Interlingual and Intralingual Interference during Gender Production
Interlingual and Intralingual Interference during Gender Production
Interlingual and Intralingual Interference during Gender Production
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong> <strong>during</strong> <strong>Gender</strong> <strong>Production</strong><br />
in Czech <strong>and</strong> German<br />
Denisa Bordag<br />
University ofLeipzig & Charles University in Prague<br />
Abstract: The purpose of this article is to explore the psycholinguistic aspects of<br />
interlingual <strong>and</strong> intralingual interference <strong>during</strong> second language grammatical<br />
gender production. Two experiments were constructed to explore whether the<br />
difficulties in acquisition <strong>and</strong> correct production of grammatical gender in a second<br />
language are influenced by competition for selection between the gender node of a<br />
second language noun <strong>and</strong> the gender node of its first language translation<br />
equivalent. The languages tested were Czech <strong>and</strong> German, which have the same<br />
number <strong>and</strong> type of grammatical gender. The results of the experiments confirmed<br />
the initial hypothesis concerning the gender competition. Furthermore, the analysis<br />
of the gender errors subjects made in both experiments yielded evidence that<br />
phonological form, especially word termination, also has an effect on gender<br />
production in the second language. The intralingual interference manifested itself<br />
especially in gender overgeneralization or misselection in cases where second<br />
language nouns had an atypical or ambiguous termination.<br />
1. Introduction<br />
<strong>Interlingual</strong> <strong>and</strong> intralingual interference have been in the center of second<br />
language research for a very long time. The period when the influence of<br />
interlingual interference on SLA was overestimated (Weinerich 1953, Lado<br />
1957) was followed by a period which was dominated by the Creative<br />
Construction Hypothesis of Dulay & Burt (1974 <strong>and</strong> later), who claimed that<br />
most errors are developmental <strong>and</strong> that interlingual interference has a minimal<br />
effect on SLA. In the late seventies <strong>and</strong> eighties the situation developed towards<br />
a more balanced approach to the interfrence phenomena, admitting that both<br />
interlingual <strong>and</strong> intralingual interference play an important role in SLA, though<br />
the strength of their influence may differ depending on various factors, e.g.<br />
language proficiency. <strong>Interference</strong>, or, as often preferred, transfer is also no<br />
longer seen as a mechanical transference of first language features or structures<br />
Cahiers linguistiques d'Ottawa, June/juin 2004, Vol. 32: 1-23<br />
ISSN 0315-3167. ©2004, Department of Linguistics, University of Ottawa
Denisa Bordag<br />
into the second language, but rather is generally viewed as a complex cognitive<br />
mechanism underlying SLA.<br />
The emphasis on the cognitive character of interference/transfer, has not,<br />
however, resulted in extensive psycholinguistic research of these phenomena so<br />
far. One of the possible reasons is that for a long time psycholinguistics has been<br />
concerned mainly with the processing <strong>and</strong> acquisition of the first language.<br />
In this paper I want to investigate the influence of interlingual <strong>and</strong><br />
intralingual interference on second language acquisition <strong>and</strong> production. The<br />
subject of my research is grammatical gender. Though a very challenging topic<br />
(as numerous studies exploring grammatical gender production <strong>and</strong><br />
representation in first language show, see below), it has hardly been addressed by<br />
psycholinguists so far. A probable explanation might be the fact that one of the<br />
two languages used in second language psycholinguistic research is usually<br />
English, a language without grammatical gender <strong>and</strong> thus intrinsically unsuitable<br />
for research on the role of interlingual interference in second language gender<br />
acquisition <strong>and</strong> production.<br />
Although there are no studies directly concerned with the topic of this<br />
article, a lot of psycholinguistic research has been done to explore how a second<br />
language is stored <strong>and</strong> processed <strong>and</strong> how it interacts with the first language. In<br />
the first part of this article the relevant findings of this research will be reviewed,<br />
together with findings concerning grammatical gender in the first language. In<br />
the second part, a hypothesis about interlingual interference in second language<br />
gender production will be formulated <strong>and</strong> tested in two experiments. Data from<br />
these experiments will also be used to explore intralingual interference. At the<br />
end of each section, the results of the experiments <strong>and</strong> analyses will be discussed<br />
<strong>and</strong>, where appropriate, compared with results of similar experiments in first<br />
languages.<br />
1.1 Lexical processing by bilinguals<br />
In the human mind, information about words is stored in a mental lexicon.<br />
Lexical units are retrieved on the basis of their meaning. First, a preverbal<br />
message is formulated, which involves the activation <strong>and</strong> retrieval of relevant<br />
2
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
concepts. From these the activation spreads to so-called lemmas, lexical units<br />
which are already specified semantically <strong>and</strong> syntactically, but not yet<br />
phonologically. Phonological encoding takes place at the next level of speech<br />
production, where the units are called phonological forms. In the final stage,<br />
articulation is initiated <strong>and</strong> executed (Levelt et al. 1999).<br />
The first attempt to answer the question of whether the words of two<br />
languages are stored in one or two lexicons was made already in 1963 by Kolers<br />
(1963). Today it is generally assumed that whereas the units at the conceptual<br />
level are shared across languages, lemmas <strong>and</strong> phonological forms (with the<br />
exception of cognates) are language-specific, but stored together in one lexicon.<br />
The preferred hypothesis in psycholinguistic research (De Bot 1992,<br />
Poulisse 1997) about lexical storage in the bilingual mind is the Subset<br />
Hypothesis, first formulated by Paradis (1987). According to this hypothesis,<br />
there is only one storage system. The links between its units are strengthened by<br />
repeated usage. Consequently, the units of one language are connected with each<br />
other more strongly than they are with the units from another language, which<br />
results in emergence of subsets. Each subset contains units from one <strong>and</strong> the<br />
same language <strong>and</strong> can be accessed separately.<br />
Most theories of bilingual <strong>and</strong> second language lexical access further<br />
assume that speakers do not initially select the language-appropriate system<br />
(subset) before making a lexical search, but that lexical access is language<br />
independent <strong>and</strong> both systems are activated <strong>and</strong> searched simultaneously (De Bot<br />
1992, Green 1993, Poulisse & Bongaerts 1994, Hermans et al. 1998, Grosjean,<br />
1998). Hence the lexical processes operating on a bilingual lexicon are<br />
principally the same as those operating on a monolingual one. The relationship<br />
between translation equivalents could be compared to that of near-synonyms in<br />
one language: The not-to-be-selected first language lemma is activated <strong>during</strong> the<br />
lemma selection process of its foreign language equivalent, just as semantically<br />
related first language lemmas are. Hermans <strong>and</strong> his colleagues (Hermans et al.<br />
1998) found evidence that the competition between the to-be-selected foreign<br />
language lemma <strong>and</strong> its first language translation equivalent is as strong as the<br />
3
Denisa Bordag<br />
competition between near-synonyms, because both lemmas turned out to be<br />
phonologically encoded <strong>during</strong> picture naming (with distractors) in a foreign<br />
language (similarly to the way near synonyms are: see Jescheniak & Schriefers<br />
1998, Peterson & Savoy 1998).<br />
The subjective frequencies of foreign language lemmas are, however,<br />
generally low, <strong>and</strong> the links between them weak in comparison with the<br />
frequencies of first language lemmas <strong>and</strong> the strength of connections between<br />
them. It is therefore difficult to overcome their activation <strong>during</strong> second language<br />
production (Hermans et al. 1998). The competition with interfering first language<br />
lemmas then could be one of the reasons why the production of a word in a<br />
foreign language requires considerably more time than in a first language (e.g.<br />
Kroll & Curley 1988, Potter et al.1984) <strong>and</strong> why unintended code switching<br />
occurs, as well as cross-linguistic interference in general.<br />
Theories concerned with how second language speakers (<strong>and</strong> bilinguals)<br />
control their language production (Green 1998, Grosjean 1997, 1998) assume<br />
that language systems can be at different levels of activation <strong>and</strong> that in order to<br />
speak one language rather than another its activation level must exceed that of<br />
the other language (Paradis 1981, Grosjean 1988 <strong>and</strong> later). A further notion is<br />
that regulation is achieved through the modification of levels of activation of<br />
language networks, or items within these networks, rather than via a simple<br />
switch mechanism (De Bot & Schreuder 1993, Grainger & Dijkstra 1992,<br />
Grosjean 1988, Paradis 1981).<br />
Grosjean (1997, 1998) proposes that the activation <strong>and</strong> deactivation of<br />
language systems allows bilinguals to achieve different language modes. In the<br />
monolingual mode, one language is the base language <strong>and</strong> the other is<br />
deactivated at least partially. In contrast, in the bilingual mode (the other extreme<br />
of the bilingual continuum), when individuals are speaking with somebody with<br />
whom they can code-switch or mix languages, both languages are relatively<br />
active, but the activation of the base language is greater. As for the second<br />
language speakers, it is assumed that they cannot prevent their first language<br />
from being at least partly activated <strong>and</strong> thus interfering with the production of<br />
4
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
their speech in the foreign language (Green 1986, 1993, De Bot & Schreuder<br />
1993).<br />
2. <strong>Interlingual</strong> <strong>Interference</strong><br />
On the basis of this information it is possible to hypothesize that one of the<br />
reasons why acquisition <strong>and</strong> production of grammatical gender causes so many<br />
difficulties for second language speakers at all levels of proficiency is the<br />
interference from their first language. During second language production, both<br />
language systems are activated <strong>and</strong> searched in parallel: If a second language<br />
speaker wants to produce a nominal phrase in his or her second language, the<br />
lemmas of the corresponding first language translation equivalents <strong>and</strong> their<br />
grammatical features are also activated <strong>and</strong> may thus interfere with second<br />
language production.<br />
The prominent models of first language production (Levelt 1989, Levelt et<br />
al. 1999, Dell 1986) assume that the information concerning grammatical gender<br />
is stored in the form of generic gender nodes at the lemma level. Lemmas of all<br />
nouns of the same gender are thus connected to one gender node common to<br />
them all. If gender information is necessary for further encoding, all activated<br />
gender nodes compete for selection <strong>and</strong> the most highly activated one (usually<br />
the one of the target lemma) is finally selected. If we exp<strong>and</strong> this hypothesis to<br />
the representation of second language grammatical gender, we might assume that<br />
if two languages have the same number <strong>and</strong> type of grammatical gender, their<br />
lemmas are all mapped onto one set of gender nodes. This assumption is<br />
supported by theories stressing the fact that the incomplete second language<br />
system takes over, or inherits, all it possibly can from the first language<br />
(Competition Model by Mac Whinney 1997), i.e. possibly including the generic<br />
gender nodes as well.<br />
The hypothesis to be tested, concerning the interlingual interference <strong>during</strong><br />
the second language production of grammatical gender, thus reads as follows: If<br />
the gender of a first language noun differs from that of its to-be-selected second<br />
language translation equivalent, their lemmas compete for selection <strong>and</strong> with<br />
5
Denisa Bordag<br />
them their corresponding gender nodes, which they automatically activate.<br />
Selection of the second language lemma should thus be delayed <strong>and</strong> subjects<br />
should produce more gender errors compared to the condition where both<br />
translation equivalents have congruent genders. We will use the term gender<br />
interference effect to refer to the negative effect of the first language gender on<br />
the selection of the gender of its second language translation equivalent.<br />
Conceptualization<br />
<br />
Grammatical<br />
encoding<br />
(lemmas)<br />
I<br />
I<br />
I<br />
I<br />
, I Phonological<br />
<br />
encoding<br />
(phonological forms)<br />
[hrad]<br />
Articulation<br />
Figure I: <strong>Gender</strong> interference effect: The gender node of the target second language word "hrad"<br />
(castle, masculine in Czech) competes for selection with the gender node of its automatically<br />
activated first language translation equivalent "burg" (feminine in German).<br />
2.1 Experiments<br />
Two experiments were constructed to explore whether such a gender interference<br />
effect indeed exists. Two languages, German <strong>and</strong> Czech, were chosen to be<br />
employed in these experiments. Both these languages have grammatical gender<br />
<strong>and</strong> the number <strong>and</strong> type of their genders is identical: masculine, feminine <strong>and</strong><br />
6
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
neuter. Despite these similarities, the languages differ in how the three genders<br />
are assigned to their nouns: Approximately half of their translation equivalents<br />
have identical genders while the other half does not. There is also no 1:1<br />
mapping between the genders of Czech <strong>and</strong> German nouns: It is impossible to<br />
say that e.g. most German feminines correspond to Czech neuters or the other<br />
way around. German learners of Czech (<strong>and</strong> vice versa) thus have to learn the<br />
appropriate gender with every single noun they acquire. Sometimes the gender<br />
(especially of Czech nouns) can be guessed from their terminations, but this<br />
method is not very reliable as there is no ending in the Czech or German<br />
language (apart from some derivational suffixes), which would unambiguously<br />
mark e.g. only feminine nouns <strong>and</strong> not any other.<br />
2.1.1 Experimental design<br />
In both experiments, subjects had to name pictures in their second language in<br />
two conditions: either with a bare noun (short condition) or with an adjective of<br />
size (big or small) <strong>and</strong> a noun (long condition). Noun phrases with adjective +<br />
noun <strong>and</strong> not solely article + noun were chosen, because Czech as contrasted to<br />
German does not have any articles. On the other h<strong>and</strong>, adjectives in both<br />
languages have to agree in gender with their head nouns. Half of the nouns<br />
chosen for the experiment had congruent gender in both languages, the other half<br />
did not. The assumption was that production of noun phrases (<strong>and</strong> nouns) with<br />
congruent gender in both languages would be faster than production of those<br />
with different gender in German <strong>and</strong> Czech. To create optimal conditions in<br />
which the first language is most likely (<strong>and</strong> most strongly) activated, it was<br />
decided to keep subjects as close to the bilingual end of the bilingual continuum<br />
(Grosjean 1997) as possible. First language fillers were therefore included into<br />
the experiment, so that one third (22) of the pictures was to be named in the<br />
subjects' first <strong>and</strong> two thirds (44) in their second language.<br />
2.1.2 Participants<br />
Eighteen subjects participated in each experiment. The German subjects tested in<br />
the Czech experiment (Czech was the second language) were mostly students of<br />
West Slavic studies, started to learn Czech after their 16th year (at 20 years on<br />
7
Denisa Bordag<br />
average) <strong>and</strong> had been learning Czech for 4.6 years on average. The Czech<br />
participants tested in the German experiment (German was the second language)<br />
started to learn German after their 8th year (at 14 on average) <strong>and</strong> had been<br />
learning German for 8.4 years on average. Many of them experienced long<br />
periods when they were not using German at all, so although they had been<br />
learning German almost twice as long as the German subjects, their proficiency<br />
in the foreign language was on average not exceedingly higher than that of the<br />
German subjects. The proficiency of the subjects was judged on the basis of an<br />
informal interview in subjects" second language at the beginning of the<br />
experiment, the information they provided in the questionnaire <strong>and</strong> their rating of<br />
their familiarity with the experimental items. Most German subjects have already<br />
passed or were just about to take the German equivalent of the English B.A.<br />
exams in Czech Studies (Zwischenpriifung). All Czech participant had sufficient<br />
knowledge of German to study at this language at the University of Leipzig.<br />
Highly advanced subjects or subjects below the level of intermediate were not<br />
tested or not included in the analysis. All subjects were paid for their<br />
participation.<br />
2.1.3 Materials<br />
The target stimuli were 72 line drawings in the Czech <strong>and</strong> 70 in the German<br />
experiment. Pictures of forty-four nouns were chosen as experimental items,<br />
none of them were cognates. The experimental items were partly different for<br />
each experiment to match all conditions (see below). The gender of half of the<br />
nouns was congruent with that of their translation equivalents (e.g. strom <strong>and</strong><br />
Baum, both rnasculines), the gender of the other half differed from that of their<br />
translation equivalents (e.g. obraz (masculine) <strong>and</strong> BUd (neuter». They were all<br />
easily-depictable concrete nouns with high frequency both in Czech <strong>and</strong> in<br />
German. The majority of the experimental items were monomorphemic <strong>and</strong><br />
inanimate. Each set of 22 nouns consisted of 8 masculines, 8 feminines <strong>and</strong> 6<br />
neuters. Their length ranged between one <strong>and</strong> three syllables <strong>and</strong> each set<br />
contained the same number of mono-, di- <strong>and</strong> trisyllabic masculine, feminine <strong>and</strong><br />
8
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
neuter nouns. The sets were also matched for final letter to control for the<br />
possible influence of phonological form on gender selection.<br />
Next to 44 experimental items, there were also 22 first language filler items<br />
<strong>and</strong> 14 practice items. They were also very frequent in both languages. Some of<br />
the fillers were used as practice items as well.<br />
2.1.4 Procedure<br />
The experimental session had three parts <strong>and</strong> lasted approximately 40 minutes.<br />
Participants were tested individually.<br />
The first part started with a short informal dialogue in the second language<br />
to test whether the subject is a suitable c<strong>and</strong>idate for the experiment. Then,<br />
subjects were presented with instructions emphasizing both the speed <strong>and</strong><br />
accuracy of their responses <strong>and</strong> with a booklet containing all 72 (70) pictures,<br />
which would later appear in the experiment. Under each picture, its first language<br />
name <strong>and</strong> its second language translation equivalent were printed. Subjects were<br />
asked to rate on a 7-point scale 1) how well the picture depicted the first<br />
language word, 2) how well it depicted the second language word <strong>and</strong> 3) how<br />
familiar they were with the second word. The ratings were done to ensure that<br />
participants really study the names of the pictures <strong>and</strong> to obtain data for<br />
evaluations of the stimuli materials <strong>and</strong> of subjects' familiarity with the second<br />
language words. The ratings also allowed the experimenter to control for whether<br />
the words in either the gender congruent or gender-incongruent set were not less<br />
familiar or less well-depicted than the words in the other set. Later analyses<br />
showed that all words were equally well-depicted <strong>and</strong> very well known to all<br />
participants. (Total rating in the Czech experiment was 6.48 in the congruent <strong>and</strong><br />
6.45 in the incongruent group; in the German experiment 6.58 <strong>and</strong> 6.55 in the<br />
congruent <strong>and</strong> incongruent group, respectively).<br />
In the second part of the experimental session the experiment proper was<br />
run. Before the experiment started, participants were presented with a sheet<br />
containing Czech <strong>and</strong> German examples of noun phrases, which were to be used<br />
in the experiment's long condition. Attention was not explicitly drawn to<br />
grammatical gender.<br />
9
Denisa Bordag<br />
The experimental session started with a block of 14 practice items, which<br />
was the same for all subjects in each experiment. Each trial started with a visual<br />
fixation point (= * =) presented on a grey background for 600 ms on the right of<br />
the screen in the middle of an imaginary square in which the target picture was<br />
displayed later. Then, either a yellow or a blue background appeared for 300 ms.<br />
The colour of the background defined the language in which the response was<br />
required. Half of the subjects in each experiment had to respond in Czech when a<br />
picture was displayed on a blue background <strong>and</strong> in German when a yellow<br />
background appeared, the other half of the subjects had the colours reversed.<br />
Following this pause, a target picture was presented in the imaginary square on<br />
the right of the computer screen <strong>and</strong> marked with a black arrow pointing at it<br />
from the upper right part of the screen. In the long condition, two pictures of the<br />
same object appeared simultaneously on the screen: The target picture marked<br />
with an arrow on its usual position on the right part of the screen <strong>and</strong> either a<br />
larger or a smaller control picture of the same object on the left. When the left<br />
control picture was larger than the target picture, the subjects had to name the<br />
target picture with the adjective "small"; if the left control picture was smaller<br />
than the target picture, subjects had to name the target picture with the adjective<br />
"big". Naming latencies were measured from the right or both pictures onset by a<br />
voice key.<br />
The display colour of the arrow was contingent on the participant's<br />
response. It turned white as soon as a vocal response was initiated. However, if<br />
no response was registered within 5000 ms (time out) the arrow turned red.<br />
When the participant was ready for the next trial, he or she pressed the space bar<br />
on the keyboard <strong>and</strong> after a pause of 300 ms the fixation point of the new trial<br />
was displayed.<br />
Subjects were r<strong>and</strong>omly assigned to one of two main r<strong>and</strong>omisation<br />
versions in each experiment. Each item was presented in a long <strong>and</strong> in a short<br />
condition. Half of the subjects in each experiment saw the items in the long<br />
condition first; the other half saw the items in the short condition first. Within<br />
each subject, half of the items were long first, the other half were short first. The<br />
10
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
two main versions of the experiment differed with respect to which items were<br />
long first.<br />
Individual r<strong>and</strong>omisations within the two main versions were prepared for<br />
each subject with the following restrictions: No more than 3 trials of the same<br />
condition (long vs. short), the same size (big vs. small) <strong>and</strong> the same gender<br />
(masculine, feminine, neuter) were allowed to follow each other. The same<br />
language of response was allowed in maximally 4 successive trials.<br />
In the third part of the experimental session, subjects filled in a<br />
questionnaire concerning their proficiency in the second language <strong>and</strong> their<br />
language background.<br />
2.1.5 Results <strong>and</strong> discussion<br />
A total of 352 (14.80/0) observations in the Czech experiment <strong>and</strong> 443<br />
(18.60/0) in the German experiment were marked as incorrect. Observations were<br />
discarded mainly due to voice-key errors, gender errors <strong>and</strong> task errors (picture<br />
named in the wrong language, with the wrong adjective etc.). The cut-off was<br />
two st<strong>and</strong>ard deviations from a subject's mean response time in the long <strong>and</strong><br />
short condition.<br />
Analyses of variance (ANOYAs) were performed on the mean naming<br />
latencies per subject (FI) <strong>and</strong> per item (F2) in each experiment. Overall, naming<br />
latencies for pictures with names from the gender-incongruent set were slower<br />
than naming latencies for pictures with congruent gender in both languages. Both<br />
analysis with subjects as a r<strong>and</strong>om factor <strong>and</strong> with items as a r<strong>and</strong>om factor<br />
showed a significant effect on condition (congruent vs. incongruent) at the<br />
p
Denisa Bordag<br />
Table 1.<br />
Results of the Czech experiment with reaction times in ms<br />
<strong>and</strong> % of errors in each experimental conditon.<br />
Czech exp, Congruent Incongruent Interf, effect<br />
RT %e RT %e<br />
Short 1202 1.4 1276 1.3 +74<br />
Long 1642 3.9 1703 4.1 + 61<br />
Table 2.<br />
Results of the German experiment with reaction times in ms<br />
<strong>and</strong> % of errors in each experimental condition.<br />
German exp, Congruent Incongruent Interf. effect<br />
RT %e RT %e<br />
Short 1136 1.8 1240 2.4 +104<br />
Long 1410 3.9 1489 5.9 +79<br />
As the Tables 1 <strong>and</strong> 2 show, reaction times in the Czech experiment were<br />
numerically slower <strong>and</strong> the interference effect smaller compared to the German<br />
experiment (though no statistic comparison can be made, because different<br />
subjects were tested in both experiments). Rather than difference in proficiency<br />
(subjects in the Czech experiment made less errors than those in the German<br />
experiment), this might suggest different experimental strategies employed by<br />
the Czech <strong>and</strong> German subjects: Whereas German subjects preferred a correct<br />
answer to speed (which might have caused that the interference effect partly<br />
"disappeared" in the long naming latencies), Czech subjects probably mostly<br />
opted for the reversed strategy.<br />
To examine whether the gender interference effect obtained in the RT<br />
analysis also showed up in the error analysis, a Qui-square test for the given<br />
probabilities was performed on the number of gender errors in the gendercongruent<br />
<strong>and</strong> gender-incongruent conditions. The test reached significance in<br />
both experiments (p
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
induces more gender errors in the adjective + noun condition than when the<br />
genders of the translation equivalents are congruent in both languages.<br />
Table 3.<br />
Number of gender errors in both experiments in the long condition.<br />
number of gender errors<br />
Czech experiment<br />
German experiment<br />
Congruent condition 22 48<br />
Incongruent condition 52 90<br />
Three important results were obtained in the experiment. First, the<br />
response latencies were longer if the gender of a second language noun differed<br />
from that of its first language translation equivalent. Second, more gender errors<br />
were produced if genders of translation equi valents were incongruent. These two<br />
results support the initial hypothesis that there is a gender interference effect from<br />
the first language <strong>during</strong> second language production of grammatical gender.<br />
Interestingly, <strong>and</strong> this is the third important result, the gender interference effect<br />
was obtained not only in the long condition, where the information about<br />
grammatical gender was necessary to encode the adjecti ve agreement ending, but<br />
also in the short condition, when only a bare noun was named.<br />
In this aspect the gender interference effect differs from a gender<br />
distraction effect obtained in picture-word distractor experiments in some native<br />
languages, e.g. in German <strong>and</strong> Dutch (German: Schriefers & Truel 2000, van<br />
Berkum 1997, Dutch: Schriefers 1993, La Heij et al. 1998; for overview:<br />
Schriefers & Jescheniak 1999). In these experiments subjects had to name<br />
pictures either when gender congruent or a gender incongruent word distractor<br />
was present. In these experiments, similarly to those described above, slower<br />
reaction times <strong>and</strong> more gender errors were obtained when the target picture <strong>and</strong><br />
word distractor had different genders than when their genders agreed. This effect<br />
was, however, consistently obtained only when the gender was required by the<br />
syntactic environment, i.e. by naming with a gender-marked nominal phrase. The<br />
difference between the two effects may be due to the fact that whereas the picture<br />
13
Denisa Bordag<br />
name <strong>and</strong> word distractor in the first language experiments were usually<br />
semantically unrelated words, the two translation equivalents are semantically<br />
identical (or at least very close) <strong>and</strong> share a common concept. Consequently, it<br />
might be the case that the competition between gender nodes is stronger if the<br />
two relevant lemmas share the same concept <strong>and</strong> that under this condition the<br />
not-to-be-selected gender node interferes <strong>and</strong> competes for selection even when<br />
it is not needed by the syntactic environment. More experiments are, however,<br />
necessary in order to explain the difference between the two effects satisfactorily.<br />
3. <strong>Intralingual</strong> interference<br />
In the next step, the role of intralingual interference <strong>during</strong> second language<br />
production of grammatical gender shall be explored. According to Richards<br />
(1971) intralingual interferece "reflects the general characteristics of rule<br />
learning such as a faulty generalisation, incomplete application of rules <strong>and</strong><br />
failure to learn conditions under which rules apply". A typical manifestation of<br />
an intralingual interference is thus overgeneralization.<br />
As already mentioned above, the gender of German <strong>and</strong> especially Czech<br />
nouns can be partly derived from a word's termination. In Czech, there is a<br />
typical termination for each gender (monomorphemic words): -a for feminines, <br />
o for neutrals <strong>and</strong> a consonant for masculines, in German this is only partly true<br />
(see below). In both languages it is, however, possible to formulate rules on the<br />
basis of which a large number of nouns can be assigned a particular gender<br />
according to their termination. One way in which intralingual interference <strong>during</strong><br />
second language production can be manifested is hence overgeneralization of<br />
these regularities between a noun's grammatical gender <strong>and</strong> its termination. To<br />
find out whether this kind of intralingual interference appears in the data, the<br />
gender errors made in the long conditions in both experiments were analyzed.<br />
4.1 <strong>Gender</strong> errors analysis<br />
A detailed analysis of gender errors made by subjects in both experiments in the<br />
long condition revealed a consistent pattern: Subjects made most gender errors in<br />
agreement with nouns which had either an ambiguous or an atypical termination<br />
14
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
for their gender. In these cases, subjects mostly assigned to these nouns a gender<br />
which they "should have had" according to their termination. Figure 1 presents<br />
an overview of average number of agreement errors for items with particular<br />
termination in the first experiment (Czech).<br />
Figure 1.<br />
Average number of errors for particular terminations within each gender<br />
in the Czech experiment.<br />
[I-c<br />
.-e<br />
D-a<br />
ml-o<br />
mase. I-cl fern. [-a] neut. [-0]<br />
A typical termination for masculine nouns in Czech is a consonant (e.g. strom<br />
(M) - tree), for feminine -a (e.g. kniha (F) - book) <strong>and</strong> for neutral -0 (e.g. m sto<br />
(N) - town). Figure 1 shows that if the experimental nouns had a gender<br />
predictable from their terminations, subjects made almost no agreement errors.<br />
The vowel -e is from this aspect an ambiguous termination in Czech, because<br />
nouns ending with -e are usually either feminine or neuter (e.g. iidle (F) - chair<br />
<strong>and</strong> slunce (N) - sun). As obvious from the table, the number of agreement<br />
errors with these nouns was higher. Typically, if the participants made a gender<br />
error in agreement with a feminine noun ending with -e, it was because they took<br />
it for a neuter (e.g. *male(N) iidle - small chair), <strong>and</strong> vice versa (e.g. *mald(F)<br />
slunce - small sun). Most errors were made in agreement with feminine nouns<br />
ending with a consonant (mys(F) - mouse). As mentioned above, a final<br />
consonant is a termination typical for masculine nouns (there are only a few<br />
derived Czech masculine nouns which end with a vowel) <strong>and</strong> though the number<br />
of Czech feminine nouns ending with a consonant is quite large, it is still<br />
15
Denisa Bordag<br />
considered an atypical feminine ending. Characteristically, almost all feminine<br />
nouns ending with a consonant were considered masculine by the subjects<br />
(*maly(M) mys - small mouse),<br />
The situation was similar in the second experiment (German) as<br />
summarized in Figure 2.<br />
Figure 2.<br />
Average number of errors for particular terminations within each gender<br />
in the German experiment.<br />
5<br />
4,5<br />
4<br />
3,5<br />
3<br />
2,5<br />
2<br />
1,5<br />
1<br />
0,5<br />
o<br />
rnasc. [-cl neut. [-cl fern. [-e]<br />
In this experiment, the variety of the noun terminations was not so large as in the<br />
Czech experiment: The experimental nouns ended either with a consonant or<br />
with -e. This reflects the situation in German inflectional morphology, which is<br />
less rich than in Czech. The only typical termination in this experiment was -e,<br />
marking predictably German feminine nouns (e.g. Brucke'F) - bridge). As<br />
clearly seen from the table, almost no gender errors were made with these nouns<br />
(one error on each such item on average in both conditions in the whole<br />
experiment). A consonant must be considered an ambiguous termination in<br />
German, because it is a usual ending for both masculine <strong>and</strong> neuter nouns (e.g.<br />
Kopf(M) - head <strong>and</strong> Herz(N) - heart). It is thus not surprising that subjects often<br />
confused German masculine nouns with neuter ones (e.g. *kleines(N) Kopfsmall<br />
head) <strong>and</strong> vice versa (*kleiner(M) Her: - small heart). (The average<br />
16
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
number of gender errors for neuter nouns in both conditions was 3.75, for<br />
masculine nouns 4.8.). A possible interpretation of the fact that subjects made<br />
more gender errors in this experiment (138 compared to 74 in the Czech<br />
experiment), thus might be that in German participants could rely less on a word<br />
termination as a source of information about grammatical gender. If they did, it<br />
often resulted in agreement errors, because the correlation between word<br />
termination <strong>and</strong> grammatical gender is generally lower in German than in Czech.<br />
The results of the gender errors analysis clearly shows that intralingual<br />
interference affected the gender production in both experiments as well: Subjects<br />
clearly overgeneralized the phonological rules for gender assignment in both<br />
languages <strong>and</strong> applied them to nouns with different gender as well. This effect<br />
was stronger in Czech than in German, because the correlation between<br />
grammatical gender <strong>and</strong> word termination is more straightforward in this<br />
language. The analysis further showed that phonological form of a word plays an<br />
important role in second language grammatical gender production <strong>and</strong> that<br />
subjects take word termination into account <strong>during</strong> gender selection.<br />
It is noteworthy that the way the phonological form of a noun affects the<br />
gender assignment in a second language is similar to the way a noun termination<br />
influences gender assignment in the child language. Overgeneralizations of<br />
typical terminations <strong>and</strong> difficulties with gender of nouns with ambiguous or<br />
atypical terminations have been observed in the language of both Czech <strong>and</strong><br />
German children. Henzel (1975), who published probably the only article on this<br />
topic concerning acquisition of the first language Czech reports error free gender<br />
production with nouns with a "clear-cut phonological form" <strong>and</strong> states that:<br />
"the child learning to speak Czech gives preference to those gender categories<br />
which are least ambiguous rather than to those that are most frequent; he<br />
attends to the consonant ending with masculine, to the a-ending with feminine,<br />
<strong>and</strong> to the a-ending with neuter" (Henzel 1975, 189)<br />
More evidence about first language gender acquisition has been collected about<br />
German. In her monography on gender acquisition by German <strong>and</strong> English<br />
17
Denisa Bordag<br />
children, Mills (1986) observes that similarly to the second language learners,<br />
also German children have least problems with acquisition of feminine gender of<br />
nouns ending with -e <strong>and</strong> show difficulties with acquisition of gender of nouns<br />
with consonantal termination.<br />
4. Conclusion<br />
According to the results of this study there seem to be at least three factors<br />
affecting gender production in a second language:<br />
(1) The gender of the corresponding first language translation equivalent<br />
(interlingual interference/transfer).<br />
(2) The phonological form of the to-be-selected second language noun<br />
(intralingual interference/transfer),<br />
(3) <strong>Gender</strong> information saved on the second language lemma (otherwise<br />
subjects would not be able to produce the correct gender of nouns whose<br />
gender differs from that of their first language translation equivalent<br />
<strong>and</strong>/or has an atypical termination).<br />
The gender errors reported in the language of Czech <strong>and</strong> German children<br />
are also in accordance with the hypothesis that at least in the early stages of<br />
language acquisition children assign gender on the basis of phonological form of<br />
a noun. As native speakers later achieve full competence in gender production<br />
<strong>and</strong> can retrieve grammatical gender very fast without making errors, it is<br />
assumed by the models of adult language speech production that the information<br />
about grammatical gender is stored.<br />
Inlight of the findings reported in this paper it however does not seem to be<br />
necessarily the case that grammatical gender of second language nouns is stored<br />
as their inherent property in the same way as it is assumed for native languages.<br />
It is more likely that the grammatical gender of second language nouns is<br />
computed each time it is needed anew on the basis of stored information,<br />
information coming from the first language lemmas <strong>and</strong> information coming<br />
from the phonological form. In this aspect the second language gender<br />
production seems to be closer to the child than to the adult language processing.<br />
18
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
Interestingly, effects similar to those in the child language were observed with<br />
quite proticient second language learners <strong>and</strong> with nouns belonging to the core of<br />
basic vocabulary in our experiments. It would be interesting to see whether the<br />
same interference effects could be obtained with second language speakers of<br />
near-native proficiency, or whether it is possible for the second language learners<br />
to reach a processing stage equivalent to that of adult native language speakers.<br />
The critical period is likely to play a role here.<br />
The influence of phonological form is of crucial importance for the models<br />
of speech production, because it would assume a feedback activation between the<br />
level of phonological forms <strong>and</strong> the lemma level, where gender selection takes<br />
place. Whereas this idea is compatible with models permitting bottom-up<br />
processes like the Interactive Activation Model of Dell (1986), it is incompatible<br />
with strictly serial models without feedback like the one of Levelt (1989, Levelt<br />
et al. 1999). However, it is not possible to decide on the basis of our data,<br />
whether the effect of phonological form could not be also a post-lexical strategy<br />
which the subjects use to solve the problem of gender assignment in cases where<br />
the usual mechanisms fail. If the feedback from the phonological form is<br />
however an automatic process, speech production models which aspire to explain<br />
both the first <strong>and</strong> second language production <strong>and</strong> which do not permit bottomup<br />
processes would have to reconsider their claims. Further experiments are<br />
clearly necessary to answer this question <strong>and</strong> many others concerning gender<br />
production <strong>and</strong> representation in second languages.<br />
References<br />
Bordag, D. (2003) Gramaticky rod a e ova produkce v mate skern a cizim<br />
jazyce. asopis eskoslovenske psychologie 2:105-122.<br />
De Bot, K. (1992) A bilingual production model: Levelt's speaking model<br />
adapted. Applied Linguistics 13:1-24.<br />
Dell, G. S. (1986) A spreading-activation model of retrieval in sentence<br />
production. Psychological Review 93:231-241.<br />
Dulay, H.C. <strong>and</strong> M.K. Burt (1974a) Errors <strong>and</strong> strategies in child second language<br />
acquisition. TESOL Quarterly 8:129-136.<br />
Dulay, H.C. <strong>and</strong> M.K. Burt (1974b) Natural sequences in child second language<br />
acquisition. Language learning 24:37-53.<br />
19
Denisa Bordag<br />
Grainger, J. <strong>and</strong> T. Dijkstra (1992) On the representation <strong>and</strong> use of language<br />
information in bilinguals. In: R. J. Harris (ed), Cognitive Processing in<br />
Bilinguals. Amsterdam: Elsevier, 207-220.<br />
Green, D. W. (1993) Towards a model of second language comprehension <strong>and</strong><br />
production. In: R. Schreuder <strong>and</strong> B. Weltens (eds), The Bilingual Lexicon.<br />
Amsterdam: John Benjamins, 249-277.<br />
Green, D. W. (1998) Mental control of the bilingual-semantic system.<br />
Bilingualism: Language <strong>and</strong> Cognition 1:67-81.<br />
Grosjean, F. (1988) Exploring the recognition of guest words in bilingual speech.<br />
Language <strong>and</strong> Cognitive Processes 3:233-274.<br />
Grosjean, F. (] 997) Processing mixed language: Issues, findings <strong>and</strong> models. In:<br />
A. De Groot <strong>and</strong> J. Kroll (eds), Tutorials in Bilingualism: Psycholinguistic<br />
Perspectives. Mahwah, NJ: Lawrence Erlbaum Associates, 225-254.<br />
Grosjean, F. (I 998) Studying bilinguals: Methodological <strong>and</strong> conceptual issues.<br />
Bilingualism: Language <strong>and</strong> Cognition 1:131-]49.<br />
Henzel, V. M. (1975) Acquisition of grammatical gender in Czech. Reports on<br />
Child Language Development 10:188-200.<br />
Hermans, D., Bongaerts, T., De Bot, K. <strong>and</strong> R. Schreuder (1998). Producing<br />
words in a foreign language: Can speakers prevent interference from their<br />
first language Bilingualism: Language <strong>and</strong> Cognition 1:213-229.<br />
Jescheniak, J. D. <strong>and</strong> H. Schiefers (1998) Discrete serial versus cascaded<br />
processing in lexical access in speech production: Further evidence from the<br />
coactivation of near synonyms. Journal of Experimental Psychology:<br />
Learning, Memory <strong>and</strong> Cognition 24:1256-1274.<br />
Kolers, P.A. (1963) <strong>Interlingual</strong> word associations. Journal of Verbal Learning<br />
<strong>and</strong> Verbal Behavior 2: 291-300.<br />
Kroll, J. F. <strong>and</strong> J. Curley (1988) Lexical memory in novice bilinguals: The role of<br />
concepts in retrieving second language words. In: M. Gruneberg, P. Morris<br />
& R. Sykes (eds), Practical Aspects of Memory: Vol.2. New York: Wiley,<br />
389-395.<br />
Lado, R. (1957) Linguistics across cultures. Applied linguistics for language<br />
teachers. Ann Arbor: University of Michigan.<br />
La Heij, W., Mak, P., S<strong>and</strong>er, J. <strong>and</strong> E. Willeboordse (1998) The gender<br />
congruency effect in picture-word tasks. Psychological Research 61:209<br />
219.<br />
Levelt, W. J. M. (1989) Speaking: From intention to articulation. Cambridge,<br />
MA: MIT Press.<br />
Levelt, W. J. M., Roelofs, A. <strong>and</strong> A.S. Meyer, A. S. (1999) A theory of lexical<br />
access in speech production. Behavioral <strong>and</strong> Brain Sciences 22:1-75.<br />
MacWhinney, B. (1997) Second language acquisition <strong>and</strong> the competition model.<br />
In: De Groot, A. M. B., Kroll J. F. (eds), Tutorials in Bilingualism:<br />
Psycholinguistic Perspective. Mahvah, NJ: Lawrence Erlbaum, 1997, 113<br />
142.<br />
Mills, A. (1986) The Acquisition of <strong>Gender</strong>. A Study of English <strong>and</strong> German.<br />
Berlin Heidelberg: Springer Verlag<br />
20
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
Paradis, M. (1981) Neurolinguistic organisation of bilinguals: two languages. In:<br />
J. E. Copel<strong>and</strong> <strong>and</strong> P. W. Davis (eds), The Seventh LACUS Forum.<br />
Columbia, SC: Hornbeam Press, 486-494.<br />
Paradis, M. (1987) The Assessment of Bilingual Aphasia. Hillsdale, NJ:<br />
Lawrence Erlbaum Associates.<br />
Peterson, R. R. <strong>and</strong> P. Savoy (1998) Lexical selection <strong>and</strong> phonological encoding<br />
<strong>during</strong> language production: Evidence for cascaded processing. Journal of<br />
Experimental Psychology: Learning, Memory <strong>and</strong> Cognition 24:539-557.<br />
Potter, M. C., So, K.-F., von Eckhardt, B. <strong>and</strong> L.B. Feldman (1984) Lexical <strong>and</strong><br />
conceptual representations in beginning <strong>and</strong> more proficient bilinguals.<br />
Journal of Verbal Learning <strong>and</strong> Verbal Behaviour 23:23-38.<br />
Poulisse, N. (1997) Language production in bilinguals. In: De Groot, A. M. B.,<br />
Kroll J. F. (eds), Tutorials in Bilingualism: Psycholinguistic Perspective.<br />
Mahvah, NJ: Lawrence Erlbaum, 201-224.<br />
Poulisse, N. <strong>and</strong> T. Bongaerts (1994) First language use in second language<br />
production. Applied Linguistics 15:36-57.<br />
Richards, J. (1971) Error analysis <strong>and</strong> second language strategies. English<br />
Language Teaching 25:115-135.<br />
Schriefers, H. (1993) Syntactic processes in the production of noun phrases.<br />
Journal of Experimental Psychology: Learning, Memory <strong>and</strong> Cognition 19:<br />
841-850.<br />
Schriefers, H. <strong>and</strong> J. D. Jescheniak (1999) Representation <strong>and</strong> Processing of<br />
Grammatical <strong>Gender</strong> in Language <strong>Production</strong>: A Review. Journal of<br />
Psycholinguistic Research 28:575-600.<br />
Schriefers, H. <strong>and</strong> E. Teruel (2000) Grammatical gender in noun phrase<br />
production: The gender interference effect in German. Journal of<br />
Experimental Psychology: Learning, Memory <strong>and</strong> Cognition 26:1368-1377.<br />
Van Berkum, J. J. A. (1997) Syntactic processes in speech production: The<br />
retrieval of grammatical gender. Cognition 64: 115-152.<br />
Weinreich, U. (1974) Languages in contact. Findings <strong>and</strong> Problems. The Hague,<br />
Paris: Mouton.<br />
Appendix<br />
Items used in Experiments 1 <strong>and</strong> 2 in both conditions. The lists contain the target<br />
L2 words, their L1 translation equivalents (with their corresponding genders) <strong>and</strong><br />
the English translation.<br />
Experiment 1<br />
<strong>Gender</strong> congruent group<br />
<strong>Gender</strong> incongruent group<br />
Czech (L2) German (L1) English Czech (L2) German (L1) English<br />
Mf m Ball m ball Nos m Nase f nose<br />
Pes m Hund m dog Byt m Wohnung f flat<br />
Vlak m Zug m train hrad m Burg f castle<br />
21
Denisa Bordag<br />
Strom m Baum m tree most m Brticke f bridge<br />
Dopis m Brief m letter obraz m Bild n picture<br />
Kabat m Mantel m coat kostel m Kirche f church<br />
M sic m Mond m moon darek m Geschenk n present<br />
Koberec m Teppich m carpet pomeran m Orange f orange<br />
Mys F Maus f mouse Lo f Schiff n ship<br />
Ruka F H<strong>and</strong> f h<strong>and</strong> noha f FuB m foot<br />
Svf ka F Kerze f c<strong>and</strong>le kniha f Buch n book<br />
Mapa F Karte f map hlava f Kopf m head<br />
Sprcha F Dusche f shower ryba f Fisch m fish<br />
Ko ka F Katze f cat zidle f Stuhl ill chair<br />
Tabule F Tafel f blackboard vesnice f Dorf n village<br />
Koruna F Krone f crown zmrzlina f Eis n ice-cream<br />
Srdce n Herz n heart raj e n Tomate f tomato<br />
Pole n Feld n field slunce n Sonne f sun<br />
Oko n Auge n eye m sto n Stadt f city<br />
Ucho n Ohr n ear vino n Wein m wine<br />
Okno n Fenster n window k eslo n Sessel m arm-chair<br />
Letadlo n Flugzeug n plane zrcadlo n Spiegel m mirror<br />
Experiment 2<br />
<strong>Gender</strong> congruent group<br />
<strong>Gender</strong> incongruent group<br />
English<br />
German (L2) Czech (Ll) German (L2) Czech (Ll) English<br />
Ball m Mf m ball Fuss m noha f foot<br />
Hund m Pes m dog Stuhl m zidle f chair<br />
Zug m Vlak m train Kopf m hlava f head<br />
Baum m Strom m tree Wein m vino n wine<br />
Brief m Dopis m letter Fisch m ryba f fish<br />
Zahn m Zub m tooth Schrank m sk f f wardrobe<br />
22
<strong>Interlingual</strong> <strong>and</strong> <strong>Intralingual</strong> <strong>Interference</strong><br />
Teller m Tall m plate Sessel m k eslo n arm-chair<br />
Teppich m Koberec m carpet Spiegel m zrcadlo n mirror<br />
Maus f Mys f mouse Burg f hrad m castle<br />
H<strong>and</strong> f Ruka f h<strong>and</strong> Stadt f m sto n city<br />
Kerze f Svf ka f c<strong>and</strong>le Nase f nos m nose<br />
Rose f R ze f map Sonne f slunce n son<br />
Dusche f Sprcha f shower Brticke f most m bridge<br />
Katze f Ko ka f cat Kirche f kostel m church<br />
Tafel f Tabule f blackboard Butter f maslo n butter<br />
Krone f Koruna f crown Tomate f raj e n tomato<br />
Herz n Srdce n heart Schiff n 10 f ship<br />
Feld n Pole n field Bett n postel f bed<br />
Ei n Vejce n egg Pferd n k m horse<br />
Ohr n Ucho n ear Dorf n vesnice f village<br />
Fenster n Okno n window Eis n zmrzlina f ice-cream<br />
Flugzeug n Letadlo n plane Geschenk n darek m present<br />
Denisa Bordag<br />
lnstitut fur Linguistik<br />
Universitat Leipzig<br />
BeethovenstraBe 15<br />
04107 Leipzig<br />
Germany<br />
e-mail: denisav@gmx.net<br />
23