Non-selective access to L2 phonological representations in Cantonese picture naming
Master thesis July 2014
Cheung, W.S. (0420506)
Prof. dr. A.M.B. de Groot, Dr. P.A. Starreveld Brain & Cognition department
Faculty of Psychology University of Amsterdam Amount of words: 7741
Abstract
Ample research shows that bilingual language production is language-nonselective; that is, a bilingual cannot de-activate their second language during the use of one language only. However, most of these studies involved languages that share the same script, leaving this conjecture undecided for languages that are dissimilar in script (e.g. Chinese vs Dutch). Using a phoneme monitoring task (Colomé, 2001), the present study investigated whether in
unbalanced Cantonese-Dutch bilinguals Dutch is co-activated during covert Cantonese picture naming. The participants were asked to indicate whether a specific phoneme appeared in the Cantonese name of a picture. Each picture was paired with a visually presented letter
representing a phoneme of the Cantonese picture name, a phoneme of the Dutch picture name, or an unrelated phoneme. In the second experiment, the phonemes were presented aurally to maintain a strict unilingual setting. In both experiments, evidence for non-selective processing up to the phonological level has been found, suggesting that languages are non-selectively processed even in different-script bilinguals.
Lexical access in different-script bilinguals
Bilinguals are able to use two or more languages accurately. A large body of literature suggests that during language comprehension in one language, the language representations of a bilingual’s other language are activated as well (e.g. Dijkstra, Van Jaarsveld, & Ten Brinke, 1998; Grainger & Dijkstra, 1992; Spivey & Marian, 1999). However, less evidence is
available concerning the issue whether the other language is also activated during bilingual language production. For instance, when a Cantonese-Dutch bilingual names a table in Cantonese Chinese, could it be that the word tafel (the Dutch translation of table) also becomes activated? This question has been a main topic in bilingual language production research. Whether the language not in use is co-activated has been investigated by focusing on a core component of the speech production process: lexical access.
During lexical access, the message one wants to convey becomes phonologically encoded. It has been assumed that during this process the semantic nodes that correspond to the message become activated first (e.g. Costa, Caramazza & Sebastián-Gallés, 2000). Second, a lexical node that corresponds to the intended meaning (the target) will be selected from the lexical memory. This selected lexical node will then be phonologically encoded and hence, the message will be ready for articulation, see Figure 1.
Figure 1. Schematic representation of lexical access during (monolingual) word production. Adapted from Costa
et al (2000).
In monolingual speech production research, it is generally assumed that during this process not just the lexical node of the target but a set of lexical nodes becomes activated. Lexical items that are semantically related to the target are assumed to be activated as well. There has been some debate on whether phonological encoding is restricted to only the target (e.g. Levelt, 1989), or whether this also involves those of the semantically related items (e.g. Dell, 1986). Hence, the question raised in monolingual speech production research was whether non-selected lexical items become phonologically encoded or not. Most researchers now agree that the phonology of a set of lexical items becomes activated (ref) during
monolingual word production.
In a bilingual, the process of lexical access might even be more complex. The set of lexical items activated by the semantic representation of the target might not just involve the semantically related items in the intended language but also the corresponding items in the second language, including the translation equivalent of the target. If this were the case, the
translation equivalent will probably be the most highly activated lexical item in the second language. The current study is aimed at detecting the possible phonological activation of the translation equivalent of the target. Such evidence has been obtained in the past and is interpreted as showing that language processing in bilinguals is language-nonselective. That is, bilinguals cannot de-activate a non-target language.
One of the methods that has been used to investigate language-nonselective processing is the occurrence of a cognate facilitation effect. Cognates are words that share word form and phonology across languages (e.g. gat-gato; ‘cat’ in Catalan and Spanish, respectively). Given the similarities between this kind of words, it has been hypothesized that pictures with
cognate names might be named faster than pictures with non-cognate names, which are just translations of each other and thus only share meaning. Such a facilitation in naming for pictures with cognate names compared to pictures with non-cognate names might arise when words from the two languages of a bilingual are simultaneously activated up to the
phonological level (i.e., when language processing is non-selective). If that were the case, naming a picture with a cognate name in the second language (L2) would result in the
activation of the phonological nodes of the corresponding name in the first language (L1). For example when naming the picture of a cat in L2 Spanish, some of the phonological nodes corresponding to the target word gato would also be activated by the L1 translation equivalent
gat (see Figure 2). As a result, the phonological nodes belonging to a cognate target word can
be selected earlier compared to those belonging to a non-cognate target word for which such co-activation of phonological nodes does not occur. Thus, the occurrence of a cognate facilitationeffect in picture naming can be interpreted as evidence of language-nonselective processing.
Figure 2. Schematic representation of non-selective processing
Costa, Caramazza, and Sebastián-Gallés (2000) presented such evidence. In their study, Catalan-Spanish bilinguals were asked to name a set of pictures with cognate and non-cognate names in their non-dominant language Spanish. It was found that Catalan-Spanish bilinguals named the cognate words faster than the non-cognate words. This facilitation in naming was not found with monolingual Spanish speakers. Even when bilinguals performed in their dominant language Catalan, a cognate facilitation effect was found. The authors interpreted these results as evidence that during picture naming in an intended language, words from the non-target language were also activated. Recently, similar results have also been reported for Dutch-English bilinguals (Starreveld, De Groot, Rossmark, & Van Hell, 2013).
Interestingly, the cognate facilitation effect in language production was also found in languages that do not share orthography, such as Japanese and English. Hoshino and Kroll
(2008) compared Japanese-English and Spanish-English bilinguals on cognate and
non-cognate picture naming. The authors used three kinds of non-cognates: English-Japanese non-cognates, English-Spanish cognates, and English-Spanish-Japanese cognates. They found similar results for the two bilingual groups on response latency and accuracy: both bilingual groups were faster and more accurate in naming pictures with cognate names relative to pictures with non-cognate names. Moreover, Japanese-English bilinguals showed this facilitation only for Japanese-English cognates and English-Japanese cognates, but not for Spanish-English cognates. Spanish-Spanish-English bilinguals, on the contrary, showed only facilitation for Spanish-English cognates and Spanish-Japanese cognates, but not for English-Japanese cognates. The finding of a cognate facilitation effect in different-script languages has been interpreted as that script has no influence on the activation of the phonological representations in L2 (Hoshino & Kroll, 2008; Schwartz, Kroll & Diaz, 2007). In other words, even in languages that do not share script, phonological representations of the non-target language are accessed in picture naming.
Although the cognate facilitation effect in picture naming seems robust, it has been argued that a paradigm involving cognate words might not be conclusive about language-nonselective processing. Having more in common than just the translation in the other language, some authors argue that cognate words possess a special status in the bilingual language system. Cognate words are assumed to have a larger conceptual overlap than regular translation words (Van Hell & De Groot, 1998), which might lead to an advantage in
retrieving these words (Colomé & Miozzo, 2010). Therefore, more evidence regarding this issue is needed from a method that does not require the use of cognate words. Such a method is phoneme monitoring. The phoneme monitoring task requires bilinguals to indicate whether the name of a presented picture in one language contains a specific phoneme or sound. The task is completely unilingual. Colomé (2001), for example, conducted this task on fluent
Catalan-Spanish bilinguals in Catalan. The to-be-monitored phonemes that were visually presented as letters were either part of the Catalan name of the picture, or the Spanish name of the picture, or they were unrelated to both the Catalan and the Spanish name of the picture. Thus, when a Catalan-Spanish bilingual was presented with a picture of a table, the presented phonemes could be either /t/ (from taula, the Catalan word for table), /m/ (from mesa, the Spanish word for table), or /f/ (totally unrelated to either the Catalan or the Spanish name for table). If the latencies in rejecting /m/ compared to /f/ do not differ, this indicates that “mesa” is not active and thus that Catalan and Spanish are not simultaneously activated. If speech is language-nonselective, however, then “mesa” would be co-activated together with “taula” and the subjects in this experiment would need more time to reject /m/ compared to /f/. The rationale is that /m/ is present in the translation equivalent of the picture’s name, and thus would trigger a “yes” response. The presence of such phonemes would then interfere with decision making (/m/ is not part of the Catalan name, “ no” is the correct response in this case). Indeed, the author found longer latencies in the trials with the letters representing the Spanish phonemes than the trials with letters representing the unrelated phonemes.
Recently, the phoneme monitoring paradigm has also been applied to Korean-English bilinguals to analyze non-selective processing in different-script languages (Moon & Jiang, 2012). The task was conducted in Korean as well as in English. In both experiments, the non-target language was found to be active during phoneme monitoring. So, even in a task without cognate words the simultaneous activation of L2 occurred for orthographically dissimilar languages.
It is worth mentioning though, that the above mentioned studies with different-script languages are in one respect very similar to alphabetic languages. Moon and Jiang used one of the Korean writing systems that resembles an alphabetic language in the way how the
represent individual sounds such as alphabetic letters do. So, although the graphemes of the Hangul script and the Latin script look different, the writing systems are in essence similar in that there is a strong correspondence between its graphemes and its phonemes. The findings of the Moon and Jiang study do suggest however that language-nonselective processing occurs in different-script languages, but yet this conclusion only concerns different-script languages in which the script has strong grapheme-phoneme associations such as in the Latin script or other alphabetic languages. The same argument also applies to the study of Hoshino and Kroll (2008).
Given the weakness of previous production studies with different-script languages, the present study again addressed this matter but investigated the issue in bilinguals whose second language do not show strong correspondences between script and language sounds. The subjects who participated in this study were Cantonese-Dutch bilinguals. Cantonese is one of the dialects of the Chinese language. Although Cantonese speakers have their ‘own
language’, there is only one Chinese writing system with characters that Chinese people learn to read and write. There is no orthographic overlap between Chinese and Dutch, and neither is Chinese an alphabetic language. Unlike alphabetic letters or graphemes, the characters used in this language have a highly arbitrary character-to-sound correspondence (see Zhao, La Heij, & Schiller, 2012 for a more elaborated description on this).
Another reason to investigate the possible co-activation of non-target language representations during word production by bilinguals whose languages use more distinct language representations is that for these bilinguals picture processing itself seems to involve distinct visual processing patterns. Native Chinese speakers, for example, showed similar ERP patterns for processing Chinese characters and pictures, whereas native English speakers showed processing patterns for pictures and (alphabetic) words that were significantly
between different-script language speakers, it would be relevant to find out whether these differences also exist in the production of the names of the pictures.
The aim of the current study was to evaluate whether non-cognate words of languages from different language families become co-activated during language production in a
unilingual task. To that end, we investigated the question of language-nonselective processing in Cantonese-Dutch bilinguals using the phoneme monitoring paradigm.
In the present study, unbalanced Cantonese-Dutch bilinguals with Cantonese as their mother tongue but Dutch as their dominant language were asked to indicate whether a
particular phoneme appeared in the Cantonese name of a picture. We employed the paradigm in two modalities: visually and aurally. The visual experiment fundamentally resembled Colomé’s study (2001). Every picture was paired with three different letters that represented the phonemes: a phoneme that was part of the Cantonese picture name; a phoneme that was part of the Dutch translation; or a phoneme that was unrelated to either the Cantonese picture name or the Dutch picture name. To illustrate, a picture of a ‘tsung’ (Cantonese word for ‘window’) was accompanied by the phonemes /t/ (of the Cantonese ‘tsung’), /r/ (of the Dutch translation ‘raam’), or /f/ (neither part of ‘tsung’ nor ‘raam’). For the auditory experiment, the to-be-monitored phonemes were recorded and were presented instead of the letters. The presentation of the Latin letters in the first experiment might have had the unintended
consequence of activating the Dutch language system. The objective of the second experiment was therefore to make the experiment more unilingual. The strongest conclusions can namely be drawn from results obtained with unilingual tasks or settings (see Costa, La Heij, &
Navarrete, 2006 and Grosjean, 2001 for a discussion).
To determine whether Dutch was activated during Cantonese picture naming, two effects were analyzed: the response latencies to reject Dutch phonemes compared to the response latencies to reject unrelated phonemes, and the amount of errors on the Dutch
phonemes compared with those on the unrelated phonemes. The rationale behind this is that if Dutch is co-activated with Cantonese during phoneme monitoring, seeing the picture of a window would activate both /t/ and /r/. If the picture’s names in both languages are co-activated then, despite the unilingual nature of the task, it is expected that subjects are slower in rejecting /r/ compared to rejecting /f/, and that they make more errors when rejecting /r/ compared to /f/.
Experiment 1: Visual Stimuli Method
Participants. Twenty-nine Chinese-Dutch bilinguals (14 males, 15 females; mean age = 26, SD = 3.82), all living in the Netherlands participated in this experiment. All had been born into monolingual Cantonese speaking families and learned Dutch in their early
childhood. However, three of the 29 participants were excluded from further analyses, since they had stated that Dutch was not their dominant language. The data of the remaining participants (13 males, 13 females; mean age = 26,1, SD = 3.72) were used in the analyses. Participants received €7 as incentive.
An exit-questionnaire was designed to obtain more information about the participants’ language abilities. For each language and every self-rated language ability (Speaking and Reading), a mean score and a deviation score were calculated. The self-rated reading ability and speaking ability were taken together for Dutch and Cantonese separately to calculate the average knowledge for both languages. These average knowledge rates were compared to determine Dutch dominancy, see Table 1 for a summary.
Two paired-samples t-tests were performed on the averages of the self-rated reading and speaking ability in Cantonese and Dutch respectively. It was found that participants’ self-rated reading ability for Dutch was clearly higher than their reading abilityfor Cantonese,
t(25) = 10.510, p < .001. The participants also rated their Dutch speaking ability higher than
their Cantonese speaking ability, t(25) = 6.082, p < .001. Finally, a paired-samples t-test was performed on the averages of the self-reported language abilities scores to determine Dutch dominancy. The participants rated their average knowledge of Dutch higher than their average knowledge of Cantonese making them unbalanced bilinguals, t(25) = 10.067, p < .001, two-tailed.
Table 1
Means and Standard Deviations (in Parentheses) of Participants’ Self-rated Reading and Speaking Abilities for Dutch and Cantonese
Reading Speaking Total Dutch 6.31 (0.74) 6.08 (0.74) 6.19 (0.68) Cantonese 2.96 (1.59) 4.85 (0.92) 3.90 (1.11)
Materials and Design. A set of 58 black-on-white line drawings was selected from Bates et al. (2003) to serve as stimuli: 8 of them were used as practice items, 25 as
experimental items, and the remaining 25 as fillers. It was made sure that all items were non-cognate words in Dutch and Cantonese. With the exception of one item (‘octopus’), all the stimuli started with a consonant in both languages and consisted of two or more syllables. All stimuli were paired with three black-on-white letters (Arial, 20, bold) representing the
phonemes that had to be monitored. For each experimental item, the letters represented (a) the first phoneme of the Cantonese name of the picture, or (b) the first phoneme of the Dutch name of the picture, or (c) a phoneme that was neither present in the Cantonese name, nor the Dutch name of the picture. For instance, the letters presented with the picture of a POEI LONG (‘backpack’, Dutch translation: ‘rugzak’) were respectively /p/, /r/, and /w/. To avoid
confusion, picture names that started with the letter ‘c’ were left out as much as possible because in Dutch the letter ‘c’ can be pronounced either as ‘c’ or as ‘k’. Although it was tried avoiding picture names that start with a ‘c’ there was one item included which did (“citroen”, meaning “lemon”), because of the limited alternatives. This item was used as a filler item. The Cantonese nouns were transliterated according to Dutch phonetics. The experimental items are listed in Appendix A.
Each experimental picture was presented three times, once for each experimental
condition. For the experimental stimuli, there was one affirmative condition which required a ‘yes’ response (i.e. the letter appeared in the Cantonese picture name) and two negative conditions, in which the letters were part of the Dutch translation or did not appear in the Cantonese nor the Dutch picture names.
Fillers also appeared three times, but were arranged over two affirmative conditions (‘Yes1’ and ‘Yes2’) and one negative condition. The main purpose of this arrangement was to outbalance the number of yes/no responses in the experimental condition. In both
‘yes’-conditions, the pictures were paired with a letter representing a phoneme of their Cantonese name. These letters were chosen from positions other than word onset to ensure that
participants covertly named the whole word instead of focusing on word onset only. The letters used in the ‘no’-condition were related to neither the Cantonese nor the Dutch picture names of the filler items.
Practice items were shown three times as well. In line with the experimental items and fillers, one half of the practice pictures required two affirmative and one negative response, whereas the other half of the practice items required two negative and one affirmative response. Half of the letters were taken from word onset, whereas the other half were taken from positions other than word onset.
The experiment consisted of 150 trials (three conditions x 50 pictures), and the whole experiment was run twice for each participant, making 300 trials in total. After a series of every 50 pictures, the participants were able to pause for a short period of time. Practice pictures were presented before the actual experiment started and were rerun until all practice items had had at least one correct answer.
Procedure. Participants were tested individually in a quiet surrounding, such as a library. Only Cantonese was used for the instructions and in the conversations between the experimenter and the participants. It was explained that the study was on how Cantonese speaking people produce language. They were verbally instructed that a picture would appear on the laptop screen and they were asked to covertly name the picture in Cantonese. The picture would be followed by a letter and they had to decide whether the sound of that letter appeared in the Cantonese name of the picture. It was emphasized that they should focus on the phonology of the letters instead of their graphemic form. They were then given a booklet with the pictures and their Cantonese names written under it in Latin script and were asked to study them for 10 minutes in order to ensure the same words would be used during the
experiment. After they had studied the booklet, the instructions were given another time and it was explained that they could indicate their choice by pressing on either the ‘z’ (for a ‘yes’ response) or the ‘/’ button (for a ‘no’ response). The experiment started with a practice phase, in which the participants were able to get used to the procedure and the task. After the
practice phase, the experiment began. Each trial started with a fixation point (a plus sign) in the middle of the screen for 500 ms, followed by a blank screen for 200 ms, then the picture was presented for 200 ms, and after that the letter was shown for 200 ms. This was followed by a black screen so the participants could enter their response. If the response was wrong, feedback was given by showing the word ‘wrong!’ in Chinese for 1 s. Intertrial latency was 1500 ms. The next trial started automatically if participants did not react within 2500 ms.
Reaction times were measured from the moment the letter appeared on the screen. After completing the experiment, the participants filled out an exit-interview. Presentation software (from Neurobehavioral Systems, Inc.) was used to run the experiment.
Results
Trials answered out of time (before 300 ms or after 2500 ms) were removed before the data were analyzed. This accounted for 5.3% of the trials. The remaining RTs were used to calculate the means for the two conditions for each series, see Table 2.
Two paired-samples t-tests by subjects (t1) and by items (t2) were conducted on the reaction times of both series averaged together to determine any difference between the two negative conditions. It was found that participants were slower to reject the letters belonging to the Dutch translation items compared to rejecting the letters belonging to the control items,
t1(25) = 3.84, p = .001; t2(24) = 2.32, p = .029. The same tests were run on the corresponding error rates. It was found that the participants made more errors on the phonemes that belonged to the Dutch translation items compared to the phonemes that belonged to the control items,
t1(25) = 7.24, p < .001; t2(24) = 3.64, p = .001.
To analyze the pattern of results of the first and the second series, we performed two analyses of variance, one by subjects (F1) and one by items (F2), each with two within-subjects variables condition (translation vs unrelated) and series (first vs second). The main effect of condition was found to be significant, F1(1, 25) = 14.780, MSE = 5,951, p = .001;
F2(1, 24) = 5.378, MSE = 20,140, p =.029. Trials in which the pictures were accompanied by phonemes that were present in the translation equivalent of the pictures’ names were
responded to more slowly than trials in which the pictures were accompanied by unrelated phonemes. The main effect of series was also significant, F1(1, 25) = 7.146, MSE = 23,876, p = .013; F2(1, 24) = 16.548, MSE = 10,552, p < .001. Participants were faster in the second
series than in the first series. The interaction between condition and series was not significant, indicating that the experimental manipulation showed similar results for both series.
Table 2
Mean Reaction Times and Error Rates with accompanying Standard Deviations (in Parentheses) for Each Condition
Condition RT Error
Dutch Translation equivalent 1164 (261) 6.2 (2.4)
Unrelated phoneme 1106 (251) 2.9 (2.2)
The same analyses were performed using the error data. For the error analyses, a main effect of condition was found, F1(1, 25) = 52.344, MSE = 5,185, p < .001; F2(1, 24) = 13.249,
MSE = 21,303, p = .001. In the trials in which the pictures were accompanied by phonemes
that were present in the translation equivalent of the pictures’ names, participants made more errors than in the trials in which the pictures were accompanied by unrelated phonemes. The main effect of series was also significant, F1(1, 25) = 11.280, MSE = 3,942, p = .003; F2(1, 24) = 10.747, MSE = 4,303, p = .003. Participants made more errors in the first series compared to the second series. The interaction between the two factors was not significant, suggesting that the experimental manipulation had the same effect for both series. The reaction times and the error rates are summarized in Table 3.
Summarizing, the two negative conditions differed from each other on reaction time. In addition, more errors were made on the phonemes of the Dutch translation items than on the unrelated phonemes. There was no interaction between the factors condition and series, showing that the first and the second series produced the same effect in the two negative conditions.
Table 3
Subjects’ Mean Score, Standard Deviation and Percentages of Error Rates per Condition, per Series
Series and Condition RT SD %Error
1st series Translation phoneme 1201.9 292.2 4.7 Unrelated phoneme 1148.6 280.3 2.3 2nd series Translation phoneme 1125.8 263.6 3.6 Unrelated phoneme 1062.7 245.8 1.6 Discussion
It was found that participants took longer to reject letters representing phonemes that were present in the Dutch names of the pictures compared to the letters representing unrelated phonemes. Moreover, participants made more errors when rejecting the Dutch phonemes compared to the unrelated phonemes. These results were found in the first series but also in the second series. These results suggest that L2 Dutch was activated during phoneme monitoring in L1 Cantonese.
Although the findings seem straightforward, one important issue should be addressed. As was done in the initial phoneme monitoring paradigm, we used alphabetic letters to represent the phonemes that subjects had to monitor. Indeed, in Colomé’s study (2001) this was a logical step given that the languages that were being studied use the Latin script. However, Chinese words do not consist of individual sounds (phonemes) such as letters do and can therefore not be represented by alphabetic letters. This raises three problems: First,
grapheme-phoneme associations in languages with an alphabetic writing system do not exist in Chinese which makes the use of alphabetic letters to represent the phonemes problematic. Second, with the use of alphabetic letters in this context, it cannot be assured that the Dutch language system has not been accidentally activated. Third, it is very uncommon to spell out or to pronounce individual sounds in languages in which there is no systematic
correspondence between the orthographic and phonologic units. Although this is sometimes done by the Chinese-Dutch youth, this is not common and moreover unstandardized.
With the above notions in mind, a second experiment was designed to address these shortcomings. In the second experiment, the phonemes were presented aurally instead of visually. In this way, a better unilingual setting was retained, and the problem of spelling out Chinese words was also accounted for. Moreover, the auditory experiment resembles a natural setting (i.e. such as in conversations) more than in Experiment 1, where the phonemes were presented visually.
Experiment 2: Auditory stimuli Method
Participants. Twenty-seven Cantonese-Dutch bilinguals (13 males, 14 females; mean age = 24.8, SD = 7.35) living in the Netherlands participated in this experiment. All had been born into monolingual Cantonese speaking families and learned Dutch in their early
childhood. However, four of the twenty-seven participants were removed before data analyses: one participant had also taken part in the first experiment; two participants were removed because they had stated that Dutch was not their dominant language; one participant did not follow the instructions. Another two participants had been removed before the data were analysed because they answered more than half of the items incorrectly. The data of the
remaining twenty-one Cantonese-Dutch bilinguals (11 males, 10 females; mean age = 24.1,
SD = 5.93) were further analysed. Participants received €7 as incentive.
For each language and each self-rated language ability (Reading and Speaking), a mean score and the standard deviation were calculated. The participants’ self-rated reading ability and speaking ability were then taken together to calculate the participants’ average knowledge for Cantonese and Dutch respectively. A t-test was run on these average rates to determine Dutch dominancy. A summary of the results is provided in Table 4.
Paired-samples t-tests were performed on the averages of the self-rated reading and speaking ability in Cantonese and Dutch respectively. The participants rated their Dutch reading ability higher than their Cantonese reading ability, t(20) = 6.136, p < .001. This result was also found for speaking ability: the participants rated their Dutch speaking ability higher than their Cantonese speaking ability, t(20) = 4.280, p < .001. To determine Dutch
dominancy, the averages of the self-rated language abilities were taken together per language. A paired-samples t-test was run on the average of these averages of the self-rated language abilities. The participants rated their average knowledge of Dutch higher than their average knowledge of Cantonese, making them unbalanced bilinguals, t(20) = 5.675, p < .001, two-tailed.
Table 4
Means and Standard Deviations (in Parentheses) of Participants’ Self-rated Reading and Speaking ability scores for Dutch and Cantonese
Reading Speaking Total
Dutch Cantonese 6.19 (0.75) 3.90 (1.87) 6.29 (0.56) 4.76 (1.48) 6.23 (0.58) 4.33 (1.51)
Materials and Design. The same set of pictures and phonemes as in Experiment 1 was used for Experiment 2. For each picture, the Cantonese and the Dutch names were recorded. The picture names were named by native speakers and they were instructed to articulate well. The voice samples were recorded in a soundproof room. The audio-editor program Audacity 2.0 was used for optimalizing the voice samples and the segmenting of the individual phonemes. For the experimental pictures, each phoneme was cut out of its original word to maintain the actual co-articulation of the sounds. For the affirmative condition, 50 phonemes were cut out of the recorded Cantonese picture names. For the negative conditions, the remaining 100 phonemes were cut out of the recorded Dutch picture names. For the fillers, the opposite was the case: 100 phonemes were cut out of the recorded Cantonese picture names (affirmative conditions) and the remaining 50 phonemes were cut out of the recorded Dutch picture names (negative condition). This counterbalancing of Cantonese and Dutch phonemes was done to reduce any possible influence of differences between the Cantonese and Dutch phonemes (i.e. voice pitch, phoneme length, clarity, etc). A Sennheiser HD 201 was used as headphone during the experiment. The number of trials was identical to that of in Experiment 1, namely 300.
Procedure. The experimental procedure in Experiment 2 followed the procedure used in Experiment 1. However, here it was explained that the participants would hear sounds that they had to monitor. As was done in Experiment 1, the phonemes were presented after the pictures were shown. The participants were instructed to respond as soon as possible, even if the sounds were too short to identify. After completion of the task, the participants filled out an exit interview.
Results
Trials answered out of time (before 200 ms and after 2500 ms) were removed before data analyses. This accounted for 2.7% of the trials. Reaction times were measured from the moment the phoneme was presented.
Two paired-samples t-tests by subjects (t1) and by items (t2) were conducted on the reaction times of both series together to determine differences between the two negative conditions. The analyses on the reaction times showed that the difference between the translation condition and control condition was significant, t1(20) = 2.258, p = .035; t2(24) = 3.481, p = .002. The same test was conducted on the corresponding error rates. It was found that the participants made more errors on the Dutch translation items than on the control items, but this effect was significant by subjects and not significant by items, t1(20) = 2.663, p = .015; t2(24) = 0.935, p = .359.
Table 5
Mean Reaction Times and Error Rates with accompanying Standard Deviations (in Parentheses) for Each Condition
Condition RT Error
Dutch Translation equivalent 1128 (342.1) 10.2 (2.97) Unrelated phoneme 1082 (308.4) 8.95 (2.61)
An additional analysis was performed to determine whether there was a difference in the pattern of results between the two series. A two-way ANOVA was performed on the reaction times by subjects (F1) and by items (F2), taking the factors series (first vs second) and condition (translation vs control) into account. The analyses of the reaction times showed an effect of condition, F1(1, 20) = 5.098, MSE = 8,712, p = .035; F2(1, 24) = 12.120, MSE = 13,746, p = .002, which shows that the participants responded slower on the Dutch translation
items than on the control items. We also found a main effect of series, F1(1, 20) = 16.481,
MSE = 33,067, p = .001; F2(1, 24) = 47.157, MSE = 15,293, p < .001. The participants were faster in the second series than in the first series. The interaction between condition and series was found to be significant by subjects, F1(1, 20) = 10.600, MSE = 6,076, p = .004; F2(1, 24) = 1.944, MSE = 13,412, p = .176. Inspection of Table 6 shows that the difference between the translation condition and control condition was numerically much larger for the second series than for the first series. For the error analyses, a main effect of condition was found by subjects but not by items, F1(1, 20) = 7.090, MSE = 4,540, p = .015; F2(1, 24) = .874, MSE = 30,936, p = .359. The participants made more errors when rejecting Dutch phonemes than when rejecting control phonemes. The main effect of series was significant by both subjects and items, F1(1, 20) = 27.536, MSE = 6,648, p < .001; F2(1, 24), = 46.866, MSE = 3,281, p < .001, showing that participants were faster in the second series than in the first series. The interaction between the two factors was not significant, showing that the amount of errors made on the Dutch phonemes compared to control phonemes was similar across series.
To further examine the above found patterns of results, two paired-samples t-tests were done on the two series separately. In the first series, it was found that the reaction times of the two negative conditions did not differ significantly from each other, t(20) = -.398, p = .695. The corresponding error rates were found to be different from each other, t(20) = 2.649,
p = .015, two-tailed. In the second series, the reaction times of the two conditions were found
to be different, t(20) = 3.473, p = .002. The corresponding error rates did not differ statistically from each other, t(20) = 1.353, p = .191.
Summarizing, the effect of condition was found in the second series, but not in the first series. The opposite pattern of results was found for the error rates: the effect of errors was only significant in the first series, but not in the second series.
Table 6
Subjects Mean Score, Standard Deviation and Average Amount of Errors per Condition, per Series
Series and Condition RT SD %Error
1st series Translation phoneme 1181 372.1 7.9 Unrelated phoneme 1190 355.9 6.8 2nd series Translation phoneme 1075 344.8 5.7 Unrelated phoneme 974 281.8 5.1 Discussion
It was found that the participants needed more time to reject Dutch phonemes than unrelated phonemes. This effect was found for the second series but not for the first series. Additionally, the participants made more errors on the Dutch phonemes than on the unrelated phonemes. Yet, this difference was only found in the first series and not in the second series.
Although the expected effects did not emerge in both series of the experiment, the results do suggest that Dutch was activated during Cantonese picture naming. The error rates, for example, were not found to be statistically different, but numerically more errors were made when the Dutch phonemes were presented than when the unrelated phonemes were presented. In general, this task was rated more difficult than Experiment 1 as can be seen by the relatively high error rates. Yet, the manipulation did have an effect on the reaction times, but apparently this needed more trials to become visible.
General discussion
The aim of the present study was to examine whether non-cognate words from languages with different orthographies become phonologically co-activated during word production. Two groups of Cantonese-Dutch bilinguals were presented with a phoneme monitoring task in which they had to indicate whether a specific phoneme appeared in the Cantonese name of a picture. The phonemes were presented in the visual (Experiment 1) and auditory modality (Experiment 2).
In both series of Experiment 1, longer latencies were found when subjects rejected Dutch translation items compared to unrelated items. Moreover, the error rate was higher on these Dutch translation items than on the unrelated items. In Experiment 2, the results were in a similar fashion. The expected delay in rejecting the Dutch translation items was found in the second series, while the difference in error rates was found in the first series.
The findings of the current study can be interpreted as evidence for non-selective phonological processing in different-script languages. This study provides a stronger test than other studies available in the literature in showing non-selective phonological access in different-script languages (e.g. Hoshino & Kroll, 2008; Moon & Jiang, 2012). As pointed out above, previous studies which investigated different-script languages involved languages with scripts that are basically similar to alphabetic languages. In Hoshino and Kroll’s for example, the Japanese Katakana script that was used has characters that represent sounds or
combinations of sounds. The Korean Hangul script in Moon and Jiang’s resembles the
alphabetic script even more, as these characters represent individual sounds such as phonemes do. Those different-script languages use graphemes that look different from those of the Latin script, but in essence the strong correspondence between graphemes and phonemes seen in alphabetic languages are also found in those languages.
More evidence from language production research is needed to validate the findings of the current study and thus language-nonselective phonological processing in different-script languages. Evidence found by recent priming studies are in line with the findings of the present study. Zhou, Chen, Yang and Dunlap (2010), for example, found cross-lingual phonological priming effects in a word-naming task and a lexical decision task with highly proficient and less proficient Chinese-English bilinguals. The bilinguals were primed with words from the non-target language that were phonologically similar or dissimilar to the target words in the intended language. Both tasks were employed in Chinese and English. Target words that were preceded by phonologically similar primes were responded faster to than to target words that were primed with phonologically dissimilar words, regardless of priming direction (L1-to-L2 direction or L2-to-L1 direction) and level of L2 proficiency.
Another priming study using ERP also found evidence for cross-lingual phonological access. In a semantic judgment task, highly proficient Chinese-English bilinguals had to indicate whether the presented word pairs in English were related (Wu & Thierry, 2010). Word pairs that were unrelated in English were actually repeated in either sound or spelling (i.e. character) in their Chinese translations. The word pairs were presented visually as well as aurally. In both modalities, the authors found a phonological priming effect, while they did not find an orthographical priming effect. More interestingly, these effects were not seen in the behavioural results (i.e. response times), but were seen in the ERP recordings only. The latter result indicates that processing an L2 activates its phonological representations but not necessarily its orthographical representations. Given that the task explicitly required the processing of sublexical phonological representations, these results further indicate that the co-activation of the non-target’s phonological representations occurs automatically in bilingual language processing. Together with Zhou and colleagues’ study, this study found
convincing evidence that in language recognition, the phonological representations of different-script languages are non-selectively accessed.
Although we did not take this into account, our subjects varied in their level of
Cantonese proficiency. Some subjects rated their L1 proficiency as poor on the exit-interview while others indicated to have a higher level of proficiency of their L1Cantonese, especially on the reading ability scale. The subjects of the two priming studies mentioned above were more balanced in their languages than our subjects were (although Zhou and colleagues also included bilinguals that were less proficient in their L2). If our subject groups were larger it might have been meaningful to see whether level of proficiency might have an effect on phoneme monitoring, or more generally, on bilingual phonological access. Anyhow, Zhou and colleagues did not find effects of L2 proficiency on the occurrence of the phonological
priming effect, suggesting that cross-lingual phonological access in language recognition is independent of L2 proficiency.
The finding of automatic co-activation of a bilingual’s non-target language’s
phonological representations has been interpreted as evidence for a shared mental lexicon of a bilingual’s languages (i.e. Moon & Jiang, 2012; Colomé & Miozzo, 2010). Indeed, if the language representations of the non-target language are completely independent from the language representations from the target language it would be unlikely to find activation of the former language. Here we showed that languages can have shared phonological
representations when languages do not share representations at the orthographical level (see also Hoshino & Kroll, 2008). The ERP study by Wu and Thierry (2010) supports this idea.
Given the finding that the non-target language is simultaneously activated with the target language, it remains open to question how a bilingual manages to select the language for production that the setting requires. If in a unilingual setting both languages of a bilingual are activated, this would make selecting the contextually appropriate language difficult as the
language representations from each language will compete for production. One might think that the activation level of the two languages differs, such as that the target language is activated slightly stronger compared to the non-target language. If so, selecting the appropriate language for production would go effortlessly (e.g. Starreveld, De Groot,
Rossmark, & Van Hell, 2013). If, however, the language representations of the target and the non-target language are activated equally strongly, a competition for selection is likely to take place. This competition needs to be resolved in order to be able to produce the target
language. This solution could be achieved by suppressing the language representations of the non-target language in order to select the language representations of the target language. In the latter case, inhibition mechanisms are likely to operate (i.e. Dijkstra, 2005 and Green, 1998). Alternatively, although there is non-selective activation of the target language and the non-target language, lexical selection might be confined to the lexical representations of the target language (i.e. language-selective selection). Since lexical selection is limited to the target language, the translation equivalent of the target (and related semantic representations in the non-target language) will not interfere with the selection of the target. As a result, this language-specific selection will lead to a facilitation in naming the target, because it becomes activated additionally through the activated translation equivalent but will not compete, and therefore will not interfere, for lexical selection (Costa, Miozzo, Caramazza, 1999).
Although the present study is significant for understanding bilingual language
production, our conclusions are limited to the production of isolated words. In conversational speech, words are surrounded by a meaningful context such as sentences and are rarely uttered in isolation. Evidence suggesting that sentence context has a modulating effect on word processing comes mainly from language recognition studies. In the word production field, context studies are scarce but are needed for better understanding of bilingual language processing. A recent attempt to address this issue in bilingual speech has been done by
Starreveld, De Groot, Rossmark, and Van Hell (2013). The authors employed a picture-naming task, in which the cognate status of the picture names was controlled for. The picture names (target words) were embedded in high- and low-constraining sentences. The
experiment was employed in L1 Dutch and in L2 English. In general, the authors found smaller cognate facilitation effects when the target words were presented in sentences compared to when presented in isolation. In other words, when the target words were
presented in a meaningful context, naming cognate words compared to non-cognate words did not lead to shorter response times. This modulating effect of context was strongest in high-constraining sentences (i.e. sentences that strongly constrained the possible target words) and when performed in the subjects’ (stronger) L1. According to the authors, sentence context modulates word production in two ways. First, sentence context leads to an overall increase in the activation level of the target language. The language representations of the target language will therefore have a slight advantage on those of the non-target language. Second, the degree of sentence constraint leads to the activation of more general semantic representations (in the case of low-constraining sentences) or more specific semantic representations (in high-constraining sentences) of the target picture’s name. The latter will also lead to the activation of (all or parts of) the semantic representations of the target (and eventually its corresponding phonological representations) so the target name becomes activated before it has been
presented pictorially. In other words, speakers are able to predict upcoming words if these words are surrounded by a highly meaningful context. Still, a cognate effect occurred in the high-constraining L2 condition. This finding has been explained as a result of the relatively lower English proficiency of the bilinguals compared to their L1 Dutch. The activation of the language representations of the bilinguals’ L2 might have taken more time than it would take in their L1. This is likely to result in a longer processing time and therefore the inability to name the picture before it is presented. To be more certain about the effect of context on
bilingual speech, other bilingual language combinations need to be considered, including different-script languages.
To conclude, the current study found evidence for non-selective processing of phonological representations in non-alphabetic, different-script languages. During covert picture naming in Cantonese Chinese, Dutch phonemes were found to be activated along with Cantonese phonemes. The findings were similar for visually presented phonemes as well as auditory presented phonemes. The current study extends previous studies in understanding bilingual language production as it showed that even in different-script languages, the phonological representations are non-selectively accessed.
References
Bates, E., D’Amico, S., Jacobsen, T., Szekely, A., Andonova, E., Devescovi, A., Herron, D., … Tzeng, O. (2003). Timed picture naming in seven languages. Psychonomic Bulletin
& Review, 10, 2, 344-380.
Colomé, A., (2001). Lexical activation in bilinguals’ speech production: Language-specific or language-independent? Journal of Memory and Language, 45, 721-736.
Colomé, A., & Miozzo, M. (2010). Which words are activated during bilingual word production? Journal of Experimental Psychology, 36, 96-109.
Costa, A., Caramazza, A., & Sebastián-Gallés, N. (2000). The cognate facilitation effect: Implications for the models of lexical access. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 26, 1283-1296.
Costa, A., La Heij, W., & Navarrete, E. (2006). The dynamics of bilingual lexical access.
Bilingualism: Language and Cognition, 9, 137-151.
Costa, A., Miozzo, M., & Caramazza, A. (1999). Lexical selection in bilinguals: Do words in the bilingual’s two lexicons compete for selection? Journal of Memory and Language,
41, 365-397.
Dell, G. S. (1986). A spreading activation theory of retrieval in sentence production.
Psychological Review, 93, 283-321.
Dijkstra, T. (2005). Bilingual word recognition and lexical access. In J. F. Kroll & A. M. B. de Groot (eds.), Handbook of bilingualism: Psycholinguistic approaches, pp. 179-201. New York: Oxford University Press.
Dijkstra, T., Van Jaarsveld, H., & Ten Brinke, S. (1998). Interlingual homograph recognition: Effects of task demands and language intermixing. Bilingualism: Language and
Grainger, J., & Dijkstra, T. (1992). On the representation and the use of language information in bilinguals. In R. J. Harris (Ed.). Cognitive processing in bilinguals (pp. 207-220). Amsterdam: Elsevier Science Publishers.
Green, D. W. (1998). Mental control of the bilingual lexico-semantic system. Bilingualism:
Language and Cognition, 1, 67-81.
Grosjean, F. (2001). The bilingual’s language models. In Nicol, J. (Ed.), One Mind, Two
Languages: Bilingual Language Processing (pp. 1-22). Oxford: Blackwell.
Hoshino, N., & Kroll, J. F. (2008). Cognate effects in picture naming: Does cross-language activation survive a change of script? Cognition, 106, 501-511.
Levelt, W. J. M. (1989). Speaking: From intention to articulation. Cambridge, MA: MIT Press.
Moon, J., & Jiang, N. (2012). Non-selective lexical access in different-script bilinguals.
Bilingualism: Language and Cognition, 15, 173-180.
Schwartz, A. I., Kroll, J. F., & Diaz, M. (2007). Reading words in Spanish and English: Mapping orthography to phonology in two languages. Language and Cognitive
Processes, 22, 106-129.
Spivey, M. J., & Marian, V. (1999). Cross talk between native and second languages: Partial activation of an irrelevant lexicon. Psychological Science, 10, 281-284.
Starreveld, P. A., De Groot, A. M. B., Rossmark, B. M. M., Van Hell, J. G. (2013). Parallel language activation during word processing in bilinguals: Evidence from word production in sentence context. Bilingualism: Language and Cognition
doi:10.1017/S1366728913000308
Van Hell, J. G., & De Groot, A. M. B. (1998). Conceptual representation in bilingual
memory: Effects of concreteness and cognate status in word association. Bilingualism:
Wu, Y. J., & Thierry, G. (2010). Chinese-English bilinguals reading English hear Chinese.
The Journal of Neuroscience, 30, 7646-7651.
Yum, Y. N., Holcomb, P. J., & Grainger, J. (2011). Words and pictures: An
electrophysiological investigation of domain specific processing in native Chinese and English speakers. Neuropsychologia, 49, 1910-1922.
Zhao, H., Chen, B., Yang, M., & Dunlap, S. (2010). Language nonselective access to
phonological representations: Evidence from Chinese-English bilinguals. The quarterly
Journal of Experimental Psychology, 63, 2051-2066.
Zhao, H. R., La Heij, W., & Schiller, N. O. (2012). Orthographic and phonological facilitation in speech production: New evidence from picture naming in Chinese. Acta
Appendix A
Experimental stimuli
Picture-Dutch name (English) Cantonese onset Dutch onset Unrelated
hee khau-ballon (balloon) /h/ /b/ /i/
phien fok-vleermuis (bat) /p/ /v/ /t/
mong juun keng-verrekijker (binoculars) /m/ /v/ /p/
woe tiep-vlinder (butterfly) /w/ /v/ /s/
hong loh paat-wortel (carrot) /h/ /w/ /s/
jien tjai-sigaret (cigarette) /j/ /s/ /b/
tsung liem-gordijn (curtain) /t/ /g/ /v/
suu thoi-bureau (desk) /s/ /b/ /v/
hoi thuun-dolfijn (dolphin) /h/ /d/ /z/
tsoi hong-regenboog (rainbow) /t/ /r/ /w/
sau tjie-vinger (finger) /s/ /v/ /h/
siew fong juun-brandweer (fireman) /s/ /b/ /k/
tsing wah-kikker (frog) /t/ /k/ /b/
thai tjie-druiven (grapes) /t/ /d/ /v/
tseu tjai-hamer (hammer) /t/ /h/ /p/
fong tjan-vlieger (kite) /f/ /v/ /b/
jien thong-schoorsteen (chimney) /j/ /s/ /d/
tho pa-zwabber (mop) /t/ /z/ /g/
mo koe-paddestoel (mushroom) /m/ /p/ /n/
mat yu-inktvis (octopus) /m/ /i/ /r/
thai to-scheermes (razor) /t/ /s/ /v/
jie tse-naaimachine (sewing machine) /j/ /n/ /v/
lo foe-tijger (tiger) /l/ /t/ /d/
faan khe-tomaat (tomato) /f/ /t/ /s/
kong kham-piano (piano) /k/ /p/ /t/
