Behavioural and ERP correlates of bilingual language control and general-purpose inhibitory control predicted by L1 and L2 proficiency

Highlights

• We examined neurocognitive correlates of inhibitory control with linguistic and non-linguistic stimuli using the negative priming paradigm in bilingual adults.

• Inhibition effects were modulated by the weight of a particular language with greater inhibition effect for L1 compared to L2.

• We find significant modulations of the N200 amplitudes as a function of demands on inhibitory control and language.

• Bilingual language control is related to the overall L2 proficiency and General-purpose cognitive control is related to the overall L1 proficiency.

Abstract

Cognitive control is the ability to adapt flexibly to current demands by promoting task-relevant information in the face of interference, and this has been asserted as an advantage with bilinguals. Bilingual language control and general-purpose cognitive control are discussed in the literature using different types of stimuli, cognitive tasks, component processes (selection, interference, inhibition, switching) and groups (bilingual vs. monolingual; high proficient vs. low proficient bilingual). The present study was designed to investigate the neurocognitive correlates of inhibitory control with linguistic (i.e., words) and nonlinguistic (i.e., line drawings of objects) stimuli in Hindi-English bilingual adults. We conducted the behavioural experiment first to establish the linguistic version of identity negative priming paradigm followed by the Event-related potential (ERP) experiment using the linguistic and nonlinguistic negative priming task. Results show the presence of inhibition effect using mean reaction times as well as ERP data while comparing the control and ignored repetition conditions, and this pattern varied with different stimuli – linguistic vs. nonlinguistic. Thus, the current study suggests that bilingual language control is not entirely subsidiary to general-purpose cognitive control and shows differences based on the stimulus type. Also, proficiency in L1 and L2 differentially predicts performance on the nonlinguistic and linguistic negative priming paradigm, respectively. These results are indicative of a dynamic cognitive control system associated with bilingualism, which varies as a function of stimulus type as well as language proficiency.

Introduction

Over the years, it has been established that speaking two or more languages appears to have a positive effect on cognitive control in bilinguals. Bilingual cognitive advantage is usually defined as the individual differences in cognitive task performances resulting in a better response time and fewer errors in different paradigms like Go/No-Go, Flanker task, Simon task, and Attention network task (Rodrigues-Fornells, Balaguer & Munte, 2007; Costa, Hernandez & Sabastian-Galles, 2008; Bialystok, Craik, Klein & Viswanatha, 2004). This advantage develops over time with the juggling of the two languages resulting in Bilingual language control (bLC) ability, which is transferred to the General-purpose cognitive control ability (GPCC¹; Green, 1998; Abutalebi & Green, 2007; Calabria, Costa, Green, & Abutalebi, 2018). Bilingual language control involves similar cognitive processes that are proposed for the general-purpose cognitive control system and consist of an overlapping neural correlate (Calabria, Baus, & Costa, 2019). However, the association between general-purpose cognitive control and bilingual language control (bLC) is debatable. Some studies suggest an overlap between the bLC and GPCC ( Dash & Kar., 2014; Declerck, Grainger, Koch, & Philipp, 2017; Green & Abutalebi, 2017; Prior & Gollan, 2011), and other studies consider both being distinct processes – lack of overlap between bLC and GPCC (Branzi, Calabria, Boscarino, & Costa, 2016; Calabria, Hernández, Branzi, & Costa, 2012; Jylkkä, Lehtonen, Lindholm, Kuusakoski, & Laine, 2018; Segal, Stasenko, & Gollan, 2019). For instance, Prior and Gollan (2011) showed similar language and task switch costs in Spanish English bilinguals, suggesting that language switching affects domain-general cognitive control. On the contrary, Declerck, Eben, and Grainger (2019) found no overlap between language control and executive control using a bilingual and a nonlinguistic flanker task. Interestingly, these studies compared linguistic and nonlinguistic versions of the cognitive control tasks, measuring different subcomponents of cognitive control (i.e., switching or goal shifting, inhibition, monitoring etc), but none of these studies compared the performance within the bilingual population based on the level of language proficiency as a measure of bilingualism. The current study examined the influence of language proficiency (i.e., L1-Hindi and L2-English) on bLC and GPCC – using linguistic and nonlinguistic stimuli – in an identity negative priming paradigm that measures inhibitory control.

The interaction between bilingualism and cognitive control is studied extensively in different language groups using electrophysiological measurement (Abutalebi, Annoni, Zimine, Pegna, Seghier, Jahnke et al., 2008; Crinion, Turner, Grogan, Hanakawa, Noppeney, Devlin et al., 2006; Hernandez, Dapretto, Mazziotta, & Bookheimer, 2001; Rodriguez-Fornells, Schmitt, Kutas, & Munte, 2002). Rodriguez-Fornells, Balaguer, and Münte (2006) have demonstrated the involvement of control mechanisms through frontal-central negativity observed at about 400 ms (ms) for cross-linguistic interference at the phonological and syntactic level. Jackson et al. (2001) conducted ERP studies with 20 native English speakers and with different languages as L2. Larger left frontocentral negativity was observed in the switch trials as compared to non-switch trials. Bilingualism does not have an equal impact on all the subcomponents of cognitive control – selection, inhibition, and switching (Hernández, Costa, & Humphreys, 2012; Bialystok, Craik, & Luk, 2012). Therefore, the selection of appropriate experimental task is crucial. Negative priming paradigm – measuring inhibitory control – is an ideal task to examine the effect of the previous trial on the current trial, thus introducing a high monitoring demand on each trial.

Negative priming occurs when responses to a target stimulus is slower in a situation when the same target stimulus is present as a distractor in the previous trial. Although, negative priming is a well-established paradigm to measure the effect of inhibitory control that is viewed as a consequence of the competing irrelevant stimuli from the previous trial that needs to be inhibited for current trial target selection (D'Angelo, Thomson, Tipper & Milliken, 2016), there are limited studies measuring the effect of bilingualism using a location-based negative priming paradigm (Blumenfeld & Marian, 2011; Treccani, Argyri, Sorace & Della Sala., 2009). The findings from these studies suggest that bilinguals have a more efficient inhibition mechanism when compared to monolinguals. Moreover, there are fewer studies that supports the inhibitory account of the cross-language negative priming effects using word stimuli monolingual context; Macizo, Bajo, & Martín, 2010; Martín, Macizo, & Bajo, 2010; (bilingual context; Neumann, McCloskey, & Felio, 1999; Neumann, Nkrumah, & Chen, 2018; Nkrumah & Neumann, 2018). The goal of these studies, however, was not to look at the bilingual advantage in inhibitory control mechanism but rather look at the bilingual language processing and representation. Interestingly, none of these studies look at the effect of bilingual language proficiency on the inhibitory control mechanism. Moreover, the bilingual cognitive control studies are often challenged and criticized due to various methodological issues – selection criteria of participants, stimulus types, task demands, experimental designs or subcomponent processes under study (Costa, Hernander, & Sabastia-Galle, 2008; Grosjean, 1998; Hilchey & Klein, 2011; Paap & Greenberg, 2013; Saidi & Ansaldo, 2015). With the overwhelming criticism on bilingual benefits (Costa, Hernández, Costa-Faidella, & Sebastián-Gallés, 2009; Hilchey & Klein, 2011; Paap & Greenberg, 2013; Saidi & Ansaldo, 2015), there is a need for methodological precision, more so, while defining and measuring bilingualism. Bialystok (2001) and Grosjean (1998) advocate for the notion of language proficiency as a continuous variable to address individual variability in performance, which was found to be particularly true in the Indian context with a complex linguistic environment (Dash & Kar, 2012). There is a gradual shift from the traditional ways of categorizing the bilingual population (Bialystok et al., 2012; Bialystok & Feng, 2009; Costa, Hernender & Sabastia-Galle, 2008; Siegal, Lozzi, & Lurian, 2009) to considering measures of bilingualism as a continuous variable (Anthony & Blumenfeld, 2018; Dash, Berroir, Joanette, & Ansaldo, 2019; Dash & Kar., 2012; Goral, Campanelli, & Spiro, 2015; Incera & McLennan, 2018; Yow & Li, 2015). Dash and Kar (2012) highlighted another important implication of a complex linguistic environment. Different patterns of clustering were observed in both the languages on proficiency tasks (i.e., L1-Hindi and L2-English), i.e., L2 tasks are clustered at the domain level (speaking/understanding tasks forming one cluster, for example) whereas L1 tasks showed skill-specific clusters (i.e., metalinguistic skills grouped as one component) (Dash & Kar, 2012). Studies on bilingual cognitive control are primarily based on comparisons of bilinguals and monolinguals, which may or may not generalize while considering language proficiency as a continuous variable.

Research on bilingualism often considers only L2 proficiency to interpret the effects of bilingualism on control processes. Given the variability in the organization of both the languages, both L1 and L2 skills need to be considered while interpreting experimental data. The current study is an investigation of the behavioural and electrophysiological correlates of inhibitory control among bilinguals varying in the level of L1 and L2 language proficiency. The negative priming paradigm with linguistic and nonlinguistic stimuli was employed to look at the inhibition mechanisms operating in the current trial as a function of previous trial effects. We used comparable identity negative priming tasks for both linguistic and nonlinguistic stimuli in two separate experiments to examine the influence of language proficiency on bLC and GPCC. We also intended to examine the relationship between L1 and L2 proficiency – treated as a continuous variable. The current study consists of two parts. In Experiment 1, the linguistic version of identity negative priming was validated in the behavioural paradigm. Study 2 consists of the ERP version of identity negative priming with both linguistic and nonlinguistic stimuli. The change in the amplitudes of the N200 component was examined as a measure of inhibition and conflict monitoring. We expected an inhibitory control effect for both linguistic and nonlinguistic tasks suggesting commonality between the bLC and GPCC. We also expected to find a significant relationship between L1 and L2 proficiency and inhibition effect with nonlinguistic stimuli in Experiment 2, given the findings supporting the relationship between general cognitive control mechanisms and language control.

A cross-linguistic negative priming task was designed with overlapping words in Hindi and English, where one of the two overlapping words was attended (and the other was ignored) on a given trial. Current trial reaction times were expected to vary as a function of the previously activated/suppressed language. It was hypothesized that the inhibition effect would be higher for L1 as compared to L2.

Twenty, right-handed Hindi-English bilingual adults (Mean age: 21.75 years, ±3.2 SD, with 7 males and 13 females) participated in the experiment. All participants were native speakers of Hindi (L1) and learned English (L2) in a more formal setting with at least 7 years of education in both languages, used both the languages in daily use, and with no significant history of sensory/motor/neurological disorders, were taken for the study. Participants were selected randomly from the University of Allahabad and provided written informed consent. The study was approved by the Institutional Ethics Committee, University of Allahabad.

Language history questionnaire (Vasanta, Suvarna, Sireesha, & Bapi Raju, 2010) was administered to collect information related to language acquisition, context of language acquisition, present language use (in percentage), language preference, use of language with family, friends, extended family, neighbours and respective hours of usage (per day), medium of instruction and self-reported proficiency level on different domains (5 point rating, where 1 represented “poor” and 5 represented “excellent”). All this information was organized under three major headings: age-related self-reported information; use related self-reported information and proficiency related self-reported information. The administration of the questionnaire took approximately 20–30 min for each participant (Appendix 1a).

Test of language proficiency in Hindi and English: An indigenously developed test of language proficiency in Hindi and English consisting of different tasks under the domains of speaking/understanding and reading/writing was employed (Dash & Kar, 2012). Speaking/understanding domain consisted of both production and comprehension tasks, which measured the level of performance at lexical, syntactic as well as discourse levels. These tasks included confrontation naming task, discourse analysis, convergent production/synonym tasks, and auditory comprehension. Similarly, the reading/writing domain consisted of tests for reading comprehension, fluency, phonological awareness, and written discourse. Performance on each test was measured in terms of accuracy percentage and was added to the composite score for each domain, namely speaking/understanding and reading/writing. The questionnaire and the proficiency test were individually administered in a quiet, well-lit room (Appendix 1b).

Stimuli and Procedure: The stimuli for identity negative priming consisted of a display containing overlapping linguistic stimuli (words in Hindi and English) in two shades of gray at the center of the screen. The words were taken from a set of 303 words (both language translation of the pictures from Abbate & LaChapelle, 1984) in both languages out of which a total of 120 words were selected (60 in each language), they were matched in length in both the languages and consisted of both animate and inanimate words (60 animate as well as inanimate), which resulted in 240 stimuli after overlapping these words. Each word was repeated twice in different combinations, thus resulting in 60 combinations for language and animacy (Hindi Animate = 60, Hindi Inanimate = 60, English Animate = 60, English Inanimate = 60). Words were selected based on the ratings on frequency, imageability, and familiarity levels of each word on a 5 point rating scale (where 1 represented “least” and 5 represented “most”. 20 high proficient Hindi-English Bilinguals (Mean = 21.34 years, SD = ±2.3) participated in the rating process and words with average rating were selected (Frequency: Mean = 3.67, SD = 0.75; Imageability: Mean = 4.24, SD = 0.61; Familiarity: Mean = 4.56, SD = 0.64).

The pairs of trials were structured according to a prime-probe schema. The stimuli consisting of overlapping meaningful words were presented in such a manner that the current stimulus acted as a prime for the upcoming stimulus and probe for the previous stimulus. There were four kinds of prime-probe combination trials with both prime and probe stimuli being monolingual, monolingual prime stimulus – bilingual probe stimulus, bilingual prime stimulus –monolingual probe stimulus, and both prime and probe with bilingual stimuli (see Fig. 1 and Table 1). RGB coordinates (157,157,157) defined the target. The assignment of color cue for the target word was counterbalanced across participants, i.e., half of the participants responded to the stimulus in the shade of light gray and other half responded to words in the shade of dark gray. 2 different pseudo-randomized lists were created for both dark and light versions of the experiment. None of the pairs of words presented in a particular trial or successive trials were semantically or phonologically related.

The stimuli were presented on a 17″ monitor in a quiet, dimly lit room. The stimuli appeared at the center of the screen, measuring within the frame of 106 pixels * 52 pixels. A horizontal and vertical resolution was fixed to 71dpi. Each trial began with a fixation point for 400 ms at the center of the screen followed by the pair of words against a white background for 200 ms, after which a blank screen was presented, and it stayed until response followed by a fixation point for the next trial. Participants performed an animacy judgement task once the stimuli appeared on the screen and responded by pressing the left arrow key if the target word represented an animate object and right arrow key if the target word represented an inanimate object. Participants responded using the first and second fingers of their dominant hand. Participants were instructed to respond as quickly and as accurately as possible. Overlapping words were presented in different combinations, i.e., Hindi over Hindi, English over English, Hindi over English, and English over Hindi (see Table 1). The stimuli were presented in the ratio of 3:1 for attended repetition and ignored repetition conditions. Attended repetition condition (336 trials, 168 monolingual, and bilingual trials each) consisted of trials in which the target language in the current trial was the same as the one in the previous trial. Ignored repetition condition (144 trials, 72 monolingual and bilingual trials each) consisted of trials in which the target language in the current trial was a distractor and hence was ignored on the previous trial. Other than the two priming conditions, there were four possible prime-probe combinations, as mentioned above, due to the two languages in which stimuli were presented. Also, the pure switch trials with a monolingual prime stimulus followed by a monolingual probe stimulus with one in L1 and another in L2 or vice versa were retained to partial out the effect of pure switch between the two languages and to bring out the true effect of persistent inhibition.

Reaction time (RT) data were analysed for the two priming conditions (attended repetition and ignored repetition), each having four types of prime-probe combination trials (monolingual-monolingual, monolingual-bilingual, bilingual-monolingual and bilingual-bilingual) for both the languages. Error trials (9.2%) and those with RTs that were more than 3SD (1.5%) were excluded from further analysis. Cross-linguistic inhibition effect is said to occur when responses to target language are slower in a situation when the target language was presented as a distractor on the previous trial. To examine the cross-linguistic inhibition effect, we computed difference scores using RTs across the prime-probe combination. Since the stimuli included words from both the languages, switching effects were bound to occur in addition to the cross-linguistic inhibition effect. To rule out the effect of language switching effect due to the pure-switch trials, simple subtractions were performed. Difference scores for the bilingual prime trial followed by monolingual probe trial in the attended repetition condition (for example, EH followed by EE where English is the target language in both the combinations) were further subtracted from the pure switch condition (HH followed by EE), same for the ignored repetition condition. Similarly, difference scores for the bilingual prime trial followed by bilingual probe trial in the attended repetition condition (for example, EH followed by EH where English is the target language in both the combinations) were further subtracted from the no switch condition (EE followed by EE); same for ignored repetition (see Table 1 for details). This resulted in 2 versions of priming conditions for both attended repetition and ignored repetition – monolingual-bilingual prime-probe combination (i.e., H1 and E1) and bilingual prime probe combination (i.e., H2 and E2) Means of the difference scores of both prime-probe combinations (with standard error of mean) as a function of language and priming Bilingual-monolingual prime-probe combination was not considered for analysis based on the non-significant effect of this combination of trials on the inhibition effect as observed in the pilot study.

Bivariate correlation and regression analysis were performed to examine the relationship between measures of bilingualism and negative priming task performance. Multiple regression analysis was performed by using the simultaneous method. Results of correlation and regression analysis are discussed for the speaking/understanding and reading/writing domains of language skills (measured using the test of language proficiency).

The difference scores were then subjected to a three-way, 2 languages (Hindi/English) * 2 priming conditions (Attended repetition/ignored repetition) * 2 prime-probe combinations (H1/E1 and H2/E2) repeated measures ANOVA (see Fig. 2). There was a significant main effect of priming condition, indicating that ignored repetition trials were slower than attended repetition trials [F(1,19) = 6.38, p = 0.021]. The main effect of prime-probe combination and the interaction between language and prime-probe combination was not significant, [F(1,19) = 2.54, p = 0.12 and F(1,19) = 1.17, p = 0.29 respectively]. However, the interaction effect between priming condition and prime-probe combination was significant, F(1,19) = 8.77, p = 0.008. Post hoc comparisons showed a significant effect for the bilingual prime-probe condition only. The interaction between language and priming condition showed a trend of significance, [F(1,19) = 4.37, p = 0.05. The three-way interaction of language, priming condition and prime-probe combination was significant [F(1,19) = 12.81, p = 0.008]. Tukey's post hoc test for the three-way interaction further confirmed significant inhibition effect for the set of bilingual prime-probe combination (H2/E2) in both the languages (p = 0.043 for L1 and p = 0.0002 for L2), which indicates that when the weight of the prime and probe trial was the same (when both prime and probe trials had bilingual stimuli), inhibition effect was seen in both the languages and magnitude of this effect was more for L1 as compared to L2. This is to say that the sustained effect of inhibition was found to be greater for L1 as compared to L2.

In the domain of speaking/understanding, there was a significant correlation between the total score of L2 proficiency and the inhibition effect in L2 for the monolingual-bilingual prime probe combination trials, r = 0.514, p = 0.02. The total score of L2 predicted 22.3% of the variance in performance on the negative priming task (R square = 0.264, adjusted R square = 0.223), and the model was significant, F(1, 18) = 6.46, p = 0.02. The total score of L1 did not significantly predict the performance on the linguistic version of the negative priming task. However, synonym task in L1 in the speaking/understanding domain, predicted performance on the negative priming task for L1 and accounted for 16.2% of the variance (R square = 0.206, adjusted R square = 0.162) and the model was significant, F(1, 18) = 4.671, p = 0.044.

In the domain of reading/writing, reading speed in L1 and reading fluency in L2 were correlated with the difference scores in L2 for the monolingual prime-probe combination trials. Both accounted for 21.2% of variance (R square = 0.295, adjusted R square = 0.212) and the model showed a trend of significance, F (2, 17) = 3.55, p = 0.051. On the other hand, reading comprehension in L1 significantly predicted the negative priming task performance in L1. There was a negative correlation between reading comprehension in L1 and attended repetition condition in L1 only for monolingual-bilingual combination trials. Reading comprehension accounted for 29.8% of variance (R square = 0.331, adjusted R square = 0.298) and the model was significant, F (1, 18) = 6.88, p = 0.006. In case of bilingual prime-probe combination trials, reading comprehension in L1 predicted 27.0% of variance (R square = 0.309, adjusted R square = 0.270) in the difference scores of both attended and ignored repetition conditions for L1 and the model was significant, F (1, 18) = 8.042, p = 0.011.

Thus, results based on regression analysis suggest that L2 proficiency for the speaking/understanding domain predicted performance on the linguistic negative priming task. In contrast, only task-specific relationship was observed between L1 proficiency (synonym task) and negative priming task performance. Since the negative priming task involved animacy judgement with words as stimuli, reading speed and reading comprehension emerged as significant predictors of performance on the negative priming task.

To sum up the findings of the experiment in Experiment 1, inhibitory control effects were evident in both languages when prime and probe trials had bilingual stimuli. L1 proficiency showed a task (reading comprehension) specific relationship, whereas L2 proficiency across domains, including both speaking/understanding and reading/writing skills, showed a significant correlation with performance on the negative priming task.