continuity of occupation between the late Roman and pagan Saxon periods” (Jackson 1953: 212), there is no historical evidence for a kingdom of Deira before the second half of the sixth century. During the latter period Deira must have gained considerable strength in view of the spectacular expansion after the battle of Catterick about 600 which is described in the Gododdin. It seems that the battle of Catterick can be viewed as the northern equivalent of the battle of Mount Badon, except for the fact that it was won by the other side. Thus, I suggest that the “Saxon” invasion yielded the conquest of Kent and Sussex in the fifth century, whereas the “Anglian” invasion can be connected with the subjugation of the north which started around the middle of the sixth century (cf. also Alcock 1971: 107-121). There is no linguistic evidence for a different continental homeland, especially because the shared innovations of Anglian and Old Saxon 284 English point to geographical contiguity after the early migrations. Chronologically, the Kentish invasion can be identified with the “Anglian” invasion. The spread of “Saxon” features to the north can be identified with the West Saxon expansion. It is important to realize that at the time under consideration we have to reckon with small numbers of highly mobile people. Milroy & Milroy have established that linguistic change is slow to the extent that the relevant populations are well established and bound by strong ties, whereas it is rapid to the extent that weak ties exist in populations (1985: 375). They point out that “societies undergoing social processes which entail social and geographical mobility and the dissolution of close-knit networks [...] provide the conditions under which innovations can be rapidly transmitted along considerable social and geographical distances” and that “in situations of mobility or social instability, where the proportion of weak links in a community is consequently high, linguistic change is likely to be rapid” (1985: 370, 380). These observations are fully applicable to the Anglo-Saxon invaders of Britain in the 5th and 6th centuries. THE OLD NORSE i-UMLAUT 1. Timothy Reid has reviewed the problem of ON gestr ‘guest’ versus staðr ‘place’ (1990). Since his explanation is, to my mind, no more satisfactory than previous solutions, I venture to present my own view here. Reid’s treatment relieves me of the duty to discuss the earlier literature in any detail (cf. also Steblin-Kamenskij 1959 and especially Bibire 1975 for an assessment of previous work). 2. As far as I can see, the problem has essentially been solved by Axel Kock (1888, 1892). The main objection which has been raised against his explanation is that it is unclear why *iR should cause umlaut in a preceding short syllable while *i did not do so. This objection has been countered by Seip, who pointed out that when *i was syncopated after a long syllable, it may have been reduced to *ǩ after a short syllable unless it was followed by *R, which was “en palatallyd” and “kunde endog palatalisere foregående vokal” (1919: 88). This eliminates Kock’s umlautless period: “Vi får én omlydsperiode, som begynner med omlyd virket av synkopert vokal og fortsetter med omlyd virket av bevart vokal”. 3. The problem which remains is: why did the earlier umlaut affect long syllables only? In my view, the key to the solution of this problem is “Kock’s failure to distinguish between vowel-length and syllable-length” (Bibire 1975: 201). While *i “was lost after long root-syllables earlier than after short ones (and evidence for this seems to be irrefutable)” (Steblin-Kamenskij 1959: 109), it is reasonable to assume that the umlaut affected long vowels and diphthongs earlier than short vowels. The prominent place which the word gestr, which has a short vowel in a long syllable, has occupied in the discussion from Kock (1888) till Reid (1990) may have prevented scholars from confronting the problem in an adequate way. 4. Thus, I arrive at the following reconstruction of glǿpr ‘crime’, gestr and staðr after the earlier syncope: nom.sg. gen.sg. dat.sg. acc.sg. glǿpR glópaR glǿp glǿp gastR gastaR gast gast staðĭR staðaR staðǩ staðǩ nom.pl. gen.pl. dat.pl. acc.pl. glǿpiR glópa glǿpimR glǿpi gastiR gasta gastimR gasti staðiR staða staðimR staði 286 Scandinavian At this stage, the forms *glópaR, *glópa, *glǿpimR were replaced by glǿps, glǿpa, *glǿpumR on the analogy of the a-stems. When umlaut was phonemicized in short vowels, the umlauted root vowel of the plural forms *gestiR and gesti spread to the singular, which had the regular endings of the long stem paradigm. The analogical ending of gen.sg. gests suffices to show that the word gestr cannot be used as a representative example of an i-stem paradigm. 5. If the theory advanced here is correct, we expect lack of umlaut in i-stems with a short vowel in a long syllable, and this is indeed what we find, e.g. burðr ‘birth’, fundr ‘meeting’, kostr ‘choice’, skurðr ‘cut’, stulðr ‘theft’, sultr ‘hunger’, þróttr < *þruhti- ‘strength’, *þurðr ‘diminution’ (Noreen 1970: 267), also urðr ‘fate’ and the feminines. There is an umlauted root vowel in feldr ‘cloak’ and gestr, possibly in brestr ‘crash’ (u-stem?), dyttr ‘din’, pyttr ‘pit’ (loanword?), skellr ‘clash’, and in a number of plural names (cf. Noreen 1970: 266f.). Thus, it appears that short vowels were not umlauted by syncopated *i but only by *j, *iR, *R, palatalized k/g, and retained i (cf. Bibire 1975: 189 and passim). 6. This raises a problem in connection with the development of the ja-stems. Since short root vowels were regularly umlauted in this category, the *j must have been preserved when short vowels were lost in the endings. I therefore reconstruct *harĭR, *vaðĭ beside the a-stems *harmR, *barn and *dagǩR, *bakǩ for herr ‘army’, veð ‘pledge’, harmr ‘grief ’, barn ‘child’, dagr ‘day’, bak ‘back’ at the stage after the earlier syncope. Whether one considers *ĭ to be a reduced vowel or a syllabic consonant is only a matter of taste. For bekkr ‘bench; brook’ I reconstruct *bakkĭR, acc.sg. *bakkĭ because this type behaves in the same way as the short ja-stems. 7. The difference between gestr and staðr has a parallel in the weak preterits of fella ‘to fell’ and velja ‘to choose’, viz. 1st sg. felda vs. valða. The different suffix can be explained by the assumption that intervocalic *d became a fricative between the earlier and the later syncope (cf. K102: 4). As in the case of gestr, I assume that the umlaut in felda was introduced on the analogy of the long vowel stems, e.g. dǿmda ‘I judged’. There is unmistakable evidence for such analogical influence in the case of selda ‘I sold’ (OSw. salde), where the suffix -d- shows that the form “schon in urnordischer Zeit zweisilbig gewesen ist und zusammenstoßendes ld gehabt hat” (Kock 1894: 452, cf. OE sealde), similarly bygða ‘I lent’ (OE bohte), cf. also keypta ‘I bought’ (inf. kaupa), and the later analogy in lykta ‘I shut’ (beside original lukða), sekta ‘I sentenced’, and setta ‘I set’ (OSw. satte), where the sequence *-tð- may have yielded -tt- immediately after the later syncope, cf. Runic satido ‘I set’ (Rö stone, AD 400), sAte ‘he set’ (Gummarp stone, 7th century). 8. Up to now I have assumed that the phonemicization of umlaut in long vowels can be identified with the earlier syncope. There is no reason why this The Old Norse i-umlaut 287 should be so. It is actually much more probable that unstressed short vowels were reduced after long and short syllables alike when unstressed long vowels were shortened. This vowel reduction must have preceded the earlier syncope but cannot have preceded the rise of umlaut in long vowels. Thus, we arrive at the following relative chronology: (1) (2) (3) (4) (5) Umlaut of long vowels. Reduction of unstressed vowels. Syncope after long syllables. Umlaut of short vowels. Syncope after short syllables. The gemination of velars before *j can be dated to stage (2) if it is viewed as a compensation for the reduction of the following syllable, e.g. *bakjaR > *bakkjǩR > *bakkĭR > *bekkĭR > bekkr ‘brook’. ON BREAKING As H.F. Nielsen points out, for Old English “it is fairly certain that breaking takes place prior to i-mutation,1 which itself precedes back umlaut.2 [...] On the other hand, OE breaking must be later than OE fronting of a > æ,3 which is most likely to be an independent development” (1984: 75, 80). This chronology suffices to show that the Old English breaking cannot be identified with the Scandinavian breaking. Moreover, the conditions of the two were quite different. Since the Old Frisian breaking “took place only before ht and hs, and not before intervocalic h, such forms as siucht (‘sees’) show that it must have taken place later than i-mutation, for the i of the 3rd pers. sg. pres. indic. was not syncopated till after it had caused mutation” (Campbell 1939: 105). Thus, we find similar, yet quite different developments in the three languages. Nielsen quotes with approval Fourquet’s view that in Old English “les produits de la fracture des voyelles brèves sont venus occuper dans le système des brèves la même place que les diphtongues d’origine ancienne occupaient dans le système des longues” (1959: 151), which is in accordance with Krupatkin’s observation that “every time the initial shifts in the field of the long vowels raised similar transformations in the field of the short vowels” (1970: 63). This explains the fact that the “fractured reflexes of i and e have rounded second elements in OE, in early OE -u”, and possibly “the second element of the broken reflex of æ was also -u at first” (Nielsen 1984: 76). Thus, io, eo, ea were the short counterparts of īo, ēo, ēa from the time when these were still u-diphthongs. Nielsen can now explain the fact that the reflex of *a is not broken in Old Frisian: “Gmc. au was monophthongized to ā in OFris., and consequently there was no systematic pressure in terms of creating another short diphthong in the way that iu (io) was phonemicized as a short counterpart of Gmc. eu/iu” (1984: 77f.). This leads me to reconsider the Scandinavian breaking against the same background. As K.M. Nielsen has convincingly argued, “the diphthong arisen by breaking was ia both before a and before u”in Scandinavian, and this diphthong “is found on the stones of Sparlösa and Rök; by u-mutation it passes into iNj in WN, as appears from scaldic rhymes and vowel harmony; at the further development into ö in Icelandic, bNjrn and biNjrn go together. In EN the development into iNj only takes place with lost u, a stage which perhaps is expressed in biaurn in the runic inscriptions; iNj is developed into io, which appears in the biurn of the runic inscriptions and in the biorn of the medieval MSS” (1961: 40f.). The further development of iNj in Old Icelandic þiokkr ‘thick’, miólk ‘milk’, Old Norwegian þiukkr, neuter fiugur ‘four’, Old Swedish fiughur is 290 Scandinavian secondary.4 Since *e was the short counterpart of *ē2 in North and West Germanic, we must look into the origin of the latter. The origin of *ē2 is the subject of an article by the regretted Tocharologist, Baltologist, Germanicist and Indo-Europeanist Jörundur Hilmarsson (1991). With his characteristic care and acute sense of etymology, the author divides the instances of *ē2 into seven groups: (1) *hē2r, (2) *mē2da-, (3) class VII preterits, (4) Latin loanwords, (5) *fē2rō, (6) *kē2na- and *lē2ba-, (7) Continental Germanic residue. For the present purpose, groups (4)-(7) can be regarded as a residue and will be left out of consideration. The same holds for *mē2da-, which does not occur in Scandinavian. For *hē2r we must start from a deictic particle *hi ‘here’ (cf. K063), which was extended by -ar from þar ‘there’, jainar ‘yonder’, aljar ‘elsewhere’.5 The regular lowering of *i to e before a yielded OHG hear (Isidor), later hiar, hier. As I have indicated elsewhere (K115), I think that *ē2 in the class VII preterits of strong verbs represents *ea, which was preserved in OHG geang ‘went’, feang ‘seized’, feal ‘fell’. The model for the development of this formation was provided by the preterits *eauk ‘increased’, *eaus ‘poured’, *eaud ‘granted’, *ear ‘ploughed’, *ealþ, ‘grew old’, *eaik ‘claimed’, and especially *eaj ‘went’. The spread of *ea as a preterit marker yielded *hleaup ‘leaped’, *heald, ‘held’, *heait ‘called’, also *beauw ‘dwelt’, *feāh ‘seized’, *leǃt ‘let’, plural *hleup-, *held-, *heit-, *beuw-, *feng-, *let-, ON hlióp, helt, hét, bió, biogg-, fekk, fing-, lét, lit- (cf. Noreen 1970: 338-340). The identification of *ē2 as *ea now explains the Scandinavian breaking of *e to *ĕă in accordance with the considerations cited above. It is remarkable that there is no evidence for breaking before a front vowel in the following syllable, where the model *ea was lacking, and that breaking is less frequent in light than in heavy syllables. Even more strikingly, breaking was blocked by a preceding *w, e.g. verpa ‘to throw’, huelpr ‘whelp’, while the preterit sueip, pl. suip- ‘swept’ shows absence of *ea after *w, which is a natural restriction because the form contains a triphthong already. The broken vowel *ĕă either developed into ja by “coinciding in its onset with the non-syllabic allophone of /i/” (SteblinKamenskij 1957: 91) or lost its diphthongal character and merged with the reflex of umlauted *a. The latter development may have been conditioned by the monophthongization of *ea to ē, which probably took place under the pressure of the rise of ǃ from umlauted *ā. It is probably no accident that *ea is best preserved in Old High German, where the umlaut of *ā was late. Analogical developments have rendered the original conditions of breaking opaque. The Old Icelandic paradigm of hialpa ‘to help’ closely follows that of falla ‘to fall’, reflecting the conditions of umlaut, not of breaking, e.g. 2nd pl. hialpeþ like falleþ < *-ed, *-aid. The purely phonological development is perhaps most faithfully preserved in the word for ‘si’x, where the cardinal is not broken On breaking 291 while the ordinal is in East Norse, cf. Swedish sex, sjätte, Danish seks, sjette, Latin sex, sextus. NOTES “Cf. forms like OE wierpþ (< *wiorpiþ < *wirpiþ) and nīehst (< *nēahist < *nǃhist), which would have come out *wiorpþ, *weorpþ and *nēahst if the reverse relative chronology had been true”. Differently Collier (1987), who disregards the fact that the Old High German umlaut took place before preserved i but not before lost *i and cannot therefore be identified with the Old English umlaut. 1 “This is shown e.g. by eosol, whose diphthong is due to back mutation of e which again reflects an i-mutated a, cf. eosol (suffix substitution) < *esil < *asiluz”. 2 “Thus OE seah, *neahti (> nieht) and healp presuppose the intermediate stage -æ-, cf. Gmc. *sah, *nahtiz and *halp”. 3 Cf. Benediktsson 1963: 428-31 and 1982: 38-41. On doublets such as biarg, berg ‘rock’ and fiall, fell ‘mountain’, cf. Hoff 1949: 195-202. On Dyvik’s theory (1978), see Benediktsson’s review (1982: 41-55). 4 This was already suggested by Mahlow (1879: 163) and more recently by Meid (1971: 94). Ringe’s objection that we should expect *hir instead of *hiar (1984: 140) is proved invalid by *hwar ‘where’, ON huar, Lith. kurɶ. 5 GLOTTALIZATION, PREASPIRATION AND GEMINATION IN ENGLISH AND SCANDINAVIAN Docherty et alii have “noted that several sociolinguistic accounts have shown a sharp distinction between the social trajectories for glottal replacement as opposed to glottal reinforcement, which have normally been treated by phonologists as aspects of ‘the same thing’. It may therefore not always be appropriate to treat the two phenomena as manifestations of a single process or as points on a single continuum (presumably along which speakers move through time). From the speaker’s point of view (as manifested by different patterns of speaker behavior ) they appear as independent phenomena” (1997: 307). In particular, “while the glottal stop is spreading rapidly in mainstream English, glottal reinforcement (especially of /p/ and /k/ in intervocalic positions) is possibly recessive. It is characteristic not only of Tyneside male speech but also of rather conservative rural varieties, such as those of south-west Scotland and much of Northern Ireland” (o.c.: 306). This supports my view that the “reinforcing” glottal closure of [’p], [’t], [’k] is ancient, in spite of the recent spread of the replacing glottal stop in mainstream English (cf. K142). The two types of glottal variant are clearly distinct in Newcastle English, where they moreover exhibit quite different sociolinguistic patterns. The replacing glottal stop “is variably substituted for non-initial pre-vocalic /t/ (e.g. in set off, water) by younger speakers, especially middle-class females, and as such appears to be a non-local form entering Newcastle English” whereas the preglottalized variants “by contrast, are largely the preserve of older males” (Docherty & Foulkes 1999: 54). It appears that the preglottalized stops “differ from the ‘pure’ glottal variants in terms of the presence of movement of the second formant in the previous vowel: formant transitions are caused by gestures involving the supralaryngeal vocal organs. 79% of our tokens contained F2 transitions, the exceptional cases sounding clearly like glottal stops” (o.c.: 57). It turns out that “older males appear to be producing glottalised tokens with a different articulatory co-ordination than other members of the speech community: they have a greater tendency to time the oral gesture such that it lags behind the accompanying glottal articulation” (o.c.: 61f.). Thus, preglottalization is disappearing from the language while the replacing glottal stop is spreading in the speech of the younger generation. While the “increasing space given by phoneticians from about 1920 onwards to the treatment of the glottal stop” (Andrésen 1968: 34) can be explained by the phonemic character of the glottal replacement, the earlier preglottalization of /p/, /t/, /k/ went unnoticed because it was not distinctive. Glottalization is pervasive in pre-1930 audio recordings of people born in the second half of the 294 Scandinavian 19th century, even in formal delivery (cf. K142, with ref.). It follows that glottalization was well-established in upper-class English speech in the 19th century and must have been widespread in the standard language of that time. The lack of attention to this phenomenon can be explained not only by the subphonemic character of preglottalization but also by its loss in pre-pausal position. While “glottal variants are widespread in various phonological contexts in Newcastle, they are almost categorically prohibited in pre-pausal position. Tokens before a pause are instead – from an auditory perspective – clearly ‘released’ voiceless alveolars” (Docherty & Foulkes 1999: 62). It appears that either the glottalization or the buccal features could be lost in pre-pausal position: “In Derby glottal stops in pre-pausal position are far more widespread, but in the self-conscious context of word-list readings most speakers produce what sound like ‘released’ [t]s, just as in Newcastle” (l.c.). This suggests that prepausal [t] is due to restoration and that the spread of the replacing glottal stop in mainstream English may have started from pre-pausal positions. This brings the original distribution of the English glottalization closer to its Danish counterpart, the so-called vestjysk stød, which is found immediately before the plosives p, t, k “wherever these stand in an original medial position, following a voiced sound in a stressed syllable” (Ringgaard 1960: 195). The vestjysk stød cannot possibly be connected with the Jylland apocope because it is also found in the northeastern part of the vestfynsk dialects, where the apocope did not take place. The vestjysk stød in these isolated dialects suggests that it is a retention rather than an innovation. Moreover, Hansson has drawn attention to the fact that vestjysk stød is found on original monosyllables and polysyllables alike in the most remote and isolated villages on the island of Als, where it coexists with true pitch accents representing the original accents 1 and 2 from which the Common Danish stød opposition developed (2001: 166). We must conclude that outside these archaic dialects the vestjysk stød was lost in monosyllables, as was the case with preglottalization in Newcastle English. The preglottalized stops of English and Danish cannot be separated from the preaspirated stops in the northern Scandinavian languages. The geographical distribution of preaspiration has recently been examined in detail by Hansson (2001: 158-164), who concludes that it is a peripheral archaism to be identified historically with the vestjysk stød. Hansson points out that we find preaspiration proper in Icelandic, Faroese, the Norwegian dialects of Jæren, North Gudbrandsdal, most of Härjedalen, and the island of Senja, the Swedish dialects of northeastern Uppland (Gräsö, Valö, Hållnäs, Forsmark), of the Åland island of Kökar and some nearby dialects, the Estonian Swedish dialect of Ormsö, the Lapland dialects of Vilhelmina and Arjeplog (where it may be attributed to Saami influence), and in a spoken corpus from Central Standard Swedish. Most importantly, preaspiration is found not only in original geminates and in stops before sonorants, as in Icelandic, but also in single /p/, Glottalization, preaspiration and gemination in English and Scandinavian 295 /t/, /k/ between vowels and word-finally after a vowel, especially in Faroese (cf. already K182: 61) and in the dialects of Uppland (Gräsö). We find extensive sonorant devoicing before /p/, /t/, /k/ in the same areas as preaspiration proper while devoicing of /l/ and especially /r/ is more widespread, covering most of the Norwegian and Swedish territory. The strong geographical correlation between preaspiration proper and extensive sonorant devoicing and the much wider distribution of the devoicing of /r/ and /l/ suggest that preaspiration was lost after a vowel in the larger part of Scandinavia. We now turn to the origin of the Scandinavian preaspiration. Hansson assumes that /p/, /t/, /k/ were “phonetically preaspirated in all non-initial positions in Late Proto-Scandinavian” (2001: 167) but takes an agnostic position on the ultimate origins of preaspiration, proposing that /p/, /t/, /k/ “became either preaspirated or preglottalized through a simple sound change”, viz. “a slight misalignment of articulatory gestures”, and assuming that “preglottalization developed out of preaspiration” without adducing any arguments for this assumption (o.c.: 169). He thinks that preaspiration was phonemicized in Icelandic as a result of devoicing of voiced stops (o.c.: 168) but suggests in a footnote that it may have resulted from the shortening of long vowels in closed syllables if /b/, /d/, /g/ were already voiceless at the time. The latter view comes close to my own. In fact, there is no evidence for voiced obstruents in Proto-Scandinavian because the voiceless stops of e.g. Icelandic henda ‘throw’, vagga ‘cradle’, bera ‘carry’ (o.c.: 164) may go all the way back to Proto-Germanic. Indeed, the “hardening” of voiced fricatives to voiceless stops (l.c.) suggests that the fricatives were actually voiceless at that time. Elsewhere I have proposed that intervocalic *-d- became a fricative between the earlier and the later syncope (K102: 4) and that all obstruents were voiceless in North-West Germanic (K138: 54). I think that the rise of phonemic voicedness was a more recent development which took place under Romance and subsequently Low German influence and eventually led to the loss of preaspiration in South and Central Scandinavian. The close correspondence between preglottalization in English and Danish suggests that preaspiration developed out of it in the northern Scandinavian languages. I am inclined to identify the development with the fricativization of postvocalic *d [t] between the earlier and the later syncope, which can be dated to the 7th century. The loss of glottal constriction in the preglottalized stops yielded preaspiration in the same way as the loss of occlusion in postvocalic b [p], d [t], g [k] yielded fricatives. This account of the facts offers parallel explanations for the Scandinavian preaspiration and the High German consonant shift, where the new affricates [pf], [ts], [kx] can be derived from preglottalized stops [’p], [’t], [’k] by loss of the buccal occlusion with concomitant oralization (klusilspring) of the glottal constriction (cf. K138a). Thus, I think that OHG helpfan, English hel’p, Vestjysk hjæl’b and Icelandic 296 Scandinavian hjálhpa all developed from Proto-Germanic *hel’p- and that conservative English dialects have best preserved the original sound structure. The reconstruction of preglottalized stops sheds new light on the quantity shift in Scandinavian because the phonetic difference between preglottalized and geminated stops is slight, the glottal constriction preceding the buccal closure in the former but not in the latter instance. As a result, preglottalization could easily be reanalyzed as gemination, entailing the rise of new closed syllables. This mechanism accounts in a principled way for the existence of several layers of gemination in North and West Germanic (cf. K102: 7). It also explains the fact “that in Härjedalen, only the more innovative dialects (as opposed to that of Vemdal) preaspirate these secondary geminates, and on Kökar, only younger speakers preaspirate them. Both are obvious cases of later generalization” (Hansson 2001: 1723), similar to the generalization of sonorant devoicing before /s/ in Faroese (o.c.: 162f.). While preaspiration was lost after lengthened short vowels before single stops in most of the West and North Norwegian, Faroese and Icelandic dialects, the lengthening was evidently anticipated by the rise of geminates from preglottalized stops in East Norwegian and Central and North Swedish dialects, except after the low vowels a and æ, which were lengthened (cf. Perridon 2002: 73), e.g. Swedish vecka ‘week’, droppe ‘drop’, skepp ‘ship’, but äta ‘eat’, cf. ON vika, dropi, skip, OE wice, dropa, scip. As in the case of preaspiration in Härjedalen and Kökar and sonorant devoicing in Faroese, gemination could be generalized by lengthening /s/ and /m/ after short vowels, and even other resonants in the Norwegian dialect of Bergen (o.c.: 76). Elsewhere I have argued that the preglottalization which I have reconstructed for Proto-Germanic can be identified with the preglottalization which must be reconstructed for other Indo-European languages on the basis of direct evidence from Baltic and Indic and indirect evidence from Indo-Iranian, Greek, Latin and Slavic (see especially K075). Hansson’s view that this hypothesis “hinges on the validity of the Glottalic Theory as such” (2001: 169) is mistaken because my reconstruction of Proto-Germanic preglottalization is based on its actual attestation in English and western Danish and on the derivation of preaspiration in northern Scandinavian, of various layers of gemination in North and West Germanic, and of affrication in High German (and English dialects) from natural developments of the same, irrespective of its origins. Perridon asks rhetorically: “If for instance the geminated fricative in High German essen is the reflex of a preglottalized stop in Pgerm., why then is there in this case no preaspiration in Icelandic, no (vestjysk) stød in the western dialects of Danish, no glottal stop in English, nor gemination in Central and North Swedish?” (2002: 743). The answer in simple: preaspiration was lost after a lengthened short vowel in Icelandic (but not in northern Faroese), preglottalization was usually lost in monosyllables in vestjysk and English (but preserved in polysyllabic forms of such words), and gemination did not arise Glottalization, preaspiration and gemination in English and Scandinavian 297 after lengthened a, æ in Central and North Swedish (as Perridon remarks himself, o.c.: 73). Note that preaspiration is actually attested in Faroese eta [e:hta] (K182: 61) and preglottalization in Old Northumbrian eatta (K142: 177), where the double consonant cannot denote either a preceding short vowel (because the attested form is earlier than the lengthening of short vowels in northern English) or a true geminate (because the short vowel is regularly lengthened at a later stage in these dialects). The hitherto unexplained double consonants in the Lindisfarne Gospels and Rushworth glosses are another phenomenon which is accounted for by the reconstruction of preglottalized stops for Proto-Germanic (l.c.). EARLY RUNIC CONSONANTS AND THE ORIGIN OF THE YOUNGER FUTHARK Elsewhere I have argued that all obstruents were voiceless in North-West Germanic (K102: 9, K138: 54, K192: 8f.). H.F. Nielsen’s comprehensive study of Early Runic (2000) now provides a welcome opportunity to compare my reconstruction with more traditional views of the early Germanic consonant systems and to specify its implications for the interpretation of the Runic evidence. While I generally agree with the common interpretation of the Gothic consonant system (as modified by Roberge 1983), I think that it originated from an early fixation of the stress on the initial syllable which forestalled the devoicing of voiced stops and rhotacism of *z found in the other Germanic languages. Unlike Gothic, North-West Germanic preserved the preglottalized stops which were inherited from the proto-language and later developed into preaspirated stops in northern Scandinavia and into affricates in High German (cf. K102, K138, K192). This leads to the following comparison of my reconstruction (in square brackets) with Nielsen’s interpretation (between slashes, cf. 2000: 122f.) of the Early Runic consonant system: b d g p t k f þ h s R m n l r w j /b/ /d/ /g/ /p/ /t/ /k/ /f/ /þ/ /h/ /s/ /z/ /m/ /n/ /l/ /r/ /w/ /j/ [p] [t] [k] [’p] [’t] [’k] [f] [þ] [h] [s] [R] [m] [n] [l] [r] [w] [j] Nielsen assumes that the voiced stops had fricative allophones in non-initial positions. I find no evidence for this hypothesis, which cannot be separated from the common view that Grimm’s law preceded Verner’s law in ProtoGermanic. Elsewhere I have argued that this view is mistaken (K102: 5f. and 300 Scandinavian K119: 2f.). If Verner’s law preceded Grimm’s law, there is no reason to assume voiced fricatives for any stage of Proto-Germanic except for the allophone *z of /s/. In particular, West Germanic *d and High German *b and *g never had fricative allophones in prehistoric times. Moreover, I think that they remained voiceless lenes up to a comparatively recent stage. In Old Norse, Nielsen’s /z/ merged with /r/, e.g. gestr ‘guest’ versus Early Runic -gastiR. Nielsen assumes that the non-initial allophones of /f/ and /þ/ became voiced and that the alleged voiced fricative allophones of /b/ and /d/ were rephonemicized as allophones of /f/ and /þ/ (2000: 125). This assumption creates several problems. First of all, “a contrast depending on the presence and absence of voice is retained by /k/ ≠ /g/, the latter phoneme having stop as well as fricative allophones” (o.c.: 14563), e.g. voiced stop in ganga ‘to go’, voiced fricative in auga ‘eye’ and acc.sg. dag ‘day’, palatal glide [j] in dat.sg. degi, voiceless fricative [x] in gen.sg. dags and neuter heilagt ‘holy’ (o.c.: 125). It remains unclear why [j] and [x] are identified as allophones of /g/, not of /j/ and /k/, respectively. It also remains unclear why the voiced fricative allophones of /b/ and /d/ should be rephonemicized as allophones of /f/ and /þ/ if this did not hold for the voiced fricative allophone of /g/. Alternatively, one could assume that the voiced allophones of /f/ and /þ/ were rephonemicized as fricative allophones of /b/ and /d/, which would bring them into conformity with the corresponding allophone of /g/. Secondly, it remains unclear why the voiced allophone [v] of /f/, e.g. in grafa ‘to dig’, preterit gróf < *-b-, and in nefi ‘nephew’ < *-p- (ibidem) is not identified with /v/, e.g. in vápn ‘weapon’, instead of /f/. The separation of non-initial [j] and [v] from initial /j/ and /v/ looks like an undesirable artefact of the method. Thirdly, Nielsen notes that “there are no voiced allophones of /s/, not even in medial voiced surroundings, cf. leysa. This is surprising in view of the voiced allophones of the other fricatives in medial and final position” (o.c.: 125). Surprising indeed! It is a strong indication that there were no voiced fricatives at all at the stage under consideration and that all obstruents were voiceless. Nielsen remarks that “even the earliest Icelandic manuscripts had alternative spellings in u or v” for word-medial /f/ (o.c.: 14564), but this reflects a more recent stage. “Originally þ was used in all positions. When ð was borrowed from Norwegian, the two letters occurred in free variation until eventually ð was reserved for non-initial position” (l.c.65). This may offer an indication for the chronology and geographical origin of the rise of voiced fricatives in Scandinavian. Fourthly, the phonological status of long consonants remains to be specified, e.g. leggja ‘to lay’, lykkja ‘loop’, þakka ‘to thank’ (o.c.: 126). What exactly is the difference between leggja and dat.sg. degi in terms of distinctive features? Fifthly, the rhotacism is an unsolved problem. Nielsen’s identification of R as /z/ is hard to reconcile with the fact that apart from the By stone (Norway, 6th Early Runic consonants and the origin of the younger futhark 301 century) it is almost or wholly limited to word-final position (o.c.: 214). In my view, R stands for voiceless r < *z and originated from the general devoicing of obstruents in North-West Germanic as a result of Grimm’s law (cf. K138: 54). Now we turn to the 7th century evidence of the Blekinge and Eggja stones, which appears to be crucial for an interpretation of the Runic consonant system. Nielsen notes that initial /j/ “must have disappeared in the language of the Blekinge inscriptions, seeing that the jāra-rune has come to designate A” (2000: 126) and suggests on the basis of Istaby AfatR ‘after’, with -R for -r, that “by this time the reflex of Gmc. *-z had coalesced with -r in final position in North Germanic when following an alveolar obstruent” (o.c.: 96). He thinks that the obstruent system of Early Runic was otherwise retained, to judge from examples such as Gummarp hAþuwolAfA, Stentoften hAþuwolAfR (personal name), Björketorp uþArAbA ‘harmful’, where f and þ are clearly preserved, as opposed to Gummarp stAbA ‘staves’, Stentoften hAborumR ‘he-goats’, hederA ‘hither’, Björketorp hAidR ‘brightness’, which exhibit allophones of /b/ and /d/. I find it very difficult to assume that *-z was retained in Stentoften -wolAfR (2×), Istaby -wulafR, also Eggja fiskR ‘fish’, as in Stentoften dat.pl. hAborumR ‘he-goats’, hagestumR ‘stallions’, Eggja nom.pl. manR ‘men’, while it became -r in Björketorp -lAusR ‘loose’, perhaps hAidR ‘brightness’, also bArutR ‘breaks’, cf. Stentoften -lAsAR, hideR, bAriutiþ. This looks like paper phonetics. If -R was the voiceless counterpart of -r, all of these examples receive a natural explanation. The final -r of *aftr was devoiced in Istaby AfatR, ON aptr, perhaps similarly in Björketorp hAidR [tR], while the final -r of Björketorp bArutR [’tR], ON brýtr ‘breaks’ developed phonetically from -iþ, Stentoften bAriutiþ, with -iþ from word-final *-id with voiceless *-d [t] < *-ti. Nielsen states that things may be less clear-cut in the case of Stentoften gAf ‘gave’, where he assumes devoiced /b/ in final position, cf. Sjælland bracteate 2 gibu ‘(I) give’ (o.c.: 126). Interestingly, the consistent spelling of Gothic gif, gaf (beside giban, gibis, gebum, gebun) as opposed to grob ‘dug’ and gadob ‘was fitting’ (cf. Roberge 1983: 129) suggests that we have a Verner alternation here. I have therefore proposed to derive the verb ‘to give’ from ga- plus *ep-, cf. Hittite epzi ‘seizes’, Latin apīscor ‘reach’, coēpī ‘have begun’ (K120: 104f.), like OHG gezzan beside ezzan ‘to eat’ and MHG gan, gunnen ‘grant’ beside OHG an, unnun. The contrast between /þ/ and /d/ [t] in medial position was clearly retained in Eggja moþA ‘tired’ versus mAde ‘rubbed off ’ (Nielsen 2000: 127), so that the spelling of nAkdan ‘naked’ < *nakudan represents the expected reflex of [’kt], not [kð] (as assumed by Nielsen, l.c.), similarly in final position niþ ‘waning of the moon’ beside ob [op] < *uba, ON of ‘over’. The use of k and t instead of g and d in Eggja fokl ‘bird’ and lat ‘land’ suggests that preglottalization was lost in western Norway around 700 because it developed into preaspiration at that time. This development must evidently be connected with the rise of the younger futhark. 302 Scandinavian Elsewhere I have argued that the simplest way to account for the difference between the allomorphs of the weak preterit suffix in Old Norse, e.g. in deilda ‘divided’ < *dailidō and valða ‘chose’ < *walidō, is the hypothesis that intervocalic *-d- became a fricative between the earlier and the later syncope (K102: 4). If all obstruents were voiceless at that time, it follows that intervocalically b [p], d [t], g [k] became [f], [þ], [x] and merged with the earlier fricatives f, þ, but not h (which disappeared in non-initial positions). In northern Scandinavia, the loss of glottal constriction in the preglottalized stops p, t, k yielded preaspiration in the same way as the loss of occlusion in the noninitial lenes stops yielded fricatives. The preaspiration was often realized as devoicing of a preceding resonant and was thereby dissociated from the following stop. Note that there were also devoiced resonants without a following stop, e.g. in Stentoften and Björketorp welA ‘deceitful’ < *wihla-, ON véla ‘betray’, Finnish (loanword) vihlata ‘delude’, also ON mǃla ‘to speak’ < *-þl-, rǃna ‘to rob’ < *-hn-, where the weak preterit in -ta instead of -da points to a voiceless resonant, and in my view of course -R for voiceless -r, cf. also rh- for *hr- in Helnæs rhuulfR and Vatn (Norway, 8th century) rhoAltR < *hrōþuwaldaz (Nielsen 2000: 257ff.). On the other hand, -h- was not written before -tin Glavendrup trutin ‘husband, lord’, ON dróttinn, OE dryhten. These developments immediately explain the rise of the younger futhark, which does not denote voicedness because there were no voiced obstruents at that time. The choice of b rather than p is a consequence of the low frequency of the latter. Thus, we find original þ in Rønninge bruþur ‘brother’ beside þ < *d [t] in rauþum ‘red’ and Glavendrup faþur ‘father’ (Fyn, around 900), corresponding to OE brōþor, rēad, fæder, OHG bruoder, rōt, fater (cf. o.c.: 129). The new consonant system did not arise everywhere at the same time. While it is already attested in the Ribe skull fragment (southwestern Jutland, around 725) in the form uþin, OE Wōden, the language of the Rök stone (Östergötland, around 825) has preserved the distinction between b [p] and f in ualraubaR ‘spoils of war’ versus -ulfaR ‘wolves’ (o.c.: 14571). However, note that the latter inscription has also preserved the form sitiR ‘sits’, with -iR < -iþ (attested in Stentoften, 7th century) < *-id with final [t], later Swedish sitr (o.c.: 260). It thus appears that the lenition was earlier word-finally after unstressed vowels. In a similar vein, I assume that b represents a bilabial stop in Old Frisian habuku ‘hawk’ (Oostum comb, 8th century) and in early Old English heben ‘heaven’, gibaen ‘given’, -hebuc ‘hawk’, halb- ‘half ’, salb ‘ointment’, scribun ‘they decreed’, as opposed to gen.sg. wulfes ‘wolf ’ (o.c.: 135, cf. Campbell 1959: 179). In North-Sea Germanic the original fricatives developed voiced allophones in medial positions before the syncope. The same may have happened in Scandinavia in the 12th century. Nielsen writes: “It is interesting and puzzling (a) that the (Old) Norse reflex of Gmc. *s remained voiceless in all positions, and (b) that the reflex of Gmc. *z was not devoiced finally in the Blekinge and Eggja Early Runic consonants and the origin of the younger futhark 303 language, cf. the final fricative in Stentoften gAf vs. hAborumz” (o.c.: 137). I claim that all obstruents remained voiceless throughout the Viking Age, that -f in gAf represents an original voiceless fricative from Indo-European *-p-, and that -R was voiceless -r, not -z. BJÖRKETORP AND STENTOFTEN In his discussion of the Björketorp and Stentoften stones, Elmer Antonsen claims that “all the forms of these two inscriptions represent the same stage in the development of Old Norse, and that the few instances of syntactic and morphological variation present no real evidence for a difference in chronology” (2002: 313). This conclusion is far from obvious. In the following I shall review the evidence against the background of my reconstruction of the Early Runic consonant system (K212), where all obstruents are voiceless and R is a voiceless trill which I shall write [R]. For the sake of convenience, the two inscriptions are reproduced here in the usual transcription (cf. Krause 1966: 209-217, Antonsen 1975: 85-88, 2002: 304). BJÖRKETORP A B:I II III IV V VI uþArAbAsbA hAidRrunoronu fAlAhAkhAiderAg inArunARArAgeu hAerAmAlAusR utiARwelAdAude sARþAtbArutR “Harm(ful) prophecy. To the sequence of clear runes I commit hither mighty runes; through baseness protectionless abroad, (condemned) to insidious death, (is) he who breaks this.” STENTOFTEN I II III IV V VI niuhAborumR niuhagestumR hAþuwolAfRgAfj hAriwolAfRmAgiusnuhle hideRrunonofelAhekAhederAginoronoR herAmAlAsARArAgeuwelAdudsAþAtbAriutiþ “(With) nine he-goats, nine stallions, Haþuwolf gave good-year, Hariwolf.... To the (sequence) of clear runes I commit hither mighty runes; protectionless through baseness, (condemned) to insidious death, (is) he (who) breaks this.” 306 Scandinavian In his identification of Bj. BI-VI with St. V-VI, Antonsen equates the following forms (2002: 305-312): (1) Bj. hAidR = St. hideR = /hǃdr-/ ‘clear’. In my reconstruction, the sequence dR is voiceless [tR], with a voiceless trill. Antonsen assumes that Stentoften, unlike Björketorp, has an epenthetic e in this sequence and that *ai and *au had been monophthongized both in St. hideR, lAs, dud and in Bj. hAidR, lAusR, dAude. Alternatively, one may assume that unstressed *i was lost in Bj. hAidR but not (yet) in St. hideR < *haidiz ‘brightness’ (cf. Krause 1966: 215, Schulte 1998: 113-118, Nielsen 2000: 96) and that St. i, A, u are deficient spellings representing diphthongs in these words. (2) Bj., St. runo = /-rūnā/ ‘of runes’. Here Antonsen’s interpretation is at variance with the spelling, which points to *rūnō. (3) Bj. ronu = St. no = /(ru)nū/ ‘to the (sequence)’. Here again, the discrepancy between the vowels remains unexplained. (4) Bj. fAlAhAk = St. felAhekA = /fælh-æk(a)/ ‘I commit’, where I reconstruct preglottalized k [’k]. It may be assumed that the final vowel was lost after a light syllable in Bj. Ak but not in St. ekA ‘I’. The vowel of Bj. Ak is best explained as a reduction of *e, which also points to a chronological difference between the two forms (cf. Schulte 1998: 134). (5) Bj. hAiderA = St. hederA = /hædra/ ‘hither’. These forms have an epenthetic e in the sequence /dr/, in my view [tr] with voiced r. (6) Bj. ginA = St. gino = /ginn-/ ‘mighty’. Antonsen’s assumption of an epenthetic vowel does not explain the timbre of St. o, which must represent earlier *u and points to a chronological difference between the two forms (cf. Schulte 1998: 124-128). (7) Bj. runAR = St. ronoR = /-rūnāR/ ‘runes’. Here again, Antonsen’s interpretation leaves the discrepancies between the vowels unexplained. The spellings rather point to a difference between an older form *rūnōR and a younger form *rūnaR (cf. Nielsen 2000: 96, 263). (8) Bj., St. ArAgeu = /ærgiū/ ‘through baseness’. (9) Bj. hAerAmA = St. herAmA = /hjærm-/ ‘protection’. Here we find breaking in Björketorp but not in Stentoften, which again may reflect a chronological difference. I reconstruct Bj. [hĭĕr-], also in Istaby hAeru ‘sword-’, later [ĭă] in Ribe hiAlb ‘help’ and Rønninge skialta ‘shield’ (cf. Nielsen 2000: 109), with the short counterpart of *ē2 = [ea], as I have argued elsewhere (K124). (10) Bj. lAusR = St. lAs = [-lǿss] = /-lǿsr/ ‘-less’. This peculiar analysis smacks of paper phonetics. We must rather identify Bj. lAusR with St. lAsAR (cf. Krause Björketorp and Stentoften 307 1966: 215, Schulte 1998: 136, Nielsen 2000: 96, 259), with loss of unstressed *a in Björketorp but not in Stentoften. (11) Bj. uti = /ūtē/ ‘outside, abroad’ has no counterpart in Stentoften. (12) Bj. AR = St. AR = /æR/ ‘is’, which Antonsen derives from 2nd sg. *iz < *ezi < *esi with leveling of *e from the plural forms and lowering of *e to /æ/. I find this derivation highly improbable. First of all, St. AR is not a copula but part of the preceding word lAsAR (see above). Since there is no copula in Stentoften, there is no reason why there should be one in Björketorp. Moreover, the substitution of a 2nd sg. for a 3rd sg. form is not a natural development (see below) and does not yield the required output: “ikke blot R, men også A er umotiveret” (Jacobsen 1935: 30). I can only conclude that the sequence utiAR remains to be explained (in spite of Schulte 2006a: 24-34, 2006b: 406-410). (13) Bj., St. welA = /wǃl-/ ‘insidious’, for which I reconstruct [wēL-], with a voiceless lateral in view of the preterit vélta and the Finnish borrowing vihlata ‘delude’. (14) Bj. dAude = St. dud = /-dǿd(ē)/ ‘to death’, where Antonsen thinks that the original nonzero ending was replaced by a zero ending under the influence of the consonant stems. This is an arbitrary solution because such a development is entirely unmotivated. The discrepancy is best explained by the hypothesis that we are dealing with an n-stem where the nom.sg. ending was lost phonetically in Stentoften and replaced by the ending of the ian-stems in Björketorp (cf. Lid 1952, Syrett 1994: 151, Schulte 1998: 141, Grønvik 1998: 129-132, K219). Nedoma’s outdated view (2005) can safely be discarded. (15) Bj. sAR = St. sA = /sa(R)/ ‘he (who)’. (16) Bj., St. þAt = /þat/ ‘this’, in my view with preglottalized t [’t]. (17) Bj. bArutR = /brȳtr/, St. bAriutiþ = /brȳtiþ/ ‘breaks’, where I reconstruct voiceless b and preglottalized t. Here again, the proposed substitution of a 2nd sg. for a 3rd sg. form is unacceptable to me. It is much more probable that final *-þ became [R] when the preceding vowel was devoiced before the syncope. There is a parallel in Nivkh, where *t yielded [R] after a devoiced vowel which was syncopated (cf. K205: 285). Thus, we have the following indications for a chronological difference between Stentoften and Björketorp: (1) Bj. hAidR = St. hideR. One may assume that unstressed *i was lost in Bj. hAidR but not in St. hideR < *haidiz ‘brightness’ and that St. i, A, u are deficient spellings representing diphthongs. 308 Scandinavian (4) Bj. fAlAhAk = St. felAhekA. It may be assumed that the final vowel was lost after a light syllable in Bj. Ak but not in St. ekA ‘I’. The vowel of Bj. Ak is best explained as a reduction of *e, which also points to a chronological difference between the two forms. (6) Bj. ginA = St. gino. The timbre of St. o must represent earlier *u and points to a chronological difference between the two forms. (7) Bj. runAR = St. ronoR. Here the spellings point to a difference between an older form *rūnōR and a younger form *rūnaR. (9) Bj. hAerAmA = St. herAmA. Here we find breaking in Björketorp but not in Stentoften, which again may reflect a chronological difference. (10) Bj. lAusR = St. lAs. We must identify Bj. lAusR with St. lAsAR, with loss of unstressed *a in Björketorp but not in Stentoften. (14) Bj. dAude = St. dud. This discrepancy is best explained by the hypothesis that we are dealing with an n-stem where the nom.sg. ending was lost phonetically in Stentoften and replaced by the ending of the ian-stems in Björketorp. (17) Bj. bArutR, St. bAriutiþ. Here final *-þ became [R] when the preceding vowel was devoiced before the syncope. I conclude that Antonsen’s theory cannot be correct. In an earlier study I proposed the following relative chronology (K117: 30): (A) Umlaut of long vowels. (B) Reduction of unstressed vowels. (C) Syncope after heavy syllables. (D) Umlaut of short vowels. (E) Syncope after light syllables. Moreover, I have argued that the difference between deilda < *däildō < *dailidō ‘I divided’ and valþa < *walþō < *walidō ‘I chose’ is best explained by the assumption that intervocalic *d [t] became a fricative [þ] between the earlier and the later syncope (K102: 4) and can therefore be dated to stage D of this chronology. It now turns out that Stentoften hideR, lAsAR, bAriutiþ can be dated before and Björketorp hAidR, lAusR, bArutR after the syncope at stage C. The vocalism of the second syllables in St. hideR, felAhekA, gino, ronoR versus Bj. hAidR, fAlAhAk, ginA, runAR points to a date before and after the reduction of unstressed vowels at stage B, respectively. Syncope of original final vowels had evidently taken place already in St. welAdud and bAriutiþ and can therefore be dated before stage B. The lenition of final *-d [t] to [þ] in the unstressed ending -iþ can now also be dated before stage B while its further development to -R can be dated after the devoicing of the vowel which initiated Björketorp and Stentoften 309 the syncope at stage C. There is no reason to assume that the umlaut of stage A had already taken place in Stentoften though it must have affected Björketorp hAidR. There is no evidence for an early monophongization in St. bAriutiþ but the explicit spelling of the diphthongs in Bj. hAidR, lAusR, dAude suggests that it had taken place in bArutR. If the monophthongization can be identified chronologically with the umlaut at stage A, this offers a terminus ante quem for Stentoften. The rise of breaking in Bj. hAerAmA can perhaps be identified with the reduction of unstressed vowels at stage B. The preservation of the compositional vowels in both St. gino-ronoR, herAmA-lAsAR, welA-dud and Bj. ginA-runAR, hAerAmA-lAusR, welA-dAude suggests that the loss of these vowels, which may perhaps be identified chronologically with the syncope at stage E, had not yet taken place in the two inscriptions, especially because epenthetic vowels are not to be expected at morpheme boundaries (cf. Perridon 1991: 13). This puts Stentoften before stage A (or at least before stage B) and Björketorp between stages C and E. As I have argued elsewhere (K212: 74), the loss of glottal constriction in the preglottalized stops yielded preaspiration in the same way as the lenition of intervocalic *d [t] yielded a fricative [þ] and can therefore be dated between the earlier and the later syncope, i.e. to stage D. Since the distinction between plain d [t], g [k] and preglottalized t [’t], k [’k] is well preserved in both inscriptions, not only Stentoften but also Björketorp must be dated before stage D. This is in conformity with the apparent absence of umlaut in ArAgeu. One of the most remarkable features of the two inscriptions is their agreement in the presence of epenthetic vowels (cf. Schulte 1998: 117). As the compositional vowels of St. gino-, herAmA-, welA- and Bj. ginA-, hAerAmA-, welA- were preserved, they are not epenthetic. This leaves us with the following instances: (4) St. felAh-ekA, Bj. fAlAh-Ak < *felh-, possibly Bj. *falh-. (5) St. hederA, Bj. hAiderA, Gothic hidre ‘hither’, where the epenthetic e in the sequence *dr was probably conditioned by the preceding *i and must therefore be older than the lowering of the root vowel. (8) St., Bj. ArAgeu < *argiū. (9) St. herAmA-, Bj. hAerAmA- < *herma-. (17) St. bAriutiþ, Bj. bArutR < *breutid. 310 Scandinavian The epenthetic vowel was A [ǩ] except in (5), where e represents [ĭ]. There is no epenthetic vowel in voiceless clusters and at morpheme boundaries: (1) Bj. hAidR, where the sequence dR is voiceless [tR]. (10) Bj. lAusR, with voiceless [sR]. (17) Bj. bArutR, with voiceless [’tR]. (1-2) Bj. hAidR-runo, St. hideR-runo. (4-5) Bj. fAlAhAk-hAiderA. (12-13) Bj. utiAR-welA-. (14-15) St. -dud-sA. (15-16) Bj. sAR-þAt. (16-17) Bj. þAt-bArutR, St. þAt-bAriutiþ. The consistency of the presence versus absence of epenthetic vowels in these instances leaves no doubt about their phonetic reality. It follows that the rise of epenthetic vowels must have been early, at least before stage B. The reduction of St. ekA to Bj. Ak ‘I’ is now explained by the chronology of the univerbation, which can be dated between Stentoften and Björketorp (cf. Schulte 1998: 133). We can now test the relative chronology advocated here, first against the remaining parts of Björketorp and Stentoften and second against other inscriptions from the same period. Bj. uþArAbA-sbA < *ūþarba-spahō ‘harm prophecy’ shows voiced r and voiceless b [p] with an epenthetic A [ǩ] in rAb, preservation of the compositional vowel A [ǩ], no preglottalization and no epenthetic vowel in the voiceless cluster sb [sp], loss of *h with lengthening and syncope of the final vowel with u-coloring in view of the later form spǍ < *spāu < *spahu (cf. Nielsen 2000: 96). St. niu hAborumR niu hagestumR hAþuwolAfR gAf j hAriwolAfR ... ‘(with) nine he-goats, nine stallions, Haþuwolf gave good-year, Hariwolf ...’ shows voiceless b [p] and epenthetic o [ŭ] before the ending umR in *habrumR, voiceless g [k] and preserved e [ĭ] in *hangistumR, preservation of the first and syncope of the second vowel in the original ending *-umuR (cf. K219: 4), preservation of *u in hAþu- and of *i in hAri- without umlaut of the preceding vowel, twice epenthetic A [ǩ] and syncope in the following syllable in -wolAfR, voiceless g [k] and preservation of the original fricative in gAf (cf. K212: 74). All of these features are in agreement with the chronology established above, with early epenthesis and syncope in polysyllables and late preservation of compositional vowels. Björketorp and Stentoften 311 Now we turn to the evidence of the other Blekinge stones (cf. Krause 1966: 205-208, 218-220, Antonsen 1975: 83f.). Gummarp: hAþuwolAfA sAte stAbA þria fff ‘Haþuwolf set three staves, fff ’. Here we find syncope without umlaut in sAte < *satidē, which puts the inscription after stage E. The compositional vowel was preserved in hAþu- and the epenthetic vowel in -wolAfA, where the final A must evidently be corrected to R. The absence of an epenthetic vowel in þria also points to a recent date. Istaby: AfatR hAriwulafa hAþuwulafR hAeruwulafiR warAit runAR þAiAR ‘In memory of Hariwolf, Haþuwolf Heruwolfsson wrote these runes’. Here we find epenthetic a [ǩ] in the first five words as well as compositional i and u in the three names, also breaking in hAeru- and reduction of the long ending in runAR, which point to the same stage as Björketorp. The final *r of Istaby AfatR ‘after’ may have been devoiced by the preceding consonant, which suggests a more recent date than both St. hederA and Bj. hAiderA, with preservation of the epenthetic e [ĭ]. Thus, I date Stentoften around stage A, Björketorp between stages C and D, Istaby around stage D, and Gummarp after stage E. It must be emphasized that this sequence reflects the language of the carvers, which may rather have been characteristic of their generations than of their dialectal background. THE ORIGIN OF THE VESTJYSK STØD In his recent discussion of the vestjysk stød, Harry Perridon rejects my view that glottalization is ancient in Germanic (2006: 45). It may therefore be useful to specify the source of our disagreement. According to the view which I have put forward on a number of occasions (e.g. K211, K212, K230), the absence of voicedness in Danish stops was inherited from Proto-Germanic, where fortes stops were preglottalized and lenes stops were plain voiceless. Proto-Germanic geminates had largely arisen from Kluge’s law (cf. now K235). Other geminates arose from *-jH- and *-wH- (cf. K072: 356), e.g. tueggia ‘of two’, hNjggua ‘to hew’, also from velar stops before *j and *w, e.g. leggia ‘to lay’, bekkr ‘brook’, røkkr ‘dark’, further in sequences of nasal plus fortis stop, e.g. drekka ‘to drink’, and as a result of the syncope, e.g. leidda ‘I led’ with [tt] < *laididō [laititō] and latta ‘I hindered’ with [Ȥtt] < *latidō [laȤtitō]. Thus, we have an opposition between þ [þ], d [t], t [Ȥt], dd [tt] and tt [Ȥtt]. This system was simplified along different lines in the separate languages. The rise of voicedness in postvocalic simple obstruents can probably be dated to the 12th century, when the distinction between þ and ð was introduced (cf. Haugen 1976: 195, K212: 73). In Danish, the lenition affected not only fricatives and lenes stops, but also intervocalic and word-final p, t, k, which lost their glottalization and became simple voiceless stops. Word-initial p, t, k had become aspirates [ph], [th], [kh] at an earlier stage already, perhaps in the 7th century. As a result of the lenition, the corresponding geminates pp, tt, kk lost their distinctive length and became simple [Ȥp], [Ȥt], [Ȥk]. The voiced fricatives were further lenited to semivowels [w], [j] and the voiced stops to fricatives v [β], th/dh [ð], gh [γ] in the 13th century and later to semivowels [w], [j] in the central dialects (cf. Haugen 1976: 205). The remaining geminates bb [pp], dd [tt], gg [kk] were shortened intervocalically and became [Ȥp], [Ȥt], [Ȥk] word-finally, where original p, t, k had lost their glottalization. The original glottalization of intervocalic t [Ȥt] in fátǿkr ‘poor’ was preserved in Danish fattig because the word was still a compound at the time of the lenition. Harry Perridon assumes that the vestjysk stød arose from glottal reinforcement of unaspirated stops after the Jylland apocope in the 14th century (2006: 46-48). His argumentation is based on the Stockholm manuscript C37 of Jyske Lov from around 1280, which has numerous examples of apocopated forms, e.g. fyllægh ‘to follow’, sald ‘sold’, but none of the weakening of postvocalic p, t, k, while we find vestjysk stød e.g. in kjøvȤd ‘bought’, bruwȤd ‘used’, Danish købte, brugte. The argumentation does not hold because these words evidently continue keypta [käüȤpta], *brūkta [brūȤkta], cf. German kaufen, brauchen. Like the Proto-Germanic geminate *tt, the original clusters *pt and *kt preserved the 314 Scandinavian glottal closure of the unreleased stop, e.g. in fæmȤd ‘fifth’, oȤd ‘eight’, næȤdǩ ‘nights’, Danish femte, otte, nætter. Moreover, the suffixed article, which dates perhaps from the 11th century and was generalized in the 14th century, has no influence on the presence or absence of vestjysk stød, e.g. in Himmerland sdork ‘stork’, sdorkǩn ‘the stork’, kerȤg ‘church’, kerȤgǩn ‘the church’, Danish stork, kirke (cf. Ringgaard 1960: 49), which shows that the vestjysk stød is older. Contrary to Perridon’s statement (fn. 10), my theory does account for the difference beween sg. storkǩn and pl. storȤgǩn, where t and k stand for non-glottalized stops, because final *k had lost its glottalization before the univerbation with the suffixed article and before the apocope in the plural form. It is clear that the vestjysk stød must be older than the Jylland apocope because it is also found in the northeastern part of the vestfynsk dialects, where the apocope did not take place. Gunnar Hansson has drawn attention to the fact that vestjysk stød is found in original monosyllables and polysyllables alike in the most remote and isolated villages on the island of Als (2001: 166). We must conclude that outside these archaic dialects the vestjysk stød was lost in monosyllables, as was the case with preglottalization in Newcastle English. There is simply no evidence for glottal reinforcement of unaspirated stops. While I claim that preglottalization is ancient and that non-initial aspiration is recent, Perridon maintains the contrary. Since Jul Nielsen “found that in the dialects of the Bjerre district in SE Jutland all final stops are fully aspirated and all medial stops unaspirated” and Ejsing “writes that all medial and final stops, be they preglottalised or not, are unaspirated (pronounced as a stop after s) in the dialect of Salling (NW Jutland)” (thus Perridon 2006: 47), my view that the latter dialect directly reflects the original situation is more economical than Perridon’s, which presupposes large-scale secondary loss of aspiration. Both the alleged reinforcement of unaspirated stops and the following suppression of aspiration in recent centuries are quite unmotivated and unnatural. In my view, the whole development of obstruents from Proto-Germanic times up to the modern dialects can be viewed as a continuous process of lenition. It has been pointed out that in Danish, phonetic studies “have found lenis stops to have a greater tension than the fortes, which sounds like a terminological paradox” and that “the lack of firm closure associated with the aspirated stops was considered the decisive factor” (Goblirsch 1994: 9). This is in agreement with my view that the aspirated stops originated from a lenition process. According to Perridon’s chronology (2006: 48f.), we have to start from an alternation between medial b, d, g, bb, dd, gg, ββ, ðð, γγ and final p, t, k, pp, tt, kk, bb, dd, gg, respectively. The position of the voiced geminates in this phonological system is peculiar (cf. K212: 73). The distinctively unaspirated geminate stops, which were later reduced and allegedly reinforced by vestjysk stød, were only found in monosyllables, e.g. æȤg ‘egg’, neȤb ‘beak’, røȤg ‘back’, Danish æg, næb, ryg, Swedish ägg, näbb, rygg (cf. Ringgaard 1960: 13). The lax The origin of the vestjysk stød 315 fricatives β, ð, γ appear as glides in one of the oldest manuscripts of Jyske Lov (from around 1325), e.g. sauthæ ‘said’, lauth ‘laid’, where other manuscripts have sagthe, lagth, laght. After the Jylland apocope, the allophonic variations became phonemic, e.g. drigg < drikke ‘to drink’ versus drikk ‘drink!’, later dreȤg versus drek (Ringgaard 1960: 21). Then long obstruents were degeminated and postvocalic stops were weakened: aspirated p, t, k became lenes b, d, g and subsequently fricatives β, ð, γ or glides w, j; word-final pp, tt, kk became p, t, k and bb, dd, gg became b, d, g, while the fricatives ββ, ðð, γγ changed to β, ð, γ or w, j. At some stage, all unaspirated stops were allegedly strengthened by vestjysk stød if they followed a stressed vowel or vowel plus sonorant. These developments resulted in alternations such as væȤg ‘wall’, væγǩ ‘walls’ and træȤgǩ ‘pulls’, træk ‘pull!’, Danish væg, vægge, trækker, træk (Ringgaard 1960: 13, 24). In my chronology, g [k], k [Ȥk], gg [kk], kk [Ȥkk] became g [γ], k [k], gg [kk], kk [Ȥk] in the 12th century, then [γ], [k], [kk], [Ȥk], word-finally [γ], [k], [kk], [kh] in the 13th century, then [w], [γ], [Ȥk], word-finally [w], [k], [Ȥk], [kh] in the 14th century, with eventual generalization of [k] in the north and [kh] in the south. As a result, we find vestjysk stød immediately before the plosives p, t, k wherever these stand in an original medial position following a voiced sound in a stressed syllable (thus Ringgaard 1960: 10, 195) and in original monosyllables with final gg [kk] and bb [pp] (cf. Ringgaard 1960: 13) since the 14th century. VESTJYSK STØD AGAIN My recent article about the origin of the vestjysk stød (K251) evoked an immediate reaction from Harry Perridon (2009), which may stimulate further discussion of the subject. It is therefore important to specify the nature and origin of our disagreement. First of all I would like to emphasize that my analysis of the Germanic consonant system is in no way dependent on any theories about the IndoEuropean proto-language. My view that the vestjysk stød is ancient is based on its correspondence with preglottalization, preaspiration, gemination and affrication in other Germanic languages (cf. K211, K212, K230). The hypothesis that the vestjysk stød is a spontaneous dialectal innovation (thus Kock 1891: 368 fn., Jespersen 1913: 23, Ringgaard 1960: 108) implies the equally spontaneous independent development of preglottalization in English, preaspiration in western Scandinavia, gemination in eastern Scandinavia (and elsewhere), and affrication in High German (and elsewhere), while nothing comparable is found in the surrounding Celtic, Romance, Slavic and Finnic languages except for the development of preaspiration in Scottish Gaelic under the influence of a Scandinavian substratum (cf. Marstrander 1932: 298) and independently in Saami (cf. Sammallahti 1998: 54f.). My theory that the vestjysk stød and its English counterpart are of Proto-Germanic origin and developed into preaspiration, gemination and affrication elsewhere offers a principled explanation of these phenomena. Moreover, it accounts for the curious observation that Proto-Germanic *b, *d, *g allegedly became devoiced independently in High German, Danish (and neighbouring dialects of Swedish, Norwegian, Low German and Frisian) and Icelandic (and Faroese), and then word-initially in Swedish, Norwegian, English and Low German (thus Goblirsch 2005: 17). In a strict comparative analysis one would rather assume an archaism in High German, Danish and Icelandic and substratum influence from Celtic, Romance, Slavic and Finnish in northern English, Low and Central Franconian, northern Middle German and eastern Swedish, in spite of Goblirsch’s statement to the contrary (2005: 79). In my reconstruction, the absence of voicedness in Icelandic, Danish and Upper German obstruents represents the original state of affairs (cf. K192), which also explains the rise of the younger futhark in Scandinavia (cf. K212). I have claimed that the vestjysk stød in kjøvȤd ‘bought’ and bruwȤd ‘used’ (Danish købte, brugte) continues the glottal closure of the unreleased stops in the original clusters *pt [Ȥpt] and *kt [Ȥkt], cf. German kaufen < *-p- [Ȥp], brauchen < *-k- [Ȥk], just as the vestjysk stød in fæmȤd ‘fifth’, oȤd ‘eight’, næȤdǩ ‘nights’ (Danish femte, otte, nætter) reflects the preserved glottal closure of the 318 Scandinavian Proto-Germanic geminate *tt. Perridon objects that the verb bruge “is a loanword from Middle Low German (brûken) which entered the language/dialects in the course of the 13th or 14th century, i.e. one or two centuries after an alleged change from non-geminated [Ȥk] to [k] which Kortlandt [K251] suggests took place in the 12th century” (2009: 6). The objection is mistaken because the glottalization was preserved in the geminate and became a concomitant feature of the unlenited stop, yielding the vestjysk stød (cf. K251: 3f.). As a result, kjøvȤd < køȤftæ < keyȤpta with preservation of the glottal closure after the voiced segment is the expected development. When the glide v was devoiced to f in southern Jutland (cf. Perridon 2009: 6 fn.), the glottal stop was lost because it was no longer preceded by a voiced segment: kjøft < kjøvȤt. The vestjysk stød in the preterit bløȤd ‘bled’ (Ringgaard 1960: 25) < *-dd- is evidently of morphological origin. Perridon argues that the loss of final shwa preceded the lenition of postvocalic p, t, k, e.g. in the preterit døpt ‘baptized’ < døptæ (2009: 8). It follows that the vestjysk stød which distinguishes the preterit døvȤt < døptæ from the participle døvt < døpt is not only older than the apocope (thus already Ringgaard) but also older than the lenition. Perridon’s presupposition that this cannot be the case (2009: 5) forces him to assume an early rise and later loss of aspiration in final stops (stages 1 and 6 of Perridon 2009: 9) before and after the apocope (stage 3) and the spontaneous rise of the vestjysk stød (stage 5) in order to keep the two forms apart. This scenario is quite unmotivated. Moreover, it is at variance with the fact that the vestjysk stød is also found in the northeastern part of the vestfynsk dialects, where the Jutland apocope never took place (cf. also Ejskjær 1990). Hansson has drawn attention to the fact that the vestjysk stød is found on original monosyllables and polysyllables alike in the most remote and isolated villages on the island of Als, where it coexists with true pitch accents representing the original accents 1 and 2 from which the Common Danish stød opposition developed (2001: 166). It follows that outside these archaic dialects the vestjysk stød was lost in original monosyllables, as was the case with preglottalization in Newcastle English (cf. K211: 6). Thus, I conclude that preglottalization was inherited from Proto-Germanic, generally lost wordfinally before the apocope, and preserved as vestjysk stød elsewhere. PROTO-INDO-EUROPEAN *s IN ALBANIAN 1. Recent publications by Huld (1984) and Orel (1985) provide an incentive to reconsider the fate of PIE *s in Albanian. Though the problem was largely solved by Meyer (1892) and Pedersen (1900b), a number of unclear points have remained. In the following I intend to reconsider the evidence in order to arrive at an identification of what I see as the main difficulty. I shall not dwell upon the points which I regard as solved. 2. Initial *s- before a stressed vowel yielded gj-, e.g.: gjarpër ‘snake’, Latin serpēns; gjashtë ‘si’x, Latin sex, Greek ἕξ; gjalpë ‘butter’, Greek ἔλπος, Toch. A ṣälyp, B ṣalype; gjumë ‘sleep’, Greek ὕπνος, OCS sъnъ; gjallë ‘living’, Greek ὅλος, Skt. sárva-; gjak ‘blood’, Greek ὀπός, OCS sokъ. Orel rejects the accent as a conditioning factor on the basis of the last example (1985: 279). His argument is not valid because the accentuation of the Greek word is irregular, as Lubotsky has pointed out (1987: 167), and the Slavic evidence is ambiguous. 3. Initial *s- was dissimilated to th- before a following *s in thi ‘pig’, Latin sūs, and in thanj ‘I dry’, Lith. saũsas (on the latter etymon cf. Lubotsky 1985). Initial *sw- before a stressed vowel yielded d- in the following words: diell ‘sun’ < *swel-, Greek ἕλη, εἵλη; dergjem ‘I am ill’ < *sworgh-, Lith. sirɶgti; dirsë ‘sweat’ < *swidr-, Greek ἱδρώς, Skt. svéda-. Huld adds to this list dosë ‘brood sow’ < *swātjā and derr ‘boar’ < *swoinro(1984: 148), which are questionable examples. Before an unstressed vowel, *swyielded v- in the following instances: vjehërr ‘father/mother-in-law’, Greek ἑκυρός, -³ < *swe-; vetë ‘self ’ < *swe-; vjerr ‘I hang’, Lith. sverɶti. The clitic u- ‘self ’ may be derived from *swe-. 320 Albanian 4. PIE *s was lost before r, l, n, m, e.g.: dorë ‘hand’ < *gʄhēsrā; kollë ‘cough’ < *kwāslā, Russian kášeľ; thanj ‘I dry’ < *sousnjō; notim ‘swimming’ < *snāt-, Skt. sn¡ti; mjekër ‘beard’ < *smekʄrā, Lith. smãkras; jam ‘I am’ < *esmi. Final *-s was also lost, e.g.: mi ‘mouse’ < *mūs; thi ‘pig’ < *sūs. It is clear from the latter word that the loss of final *-s was posterior to the dissimilation of the initial *s- to th-. It was apparently anterior to the delabialization of final *-ū, cf. ti ‘you’ < *tū. 5. Initial *sp- yielded f-, e.g.: farë ‘seed’, Greek σπορ³; fjalë ‘word’, OE spell. Initial and medial *sk yielded h, e.g.: hie ‘shade’, Greek σκι³, Toch. B skiyo; hedh ‘I throw’ < *skeudō, OE scēotan; ah ‘beech’ < *osko-, OE æsc; -h < *-skō. Medial *-sd(h)- yielded the same reflex as *-d(h)-, viz. -dh-, which was sometimes devoiced to -th- (cf. Jokl 1912: 198-210), e.g.: pidh ‘vulva’, Old Prussian peisda; ledh ‘wall’, OHG līsta; gjeth ‘leaf, OHG questa; drithë ‘grain’, OHG gersta. 6. It seems that PIE *s yielded sh in all other positions, e.g.: shoh ‘I see’ < *sēkwskō (Pedersen 1900b: 283) or *sokwēskō (Huld 1984: 115), OHG sehan; shtatë ‘seven’ < *septḿ-; shtërpinj ‘vermin’ < *serpén-; shi ‘rain’, Old Prussian soye; shosh ‘I sieve’, Lith. sijóti; vesh ‘ear’, Lith. ausìs; breshër ‘hail’ < *bhreus-, Latin frustum; Proto-Indo-European *s in Albanian 321 desha ‘I loved’ < *gʄeus-, Greek γεύομαι; mish ‘meat’, Skt. māṃsám; dhashë ‘I gave’, pashë ‘I saw’, rashë ‘I fell’; abl.pl. -sh < *-su; shteg ‘path’, Greek στοῖχος; asht ‘bone’, Greek ὀστέον; është ‘is’ < *enesti; jashtë ‘outside’ < *egʄhstos, Greek ἐχθός; gjashtë ‘si’x < *sekʄs-t-; shkonj ‘I go’ < *st(o)igh-; shpend ‘bird’ < *su-petno-t- (Huld 1984: 154); shpreh ‘I utter’ < *(sm)-spregʄ-skō (Huld 1984: 64) or rather *egʄhspregʄskō, OHG sprehhan; ashtë ‘beech grove’, cf. ah ‘beech’ < *osko-; kashtë ‘straw’ < *kolstā, cf. kall ‘ear (of cereals)’; viç ‘calf ’ < *wetes-, Skt. vatsá-. The absence of sh in djathtë ‘right’ precludes its identification with Latin dexter. The word can be compared with OCS desnъ and derived from *dekʄsn-, where *s was lost before *n, with later addition of the suffix -të. Orel’s derivation of -th in ankth ‘anxiety’ and zjarrth ‘fever’ from *-st- (1985: 282) is arbitrary. The suffix denotes physical affections (cf. Camaj 1966: 122) and must probably be connected with Slavic -ota, Skt. -atā. 7. It follows from the position taken here that a number of etymologies where PIE *s is allegedly reflected as Alb. h or zero cannot be maintained. Thus, I withdraw the view that the intervocalic reflex of PIE *s is zero (K085: 42). The most important instances are the following: (h)yll ‘star’, which Meyer connected with the PIE word for ‘sun’ and derived from *sulno- or *sūli- (1891: 460). Though Pedersen remarked that “die bedeutungsentwicklung ist zwar recht auffällig” (1900b: 278), he stuck to the connection and derived the word from *sūlo- or *sūli-. The etymology was rejected by Huld, who pointed out the secondary character of the initial h-, denied the likelihood of an association with the word for ‘sun’, and proposed to equate Alb. yll with OE ysle ‘spark’ < *usli- (1976: 180 and 1984: 132). This etymology is certainly preferable. he(l)q ‘I pull’, which Meyer identified with Greek ἕλκω and Latin sulcus (1891: 151). Pedersen derived the verb from *solkejō, assuming that it adopted the flexion of primary verbs at a later stage (1900b: 278). Rejecting the derivation of h- from *s-, Hamp posited a Proto-Albanian verb *Hwolkejō to be compared with Lith. viÞkti (1965: 132). Finally, Huld connected the word with OE ealh ‘temple’ and reconstructed *Holkejō (1984: 73). I wonder if the word may rather 322 Albanian be cognate with OE sceolh ‘wry’. In any case, it cannot be used as evidence for a development of h- from *s-. kohë ‘time’, which according to Meyer “ist vielleicht mit asl. časъ ‘Zeit, Stunde’ verwandt” (1891: 194), a comparison which he characterized as “zweifelhaft” (1892: 86). According to Pedersen, the comparison “lässt sich kaum bezweifeln”, in spite of the fact that it is “das einzige beispiel für inlautendes h aus idg. s; sonst herrscht š” (1900b: 279). It is certainly not “eine evidente Wortgleichung” (Jokl 1937: 159n.) because the two words have little in common. First, we should expect palatalization of the initial k- in view of zorrë ‘intestines’ < *gwērnā. Second, the derivation of the medial -h- from *s is not supported by other examples. Third, the final -ë must be derived from *-ā, which points to a different flexion class from the Slavic word. The equation must therefore be abandoned. ai ‘he’, ajo ‘she’, plural ata, ato, also ‘that, those’; ky, kjo, këta, këto ‘this, these’. These words consist of a deictic element a-, kë- plus an anaphoric pronoun -i, -jo, -ta, -to. Pedersen derived the latter from PIE *so, *sā, *to- (1897: 288 and 1900b: 282), a view upheld even by Huld (1984: 148). The key argument is the observation that the distribution of -j- and -t- corresponds with PIE *so, *sā, *to-. This does not substantiate the derivation of -i, -jo from *so, *sā, however. Firstly, the derivation does not explain the vowel -i, for which Pedersen assumes a development of final *-o to *-ü with subsequent delabialization, adducing dy ‘two’ < *dwo and acc. ty ‘you’ < *twe in support of his view (1900b: 282). But ty must be derived from the north-east Gheg form tye < *twēm, cf. mue ‘me’ < *mēm (Jokl 1963: 142), and Huld derives dy from *duwai, OCS dъvě (1984: 57), which is preferable. Secondly, Pedersen assumes that -j- is a simple hiatus filler because *j is normally reflected as gj (1900b: 313). It is unclear how an epenthetic -j- could originate between a- and -o, however. Thirdly, I think that the expected reflex of PIE *so is actually attested in the interrogative pronoun kush ‘who’ < *ku-so, obl. kujt, cf. OCS kъto. It seems that Pedersen’s view of the demonstrative pronoun prevented him from considering this interpretation of the interrogative: “der nominativ ist ku-š zu zerlegen; š muss rest eines nicht näher zu bestimmenden pronominalen elementes sein; durch analogische anfügung der genitivendung -i an den stamm ku- entstand kuj, mit dem postpositiven artikel kujt” (1900b: 317). In view of these difficulties, it is necessary to reconsider the system of demonstrative pronouns which can be reconstructed for Proto-Albanian. 8. In an earlier study I pointed out that a reconstruction of the Balto-Slavic demonstrative pronouns leads to the establishment of a single demonstrative *so, *to-, an anaphoric pronoun *e/i-, and three deictic particles, *kʄi ‘hic’, *au ‘istic’, and *an ‘illic’, and that this system also accounts for the demonstrative pronouns of Armenian and Tocharian (K063). The attested paradigms resulted Proto-Indo-European *s in Albanian 323 from various conflations of these elements. One may wonder if the Albanian forms can be derived from the same system. The deictic element a- may represent *au, but the element kë- cannot be derived from *kʄi, which is perhaps found in sot ‘today’, sonte ‘tonight’ < *kʄjā(Huld 1984: 112), cf. ditë ‘day’, natë ‘night’, Greek σήμερον ‘today’, though a comparison with sivjet ‘this year’ rather supports a derivation from *tjā(Pedersen 1900b: 311), cf. abl. masc. kësi, fem. këso ‘this’ < *-tj-. It seems evident to me that kë- reflects Latin eccum, Italian ecco, which played a dominant role in the formation of the Romance demonstratives, e.g. Italian qui ‘here’ < *eccu-hic. This derivation explains the labialization in Alb. ky < *ku-i. It suggests that Alb. a- must perhaps be connected with Romance a- (cf. Meyer 1891: 1). The second component of the demonstrative pronouns -i, -jo, -ta, -to may represent a conflation of the PIE demonstrative *so, *to- with the anaphoric pronoun *e/i-. Indeed, Alb. -i, -jo can be directly compared with Latin is, ea (Meyer 1892: 79), or rather with Skt. ayám, iyám < *ei-om, *iH-om (cf. Beekes 1983: 209), with added *-ā in the feminine. The unstressed variants are found in the article i, e. Similar conflations took place in Germanic, Baltic, Slavic, and Tocharian. The Germanic paradigm of *so, *to- has *te- in the genitive forms and in the feminine dat.sg. form. Since there is no motivation for the analogical introduction of e/i-forms into the paradigm of *so, *to-, I think that it actually represents the paradigm of *e/i- which took *t- and suppletive nominative and accusative forms from *so, *to-. The original nom. and acc. forms with added *s/t- are found in the Viking age paradigm of ON siá ‘this’ (cf. Beekes 1983: 219). The origin of these forms can be dated to an earlier period in view of the Vedic evidence (ibidem: 216); they do not belong to the same paradigm. In Prussian, the paradigm of *e/i- adopted *an- in order to create a form meaning ‘he there’, which subsequently gave rise to correlating forms with *tand *si- (cf. K063: 317). The further development of these forms yielded an anaphoric pronoun *tan(e/i)-, an article *st(e/i)-, and demonstrative pronouns *sta- and *ši(a)- (ibidem: 312-314, cf. K050: 9). The Slavic and Old Lithuanian paradigms of *e/i- have suppletive nominative forms with *an- (cf. Van Wijk 1918: 116), which evidently have the same origin as the Prussian forms. Note in this connection the parallel in Greek, which has a suppletive nominative for αὐτό- as a 3rd person pronoun. The Slavic demonstrative sь ‘this’ is inflected as a soft stem in spite of the fact that the s- was hard, as is clear from the West Slavic reflex s-, not š-. The paradigm must therefore be derived from the addition of sfrom *si ‘hic’ to the inflected forms of *e/i-. This derivation also accounts for the deviant masc. nom.pl. form sii, which may represent *kʄi plus *ei. 324 Albanian The paradigm of the West Tocharian word for ‘this’ is the following: nom.sg. obl.sg. nom.pl. obl.pl. masc. se ce cey ceṃ fem. sā tā toy toy neuter te te Since e is the phonetic reflex of PIE *o, the masc. forms show the expected development of PIE *so, *tom, *toi, *tons, except for the fact that c- is the phonetic reflex of *t before a front vowel. It follows that there must have been a stem *te- with a suppletive nom.sg. form and that the initial consonant was adopted in the demonstrative pronoun. Elsewhere I have argued for the reconstruction of a Proto-Tocharian anaphoric pronoun *ä- < *e/i- which adopted *t- from the demonstrative (K063: 321). I think that we must assume the same conflation for Proto-Albanian. Thus, I propose to derive Alb. -i, -jo from *is, *ijā or *ei, *ejā, acc. -të < *tom, *tām, pl. -ta, -to < *tons, *tās, gen.dat. -ti(j), -saj < *tei(-), *t(e)jāi, abl. -si(sh), -so(sh) < *t(e)jei(su), *t(e)jā(su) (cf. Beekes 1983: 209, Pedersen 1900b: 314). The unstressed variants of these forms are found in the article i, e, të, së. In any case, the demonstrative pronoun cannot be used as evidence for the alleged loss of PIE *s in Albanian. PROTO-INDO-EUROPEAN *j IN ALBANIAN According to Meyer (1892: 39f.), PIE initial *j- can be reflected as either gj- or jin Albanian: gjesh ‘knead’, Skt. yásati, Greek ζέω ‘seethe’, OHG jesan ‘foam’. ngjesh ‘gird’, Avestan yāsta-, Greek ζωστός, Lith. júosti. gjër ‘soup’, Skt. yáuti, Lith. jáuti ‘mi’x. ju ‘you’, Avestan yūš, Gothic jūs, Lith. js. jē ‘permission’, Skt. yóḥ ‘welfare’, Latin iūs ‘justice’. a-jo ‘she’, kë-jo ‘this’ < *jā , Skt. yā. Pedersen regarded gj- as the only regular outcome of PIE *j- in Albanian and derived j- in ju and -jo from a hiatus filler (1900a: 103, 1900b: 313). Jokl agrees with Pedersen and connects jē ‘permission’ with Skt. áviḥ ‘favorable’ (1911: 32). Çabej has argued in favor of a reflex z- from PIE *j- (1956, cf. 1972: 139). This view cannot be maintained, as Orël has made quite clear (1989: 41f.). Orël discusses the available evidence in detail and presents an extensive account of the scholarly literature (to which Jokl 1911 and Rusakov 1987 should be added). He rejects Pedersen’s suggestion of a hiatus filler and proposes additional instances of both gj- and j- from PIE *j- (1989: 43f.): gjaj ‘happen’ and gjaj ‘resemble’ from *jā ‘go’, Skt. y¡ti. gjem ‘bridle’, Skt. yámaḥ. gjer ‘till’ < *ajeri, Greek ἤρι ‘early’, Avestan ayar- ‘day’. gjymësë, gjysmë ‘half ’, gjymtë ‘defective’, Skt. yamáḥ ‘twin’. josh ‘curl, fondle, caress’, Skt. yódhati ‘fight’, Lith. jáudinti ‘excite’. juzi, juci ‘(thin) silt’, Lith. jáudra ‘swamp, marsh’. Orël suggests that PIE *j- yielded j- before back vowels in Proto-Albanian and developed into gj- elsewhere. I find this account unsatisfactory for a number of reasons. First of all, the new comparisons are far from compelling semantically. Though meanings sometimes change in unexpected ways, it seems obvious to me that the proposed etymologies for josh and juzi, juci cannot be used as independent evidence for the establishment of a sound law. Note that the latter instance may be related to llucë ‘Jauchengrube’ (Kristoforidhi apud Orël). Secondly, the phonetic motivation for the retention of *j- before back vowels, as opposed to the rise of gj- elsewhere, remains unclear. One would rather expect the opposite. Thirdly, the number of examples is really too small to establish a distribution of gje-, gjë-, gja-, gjy- versus jo-, ju-. Moreover, the clear instances of 326 Albanian j- < PIE *j- are pronominal stems, as Orël remarks himself (1989: 43). It therefore seems appropriate to assume that gj- is the only normal reflex of PIE *j- and to look for a special development in the pronouns. Elsewhere I have argued that -jo in ajo ‘she’ and k(ë)jo ‘this’ represents *ijā or *ejā, Latin ea, Skt. iyám < *iH-om, cf. masc. -i in ai ‘he’, ky < *ku-i ‘this’, Latin is, Skt. ayám < *ei-om (K094: 224f.). It follows that -jo reflects the intervocalic development of PIE *j in Albanian. We may therefore surmise that ju ‘you’ represents a cliticized form *ju of the PIE pronoun *jū < *iuH (cf. Beekes 1995: 208) after the verbal ending *-te, cf. na ‘we’ < *nos, where the short vowel also points to a clitic. This leads us to reconsider the reflex of PIE intervocalic *-j- in Albanian. According to Orël, “while in the majority of Slavonic borrowings Slav. *-j- is reflected as Alb. -j-, there exist a few Slavonic elements (presumably belonging to the earliest stratum of Slavonic words in Albanian) which display the development of *-j- to Alb. -h-” (1989: 46), e.g. krahinë ‘region, area’, OCS kraina. It appears that Alb. -h- is a hiatus filler here. Orël further adduces krah ‘arm, shoulder, wing, side’ and llohë ‘rain with snow, snow broth, dampness’, allegedly from Slavic *kraj- and *loj-, respectively. These examples are not convincing for both formal and semantic reasons. In the inherited material, Orël proposes Alb. -h- < *-j- in bahë ‘sling’ < h *b ojā, pah ‘scab, dust’ < *poujo-, prehër ‘lap’ < *prōjenos, shtrohë ‘kennel’ < *strējā, vehte ‘self ’ < *svojeti- beside vete and vetë, and in the verbs ftoh ‘cool’ and ngroh ‘warm’, with *-jō where other scholars reconstruct *-skō. While vehte beside vete and vetë is unclear, there is little reason to assume *-j- in the other words. Orël finds a zero reflex of intervocalic *-j- in brie ‘caries’ < *bhrejā, di ‘know’ < *dhejō or *dhijō, fle ‘sleep’ < *owo-lejō, hie ‘shadow’ < *skijā, kri ‘worm’ < *kwr(i)jo-, ve egg’ < *ōwijom, bie ‘fall, strike, beat’ < *bhejō, dhi ‘she-goat’ < *aigʄijā, fli ‘offering, sacrifice’ < *owo-lejos, tre ‘three’ < *trejes. All of these examples point to loss of intervocalic *-j- after a retained front vowel. Thus, I think that the phonetic reflex of intervocalic *-j- is Alb. -j-, which was lost after a front vowel at a recent stage (but earlier than the rise of new -jfrom -lj- and -nj-). As the reflex of intervocalic *-s- is -sh-, not -h- (cf. K094: 221f.), the two never merged except word-initially before a stressed vowel, where both yielded gj-. New initial j- arose from breaking of *e-, e.g. jam ‘am’ < *esmi, jashtë ‘outside’ < *egʄhstos. Orël’s chronology (1989: 48) must therefore be revised as follows: (1) (2) (3) (4) (5) s > sh, sh- > zh- before stressed vowels, j- > gj-, zh- > gj-, rise of new j- and zh-. Proto-Indo-European *j in Albanian 327 It is possible that the development of *j- to gj- at stage (3) was limited to the position before a stressed vowel. If the derivation of gjer ‘till’ < *ajeri is correct, the rise of gj- must be dated after the reduction of pretonic vowels. However, I find it difficult to separate gjer from deri ‘until’, which precludes a derivation from *j-. It seems possible to me that gjer took its initial consonant from gjerë, gjërë ‘broad, wide, far’, for which no convincing etymology is available. REFLEXES OF INDO-EUROPEAN CONSONANTS IN ALBANIAN In his study of Albanian etymologies, Bardhyl Demiraj briefly summarizes his views on the development of the Indo-European phonemic system in this language (1997: 41-67). Here I shall review the reflexes of the IE consonants on which I have written myself (K022, K085, K094, K146). 1. I have argued that the IE plain velars developed from neutralization between palatovelars and labiovelars after *s and *u in the proto-language, from delabialization of labiovelars before rounded vowels in western IE languages, and from depalatalization of palatovelars before resonants in eastern IE languages (K022). Demiraj lists the following instances of plain velars (1997: 64): (1) kap ‘(an)fassen, ergreifen’ < *kH2p-, Greek κάπτω, where I reconstruct a palatovelar which was depalatalized before the following laryngeal. (2) thek ‘(er)wärmen, rösten’ < *kʄeuk-, Skt. śócati, with neutralization in *-uk-. (3) ag ‘Dämmerung, Strahlen’ < *H2eug-, Greek αὐγή, with neutralization in *-ug-. (4) gardh ‘Zaun, Gehege’ < *ghordho-, Lith. garɶdas, Gothic gards, Skt. gṛhá- with depalatalization of the palatovelar in the zero grade form. For other examples I refer to my earlier treatment (K022). 2. I have suggested that IE *H2e- and *H3e- developed into Albanian hawhereas IE *Ho- and syllabic *H2R- and *H3R- developed into a-, aR- (K085: 43-45). Demiraj lists the following examples (1997: 59, 51): (1) (h)éthe ‘Fieber’ < *H2eidh-, Greek αἶθος, with (h)e- < *hai-. (2) hérdhe ‘Hoden’ < *H3ergʄhi-, Greek ὄρχεις, with umlauted he- < *ha-, but Arm. orjik‘ < *H3rgʄhi- with zero grade. (3) (i) hídhët, (h)ídhtë ‘bitter’ < *H2idh-, Greek ἰθαρός, αἴθω, with zero grade *H2i- and h- from *H2ei-? (4) hut ‘vergeblich, leer, eitel’ < *H2uto-, Greek αὔτως, with zero grade *H2u- and h- from *H2eu-? (5) ah ‘Rotbuche’ < *H3osk-, OHG asc, Greek ὀξύη, but Arm. hac‘i < *H3esk- (cf. Schrijver 1991: 77). (6) (i) áthët, áthtë ‘herb, sauer, scharf ’ < *H2okʄ-, Latin acidus < *H2ekʄ- (cf. Schrijver 1991: 132). (7) (i) épër ‘oberer’ < *H1opi-ro-, Greek ὄπιθε(ν), with umlauted e- < *a-. (8) ag ‘Morgendämmerung, Strahlen’ < *H2eug-, Greek αὐγή, with dissimilation of *h- before *-w- (thus Demiraj)? I would rather assume loss of *h- after a preposition, as in Arm. ayg ‘dawn’ < *H2eusi (cf. Clackson 1994: 223, K194: 119). 330 Albanian (9) aft ‘Luftzug des Blasebalgs, des Windes, Duft, heißer Anhauch des Feuers’ < *H2euH1-et-, afsh ‘(heißer oder kalter) Atem, Duft’ < *H2euH1-es-, Greek ἀϋτμή, with dissimilation of *h- before *-w ? I have no opinion on these words. Note that the Greek form is also unexplained (cf. Fritz 1993). (10) a- deictic particle < *H2eu-? I think that this is a borrowing from Latin because the same holds for the corresponding particle kë- < *ku < *eccum (cf. K094: 223). (11) a question particle < *H2en, Latin an. Initial *h- and final *-n can easily have been lost in this proclitic element. (12) arí ‘Bär’ < *H2rtkʄo-, Greek ἄρκτος, Arm. arǐ. (13) árë ‘bebautes Feld’ < *H2rH3uo-, Latin arvum < *H2erH3- (cf. Schrijver 1991: 250). (14) émër Name < *H1nH3men- < *H3nH3-men-, Greek ὄνομα, with dissimilation of the initial laryngeal. Demiraj’s suggestion that *h- was dissimilated before *-w- in the reflex of IE *H2eu- (1997: 59) seems improbable to me, not only because this is a phonetically unmotivated development, but especially because there is counterevidence in the following instances: (4) hut ‘empty’ if h- was taken from the full grade form (see above), (15) ha ‘eat’ < *H2eu-(H1ed-), aor. hëngra < *-gwerH3-, Skt. áva-(gir-), Arm. utem, aor. keray (cf. K085: 40). 3. I have argued that IE *s yielded sh in all positions except the following (K094, cf. Demiraj 1997: 56-58). 3.1. Initial *s- yielded gj- before a stressed vowel. Demiraj’s examples: gjálpë ‘Butter’, gjak ‘Blut’, gji ‘Brust, Busen, Bucht’. 3.2. Initial *s- was dissimilated to th- before a following *s in thi ‘pig’ <*sūs and thaj ‘I dry’ < *sausnjō. Demiraj is doubtful about this dissimilation. 3.3. Initial *sw- yielded d- before a stressed vowel and v- before an unstressed vowel. Demiraj assumes a development to v- before a stressed and to h- before an unstressed vowel. His examples are the following (1997: 48): (1) vétë ‘selbst’ < *swoi-, Slavic svoj- ‘eigen’. Here I reconstruct original final stress on the basis of Skt. svayám ‘self ’. The clitic u ‘self ’ represents unstressed *swe. (2) vjéhërr ‘Schwiegervater’ < *swekʄwer- (thus Demiraj). Greek ἑκυρός has final stress. (3) he(l)q ‘ziehen’ < *swelkō, Greek ἕλκω. I have discussed this word in detail (K094: 222) and connected it with Old English sceolh ‘wry’ < *skelk-. It cannot be used as evidence for a reflex h- < *s(w)-. Reflexes of Indo-European consonants in Albanian 331 The prime examples of d- < *sw- before a stressed vowel are the following: (4) díell ‘sun’ < *swel-, which Demiraj does not discuss. (5) dérgjem ‘am ill’ < *swerghō = Lith. sergù, which Demiraj prefers to compare with Lith. dìrgti ‘get spoiled’ (1997: 131). The latter etymology is semantically less convincing. (6) dírsë, djérsë ‘sweat’ < *swidro-tjā, Greek ἱδρώς. Demiraj objects that this etymology does not explain the loss of *-d- before -r- in Albanian (1997: 139). However, I am not aware of any clear counter-example against this development. Since medial *-d- became a fricative and intervocalic *-d-, like *-b-, may have been lost (cf. Demiraj 1997: 62), I think that the objection cannot be maintained and that the etymology is perfect. 3.4. IE *s was lost before r, l, n, m. Demiraj assumes compensatory lengthening before r, l and gemination before n, m (1997: 44, 55), and I think that he is right: (7) dórë ‘Hand’ < *gʄhesr-, Greek χείρ, with o < *ē. (8) (h)yll ‘Stern’ < *H1usl-, ON usli, with y < *ū. (9) thom ‘sage’ < *kʄēH1smi, thën ‘gesagt’ < *kʄH1sno-, not -rë. 3.5. Final *-s was lost, e.g. mi ‘mouse’ < *mūs, thi ‘pig’ < *sūs. I disagree with Demiraj’s reconstruction *dekʄs- ‘right’ (1997: 58) and attribute the loss of *s in (i) djáthë, djáthtë to an earlier suffix *-no-, as in Slavic desn- <*dekʄsno- (K094: 221). 3.6. Initial *sp- yielded f-, while initial and medial *sk yielded h-, -h-. (10) fárë ‘Samen, Saat’ = Greek σπορ³. (11) híe ‘Schatten’ = Greek σκι³. (12) -h inchoative suffix < *-skō. The evidence is obscured by IE *s- mobile and by the Albanian prefix sh-: (13) hálë ‘Granne, Gräte, Nadel’ < *skolnā, kallí ‘Ähre’ < *kol-, Lith. skélti, Russ. kólos. (14) harr ‘(aus)jäten’ < *skor-nō, ther ‘stechen, reißen’ < *kʄer-, Greek κείρω. (15) korr ‘absicheln’ < *kos-r-nō, Slavic kosa. (16) shqerr ‘zerreißen’ < *egʄhs-sker-nō, Greek κείρω. 3.7. Medial *-sd(h)- appears to have yielded -th-: (17) gath ‘Kätzchen’ < *gwosdo-, with umlaut gjéthe ‘Laub, Blatt’, OHG questa. (18) dríthë ‘Getreide’ < *gʄhṛsdh-, OHG gersta. The alternative development to -dh- (K094: 220) must probably be abandoned: (19) pidh beside pith ‘vulva’ may be secondary (thus Demiraj 1997: 319f.). (20) ledh ‘wall’ must be derived from ledh ‘Lehm’ = Old Prussian laydis (cf. Demiraj 1997: 235). 332 Albanian 3.8. IE *s never yielded h in Albanian, and the comparison of kóhë ‘time’ with Slavic časъ must be abandoned (cf. K094: 222). Demiraj derives kóhë from *kēswā (1997: 57, 222), which is arbitrary. I have suggested that the word must be connected with qoj ‘awaken’ (K022: 248). 4. Following Pedersen, I have argued that initial *j- yielded gj- (K146). Demiraj is doubtful and mentions the following instances (1997: 47): (1) (n)gjesh ‘gürten, umschnallen’ < *(H)ieH3s-, Greek ζώννῡμι. (2) ju 2nd pl. < *ju- is a clitic pronoun. For other examples I refer to my earlier treatment (K146). I see no evidence for a distinct reflex of *Hj- in any Indo-European language. Thus, I find myself in agreement with Demiraj on most issues. Our main differences concern the reflexes of *H2u-, *H2eu- and *sw- and the etymology of the word kóhë. I gratefully accept his proposal of compensatory lengthening in *-sr-, *-sl- versus gemination in *-sn-, *-sm- and his elimination of -dh- as a phonetic reflex of *-sd(h)-. ARMENIAN ewł ‘OIL’ Antoine Meillet has established an alternation between stressed ew and unstressed iw, e.g. in ewł ‘oil’, gen. iwłoy, inst. iwłov, later analogical iwł (1903: 493), which replaced the earlier paradigm of ewł, gen. ełoy, and analogical eł (p. 496). He maintained that the -w- developed phonetically in final -ił and -eł, listing giwł ‘village’, iwł ‘oil’, erkiwł ‘fear’ and čiwł ‘branch’ as examples and citing Cilician geł ‘village’ as analogical on the basis of gen. gełǐ (1904: 27). This resulted in the fourfold shape of the word eł, ił, ewł, iwł, all meaning ‘oil’. Holger Pedersen accepted Meillet’s view and added p‘ił, p‘eł, p‘iwł, gen. p‘łi ‘elephant’, zambił, zambiwł ‘basket’, piłc, piwłc, pełc, gen. płcoy, also płtor ‘dirty’, ceł, cił, ciwł, gen. cłi, also cławt ‘stalk’, ƽnciwł, ƽnjiwł, ƽnjił ‘shoot’, šił, šiwł, šeł, gen. šłi, šłoy, šiwłi, šiwłoy, šełi ‘twig’ (1906: 402). The diphthongs ew and iw never arose phonetically in non-final syllables, where ew does not occur and iw is always the result of an analogical development. The converse analogy is found e.g. in geł ‘beauty’, gełec‘ik ‘beautiful’. The phonetic development must obviously be dated after the apocope (thus already Pedersen 1906: 403). The development of -w- before final -ł established by Meillet and Pedersen has been disregarded by Godel (1975), Schmitt (1981), Klingenschmitt (1982) and Rasmussen (1985), but was accepted by Beekes (apud K194: 205). Joachim Matzinger has recently proposed to derive ewł ‘oil’ from *seiblo- or *soiblo-, cognate with German Seife ‘soap’ (2006). This is unfortunate because the German root alternates with *seip- (Sieb ‘sieve’), *seikw- (seichen ‘to pee’) and *seim- (Seim ‘treacle’), which renders the comparison quite useless. I would rather identify ewł with Greek ἔλπος, Albanian gjalpë ‘butter’, Sanskrit sarpís, German Salbe ‘ointment’, Tocharian A ṣälyp, B ṣalype, with regular loss of *p before *o between stages 10 and 12 of my chronology (K194: 28f.). Note that there is no evidence for an epenthetic -w- from apocopated *u (contra Pedersen 1906: 408f., Meillet 1936: 55, Godel 1975: 88, Klingenschmitt 1982: 153f. and later authors) but only from syncopated *u (cf. Pedersen 1906: 410f. on giwt ‘finding’ and mawt ‘near’). The instances of -w- which have erroneously been adduced in this connection are the result of consonantal developments (cf. K194: 60 on artawsr ‘tear’ and awr ‘day’ and p. 27 on awj ‘snake’ and awcanem ‘anoint’). THE BALTIC WORD FOR ‘IN’ Professor Zigmas Zinkevičius has again drawn our attention to the presence of an acute in the inessive ending *-ęɴ as opposed to the absence of an acute in the preposition  ‘in’ (2008, cf. Stang 1966: 182). As he points out, we also find an acute in the nominal prefix -, e.g. lanka ‘bay’, pėdinis ‘heir’, sūnis ‘adopted son’, ìndėvė ‘poison’, as opposed to - in linkas ‘concave’, prastas ‘usual’, suka ‘screws in’, iñdas ‘dish’. As Zinkevičius correctly observes, this is the same alternation as in pókalbis ‘conversation’, prótėvis ‘ancestor’, príetėmis ‘twilight’, pérpykis ‘anger’ beside põ ‘about’, prõ ‘through’, priẽ ‘at’, perɶ ‘across’. He does not mention the comparable alternations in núo-, nuõ- and sąɴ-, sąɶ- (e.g. in sado ‘hires’, sañdas ‘component’), nor the short prefixes pa-, pra-, pri-, nu-, su-. The big question is: how did these alternations originate? The Balto-Slavic acute was a glottal stop which developed from an IndoEuropean laryngeal or preglottalized stop after an original short vowel or diphthong (cf. K025 and K064). It follows that -, pó-, pró-, príe-, pér-, sąɴ-, núoare the expected variants of  < *in, pa-, pra-, priẽ < *prei, perɶ, sa-, *na (Prussian na ‘on’) before an Indo-European word-initial laryngeal or preglottalized stop, e.g. in nèšti ‘to carry’, dúoti ‘to give’, cf. Greek ἤνεγκον ‘I brought’, δίδωμι ‘I give’. Thus, the rise of the acute in the prefixes is the same as in the reduplication syllable of dúodu as opposed to dedù ‘I put’, Greek τίθημι (cf. K025: 323). The acute nominal prefixes are also attested in Slavic, e.g. Russian páguba ‘ruin’, pásynok ‘stepson’, prádedy ‘ancestors’, súdoroga ‘cramp’, súmerki ‘twilight’, which clearly show that the formation can be dated to the Balto-Slavic period. Apart from the tonal difference, there is an apophonic distinction between zero grade in Lith. į-, pri-, nu-, su-, also Slavic vъ ‘in’, sъ ‘with’, and full grade in Lith. pa-, pra-, prie-, nuo-, są-, also inessive *-ęɴ, Latvian ìe- ‘in’ < *en, Slavic po ‘after’, pro ‘through’, pri ‘at’ < *prei, na ‘on’ < *noH, sNj- ‘together’ < *som, Nj- ‘in’ < *on- in Njtrь ‘inside’, Njtroba ‘entrails’, Russian vnutrí, utróba (cf. Derksen 2008: 387). It now appears that the vowel of Lith. nu-, su- and Slavic vъ, sъ represents a secondary zero grade on the analogy of the o-grade in Lith. nuo-, są-, Slavic Nj-, sNj- (cf. Trautmann 1923: 4, Vaillant 1950: 173, K245: 10). Conversely, Prussian has introduced a secondary front vowel in the preposition sen ‘with’ beside san- and in the prefixes ep- and et-, East Baltic ap-, at-, Slavic ob-, ot-. This is a result of the fact that the initial vowel of Prussian en ‘in’ and esse ‘from’ represents a secondary zero grade going back to an originally pretonic reduced grade *i- < *e- reflected in Lith. į, iš and Slavic vъ, iz < *ьz (Greek ἐν, ἐξ) whereas the regular phonetic reflex of Balto-Slavic initial *e- is a- in Prussian, e.g. addle ‘spruce’, alne 336 Balto-Slavic ‘hind’, as ‘I’, asmai ‘am’, assaran ‘lake’, also an ‘in’ (7×) and assa ‘from’ (6×) in the First Catechism (cf. K190). It follows that the acute of the inessive ending *-ęɴ cannot have the same origin as the acute in the nominal prefixes -, pó-, pró-, príe-, pér-, núo-, sąɴ-. Elsewhere I have argued that the inessive was created by the addition of stressed *en to the East Baltic pronominal locative forms masc. *tàmi, fem. *tàjai, pl. *tàisu (Slavic tomь, toi, těxъ), yielding *tamęɵ, *tajęɵ, *taisęɵ with glottalization from the hiatus before *en (K221: 68). The absence of an acute tone in the stem of the pronoun (which is clear from the Serbo-Croatian evidence) explains the root stress in the forms rañkoje ‘hand’, pl. rañkose, rãtuose ‘wheels’, turɶguose ‘markets’, where Saussure’s law did not operate. Note that the sg. form ratè reflects *ràtęɵ, not **ratajęɵ < *ratai’en, and must therefore have been built on the analogy of *tamęɵ. The earlier view “dass butè aus *butẹɷ + *en (*ę) entstanden ist” (Stang 1966: 183, followed by K014: 49) must be rejected because the monophthongization of *oi to *ẹɷ was limited to stressed syllables (cf. K025: 323). ALL’S WELL THAT ENDS WELL A few years ago, Jasanoff adopted the central tenet of my accentological theory, viz. that the Balto-Slavic acute was a stød or glottal stop, not a rising tone (cf. K014, K025, K218, Jasanoff 2004a). Of course, nobody will believe Jasanoff ’s claim that he arrived at the same result independently thirty years after I published it and ten years after we discussed it when he came to Leiden to visit us. Though at the time he haughtily dismissed “the tangle of secondary hypotheses and “laws” that clutter the ground in the field of Balto-Slavic accentology” (Jasanoff 2004b: 171), he has now recognized the importance of Pedersen’s law, Hirt’s law, Winter’s law, Meillet’s law, Dolobko’s law, Dybo’s law and Stang’s law and largely accepted my relative chronology of these accent laws, including the loss of the acute shortly before Stang’s law (cf. Jasanoff 2008). He has also accepted my split of Pedersen’s law into a Balto-Slavic and a Slavic phase (to which a Lithuanian phase must be added), my thesis that the tonal contours of Baltic and Slavic languages are post-Balto-Slavic innovations (cf. Jasanoff 2008: 34410), and the rise of a tonal distinction on non-acute initial syllables before Dybo’s law which I discussed at some length in my review (K028) of Garde’s monograph (1976). This is great progress. Though Jasanoff has come a long way in the last few years, he has not yet understood the origin of the Balto-Slavic glottalization, nor the origin of the Baltic and Slavic tonal contours, nor the origin of distinctive vowel length in Slavic. He has not yet understood the exact conditions of Hirt’s law, nor of Stang’s law, nor of the distribution of the o-stems over the accent classes. He evidently has not grasped the basic problem of Proto-Slavic quantity which is central to a correct understanding of the developments and their chronology. Perhaps it is only a matter of time before such insights get through to the IndoEuropeanist scholarly community. A major problem will be that much of the relevant literature is in Baltic and Slavic languages and therefore not easily accessible to scholars without at least a reading knowledge of these languages. Some news travels slowly. For the time being, Jasanoff ’s contribution to our knowledge of Baltic and Slavic accentuation is zero. He calls his recent article “programmatic” (Jasanoff 2008: 339 and 371), which appears to be newspeak for a shot in the dark without calculating the consequences. Following the example of Ebeling’s work (1967: 580, cf. Jasanoff 2008: 360), he offers an effort to reformulate Pedersen’s law and Dolobko’s law as a basic principle generating lateral mobility from stress on medial syllables. He proposes that Pedersen’s law “moved the accent one syllable to the left, producing a contrastive intonation on the newly accented syllable” whereas Dolobko’s law (in his jargon “Proto-VDL”) moved the accent to the 338 Balto-Slavic final syllable in sequences of four or more syllables when the initial syllable had such a contrastive accent (Jasanoff 2008: 349 and 367f.). There are three strategies to deal with counter-evidence in Jasanoff ’s methodology: (1) ignore it, (2) assume irregular analogical leveling, (3) propose additional specific rules for specific instances (cf. already K218). Thus, Jasanoff dismisses “late and productively formed [Lith.] stems in -ùmas, -ìnis, and the like” (p. 349), “a word like Lith. sūnùkas” and Slavic *vьdovà (p. 350), Slavic *vÈdNj for **vèdNj and *vedetьɳ for **vèdetь (here I substitute the usual accent marks for Jasanoff ’s idiosyncratic notation) but Lith. nèvedame, nèvedate for his expected final stress (p. 367), Slavic *prošNjɳ, *pròsitь for his **prΚNj, **prÎsitь (p. 369), similarly in the nasal presents (p. 371), and so on and so forth. He admits that it “is not clear, however, why non-mobile presents are as numerous as mobile presents” in the stative i-presents “or how the non-mobile forms came to be accented on the root syllable rather than the endings” (p. 372). He does not mention the word for ‘mother’, which escaped Meillet’s law, and arbitrarily assumes restoration of accentual mobility in the words for ‘son’ and ‘alive’, which escaped Hirt’s law (p. 353). He simply does not explain the data as we have them. Note that Jasanoff ’s adaptation of Pedersen’s law and Dolobko’s law is the exact opposite of Olander’s (2006), known to him at least from my publications but not mentioned by him, where Pedersen’s law is reformulated as loss of accent on a non-acute final syllable with rise of contrastive tone on the initial syllable and Dolobko’s law is reformulated as a part of Dybo’s law, which moved the accent one syllable to the right. The main problem with Jasanoff ’s reformulation of Pedersen’s law as a leftward accent shift is that we would expect a rising tone on the newly accented syllable, as in SCr. vòda ‘water’ < *vodà (cf. Jasanoff 2008: 348), whereas we actually find a falling tone as its Slavic reflex, e.g. in acc.sg. vÎdu. Jasanoff ’s solution to this problem is that he simply disregards the data, stating that no inference should “be drawn about the nature of the phonetic difference between the left-marginal [retracted] and in situ [unretracted] accents, other than that such a difference existed” (p. 351). The more unspecified distinctions one assumes, the more different forms one can “explain”. Jasanoff reconstructs a Proto-Balto-Slavic prosodic system with nine different possibilities (p. 350f.): short, long acute, and long non-acute syllabic nuclei combined with retracted, unretracted, and no accent. His use of the grave accent mark for the retracted accent is particularly unfortunate because the grave accent is the conventional symbol for a short rising tone in Slavic, where the retracted accent is reflected as a (short or long) falling tone. Jasanoff states that the acute became a rising tone in Slavic (p. 352) without explaining why it did not merge with the other (neoacute) rising tone. He states that in unstressed syllables “the glottal component of acuteness was lost without a trace” (p. 353) without explaining the rise of the Slovene neo-circumflex. All’s well that ends well 339 Jasanoff ’s treatment of the Balto-Slavic verb is so full of mistakes that it would be pointless to subject his account to a detailed critique. It is not true that extra-presential forms “tend (at least in Slavic) to derive their accentual properties from the present” (p. 35427). Jasanoff ignores the athematic origin of the i-flexion (p. 356, cf. K033, K087, K099). It is not true that “the overwhelming majority of athematic presents in Balto-Slavic are conspicuously non-mobile” (p. 358). It is not true that the verb ei- ‘go’ had an immobile present in Baltic, as is clear from Latvian (Varakļāni) 1st sg. èimu, 2nd sg. èi, 3rd ît (cf. K025: 327). The concept of “Narten” present is an outdated phantom (cf. de Vaan 2004). It is not true that the Slavic copula owes its oxytone forms to Dybo’s law (p. 359), as is clear from the long rising vowel in Čakavian and Posavian jẽ ‘is’ (e.g. Jurišić 1973: 24f.). Lith. nèvedu does not continue the Balto-Slavic place of the accent (as suggested on p. 363) because the stressed vowel is not lengthened. It is not true that the “word-final accent in Proto-Slavic was non-contrastively falling” (p. 36447) because it is rising in the languages which have preserved distinctive tone. It is not true that the Baltic verb *ded- ‘put’ had an immobile present (p. 37261). It is not true that aorists of Slavic verbs with mobile presents have “originally accented endings in the sigmatic forms” (p. 373). Jasanoff even goes so far as to invent his own “data” in order to support his ill-conceived proposal, positing a Slavic paradigm of the present participle with end-stressed masc. acc.sg. **vedNjtjьɳ and gen.sg. **vedNjtjà (p. 361) for which there is simply no evidence whatsoever. Contrary to Jasanoff ’s statement, Lithuanian does not have the accent “on the root syllable in the longer forms” but shows the regular accent patterns (1) and (3) with final stress in such forms as gen.pl. vedančiųɶ, loc.sg. vedančiamè, vedančiojè, loc.pl. vedančiuosè, vedančiosè, fem. nom.sg. vedantì, dat.pl. vedančióms, with recent transfer to accent class (1) in the standard language (e.g. Endzelynas 1957: 201ff., Zinkevičius 1981: 149). There is an older pattern in both East and West Baltic with nom.sg. *esints ‘being’, *eints ‘going’, other cases *sent-, *jent- (cf. K195: 71), corresponding to Latin iens, eunt- from Indo-European *eints, acc.sg. *ientm, gen.sg. *intos (cf. Beekes 1985: 70). It follows that Lith. ėdąɶs ‘eating’, duodąɶs ‘giving’ replace earlier *ẻdints, *dẻdints, adopting the suffixal accentuation of the stem form *dent-, *dỉdont-. The original accentuation of the masc. and fem. nom.sg. forms has been preserved in the Slavic gerund, e.g. Old Russian stója and stojačí ‘standing’ (cf. Stang 1957: 140). It will take some more time before Jasanoff will be in a position to make a contribution to the study of Balto-Slavic accentuation. The good news is that he has now understood the importance of at least some of the previous work in the field, even if he is reluctant to acknowledge his debt to earlier scholarship. BALTO-SLAVIC ACCENTUATION REVISITED There is every reason to welcome the revised edition (2009) of Thomas Olander’s dissertation (2006), which I have criticized elsewhere (K234). The book is very well written and the author has a broad command of the scholarly literature. I have not found any mistakes in Olander’s rendering of other people’s views. This makes the book especially useful as an introduction to the subject. It must be hoped that the easy access to a complex set of problems which this book offers will have a stimulating effect on the study of Balto-Slavic accentology. The purpose of the following observations is twofold. On the one hand, I intend to show that what the author evidently regards as his main result, the “mobility law”, cannot be accepted because it is incompatible with the data. On the other hand, my aim is to pinpoint the essential differences between Olander’s theory and mine (e.g. K066, K222, K249) in order to clarify where progress can be made. In the following, bracketed numbers which do not denote the Lithuanian accent classes (1) through (4) will refer to the pages of the book under discussion (Olander 2009). The origin of the mistaken analysis which has resulted in Olander’s “mobility law” must be sought in his reconstructions of Proto-Indo-European and Proto-Slavic. Following the German (Brugmannian, pre-structuralist) tradition, Olander reconstructs five short and five long vowel phonemes *i, *e, *a, *o, *u, *ī, *ē, *ā, *ō, *ū, of which *i and *u had non-syllabic variants which were “probably in complementary distribution” but are nevertheless distinguished in the reconstructions, four resonants *r, *l, *m, *n with “syllabic realisation between consonants” distinguished by a ring underneath, “four fricatives” *s, *h1, *h2, *h3, the latter of which had “vocalic variants” *ǩ1, *ǩ2, *ǩ3, three labial stops *p, *b, *bh, three dental stops *t, *d, *dh, three palatal stops *kʄ, *gʄ, *gʄh, three velar stops *k, *g, *gh, and three labiovelar stops *kw, *gw, *gwh (83), i.e. a total of 10 vowels, two of which had consonantal realizations, and 23 consonants, seven of which had “vocalic variants”. This large and complex phonological system, which allows an impressive number of 425 CV and 10625 CVC sequences, is clearly at variance with Olander’s professed “methodological choice to attach considerable weight to simplicity” of reconstructed synchronic systems (2009: 4). The larger the inventory of the input, the easier it is to “explain” almost any actually occurring word form. In my view, the great merit of the laryngeal theory is that it enables us to reduce the inventory of ProtoIndo-European phonemes in a substantial way. Olander recognizes that long vowels “had a very limited distribution in the proto-language; most long vowels in the Indo-European languages are the result of contraction of a short vowel with a following laryngeal” (83), which raises the question if they must be 342 Balto-Slavic reconstructed at all. The same holds for the “consonantal and vocalic realisations” of *i, *u, *r, *l, *m, *n and the laryngeals. In my view, syllabification developed in the separate branches of Indo-European, e.g. Latin sine and Tocharian B snai ‘without’ from PIE *snHi, where the laryngeal was consonantal in Latin and became syllabic in Tocharian. Interestingly, Olander’s definition of syllables is language-specific (13), and I agree: there is no such thing as a universal syllable. From a phonological point of view, the coloring of vowels by contiguous laryngeals also does not go back to the proto-language (cf. Lubotsky 1989, 1990). The velar stops developed in dialectal Indo-European times from depalatalization of the palatovelars and delabialization of the labiovelars of the proto-language (cf. Steensland 1973). Thus, I reconstruct two vowels *e and *o with lengthened variants *ē and *ō in monosyllables and before final resonants (cf. Wackernagel 1896: 66-68), six resonants *i, *u, *r, *l, *m, *n, three laryngeals *Ȥ, *ȥ, *ȥw, one fricative *s, and twelve stops. As I have argued elsewhere (e.g. K075), the “voiceless” and “voiced aspirated” stops were fortes and lenes, respectively, and the “plain voiced” stops were glottalized. While Olander’s reconstruction does not allow for a chronology of dialectal Indo-European developments, such a chronology is an essential part of my reconstructions. In contrast with the large inventory of Proto-Indo-European phonemes, Olander reconstructs a minimal system for Proto-Slavic: six stops *p, *b, *t, *d, *k, *g, three fricatives *s, *z, *x, four resonants *m, *n, *l, *r, four short vowels *a, *e, *i, *u and four long vowels *ā, *ē, *ī, *ū (127). The two semivowels *j and *w “were probably variants” of the vowels *i and *u and “did not have independent phonological status” (ibidem). Since Olander reconstructs neither *w nor *H for this stage, it remains unclear how he accounts for the difference between e.g. zvěrь ‘beast’ and zъvati ‘to call’. The smaller the inventory of the output, the easier it is to “explain” almost any word form thus reconstructed. Olander dismisses all instances of shortened “long” vowels, as in Serbo-Croatian jÇgoda and Czech jahoda ‘strawberry’, and all instances of lengthened “short” vowels, as in SCr. bÈrē ‘gathers’ and Czech vůle ‘will’, as recent developments and does not take them into consideration in his analysis. This greatly simplifies his task because accounting for the new quantitative distinctions such as between *a and *ā and between *o and *ō is by far the trickiest part of Slavic accentology. It also puts him on the wrong track because his “mobility law” is irreconcilable with the quantitative reflexes of vowels and diphthongs in Slavic. The same holds for his early date of Dybo’s law (ibidem), for which he offers no argumentation. Note that Olander’s reconstruction of “Proto-Slavic” corresponds to stage 6.0 of my chronology (e.g. K066: 47, K222: 119), which differs from his system in the presence of a glottal stop. The loss of the glottal stop gave rise to shortened “long” vowels (stages 7.13 and 9.2 of my chronology) and paved the way for the rise of lengthened “short” vowels (my stages 7.15, 8.1, 8.2 and 8.8). Thus, Olander omits a large and essential part of the evidence. Balto-Slavic accentuation revisited 343 There is a lot of misunderstanding resulting from terminology and definitions in Balto-Slavic accentology. It is therefore appropriate that Olander devotes an introductory section to these problems (7-14). Unfortunately, his choices are not always felicitous. His īs- and ūs-stems are actually iH- and uH-stems with a sigmatic nominative. He regards stems in semivowels and laryngeals as vowel stems, not consonant stems. His “desinence” refers to the complex of stem-forming suffix and case marker while his “ending” refers to the case marker only, which is the opposite of common practice. He defines ProtoIndo-European as “the language spoken at the end of the period that precedes the oldest innovation not shared by all (known) Indo-European languages”, which corresponds to stage 2.0 of my chronology, not to stage 3.0, which is the end of the dialectal Indo-European period preceding the earliest Balto-Slavic innovations. This is at variance with his reconstruction of the “Proto-IndoEuropean phonological system” (83), which represents a subphonemic diasystem between my stages 2.1 and 2.2 (see above). His Proto-Balto-Slavic and his Proto-Baltic correspond to my stage 5.0 (see also K253). His Proto-Slavic is defined as the stage immediately before the monophthongization of oral diphthongs, which took place at my stage 6.5 (e.g. K066: 48, K222: 119). In fact, his reconstructed “Proto-Slavic” phonological system (127) corresponds to my stage 6.0, before the umlaut (6.1) and the first palatalization of velars (6.2). This is long before the earliest Slavic dialectal developments, which arose around stage 7.0 (cf. K208: 231), and many centuries before the last common Slavic innovations and the disintegration of the common language (stage 10.0). Olander’s discussion of the prosodic terminology (10-14) is also less than satisfactory. He states that Štokavian “in a superficial analysis has contrasting tones, e.g. gen. sg. sèla with rising tone vs. nom.-acc. pl. sÈla with falling tone. In a somewhat deeper analysis, however, where rising tone is interpreted as accent on a following syllable, i.e. seɑla vs. ɑsela, Štokavian may be viewed as a non-tonal language”. In fact, the “somewhat deeper analysis” is simply wrong because noninitial falling tones are frequent in Štokavian and the normative system is artificial and probably never existed in any authentic dialect (cf. Vermeer 1985). Olander defines the term “accented” as referring to “the prominent syllable of a word in prosodic systems where no more than one syllable of a word is prominent relative to its neighbouring syllables. [...] Unaccented word-forms are found in languages like Vedic and Japanese [...]. Automatic, i.e. non-distinctive, prominence of a certain syllable in a phonological word” is referred to as “ictus”, which “in the case of unaccented words” falls on the initial syllable (11). This again is contrary to common practice, where ictus (‘beat’) refers to dynamic prominence and accent (=προσῳδία ‘pitch’) to tonal characteristics. It would be preferable to restate these definitions in terms of High and Low tone (where I use capital letters with reference to tone levels), making clear that the “automatic” Low tone on the initial syllable of an “unaccented word-form” is in 344 Balto-Slavic fact distinctively Low as opposed to a High tone on the initial syllable of an “accented” word form with initial stress. Olander’s identification of the Vedic and (Tokyo) Japanese systems as having “unaccented word-forms” is a source of confusion. Vedic has a full-fledged tone system with any sequence of High and Low tones, e.g. RV 1.1.6 távét tát satyám on one hand and 10.75.5 imáṃ me gaṅge yamune sarasvati śútudri on the other (cf. K080: 156). On the other hand, contrary to Olander’s presentation (12), Japanese hasi ‘edge’ and hasi ‘bridge’ are homophonous (Low-High) in the Tokyo dialect, where a following enclitic particle is High after ‘edge’ but Low after ‘bridge’, just as Russian kod ‘code’ and kot ‘tom-cat’ are homophonous but differ accentually in the genitive kóda versus kotá. While Vedic has distinctive tone, Tokyo Japanese has only lexical pitch accent. A final source of confusion is the term “circumflex” (14), which usually refers to a falling tone in Slavic and to the absence of an “acute” in Baltic. While I am happy to see that Olander has accepted my view that the Balto-Slavic acute can be identified as glottalization, I am sorry that he has not (yet) seen why glottalized vowels must have remained distinct from earlier long vowels in Slavic. As is clear from Olander’s Table 2 (46), the two pillars of modern Slavic accentology are Stang’s demonstration (1957) that Saussure’s law did not operate in Slavic and Dybo’s establishment of a progressive accent shift from non-acute non-falling vowels in flexion (1962) and derivation (1968). The views of authors who do not accept these two fundamental discoveries (including Klingenschmitt and Stankiewicz) are by now primarily of historical interest. The major question which remains, in Olander’s framework, is whether original accentual mobility was inherited from the Indo-European proto-language or developed in Balto-Slavic times. While original accentual mobility in consonant stems such as Greek θυγάτηρ, θυγατέρα, θυγατρός ‘daughter’ and Vedic ātm¡, ātm¡nam, tmánā, tmáne ‘soul’ can hardly be doubted, the absence of accentual mobility in original o-stems is equally certain. Olander challenges the classic view that vowel stems adopted the accentual mobility of the consonant stems (“Pedersen’s law”) and claims that the earlier accentual mobility which is reflected in alternating ablaut grades had almost wholly been eliminated in the Indo-European proto-language already. Here I disagree: nobody would maintain on the basis of the Welsh and Armenian evidence that the accent was fixed on the penultimate syllable in the proto-language, and the same holds true for the fixation of the stress in Vedic and Greek, where many traces of accentual mobility have been preserved. Indeed, it seems to me that the alternating ablaut grades provide a much more faithful piece of evidence than the attested place of the ictus, which does not go back more than 2500 years at most. Olander’s analysis in terms of syllables is less appropriate than an analysis in terms of tonebearing morphemes, where e.g. Greek θυγατέρα is stressed on the suffix and θυγατρός on the desinence. In my view, the alternation between stem-stressed Balto-Slavic accentuation revisited 345 and end-stressed forms can easily have been generalized from consonantal to vocalic stems. The argument that “the consonant stems constitute an unproductive group of words that are gradually disappearing in Baltic and Slavic” and that they cannot therefore “have influenced the large and productive group of vowel stems in such a profound manner as an imitation of the accent curves would imply” (51) is not valid because we are dealing with the earliest Balto-Slavic developments here, taking place at a stage when consonantal stems may still have been in the majority. Note that the ā-, ē-, ī-, ū-, i- and u-stems were also consonant stems at the outset. While I agree with Olander that analogical change requires a motivation and that a “typical example of a motivation would be the simplification of a complicated system” (50), I think that a morphological generalization is no less probable than the automatic fixation of the stress on a non-initial syllable. Thus, I stick to Pedersen’s law; for the oxytonesis see below. Olander states that the analogical changes which account for the lateral mobility in Slavic paradigms “constitute the backbone of Kortlandt’s theory of Balto-Slavic accentuation” (49). This is not correct. Starting from the assumption that the lateral mobility was not directly inherited from the IndoEuropean proto-language but could have developed by analogy with the consonant stems, this was not my major concern. The backbone of my theory is the thesis that the Balto-Slavic acute was a glottal stop which developed from the Indo-European laryngeals and from Winter’s law and is reflected as glottalization in Latvian and Lithuanian and that the gradual loss of this glottal stop accounts for the development of vocalic timbre and quantity distinctions in Slavic. The red thread which runs through these developments is a series of sound changes: Hirt’s law (4.1), Winter’s law (4.3), retraction of the stress from final open syllables (4.4), loss of the glottal stop in pretonic and post-posttonic syllables (5.3), loss of the glottal stop in the remaining posttonic syllables (7.13), Van Wijk’s law (7.15), contractions in posttonic syllables (8.1), retraction of the stress from final jers (8.2), Dybo’s law (8.7), lengthening of short falling vowels in monosyllables (8.8), loss of glottalization in stressed syllables (9.2), Stang’s law (9.3), shortening of long falling vowels (9.4), lengthening of short vowels and retractions of the stress in the daughter languages (10.4-10.12). These phonetic laws were followed by analogical levelings which account for the distribution of accent, timbre and quantity in the attested Slavic material. In this framework, the loss of Indo-European accentual mobility (3.1) was never intended to be more than a working hypothesis in order to simplify the analysis. As more traces of Indo-European accentual mobility in Balto-Slavic became clear to me (cf. K082, K087, K099, K173, K195, K201), I have finally abandoned this hypothesis (K236), see below. Olander makes a distinction between long and hiatal final syllables, the latter “containing two contiguous vowels, possibly separated by a laryngeal” (8). 346 Balto-Slavic In Greek, “sequences of two vowels behave alike whether separated by a laryngeal or not”, e.g. dat.sg. ἀγρῷ ‘field’ < *-oei and φυγῇ ‘flight’ < *-aHai (89). I agree that the Greek circumflex reflects a sequence of two vowels possibly separated by a laryngeal here and reconstruct *-oȤei and *-aȤai < *-eȥei, similarly dat.sg. οἴκῳ ‘house’ < *-oȤei, loc.sg. οἴκοι < *-oȤi but nom.pl. οἶκοι < *-oi like ἀγροί < *-oi, gen.pl. ἀγρῶν < *-oȤom, dat.pl. ἀγροῖς < *-oȤois but acc.pl. ἀγρούς < *-ons, βοῦς ‘o’x < *gwoȤus < *gweȥwus but Ζεύς without a laryngeal, optative 3rd sg. παιδεύοι ‘educate’ < *-oī < *-oiȤt, 3rd pl. τιθεῖεν ‘put’ < *-eȤiy- but aor. λύσειαν ‘loosen’ < *-eiy- (cf. K097). In Indo-Iranian, intervocalic laryngeals were lost at an early stage with contraction yielding long vowels but later restored at morpheme boundaries after the rise of new intervocalic laryngeals from the vocalization of the syllabic nasals, e.g. *maȤas ‘moon’ < *meȤns, *vaȤatas ‘wind’ < *ȥueȤntos, which was followed by the rise of the new gen.pl. ending *-aȤam (cf. K240). The same introduction of the formative suffix before the PIE ending *-om is found in Germanic, where the laryngeals were lost at an early stage, e.g. Gothic -o < *-āom, later again in Old High German -ōno, similarly in Sanskrit -ānām, -īnām, -ūnām, also Greek -³ων < *-āsōm and Latin -ārum, -ōrum. Olander’s dismissal of this explanation as “quite unnatural” (78) is quite incomprehensible to me. The original PIE ending *-om is attested in Slavic -ъ, Lith. -ų, Prussian -on, and in Germanic, Celtic, Italic, Indo-Iranian and Anatolian (cf. K030). I am happy to see that Olander has accepted (90) my derivation of the Lith. nom.pl. ending *-íe from *-aȤi < *-eȥ-i (without giving the argumentation for this reconstruction, cf. K110). I cannot accept Olander’s derivation of Slavic -i from *-oi and *-ai (90), which is at variance with loc.sg. *vьlcě ‘wolf ’, for which he arbitrarily assumes replacement by the ending of the ā-stems, and with the pronominal forms mьně, tebě, sebě and 1st sg. vědě ‘know’, which he does not discuss (cf. K055: 178). In my view, oral and nasal diphthongs were raised before final *-s (stage 5.9 of my chronology), e.g. imperative nesi ‘carry’ < *-oiȤs, inst.pl. raby ‘slaves’ < *-ōis, acc.pl. raby < *-ons, ženy ‘women’ < *-aȤns, as opposed to loc.sg. rabě < *-oi, dat.sg. rabu < *-ōi, Old Russian nesa ‘carrying’ < *-onts, soft ending -ja < *-jonts, but raising in acc.pl. koně ‘horses’ < *-jons, OCS konję, and also in nom.pl. *vьlci < *-oi-s. As for the circumflex tone in the Lith. optative tesukiẽ ‘turn’ < *-oiȤt, as opposed to the acute of Slavic *nesì < *-oiȤs, I assume that final *-Ȥt merged with preglottalized *-d < *-t (as in Latin quod and Old High German hwaz ‘what’) before the latter was lost in BaltoSlavic times (my stage 3.7). This development is comparable to the Indo-Iranian loss of a laryngeal before a preglottalized stop followed by another consonant which was established by Lubotsky (1981), e.g. in Vedic pajrás ‘firm’ < *peȥgʄros beside p¡jas ‘frame’ < *peȥgʄos. It has long been recognized that the traditional PIE voiceless and voiced aspirated stops could not co-occur in the same root, so that roots of the type *te(r)dh- and *dhe(r)t- are excluded. It follows that the distinction between fortes Balto-Slavic accentuation revisited 347 and “aspirates” was a prosodic feature of the root as a whole, which may be called “strong” (or “high”) if it contained e.g. *t and “weak” (or “low”) if it contained e.g. *dh. Dybo has shown (1968) that Baltic and Slavic morphemes can be divided into two prosodic classes, viz. “strong” (“high”, “dominant”) morphemes which attract the accent and “weak” (“low”, “recessive”) morphemes which repel the accent, and that the stress falls on the first strong morpheme of a word form (cf. Olander 2009: 33). This system can be explained in a straightforward way from an earlier system with distinctive High and Low tones. Lubotsky has shown that there is a highly peculiar correlation between IndoEuropean root structure and accentuation (1988a: 170), which again points to an earlier level tone system. I have proposed (K213) that we must rather start from “strong” and “weak” syllables which originated from an Indo-Uralic consonant gradation. In any case, the prosodic system reconstructed for Proto-IndoEuropean was very close to the system actually attested in Vedic Sanskrit. The question whether “each constituent morpheme of a Proto-Indo-European word had a surfacing distinctive high or low tone, which would render the accent redundant, or that the tones were distinctive only at a pre-stage of Proto-IndoEuropean, having become redundant in the proto-language because of the distinctive accent” (Olander 2009: 85) receives the same answer as in the case of Vedic, depending on one’s theoretical predilections (see above). Holger Pedersen made a distinction between “proterodynamic” and “hysterodynamic” accentual mobility (1926: 25, 1933a: 21), the former between root and suffix and the latter between suffix and desinence, and stated that these accent movements “n’ont joué aucun rôle pour le développement lituanien” (1933a: 22). The idea that there were two types of PIE accentual mobility was elaborated by Kuiper (1942) and other scholars, most consistently by Beekes (1985), who demonstrated that the nom.sg. form of the hysterodynamic paradigm was originally stressed on the root, not on the suffix, and that the proterodynamic and hysterodynamic paradigms have a common origin. Olander again follows the German (Hoffmannian) tradition, using “-kinetic” for “-dynamic” (92) and disregarding Beekes’ findings. He dismisses “the presence of ablaut alternations in the root of a word in one or more languages as an indication of paradigmatic mobility in that word in the proto-language” (93), which puts an end to the discussion. He accepts Rasmussen’s idiosyncratic view that the “distribution of *-e- and *-o- in the thematic suffix was not dependent on the accent but on a following segment, *-e- and *-o- appearing before an unvoiced and voiced segment respectively” (94). Disregarding Beekes’ argumentation (1995: 128f.), he ignores the evidence for paradigmatic mobility in the ā-stems (95) and dismisses the evidence in the i- and u-stems (96). He rejects the accentuation of Greek θυγάτηρ and μήτηρ as secondary (72f.) without mentioning Lycian kbatra ‘daughter’ < *dhuegȥtr (cf. Kloekhorst 2008: 904) and attaches no importance to the accentual mobility in Greek ὄργυια, 348 Balto-Slavic ὀργυιᾶς ‘fathom’ and Vedic pánthās, pánthām, pathás, pathíṣu ‘path’ (97). In my view, Vedic pātár- ‘protector’ and pitár- ‘father’ represent a single PIE paradigm with nom. *peȥtr, acc. *pȥterm, gen. *pȥtros, cf. Maltese missier ‘father’ < French monsieur < mon seigneur ‘my lord’. I shall not discuss the accentuation of verbal paradigms here. Since “the curves of the Proto-Baltic mobile paradigms are virtually reconstructed on the basis of Lithuanian material only” (101), Olander does not discuss Proto-Baltic. I have argued that Proto-Baltic and Proto-Balto-Slavic are the same thing because there are no common innovations of West and East Baltic which were not shared by Slavic (e.g. K025, K253). Olander does not discuss métatonie douce in Lithuanian (102) and does not even mention métatonie rude, nor metatony in Latvian. In my view, the rise of metatony coincides with the rise of syllabic tones in East Baltic (e.g. K025: 324-328, cf. also Derksen 1996). Olander regards the accentuation of secondary case forms in -n(a) and -p(i) as evidence for “unaccented word-forms” in Lithuanian, e.g. illative galvõn, galvósna ‘head(s)’, allative darbóp ‘work’ (103). I agree with Seržants’ view (2004) that the original accentuation of the illative was that of the accusative (cf. K221). The Latvian locative represents the illative, not the inessive (cf. Vanags 1994), and also has the accentuation of the accusative. The short vowel of Lith. nèveda ‘does not lead’ shows that the accentuation of this form is more recent than the lengthening in vẽda, which is limited to Lithuanian (cf. K025: 326), and cannot therefore go back to Proto-Baltic times. Nieminen’s law (105) represents two distinct developments: an East Baltic retraction from final *-à and a later retraction from final *-às in Lithuanian (cf. Derksen 1996: 96-128 and 229-232). The “analogical spread of accentual mobility at the expense of the immobile paradigms” in Lithuanian (107) is largely reversed in the modern language. The Old Lithuanian mobile accentuation in žinóti ‘to know’ (108) is supported by the Slavic evidence (cf. K082). Accentual mobility is also found in Lith. duodąɶs ‘giving’, Latvian duômu ‘I give’, sêžu ‘I sit’, èimu ‘I go’ (Varakļāni, cf. K025: 321, 323, 327), and Slavic athematic verbs. Olander has accepted (114f.) my view that original PIE long vowels are non-acute in Balto-Slavic in the case of Lith. akmuõ ‘stone’ and duktÄ ‘daughter’ but does not mention the evidence of Latvian âbuõls ‘apple’, SCr. žÈrāv, Czech žeráv ‘crane’ (e.g. K141: 26). He regards the Lith. 1st and 2nd sg. circumflex endings -aũ, -aĩ, -eĩ as “exceptions to Leskien’s Law” (115) without mentioning that they are acute in the Žemaitian dialects and regularly became circumflex in Aukštaitian. He rightly concludes that Saussure’s law was limited to Lithuanian (116f.) but does not mention the chronological argument (cf. K025: 327). Olander rejects “Kortlandt’s Law” (124f.) without explaining the accentuation of such forms as Old Prussian semmē, weddē, Lith. žẽmė ‘earth’, vẽdė ‘led’, also OPr. twaiā, twaiāsmu, swaiāsmu, tennā, tennēismu, tennēison, tennēimans, gennāmans, widdewū, widdewūmans. He adduces the absence of a macron in such words as OPr. deiws ‘god’, deinan ‘day’ Balto-Slavic accentuation revisited 349 as evidence for “unaccented word-forms” (125f.) without explaining the presence of a macron in e.g. mērgan ‘maid’, āntran ‘other’, āusins ‘ears’ (cf. K234: 363). Now we turn to Slavic. “If a word-form containing only syllables with low pitch, i.e. an unaccented word-form, was followed by an enclitic, the enclitic received an automatic ictus” (128). This is Dolobko’s law (my stage 7.2). “If there was no enclitic, the phonological word, i.e. the morphological word-form preceded by zero or more proclitics, received an automatic ictus on the initial syllable”, i.e. a distinctive Low tone (my stage 6.10). “The unaccented wordforms, which at later stages of Slavic often received initial accentuation, were realised differently from initially accented word-forms”, viz. by the distinction between Low and High tone. Apart from the unfortunate terminology (“unaccented” = Low tones only, “automatic ictus” = High tone on an enclitic but Low tone on an initial syllable, “initially accented” = High tone on an initial syllable), Olander and I seem to be in agreement here. The phrase “which at later stages of Slavic often received initial accentuation” is confusing because we already have “automatic ictus” and distinctive tone on the initial syllable at this stage, Low tone being regularly reflected as falling tone in Serbo-Croatian. Olander’s biggest mistake (cf. K234: 364) is his assumption that the distinction between glottalized and non-glottalized vowels had disappeared at this stage, the former yielding “long” vowels. As a result, he is unable to account for such quantitative distinctions as *a versus *ā in Slavic. Contrary to Olander’s statement (128), the Proto-Slavic prosodic system was not typologically similar either to that of Vedic (which had High and Low particles and syntactic conditioning of tone alternations) or to that of Tokyo Japanese (where accented Low tones surface as High tones). Greatly adding to the confusion, Olander now changes his terminology: “syllables are acute if they are accented and contain a long vowel, circumflex if they are unaccented or contain a short vowel” (129). Disregarding the instances where a long vowel lost the stress to the following syllable as a result of Dybo’s law, e.g. Czech bílý ‘white’, poutník ‘traveler’, tráva ‘grass’, trávní ‘grassy’, trávník ‘pasture’, národ ‘people’, zákon ‘law’, trouba ‘trumpet’, zábava ‘fun, party’, útroba ‘intestine’, as opposed to jazyk ‘tongue’, malina ‘raspberry’, chladný ‘cold’, těžký ‘heavy’, suchý ‘dry’, ruka ‘hand’, ruční ‘hand-’, ručník ‘towel’, sukno ‘cloth’, humno ‘threshing-floor’, where an original pretonic long vowel was shortened, Olander reconstructs long vowels in “Proto-Slavic” *dārgā ‘road’, *gālwā ‘head’, acc. *gālwān for the short vowels of SCr. drÇga, gláva but obl.pl. glàvama, adj. glàvnī, Czech hlava, hlavu, hlavní, Polish droga, głowa, głowę, Upper Sorbian dróha, hłowa, also long *ā in *genā ‘woman’ for SCr. žèna, dial. ženÇ, and in inst.pl. *genāmī, SCr. žènama, Slovene ženÍmi < *ženàmī (cf. K222: 124). Similarly, Olander does not account for the quantitative difference in the suffix between Czech pekař ‘baker’ and rybář ‘fisherman’ or between SCr. dvòrište ‘yard’ and blÇtīšte ‘mud-pit’ or Čakavian potēgnÏt ‘to pull’ and dvËgnūt ‘to lift’ (cf. K222: 129). 350 Balto-Slavic According to Šaxmatov’s law (1915: 84), medial syllables lost a falling tone to a preceding short vowel but not to a preceding long vowel, where the falling vowel was shortened instead, e.g. SCr. prÎdāli ‘sold’ < *prodÍli, náuka ‘science’ < *nā(ka. We now know that the latter formation received the medial stress as a result of Dybo’s law (cf. K222: 122). I have subsumed the former type under Pedersen’s law because it represents a retraction of the stress in mobile paradigms. Šaxmatov’s law must not be confused with Stang’s law, according to which the stress was retracted from long falling vowels after Dybo’s law (cf. K222: 123). Olander redefines Šaxmatov’s law as “an accent retraction from a word-initial syllable with falling tone to a proclitic or prefix” (130). He regards the retracted accent as “the result of the automatic ictus placement rule inherited from Proto-Balto-Slavic” (ibidem). The problem with this interpretation is that the retraction of the stress and the rise of distinctive tone (my stage 6.10) must have been more recent than the generalization of accentual mobility in the masculine o-stems which did not have an acute root vowel (“Illič-Svityč’s law”, stage 6.9), e.g. in SCr. z(b ‘tooth’, Greek γόμφος (cf. K222: 119), which was conditioned by the identity of the paradigms (b) and (c) except in the oblique plural case forms and by the absence of barytone forms with an acute root vowel in paradigms with mobile stress after Meillet’s law (stage 5.4), while original barytone neuters are continued as masculine o-stems of paradigm (b). Olander evidently does not appreciate the problem when he states that “substantial coincidence of the accent curves of two paradigms is not a necessary prerequisite for the transfer of lexemes among the paradigms” (145, cf. also K234: 364). Note that in Olander’s reconstruction only the gen.pl. and inst.pl. forms of the paradigms (b) and (c) have the same accentuation while the original (i.e. pre-Hirt) barytone masculines and barytone neuters only differed in the nom.pl. and acc.pl. forms. It appears that the alleged paradigm (d) reconstructed by some scholars (135) never existed (cf. Langston 2007 and K226: 231f.). Olander objects to my view of Meillet’s law as an analogical elimination of glottalization from the barytone forms of mobile paradigms (stage 5.4) after the phonetic loss of glottalization in pretonic syllables (stage 5.3), e.g. in SCr. acc.sg. glÍvu, Lith. gálvą, that such a development is unexpected in o-stem singularia tantum (131). He does not mention SCr. mÇti ‘mother’, which combines an acute root vowel with accentual mobility (cf. Jurišić 1973: 116) and thereby shows that Meillet’s law must have been an analogical development. As I pointed out earlier (K234: 363f.), the high frequency of pluralia tantum with derived singulars is characteristic of Balto-Slavic, e.g. Lith. mėsà ‘meat’, taukaĩ ‘fat’, Latvian mìesa, OPr. mensā, crauyo ‘blood’, Slavic męso ‘(piece of) meat’. Olander redefines Stang’s law (my stage 9.3) as a retraction of the accent “from short medial diphthongs, from reduced vowels in weak position, and from contracting syllables” (131). This is quite unsatisfactory, first of all because the retraction Balto-Slavic accentuation revisited 351 from weak jers yielded a different reflex and therefore belongs to a different chronological layer (e.g. stage 8.2 of K222: 122), secondly because “short medial diphthongs” include *-īna, *-īk-, *-īn-, from which the accent was not retracted to (b) roots (cf. Dybo 1968), and do not include e.g. loc.sg. -ū < *-ēu and inst.pl. -ȳ < *-ōis, from which the accent was retracted to (b) roots but not to (c) roots, thirdly because the accent was not retracted in such contracted forms as Čakavian kopÍ ‘digs’ < *kopàje, and fourthly because the accent was retracted from long vowels which originated from Van Wijk’s law, which Olander does not mention, e.g. SCr. p)šē ‘writes’, vÎlja ‘will’ (stage 7.15 of K222: 121). I am glad to see that Olander has accepted (132) my derivation of -ī- from athematic *-ei- in the Slavic i-presents, e.g. SCr. nÎsī ‘he carries’. I shall not discuss Olander’s reconstruction of “Proto-Slavic” paradigmatic accent in detail but limit myself to a few remarks. The derivation of Slavic -ъ from PIE *-os (135) cannot be upheld (cf. K055: 181f. and Vermeer 1991). Contrary to Olander’s statement (136f.), the OCS verbs jesmь ‘am’, damь ‘give’ and jamь ‘eat’ had final stress before Dybo’s law, as is clear from the long rising vowel in Čakavian (Vrgada) jẽ ‘is’, jesũ ‘(they) are’, dådũ ‘(they) give’ (cf. Jurišić 1973: 24, 42), (Hvar) jé, jesú, (Novi) jé, sú, dá, Posavian dādũ, Slovene dá, dadNjɴ, j¦, jedNjɴ, also v¦, vedNjɴ ‘know’. The verbs imamь ‘have’ and *zьnamь ‘know’ also had mobile stress (cf. K082). The sigmatic aorist had fixed stress on a non-acute long vowel, e.g. SCr. dònijeh ‘I brought’, inf. dònijeti with the long vowel of the aorist beside nèsti ‘to carry’, similarly (Dubrovnik) rËjet beside rèći ‘to say’ (cf. K234: 365). There is no evidence for any type of change in the aorist “by analogy with the infinitive” (138). Olander does not discuss the thematicization of athematic verbs which in my opinion gave rise to accentual mobility in the thematic presents (cf. K226: 229f.). As a result of Olander’s erroneous assumption that “after Dybo’s Law the glottalisation disappeared and long glottalised vowels merged with long non-glottalised vowels” (142), he is unable to explain the numerous formations where Dybo’s law introduced new long vowels in pretonic syllables. He was evidently misled by Rasmussen’s suggestion that the final accentuation of *notjь sьɳ ‘this night’ and *nesetè ‘you carry’ originated from Dybo’s law (142). As is clear from the final stress of Slovene gen.sg. lahkegà, dat.sg. lahkemù ‘light’ and from the long vowel of Slovak nesie < *nesetьɳ (cf. Stang 1952), the final accentuation is older than Dybo’s law here. The long rising vowel in Slovene inst.pl. kostmí ‘bones’, možmí ‘men’, stǩbrí ‘pillars’, Posavian sa sinoví ‘with sons’, and in Slovene loc.pl. mož¦h, 3rd pl. nesNjɴ < *nesNjtьɳ also shows that the final accentuation in these forms is older than Dybo’s law, which in these endings yielded a long falling vowel from which the accent was retracted in accordance with Stang’s law (cf. K222: 123, Stang 1957: 70f., 117). The quantitative difference between Slovene kònj ‘horse’ < *kònjь and gen.pl. gọɴr ‘mountains’ < *gorъɳ shows that Dybo’s law did not shift the accent onto final jers (cf. K222: 122). 352 Balto-Slavic I can also be brief about Olander’s reconstruction of Proto-Balto-Slavic, which is an intermediate stage between his “Proto-Indo-European” and his “Proto-Slavic”. His Proto-Balto-Slavic phonological system (144) differs from mine (stage 5.0 of my chronology) in the absence of a fifth short vowel *o, phonemic *j, *w and glottal stop *Ȥ, and in the presence of “unaccented wordforms”. My primary reason for rejecting the latter category (apart from the terminological confusion, see above) is the discrepancy between the Slavic accentual mobility between pre-phrasal and phrase-final syllables and the Lithuanian accentual mobility between pre-radical and word-final syllables, both of which originated after specific developments limited to the separate branches of Balto-Slavic (cf. already K028: 73-75). Olander ignores the fact that PIE lengthened grade vowels are never acute and (unlike the acute vowels) did not lose their length when the new timbre distinctions arose in Slavic (e.g. K064, Vermeer 1992). He does not discuss the dozens of instances which I have adduced but limits himself to two isolated examples “suggesting that plain long vowels merged prosodically with long vowels of laryngeal origin and long vowels from Winter’s Law” (148). The first example is Lith. žvėrìs (3), Slavic zvěrь (c) ‘beast’, which must be reconstructed as *źweȤr- (cf. Derksen 2008: 550). The second example is Lith. várna (1), SCr. vrÇna (a), Upper Sorbian wróna < *worȤnaȤ ‘crow’ (cf. Derksen 2008: 528), which Olander regards as a vṛddhi formation of Lith. varɶnas (4), SCr. vrÍn (c) ‘raven’ (147) though the word pair clearly represents an alteration of the corresponding forms in Latin cornīx, corvus and Greek κορώνη, κόραξ with substitution of *wor- for *kor-, both meaning ‘burn’, cf. also Lith. širɶvas ‘grey’, muÞvas ‘reddish’ beside Russian sérna (a) ‘roe deer’, Latvian m+Þns ‘black’, OPr. sirwis, Greek μέλᾱς (cf. K064: 121). Just for the record I mention one of the very few printing errors in the book under discussion: Lith. “nešąɴs” (154, 260) must be corrected to nešąɶs (183). Now we come to Olander’s “Mobility Law” (156): a High tone on a wordfinal mora became Low in Balto-Slavic. This is an improvement in comparison with his earlier formulation (2006: 133) but presupposes an ad hoc oxytonesis in VHV sequences, where the accent shifted from the first to the second mora while it remained on the first mora of long vowels and diphthongs (156). As a result of the mobility law, end-stressed word forms became “unaccented” and received an automatic Low tone “ictus”, either on the final syllable before an enclitic particle or (if there was none) on the first proclitic element or initial syllable of the phonological word (157). This is a peculiar compromise satisfying neither the Slavic mobility between pre-phrasal and phrase-final syllables nor the Lithuanian mobility between pre-radical and word-final syllables. The rule does not explain the accentuation of case forms in *-s, e.g. Lith. širdìs ‘heart’, lietùs ‘rain’, šaltàsis ‘the cold one’, arklỹs ‘horse’, gen.sg. galvõs, sirdiẽs, žvėriẽs, dešimtiẽs ‘ten’, lietaũs, dukterɶs, nom.pl. šìrdys, líetūs, dùkterys, nor langaĩ ‘windows’, inst.sg. širdimì, lietumì, gen.pl. langųɶ, galvųɶ, loc.pl. akisù ‘eyes’, šakosù Balto-Slavic accentuation revisited 353 ‘branches’, (Old) Russian gen.sg. plotí ‘flesh’, smertí ‘death’, desjatí (cf. Stang 1957: 87f.), inst.sg. krugóm ‘round’, včerá ‘yesterday’, dat.pl. détjam ‘children’, ljúdjam ‘people’, loc.pl. détjax, ljúdjax, Slovene gen.pl. ọɴvǩc ‘sheep’, dat.pl. možẹʄm, kostẹʄm, inst.pl. možmí, kostmí. Moreover, it does not account for the category of endstressed neuters established by Derksen (cf. K236: 77f.), e.g. Lith. aũkštas (2) ‘floor‘ with metatony, Czech vědro, SCr. vjèdro (b) ‘bucket’ with a shortened root vowel (cf. Derksen 2008: 518f.). One cannot escape the impression that Olander’s reconstruction of “Proto-Indo-European” endings is strongly influenced by the outcome of his mobility law: he accepts the Greek circumflex as evidence for a hiatal ending in dat.sg. -ῷ, -ῇ (173), inst.sg. -ῆ (175), where the accent was retracted in Balto-Slavic, but cannot use the hiatus in gen.sg. -ῆς (170), loc.sg. -οῖ (177), gen.pl. -ῶν (186), inst.pl. -οῖς (190); conversely, the absence of a hiatus is welcome in nom.sg. -ή (167) but not in acc.sg. -ήν (169), dual -ώ (179), nom.pl. -οί (181), acc.pl. -ούς (183), -³ς (184). I shall not discuss the separate case forms (166-194) because I have done that earlier (K234: 366-368). Let me only add that my reconstruction of dat.pl. *-mus is based on Old Lithuanian -mus, Slavic -mъ, u-infection in Old High German tagum and Old Norse dNjgom ‘days’ (cf. van Helten 1891: 460-462), and the zero reflex in Armenian (cf. K194: 49). My retraction of the stress from this ending to the initial syllable in the i- and u-stems at a stage (8.2) when pretonic jers in medial syllables could no longer receive the stress, e.g. in Russian détjam < *dětьmъɳ, is supported by Slovene dánǩs ‘today’ < *dьnьsьɳ and gen.pl. ọɴvǩc ‘sheep’ < *owьcьɳ. Note that the rise of final accentuation in the polysyllabic case forms of the i- and u-stems must have preceded Hirt’s law because accentual mobility was preserved in Slavic klětь ‘store-room’, kyjь ‘stick’, synъ ‘son’, darъ ‘gift’, stanъ ‘stand’, cf. Lith. kl°tis, kjis, sūnùs, all of which would have received root stress (1) if the accent had been fixed on the second syllable before Hirt’s law (cf. K234: 366). This already suffices to show that Olander’s mobility law cannot be maintained. Olander regrettably follows Andersen’s unfortunate suggestion to compare his mobility law with the rise of initial accentuation in the Podravian dialects of Croatia (159f.). In these dialects, which did not share the neo-Štokavian retraction of the stress, there is a long rising vowel in krãlj je došo ‘the king has come’ and a short stressed vowel in rūkÇ me boli ‘my hand aches’ (cf. Klaić 1936: 182). When a phrase ends in a syllable with a long rising or short vowel, the last word receives initial stress with a falling tone on a long vowel, e.g. došo je krÍlj, boli me rûka, where the accent of rûka stands for a falling tone followed by a trace of the original final stress: r(kÇ, similarly imperative pîši = p)šË for pīšË ‘write’, krâdi for krādË ‘steal’, pîsmo for pīsmÎ ‘letter’, also mÏškārÇc for muškārÇc je dΚo, ali cËgānka je kāzÇla ‘the man came but the gypsy woman said’ and svīrÇće tÇmburãš for tamburãš će svīrÇti ‘the mandolinist will play’, with the main stress on the initial syllable of the word. Klaić emphasizes the difference between gen.sg. sÈljâka for seljākÇ (b) ‘peasant’ and cËgānka (a) and between Ï Beničânce 354 Balto-Slavic for u BeničāncÈ (b) ‘to Beničanci’ and u ŠljËvošēvce (a) ‘to Šljivoševci’. It is clear that the initial accentuation did not arise from a phonetic retraction of the stress but developed as an autonomous word-initial boundary signal. Contrary to Olander’s statement (160), such forms with initial accentuation are not “phonologically unaccented” but doubly accented. Olander is evidently unaware of the existence of similar systems with double accentuation in Slovak and Polish dialects along the river Orava (cf. Topolińska 1961: 86-89). In the Karelian dialects of Russian, we find variation between original final stress and new initial accentuation (cf. Ter-Avanesova 1989: 218). In Polabian and in the Pannonian dialect of the Kiev Leaflets we find both retraction of the stress from final syllables and rise of initial accentuation, which are clearly independent developments (cf. K027, K107). None of these phenomena can be adduced as a typological parallel in support of Olander’s rise of “unaccented word-forms”. In his reply to my earlier criticism (K234), Olander discards objections which “are only valid from the point of view of [Kortlandt’s] own theory” because such criticism “is interesting insofar as it illustrates differences between our theories, but it does not bring to light weaknesses in my theory that need to be dealt with” (206). This will not do. Olander’s interpretation of Meillet’s law cannot be correct because the prosodic merger of acute and circumflex in Slavic was limited to pretonic and post-posttonic syllables while the distinction between acute and non-acute was preserved under the stress and in the first posttonic syllable, where it is reflected as short versus long in the historical languages. Similarly, Dybo’s law did not shift the accent to final jers because we find a short vowel in Slovene kònj ‘horse’ < *kònjь but a long vowel in gen.pl. gọɴr ‘mountains’ < *gorъɳ, and similarly in other languages. The point is that Olander’s theory simply does not account for the evidence. My chief objection is not the typological improbability of Olander’s mobility law but the fact that it is contradicted by the evidence. I agree with Olander that the lack of typological evidence is not crucial. As was pointed out above, contrary to Andersen’s mistaken analysis, it is simply not true that a final High tone was lost in Podravian or Karelian dialects of Slavic which developed an initial High tone under the influence of neighbouring languages. There is no typological parallel for Olander’s phonological (as opposed to a syntactic) rise of “unaccented wordforms”. In my earlier criticism I drew attention to the following words (K234: 361): (a) SCr. krÇva ‘cow’, Slovak krava, Polish krowa, Czech kráva, Upper Sorbian kruwa < krówa; (b) SCr. brázda ‘furrow’, Slovak brázda, Polish bruzda < brózda, Czech brázda, Upper Sorbian brózda; (c) SCr. bráda ‘beard’, Slovak brada, Polish broda, Czech brada, Upper Sorbian broda. Balto-Slavic accentuation revisited 355 It is clear that we have a distinction between acute (a), long (b) and short (c) vowels here, all of which are reflected as a in South Slavic and Czecho-Slovak and as o in Polish and Sorbian. The acute vowels were lengthened in Czech and Upper Sorbian (stage 10.6) after the rise of the new timbre distinctions (stage 7.13) while they remained short in Serbo-Croatian, Slovak and Polish. In SerboCroatian we find a long vowel in disyllabic forms of paradigm (c) but a short vowel in polysyllabic forms and in derivatives (cf. K222: 125-128). Olander obscures the evidence by removing the pretonic vowels from the discussion through substitution of the accusative for the nominative in my examples, asserting that I do not take “the role played by the position of the accent” into account (207f.). He reconstructs a long vowel for the short vowel in (a) and a short vowel for both the long vowel in (b) and the short vowel in (c). He relegates the discrepancy between the long vowel in (b) and the short vowel in (c) to a footnote, where he claims that “we do not avoid the assumption of some subsequent analogical levelling in the Slavic languages” and that in the reflexes of (c) “in most cases the short root vowel has been generalised throughout the paradigm” (208), ignoring the universal short vowel reflex in flexion and derivation except in disyllabic word forms in Serbo-Croatian paradigms, e.g. obl.pl. rùkama ‘hands’, rùčnī ‘hand-’, rùčnīk ‘towel’, gràdskī ‘urban’, hlàdnī ‘cold’, tèškī ‘heavy’, sùhī ‘dry’, Czech ruka, ruční, ručník, chladný, těžký, suchý (c), as opposed to SCr. tráva ‘grass’, trÍvnī ‘grassy’, trÍvnīk ‘pasture’, bËjelī ‘white’, p(tnīk ‘traveler’, Czech tráva, trávní, trávník, bílý, poutník (b). Pretonic long vowels in paradigm (a) are always shortened, e.g. SCr. jèzik ‘tongue’, màlina ‘raspberry’, Czech jazyk, malina, Polish język, similarly sędzia ‘judge’ < *sNjdьjà (< pre-Dybo *sNjdьɳja, cf. Russian sud'já with final stress) as opposed to wątroba ‘liver’, Czech útroba < *Ǎtròba, where the initial vowel lost the stress as a result of Dybo’s law. Following Holger Pedersen (1933a: 22), I started from the assumption that Proto-Indo-European accentual mobility had largely been eliminated at the beginning of the Balto-Slavic period (stage 3.1). This is not only because the loss of PIE accentual mobility also affected Vedic and Greek, but especially because Illič-Svityč (1963) did not distinguish between mobile and oxytone paradigms and because I wanted to avoid circular reasoning when directly comparing Balto-Slavic with Indo-European accentual mobility. Dropping the assumption that accentual mobility had been lost at an early stage, I reconsidered the BaltoSlavic accent laws against the background of an independent reconstruction of Proto-Indo-European accent patterns on the basis of the apophonic alternations in the most archaic attested paradigms (K236: 76f., cf. Beekes 1985: 150). This enabled me largely to remove the barytonesis (3.3) and the oxytonesis (3.4) from my chronology. It also opened the way to explain the origin of Dybo’s “dominant” suffixes on the basis of Derksen’s end-stressed paradigms (ibidem, 77-79) and thereby to reformulate Pedersen’s law (3.2) as a phonetic development, eliminating Stang’s counter-examples (1957: 12). This does not, 356 Balto-Slavic however, prove that Pedersen’s law was indeed a phonetic development, and I do no think that it was. When we look at accent retractions in South and West Slavic languages, we see that they are always part of a gradual process. In Bulgarian, the stress was retracted from a final short vowel to a preceding open syllable (cf. K052). In Serbo-Croatian, the stress was retracted earlier from a final than from a non-final syllable, earlier from an open than from a closed syllable, earlier from a short than from a long vowel, and earlier to a preceding long than to a preceding short vowel (cf. Ivić 1958: 105). In Slovene, the stress was retracted from a final short vowel to a preceding long vowel, and later also to a preceding short vowel (cf. K017: 6f., Greenberg 2000: 120, 143). In the Pannonian dialect of the Kiev Leaflets, the stress was retracted from a final open syllable (cf. K027). In Polabian, the stress was retracted from a short vowel in a final syllable (cf. K107). In Slovincian, the stress was retracted first from a final syllable to a preceding long vowel, then from a final syllable in polysyllabic word forms and analogically from medial syllables in paradigms with fixed stress, and later from a final short vowel in disyllabic word forms (cf. K028: 77). The fixation of the stress on the initial syllable in West Slavic languages first affected polysyllabic word forms in Polabian, Kashubian, Polish and Slovak dialects and the Pannonian dialect of the Kiev Leaflets, and end-stressed word forms in Podravian and in Karelian dialects of Russian. After the fixation of the stress on the initial syllable, it may look like this was the result of a single phonetic process, but this conclusion is clearly wrong. In view of the attested retractions of the stress in West and South Slavic languages, it seems to me that a retraction of the stress from medial syllables can more easily have been an analogical than a phonetic development and I therefore stick to the term “Pedersen’s law”. While retractions of the stress can often be described as phonetic developments, the fixation of the stress on the initial syllable requires the existence of a morphosyntactic unit with an initial syllable. This renders the distinction between sound law and analogy disputable (cf. Olander 2009: 21021). Olander finds it “difficult to see the motivation behind” (211) the Slavic extension of Pedersen’s law (which he calls “Šaxmatov’s Law”, 130). In my view, generalization of the Low tone of pretonic syllables to barytone forms of mobile accent paradigms gave rise to Olander’s “unaccented word-forms” with distinctive Low tone on the initial syllable (stage 6.10). This introduction of a distinctive Low tone is an essentially syntactic development with a perfect analogue in Vedic. It created the possibility of lexical clitics, e.g. Russian četýrnadcat' ‘fourteen’, (byliny) belý grudi ‘white breasts’, Slovincian jáu robją ‘I work’, Bulgarian Čérno more ‘Black Sea’ (cf. already K028: 74), also Slovene gen.sg. lahkegà, dat.sg. lahkemù ‘light’ (Dolobko’s law, stage 7.2), where the final stress marks the end of the “phonological word”, as Olander calls it. The Low tone had a falling contour after a preceding High tone, as a result of which the High tone received a rising contour after a preceding Low tone. At a later stage Balto-Slavic accentuation revisited 357 (8.7), the rising contour shifted the High tone to the following syllable (Dybo’s law). Suppression of the contour could probably be used for a contrastive interpretation, as in modern Serbo-Croatian od brÇta (Low-High-Low) ‘from the brôther’ as opposed to regular òd brata (rising-falling) ‘from the brother’ (cf. Ivić & Lehiste 1967: 75f.). In Slovene, the distinctive Low tone became High when the falling contour shifted to the right while the High tone became Low, e.g. in kọʄst (High) ‘bone’ versus pNjɴt (Low) ‘way’ (cf. Greenberg 2007: 77 and Pronk 2009: 20). The rise of a distinctive Low tone was not an automatic consequence of the retraction of the stress to a preposition or prefix, as is clear e.g. from Russian ottúda ‘from there’, donél'zja ‘as can be’, cf. tudá, nel'zjá, also SCr. nà vrāta beside vráta (b) ‘door’, all of which received non-initial stress as a result of Dybo’s law, similarly in the verb nàlomīm, slÎmīm beside lòmīm (c) ‘I break’. The latter accentuation recalls the Vedic loss of accent on finite verb forms in main clauses, e.g. ¡ gamat ‘may he come’. As to Olander’s rhetorical “final remark” on my “methodological approach” (212), it must be regretted that he does not draw the reader’s attention to the fact that his “Proto-Slavic” phonological system (127) differs from my Early Slavic system (6.0) in the absence of a glottal stop which accounts for the rise and development of the new timbre and quantity distinctions, which he does not discuss. He also does not discuss the rise and development of the nasal vowels, nor of *j, which have considerable impact on the reconstruction of the phonological system. I am sorry that he has not replaced his confusing analysis in terms of “accent” and “ictus” by a consistent treatment in terms of High and Low tone, which could have avoided a number of misunderstandings. Had he done these things, he would probably have found very few substantial differences between his analysis and mine. But it is difficult to see how his mobility law could have survived the evidence of Slavic vowel length. As it is, Olander’s free choice between “Proto-Indo-European” alternatives as input and his limitation of the Slavic evidence by excluding the new timbre and quantity distinctions as output enable him to explain almost any reconstructed word form, either as a regular phonetic reflex or as the result of influence from alternating forms. This is in stark contrast with my aim to explain the actual distribution of accent, tone and quantity as they are attested in the Slavic languages. LITHUANIAN žinóti ‘TO KNOW’ Elsewhere I have argued that an apophonic difference between singular and plural forms of present tense suffixes such as *-ei/i-, *-ā/i-, *-nā/n(a)- was quite common in Balto-Slavic times (K087, K099, also K263: 151-179 and 275-296), e.g. Prussian 1st sg. posinna (4×) ‘bekenne’ < *-zinā, 1st pl. posinnimai < *-zini(n)ma-, 3rd pl. posinna < *-zin(n)a. I have identified this flexion type with Lith. žìno and Vedic jān¡ti ‘knows’, Latvian zinim beside zinãm ‘we know’, Tocharian A knānat ‘you know’, and Slavic *zьnāmь (K082). The Slavic verb had mobile stress (c), as is clear from Serbo-Croatian (Dubrovnik) nÈ znām, ne známo, pÎznām, poznámo, (Sarajevo) dÇ znāš, nÈ znāš, (Posavian) nÈ znām, pÎznām, Slovene poznÍm, also OLith. (Daukša) żîno, żinomê, żinotê. The Slavic verb znati and its derived noun znamę ‘sign’, which are based on the root aorist *gʄneH3-, have fixed stress (a), as is clear from SCr. znÇti, znÇmēn (e.g. Derksen 2008: 546). The initial palatovelar was evidently restored on the basis of *zьnāmь in these words because the phonetic reflex of the root aorist would be *gna- (cf. K263: 47), which would have merged with gna- < *gъna-, SCr. gnÇti ‘chase’, OPr. guntwei. It follows from both the mobile accentuation and the preservation of the initial palatovelar that SCr. znÍm represents *zьnāmь and cannot be derived from the root aorist or from the perfect *-gʄnōu, Vedic jajñáu, which is found in SCr. poznávati with restored palatovelar and long -āreflecting the lengthened grade vowel. Miguel Villanueva Svensson has raised two objections to the derivation of Lith. žinóti from a nasal present (2008: 176-181). He points out correctly that the vowel -ó- points to *-eH2-, which is at variance with a reconstruction *gʄn-neH3-. However, OLith. (Daukša) żîno, żinomê, żinotê shows that the present tense had lateral stress and, consequently, that the -o- was unstressed and may therefore represent either *-ā- or *-ō- (e.g. K263: 6, 46). Since the rise of lateral mobility in Balto-Slavic accent paradigms preceded the East Baltic merger of *-ā- and *-ōin unstressed syllables, the Lithuanian present tense directly continues the nasal present *gʄn-neH3- with analogical loss of the acute in the tense suffix. The acute in the infinitive was evidently taken from the preterit suffix *-eH2 before the latter lost its acute on the analogy of the preterit in *-ē (cf. K263: 187). The other objection put forward by Villanueva regards the Latvian forms 1st pl. zinim, 2nd pl. zinit beside zinãm, zinãt, which are difficult to explain on the basis of the reconstructions *žinmē, *žintē, allegedly from *gʄn-nH3- before consonant. It follows that the reconstructed development is incorrect. The solution to this problem is provided by the Prussian forms -sinnimai, -sinnati, which Villanueva does not explain (cf. K263: 287-296). While 2nd pl. -sinnati can easily have replaced *zinte < *zinnte on the analogy of 3rd pl. -sinna < *zina 360 Balto-Slavic < *zinna, 1st pl. -sinnimai evidently reflects *zinima < *zininma (cf. K263: 280). We must conclude that the phonetic development of *-nnHm-, with three nasal resonants in succession, differed from that of *-nnHt-, where the nasal geminate was simplified. An imperfect parallel is offered by Greek ἐξελαύνοια ‘I may drive out’ < *-oyyṃ < *-oiH1m versus κελεύοι ‘he may order’ < *-oï < *-oiH1t (cf. K097: 237). Villanueva’s suggestion that “-sinnat built a thematic present in Old Prussian” (2008: 175) is clearly mistaken in view of the regular 1st pl. thematic endings -ammai, -emmai, as opposed to -imai in je-presents and athematic formations (cf. K087). Thus, the derivation of Lith. žinóti from a nasal present is straightforward if the Prussian evidence is taken seriously. Villanueva’s own proposal is to derive žinóti from the weak perfect stem form *žini-, to which the preterit suffix *-eH2 was added (2008: 194). He disregards both the mobile stress of żîno, żinomê, żinotê and the Slavic formations. The accentuation of SCr. pòznati, poznávati, pÎznām points to fixed stress in the aorist (a) and the perfect (b) and mobile stress in the present tense (c). It follows that in this verb the apophonic alternation between singular and plural forms had already been eliminated in the aorist and the perfect, but not in the present tense, before the characteristic system of accent paradigms was established in early Balto-Slavic times (e.g. K263: 43). It is therefore highly unlikely that the stem form *žin- originated in the aorist or the perfect. Moreover, it is unclear how the addition of the preterit suffix *-eH2 could yield a present tense. Anyway, the addition of *-eH2 instead of *-eH1, as in dėv°ti ‘to wear’ and stov°ti ‘to stand’, is quite unexpected and unmotivated. I conclude that Villanueva’s proposal does not solve the problems which he raised himself. I find no evidence for Babik’s reconstruction of a thematic present *žineti ‘makes acquaintance of ’ (2004: 79) beside Lith. pažsta, Proto-Indo-European *gʄnH3ske/o- (with depalatalized *k, cf. Lubotsky 2001, Villanueva 2009). MORE ON THE CHRONOLOGY OF CELTIC SOUND CHANGES Graham Isaac’s recent monograph (2007) deals with the chronology of Celtic sound changes. Remarkably, the author completely disregards the relative chronology which I published 28 years earlier (K035). In the following I shall discuss the main issues on which our views differ. 1. Following Thurneysen (1946: 435), Isaac derives Old Irish -ír ‘granted’ from *peper- or *pepor- (p. 15). I have argued that *p developed into a bilabial fricative [φ] which was preserved into the separate languages (K051: 74-76 = K239: 53-55). It merged with *w after *s in Irish and coalesced with a preceding *s into *f in British, e.g. OIr. seir ‘heel’, dual di pherid, Welsh ffer ‘ankle’. The loss of the labial articulation was obviously posterior to the lenition, as it was in OIr. secht ‘seven’, necht ‘niece’, lassar ‘flame’, Welsh saith, nith, llachar. The voiceless bilabial fricative [φ] merged with its voiced counterpart [β] in OIr. -ebra ‘he will bestow’ < *piprā- and -ebla ‘he will drive’ < *piplā- (cf. Thurneysen 1946: 403). Before *n, the reflex of *p merged with *u after *o and *a but was lost after *e, e.g. súan ‘sleep’ < *sopno-, cúan ‘harbour’ < *kapno-, tene ‘fire’ < *tepnet-. These developments were more recent than the merger of *eu with *au and *ou into *ou, which was more recent than the loss of intervocalic *s (cf. Greene 1976: 27) but earlier than the monophthongization of the latter into *ō2, so that the merger of *p with *u before *n can be dated between stages 2 and 3 of my chronology (K035: 39). Intervocalic *p was lost, e.g. saer ‘artificer’ < *sapero-, té ‘hot’ < *tepent-, ni(a)e ‘nephew’ < *nepot-, íar ‘after’ < *epi-. Since *epi- shared the monophthongization of *ei to *ē2, the loss of intervocalic *p can be dated before stage 3 of my chronology (K035: 40). In other positions, the reflexes [h] < *s and [φ] < *p were preserved up to a later stage, e.g. in lenited ph- < *sp-. As I pointed out earlier (K051: 75), OIr. timme ‘heat’ represents *tepsmiā, not *tepesmiā. The sequence *epe merged with *ese, yielding *ē3 (cf. K035: 41), not *ī. I have dated the hiatal raising of *e to *i before back vowels, e.g. in ni(a)e ‘nephew’, gen.sg. niath, niad < *nepotos, siur ‘sister’, acc.sg. sieir < *swesoram, dual sieir < *swesore, between stages 3 and 6 of my chronology (K239: 140), after the lenition (1) but before the (first) palatalization (7), the raising of *e and *o before a high vowel in the following syllable (8), and the lowering of *i and *u before a non-high vowel in the following syllable (11). Isaac mistakenly posits a raising of unstressed *e before *s anterior to the lenition in 2nd sg. biri ‘carry’ < *berisi < *beresi (p. 16), ignoring the original endings of the Indo-European thematic inflection (cf. K035, K068, K150 and K239 passim). He mistakenly limits the raising of *e to *i to the stressed syllable (cf. Schrijver 1995: 387 and K239: 140). Following McCone, he mistakenly dates the shortening of long vowels in medial syllables (my stage 10) before the raising (my stage 8) in his 362 Italo-Celtic discussion of niad < *nepotos and cucann ‘kitchen’, Welsh cegin < Latin coquīna (p. 19, cf. K167 = K239: 117-120). 2. While Isaac dates the loss of *p and the raising of *ē1 to *ī to Proto-Celtic (p. 21), I have dated these developments to early stages of Proto-Irish postdating the lenition (K035: 40). In another study (K046: 1-16 = K239: 25-44) I discussed the Italo-Celtic shortening of pretonic long vowels discovered by Dybo (1961) and its interpretation by Illič-Svityč (1962). I have claimed that the Italo-Celtic shortening affected pretonic *ē, *ā, *ō but not sequences of resonant plus laryngeal, where the long vowel had not yet originated (K046: 13f. = K239: 40f.), and that the long vowel never originated in pretonic sequences of laryngeal plus resonant such as *gwHi- ‘live’ and *bhHu- ‘be’ (K046: 15f. = K239: 43f.). The discussion was continued by Schrijver (1991: 334-357, 512-536 and 1995: 168-191). Isaac rejects the derivation of OIr. sith- ‘long’, Welsh hyd ‘length’ < *sH1i- beside OIr. sír, Welsh hir < *seH1- (p. 24) in spite of such obvious parallels as PIE *dheH1(i)- ‘to suck’, *leH1(i)- ‘to pour’, *peH3(i)- ‘to drink’. He subscribes to the criticism of pretonic shortening by Irslinger and Rasmussen (p. 25). In fact, Irslinger misrepresents the issue by substituting *RH for *HR and by rejecting a metathesis *CeHU > *CeUH which nobody ever proposed (2002: 28). Rasmussen’s complaint that my view “is at variance with everything we know about IE syllabification” (1999: 1701) should be remedied by a conscientious study of Werner Winter’s pertinent article (1965) and by abandonment of his preconceived ideas about Indo-European, followed by a discussion of my arguments. Isaac’s dismissal of my reconstruction *HR in cases of shortening as circular is mistaken because this reconstruction is supported by full grade forms with *VHR. His peculiar comparison of Latvian dzîvs, Lith. gývas ‘living’ with Latvian plâns, Lith. plónas ‘thin’ obscures the issue because the latter word has a full grade vowel *VH. Eventually he devises a highly unnatural, complex and unconvincing phonetic rule on the basis of a host of additional assumptions in order to explain the data without recourse to analogy. It is a typical example of paper phonetics. Isaac’s proposal to derive the short vowel of Latin vir ‘man’ from nom.sg. *vĭrz < *vīrz < *vīros (p. 57f.) does not explain the short vowel of the Old Irish cognate fer. 3. Isaac presents a relative chronology of 25 sound changes from Proto-IndoEuropean to Celtic (pp. 62-64). He mistakenly thinks that the circumflex of Lith. taũras ‘aurochs’ is compatible with a reconstruction **taHuros (p. 65), cf. acc.sg. díeverį ‘brother-in-law’ < *daHiuerm, píemenį ‘shepherd’ < *poHimenm. The word is an early European borrowing from Semitic. I cannot accept all stages of Isaac’s relative chronology (pp. 69-74). I have dated the split of *ō1 into *ā and *ū (Isaac 22) before the rise of *ē3 (Isaac 21) and the raising of *ē1 to *ī (Isaac 17) between these two developments (K035: 39-41 = K239: 6-9). I think that the loss of the laryngeals (Isaac 7 and 18) can largely be dated to the Italo-Celtic period. More on the chronology of Celtic sound changes 363 Isaac’s anaptyctic shwa (5, 13a, 20) is a heterogeneous phenomenon which can partly be dated to the Italo-Celtic period (cf. K239: 88). There was no phonetic reduction of *ye to *i between consonants (Isaac 19, cf. K239: 137). The development of *p (Isaac 12a, 14, 15, 16) was discussed above. Isaac’s earlier rules are Proto-Celtic (4, 6, 10, 11), Italo-Celtic (5, 7, 8, 9), dialectal Indo-European (3), or mistaken (2, cf. Puhvel 1987). I am sorry that Isaac has found it unnecessary to discuss the arguments for the chronology which I put forward earlier. 4. Isaac reconstructs *gʄhdies > *gʄhdhyes > *ghdhes > Greek χθές ‘yesterday’, with loss of the dental stop in Sanskrit hyáḥ, Latin herī and German gestern but loss of the palatal stop in Welsh doe and Albanian dje (p. 75). He assumes metathesis in Greek *dhgʄhomios > *ghdhonyos > χθόνιος ‘of the earth’ and its cognate OIr. duine, Welsh dyn ‘man’, Gaulish gen.pl. -XTONION (p. 78), but not in Sanskrit kṣám- ‘earth’, Gothic guma ‘man’, Latin humus, homō, Albanian dhe (p. 81), to which Lith. žẽmė, Slavic zemlja, Thracian Σεμέλη and Phrygian ζέμελεν can be added. It seems to me that none of these examples supports the hypothesis of an early dialectal Indo-European development or later language contact. All forms can be derived from *tkʄ-, *dhgʄh-, with metathesis in Proto-Greek and ProtoCeltic. I also reject the derivation of Latin sitis ‘thirst’, situs ‘mould’ < *dhgwhiti/uand situs ‘located, site’ < *tkʄito/u- (p. 79). The former words may be related to Sanskrit jásate ‘be exhausted’, Greek σβέννῡμι ‘extinguish’ and the latter to Latin sileō ‘be silent’ (cf. de Vaan 2008: 563f.). No conclusions can be based on the word ursus ‘bear’. Following Melchert (1994: 251f.), Isaac assumes three series of velar stops in Anatolian and Proto-Indo-European (p. 83). I have argued that the plain velar series developed from depalatalization of the palatovelars and delabialization of the labiovelars (e.g. K263: 27-32, cf. already Meillet 1894 and Steensland 1973). Melchert lists three examples of plain velar *k in Luwian, viz. karš- ‘cut’ < *krs(cf. Kloekhorst 2008: 455), kattawatnalli- ‘plaintiff ’ < *kH2et- (cf. Kloekhorst 2008: 466), and kiš- ‘comb’ < *ks- (cf. Kloekhorst 2008: 481f.), all of them with depalatalization of an original palatovelar before the following consonant. Contrary to Isaac’s “neutralisation of distinctive aspiration in the voiced occlusives” during a “period of contact between Celtic, Balto-Slavic and IndoIranian around 2,000 BC” (p. 90), I assume loss of PIE glottalization in ProtoCeltic and its preservation in Indo-Iranian, Balto-Slavic, Germanic, Italic, Greek and Armenian (e.g. K075 and K239: 149-151). In my view, Italo-Celtic was the first branch of Indo-European which separated from the proto-language after Anatolian and Tocharian and did not therefore participate in the more recent innovations of the central dialects such as the extended development of the subjunctive, the optative and the middle voice (cf. K239: 151-157). Isaac’s view that the sigmatic formations “are patent innovations of the late Proto-IndoEuropean period” (p. 93) is surely mistaken. Contrary to his statement, the 364 Italo-Celtic relative pronoun *yos was not a common innovation of Indo-Iranian, BaltoSlavic, Graeco-Phrygian and Celtic but was simply replaced by *kwo- in Germanic and Italic, just as it recently was in most Slavic languages. Celtic never was a central Indo-European language. 5. Isaac’s final chapter deals with palatalization in Irish, which I have discussed earlier (K035: 41-48 = K239: 9-17 and K167 = K239: 117-120). He starts from McCone’s treatment (1996) without taking my criticism into consideration. Under these circumstances it seems pointless to repeat what I have said earlier about the mistakes in McCone’s account and I simply refer to my earlier work. Just for the record I only mention that McCone’s first, second and third palatalizations correspond roughly to my stages 7, 12 and 18 and the labialization of *i to *u in cruth ‘shape’ and gen.sg. cruimthir < QRIMITIR ‘priest’ to my stages 9 and 16, respectively. Eventually Isaac arrives at a chronology which is very close to mine (p. 102): first palatalization (7), lowering (11), second palatalization (12), apocope (15), labialization in cruimthir (16), third palatalization (18), syncope (19). Isaac’s effort to conflate “the various palatalisations of Proto-Irish” (p. 104), which is evidently meant to remedy the inconsistencies in McCone’s account, does not contribute to a better understanding of the chronological processes. There can be little doubt that the first palatalization affected initial consonants but not *kwr- and *gw- (cf. K239: 119f.), just as the intervening *w blocked palatalization of the velar obstruent by the following front vowel in Czech květ ‘flower’, hvězda ‘star’, unlike Russian cvet, zvezdá. The lowering of *e in Old Irish daig ‘flame’, dat.sg. taig ‘house’, laigid ‘lies’, but not in gen.sg. and nom.pl. tige, verbal noun lige, is a result of palatal dissimilation and must be dated after the general raising and lowering (cf. K239: 141). I conclude that Isaac’s discussion has given me no reason to change my mind on any of the issues involved. INITIAL LARYNGEALS IN ANATOLIAN Elsewhere I have argued that initial *H2- and *H3- yielded h- before *-e- and zero before *-o- in Armenian and Albanian and suggested that the same development may be established for Hittite, e.g. harp- ‘separate’ < *H3erbh- versus ark- ‘mount’ < *H3orgʄh-ey-, Gr. ὀρφανός, ὄρχις (cf. K073: 42). The new monographs by Kimball (1999) and Rieken (1999) have strengthened my view that this is indeed correct. In his classic study of *H3 in Anatolian, Melchert lists seven examples of ha< *H3e- (1987: 21): (1) (2) (3) (4) (5) (6) (7) haran- ‘eagle’ < *H3eron-, Gr. ὄρνις; harp- ‘change one’s group’ < *H3erbh-, Latin orbus; happar ‘transaction’, happinant- ‘rich’ < *H3ep-, Latin opus; haštāi ‘bone(s)’ < *H3ešt(H2)ōi, Gr. ὀστέον; hark- ‘perish’ < *H3erg-, Old Irish orgaid ‘slays’; hawi- ‘sheep’ < *H3ewi-, Latin ovis; haliya- ‘bow’, halhaltumar ‘corner’ < *H3el-, Gr. ὠλένη ‘elbow’. As Melchert points out, proponents of the view that *H3- was lost in Anatolian assume ha- < *H2o- in these words, which implies that the absence of nonAnatolian a- < *H2e- in these roots must be ascribed to chance. This is clearly a circular argument. Moreover, initial *H3- is reflected as h(a)- before a consonant in Hittite harg(a)nāu ‘palm, sole’ < *H3rgʄ-, Gr. ὀρέγω ‘stretch out’ (and allegedly hapuš- ‘shaft, penis’ < *H3pus-, Gr. ὀπυίω ‘marry’, which seems an improbable etymology to me), as Melchert points out, concluding that *H3-, like *H2-, is generally preserved as h-. In his more recent monograph, Melchert silently abandons *H3- for *H2- in ‘bone’ and ‘sheep’, to my mind for no good reason, and stealthily adds hanna- ‘litigate’ < *H3enH2o- (1994: 145, 235). In his article, Melchert lists three examples where initial *H3- may have been lost in Hittite (1987: 20): (1) arta ‘stands’ < *H3erto, Gr. ὦρτο; (2) arki- ‘testicle’ < *H3ergʄhi-, Gr. ὄρχις; (3) aniya- ‘carry out, execute’ < *H3en-, Latin onus. As he points out, these instances might reflect *H1o- if the necessary o-grade could be justified. In his book, Melchert posits *H1e- with lowering of *e- to abefore the following resonant in these forms (1994: 85, 137). This is clearly unsatisfactory in view of the non-Anatolian evidence. Moreover, he follows Kimball’s suggestion (1987) that initial *H3-, unlike *H2-, was lost in Lycian (on which see below). 366 Anatolian In her monograph, Kimball adduces three alleged instances of Hittite ha- < *H2o- (1999: 142): (1) hāwi- ‘sheep’, Hom. ὄϊς, Latin ovis; (2) haštāi ‘bone(s)’, Gr. ὀστέον; (3) hašduēr ‘twigs, brush’, Gr. ὄζος. Since these are classic examples of non-apophonic o- in Indo-European (e.g. Beekes 1969: 130f., 139f.), I reconstruct *H3e- here. Kimball lists six possible examples of ha- < *H3e- (1999: 393f.): (1) hāppar ‘price deal’, hapzi ‘is rich’, hāpperiya- ‘city, settled place’, Latin opus; (2) hāraš, hāran- ‘eagle’, Gr. ὄρνις; (3) hāriya- ‘bury’, Gr. ὀρύσσω ‘dig’; (4) hāri- ‘valley’, Hom. οὖρος ‘boundary’ (which seems an improbable etymology to me); (5) happēna- ‘baking kiln’, Gr. ὀπτός ‘roasted’; (6) hāliya- ‘kneel’, Gr. ὠλένη ‘elbow’. Kimball rejects the etymological connection of aniya- ‘work’ with Latin onus ‘burden’ and assumes ā- < *H1o- in ārra- ‘arse’ < *H1orso-, Gr. ὄρρος, and in ār‘arrive’, Gr. ὄρωρα, versus ar- < *H1r- in arki- ‘testicle’ and ar- ‘stand, place oneself ’ (1999: 387, 389). I would rather assume *H3o- in arki- ‘testicle’ and distinguish between *H1er- ‘move’ and *H3er- ‘rise’ (cf. Oettinger 1979: 403f., 523f.). While ār- ‘arrive’ < *H1ōr- reflects the vowel of the perfect, ar- ‘stand’ < *H3or- has the root vowel of the original causative and iterative presents which spread to the middle flexion (cf. Oettinger 1979: 526). It is important to note that Hittite does not tolerate an alternation between initial h- and zero within the paradigm while the vowel alternation between -eand -a- in the root is productive. Consequently, a methodology which does not reckon with the possibility that initial h- was restored or analogically eliminated leads to a proliferation of reconstructed phonemes, and this is precisely what we see in Anatolian studies. My reconstructions differ from the ones proposed by Melchert and Kimball in two respects. First, they are much more constrained because I do not find evidence for more than four distinct sequences (three laryngeals before *-e- and neutralization before *-o-) whereas they start from 24 possibilities (zero and three laryngeals before three vowels *e, *a, *o which may be short or long, cf. Melchert 1994: 46f., Kimball 1999: 119f.). Second, my reconstructed laryngeals are based on independent evidence from the nonAnatolian languages, especially Greek, whereas theirs are based on the internal evidence of the Anatolian languages, especially Hittite. In her excellent new monograph, Rieken is quite candid about adopting the latter methodology when she concedes that her view that *H3 was always lost in Anatolian is “nicht viel mehr als eine Arbeitshypothese” (1999: 5). Initial laryngeals in Anatolian 367 Any proponent of a scientific theory should indicate the type of evidence required for its refutation. While it is difficult to see how a theory which posits *H2- for Hittite h- and a dozen other possible reconstructions for Hittite a- can be refuted, it should be easy to produce counter-evidence for a theory which allows no more than four possibilities which are moreover based on independent, non-Anatolian evidence. The fact that no such counter-evidence has been forthcoming suggests that my theory is correct. In particular, the alleged instances of ha- < *H2o- all show non-apophonic o- in the non-Anatolian languages and Melchert reconstructs e-grade in these words (cf. 1987: 21 and 1994: 106, 145, 235, 257). For Hittite aiš ‘mouth’, Latin ōs, I reconstruct *H1eH3s (cf. Melchert 1994: 115f., Rieken 1999: 186). My theory makes three more predictions which could but have not been refuted: the absence of an Indo-European alternation between *e- and *a- (not *o-), the absence of Indo-European etymologies with *a- not from *H2e-, and the absence of an Indo-European origin of Hittite he- (unless -e- represents an i-diphthong). There was no phoneme *a in Indo-European (cf. Lubotsky 1981 and 1989). The only example of Hittite a- < *a- which looks ancient is alpā‘cloud’, Gr. ἀλφός, Latin albus ‘white’, also Gr. ἄλφι ‘barley’, Old High German albiz ‘swan’ (cf. Melchert 1994: 147, Rieken 1999: 98, rejected by Kloekhorst 2008: 169). I think that this root was borrowed from a European substratum language because it is not found in Indo-Iranian or Tocharian, has a variant *elbh- in Slavic, has an alternating suffix -it-, -ut- in Germanic (cf. Boutkan 1998b: 127) and the same suffix with an infixed nasal in Slavic in the word for ‘swan’, plays a role in Germanic mythology (cf. English elf) and is frequent in European geographical names (e.g. Alba, Albion, Elbe, the Alps). It supports the view that the Anatolians preceded the Greeks and the Phrygians in their migration from the Ukraine into the Balkans and then into Anatolia. The assumption that original long vowels were not colored by adjacent laryngeals is still maintained by Melchert (1994: 47, 68) and Kimball (1999: 120, 144f.). Since the color of the laryngeals is under discussion and original vowel length is notoriously difficult to establish, it adds another degree of freedom to available loose reconstructions. Interestingly, the harvest of this free lunch is extremely small. Kimball still subscribes to Melchert’s obsolete view that *H2ēand perhaps *H3ē- yielded hi- in Hittite (1999: 144f.), a view which Melchert himself has fortunately withdrawn (1994: 143). Both Melchert (1994: 144) and Kimball (1999: 145) now recognize that hekur ‘crag, rock’ is a loanword from Hurrian. Incidentally, šēhur ‘urine’ looks like a loanword from Semitic (cf. Orel & Stolbova 1995: 125, #533; differently Kloekhorst 2008: 742). The derivation of henk- ‘offer, grant’, middle voice ‘bow’ < *H2ē- cannot be correct in view of the Old Hittite spellings 3sg. ha-ik-ta, 3pl. ha-in-kán-ta which “cannot be dismissed as hypercorrections” (Melchert 1994: 144, cf. Oettinger 1979: 172, 177, Rieken 368 Anatolian 1999: 336). It follows that there is simply no evidence for Hittite he- < *H2ē- or *H3ē- (cf. now Kloekhorst 2008: 98). Kimball claims that initial *H3-, unlike *H2-, was lost in Lycian epirije- ‘sell’, Hittite happariye- < *H3ep-, Latin opus (1987: 187f. and 1999: 385), and Melchert follows her (1994: 72). Rieken has proposed to derive the verb from a thematic stem *Hopro-, not directly from hāppar ‘transaction’ (1999: 315). The initial laryngeal may have been lost in the verb and later restored in Hittite on the basis of the heteroclitic *H3ep-r/n-. Note that Oettinger already proposed two different chronological layers for hap(pa)rae- and hap(pa)rie- and an anaptyctic -i- in Lycian epirije- (1979: 352f., cf. Rieken, l.c.). However, the meaning of the Lycian verb is unclear and it should therefore be excluded from the discussion (cf. Kloekhorst 2008: 296). I conclude that the material adduced by Oettinger, Melchert, Kimball and Rieken is fully compatible with my view that initial *H2and *H3- were preserved before *-e- and lost before *-o- in Anatolian. POSTSCRIPT In a recent study (2006a), Kloekhorst argues that initial *H3- is preserved before *e in Hittite hā- ‘believe’, hanna- ‘sue’, hāppar- ‘trade’, hāran- ‘eagle’, hark- ‘perish’, harp- ‘separate oneself ’, haštāi- ‘bone’, hāwi- ‘sheep’, hallanna/i- ‘lay waste’, harnāu- ‘birthing chair’, hartu- ‘descendant’ and lost before *o and before resonants in ārk- ‘mount’, aru- ‘high’, arki- ‘testicle’, ar- ‘stand’, arai- ‘(a)rise’, arnu- ‘make go’, aniya- ‘carry out’ and lāman ‘name’. He rejects the etymologies of haliya- ‘kneel’, *hapuš- (recte: hāpūša- ‘shaft’, cf. Kloekhorst 2005) and hašduer- ‘brushwood’ and proposes *H2- in harg(a)nau- ‘palm, sole’ and *H1- in aiš- ‘mouth’. I agree that zero grade is preferable to o-grade in arta < *H3r-to, arai- < *H3r-oi- and aniya- < *H3n-ie/o- and that his alternatives for hallanna/i-, harnāu-, hartu-, aru-, arki-, harg(a)nau- and aiš- are attractive possibilities. HITTITE ammuk ‘ME’ In the Indo-European department of Leiden University, Alwin Kloekhorst has initiated a discussion on Hittite ammuk ‘me’. The central question is: where did the geminate come from? This has led me to reconsider the origin of the IndoEuropean personal pronouns against the background of my reconstruction of Indo-Uralic (K203: 221-225). For the historical data I may refer to Schmidt (1978). On the basis of the Indo-European evidence, the personal pronouns can be reconstructed as follows (cf. Beekes 1995: 207-211, Cowgill 1965: 169f.): singular nom. acc. gen. abl. dat. loc. poss. 1st *Ȥeg*Ȥme *Ȥmene *Ȥmed *Ȥmighi *Ȥmoi *Ȥmos 2nd *tu*tue *teue *tued *tubhi *toi *tuos dual nom. acc. gen. abl. dat. loc. 1st *ueȤ*nȤue *noȤ *nȤued *nȤuebhi *nȤui 2nd *iuȤ*uȤe *uoȤ *uȤed *uȤebhi *uȤi plural nom. acc. gen. abl. dat. loc. poss. 1st *ue*nsme *nos *nsmed *nsmei *nsmi *nsos 2nd *iu*usme *uos *usmed *usmei *usmi *usos reflexive *sue *seue *sued *subhi *soi *suos Here *Ȥ stands for the glottal stop *H1. The reconstruction of initial *Ȥm- is based on Greek and Armenian. The reconstruction of the dative forms *Ȥmighi, *tubhi, *subhi is based on Italic *mihei (cf. Oscan sífeí), Sanskrit túbhyam, and the BaltoSlavic forms, which point to dat.-loc. *minoi, *tubhoi, *subhoi, e.g. Polish mnie, tobie, sobie (similarly in Czech and Old Russian, but te-, se- from the gen.-acc. 370 Anatolian form in South Slavic and modern Russian), East Baltic dial. mu- from *tu- (cf. Endzelin 1971: 187), Old Prussian subs ‘self ’. While Sanskrit preserved the dual paradigms remarkably well, Greek evidently generalized full grade *noȤ-. The same generalization in the corresponding form *uoȤ- and the reanalysis of the pronominal stem as *w- may have provoked the replacement of the latter by s- < *tw- from the singular and subsequently by σφ- on the basis of the dat.-inst. form σφι, which apparently replaced first *subhi on the model of loc. *soi and later *tubhi after dat.-loc. σοι < *twoi. As a result, we find Greek σφ- in the 2nd dual and 3rd plural forms. The reconstructed accusatives *Ȥme (or *mme), *tue (or *twe), *nȤue (or *nHwe), *uȤe (or *uwe), *nsme, *usme have given rise to different interpretations. It has been proposed that the distinction between *-me and *-ue reflects plural versus dual number (Cowgill 1965: 169), first versus second person (Katz 1998: 279), or a phonological split (Meyer 1997: 101-104). None of these proposals explains the actual distribution of the endings, in particular the short root vowel -u- in all case forms of Sanskrit yuvám ‘ye two’ and the fact that we never find *u twice in the same pronominal form. This strongly suggests that all instances of *u in the personal pronouns have a single origin. For Indo-Uralic we can reconstruct the pronouns *mi ‘I’, *me ‘we’, *ti ‘thou’, *te ‘you’, demonstratives *e/i, *t-, *s-, reflexive *u/w, dual *-ki, plural *-t in heads and *-i in dependent forms, genitive *-n, case particles acc. *m, loc. *i, abl. *t, adessive *pi (cf. Collinder 1960: 237, 243, K203 and K205), where acc. *m and abl. *t may represent earlier *me and *te. Sound laws which are relevant in the present context include IE *s < IU *ti (e.g. in nom.pl. *-es beside *-i and abl.sg. *-os beside *-d), IE *e < all IU vowels under the stress except word-final high vowels and zero grade elsewhere, then IE *o < *u (syllabic *w) in unstressed syllables, then IE *o < *e in cliticized forms, and the (Indo-Uralic) lenition of *p, *t, *k to *b(h), *d(h), *g(h) in weak syllables and word-finally (cf. K203: 221f. and K213). The Indo-European thematic flexion was built on an ergative case form in *-os (cf. Beekes 1985: 191-195) which still functions as the gen.-abl. form of the consonant stems in the historical languages. The possessives in *-os thus represent the earlier ablative of the personal pronouns while the new ablative in *-ed is built on the accusative, like the locatives *nsmi and *usmi and the datives *nsmei and *usmei as well as the corresponding dual forms. This essentially reduces the problem of the stem formation to the nom. acc. gen. forms and to the dat. and loc. forms of the singular. Here the question arises: how did the apparently simple and transparent pronominal system of Indo-Uralic develop into the much more complicated and opaque Indo-European system? The answer to this question lies primarily in the assibilation of *ti to *si and the rise of ablaut which reduced all non-final vowels to *e under the stress and zero grade elsewhere. As a result, we expect the following outcome: Hittite ammuk ‘me’ 371 independent ‘I, me’ ‘myself ’ ‘we, us’ ‘thou, thee’ ‘yourself ’ ‘ye, you’ stressed *mi, *me*mu, *me*me, *me*si, *se*tu, *te*te, *te- unstressed *m*m*m*s*t*t- Indo-Uralic *mi *mu *me *ti *tu *te dependent ‘I, me’ ‘myself ’ ‘we, us’ ‘thou, thee’ ‘yourself ’ ‘ye, you’ stressed *men *men *men *sen *ten *ten unstressed *mn*mn*mn*sn*tn*tn- Indo-Uralic *min *mun *men *tin *tun *ten It is clear that this system could not be maintained. Moreover, the stem form *s< *ti for the second person interfered with the Indo-Uralic demonstrative *s-, which is preserved in the Indo-European anaphoric pronoun *so. The largescale homophony was eliminated by the use of deictic *Ȥe ‘this’ for the first person singular and *ue ‘self ’ for a person who is contrasted with another (third) person and by the suffixation of *-Ȥ < *-ki for the dual and *-i, later *-s < *-ti for the plural. This resulted in such forms as *Ȥme ‘this-me’, *tue ‘thee-self ’, *sue ‘him-self ’ (cf. K203: 225), also *ueȤ, *uei ‘(our)selves’ in contrast with outsiders (inclusive meaning) versus *(m)neȤ, *(m)nes ‘ours’ in contrast with your people (exclusive meaning), *ueȤ, *ues ‘yours’ in contrast with other people, then *uȤe ‘you two’ in contrast with ‘them’ and *nȤue ‘we two’ in contrast with both ‘you’ and ‘them’. These forms must have existed at an early stage already because the o-vocalism of *noȤ, *nos, *uoȤ, *uos originated in their use as clitics and we find the corresponding zero grade in acc.pl. *nsme, *usme, where *-me can hardly be anything else than the full grade IU case particle *me. On the other hand, the forms *teue and *seue show the continued existence of *te, *se, *ue as separate words at the stage when full grade *e in unstressed syllables became possible. Now we turn to the case endings. It appears that gen. *men ‘me’ was remodeled to *mene on the basis of *teue and *seue. Contrary to my earlier view (K203: 223), I now think that dat. *mighi represents original *mibhi with dissimilation of the labial articulation because I cannot otherwise explain the differentiation from *tubhi and *subhi. These forms seem to preserve IU *mi ‘I’, *tu ‘thou-self ’, and *pi ‘at’. There is no reason to assume an initial laryngeal in *mighi and *mene if these directly continue IU nom. *mi and gen. *min. Indeed, the laryngeal may have been limited to acc. *Ȥme, loc. *Ȥmoi, and poss. *Ȥmos, 372 Anatolian where it is preserved in Greek. It was evidently introduced at an early stage from the nominative *Ȥeg-, which contains the deictic element *Ȥe- ‘this’ and the emphatic particle *g(e) which is also found in acc. Greek ἐμέγε, Gothic mik, Hittite ammuk. The absence of *-u- in the locatives *toi and *soi suggests to me that we have *o < *u rather than *o < *e here, while the form *Ȥmoi can easily be analogical. Hittite appears to have preserved the original IU pronoun *ti (with restored *t-) in nom. zik ‘thou’, as opposed to *tu ‘thou-self ’ in the other IE languages. On the other hand, nom. uk ‘I’ seems to reflect *Ȥe-u- ‘this-self ’, which is not found elsewhere, similarly ammuk ‘me’ beside tuk ‘thee’. The latter form may have originated from generalization of the zero grade in *tu-g(e), which suggests the possibility that *Ȥme-g(e) and *Ȥu-g(e) were conflated into *Ȥm-Ȥu-g(e), yielding ammuk. POSTSCRIPT Kloekhorst now derives amm- from *Ȥmn- (2008: 114) with a cluster *-mn- for which I see no evidence in the Indo-European material. In my view, original *mn- yielded *n- in the 1st pl. forms whereas the 1st sg. paradigm has *Ȥme(-). HITTITE hi-VERBS AND THE INDO-EUROPEAN PERFECT In an earlier study (K049) I argued that unlike aorists and athematic presents, Indo-European perfects and thematic presents originally had a dative subject, as in German mir träumt ‘me dreams’ for ich träume ‘I dream’, e.g. Greek οἶδα ‘I know’ < ‘it is known to me’, ἔδομαι ‘I will eat’ < ‘it is eatable to me’. On the basis of Oettinger’s epoch-making book (1979), I proposed that the Hittite hi-flexion originated from a merger of the perfect, where *-i was added to 3rd sg. *-e in order to supply a new present, with the thematic flexion of causatives and iteratives, where the final *-e of 3rd sg. *-eie was dropped before the loss of intervocalic *-i- (K049: 315). This view must now be reconsidered against the background of Kloekhorst’s dissertation (2007), which marks another turningpoint in the history of Hittite studies. For convenience’s sake I shall write h, j, w for ḫ, â, uʢ and *q for any Indo-European laryngeal. Kloekhorst has demonstrated that apart from the factitives in -ahh-, Hittite hi-verbs show an alternation between *-o- in the singular and zero in the plural both in the root of underived stems and in the suffix of derived stems, e.g. au-, u- < *q(o)u- ‘to see’, āk-, akk- < *q(o)k- ‘to die’, ārr-, arr- < *q(o)r-q ‘to wash’, ištāp-, ištapp- < *st(o)p- ‘to shut’, tarna-, tarn- < *trk-n(o)-q ‘to let go’, hamank-, hame/ink- < *qm(o)-n-gh- ‘to tie’, dai-, ti- < *dh-q(o)i- ‘to put’, pai-, pi- < *qp-(o)i‘to give’, mēma-, mēmi- < *me-m(o)i- ‘to speak’, lilhuwa-, lilhui- < *li-lqu-(o)i- ‘to pour’. It follows that all of these must be derived from Indo-European perfects. Note that Kloekhorst has conclusively refuted Jasanoff ’s ill-conceived theory (2003), which can now safely be discarded. Two questions remain: how did the Hittite hi-verbs develop semantically from original perfects, and where do the causatives and iteratives fit in with the new reconstruction? Here I would like to call attention to an important but largely forgotten article by Herman Kølln (1968), who points to the threefold opposition between Greek ἀποθνῄσκω ‘struggle with death’, ἀπέθανον ‘passed away’, and τέθνηκα ‘am dead’, which represent three successive stages of a single event. The same threefold opposition is found in Czech, e.g. imperfective klekat, perfective kleknout ‘to kneel down’, stative (resultative) klečet ‘to be on one’s knees’, also sedat (si), sednout (si) ‘to sit down’, sedět ‘to sit’, lehat (si), lehnout (si) ‘to lie down’, ležet ‘to lie’, blýskat (se), blýsknout (se) ‘to flash’, blyštět se ‘to shine’, zmlkat, zmlknout ‘to fall silent’, mlčet ‘to keep silent’, vstávat, vstát ‘to get up’, stát ‘to stand’. Similar triplets are found in the other Slavic languages. The Slavic stative verbs in -ěti such as Czech klečet ‘to kneel’, vidět ‘to see’, držet ‘to hold’ correspond to the Greek perfect, denoting an event where the non-agentive subject has no effect on an outside object. 374 Anatolian When there is no stative verb, the imperfective member of an aspectual pair may take its place, e.g. opíral se o strom ‘er lehnte sich an den Baum zurück’ or ‘er sass an den Baum zurückgelehnt’, obklopovali svého přítele ‘sie stellten sich in einem Kreis um ihren Freund’ or ‘sie standen in einem Kreis um ihren Freund’, skrýval peníze ve skříni ‘er versteckte das Geld im Schrank’ or ‘er hielt das Geld im Schrank verborgen’, hosté zaujímali svá místa ‘die Gäste nahmen ihre Plätze ein’ or ‘die Gäste sassen auf ihren Plätzen’ (Kølln 1968: 133). Similar instances can be found in Polish, e.g. Jan rozchyla drzwi ‘John sets or is keeping the door ajar’, Jan obejmuje Marię wpół ‘John puts or is holding his arm around Mary’s waist’, Jan wyciąga rękę ‘John stretches out his arm or is holding his arm stretched out’, Jan się nachyla ‘John leans forward or is leaning forward’ (Proeme 1980: 312), and in Russian, e.g. sneg pokryvaet kryši ‘snow covers the roofs’, which may refer either to the process or to the resulting state. While the Slavic stative verbs in -ěti generally correspond semantically to the Greek perfect, this is not always the case. Kølln calls attention to Czech pučet ‘to swell, to bud, to sprout’, which denotes a development leading up to the event of pukat ‘to become cracked, to break (into leaf)’, perfective puknout. Here pučet describes the stage preceding pukat and puknout, whereas klečet describes the stage following klekat and kleknout. Other stative verbs denote continuous sound or movement, e.g. pištět ‘to whistle, to pipe’ beside pískat and písknout which depict the course of action and its conclusion, e.g. neslyšels že jsem na tebe pískal (ipf.) ‘hast du nicht gehört, dass ich dir pfiff (einmal oder mehrmals)’ and pískl (pf.) jsem jen jednou, ale i kdybych pískl (pf.) víckrát, asi bys to neslyšel ‘ich pfiff nur einmal, aber hätte ich mehrmals gepfiffen, hättest du es wohl auch nicht gehört’ (Kølln 1968: 136). Here the stative verb pištět does not denote another stage in the development of the action but expresses its continuousness. Similarly, the Greek perfect κέκληγα ‘scream continually’ denotes incessant action, as opposed to the aorist ἔκλαγξα ‘let out a scream’. Stative verbs in -ěti like Czech klečet, pučet and pištět are intransitive and denote either inactivity or continuous action. Other stative verbs in -ěti lack an aspectual pair denoting the same event, e.g. bolet ‘to ache’, šumět ‘to make a noise’, or at least its perfective member, e.g. letět ‘to fly’, běžet ‘to run’. A few of them have developed into regular imperfective verbs, e.g. hořet ‘to burn’, cf. zahořet ‘to catch fire’, shořet ‘to burn down’, and křičet ‘to shout’, which generally describes constant screaming but may also refer to a single cry. All of these are intransitive, e.g. letět, hořet, křičet, or at least denote an event where the nonagentive subject has no effect on an outside object, e.g. držet ‘to hold’, vidět ‘to see’, slyšet ‘to hear’. The only exception Kølln mentions is vrtět máslo ‘Butter schlagen, to churn’, which must be recent in view of the usual construction with an instrumental object in vrtět hlavou ‘to shake one’s head’, vrtět ocasem ‘to wag one’s tail’. This development of stative verbs into regular imperfectives and subsequently into transitive verbs offers a model for the development of the Hittite hi-verbs and the Indo-European perfect 375 Indo-European perfect in Hittite. As Kølln put it in an earlier article (1966: 75): “Vom historischen Gesichtspunkt aus kann dieser Befund so interpretiert werden, dass gewisse Deverbativa, die ursprünglich nur das Zuständliche ausdrückten, ihren Anwendungsbereich allmählich erweiterten und sich auch das imperfektive Bedeutungsgebiet unterworfen haben.” Turning now to the Hitttite material, we may wonder if the hi-verbs can semantically be derived from Indo-European perfects along the lines indicated by Kølln for the Slavic stative verbs in -ěti such as Czech klečet, pučet, pištět, bolet, šumět, letět, běžet, hořet, křičet, vidět, držet, vrtět. An important point which must be taken into account is the syntactic change from dative subject to nominative subject, which may have occurred as early as the Anatolian exodus from the Indo-European homeland in the Ukraine and given rise to transitive usage, as in Czech vidět, držet, vrtět. This development can be illustrated with the following example from Georgian (cf. Tschenkéli 1958: 488): kurdi gaep’ara p’olicielebs ‘the thief (nom.) escaped the policemen (dat.)’, p’olicielebs gaep’arat kurdi ‘idem’, where the added plural marker -t in the second variant is coreferential with the dative subject, the hapless policemen. The substitution of the nominative for the dative subject, as in German er träumte for ihm träumte ‘he dreamt’, might yield the meaning ‘the policemen let the thief escape’, which could easily develop into a causative. This offers an explanation for the development of the Hittite hi-flexion. The following analysis is entirely based on the data presented by Kloekhorst in his dissertation (2007). I shall first leave suffixed and reduplicated formations and compounds out of consideration here. 1. āk-, akk- ‘to die, to be killed, to be eclipsed (of sun and moon)’. Note that ‘to be eclipsed’ like ‘to obscure, to conceal, to hide’ may refer either to the process or to the resulting state, as in Czech skrýval (peníze ve skříni) ‘hid’ or ‘kept hidden’ cited above. 2. ār-, ar- ‘to come (to), to arrive (at)’. The cognates Gr. ἔρχομαι and Skt. ṛccháti with the imperfective suffix *-ske/o- relate to the Hittite verb like the Czech atelic imperfectives létat and běhat to the original statives denoting continuous movement letět ‘to fly’ and běžet ‘to run’. 3. ārr-, arr- ‘to wash’, Toch. A yär- ‘to bathe’ is a typical verb denoting continuous action. 4. ārk-, ark- ‘to mount, to copulate’ denotes continuous action. 5. au-, u- ‘to see, to look’. This verb is immediately comparable to Czech vidět ‘to see’. 6. hān-, han- ‘to draw (liquids)’ denotes continuous action. 7. harra-, harr- ‘to grind, to splinter up (wood), to crush (bread)’, Gr. ἀρόω ‘plough’. This verb denotes continuous action. 376 Anatolian 8. hāš-, hašš- ‘to give birth (to), to beget, to procreate’ is reminiscent of Czech pučet ‘to swell, to bud, to sprout’, which describes the state preceding the event of pukat ‘to break (into leaf), to become cracked’. 9. hāt-, hat- ‘to dry up, to become parched’, Gr. ἄζω. This verb denotes continuous change. 10. hatk- ‘to shut, to close’, Gr. ἄχθομαι ‘be burdened, be depressed’. Like ‘to hide’ and ‘to cover’, the Hittite verb can easily encompass both the process and the resulting state, like Polish rozchyla ‘sets ajar’ or ‘is keeping ajar’, obejmuje ‘puts around’ or ‘is holding around’, wyciąga ‘stretches out’ or ‘is holding stretched out’ cited above, or may be a causative (see below). 11. huwapp-, hupp- ‘to be hostile towards, to do evil against, to hurl, to throw’, Skt. vap- ‘to strew, to scatter’. This verb denotes continuous action. 12. huwart-, hurt- ‘to curse’ denotes incessant action and may refer both to the process and to the resulting state. 13. iškalla-, iškall- ‘to slit, to split, to tear’, Gr. σκέλλω ‘hoe’. This verb denotes continuous action or may be a causative. 14. išpār-, išpar- ‘to spread (out), to strew’, Gr. σπείρω. This verb denotes continuous action. 15. išparra-, išparr- ‘to trample’ denotes continuous action. 16. lā-, l- ‘to loosen, to release, to untie, to relieve’, Gothic letan. The action entails a resulting state where the subject has no effect on the outside object. 17. malla-, mall- ‘to mill, to grind’ denotes continuous action. 18. mālk-, malk- ‘to spin’ denotes continuous action. 19. māld-, mald- ‘to recite, to make a vow’ denotes continuous action. 20. mau-, mu- ‘to fall’, Latin moveō ‘move’. This verb denotes continuous movement, cf. Czech letět ‘to fly’. 21. nāh-, nahh- ‘to fear, to be(come) afraid, to be respectful, to be careful’. This is a typical perfect. 22. nai-, ni- ‘to turn, to send’, Skt. nay- ‘to lead’. This verb is immediately comparable to Czech vrtět ‘to turn’. 23. para-, par- ‘to appear, to emerge’ may refer either to the process or to the resulting state. 24. pāšk-, pašk- ‘to stick in, to fasten, to plant, to set up’, which appears to contain a suffix *-sk-, does not conform to the semantic pattern discussed here and may be a causative (see below). 25. padda-, padd- ‘to dig (the ground)’, Latin fodiō. This verb denotes continuous action. 26. šāh- ‘to clog, to stuff, to fill in, to plug up’, Toch. B soy- ‘be satisfied’. This verb denotes continuous action or may be a causative. It is reminiscent of Czech pučet ‘to swell, to bud, to sprout’. 27. šākk-, šakk- ‘to know (about), to experience, to recognize, to remember’, Latin sciō. This is a typical perfect. Hittite hi-verbs and the Indo-European perfect 377 28. šārr-, šarr- ‘to divide up, to distribute, to split, to separate’ denotes continuous action or may be a causative. 29. šarta-, šart- ‘to wipe, to rub’ denotes continuous action. 30. šuhha-, šuhh- ‘to scatter’, Gr. ὕω ‘to rain’. This verb denotes continuous action. 31. dā-, d- ‘to take, to wed, to decide’, Skt. dádāti, Gr. δίδωμι. The compounds uda-, ud- ‘to bring (here)’ and peda-, ped- ‘to take (somewhere), to carry, to transport, to spend (time)’ suggest continuous movement, like Czech běžet ‘to run’, cf. also ūnna-, ūnni- ‘to send (here), to drive (here)’ and penna-, penni- ‘to drive (there)’ belonging with #22 above. The verb may be identified with Finnish tuo- ‘bring’, Hung. toj- < Proto-Uralic *toqi- (cf. K112: 82 and K203: 220, Sammallahti 1988: 550), which has evidently preserved the original meaning of continuous movement. 32. wai-, wi- ‘to cry (out)’ is immediately comparable to Czech křičet ‘to shout’. 33. wāk-, wakk- ‘to bite’, Gr. ἄγνῡμι ‘break’, Toch. AB wāk- ‘split, burst’. This verb suggests continuous action but may be a causative. 34. wašta-, wašt- ‘to sin, to offend’ may be compared to #12 above, denoting incessant action. 35. zāh-, zahh- ‘to hit, to beat’ denotes continuous action or may be a causative. Oettinger classifies the following verbs as original causatives and iteratives to be compared with the Sanskrit presents in -áya- (1979: 414-430). Here again I give Kloekhorst’s translations. The rise of the causative can be attributed to the substitution of a nominative for a dative subject as illustrated with the Georgian example cited above. 36. ārk-, ark- ‘to cut off, to divide’, Latin (h)ercīscō ‘to divide (an estate)’. 37. iškār-, iškar- ‘to sting, to stab, to pierce’, Gr. κείρω ‘cut (off)’. 38. išpānt-, išpant- ‘to libate, to pour, to sacrifice’, Gr. σπένδω, Latin spondeō ‘pledge, promise’. 39. ištāp-, ištapp- ‘to plug up, to block, to enclose, to shut’, cf. #10 and #26 above. 40. kānk-, kank- ‘to hang, to weigh’, Gothic hahan. 41. karāp-, kare/ip- ‘to devour, to consume’, Skt. grabh- ‘to seize’. This verb may actually denote continuous action, like Czech hořet ‘to burn’. 42. lāhu-, lahu- ‘to pour, to cast (objects from metal), to (over)flow’. The intransitive meaning suggests continuous movement, like Czech běžet ‘to run’. 43. lāk-, lak- ‘to knock out (a tooth), to turn (one’s ears or eyes towards), to train (a vine)’, Gothic lagjan ‘to lay down’. This verb looks like a typical causative. 44. mārk-, mark- ‘to divide, to separate, to distribute, to cut up’ may be compared to #28 and to #36. 45. šarāp-, šare/ip- ‘to sip’, Latin sorbeō. 46. dākk-, dakk- ‘to resemble’, Gr. δοκεῖ ‘seems’. 378 Anatolian 47. warš- ‘to reap, to harvest, to wipe’, Old Latin vorrō ‘wipe’, but cf. #29 above. 48. wāš- ‘to buy’, Latin vēnum dare ‘to sell’. Unlike the other Indo-European languages, Anatolian evidently created perfects from (imperfective) nasal presents and (perfective) sigmatic aorists: 49. tarna-, tarn- ‘to let go, to allow, to leave’ < *trk-n(o)-q. 50. šunna-, šunn- ‘to fill’ < *su-n(o)-q. 51. šanna-, šann- ‘to hide, to conceal’ < *sn-n(o)-q. 52. hamank-, hame/ink- ‘to tie, to betroth’ < *qm(o)ngh-. 53. kalank- ‘to soothe, to satiate, to satisfy’ < *gl(o)ngh-. 54. ānš- ‘to wipe’ < *qom-qs-, Gr. ἀμάω ‘mow, reap’. 55. hārš- ‘to till (the soil)’ < *qor-qs-, Gr. ἀρόω ‘plough’, cf. #7 above. 56. maz- ‘to withstand, to resist’ < *m(o)qdh-s-. 57. pahš- ‘to protect, to guard, to defend’ < *p(o)-qs-, Latin pāscō, pāvī ‘graze’. 58. pāš-, paš- ‘to swallow’ < *p(o)-qs-, Gr. πÕνω ‘drink’. The most frequent suffix of derived perfects in Hittite is *-(o)i-, which apparently contributed a sense of directionality and which may (or may not) be identical with Proto-Uralic *-j- found in inchoative, terminative, passive, frequentative and continuative verbs (cf. Collinder 1960: 275). 59. āppa-, āppi- ‘to be finished, to be done’ is a typical perfect. 60. arai-, ari- ‘to arise, to lift, to raise’, Latin orior. This verb denotes directed movement, cf. Czech běžet ‘to run’. 61. halai-, hali- ‘to set in motion’, Gr. ἰάλλω ‘send off ’. This verb also suggests directed movement, cf. #43 above. 62. halzai-, halzi- ‘to cry out, to shout, to invoke, to recite’, Gothic laþon. This verb is comparable to Czech pištět ‘to whistle’ and křičet ‘to shout’. 63. huwai-, hui- ‘to run, to hurry, to spread (of vegetation)’, Skt. v¡ti ‘to blow (of wind)’. This verb is comparable to Czech běžet ‘to run’. 64. išhai-, išhi- ‘to bind, to wrap, to obligate with, to impose upon’, Skt. sā-, si-, perfect siṣ¡ya. This verb may be compared to Czech držet ‘to hold’. 65. išhamai-, išhami- ‘to sing’ may be compared to Czech pištět ‘to whistle’. 66. išhuwai-, išhui- ‘to throw, to scatter, to pour’, Gr. ὕω ‘to rain’, cf. #38 and #42 above. 67. išpai-, išpi- ‘to get full, to be satiated’, Skt. sphā(ya)- ‘to become fat, to increase’. This verb is reminiscent of Czech pučet ‘to swell, to bud, to sprout’. 68. mai-, mi- ‘to grow (up), to thrive, to prosper, to be born’ is again reminiscent of Czech pučet, cf. #8 above. 69. pai-, pi- ‘to give, to pay, to grant, to hand over’, which is cognate with Hitt. epp-, app- ‘to take, to seize’, Skt. āpnóti ‘to reach, to gain, to obtain’, Latin apīscor ‘reach, get’ (cf. also Kloekhorst 2006b). This verb is of the type exemplified by Czech (hosté) zaujímali (svá místa), ‘(the guests) took (their places)’ or ‘were Hittite hi-verbs and the Indo-European perfect 379 sitting’, where English sat covers both variants. The meaning ‘to give’ evidently developed from ‘to take there/somewhere’, cf. #31 above. Note that the Sanskrit and Latin verbs are derived imperfectives from the Hittite formations epp- and pai-, respectively (cf. K239: 136). 70. parai-, pari- ‘to blow (a horn), to blow on (a fire), to blow up’, Gr. πίμπρημι. This verb may be compared to Czech pištět ‘to whistle’. 71. pattai-, patti- ‘to run, to race, to flee, to fly’, Skt. pátati, Gr. πέτομαι. This verb is comparable to Czech běžet ‘to run’ and letět ‘to fly’. 72. šai-, ši- ‘to impress, to seal, to sting, to shoot, to throw’, Latin serō, Gothic saian ‘to sow’. This verb denotes directed action and is reminiscent of Czech vrtět ‘to turn, to shake, to churn’. 73. dai-, ti- ‘to lay, to put, to place’ < *dh-q(o)i- denotes directed action, cf. also #31 and #69 above. 74. tarai-, tari- ‘to exert oneself, to become tired’ denotes directed action and is reminiscent of Czech běžet ‘to run’. 75. zai-, zi- ‘to cross (over)’, Skt. at- ‘to wander, to roam’. The Hittite verb denotes directed movement, like Czech běžet ‘to run’. Reduplication appears to have added intensive meaning, but the number of examples is limited. There are three verbs which look like Indo-European reduplicated perfects and aorists, all of which are formally and semantically comparable to Skt. uv¡ca, vav¡ca, ávocat ‘spoke’ < *we-w(o)kw-. 76. mēma-, mēmi- ‘to speak, to recite, to tell’ < *me-m(o)i-, perhaps cognate with Skt. minóti ‘to establish’. 77. wewakk- ‘to demand, to ask’ < *we-wok-, which is cognate with wekk- ‘to wish, to desire, to ask for’, Skt. vaś-. 78. hanna-, hann- ‘to sue, to judge’ < *qe-qn(o)-q, cognate with Gr. ὄνομαι ‘blame’. There are eight verbs with -i- in the reduplication syllable, which suggests that they were derived from reduplicated presents like Skt. vívakti ‘speaks’: 79. halihla-, halihli- ‘to genuflect, to make obeisance to’ < *qli-ql(o)i-, which is cognate with halije/a- ‘to kneel down’ and probably with #61 halai-, hali- ‘to set in motion’, cf. Czech klečet, klekat cited above. 80. lilhuwa-, lilhui- ‘to pour’ < *li-lqu(o)i-, which is cognate with #42 lāhu-, lahu-, also lilahu- ‘to pour’. 81. mimma-, mimm- ‘to refuse, to reject’ < *mi-m(o)-q. 82. parip(p)ara-, parip(p)ari- ‘to blow a horn’ < *pri-prq(o)i- is a derivative of #70 parai-, pari-, also papra-, papri- ‘to blow’, Gr. πίμπρημι. 83. pippa-, pipp- ‘to knock down, to tear down, to destroy, to throw up’ < *pi-p(o)-q may be cognate with #58 pāš-, paš- ‘to swallow’ (cf. ‘the earth swallowed them up’). 380 Anatolian 84. šišha-, šišh- ‘to decide, to appoint’ < *si-sq(o)i- is cognate with #64 išhai-, išhi- ‘to bind’, Skt. sā-, si-. 85. titta-, titti- ‘to install, to assign’ < *dhi-dhq(o)i- is cognate with #73 dai-, ti-, also tāišta-, tāišti- ‘to load’ < *dhoqes-dhq(o)i-, Gr. τίθημι. 86. wiwa-, wiwi- ‘to cry’ < *wi-w(o)i- is cognate with #32 wai-, wi-. Both the combined presence of i-reduplication and i-suffixation in most of these verbs and their coexistence with simpler formations show that this was a productive type in pre-Hittite. The derivation of hi-verbs from nasal presents and sigmatic aorists points in the same direction. We may therefore conclude that the exclusive derivation of perfects from the root attested in the other IndoEuropean languages represents a more archaic state of affairs. After the loss of the dative subject construction, the hi-flexion evidently became a device to supply imperfective verbs in the way envisaged by Kølln as formulated in the quotation above. Since the Slavic verbs in -ěti clearly represent the IndoEuropean perfect, we may wonder if the same holds for the verbs in -iti. Both verb classes have an i-present reflecting an athematic flexion type with full grade *-ei- in the singular and zero grade *-i- in the plural (cf. K033: 61 and K087: 107). This type can be identified with the flexion of Latin capiō ‘take’ (cf. K239: 134), Gothic hafjan ‘to raise’. It has a twofold origin. On the one hand, the derivation of Hittite hi-verbs from reduplicated and nasal presents suggests that hi-verbs with the suffix *-(o)i- may similarly have been derived from athematic i-presents. On the other hand, the intransitive Slavic verbs in -ěti clearly correspond to an original perfect, which can now be identified with the Hittite hi-verbs in *-(o)i-. It follows that the latter formation must be reconstructed for the Indo-European proto-language. It is reflected in Skt. kupya- ‘be angry’, tuṣya- ‘be content’, tṛɴṣya- ‘be thirsty’, dṛɴhya- ‘be firm’, búdhya- ‘be awake’, mánya‘think’, yúdhya- ‘fight’, lúbhya- ‘be confused’, hṛṣya- ‘be exited’, Gr. μαίνομαι ‘be furious’, φαίνομαι ‘appear’, χαίρω ‘rejoice’, Latin cupiō ‘desire’, fugiō ‘flee’, patior ‘suffer’, Old Irish do-moinethar ‘think’. What happened to the original athematic i-presents? Latin faciō ‘make’ and iaciō ‘throw’ show that the type must have been productive at an early stage. At that time, the stem-final vowel of the thematic flexion appears to have been an object marker (cf. K049). Traditional Sanskrit grammar distinguishes between 1st class presents with an accented full grade root and unaccented thematic -a-, which are historically identical with the subjunctives of athematic verbs (cf. Renou 1932), 4th class presents with a mostly accented zero grade root and the suffix -ya-, 6th class presents with an unaccented zero grade root and accented thematic -á-, e.g. tudáti ‘to thrust’, which are characteristically accompanied by an implicit or explicit totally affected definite object experiencing a change of state as a result of the action (cf. Renou 1925), and 10th class presents, especially causatives with an original o-grade root and the accented suffix -áya-. The 1st Hittite hi-verbs and the Indo-European perfect 381 class can be explained from the original syntax with a dative subject and a nominative object (cf. K049: 319). The 4th class can now be identified with the original perfect in *-(o)i- reflected in the Hittite material. The 6th class represents an original transitive construction with an ergative subject and a nominative object. The 10th class causatives combine the o-grade of the Hittite simple hi-verbs with the full grade suffix *-ei- expected in the objective (=thematic) flexion of the athematic i-presents. If this suffix contributed a sense of directionality and the thematic flexion reflects its transitivization, their combination is an appropriate device to derive causatives from o-grade perfects. There is no evidence for Anatolian formations corresponding to the Sanskrit 1st and 10th class presents (or thematic subjunctives of athematic verbs) with an accented e-grade in the root or the suffix preceding the thematic vowel. I therefore think that these formations originated after the exodus of the Anatolians from the Indo-European homeland in the Ukraine. At that time, the thematic vowel was evidently added to a stem with zero grades only. It follows that the original objective flexion of athematic i-presents should appear as Sanskrit 6th class presents with the accented suffix -yá-. This is the type syáti ‘to bind’ recently discussed by Kulikov (2000, cf. also 2001: 493-508). While the ergative construction was lost in Proto-Indo-European times already, giving rise to the sigmatic nominative and to thematic nouns with a nominative in *-os and an accusative in *-om (cf. Beekes 1985: 172-195), the dative construction evidently persisted until after the rise of the thematic subjunctive in the nonAnatolian languages. After the loss of the ergative construction, the accented suffix *-ie/o- could easily spread as a suitable device to derive imperfective presents, primarily of transitive verbs. The introduction of full grade stems before the thematic vowel (or the addition of the thematic vowel to full grade stems) now differentiated the original thematic present flexion with a dative subject from the new thematic present flexion with a nominative subject in the non-Anatolian languages, e.g. Skt. dáya- ‘distribute’ beside dyá- ‘cut’ (cf. Kulikov 2000: 277f.). The new suffix *-eie/o- then spread to o-grade perfects before the substitution of a nominative for a dative subject which gave rise to 10th class causatives as illustrated with the Georgian example cited above. The remaining question is: what happened to the athematic i-presents and to the simple thematic flexion (Sanskrit 6th class presents) in Anatolian? While the former can easily have been thematicized and appear as mi-verbs in Hittite, the latter seem to correspond to the causative hi-verbs. When the dative subject construction was lost at an early stage in Anatolian, the nominative construction which had replaced the pre-Indo-European ergative construction with transitive thematic verbs was semantically closer to the perfect than to the formally similar suffixed thematic presents, cf. Skt. ávocat ‘spoke’ < ‘he uttered the word’ beside uv¡ca, vav¡ca ‘spoke’ < ‘the word escaped him’. At this stage, the hi-flexion may have replaced the simple thematic flexion in pre-Hittite. The 382 Anatolian original type may have been preserved as a distinct class of hi-verbs in the nonablauting factitives in -ahh- < *-eqe/o-, e.g. happinahhahhi ‘I enrich’, which were replaced by verbs in *-eqie/o- in the non-Anatolian languages. Summarizing we arrive at the following picture. The elimination of the preIndo-European ergative construction was a common development of the IndoEuropean languages which gave rise to sigmatic nominatives and to the new category of nominal o-stems. As a result, thematic verb forms now had a nominative subject and an accusative object (as did athematic verb forms) if the verb was transitive but an indirect object which could be reanalyzed as a dative subject if the verb was intransitive (as was also the case in the perfect). This ambiguity gave rise to middle paradigms, which supplied intransitive verb forms to transitive verbs. The creation of derived perfects from athematic i-presents also belongs to the Indo-European proto-language, but the derivation of perfects from reduplicated and nasal presents and sigmatic aorists was evidently limited to the Anatolian branch. When the nominative replaced the dative subject of perfects and thematic presents in Anatolian, the older transitive thematic verbs (corresponding to the Sanskrit 6th class presents) adopted the endings of the perfect, giving rise to the hi-flexion. On the basis of both athematic presents and perfects, the non-Anatolian languages created a new class of indeterminate presents with a dative subject, a full grade stem and thematic endings, which developed into the Sanskrit 1st and 10th class presents and the subjunctive of athematic presents. When the nominative replaced the dative subject in these languages, the indeterminate presents which were derived from o-grade perfects developed into causatives and iteratives in the way discussed above. The central Indo-European languages (at least Indo-Iranian and Greek, but not Italo-Celtic, cf. K239: 151-154) finally created a subjunctive of thematic presents by inserting another thematic vowel before the endings. A final question is: where did the various stem formatives come from? As I have indicated elsewhere (K203: 219), I think that nominalizing *-i-, *-m-, *-s-, participial *-l-, *-n-, *-t-, *-nt- and conative *-sk- are of Indo-Uralic origin. It is attractive to identify the last of these with Tocharian A ske-, B skai- ‘to attempt’, which suggests that other verbal suffixes may also go back to simple verbs. I am therefore inclined to identify the Indo-European present stem formatives *-(e)i-, *-(e)m-, *-(e)s-, *-n-, *-t/dh- with the roots ‘to go’, ‘to take’, ‘to be’, ‘to lead’, ‘to put’, like *-sk- with the root ‘to try’. A comparison of these elements with the Uralic verbal suffixes *-j-, *-m-, *-n-, *-t- (cf. Collinder 1960: 272-281) remains a task for the future. STATIVE AND MIDDLE IN HITTITE AND INDO-EUROPEAN In his dissertation (2007), Alwin Kloekhorst lists the following classes of middle verbs (I write h, j, w for ḫ, â, uʢ and *q1, *q2, *q3 for the Indo-European laryngeals). (a) original *CeC-o: ā(i)- ‘to be hot’, eš- ‘to sit down’, happ- ‘to work out’, hatt- ‘to pierce, to prick’, huetti- ‘to draw, to pull’, kīš-, kiš-, kikkiš- ‘to happen, to occur’, nē- ‘to turn (oneself)’, pahš- ‘to protect’, šalīk- ‘to touch’, weh- ‘to turn (oneself)’, zē- ‘to cook (intr.)’. (b) original *CeC-to: harp- ‘to change allegiance’, huett- ‘to draw, to pull’, ki- ‘to lie’, lukk- ‘to get light’, park- ‘to rise’, warš- ‘to lift oneself ’, wešš- ‘to be dressed’. (c) original *CC-o: ark- ‘to mount, to copulate’, halzi- ‘to cry out’, parši-, parš- ‘to break’, tuhš- ‘to be cut off ’. (d) original *CC-to: ar- ‘to stand’, karp- ‘to be angry’, tarupp- ‘to collect oneself ’. (c/d) original *CC-o or *CC-to: pukk- ‘to be hateful’, šar- ‘to embroider’, šupp- ‘to sleep’. (e) tith-a ‘thunders’, which will be left out of consideration here. (f) ištu-āri ‘is exposed’, kišt-āri ‘perishes’, lag-āri ‘falls’, mi-āri ‘is born’, dukk-āri ‘is visible’, ur-āri ‘burns’, wakk-āri ‘is lacking’. (g) verbs in -je/a-, which will be left out of consideration here. (h) other verbs, which will also be left out of consideration here. Kloekhorst lists the following analogical developments after the Old Hittite period. (a) >> (b): ā(i)- ‘to be hot’, eš- ‘to sit down’, happ- ‘to work out’, kikkiš- ‘to happen, to occur’, pahš- ‘to protect’, weh- ‘to turn (oneself)’. (a) >> (g): hatt- ‘to pierce, to prick’, huetti- ‘to draw, to pull’. (b) >> (g): huett- ‘to draw, to pull’, park- ‘to rise’, wešš- ‘to be dressed’. (c) >> (d): ark- ‘to mount, to copulate’. (c) >> (g): halzi- ‘to cry out’, parši-, parš- ‘to break’. (d) >> (g): karp- ‘to be angry’. It follows that class (a) must be ancient and that class (b) can easily be analogical. Moreover, the radical vocalism is not always unambiguous: lukk- and warš- may reflect an original zero grade while pukk-, tarupp- and tuhš- may represent an earlier e-grade, as may have been the case with ark- ‘to mount, to copulate’ if this verb eliminated the initial h- on the analogy of the active paradigm ārk-, ark-. Alwin Kloekhorst points out to me that the Old Hittite middle forms of halzi- and parši- are limited to 3rd sg. halzija and paršija beside active halzāi-, halzi- and thematic middle halzija-, paršija- whereas 1st sg. middle paršha(ri) may represent an original e-grade, like 3rd sg. active paršzi < 384 Anatolian *bhersti. Thus, there is no compelling reason to assume that classes (b) and (c) are original, so that we may have to deal with classes (a), (d) and (f) only. Kloekhorst finds no semantic difference between the 3rd sg. endings *-o and *-to in Hittite, for which Norbert Oettinger proposed a distinction between “stative” and “middle” thirty years ago. The latter author listed four correspondences between Indo-Iranian and Anatolian statives (1976: 112f.): (1) Vedic stáve ‘is announced’, Hittite ištuāri ‘is exposed’. (2) Vedic ¡ste ‘is sitting’, Hittite ēša(ri), Greek ἧσται ‘sits’. (3) Vedic śáye ‘is lying’, Hittite kitta(ri), Palaic kītar ‘lies’. (4) Vedic duhé ‘gives milk’, Hittite dukkāri ‘is visible’, Gothic daug ‘is useful’. It seems to me that the zero grade of Vedic duhé can easily be analogical after the regular middle paradigm whereas the formation of Hittite ištuāri and dukkāri requires a separate explanation. This leaves us with two types of “stative” (a) and (f) and one type of “middle” (d). Elsewhere I have proposed the following reconstruction of the Proto-IndoEuropean perfect, stative and middle endings (K033: 66-68, K044: 128, K049: 312, K203: 218). 1st sg. 2nd sg. 3rd sg. 3rd pl. perfect -q2e -tq2e -e -(ē)r stative -q2 -tq2o -o -ro middle -mq2 -stq2o -to -ntro I now think that the final *-o of the 2nd sg. and 3rd pl. stative endings is analogical and was taken from the 3rd sg. ending (cf. already K203: 224). It follows that the stative differed from the perfect in the e-grade of the root, the 3rd sg. ending *-o, and the absence of a vowel from the other endings. I have proposed that *-q2 and *-tq2 reflect original datives ‘to me, to you’, that *-e represents a subject marker and *-o a reflexive ending, that the formation in *-r can be identified as an original nomen loci, and that the middle endings represent a combination of agent markers with stative endings (K203: 225f.). The expected 1st and 2nd pl. endings *-mq2 and *-q2 or *-tq2 were evidently replaced by the more distinctive pronominal forms *medhq2 ‘to us’ and *(t)dhq2ue ‘to yourselves’ in the stative and the middle (ibidem). While the perfect and the stative were originally intransitive formations with or without the 3rd sg. subject marker *-e and a dative beneficiary, the addition of the stative endings after the agent markers *-m, *-s, *-t, *-nt gave rise to a middle paradigm where the beneficiary was coreferential with the agent. This formation is best compared with the Georgian subjective version, e.g. me vimzadeb sadils ‘I prepare myself a dinner’, as opposed to the neutral version me vamzadeb sadils ‘I prepare a dinner’ and the intransitive middle vemzadebi ‘I Stative and middle in Hittite and Indo-European 385 prepare myself ’, where v- is the 1st sg. subject marker and the following vowel refers to an indirect object (cf. Vogt 1971: 119ff., K049: 321). Though Kloekhorst finds no correlation between Hittite -hi and -mi on the one hand and the endings *-o and *-to on the other, the historical connection seems unmistakable to me in view of the agent marker which is present in the latter but absent from the former members of these pairs. Elsewhere I have argued that the Hittite hi-flexion comprises original perfects, new perfects created on the basis of derived presents, and transitive zero grade thematic formations (K241). This merger obliterated the semantic distinction between original intransitive perfects and transitive verbs in the Hittite hi-flexion and similarly between the 3rd sg. endings *-o and *-to in the stative and the middle. As a result, the original distribution can no longer be established on the basis of the Hittite evidence. The only exception are the verb forms in -āri of class (f), which have clearly preserved the meaning of a stative. Formally, these look like analogical 3rd sg. perfect forms in *-orei created on the basis of a 3rd pl. form in *-r in a similar way as tarnāi ‘lets go’ < *-noqei and dāi ‘puts’, pāi ‘gives’ < *-oiei were created on the basis of derived presents (cf. Kloekhorst 2007, K241). This suggests that, as in the case of the perfect, new statives in *-o could be derived from athematic verbs while new perfects were derived from earlier statives. The peculiar fact that there is no interchange between class (f) and the other classes of middle verbs in Hittite can be explained by the assumption that the creation of the r-perfect was early and preceded the merger of the perfect with the transitive thematic flexion. At a later stage, the derivation of new statives from athematic verbs blurred the distinction between the stative and the middle, which are still reflected in classes (a) and (d), respectively. This leads me to the following chain of events. At the earliest reconstructible stage we expect e-grade of the root in the stative but zero grade before the ending *-e in the perfect. If the apophonic alternation between e- and zero grade was still automatic at the stage when the new 1st and 2nd pl. endings *-medhq2 and *-(t)dhq2ue were introduced into the stative paradigm (stage A of K203: 221), the new forms must have had zero grade in the root. The original 3rd sg. stative ending *-o arose phonetically from lowering of Indo-Uralic *-u (stage B, cf. K203: 224). The paradigmatic alternation between full and zero grade was then evidently introduced from the stative into the perfect at a stage when the alternation between stressed *e and unstressed *o was automatic (stage C of my chronology). The stress was eventually retracted in the singular forms of the perfect when stressed *o and unstressed *e had become possible (stages D and E, respectively), probably on the analogy of the athematic present. The rise of lengthened grade in the 3rd pl. ending *-ēr < *-er was most recent (stage F of my chronology). Reduplicated perfects were probably created in order to supply perfects to reduplicated aorists at a comparatively recent stage (cf. K241) and 386 Anatolian adopted the zero grade 3rd pl. ending *-r because the accent was on the reduplication syllable. The developments proposed here can be summarized in the following tables. stative 1st sg. 2nd sg. 3rd sg. 1st pl. 2nd pl. 3rd pl. stage A1 CeC-q2 CeC-tq2 CeC-u CeC-mq2 CeC-(t)q2 CeC-r stage A2 CeC-q2 CeC-tq2 CeC-u CC-medhq2 CC-(t)dhq2ue CeC-r stages B-F CeC-q2 CeC-tq2 CeC-o CC-medhq2 CC-dhq2ue CeC-r perfect 1st sg. 2nd sg. 3rd sg. 1st pl. 2nd pl. 3rd pl. stages A-B CC-q2e CC-tq2e CC-e CC-mq2e CC-(t)q2e CC-er stages C-E CoC-q2e CoC-tq2e CoC-e CC-mq2e CC-q2e CC-er stage F CoC-q2e CoC-tq2e CoC-e CC-mq2e CC-q2e CC-ēr EIGHT INDO-URALIC VERBS? Károly Rédei (1986) lists 64 words which were supposedly borrowed from IndoEuropean into Uralic at an early date. The material is divided into three groups: 7 Proto-Uralic (PU) etymologies, 18 Finno-Ugric (FU) etymologies, and 39 Finno-Permian (FP) and Finno-Volgaic (FV) etymologies. The source of the borrowings is specified as “vorarisch” for the PU words, “vorarisch oder frühurarisch” and “urarisch” for the FU words, and “frühurarisch” through “uriranisch” for the FP and FV words (Rédei 1986: 26). There are several reasons to call this account into question. Firstly, it is difficult to determine a place and a time which are suitable for borrowings from Indo-European into Proto-Uralic. We can probably identify the Proto-Indo-Europeans with the Sredny Stog culture in the eastern Ukraine around 4000 BC (cf. Mallory 1989 and K111). This clashes with the concept of direct borrowings from Indo-European into Proto-Uralic: “All that seems to be certain is that in the fourth millennium B.C. the ancestors of the Finno-Ugrians and the Samoyeds had lived on the eastern side of the Urals” (Fodor 1976: 50). The earliest contacts between Indo-European and Uralic languages must probably be identified with the eastward expansion of the “vorarische oder frühurarische” Yamnaya culture around 3000 BC and the simultaneous spread of the Finno-Ugric Ural-Kama neolithic culture to the southwest. Even if we were to assume an Uralic homeland west of the Ural mountains, earlier borrowings could only have been taken from the Samara and Khvalynsk cultures on the Middle Volga. Though it is conceivable that the languages spoken in that area were genetically related to Indo-European, or to Uralic, they cannot be identified with the language of the Sredny Stog culture. Secondly, the number of verbs in the oldest material is too large to support the hypothesis that they were borrowed: 3 out of 7 (43%) in the first group, 5 out of 18 (28%) in the second group, and 2 out of 39 (5%) in the third group. Moreover, the two verbs from the third group have questionable etymologies. The verb *kara- ‘graben’ (Rédei 1986: 51) is attested in the Volgaic languages (Mordvin and Cheremis) only. The corresponding words in the Permian languages (Votyak and Ziryene) and in Ob-Ugric (Ostyak) require a reconstruction *kur -, which is incompatible both with the Volgaic forms and with the alleged (Indo-)Iranian source. The verb *niδa- ‘befestigen, heften, binden’ (Rédei 1986: 53) is limited to Finno-Volgaic, e.g. Finnish nito-, Rédei doubts the connection with Skt. náhyati ‘binds’ himself: “Zufälliger Gleichklang?” If we eliminate these two items from the list, the presence of eight verbs in the older material becomes even more significant. 388 Indo-Uralic Thirdly, the derivation of the Proto-Uralic forms from their alleged IndoEuropean sources involves considerable formal difficulties. I shall briefly discuss the four nouns of the first group (Rédei 1986: 40-43). PU *nime ‘Name’, Finnish nimi, Mordvin ľem, Votyak and Ziryene ńim, Ostyak nem, Hung. név, Tavgi (Samoyed) ńim, etc. The PIE word must be reconstructed as *H3neH3mn, Latin nōmen, Hitt. lāman, Skt. n¡ma, Arm. anun, oblique stem *H3nH3men-, Gr. ὄνομα, OIr. ainm, OPr. emmens, Russ. ímja, Alb. emër (cf. K073: 42, K090: 63). The only Indo-European language which has a front vowel in the root is Tocharian, where A ñom and B ñem point to a reconstruction *nēmn, with delabialization of the second laryngeal. But even this form does not account for the high front vowel of the Uralic words. The PIE word is probably a derivative of the verbal root *H3neH3-, Greek ὄνομαι ‘blame’ (cf. Kloekhorst 2008: 283). PU *seʡne (*sōne) ‘Ader, Sehne’, Finnish suoni, Mordvin san, Votyak and Ziryene seʡn, Hung. ín, Tavgi taŋa, etc. The word is compared with PIE *sneH1ur, obl. -en-, Skt. sn¡va, Toch. B ṣñor, Arm. neard, Gr. νεῦρον. Here again, the IndoEuropean forms do not explain the Uralic vocalism, which may be original if the words are related at all, whether the PIE word is a derivative of the root *sneH1or not. A comparison with English sinew from *sH1inu- is no better. It is actually worse because the meaning of the latter word is the result of a Germanic innovation. PU *waśke ‘irgendein Metall, ?Kupfer’, Finnish vaski, Mordvin uśke, viśkä, Votyak veś, Hung. vas, Tavgi basa, etc. This is the only “Kulturwort” in the list. It may be compared with Toch. A wäs, B yasa ‘gold’, which point to earlier *wesa. The latter word cannot be identified with Latin aurum, Lith. áuksas, and besides does not explain the Uralic vocalism. It is much more probable that the Tocharian word was borrowed from Samoyed *wesä (Janhunen 1983: 120). PU *wete ‘Wasser’, Finnish vesi, Mordvin veď, Votyak vu, Hung. víz, Tavgi bēȤ, beda-, etc. In Indo-European, the e-grade is attested in Hittite obl. weten-, Phrygian βεδυ, Arm. get, and in Germanic and Slavic derivatives. If the word was actually borrowed int0 Uralic, this must have occurred at a very early stage. But it is not the kind of word that is easily borrowed, and the Indo-European forms rather look like derivatives of the (Indo-)Uralic word. Against this background, we must consider the possibility that the eight verbs in Rédei’s first and second groups were inherited from Proto-Indo-Uralic. I shall give a brief summary of the material (cf. Rédei 1986: 40-48). PU *miγe- ‘geben, verkaufen’, Finnish myy-, myö-, Mordvin mije-, Vogul (ObUgric) mä(j)-, mi-, maj-, Yenisei (Samoyed) miȤe-, PIE *mei-, Skt. min¡ti ‘exchanges’, Latvian mît. Eight Indo-Uralic verbs? 389 PU *muśke- (*mośke-) ‘waschen’, Estonian mõske-, Mordvin muśke-, muśko-, Votyak miʡśk-, Hung. mos-, Yenisei musua-, PIE *mesg-, Skt. májjati ‘sinks’, Latin mergere, Lith. mazgóti ‘to wash’. PU *toγe- ‘bringen, holen, geben’, Finnish tuo-, Mordvin tuje-, Ostyak tu-, Yurak (Samoyed) tā-, PIE *deH3-, Skt. dádāti ‘gives’, Hitt. dā- ‘take’. FU *aja- ‘treiben, jagen’, Finnish aja-, Ziryene voj-, Vogul wujt-, wojt-, PIE *H2egʄ-, Skt. ájati ‘drives’, Latin agere. FU *kan - ‘streuen, schütten, werfen, graben’, Ziryene kundiʡ-, Ostyak kiʡṇ-, Vogul kōn-, Hung. hány-, PIE *kH2en-, Skt. khánati ‘digs’. FU *teke- ‘tun, machen’, Finnish teke-, Mordvin ťeje-, ťije-, Hung. të(v)-, tësz-, PIE *dheH1-, Skt. dádhāti ‘puts’, Hitt. dāi-, Latin facere. FU *wetä- ‘führen, leiten, ziehen’, Finnish vetä-, Mordvin veďa-, veťa-, viťi-, väďa-, väťe-, Hung. vezet-, PIE *uedh-, OIr. fedid ‘leads’, Lith. vèsti. FU *wiγe- ‘nehmen, tragen’, Finnish vie-, Mordvin vije-, Votyak and Ziryene vaj-, Hung. vi(v)-, visz-, vë(v)-, vësz-, PIE *uegʄh-, Skt. váhati ‘carries’, Latin vehere, Lith. vèžti. Apart from Skt. khánati, all of the Indo-European words are basic verbs with impeccable etymologies. This is a strong argument against borrowing and in favor of an original genetic relationship. As I have indicated elswhere (K111), we may conceive of Indo-European as a language of the Uralic type which was transformed under the influence of a Caucasian substratum. Following this line of thought I tentatively reconstruct Proto-Indo-Uralic *miye-, *muske-, *tagu-, *gaki-, *deka-, *weda-, *wige- (but cf. K203: 220). Thus, I think that the PIE laryngeals developed from velars in the neighbourhood of back vowels, as did Yukagir h- (Collinder 1965: 168) and the uvulars in Turkic and Mongolian. It has been argued that the small number of Indo-Uralic etymologies favors the assumption of borrowing rather than genetic relationship (e.g. Rédei 1986: 10, 20). I am afraid that I fail to understand this reasoning. When we are dealing with distant linguistic affinity, we cannot expect to find large numbers of obvious cognates, which would be contrary to the idea of distant affinity. What we do expect to find is morphological correspondences and a few common items of basic vocabulary. I think that this is precisely what we find in the case of Indo-European and Uralic. Advocates of the alternative hypothesis, viz. that the verbs listed above were borrowed into Uralic, are faced with two insurmountable problems. First, they have to explain the prominent place of basic verbs among the oldest borrowings. Second, they do not account for the differences in the Uralic vocalism, e.g. *nime-, *miγe-, *wiγe- versus *wete, *teke-, *wetä-. It therefore seems to me that the burden of proof is now on the opponents of the Indo-Uralic theory. 390 Indo-Uralic Uhlenbeck (1935: 9ff.) makes a distinction between two components of PIE, which he calls A and B. The first component comprises pronouns, verbal roots, and derivational suffixes, and may be compared with Uralic, whereas the second component contains isolated words, such as numerals and most underived nouns, which have a different source. This is a simplification because we can find good Uralic etymologies for some B words, e.g. Finnish käly ‘sister-in-law’, Gr. γάλως, Russ. zolóvka, but I think that the distinction is basically correct. The wide attestation of the Indo-European numerals must be attributed to the development of trade resulting from the increased mobility which was the primary cause of the Indo-European expansions. Numerals do not belong to the basic vocabulary of a neolithic culture, as is clear from their absence in ProtoUralic and from the spread of Chinese numerals throughout East Asia (cf. also Collinder 1965: 113 and Pedersen 1906: 369 on Swedish kast ‘4’, val ‘80’, Danish snes ‘20’, ol ‘80’, German Stiege ‘20’, Russ. sórok ‘40’, kopá ‘50, 60’). Though Uhlenbeck objects to the term “substratum” for his B complex, I think that it is a perfectly appropriate denomination. The concept of “mixed language” has done more harm than good to linguistics and should be abandoned. THE INDO-URALIC VERB C.C. Uhlenbeck made a distinction between two components of Proto-IndoEuropean, which he called A and B (1935a: 133ff.). The first component comprises pronouns, verbal roots, and derivational suffixes, and may be compared with Uralic, whereas the second component contains isolated words, such as numerals and most underived nouns, which have a different source. The wide attestation of the Indo-European numerals must be attributed to the development of trade resulting from the increased mobility which was the primary cause of the Indo-European expansions. Numerals do not belong to the basic vocabulary of a neolithic culture, as is clear from their absence in ProtoUralic (cf. also Collinder 1965: 112) and from the spread of Chinese numerals throughout East Asia. Though Uhlenbeck objects to the term “substratum” for his B complex, I think that it is a perfectly appropriate denomination. The best candidate for the original Indo-European homeland is the territory of the Sredny Stog culture in the eastern Ukraine (cf. Mallory 1989). If we can identify Indo-Hittite and nuclear Indo-European with the beginning and the end of the Sredny Stog culture, respectively (cf. K111: 138), Uhlenbeck’s view can be unified with Gimbutas’ theory of a primary homeland north of the Caspian Sea and a secondary homeland north of the Black Sea (cf. 1985). What we have to take into account is the typological similarity of Proto-Indo-European to the North-West Caucasian languages. If this similarity can be attributed to areal factors (cf. K130: 94), we may think of Indo-European as a branch of IndoUralic which was transformed under the influence of a Caucasian substratum connected with the Maykop culture in the northern Caucasus. We may then locate the Indo-Uralic homeland south of the Ural Mountains in the seventh millennium BC (cf. Mallory 1989: 192f.) and perhaps identify the Khvalynsk culture on the middle Volga as an intermediate stage before the rise of the Sredny Stog culture in the fifth millennium BC. The Indo-European verbal system appears to combine Uralic flexional morphemes with Caucasian syntactic patterns. Holger Pedersen already argued that the subject of a transitive verb was in the genitive (= sigmatic nominative) case if it was animate and in the instrumental case if it was inanimate while the subject of an intransitive verb and the object of a transitive verb were in the absolutive (= asigmatic nominative) case form (1907: 152), that the endings of the perfect and the thematic present originally belonged to the flexion of intransitive verbs and the “normal”, mostly athematic endings to the flexion of transitive verbs (1933: 311-315), and that the intransitive and transitive flexion types correspond to the Hittite flexional paradigms in -hi and -mi (1938: 80-85). Beekes has shown that this theory explains the origin of the Indo-European 392 Indo-Uralic nominal flexion in its entirety (1985). Knobloch however identified the IndoEuropean thematic vowel in verbal paradigms *-e/o- with an object marker (1953). Elsewhere I have integrated these findings into a coherent whole, arguing that the Indo-European thematic flexion of the verb can be compared with the objective conjugation of the Uralic languages and that this hypothesis explains the distribution of the thematic flexion in Hittite and Sanskrit as well as the rise of the thematic subjunctive (K049, cf. Nikolaeva 1999 on the remarkably similar system in Ostyak). In the following I intend to examine the Indo-Uralic origins of the Proto-Indo-European verbal system which has thus been reconstructed. Since the Indo-European laryngeals apparently developed from uvular obstruents, I shall write *q1, *q2, *q3 in order to facilitate comparison with the Uralic data. Note that *dh stands for a lenis dental stop. My reconstruction contains the following Indo-European verbal paradigms (K033: 67, K049: 312, also Beekes 1995: 252, for the dual endings see K145): I. athematic present (dynamic, subjective, imperfective) 1st sg. 2nd sg. 3rd sg. -mi -si -ti 1st pl. 2nd pl. 3rd pl. -mes -tq1e -(e)nti II. athematic aorist (dynamic, subjective, perfective) 1st sg. 2nd sg. 3rd sg. -m -s -t 1st pl. 2nd pl. 3rd pl. -me -te -(e)nt III. thematic aorist (dynamic, objective, perfective) 1st sg. 2nd sg. 3rd sg. -om -es -et 1st pl. 2nd pl. 3rd pl. -omo -ete -ont IV. thematic present (dynamic, objective, imperfective) 1st sg. 2nd sg. 3rd sg. -oq1 -eq1i -e 1st pl. 2nd pl. 3rd pl. -omom -etq1e -o 1st pl. 2nd pl. 3rd pl. -me -e -(ē)r V. perfect (static, perfective) 1st sg. 2nd sg. 3rd sg. -q2e -tq2e -e The Indo-Uralic verb 393 VI. stative (static, imperfective) 1st sg. 2nd sg. 3rd sg. -q2 -tq2o -o 1st pl. 2nd pl. 3rd pl. -medhq2 -dhwe -ro The six paradigms were originally interconnected by a network of derivative, not flexional relationships. While the stative supplied a middle paradigm to intransitive verb stems, the transitive middle paradigm combined the endings of sets II and VI (cf. K044: 128): VII. transitive middle 1st sg. 2nd sg. 3rd sg. -mq2 -stq2o -to 1st pl. 2nd pl. 3rd pl. -medhq2 -tdhwe -ntro Note that the system looks like the remains of a much more elaborate, but perhaps more regular structure. The most probable grammatical correspondences between Indo-European, Uralic, and other possibly related language families have conveniently been summarized by Joseph Greenberg (2000). The following items from his list are in my view definitely Indo-Uralic (I retain Greenberg’s numbering): 1. 4. 8. 11. 12. 14. 15. 16. 24. 25. 26. 29. 30. 31. 33. 36. 38. 39. 42. 43. 44. first person *m, second person *t, demonstrative *i/e, demonstrative *t, demonstrative *s, dual *ki, plural *t, plural *i, accusative *m, genitive *n, dative *ka, locative *ru, locative *n, locative *i, ablative *t, diminutive *k, nominalizer *i, nominalizer *m, participle *n, participle *t, participle *nt, 394 Indo-Uralic 45. 46. 53. 54. 56. 60. participle *l, verbal noun *s, conative *sk, reflexive *u/w, negative *n, interrogative *k. After this rather lengthy introduction, I now come to the chief part of my contribution, which is a comparison of the reconstructed Indo-European verbal system with its Uralic counterpart. There are two major problems involved here. On the one hand, the shallow time depth of the Uralic data does not allow a reconstruction of the Proto-Uralic verbal system but only of (some of) its components. This deficiency is mitigated by the relatively conservative character of the Uralic languages. On the other hand, the great antiquity of the earliest Indo-European evidence is to some extent invalidated by the radical changes which took place under the influence of the presumably Caucasian substratum. I start from the assumption that the Proto-Indo-Uralic vowel system was identical with the one which has been reconstructed for Proto-Uralic (cf. Sammallahti 1988: 481): i e ä ü ï u o a This system was reduced in non-initial syllables: i ä ï a Moreover, “front and back vowels could not occur together in a (noncompound) word” (Sammallahti, l.c.), so that we can write /i/ for [i, ï] and /a/ for [ä, a] in non-initial syllables. The Proto-Indo-Uralic consonant system cannot easily be reconstructed because the gap between Uralic and Indo-European is huge. I reconstruct ProtoUralic palatalized /ŕ/ and /ľ/ instead of Sammallahti’s spirants /d/ and /ď/ because they pattern like resonants and are reflected as *r and *j in Samoyedic and as *l and *ľ in Finno-Ugric (cf. Sammallahti 1988: 485, 511f., 518, 532), cf. also the variation between Proto-Finno-Permic *śülki and Proto-Ugric *süľki ‘saliva’. I prefer to write Proto-Uralic *q for Sammallahti’s /x/, which is strongly reminiscent of the Indo-European laryngeals (being lost before a vowel and vocalized before a consonant in Samoyedic and lengthening a preceding vowel before a consonant in Finno-Ugric) and may represent more than a single phoneme. Thus, I arrive at the following Proto-Uralic consonant system: The Indo-Uralic verb p m w 395 t s n r l c ś ń ŕ ľ j k q ŋ Unlike Uralic, Indo-European had three series of stops, conventionally written *t, *d, *dh, etc. The difference between fortis *t and lenis *dh is in my view the result of a secondary development, conditioned by the tonal patterns of strings of morphemes (cf. Lubotsky 1988a: 4-7). It is reminiscent of Verner’s law in Germanic and similar phenomena in other languages. Though I do not intend to discuss lexical correspondences here, I would like to adduce seven IndoUralic etymologies which seem particularly attractive to me (cf. K112, Sammallahti 1988: 538, 542, 550f.): (1) (2) (3) (4) (5) (6) (7) *meqi- ‘give, sell’, PIE *mey- ‘exchange’; *mośki- ‘wash’, PIE *mesg- ‘sink, wash’; *(q)aja- ‘drive’, PIE *q2egʄ- ‘drive’; *teki- ‘do’, PIE *dheq1- ‘put’; *toqi- ‘bring’, PIE *deq3- ‘give’; *weta- ‘pull’, PIE *wedh- ‘lead’; *wiqi- ‘take’, PIE *wegʄh- ‘carry’. It appears that no simple sound laws can be established. While it is probable that the Indo-European distinction between palatovelars *kʄ, *gʄ, *gʄh and labiovelars *kw, *gw, *gwh arose when the distinctive timbre of the following vowel was lost (as happened in Ethiopic), the relation between velars and uvulars remains unclear. In particular, the correspondence of Uralic *mośki-, *teki-, *toqi-, *wiqiwith Indo-European *mesg-, *dheq1-, *deq3-, *wegʄh- suggests that the distinction between velars and uvulars is due to a secondary development. If we look beyond Indo-Uralic to the Altaic languages, we should expect that the uvulars developed from velars before back vowels and that the original distribution was obscured by the reduction of the vowel system in non-initial syllables. While Indo-European looks like the development of a Uralic system, the latter looks like having developed from an Altaic system. I therefore take the Uralic distribution of *k and *q to be probably more original and assume for IndoEuropean secondary fronting in *wegʄh- < *wiqi- and secondary retraction in *dheq1- < *teki- (see further below). The rounded laryngeal *q3 of Indo-European *deq3- < *toqi- suggests that the non-initial vowel was rounded as a result of Indo-Uralic vowel harmony in this root. Greenberg rightly points out that Indo-European *i and *u represent not only syllabic *y and *w but also original vowels which alternated with *e and *o 396 Indo-Uralic (2000: 34-39), though his examples are largely incorrect (cf. K069 and K065: 222). For the present purpose it suffices to adduce the relevant instances from Greenberg’s list of Indo-Uralic morphemes (see above): 1. 4. 8. 14. 16. 26. 31. 33. 38. 54. 60. first person *-mi beside *m, second person *-si beside *t (see below), demonstrative *i- beside *e-, dual *-i beside *-e beside *-q1 (cf. K118), plural *-i beside *-es (see below), dative *gʄhi beside *q (see below), locative *-i, ablative *-os beside *-t (see below), nominalizer *-i, reflexive *-o (see below), interrogative *kwi- beside *kwe-. Beekes distinguishes three stages in the development of the Indo-European vowel system (1985: 157): I. full grade (i.e. non-high) vowels *e and *o in stressed syllables only; II. introduction of *o in unstressed syllables; III. introduction of *e in unstressed syllables. This theory accounts for all types of vowel alternation in the Indo-European nominal inflexion (cf. especially Beekes 1985: 161, 169, 207). However, as I doubt the possibility of o-grade in stressed syllables at stage I when all unstressed syllables had zero grade, I would propose the following alternative chronology: A. Indo-European vowel reduction, giving rise to full grade *e under the stress and zero grade elsewhere; B. phonetic lowering of *u (= syllabic *w) to *o, giving rise to a full grade (= non-high) vowel in unstressed syllables; C. analogical introduction of a full grade vowel in unstressed syllables (e.g. in compounds), which automatically yielded new *o; D. introduction of *o in stressed syllables (e.g. by decompounding), resulting in a phonemic opposition between /e/ and /o/ under the stress; E. analogical introduction of full grade *e in unstressed syllables, generalizing the opposition between /e/ and /o/; F. rise of lengthened grade vowels *ē and *ō, yielding the conventional ProtoIndo-European vowel system. This chronology has the advantage of providing an explanation for the successive stages in the development of the vowel system. It also accounts for Beekes’ “difficulty which I cannot explain” (1985: 196) that neuter i- and u-stems as a rule have o-grade whereas masculines and feminines have e-grade in the The Indo-Uralic verb 397 root because the uninflected neuter form was found in compounds, unlike the nominative in *-s and the accusative in *-m of masculines and feminines. Moreover, it accounts for the frequent instances of *wo after a consonant where the semivowel was restored on the basis of an alternating *w, especially before *i and *r, which were syllabic in the zero grade, e.g. in the words for ‘two’ and ‘four’. We now come to the crucial sound law which identifies Indo-European as a branch of Indo-Uralic: *ti was assibilated to *si (as later happened in Finnish). The principal evidence for this sound law consists of three pieces, viz. the 2nd sg. ending *-si beside *t-, the plural ending *-es beside *-i, and the ablative ending *-os beside *-t. A fourth piece of evidence is the isolated pronoun *sim for *tim (cf. Beekes 1983: 219-224). A fifth piece of evidence is the perfect participle, cf. Greek masc. εἰδότ-, fem. ἰδυῖα < *-us-iq2 < *-ut-iq2 ‘knowing’, Vedic neuter -vát beside -úṣ-. The Proto-Uralic pronouns 1st sg. *mi, 2nd sg. *ti (later *mu, *tu with the suffix *-u ‘self ’), 1st pl. *me, 2nd pl. *te (later *me-i, *te-i with the plural ending *-i) are attested in the corresponding personal endings *-mi, *-ti, *-me, *-te (cf. Collinder 1960: 243, 308, Raun 1988: 562), which can be identified with the corresponding Proto-Indo-European athematic endings *-mi, *-si, *-me, *-te. These endings are directly preserved as *-m, *-s, *-me, *-te in the athematic aorist (II), where the final *-i was lost because it was unstressed. In the athematic present (I) the final *-i was restored on the basis of the independent pronouns at an early stage, while 1st pl. *-me received the additional plural marker *-s and 2nd pl. *-tq1e was taken from the thematic present (IV). When the latter substitution took place, it was evidently more important to distinguish between the present (I, IV) and the aorist (II, III) than between the athematic (I, II) and the thematic (III, IV) flexion, which were already differentiated by the thematic vowel in the latter paradigms. The Proto-Uralic plural suffix was *t in the nominative and *i in the oblique cases (cf. Collinder 1960: 237, 297f., Raun 1988: 557f.). The ending *-i is preserved in the Proto-Indo-European 3rd pl. ending *-nti of the athematic present (I), which evidently represents the original nom.pl. ending of the nt-participle, like Finnish laulavat ‘they sing’ (cf. Collinder 1960: 243), and in the Proto-Indo-European pronoun, e.g. nom. *to-i, gen. *to-i-s-om, dat. *to-i-mus, abl. *to-i-os, inst. *to-i-bhi, loc. *to-i-su (cf. K065: 222). The ending *-i was apparently added to the original nom.pl. ending *-t, which after the loss of unstressed *-i yielded *-s < *-si < *-ti. Thus, the Indo-European ending *-es represents *-eti. The correspondence between Uralic and Indo-European is even closer if Janhunen is right that Proto-Uralic *-i was originally a conjunctive rather than an oblique ending (1982: 29f.) because this explains the IndoEuropean distribution of *-i in the pronoun and the participle versus *-es in the noun. The Indo-European acc.pl. ending *-ns looks like the Proto-Uralic gen.sg. 398 Indo-Uralic ending *-n plus the new plural ending *-s < *-ti. This suggests that it was created as a definite oblique plural ending after *-n had developed into a general oblique singular ending (subsequently yielding n-stems) in Indo-European. Proto-Uralic gen.sg. *-n and acc.sg. *-m were probably limited to definite nouns (cf. Janhunen 1982: 31) and the same must be assumed for the Indo-European acc.sg. ending *-m. Note that the 3rd pl. ending *-nti must be due to restoration because both *t and *-i have been preserved. It was evidently built on the 3rd pl. ending *-nt of the athematic aorist (II), which will be discussed below. The Proto-Uralic ablative suffix *-ta developed into a partitive in Finnish and into an instrumental -l in Ugric, though the latter may have lacked the final vowel (cf. Collinder 1960: 287f., Raun 1988: 559). In Indo-European there is an ablative in *-t which functions as an instrumental in Hittite, which has an ablative in -z < *-t-i. There is another ablative in *-os which also functions as a genitive and earlier apparently as an ergative which became the nom.sg. form of the nominal thematic flexion (cf. Beekes 1985: 176-195). Finally, there is an ablative in *-tos which evidently represents *-t-os (cf. Beekes 1985: 181f.). The abl.pl. ending was probably *-ios (cf. Beekes 1985: 144f.), which reflects plural *-i plus ablative *-os. The simplest explanation of all these endings is that the original Indo-Uralic ablative ending *-ta was replaced by *-ti in its local use in order to differentiate it from its instrumental use and then developed into *-s. This explains why *-t is found as a relic in the ablative of the personal pronouns and the o-stems (where it had to be distinct from the nominative ending *-s) and in the Hittite instrumental, whereas we find *-os in the ablative and genitive of the consonant stems and in the nominative of the o-stems. I think that the same *-t survives in the pronominal ending of the neuter o-stems, reflecting the substitution of the instrumental for the ergative with inanimate agents in transitive constructions. We now return to the Indo-European verbal paradigms cited above. The 3rd sg. ending *-t of the athematic aorist (II) evidently represents the Indo-Uralic demonstrative *t (no. 11), cf. Indo-European *to- (with o-grade from stage C, see above), which was added to the original zero ending. In Uralic (or rather UraloSiberian, cf. Uhlenbeck 1935b, Fortescue 1998, Seefloth 2000) the 3rd sg. pronoun was supplied by the demonstrative *s (no. 12), e.g. Finnish hän, which corresponds to the Indo-European nominative *so (again with o-grade from stage C). The formative suffix of the sigmatic aorist must be derived from the verbal noun in *-s (no. 46, cf. Janhunen 1982: 36). The 3rd sg. ending *-ti of the athematic present (I) is evidently analogical after 1st sg. *-mi and 2nd sg. *-si. The 2nd sg. imperative ending *-dhi may represent the original pronoun *ti with restored *t-. Elsewhere I have compared the difference between the athematic present (I), e.g. Vedic ád-mi ‘I eat’, and the thematic present (IV), e.g. Greek ἔδ-ο-μ-αι ‘I will eat’, with the distinction between Bulgarian spj-a ‘I sleep’ and spi mi se ‘I am The Indo-Uralic verb 399 sleepy’ (K049: 319). While the athematic (subjective) flexion has an agent marker (Vedic -mi, Bulg. -a), the thematic (objective) flexion has a patient marker (Gr. -o-, Bulg. zero), an experiencer (Gr. -m-, Bulg. dative mi), and a reflexive marker (Gr. -ai, Bulg. acc. se). It has long been recognized that there is a correlation between thematic flexion and middle voice, as opposed to an athematic active paradigm, in the oldest Indo-European material (cf. Thieme 1929: 53, Renou 1932: 21). I therefore think that the thematic present endings (IV) represent a combination of object, recipient, and reflexive marking. The thematic aorist endings (III) evidently combine the object marker *-e/o- with the agent markers of the athematic aorist (II). When we compare the thematic present endings with the Indo-Uralic morphemes listed above, the obvious candidate for the 3rd sg. ending *-e is the demonstrative *e beside *i (no. 8). The characteristic laryngeal *q1 of the nonthird persons may perhaps be compared with the dative marker *ka (no. 26). Note that the 1st pl. ending *-omom may actually represent *-omq1om, so that the non-third person plural endings may contain Indo-Uralic *me-ka ‘to us’, *te-ka ‘to you’. Since the double full grade vowel in the endings *-omq1om, *-etq1e cannot be original, the addition of final *-om and *-e must have been recent. The final vowel of the 2nd sg. ending *-eq1i may have been taken from the athematic present, perhaps in order to disambiguate it from the derivative suffix *-eq1 which is found e.g. in the Greek passive aorist. The addition of this final *-i must obviously have been more recent than the grammatical differentiation between athematic present and aorist. Thus, we may reconstruct the following paradigm for the thematic present at an early stage (IVa): 1st sg. 2nd sg. 3rd sg. -o-q1 -e-q1 -e 1st pl. 2nd pl. 3rd pl. -o-mq1 -e-tq1 -o This paradigm must be examined in relation to the perfect (V) and the stative (VI). We may wonder if the thematic present must properly be called transitive or intransitive. I think that this is largely a matter of terminology. In the Bulgarian example spi mi se ‘I am sleepy’, which contains three person markers, a clearly intransitive situation is described by the reflexivization of a 3rd sg. intransitive verb form with the sole real participant in the dative. I claim that the same construction is found in Indo-European not only in the thematic present, but also in the perfect and the stative. As in the thematic present, I think that we have a patient marker and an experiencer in the perfect. If the agent was mentioned, it was probably in the dative if it was animate and in the instrumental if it was inanimate (cf. K049: 321). Here again, the obvious candidate for the 3rd sg. ending *-e is the demonstrative *e and the characteristic laryngeal *q2 of the non-third persons 400 Indo-Uralic may be compared with the dative marker *ka. The reconstructed endings 1st pl. *-me, 2nd pl. *-e may actually represent *-mq2e, *-q2e (cf. K033: 68), which yields the following paradigm for the perfect at an early stage (Va): 1st sg. 2nd sg. 3rd sg. -q2-e -tq2-e -e 1st pl. 2nd pl. 3rd pl. -mq2-e -q2-e -r This paradigm differs from the thematic present first of all in the order of the constituent morphemes. It is reasonable to assume that the first part of the ending belongs more closely with the preceding stem while the second part has a more independent status. If we simplify matters by substituting ‘I have’ for the dative ‘to me’, we may paraphrase the thematic present as “I have it being changed” and the perfect as “it is me having changed”. The distribution of *m and *t suggests that these are person markers and that number was originally unmarked, perhaps because the following vowel was lost by the Indo-European vowel reduction (stage A). In the thematic present, *m and *t could be confined to the plural on the analogy of the athematic flexion because first and second person were already distinguished by the thematic vowel. In the perfect, the same distribution is found in the first person, but not in the second, where the 2nd sg. form was obviously much more frequent than its plural counterpart. The remarkable elimination of the person marker in the plural ending suggests that it was disambiguated from the singular ending, which then must have been homophonous at the time. This brings us back to the distribution of velars and uvulars in Indo-Uralic. If the Indo-European distinction between palatovelars and labiovelars arose when the distinctive timbre of the following vowel was lost and the uvulars developed from velars before back vowels, we expect e.g. *kʄ < *ki, *kw < *kü, *q2 < *kï, *q3 < *ku. Note that *q1 has a special position because it does not color a contiguous vowel and is automatic if there is no other word-initial consonant. It has often been identified with a glottal stop. We may then hypothesize that it developed from *k if no vowel followed. Interestingly, there is some evidence for reduction of laryngeals in word-final position. The Indo-European vowel reduction changed the root structure from *CV(C)CV- into *CV(C)C- and, consequently, the suffixal structure from *-CV- into *-VC-, with full grade *e under the stress, shwa secundum in unstressed closed syllables, and zero in unstressed open syllables. Final clusters ending in a laryngeal may have originated from medial clusters of any consonant plus *k, which were particularly frequent (cf. Sammallahti 1988: 492). This accounts for the peculiar loss of laryngeals in compounds and o-grade formations, where the final laryngeal was lost before the initial consonant of the second component (cf. Hirt 1921: 185-187). Thus, I think that the particle *gʄhi, the k-perfect of Greek and Latin, and the laryngeals *q1 in the thematic present and *q2 in the perfect all go The Indo-Uralic verb 401 back to the same element, which appears as -k or -ka in Uralic, often followed by other suffixes (cf. Collinder 1960: 296, Raun 1988: 560, also Fortescue 1998: 115). The principal difference between the stative (VI) and the perfect (V) is the 3rd sg. ending *-o instead of *-e. Since the stative was used to supply a middle paradigm, I think that the ending can be identified with the Indo-Uralic reflexive *u/w (no. 54), which yielded *-o in Indo-European (stage B). If *-e was a patient marker and the preceding element an experiencer in the perfect, the stative is structurally comparable with the Bulgarian example spi mi se ‘I am sleepy’. Since the *-o is absent from the 1st sg. as well as the 1st pl. and 2nd pl. endings, we must conclude that it was originally limited to the third person. The final vowel of the 2nd sg. ending *-tq2o can easily have been taken from the 3rd sg. form. The reconstructed 2nd pl. ending *-dhwe may actually represent *-dhq2we (cf. Melchert 1984: 26 and Kloekhorst 2008: 899), which yields the following paradigm for the stative at an early stage (VIa): 1st sg. 2nd sg. 3rd sg. -q2 -tq2 -o 1st pl. 2nd pl. 3rd pl. -medhq2 -dhq2-we -r-o The corresponding transitive paradigm, where the endings were preceded by an agent marker, is the following (VIIa): 1st sg. 2nd sg. 3rd sg. -m-q2 -s-tq2 -t-o 1st pl. 2nd pl. 3rd pl. -me-dhq2 -t-dhq2-we -nt-r-o This explains the correlation between thematic flexion and middle voice, as opposed to the athematic active paradigm, in the oldest Indo-European material. The suffixation of the Indo-Uralic reflexive element *u/w to verbal stems yielded intransitives, middles and passives in Uralic (cf. Collinder 1960: 281). In Indo-European, it seems to have developed an oppositional meaning in relation to first person *m, as is especially clear in the pronouns, cf. acc. *q1-me ‘this-me’ versus *t-we ‘thee-self ’, *s-we ‘him-self ’, possessive *q1-mos ‘this-my’ versus *t-wos ‘thy-own’, *s-wos ‘his-own’, also nom. *q1e-gʄ- ‘I’ versus *t-u- ‘thou’. This explains why *-o spread to the 2nd sg. ending *-tq2o but not to the 1st sg. ending *-q2. It also explains the addition of *-we in the 2nd pl. ending. The elements 1st pl. *-medh- and 2nd pl. *-(t)dh- can be understood as replacements of earlier *-m- and *-(t)- before *-q2 in order to mark the plural subject of the stative. They can be identified as the absolutives (intransitive nominatives) *me-t and *te-t which were introduced when the laryngeal had lost its original function and become a simple voice marker. This development could not take place in the perfect as long as *-e functioned as a subject marker. 402 Indo-Uralic In the first person, *u/w is found instead of *m in the dual endings (cf. K145). It is probable that the difference between these two morphemes reflects an original distinction between inclusive and exclusive first person forms, *u/w meaning ‘you and I, ourselves’ and *m meaning ‘we as opposed to you’ (cf. Ivanov 1981: 21 and K223). We also find *u as a deictic element connected both with the meaning ‘self ’, as in Greek αὐτός, and with the second person, in opposition to *kʄi for the first person (cf. K063). This meaning of *u/w as a person marker which distinguishes its referent from the first person *m now explains the substitution of *o for *e as the patient marker in first person forms of the thematic flexion. Thus, the meaning of the 1st sg. thematic endings *-om and *-oq1 can be paraphrased as ‘other by/to me’ whereas 2nd sg. *-es, *-eq1i and 3rd sg. *-et, *-e represent ‘this by/to thee/him/her’. The final vowel of the thematic aorist endings 1st pl. *-omo and 2nd pl. *-ete looks like a copy of the thematic vowel, and a similar origin may be suspected for the addition of final *-om and *-e in the corresponding thematic present endings. If we call the thematic vowel *-e/o- an object marker and the perfect ending *-e a subject marker, we can now summarize the general structure of the seven paradigms discussed above as follows: Stem-object-agent-recipient-subject-reflexive This structure may reflect the original order of the clitics from which the endings developed. The chronology of the developments can largely be deduced from the vocalism of the endings. The athematic aorist endings 1st pl. *-me, 2nd pl. *-te probably replaced *-m, *-t shortly after the Indo-European vowel reduction (stage A) on the basis of the independent pronouns and thereby introduced mobile stress in the verbal paradigm. The athematic present endings apparently developed in order to distinguish the actual present from the timeless aorist in imperfective verbs. The thematic aorist was the corresponding objective flexion, indicating a definite object. The substitution of *o for *e in the first person cannot have taken place before the introduction of *o in stressed syllables (stage D). The thematic present supplied an actual present for the objective flexion of imperfective verbs. It supplied a subjunctive after the introduction of *e in unstressed syllables (stage E) because this category has e-grade in the root. The perfect had final stress but introduced full grade in the root (stage C) and subsequently retracted the stress (stage D) in the singular forms, evidently on the analogy of the athematic present and aorist. The stative had e-grade in the root and developed the ending *-o by the phonetic lowering of *u in unstressed syllables (stage B). The middle aorist cannot have originated before the introduction of *o in stressed syllables (stage D) because it has zero grade in the root. The middle present supplied a dynamic counterpart to the stative after the introduction of *o in stressed syllables (stage D) and a subjunctive after the introduction of *e in unstressed syllables (stage E). The Indo-Uralic verb 403 The 3rd pl. endings have not yet been discussed because their deviant accent pattern betrays a separate origin (cf. K065: 222). Pedersen already pointed to the possibility of identifying 3rd pl. “intransitive” *-r and “transitive” *-nt with the formative suffix of Greek nom. ὕδωρ ‘water’, oblique ὕδατ- < *-nt- (1933b: 313). Both *r and *nt are found as formatives in neuters, collectives, and adjectives (cf. Benveniste 1935: 123-128). Interestingly, the accent of the 3rd pl. forms agrees with the oblique cases of the neuter, not with the nominative (cf. K195: 71). I therefore think that the 3rd pl. forms may be compared with English awry < on wry rather than wry ‘turned, twisted’. This explains why the stop in *-nt(i) was not assibilated to *s. The same construction is found in the singular of the perfect in *-ēu (cf. K099: 111), which represents the locative form of the u-stem from which the participle in Vedic -uṣ-, Greek -οτ- < *-ut- is derived. When we compare the 3rd pl. ending *-(e)nt with English -ing in agoing ‘in motion’, the corresponding perfect form in *-(ē)r can be compared with English asleep and identified as a nomen loci with the locative suffix *ru (no. 29). The Avestan ending -rš apparently added the nom.pl. ending *-s after *-r. No such explanation is possible for the thematic present ending *-o, which must be derived from the reflexive marker *u/w. There evidently was an impersonal form with the reflexive *u in object position which supplied a 3rd pl. form to the thematic present. If we may paraphrase the original meaning of the 1st sg. thematic present as “I have it being changed” and of the 1st sg. perfect as “it is me having changed”, the 3rd sg. forms can be derived from “there is it being changed” and “there is it having changed”. We can then identify the 3rd pl. form of the thematic present as “there is being changed”, with reflexive *-o replacing deictic *-e and thereby eliminating the 3rd sg. reference. There appear to have been no original 3rd pl. verb forms in Indo-Uralic. I conclude that the Indo-European verbal system can be understood in terms of its Indo-Uralic origins. Most importantly, the reconstructed endings can be derived from combinations of Indo-Uralic morphemes by a series of well-motivated phonetic and analogic developments. The component parts of the endings either represent original morphemes (-m, -s, -t, -me, -te, -nt, -q1, -q2, -e, -o, -r, -t-, -dh-) or were introduced for disambiguation from other endings. NIVKH AS A URALO-SIBERIAN LANGUAGE In his magnificent book on the language relations across Bering Strait (1998), Michael Fortescue does not consider Nivkh (Gilyak) to be a Uralo-Siberian language. Elsewhere I have argued that the Indo-European verbal system can be understood in terms of its Indo-Uralic origins (K203). All of these languages belong to Joseph Greenberg’s Eurasiatic macro-family (2000). In the following I intend to reconsider the grammatical evidence for including Nivkh into the Uralo-Siberian language family. The Indo-Uralic evidence is of particular importance because it guarantees a time depth which cannot otherwise be attained. Nivkh initial consonants are subject to mutations which are strongly reminiscent of Celtic. Adopting Ekaterina Gruzdeva’s transcription (1998) except for the uvulars q, q‘, gɷ, xɷ, γɷ and the velar nasal ŋ, we can summarize the alternations as follows: (1) p, t, ť, k, q become v, r, z, γ, γɷ after a vowel (which may have been lost) and analogically in certain categories; (2) p, t, ť, k, q become b, d, ď, g, gɷ after a nasal (which may or may not be lost); (3) p, t, ť, k, q become f, ř, s, x, xɷ after the 2nd sg. prefix č- and the reflexive prefix p‘-; (4) p‘, t‘, č, k‘, q‘ become f, ř, s, x, xɷ under the same conditions as (1) and (3) but remain unchanged under the conditions of (2); (5) p‘, t‘, č, k‘, q‘ become v, r, z, γ, γɷ after the 3rd sg. prefix i-/e- in ablauting and cluster-initial verbs, e.g. iγď ‘kills’ of the verb k‘u- ‘kill’ (cf. Krejnovič 1958: 23f., Mattissen 2001: 142-146). On the basis of these alternations I reconstruct *VC for the voiced fricatives, *VNC for the voiced stops, *VHC for the voiceless fricatives, and *HC for the aspirated stops, where *C represents p, t, ť, k, q. Moreover, *VNHC > *VHC and *VHCC > *VCC. If *H developed from *h < *s, this brings the original consonant system rather close to the one reconstructed for Proto-Uralic (cf. K203: 220). Among the morphological elements for which I have suggested an IndoUralic origin (K203: 218f.), the following are likely candidates for a comparison with Nivkh (I retain the numbering of Greenberg 2000): 1. first person *m, 4. second person *t, 8. demonstrative *i/e, 11. demonstrative *t, 12. demonstrative *s, 406 Indo-Uralic 14. dual *ki, 15. plural *t, 25. genitive *n, 44. participle *nt, 45. participle *l, 46. verbal noun *s, 54. reflexive *u/w. Other possible connections are less convincing. Nivkh case markers and postpositions appear to have a lexical source (cf. Panfilov 1962: 143-156 and Mattissen 2001: 93). The Uralic participle in *-pa (Collinder 1960: 270) and the Nivkh gerund in *-pa (Panfilov 1965: 145) may represent the only Uralo-Nivkh formation without an Indo-European cognate unless they are related to the root of the English verb ‘to be’. The reconstruction of interrogative *k and relative *j is highly questionable. The principal evidence for the Uralo-Siberian character of the Nivkh language is provided by the pronominal elements *m, *t, *i/e, *t, *s, *u/w. The personal pronouns are the following (cf. Gruzdeva 1998: 25f.; my reconstructions): 1st sg. ńi: *ńi, 1st du. megi/mege, meŋ, memak: *meŋki, 1st pl. incl. mer/mir, meřn/miřn, min: *mer, 1st pl. excl. ńyŋ, ńin: *ńiŋ, 2nd sg. či: *či, 2nd pl. čyŋ, čin: *čiŋ, 3rd sg. if, i, jaŋ: *iw, 3rd pl. imŋ, ivŋ, imγ, iřn, in: *iwŋ. These paradigms can be derived from the Indo-Uralic pronouns 1st sg. *mi, 1st du. *men-ki (‘the two of us’), 1st pl. *me-t, 2nd sg. *ti, 3rd sg. *i/e (cf. K203: 221f.), where sg. *mi, *ti, pl. *me are the reconstructed stems and *-n, *-ki, *-t are the suffixes for genitive, dual and plural mentioned above. The Indo-Uralic cognates suggest that Nivkh ńi and či developed phonetically from *mi and *ti, respectively. While the latter development is commonplace, the former is reminiscent of Czech [mń] < mj-, e.g. in město [mńesto] ‘city’, also (as Jos Schaeken reminds me) North Russian [ń] < [mń] < *-mj-, e.g. na zeni ‘on the ground’, na zeń ‘to the ground’ (Zaliznjak 1995: 62, Honselaar 2001: 23). It has been suggested that the pronominal stem me- must be derived from the numeral stem me- ‘two’ (cf. Austerlitz 1959: 109, Panfilov 1962: 205f.). This is highly improbable because it does not explain the occurrence of me- in the plural, the semantic contribution of the suffixes, and the unexpected order of the pronominal and the numeral element (cf. Greenberg 1997: 192), cf. also megi Nivkh as a Uralo-Siberian language 407 men ‘we two’ (Panfilov 1962: 233), which cannot possibly be glossed as *‘two-du. two’. I reconstruct *iw for 3rd sg. if, oblique stem iv-, because this pronoun is limited to the Amur dialect, where *w > v (cf. Gruzdeva 1998: 11), and corresponds to i, oblique stem j- in Sakhalin (cf. Mattissen 2001: 20). It seems attractive to derive this *-w from the Indo-Uralic reflexive element *u/w, which may have been used as a reinforcement of the 3rd sg. pronoun *i/e. The IndoUralic demonstratives *t- and *s- are reflected in the Nivkh demonstrative stems t- ‘this’ and h- ‘that’ (cf. Gruzdeva 1998: 26, Mattissen 2001: 21). Apart from the personal pronouns, there are personal prefixes which denote the possessor of a following noun or the undergoer of a following verb form (cf. Mattissen 2001: 62ff. for the distribution of the allomorphs): 1st sg. ń-, ńi-, ńe-, n-, 2nd sg. č-, či-, če-, t‘-, 3rd sg. i-, v(i)-, j-, e-, reflexive p‘-, p‘i-, p‘e-, reciprocal u-, v-, o-. On the basis of the alternations in the root-initial consonant I reconstruct the following paradigms (cf. Mattissen 2001: 66-69): paxɷ ‘stone’, ńvaxɷ ‘my stone’ < *mi-, čfaxɷ ‘your stone’ < *tis-, p‘faxɷ ‘one’s own stone’ < *pis-, ibaxɷ ‘his/her stone’ < *in-, ńyŋbaxɷ ‘our stone’ < *minkun-, ńzaď ‘beats me’ < *mi-, čsaď ‘beats you’ < *tis-, p‘saď ‘beats him/herself ’ < *pis-, zaď ‘beats someone’ < *i-, iďaď ‘beats him/her’ < *in-, ńyŋďaď ‘beats us’ < *minkun-, where *-kun is the plural suffix (cf. Gruzdeva 1998: 16, Greenberg 2000: 116). The 3rd sg. possessive prefix v(i)-, which is limited to the Amur dialect, apparently represents *iwin- and may have been introduced when initial *i- was lost. The final nasal of possessive *in- may represent the original genitive suffix *-n and may have been introduced into the verbal prefix for emphasis (cf. Mattissen 2001: 65). The reconstruction of final *-s in *tis- and *pis- accounts both for the following voiceless fricative and for the aspiration of the preceding plosive after the syncope of the intervening vowel. The labial element of p‘i- has no obvious etymology. It is reminiscent of Latin ipse ‘self ’, but also of dative sibī, Greek pl. σφι < *sbhi beside sg. οἱ < *swoi 408 Indo-Uralic (for earlier *sui, cf. K203: 221), Slavic sebě, of which p‘i- could be the phonetic reflex, further Prussian sups ‘self ’, Gothic sibja ‘clan’, silba ‘self ’, cf. Greek φίλος ‘dear’, Indo-European *bhi ‘near’, which is perhaps related to Nivkh fiď ‘be in a place’, p‘iŋ ‘inhabitant’ (Greenberg 2000: 146). The prefix p‘i- may have ousted ufrom its original reflexive function into secondary reciprocal use, as in Russian oni celujutsja ‘they kiss each other’ but oni celujut sebja ‘they kiss themselves’, cf. Nivkh ozmuď ‘love each other’ versus p‘ezmuď ‘love him/herself ’ (Panfilov 1965: 52). This idea is supported by the possibility of identifying the labial of the 3rd sg. possessive prefix v(i)- < *iwin- with the reciprocal verbal prefix u-, as was suggested above. It allows us to identify the latter with the Indo-Uralic reflexive morpheme *u/w (cf. K203: 224). The verbal ending -ď, -ť, -d, -nd, -nt, -t < *-nt(i) (Gruzdeva 1998: 22, 33) can be identified with the Indo-Uralic participial suffix *nt (cf. Collinder 1960: 269f., 277f., Greenberg 2000: 184f.). It is found in finite and infinite verb forms and in participles and verbal nouns in Nivkh, Uralic, and Indo-European (cf. Panfilov 1962: 64-68, 1965: 153f., K203: 226) and is therefore a strong piece of evidence for a common origin. Another participial formation which may have been inherited are the verbal adjectives in *-l- (cf. Panfilov 1965: 85-88, Greenberg 2000: 190), e.g. Latin bibulus, Nivkh raxyla ńivx ‘drinker, drunkard’, as opposed to ra ńivx ‘drinker, person drinking’. This suffix is found in iterative verbs in Uralic (cf. Collinder 1960: 275f.). Finally, the Indo-Uralic verbal noun in *-s- (cf. Collinder 1960: 271, Greenberg 2000: 191f.) may be reflected in the Nivkh deverbal nouns in -s/-ř < *-s-t(i) denoting subject, object, instrument or result and in -f < *-s-p(i) denoting place of action or result (Panfilov 1962: 41-48, Gruzdeva 1998: 22). The labial of the latter suffix may again be identified with Indo-European *bhi ‘near’ and with the root of the Nivkh verb fiď ‘be in a place’. Thus, I think that we have strong indications of a close relationship between Nivkh and IndoUralic. The relations between these and the other Uralo-Siberian languages remain to be clarified (cf. also Bouda 1960, Tailleur 1960, Naert 1962). INDO-URALIC CONSONANT GRADATION Koivulehto and Vennemann have recently (1996) revived Posti’s theory (1953) which attributed Finnic consonant gradation to Germanic influence, in particular to the influence of Verner’s law. This theory disregards the major differences between Finnic and Saami gradation (cf. Sammallahti 1998: 3) and ignores the similar gradation in Nganasan and Selkup (cf. Kallio 2000: 92). Janhunen recognizes that the Proto-Uralic stress pattern “divided the word in two-syllable sections with initial stress, with the main stress on the first section of the word”: (C)É(C)-CE(C)-CÈ(C)-CE(C) and asserts that this phenomenon “has convergently led to important phonotactic and morphophonemic developments”, especially the consonant gradation (1981: 27). I rather agree with Helimski, who maintains that “we are left with only two options: to believe in wonders capable of producing most incredible coincidences in related or unrelated languages – or to regard the consonant gradation found in FinnicLapp and in Nganasan, both in its rhythmic and syllabic forms, as a PUr. phenomenon” (1995: 28 = 2000: 176). On the basis of Janhunen’s Proto-Uralic stress pattern cited above, we may call odd syllables “strong” and even syllables “weak”, counting from the beginning of a word form. Helimski’s “rhythmic” and “syllabic” gradations can now be defined as follows (cf. 1995: 24-26 = 2000: 172-174): I. A consonant which follows the vocalic nucleus of a weak syllable is weakened. II. A consonant which precedes the vocalic nucleus of a closed weak syllable is weakened. These two rules are ordered because a closed weak syllable which becomes open by losing its coda as a result of I is no longer subject to II. Helimski shows that intervocalically the first rule yielded voiced fricatives and the second voiced stops both in Nganasan and in Finnic (1995: 31-33 = 2000: 178-179). It appears that the original situation is best preserved in North Saami, where non-weak consonants were strengthened (cf. Sammallahti 1998: 47-50). Note that the rules I and II yielded a subphonemic alternation between strong and weak consonants which was dependent on the stress pattern and could be either phonemicized or lost as a result of later developments. Elsewhere I have argued that the Indo-European verbal system can be understood in terms of its Indo-Uralic origins because the reconstructed IndoEuropean endings can be derived from combinations of Indo-Uralic morphemes by a series of well-motivated phonetic and analogic developments (2002). In the same vein I claim that the Proto-Uralic consonant gradation 410 Indo-Uralic accounts for the peculiar correlations between Indo-European root structure and accentuation discovered by Lubotsky (1988a). The facts to be explained are the following: (1) Proto-Indo-European had three series of stops, which are traditionally considered to be voiceless, voiced, and voiced aspirated. There is reason to assume that the plain voiced stops were actually preglottalized (cf. K075) while the voiced aspirates may not have been aspirated. In order to avoid confusion I shall write *T, *ȤD, *Dh for the three series and call them fortes, glottalics and aspirates, respectively. (2) Proto-Indo-European roots with two stops could not contain two glottalics, so that **ȤDE(R)ȤD- is an impossible root structure. Moreover, fortes and aspirates could not co-occur in the same root, so that **TE(R)Dh- and **DhE(R)T- are also excluded. It follows that the distinction between fortes and aspirates was a prosodic feature of the root as a whole, which may be called “strong” if it contained *T and “weak” if it contained *Dh. (3) Dybo has shown (1968) that Baltic and Slavic morphemes can be divided into two prosodic classes, viz. “strong” morphemes which attract the accent and “weak” morphemes which repel the accent, and that the stress falls on the first strong morpheme of a word form. If a word form contains weak morphemes only, it has initial stress unless it can be cliticized as a whole to the preceding word form (cf. Lubotsky 1988a: 3). This rule was perhaps inherited from ProtoIndo-European. It raises the question if the “strong” and “weak” consonants and morphemes of Indo-European can be related to the “strong” and “weak” consonants and syllables of Proto-Uralic. I think that this is indeed the case. Lubotsky divides the Indo-European roots into four categories, viz. roots without stops, roots with a single stop and no initial laryngeal, roots with a single stop and an initial laryngeal, and roots with two stops (1988a: 14). It turns out that derivatives of roots without stops and derivatives of roots with an initial laryngeal and a stop which is contiguous to the syllabic nucleus are either barytones with full grade in the root or oxytones with zero grade in the root, which points to an ancient correlation between ablaut and accentuation. However, in the case of derivatives of roots with a stop which is contiguous to the syllabic nucleus but without an initial laryngeal, it becomes apparent that o-stems are barytone if the root contains *T and oxytone if the root contains *ȤD or *Dh whereas i- and u-stems are oxytone if the root contains *T and barytone if the root contains *ȤD or *Dh, regardless of the ablaut grade of the root (Lubotsky 1988a: 169-170). This highly peculiar distribution requires an explanation. Indo-Uralic consonant gradation 411 Elsewhere I have proposed the following relative chronology for the IndoEuropean branch of Indo-Uralic on the basis of the internal evidence (2002: 221): A. Indo-European vowel reduction, giving rise to full grade *e under the stress and zero grade elsewhere; B. phonetic lowering of *u (= syllabic *w) to *o, giving rise to a full grade (= non-high) vowel in unstressed syllables; C. analogical introduction of a full grade vowel in unstressed syllables (e.g. in compounds), which automatically yielded new *o; D. introduction of *o in stressed syllables (e.g. by decompounding), resulting in a phonemic opposition between /e/ and /o/ under the stress; E. analogical introduction of full grade *e in unstressed syllables, generalizing the opposition between /e/ and /o/; F. rise of lengthened grade vowels *ē and *ō, yielding the conventional ProtoIndo-European vowel system. The remaining problems are the original place of the stress, the rise of new consonant clusters, and the distribution of the stops. These problems can be solved by the following rule, which must be inserted after I and II but before A-F and thereby separates Indo-European from the Indo-Uralic proto-language: III. An open strong syllable becomes weak and loses its (primary or secondary) stress to the following syllable, which becomes strong if it is closed (but not if it is open). As a result, rule A yields full grade *e under the stress, which falls on the first strong syllable of a word form, shwa secundum in unstressed closed syllables, and zero in unstressed open syllables. The loss of initial and medial open syllables gave rise to new consonant clusters while full vowels in open syllables could only be preserved word-finally. The expected distribution of fortes and aspirates can now be specified as follows. The two types of Uralic weak stops (before and after the vocalic nucleus of a weak syllable) apparently merged into the Indo-European aspirates while the Uralic strong stops (before the vocalic nucleus of an open syllable) became the Indo-European fortes. Initial stops adopted the same pattern, which resulted in a consonant alternation in roots with fortes before zero grade suffixes and aspirates before full grade suffixes, e.g. *tekm, *dhghem- ‘earth’. After the analogical introduction of *o in unstressed syllables at stage C, we obtain paradigms like the following (cf. Beekes 1995: 178): 412 Indo-Uralic nom. acc. loc. abl. inst. ‘winter’ *gheiom *ghiem(m) *ghiem(i) *ghimes *ghimet ‘grandson’ *nepot *nepot(m) *nepot(i) *neptos *neptot In the nominative *gheiom, which replaced *keim, the full grade suffix was apparently introduced from the oblique form at this stage, while nom. *nepot and obl. *nepot- may represent an earlier paradigm *nept, *nbhedh-, with generalization of fixed stress on the initial syllable. These examples show how fortes and aspirates could become associated with fixed and mobile stress patterns, respectively. As a result of developments in the verbal system, the nominative had now been replaced by the ablative of animate nouns and the instrumental of inanimate nouns occupying the subject position of transitive verbs, so as to yield an ergative system (cf. K203, with references). After the analogical introduction of stressed *o at stage D, the ergative in *-os, with generalized o-grade replacing e-grade in paradigms with mobile stress, developed its own paradigm, which resulted in the thematic flexion (cf. Beekes 1985: 191-195). This paradigm had zero grade vocalism in the root at that stage. Since fortes and aspirates were now associated with barytone and oxytone stress, respectively, we find a discrepancy between the ablaut grade (which was determined by the derivation) and the accentuation (which was determined by the root structure). The Indo-European proto-language developed an opposition between agent nouns with final stress (reflecting the original ergative) and action nouns with radical stress (representing earlier root nouns), both with o-grade in the root. We can assume that this development started before unstressed e-grade was introduced at stage E. In the historical material, the ablaut grade is evenly distributed over barytones and oxytones while the accentuation is still closely linked to the consonantal root structure. Counting the certain examples of o-stem derivatives with a single contiguous stop and no initial laryngeal in Greek, we find 4 instances of e-grade in the root, all of them oxytones with *ȤD or *Dh and therefore clearly secondary, 11 instances with o-grade, 10 instances with zero grade, and 3 oxytones with an a-diphthong in the root (cf. Lubotsky 1988a: 138). Thus, we clearly have to assume original zero grade in this formation, regardless of the root structure. The situation with the i- and u-stems was different because the oblique form of the suffix *-ey, *-ew was evidently strong at the outset, so that the analogical introduction of unstressed *o at stage C yielded paradigms like the following: Indo-Uralic consonant gradation nom. acc. loc. abl. inst. ‘arm’ *bheȥghu *bhȥgheu(m) *bhȥghew(i) *bhȥgheus *bhȥgheut 413 ‘thin’ *tenȥu *tenȥou(m) *tenȥow(i) *tenȥous *tenȥout These paradigms may have replaced earlier *peȥku, *bhȥghew- and *tenȥu, *dhnȥew-, respectively. The differentiation between an ergative in *-s and an ablative in *-os in paradigms with fixed stress after stage C now gave rise to a new ablative *tenȥuos beside the ergative *tenȥous whereas paradigms with mobile stress created a new ergative in *-is, *-us beside the ablative in *-eis, *-eus, later also a new accusative in *-im, *-um. After stage D, accentual mobility with radical ablaut could be restored in paradigms with fortes in the root, yielding stressed *-oi-, *-ou- in the accusative and the ergative and zero grade *-i-, *-u- before the stressed ending of the ablative *-os (cf. Beekes 1995: 181). There was no differentiation between ergative and ablative (nor between nominative and accusative) in the neuter gender, where the instrumental was used instead of the ergative with transitive verbs. We have now arrived at the paradoxical stage where original paradigms with fixed stress have developed full accentual mobility through the creation of a new ablative in *-os beside the ergative in *-s after a full grade suffix while original paradigms with mobile stress tend to become barytones through the creation of a new ergative in *-s and an analogical accusative in *-m on the basis of the rootstressed nominative beside the original ablative in *-s after a stressed suffix. This explains Lubotsky’s remarkable discovery that i- and u-stems are oxytone if the root contains *T and barytone if the root contains *ȤD or *Dh. It is an indirect consequence of the Indo-Uralic consonant gradation. Lubotsky states that 12 of the 14 i- and u-stems with *T in the root have radical zero grade in Sanskrit (1988a: 174). In fact, all 26 certain examples of i- and u-stem derivatives of triconsonantal roots with a single contiguous stop and no initial laryngeal in Sanskrit have radical zero grade except 2× harṣ- beside hṛṣ- ‘rejoice’ (cf. Lubotsky 1988a: 55). The 16 instances with biconsonantal roots also have zero grade except 2× tan- ‘spread’ and carú-, hári-, hánu-, jásu-, sáhyu-, mádhu-, váhni-, and yájyu- beside íṣṭi- ‘sacrifice’. All of these formations would be morphologically awkward if they had radical zero grade, so that we can safely assume analogical restoration of full grade in the root in these instances. There was no such reason to introduce an analogical full grade of the root in bhṛtí-, dṛɴti-, ghṛɴṇi-, jíti-, gáti- or derivatives of triconsonantal roots. It follows that all of these i- and u-stem derivatives may have had original zero grade in the root, regardless of the root structure, as was the case with the o-stem derivatives. This supports the view expressed above that the case forms in *-is, *-us, *-im, *-um 414 Indo-Uralic replaced earlier ergatives and accusatives with full grade suffixes on the analogy of an earlier root-stressed nominative. Thus far I have left the rise of the glottalic consonants out of consideration because this problem requires a separate treatment. I suspect that the root-final glottalics reflect original consonant stems (cf. in this connection Helimski 1995: 31 = 2000: 178). This is in accordance with the word-final neutralization of the Indo-European stops into glottalics, e.g. Latin quod, Old High German hwaz ‘what’, and would explain the virtual absence of the glottalics in Indo-European word formation. For the root-initial glottalics I think of prefixes which may have left a trace in the glottalization (cf. in this connection Rousseau 1990). The matter cannot be pursued here. The theory of Indo-Uralic consonant gradation proposed here offers an explanation for several other sets of data which remain to be explored. An obvious example is the alternation between fortes and aspirates in IndoEuropean word formation, e.g. *-tro-, *-tlo- beside *-dhro-, *-dhlo-. A less obvious example is the alternation between the suffixes -ok- and -k- in Russian vysókij ‘high’, širókij ‘broad’, glubókij ‘deep’, dalëkij ‘distant’ and nízkij ‘low’, úzkij ‘narrow’, mélkij ‘shallow’, blízkij ‘near’, which are accentually strong and weak, respectively (cf. Dybo 1968: 155-158 on the latter). The difference between original *-o(k)- and *-u(k)- can be derived from an Indo-Uralic alternation between anteconsonantal (IE strong) *-w- which was syllabified to *-u- at stage A and lowered to *-o- at stage B and antevocalic (IE weak) *-w- which remained consonantal at stage A and was syllabified to *-u- at a later stage. The semantic differentiation is secondary, as is clear from Lithuanian platùs ‘broad’, gilùs ‘deep’ versus siaũras ‘narrow’, žẽmas ‘low’. The corresponding front vowel suffix *-ik(Russian -c-) is strong in Slavic (cf. Dybo 1968: 174-181), evidently because *-iwas not lowered to *-e- at stage B and the strong variant was generalized. The suffixes *-in- and *-isk- were weak in Baltic and Slavic (ibidem: 152-155 and 214-216). INDO-URALIC AND ALTAIC Elsewhere I have argued that the Indo-European verbal system can be understood in terms of its Indo-Uralic origins because the reconstructed IndoEuropean endings can be derived from combinations of Indo-Uralic morphemes by a series of well-motivated phonetic and analogic developments (K203). Moreover, I have claimed (K213) that the Proto-Uralic consonant gradation accounts for the peculiar correlations between Indo-European root structure and accentuation discovered by Lubotsky (1988a). My reconstruction of the Indo-Uralic phonological system is essentially the same as Sammallahti’s for Proto-Uralic (1988), except for the fact that I reconstruct palatalized resonants *r' and *l' for his dental spirants *ð and *ð'. In particular, I think that the large number of Indo-European plosives is the result of a secondary development. Though it is quite possible that Indo-Uralic had a larger number of consonants than can be reconstructed for Proto-Uralic, I see no compelling evidence for this. The simplest assumption is that the Indo-Uralic protolanguage was identical with Proto-Uralic. Indeed, it seems possible to derive Nivkh (Gilyak) from the same proto-language, as I have indicated elsewhere (K205). As far as I can see, both Indo-Uralic and Nivkh (Gilyak) belong to the larger Uralo-Siberian language family which is now partly reconstructed by Fortescue (1998) and Seefloth (2000) on the basis of evidence from Uralo-Yukagir, Chukotko-Kamchatkan and Eskimo-Aleut. All of these languages belong to Greenberg’s Eurasiatic language family (2000), which in addition comprises Altaic (Turkic, Mongolian, Tungusic), Korean, Japanese and Ainu. There can nowadays be little doubt about the reality of an Altaic language family including Korean and Japanese (see especially Robbeets 2003), though the reconstruction of Proto-Altaic is extremely difficult because of its very large time depth. The position of Ainu remains unclear (at least to me). It is easy to criticize Greenberg’s methodology, but this leaves the basic question about the correctness of his Eurasiatic hypothesis open (see now Georg & Vovin 2003). One should regard his list of grammatical elements, like Pokorny’s Indo-European dictionary (1959) and Starostin’s Altaic dictionary (2003), as a collection of possible rather than actual cognates which must be subjected to further analysis. My reconstruction of Indo-Uralic retains 27 of Greenberg’s 68 grammatical elements, and I find 12 out of these 27 in Nivkh (Gilyak). We may wonder how many of these can now be reconstructed for Altaic. Here I shall take Starostin’s list of Proto-Altaic grammatical elements (2003: 221-229) as my point of departure. I shall refer to Greenberg’s numbering as G1-G60. 416 Indo-Uralic Starostin reconstructs personal pronouns 1sg. *bi, 1pl. *ba, *bu, obl. *min-, *man-, *mun-, 2sg. *si, 2pl. *su, obl. *sin-, *sun-, Mongolian 2sg. či < *thi and 2pl. ta < *tha. Besides, there are forms with a velar nasal in the first and a dental nasal in the second person, which “may have originally been restricted to some oblique cases” (Starostin 2003: 225). It seems to me that these forms are compatible with Indo-Uralic G1 *mi ‘I’, *me ‘we’, G4 *ti ‘thou’, *te ‘you’, G54 *-u ‘self ’ and G25 *-n genitive (cf. K203: 221, 225). The Indo-Uralic *m- may have spread from the genitive if it was not the phonetic reflex of an original labial plosive. The Altaic forms with a velar and a dental nasal may reflect *mn- and *tn- with syncope before a following suffix. The alternation between *s- and *thin Altaic suggests that we must start from 2sg. *si < *thi and 2pl. *tha, with restoration of the plosive in Mongolian and generalization of the fricative in Tungusic; the form is limited to the singular in Turkic (where the plural is *siŕ) and Japanese and unattested in Korean. The assibilation of *ti to *si is also found in the Indo-European branch of Indo-Uralic (cf. K203: 221) and in modern Finnish. Starostin reconstructs demonstrative pronouns *sV, *ko, *la, *o ‘this’ and *čha, *e, *i, *tha, *the ‘that’; I reconstruct Indo-Uralic demonstratives G8 *i/e, G11 *t- and G12 *s-. If the reconstruction of Proto-Uralic *ti ‘this’ and *tu ‘that’ is correct, it is possible that Proto-Altaic *sV ‘this’ and *tha ‘that’ are the result of a secondary development. However this may be, the identification of IndoEuropean *so with Finnish hän ‘he, she’ < *s- seems to be perfect. Thus, Altaic *e, *i, *tha, *the, *sV can be identified with Indo-Uralic *i/e, *t-, *s- while Altaic *o may be compared with G54 *u (cf. K203: 225). It is possible that the latter is also found as a suffix in the stem *meno ‘self, body’ which functions as a reflexive pronoun. The Altaic interrogative pronoun *kha ‘who’ may be identical with the Indo-Uralic interrogative G60 *k-. The Altaic plural suffix *-th- can be identified with the Indo-Uralic plural suffix G15 *-t. The Altaic accusative suffix *-be may be identical with the IndoUralic accusative G24 *-m if the latter is the phonetic reflex of an original labial plosive, as in the first person pronoun. The Altaic genitive has a velar, dental or palatal nasal, which points to *-n followed by other suffixes. This is supported by the fact that *-nV is also found as a dative, locative and instrumental case suffix. As in Indo-European (cf. K203: 222), it appears that the genitive G25 *-n developed into a general oblique singular ending in Altaic. Alternatively, it may have merged with the locative G30 *-n which may be compared with the Altaic dative, locative and instrumental suffix *-nV. The locative G29 *-ru can be identified with the Altaic directive suffix *-rV. Other case suffixes may be compared with the dative G26 *-ka and the ablative G33 *-t. Starostin reconstructs partitive *-ga, dative or directive *-khV, and allative *-gV, all of which may be related to the Indo-Uralic dative suffix *-ka. Since the Indo-European evidence points to a number of different vowels Indo-Uralic and Altaic 417 after the velar consonant (cf. K203: 224), it is quite possible that several cognates of the Altaic suffixes merged in Indo-European. Similarly, the Altaic dative or locative *-du, *-da, comitative or equative *-čha, and instrumental or ablative *-ǐV (which function as an ablative in Turkic, Mongolian and Japanese, respectively) may all be related to the Indo-Uralic ablative suffix *-t, which could be followed by other suffixes (cf. K203: 222). Indeed, the distinction between Japanese genitive no and dative or locative ni and between Turkic dative *-ka and Tungusic directive *-ki suggests that the locative G31 *-i may have been added to other suffixes so as to provide a (stronger) locative meaning, in the same way as Indo-European replaced the original ablative ending by *-ti in its local use in order to differentiate it from its instrumental use (cf. K203: 222). The Altaic deverbative nominal suffixes *-m- and *-l- (Starostin 2003: 177, 187) can be identified with the Indo-Uralic nominalizer G39 *-m- and participle G45 *-l-. The Altaic gerund *-jV and past tense *-thV (Starostin 2003: 227) may be identical with the Indo-Uralic nominalizer G38 *-i- and participle G43 *-t-. The combination of the Altaic desiderative or inchoative *-s- and factitive or intensive *-g- (Starostin 2003: 207, 209) may be found in the Indo-Uralic conative G53 *-sk-. The Altaic *-s- is also found by itself in Indo-European and may be identical with the IE root *es- ‘to be’ (cf. K161: 169). I am inclined to identify the Altaic negative verb *e-, Mongolian ese ‘not’, with the Uralic negative verb *e- (cf. Collinder 1960: 247) and the IE root *es-, with loss of the original negative particle G56 *ne in Altaic (as in modern French). The Proto-Altaic verb *era ‘to be’ (Starostin 2003: 515) is evidently a derivative of the same root. Thus, I find evidence for 20 Indo-Uralic grammatical elements in Altaic: first person G1 *m, second person G4 *t, demonstratives G8 *i/e, G11 *t, G12 *s, plural G15 *t, accusative G24 *m, genitive G25 *n, dative G26 *ka, locatives G29 *ru, G30 *n, G31 *i, ablative G33 *t, nominalizers G38 *i and G39 *m, participles G43 *t and G45 *l, conative G53 *sk, reflexive G54 *u/w, and interrogative G60 *k. I conclude that the reality of an Eurasiatic language family is very probable. The historical relationship between the Altaic and Uralo-Siberian language families remains to be specified. We must reckon with the possibility that these are the two main branches of the Eurasiatic macro-family. Further research should therefore aim at separate reconstructions of Proto-Altaic and ProtoUralo-Siberian before other possible inner and outer connections are taken into consideration. Fortescue dates the dissolution of the Uralo-Siberian and Uralo-Yukagir language families to 8000 and 6000 BP or 6000 and 4000 BC, respectively (1998: 182, 219, and maps 3 and 4). Sammallahti dates the dissolution of ProtoUralic and Proto-Finno-Ugric to the 5th and 4th millennia BC, respectively (1988: 480), and these are identical with my datings for the dissolution of IndoHittite and of nuclear Indo-European (K203: 217). While I date the dissolution 418 Indo-Uralic of Indo-Uralic to the end of the 7th millennium (ibidem), Starostin dates the earliest split of Proto-Altaic to the 6th millennium (2003: 236). We may conclude that Proto-Indo-Uralic and Proto-Altaic may have been contemporaries (6000-5500), that Proto-Uralic and Proto-Uralo-Yukagir may have been the same thing and contemporaneous with Proto-Indo-Hittite (4500-4000), and that Proto-Finno-Ugric and nuclear Proto-Indo-European may again have been contemporary languages (3500-3000). This puts the dissolution of the Uralo-Siberian language family in the 7th millennium. It now becomes attractive to identify the latter with the abrupt climate change of 8200 BP or 6200 BC, when severe cold struck the northern hemisphere for more than a century. The catastrophic nature of this disastrous event agrees well with the sudden dispersal and large-scale lexical replacement which are characteristic of the Uralo-Siberian languages. INDO-URALIC AND ALTAIC REVISITED After Dybo & Starostin’s comprehensive rebuttal (2008) of Vovin’s critique (2005), one may wonder if it is useful to continue a debate which seems to lead nowhere and can only deter younger scholars from entering the field of Altaic studies. Yet I think that progress can be made by ending the controversy and developing a positive attitude to new perspectives. On the one hand, one cannot expect radical breakthroughs in a field where very few scholars are working on a number of extremely heterogeneous cultural traditions. On the other hand, the dramatic progress of Indo-Uralic studies in recent decades shows that there is room for unexpected results. Dybo & Starostin’s civil tone and admirable restraint contrast starkly with Vovin’s vicious rhetoric and personal insults. In my view, the Altaic controversy can be ended by abandoning emphasis on separate etymologies and shifting the attention to morphological correspondences and questions of chronology. The Vovin controversy can perhaps be ended by temporarily excluding this author from the debate and giving him a chance to reconsider the volatility of his position and the damage he has inflicted on the field by his offensive style of writing. In the meantime it is important to stimulate younger scholars to take part in a discussion which is traditionally dominated by an elderly generation. Dybo & Starostin claim (2008: 135) that “if genetic relationship between two or more languages can be demonstrated on morphological evidence, it will inevitably show up in the basic lexicon as well” whereas “if genetic relationship can be demonstrated on lexical evidence, it will not necessarily be detected within the compared languages’ morphology as well”. The problem is that critics of the Altaic hypothesis find most etymologies unattractive or suspicious and feel that the corpus of Altaic comparisons comprises not only possible cognates but also obvious loanwords, accidental lookalikes and even totally irrelevant non-lookalikes (as one colleague put it in an e-mail message to me). This assessment of the Altaic etymological dictionary is partly based on a misunderstanding. Like Pokorny’s etymological dictionary of the IndoEuropean languages (1959), Starostin’s dictionary is an essentially achronic collection of materials which can be used for analysis and reconstruction. Nobody today would subscribe to Pokorny’s reconstructed forms, and the same may be the fate of Starostin’s in the future. The quality of an etymology becomes more difficult to assess as we move deeper into the past because more unforeseen things may have happened. This is why there is reason to attach less value to separate etymologies at a larger time depth, unless there is independent evidence for the chronological layer to which they may belong, such as geographical distribution or relative chronology of specific changes. Note that 420 Indo-Uralic Starostin et al. date the earliest split of Proto-Altaic to the sixth millennium BC (2003: 236), which means that Proto-Indo-Uralic and Proto-Altaic may have been spoken around the same time. When we look at language interference in bilingual communities, it appears that there is a marked difference in the ease of linguistic borrowing between grammar and lexicon, between bound and free morphemes, and between verbs and nouns. As a result, the older strata of a language are better preserved in the grammatical system than in the lexical stock, better in morphology than in phonology or syntax, better in verb stems and pronouns than in nouns and numerals. The wide attestation of the Indo-European numerals must be attributed to the development of trade which accompanied the increased mobility of the Indo-Europeans at the time of their expansions. Numerals do not belong to the basic vocabulary of a neolithic culture, as is clear from their absence in Proto-Uralic and from the spread of Chinese numerals throughout East Asia. The inequality between different parts of a language in linguistic borrowing is of particular importance when we are dealing with distant affinity. In a study of the earliest contacts between the Indo-European and Uralic language families (1986), Rédei lists 64 words which were supposedly borrowed from Indo-European into Uralic at an early date. The material is divided into three groups: 7 Indo-European words which are attested in both Finno-Ugric and Samoyedic, 18 Indo-European or Indo-Iranian words which are attested in Finno-Ugric but not in Samoyedic, and 39 Indo-Iranian words which are found neither in Ugric nor in Samoyedic. Now it turns out that the number of verbs in the oldest material is too large to support the hypothesis that they were borrowed: verbs constitute 43% of the first group, 28% of the second group, and 5% of the third group. This is strong evidence for the thesis that the oldest layer was in fact inherited from an Indo-Uralic proto-language. Though the material is very small, the case for an original genetic relationship is particularly strong because we are dealing with basic verbs meaning ‘to give’, ‘to wash’, ‘to bring’, ‘to drive’, ‘to do’, ‘to lead’, ‘to take’ (cf. K112). Moreover, it is difficult to see how Proto-Indo-European words could have been borrowed into Proto-Uralic if the Indo-Europeans lived in the South Russian steppe when the ancestors of the Finno-Ugrians and the Samoyeds lived on the eastern side of the Ural mountains. The earliest contacts between Indo-European and Uralic languages must probably be identified with the eastward expansion of the Indo-Iranians and the simultaneous spread of the Finno-Ugrians to the southwest. Thus, it appears that we do not need a large number of obvious cognates, which cannot be expected in the case of distant linguistic affinity, in order to establish a genetic relationship between languages. Dybo & Starostin argue (2008: 128) that “it is unreasonable to expect to be able to reconstruct paradigmatic morphology when dealing with macrofamilies” because a morphological system can undergo an overwhelming collapse over a Indo-Uralic and Altaic revisited 421 relatively short period of time, as happened in the case of Classical Latin. However, this does not generally hold for the separate elements which make up the morphological system. The advantage of morphology over the lexicon is that it offers two types of chronological clue: in addition to sound changes which affect both lexical and morphological elements, the development of morphosyntactic categories poses obvious restrictions on the genesis and development of paradigmatic systems. In my reconstruction of the Indo-Uralic verb (K203) I have argued that the Indo-European verbal system can be derived from combinations of Indo-Uralic morphemes by a series of well-motivated phonetic and analogic developments. It is precisely the explanation of the IndoEuropean system of paradigms in terms of its Indo-Uralic origins that corroborates the reconstruction of the original morphemes. There is additional evidence for Indo-Uralic in the relation between ProtoIndo-European root structure and accentuation discovered by Lubotsky (1988a: 169-170). It appears that in the case of derivatives of roots with a stop which is contiguous to the syllabic nucleus but without an initial laryngeal, o-stems are barytone if the root contains a voiceless obstruent and oxytone if the root contains a voiced obstruent whereas i- and u-stems are oxytone if the root contains a voiceless obstruent and barytone if the root contains a voiced obstruent, regardless of the ablaut grade of the root. This highly peculiar distribution can be explained by the assumption that Indo-European underwent the “rhythmic” and “syllabic” consonant gradations reconstructed for ProtoUralic (cf. Helimski 1995: 24-26 = 2000: 172-174) followed by a vowel gradation which shifted the stress toward the end of a word form and gave rise to the ablaut system (cf. K213: 165). Here again, Indo-Uralic offers an explanation for a state of affairs attested in Indo-European which remains unexplained if the Uralic data are not taken into account. Since the two Uralic consonant gradations were phonetic developments, one could suggest that their operation in Indo-European might be the result of substratum influence, or conversely. This suggestion meets with two difficulties. Firstly, the common chronology of the consonant gradations rather points to a shared innovation at a time of structural similarity. Secondly, the hypothesis of substratum influence before the Indo-Europeans arrived in Europe and acquired their highly characteristic linguistic features is arbitrary. It is definitely more probable that we are dealing with a single language family which split up when the Indo-Europeans moved westwards while their relatives stayed behind. My reconstruction of the IndoUralic phonological system is essentially the same as Sammallahti’s for ProtoUralic (1988), except for the fact that I reconstruct palatalized resonants *r' and *l' for his dental spirants *ð and *ð'. In particular, I think that the large number of Indo-European plosives is the result of a secondary development. The simplest assumption is that the Indo-Uralic proto-language was identical with 422 Indo-Uralic Proto-Uralic. Indeed, it seems possible to derive Nivkh (Gilyak) from the same proto-language, as I have indicated elsewhere (K205). Uhlenbeck has argued (1935a) that Proto-Indo-European consisted of two unrelated components, which he calls A and B. The first component comprises pronouns, verbal roots, and derivational suffixes, whereas the second contains isolated words which are not related to verbal roots, such as numerals, some kinship terms, and many names of body parts, animals and trees. Uhlenbeck compares A with Uralic and Altaic and attributes irregular features such as heteroclitic inflection and grammatical gender to B. The Indo-European verbal system appears to combine Uralic flexional morphemes with Caucasian syntactic patterns. The rise of the ergative construction (which gave rise to the paradigm of the nominal o-stems, cf. Beekes 1985), grammatical gender and adjectival agreement can be attributed to North Caucasian influence and may have proceeded as indicated by Pedersen (1907). These views can be unified with Gimbutas’ theory (e.g. 1985) that the Indo-Europeans moved from a primary homeland north of the Caspian Sea to a secondary homeland north of the Black Sea. What we have to take into account is the typological similarity of Proto-Indo-European to the North-West Caucasian languages. If this similarity can be attributed to areal factors (cf. K130: 94), we may think of Indo-European as a branch of Indo-Uralic which was transformed under the influence of a North Caucasian substratum. We may then locate the Indo-Uralic homeland south of the Ural Mountains in the seventh millennium BC (cf. Mallory 1989: 192f.). Having established the probability of an Indo-Uralic proto-language, we can now turn to the question if the reconstructed morphemes can be identified in other languages as well. This is indeed plausible for Eskimo (cf. Uhlenbeck 1935b, Fortescue 1998, Seefloth 2000) and Nivkh. It may therefore be appropriate to look for the same elements in the Altaic languages. Here I shall first list those items adduced by Greenberg (2000) as grammatical evidence for Eurasiatic which I reconstruct for Proto-Indo-Uralic: first person *m, second person *t, demonstrative *i/e, demonstrative *t, demonstrative *s, dual *ki, plural *t, plural *i, accusative *m, genitive *n, dative *ka, Indo-Uralic and Altaic revisited 423 locative *ru, locative *n, locative *i, ablative *t, diminutive *k, nominalizer *i, nominalizer *m, participle *n, participle *t, participle *nt, participle *l, verbal noun *s, conative *sk, reflexive *u/w, negative *n, interrogative *k. I have identified 12 of these 27 elements in Nivkh (K205), viz. first person *m, second person *t, demonstrative *i/e, demonstrative *t, demonstrative *s, dual *ki, plural *t, genitive *n, participle *nt, participle *l, verbal noun *s, reflexive *u/w. Moreover, I have suggested that we can add adessive *pi here on the basis of Indo-European *bhi ‘near’, Nivkh fiď ‘be in a place’, p‘iŋ ‘inhabitant’. For the 1st and 2nd person pronouns I reconstruct the following Indo-Uralic paradigms: nom. gen. ‘I/me’ *mi *min ‘myself ’ *mu *mun ‘we/us’ *me *men ‘thou/thee’ ‘yourself ’ ‘ye/you’ *ti *tu *te *tin *tun *ten In Indo-European, the assibilation of *ti to *si and the rise of ablaut which reduced all non-final vowels to *e under the stress and zero grade elsewhere resulted in the following outcome: independent dependent ‘I/me’ *mi, *me-, *m*men, *mn- ‘myself ’ *mu, *me-, *m*men, *mn- ‘we/us’ *me, *me-, *m*men, *mn- independent dependent ‘thou/thee’ *si, *se-, *s*sen, *sn- ‘yourself ’ *tu, *te-, *t*ten, *tn- ‘ye/you’ *te, *te-, *t*ten, *tn- It is clear that this system could not be maintained. Moreover, the stem form *s< *ti for the second person interfered with the Indo-Uralic demonstrative *s-, which is preserved in the Indo-European anaphoric pronoun *so. The largescale homophony was eliminated by the use of deictic *Ȥe ‘this’ for the first 424 Indo-Uralic person singular and *ue ‘self ’ for a person who is contrasted with another (third) person and by the suffixation of *-Ȥ < *-ki for the dual and *-i, later *-s < *-ti for the plural. This resulted in such forms as *Ȥme ‘this-me’, *tue ‘thee-self ’, *sue ‘him-self ’ (cf. K203: 225 and K223: 9), also *ueȤ, *uei ‘(our)selves’ in contrast with outsiders (inclusive meaning) versus *(m)neȤ, *(m)nes ‘we’ in contrast with your people (exclusive meaning), *ueȤ, *ues ‘you’ in contrast with other people, then *uȤe ‘you two’ in contrast with ‘them’ and *nȤue ‘we two’ in contrast with both ‘you’ and ‘them’. These forms must have existed at an early stage already because the o-vocalism of *noȤ, *nos, *uoȤ, *uos originated in their use as clitics and we find the corresponding zero grade in acc.pl. *nsme, *usme, where *-me can hardly be anything else than the full grade Indo-Uralic case particle *me. On the other hand, the forms *teue and *seue show the continued existence of *te, *se, *ue as separate words at the stage when full grade *e in unstressed syllables became possible. It appears that gen. *men ‘me’ was remodeled to *mene on the basis of *teue and *seue. I think that dat. *mighi represents original *mibhi with dissimilation of the labial articulation because I cannot otherwise explain the differentiation from *tubhi and *subhi. These forms seem to preserve Indo-Uralic *mi ‘I’, *tu ‘thou-self ’, and *pi ‘at’. In Nivkh we find 1sg. *mi, 1du. *men-ki (‘the two of us’), 1pl. *me-t, 2sg. *ti, 3sg. *i/e, *i-w, reflexive *pi-, reciprocal *u- (cf. K205). We now turn to the Altaic languages. Starostin et al. reconstruct personal pronouns 1sg. *bi, 1pl. *ba ~ *bu, obl. *min-, *man- ~ *mun-, 2sg. *si, 2pl. *su, obl. *sin-, *sun-, adding Mongolian 2sg. či < *thi, 2pl. ta < *tha, which are “no doubt archaic” (2003: 225). These forms are strongly reminiscent of Indo-Uralic 1st person *mi, *me, *mu, gen. *min, *men, *mun, 2nd person *ti, *tu, gen. *tin, *tun, and 2sg. *ti, 2pl. *te, respectively. The alternation between *s- and *th- in Altaic suggests that we must start from 2sg. *si < *thi and 2pl. *tha, with restoration of the plosive in Mongolian and generalization of the fricative in Tungusic; the form is limited to the singular in Turkic (where the plural is *siŕ) and Japanese and unattested in Korean. The assibilation of *ti to *si is also found in the Indo-European branch of Indo-Uralic (cf. K203: 221). In the 1st person form, Indo-Uralic *m- may have spread from the genitive if it was not the phonetic reflex of an original labial plosive, e.g. prenasalized *mb or preglottalized *Ȥb. Besides, Starostin et al. reconstruct 1st person *ŋa and 2nd person *na, which “may have originally been restricted to some oblique cases” (2003: 225), largely on the basis of the Korean and Japanese evidence. These forms may reflect *mn- and *tn- with syncope before a following suffix, as in the Indo-European forms reconstructed above. If these considerations are correct, we arrive at the following reconstruction of the original personal pronouns in Indo-Uralic, Turkic, Mongolic, Tungusic, Koreanic and Japonic (cf. also Janhunen 2003: 18, Gorelova 2002: 216, Benzing 1955: 107, Robbeets 2005 s.v.): Indo-Uralic and Altaic revisited PIU *mi *min *mu *mun *me *men *ti *tin *tu *tun *te *ten PTk *bi- *si- PMo *bi *min-, *n*ba *man*či *čin- 425 PTg *bi *min *bu *mun *ba *man *si *sin *su *sun PK PJ *n*u- *a *ba *n- *si *na *ta *tan- From this table it appears that the Altaic personal pronouns can largely be derived from the ones reconstructed for Indo-Uralic except for the initial *b- in the first person forms. Starostin et al. reconstruct demonstrative pronouns *s-, *ko, *la, *o ‘this’ and *čha, *e, *i, *tha (*the) ‘that’. It appears that Altaic *e, *i, *tha (*the), *s-, *o may be identical with the Indo-Uralic demonstratives *i/e, *t-, *s-, and reflexive *u: PIU *i PTk *ın- *e *t- *an*ti- *s*u/w *-sı *o(l) PMo *i *in*e*en*te*ten*on- PTg *i *in*e- PK *i PJ *i *a- *a- *ta- *tj- *to- *u- *so*o- The Altaic interrogative pronoun *kha- ‘who’, PTk *ka-, *ke-, PMo *ka-, *ke-, PTg *xa-, PK *ka, PJ *ka, may be identical with the Indo-Uralic interrogative *k-. The Altaic plural suffix *-th- can be identified with the Indo-Uralic plural suffix *-t. The Altaic accusative suffix *-be may be identical with the Indo-Uralic accusative *-m if the latter is the phonetic reflex of an original labial plosive, as in the first person pronoun. The Altaic genitive has a velar, dental or palatal nasal, which points to *-n followed by other suffixes. This is supported by the fact that *-n- is also found as a dative, locative and instrumental case suffix. As in Indo-European (cf. K203: 222), it appears that the genitive *-n developed into a general oblique singular ending in Altaic. Alternatively, it may have merged with the locative *-n which may be compared with the Altaic dative, locative and 426 Indo-Uralic instrumental suffix *-n-. The locative *-ru can be identified with the Altaic directive suffix *-r-. Other case suffixes may be compared with the dative *-ka and the ablative *-t. Starostin et al. reconstruct partitive *-ga, dative or directive *-kh-, and allative *-g-, all of which may be related to the Indo-Uralic dative suffix *-ka. Since the Indo-European evidence points to a number of different vowels after the velar consonant (cf. K203: 224), it is quite possible that several cognates of the Altaic suffixes merged in Indo-European. Similarly, the Altaic dative or locative *-du, *-da, comitative or equative *-čha, and instrumental or ablative *-ǐ- (which function as an ablative in Turkic, Mongolian and Japanese, respectively) may all be related to the Indo-Uralic ablative suffix *-t, which could be followed by other suffixes (cf. K203: 222). Indeed, the distinction between Japanese genitive no and dative or locative ni and between Turkic dative *-ka and Tungusic directive *-ki suggests that the locative *-i may have been added to other suffixes so as to provide a (stronger) locative meaning, in the same way as Indo-European replaced the original ablative ending by *-ti in its local use in order to differentiate it from its instrumental use (cf. K203: 222). This results in the following comparisons (cf. Starostin et al. 2003: 221, Erdal 2004: 168-179, Janhunen 2003: 14, Benzing 1955: 78-89, Robbeets 2005 s.v.): PIU pl. *-t acc. *-m gen. *-n dat. *-ka loc. *-ru loc. *-n loc. *-i abl. *-t PTk *-t PMo *-d *-ŋ *-g *-ka *-ga *-ru *-n *-n *-da *-ča *-dur *-ča *-ga *-ru PTg *-ta, *-te *-ba, *-be *-ngī *-ga *-kī *-gī PK *-tır *-ń PJ *-tati *-bo *-n *-nka *-ro *-du *-ǐi *-ni *-tu *-to *-du Here the large variety of case endings in the Altaic languages cannot simply be derived from the ones reconstructed for Indo-Uralic, which may perhaps reflect a reduction of the Altaic system. Though some of the comparisons may have to be abandoned (cf. especially Robbeets 2005: 170-173 on PJ *-tati, *-to, *-tu, *-du), the considerable agreement between form and meaning of the suffixes suggests a common origin of plural *-t-, accusative *-m/b-, genitive *-n, dative *-ka, *-ga, and local cases *-ru, *-n-, *-i, *-t-, *-du. If the equations adduced above are correct, they render a genetic relationship between Indo-Uralic and the separate Altaic languages probable. Since Proto-Indo-Uralic seems to be both phonologically and morphologically Indo-Uralic and Altaic revisited 427 simpler than what we find in the Altaic languages, Indo-Uralic may have been either a sister or a daughter of an Altaic proto-language. In order to establish a possible chronology we now turn to the verb in the Altaic languages. As was indicated above, I reconstruct Proto-Indo-Uralic nominalizers *i and *m, participles *n, *t, *nt, *l, verbal noun *s, and conative *sk. The following deverbal nominals appear to have correspondences in the Altaic languages (cf. Starostin et al. 2003: 177, 187, 227): PIU *-i *-m *-t *-l PTk *-ja *-m *-t*-l PMo *-ja *-m *-l PTg PK *-ja *-m *-t- PJ *-i *-t- *-l Besides, I have suggested (K241 in fine) that the Indo-European present stem formatives *-(e)i-, *-(e)m-, *-(e)s-, *-n-, *-t/dh-, *-sk- represent original roots of simple verbs meaning ‘to go’, ‘to take’, ‘to be’, ‘to lead’, ‘to put’, ‘to try’, cf. Latin i-, em-, es-, Sanskrit nī-, dhā-, Tocharian A ske-, B skai-, and may be compared with Uralic inchoative *-j-, fientive *-m-, *-n-, causative and momentaneous *-t- (cf. Collinder 1960: 272-281). The suffix *-(e)s- is strongly reminiscent of the Altaic desiderative/inchoative *-s- (cf. Starostin et al. 2003: 206f. and K254) while the Indo-European root *es- may be identical with Altaic *a- (Robbeets 2005: 380, 468) and *er- ‘to be’ (Starostin et al. 2003: 515), PTk *er-, PMo *a-, *ere-, PTg *eri-, PK *a-, PJ *a-, *ar- (cf. K161). I am inclined to identify the Altaic negative verb *e-, PTg *e-, Mong. ese ‘not’ (Starostin et al. 2003: 488) with the Uralic negative verb *e- (cf. Collinder 1960: 247) and the Indo-European root *es-, with loss of the original negative particle *ne in the Altaic languages (as in modern French, e.g. c’est pas vrai). Robbeets has recently (2007) argued that the relative order of verbal stem formatives in Japanese overlaps with the distributional characteristics of related suffixes in other Altaic languages. Her conclusions are summarized in the following table: effort transform process iconic intention inchoative PTk *-la*-d*-n*-ki*-k- PMo *-la*-d*-n*-ki*-ma*-gi- PTg *-lā*-dā*-na*-ki*-m*-ga- PK *-no*-ki*-m*-k- PJ *-ra*-da*-na*-ka*-ma*-ka- The shape of attested chains of suffixes generally follows the order which can be reconstructed as *-la-da-na-ki-ma-ga-. If this is correct, it provides strong evidence for an Altaic proto-language which differed from Indo-Uralic. While 428 Indo-Uralic the suffix *-la- can be compared with Uralic iterative *-l- (cf. Collinder 1960: 275f.), the other suffixes appear to be limited to the Altaic languages. After this discussion of the morphological evidence, we may return to the problem of the lexicon. Arguing against a genetic relationship between the Mongolic and Tungusic languages, Doerfer has presented a detailed analysis of their common vocabulary (1985). Elsewhere I have shown that his material allows of a quite different conclusion (K156). Doerfer’s classification of the Tungusic languages into dialectal areas from west to east differs sharply from the genetic classification of the Tungusic languages. As a result, his Central Tungusic is much more heterogeneous than the other groups. For Central Tungusic, Doerfer removes the words which are found in both North and South Tungusic from the material and lists those words which are found in either North or South Tungusic only. The high number of ancient words in this part of the material casts grave doubts on Doerfer’s thesis that all of them were borrowed from Eastern Evenki, Solon or Manchu at a recent stage. It seems to me that the semantic distribution of the ancient Central Tungusic words with cognates in either North or South Tungusic points to genetic relationship rather than borrowing. In particular, the relatively large number of verbs is difficult to explain under the assumption of borrowing. In her magnum opus (2005), Robbeets eliminates the large majority of etymologies which have been proposed for Japanese words because they may be suspect for a variety of reasons, reducing a corpus of 2055 lexical entries to 359 core etymologies representing 4 pronouns, 170 verbs, 46 adjectives or quality nouns, 83 basic nouns and 56 non-basic nouns. Here again, the large number of verbs requires an explanation if one does not accept her analysis as proof of a genetic relationship between Japanese and the other Altaic languages. It is quite possible, and even probable, that some of the remaining etymologies will have to be abandoned in the future, especially because their number seems to be at variance with the large time depth assumed for the Altaic proto-language. On the other hand, the huge number of etymologies which were rejected out of hand because they might be suspect for one reason or another may comprise many instances where judgment has been too rash. We can only hope that future research will bridge the gap between the historical data of the attested languages and their reconstructed origins. This can only be achieved by training a new generation of scholars with an interest in the chronological aspects of linguistic diversity. A PARASITOLOGICAL VIEW OF NON-CONSTRUCTIBLE SETS “The genetic code, the primary manifestation of life, and, on the other hand, language, the universal endowment of humanity and its momentous leap from genetics to civilization, are the two fundamental stores of information transmissible from the ancestry to the progeny, the molecular succession, which ensures the transfer of hereditary messages from the cells of one generation to the next generation, and the verbal legacy as a necessary prerequisite of cultural tradition.” (Jakobson 1971: 681) “Divergent terminologies direct attention to different patternings; and finding a logically convincing test, acceptable all around, that can determine whether one such system of terms is superior to its rivals, is often impossible. Yet the slow processes of evolution presumably apply to human societies and their symbolic systems as much as to human bodies, so that when logic cannot decide, survival eventually will.” (McNeill 1976: 8) πάντα ῥεῖ As McNeill points out in his remarkable book on the role of infectious disease in the history of mankind, “one can properly think of most human lives as caught in a precarious equilibrium between the microparasitism of disease organisms and the macroparasitism of large-bodied predators, chief among which have been other human beings” (1976: 5). This view, which is abundantly illustrated in the book, leaves several questions open. Firstly, it is noteworthy that man has surpassed other large-bodied predators like lions and wolves in his ability to command the environment. What was the device that enabled man to achieve higher efficiency in hunting? The obvious answer is: the use of language. Secondly, it must be noted that man is curiously insensitive as compared with other hunting species. The conjecture that the use of language diminishes the need for direct observation does not explain the rapid disappearance of the 430 Appendix hunter’s senses. More probably, the faculty of language has an adverse effect on the perceptual capacity of the brain. Thirdly, the macroparasitism among human beings differs in kind from the relation between predator and prey. Its function is to shorten the food chain. Even cannibalism generally serves a legal purpose and is not merely a way of feeding. Human beings are driven by ideas. In recent years there has been some debate whether language must be viewed primarily as a means of communication or as a form of self-expression of the human mind. Both of these views start from an anthropocentric conception of language. According to the view advanced in the present article, language is the means of communicating natural forms of self-expression through the human mind. The influence of concepts is particularly striking in man’s economic behavior . The following example is typical: “What had always seemed commonplace and respectable became, after Veblen, fraudulent, ridiculous and (a favorite word of his) barbaric. This is high art. The American rich never recovered from the sardonic disdain with which Veblen analyzed their behavior. The manners of an entire society were altered as a result. After he made the phrase “conspicuous consumption” a part of the language, the real estate market in Newport was never again the same. What had been the biggest and best was henceforth the most vulgar. “Conspicuous leisure” made it difficult even for the daughters of the rich to relax. Their entertainment had thereafter to be legitimatized by charitable, artistic or even intellectual purpose or, at minimum, sexual relief.” (Galbraith 1972: 35f.) It is hard to think of child labour, war, totalitarianism, or massive unemployment without the driving force of a system of beliefs in conjunction with a blunted sense of perception. The observation that in all Yuman languages the word for ‘work’ is a loan from Spanish (Werner Winter, personal communication) should be a major blow to any current economic theory. The force of language is indeed comprehensive. “Even in the most primitive cultures the strategic word is likely to be more powerful than the direct blow” (Sapir 1949: 18). This has little to do with truth or logic because “the normal speaker does not actually feel the clash which the logician requires” (ibidem, 27). The view of language as a tool of the human species is less well-founded than its converse. The question is, in Humpty Dumpty’s words, which is to be master. The relation of a language to its carrier bears a strong resemblance to that of a parasite to its host. It invades the left hemisphere, diminishing the perceptual capacity of the brain. As a result, man’s major capability to change his environment is matched by a minor capability to gain insight from direct observation. An excessive attack may lead to autistic phenomena. If the brain strikes back, it may yield a form of epilepsy. Cruelty is a human characteristic A parasitological view of non-constructible sets 431 because it results from the substitution by the linguistic parasite of conception for perception. The formalist philosophy of language, of which generative grammar is but the latest variety, is futile because its subject matter is more easily accessible to the biochemist. The proper subject of the humanities is the behavior of the linguistic parasite. As in the natural sciences, advance in a humanist discipline springs from observation of what happens under changing circumstances, not from reflection on what is generally known. Language differs from viral diseases such as measles or smallpox in two respects. First, it is transmitted through sound waves, not through bodily contact. It can therefore be assumed that the fundamental structure of language is much simpler than that of a regular childhood disease. Second, the pattern of mutual adaptation between the individual and his language is much more stable than in the case of a more virulent infection. The process of adjustment alters not only the individual’s behavior and his language, but also the behavior of the group and the structure of the society at large. The relative ease with which the individual’s language responds to a change in the environment and the violent reactions which the use of language provokes among larger groups show that language is ancient in the individual and unfit for the large-sized communities of modern times. The fast rate of change which language exhibits can be compared with the instability of the influenza virus. It exemplifies a type of change which differs qualitatively from what we are accustomed to regard as the normal type of biological reproduction. In order to clarify the matter I define: (1) Organic reproduction yields an image which resembles the model to a large extent. (2) Symbolic reproduction yields an image which resembles the model to a small extent. Both types of reproduction must be distinguished from physical growth, which changes the size of an object without affecting its internal relationships. A crystal may grow, but does not reproduce itself. Werner Winter once compared the work of a translator with that of an artist who is asked to create an exact replica of a marble statue, but who cannot secure any marble. This is an apt characterization. Other linguistic activities differ from the translator’s work in the absence of the intention to create a replica. They share the feature of symbolic reproduction and involve the creation of objects with unforeseen properties. Logical analysis requires the identifiability of distinguishable elements as belonging to the same set. In the case of an extensional definition, it presupposes a sufficient degree of similarity between the indicated and the intended elements. In the case of an intensional definition it presupposes the 432 Appendix applicability of a criterion, which depends on the degree of similarity between the indicated property and the perceptible characteristics of the intended objects. The constructibility of a set is determined by the identifiability of its elements. Language does not generally satisfy the fundamental requirement of logic. Consider the following example: (3) Man is numerous. (4) Socrates is a man. (5) Socrates is numerous. From a linguistic point of view, the inference is equivalent to the first syllogism of traditional logic. The point is that a linguistic meaning thrives by virtue of its applications, which cannot be deduced from its implications. The latter must be derived from its applicability, rather than the other way round. Thus, a linguistic meaning has the properties of a non-constructible set. Now I define: (6) Existence is the capacity of an element being distinguished. (7) Truth is the capacity of an element belonging to a set. (8) Meaning is the capacity of being a set. (9) The power of a capacity is the set of elements with that capacity. (10) Symbolization is the power of existence. (11) Generalization is the power of truth. (12) Abstraction is the power of meaning. (13) Mathematics is the study of symbolization. (14) Logic is the study of generalization. (15) Philosophy is the study of abstraction. Thus, a mathematician is typically concerned with the problem of existence, a logician with the problem of truth, and a philosopher with the problem of meaning. (16) Physics is the phenomenology of existence. (17) Anthropology is the phenomenology of truth. (18) Linguistics is the phenomenology of meaning. The following statement can easily be verified: (19) Symbolization is a finite simple group of extremely large order. – If x can be distinguished and y can be distinguished, then x and y can be distinguished. – If x and y can be distinguished and z can be distinguished, then x can be distinguished and y and z can be distinguished. A parasitological view of non-constructible sets 433 – There is an identity element, viz. nothing, which combines with any distinguishable element without affecting what is distinguished. – For every distinguishable element x there is an inverse element x–1, which is the absence of x. – There is no proper subset of symbolization such that every distinguishable element can be distinguished as an element of the subset. – The number of distinguishable elements is limited by the finite ability of the senses. (20) Generalization is the Cartesian product of symbolization and abstraction. I further define: (21) Sense is the applicability of meaning. (22) Formalization is the reduction of meaning to truth. (23) A contradiction is an element of generalization which is both true and false. (24) A confusion is an element of abstraction with a contradictory formalization. The existence of non-constructible sets offers a solution for the problem of the philosopher’s stone. I think that the philosopher’s stone is a 4-dimensional object and that it is crossing the 3-dimensional surface of a 4-dimensional pond called history. The universe is the slice of the stone which is at the level of the water. The universe originated with a big bang when the stone hit the surface. It is finite and expands as the stone sinks into the water. Why did the stone hit the surface? We shall never know because it is beyond human observation. The parasitological view of non-constructible sets outlined here has important consequences for the daily practice of human affairs. Language has enabled humanity time and again to discover new techniques, allowing easy exploitation and rapid depletion of hitherto inaccessible resources and thereby renewing or intensifying damage to other forms of life. Its fast rate of change has not permitted a stable, chronic relationship to establish itself. “A stable new disease pattern can arise only when both parties manage to survive their initial encounter and, by suitable biological and cultural adjustments, arrive at a mutually tolerable arrangement. [...] historical experience of later ages suggests that something like 120 to 150 years are needed for human populations to stabilize their response to drastic new infections” (McNeill 1976: 51). Language may be lethal if time does not suffice for humans to adjust to changing conditions. THE ORIGIN AND NATURE OF THE LINGUISTIC PARASITE “Whatever these difficulties may be, and whatever their biological origin, it is clear that, at the level of concepts, categorization is carried out neither by rigorous, nor by logical, nor by universal criteria. Indeed, there may be no general means by which categories are formed at this level.” (Edelman 1987: 246) The philosophy of language comes in three varieties. 1. The functionalist’s view: linguistic forms are instruments used to convey meaningful elements. This is the basis of European structuralism. 2. The formalist’s view: linguistic forms are abstract structures which can be filled with meaningful elements. This is the basis of generative grammar. 3. The parasitologist’s view: linguistic forms are vehicles for the reproduction of meaningful elements. This is the view which I advocated twenty years ago in the Festschrift for Werner Winter’s 60th birthday (K067). Here I intend to discuss the evolutionary origin and the physiological nature of the linguistic parasite. My theory of language is wholly consistent with Gerald Edelman’s theory of neuronal group selection. This author makes three fundamental claims (Edelman 1987: 5): “1. Diversification of anatomical connectivity occurs epigenetically during development, leading to the formation by selection of primary repertoires of structurally variant neuronal groups. The diversification is such that no two individual animals are likely to have identical connectivity in corresponding brain regions. […] 2. A second selective process occurs during postnatal behavior through epigenetic modifications in the strength of synaptic connections within and between neuronal groups. As a result, combinations of those particular groups whose activities are correlated with various signals arising from adaptive behavior are selected. […] 3. Coherent temporal correlations of the responses of sensory receptor sheets, motor ensembles, and interacting neuronal groups in different brain regions occur by means of reentrant signaling. Such signaling is based on the existence of reciprocally connected neural maps. […]” 436 Appendix Linguistic meanings are combinations of neuronal groups whose activities are correlated with the responses of sensory reception sheets connected with hearing and motor ensembles connected with speaking and their interactions. Language differs from bird song in allowing continuous and coherent correlation of various temporal and spatial aspects of a neural construct with at least some features of a real-world object which is not speech (cf. Edelman 1987: 108). Linguistic meanings are instances of categorical memory, combining relatively long-term changes at the cellular level with continuing creation of variants in certain synapses (cf. Edelman 1987: 205). They are subject to a Darwinian competition between various groups for cortical representation space as different stimuli are successively encountered; the most competitive groups are those that are associated with the most frequently stimulated peripheral locations (cf. Edelman 1987: 171). Local movement of map borders is accounted for by the trading of cells between adjacent groups; continuous alteration in map boundaries is the physiological correlate of the nonconstructibility of linguistic meanings which I discussed in my earlier contribution (K067). The sensorimotor channels of speaking and hearing can be regarded as the male and female sex organs of the linguistic parasite. The successful transmission of a message from a speaker to a hearer produces a mapping which correlates various aspects of a neural construct with identifiable features of realworld objects in the environment. A linguistic analysis must therefore start from a correlation of physical aspects of the speech flow with identifiable features of objects and events in the real world. The physical world is perceived as disjunctively partitioned in polymorphous sets, and neuronal groups are disjunctively partitioned by selection as a result of reentrant mapping of disjunctions of partitions in polymorphous sets of signals (cf. Edelman 1987: 262). As a result, there is no isomorphism with the signal domain in global mappings. The combinability of linguistic meanings presupposes the decomposability of neuronal groups and their interconnections as well as the possibility of creating new subcircuits by a variety of neurotransmitters. These allow the speech flow to convey a representation of identifiable features of objects and events in the real world from the speaker to the hearer, creating in the latter a neural construct which is isofunctional with an image constructed in the former. The isofunctional character of the representation, which correlates various aspects of a neural construct with identifiable features of real-world objects, detaches the image from its carrier and thereby gives rise to an independent organism which is parasitic upon the human brain, competing for cortical representation space. It must be realized that the concept of language as a system of neural constructs which are correlated with identifiable features of real-world objects and move from one brain to the next by means of a device which resembles bird The origin and nature of the linguistic parasite 437 song is at variance with the functionalist view of language as an instrument used by a speaker to express his thoughts, which does not account for the adverse effects of linguistic behavior (cf. K067), and is opposed to the formalist view of language as a set of abstract rules and representations, which does not explain categorization and is irreconcilable with biological reality (cf. Edelman 1987: 38). The functionalist paraphrase of the statement S that X is the case as “I want you to think that I think that X is the case” can now be reformulated as “my linguistic parasite tries to create in your brain a neural construct which is isofunctional with the neural image of S constructed in my brain” (cf. also Grace 1987). In accordance with the theory advocated here, the exploratory behavior of linguistic meanings in the human brain bears a strong resemblance to ant foraging (cf. Gerhart and Kirschner 1997: 146-151). The complex large-scale pattern of ant movements is a consequence of many simpler responses, viz. the individual responses of single ants to the distribution of food. Ants leaving the nest secrete a pheromone trail which they follow back to the nest. When an ant finds food, it secretes a stronger pheromone trail. Ants leaving the nest tend to follow existing trails; however, some wander off randomly because the volatile pheromone of unreinforced trails is weak and evaporates rapidly. “Exploration rather than hardwiring specific contingencies seems like the only practical means of responding to the variability or complexity of the environment. The ant cannot anticipate where food may be; the centrosome has no way of detecting the position of the chromosomes and directing the microtubules toward them; the nerve cell cannot maintain or express all the information necessary for the fine-grained pathfinding decisions required to find its many targets and to cope with physiological variability” (Gerhart and Kirschner 1997: 193). Environmental changes produce coordinated changes in neuronal groups, and neuronal exploration allows the development of new linkages among neuronal groups, extending the opportunities for generating new contingencies and thereby serving an ongoing physiological function where the environment remains forever changeable. This mechanism creates a wide variety of linguistic meanings upon which selection eventually acts at the social level. REFERENCES Adams, Douglas Q. 1988 Tocharian historical phonology and morphology (New Haven: American Oriental Society). Adjarian, Hratch 1909 Classification des dialectes arméniens (Paris: Honoré Champion). Agajan, Èduard B. 1960 O genezise armjanskogo konsonantizma. Voprosy Jazykoznanija 1960/4: 37-52. Agard, Frederick B. 1984 A course in Romance linguistics II: A diachronic view (Washington DC: Georgetown UP). Alcock, Leslie 1971 Arthur’s Britain: History and archaeology (New York: St. Martin’s Press). Allen, W. Sidney 1950 Notes on the phonetics of an Eastern Armenien speaker. Transactions of the Philological Society 1950, 180-206. 1951 Phonetics and comparative linguistics. Archivum Linguisticum 3: 126-136. 1953 Phonetics in Ancient India (London: Oxford UP). Andreev, Nikolaj D. 1957 Periodizacija istorii indoevropejskogo prajazyka. Voprosy Jazykoznanija 1957/2, 3-18. Andrésen, Bjørn Stålhane 1968 Pre-glottalization in English standard pronunciation (Oslo: Norwegian Universities Press). Anthony, David W. 1986 The ‘Kurgan culture’, Indo-European origins, and the domestication of the horse: A reconsideration. Current Anthropology 27, 291-304. 440 References Antonsen, Elmer H. 1970 Old High German and the laws of final syllables. Studies in Linguistics 21, 55-76. 1975 A concise grammar of the older Runic inscriptions (Tübingen: Niemeyer). 2002 Runes and Germanic linguistics (Berlin: Mouton de Gruyter). Austerlitz, Robert 1959 Semantic components of pronoun systems: Gilyak. Word 15, 102-109. Austin, William M. 1946 A corollary to the Germanic Verschärfung. Language 22, 109-111. 1958 Germanic reflexes of Indo-European -Hy- and -Hw-. Language 34, 203-211. Babik, Zbigniew 2004 Morphonology of the Slavic present stem *jьmamь. Rocznik Slawistyczny 54, 65-85. Bakker, Peter, & Maarten Mous (eds.) 1994 Mixed languages: 15 case studies in language intertwining (Amsterdam: IFOTT). Ball, C.J.E. 1968 The Germanic dental preterite. Transactions of the Philological Society, 162-188. Bally, Charles 1945 Manuel d’accentuation grecque (Berne: Francke). Bammesberger, Alfred 1986a Der Aufbau des germanischen Verbalsystems (Heidelberg: Carl Winter). 1986b Gotisch (ni) ogs (þus) und althochdeutsch ni kuri. O-O-PE-RO-SI: Festschrift für Ernst Risch zum 75. Geburtstag (Berlin: Mouton de Gruyter), 673-677. 1987 Das Paradigma der ē-Verben im Urgermanischen. Beiträge zur Geschichte der deutschen Sprache und Literatur (Tübingen) 109, 341-349. Barth, Fredrik 1981 Selected essays, 2 vols. (London: Routledge & Kegan Paul). References 441 Bech, Gunnar 1963 Die Entstehung des schwachen Präteritums (København: Munksgaard). 1969 Das germanische reduplizierte Präteritum (København). Beekes, Robert S.P. 1969 The development of the Proto-Indo-European laryngeals in Greek (The Hague: Mouton). 1972 H2O. Die Sprache 18, 117-131. 1981 The subjunctive endings of Indo-Iranian. Indo-Iranian Journal 23, 21-27. 1982 GAv. må, the PIE. word for ‘moon, month’, and the perfect participle. Journal of Indo-European Studies 10, 53-64. 1983 On laryngeals and pronouns. Zeitschrift für vergleichende Sprachforschung 96, 200-232. 1985 The origins of the Indo-European nominal inflection (Innsbruck: Institut für Sprachwissenschaft). 1987 On Indo-European ‘wine’. Münchener Studien zur Sprachwissenschaft 48, 21-26. 1988 A Grammar of Gatha-Avestan (Leiden: Brill). 1995 Comparative Indo-European linguistics: An introduction (Amsterdam: Benjamins). 2010 Etymological dictionary of Greek (Leiden: Brill). Behaghel, Otto 1924 Zur formenbildung vocalisch auslautender oder anlautender stämme. Beiträge zur Geschichte der deutschen Sprache und Literatur 48, 128-130. Benediktsson, Hreinn 1963 Some aspects of Nordic umlaut and breaking. Language 39, 409-31. 1982 Nordic umlaut and breaking: Thirty years of research (1951-1980). Nordic Journal of Linguistics 5, 1-60. Benveniste, Émile 1935 Origines de la formation des noms en indo-européen (Paris: Maisonneuve). 1959 Sur la phonétique et la syntaxe de l’arménien classique. Bulletin de la Société de Linguistique de Paris 54: 46-68. 442 References Benzing, Johannes 1955 Die tungusischen Sprachen: Versuch einer vergleichenden Grammatik (Wiesbaden: Steiner). Berg, Nils 1977 Der Ursprung des altgriechischen aktiven Plusquamperfekts und die Entwicklung der alphathematischen Flexion. Norsk Tidsskrift for Sprogvidenskap 31, 205-263. Bibire, Paul 1975 Some notes on the Old Icelandic front mutations. Arkiv för Nordisk Filologi 90, 183-212. Binchy, Daniel 1956 Some Celtic legal terms. Celtica 3, 221-231. Boeder, Winfred 1979 Ergative syntax and morphology in language change: the South Caucasian languages. Ergativity (London: Academic Press), 435-480. Bouda, Karl 1960 Die Verwandtschaftsverhältnisse des Giljakischen. Anthropos 55, 355-415. Boutkan, Dirk F.H. 1995 The Germanic ‘Auslautgesetze’ [Leiden Studies in Indo-European 4] (Amsterdam: Rodopi). 1996 A concise grammar of the Old Frisian dialect of the First Riustring manuscript (Odense). 1998a On labial mutation and breaking in Old Frisian. Approaches to Old Frisian Philology = Amsterdamer Beiträge zur älteren Germanistik 49, 77-88. 1998b On the form of North European substratum words in Germanic. Historische Sprachforschung 111, 102-133. Boutkan, Dirk, & Maarten Kossmann 2001 On the etymology of ‘silver’. North-Western European Language Evolution 38, 3-15. Boutkan, Dirk, & Sjoerd M. Siebinga 2005 Old Frisian etymological dictionary (Leiden: Brill). References 443 Bradley, David 1979 Proto-Loloish (Copenhagen Scandinavian Institute of Asian Studies). Bradshaw, Joel 1979 Obstruent harmony and tonogenesis in Jabêm. Lingua 49, 189-205. Braune, Wilhelm, & Hans Eggers 1975 Althochdeutsche Grammatik (13. Auflage, Tübingen: Niemeyer). Broch, Ingvild, & Jahr, Ernst H. 1984 Russenorsk: Et pidginspråk i Norge (2. utgave, Oslo: Novus). Brown, Gillian 1977 Listening to spoken English (London: Longman). Brozović, Dalibor 1989 Review of Holzer 1989. Wiener Slavistisches Jahrbuch 35, 223-225. Brugmann, Karl 1904 Kurze vergleichende Grammatik der indogermanischen Sprachen (Strassburg: Trübner). Brunner, Karl 1965 Altenglische Grammatik (3. Auflage, Tübingen: Niemeyer). Būga, Kazimieras 1959 Rinktiniai raštai II (Vilnius: Valstybinė politinės ir mokslinės literatūros leidykla). Bybee, Joan, Revere Perkins & William Pagliuca 1994 The evolution of grammar: Tense, aspect, and modality in the languages of the world (Chicago: University Press). Çabej, Eqrem 1956 Über einige mit z- anlautende Wörter des Albanischen. Zeitschrift für Phonetik und allgemeine Sprachwissenschaft 9, 203-229. 1972 Über einige Lautregeln des Albanischen. Die Sprache 18, 132-154. Camaj, Martin 1966 Albanische Wortbildung: Die Bildungsweise der älteren Nomina (Wiesbaden: Harrassowitz). Campbell, Alistair 1939 Some Old Frisian sound-changes. Transactions of the Philological Society 1939, 78-107. 444 References 1959 Old English grammar (Oxford: University Press). Castrén, M. Alexander 1854 Grammatik der samojedischen Sprachen (herausgegeben von Anton Schiefner, St. Petersburg: Akademie der Wissenschaften; reprint 1966, Bloomington: Indiana University). Cercignani, Fausto 1979 Proto-Germanic */i/ and */e/ revisited. Journal of English and Germanic Philology 78, 72-82. Chantraine, Pierre 1961 Morphologie historique du grec (Paris: Klincksieck). 1967 Morphologie historique du grec (2-ème édition, Paris: Klincksieck). 1973 Grammaire homérique I: Phonétique et morphologie (Paris: Klincksieck). Chapman, Kenneth G. 1962 Icelandic-Norwegian linguistic relationships (Oslo: Universitetsforlaget). Christophersen, Paul 1952 The glottal stop in English. English Studies 33, 156-63. Clackson, James 1994 The linguistic relationship between Armenian and Greek (Oxford: Blackwell). Collier, L.W. 1987 The chronology of i-umlaut and breaking in pre-Old English. North-Western European Language Evolution 9, 33-45. Collinder, Björn 1960 Comparative grammar of the Uralic languages (Stockholm: Almqvist & Wiksell). 1965 Hat das Uralische Verwandte? Acta Universitatis Upsaliensis N.S. 1/4 (Uppsala: Almqvist & Wiksell), 109-180. 1974 Indo-Uralisch – oder gar Nostratisch? (Vierzig Jahre auf rauhen Pfaden) Antiquitates Indogermanicae (Innsbruck: Institut für Sprachwissenschaft), 363-375. Collins, Beverley, & Inger M. Mees 1994 Though’ts on the glo’ttal sto’p. Knowing the words: Liber amicorum for Robert Druce (Leiden: Academic Press), 68-81. References 445 Collitz, Hermann 1891 Die behandlung des urspr. auslautenden ai im Gotischen, Althochdeutschen und Altsächsischen. Beiträge zur kunde der indogermanischen sprachen 17, 1-53. Cowgill, Warren 1959 The inflection of the Germanic ō-presents. Language 35, 1-15. 1960 Gothic iddja and Old English eōde. Language 36, 483-501. 1965 Evidence in Greek. Evidence for laryngeals (The Hague: Mouton), 142-180. 1968 The first person singular medio-passive of Indo-Iranian. Pratidānam (The Hague: Mouton), 24-31. 1980 The etymology of Irish guidid and the outcome of *gwh in Celtic. Lautgeschichte und Etymologie (Wiesbaden: Reichert), 49-78. 1985 PIE *duuʢo ‘2’ in Germanic and Celtic, and the nom.-acc. dual of non-neuter o-stems. Münchener Studien zur Sprachwissenschaft 46, 13-28. Čekman, Valerij N. 1974 O refleksax indo-evropejskix *kʄ, *gʄ v balto-slavjanskom jazykovom areale. Balto-slavjanskie issledovanija (Moskva: Nauka), 116-135. Dahl, Ivar 1938 Substantival inflexion in early Old English: vocalic stems [Lund Studies in English 7] (Lund: Gleerup). Demiraj, Bardhyl 1997 Albanische Etymologien: Untersuchungen zum albanischen Erbwortschatz (Amsterdam: Rodopi). Derksen, Rick 1996 Metatony in Baltic (Amsterdam: Rodopi). 2003 Slavic *jь-. Studies in Slavic and General Linguistics 30: Dutch Contributions to the Thirteenth International Congress of Slavists (Ljubljana, 2003), Linguistics, 97-105. 2008 Etymological dictionary of the Slavic inherited lexicon (Leiden: Brill). Dishington, J. 1978 Arguments for an ai/ja-paradigm in the 3rd weak class of ProtoGermanic. Indogermanische Forschungen 83, 301-323. 446 References Djahukian, Gevork B. 1967 Očerki po istorii dopis’mennogo perioda armjanskogo jazyka (Erevan: AN Arm. SSR). Docherty, Gerard J., & Paul Foulkes 1999 Derby and Newcastle: instrumental phonetics and variationist studies. Urban voices: accent studies in the British Isles (London: Arnold), 47-71. Docherty, Gerard J., Paul Foulkes, James Milroy, Lesley Milroy & David Walshaw 1997 Descriptive adequacy in phonology: a variationist perspective. Journal of Linguistics 33, 275-310. Doerfer, Gerhard 1985 Mongolo-Tungusica (Wiesbaden: Harrassowitz). Dunkel, George 1981 Typology versus reconstruction. Bono Homini Donum: Essays in historical linguistics in memory of J. Alexander Kerns II (Amsterdam: Benjamins), 559-569. Dybo, Anna V., & George S. Starostin 2008 In defense of the comparative method, or the end of the Vovin controversy. Aspects of comparative linguistics 3 (Moscow: Russian State University for the Humanities), 119-258. Dybo, Vladimir A. 1961 Sokraščenie dolgot v kel’to-italijskix jazykax i ego značenie dlja balto-slavjanskoj i indo-evropejskoj akcentologii. Voprosy slavjanskogo jazykoznanija 5, 9-34. 1962 O rekonstrukcii udarenija v praslavjanskom glagole. Voprosy slavjanskogo jazykoznanija 6, 3-27. 1968 Akcentologija i slovoobrazovanie v slavjanskom. Slavjanskoe jazykoznanie: VI meždunarodnyj s”ezd slavistov, Praga, 1968 (Moskva: Nauka), 148-224. 1973 Baltoslavjanskaja akcentnaja sistema s tipologičeskoj točki zrenija i problema rekonstrukcii indoevropejskogo akcenta. Kuznecovskie čtenija 1973: Istorija slavjanskix jazykov i pis’mennosti, 8-10. References 447 Dybo, V., Nikolayev S. & Starostin S. 1978 A tonological hypothesis on the origin of paradigmatic accent systems. Estonian Papers in Phonetics (Tallinn: Academy of Sciences), 16-20. Dyvik, H.J.J. 1978 Breaking in Old Norse and related languages: A reassessment of the phonetic conditions. Arkiv för Nordisk Filologi 93, 1-37. Ebeling, Carl L. 1966 The grammar of literary Avar. Studia Caucasica 2, 58-100. 1967 Historical laws of Slavic accentuation. To Honor Roman Jakobson (The Hague: Mouton), 577-593. Edelman, Gerald M. 1987 Neural Darwinism (New York: Basic Books). Eichner, Heiner 1973 Die Etymologie von heth. mehur. Münchener Studien zur Sprachwissenschaft 31, 53-107. Einarsson, Stefán 1941 Some notes on E. Prokosch’s A Comparative Germanic Grammar, with special reference to his treatment of the Scandinavian languages. The Journal of English and Germanic Philology 40, 38-47. Ejskjær, Inger 1990 Stød and pitch accents in the Danish dialects. Acta Linguistica Hafniensia 22, 49-75. Elizarenkova, Tat’jana Ja. 1982 Grammatika vedijskogo jazyka (Moskva: Nauka). Emeneau, Murray B. 1966 The dialects of Old Indo-Aryan. Ancient Indo-European dialects (Berkeley and Los Angeles: University of California Press), 123-138. Endzelin, Jan 1957 Baltų kalbų garsai ir formos (Vilnius: Valstybinė politinės ir mokslinės literatūros leidykla). 1971 Comparative phonology and morphology of the Baltic languages (The Hague: Mouton). 448 References Erdal, Marcel 2004 A grammar of Old Turkic (Leiden: Brill). Fagan, Sarah M.B. 1989 Geminates in intensive and iterative Germanic class II weak verbs. Beitraege zur Geschichte der deutschen Sprache und Literatur 111, 35-58. Feist, Sigmund 1910 Europa im Lichte der Vorgeschichte (Berlin: Weidmann). Flasdieck, Hermann M. 1935 Untersuchungen über die germanischen schwachen Verben III. Klasse (unter besonderer Berücksichtigung des Altenglischen). Anglia 59, 1-192. 1936 Die reduplizierenden Verben des Germanischen (unter besonderer Berücksichtigung des Altenglischen). Anglia 60, 241-365. Fodor, István 1976 The main issues of Finno-Ugrian archaeology. Ancient Cultures of the Uralian Peoples (Budapest: Corvina), 49-78. Forbes, K. 1958 The formation of the so-called Aeolic optative. Glotta 37, 165-179. Fortescue, Michael 1988 The Eskaleut-Yukagir relationship: An alternative to the genetic/contact dichotomy. Acta Linguistica Hafniensia 21/1, 21-50. 1998 Language relations across Bering Strait: Reappraising the archaeological and linguistic evidence (London-New York: Cassell). Fourquet, Jean 1959 Le système des éléments vocaliques longs en vieil-anglais. Mélanges de linguistique et de philologie: Fernand Mossé in memoriam (Paris), 148-160. Fraenkel, Ernst 1950 Die baltischen Sprachen (Heidelberg: Carl Winter). Frisk, Hjalmar 1973 Griechisches etymologisches Wörterbuch I (Heidelberg: Carl Winter). References 449 Fritz, Matthias 1993 Griechisch ἀϋτμή ‘Dampf, Duft’. Historische Sprachforschung 106, 288-301. Fulk, Robert D. 1987 Reduplicating verbs and their development in northwest Germanic. Beiträge zur Geschichte der deutschen Sprache und Literatur (Tübingen) 109, 159-178. 1998 The chronology of Anglo-Frisian sound changes. Approaches to Old Frisian Philology = Amsterdamer Beiträge zur älteren Germanistik 49, 139-154. Galbraith, John Kenneth 1972 Economics, peace and laughter (New York: The New American Library). Gamkrelidze, Thomas V., & Vjačeslav V. Ivanov 1972 Lingvističeskaja tipologija i rekonstrukcija sistemy indoevropejskix smyčnyx. Konferencija po sravnitel’no-istoričeskoj grammatike indoevropejskix jazykov: Predvaritel’nye materialy (Moskva: Nauka), 15-18. 1973 Sprachtypologie und die Rekonstruktion der gemeinindogermanischen Verschlüsse. Phonetica 27, 150-156. 1980 Rekonstrukcija sistemy smyčnyx obščeindoevropejskogo jazyka: Glottalizovannye smyčnye v indoevropejskom. Voprosy Jazykoznanija 1980/4, 21-35. Garde, Paul 1976 Histoire de l’accentuation slave (Paris: Institut d’études slaves). Garibjan, Ararat S. 1958 Novaja gruppa dialektov armjanskogo jazyka. Voprosy Jazykoznanija 1958/6: 95-101. 1959 Ob armjanskom konsonantizme. Voprosy Jazykoznanija 1959/5: 81-90. Geldner, Karl F. 1951 Der Rig-Veda, aus dem Sanskrit ins Deutsche übersetzt und mit einem laufenden Kommentar versehen (Cambridge, Mass.). Georg, Stefan, & Alexander Vovin 2003 Review of Greenberg 2000. Diachronica 20/2, 331-362. 450 References Georgiev, Vladimir 1977 Trakite i tehnijat ezik (Sofija: Bъlgarska Akademija na Naukite). Gerhart, John, & Marc Kirschner 1997 Cells, embryos, and evolution (Malden: Blackwell). Gimbutas, Marija 1985 Primary and secondary homeland of the Indo-Europeans. Journal of Indo-European Studies 13, 185-202. Goblirsch, Kurt Gustav 1994 Consonant strength in Upper German dialects [= North-Western European Language Evolution, Supplement vol 10] (Odense: University Press). 1999 The correlation of voice in Germanic. North-Western European Language Evolution 35, 115-140. 2005 Lautverschiebungen in den germanischen Sprachen (Heidelberg: Carl Winter). Godel, Robert 1975 An introduction to the study of Classical Armenian (Wiesbaden: Reichert). Gorelova, Liliya M. 2002 Manchu grammar (Leiden: Brill). Grace, George W. 1987 The linguistic construction of reality (London: Croom Helm). Green, Dennis Howard 1998 Language and history in the early Germanic world. Cambridge: University Press. Greenberg, Joseph H. 1970 Some generalizations concerning glottalic consonants, especially implosives. International Journal of American Linguistics 36, 123-145. 1997 The Indo-European first and second person pronouns in the perspective of Eurasiatic, especially Chukotkan. Anthropological Linguistics 39, 187-195. 2000 Indo-European and its closest relatives: The Eurasiatic language family I: Grammar (Stanford: University Press). References 451 Greenberg, Marc L. 2000 A historical phonology of the Slovene language (Heidelberg: Carl Winter). 2007 Phonetic evidence for the development of the “acute” tone in Slavic. Tones and theories: Proceedings of the international workshop on Balto-Slavic accentology (Zagreb: Institut za hrvatski jezik i jezikoslovlje), 75-87. Greene, David 1974 The growth of palatalization in Irish. Transactions of the Philological Society 1973, 127-136. 1976 The diphthongs of Old Irish. Ériu 27, 26-45. Grønvik, Ottar 1998 Untersuchungen zur älteren nordischen und germanischen Sprachgeschichte [Osloer Beiträge zur Germanistik 18] (Frankfurt: Peter Lang). Gruzdeva, Ekaterina 1998 Nivkh (München: Lincom Europa). Gysseling, Maurits 1962 Het oudste Fries. It Beaken 24, 1-26. Hachmann, Rolf 1970 Die Goten und Skandinavien (Berlin: Walter de Gruyter). Hajdú, Péter 1975 Finno-Ugrian languages and peoples (London: André Deutsch). Hamp, Eric P. 1965 Evidence in Albanian. Evidence for laryngeals (The Hague), 123-141. 1985 German Bein, Old English bān, Slavic kostь. North-Western European Language Evolution 6, 67-70. Hansen, Aage 1943 Stødet i Dansk (København Munksgaard). Hansson, Gunnar Ólafur 2001 Remains of a submerged continent: Preaspiration in the languages of northwest Europe. Historical linguistics 1999 (Amsterdam: Benjamins), 157-173. 452 References Harðarson, Jón Axel 1993 Studien zum urindogermanischen Wurzelaorist (Innsbruck: Institut für Sprachwissenschaft). Haudricourt, André-Georges 1975 Les mutations consonantiques (occlusives) en indo-européen. Mélanges linguistiques offerts à Émile Benveniste (Louvain: Peeters), 267-272. Haudry, Jean 1977 L’emploi des cas en védique: Introduction à l’étude des cas en indoeuropéen (Lyon: L’Hermès). Haugen, Einar 1941 On the consonant pattern of modern Icelandic. Acta Linguistica 2, 98-107. 1958 The phonemics of modern Icelandic. Language 34, 55-88. 1976 The Scandinavian languages (London: Faber & Faber). Heather, Peter 1991 Goths and Romans 332-489 (Oxford: Clarendon Press). Heenen, François 2006 Le désidératif en védique (Amsterdam: Rodopi). Helimski, Eugene 1995 Proto-Uralic gradation: Continuation and traces. Congressus Octavus Internationalis Fenno-Ugristarum I: Orationes plenariae et conspectus quinquennales (Jyväskylä: Moderatores), 17-51. Heuser, Wilhelm 1903 Altfriesisches Lesebuch mit Grammatik und Glossar (Heidelberg: Carl Winter). Heusler, Andreas 1967 Altisländisches Elementarbuch (Heidelberg: Carl Winter). Hilmarsson, Jörundur 1986 Studies in Tocharian phonology morphology and etymology, with special emphasis on the o-vocalism (Diss. Leiden, Reykjavík). 1989 The dual forms of nouns and pronouns in Tocharian [=Tocharian and Indo-European Studies, suppl. 1] (Reykjavík: Málvísindastofnun Háskóla Íslands). 1991 On ē2 in Germanic. Acta Linguistica Hafniensia 23, 33-47. References 1996 453 Materials for a Tocharian historical and etymological dictionary [=Tocharian and Indo-European Studies, suppl. 5] (Reykjavík: Málvísindastofnun Háskóla Íslands) Hirt, Herman 1892 Die Urheimat der Indogermanen. Indogermanische Forschungen 1, 464-485. 1895 Der Ackerbau der Indogermanen. Indogermanische Forschungen 5, 395-402. 1921 Indogermanische Grammatik II: Der indogermanische Vokalismus (Heidelberg: Carl Winter). 1929 Indogermanische Grammatik V: Der Akzent (Heidelberg: Carl Winter). Hirt, Herman, & Helmut Arntz 1939 Die Hauptprobleme der indogermanischen Sprachwissenschaft (Halle/Saale: Niemeyer). Hockett, Charles F. 1955 A manual of phonology (Baltimore: Waverly Press). Hoff, Ingeborg 1949 Vilkårene for brytning av germansk e till ia, io i vestnordisk. Arkiv för Nordisk Filologi 64, 177-210. Hoffmann, Karl 1957 Zur vedischen Verbalflexion. Münchener Studien zur Sprachwissenschaft 22, 121-137. 1967a Der Injunktiv im Veda: Eine synchronische Funktionsuntersuchung (Heidelberg: Carl Winter). 1967b Der vedische Prekativtyp yeṣam, jeṣma. Münchener Studien zur Sprachwissenschaft 20, 25-37. 1968 Zum Optativ des indogermanischen Wurzelaorists. Pratidānam (The Hague: Mouton), 3-8. Hofmann, Dietrich 1964 “Germanisch” ē2 im Friesischen. Festschrift für Jost Trier zum 70. Geburtstag (Köln: Böhlau), 160-185. 1989 Die spätgermanische Silbenquantitätsverschiebung und die Doppelschreibung alter kurzer Konsonanten in den altwestfriesischen Quellen. Gesammelte Schriften II: Studien zur 454 References friesischen und niederdeutschen Philologie (Hamburg: Buske), 206-14. Hogg, Richard M. 1979 Old English palatalization. Transactions of the Philological Society 1979, 89-113. Hollifield, Patrick H. 1980 The phonological development of final syllables in Germanic. Die Sprache 26, 19-53 and 145-178. Holzer, Georg 1989 Entlehnungen aus einer bisher unbekannten indogermanischen Sprache im Urslavischen und Urbaltischen (Wien: Akademie der Wissenschaften). Honselaar, Zep 1998 The dialect of Ostrovcy in the Pskov oblast. Dutch Contributions to the Twelfth International Congress of Slavists (Amsterdam: Rodopi) = Studies in Slavic and General Linguistics 24, 283-307. 2001 Govor derevni Ostrovcy pskovskoj oblasti (Amsterdam: Rodopi). Hopper, Paul J. 1973 Glottalized and murmured occlusives in Indo-European. Glossa 7, 141-166. Hughes, Arthur, & Peter Trudgill 1987 English accents and dialects: An introduction to social and regional varieties of British English (London: Edward Arnold). Huld, Martin E. 1976 Albanian yll ‘star’. Zeitschrift für vergleichende Sprachforschung 90, 178-182. 1984 Basic Albanian etymologies (Columbus, Ohio). Hyman, Larry M. 1975 Phonology: Theory and analysis (New York: Holt, Rinehart and Winston). Illič-Svityč, Vladislav M. 1962 K istolkovaniju akcentuacionnyx sootvetstvij v kel’to-italijskom i balto-slavjanskom. Kratkie soobščenija Instituta slavjanovedenija AN SSSR 35, 63-72. 1963 Imennaja akcentuacija v baltijskom i slavjanskom: Sud’ba akcentuacionnyx paradigm (Moskva: AN SSSR). References 1971 455 Opyt sravnenija nostratičeskix jazykov: Vvedenie, sravnitel’nyj slovar’ b-Ḳ (Moskva: Nauka). Insler, Stanley 1968 The origin of the Sanskrit passive aorist. Indogermanische Forschungen 73, 312-346. 1972 On proterodynamic root present inflection. Münchener Studien zur Sprachwissenschaft 30: 55-64. 1975a The Gāthās of Zarathustra (Leiden: Brill). 1975b The Vedic type dheyām. Die Sprache 21, 1-22. Irslinger, Britta S. 2002 Abstrakta mit Dentalsuffixen im Altirischen (Heidelberg: Winter). Isaac, Graham R. 2007 Studies in Celtic sound changes and their chronology (Innsbruck: Institut für Sprachwissenschaft). Ivanov, Vjačeslav V. 1981 Slavjanskij, baltijskij i rannebalkanskij glagol: Indoevropejskie istoki (Moskva: Nauka). Ivić, Pavle 1958 Die serbokroatischen Dialekte: Ihre Struktur und Entwicklung (The Hague: Mouton). Ivić, Pavle, & Ilse Lehiste 1967 Prilozi ispitivanju fonetske i fonološke prirode akcenata u savremenom srpskohrvatskom jeziku. Zbornik za Filologiju i Lingvistiku 10, 55-93. Jackson, Kenneth 1953 Language and history in early Britain (Edinburgh: University Press). Jacobsen, Lis 1935 Forbandelsesformularer i nordiske runeindskrifter. Stockholm: Akademiens Förlag. Jakobson, Roman 1932 Zur Struktur des russischen Verbums. Charisteria Vilhelmo Mathesio oblata (Prague: Cercle Linguistique de Prague), 74-83. 1941 Kindersprache, Aphasie und allgemeine Lautgesetze (Uppsala: Lundequist). 1962 Selected writings I: Phonological studies (The Hague: Mouton). 456 References 1971 Selected writings II: Word and language (The Hague: Mouton). Jamison, Stephanie W. 1983 Two problems in the inflection of the Vedic intensive. Münchener Studien zur Sprachwissenschaft 42, 41-73. Janhunen, Juha 1982 On the structure of Proto-Uralic. Finnisch-Ugrische Forschungen 44, 23-42. 1983 On early Indo-European-Samoyed contacts. Symposium Saeculare Societatis Fenno-Ugricae [=Memoires de la Societe Finno-Ougrienne 185], 115-127. (Helsinki: Suomalais-Ugrilainen Seura) 2003 The Mongolic languages (London: Routledge). Jasanoff, Jay H. 1976 Gr. ἄμφω, lat. ambō et le mot indo-européen pour ‘l’un et l’autre’. Bulletin de la Société de Linguistique de Paris 71/1, 123-131. 1978 Stative and middle in Indo-European (Innsbruck: Institut für Sprachwissenschaft). 1988 The sigmatic aorist in Tocharian and Indo-European. Tocharian and Indo-European Studies 2, 52-76. 2003 Hittite and the Indo-European verb (Oxford: University Press). 2004a Acute vs. circumflex: Some notes on PIE and post-PIE prosodic phonology. Per aspera ad asteriscos [Fs. Rasmussen] (Innsbruck: Institut für Sprachwissenschaft), 247-255. 2004b Balto-Slavic accentuation: telling news from noise. Baltistica 39/2, 171-177. 2008 The accentual type *vèdō, *vedetì and the origin of mobility in the Balto-Slavic verb. Baltistica 43/3, 339-379. Jellinek, Max Hermann 1926 Geschichte der gotischen Sprache (Berlin: Walter de Gruyter). Jespersen, Otto 1913 Det danske stød og urnordisk synkope. Arkiv för Nordisk Filologi 29, 1-32. Jokl, Norbert 1911 Studien zur albanesischen Etymologie und Wortbildung (=Sitzungsberichte der philosophisch-historischen Klasse der Kaiserlichen Akademie der Wissenschaften 168/1, Wien). References 1912 1937 1963 Jurišić, Blaž 1973 Kallio, Petri 2000 457 Beiträge zur albanesischen Grammatik. Indogermanische Forschungen 30, 192-210. Ein Beitrag zur Lehre von der alb. Vertretung der idg. Labiovelare. Mélanges linguistiques offerts à M. Holger Pedersen (Aarhus-København), 127-161. Die Verwandtschaftsverhältnisse des Albanischen zu den übrigen indogermanischen Sprachen. Die Sprache 9, 113-156. Rječnik govora otoka Vrgade II: Rječnik (Zagreb: JAZU). Posti’s superstrate theory at the threshold of the new millennium. Facing Finnic: Some challenges to historical and contact linguistics (Helsinki: Suomalais-Ugrilainen Seura), 80-99. Katz, Joshua T. 1998 Archaische keltische Personalpronomina aus indogermanischer Sicht. Sprache und Kultur der Indogermanen (Innsbruck: Institut für Sprachwissenschaft), 265-291. Kaufmann, F. 1887 Zur geschichte des germanischen consonantismus. Beitraege zur Geschichte der deutschen Sprache und Literatur 12, 504-547. Kern, Johan H. 1906 Zum nom. und acc. plur. der ā-stämme im ags. Beiträge zur Geschichte der deutschen Sprache und Literatur 31, 272-276. Kimball, Sara E. 1987 *H3 in Anatolian. Festschrift for Henry Hoenigswald (Tübingen: Gunter Narr), 185-192. 1999 Hittite historical phonology (Innsbruck: Institut für Sprachwissenschaft). Klaić, A. 1936 O podravskom akcentu i kvantitetu. Južnoslovenski Filolog 15, 181-183. Klein, Thomas 1990 Die Straubinger Heliand-Fragmente: Altfriesisch oder Altsächsisch? Aspects of Old Frisian Philology = Amsterdamer Beiträge zur älteren Germanistik 31-32, 197-225. 458 References Klingenschmitt, Gert 1975 Tocharisch und Urindogermanisch. Flexion und Wortbildung (Wiesbaden: Reichert), 148-163. 1982 Das altarmenische Verbum (Wiesbaden: Reichert). 1994 Das Tocharische in indogermanistischer Sicht. Tocharisch: Akten der Fachtagung der Indogermanischen Gesellschaft [=Tocharian and Indo-European Studies, suppl. 4] (Reykjavík: Málvísindastofnun Háskóla Íslands), 310-411. Kloekhorst, Alwin 2005 Hittite ḫāpūša(šš)- (formerly known as ḫapuš- ‘penis’). Journal of Indo-European Studies 33, 27-39. 2006a Initial laryngeals in Anatolian. Historische Sprachforschung 109, 77-108. 2006b Hittite pai-/pi- ‘to give’. Indogermanische Forschungen 111, 110-119. 2007 The Hittite inherited lexicon (Diss. Leiden). 2008 Etymological dictionary of the Hittite inherited lexicon (Leiden: Brill). Kluge, Friedrich 1884 Die germanische consonantendehnung. Beitraege zur Geschichte der deutschen Sprache und Literatur 9, 149-186. Knobloch, Johannes 1953 La voyelle thématique -e/o- serait-elle un indice d’objet indoeuropéen? Lingua 3, 407-420. Kock, Axel 1888 I-omljudet och den samnordiska förlusten af ändelsevokaler. Arkiv för Nordisk Filologi 4, 141-62. 1891 Fornnordiska kvantitets- och akcentfrågor. Arkiv för Nordisk Filologi 7, 334-377. 1892 Språkhistoriska bidrag. Arkiv för Nordisk Filologi 8, 256-74. 1894 Kritische bemerkungen zur frage nach dem i-umlaut. Beiträge zur Geschichte der deutschen Sprache und Literatur 18, 417-64. 1901 Die alt- und neuschwedische Accentuierung (Strassburg: Trübner). 1911-16 Umlaut und Brechung im Altschwedischen: Eine Übersicht (Lund: Gleerup). References 459 Köhnlein, Björn 2005 Tonakzente im rechtsrheinischen Regelumkehrgebiet: Eine phonetisch-phonologische Analyse 84 Jahre nach Adolf Bachs Entdeckung (Marburg: MA thesis). Koivulehto, Jorma, & Theo Vennemann 1996 Der finnische Stufenwechsel und das Vernersche Gesetz. Beiträge zur Geschichte der deutschen Sprache und Literatur 118, 163-182. Kølln, Herman 1966 Aspekt und Diathese im Slavischen. Scando-Slavica 12, 57-79. 1968 Zur Definition des Verbalaspekts. Scando-Slavica 14, 131-139. Kortlandt, Frederik [numbering of www.kortlandt.nl] K014 1975 Slavic accentuation: A study in relative chronology (Lisse: Peter de Ridder). K017 1976 The Slovene neo-circumflex. The Slavonic and East European Review 54/1, 1-10. K022 1980 Albanian and Armenian. Zeitschrift für vergleichende Sprachforschung 94, 243-251. K023 1978 IE palatovelars before resonants in Balto-Slavic. Recent developments in historical phonology (The Hague: Mouton), 237-243. K025 1977 Historical laws of Baltic accentuation. Baltistica 13/2, 319-330. K027 1980 Zur Akzentuierung der Kiever Blätter. Zeitschrift für slavische Philologie 41/1, 1-4. K028 1978 A history of Slavic accentuation: Review of Garde 1976. Lingua 44/1, 67-91. K030 1978 On the history of the genitive plural in Slavic, Baltic, Germanic, and Indo-European. Lingua 45, 281-300. K031 1978 Notes on Armenian historical phonology II. Studia Caucasica 4, 9-16. K032 1978 Proto-Indo-European obstruents. Indogermanische Forschungen 83, 107-118. K033 1979 Toward a reconstruction of the Balto-Slavic verbal system. Lingua 49, 51-70. K034 1980 H2o and oH2. Lingua Posnaniensis 23 [Fs. Kudzinowski], 127-128. [cf. this volume, 51-52] 460 K035 K036 K038 K044 K046 K049 K050 K051 K052 K055 K058 K060 K061 K062 K063 K064 K065 References 1979 The Old Irish absolute and conjunct endings and questions of relative chronology. Ériu 30, 35-53. 1979 On the history of the Slavic nasal vowels. Indogermanische Forschungen 84, 259-272. 1981 Glottalic consonants in Sindhi and Proto-Indo-European. Indo-Iranian Journal 23: 15-19. [cf. this volume, 121-124] 1981 1st sg. middle *-H2. Indogermanische Forschungen 86, 123-136. [cf. this volume, 81-90] 1981 More evidence for Italo-Celtic. Ériu 32 (1981), 1-22. 1983 Proto-Indo-European verbal syntax. Journal of IndoEuropean Studies 11, 307-324. [cf. this volume, 91-103] 1982 Innovations which betray archaisms. Baltistica 18/1, 4-9. 1982 Phonemicization and rephonemicization of the Old Irish mutations. Ériu 33, 73-83. 1982 Sravnitel’no-istoričeskie kommentarii k bolgarskomu udareniju. Zbornik za Filologiju i Lingvistiku 25/1, 91-96. 1983 On final syllables in Slavic. Journal of Indo-European Studies 11, 167-185. 1982 Early dialectal diversity in South Slavic I. Studies in Slavic and General Linguistics 2: South Slavic and Balkan Linguistics, 177-192. 1983 Greek numerals and PIE glottalic consonants. Münchener Studien zur Sprachwissenschaft 42: 97-104. [cf. this volume, 105-109] 1983 Notes on Armenian historical phonology III: h-. Studia Caucasica 5, 9-16. 1984 Proto-Armenian case endings. International Symposium on Armenian linguistics: Reports (Erevan: Armenian Academy of Sciences), 97-106. 1983 Demonstrative pronouns in Balto-Slavic, Armenian, and Tocharian. Studies in Slavic and General Linguistics 3, 311-22. 1985 Long vowels in Balto-Slavic. Baltistica 21/2: 112-124. 1987 Archaic ablaut patterns in the Vedic verb. Festschrift for Henry Hoenigswald (Tübingen: Gunter Narr), 219-223. [cf. this volume, 125-129] References K066 K067 K068 K069 K070 K071 K072 K073 K075 K075a K080 K082 K084 K085 461 1989 Od praindoevropskog jezika do slovenskog (Fonološki razvoj). Zbornik za Filologiju i Lingvistiku 32/2, 41-58. 1985 A parasitological view of non-constructible sets. Studia linguistica diachronica et synchronica: Werner Winter sexagenario oblata (Berlin: Mouton de Gruyter), 477-483. [cf. this volume, 431-435] 1984 Old Irish subjunctives and futures and their Proto-IndoEuropean origins. Ériu 35, 179-187. 1985 On reduced vowels in Slavic. Zbornik za Filologiju i Lingvistiku 27-28, 367-368. 1986 Posttonic *w in Old Irish. Ériu 37, 89-92. 1986 The origin of the Slavic imperfect. Festschrift für Herbert Bräuer zum 65. Geburtstag am 14. April 1986 (Köln: Böhlau), 253-258. 1988 Vestjysk stød, Icelandic preaspiration, and Proto-IndoEuropean glottalic stops. Languages and cultures: Studies in honor of Edgar C. Polomé (Berlin: Mouton de Gruyter), 353-357. [cf. this volume, 165-168] 1984 PIE *H- in Armenian. Annual of Armenian Linguistics 5, 41-43. 1985 Proto-Indo-European glottalic stops: The comparative evidence. Folia Linguistica Historica 6/2, 183-201. [cf. this volume, 53-65] 1985 Praindoevropejskie glottalizovannye smyčnye (Sravnitel’noistoričeskie dannye). Voprosy Jazykoznanija 1985/4, 43-53. 1986 Proto-Indo-European tones? Journal of Indo-European Studies 14, 153-160. [cf. this volume, 67-72] 1985 Slavic imamь. International Journal of Slavic Linguistics and Poetics 31-32 [Fs. Birnbaum], 235-239. 1986 The origin of the Old English dialects. Linguistics across historical and geographical boundaries I: Linguistic theory and historical linguistics [Fs. Fisiak] (Berlin: Mouton de Gruyter), 437-442. [cf. this volume, 259-263] 1986 Armenian and Albanian. La place de l’arménien dans les langues indo-européennes (Leuven: Peeters), 38-47. 462 K086 K087 K088 K090 K094 K095 K096 K097 K098 K099 K101 K102 K104 K105 K107 K109 K110 References 1989 Lachmann’s law. The new sound of Indo-European: Essays in phonological reconstruction (Berlin: Mouton de Gruyter), 103-105. 1987 The formation of the Old Prussian present tense. Baltistica 23/2, 104-111. 1986 The Germanic first class of weak verbs. North-Western European Language Evolution 8, 27-31. [cf. this volume, 201-203] 1987 Notes on Armenian historical phonology V. Studia Caucasica 7, 61-65. 1987 PIE *s in Albanian. Studies in Slavic and General Linguistics 10, 219-226. [cf. this volume, 319-324] 1988 On the development of PIE final syllables in Tocharian. Tocharian and Indo-European Studies 2, 80-88. [cf. this volume, 143-147] 1988 The Tocharian word for ‘woman’. Tocharian and IndoEuropean Studies 2, 77-79. [cf. this volume, 149-150] 1992 The Aeolic optative. Rekonstruktion und relative Chronologie: Akten der VIII. Fachtagung der Indogermanischen Gesellschaft, Leiden, 1987 (Innsbruck: Institut für Sprachwissenschaft), 235-239. [cf. this volume, 111-115] 1988 The Greek 3rd pl. endings. Münchener Studien zur Sprachwissenschaft 49, 63-69. [cf. this volume, 117-120] 1989 Lithuanian statýti and related formations. Baltistica 25/2, 104-112. 1988 The Thraco-Armenian consonant shift. Linguistique Balkanique = Balkansko Ezikoznanie 31, 71-74. 1988 Proto-Germanic obstruents. Amsterdamer Beiträge zur älteren Germanistik 27, 3-10. [cf. this volume, 169-174] 1989 The making of a puzzle. Annual of Armenian Linguistics 10, 43-52. 1990 The Germanic third class of weak verbs. North-Western European Language Evolution 15, 3-10. [cf. this volume, 205-208] 1989 Der polabische Wortakzent. Zeitschrift für slavische Philologie 49/1 (1989), 163-170. 1989 The Germanic weak preterit. Amsterdamer Beiträge zur älteren Germanistik 28, 101-109. [cf. this volume, 227-233] 1993 Tokie šalti rytai. Baltistica 28/1, 45-48. References K111 K112 K112a K114 K115 K117 K118 K119 K120 K121 K122 K123 K124 K128 K130 463 1990 The spread of the Indo-Europeans. Journal of IndoEuropean Studies 18, 131-140. [cf. this volume, 1-6] 1989 Eight Indo-Uralic verbs? Münchener Studien zur Sprachwissenschaft 50, 79-85. [cf. this volume, 389-392] 1992 Vosem’ indo-ural’skix glagolov? Voprosy Jazykoznanija 1992/1, 101-104. 1994 The fate of the sigmatic aorist in Tocharian. Tocharisch: Akten der Fachtagung der Indogermanischen Gesellschaft, Berlin, 1990 [=Tocharian and Indo-European Studies, suppl. 4] (Reykjavík: Málvísindastofnun Háskóla Íslands), 61-65. [cf. this volume, 151-153] 1991 The Germanic seventh class of strong verbs. North-Western European Language Evolution 18, 97-100. [cf. this volume, 209-210] 1992 The Old Norse i-umlaut. North-Western European Language Evolution 20, 27-31. [cf. this volume, 285-287] 1991 A note on the Tocharian dual. Tocharian and Indo-European Studies 5, 5-10. [cf. this volume, 155-157] 1991 Kluge’s law and the rise of Proto-Germanic geminates. Amsterdamer Beiträge zur älteren Germanistik 34, 1-4. [cf. this volume, 175-177] 1992 The Germanic fifth class of strong verbs. North-Western European Language Evolution (Odense University Press) 19, 101-107. [cf. this volume, 211-214] 1992 Le statif indo-européen en slave. Revue des Études Slaves 64/3, 373-376. 1994 The Germanic sixth class of strong verbs. North-Western European Language Evolution 23, 69-73. [cf. this volume, 215-217] 1993 Old High German umlaut. Amsterdamer Beiträge zur älteren Germanistik 37, 19-20. [cf. this volume, 247-248] 1994 On breaking. North-Western European Language Evolution 24, 15-19. [cf. this volume, 289-291] 1994 The Proto-Germanic pluperfect. Amsterdamer Beiträge zur älteren Germanistik 40, 1-5. [cf. this volume, 235-238] 1995 General linguistics and Indo-European reconstruction. Rask 2, 91-109. [cf. this volume, 7-20] 464 K135 K138 K141 K142 K143 K145 K146 K149 K150 K156 K161 K167 K173 K180 K182 References 1995 The Germanic fourth class of weak verbs. North-Western European Language Evolution (Odense University Press) 25, 137-139. [cf. this volume, 219-220] 1996 The High German consonant shift. Amsterdamer Beiträge zur älteren Germanistik 46, 53-57; Korrektur in Amsterdamer Beiträge zur älteren Germanistik 47, 231. [cf. this volume, 249-253] 1997 PIE lengthened grade in Balto-Slavic. Festschrift for Eric P. Hamp II (Washington: Institute for the Study of Man), 26-31. 1997 How old is the English glottal stop? North-Western European Language Evolution 31-32 [Germanic studies in honor of Anatoly Liberman], 175-179. [cf. this volume, 265-268] 1997 Labials, velars and labiovelars in Germanic. North-Western European Language Evolution 30, 45-50. [cf. this volume, 179-183] 1998 The dual endings of the Indo-European verb. Studia Indogermanica Lodziensia 2, 71-73. 1996 PIE *j in Albanian. Studies in Slavic and General Linguistics 23, 173-176. [cf. this volume, 325-327] 1996 The Tocharian imperfect. Historische Sprachforschung 109, 169-174. [cf. this volume, 159-163] 1997 Thematic and athematic verb forms in Old Irish. Sound law and analogy: Papers in honor of Robert S.P. Beekes on the occasion of his 60th birthday (Amsterdam: Rodopi), 133-137. 1998 Are Mongolian and Tungus genetically related? Acta Orientalia [Academiae Scientiarum Hungaricae] 51, 235-237. 1997 Japanese aru, iru, oru ‘to be’. Studia Etymologica Cracoviensia 2, 167-170. 1997 On the relative chronology of Celtic sound changes. Historische Sprachforschung 110/2, 248-251. 1997 Baltic ē- and ī/jā-stems. Baltistica 32/2, 157-163. 1999 The origin of the Old English dialects revisited. Amsterdamer Beiträge zur älteren Germanistik 51, 45-51. [cf. this volume, 269-273] 2000 Old Norse taka, Gothic tekan, Greek τεταγών. NorthWestern European Language Evolution 36, 59-65. [cf. this volume, 221-225] References K183 K187 K188 K190 K192 K194 K195 K197 K198 K201 K202 K203 K205 K206 K208 465 1998 Reflexes of Indo-European consonants in Albanian. Orpheus 8, 35-37. [cf. this volume, 329-332] 2003 An Indo-European substratum in Slavic? Languages in prehistoric Europe (Heidelberg: Winter), 253-260. [cf. this volume, 73-80] 2004 Accent and ablaut in the Vedic verb. Indo-Iranian Journal 47/1, 7-15. [cf. this volume, 131-137] 2000 Initial a- and e- in Old Prussian. Linguistica Baltica 8, 125-127. 2000 Preaspiration or preglottalization? Amsterdamer Beiträge zur älteren Germanistik 53, 7-10. [cf. this volume, 185-187] 2003 Armeniaca: Comparative notes (Ann Arbor: Caravan Books). 2000 Old Prussian participles. Res Balticae 6, 69-75. 2000 On Russenorsk. Amsterdamer Beiträge zur älteren Germanistik 54, 123-127. [cf. this volume, 21-25] 2001 The origin of the Goths. Amsterdamer Beiträge zur älteren Germanistik 55, 21-25. [cf. this volume, 27-30] 2001 OPr. -snā, Lith. -sena, Latv. -šana. Mvnera lingvistica et philologica [Fs. Hasiuk] (Poznań: UAM), 137-139. 2004 Initial laryngeals in Anatolian. Orpheus 13-14, 9-12. [cf. this volume, 365-368] 2002 The Indo-Uralic verb. Finno-Ugrians and Indo-Europeans: Linguistic and literary contacts (Maastricht: Shaker), 217-227. [cf. this volume, 393-405] 2004 Nivkh as a Uralo-Siberian language. Per aspera ad asteriscos [Fs. Rasmussen] (Innsbruck: Institut für Sprachwissenschaft), 285-289. [cf. this volume, 407-410] 2003 The origin and nature of the linguistic parasite. Language in time and space [Fs. Winter] (Berlin: Mouton), 241-244. [cf. this volume, 437-439] 2003 Early dialectal diversity in South Slavic II. Dutch contributions to the 13th international congress of slavists, Ljubljana: Linguistics [=Studies in Slavic and General Linguistics 30], 215-235. 466 K211 K212 K213 K216 K217 K218 K219 K221 K222 K223 K225 K226 K228 K229 K230 References 2003 Glottalization, preaspiration and gemination in English and Scandinavian. Amsterdamer Beiträge zur älteren Germanistik 58, 5-10. [cf. this volume, 293-297] 2003 Early Runic consonants and the origin of the younger futhark. North-Western European Language Evolution 43, 71-76. [cf. this volume, 299-303] 2004 Indo-Uralic consonant gradation. Etymologie, Entlehnungen und Entwicklungen [Fs. Koivulehto] (Helsinki: Société Néophilologique), 163-170. [cf. this volume, 411-416] 2004 Indo-Uralic and Altaic. [this volume, 417-420] 2005 The inflexion of the Indo-European ā-stems in Germanic. Amsterdamer Beiträge zur älteren Germanistik 60, 1-4. [cf. this volume, 239-241] 2004 Balto-Slavic accentuation: Some news travels slowly. Baltistica 39/1, 13-17. 2006 The inflexion of the Germanic n-stems. North-Western European Language Evolution 48, 3-7. [cf. this volume, 243-245] 2005 On the accentuation of the illative. Baltu Filoloģija 14/1, 67-69. 2005 From Serbo-Croatian to Indo-European. Wiener slavistisches Jahrbuch 51, 113-130. 2005 Hittite ammuk ‘me’. Orpheus 15, 7-10. [cf. this volume, 369-372] 2006 Germanic *ē1 and *ē2. North-Western European Language Evolution 49, 51-54. [cf. this volume, 189-191] 2007 Miscellaneous remarks on Balto-Slavic accentuation. Tones and theories: Proceedings of the international workshop on BaltoSlavic accentology (Zagreb: Institut za hrvatski jezik i jezikoslovlje), 229-235. 2005 Lithuanian tekėti and related formations. Baltistica 40/2, 167-170. 2007 The origin of the Franconian tone accents. Amsterdamer Beiträge zur älteren Germanistik 63, 1-3. [cf. this volume, 255-257] 2007 Proto-Germanic obstruents and the comparative method. North-Western European Language Evolution 52, 3-7. [cf. this volume, 193-196] References K234 K235 K236 K238 K239 K240 K241 K244 K245 K247 K248 K249 K250 K251 K253 K254 467 2006 Balto-Slavic accentual mobility. Baltistica 41/3, 359-369. 2007 English bottom, German Boden, and the chronology of sound shifts. Amsterdamer Beiträge zur älteren Germanistik 63, 5-8. [cf. this volume, 197-199] 2009 Accent retraction and tonogenesis. Stressing the past: Papers on Baltic and Slavic accentology [=Studies in Slavic and General Linguistics 35], 75-82. 2008 Anglo-Frisian. North-Western European Language Evolution 54-55, 265-278. [cf. this volume, 275-284] 2007 Italo-Celtic origins and prehistoric development of the Irish language (Amsterdam: Rodopi). 2007 Gothic gen.pl. -e. Historische Sprachwissenschaft 120, 237-240. 2008 Hittite hi-verbs and the Indo-European perfect. [this volume, 373-383] 2008 Stative and middle in Hittite and Indo-European. [this volume, 385-388] 2007 The development of the Indo-European syllabic resonants in Balto-Slavic. Baltistica 42/1, 7-12. 2007 C.C. Uhlenbeck on Indo-European, Uralic and Caucasian. [this volume, 31-36] 2008 Arm. ewł ‘oil’. Aramazd 3/1, 40-41. [cf. this volume, 333] 2008 Slavic historical morphology: Nominal paradigms. Dutch contributions to the 14th international congress of slavists, Ohrid: Linguistics [=Studies in Slavic and General Linguistics 34], 397-429. 2008 Björketorp and Stentoften. North-Western European Language Evolution 53, 19-27. [cf. this volume, 305-311] 2009 The origin of the vestjysk stød. Amsterdamer Beiträge zur älteren Germanistik 65, 1-4. [cf. this volume, 313-315] 2008 Balto-Slavic phonological developments. Baltistica 43/1, 5-15. 2008 The origin of the Indo-Iranian desiderative. Indologica [=Gs. Elizarenkova] (Moskva: RGGU), 227-230. [cf. this volume, 139-142] 468 K256 References 2010 Indo-Uralic and Altaic revisited. Transeurasian verbal morphology in a comparative perspective: genealogy, contact, chance [=Turcologica 78] (Wiesbaden: Harrassowitz), 153-164. [cf. this volume, 421-430] K261 2009 The Baltic word for ‘in’. Baltistica 44/1, 33-35. [cf. this volume, 335-336] K263 2009 Baltica & Balto-Slavica (Amsterdam: Rodopi). K264 2009 All’s well that ends well. Baltistica 44/1, 59-63. [cf. this volume, 337-339] K265 2010 Glottalization and tonogenesis in Athabaskan, Balto-Slavic and Germanic. Suvremena Lingvistika 10, 00-00. K266 2010 Vestjysk stød again. Amsterdamer Beiträge zur älteren Germanistik 66, 29-32. [cf. this volume, 317-318] K267 2009 Balto-Slavic accentuation revisited. [this volume, 341-358] K268 2009 More on the chronology of Celtic sound changes. [this volume, 361-364] K269 2010 An outline of Proto-Indo-European. [this volume, 37-45] K270 2010 Lithuanian žinóti ‘to know’. Baltistica 45/1, 00-00. [cf. this volume, 359-360] K271 2010 Schleicher’s fable. [this volume, 47-50] Krause, Wolfgang 1953 Handbuch des Gotischen (München: Beck). 1966 Die Runeninschriften im älteren Futhark I: Text (Göttingen: Vandenhoeck & Ruprecht). Krause, Wolfgang, & Werner Thomas 1960 Tocharisches Elementarbuch I: Grammatik (Heidelberg: Carl Winter). Krejnovič, Eruxim A. 1958 Ob inkorporirovanii v nivxskom jazyke. Voprosy Jazykoznanija 7/6, 21-33. Kroonen, Guus 2006 Gemination and allomorphy in the Proto-Germanic mn-stems: bottom and rime. Amsterdamer Beiträge zur älteren Germanistik 61, 17-25. 2009 Consonant and vowel gradation in the Proto-Germanic n-stems (Diss. Leiden). References 469 Krupatkin, Y.B. 1970 From Germanic to English and Frisian. Us Wurk 19/3, 49-71. Kuiper, Franciscus B.J. 1934 Zur Geschichte der indoiranischen s-Präsentia. Acta Orientalia 12, 190-306. 1942 Notes on Vedic noun-inflexion (Amsterdam: KNAW). 1995 Gothic bagms and Old Icelandic ylgr. North-Western European Language Evolution 25, 63-88. Kuipers, Aert H. 1962 The Circassian nominal paradigm: A contribution to casetheory. Lingua 11, 231-248. 1968 Unique types and typological universals. Pratidānam: Indian, Iranian and Indo-European studies presented to F.B.J. Kuiper on his 60th birthday (The Hague: Mouton), 68-88. 1974 The Shuswap Language (The Hague: Mouton). Kulikov, Leonid 2000 The Vedic type syáti revisited. Indoarisch, Iranisch und die Indogermanistik (Wiesbaden: Reichert), 267-283. 2001 The Vedic -ya-presents (Diss. Leiden). 2005 Reduplication in the Vedic verb: Indo-European inheritance, analogy and iconicity. Studies on Reduplication (Berlin: Mouton de Gruyter), 431-454. Kuryłowicz, Jerzy 1964 The inflectional categories of Indo-European (Heidelberg: Carl Winter). Kuz’mina, Elena E. 2007 The origin of the Indo-Iranians (Leiden: Brill). Lane, George S. 1953 Imperfect and preterit in Tocharian. Language 29, 278-287. 1959 The formation of the Tocharian subjunctive. Language 35, 157-179. 1963 Bimoric and trimoric vowels and diphthongs: Laws of Germanic finals again. Journal of English and Germanic Philology 62, 155-170. 1976 Notes sur le sort des syllabes finales i.e. en tokharien. Bulletin de la Société de Linguistique de Paris 71/1, 133-164. 470 References Langston, Keith 2007 Common Slavic accentual paradigm (d): A reevaluation of evidence from Čakavian. Tones and theories: Proceedings of the international workshop on Balto-Slavic accentology (Zagreb: Institut za hrvatski jezik i jezikoslovlje), 121-150. Laroche, Emmanuel 1956 Études de linguistique anatolienne. Revue Hittite et Asianique 23, 33-54. Lehmann, Winfred P. 1965 Germanic evidence. Evidence for laryngeals (The Hague: Mouton), 212-223. 1986 A Gothic etymological dictionary (Leiden). Lehmann, Winfred P., & Ladislav Zgusta 1979 Schleicher’s tale after a century. Festschrift for Oswald Szemerényi on the occasion of his 65th birthday I (Amsterdam: Benjamins), 455-466. Lejeune, Michel 1972 Phonétique historique du mycénien et du grec ancien (Paris: Klincksieck). Leumann, Manu 1952 Morphologische Neuerungen im altindischen Verbalsystem (Amsterdam: KNAW). Liberman, Anatoly S. 1972 The glottal stop in English as viewed against its Germanic background. Kalbotyra 23/3, 45-57. 1982 Germanic Accentology I: The Scandinavian Languages (Minneapolis: University of Minnesota Press). Lid, Nils 1952 Den nordiske nominativ singularis av maskuline an-stammer. Norsk Tidsskrift for Sprogvidenskap 16, 237-240. Loma, Aleksandar 1990 Pozajmljenice iz nepoznatog jezika u praslovenskom [Review of Holzer 1989]. Južnoslovenski Filolog 46, 87-122. References 471 Lubotsky, Alexander 1981 Gr. πήγνυμι : Skt. pajrá- and loss of laryngeals before mediae in Indo-Iranian. Münchener Studien zur Sprachwissenschaft 40, 133-138. 1985 The PIE word for ‘dry’. Zeitschrift für vergleichende Sprachforschung 98, 1-10. 1987 Nominal accentuation in Sanskrit and Proto-Indo-European (Diss. Leiden). 1988a The system of nominal accentuation in Sanskrit and Proto-IndoEuropean (Leiden: Brill). 1988b The Old Phrygian Areyastis inscription. Kadmos 27/1, 9-26. 1989 Against a Proto-Indo-European phoneme *a. The new sound of Indo-European: Essays in phonological reconstruction (Berlin: Mouton de Gruyter), 53-66. 1990 La loi de Brugmann et *H3e-. La reconstruction des laryngales (Paris: Les Belles Lettres), 129-136. 1994 RV. ávidhat. Früh-, Mittel-, Spätindogermanisch (Wiesbaden: Reichert), 201-206. 1997 Remarks on the Vedic intensive [Review of Schaefer 1994]. Journal of the American Oriental Society 117, 558-564. 1998 New Phrygian metrics and the δεως ζεμελως formula. Mír Curad: Studies in honor of Calvert Watkins (Innsbruck: Institut für Sprachwissenschaft), 413-421. 2001 Reflexes of Proto-Indo-European *sk in Indo-Iranian. Incontri Linguistici 24, 25-57. Lühr, Rosemarie 1979 Das Wort ‘und’ im Westgermanischen. Münchener Studien zur Sprachwissenschaft 38, 117-154. 1980 Zu einem urgermanischen Lautgesetz. Lautgeschichte und Etymologie (Wiesbaden: Reichert), 248-259. 1984 Reste der athematischen Konjugation in den germanischen Sprachen. Das Germanische und die Rekonstruktion der indogermanischen Grundsprache, 25-90. 1988 Expressivität und Lautgesetz im Germanischen (Heidelberg: Carl Winter). 472 Luick, Karl 1964 References Historische Grammatik der englischen Sprache (Stuttgart: Tauchnitz). Lunden, Siri Sverdrup 1978 Tracing the ancestry of Russenorsk. Slavia Orientalis 27/2, 213-217. Mahlow, Georg H. 1879 Die langen Vocale ā ē ō in den europäischen Sprachen (Berlin: Hermann). Mallory, James P. 1989 In search of the Indo-Europeans: Language, archaeology and myth (London: Thames & Hudson). Mańczak, Witold 1982 Kamen die Goten aus Skandinavien? Indogermanische Forschungen 87, 127-137. 1984 Origine méridionale du gotique. Diachronica 1, 79-102. 1987a On the Ausgliederung of Germanic languages. Journal of IndoEuropean Studies 15, 1-17. 1987b L’habitat primitif des Goths. Folia Linguistica Historica 7/2, 371-380. 1992 De la préhistoire des peuples indo-européens. Kraków: Uniwersytet Jagelloński. Marstrander, Carl J.S. 1932 Okklusiver og substrater. Norsk Tidsskrift for Sprogvidenskap 5, 258-314. Martinet, André 1953 Remarques sur le consonantisme sémitique. Bulletin de la Société de Linguistique de Paris 49: 67-78. Mattissen, Johanna 2001 Dependent-head synthesis in Nivkh and its contribution to a typology of polysynthesis (Köln: Institut für Sprachwissenschaft). Matzinger, Joachim 2006 Altarmenisch ewł/iwł ‘Öl’. In the Orient where the gracious light... [Fs. Pisowicz] (Kraków: Księgarnia Akademicka), 71-72. References 473 Mayrhofer, Manfred 1983 Sanskrit und die Sprachen Alteuropas: Zwei Jahrhunderte des Widerspiels von Entdeckungen und Irrtümern (Göttingen: Vandenhoeck & Ruprecht). McCawley, James D. 1978 Notes on the history of accent in Japanese. Recent developments in historical phonology (The Hague: Mouton), 287-307. McCone, Kim 1996 Towards a relative chronology of ancient and medieval Celtic sound change (Maynooth: Department of Old Irish). McNeill, William H. 1976 Plagues and peoples (Garden City: Anchor Press). Meid, Wolfgang 1971 Das germanische Praeteritum (Innsbruck: Institut für Sprachwissenschaft). Meillet, Antoine 1894 De quelques difficultés de la théorie des gutturales indoeuropéennes. Mémoires de la Société de Linguistique de Paris 8, 277-304. 1903 Observations sur la graphie de quelques anciens manuscrits de l’évangile arménien. Journal Asiatique 10/2, 487-507. 1904 Remarques sur la grammaire historique de l’arménien de Cilicie de M. J. Karst. Zeitschrift für armenische Philologie 2, 18-28. 1908 Les dialectes indo-européens (Paris: Champion). 1918 À propos de latin formīca. Mémoires de la Société de Linguistique de Paris 20, 115. 1920 Sur le rythme quantitatif de la langue védique. Mémoires de la Société de Linguistique de Paris 21: 193-207. 1931a Essai de chronologie des langues indo-européennes: La théorie du féminin. Bulletin de la Société de Linguistique de Paris 32, 1-28. 1931b Caractère secondaire du type thématique indo-européen. Bulletin de la Société de Linguistique de Paris 32, 194-203. 1936 Esquisse d’une grammaire comparée de l’arménien classique (Vienne: Mékhitharistes). 1937 Introduction à l’étude comparative des langues indo-européennes (Paris: Klincksieck). 474 1964 References Introduction à l’étude comparative des langues indo-européennes (Alabama UP). Melchert, H. Craig 1977 Tocharian verb stems in -tk-. Zeitschrift für vergleichende Sprachforschung 91, 93-130. 1984 Studies in Hittite historical phonology (Göttingen: Vandenhoeck & Ruprecht). 1987 Reflexes of *h3 in Anatolian. Die Sprache 33, 19-28. 1994 Anatolian historical phonology (Amsterdam: Rodopi). Melchert, H. Craig, & Norbert Oettinger 2009 Ablativ und Instrumental im Hethitischen und Indogermanischen: Ein Beitrag zur relativen Chronologie. Incontri Linguistici 32, 53-73. Meyer, Denise P. 1997 Greek pronouns in σφ- and the PIE personal pronominal system. Historische Sprachforschung 110, 93-108. Meyer, Gustav 1891 Etymologisches Wörterbuch der albanesischen Sprache (Strassburg: Trübner). 1892 Albanesische Studien III: Lautlehre der indogermanischen Bestandtheile des Albanesischen (=Sitzungsberichte der philosophisch-historischen Classe der Kaiserlichen Akademie der Wissenschaften 125/11, Wien). Migron, S. 1975 The Rgvedic passive aorist in -i: A functional study. Folia Linguistica 8, 271-310. Milroy, James, & Lesley Milroy 1985 Linguistic change, social network and speaker innovation. Journal of Linguistics 21, 339-384. Monna, Maria C. 1978 The Gathas of Zarathustra: A reconstruction of the text (Amsterdam: Rodopi). Moszyński, Leszek 1992 Review of Holzer 1989. Rocznik Slawistyczny 48/1, 88-95. References 475 Mottausch, Karl-Heinz 1993 Zwei verkannte germanisch-italische Isoglossen. Historische Sprachforschung 106, 148-175. 1994 Idg. *h1ei- ‘gehen’ im Germanischen. Historische Sprachforschung 107, 123-140. Moulton, William G. 1954 The stops and spirants of early Germanic. Language 30, 1-42. Munske, Horst Haider 1986 What are mixed languages? Language contact in Europe [Linguistische Arbeiten, 168] (Tübingen: Niemeyer), 81-95. Naert, Pierre 1962 Contacts lexicaux aïnou-gilyak. Orbis 11, 199-229. Narten, Johanna 1964 Die sigmatischen Aoriste im Veda (Wiesbaden: Harrassowitz). Nedoma, Robert 2005 Urnordisch -a im Nominativ Singularis der maskulinen n-Stämme. North-Western European Language Evolution 46/47, 155-191. Nielsen, Hans F. 1981a Old English and the Continental Germanic languages (Innsbruck: Institut für Sprachwissenschaft). 1981b Old Frisian and Old English dialects. Us Wurk 30/2, 49-66. 1984 A note on the origin of Old English breaking and back mutation. Amsterdamer Beiträge zur älteren Germanistik 22, 73-81. 2000 The Early Runic language of Scandinavia (Heidelberg: Carl Winter). 2001 Frisian and the grouping of the older Germanic languages. Handbuch des Friesischen = Handbook of Frisian studies, (Tübingen: Niemeyer), 512-523. Nielsen, Karl M. 1961 Scandinavian breaking. Acta Philologica Scandinavica 24, 33-45. Nihalani, Paroo 1974 An aerodynamic study of stops in Sindhi. Phonetica 29, 193-224. Nikolaeva, Irina 1999 Object agreement, grammatical relations, and information structure. Studies in Language 23/2, 331-376. 476 References Noreen, Adolf 1970 Altnordische Grammatik I: Altisländische und altnorwegische Grammatik (Laut- und Flexionslehre) unter Berücksichtigung des Urnordischen (5. Auflage, Tübingen: Niemeyer). O’Connor, Joseph D. 1952 RP and the reinforcing glottal stop. English Studies 33, 214-18. Oettinger, Norbert 1976 Der indogermanische Stativ. Münchener Studien zur Sprachwissenschaft 34, 109-149. 1979 Die Stammbildung des hethitischen Verbums (Nürnberg: Hans Carl). 1988 Der indogermanische Nominativ Dual aus laryngalistischer Sicht. Die Laryngaltheorie und die Rekonstruktion des indogermanischen Laut- und Formensystems (Heidelberg: Carl Winter), 355-359. Oftedal, Magne 1947 Jærske okklusivar. Norsk Tidsskrift for Sprogvidenskap 14, 229-235. Olander, Thomas 2006 Accentual mobility: The prehistory of the Balto-Slavic mobile accent paradigms (Diss. Copenhagen). 2009 Balto-Slavic accentual mobility (Berlin: Mouton de Gruyter). Orel, Vladimir E. 1985 PIE *s in Albanian. Die Sprache 31, 279-285. 1989 PIE *j in Albanian. Folia Linguistica Historica 8, 37-49. Orel, Vladimir E., & Olga V. Stolbova 1995 Hamito-Semitic etymological dictionary (Leiden: Brill). Orton, Harold, Stewart Sanderson & John Widdowson (eds.) 1978 The linguistic atlas of England (London: Croom Helm). Osthoff, Hermann 1882 Ueber aoristpraesens und imperfectpraesens. Beitraege zur Geschichte der deutschen Sprache und Literatur 8, 287-311. Page, B.R. 1997 On the origin of preaspiration in Scandinavian. American Journal of Germanic Linguistics and Literatures 9/2, 167-190. Panfilov, Vladimir Z. 1962 Grammatika nivxskogo jazyka I (Moskva-Leningrad: AN SSSR). References 477 1965 Grammatika nivxskogo jazyka II (Moskva-Leningrad: Nauka). Paul, Hermann 1880 Beitraege zur geschichte der lautentwickelung und formenassociation. Beitraege zur Geschichte der deutschen Sprache und Literatur 7, 105-170. Pedersen, Anders 1912 Dansk og urnordisk akcentuering. Arkiv för Nordisk Filologi 28, 1-53. Pedersen, Holger 1895 Das indogermanische s im Slavischen. Indogermanische Forschungen 5, 33-87. 1897 Das albanesische neutrum. Zeitschrift für vergleichende Sprachforschung 34, 283-291. 1900a Wie viel laute gab es im Indogermanischen? Zeitschrift für vergleichende sprachforschung 36, 74-110. 1900b Die gutturale im Albanesischen. Zeitschrift für vergleichende Sprachforschung 36, 277-340. 1906 Armenisch und die nachbarsprachen. Zeitschrift für vergleichende Sprachforschung 39: 334-484. 1907 Neues und nachträgliches. Zeitschrift für vergleichende Sprachforschung 40, 129-217. 1921 Les formes sigmatiques du verbe latin et le problème du futur indoeuropéen (København: Høst & Søn). 1926 La cinquième déclinaison latine (København: Høst & Søn). 1933a Études lituaniennes (København: Levin & Munksgaard). 1933b Zur Frage nach der Urverwandtschaft des Indoeuropäischen mit dem Ugrofinnischen. Liber Semisaecularis Societatis FennoUgricae (Helsinki: Suomalais-Ugrilainen Seura) = Mémoires de la Société Finno-Ougrienne 67, 308-325. 1938 Hittitisch und die anderen indoeuropäischen Sprachen [Historiskfilologiske Meddelelser 25/2] (København: Levin & Munksgaard). 1941 Tocharisch vom Gesichtspunkt der indoeuropäischen Sprachvergleichung [Historisk-filologiske Meddelelser 28/1] (København: Munksgaard). 1942 Er stødet en konsonant? Acta Philologica Scandinavica 16, 111-120. 478 1944 References Zur tocharischen Sprachgeschichte [Historisk-filologiske Meddelelser 30/2] (København). 1951 Die gemeinindoeuropäischen und die vorindoeuropäischen Verschlusslaute [Historisk-filologiske Meddelelser 32/5] (København: Munksgaard). 1962 The discovery of language: Linguistic science in the 19th century (Bloomington: Indiana University Press) Penney, John H.W. 1989 Preverbs and postpositions in Tocharian. Transactions of the Philological Society 87, 54-74. Perridon, Harry 1991 The Reistad stone and the history of the Scandinavian languages. Amsterdamer Beiträge zur älteren Germanistik 34, 5-17. 2002 The quantity shift in North Germanic. Amsterdamer Beiträge zur älteren Germanistik 56, 69-77. 2006 On the origin of the vestjysk stød. Amsterdamer Beiträge zur älteren Germanistik 62, 41-50. 2008 Reconstructing the obstruents of Proto-Germanic. Evidence and counter-evidence [Fs. Kortlandt], vol. 1 [=SSGL 32]: Balto-Slavic and Indo-European linguistics, 415-429. 2009 How old is the vestjysk stød? Amsterdamer Beiträge zur älteren Germanistik 65, 5-10. Petersen, Hjalmar P. et al. 1998 Faroese: An overview for students and researchers (TórshavnReykjavík). Petersen, Walter 1936 The personal endings of the middle voice. Language 12, 157-174. Peterson, R.E. 1980 Russenorsk: A little known aspect of Russian-Norwegian relations. Studies in language 4/2, 249-256. Peyrot, Michaël 2008 Variation and change in Tocharian B (Amsterdam: Rodopi). 2009 On the formation of the Tocharian preterite participle. Historische Sprachforschung 121, 00-00. References 479 Pinault, Georges 1989 Introduction au tokharien. LALIES: Actes des sessions de linguistique et de littérature d'Aussois (Presses de l’École normale supérieure, Paris) 7, 5-224. Pisowicz, Andrzej 1976a Le développement du consonantisme arménien (Wrocław: Polska Akademia Nauk). 1976b Matériaux pour servir à la recherche du consonantisme arménien. Folia Orientalia 17: 197-216. Pokorny, Julius 1959 Indogermanisches etymologisches Wörterbuch I (Bern: Francke). Polomé, Edgar C. 1987 Initial PIE *gwh- in Germanic. Festschrift for Henry Hoenigswald (Tübingen: Gunter Narr), 303-313. Posti, Lauri 1953 From Pre-Finnic to Late Proto-Finnic: Studies on the development of the consonant system. Finnisch-Ugrische Forschungen 31, 1-91. Pottier, Bernard 1968 L’emploi de la préposition a devant l’objet en espagnol. Bulletin de la Société de Linguistique de Paris 63, 83-95. Price, Glanville 1979 The French language: present and past (London: Arnold). Proeme, Henk 1980 On aspectual pairs in Polish. Studies in Slavic and General Linguistics 1, 299-314. 1984 Over de Nederlandse imperativus. Forum der Letteren 25, 241-258. Pronk, Tijmen C. 2009 The Slovene dialect of Egg and Potschach in the Gailtal, Austria (Diss. Leiden). Puhvel, Jaan 1987 All our ‘yesterdays’. Festschrift for Henry Hoenigswald (Tübingen: Gunter Narr), 315-318. 480 References Pulleyblank, Edwin G. 1984 Middle Chinese: A study in historical phonology (Vancouver: The University of British Columbia Press) Rasmussen, Jens E. 1974 Haeretica Indogermanica: A selection of Indo-European and preIndo-European studies [Historisk-filosofiske Meddelelser 47/3] (København: Munksgaard). 1985 Miscellaneous morphological problems in Indo-European languages. Lingua Posnaniensis 28, 27-62. 1999 Selected papers on Indo-European linguistics (Copenhagen: Museum Tusculanum). Rauch, Irmengard 1967 The Old High German diphthongization (The Hague: Mouton). Raun, Alo 1988 Proto-Uralic comparative historical morphosyntax. The Uralic languages: Description, history and foreign influences (Leiden: Brill), 555-571. Rédei, Károly 1986 Zu den indogermanisch-uralischen Sprachkontakten [=Sitzungsberichte der Österreichischen Akademie der Wissenschaften, Philosophisch-historische Klasse, 468] (Wien: Akademie der Wissenschaften). Reid, Timothy G. 1990 The lack of i-mutation in short-stemmed, syncopated forms in Old Icelandic. North-Western European Language Evolution 15, 23-48. Renou, Louis 1925 Le type védique tudáti. Mélanges linguistiques offerts à M. J. Vendryes par ses amis et ses élèves (Paris: Champion), 309-316. 1932 À propos du subjonctif védique. Bulletin de la Société de Linguistique de Paris 33, 5-30. Rieken, Elisabeth 1999 Untersuchungen zur nominalen Stammbildung des Hethitischen (Wiesbaden: Harrassowitz). Ringe, Donald A. 1984 Germanic “ē2” and *r. Die Sprache 30, 138-155. References 481 Ringgaard, Kristian 1960 Vestjysk stød (Aarhus: Universitetsforlaget). Risch, Ernst 1962 Das indogermanische Wort für ‘hundert’. Indogermanische Forschungen 67, 129-141. 1975 Remarques sur l’accent du grec ancien, Mélanges linguistiques offerts à Émile Benveniste (Louvain: Peeters), 471-479. 1982 Ein Problem des griechischen Verbalparadigmas: Die verschiedenen Formen der 3. Person Plural. Serta Indogermanica [Fs. Neumann] (Innsbruck: Institut für Sprachwissenschaft), 321-334. Rix, Helmut 1976 Historische Grammatik des Griechischen: Laut- und Formenlehre (Darmstadt: Wissenschaftliche Buchgesellschaft). Robbeets, Martine I. 2003 Is Japanese related to the Altaic languages? (Diss. Leiden). 2005 Is Japanese related to Korean, Tungusic, Mongolic and Turkic? (Wiesbaden: Harrassowitz). 2007 How the actional suffix chain connects Japanese to Altaic. Turkic languages 11, 3-58. Roberge, Paul T. 1983 Those Gothic spirants again. Indogermanische Forschungen 88, 109-155. Rousseau, André 1990 L’alternance *k-/Ø- à l’initiale des mots en indo-européen: Essai d’interprétation sémantique. La reconstruction des laryngales (Paris: Les Belles Lettres), 149-180. Ruijgh, Cornelis J. 1967 Études sur la grammaire et le vocabulaire du grec mycénien (Amsterdam: Hakkert). 1971 Review of Beekes 1969. Lingua 26, 181-198. 1998 Review of Harðarson 1993. Mnemosyne 51, 216-227. Ruipérez, Martín S. 1952 Desinencias medias primarias indoeuropeas sg. 1a *-(m)ai, 2a *-soi, 3a *-(t)oi, pl. 3a *-ntoi. Emerita 20, 8-31. 482 References Rumsey, Alan 1987a Was Proto-Indo-European an ergative language? Journal of IndoEuropean Studies 15, 19-37. 1987b The chimera of Proto-Indo-European ergativity. Lingua 71, 297-318. Rusakov, A.Ju. 1987 K voprosu o rekonstrukcii drevnealbanskogo konsonantizma. Acta Baltico-Slavica 17, 171-184. Sammallahti, Pekka 1988 Historical phonology of the Uralic languages. The Uralic languages: Description, history and foreign influences (Leiden: Brill), 478-554. 1998 The Saami languages: An introduction (Kárášjohka: Davvi Girji). Sapir, Edward 1949 Selected writings in language, culture and personality (ed. by David G. Mandelbaum, Berkeley-Los Angeles: University of California Press). Schaefer, Christiane 1994 Das Intensivum im Vedischen (Göttingen: Vandenhoeck & Ruprecht). Schindler, Jochem 1966 Idg. *duʢ- im Tocharischen. Indogermanische Forschungen 71, 236-238. 1967 Tocharische Miszellen. Indogermanische Forschungen 72, 239-249. Schleicher, August 1868 Eine fabel in indogermanischer ursprache. Beiträge zur vergleichenden Sprachforschung 5, 206-208. Schmid, Wolfgang P. 1963 Studien zum baltischen und indogermanischen Verbum (Wiesbaden: Harrassowitz). Schmidt, Gernot 1978 Stammbildung und Flexion der indogermanischen Personalpronomina (Wiesbaden: Harrassowitz). Schmidt, Karl H. 1979 Reconstructing active and ergative stages of Pre-Indo-European. Ergativity (London: Academic Press), 333-345. References 483 Schmidt, Klaus T. 1980 Zu Stand und Aufgaben der etymologischen Forschung auf dem Gebiete des Tocharischen. Lautgeschichte und Etymologie (Wiesbaden: Reichert), 394-411. Schmitt, Rüdiger 1974 Proto-Indo-European culture and archaeology: Some critical remarks. Journal of Indo-European Studies 2, 279-287. 1981 Grammatik des Klassisch-Armenischen mit sprachvergleichenden Erläuterungen (Innsbruck: Institut für Sprachwissenschaft). Schrader, Otto 1883 Sprachvergleichung und Urgeschichte (Jena: Costenoble). 1890 Sprachvergleichung und Urgeschichte (2. Auflage, Jena: Costenoble). Schrijver, Peter 1991 The reflexes of the Proto-Indo-European laryngeals in Latin (Amsterdam: Rodopi). 1995 Studies in British Celtic historical phonology (Amsterdam: Rodopi). 2003 Early developments of the vowel systems of North-West Germanic and Saami. Languages in prehistoric Europe (Heidelberg: Carl Winter), 195-226. Schulte, Michael 1998 Grundfragen der Umlautphonemisierung (Berlin: Walter de Gruyter). 2006a Zum Nachweis von Allegro- und Presto-Sprachregeln in den Übergangsinschriften. North-Western European Language Evolution 48, 9-44. 2006b Ein kritischer Kommentar zum Erkenntnisstand der Blekinger Inschriften. Zeitschrift für deutsches Altertum und deutsche Literatur 135, 399-412. Seebold, Elmar 1970 Vergleichendes und etymologisches Wörterbuch der germanischen starken Verben (The Hague: Mouton). Seefloth, Uwe 2000 Die Entstehung polypersonaler Paradigmen im UraloSiberischen. Zentralasiatische Studien 30, 163-191. 484 References Seip, Didrik Arup 1919 Review of Kock 1911-16. Maal og Minne, 85-90. Seržants, Iļja 2004 Einige Bemerkungen zur Geschichte des Illativs. Baltu Filoloģija 13/1, 113-120. Shevelov, George Y. 1964 A prehistory of Slavic: The historical phonology of Common Slavic (Heidelberg: Carl Winter). Sievers, Eduard 1878 Zur accent- und lautlehre der germanischen sprachen. Beitraege zur Geschichte der deutschen Sprache und Literatur 5, 63-163. Skautrup, Peter 1928 Klusiler og ‘yngre’ stød i vestjysk. Acta Philologica Scandinavica 3, 32-51. Stang, Christian S. 1942 Das slavische und baltische Verbum (Oslo: Universitetsforlaget). 1952 Eine Bemerkung zur slowakischen Präsensflexion. Norsk Tidsskrift for Sprogvidenskap 16, 271-275. 1957 Slavonic accentuation (Oslo: Universitetsforlaget). 1966 Vergleichende Grammatik der baltischen Sprachen (Oslo: Universitetsforlaget). Starostin, Sergej A. 2007 Indoevropejsko-severnokavkazskie izoglossy. Trudy po jazykoznaniju (Moskva: Jazyki slavjanskix kul’tur), 312-358. Starostin, Sergei, Anna Dybo & Oleg Mudrak 2003 Etymological dictionary of the Altaic languages I (Leiden: Brill). Steblin-Kamenskij, Mixail 1957 Scandinavian breaking from a phonemic point of view. Studia Linguistica 11, 84-91. 1959 Concerning the three periods in the Scandinavian i-umlaut. Arkiv för Nordisk Filologi 74, 105-11. Steensland, Lars 1973 Die Distribution der urindogermanischen sogenannten Gutturale (Diss. Uppsala: Universitetsforlaget). References 485 Stender-Petersen, Adolf 1927 Slavisch-germanische Lehnwortkunde. Göteborg: Wettergren & Kerber. Stiles, Patrick V. 1995 Remarks on the “Anglo-Frisian” thesis. Friesische Studien II (Odense University Press), 177-220. 2004 Place-adverbs and the development of Proto-Germanic long *ē1 in early West Germanic. Etymologie, Entlehnungen und Entwicklungen [Fs. Koivulehto] (Helsinki: Société Néophilologique), 385-396. Streitberg, Wilhelm 1896 Urgermanische Grammatik: Einführung in das vergleichende Studium der altgermanischen Dialekte (Heidelberg: Carl Winter). 1974 Urgermanische Grammatik: Einführung in das vergleichende Studium der altgermanischen Dialekte (4th edition, Heidelberg: Carl Winter). Stumpf, Peter 1990 Die Erscheinungsformen des Westtocharischen: Ihre Beziehungen zueinander und ihre Funktionen [=Tocharian and Indo-European Studies, suppl. 2] (Reykjavík: Málvísindastofnun Háskóla Íslands). Swadesh, Morris 1971 The origin and diversification of language (Chicago: AldineAtherton). Syrett, Martin 1994 The unaccented vowels of Proto-Norse [North-Western European Language Evolution Supplement 11] (Odense: University Press). Szemerényi, Oswald 1960 Studies in the IE System of Numerals (Heidelberg: Carl Winter). Šaxmatov, Aleksej A. 1915 Očerk drevnějšago perioda istorii russkago jazyka (Petrograd: Tipografija Imperatorskoj Akademii Nauk). Tailleur, Olivier Guy 1960 La place du ghiliak parmi les langues paléosibériennes. Lingua 9, 113-147. 486 References Tauli, Valter 1966 Structural tendencies in Uralic languages (The Hague: Mouton). Ter-Avanesova, Aleksandra V. 1989 Ob odnoj slavjanskoj akcentnoj innovacii. Slavjanskoe i balkanskoe jazykoznanie: Prosodija (Moskva: Nauka), 216-250. Thieme, Paul 1929 Das Plusquamperfektum im Veda (Göttingen: Vandenhoeck & Ruprecht). Thomas F. 1957 Autour de l’optatif grec dit “éolien”. Revue des Études Anciennes 59, 250-274. Thurneysen, Rudolf 1883 Urspr. dn tn cn im lateinischen. Zeitschrift für vergleichende Sprachforschung 26: 301-314. 1908 Lecture, reported in Indogermanische Forschungen: Anzeiger 22, 65. 1946 A Grammar of Old Irish (Dublin: Institute for Advanced Studies). Topolińska, Zuzanna 1961 Z historii akcentu polskiego od wieku XVI do dziś (Wrocław: Wydawnictwo Polskiej Akademii Nauk). Tops, Guy A.J. 1974 The origin of the Germanic dental preterit (Leiden: Brill). Trautmann, Reinhold 1923 Ein Kapitel aus der Lautlehre der baltisch-slavischen Sprachen. Slavia 2, 1-4. Tschenkéli, Kita 1958 Einführung in die georgische Sprache I (Zürich: Amirani). Turner, Ralph L. 1924 The Sindhi recursives or voiced stops preceded by glottal closure. Bulletin of the School of Oriental Studies 3/2, 301-315. 1975 The Sindhi recursives or voiced stops preceded by glottal closure. Collected Papers 1912-1973 (London: Oxford UP), 192-205. Uhlenbeck, Christiaan Cornelis 1895 Waar werd de Indogermaansche stamtaal gesproken? Tijdschrift voor Nederlandsche Taal- en Letterkunde 14, 69-74. References 1897 1901 1904 1905a 1905b 1906 1907 1913 1923 1933 1934-35a 1934-35b 1937a 1937b 487 De voorgeschiedenis der Indogermaansche volken. Taal en Letteren 7, 1-25. Agens und Patiens im Kasussystem der indogermanischen Sprachen. Indogermanische Forschungen 12, 170-171. Eine baskische Parallele. Indogermanische Forschungen 17, 436-441. Uralische Anklänge in den Eskimosprachen. Zeitschrift der Deutschen Morgenländischen Gesellschaft 59, 757-765. Baskisch und Uralaltaisch. Beiträge zur Kunde der indogermanischen Sprachen 29, 305-307. Zur Eskimogrammatik. Zeitschrift der Deutschen Morgenländischen Gesellschaft 60, 112-114. Zur Eskimogrammatik. Zeitschrift der Deutschen Morgenländischen Gesellschaft 61, 435-438. Baskisch und Indogermanisch. Indogermanische Forschungen 33, 171-172. Over een mogelijke verwantschap van het Baskisch met de Palaeo-Kaukasische talen. Mededeelingen der Koninklijke Akademie van Wetenschappen, Afdeeling Letterkunde, A 55/5, 105-137. Eine Bemerkung zur Frage nach der Urverwandtschaft der uralischen und indogermanischen Sprachen. Liber Semisaecularis Societatis Fenno-Ugricae (Helsinki: Suomalais-Ugrilainen Seura), 396-397. Oer-Indogermaansch en Oer-Indogermanen. Mededeelingen der Koninklijke Akademie van Wetenschappen, Afdeeling Letterkunde 77-79, serie A, 125-148. Eskimo en Oer-Indogermaansch. Mededeelingen der Koninklijke Akademie van Wetenschappen, Afdeeling Letterkunde 77-79, serie A, 179-196. Über den Wert eskimoisch-indogermanischer Wortähnlichkeiten. Mélanges linguistiques offerts à M. Holger Pedersen (Aarhus: Universitetsforlaget), 110-114. The Indogermanic mother language and mother tribes complex. American Anthropologist 39, 385-393. 488 1941 References Oude Aziatische contacten van het Eskimo. Mededeelingen der Nederlandsche Akademie van Wetenschappen, Afdeeling Letterkunde, NR 4/7, 201-227. 1942 Ur- und altindogermanische Anklänge im Wortschatz des Eskimo. Anthropos 37-40, 133-148. 1946 Gestaafde en vermeende affiniteiten van het Baskisch. Mededeelingen der Koninklijke Nederlandsche Akademie van Wetenschappen, Afdeeling Letterkunde, NR 9/2, 13-24. Ul’janov, Grigorij 1903 Kratnoe značenie udvoennyx osnov. Russkij filologičeskij věstnik 49, 235-249. Vaan, Michiel de 1999 Towards an explanation of the Franconian tone accents. Amsterdamer Beiträge zur älteren Germanistik 51, 23-44. 2004 ‘Narten’ roots from the Avestan point of view. Per aspera ad asteriscos [Fs. Rasmussen] (Innsbruck: Institut für Sprachwissenschaft), 591-599. 2008 Etymological dictionary of Latin and the other Italic languages (Leiden: Brill). Vaillant, André 1936 L’ergatif indo-européen. Bulletin de la Société de Linguistique de Paris 37, 93-108. 1937 L’origine des présents thématiques en -e/o-. Bulletin de la Société de Linguistique de Paris 38, 89-101. 1950 Grammaire comparée des langues slaves I: Phonétique (Lyon: IAC). 1958 Grammaire comparée des langues slaves II: Morphologie (Lyon: IAC). 1966 Grammaire comparée des langues slaves III: Le verbe (Paris: Klincksieck). 1974 Grammaire comparée des langues slaves IV: La formation des noms (Paris: Klincksieck). Van Haeringen, Coenraad B. 1920 Zur friesischen Lautgeschichte. Beiträge zur Geschichte der deutschen Sprache und Literatur 44, 27-53. References 489 Van Helten, Willem L. 1891 Grammatisches. Beiträge zur Geschichte der deutschen Sprache und Literatur 15, 455-488. Van Wijk, Nicolaas 1902 Der nominale Genetiv Singular im Indogermanischen in seinem Verhältnis zum Nominativ (Zwolle: Tijl). 1911 Een Oudwestnederfrankies ǃ-dialekt. Tijdschrift voor Nederlandsche taal- en letterkunde 30, 161-189. 1918 Altpreussische Studien (Haag). Vanags, Pēteris 1994 Die Entwicklungstendenzen der Kasusendungen in den ältesten lettischen Sprachdenkmälern. Linguistica Baltica 3, 121-130. Vennemann, Theo 1984 Hochgermanisch und Niedergermanisch: Die Verzweigungstheorie der germanisch-deutschen Lautverschiebungen. Beiträge zur Geschichte der deutschen Sprache und Literatur (Tübingen) 106, 1-45. 1988 Die innergermanische Lautverschiebung und die Entstehung der germanischen und deutschen Dialekte. Languages and cultures: Studies in honor of Edgar C. Polomé (Berlin: Mouton de Gruyter), 749-761. 1991 The relative chronology of the High Germanic consonant shift and the West Germanic anaptyxis. Diachronica 8, 45-57. 1994a Linguistic reconstruction in the context of European prehistory. Transactions of the Philological Society 92, 215-284. 1994b Dating the division between High and Low Germanic: A summary of arguments. Language change and language structure (Berlin: Mouton de Gruyter), 271-303. 2006 Grimm’s Law and loanwords. Transactions of the Philological Society 104/2, 129-166. Vermeer, Willem 1985 Non-initial falling tones in neo-štokavian dialects. Zbornik za Filologiju i Lingvistiku 27-28 [Fs. Ivić], 143-149. 1991 The mysterious North Russian nominative singular ending -e and the problem of the reflex of Proto-Indo-European *-os in Slavic. Die Welt der Slaven 36, 271-295. 490 1992 References In the beginning was the lengthened grade: On the continuity of Proto-Indo-European vowel quantity in Slavic. Rekonstruktion und relative Chronologie: Akten der VIII. Fachtagung der Indogermanischen Gesellschaft, Leiden, 1987 (Innsbruck: Institut für Sprachwissenschaft), 115-136. Villanueva Svensson, Miguel 2008 Lithuanian žinóti ‘to know’. Baltistica 43/2, 175-199. 2009 Indo-European *skʄ in Balto-Slavic. Baltistica 44/1, 5-24. Vogt, Hans 1938 Arménien et caucasique du sud. Norsk Tidsskrift for Sprogvidenskap 9: 321-338. 1958 Les occlusives de l’arménien. Norsk Tidsskrift for Sprogvidenskap 18: 143-161. 1971 Grammaire de la langue géorgienne (Oslo: Universitetsforlaget). Vovin, Alexander 2005 The end of the Altaic controversy. Central Asiatic Journal 49/1, 71-132. Voyles, Joseph B. 1991 A history of OHG i-umlaut. Beiträge zur Geschichte der deutschen Sprache und Literatur (Tübingen) 113, 159-194. Wackernagel, Jakob 1896 Altindische Grammatik I: Lautlehre (Göttingen: Vandenhoeck & Ruprecht). Watkins, Calvert 1969 Indogermanische Grammatik III: Formenlehre I: Geschichte der indogermanischen Verbalflexion (Heidelberg: Carl Winter). Weitenberg, Joseph J.S. 1987 Proto-Indo-European nominal classification and Old Hittite. Münchener Studien zur Sprachwissenschaft 48, 213-230. Winter, Werner 1962 Die Vertretung indogermanischer Dentale im Tocharischen. Indogermanische Forschungen 67, 16-35. 1965 Tocharian evidence. Evidence for laryngeals (The Hague: Mouton), 190-211. 1978 The distribution of short and long vowels in stems of the type Lith. °sti : vèsti : mèsti and OCS jasti : vesti : mesti in Baltic and References 491 Slavic languages. Recent developments in historical phonology (The Hague: Mouton), 431-446. 1980a Morphological signalling of selection properties: Transiteveness in Tocharian B and A verbs. Historical morphology (The Hague), 421-442. 1980b Zum Beitrag der tocharischen Sprachen zu Problemen der lautlichen Rekonstruktion des Indogermanischen. Lautgeschichte und Etymologie: Akten der Fachtagung der Indogermanischen Gesellschaft (Wiesbaden: Reichert), 542-563. 1994 Zum tocharischen Verb. Tocharisch: Akten der Fachtagung der Indogermanischen Gesellschaft [=Tocharian and Indo-European Studies, suppl. 4] (Reykjavík: Málvísindastofnun Háskóla Íslands), 284-309. Wolfram, Herwig 1979 Geschichte der Goten (München: Beck). Wollmann, Alfred 1990 Untersuchungen zu den frühen lateinischen Lehnwörtern im Altenglischen. (München: Fink). Xelimskij, Evgenij A. 2000 Proto-Uralic gradation: Continuation and traces. Komparativistika, uralistika: Lekcii i stat’i (Moskva: Jazyki russkoj kul’tury), 167-190. Zaliznjak, Andrej A. 1995 Drevnenovgorodskij dialekt (Moskva: Jazyki russkoj kul’tury). Zinkevičius, Zigmas 1966 Lietuvių dialektologija: Lyginamoji tarmių fonetika ir morfologija (Vilnius: Mintis). 1981 Lietuvių kalbos istorinė gramatika II (Vilnius: Mokslas). 2008 Dėl postpozicijos *-én akūto. Baltistica 43/3, 380.