Lexical access in bilingual speech production: Evidence from language switching in highly proficient bilinguals and L2 learners

https://doi.org/10.1016/j.jml.2004.02.002Get rights and content

Abstract

Five experiments are reported in which the picture naming performance of bilingual speakers in a language-switching task was explored. In Experiment 1, Spanish learners of Catalan and Korean learners of Spanish were asked to perform a switching task between their first and dominant language (L1, Spanish or Korean) and their second language (L2, Catalan or Spanish). For these two groups switching from the weaker language (L2) to the more dominant language (L1) was harder than vice versa. This asymmetrical switching cost was not present when highly proficient Spanish–Catalan bilinguals performed the task either in their two dominant languages (Experiments 2 and 3) or in their dominant language (L1) and in their much weaker language (L3 English; Experiment 4). Furthermore, highly proficient bilinguals showed faster naming latencies in their weaker languages (L2 and L3) than in their dominant language (L1). Experiment 5 tested whether a bias in the triggering of lexicalization is at the basis of such a difference. Together these results reveal that the switching performance of highly proficient bilinguals does not seem to be subject to the same mechanisms as that of L2 learners.

Introduction

One of the most remarkable abilities of bilingual speakers is that of separating their two languages during the production of speech. Although the speech of highly proficient bilinguals in their second language (L2) often carries traces of the first language (L1) (e.g., L1 accent, L1 syntactic structures; Flege, MacKay, & Meador, 1999; Pallier, Colomé, & Sebastián-Gallés, 2001; Yeni-Komshian, Flege, & Liu, 2000), it rarely exhibits L1 lexical intrusions (Poulisse, 1999; Poulisse & Bongaerts, 1994). That is, when necessary, these bilinguals are quite competent at selecting and producing words from only one of their lexicons, both in L1 and, more remarkably, in their L2. What are the mechanisms controlling lexical access in bilingual speech production? A clear understanding of how two languages are handled by one mind requires, among other things, that we investigate the mechanisms that allow bilingual speakers to restrict their lexicalization process to only one language. In this article, we address this issue by investigating the naming performance of several groups of bilingual speakers in a language switching task. The main objective of this research is to assess how L2 proficiency level affects the processes of lexical selection in speech production. We will do so by comparing the patterns of language switching performance of various groups of bilinguals that vary in their proficiency levels.

The Introduction is structured as follows. First, we discuss the general functional architecture and dynamics of the speech production system. Second, we introduce the hypothesis that lexical access in bilingual speakers involves inhibitory control, and discuss how the language switching task can provide useful information to test such a hypothesis. Before presenting the overview of the experiments, we put forward the hypothesis that the language switching performance may vary depending on the bilingual's proficiency levels.

A central stage in language production is that in which speakers retrieve the words from the lexicon that match their communicative intention. The process by which this is achieved is often referred to lexical selection (Caramazza, 1997; Dell, 1986; Levelt, 1989, Levelt, 2001; Levelt, Roelofs, & Meyer, 1999). A selection mechanism is needed because several lexical representations are activated due to spreading activation from the semantic system to the lexical level. Thus, any representation activated at the conceptual level spreads a proportion of its activation to its corresponding lexical node. In this scenario, the semantic system activates not only the word that matches the intended meaning but also other semantically related items. That is, when naming the picture of a dog, not only the lexical node “dog” is activated, but also other related words, such as “cat” and “bark.” The lexical selection mechanism is in charge of deciding which of the activated lexical items needs to be prioritized for further processing. It is widely accepted that the level of activation of lexical nodes is the critical variable for deciding which element is to be selected. Thus, in general, the lexical selection mechanism would pick out the word with the highest level of activation which, in normal error-free production, corresponds to the intended meaning. However, some models of lexical access assume that this mechanism is also sensitive to the level of activation of non-target—but activated—lexical nodes that act as competitors. That is, the ease with which a word is selected depends on its level of activation relative to that of other lexical items (e.g., Roelofs, 1992).

When implementing the lexical selection mechanism in the context of bilingual speech production, there are two questions that become relevant. First, does the semantic system activate the two lexicons of a bilingual? Second, do the lexical nodes of the non-response language (the language not intended for production) act as competitors? The first question has a positive answer, and present models of lexical access assume that during the course of lexicalization in one language (e.g., L2), the lexical nodes of both languages of a bilingual receive activation from the semantic system (e.g., Colomé, 2001; Costa, Caramazza, & Sebastián-Gallés, 2000; Costa, Colomé, Gomez, & Sebastián-Gallés, 2003; De Bot, 1992; Gollan & Kroll, 2001; Hermans, Bongaerts, de Bot, & Schreuder, 1998; Poulisse, 1999).

More controversial is the answer to the second question. Some models of lexical access assume that the lexical selection mechanism is language-specific (Costa, in press; Costa & Caramazza, 1999; Costa, Miozzo, & Caramazza, 1999; Roelofs, 1998), in the sense that it only considers the activation-levels of words in the intended language. According to this idea, lexical intrusions from the non-intended language would be prevented since those words would not be included in the pool of possible candidates for production, and therefore they will not be able to interfere during lexical access. The notion that lexical selection can be sensitive to specific properties of lexical nodes, and can use them to guide selection, has already been postulated in models of monolingual lexical access. For example, in Dell's model (Dell, 1986), the lexical selection mechanism is sensitive to the grammatical class of lexical items. That is, if the speaker wants to produce a noun the selection mechanism would consider for selection only lexical items corresponding to nouns (see Mahon, Costa, Shapiro, & Caramazza, 2002; Pechmann & Zerbst, 2002, for evidence consistent with such an assumption).

In contrast, other models of bilingual lexical access assume that the lexical selection mechanism is insensitive to the language in which the speaker intends to express her ideas. In such a framework, the speaker would consider for selection all activated lexical nodes, irrespective of the language to which they belong, and successful selection of the proper lexical node (i.e., in the correct language) is achieved by creating a differential level of activation in the two lexicons of a bilingual. Thus, the question here is: How does the system produce an imbalance of activation between the two lexicons? One way to do so is to assume that the semantic system activates words in the intended language to a larger extent than words in the non-response language (La Heij, in press; Poulisse & Bongaerts, 1994). The second solution is to postulate that lexical access in bilingual speakers entails the reactive inhibition of lexical items belonging to the non-response language (Green, 1986, Green, 1998; Hermans et al., 1998).

In the present study, we test the second of these implementations. We address whether inhibitory control, per se, can account for lexical access in highly proficient bilingual speakers. We do so by studying the properties of one experimental effect that has often been interpreted as supporting the notion that lexical access entails inhibitory processes; the asymmetrical switching cost reported by Meuter and Allport (1999) (see below).

The most articulated model of inhibitory control in bilingual speech production is that proposed by Green (1998). Among several assumptions made by this model (IC model—Inhibitory Control model), two are crucial for our present purposes. First, inhibition is reactive and proportional to the level of activation of the words that are to be suppressed. The term reactive means that inhibition is only applied after the lexical nodes of the non-response language are activated from the semantic system. Importantly, the amount of inhibition applied to one language depends on the speaker's proficiency level in that language. In other words, when speaking in L1, not much inhibition is required for the less dominant language (L2) because it is assumed that the baseline level of activation of L2 lexical items is lower than that of L1 lexical items. However, when speaking in the less dominant language (L2), L1 representations must be strongly inhibited in order to ensure that L2 lexical items are selected. The second assumption crucial for present purposes refers to the time required for inhibition to be overcome. It is assumed that the suppression (or inhibition) of the activation of a given language may exert an influence on subsequent production events in which words from the suppressed lexicon need to be retrieved (see Allport, Styles, & Hsieh, 1994). Accordingly, retrieving words from a lexicon that has just been inhibited would be relatively difficult, since it would take relatively more time for that inhibition to be overcome. Therefore, the more inhibition applied to a given lexicon, the harder it will be to overcome such suppression on a subsequent trial (see Tipper, 2001; for discussion of negative priming). This sort of model can be considered language non-specific, given that the computation responsible for lexical selection considers the activation levels of all lexical nodes, irrespective of the language to which they belong.

There is one set of results that has been argued to support the two assumptions made by the IC model. Meuter and Allport (1999) conducted a language switching experiment in which bilingual speakers were asked to name aloud series of lists containing Arabic digits (from 1 to 9) either in L1 or in L2. The language in which a given number had to be named was signaled by the background color of the screen (i.e., if blue name the digit in L1, if red name the digit in L2). Experimental trials were divided into switch and non-switch trials. A non-switch trial required a response in the same language as the immediately preceding trial, while a switch trial required a response in a different language. Given that responses were produced in both L1 and L2, there were four different types of trials: Switch to L1, Switch to L2, Non-Switch in L1, Non-Switch in L2. As expected, naming latencies in switch trials were slower than in non-switch trials, revealing that switching between tasks (name in language X, name in language Y) incurred a time cost. However, the magnitude of the switching cost was larger for L1 than for L2. That is, the switching costs were asymmetrical: to switch from L2 to L1 was more costly than vice versa (see also Janssen, 1999). The asymmetrical switching cost, which at first glance seems paradoxical, finds a ready explanation in the framework of the IC model and in the Task Set Inertia hypothesis (e.g., Allport et al., 1994; Allport & Wylie, 1999). As Meuter and Allport argued, the magnitude of inhibition applied to L1 when speaking in L2 must be greater than vice versa in order to allow successful selection of L2 lexical items (assumption 1 of the IC model). That is, the relative strength of the two tasks at hand (naming in L1 or in L2) would have an effect on the strength with which each of the two languages is inhibited. The authors further assumed that the inhibition exerted in one language has effects on the subsequent trial. As a consequence, to retrieve L1 words in a switch trial will be relatively hard because the system has to overcome the large inhibition that was applied to that lexicon in the immediately preceding trial (assumption 2 of the IC model). However, to retrieve L2 words in a switch trial will be relatively less difficult since the L2 lexicon was not as strongly suppressed in the preceding trial.

Evidence converging on the assumption that access to L2 representations involves the active suppression (or inhibition) of L1 was reported by Jackson, Swainson, Cunnington, and Jackson (2001) who recorded participants' event-related electrical potentials (ERPs) in a language switching task. Switch trials in L2 were associated with an enhanced negative deflection in the ERP at frontal sites compared to L2 non-switch trials. However, such a difference was not observed for L1 trials. Given that this component seems to index the presence of response suppression (e.g., Jackson, Jackson, & Roberts, 1999), the authors interpreted its asymmetrical presence in L2 as revealing the larger suppression of L1 in the switch trials (see also Hernandez, Martinez, & Kohnert, 2000; Hernandez, Dapretto, Mazziotta, & Bookheimer, 2001; Price, Green, & Von Studnitz, 1999; for brain imaging and electrophysiological studies in which language switching has been interpreted as revealing inhibitory processing). Further support for the notion that the asymmetrical switching cost reveals different degrees of language dominance comes from the study of Hernandez and Kohnert (1999, personal communication). In that study, a larger switching cost was observed in the dominant language (English), despite the fact that for those bilinguals (Spanish–English) Spanish was the earlier acquired language.

Note, however, that the above explanation of the existence of asymmetric language switching costs is a bit more complex. There are two components contributing to the existence of asymmetrical switching costs. The first is the time it takes for the speaker to implement the new task goal, or in Green's model, to select the right “language schema” (see also Allport & Wylie, 2000; Meiran, 2000; Ruthruff, Remington, & Johnston, 2001; Sohn & Anderson, 2001 for similar arguments in general task-switching models). This component is common for both directions of switching (from L1 to L2 and vice versa), and reflects the intrinsic time it takes for speakers to reset their experimental goal in switch trials (from naming in language X to naming in language Y). This switching cost is common to many other switching tasks, and is supposed to reflect processes outside the lexical system (e.g., Von Studnitz & Green, 2002). The second component refers to the inhibition applied to the lexical representations of the language that is not involved in a given naming trial. This component is assumed to be responsible for observing the asymmetrical switching costs in the language switching task. This is because in the switching task L1 lexical representations are more inhibited than L2 lexical representations. Thus, while the first component of the switching cost accounts for the presence of switching costs per se, the second component accounts for the presence of the asymmetrical cost.

Further post hoc analyses conducted by Meuter and Allport (1999) supported the notion that the asymmetrical switching costs stem from differences in the L2–L1 proficiency levels. In particular, the asymmetrical switching costs were somewhat dependent on the bilinguals' L2 proficiency: the higher the L2 proficiency, the smaller the asymmetry. Although, such differences among groups did not turn out to be significant, the trend was in the direction predicted by the IC model, and led the authors to speculate that the asymmetrical switching cost may disappear for highly proficient bilinguals. This is because for these bilinguals, the difference in proficiency levels between the two languages would be minimal, and therefore the amount of inhibition applied to both languages would be similar (see Monsell, Yeung, & Azuma, 2000; for a discussion of the conditions in which asymmetrical switching costs are observed).

In the present article, we adopt the explanation given by Meuter and Allport (1999) of asymmetrical switching costs and test whether such an explanation can account for the switching performance of highly proficient early bilinguals.

To contrast the performance of high and low proficient bilinguals is important, because L2 proficiency seems to be one of the most relevant factors for predicting bilingual speech performance. For example, proficiency correlates negatively with the number of lexical intrusions in speech production (Poulisse & Bongaerts, 1994). That is, highly proficient bilinguals suffer fewer L1 intrusions than low-proficient bilinguals when speaking in their L2. Also, proficiency seems to be one of the factors governing the representation of L2 in the brain (e.g., Perani et al., 1998). Thus, in the experiments to be reported we pay special attention to highly proficient bilinguals that have acquired their two languages early in life. The issue of whether L2 proficiency or L2 age of acquisition is at the basis of the performance of these speakers will not be directly addressed in our study, and we defer discussion of the relevance of these two variables until the General Discussion.

The differences between various types of bilingual speakers may be explained in terms of a difference in quantitative or qualitative terms. The difference may be revealing the different degrees of mastery with which the same lexical access mechanism is handled by different bilingual populations. For example, the fact that proficiency correlates negatively with language intrusions, may be accounted for by assuming that proficiency correlates with better inhibitory control. Also, such an effect could be revealing a better control of the amount of activation that the semantic system sends to the two lexicons. In other words, becoming a highly proficient bilingual would be a consequence of applying the same mechanisms available to less-proficient bilinguals in a more skilled way. However, a difference between high and low-proficient bilinguals could also be revealing a qualitative difference between the two populations. Thus, one may speculate on whether lexical selection in bilingual speakers entails different types of cognitive processes depending on their proficiency level. In other words, a difference in the performance of highly proficient bilinguals and less fluent bilinguals may stem not from “better” processing but from “different” processing.

Note that the aim of our research is not to investigate issues related to task switching per se, but rather to assess how experiments of this sort can inform us about the way in which bilingual speakers produce fluent speech (see Monsell, 2003, for a recent review of the task switching paradigm; and Meuter, in press, for a review on language switching). Nevertheless, this research can also inform us about the validity of the switching paradigm to study issues related to language processing in bilingual speakers, if it can be shown that the paradigm is sensitive to the linguistic particularities of the speakers performing the task.

The experiments presented in this article have two main goals: first, to replicate the observations of Meuter and Allport (1999); second, and more importantly, to test whether L2 proficiency affects the pattern of switching performance. Experiment 1 is devoted to the first objective while Experiments 2–5 to the second.

In Experiment 1, we investigate the switching performance of two different bilingual populations (Spanish–Catalan and Korean–Spanish). The speakers in this experiment could be considered to be L2 learners. In Experiment 2, we analyze the performance of highly proficient Spanish–Catalan bilinguals when switching between their two dominant languages (Spanish and Catalan). In Experiment 3, we replicate the observations made in Experiment 2 under somewhat different experimental conditions (e.g., larger set of experimental materials). Experiment 4 explores the switching performance of highly proficient Spanish–Catalan bilinguals when the task involves switching between their dominant language L1 (Spanish) and their much weaker language L3 (English). Finally, Experiment 5 extends the observations made in Experiments 2 and 3 to another experimental situation that allows participants more time to prepare for the language switch.

Despite the potential limitations of the specific properties of the design used by Meuter and Allport (1999) for drawing conclusions about the performance of bilingual speakers in other contexts (e.g., the large number of repetitions of items), and to be able to compare both studies, we keep our experimental design, in the relevant dimensions (e.g., % of switching trials, % of responses in L1 and L2, etc.) as similar as possible to theirs. There are, however, several important modifications. First, the populations tested in each of our experiments are homogeneous (regarding the languages of the speakers), while the population tested in Meuter and Allport's experiment was not (French–English, German–English, etc.). Second, the number of trials, and the number of repetitions of the experimental stimuli, is much smaller in our study (from 2000 trials to 950). Third, to avoid the use of phonologically similar translations (cognates) we used pictures of common objects rather than Arabic numerals as experimental stimuli.

Section snippets

Experiment 1: Replicating the asymmetrical switching costs in L2 learners

This experiment aims at replicating the asymmetrical switching costs reported by Meuter and Allport (1999) for low-proficient bilinguals. We do so by investigating the performance of two groups of speakers who have been exposed to their L2 relatively late in life and who can be considered L2 learners. The languages of the speakers in the two groups were different. In Group 1, we tested Spanish–Catalan participants and in Group 2, Korean–Spanish participants. Given the results of Meuter and

Experiment 2: Language switching costs in highly proficient bilinguals

The objective of this experiment is to assess whether a reduction in the difference between L1 and L2 proficiency levels leads to a reduction in the magnitude of the asymmetrical switching cost. If the interpretation of the assymetrical switching cost advanced by Meuter and Allport (1999) is correct, then we should expect a reduction (or even an elimination) of such an asymmetry when highly proficient Spanish–Catalan bilinguals perform the task in their L1 and L2. That is, to switch from L2 to

Experiment 3: Symmetrical switching costs in highly proficient bilinguals

In this experiment we further analyse the switching performance of highly proficient bilinguals. We do so by asking Spanish–Catalan highly proficient bilinguals to perform the same switching task as in the previous experiment. However, in this experiment we include 40 pictures rather than 10. This manipulation allows us to assess whether the pattern of results observed in Experiment 2 generalizes to other experimental contexts (more and different experimental items, and fewer repetitions).

Experiment 4: Highly proficient bilinguals switching between L1 and L3

The main goal of this experiment is to test whether highly proficient bilinguals show asymmetrical switching costs when asked to perform a switching task in their L1 and in their much weaker L3. We assessed this issue by testing highly proficient Spanish–Catalan bilinguals who were learning English (L3) in a language switching task. Crucially, the L2 (Catalan) proficiency level of these speakers is very different from their L3 (English) proficiency level. In addition, the participants tested in

Experiment 5: Is the L2 advantage a matter of lexicalization bias?

In this experiment, we assess whether the difference in naming latencies between the dominant and the non-dominant languages is due to a bias in the naming process that prioritizes the start of lexicalization in the non-dominant language. One property of the design used in the previous experiments may have been critical for establishing such a bias: the simultaneous presentation of the language cue and the target picture. Given this simultaneous presentation, participants may have started

General discussion

We reported five experiments in which the performance of different populations of bilingual speakers in a language switching task was tested. The main issue assessed was whether a difference in proficiency levels between the two languages involved in the switching task affects the switching performance of highly proficient bilinguals and L2 learners to the same extent. Given that the asymmetrical switching costs observed in this paradigm (switching to the more dominant language is more

Conclusion

The experiments reported in this article revealed striking differences in the language-switching performance of L2 learners and highly proficient bilinguals. While the former group showed a clear asymmetrical switching cost (e.g., switching to L1 being harder than switching to L2), the latter did not, even when the difference in proficiency levels between the two languages tested in the experiment was large. Furthermore, highly proficient bilinguals were faster when naming the pictures in their

Acknowledgements

This research was supported by two grants from the Spanish Government (BSO2001-3492-C04-01; BFF2002-10379-E), and by the McDonnell grant “Bridging Mind Brain and Behavior.” Albert Costa was supported by the research program “Ramon y Cajal” from the Spanish government. Mikel Santesteban was supported by a Pre-doctoral fellowship from the Basque government (PI-1999-18). Requests for reprints may be addressed to Albert Costa. The authors are grateful to Gabi Felhosi and Jin Nam Choi for their

References (67)

  • A Roelofs

    A spreading-activation theory of lemma retrieval in speaking

    Cognition

    (1992)
  • F.X Alario et al.

    Frequency effects in noun phrase production: Implications for models of lexical access

    Language and Cognitive Processes

    (2002)
  • A Allport et al.

    Shifting intentional set: Exploring the dynamic control of tasks

  • A Allport et al.

    Task-switching: Positive and negative priming of task-set

  • D.A Allport et al.

    `Task-switching,' stimulus-response bindings, and negative priming

  • Bialystok, E. (in press). Consequences of bilingualism for cognitive development. In J. F. Kroll, A. M. B. De Groot...
  • E Bialystok

    Cognitive complexity and attentional control in the bilingual mind

    Child Development

    (1999)
  • A Caramazza

    How many levels of processing are there in lexical access?

    Cognitive Neuropsychology

    (1997)
  • A Caramazza et al.

    The specific-word frequency effect: Implications for the representation of homophones

    Journal of Experimental Psychology: Learning, Memory and Cognition

    (2001)
  • J Cenoz

    The additive effect of bilingualism on third language acquisition: A review

    International Journal of Bilingualism

    (2003)
  • J Cenoz et al.

    Additive trilingualism: Evidence from the Basque Country

    Applied Psycholinguistics

    (1994)
  • Costa, A. (in press). Lexical access in bilingual production. In J. F. Kroll, A. M. B. De Groot (Eds.), Handbook of...
  • A Costa et al.

    Another look at cross-language competition in bilingual speech production: Lexical and phonological factors

    Bilingualism: Language and Cognition

    (2003)
  • A Costa et al.

    Is lexical selection in bilingual speech production language-specific?. Further evidence from Spanish–English and English–Spanish bilinguals

    Bilingualism: Language and Cognition

    (1999)
  • A Costa et al.

    The cognate facilitation effect: Implications for models of lexical access

    Journal of Experimental Psychology: Learning, Memory and Cognition

    (2000)
  • K De Bot

    A bilingual production model: Levelt's speaking model adapted

    Applied Linguistics

    (1992)
  • G.S Dell

    A spreading activation theory of retrieval in sentence production

    Psychological Review

    (1986)
  • J.E Flege et al.

    Native Italian speakers' perception and production of English vowels

    Journal of the Acoustical Society of America

    (1999)
  • T.H Gollan et al.

    Bilingual lexical access

  • D.W Green

    Mental control of the bilingual lexico-semantic system

    Bilingualism: Language and Cognition

    (1998)
  • F Grosjean

    The bilingual's language modes

  • D Hermans et al.

    Producing words in a foreign language: Can speakers prevent interference from their first language?

    Bilingualism: Language and Cognition

    (1998)
  • A.E Hernandez et al.

    Aging and language switching in bilinguals

    Aging, Neuropsychology, and Cognition

    (1999)
  • Cited by (684)

    View all citing articles on Scopus
    View full text