Abstract

In this study, the authors show that cross-lingual phonological priming is possible not only from the 1st language (L1) to the 2nd language (L2), but also from L2 to L1. In addition, both priming effects were found to have the same magnitude and to not be related to differences in word naming latencies between L1 and L2. The findings are further evidence against language-selective access models of bilingual word processing and are more in line with strong phonological models of visual word recognition than with the traditional dual-route models. Until 1990, the general idea was that bilinguals had two mental lexicons: one for the first language and one for the second. In addition, a language switch mechanism controlled which lexicon was active. Such an architecture of language selective access seemed ideal to explain why, in general, bilinguals do not experience interference problems from one language to the other. In recent years, however, evidence has accumulated showing that the first stages of visual word recognition are largely language independent and that the assumption of independent lexicons may be incorrect. Subsequently, we give a summary of this evidence (for further discussion, see First, in lexical decision tasks, it has been shown that bilingual participants cannot suppress one of their languages, even when the task strongly urges them to do so because of interference costs. For instance, it has been shown repeatedly that if the nonword trials contain words of the nontarget language, there are large processing costs associated with these trials (e.g., Second, researchers have extended basic findings of monolingual language processing to bilingual processing. The second issue In their first experiment, However, it might be argued that Brysbaert et al.'s (1999) findings are not really strong evidence against the traditional dual-route models of visual word recognition 1 In these models, visual word recognition involves the combined activity of an orthographic lexicon (the lexical route) and a rule-based grapheme-phoneme conversion (GPC) system (the nonlexical route). The simplest extension to the bilingual situation would probably involve a bilingual lexicon (e.g., similar to the one implemented by and an augmented GPC system. Certainly in the case of sequential bilingualism (when L1 is mastered well before L2), the latter would include all GPC rules from the native language, possibly augmented with noncontradicting rules from the second language. These are rules that concern graphemes that are present in L2 but not in L1. Although the different versions of the dual-route model have not yet addressed the simulation of the 1 We discuss dual-route models because there are different versions of the general architecture, depending on how the grapheme-phoneme conversion route has been implemented. 617 WORD RECOGNITION IN BILINGUALS masked phonological priming effect, it could be conceived that such a model might predict a phonological priming effect from L1 on L2. Such a model might also be able to incorporate phonological priming from L2 on L2, as long as the rules involved do not contradict the GPC rules of L1. (Because the GPC system is rule based, deviations from the rules must be handled by the lexical route.) In agreement with this view, the reduced net phonological priming effect in L2 (8%) compared with L1 (16%; see the first three rows of A way to more clearly decide between the strong phonological models of bilingual word processing and the dual-route models is to look at whether it is possible to prime L1 target words with L2 pseudohomophones. If the GPC system does not include lettersound correspondences that deviate from the L1 graphemephoneme correspondences, then a dual-route model predicts that phonological priming from L2 to L1 will be very limited. For instance, in English-Dutch bilinguals, the nonword bleem is not expected to prime the target word BLAME, because the grapheme ee already maps to the phoneme /i/ and thus can no longer be linked to the phoneme /e/. In contrast, a strong phonological model of visual word recognition predicts very similar phonological priming from L2 on L1 as from L1 on L2, because phonological coding is mandatory in both languages and all letter-sound correspondences are activated automatically and in parallel. Thus, the evidence in favor of a strong phonological model of bilingual visual word processing would become more compelling if cross-lingual phonological priming were shown from L2 primes to L1 targets. This is what we investigated in Experiment 2. In Experiment 1, we looked at the naming latencies for words in L1 and L2. There were two reasons for doing so: First, word naming times could provide us with an interesting measure of language proficiency in bilinguals, and second, different theories of visual word recognition make different predictions about the correlation between prelexical phonological coding and word naming times. The importance of these two issues is outlined in the introductions of Experiments 1 and 2, respectively. Experiment 1 In the early days of research on bilingualism, researchers were mainly interested in so-called balanced bilinguals. These are persons who master two languages to the same degree and who can use both languages in virtually every situation. Usually, they are individuals who have been raised bilingually from their birth (e.g., because their parents spoke different languages, or because the language of the family differed from the language of the school environment). Subsequent research has indicated, however, that completely balanced bilinguals are very rare, because it is too effortful to consistently use both languages for all conversation topics (e.g., physics vs. cooking) and in all modalities (e.g., listening vs. writing; for a review, see Schaerlaekens, 1998). Furthermore, investigators have realized that research on balanced bilingualism has limited practical value, because the vast majority of people who know more than one language are unbalanced bilinguals, with less than perfect knowledge of their second language. Research with unbalanced bilinguals, however, raises the problem of determining the proficiency level in both languages, as argued most strongly by There is some evidence that bilinguals are slower to read aloud printed words in their second language than in their first language. For instance, La Heij, Hooglander, Kerling, and van der Velden (1996) reported color-word naming times of 443 ms in L1 and 456 ms in L2 (Experiment 1). For the reading of concrete words (Experiment 2), the naming times were respectively 441 ms and 473 ms. Similarly, Kroll and Stewart (1994, Experiment 3) reported that word naming in L2 took 91 ms longer than word naming in L1. So, an interesting way to assess the vocabulary proficiency in L1 and L2 may be simply to ask participants to read aloud words in both languages and to measure the voice onset times. Experiment 1 was designed to test the usefulness of such a task, by directly comparing the naming latencies of two complementary groups of bilinguals (French-Dutch vs. Dutch-French) with the same sets of stimuli. In addition, we manipulated word frequency in both languages, to gather the effect of this variable in second language reading compared with first language reading. Traditionally, the frequency effect in word naming has been considered a marker of lexical involvement in the naming process. It has been shown that the frequency effect is stronger in the naming of words with an opaque correspondence between letters and sounds than in the naming of words with a transparent relation. For instance, Frost, VAN WIJNENDAELE AND BRYSBAERT Method Participants. Participants consisted of two groups of bilinguals: one group of French-Dutch bilinguals and one group of Dutch-French bilinguals. The French-Dutch bilinguals were 39 students from the University of Leuven. Their first language was French, and most of them lived and had attended high school in the French-speaking part of Belgium. At the time of the experiment, they studied at the University of Leuven and followed university courses in Dutch. The majority had already finished a graduate program at a French-speaking university. Only 5 participants had started to learn Dutch before the age of 10. The Dutch-French bilinguals were 20 PhD students from the University of Leuven, mainly from the Department of Psychology. All had Dutch as their first language and had started to learn French at the age of 11. None of them had grown up in a bilingual environment. All participants rated their comprehension of the second language as quite good (7 or more on a 10-point rating scale, ranging from bad to excellent), but most of them were less sure about the quality of their production in the second language. Stimulus materials. Stimulus materials consisted of 154 French words and 160 Dutch words. The words of both languages were matched on frequency, number of letters, number of phonemes, and number of syllables. Words were divided in four frequency categories. Very-lowfrequency words (LLFR) had a frequency of 1-10 occurrences per million words (based on the CELEX counts for the Dutch words Procedure. Participants were seated individually in a quiet room. The instructions were presented on a computer screen. The participants were told that there would first be a practice session with 20 trials, then a short break followed by a naming task in the first language, and then another short break followed by a naming task in the second language. The words had to be read aloud as quickly and as accurately as possible. Each trial started with a blank screen for 1 s, after which two vertically aligned lines appeared at the center of the screen. The participants were asked to fixate the gap between the lines. After the appearance of the lines (500 ms), a word was presented horizontally in such a way that the second letter was situated in the gap between the lines. Previous research has indicated that the second letter is the optimal viewing position for naming short Dutch and French words Results The data of 1 French-Dutch bilingual were lost due to a computer failure. Therefore, the analyses described below are based on 38 French-Dutch and 20 Dutch-French bilinguals. Analyses of variance (ANOVAs) with three variables (language group of the participants, language of the words, and word frequency) revealed a reliable main effect of word frequency, F 1 (3, 168) ϭ 39.53, MSE ϭ 238, p Ͻ .01, and F 2 (3, 306) ϭ 19.67, MSE ϭ 713, p Ͻ .01, but this effect was embedded in a significant interaction with language of the words, F 1 (3, 168) ϭ 4.46, MSE ϭ 198, p Ͻ .01, and F 2 (3, 306) ϭ 1.81, MSE ϭ 736, p Ͼ .10, and a significant triple interaction between language group, language of the words, and word frequency, F 1 (3, 144) ϭ 14.53, MSE ϭ 198, p Ͻ .01, and F 2 (3, 306) ϭ 12.67, MSE ϭ 348, p Ͻ .01. Discussion The results of Experiment 1 confirm that participants are faster to name words in their native language than to name words in their second language. Overall, there was a 40-to 50-ms difference between L1 and L2. The L1 advantage was modulated by word frequency. It was larger for low-frequency words than for highfrequency words, because the frequency effect was more pronounced in L2 than in L1. In fact, the frequency effect in L1 was quite small both for Dutch and French (around 10 ms). This is in agreement with figures previously published for these languages (e.g., Brysbaert, Lange, & Van Wijnendaele, 2000; Peereman & WORD RECOGNITION IN BILINGUALS As indicated in the introduction, the word frequency effect is often interpreted as a marker for lexical involvement in word naming, both within the dual-route models of word naming The results of Experiment 1 also confirm that the word naming task can be used to assess language proficiency in bilinguals. In fact, we obtained a nearly symmetric pattern of naming latencies as a function of the language presented (Dutch-French) and the mother tongue of the participants (see