This art icle was downloaded by: [ Xavier Aparicio] On: 28 March 2013, At : 09: 36 Publisher: Rout ledge I nform a Lt d Regist ered in England and Wales Regist ered Num ber: 1072954 Regist ered office: Mort im er House, 37- 41 Mort im er St reet , London W1T 3JH, UK International Journal of Multilingualism Publicat ion det ails, including inst ruct ions for aut hors and subscript ion informat ion: ht t p:/ / www.t andfonline.com/ loi/ rmj m20 Recognising words in three languages: effects of language dominance and language switching a Xavier Aparicio & Jean-Marc Lavaur b a Laborat ory Vision Act ion Cognit ion – EAU01, Inst it ut e of Psychology, Universit y of Paris Descart es, 92100, BoulogneBillancourt , France b Laborat ory Epsylon, EA 4556, Universit y of Mont pellier 3, Place Albert 1er, 34000, Mont pellier, France Version of record first published: 28 Mar 2013. To cite this article: Xavier Aparicio & Jean-Marc Lavaur (2013): Recognising words in t hree languages: effect s of language dominance and language swit ching, Int ernat ional Journal of Mult ilingualism, DOI:10.1080/ 14790718.2013.783583 To link to this article: ht t p:/ / dx.doi.org/ 10.1080/ 14790718.2013.783583 PLEASE SCROLL DOWN FOR ARTI CLE Full t erm s and condit ions of use: ht t p: / / www.t andfonline.com / page/ t erm s- andcondit ions This art icle m ay be used for research, t eaching, and privat e st udy purposes. Any subst ant ial or syst em at ic reproduct ion, redist ribut ion, reselling, loan, sub- licensing, syst em at ic supply, or dist ribut ion in any form t o anyone is expressly forbidden. The publisher does not give any warrant y express or im plied or m ake any represent at ion t hat t he cont ent s will be com plet e or accurat e or up t o dat e. The accuracy of any inst ruct ions, form ulae, and drug doses should be independent ly verified wit h prim ary sources. The publisher shall not be liable for any loss, act ions, claim s, proceedings, dem and, or cost s or dam ages what soever or howsoever caused arising direct ly or indirect ly in connect ion wit h or arising out of t he use of t his m at erial. International Journal of Multilingualism, 2013 http://dx.doi.org/10.1080/14790718.2013.783583 Recognising words in three languages: effects of language dominance and language switching Xavier Aparicioa* and Jean-Marc Lavaurb a Laboratory Vision Action Cognition
EAU01, Institute of Psychology, University of Paris Descartes, 92100 Boulogne-Billancourt, France; bLaboratory Epsylon, EA 4556, University of Montpellier 3, Place Albert 1er, 34000 Montpellier, France Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 (Received 6 November 2012; final version received 20 February 2013) This study aims to examine language dominance and language switching effects in a series of monolingual and multilingual lexical decisions in which participants have to decide if the presented letter string is a word or not, regardless of language. Thirty participants (12 French
English bilinguals and 18 French
English
Spanish trilinguals) were recruited for two different experiments. In Experiment 1, 12 bilinguals processed two monolingual lists (L1 and L2) followed by a bilingual list (with L1 and L2 words). The results indicate faster answers and better accuracy for L1 words compared to L2 words in both lists, and a general slowdown of processing in the bilingual list, highlighting language switching effects (switch from L2 to L1 faster than from L1 to L2). In Experiment 2, 18 trilinguals performed successively monolingual (L1, L2 and L3), bilingual (L1L2, L1L3 and L2L3) and trilingual (L1L2L3) lists. The results show a slowdown of word processing as a function of number of languages involved, even for the dominant language (L1). Moreover, a language switching effect was found for all multilingual lists, in larger proportions when the list involves the two weakest languages (L2 and L3). Results are discussed in the light of interactive models of bilingual visual word processing. Keywords: trilinguals; visual word recognition; lexical decisions; switching cost; language dominance; language repetition; BIA-d Introduction In psycholinguistic research, it has been well demonstrated that the act of reading words, in any known language, involves cognitive processes in which the reader matches a representation of the incoming orthographic information of a given word with the words he/she knows, stored as abstract form representations in an entity called the mental lexicon (van Heuven, Dijkstra, & Grainger, 1998). The mental lexicon is familiarly compared to a huge database, containing not only the words known by a person, but also all their lexical characteristics, such as orthography, phonology, semantics and language information. Retrieving information about a word (i.e. the words identity and a useful understanding of the word) is also defined by researchers as lexical access (see Dijkstra, 2005). If lexical access appears to be an easy notion at first glance, complications arise as within a language one incoming word could activate several candidates before being identified. The situation reaches a higher level of complexity when lexical access involves multilinguals, as in this case *Corresponding author. Email: xavier.aparicio@parisdescartes.fr # 2013 Taylor & Francis Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 2 X. Aparicio and J.-M. Lavaur several lexical candidates from different languages could be activated by the input word (Dijkstra & van Hell, 2003; Lemhöfer, Dijkstra, & Michel, 2004; van Heuven et al., 1998). Lexical access is the crux of much research on multilingualism, and most of the researches agreed to consider the access to the mental lexicon as nonselective, leading to interferences between the languages known. Basically, it seems that all possible word candidates from the different languages are activated during word processing, independently of the language they belong to (Font & Lavaur, 2004; Lemhöfer et al., 2004; Orfanidou & Sumner, 2005). According to Midgley, Holcomb, and Grainger (2009), word recognition is modulated by bilinguals’ proficiency, depending on frequency of language use and Age of Acquisition (AoA). In case of late learners of an L2, proficiency is often asymmetrical, ranging from the L1 to the last language acquired. The hypothesis of an integrated lexicon also raises the question of language dominance (the mother tongue in most cases, or the L2 if it reaches a higher level of proficiency) and language switching, and the eventual costs elicited by switching from one language to another. Word recognition in multilinguals According to Dijkstra and van Hell (2003), multilinguals are most often led to speak or read in one given language, and to do so, it is necessary to reduce interference with their other languages. The capacity of control by multilinguals on the relative activation of their different lexicons seems to be quite limited (Christoffels, Firk, & Schiller, 2007). Nevertheless, for people able to speak several languages, it is sometimes useful within a conversation to use words belonging to a language different from the one being used (Meuter, 2005; Moreno, Federmeier, & Kutas, 2002). Moreno et al. (2002) suggest that the unintended intrusion of a word from a language different from the one initially spoken is often due to a problem of retrieval of the word in the initial language, leading to the substitution of a homologue (translation equivalent) from another known language, in order to retain the meaning. For the person detecting the language switch, it is difficult to process a word from the new language, primarily because the word is unexpected and the language used is irrelevant in the communication context. In a recent study, Aparicio et al. (2012) have highlighted N400 differences between the three languages known by a trilingual during a semantic categorization task involving words in three languages. Participants were asked to read silently the presented words and to press a key on a remote control if the word was the name of an animal (e.g. dog). Words were presented in a mixed list including French, English and Spanish words, allowing the authors to study language switching effects. Their results indicate that the native language (L1) possesses a very specific status within the brain compared to L2 and L3, meanwhile the larger amplitude of the N400 for L3 words compared to L2 words is supposed to reflect the difference of proficiency between these later acquired languages. According to these findings, we assume that language switching costs depend on the number of languages involved in the switch, and more precisely that language switching will be influenced by the relative dominance of one given language upon the other languages known. Language switching costs Regardless of language proficiency in each language, language switching ability is not constant and relies on individual ability to switch in a given context (reading, Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 International Journal of Multilingualism 3 speaking). Many researches (Chauncey, Grainger, & Holcomb, 2008; Christoffels et al., 2007; Grainger & Beauvillain, 1987; Macnamara, Krauthammer, & Bolgar, 1968; Thomas & Allport, 2000) focused on the eventuality of a cognitive cost elicited by code switching. The researchers argue in favour of a switch cost; however, a debate still exists to determine whether the cost is due to a comprehensive mechanism of task control, or to specific processes of language control in multilinguals (Chauncey et al., 2008). According to the developemental version of the Bilingual Interactive Activation (BIA) model (an extension of BIA model also based on Revised Hierarchical Model (RHM) model, see Grainger, Midgley, & Holcomb, 2010), the cost of language switching is attributed to a mechanism, a language node, controlling the relative activation of each lexical representation per language. Language nodes in the BIA worked as a control mechanism of attention and of the task proposed in models of cognitive control (Chauncey et al., 2008). According to Chauncey et al. (2008), when learning a second language, bilinguals may develop a general cognitive control mechanism permitting to supervise the activation of representations of each language. The BIA-d postulates that the processing cost of a language switch is the result of descending inhibition from the inappropriate language node to the lexical representations of the target language. Grainger and Beauvillain (1987; see Thomas & Allport, 2000 for similar findings) observed switch costs elicited by language switching in a generalised lexical decision task. In this study, participants responded to items in two monolingual lists, in which items from one language were presented, and in a bilingual list in which French and English items were mixed. In this task, the language to which the stimulus belongs to was supposed to have no consequences on the required response. Nevertheless, the authors demonstrated that this information was not ignored. In the bilingual list, the response time for the lexical decision for a given language was slower when the preceding trial belongs to the other language. Thus, it seems easier to switch from the dominant language to the non-dominant language than the contrary. The present study The aim of this research is the study of language dominance and language switching effects in trilinguals. We based our framework on the one used by Grainger and Beauvillain (1987) but adding one language, using several lexical decision tasks with bilinguals and trilinguals. We assume that the addition of a third language will allow us to better understand the different connections between the languages, given that most bilinguals speak a third language. It will also help us to determine the nature of the links (lexical or conceptual) bounding languages within the mental lexicon. In this purpose, we recruited 30 participants, 12 French
English bilinguals and 18 French
English
Spanish trilinguals. Our first aim is to estimate the dominance effect between different monolingual lists (constituting a baseline for each language in terms of lexical access) and in multilingual lists including two or three languages. According to the findings of Grainger and Beauvillain (1987), the response time in a bilingual list (two languages involved) is slower than in monolingual list (one language involved) due to language switching. All languages are expected to be slowed down in multilingual conditions, but dominance effect is supposed to remain, given that language competition would be in favour of the dominant language. Nevertheless, some attenuation of dominance could occur due to the difficulty of processing two or three languages simultaneously. Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 4 X. Aparicio and J.-M. Lavaur The general prediction is that monolingual lexical decision would be faster than multilingual lexical decision for all languages. Based on previous findings, we assumed that switch costs are higher for non-dominant languages (L2 and L3). In addition, we expect word recognition to be slower as a function of number of languages involved. We predict that L1 would have a higher level of activation compared to the two other languages. Moreover, given that participants’ L3 was learned consecutively to L2, L2 is predicted to be activated more easily than L3 as participants are expected to be more proficient in L2. In the bilingual lists, we aim to determine the cost of language switching in all combinations of languages, and we expect switching between L2 and L3 to be slower than other switch directions. Indeed, the route between these languages is rarely used, and probably required a necessary activation of L1 nodes. Consequently, we assume that in trilingual memory, L1
L2 and L1
L3 connections are often solicited in purpose of communication, but L2
L3 links are supposed to be weaker, which may result in a higher switch cost. In trilingual lists, we expect larger switch effects compared to bilingual conditions, because of many combinations of switches and unusual language changes. Moreover, in multilingual lists, we will compare repetition (target preceded by a word in the same language) and switching trials (target preceded by a word in a different language). Experiment 1: lexical decisions in two languages In the first experiment, we compare the processing of words specific to one language (no orthographic overlap between translation equivalents in the three languages). We select French
English bilinguals, with a high level of competence in L2, and record their answers in a series of lexical decisions in which words are presented in one language only (monolingual list), or two languages (bilingual list). Our objective is to evaluate the effects related to language switching when two languages are presented within the same list of stimuli (Grainger & Beauvillain, 1987), in order to ultimately compare the results to those obtained when a third language is introduced. Participants performed several lexical decisions involving one or both languages known by the participants. When a word belonging to L2 is presented consecutively to L1 trial, we assumed that L1 lexicon has to be inhibited (strong inhibition because L1 processing is automatic) and the inhibition previously settled on L2 has to be overcome. This switch is expected to require more time, because L1 has been strongly inhibited, and consequently it would be more difficult to overcome the inhibition. In language repetition (L1 word following a L1 word, and L2 word following a L2 word), the BIA-d predicts that the word would be identified faster, because the corresponding language node has been already activated by the preceding item, and its degree of activation is really high. Methods Participants Twelve French
English bilinguals (mean age23.9), students of English at the University of Montpellier 3 (France) were recruited. All were native speakers of French, and spoke English as a second language. They had to fill in a questionnaire to determine their skills in both languages (see Table 1). After the experiment, they International Journal of Multilingualism 5 also had to perform a post-test translating French words into English to ensure that they had a sufficient knowledge of the words used in the experiment. Table 1 shows answers to the linguistic skills questionnaire for L1 and L2. The results indicate that participants consider themselves more proficient in L1 compared to L2 t(11) 8.95, pB0.001. Moreover, a t-test shows that performance on post-test ranged from 77% to 100% correctness in English with a mean of 89% (SD 0.15). Table 1. Self-evaluation (SD in parenthesis) of linguistic skills of the participants of Experiment 1 in French (L1) and English (L2) on a 7-point Likert scale and AoA of English (in years). Languages Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 Language skills Reading Oral understanding Speaking AoA (years) LI
French 7.0 (0.0) 7.0 (0.0) 7.0 (0.0)
L2
English 6.8 6.0 6.1 9.9 (1.2) (0.8) (1.0) (1.1) Stimuli All words were non-cognates, with no orthographic overlap with their translation equivalent in the other language. Seventy words were selected per language, with length in between three and eight letters, a mean frequency in occurrences per million (OPM) of 89 (SD23) for French (Lexique 3, New, Pallier, Brysbaert, & Ferrand, 2004) and 93 (SD26) for English (CELEX, Baayen, Piepenbrock, & Gulikers, 1995). Seventy pseudo-words were created for each language (total 140) with the WordGen software (Duyck, Desmet, Verbeke, & Brysbaert, 2004), in order to match in terms of length, and be pronounceable letter strings. Controls have been made to ensure that words and pseudo-words had few orthographic neighbours and were not interlingual homographs. Apparatus and procedure Participants were seated in a sound attenuated room in front of a computer screen. For each trial, they had to decide if the presented letter string was a word or not, regardless of language (either L1 or L2), and manually press a response button on the keyboard. Stimuli were displayed on the centre of the screen with E-Prime software (Schneider, Eschman, & Zuccolotto, 2002), in white ink on a black background. An example of a trial sequence is provided in Figure 1. In the first part of the experiment, the participants were presented with two monolingual lists sequentially in a counterbalanced order (L1 then L2, or L2 then L1). The bilingual list (two languages) was performed after two monolingual lists. The stimuli were presented in a pseudo-randomised order, to get as much repetition as switched trials, and making the next trial unpredictable (in terms of language belonging). Results The data analysis was performed for errors and reaction times (RTs) on correct answers. RTs less than 200 ms and greater than 1500 ms (3% of the data) were 6 X. Aparicio and J.-M. Lavaur Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 excluded. Analyses of variances (ANOVAs) were performed to compare the results for each language in different experimental conditions. Figure 1. Presentation of an experimental sequence. The French word VOITURE (car) is followed by an ISI, then by a fixation point preceding the next English trial HOUSE. List context effects (number of languages) In this section, we compared performance to the same set of words in each language (L1 and L2) as a function of the number of languages in the list (1 or 2). ANOVAs were performed on the RT and percentage error data, with target language (L1 and L2) and list context (1 or 2 languages) as within subject factors. In the RT analysis, there was a main effect of target language F(1,11) 16.12, pB0.01, with French words processed faster (509.1 ms, SD53.8) than English words (569.8 ms, SD51.1), regardless of list context. Moreover, the analysis also revealed a main effect of list context, F(1,11) 5.32, p B0.05, indicating that the introduction of another language slows down the processing of words. Nevertheless, the interaction between these two factors was not significant (F B1), reflecting the fact that list context effect was quite similar for both languages. Indeed, the cumulative effect of the number of languages was about 36 ms for L1 words F(1,11) 4.18, p B0.05, and 30 ms for L2 words F(1,11) 4.68, pB0.05 (see Figure 2). In the error analysis, there was no main effect of language (FB1), but a significant main effect of list context F(1,11) 5.30, pB0.05, indicating that the presence of two languages in the list elicits more errors. The interaction between these two factors was not significant (F B1). Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 International Journal of Multilingualism Figure 2. 7 Mean RT (ms) for words in L1 and L2 in pure and generalised lexical decisions. Within-list language switching effects In these analyses, we compared performance to words in both languages (L1, L2) presented in the bilingual list as a function of the language of the preceding trial. ANOVAs were performed on the RT and error rates data with target language (L1, L2) and language switch (switch vs. repetition trials) as main factors. Additional analyses compared performance as a function of the direction of the switch (L1 preceded by L2, or L2 preceded by L1). In the RT analysis, there was a main effect of target language F(1,11)11.91, pB 0.01, indicating that French words (527 ms) were processed faster than English words (587 ms). Moreover, we found a main effect of language switch F(1,11)10.51, p B0.01, reflecting that words preceded by a trial in the same language were processed faster (language repetition 536 ms) compared to targets preceded by a trial in the other language (language switch, 574 ms). Here again, the interaction between these two factors was not significant (FB1), indicating that switch effect is quite similar for both languages. This could be explained by the fact that for L1 targets, the difference between repetition and switching condition was not significant F(1,11) 2.26, p1. Nevertheless, we obtain a significant effect F(1,11) 4.58, p B0.05 when comparing L2 targets in repetition (562.8 ms) and switching (607.1 ms) conditions. Interestingly, when focusing on language switching direction (L1L2 or L2L1), target words were processed more slowly in L1L2 direction compared to L2L1 direction F(1,11)  25.54, p B0.001, which highlights an asymmetry between languages (Grainger & Beauvillain, 1987; Kroll & Stewart, 1994; Von Studnitz & Green, 1997). As we can see in Figure 3, L2 words were processed significantly slower than L1 targets, regardless of the preceding trial. It also seems that language switching cost is greater towards the non-dominant language, i.e. L1L2. Discussion The aim of this experiment was to study language switching effects, knowing that they rely on the degree of proficiency in both languages. The results obtained are Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 8 X. Aparicio and J.-M. Lavaur Figure 3. Mean RT (ms) for French and English target words depending on the repetition or switching condition in the generalised lexical decision. similar to those from Grainger and Beauvillain (1987) reflecting that cost combined with language switching is demonstrated for all languages, but larger when switching from the dominant language to the non-dominant language. This cost seems modulated by the level of activation of the different languages, as predicted by the BIA-d model. Language dominance effects Both monolingual lexical decisions (L1 and L2) clearly show language dominance effects, with L1 words processed faster than L2 words. This dominance of L1 is also demonstrated in the bilingual lexical decision (L1
L2), with somehow a slowdown of processing in both languages compared to monolingual lists. This slowdown could be explained by a lateral inhibition of units from the inadequate lexicon (Grainger & Beauvillain, 1987; van Heuven et al., 1998). According to the BIA-d (Grainger et al., 2010), when a candidate from a language is activated, the phenomenon of lateral inhibition is applied to words belonging to the other languages spoken, reducing interferences between the lexicons. Switching from L1 to L2 supposes a deactivation of the L1 lexicon in order to activate the L2 lexicon, which represents a more complex processing. Language switching effects The consecutive presentation of words belonging to different languages constraints participants to process several language switching possibilities, from L1 to L2 and L2 to L1. The observed switch cost is 30 ms in L2L1 direction and 45 ms in L1L2 direction, indicating that the switch cost is larger from the dominant language to the weaker one. These results are in line with previous findings in the literature (Grainger & Beauvillain, 1987; Von Studnitz & Green, 1997). The processing of the trial preceding the language switch supposed a competition between the lexical units activated, and consequently it requires the intervention of an inhibition mechanism. The presentation of a L1 word is supposed to activate more representations compared to L2, and requires a stronger inhibition which will slow down the International Journal of Multilingualism 9 activation of the last presented L2 word. It could also explain the asymmetry observed during language switching. Our objective in the second experiment is to determine to what extent these language switching effects or more specifically the cost associated to language switch relies on the asymmetry of the lexical units in three languages. Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 Experiment 2: lexical decisions in three languages Experiment 2 is settled to examine language switching effects when the three languages known by the participants are involved. As in Experiment 1, target words are presented in a monolingual, bilingual and also a trilingual condition with words from the three languages. Here, we assume that if similar differences in processing are highlighted for the same words in these conditions, this could mean that the activation of the three lexicons simultaneously requires a higher degree of control in order to activate the relevant lexicon and inhibit non-target languages. Specifically for the trilingual list, we assume that interferences between these languages are more important compared to the other conditions. Because of language switching trials in the multilingual lexical decision, we expect (as reflected in the first experiment results) an overall slowdown of processing modulated by the number of languages involved in the task, keeping in mind that the modulation of the effects could be due to the dominance (or non-dominance) of the language concerned. These effects could also be modulated by the succession of activation-inhibition mechanisms involved during language switching processing. First, the expected effect of language dominance should be observed for the three monolingual lists (in French, English and Spanish, respectively), because the recognition of the shape of the word depends on the degree of proficiency in the presented language. This result is in line with the patterns of results obtained by Aparicio et al. (2012) with trilinguals, and corresponds to the activation of the N400 component. Moreover, according to Experiment 1, this effect could be generalised to all the languages involved, in monolingual and multilingual lists. Even if we expect a general slowdown of processing due to language switching, words belonging to the dominant language (French L1) should be processed faster than in the other languages. Indeed, the degree of activation of shapes and semantics is linked with the degree of proficiency in the different languages, at least at the start of the processing. Concerning L2 (English) and L3 (Spanish), we assumed the degree of activation of their lexical units to be weaker, and the control exerted on language switching to be harder to apply, with nevertheless an advantage of L2 above L3. Here, paired comparison between languages should give us more information on the dominance and the organisation of languages in the memory (Font & Lavaur, 2004; Thomas & Allport, 2000). Finally, by considering the slowdown observed in Experiment 1 to be due to the presence of two languages, we would be likely to observe an increase of these effects when participants have to process consecutively words from three languages, because of the complexity of language combination. Indeed, some language switches could be used less than others (i.e. from L2 to L3 and vice versa), depending on the context of use. We assumed the effects relying on language switching to be modulated by the number of languages involved in the lexical decision. More specifically, language switching should be modulated on one hand by the direction of the switch (e.g. L1 to L3 vs. L3 to L1), and on the other hand by the proficiency of the participants in the three languages. According to the Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 10 X. Aparicio and J.-M. Lavaur BIA-d model (Grainger et al., 2010), associated with previous findings, the cost associated to language switching is higher from the dominant language to a nondominant language (see Experiment 1). The study of the links between the second L2 and the third L3 languages has been barely submitted to experimentation so far. Switching between these languages could be a problem if we consider its small probability of occurrence and requires conceptual mediation by the L1. According to the BIA-d, for these two nondominant languages, we should observe results similar to Experiment 1 for L1 and L2, that is to say a slower processing of words from a non-dominant language, and a larger cost in L1L2 direction. Switching involving L1 and L3 should highlight similar patterns of results, with nevertheless longer latency for L3 words, given that participants have learned this language later in life and thus, they are supposed to have a lower degree of control of this language. Finally, concerning the two nondominant languages (L2 and L3), we expect a global slowdown of processing because these two languages are acquired later in life compared to L1. Considering the results of Experiment 1, we may observe a cost associated to language switching larger in L2L3 direction compared to L3L2 direction, if a kind of hierarchy is reflected also on the level of activation of the two non-dominant languages. Method The experiment included a series of seven lexical decisions involving the three languages known by our participants. First, participants were presented with three monolingual lists with words and pseudo-words matched per language (respectively L1, L2 and L3). In a second part they were shown three bilingual lists including words from two known languages and pseudo-words matched these languages (respectively L1L2, L1L3 and L2L3); finally, a trilingual list including all the languages known by our participants and pseudo-words matched from all three languages (L1L2L3) was presented. Participants Eighteen students of Foreign Language Studies at the University of Montpellier 3 (France) were recruited, native speakers of French, speaking English as a second language and Spanish as a third one (mean age24.3). They filled in the same questionnaire and post-test as in Experiment 1. Main results of the questionnaire are presented in Table 2. A t-test shows that performance on post-test ranged from 72% to 100% correctness in English with a mean of 86% (SD 0.11), and from 59% to Table 2. Results of the linguistic abilities questionnaire and AoA in the three languages (French L1, English L2 and Spanish L3) on a 7-point Likert scale (SD in parenthesis). Languages Language skills Reading Oral understanding Speaking AoA (years) LI
French 6.9 (0.1) 7.0 (0.0) 7.0 (0.0)
L2
English 6.2 6.0 6.0 9.8 (0.2) (0.5) (0.4) (1.2) L3
Spanish 5.8 5.2 5.1 13.6 (0.9) (0.6) (0.4) (1.0) International Journal of Multilingualism 11 93% correctness in Spanish with a mean of 77% (SD 0.15). Performance was significantly better in English than in Spanish (t (17) 4.3; pB0.05.) Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 Stimuli For each known language, 140 non-cognate words were selected from a multilingual lexical database (Laxén, Aparicio, & Lavaur, 2008). Monolingual lists in L1 and L2, as well as the bilingual list L1
L2 were identical to Experiment 1, but additional items were selected in order to have an equal number of words in each list. Every list is made up of 60 words and 60 non-words. In all the lists (monolingual and multilingual), there are as many words as non-words. For instance, there were 60 French words in the pure list, 30 French words in mixed lists in two languages and 20 French words in the list involving the three languages. The words from the other languages were also shared out this way. All pseudo-words were pronounceable strings of letters in their respective languages. Apparatus and procedure Apparatus was similar to Experiment 1. Procedure was slightly modified to avoid a bias of list orders, and the experiment was divided in three sessions. In the first part, three pure lists were established, one in each language. All participants were presented with three monolingual lists, in various order. In the second part, participants responded words and pseudo-words from two languages. In these bilingual lists, items were presented in a random order in a way that the language of the next item was unpredictable. Again, three lists were established: French
English words and pseudo-words (L1L2), French
Spanish words and pseudo-words (L1 L3), and English
Spanish words and pseudo-words (L2L3). The order of presentation of these three lists was counterbalanced across participants. The number of switches in each direction was the same as the number of repetitions. Finally, participants performed a trilingual list with words and pseudo-words from the three languages (L1L2L3). The items were randomly mixed to form a list. The number of switches in each direction was the same as the number of repetitions. The presentation of words and pseudo-words was controlled to ensure that a given item was presented only once to a given participant in the whole experiment. Results The data analysis was performed for errors and RTs for correct answers. RTs less than 200 ms and greater than 1500 ms (4% of the data) were excluded. ANOVAs were performed to compare the results for each language in the different experimental conditions. List context effects (number of languages) Here we compared performance to the same set of words in each of the three languages (L1
L3) as a function of the number of languages in the list (1
3 languages). ANOVAs were performed on the RT and percentage error data with target language (L1
L3) and list context (1, 2, or 3 languages) as main factors. Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 12 X. Aparicio and J.-M. Lavaur In the analysis, there was a main effect of target language for RTs F(2,34) 24.93, pB0.001 and errors F(2,34) 19.24, pB0.001. French words were processed faster than English words F(1,17) 9.86, pB0.001 and elicited less errors F(1,17) 5.42, pB0.01, and also than Spanish words for RTs F(1,17) 10.40, pB0.001 and errors F(1,17) 7.24, p B0.01. Concerning the two non-dominant languages, L2 words were processed faster than L3 words F(1,17) 54.27, pB0.001. The analyses also highlight a main effect of list context for RTs only F(2,16) 3.13, pB0.05, indicating that monolingual lists were processed faster than bilingual F(1,17) 4.58, pB0.05 and trilingual lists F(1,17) 6.34, p B0.05. Details are provided in Figure 4. There was also a significant interaction between list context and language, F(2,16) 3.40, p B0.05, with different list context effects for words in L1, L2, or L3. For L1 words, there was a cumulative effect of the numbers of languages, with longer RTs in the bilingual lists (L1
L2, L1
L3) F(2,34) 8.20, p B0.01, and even longer RTs in the trilingual list F(2,34) 4.10, pB0.05 (see Figure 4), without eliciting more errors (F B1). Moreover, words in bilingual lists were processed faster than those in the trilingual list F(2,34) 8.90, p B0.01. For L2 words, the results follow exactly the same pattern: L2 target words were processed faster in monolingual compared to bilingual F(2,34) 13.30, pB0.01 and trilingual lists F(2,34) 9.8, p B0.01, but not significantly for errors. Comparison between bilingual and trilingual lists show that L2 targets were processed faster F(2,34) 6.71, p B0.01 in the bilingual lists. Nevertheless, these patterns were not found for Spanish. We can observe differences suggesting a slowdown of processing for L3 targets depending on the list context, but the difference associated with this slowdown was not statistically significant for both RTs and ERs. Within-list language switching effects In these analyses, we compared performance to words in the three languages (L1
L3) presented in multilingual lists as a function of the language of the word on the directly preceding trial. ANAVOs were performed on the RT and percentage error Figure 4. Mean RTs (ms) and SDs per language (L1, L2, L3) for pure and generalised lexical decisions. Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 International Journal of Multilingualism 13 data with target language (L1
L3), language switch (switch vs. no-switch trials) and list context (2 vs. 3 languages). Additional analyses compared performance on switch trials as a function of the language of the word on the preceding trial. In the RT analysis, there was a main effect of target language, F(2,34) 61.9, pB 0.001, with French words processed faster than English F(1,17) 49.5, pB0.001 and Spanish words F(1,17) 74.43, pB0.001. There was also a significant difference between L2 and L3, F(1,17) 44.1 pB0.001, with English words processed faster than Spanish words. Moreover, there was a main effect of language switch, F(1,17)  16.31 p B0.001, with language repetition trials processed faster than language switch trials. The interaction between these two factors was not significant (F B1), nevertheless we developed paired comparison analysis to determine if this absence of effect could be due to specific combination of languages. For L1 targets, switch costs did not differ as a function of the language of the preceding item on switch trials. In the lexical decision L1
L2, the processing of L1 words in repetition (500 ms) was slowed down of 27.8 ms when they were preceded by English words (527.8 ms). In L1
L3 lexical decision we observed the same pattern of results, with L1 words in repetition (509.4 ms) processed slower (22.1 ms) when preceded by Spanish words (531.5 ms, cf. Table 3). Here, the analysis of switch cost for L1 targets when preceded by L2 and L3 shows no significant differences (FB1). Interestingly, in the trilingual list, the switch cost associated with L1 targets preceded by L2 was 10.9 ms, and 24.3 ms for L1 targets preceded by L3, with again no significant differences. For English (L2) words, in the bilingual lists, the switch cost associated with L2 targets when preceded by L1 was 22.1 ms, and 32.8 ms when preceded by L3 words, which was statistically significant F(1,17) 10.24, pB0.01. In the trilingual list, the switch cost associated with L2 targets was also larger when preceded by L3 targets (13.6 ms) than for those preceded by L1 targets (
27.7 ms), which is also significant F(1,17) 7.05, p B0.01. Finally, for L3 targets, in the bilingual lists, the switch cost associated with L3 words was significantly larger F(1,17)17.56, pB0.001 when preceded by L2 (50.2 ms) than preceded by L1 (13.1 ms). In the trilingual list, we observed a reverse effect, with a larger cost when preceded by L1 (76.1) compared to L2 (56.1), but the difference was not significant (FB1). Table 3. RTs (ms) for all languages in generalised lexical decision with two and three languages, depending on the preceding trial. Language switch Lists L1
L2 L1
L3 L2
L3 L1
L2
L3 Language repetition Preceding language Targets L1 Targets L2 Targets L3 L1 L2 L1 L3 L2 L3 L1 L2 L3
527.8 (45.4)
531.5 (64.2)


529.8 (81.1) 543.2 (86.8) 606.5 (75.3)



567.7 (79.8) 561.1 (85.6)
602.4 (68,9.2)

641.5 (76.5)
650.7 (60.1)
679.4 (91.9) 659.4 (74.6)
500 584.4 509.4 628.4 534.9 600.5 518.9 588.8 603.3 (50.1) (75.0) (56.5) (66.4) (44.1) (79.1) (68.4) (90.6) (90.7) 14 X. Aparicio and J.-M. Lavaur Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 Discussion The aim of this experiment was to determine language switching effects depending on the number of languages involved in the experimental list and the direction of the switch. First, the monolingual lists (L1
L3) clearly show effects related to language dominance, which are in line with previous findings of Aparicio et al. (2012) in a trilingual study recording event-related potentials (ERP). Here, L1 words were generally processed faster than L2 and L3 words, and L2 faster than L3, indicating a hierarchy between the three languages which could be related to the AoA, and also to the size of the different lexicons. This effect of language dominance persists in all the multilingual lists (with two or three languages). It is important to notice that the processing of any word (regardless of language) has been slowed down in the trilingual list, certainly because of the multiplicity of language combinations increasing the difficulty to perform the task. However, regardless of the condition, L1 (French) seems to benefit from the highest level of activation, even if it suffers from the presence of other languages in the list. This task requires the setting of greater mechanisms of control and inhibition compare to pure or two-language lists. These results are fully compatible with BIA-d model (Grainger et al., 2010), which predicts such modulations in terms of activation of languages. It is interesting to notice that the slowdown of processing due to the introduction of other languages in a list concerns every language, with a variable intensity. This could be related to the introduction of language switches (Grainger & Beauvillain, 1987). The slowdown is more important for all languages in the trilingual list, indicating a progressive increase of language processing, related to the number of languages to be processed. This could be explained by the numerous language switches in the different directions when the three languages are involved. Analyses on language switching effects revealed that the processing of languages is affected by the language switch, but with really small effects for L1. In L1L2 lexical decision, we found results similar to Experiment 1, with L1 words processed faster in repetition (L1L1). Moreover, language switch is quite similar in L2L1 (27.8 ms) direction compared to L1L2 (22.1 ms). Similar findings have been highlighted by Grainger and Beauvillain (1987). These effects of switching could be related to inhibitory control of lexical activation proposed in the BIA-d. The presentation of a word in a given language leads to an overall inhibition of all the other languages. Because inhibition needs to be overcome (Aparicio, Heidlmayr, & Isel, in press), processing of following trials will be more difficult after a language switching (Grainger & Beauvillain, 1987). In L1L3 lexical decision, there is a clear asymmetry in language switching direction. Language switching cost is more important in L1L3 direction compare to L3L1 direction. The results highlighted in the L2
L3 lexical decision also reflect these effects, with a higher cost in L2L3 direction, indicating that the same mechanism of inhibitory control is activated to control language activation. Results revealed that for each lexical decision the dominant language (L1 or L2 depending on the decision) is sensitive to repetition, which facilitates decision. Moreover, switch cost always seems smaller when switching from the weakest language to the strongest one. This could be due to the path followed to activate the lexical representation (lexical or conceptual, Kroll & Stewart, 1994), or to a mechanism of inhibitory control of lexical activation. Because L1 is automatic and strongly activated in unbalanced bilinguals (Meuter, 2005), it needs to be strongly inhibited, which could allow a more important cost International Journal of Multilingualism 15 observed in this direction of switch, and confirmed with results in L3. Finally, the results in the trilingual list highlight similar effects, i.e. language repetition effects (more important in L1) and language switching effects, which seem less important when switching from a non-dominant language to a dominant language. Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 General discussion In these experiments, we examined to what extent the language switch is influenced by the degree of mastering of a given language, as well as the direction of language switching. In other terms, we intend to determine if the access to lexical units relies only on language context, or if the processes of access could be modulated by the languages known by bilinguals and trilinguals. To reach these objectives, we choose to compare the visual processing of words in several linguistic contexts (monolingual, bilingual, and trilingual). In both experiments, we observe differences between the linguistic contexts, as well as language switching effects. On the whole, the strength of the results is supported by several aspects of the research. Indeed, we compared different language combinations (two or three languages in only one list), with monolingual control condition (pure lists), in both bilinguals and trilinguals participants. The number of participants recruited was appropriate to the study. Experiment 1 was dedicated to replicate previous findings from Grainger and Beauvillain (1987). In Experiment 2, the number of participants was limited given that it was fairly difficult to recruit trilingual participants with similar AoA of L2 and L3. Nevertheless, the homogeneity of the participants allowed us to obtain powerful data to support our hypothesis. Altogether, the results obtained here support a language non-selective access to the mental lexicon. Indeed, we obtain language dominance effects in both bilinguals (Experiment 1) and trilinguals (Experiment 2). Languages seem to be organised within the memory in a hierarchical mode, and the dominant language is easily activated compared to other spoken languages. In addition, a non-dominant language (L2) could be revealed as dominant upon a third language (L3), indicating the complexity of the lexical connections among them. Interestingly, the effect of dominance remains relevant even if participants have to deal with two or three languages simultaneously. However, this effect is modulated by a slowdown of processing in mixed lexical decisions. The asymmetries observed in languages are mainly due to AoA and proficiency, indicating that the effect of dominance is not settled once for all and can evolve depending on the skills of the participants in the three languages. Languages are often learned at different life stages, not practised with the same frequency or in the same context of usage; this fact is highlighted by an asymmetry between the languages at a cognitive level (Kroll & Stewart, 1994; Meuter, 2005). Moreover, we systematically observed effects due to language switching in multilingual lists. It seems that language switching reflects, in a certain way, the asymmetry observed among languages. If a switching effect is found in all switching directions, the results indicate that the cost associated with switching is larger when switching from a dominant language to a non-dominant language. This effect of language switching could be explained in the light of BIA-d model (Grainger et al., 2010), which postulates that lexical representations in the different languages are interconnected by inhibitory links, and that the selection of the adequate lexical candidate is made on the basis of the lexical units from all the languages known. 16 X. Aparicio and J.-M. Lavaur We observed similar findings for language dominance and language switching effects in both experiments, indicating that quantitative differences (in terms of number of languages spoken) have a negative impact on the quality of processing. It seems that recognition processes are slower in trilinguals compared to bilinguals in the non-dominant languages, as suggested in Table 4. Table 4. Comparison of RTs (ms) in bilinguals (Experiment 1) and trilinguals (Experiment 2) in identical conditions of pure lexical decisions in L1 (French) and L2 (English), and generalised lexical decisions (L1
L2). Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 Monolingual list Bilinguals Trilinguals Bilingual list L1 L2 L1 L2 491 497 527 561 527 534 585 628 This research also highlights the complexity of the links bounding languages inside the mental lexicon, as well as the asymmetry among languages, mainly between L2 and L3, which has not been investigated so far. It is highly possible that the importance of language switching depends on the temporal delay between the target and the preceding trial. It seems that language switching elicits a higher cost when the target word activates the language node presented in the dominant language, and this activation requires an inhibition of the language node from the non-dominant language previously activated. The asymmetry is based on the assumption that lexical representations from L1 have a higher level of residual activation compared to L2 and L3, due to its frequency of usage. Consequently, L1 words generate more bottom-up activation to the L1 language node compared to L2 and L3 towards the corresponding language node. Moreover, the cost associated with language switching is more important between L2 and L3, because these two languages have a strong level of asymmetry, indicated by a massive slowdown of processing. Conclusion To conclude, our data support the hypothesis of a multilingual lexicon integrated within the memory, in which lexical representations of words are linked by inhibitory connections. Lexical access appears to be supervised by formal aspects, but seems also sensitive to the linguistic context related to the task. In early stages of processing, it appears that trilinguals automatically activate the lexical representation of the three languages, and the competition between the lexical candidates leads to language switching costs, corroborating predictions from BIA-d model (Grainger et al., 2010). References Aparicio, X., Heidlmayr, K., & Isel, F. (in press). Effect of inhibition training on language switching and Stroop tasks: Evidence from late bilinguals and simultaneous interpreters. Frontiers in Psychology. Aparicio, X., Midgley, K., Holcomb, P., Pu, H., Lavaur, J. -M., & Grainger, J. (2012). Visual word recognition by trilinguals: An ERP study. Frontiers in Psychology, 3(402), 1
9. doi:10.3389/fpsyg.2012.00402 Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 International Journal of Multilingualism 17 Baayen, R. H., Piepenbrock, R., & Gulikers, L. (1995). The CELEX lexical database (CDROM). Philadelphia, PA: Linguistic Data Consortium, University of Pennsylvania. Chauncey, K., Grainger, J., & Holcomb, P. J. (2008). Code-switching effects in bilingual word recognition: A masked priming studies with event-related potentials. Brain and Language, 105(3), 161
174. doi:10.1016/j.bandl.2007.11.006 Christoffels, I. K., Firk, C., & Schiller, N. O. (2007). Bilingual language control: An eventrelated brain potential study. Brain Research, 1147, 192
208. doi:10.1016/ j.brainres.2007.01.137 Dijkstra, T. (2005). Bilingual visual word recognition and lexical access. In J. F. Kroll & A. de Groot (Eds.), Handbook of bilingualism: Psycholinguistic approaches (pp. 179
200). Oxford: Cambridge University Press. Dijkstra, T., & van Hell, J. G. (2003). Testing the language mode hypothesis using trilinguals. International Journal of Bilingual Education and Bilingualism, 6(1), 2
16. doi:10.1080/ 13670050308667769 Duyck, W., Desmet, T., Verbeke, L. P. C., & Brysbaert, M. (2004). WordGen: A tool for word selection and nonword generation in Dutch, English, German, and French. Behavior Research Methods, Instruments, & Computers, 36(3), 488
499. doi:10.3758/BF03195595 Font, N. & Lavaur, J. -M. (2004). Effets de la fréquence du voisinage orthographique interlangue lors de la reconnaissance visuelle de mots chez les bilingues [Effects of interlingual orthographic neighborhood frequency in bilingual visual word recognition]. L’année Psychologique, 104(3), 377
405. doi:10.3406/psy.2004.29673 Grainger, J. & Beauvillain, C. (1987). Language blocking and lexical access in bilinguals. The Quarterly Journal of Experimental Psychology, 39A, 295
319. Grainger, J., Midgley, K., & Holcomb, P. J. (2010). Re-thinking the bilingual interactiveactivation model from a developmental perspective (BIA-d). In M. Kail & M. Hickmann (Eds.), Language acquisition across linguistic and cognitive systems (pp. 267
284). New York: John Benjamins. Kroll, J. F. & Stewart, E. (1994). Category interference in translation and picture naming: Evidence for asymmetric connections between bilingual memory representations. Journal of Memory and Language, 33(2), 149
174. doi:10.1006/jmla.1994.1008 Laxén, J., Aparicio, X., & Lavaur, J. -M. (2008). Base lexicale trilingue ESF. Mesure du partage orthographique et du recouvrement sémantique interlangue [Trilingual lexical database ESF. Measurement of interlingual orthographic and semantic overlap]. Colloque de la société française de psychologie. Bordeaux, 10
12 Septembre 2008. Lemhöfer, K., Dijkstra, T., & Michel, M. (2004). Three languages, one ECHO: Cognate effects in trilingual word recognition. Language and Cognitive Processes, 19(5), 585
611. doi:10.1080/01690960444000007 Macnamara, J., Krauthammer, M., & Bolgar, M. (1968). Language switching in bilinguals as a function of stimulus uncertainty. Journal of Experimental Psychology, 78(2 Pt1), 208
215. doi:10.1037/h0026390 Meuter, R. F. I. (2005). Language selection in bilinguals: Mechanisms and processes. In J. F. Kroll & A. M. B. De Groot (Eds.), Handbook of bilingualism: Psycholinguistic approaches (pp. 349
368). Oxford: Cambridge University Press. Midgley, K. J., Holcomb, P. J., & Grainger, J. (2009). Language effects in second language learners and proficient bilinguals investigated with event-related potentials. Journal of Neurolinguistics, 22(3), 281
300. doi:10.1016/j.jneuroling.2008.08.001 Moreno, E. M., Federmeier, K. D., & Kutas, M. (2002). Switching languages, switching Palabras (words): An electrophysiological study of code switching. Brain and Language, 80(2), 188
207. doi:10.1006/brln.2001.2588 New, B., Pallier, C., Brysbaert, M., & Ferrand, L. (2004). Lexique 2: A new French lexical database. Behavior Research Methods, Instruments, & Computers, 36(3), 516
524. doi:10.3758/BF03195598 Orfanidou, E., & Sumner, P. (2005). Language switching and the effects of orthographic specificity and response repetition. Memory & Cognition, 33(2), 355
369. doi:10.3758/ BF03195323 Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime (Version 1.0) [Computer Software]. Pittsburgh, PA: Psychology Software Tools, Inc. 18 X. Aparicio and J.-M. Lavaur Downloaded by [Xavier Aparicio] at 09:36 28 March 2013 Thomas, M. S. C., & Allport, A. (2000). Language switching costs in bilingual visual word recognition. Journal of Memory and Language, 43(1), 44
66. doi:10.1006/jmla.1999.2700 van Heuven, W. J. B., Dijkstra, T., & Grainger, J. (1998). Orthographic neighborhood effects in bilingual word recognition. Journal of Memory and Language, 39(3), 458
483. doi:10.1006/ jmla.1998.2584 Von Studnitz, R. E., & Green, D. W. (1997). Lexical decision and language switching. The International Journal of Bilingualism, 1(1), 3
24. doi:10.1177/136700699700100102