Вы находитесь на странице: 1из 16

International Journal of Bilingual Education and

Bilingualism

ISSN: 1367-0050 (Print) 1747-7522 (Online) Journal homepage: http://www.tandfonline.com/loi/rbeb20

Testing the Language Mode Hypothesis Using


Trilinguals

Ton Dijkstra & Janet G. van Hell

To cite this article: Ton Dijkstra & Janet G. van Hell (2003) Testing the Language Mode
Hypothesis Using Trilinguals, International Journal of Bilingual Education and Bilingualism, 6:1,
2-16, DOI: 10.1080/13670050308667769

To link to this article: http://dx.doi.org/10.1080/13670050308667769

Published online: 26 Mar 2010.

Submit your article to this journal

Article views: 345

View related articles

Citing articles: 30 View citing articles

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=rbeb20

Download by: [University of Texas Libraries] Date: 16 February 2017, At: 11:01
Testing the Language Mode Hypothesis
Using Trilinguals
Ton Dijkstra
NICI, University of Nijmegen, The Netherlands
Janet G. van Hell
Department of Special Education, University of Nijmegen, The Netherlands

Current multilingual word recognition models differ in their account of non-linguistic


context effects, for example effects due to stimulus list composition and task demands.
Several models assume that the non-linguistic context can modulate the relative activa-
tion of words from different languages. One prominent example of such an approach is
the Language Mode view by Grosjean (1997b), according to which the relative state of
activation of a multilinguals two or more languages and language-processing mecha-
nisms depends not only on the characteristics of the stimulus input, but also on situa-
tional characteristics. However, recent studies do not support this viewpoint. It
predicts, for instance, that for DutchEnglish bilinguals performing a purely Dutch
word recognition task, English word candidates should not be activated. Nevertheless,
under precisely those circumstances, DutchEnglish and DutchFrench cognates
(words with a similar orthography and meaning across languages) were both processed
differently from Dutch control words by relatively proficient DutchEnglishFrench
trilinguals. On the basis of this and other studies, we argue in favour of a stimu-
lus-driven model for visual word recognition in which the non-linguistic context
cannot affect the relative activation of word candidates from different languages.

The Language Activation Metaphor


On our summer vacation abroad, we may discover that it takes a few days
before we can easily make use again of our high school knowledge of the coun-
trys language. At another time, we may search for a particular term in our native
language but its translation equivalent from a foreign language stubbornly keeps
coming to mind, because we have just read it in a book. To account for phenom-
ena like these and many others in the empirical literature, researchers often make
use of the so-called language activation metaphor. Depending on the circum-
stances, one language is more active than another, they say. However, the use of
this metaphor is seldom specified. Thinking a bit more about what the activation
metaphor might mean in the context of bilingual word recognition, one must
conclude that the term activated language is quite ambiguous. It may imply, for
instance, any of the following concepts:
(1) There is an active abstract representation of language-as-a-whole, which
refers to the currently important language in conversation. Such a represen-
tation might be useful for syntactic purposes, to keep track of the language
of the on-going sentence. This notion is similar to that of a language tag or
language node, found in some word recognition models (e.g. the Bilingual
Interactive Activation or BIA model by Dijkstra & Van Heuven, 1998). It has
been defended by some researchers and attacked by others.

1367-0050/03/01 0002-15 $20.00/0 2003 T. Dijkstra & J.G. van Hell


International Journal of Bilingual Education and Bilingualism Vol. 6, No. 1, 2003
2

Language Mode and Trilinguals


Language Mode and Trilinguals 3

(2) Because language is an abstract concept, the phrase active language


implies that its words are active without being used. As such, this sounds
rather odd. However, one may assume that words in the mental lexicon
have a resting level activation that can be adapted relative to the words of
another language. Words would only become really activated if they are
heard or read. This type of view is derived from monolingual word recogni-
tion research, where a words frequency and recency of usage determines its
resting level activation in Interactive Activation models (see Van Heuven et
al., 1998).
(3) Possibly, the amount of activation needed to recognise a word from the
active language is less than that to recognise a word from another
language. In other words, the recognition threshold for words from differ-
ent languages may be set separately (see Dijkstra & Van Heuven, 2002),
implying that faster word decisions can be made on words from the active
language.
(4) Activation is a relative concept. Word candidates from the target language
are activated faster than those from other languages, for instance, because
they receive extra contextual stimulation (cf. the BIMOLA model by Lwy &
Grosjean, in preparation), or because those other-language words have
been inhibited (BIA model by Dijkstra & Van Heuven, 1998; Inhibition
Control or IC model by Green, 1998).

Context Effects on Relative Language Activation


Current word recognition models differ in how they incorporate the activa-
tion metaphor into their frameworks. They differ considerably in their views
about how the relative activation of languages depends on (1) expectations by
the participants induced, for instance, by variation in the instructions; and (2)
stimulus list composition or language intermixing, for instance, the language of
the previous item in a stimulus list. The following bullet list summarises some of
the basic characteristics of four important approaches. As can be seen in this list,
there is one approach that assumes a strong sensitivity of relative language acti-
vation to top-down context factors, the Language Mode hypothesis, and there is
one approach that assumes no direct effects of (unrelated) context stimuli on the
multilingual word recognition process itself. In this paper, we will evaluate some
recent experimental evidence pertaining to this issue. Critical for the experi-
ments in our laboratory is the use of trilingual participants in order to allow a
direct test of the different theoretical viewpoints.
(1) Language Mode framework (Grosjean, 1998, 2001):
relative activation state of languages (language mode) is continuous and
sensitive to many factors;
stimulus list composition (previous items) affects language activationstate;
participant intentions/expectations and instruction affect language activa-
tion state;
context-sensitivity of language mode is partially due to such top-down
effects;
identification and decison levels interact.
4 Bilingual Education and Bilingualism

(2) Inhibitory Control model (Green, 1998):


language task schemas (specifying how a task is performed) can compete
and cooperate;
schemas can alter the activation level of lexical representations (lemmas);
stimulus list composition (previous items) affects activation state of
lemmas;
participant intentions can affect the activation state of items;
reactive top-down inhibition effects operate on lemmas, not on word
forms;
identification and decision levels interact.
(3) Bilingual Interactive Activation model (Dijkstra & Van Heuven, 1998):
resting level activation of words reflects the state of language activation as
well as proficiency;
stimulus list composition (previous items) affects activation state of word
forms;
participant expectations do not provide strong effects on the activation
state of words;
top-down inhibition effects on the non-target language arise via language
nodes;
identification and decision levels interact.
(4) BIA+ model (Dijkstra & Van Heuven, 2002):
resting level activation of words reflects the state of language activation as
well as proficiency;
stimulus list composition (previous items) affects task/decision system;
participant expectations may affect task decision system;
no top-down effects exist from task/decision system on identification
system (always full activation of lexemes);
identificationdecision: purely bottom-up information flow.

Language Mode View in Detail


Before we present recent empirical evidence that examines the relative
language activation issue using trilinguals, we will clarify the main theoretical
position that we are examining: that of Grosjeans Language Mode (LM)
hypothesis. In several influential papers on bilingual language processing,
Grosjean (1985, 1997a, 1997b, 1998, 2001) has proposed and developed the
general notion of a language mode, referring to the relative state of activation
of a bilinguals two (or more) languages and language-processing mechanisms.
Depending on such factors as the person spoken to, the situation, the content of
discourse and the function of the interaction, one language (the base language)
may be active while the state of the other language may vary from deactivated
(monolingual language mode) to relatively active (bilingual language mode).
Recently, Grosjean (1998: 137) has extended the notion of the language mode
to language perception and reading as follows (for a discussion of the LM
hypothesis as applied to language production in trilinguals, see DeWaele,
2001).
Language Mode and Trilinguals 5

During perception, if bilingual listeners who start off in a monolingual


mode determine (consciously or not) as they go along, that what they are
listening to can contain elements from the other language, they will put
themselves in a bilingual mode (at least partly), that is, activate both their
languages (with the base language being more strongly activated). This is
also true of readers, whether they are reading a continuous text or looking
at individual lexical items interspersed with items from the other language.
Simply knowing that there is a possibility that elements from the other
language will be presented (in an experiment, for example) will move the
bilingual away from the monolingual endpoint of the continuum. Just one
guest word in a stream of base language words can increase this
displacement.
In sum, according to Grosjean the bilinguals language mode affects percep-
tion and the speed of access to one or two lexicons, and the language mode itself
is affected both by the readers (linguistic) expectations and by language
intermixing.
One study that Grosjean (2001) considers as (indirect) evidence in support of
this LM hypothesis is that by Dijkstra et al. (1998). In three lexical decision experi-
ments, these authors found that interlingual homographs were recognised faster
than, slower than or as fast as monolingual control words depending on task
demands and stimulus list composition (language intermixing). The participants
in their study (and in many of the studies we will review later) were
DutchEnglish bilinguals, mostly university students who learnt English as a
second language from about 1112 years onwards.
In their first experiment, these bilinguals performed an English lexical deci-
sion task including EnglishDutch homographs and cognates, as well as mono-
lingual English control words and non-words. Participants were instructed to
press a yes button if a presented letter string was a correct English word and a
no button if it was not. Interlingual homographs are words that have the same
written form (orthography) but a different meaning (semantics) across
languages (e.g. stem is a DutchEnglish homograph with a high frequency
reading in English and a high frequency reading in Dutch, in which it means
voice). Cognates are words that share not only their orthographic form but also
(most of) their meaning(s) across languages (e.g. land is a DutchEnglish
cognate with a high frequency reading in both languages). Just like exclusively
English words, interlingual homographs and cognates require a yes response in
this task, because they have an English reading. In the experiment, no significant
reaction time (RT) differences were observed for interlingual homographs rela-
tive to monolingual control words. Thus, the frequency of the Dutch reading of
the homographs did not clearly affect the RTs to their English counterpart.
However, the RTs to the cognate items were facilitated relative to their matched
monolingual English controls.
In a second experiment, DutchEnglish bilinguals again performed an
English lexical decision task, but now the experimental stimuli not only included
interlingual homographs but also exclusively Dutch words requiring a no reac-
tion. Here, a dramatic change was observed in the RTs to homographs: relative to
English control items, strong inhibitory effects were obtained. The participants
6 Bilingual Education and Bilingualism

apparently found it difficult to give a yes response to these words on the basis of
the English language only. Furthermore, the RT difference between homographs
and matched control words was now dependent on the relative frequency of the
two readings of the homograph: slower RTs were obtained for homographs
when the Dutch reading was relatively frequent. Thus, there were inhibition
effects for interlingual homographs relative to words that belong to only one
language. This second experiment indicates that the small size of the effects in the
first experiment depended on the task situation and that it was not a straightfor-
ward consequence of the bilinguals lexical organisation of homographs.
Implications of this view were tested in a third experiment, in which the
bilinguals performed a general lexical decision task, responding yes if a word of
either language was presented. Now frequency-dependent facilitation effects
were observed for interlingual homographs relative to monolingual control
items, indicating that participants could succesfully exploit the double readings
of the homograph items to speed up their lexical decision process.
Grosjeans (2001: 9) account of these results in terms of the LM hypothesis
applied to language comprehension is as follows:
In Experiment 1 the participants were positioned towards the monolingual
end of the continuum without reaching it totally though as they knew they
were being tested as bilinguals. They only heard [must be: read, TD &
JvH] English words and non-words (although some words were homo-
graphs and cognates) and they were asked to decide whether the items
were English words or not. Thus although their Dutch was partly active
(which would explain the cognate effect) it was not sufficiently active to
create a homograph effect. However, in Experiment 3, the participants
were definitely at the bilingual end of the continuum. Not only were the
words and nonwords both English and Dutch but the participants were
asked to do general lexical decision, that is search/look-up both their lexi-
cons to accomplish the task.
In sum, the bilinguals participating in Experiment 1 were in a language mode
that was more positioned towards the monolingual end of the continuum than
the participants in Experiment 3 (and presumably, Experiment 2). Grosjean gave
two reasons for why this was so: (1) the participants read only English words
(although some words were interlingual homographs and cognates); and (2)
they were asked to make exclusively English lexical decisions and therefore
mainly kept their English lexicon active. Thus, the LM view assumes that
bottom-up factors such as stimulus list composition affect performance through
an adaptation of the language mode, which can also be affected by the partici-
pants interpretation of the experimental situation.
However, the results of Experiment 2 would seem to throw some doubt on the
second conclusion. In this experiment, participants were also required to make
exclusively English lexical decisions. Given that they had to say no to Dutch
words, they would have benefited from completely deactivating or suppressing
their Dutch lexicon. In that case Dutch words, just like non-words, would have
led to little activity in the mental lexicon, making it easier to respond yes to
English words and interlingual homographs and no to Dutch words. One might
argue, however, that perhaps the no response to Dutch words has come about
Language Mode and Trilinguals 7

by first identifying the Dutch words and then rejecting them on the basis of their
language membership. This would explain why there are inhibition effects for
the interlingual homographs after all. However, other studies indicate that to
obtain the inhibition effect for English homographs relative to controls, an
explicit response to the Dutch items is not required. Dijkstra et al. (2000b) also
obtained strong inhibition effects for interlingual homographs in the language
go/no-go task. In this task, participants press a button if a word of a predefined
target language is presented, but words from the non-target language are not
responded to. In the Dutch language go/no-go task, for example, no responses
needed to be given to English words but the responses to DutchEnglish
interlingual homographs were still slower than to purely Dutch words. This
analysis of Experiment 2 indicates that the participants in this experiment must
have been in a bilingual language mode over which they had no control. We have
already seen that the participants in Experiment 3 were thought to be in a bilin-
gual language mode.
What about the participants of Experiment 1? Several studies have shown that
even though in such an experiment no RT differences are found between
interlingual homographs and control words, word representations from both
languages are still activated (De Moor, 1998; Dijkstra et al., 1999; Van Heste,
1999). For instance, De Moor (1998) demonstrated that the L1 semantics of the
interlingual homographs was activated, even though there was a null result for
such homographs relative to controls. In the trial after the homograph appeared,
she presented the English translation of the Dutch reading of the interlingual
homograph. For instance, brand was followed by fire, which is the English
translation of the Dutch word brand. A small but reliable translation priming
effect of 11 milliseconds (ms) was found on the recognition of the second item
fire relative to a control condition. In a replication of this experiment with differ-
ent stimulus materials, Van Heste (1999) observed a reliable 35 ms difference
between translation and control trials. The Dutch reading of the interlingual
homograph in the previous trial had apparently been activated even though this
did not affect the RT to the homograph.
Furthermore, Dijkstra et al. (1999) also performed an experiment that involved
an English lexical decision task with interlingual homographs and controls, but
they manipulated the cross-linguistic similarity of the items rather than their
relative frequency in the two languages. Significant facilitation effects were
found for homographs having cross-linguistic overlap in orthography but not in
phonology (stage) and no effects for items with overlap in both(step). The
items in this study were comparable to those in Dijkstra et al. (1998), making it
likely that the null effects in the earlier study were due to mixing the two types of
items.
These studies show that word representations from both languages were acti-
vated in Experiment 1 by Dijkstra et al. (1998). This implies that the participants
cannot have been in a monolingual mode. In fact, there is no evidence that the
participants were in a different language mode in any of the three experiments.
Of course, it could be argued that the participants in all three experiments were in
a bilingual mode, because the homographs and cognates in Experiment 1 were,
after all, also Dutch words. However, this raises the question as to whether a
mode shift or mode variations can ever be observed or whether participants are
8 Bilingual Education and Bilingualism

always in a bilingual mode (i.e. words from both languages are always activated
and lexical access is profoundly language non-selective). Clearly, the usefulness
of the LM hypothesis for reading depends on the extent to which such shifts or
variability are observed across studies (e.g. could there be mode shifts when
target words are preceded by briefly presented masked prime items?). If
evidence for shift or variability in mode is never observed, the idea of a language
mode under control of the reader is only of limited use. In our view, the available
evidence does not provide strong evidence in favour of mode shifts in the word
recognition of relatively L2-fluent but still unbalanced DutchEnglish bilinguals
(Dijkstra & Van Heuven, 2002).

Testing the LM Hypothesis with Trilinguals


Grosjean (2001) has indicated that the LM hypothesis might be tested only
under very specific conditions and that currently no study has really met all
necessary constraints. With respect to the visual modality, three experiments by
Van Hell and Dijkstra (2002) may come closest to providing satisfactoryevidence
with respect to the LM issue.
In these experiments, we explicitly aimed at bringing the participants into
monolingual mode, i.e. their native language mode. Given that the second and
third languages of multilinguals are weaker than their native language, this is the
monolingual language mode that we reasoned would be easiest to establish. In
the first two experiments, this was done by (1) following a careful selection
procedure with respect to participants in the study; (2) using a task and stimulus
materials that did not explicitly refer to their foreign language knowledge; and
(3) providing an exclusive L1 communicative setting.
The goal of the selection procedure was to prevent the trilinguals from know-
ing in advance that we were interested in their foreign language knowledge. At
the beginning of the school year, all first-year psychology students of the Univer-
sity of Amsterdam complete a large battery of pen-and-paper tests. In one of
these tests, they were asked to fill in the subjects of their exams in secondary
school. On the basis of this test we wrote a letter to all students (n = 150) who had
done final school exams in Dutch, English and French (Dutch and English are
obligatory subjects and French is an optional subject in secondary school exams).
They were asked if they were interested in participating in a memory experi-
ment, without making any reference to its multilingual nature. This resulted in a
selection of DutchEnglishFrench trilinguals, who were most fluent in L1, less
fluent in L2 and least fluent in L3.
The instructions to these trilingual participants were all in their native
language and the task and stimulus materials did not make any direct reference
to the multilingual nature of the study. The participants were presented with
word stimuli from their dominant language, L1 (Dutch), and performed a task in
their L1. Half of the items in the list belonged exclusively to Dutch (e.g. the Dutch
word tuin, meaning garden in English). Half of the remaining items had a
cognate relationship with their English translations (e.g. Dutch bakker, mean-
ing baker in English) and half had a cognate relationship with their French
translations (e.g., Dutch muur, which is mur in French, meaning wall). The
Language Mode and Trilinguals 9

majority of the cognates were non-identical cognates, and hence were unambig-
uous with respect to which language they belonged.
In the first experiment of this study, a bilingual word association task was
performed by the selected student group. This production task requires the
person first to identify a word, thereby presumably activating a words ortho-
graphic, phonological and semantic information. In the subsequent activation
and production of an associate, semantic information seems to play an important
role (Van Hell & De Groot, 1998). An inspection of the actual associative
responses given, as listed by De Groot (1989) and Postman and Keppel (1970),
revealed that most associations are related in meaning to the presented stimulus
word. Using the word association task with bilinguals, Van Hell and De Groot
(1998) unexpectedly observed that word association times to L1 words that were
cognates with their translations in L2 were shorter than those to non-cognates.
We used the word association task in a new experimental context, using a differ-
ent sample of test stimuli, finely-tuned control items and different participants to
study the influence of foreign language knowledge on native language process-
ing. Mean association times to the L1 cognates with English were found to be
shorter that those to the L1 non-cognates (1641 versus 1845 ms). The mean associ-
ation times to the L1 cognates with French were also somewhat shorter than to
the L1 cognates (1809 versus 1845 ms), but this difference was not statistically
significant. Finally, the trilinguals were faster in associating the L1 cognates with
English than associating the L1 cognates with French.
In the second experiment, the same stimulus materials and a new group of
trilinguals, selected via the procedure of Experiment 1, were used. This experi-
ment involved a Dutch lexical decision task, in which the participants responded
as quickly as they could by pressing a yes button if a presented letter string was
a correct word in Dutch (e.g. huis) and by pressing a no button if it was not (e.g.
fuis). Lexical decision is probably the most frequently used experimental task to
study word recognition in both monolinguals and multilinguals. Even though
the word association and lexical decision tasks may be assumed to differ consid-
erably in their component processes, the patterns of results in the two experi-
ments were highly similar: as for word association, lexical decision times for
words that were cognates with their L2 (English) translations were shorter than
those for the non-cognates (499 versus 529 ms). Again, a non-significant cognate
advantage was obtained, however, with stimuli that were cognates with their
translation in the trilinguals third, and weakest, language (French) (519 versus
529 ms).
These results indicate that foreign language knowledge (L2) can affect lexical
processing in the native language (L1) even under circumstances in which the
participants are unaware their knowledge of other languages is relevant. At first
sight, this would seem to be impressive evidence against theoretical viewpoints
that assume a strong influence of the participants expectations on the relative
activation of their different languages. More specifically, this result seems
incompatible with the LM hypothesis formulated earlier. However, Grosjean
(personal communication, 2000) has suggested that Dutch participants may
always be in a bilingual mode, because Dutch culture is permeated with Ameri-
can-English. Dutch people use English frequently in everyday life through tele-
10 Bilingual Education and Bilingualism

vision, film, music, books and the internet. In fact, English is so widespread that
Dutch monolinguals are hard to find.
Fortunately, this argument can be tested with trilingual participants. Though
Dutch language users are in frequent contact with English for cultural reasons,
this is certainly not true for French. In the first two experiments the trilinguals
were most fluent in their native language (Dutch), somewhat less fluent in their
second language (English) and least fluent in their third and weakest language
(French, L3). This may explain why the cognate effect did not reach significance
in French. However, if we now test trilinguals with a higher level of fluency in
French, we should obtain a significant cognate effect with respect to that
language as well.
We therefore replicated the lexical decision experiment (Experiment 2) with
trilinguals, whose fluency levels in L2 and L3 were comparable (as assessed by an
independent test after the experiment). We had to follow a somewhat different
selection procedure, because we had to recruit students from the French
Language and Literature Department in order to conduct the experiments with
participants with a good knowledge of both L2 (English) and L3 (French). They
were recruited by a student of French who followed an optional course in experi-
mental psychology. She told potential participants that she ran a psychological
experiment to obtain course credits at the Psychology Department, and invited
them into the Psychology Laboratory. She did not tell participants that their
foreign language knowledge was important. After the experiment they were
asked about their expectation with respect to the nature of the experiment. All
participants thought they had participated in a psychological experiment and
were unaware of the importance of their foreign language knowledge.
The results confirmed the prediction that, for trilinguals who were fluent
enough in their L3, cross-linguistic effects could arise from the L3 on the native
language even though the participants were unaware of the relevance of their L3.
Now, not only did a cognate advantage arise in words that were cognates with
their L2 (English) translation (489 ms for English cognates versus 541 ms for
non-cognate controls) but also in words that were cognates with their L3 (French)
translation (520 ms for French cognates versus 541 ms for controls). These find-
ings provide evidence that lexical access is non-selective even for the mother
tongue under circumstances in which participants would presumably be in a
monolingual language mode.

Other Experiments Testing the LM Hypothesis


We consider our results to be fairly strong negative evidence against the LM
hypothesis. However, defenders of the hypothesis could still argue our study
leaves several ways out. First, because the experimenter in the third experiment
was a student of French who conducted the experiment with her fellow students,
one can argue that the participants may have switched on their French after all
(in addition to Dutch and English). We think that this is not a very strong argu-
ment, because it suggests that the language mode is extremely sensitive to
unemphasised contextual factors, while at the same time there is very little
evidence of variability or shifts in language mode in experimental situations
where such changes would clearly be useful (see Introduction). In addition, as
Language Mode and Trilinguals 11

we argued earlier, if participants are always in a multilingual mode, the concept


seems to be of little use. In that case, it would be more fruitful to focus on factors
that are stronger determinants of the observed RT data patterns, such as stimulus
list composition or L2/L3 proficiency.
A second possible alternative interpretation is to assume that cognates are not
the right kind of stimulus material to test the LM hypothesis. It can be argued that
cognates have developed a special representation due to the overlap in terms of
semantics, orthography and, to some extent, phonology. Of course, this option
assumes that multilinguals have, at some time, discovered this kind of overlap
and have reorganised their mental lexicon to benefit from it. According to this
view, the facilitation effect we observed arises from the special representation of
cognates, not from a parallel activation of words from different languages.
However, this position leads to a serious restriction for the LM hypothesis as
well. It has been estimated that about half of the Dutch vocabulary consist of
non-identical cognates with English (Van Hell, 1998). The cognates used in our
study therefore are not exotic words with limited relevance for daily language
use. If such items cannot be used to test the LM hypothesis, this implies that the
hypothesis can only be valid for a rather limited proportion of the Dutch
vocabulary.
An elegant recent bilingual study indicates that the ways out we just have
described may not seem to be viable in the auditory domain. Marian and Spivey
(submitted) tested the LM hypothesis in two eye-tracking experiments under
maximally controlled circumstances that should induce a strong monolingual
mode. Unacquainted monolingual speakers contacted and tested the partici-
pants, recorded the instructions, etc. A head-mounted eyetracker registered eye
movements of bilingual participants to particular objects presented on a display
with nine squares. The participants were RussianEnglish late bilinguals who
were tested in one language only (Russian or English). They received a spoken
instruction in English or Russian to move a particular target object. The name
properties of the objects on the display were varied systematically. Activation of
the unused language was indexed by eye movements to an object whose name in
the unused language was phonetically similar to the name of a target object in the
used language. For instance, it was examined whether participants looked at
matches [spichki in Russian] on the display when they were instructed to Pick
up the speaker. Between-language phonological competition, within-language
competition in the two languages as well as simultaneous competition from both
between-language and within-language phonological competitors were tested.
The results were unequivocal in showing effects of all types of related competi-
tors on the eye movements, indicating that phonologically related items in both
languages were activated in parallel. The result pattern seems to contradict the
LM hypothesis, which assumes there is a high degree of activation of the base
language and little activation in the other language under conditions that induce
a monolingual language mode (Grosjean, 2001, suggests that the other language
is probably never totally deactivated at the monolingual end of the language
mode continuum. However, this assumption in our view makes it impossible to
test the LM hypothesis unless there is any further specification of the relative
degree of language activation.)
A recent study suggests that the influence on performance exerted by
12 Bilingual Education and Bilingualism

top-down factors, such as the participants expectations or specificity of instruc-


tions, is less important than that exerted by bottom-up factors such as stimulus
list composition (language intermixing). Dijkstra et al. (2000a) instructed
DutchEnglish bilinguals to perform an English lexical decision task. They were
told explicitly that the stimulus list would consist of purely English words, but
also of interlingual homographs. If they encountered interlingual homographs,
they were to respond with a yes response, while Dutch words required a no
response. Examples were provided in the practice set. During testing, however,
exclusively Dutch words were not presented in the first half of the experiment
but only in the second half. The results showed no significant RT differences
between the interlingual homographs and matched English control items in the
first part of the experiment, but they showed strong inhibitory effects to such
items in the second half. Examination of the transition from Part 1 to Part 2
showed that, as soon as Dutch items started to come in, the RTs to interlingual
homographs suddenly slowed down considerably compared to control words.
These findings indicate that, even though the participants understand the
instruction, they await concrete examples of the different stimuli to adapt their
response strategies in order to optimise performance. In sum, the dramatic
change in performance after the transition from Part 1 to Part 2 of the experiment
indicates that stimulus list composition, rather than instruction-based expectan-
cies, determined the participants performance.

Conclusions
Even though we have concentrated on the LM hypothesis as applied to read-
ing, the experiments we discussed in this paper have consequences for all four
models described in the Introduction. The results presented support a model of
multilingual word recognition that has a very strong and automatic bottom-up
(signal-based) component. It appears that the amount of control that can be
exerted by multilingual participants over the relative activation of their two
languages during reading is quite limited (see Dijkstra & Van Heuven, 2002).
Lexical candidates from the different languages known by the proficient trilin-
gual seemed to be activated on the basis of an input word:
(1) irrespective of whether the target language was L1 (Dutch), L2 (English), or
L3 (French);
(2) in different tasks such as word association (having a production compo-
nent), lexical decision (visual comprehension) and eye-movement tracking
(auditory comprehension);
(3) in stimulus lists that were mixed or blocked with respect to language; and
(4) for cognates (Van Hell & Dijkstra, 2002) and cohort competitors (Marian &
Spivey, submitted). (Cohort competitors are those word candidates that are
activated by an input item, sharing their onsets with each other and with the
input item, e.g. captain and capital upon presentation of captive).
As such, these results go against the assumption that the relative activation
state of languages is very sensitive to participant intentions/expectations and
details of the experimenters instruction. This assumption appears to be embed-
Language Mode and Trilinguals 13

ded in the LM framework and, to some extent, in the IC model and BIA models as
well. Consequently, these models are not supported by the reported evidence.
In contrast, stimulus list composition (mixed or blocked) does affect the
observed pattern of results in the experiments (Dijkstra et al., 1998; Dijkstra et al.,
2000a). While it cannot yet be excluded on the basis of the presently available
evidence that stimulus list composition does affect the relative activation state of
languages in an experiment, there is some evidence that language switching
effects (i.e. the effect of the language of a previous item on the recognition of a
target item) do not originate from within the lexicon (Thomas & Allport, 2000;
Von Studnitz & Green, 1997). In sum, we believe that at present the best model to
account for the trilingual recognition data presented in this paper is the BIA+
model for bilingual word recognition by Dijkstra and Van Heuven (2002), if it is
generalised to include a trilingual lexicon.
The BIA+ model is graphically represented in Figure 1. To illustrate its func-
tioning, we will consider how a bilingual participant in the experiments by
Dijkstra et al. (1998) processes an interlingual homograph according to BIA+.

Figure 1 The BIA+ model for bilingual word recognition (Dijkstra & Van Heuven,
2002)
14 Bilingual Education and Bilingualism

Initially, both readings of an interlingual homograph such as ramp (meaning


disaster in Dutch) are activated. The speed of activation of each reading
depends on several factors, e.g. the subjective frequency of the word in the
language, whether it has recently been presented, the participants proficiency in
the words language, and so on.
In Experiment 1 by Dijkstra et al. (1998), participants were instructed to say
yes to English words and no to all other letter strings, including Dutch words.
However, in the stimulus list they encounter Dutch words only in the form of
interlingual homographs, which also have an English reading. Therefore, the
link between Dutch words and the no response will not be strong. Any (in fact,
unjustified) yes response to a recognised Dutch word reading will still lead to a
correct response, because the word in question will be an interlingual homo-
graph and will belong to English as well.
In Experiment 2, however, purely Dutch words appear in the stimulus list,
requiring a no response. Under these circumstances, the link between Dutch
words and the no response will be strengthened. This has a consequence for the
response to interlingual homographs, because a strong competition will arise
between the English reading of the item (associated with the yes response) and
its Dutch reading (associated with the no response). This response competition
will result in inhibition effects, i.e. slower RTs to the homographs than to their
matched one-language controls. To summarise, this account assumes a purely
bottom-up lexical activation process and a stimulusresponse binding mecha-
nism that is part of a task/decision system.
Finally, let us consider the consequences of the results presented for the
language activation metaphor described in the Introduction. Do these results
imply that the activation metaphor must be abandoned, because there is no such
thing as a relative activation of languages? Not at all! They merely show that one
must specify very clearly what is meant by the phrase relative activation of
languages, indicating how different factors affect the speed of activation and
recognition of words from different languages under different experimental
circumstances. The present papers main focus has been to show that there are
limits to the effects that participant expectations can exert on lexical activation
during the automatic multilingual word recognition process.
With these thoughts in mind, we may reconsider the two opening examples of
this paper as follows. It may well be that our use of a foreign countrys language
improves during our stay abroad because the recognition or use of specific
words leads to higher resting activation levels of those words, implying a faster
retrieval when they are encountered or repeated again. And it is not unlikely that
the problems of finding a word in our native language after reading the foreign
language equivalent may be due to a suppression of the native language equiva-
lent after recognition has been completed. Of course, these hypotheses remain to
be tested. Nevertheless, it seems to us that in cases like these and many others, the
language activation metaphor will prove to be most useful if it is applied in a
well-specified and testable way.

Acknowledgements
This paper has been presented by the first author at the Second International
Language Mode and Trilinguals 15

Conference on Third Language Acquisition and Trilingualism in Leeuwarden,


The Netherlands, 15 September 2001.

Correspondence
Any correspondence should be directed to Dr Ton Dijkstra, Katholieke
Universiteit Nijmegen, NICI, PO Box 1904, 6500 HE Nijmegen, The Netherlands
(dijkstra@nici.kun.nl).

References
De Groot, A.M.B. (1989) Representational aspects of word imageability and word
frequency as assessed through word association. Journal of Experimental Psychology:
Learning, Memory and Cognition 15, 824845.
De Moor, W. (1998) Visuele woordherkenning bij tweetalige personen [Visual word
recognition in bilinguals]. Unpublished Masters Thesis, University of Ghent.
Dewaele, J-M. (2001) Activation or inhibition? The interaction of L1, L2 and L3 on the
Language Mode continuum. In J. Cenoz, B. Hufeisen and U. Jessner (eds)
Cross-linguistic Influence in Third Language Acquisition: Psycholinguistic Perspectives (pp.
6989). Clevedon: Multilingual Matters.
Dijkstra, A., De Bruijn, E., Schriefers, H.J. and Ten Brinke, S. (2000a) More on interlingual
homograph recognition: Language intermixing versus explicitness of instruction.
Bilingualism: Language and Cognition 3, 6978.
Dijkstra, A., Grainger, J. and Van Heuven, W.J.B. (1999) Recognition of cognates and
interlingual homographs: The neglected role of phonology. Journal of Memory and
Language 41, 496518.
Dijkstra, A., Timmermans, M. and Schriefers, H.J. (2000b) Cross-language effects on
bilingual homograph recognition. Journal of Memory and Language 42, 445464.
Dijkstra, A. and Van Heuven, W.J.B. (1998) The BIA-model and bilingual word
recognition. In J. Grainger and A. Jacobs (eds) Localist Connectionist Approaches to
Human Cognition (pp. 189225). Hillsdale, NJ: Lawrence Erlbaum.
Dijkstra, A. and Van Heuven, W.J.B. (2002) The architecture of the bilingual word
recognition system: From identification to decision. Bilingualism: Language and
Cognition 5, 175197.
Dijkstra, A., Van Jaarsveld, H. and Ten Brinke, S. (1998) Interlingual homograph
recognition: Effects of task demands and language intermixing. Bilingualism: Language
and Cognition 1, 5166.
Green, D.W. (1998) Mental control of the bilingual lexico-semantic system. Bilingualism:
Language and Cognition 1, 6781.
Grosjean, F. (1985) The bilingual as a competent but specific speaker-hearer. Journal of
Multilingual and Multicultural Development 6, 467477.
Grosjean, F. (1997a) Processing mixed language: Issues, findings and models. In A.M.B.
De Groot and J. Kroll (eds) Tutorials in Bilingualism: Psycholinguistic Perspectives (pp.
225254). Hillsdale, NJ: Lawrence Erlbaum.
Grosjean, F. (1997b) The bilingual individual. Interpreting: International Journal of Research
and Practice in Interpreting 2, 163187.
Grosjean, F. (1998) Studying bilinguals: Methodological and conceptual issues.
Bilingualism: Language and Cognition 1, 131149.
Grosjean, F. (2001) The bilinguals language modes. In J.L. Nicol (ed.) One Mind, Two
Languages: Bilingual Language Processing (pp. 122). Oxford: Blackwell.
Lwy, N. and Grosjean, F. (in preparation) A computational model of bilingual lexical
access. Working paper, Neuchtel University.
Marian, V. and Spivey, M. (submitted) Competing activation in bilingual language
processing: Within- and between-language competition.
Postman, L. and Keppel, G. (1970)Norms of Word Association. New York: Academic Press.
Thomas, M.S.C. and Allport, A. (2000) Language switching costs in bilingual visual word
recognition. Journal of Memory and Language 43, 4466.
16 Bilingual Education and Bilingualism
Van Hell, J.G. (1998) Cross-language processing and bilingual memory organization.
Doctoral thesis, University of Amsterdam.
Van Hell, J.G. and Dijkstra, A. (2002) Foreign language knowledge can influence native
language performance in exclusively native contexts. Psychonomic Bulletin and Review 9
(4).
Van Hell, J.G. and De Groot, A.M.B. (1998) Conceptual representation in bilingual
memory: Effects of concreteness and cognate status in word association. Bilingualism:
Language and Cognition 1, 193211.
Van Heste, T. (1999) Visuele woordherkenning bij tweetaligen [Visual word recognition in
bilinguals]. Unpublished Masters Thesis, University of Leuven.
Van Heuven, W.J.B., Dijkstra, A. and Grainger, J. (1998) Orthographic neighborhood
effects in bilingual word recognition. Journal of Memory and Language 39, 458483.
Von Studnitz, R.E. and Green, D.W. (1997) Lexical decision and language switching.
International Journal of Bilingualism 1, 324.

Вам также может понравиться