ABSTRACT
Examining whether bilinguals can selectively activate the linguistic knowledge of their two languages when processing spoken or written words in one is a central issue of psycholinguistic research. However, although numerous studies have examined reading-related processes in bilingual adults, there is to our knowledge no published study examining these questions in bilingual beginning readers. This is surprising given that classrooms worldwide continue to become more linguistically diverse, resulting in a growing number of children who learn to read and spell in a second language (L2), before (or without) being formally instructed in their native language (L1). Moreover, phonological processing is a sine qua non of successful literacy devel-
opment (see e.g. Share, 1995; NICHHD, 2000). Therefore, it is important to examine whether bilingual children activate the grapheme-phoneme correspondences (GPCs) of their two languages when reading in one, as well as to determine the factors that influence the relative activation of the two languages. These questions have major educational implications for the teaching of reading to second-language learners, as efficient phonological processing requires the inhibition of the phonological and orthographic representations of the irrelevant language; and lack of such inhibition may lead to inadequate and non-fluent reading. Until the beginning of the 1990s, the dominant conception was that bilinguals
had their knowledge of the two languages separated, and could access one language selectively in line with the task being performed (e.g. de Groot, 1992, 1993; de Groot & Kroll, 1997; Kroll, 1993; Paradis, 1997). Although several early studies were compatible with the selective access hypothesis for phonological and orthographic information (e.g. Durgunoglu & Roediger, 1987; Gerard & Scarborough, 1989; Soares & Grosjean, 1984; Scarborough et al., 1984), recent studies show that bilinguals cannot totally inhibit the language irrelevant to the task, even when its activation is detrimental to performance (e.g. Wang et al., 2003; see Brysbaert, 1998; Brysbaert & Dijkstra, 2006). For example, based on studies showing that rare words take longer to identify
when preceded by visually similar frequent words (e.g. blur preceded by blue) than when preceded by visually dissimilar words (Segui & Grainger, 1990), Bijeljac-Babic et al. (1997) asked French-English bilinguals and French monolinguals to perform a
lexical decision task on French words (e.g. amont). The targets were preceded by masked primes of which some were visually similar and frequent words from French (e.g. amant) or fromEnglish (e.g. among). The results show that the bilinguals with high levels of proficiencies in English, but not the ones with lower levels of proficiencies, displayed an effect of inhibition of the same magnitude for targets preceded by English words as for targets preceded by French words; whereas the monolinguals only showed an effect of inhibition for the targets preceded by visually similar French words. Relying on the observation that words with many orthographic neighbours (words
of the same length differing by one letter) are identified faster than words with few neighbours (see Andrews, 1997, for a review), vanHeuven et al. (1998) further showed that Dutch-English bilinguals were influenced by the items’ density of orthographic neighbourhood in the other language in progressive demasking and general lexical decision tasks (see also Grainger & Dijkstra, 1992). According to the authors, only models that consider lexical access to be non-selective, but allow some control of the relative level of activation of the two languages, can account for these results. Further studies examining the processing of homographic words (words sharing
the same spelling but different pronunciations and meanings across languages) in lexical decision tasks also showed interference from the other language when such items were presented in mixed lists which contained also non-homographic words of the other language (de Groot et al., 2000; Dijkstra et al., 2000; Dijkstra et al., 1998a). These studies also show that the size of the interference was correlated with the written frequency of the homographs in the other language, even when it was lower than its frequency in the target language. Another study by de Bruijn et al. (2001), which combined the measure of reaction
times and of evoked potentials, further suggests that the linguistic context in which a homographic word is presented does not inhibit the activation of the words’ meaning in the other language. In this study, Dutch-English bilinguals performed a general lexical decision task on triplets of items (saying ‘no’ if any item was a pseudo-word in both languages), of which the first item was a non-homographic word from English (e.g. house) or from Dutch (e.g. zaak). This item was followed sometimes by a homographic word whose meaning was associated with the third item (e.g. angelheaven; angel meaning dart in Dutch) or not (e.g. angel-brush). The main issue was to examine whether a Dutch first item would inhibit the English meaning of the (second) homographic word, thus affecting the effect of semantic priming on the third item. Significant priming effects were observed both on reaction times and on evoked potentials (on the N400, which would reflect semantic integration) when the third item was semantically related to the second, but the effect was not modulated by the linguistic context provided by the language of the first item. Thus, semantic priming effects were similar for house-angel-heaven and for zaak-angel-heaven. These results are compatible with a strong non-selective access view of bilingualism. In the domain of phonological activation, several studies also support the non-
selective access hypothesis. For example, Nas (1983) showed that in an English lexical decision task, Dutch-English bilinguals produced more errors and responded slower when rejecting pseudo-words of the target language which were homophonous to words of their other language (e.g. snay, which is pronounced like the Dutch word snee -cut-) than when rejecting non-homophonous pseudo-words (e.g. rolm). Such effects were reproduced by Dijkstra et al. (1999) on homophonic (but
not homographic) Dutch-English words (e.g. leaf, homophonous of the Dutch word lief -gentle-). Phonological activation has also been observed between languages that do not
share the same script. Hence, Gollan et al. (1997) examined Hebrew-English and English-Hebrew bilinguals in a lexical decision task in their L2 or, in another experiment, in their L1. The targets were preceded by primes (presented for 50 msec) of which some were translations of the targets in the other language. The results show a robust effect of priming by translation, but this effect was bigger for the semantically and phonologically close pairs (cognates) than for the ones that were only semantically close (non-cognates). As Hebrew and English words never look alike orthographically, the authors suggest that the advantage observed for the cognate words results from inter-language phonological similarity. Moreover, an advantage for cognate words was observed only when the primes were presented in the participants’ dominant language, and the targets in their non-dominant language. In a bilingual adaptation of the Stroop paradigm, Tzelgov et al. (1996) asked
Hebrew-English bilinguals to name the colour of visual words either in Hebrew or (in another experiment) in English. Some stimuli were pseudo-words of the language not required in the experiment but were homophonous of colour names of the target language. The results show that the Stroop effect was as important for the English pseudo-words homophonous to colour names in Hebrew (e.g. kahol, which is homophonous of the Hebrew word for blue) as for the colour names written in Hebrew, although the Stroop effect induced by the Hebrew pseudo-words homophonous to colour names in English was less strong and statistically less robust. The results from these two last studies seem to indicate that phonological coding within one script automatically activates phonological representations of words in another script, even when the two scripts are visually dissimilar. The next issue concerns whether bilinguals who know two languages that share
the same script automatically activate their grapho-phonological knowledge of both languages when processing one language. Doctor and Klein (1992) report results compatible with this hypothesis of multiple phonological coding. The authors observed that in a general lexical decision task, English-Afrikaans bilinguals produced more errors and took longer to accept homophonous heterographic words (e.g. lakelyk) than to accept homographic heterophonous words (e.g. kind, meaning child in Afrikaans) or control words specific to either language. The authors account for these results by assuming that the presentation of a written word induces a search in parallel in both orthographic lexicon as well as a non-selective phonological coding using the GCPs of both languages. Homophonous heterographic words would activate a single phonological representation but two distinct orthographic representations, which would induce a process of (visual) verification, slow reaction times. This hypothesis is supported by post-hoc analyses showing that for the homophonous words, response latencies were faster for visually dissimilar pairs (e.g. eye-aai) than for visually similar pairs (e.g. brick-briek). The studies by Brysbaert and colleagues (Brysbaert et al., 1999; Van Wijnendaele
& Brysbaert, 2002) provide more direct evidence for automatic activation of the GCPs of both languages in bilinguals. Using the masked priming paradigm, Brysbaert et al. (1999) presented Dutch-French bilinguals and French monolinguals with French targets preceded by masked primes (both primes and targets were presented
for 57 msec), of which some were either pseudo-words homophonous with the targets if read with the French GPCs (e.g. fain-faim), or Dutch words homophonous with the targets if read with the Dutch GCPs (e.g. koel-coule). These two conditions were compared with control primes sharing the same degree of orthographic overlap with the targets as the homophonous primes (e.g. faic-faim and doel-coule, respectively). The results show that for the fain-faim pairs, the phonological priming effect
(advantage for the targets preceded by homophonous primes over targets preceded by non-homophonous primes) was of the same magnitude for the bilinguals and for the monolinguals. Conversely, for the koel-coule pairs, only the bilinguals showed a phonological priming effect, and this effect was of the same magnitude as the one observed for the French primes. From these results the authors conclude that the reading of words in the L2 seems to automatically activate the GCPs of the L1 as well. Moreover, these results were replicated with French-Dutch bilinguals (Van Wijnendaele & Brysbaert, 2002), which suggests that the reading of words in the L1 also automatically activates the GCPs of the L2. Thus, the processing of visual words would activate all the GCPs compatible with the stimuli, irrespective of language and linguistic dominance. Brysbaert et al. (1999) note that these results do not exclude the intervention of inhibition processes, which would allow avoiding confusions between languages (see de Bruijn et al., 2001; Dijkstra & van Heuven, 2002). However, such processes would occur only after an initial phase of non-selective activation of all the GCPs compatible with the stimuli. The hypothesis of non-selective access to the GCPs of both languages would lead
one to predict that inconsistent GCPs between languages should induce similar effects as the consistency effects observed in monolinguals with intra-language inconsistent GCPs, i.e. the fact that sequences with many ‘enemies’ (e.g. –ead, pronounced differently in bead and head) induce slower reaction times than sequences with few ‘enemies’ (see e.g. Jared, 1997; Peereman & Content, 1997). To examine whether consistency effects generalize across languages in bilinguals,
Jared and Kroll (2001) asked English-French and French-English bilinguals to read words with rimes either specific to English and consistent (e.g. bump, which has no enemy), specific to English but inconsistent (e.g. bead, which has enemies like head), or homographic between English and French and inconsistent across languages (e.g. bait, which has no enemy in English but enemies in French like fait). In all experiments, the procedure consisted in presenting a first block of English words with all three types of items, followed by a block of French, followed by a second block of English words (different from those of the first block) containing all types of items. The results show that for the first block of English words, the French-English
bilinguals, but not the English-French bilinguals, produced more errors (but not longer reaction times) to the words with enemies in French than for the words with no enemies. Conversely, both groups produced more errors and longer reaction times for the words with enemies in English than for the words with no enemies. According to Jared and Kroll (2001), these results show that the GCPs of English are activated to a greater extent than the French GCPs in all groups, even though the French GCPs are also activated for the group for which it is the dominant language. For the second block of English words, the English-French bilinguals with high
levels of proficiency in French took longer to read the words with French enemies than to read the words with no enemies. For the French-English bilinguals, the
participants schooled in a francophone university, but not those schooled in an English-speaking one, produced longer reaction times and slightly more errors for the words with French enemies than for the words with no enemies. To account for these results, the authors suggest that only the French-English bilinguals schooled in an English-speaking university would be able to inhibit their knowledge of the French GCPs when reading in English. The English-French bilinguals would not have adopted such a strategy because they would not be aware of the impact of their knowledge of French on their English reading. Jared and Szucs (2002) further observed that French-English bilinguals, but not
English-French bilinguals, took longer and made more errors on homographic heterophonous non cognate words of English and French (e.g. coin, meaning corner in French) than on English-specific words. On the basis of the results of these two last studies, Jared and Szucs (2002) suggest that the GCPs of the language irrelevant for the task are too weakly activated to induce noticeable interferences on participants’ performance, except when this language is their dominant language and when the inconsistencies across language regard entire words (homographs). This conclusion seems contradictory with the strong hypothesis of non-selective
access proposed by Doctor and Klein (1992) as well as by Brysbaert and colleagues (Brysbaert et al., 1999; Van Wijnendaele & Brysbaert, 2002). One factor that could account for this apparent discrepancy lies in the different materials exploited in the two sets of studies. Indeed, Jared and colleagues used homographic sequences (rimes or whole words), although Brysbaert and colleagues used heterographic sequences with graphemes specific to the language irrelevant for the task to be performed (see p. 180 for a more detailed description of these materials). Other studies have indeed showed that the make-up of the stimuli influences the relative degree of activation of the two languages in bilinguals. Hence, Grainger and Beauvillain (1987) showed that, in a general lexical decision task, French-English bilinguals produced longer reaction times for the second member of ‘mixed’ pairs displaying one word of each language (e.g. rule after pont) than for ‘pure’ pairs displaying only English words (e. g. rule after sand), except when the second item displayed specific orthographic cues to that language (e.g. white after pont, where wh-never occurs in French but is frequent in English). Such influence of orthographic specificity was also observed by Beauvillain (1989, 1992) when the specific and non-specific words of the two languages were matched for frequency distributions. She indeed showed that the negative regression slopes between word frequencies and reaction times were stronger for the non-specific than for the specific words. This pattern of results is compatible with the notion that non-specific words are selected within a cohort of words from both languages, whereas specific words are selected from a cohort from one language (see also Grainger, 1993, for a more detailed discussion and a reinterpretation of previous results based on the notion of specificity). The dominant model proposed to account for written word identification in adult
bilinguals is the Bilingual Interactive Activation model (BIA, see Dijkstra et al., 1998b). The BIA is an adaptation of the Interactive Activation model of McClelland and Rumelhart (1981) to the bilingual situation. Several layers of units represent visual features, letters and words. In bilinguals, both lexicons are integrated into a unitary system, so that bottom-up activation induces the activation of words in the two languages. The original model also comprised a supra-lexical layer with ‘language
nodes’, which would collect the activation of all words from that language and inhibit word activation in the other language via top-down connections. In the more recent version of the model, the ‘language nodes’ have been replaced by decisional and strategic processes occurring outside the recognition system (BIA+, Dijkstra & van Heuven, 2002), which could account for activation and inhibition processes driven by words’ orthographic cues to language (specific/typical graphemes). However, as stated by Dijkstra & van Heuven (2002), ‘we still ignore how these activation and inhibition processes which intervene in adults develop in children’. The present study aimed at examining whether bilingual children activate the
GPCs of their non-dominant reading language (in the present study, their L1) when naming words in their dominant reading language (L2), and explored whether such activation is influenced by the orthographic characteristics of the presented items. Hence, we compared these children’s performances to those of monolinguals in a reading task which involved items with inconsistent GCPs across languages (e.g. ‘an’, which is routinely pronounced /ã/ in French but /an/ in Dutch). To examine these questions, French-native children schooled in Dutch and Dutch-
native children schooled in French were compared to Dutch and French monolinguals, respectively. These four groups of children were asked to perform a naming task that included three types of items displaying inconsistent GCPs across languages: (1) non-cognate heterographic homographs (words that are spelled identically but have different pronunciations and meanings in the two languages) with cross-linguistically inconsistent GPCs (e.g. vent, which is pronounced /vã/ and means wind in French, but is pronounced /vEnt/ and means boy in Dutch); (2) matched words specific to French with inconsistent GPCs (e.g. dent, which is pronounced /dã/ and means tooth in French, but corresponds to the pseudo-word /dEnt/ in Dutch); (3) matched Dutch-specific words with inconsistent GPCs (went, which is pronounced /wEnt/ and means (you) get used in Dutch, but corresponds to the pseudo-word /wã/ in French). The materials also contained matched (control) pseudo-words with consistent GPCs across languages (mele, corresponding to the pseudo-word /mel(∂)/ in both French and Dutch). The strong hypothesis of non-selective access would lead one to predict that the
bilinguals should produce more errors and longer reaction times than the monolinguals for the three types of items with inconsistent GCPs, but not for the pseudowords with consistent GCPs. Conversely, under the hypothesis of selective access, bilinguals should not activate the GCPs of their irrelevant language to a sufficient extent to produce noticeable interference on their reading performance in the target language. In post-hoc analyses examining the factors that could account for the observed
pattern of results, the additional factor of typicality was considered since half of the presented Dutch-specific words contained graphemes typical in Dutch but exceptional in French (‘z’, ‘w’, ‘k’), whereas none of the French-specific words contained graphemes typical to French.
