Language switching and language competition

Abstract

This study examined the asymmetrical language switching cost in a word reading task (Experiment 1) and in a categorization task (Experiment 2 and 3). In Experiment 1, Spanish–English bilinguals named words in first language (L1) and second language (L2) in a switching paradigm. They were slower to switch from their weaker L2 to their more dominant L1 than from L1 to L2. In Experiment 2 and 3, high vs. low English proficiency bilinguals decided whether a word visually presented in their L1 or L2 referred to an animate or to an inanimate entity. In this case, bilinguals did not show asymmetrical cost when they switched between languages. These results suggest that inhibitory processes in bilingual processing as indexed by the asymmetrical language switching cost are only observed when L1 and L2 lexical representations compete for selection (e.g. word naming task). In addition, L2 proficiency did not influence the absence of asymmetrical switching cost.

Keywords

bilingualism language inhibition

I Introduction

Several studies have demonstrated that bilinguals activate both of their languages when they are performing a given task (non-selective processing; for reviews, see Macizo et al., 2010; Martín et al., 2010; see also Dijkstra, 2005; Kroll, 2006). The direct consequence of non-selective language processing is that bilinguals have to select a specific representation between competing alternatives across their languages. This competitive process has been widely corroborated (e.g. lexical decision task, Dijkstra et al., 1998; picture naming, Hermans et al., 1998). For example, Jared and Szucs (2002) found that French–English (L1–L2) bilinguals named interlingual homograph with different pronunciations in each language more slowly in L2 than control words suggesting that first language (L1) lexico-phonological forms competed during the word naming task. Similarly, other studies have shown that bilinguals activate spelling to sound correspondences in a non-target language that compete during word naming in a target language (e.g. Jared and Kroll, 2001).

Finkbeiner et al. (2006) have referred to this between-language activation and competition as the ‘hard problem’ in relation to bilingual lexical access during speech production. An appealing answer to this problem is the proposal of inhibitory processes acting to suppress non-target representations (although see Costa et al., 1999). The idea of inhibitory processes is included in several models of bilingual language processing (The Inhibitory Control model, IC model, Green, 1998; the Bilingual Interactive Activation model, BIA model, Dijkstra and van Heuven, 1998; for a review of these models, see Thomas and van Heuven, 2005). In the Inhibitory Control model (IC model; Green, 1998), language task schemas regulate the activation levels of representations in the bilinguals’ two languages according to the demands of the task that they perform. So, for example, when bilinguals are naming in their first language (L1), the language task schemas specifies the goal (to name in L1). Green claims that bilinguals experience between-language competition because of their non-selective processing and that they use inhibition to allow selection of the desired representations. This inhibitory mechanism is reactive because it takes place only after these representations have been activated. In addition, the amount of inhibition applied to competing representations is proportional to their degree of activation. When bilinguals are speaking in their weak second language (L2), candidates from their dominant L1 are strongly activated and, accordingly, they have to be strongly inhibited. Moreover, the linguistic representations that have been inhibited need time to overcome inhibition and recover their normal levels of activation. Therefore, if participants have to use the representations of a language that was previously inhibited, they will take more time to retrieve them than to retrieve the representations of a non-inhibited language.

The presence of inhibitory processes in bilingual language processing is supported in studies using language switching tasks in which participants are asked to name items (pictures, digits, words, sentences) presented in either their L1 or L2 (Ibáñez et al., 2010; Jackson et al., 2001; Meuter and Allport, 1999; for a review, see Abutalebi and Green, 2008). In these tasks there are trials in which the language of response is the same as that used in the preceding trial (non-switch trials) and trials in which the language of response is different from that used in the previous trial (switch trials). When bilinguals perform the switching tasks they are slower in switch trials as compared to non-switch trials. According to the IC model (Green, 1998), this switching cost results from the time required to change from the language schema used in the previous trial (e.g. naming in L2) to the new language schema, which specify the new task goal (e.g. to name in L1). Note that this general switching cost is not restricted to switching languages but it is common to many other switching tasks (e.g. to classify digits as either odd/even or high/low; for reviews, see Monsell, 1996; 2003). This observation suggests that the switching cost per se is not directly related to inhibitory processes, but that it originates outside rather than inside the bilingual linguistic system (e.g. Thomas and Allport, 2000). However, sometimes the cost observed in language switching studies is asymmetrical so switching from L2 to L1 is more costly than switching from L1 to L2. This asymmetrical cost can be easily interpreted in the framework of the IC model (Green, 1998) by assuming that inhibition is reactive and proportional to the level of activation of the representations that compete for selection and that have to be suppressed (Green, 1998). In the language switching task, the bilinguals have to strongly suppress their dominant L1 while they are naming in their weak L2. When bilinguals return to L1 naming, they have to overcome the large inhibition applied to those L1 lexico-semantic representations.

As we mentioned, Green (1998) suggests that inhibition is a reactive mechanism devoted to resolve between-language interference. For example, when Spanish–English bilinguals name the picture of a cat, the phonology of its Spanish translation ‘gato’ will compete for selection and inhibitory processes are articulated to select the correct phonology in L2 and say aloud the picture name (‘cat’). In monolingual studies, Anderson et al. (1994; for a review, see Anderson, 2003) have also proposed that inhibitory processes depend on the need to resolve competition when performing a given task. This premise has been confirmed in studies using the retrieval-practice paradigm. In the basic paradigm, participants study category-exemplar pairs (e.g. fruits–orange, fruits–banana, animals–dog). Afterward, they practise retrieving half of the exemplars from half of the studied categories (e.g. fruits–orange, but not fruits–banana, animals–dog) and, finally, they receive a test in which they recall all studied exemplars. The important result is that unpractised items from practised categories (e.g. fruits–banana) are recalled less often than baseline items (e.g. animals–dog). This result indicates that when participants retrieve concepts in the retrieval practice phase, inhibitory mechanisms suppress competing representations. More interesting, weak competitors (e.g. fruits–papaya) that are unlikely to compete during retrieval practice are less inhibited than stronger competitors (fruits–banana) or they may not even been inhibited (Anderson et al., 1994). Therefore, inhibition of non-target items seems to depend on whether they compete during the course of a given task. In studies with bilinguals it has been also proposed that competition is required to observe inhibition (but see Runnqvist and Costa, in press). Levy et al. (2007) adapted the retrieval-practice paradigm to the case of bilingual processing. In their study, English–Spanish bilinguals repeatedly named pictures in either their dominant language (English, e.g. snake) or their weak language (Spanish, e.g. serpiente). Afterward, participants had to retrieve the English names for the items using a rhyming test (a word was presented and participants had to provide a rhyming English word that matched a previously presented picture, e.g. break_). The retrieval of English words was poorer after repeatedly naming pictures in Spanish (retrieval-induced forgetting effect, RIF) as compared to a non-repetition control condition. This effect was explained by assuming that picture names in L1 competed for selection during the L2 picture-naming phase and, as a consequence, the corresponding L1 representations were inhibited. Thus, recall of the inhibited L1 words in the final rhyme test was impaired relative to the control condition. Moreover, this study suggests that inhibition occurred at the phonological level as evidenced in the rhyme test. In fact, when participants performed a final semantic test (to produce a word semantically related with the cue, e.g. venom_) there was no sign of inhibition.

In addition to between-language competition, previous studies have shown that inhibitory processes in bilinguals (as indexed by the asymmetrical language switching cost) depend on their L2 proficiency. Costa and Santesteban (2004) observed asymmetrical switching cost in groups of L2 learners that performed a switching naming task in their L1 and L2. The switching cost in L1 was larger than in L2 for these low proficient bilinguals. However, highly proficient bilinguals did not show asymmetry in the switching cost so that the magnitude of the cost associated to switching languages was the same for L1 and L2. Therefore, asymmetrical switching cost seems to be only observed in high proficient bilinguals.

In the current study we evaluated two factors that might determine the occurrence of inhibitory processes in between-language switching tasks: the presence of competition between L1 and L2 representations and the role of L2 proficiency. In Experiments 1 and 2, bilinguals with the same high English proficiency were evaluated. Since differences in Working Memory (WM) modulate performance in linguistic tasks (for a review, see Daneman and Merikle, 1996), these bilinguals were equated in WM amplitude. In Experiment 3 participated low L2 proficient bilinguals.

Experiment 1 was aimed to replicate and extend the pattern of asymmetrical switching cost observed in previous studies (e.g. Meuter and Allport, 1999). We used a word naming task that involves between-language competition that has not been used in previous studies. We expected larger cost when Spanish–English bilinguals switched from L2 to L1 as compared to the cost of switching from L1 to L2. In this task, the bilinguals’ two languages compete for selection (e.g. phonological coactivation in word reading tasks, Jared and Szucs, 2002). For example, when Spanish–English bilinguals have to name the English word ‘dog’, the phonological representation of its Spanish translation ‘perro’ becomes a potential candidate which should be inhibited in order to say aloud the correct English word ‘dog’. In Experiment 2 and 3 (with high and low English proficient bilinguals, respectively) we maintained the language switching paradigm and the same material of Experiment 1 in a categorization task in which L1 and L2 lexical forms (e.g. phonological representations) did not compete since they oriented to the same response based on the word meaning. In this task bilinguals were required to categorize visually presented words according to a living/nonliving dimension. To explore the inhibitory processes in bilingual language processing, words were randomly presented in the bilinguals’ L1 and L2. In this task there is no between-language competition. For example, when Spanish–English bilinguals have to categorize the English word ‘dog’ as an animate entity, its Spanish translation ‘perro’ would produce the same response (an ‘animate’ response in both cases). The predictions about the asymmetrical language switching cost in this task were straightforward: If inhibitory processes in bilinguals act regardless of between-language competition, larger switching cost would be observed when participants switch from categorizing words in L2 to L1 as compared to the cost of switching from L1 to L2. On the contrary, if inhibitory processes depend on the presence of between-language competition, no asymmetrical switching cost would be observed in the categorization task. In addition, the comparison among experiments would show the role of L2 proficiency. If asymmetrical switching cost is only observed in bilinguals with low L2 proficiency, this effect might be present in Experiment 3 (low English proficient bilinguals) but not in Experiments 1 and 2 (high English proficient bilinguals).

II Experiment 1: Word naming task

Experiment 1 was aimed to evaluate the possible asymmetrical language switching cost in a word naming task with the stimuli that we would use later in the categorization task of Experiment 2 and 3. Spanish–English bilinguals named aloud words that were pseudo-randomly presented in their two languages. Previous studies have observed asymmetrical switching cost in a picture naming task (Costa and Santesteban, 2004; Meuter and Allport, 1999). In this experiment we predicted the same pattern of results in a word naming task (larger language switching costs in L1 than in L2) because both tasks involve between-language competition. Thus, when bilinguals have to produce a word in one language, the phonology of its translation will be a competing response that has to be inhibited in order to produce the correct response in the desired language.

1 Method

a Participants

Twenty Spanish–English bilinguals from the English department at the University of Granada served as volunteers. The participants’ characteristics are reported in Table 1. The participants were asked to complete a language proficiency questionnaire on reading, writing, listening and speaking in Spanish (L1) and English (L2). The ratings ranged from 1 to 10, where 1 was not fluent and 10 was very fluent. The mean fluency in L1 was higher (9.06, SD = 0.87) than the mean fluency in L2 (7.49, SD = 1.09), p < .001. This difference suggests that participants were highly fluent in English but dominant in Spanish.

Table 1

Characteristics of participants in the experiments (standard deviations in parentheses)

	Experiment 1	Experiment 2	Experiment 3
General:
Age (years)	21.85 (1.50)	22.50 (1.85)	21.75 (1.80)
Age starting L2 learning (years)	8.95 (2.39)	9.60 (4.11)	8.30 (1.38)
Months living in L2 speaking countries	9.03 (6.71)	11.05 (6.25)	0.17 (0.37)
Language proficiency questionnaire:
L1 Speech fluency	9.05 (1.10)	8.70 (1.22)	8.50 (1.43)
L2 Speech fluency	7.70 (2.36)	8.10 (1.68)	5.10 (1.62)
L1 Speech comprehension	9.30 (1.03)	9.53 (0.64)	9.25 (0.79)
L2 Speech comprehension	7.50 (1.24)	7.75 (0.97)	5.15 (1.69)
L1 Writing proficiency	8.80 (1.24)	8.63 (1.22)	8.55 (1.10)
L2 Writing proficiency	7.15 (1.04)	7.15 (1.04)	5.70 (1.30)
L1 Reading proficiency	9.10 (0.79)	8.98 (1.08)	8.75 (1.29)
L2 Reading proficiency	7.60 (0.88)	7.40 (0.88)	6.00 (1.38)

Notes: The self-report ratings of speech fluency, speech comprehension, writing proficiency and reading proficiency in L1 (Spanish) and L2 (English) ranged from 1 to 10, where 1 was not fluent and 10 was very fluent

In order to equate WM span of participants in Experiment 1 with those participating in Experiment 2, the bilinguals were asked to complete a Spanish version of the Reading Span Test (Daneman and Carpenter, 1980) before the experiment. In this test, participants are instructed to read sets of sentences presented one-by-one and to recall the sentence-final word after reading each set. The sets range from 2 to 6 sentences. The maximum number of final words correctly recalled represents the participants reading span where 2 indicates very low span and 6 indicates very high span. The participants’ mean span score was 3.25 (SD = 1.08).

b Design and Materials

The language of response (L1-Spanish and L2-English) and the language switching (switch trials and non-switch trials) were manipulated within-participants. A set of 144 words was selected for the experiment. These words were instances of semantic categories. Six categories referred to living things (e.g. fruits, vegetables, etc.) and six categories referred to non-living things (e.g. tools, kitchen utensils, etc.). Twelve words were selected from each category. These words were non-cognates in Spanish and English. Lexical characteristics of stimuli are reported in Table 2.

Table 2

Characteristics of stimuli used in the experiments (standard deviations in parentheses)

	Spanish Words (L1)	English Words (L2)
Number of letters	6.40 ( 1.95)	5.53 ( 2.01)
Number of phonemes	6.10 ( 1.87)	5.88 (11.49)
Number of syllables	2.65 ( 0.96)	1.67 ( 0.85)
Lexical frequency	30.73 (68.63)	35.13 (71.33)
Orthographic neighbours	3.13 ( 4.97)	4.67 ( 5.21)
Phonological neighbours	5.28 ( 7.53)	10.47 ( 9.45)

Sources: English lexical frequency from Brysbaert and New (2009; one-million count); Spanish lexical frequency from Alameda and Cuetos (1995, count transformed to one-million). Orthographic and phonological neighbours from David (2005; English words) and David and Perea (2005; Spanish words).

Participants were presented four lists of words. Each list contained the 144 experimental words described above. In each list there were trials in which the language of response was the language used in the preceding trial (non-switch trial) and trials in which the language of response was different than that used in the preceding trial (switch trials). The words in a switch trial and the words in the preceding non-switch trial were always from different categories. Each word was presented one time in each list so across the experiment each word appeared in all the experimental conditions (Spanish/switch, English/switch, Spanish/non-switch, English/non-switch).

In total, each participant was presented 576 trials (50% non-switch trials and 50% switch trials). Participants responded in Spanish in half of the non-switch trials and in half of the switch trials. They responded in English in the rest of the trials. For each participant the experiment started with a filler trial to be named in either Spanish or English.

c Procedure

The experiment was controlled by a Genuine-Intel compatible 2,993 MHz PC using E-prime experimental software, 1.1 version (Schneider et al., 2002). Participants were tested individually. They were informed that words would appear in Spanish and English in an unpredictable manner and that they had to name them as fast as possible while trying not to make errors. Participants were seated approximately 600 mm from the computer screen. Stimuli were presented in lower-case black letters (Courier New font, 18 point size) on a white background. At this viewing distance, one character subtended a vertical visual angle of 0.48 degrees and a horizontal visual angle of 0.67 degrees. In each trial the word to be read aloud was presented in the middle of the screen until the participant’s response. Afterward, there was a blank interval of 1,000 ms before starting the next trial. Response latencies were collected using a microphone ATR 20 with low impedance connected to a PST Serial Response Box (Psychology Software Tools) and tape-recoded to eliminate trials with errors in the latency analyses.

Participants practised with the word naming task before performing the experiment. These practice trials contained 15 words in Spanish and 15 words in English referring to instances of categories that were not used in the main experiment. The Spanish and English practice trials were randomized so that participants practised with switch and non-switch trials before the experiment.

2 Results and discussion

For the analyses of both reaction times (RTs) and accuracy, two analyses of variance (ANOVAs) were performed, one with participants as the random variable (F ₁) and another with items as the random factor (F ₂). Only correct responses were included in the analyses of the RTs. Data points were excluded from the RT analyses if:

the participants stuttered or hesitated in reading aloud the word;

the participants misread or failed to read the word;

the participants produced nonverbal sounds that triggered the voice key.

Following these criteria, 3.34% of the data points were discarded from the RT analyses. Also, RT data that exceeded a criterion of 3 standard deviations (SD) for an individual participant’s mean (0.59% of the data) were excluded from the analyses. The observations in categories (1), (2) and (3) were submitted to the accuracy analyses.

In the analyses of RTs, the main effect of language was significant, F₁(1, 19) = 10.30, p < .01, F₂(1, 143) = 13.27, p < .001. Words in L2 were named slower (575 ms, SE = 15.02) than words presented in L1 (559 ms, SE = 16.56). The main effect of language switching was also significant, F₁(1, 19) = 21.61, p < .001, F₂(1, 143) = 55.48, p < .001. RTs were slower for switch trials (576 ms, SE = 17.38) than for non-switch trials (558 ms, SE = 14.21). Crucially, the interaction between language and switching was significant, F₁(1, 19) = 8.25, p < .01, F₂(1, 143) = 11.49, p < .001. The switching cost was significant for both the participants’ L1, F₁(1, 19) = 28.64, p < .001, F₂(1, 143) = 58.46, p < .001, and the participants’ L2, F₁(1, 19) = 4.06, p < .05, F₂(1, 143) = 7.85, p < .01. However, the magnitude of the switching cost was larger in L1 (27 ms) than in L2 (10 ms) (see Figure 1).

Figure 1

Mean reaction times in milliseconds (lines) and mean errors in percentage (bars) as a function of language switching and language of response (L1 = Spanish, L2 = English) in the word naming task (Experiment 1)

Concerning analyses performed on the error data, the effect of language was not significant, F₁(1, 19) = 1.41, p > .05, F₂(1, 143) < 1. The percentage of errors was 2.99% (SE = 0.44) and 3.70% (SE = 0.55) when participants responded in Spanish and English, respectively. The effect of language switching was not significant, F₁(1, 19) = 1.33, p > .05, F₂(1, 143) = 2.60, p > .05. The percentage of errors in switching trials (3.61%, SE = 0.41) was similar to the percentage of errors in non-switching trials (3.07%, SE = 0.50). In addition, the Language × Switch interaction was not significant, F₁ and F₂ < 1. In order to deal with a possible problem of unequal variance across experimental conditions, mean error proportions for each participant in each experimental condition were also computed, and the arcsine transformation of these values was introduced in a new analyses. The outcome of this analysis was identical to the one obtained on the raw data. Neither main effects of language and switch nor the interaction between these variables were significant (all ps > .05).

The results of Experiment 1 showed asymmetrical pattern of language switching cost. Spanish–English bilinguals were slower when they switched to L1 than when they switched to L2. The interaction between language of response and language switching was similar to the results observed in other studies using the language switching paradigm in a number naming task (e.g. Meuter and Allport, 1999) and extend them to word naming. The presence of asymmetrical switching costs is interpreted as an index of inhibitory processes acting on items from the language that is not used in a given word naming trial. In our opinion, the between-language alternation in the word naming task used in this experiment promoted a strong competition between L1 and L2 representations. When bilinguals named a word in their L2, its Spanish translation directly competed and oriented to a different response. In this situation, strong inhibition of the irrelevant L1 candidate was required to overcome interference.

However, the results obtained in this study contrast with that reported by Costa et al. (2006) which did not find asymmetrical switching cost in highly proficient bilinguals. It is possible that differences between studies were due to the balance of L1–L2 proficiency. Costa et al. evaluated balanced bilinguals (e.g. Spanish–Basque, Spanish–Catalan) with similar proficiency in their two languages. In contrast, the Spanish–English bilinguals that participated in Experiment 1 were unbalanced so they were highly fluent in L2 but dominant in L1. Thus, it might be possible that the relative balance of L1–L2 proficiency modulates the asymmetrical switching cost in language production tasks.

III Experiment 2: Categorization task with high proficient English (L2) bilinguals

The goal of Experiment 2 was to evaluate whether inhibitory processes in bilingual language processing (as indexed by the asymmetrical language switching cost) can be observed in a task that does not involve between-language competition.

The language switching paradigm has been previously used in categorization tasks but, however, the possible asymmetry in language switching has not been addressed (Caramazza and Brones, 1980; Mason, 1994; Potter et al., 1984). Caramazza and Brones presented category-exemplars pairs of words and Spanish–English bilingual participants decided whether the exemplar belonged to the category. Word pairs were in the same language or in different language. The bilinguals took similar time to perform the task with single language pairs than with mixed language pairs. Therefore, no evidence of language switching cost was observed. However, this previous study using the categorization task was not specifically designed to explore language switching cost. Von Studnitz and Green (2002) systematically explored the pattern of switching cost in a categorization task in which German-English bilinguals decided whether a word denoted an animate or an inanimate entity. In this case, switching cost effects were observed. However, the possibility of inhibitory processes in language comprehension tasks as indexed by the asymmetry in the switching cost was not considered in the Von Studnitz and Green’s study. In fact, they did not report the switching cost for the bilinguals’ L1 and L2 separately.

In Experiment 2 we addressed the possible asymmetrical switching cost in a categorization task. Spanish–English bilinguals were required to decide whether visually presented words denoted living or non-living things. To explore the asymmetrical language switching cost, the language of response (L1 and L2) and the language switching (switch vs. non-switch trials) were manipulated. Because between-language competition was not involved in this task, we expected that the pattern of switching cost would not be asymmetrical. In a living-nonliving categorization task, when participants are presented with a word in one of their languages (e.g. the correct response for ‘cat’ is ‘animated’) the response associated to their alternative language is the same (the Spanish translation of ‘cat’, ‘gato’ drives to the same response ‘animated’) and therefore, no competition is present.

1 Method

a Participants

Twenty Spanish–English bilinguals participated in Experiment 2. They did not take part in Experiment 1. We attempted to match them in L1–L2 language fluency and WM capacity to the participants in Experiment 1 (see Table 1). The participants’ mean fluency in L1 and L2 was 8.96 (SD = 0.88) and 7.40 (SD = 0.88), respectively. The mean reading span score for the total set of participants was 3.65 (SD = 1.27). A series of t-tests was performed to compare language proficiency questionnaire on reading, writing, listening and speaking in Spanish (L1) and English (L2) and WM span of the participants in the two experiments. All of these comparisons yielded p values > .05. Hence, both groups of bilinguals were similar in language proficiency and WM span.

b Design and materials

The materials and design used in this experiment were the same as those used in Experiment 1. In Experiment 2 participants performed a categorization task in which they indicated whether words referred to living or non-living entities.

c Procedure

Participants were tested individually. They were informed that some Spanish and English words would be presented and that they had to decide whether they denoted animate or inanimate entities. The computer keyboard was used to collect responses. Half of the participants had to press the ‘m’ button with the right hand for ‘animate’ and the ‘z’ button with the left hand for ‘inanimate’. The rest of participants received the animate-left hand and inanimate-right hand combination. In each trial a word was presented in the middle of the screen until the participant’s response. Afterward, a 1000 blank interval was introduced before the next trial. The experimental trials were preceded by 30 practice trials none of which appeared in the experimental trials. All other details of the procedure were identical to those in Experiment 1.

2 Results and discussion

The percentage of errors in the categorization task was 16.54%. Only correct responses were included in the analyses of the RTs. The RTs exceeding a criterion of 3 SD for an individual participant’s mean were excluded from the latency analysis (1.61% of the data).

In the analyses of RTs, the main effect of language was not significant in the participants analysis, F₁ < 1, but it was marginally significant in the items analyses, F₂(1, 143) = 3.40, p < .06. The RTs in the categorization task was 1,676 ms (SE = 71.48) and 1,688 ms (SE = 67.76) when participants performed the task in Spanish and English, respectively. The main effect of language switching was significant, F₁(1, 19) = 83.29, p < .001, F₂(1, 143) = 54.56, p < .001. RTs were slower for switching trials (1,747 ms, SE = 70.31) than for non-switching trials (1,618 ms, SE = 67.34). In contrast to the results of Experiment 1, the interaction between language and switching was not significant; F₁ and F₂ < 1 (see Figure 2). The magnitude of switching cost was 139 ms for L1 and 120 ms for L2.

Figure 2

Mean reaction times in milliseconds (lines) and mean errors in percentage (bars) as a function of language switching and language of response (L1 = Spanish, L2 = English) in the categorization task (Experiment 2)

In the accuracy analysis, the effect of language was significant, F₁(1, 19) = 26.30, p < .001, F₂(1, 143) = 25.06, p < .001. The percentage of errors was larger when participants performed the task in English (20.33%, SE = 1.95) than in Spanish (12.76%, SE = 1.38). In addition, the effect of language switching was significant, F₁(1, 19) = 42.85, p < .001, F₂(1, 143) = 69.40, p < .001. The percentage of errors was larger in switching trials (19.20%, SE = 1.66) than in non-switching trials (13.89%, SE = 1.49). As in the latency analysis, the Language × Switch interaction was not significant, F₁(1, 19) = 1.28, p > .05, F₂(1, 143) = 1.85, p > .05. The analysis performed on the arcsine transformed error proportions produced the same pattern as the analyses on the raw data.

In Experiment 2, we observed a main effect of switching languages so that participants took more time to make a semantic decision when there was a change in the language of response. However, this language switching cost was not modulated by the direction of the language switch, L1–L2 switches and L2-L1 switches produced similar cost. The main effect of switching languages is similar to that observed in other switching tasks that do not require language processing (e.g. Monsell, 2003), supporting the notion that this effect arise outside the bilingual lexicon. Therefore, no signs of inhibitory processes were observed when bilinguals performed a categorization task.

However, it could be argued that the absence of asymmetrical switching cost was due to the fact that bilinguals in Experiment 2 were very proficient in L2. Thus, it might be possible that asymmetrical switching cost as index of inhibitory processes would apply only for low proficient bilinguals in comprehension tasks. For example, in an event-related potential (ERP) study, Jackson et al. (2001) evaluated low L2 proficient bilinguals in a language switching comprehension task (parity judgment) and switching cost was only observed in the participants’ L1. In the next experiment we evaluated whether asymmetrical switching cost in a semantic categorization tasks could be observed in bilinguals with low L2 proficiency.

IV Experiment 3: Categorization task with low proficient English (L2) bilinguals

The goal of Experiment 3 was to evaluate whether the absence of asymmetrical switching cost in the categorization tasks was due to the high L2 proficiency of bilinguals that participated in Experiment 2. To this end, in Experiment 3 Spanish–English bilinguals with less L2 proficiency than those of the previous experiment performed the categorization tasks. If L2 competence determines asymmetrical switching cost in the categorization tasks, participants of Experiment 3 might show this pattern of results. Otherwise, they should behave as L2 proficient bilinguals of Experiment 2.

1 Method

a Participants

Twenty Spanish–English bilinguals participated in the experiment. They did not take part in previous experiments. Participants were matched in L1 proficiency (reading, writing, listening and speaking) to those of Experiment 2 (all ps > .05) (see Table 1). However, participants of Experiment 3 were low proficient in L2 relative to bilinguals of the previous experiment (all ps < .05). In addition, the participants of this experiment had lower WM scores (2.45, SD = 0.90) than those of Experiment 2 (3.25, SD = 1.08), t(38) = 3.44, p < .05.

b Design and materials

The materials and design used in this experiment were the same as those used in Experiment 2.

c Procedure

The procedure was the same as that in Experiment 2.

2 Results and discussion

The percentage of errors was 18.37%. Only correct responses were included in the analyses of the RTs. The RTs exceeding a criterion of 3 SD for an individual participant’s mean were excluded from the latency analysis (1.67% of the data).

The latency analyses showed a main effect of language, F₁(1, 19) = 48.32, p < .001, F₂(1, 143) = 201.26, p < .001. Participants were faster when they performed the task in L1 (797 ms, SE = 45.33) relative to L2 (919 ms, SE = 55.19). In addition, the main effect of language switching was significant by participants, F₁(1, 19) = 6.31, p < .05, and marginal by items, F₂(1, 143) = 2.79, p = .09. The RTs in switching trials and non-switching trials were 868 ms (SE = 49.08) and 848 ms (SE = 50.66), respectively. More importantly, as in Experiment 2, the interaction between language and switching was not significant, F₁ and F₂ < 1 (see Figure 3). The magnitude of switching cost in L1 was 11 ms and it was 28 ms in L2.

Figure 3

In the accuracy analysis, the effect of language was significant, F₁(1, 19) = 174.53, p < .001, F₂(1, 143) = 55.72, p < .001. The percentage of errors was larger when participants performed the task in L2 (24.93%, SE = 2.23) than in Spanish (11.74%, SE = 2.38). In addition, there was a marginal effect of language switching by participants, F₁(1, 19) = 3.90, p = .06, but not by item, F₂(1, 143) = 1.66, p > .05. The percentage of errors in switching trials and non-switching trials was 18.59% (SE = 2.25) and 18.07% (SE = 2.26), respectively. As in the latency analysis, the Language × Switch interaction was not significant; F₁ and F₂ < 1. The analysis performed on the arcsine transformed error proportions produced the same pattern as that on the raw data.

In Experiment 3 low proficient L2 bilinguals showed symmetrical switching cost as high proficient L2 bilinguals of Experiment 2. However, there were between experiment variations in the pattern of results. First, high proficient bilinguals presented slower overall RTs (1,682 ms) than low proficient bilinguals (858 ms), t(38) = 9.74, p < .001. In addition, the symmetrical language switching cost was larger in high proficient bilinguals (129 ms) relative to low proficient bilinguals (19 ms), t(38) = 6.78, p < .001. It might be possible that these differences were due to larger within-groups variability in the case of high proficient bilinguals relative to that of low proficient bilinguals. In fact, t-test comparisons were conducted on the variability shown by RT data and results showed higher variability for the group of high L2 proficient bilinguals (SD = 713) than for the group of low L2 proficient bilinguals (SD = 310), t(38) = 5.30, p < .001. However, the main point is that symmetrical switching cost was found regardless of L2 proficiency of the bilinguals.

In addition, to further evaluate that differences in overall RT did not determine the pattern of switching cost, we performed new analyses. Participants of Experiments 2 were divided in fast/slow responders and the asymmetrical switching cost was evaluated. The group (fast/slow responders) did not interact with language (Spanish and English) or Switch (switch and non-switch); and the Group × Language × Switch second-order interaction was not significant (all ps > .05). Thus differences in overall RT might not explain the absence of asymmetrical switching cost.

The absence of asymmetrical switching cost observed in Experiment 3 when low fluency bilinguals performed a comprehension task contrast with those reported by Jackson et al. (2001) in which low fluency bilingual only showed switching cost in L1. The discrepancies between studies might be due to methodological differences between them. Jackson et al. used a parity task while we used a semantic categorization tasks. More importantly, in Jackson et al.’s study switching between trials consisted of fully predictable sequences in which non-switching trials were presented in pure blocks while switching trials were presented in mixed block (in which the alternating runs were also predictable so language changed after two trials). It has been previously observed that switching cost can be reduced under some circumstances when switches are predictable (Dreisbach and Haider, 2006). On the contrary, in our study switches between languages were completely random. Nevertheless, we agree with Jackson et al.’s (2001) suggestion that switching cost observed in their comprehension task emerged outside the lexicon (i.e. the authors did not observe frontal N2 modulation in their parity task, an ERP component that has been considered an index of inhibitory processes in language switching studies; Verhoef et al., 2009). According with this interpretation, the switching cost observed in our categorization task was not associated to inhibitory processes but rather it was similar to other switching costs found in non-linguistic tasks due to task reconfiguration (Rogers and Monsell, 1995).

V General discussion

In three experiments, we examined whether competition is required to observe inhibition in language switching tasks (word naming task, Experiment 1 and categorization task, Experiment 2 and 3). Previous research has demonstrated that bilinguals non-selectively activate representations in both of their languages when they are performing linguistic tasks (for reviews, see for example Dijkstra, 2005; Kroll et al., 2008). When active representations guides to a different responses, this non-selective activation produces competition between active candidates in the bilinguals’ two languages, which seems to be solved by inhibiting non-target representations (although for an alternative non-inhibitory explanation, see Costa et al., 1999). These inhibitory processes have been incorporated in the Inhibitory Control model (IC model, Green, 1998). In this model, language task schemas regulate the level of activation of the bilinguals’ two languages. The IC model claims that when bilinguals experience between-language competition they use inhibition to select the desired representations. Moreover, inhibition is reactive, so that it is only triggered when the two representations are activated and compete. The results of Experiment 1 extended previous studies supporting the inhibitory hypothesis using a language switching paradigm (e.g. Meuter and Allport, 1999). Spanish–English bilinguals were asked to read aloud words randomly presented in either their L1 or L2. The critical effect was that bilinguals took more time to switch from their weak L2 to their dominant L1 than vice versa. This result supports predictions from the IC model.

However, the pattern of results obtained in Experiment 1 was not completely similar to that found by Meuter and Allport (1999). The authors found slower naming latencies after L1 switches relative to L2 switches, while we observed that L2 naming was slower than L1 naming in repeated (non-switching) trials. This discrepancy can be explained by an L1–repeat–benefit hypothesis (e.g. Verhoef et al., 2009). However, the main point is that asymmetrical switching cost was obtained in both studies.

Experiment 2 aimed to investigate the role of competition in constraining inhibitory processes indexed by the asymmetrical switching cost. Green (1998) suggested that inhibition is a reactive mechanism devoted to resolve interference arising from the competition between active representations. Thus, from this perspective, competition is required to trigger inhibition. This proposal correctly explains the inhibitory effect observed in the word naming (Experiment 1). In Experiment 1, the bilinguals had to strongly suppress competing candidates in their L1 while they read aloud words in their weaker L2 because the non-selective activation of their two languages drives them to different responses (‘perro’ vs. ‘dog’). When bilinguals returned to L1 word naming, they had to overcome the large inhibition applied to these competing representations.

In Experiment 2 we used a task that did not promote between-language competition. A new set of Spanish–English bilinguals performed a categorization task in which they decided whether a word referred to an animate or to an inanimate entity. In this task there is no between-language competition because the non-selective activation of a word in L1 and L2 orient to the same response in the categorization task (e.g. the English word ‘dog’ and its Spanish translation ‘perro’ guide to an ‘animate’ response). In this case, the bilinguals showed a switching cost effect that was similar in L1 than in L2. This symmetrical switching cost is similar to that observed in many switching tasks that do not require language processing so this effect seems to arise from processes outside the bilinguals’ lexicon. Thus, the results of Experiment 2 seem to suggest that competition is required in order to observe inhibition. The activation of two L1–L2 lexico-semantic representations is not enough to trigger inhibition; competition between the activated representations is required for inhibition to act. The results of Experiment 2 agree with the view that inhibition depends on the need to resolve competition (Anderson, 2003; Anderson et al., 1994): The items that strongly compete for selection are more vulnerable to inhibition. The present results support this hypothesis and extend it to the case of between-language switching tasks.

In addition, across experiments we evaluated the role of L2 proficiency in between-language tasks. There was no evidence of asymmetrical switching cost when high L2 proficient bilinguals (Experiment 2) and low L2 proficient bilinguals performed the categorization task. Therefore, these findings suggest that L2 proficiency might not modulate asymmetrical switching cost (see Gollan and Ferreira, 2009) at least in comprehension tasks.

The presence of an asymmetrical language switching cost in the word naming task (Experiment 1) and its absence in the categorization task (Experiment 2 and 3), could be explained by the BIA + model (Dijkstra and van Heuven, 2002). In the word naming tasks, lexical nodes would compete for selection and inhibitory connections (between lexical nodes and from language nodes to lexical nodes) would suppress irrelevant representations. In the categorization tasks, semantic systems would be mainly involve, lexical nodes would not compete for selection and no inhibitory processes would act.

In addition, the pattern of results can be accommodated in an modified version of the revised hierarchical model (RHM) proposed by Kroll and Stewart (1994; see also Green 1998, p. 72). In this model, there are independent levels of processing for lexical forms in L1, lexical forms in L2 and meaning (see Figure 4). All levels are connected via bidirectional excitatory links. However, the strength of these connections varies. The relation between L1 and L2 lexical representations is asymmetrical so that the link from L2 to L1 is stronger than the link between L1 and L2. In addition, the strength between lexical and semantic representations also depends on the language. There are stronger links between L1 lexical forms and their meanings than the links between L2 lexical forms and their meanings. There is strong support for the asymmetry in these excitatory connections for both lexical and semantic information. The asymmetry in the excitatory connections at the lexical level is supported by the fact that backward translation (L2-L1) is performed faster than forward translation (L1–L2) (e.g. Kroll and Stewart, 1994). The strong semantic link between L1 lexical forms and meaning is evidenced, for example, in the observation that forward translation is affected by semantic variables while backward translation remains unaffected by these semantic factors (e.g. blocking the category of words to be translated, Kroll and Stewart, 1994; Sholl et al., 1995). The asymmetrical language switching cost observed in Experiment 1 can be accommodated in the RHM by incorporating inhibitory links between L1 and L2 lexical representations with the same strength as those supposed for excitatory connections (see Figure 4).

Figure 4

An adapted version of the revised hierarchical model (Kroll and Stewart, 1994) that includes inhibitory lexical links

In Experiment 1, when Spanish–English bilinguals were reading aloud L2 words (e.g. ‘dog’), L1 lexical representations were also activated via strong excitatory links from L2 to L1. Since these representations oriented to different responses (to say ‘dog’ in English or ‘perro’ in Spanish), bilinguals experienced between-language competition. Afterward, to finally select the correct response in L2, bilinguals used inhibitory connections that strongly suppressed the activated competitors in L1. In L2 to L1 switching trials, participants needed additional time to overcome the inhibition applied to L1 representations. The stronger inhibitory effect in switching from L2 to L1 than vice versa could be easily explained in the model because of the asymmetrical strength of the inhibitory connections between the bilingual’s two lexicons (stronger L2 to L1 inhibitory links).

On the contrary, the non-selective activation of the bilinguals two-languages did not favour between-language competition in the categorization task. For example, when participants had to indicate that ‘dog’ referred to an animate entity, it is possible that its Spanish translation ‘perro’ was also activated via strong excitatory links from L1 to L2. However, these two alternatives (‘dog’ and ‘perro’) did not compete since they were linked to the same semantic property (e.g. ‘animate’ entity). Thus, there was no competition in the categorization task and no inhibitory processes were needed to perform it.

An intriguing question to pose in the current study is the level in which between language competition and inhibition occurred. In our opinion, bilinguals activated phonological information of non-target words in the language production tasks (word naming, Experiment 1) and the comprehension task (word categorization, Experiments 2 and 3). This assumption is supported by studies showing that phonological information in the non-intended language is non-selectively activated during comprehension (e.g. lexical decision task; Dijkstra et al., 1999) and production tasks (e.g. phoneme monitoring tasks; Colomé, 2001). However, while phonological representations competed in the naming tasks (i.e. only one phonological form had to be selected to be produced aloud), they did not in the categorization tasks (any lexical information of words converged to determine that they denoted a living/non-living entity). Note however, we are not defending that inhibitory processes occurs at the lexico/phonological level only. We assume that inhibitory processes in language selection act when the activated representations compete to correctly perform a linguistic task.

VI Conclusions

This study suggests that inhibitory processes in bilingual language processing as indexed by the asymmetrical switching cost depend on the presence of between-language competition. This asymmetrical cost is observed in a word naming task where L1 and L2 representations drive to different responses. On the contrary, it is not observed in a categorization task where competition is not present and L1–L2 activation drives to the same response. Thus, this study suggests that the mere activation of two languages when bilinguals perform a task does not suffice to promote inhibitory processes. Instead, between-language competition seems to be required in order to observe inhibition.

Footnotes

Acknowledgements

This research was supported by the following grants: PSI2009-11094/PSI to Pedro Macizo, EDU2008-01111 to Teresa Bajo, and CSD2008-00048 Consolider Ingenio 2010; Proyecto de Excelencia de la JA-2007 and JA-2008_HUM 360 to Teresa Bajo and Pedro Macizo.

References

Abutalebi

Green

(2008) Control mechanisms in bilingual language production: Neural evidence from language switching studies. Language and Cognitive Processes 23: 557–82.

Alameda

Cuetos

(1995) Diccionario de frecuencias de las unidades lingüísticas del castellano [Frequency dictionary of Spanish linguistic units]. Oviedo: Servicio de Publicaciones de la Universidad de Oviedo.

Anderson

(2003) Rethinking interference theory: Executive control and the mechanisms of forgetting. Journal of Memory and Language 49: 415–45.

Anderson

Bjork

(1994) Remembering can cause forgetting: Retrieval dynamics in long-term memory. Journal of Experimental Psychology: Learning, Memory and Cognition 20: 1063–87.

Brysbaert

New

(2009) Moving beyond Kučera and Francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for American English. Behavior Research Methods 41: 977–90.

Caramazza

Brones

(1980) Semantic classification by bilinguals. Canadian Journal of Psychology 34: 77–81.

Colomé

(2001) Lexical activation in bilinguals’ speech production: Language-specific or language-independent? Journal of Memory and Language 45: 721–36.

Costa

Santesteban

(2004) Lexical access in bilingual speech production: Evidence from language switching in highly proficient bilinguals and L2 learners. Journal of Memory and Language 50: 491–511.

Costa

Miozzo

Caramazza

(1999) Lexical selection in bilinguals: Do words in the bilingual’s two lexicons compete for selection? Journal of Memory and Language 41: 365–97.

10.

Costa

Santesteban

Ivanova

(2006) How do highly proficient bilinguals control their lexicalization process? Inhibitory and language-specific selection mechanisms are both functional. Journal of Experimental Psychology 32: 1057–74.

11.

Daneman

Carpenter

(1980) Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behaviour 19: 450–66.

12.

Daneman

Merikle

(1996) Working memory and language comprehension: A meta-analysis. Psychonomic Bulletin and Review 3: 422–33.

13.

David

(2005) N-Watch: A program for deriving neighborhood size and other psycholinguistic statistics. Behavior Research Methods 37: 65–70.

14.

David

Perea

(2005) BuscaPalabras: A program for deriving orthographic and phonological neighborhood statistics and other psycholinguistic indices in Spanish. Behavior Research Methods 37: 665–71.

15.

Dijkstra

(2005) Bilingual visual word recognition and lexical access. In: Kroll

de Groot

AMB

(eds) Handbook of bilingualism: Psycholinguistic approaches. New York: Oxford University Press, 179–201.

16.

Dijkstra

van Heuven

WJB

(1998) The BIA model and bilingual word recognition. In: Grainger

Jacobs

(eds) Localist connectionist approaches to human cognition. Mahwah, NJ: Erlbaum, 189–225.

17.

Dijkstra

van Heuven

WJB

(2002) The architecture of the bilingual word recognition system: From identification to decision. Bilingualism: Language and Cognition 5: 175–97.

18.

Dijkstra

Grainger

van Heuven

WJB

(1999) Recognizing cognates and interlingual homographs: The neglected role of phonology. Journal of Memory and Language 41: 496–518.

19.

Dijkstra

van Jaarsveld

Ten

Brinke S

(1998) Interlingual homograph recognition: Effects of task demands and language intermixing. Bilingualism: Language and Cognition 1: 51–66.

20.

Dreisbach

Haider

(2006) Preparatory adjustment of cognitive control in the task switching paradigm. Psychonomic Bulleting and Review 13: 334–38.

21.

Finkbeiner

Almeida

Janssen

Caramazza

(2006) Lexical selection in bilingual speech production does not involve language suppression. Journal of Experimental Psychology: Learning, Memory and Cognition 32: 1075–89.

22.

Gollan

Ferreira

(2009) Should I stay or should I switch? A cost-benefit analysis of voluntary language switching in young and aging bilinguals. Journal of Experimental Psychology: Learning, Memory and Cognition 35: 640–65.

23.

Green

(1998) Mental control of the bilingual lexico-semantic system. Bilingualism: Language and Cognition 1: 67–81.

24.

Hermans

Bongaerts

De Bot

Schreuder

(1998) Producing words in a foreign language: Can speakers prevent interference from their first language? Bilingualism: Language and Cognition 1: 213–29.

25.

Ibáñez

Macizo

Bajo

(2010) Language access and language selection in professional translators. Acta Psychologica 135: 257–66.

26.

Jackson

Swainson

Cunnington

Jackson

(2001) ERP correlates of executive control during repeated language switching. Bilingualism: Language and Cognition 4: 169–78.

27.

Jared

Kroll

(2001) Do bilinguals activate phonological representations in one or both of their languages when naming words? Journal of Memory and Language 44: 2–31.

28.

Jared

Szucs

(2002) Phonological activation in bilinguals: Evidence from interlingual homograph naming. Bilingualism, Language, and Cognition 5: 225–39.

29.

Kroll

Stewart

(1994) Category interference in translation and picture naming: Evidence for asymmetric connections between bilingual memory representations. Journal of Memory and Language 33: 149–74.

30.

Kroll

Bobb

Wodniecka

(2006) Language selectivity is the exception, not the rule: Arguments against a fixed locus of language selection in bilingual speech. Bilingualism: Language and Cognition 9: 119–35.

31.

Kroll

Bobb

Misra

Guo

(2008) Language selection in bilingual speech: Evidence for inhibitory processes. Acta Psychologica 128: 416–30.

32.

Levy

McVeigh

Marful

Anderson

(2007) Inhibiting your native language: The role of retrieval-induced forgetting during second-language acquisition. Psychological Science 18: 29–34.

33.

Macizo

Bajo

Martín

(2010) Inhibitory processes in bilingual language comprehension: Evidence from Spanish–English interlexical homographs. Journal of Memory and Language 63: 232–44.

34.

Martín

Macizo

Bajo

(2010) Time course of inhibitory processes in bilingual language processing. British Journal of Psychology 101: 679–93.

35.

Mason

WTM

(1994) Language switching in bilinguals: Evidence from the semantic categorization task. Unpublished undergraduate project, University of Cambridge, Cambridge, UK.

36.

Meuter

RFI

Allport

(1999) Bilingual language switching in naming: asymmetrical costs of language selection. Journal of Memory and Language 40: 25–40.

37.

Monsell

(1996) Control of mental processes. In: Bruce

(ed.) Unsolved mysteries of the mind: Tutorial essays in cognition. Hove: Erlbaum, 93–148.

38.

Monsell

(2003) Task switching. Trends in Cognitive Neuroscience 7: 134–40.

39.

Potter

von Eckhardt

Feldman

(1984) Lexical and conceptual representation in beginning and fluent bilinguals. Journal of Verbal learning and Verbal Behavior 23: 23–38.

40.

Rogers

Monsell

(1995) Costs of a predictable switch between simple cognitive tasks. Journal of Experimental Psychology: General 124: 207–31.

41.

Runnqvist

Costa

(in press) Is retrieval-induced forgetting behind the bilingual disadvantage in word production. Editorial Office of Blc 1: 289–812.

42.

Runnqvist

Costa

(in press) Is retrieval-induced forgetting behind the bilingual disadvantage in word production. Bilingualism: Language and Cognition. DOI: 10.1017/S136672891100003410.1017/S1366728911000034.

43.

Schneider

Eschman

Zuccolotto

(2002) E-Prime user’s guide Version 1.1. [software]. Pittsburgh: Psychology Software Tools.

44.

Sholl

Sankaranarayanan

Kroll

(1995) Transfer between picture naming and translation: A test of asymmetries in bilingual memory. Psychological Science 6: 45–49.

45.

Thomas

Allport

(2000) Language switching costs in bilingual visual word recognition. Journal of Memory and Language 43: 44–66.

46.

Thomas

MSC

van Heuven

WJB

(2005) Computational models of bilingual comprehension. In: Kroll

de Groot

AMB

(eds) Handbook of bilingualism: Psycholinguistic approaches. New York: Oxford University Press, 202–25.

47.

Verhoef

Roelofs

Chwilla

(2009) Role of inhibition in language switching: Evidence from event-related brain potentials in overt picture naming. Cognition 110: 84–99.

48.

Von Studnitz

Green

(2002) The cost of switching languages in a semantic categorization task. Bilingualism: Language and Cognition 5: 241–51.