The effect of inhibitory control and its training on L1 and a new learned language switching

Abstract

Aims:

This study aimed to investigate how individuals’ inhibitory control (IC) ability affects language switching in the initial period of language learning.

Design:

Using a pretest/posttest design and event-related potential (ERP) methodology, we investigated the effect of IC on Chinese–English bilinguals during their language switching between Chinese (L1) and Korean (a language new to the participants, Lnew). All participants were required to name pictures (picture-naming task) in their L1 and Lnew in the pretest and posttest. Low-IC participants received an IC task training between the pretest and the posttest, while the high-IC group did not.

Data and analysis:

Analyses of both response latencies and ERP data were conducted by repeated-measures ANOVA.

Findings:

Results showed that the high-IC group exhibited symmetrical switch costs in both the pretest and the posttest. Besides, a more obvious late positive component (LPC) was observed when the high-IC participants switched from L1 to Lnew than the other way around, indicating their ability to inhibit cross-language interference. In contrast, the low-IC group exhibited asymmetrical switch costs, and no amplitude difference when switching between Lnew and L1 in the pretest. However, in the posttest, the switch costs pattern and the LPC results of the low-IC group became similar to those of the high-IC group.

Innovation:

The present study was a first attempt to provide electrophysiological evidence that IC ability plays an important role during L1–Lnew switching.

Significance:

These findings support the hypothesis that individuals’ IC ability plays a role of suppressing the non-target lexical access during language switching in the initial period of second language learning. The results also indicate that the relevant training in IC ability could contribute to the improvement of the language-switching efficiency in the initial period of language learning.

Keywords

Language switching a new language inhibitory control training LPC

Introduction

During language switching, bilinguals often attempt to suppress the cross-language interference, which would consume some cognitive resources. Thus, switch costs are incurred, resulting in slower or delayed processing on the switch trials in which participants are asked to switch rapidly between different languages (Costa & Santesteban, 2004; Costa, Santesteban, & Ivanova, 2006; De Bruin, Roelofs, Dijkstra, & FitzPatrick, 2014; Green, 1998; Jackson, Swainson, Cunnington, & Jackson, 2001; Linck, Schwieter, & Sunderman, 2012). A prevailing model of inhibitory control (IC) argues that the lexical access is controlled by inhibition. Specifically, the amount of inhibition applied in a given language is proportional to the level of the activation of that language, and the strength needed for the reactivation of a suppressed language is proportional to the amount of inhibition (Green, 1998). For low-proficient bilinguals, who are much less proficient in their second language (L2) than in their native language (L1), high inhibition is needed to suppress the dominant L1 when switching into the weaker L2. However, when switching back from L2 to L1, they have to recover from the high inhibition, which would cause longer processing time than switching from L1 to L2, namely, asymmetrical language-switch costs (i.e. L1 switch costs are larger than those of L2). In contrast, for balanced bilinguals with similar proficiency of L1 and L2, inhibition levels for L1 and L2 are almost the same, which would produce symmetrical switch costs (i.e. L1 switch costs are similar to those of L2) instead of asymmetrical switch costs (Costa & Santesteban, 2004; Costa et al., 2006). Furthermore, trilingual switching also revealed the effect of inhibition shown by asymmetrical switch costs. Philipp, Gade and Koch (2007) required trilinguals to switch between two languages (i.e. switching between L1 and L2, or L1 and L3, or L2 and L3). Results showed that language-switch costs were always larger for the dominant language than for the weak language. It seems that the inhibition mechanism during language switching could be varied to different language proficiency levels. However, it is not clear for bilinguals with low L2 proficiency but high IC ability how the inhibition would affect the language switching in the initial period of language learning.

Evidence from previous studies suggests that language proficiency is not the only factor that determines the patterns of language-switch costs (Christoffels, Firk, & Schiller, 2007; Liu, Rossi, Zhou, & Chen, 2014; Verhoef, Roelofs, & Chwilla, 2009). For example, even low-proficient bilinguals can exhibit symmetrical switch costs under certain conditions, such as when the cross-language stimuli are cognates (Christoffels et al., 2007), or when sufficient preparation time between the cue and the stimulus is allowed (Verhoef et al., 2009), or in cases where low-proficient bilinguals have high IC ability (Liu et al., 2014).

Recently, some studies have found that trilinguals’ personal characteristics (e.g. IC ability) may also affect their performance on language switching (De Bruin et al., 2014; Linck et al., 2012). Inhibition, as measured by the difference in response latencies between incongruent and congruent trials in a Simon task, has been shown positively correlated with language switching, as tested by a picture-naming task (De Bruin et al., 2014; Linck et al., 2012). Specifically, the better IC ability is, the better language-switching performance is. It thus seems that individuals’ IC ability affects language switching. Liu et al. (2014) chose low-proficient bilinguals with different levels of IC as participants. They observed that the bilinguals with different IC abilities exhibited different patterns of L1 and L2 switch costs, indicating that even low-proficient bilinguals could benefit from their own high IC ability and exhibit symmetrical language-switch costs.

However, Liu and colleagues’ study could not guarantee that all participants learned their L2 at the same time. Previous studies demonstrated that the age of acquisition (AoA) of L2 might affect lexical retrieval (Perani et al., 2003; Waldron & Hernandez, 2013). The lexical retrieval in an early acquired language showed less extensive brain activations during production in the verbal fluency task (Perani et al., 2003). Moreover, early L2 learners preferentially recruited sensorimotor and motor control brain regions to retrieve words when producing the past tense in L2, while later L2 learners relied on executive processing regions when performing the same task (Waldron & Hernandez, 2013). In this line of reasoning, AoA might further affect the performance of language switching which involves lexical retrieval. In order to remove the confounding effect of AoA, the current study introduced a new language, one that no participant had learned prior to this study. By doing so, we ensured that all the participants started to learn the new language (Lnew) at the same time, and for a similar period, thus controlling the influence of AoA as much as possible. The present study selected low-proficient bilinguals¹ with different levels of IC in order to test the importance of inhibition in L1–Lnew switching. More importantly, we attempted to clarify the influence of inhibition on L1–Lnew switching by means of enhancing the IC ability of the low-IC participants.

According to the IC model (Green, 1998), inhibition may occur at the first phase of language task schema, where bilinguals are required to select an appropriate language to activate (i.e. L1 or L2/L3). Furthermore, inhibition may play the role of suppressing non-target lexical entries. On the one hand, some event-related potential (ERP) studies have found that inhibition interacts with language switching at the language task schema phase as evidenced by an N2 component. It is recorded over the fronto-central region of the scalp and peaks around 310 ms after the stimuli onset (Jackson et al., 2001). Generally, the N2 component is more negative when switching to a less dominant language (i.e. L2/L3) (recruiting more inhibition to suppress L1) than switching to L1 (recruiting less inhibition to suppress L2/L3; Jackson et al., 2001). Furthermore, Misra et al. (2012) found that a greater N2 component was observed for the L2 naming in a block after naming in L1 in the previous block. On the contrary, no obvious N2 component was obtained for the L1 naming after naming in L2. This finding indicated that the weak language (L2) was spoken via inhibiting the dominant language (L1) during the early phase of language task schema competition. However, Verhoef, Roelofs and Chwilla (2010) observed the N2 (200–350 ms) was more negative in switch trials compared with repeat trials for L2, but not for L1 when naming pictures in different languages. They argued that the N2 reflects attentional control rather than inhibition.

On the other hand, a behavioral study suggested that inhibition may occur at the level of language-specific mental lexicon (i.e. lemmas), since the findings show that predictable concepts alone could reduce L1 switch costs (Declerck, Koch, & Philipp, 2014). The ERP study by Guo, Ma, and Liu (2013) observed that the n-2 repetition trials elicited a more negative ERP component (not late positive component (LPC)) than the n-2 non-repetition trials in the lexical selection response phase when the trilinguals finished a picture-naming task. They proposed that suppression of the non-target languages occurs in the lexical selection phase. Furthermore, two studies found that the LPC is closely associated with the selection of the lexical lemma in the intended language (Jackson et al., 2001; Martin et al., 2013). For instance, Martin et al. (2013) found a larger LPC in early L1–L2 bilinguals than in early L1–L3 bilinguals, and also in switch rather than non-switch trials where responses were given in the weak language L3 during a picture-naming task. Based on their findings, they claimed that the LPC may be related to the control of lexical representations. Overall, inhibition may function as inhibiting the non-target language task schema at the first phase, which is reflected by N2 (Jackson et al., 2001; Misra et al., 2012), while inhibition perhaps works by suppressing the non-target lemma at the second phase, which may be evidenced by LPC (Jackson et al., 2001; Martin et al., 2013). It still remains unclear at which phase(s) inhibition would come to play a role in language switching.

Considering the existing evidence so far, most previous studies investigated the role of inhibition during language switching by comparing the performances of high and low-proficient bilinguals (e.g. Costa & Santesteban, 2004; Costa et al., 2006). However, fewer studies directly examined the effect of individuals’ IC ability during language switching. Meanwhile, most previous studies investigated the mechanism of L1–L2 switching without removing the influence of L2 AoA. Hence, in the current study, we asked participants to learn words in a new language to remove the influence of AoA. Besides, it is still unknown at which level inhibition plays its role. To examine this issue, we then compared the language-switching performances of individuals with different IC ability along with inhibition training. For all the above reasons, the main purpose of the current study was to investigate the effect of IC ability on L1 and a new learned language (Lnew) switching and the change of this effect after training on the IC ability.

The IC ability of participants was measured by the Simon switch task (for a detailed description please refer to the section of Measure of IC ability of participants). Based on the scores of the Simon switch task, participants were divided into two groups: high inhibitory control (high-IC group) and low inhibitory control (low-IC group). For the behavioral data, we expected to observe in the high-IC group symmetrical or small asymmetrical language-switch costs, but in the low-IC group asymmetrical switch costs. For electrophysiological data, we would assume that inhibition occurs during the language task schema competition phase if the high-IC group showed a larger N2 component in Lnew switch trials than in L1 switch trials (recruiting more inhibition to suppress the strong interference from L1 schema) in an early time window in both the pretest and the posttest, while in comparison the low-IC group showed a similar pattern of N2 (i.e. larger N2 in Lnew trials than that in L1 ones) only after training in the posttest but not in the pretest. Likewise, if the high-IC group showed a larger LPC in Lnew switch trials than in L1 switch trials (more inhibition being employed to suppress the interference from L1 lemma) in both the pretest and posttest, while the low-IC group showed a similar pattern of LPC only in the posttest, we would suggest that the inhibition occurs during the lexical selection phase. From previous studies we know that the larger the N2 component or LPC, the stronger the inhibition should be, and consequently the less cognitive resource is needed to reactivate the suppressed lexicon, which would lead to symmetrical or small asymmetrical language-switch costs.

Method

We used a pretest/posttest design and the ERP methodology to test Chinese–English bilinguals (see Figure 1). First, participants were divided into two groups (high-IC and low-IC group) based on their performance in the IC ability test, which lasted for a week. Two days after this test, all participants started to learn a small set of Korean (Lnew) words in an artificial context. This learning session lasted for 6 days. A language-switch task (the pretest) was performed on both the high and low-IC groups 2 days after the learning session; both L1 and Lnew stimuli were used while ERPs were recording. This pretest lasted for a week. Next, in order to pick out invalid EEG data, we checked each participant’s data carefully over 2 weeks. After that, we ran a 6-day Simon switch task training on the low-IC group to improve their IC ability while learning a second set of 16 words in Korean simultaneously. Meanwhile, the high-IC group also learned these 16 words, but received no Simon switch task training. After a break of 2 days, a language-switch task (the posttest) was run on both high and low-IC groups while ERPs were recorded. This posttest lasted for a week. Overall, this study took 52 days for data collection at Beijing Normal University of China.

Participants

Participants were 45 right-handed, native Chinese speakers with low proficiency in English (L2) from Beijing Normal University of China. They had no previous exposure to Korean, which was the new language the current study required them to learn. None of the participants reported having neurological or psychological impairments or taking psychoactive medication. Prior to the study, ethical approval was obtained from the Committee of Protection of Subjects at Beijing Normal University. All participants were required to read and approve the informed consent before participating the study. Data from five participants were eliminated: one due to his/her low accuracy and four due to the excessive EEG artifacts. The final sample consisted of 40 participants (30 female), aged from 17 to 24 years old (M = 18.82 ± 1.85 years).

Proficiency ratings of participants

In order to confirm that our participants’ English proficiency was low, we administered a self-rating questionnaire. Participants were asked to rate how well their L2 (English) listening, speaking, reading, and writing skills were compared with their L1 (Chinese) skills. Ratings were done by using a 5-point scale, in which 5 indicates that L2 skills are comparable with L1 skills, and 1 indicates very low L2 skills, much lower than L1 skills. The average proficiency ratings of L1 listening, speaking, reading, and writing of all participants were 4.38 ± .54, 4.43 ± .50, 4.53 ± .51, 4.45 ± .55, respectively, and of L2 were 2.30 ± .65, 2.45 ± .60, 2.38 ± .67, 2.53 ± .55, respectively. Paired-samples t-tests revealed a statistically significant difference between the proficiency ratings of L1 and L2 for all four skills, t(39) = 16.47, p < .001, t(39) = 20.16, p < .001, t(39) = 18.49, p < .001, and t(39) = 16.68, p < .001. Moreover, all the participants live in a Chinese-dominant environment. They started to learn English in middle school after age 10, with an average of 6 hours per week spent in middle- and high-school English classes. Since entry to college, they have been receiving similar English classes, and language switching mainly occurs during the class for they have little chance to communicate with a foreigner in English during daily life. This evidence showed that the participants were unbalanced bilinguals.

Furthermore, participants were subdivided into high and low-IC groups using the Simon switch task. (More details on the Simon switch task are provided in the Procedure section.) Table 1 shows the AoA of L2 obtained through the self-rating questionnaire, the self-ratings for L1 and L2 language skills, and the average score on the Oxford Placement Test (this test includes 25 multiple choice questions and a cloze test, and the total score is 50 points). Independent-samples t-tests showed no significant difference between the high- and low-IC participants on the AoA, language skills, and the Oxford Placement Test.

Table 1.

Means (and standard deviations) of the age of acquisition, language skill self-ratings, and the Oxford Placement Test scores for the high- and low-IC groups.

Characteristic	L1 (Chinese)		t	L2 (English)		t
Characteristic	High-IC	Low-IC	t	High-IC	Low-IC	t
AoA				11.00 (.86)	10.70 (.98)	.72
Listening	4.40 (.50)	4.37 (.60)	.18	2.30 (.66)	2.32 (.67)	−.07
Speaking	4.50 (.51)	4.37 (.50)	.81	2.45 (.61)	2.42 (.61)	.15
Reading	4.55 (.51)	4.53 (.51)	.14	2.30 (.66)	2.42 (.69)	−.56
Writing	4.50 (.51)	4.42 (.61)	.44	2.50 (.61)	2.53 (.51)	−.15
OPT				40.80 (3.95)	39.47 (4.16)	1.02

Measure of IC ability of participants

A modified version of the Simon switch task was chosen as a measure of IC ability (Liu et al., 2014; Liu, Fan, Rossi, Yao, & Chen, 2016). The task-set inertia theory posits that participants need to use more effort while performing the difficult task than the easy one, in order to inhibit the task-set that prepares them to finish the task more quickly. This inhibition then carries over to the following trial (i.e. switch trial), where an extra-long reaction time is taken to release the previous inhibition. As a result, participants produce larger switch costs for an easy task than for a difficult one (i.e. asymmetrical switch costs). According to the task-set inertia theory, switch costs are caused by persisting inhibition, and the ease of releasing inhibition and applying inhibition are related; that is, if one can apply inhibition with ease, he/she thus can release inhibition easily, and vice versa (Allport, Styles, & Hsieh, 1994; Allport & Wylie, 1999; Liu et al., 2016). The traditional Simon task emphasizes inhibition execution which is reflected by the differences in response latencies between incongruent and congruent trials (e.g. Bialystok, 2006; Hilchey & Klein, 2011), and ignores the possibility that changes from congruent to incongruent trials and the other way around would also cause some switching. In contrast, the Simon switch task focuses on the switching between congruent and incongruent trials, and forms four different trials: congruent repeat/switch trials and incongruent repeat/switch trials. In this sense, the Simon switch task measures both the processes of applying inhibition (i.e. incongruent switch costs which are the differences between incongruent switch and incongruent repeat trials) and releasing inhibition (i.e. congruent switch costs which are the differences between congruent switch and congruent repeat trials). Therefore, Simon switch costs (i.e. the difference between congruent and incongruent switch costs) can be treated as an index of IC ability. We would expect individuals with a strong ability of suppression (i.e. high-IC group) to show different magnitudes of switch costs as compared with individuals with a weak ability of suppression (i.e. low-IC group).

A trial began with a cue (red or blue square) lasted for 250 ms, and then a blank screen was presented for 500 ms, followed by an arrow for 250 ms. On the disappearance of the arrow, the symbol “******” appeared, and participants were asked to respond as accurately and quickly as possible. When the cue was red, participants were required to press the button corresponding to the pointing direction of the arrow, which means a congruent spatial response was required (congruent condition). When the cue was blue, participants had to press the button opposite to the pointing direction of the arrow, which means an incongruent spatial response was required (incongruent condition). This color–response association was counterbalanced across participants. If the participant did not respond within 2000 ms, the symbol would disappear and the screen would be blank for 1000 ms before the onset of the following trial. The experiment included five blocks. Each block contained 24 congruent repeat trials and 24 incongruent repeat trials, as well as 24 congruent switch trials and 24 incongruent switch trials, for a total of 96 trials. All stimuli were presented pseudo-randomly.

Accuracy was greater than 95% for all conditions, so it was not analyzed further. According to the task-set inertia theory, the larger the differences between congruent and incongruent switch costs (i.e. Simon switch costs), the more asymmetrical Simon switch costs will be. Participants were divided into two groups (i.e. high-IC and low-IC) based on a median split of their Simon switch costs. There were 20 participants (16 female) in the high-IC and 20 participants (14 female) in the low-IC group. Independent-samples t-tests showed no significant age differences between the high-IC (M = 18.95 ± 2.11 years) and low-IC (M = 18.70 ± 1.59 years) participants, t(38) = .42, p > .05. The high-IC group was expected to show small Simon switch costs (i.e. the difference between congruent switch costs and incongruent switch costs), while the low-IC group was expected to exhibit large Simon switch costs. Mean response latencies and switch costs of the high- and low-IC groups are shown in Figure 2.

In order to validate our subdivision of subjects into high- and low-IC groups and to test whether the expected different magnitudes of Simon switch costs could be observed in the high- and low-IC groups, we performed a three-way repeated-measures ANOVA with group (high-IC, low-IC), congruency (congruent, incongruent), and task sequence (repeat, switch) as factors. Results revealed a significant interaction of group × congruency × task sequence, F(1, 38) = 35.55, MSE = 194.64, p < .001, η²_P = .48. Further analyses on separate groups showed the congruency × task sequence interaction was not significant for the high-IC group, F(1, 19) = 1.44, p > .05. Congruent switch costs (15 ms) were similar to those of incongruent switch costs (20 ms), which we deemed symmetrical switch costs. However, the congruency × task sequence interaction was significant for the low-IC group, F(1, 19) = 38.54, p < .001, such that congruent switch costs (39 ms) were larger than those of the incongruent switch costs (–8 ms), which can be treated as asymmetrical switch costs. These results suggest that the magnitude of the switch costs was symmetrical for the high-IC group, but asymmetrical for the low-IC group, as expected (see Figure 2).

Materials

Language learning materials

The current study used Korean words as learning materials. Chinese, English, and Korean, with distinct orthography, phonology, and syntax characteristics, belong to different language families (Chinese is a Sino-Tibetan language, English is an Indo-European language, and Korean is Altaic). We thus could rule out the transfer effect during learning Korean. Before the language-switching pretest, the word form of 16 very simple Korean words² (Lnew) was presented. Participants spent 15 minutes for six consecutive days learning the pronunciation and meaning of these words with the help of an electronic dictionary. Along the same pattern, after the pretest, all participants learned another 16 words. Altogether, 32 Korean words were used in the current study. The pretest and posttest words were counterbalanced across participants.

Language-switching materials

For the picture-naming task, the 32 words were paired with black-and-white line drawings, with a size of 15 cm × 15 cm, selected from the Snodgrass and Vanderwart’s (1980) photo gallery standardized for Chinese L1 by Zhang and Yang (2003). These 32 words were divided into two groups, one for the pretest, the other for the posttest. After learning these new words, participants were required to rate the subjective familiarity with a set of L1 words and the learned Lnew words by using a 5-point scale questionnaire (1 = “very unfamiliar,” 5 = “very familiar”). Paired-samples t-tests showed that the average familiarity with Lnew names (4.64 ± .22) was not different from that of L1 words (4.65 ± .21), t(79) = –.52 p > .05. Independent-samples t-tests showed that there was no difference between the pretest and posttest in the average subjective familiarity with Lnew words, t(78) = .64, p > .05, 4.66 ± .22 vs. 4.63 ± .23, nor in the familiarity with L1 words, t(78) = –.27, p > .05, 4.65 ± .25 vs. 4.66 ± .16. These results indicated that the participants have reached a similar proficiency level in processing the L1 and Lnew words.

Procedure

Language-switching task pretest

A picture-naming paradigm was used for the language-switching task. Participants were familiarized with L1 and Lnew picture names by seeing a picture that was paired with corresponding L1 words, e.g. “蚂蚁” (ant) and Lnew words, e.g. 개미. The stimuli were presented at the center of a 17-inch computer screen with 1024 × 768 pixel resolution. A trial began with a red or blue square for 250 ms, which was a cue for naming in L1 or in Lnew. The color–language association was counterbalanced across participants. Then, the screen was blank for 500 ms. Next, a picture was presented for 250 ms, followed by a blank screen for 1000 ms. Afterwards, when the symbol “******” appeared, participants were required to name the picture as accurately and quickly as possible. Participants were instructed not to respond until the symbol appeared. This delay was done in order to avoid contamination of the EEG signal with myoelectric artifacts of language articulation (Jackson et al., 2001; Liu et al., 2014; Martin et al., 2013). If participants did not respond within 2000 ms, the symbol would disappear, and a 1000 ms blank screen would appear before the onset of next trial. There were five blocks, and 96 trials per block. Each block contained 24 repeat trials of L1–L1 and 24 of Lnew–Lnew, as well as 24 switch trials of L1–Lnew and 24 of Lnew–L1. All stimuli were presented pseudo-randomly. Response times were recorded by a PSTSR-BOX connected to a microphone.

Inhibition-related training

Two weeks after the pretest, the low-IC group received six training sessions of the Simon switch task for six consecutive days. The procedure of the Simon switch task was the same as described above. Each training session consisted of six blocks. In order to improve training efficiency, there were not only switches between congruent and incongruent trials in each block, but also changed rules across blocks. For example, in the first block, the red square was a cue for congruent response, and the blue square for incongruent response. On the contrary, the rule was opposite in the second block, that is, blue for congruent response and red for incongruent response. The following blocks were in similar alternation. This change was implemented to discourage participants from forming any specific strategy. Mean response latencies and switch costs of each training session are shown in Figure 3.

To ensure that the IC abilities of the low-IC group was improved through the training, a three-way repeated-measures ANOVA was performed on the data gathered in the six training sessions with session (the first through sixth sessions), congruency (congruent, incongruent) and task sequence (repeat, switch) as within-subjects factors. There were significant main effects of task sequence, F(1, 19) = 21.62, MSE = 1276.31, p < .001, η²_P = .53, and training sessions, F(5, 95) = 21.15, MSE = 9848.82, p < .001, η²_P = .53. Furthermore, there were significant two-way interactions of congruency × task sequence, F(1, 19) = 21.15, MSE = 9848.82, p < .001, η²_P = .43, and training session × task sequence, F(5, 95) = 6.11, MSE = 189.44, p < .001, η²_P = .24. More importantly, the interaction of training sessions × congruency × task sequence reached significance, F(5, 95) = 3.12, MSE = 109.56, p < .05, η²_P = .14. Our main focus is the change of the symmetry of Simon switch costs across sessions. The subsequent separate analyses by training session showed a significant interaction of congruency × task sequence for the first training session, F(1, 19) = 14.38, p < .01, such that congruent switch costs (34 ms) were larger than those of the incongruent (14 ms). The same was found in the second training session, F(1, 19) = 6.68, p < .05, such that congruent switch costs (21 ms) were larger than those of the incongruent (10 ms). However, this interaction was not significant for the remaining four training sessions, F(1, 19) = 3.87, p > .05, F(1, 19) = 3.48, p > .05, F(1, 19) = 3.00, p > .05, and F(1, 19) = 3.48, p > .05, respectively, indicating that the differences between congruent and incongruent switch costs started to decline since the third training session, ending up with symmetrical switch costs (see Figure 3). Therefore, the inhibitory performance of the low-IC group was significantly improved by training through the Simon switch task.

Language-switching task posttest

After the low-IC group finished inhibition-related training, both IC groups performed the language-switching task posttest. The procedure of posttest was the same as the language-switching task in the pretest, but the words used in the posttest differed from those in the pretest.

Electrophysiological recordings

Electrophysiological data were recorded from 64 Ag/AgCl electrodes placed according to the extended 10–20 positioning system. The signal was recorded with 1 kHz sampling rate and referenced online to the right mastoid (M2). Impedances were kept below 5 kΩ. Electroencephalographic activity was filtered online within a bandpass between 0.1 and 100 Hz and refiltered offline with a 30 Hz, low-pass, zero-phase shift digital filter. Eye blinks were mathematically corrected (Gratton, Coles, & Donchin, 1983) and remaining artifacts were manually rejected. Continuous recordings were segmented into cue-locked −100 to 500 ms epochs, and stimulus-locked −200 to 1000 ms epochs. This was done so that we could dissociate each step of language switching. Correspondingly, the epochs were referenced to the 100 ms pre-cue baseline, and the 200 ms pre-stimulus baseline. Signals exceeding ±80 μV in any given epoch were automatically discarded.

Results

Behavioral data analyses

Accuracy was quite high overall (>95%). Only latencies of correct trials were further analyzed. Data from the first two trials of each block were excluded because they may not reflect participants’ true performance, as the participants were not ready for the experiment. The naming latencies beyond M ± 3SD were also excluded. In total, 11.47% of the data were removed. For naming latencies, a three-way repeated-measures ANOVA was conducted with group (high-IC group, low-IC group) as between-subject factor and language (L1, Lnew) and language sequence (repeat, switch) as within-subject factors.

Behavioral results

Mean naming latencies and switch costs for the high- and low-IC groups in the pretest and posttest are shown in Figure 4.

The comparison of pretest between the high- and low-IC groups

The main effect of language sequence reached significance, F(1, 38) = 36.75, MSE = 282.69, p < .001, η²_P = .49. The interaction between group × language sequence reached significance, F(1, 38) = 7.99, MSE = 282.69, p < .01, η²_P = .17. And the interaction between language × language sequence was significant, F(1, 38) = 13.91, MSE = 314.35, p < .01, η²_P = .27. The three-way interaction of group × language × language sequence was also significant, F(1, 38) = 6.96, MSE = 314.35, p< .05, η²_P = .16. A simple effect analysis of the three-way interaction showed that L1 repeat trials and Lnew repeat trials did not differ for both high-IC and low-IC groups, F(1, 38) = .45, p > .05, F(1, 38) = 3.44, p = .07. Furthermore, L1 switch trials and Lnew switch trials did not differ for the high-IC group, F(1, 38) = .00, p > .05, whereas, the name latencies of L1 switch trials is slower than Lnew switch trials for the low-IC group, F(1, 38) = 4.85, p < .05 (see Figure 4). In addition, to address the symmetry of language-switch costs, further separate analyses by group showed no significant interaction of language × language sequence for the high-IC group, F(1, 19) = 1.48, p > .05, suggesting that L1 switch costs (12 ms) were similar to those of Lnew (5 ms). However, the low-IC group showed a significant interaction of language × language sequence, F(1, 19) = 18.63, p < .001, such that L1 switch costs (42 ms) were larger than those of Lnew (6 ms; see Figure 4). Overall, the symmetrical switch costs for the high-IC group and the asymmetrical switch costs for the low-IC group showed that inhibitory control ability in and of itself played a crucial role during language switching.

Training benefits: the comparison between pretest and posttest for the low-IC group

The main effect of test reached significance, F(1, 38) = 4.27, MSE = 135971.86, p < .05, η²_P = .10. The naming latencies in the posttest were smaller than those in the pretest, indicating that the training was effective. Furthermore, the main effect of language was significant, F(1, 38) = 39.36, MSE = 429.62, p < .001, η²_P = .51. The interaction between language × language sequence reached significance, F(1, 38) = 7.84, MSE = 541.22, p < .01, η²_P = .17. The three-way interaction of test × language × language sequence reached significance, F(1, 38) = 4.21, MSE = 541.22, p < .05, η²_P = .10. A simple effect analysis of the three-way interaction showed that L1 and Lnew repeat trials were not significantly different in the pretest, F(1, 38) = 4.08, p = .05, nor for the posttest, F(1, 38) = 4.53, p > .05. However, L1 switch trials showed longer latencies than Lnew switch trials in the pretest, F(1, 38) = 5.88, p < .05, but not in the posttest, F(1, 38) = .75, p > .05. Furthermore, a separate analysis by group was performed to see the symmetry of language-switch costs. The pretest for the low-IC group showed a significant interaction of language × language sequence, F(1, 19) = 12.68, p < .01, reflecting asymmetrical language-switch costs. However, the interaction failed to reach significance in the posttest for the low-IC group, F(1, 19) = .26, p > .05, reflecting symmetrical language-switch costs (see Figure 4). These findings may suggest that enhanced inhibition improved the efficiency of language switching.

The comparison between pretest and posttest for the high- IC group

The main effect of language reached significance, F(1, 38) = 4.32, MSE = 1456.96, p < .05, η²_P = .10, as did language sequence, F(1, 38) = 21.81, MSE = 138.07, p < .001, η²_P = .37. The interaction between language × language sequence reached significance, F(1, 38) = 4.52, MSE = 105.24, p < .05, η²_P = .11. The three-way interaction of test × language × language sequence was not significant, F(1, 38) = .06, MSE = 105.24, p > .05, η²_P = .002. Thus, there were no differences between L1 and Lnew repeat trials, or between L1 and Lnew switch trials in the pretest, nor in the posttest. Furthermore, there was no significant interaction of language × language sequence in the pretest, F(1, 19) = 1.48, p > .05, and the posttest, F(1, 19) = 3.50, p > .05, reflecting symmetrical language-switch costs for the high-IC group in both the pretest and the post test (see Figure 4).

The comparison of posttest between the high- and low-IC groups

There were main effects of language, F(1, 38) = 4.26, MSE = 1117.54, p < .05, η²_P = .10, and language sequence, F(1, 38) = 22.74, MSE = 263.78, p < .001, η²_P = .37.The interaction between language × language sequence reached significance, F(1, 38) = 5.23, MSE = 202.65, p < .05, η²_P = .12. The three-way interaction of group × language × language sequence did not reach significance, F(1, 38) = .35, MSE = 202.65, p > .05, η²_P = .009, indicating both the high-IC group and the low-IC group exhibited symmetrical switch costs. This finding further suggests that the strengthened inhibitory control ability enables the low-IC group to function equivalently to the high-IC group during language switching.

To briefly summarize the behavioral results, strong inhibitory control ability benefited the high-IC group during switching between L1 and Lnew in both the pretest and the posttest. However, the low-IC group only showed similar gains after receiving inhibition training, i.e. in the posttest.

Event-related brain potential analyses

ERP components were defined based on the grand averages and were analyzed in time windows classically used to explore the N2 and LPC. N2, through cue-locked and stimulus-locked data, is used to investigate whether and how inhibition occurs during the language task schema competition phase (Jackson et al., 2001; Verhoef et al., 2009, 2010). On the other hand, LPC from stimulus-locked data represents the possibility that inhibition functions during the lexical selection response phase (Jackson et al., 2001; Martin et al., 2013). Repeated-measures ANOVAs were performed on mean amplitudes in the following intervals: 250 to 350 ms for the cue-locked N2, 270 to 370 ms for the stimulus-locked N2, and 450 to 650 ms for the stimulus-locked LPC. Topographical analyses were based on mean amplitudes measured over 64 electrodes distributed over the entire scalp. According to previous studies on language switching (Jackson et al., 2001; Liu et al., 2014; Martin et al., 2013; Verhoef et al., 2009) and visual inspection of our data, we focused on the N2 over three regions of interest (ROIs)—frontal (F5, F3, F1, Fz, F2, F4, F6), fronto-central (FC5, FC3, FC1, FCZ, FC2, FC4, FC6) and central (C5, C3, C1, CZ, C2, C4, C6). Furthermore, we focused on the LPC over five ROIs—frontal (F5, F3, F1, Fz, F2, F4, F6), fronto-central (FC5, FC3, FC1, FCZ, FC2, FC4, FC6), central (C5, C3, C1, CZ, C2, C4, C6), central-parietal (CP5, CP3, CP1, CPZ, CP2, CP4, CP6) and parietal (P5, P3, P1, PZ, P2, P4, P6).

We analyzed the differences between the high- and low-IC groups in N2 and LPC components. The data from the first two trials of each block, naming errors, as well as trials contaminated by artifacts were deleted from the analyses (12.61%). A five-way repeated-measures ANOVA was conducted on the mean amplitudes with group (high-IC, low-IC group) as a between-subject factor, and language (L1, Lnew), language sequence (repeat, switch), hemisphere (left, midline, right), and brain region as within-subject factors. Significance levels of the F ratios were adjusted with the Greenhouse–Geisser correction. It should be noted that we did not find any differences between hemispheres or the different regions in the analysis, so they are not reported in the results.

ERP results

Cue-locked N2 time window (250–350 ms)

The cue-locked grand average waveforms and topographic maps of L1 and Lnew trials for the high- and low-IC groups in the pretest and posttest are displayed in Figures 5 and 6.

First, in the pretest, there was a significant main effect of language sequence, F(1, 38) = 13.43, MSE = 8.24, p < .01, η²_P = .26, with larger N2 amplitudes in switch trials than in repeat trials. However, there was no significant three-way interaction of group × language × language sequence, F(1, 38) = 1.89, MSE = 19.30, p > .05, η²_P = .05.

Second, regarding the training benefits between the pretest and posttest for the low-IC group, there was a significant main effect of language sequence, F(1, 38) = 19.37, MSE = 7.20, p < .001, η²_P = .34, with larger N2 amplitudes in switch trials than in repeat trials. But the three-way interaction test × language × language sequence did not reach significance, F(1, 38) = 1.76, MSE = 14.35, p > .05, η²_P = .04.

Third, regarding the comparison between the pretest and posttest for the high-IC group, there was a significant main effect of language, F(1, 38) = 5.14, MSE = 14.87, p < .05, η²_P = .12. Furthermore, the interaction between language × language sequence reached significance, F(1, 38) = 6.33, MSE = 14.22, p < .05, η²_P = .12. Further analysis showed that Lnew repeat trials elicited larger amplitudes than L1 ones, F(1, 39) = 12.14, p < .01, but L1 switch trials did not differ from Lnew ones in amplitudes, F(1, 39) = .25, p > .05. The three-way interaction of test × language × language sequence did not reach significance, F(1, 38) = .54, MSE = 14.22, p > .05, η²_P = .01.

Finally, in the posttest, there was a significant main effect of language, F(1, 38) = 5.67, MSE = 9.17, p < .05, η²_P = .13. Moreover, the interaction between language × language sequence reached significance, F(1, 38) = 7.74, MSE = 9.26, p < .01, η²_P = .17. Further analysis revealed that Lnew repeat trials elicited larger amplitudes than L1 ones, F(1, 39) = 7.81, p < .01; while L1 switch trials did not differ from Lnew ones in amplitudes, F(1, 39) = .02, p > .05. However, the three-way interaction between group × language × language sequence was not significant, F(1, 38) = .33, MSE = 9.26, p > .05, η²_P = .01.

Stimulus-locked N2 time window (270–370 ms)

The stimulus-locked grand average waveforms and topographic maps of L1 and Lnew trials for the high- and low-IC groups in the pretest and posttest are displayed in Figures 7 –10.

First, in the pretest, there was a significant interaction between group × language sequence, F(1, 38) = 7.92, MSE = 7.64, p < .01, η²_P = .17. Further analysis showed that the high-IC group elicited larger amplitudes in switch trials than in repeat trials, F(1, 19) = 5.59, p < .05, while the low-IC group did not show any such difference, F(1, 19) = 2.83, p > .05. The interaction between language × language sequence was significant, F(1, 38) = 15.52, MSE = 6.35, p < .001, η²_P = .29. Further analysis exhibited that L1 and Lnew repeat trials did not differ in amplitudes, F(1, 39) = .14, p > .05, while Lnew switch trials elicited larger amplitudes than L1 ones, F(1, 39) = 13.84, p < .01. However, there was no significant three-way interaction of group × language × language sequence, F(1, 38) = 3.09, MSE = 6.35, p > .05, η²_P = .08.

Second, regarding the training benefits for the low-IC group, there was a significant interaction between language × language sequence, F(1, 38) = 29.76, MSE = 6.76, p < .001, η²_P = .44. Further analysis showed that L1 and Lnew repeat trials did not differ in amplitudes, F(1, 39) = 2.00, p > .05, but Lnew switch trials elicited larger amplitudes than L1 ones, F(1, 39) = 25.40, p < .001. However, the three-way interaction of test × language × language sequence was not significant, F(1, 38) = .01, MSE = 6.76, p > .05, η²_P = .00.

Third, regarding the comparison between the pretest and posttest for the high-IC group, there was a significant main effect of language sequence, F(1, 38) = 4.88, MSE = 21.98, p < .05, η²_P = .11, with larger amplitudes in switch trials than in repeat trials. However, the three-way interaction of test × language× language sequence was not significant, F(1, 38) = .83, MSE = 10.47, p > .05, η²_P = .02.

Finally, in the posttest, there were no significant interaction of test × language× language sequence, F(1, 38) = 4.08, MSE = 10.88, p > .05, η²_P = .11.

Altogether, inhibition may not play a significant role during the language task schema competition phase, as indicated by cue-locked N2, or during the early processing of stimulus as evidenced by stimulus-locked N2. Moreover, no training benefits were observed in cued-locked N2 or stimulus-locked N2.

Stimulus-locked LPC time window (450–650 ms)

The comparison of pretest between the high- and low-IC groups

The main effect of language reached significance, F(1, 38) = 4.75, MSE = 50.92, p < .05, η²_P = .11. The interaction between language × language sequence was significant, F(1, 38) = 4.50, MSE = 43.73, p < .05, η²_P = .11. The three-way interaction of group × language × language sequence was significant, F(1, 38) = 5.54, MSE = 56.52, p < .05, η²_P = .13. A simple effect analysis of this three-way interaction showed that the amplitude in L1 repeat trials was not significantly larger than that in Lnew repeat trials for both the high-IC group, F(1, 38) = .01, p > .05 (see Figure 6), and the low-IC group, F(1, 38) = .46, p > .05 (see Figure 9). The amplitudes in Lnew switch trials were larger than those in L1 trials for the high-IC group, F(1, 38) = 137.59, p < .001 (see Figure 8), but again no difference was observed for the low-IC group, F(1, 38) = 3.48, p > .05 (see Figure 10). These findings suggest that the high-IC group, due to their strong inhibitory control ability, can effectively suppress interference from L1 when switching to Lnew.

Training benefits: the comparison between pretest and posttest for the low-IC group

The main effect of language reached significance, F(1, 38) = 8.88, MSE= 57.76, p < .01, η²_P = .19. The interaction between language × language sequence was significant, F(1, 38) = 4.49, MSE = 43.90, p < .05, η²_P = .11. The three-way interaction of test × language × language sequence was significant, F(1, 38) = 4.64, MSE = 65.32, p < .05, η²_P = .11. A simple effect analysis of the three-way interaction showed that, for the pretest, L1 and Lnew repeat trials did not exhibit difference in amplitudes, F(1, 38) = .38, p > .05, and the same was true for the posttest, F(1, 38) = .33, p > .05 (see Figure 9). Furthermore, L1 and Lnew switch trials failed to show a difference for the pretest, F(1, 38) = 1.85, p > .05, while Lnew switch trials elicited larger amplitudes than those of L1 ones for the posttest, F(1, 38) = 41.67, p < .001 (see Figure 10). Importantly, the interaction of language × language sequence reached significance for the low-IC group in the posttest, F(1, 19) = 8.76, MSE = 64.28, η²_P = .32, p < .01. These findings suggest that inhibition was strengthened significantly after training, and the low-IC group could easily suppress interference from L1.

The comparison between pretest and posttest for the high-IC group

The main effect of language reached significance, F(1, 38) = 5.31, MSE = 68.66, p < .05, η²_P = .12. The interaction between language × language sequence was significant, F(1, 38) = 13.67, MSE = 53.59, p < .01, η²_P = .27. Further analysis showed that L1 repeat trials did not differ from Lnew ones, F(1, 39) = .22, p > .01, while Lnew switch trials elicited larger amplitudes than L1 ones, F(1, 39) = 122.87, p < .001. The high-IC group did not show a significant three-way interaction of test × language × language sequence, F(1, 38) = .04, MSE = 69.68, η²_P = .001, p > .05. Furthermore, the interaction of language × language sequence reached significance for the high-IC group in the posttest, F(1, 19) = 5.05, p < .05, indicating that the high-IC group could suppress cross-language interference effectively.

The comparison of posttest between the high- and low-IC groups

The two-way interaction between language × language sequence reached significance, F(1, 38) = 13.64, MSE = 53.77, p < .01, η²_P = .26. A simple effect analysis showed that L1 repeat trials did not differ from Lnew ones, F(1, 39) = .91, p > .05, while Lnew switch trials elicited larger amplitudes than L1 ones, F(1, 39) = 69.64, p < .001. The three-way interaction of group × language × language sequence was not significant, F(1, 38) = .03, MSE = 78.79, η²_P = .001, p > .05. This finding further demonstrated that the IC ability of the low-IC group was significantly improved after intensive training, and such improvements enabled the low-IC group to perform as well as the high-IC group.

Discussion

Using a pretest/posttest design and ERP methodology, we investigated the effect of different levels of IC abilities in language-switching tasks between Chinese (L1) and Korean (Lnew) on Chinese–English bilinguals. Low-IC participants received training in an inhibitory control task between the pretest and the posttest. Consistent with the previous study (Liu et al., 2014), there were some significant differences in the amplitude of LPC in the pretest between high-IC and low-IC groups, indicating that inhibition may occur at the lexical selection response phase during L1–Lnew switching. Moreover, inhibition can be improved for the low-IC group via intensive training, as shown by the LPC results in the posttest, also suggesting that inhibition may occur at this later phase. Overall, we find evidence to support the hypothesis that IC ability plays a role in suppressing non-target lexical access during initial language switching. More important, the inhibition training contributes to such suppression.

In the present study, the high-IC group showed symmetrical switch costs in the pretest, and a more obvious LPC when switching into Lnew than switching into L1, while the low-IC group did not show similar results in the pretest. We speculated that this should due to the influence of inhibition on language switching. Specifically, the high-IC group is able to effectively recruit inhibition to suppress the cross-language interference, showing a larger LPC in switching to Lnew, and correspondingly they can easily release the previously suppressed L1 lemma when switching back to L1 with a smaller LPC. In contrast, the low-IC group is not able to effectively and quickly recruit inhibition. They struggled to suppress cross-language interference, which cause a persisting inhibition. Subsequently, in order to get rid of such persisting inhibition, more time is needed for them to release the previously suppressed lemma, leading to asymmetrical switch costs and similar LPCs when switching into L1 and Lnew.

Our findings are consistent with De Bruin et al.’s (2014) results. They found that compared with non-switch trials, switching to L2 and L3 would activate the domain-general inhibition areas such as the right inferior frontal gyrus and the pre-supplementary motor area, and exhibit similar switch costs in L1, L2, and L3. Thus, effective inhibition should elicit a larger LPC. Finally, although we both examined the language switching on low-proficient bilinguals, Jackson et al. (2001) did not measure the IC ability of participants, and they obtained asymmetrical language-switch costs in a single group of participants. Yet, we took such ability into account and found symmetrical language-switch costs for the high-IC group and asymmetrical language-switch costs for the low-IC group, suggesting that inhibitory control ability per se plays a crucial role during language switching. Therefore, we claim that the IC advantage could contribute to suppressing the interference from the non-target lemma during L1 and Lnew switching for the low-proficient bilinguals, indicating that inhibition probably works at the lexical selection phase.

This finding that inhibition may occur in the lexical selection phase is consistent with previous studies. For example, switch trials elicited a larger LPC than repeat trials, so the authors speculated that the LPC correlates closely with the selection of lemma in the desired language (Jackson et al., 2001; Martin et al., 2013). Altogether, based on the previous evidence and our findings, the LPC effect in the current study could be interpreted as inhibition of non-target lemmas in the lexical selection response phase.

Interestingly, the low-IC group, after receiving domain-general inhibition training, showed similar LPC effects as the high-IC group. This suggests that domain-general inhibition training enables the low-IC group to effectively suppress the L1 lemma when switching into Lnew, thereby eliciting a large LPC. The increased LPC reflects training benefits on language switching, consistent with previous studies. For example, Benikos, Johnstone, and Roodenrys (2013) found the group that received training with more difficult tasks exhibited a larger P3 in a no-go task than that received training with less difficult tasks, indicating that inhibition was strengthened via training. Furthermore, Berkman, Kahn, and Merchant (2014) observed that the training group, after receiving the stop-signal task training, exhibited a stronger activation of the right inferior frontal gyrus compared with the sham-training group. Importantly, the benefit of domain-general executive function training could be transferred to domain-specific language processing (Novick, Hussey, Teubner-Rhodes, Harbison, & Bunting, 2014). Thus, we ensured that the low-IC group indeed benefited from the domain-general inhibition training.

Inhibition benefits switching between one’s native language and a new learned language. To a larger extent, this finding is in line with the studies concerning the relationship between domain-general inhibition and language learning; that is, the stronger the domain-general inhibition, the better the performance in learning a new language (Kapa & Colombo, 2014; Yoshida, Tran, Benitez, & Kuwabara, 2011). Importantly, even after the new language has been learned for a period, such inhibition still plays an important role during L1–Lnew switching, just like during L1–L2 switching (e.g. Costa & Santesteban, 2004; Jackson et al., 2001; Liu et al., 2014; Martin et al., 2013). This finding suggests that educators and new-language learners should not only focus on strengthening the language proficiency, but also pay attention to improving the learner’s IC ability.

The N2 amplitude did not show a three-way interaction of group/test, language, and language sequence, so we argued that inhibition may not function during the phase of language task schema competition. According to previous studies, N2 reflects attentional control during language switching (Verhoef et al., 2010). We speculate that N2 is probably a reflection of attentional control processes during which the appropriate language (e.g. L1 or L2) is selected.

Considering our findings, we would like to supplement the IC model as follows. Specifically, the degree of inhibition for language is not only proportional to the level of activation of a language, but also related to the IC ability. This means that low-proficient bilinguals with high IC ability could show symmetrical switch costs due to their inhibition advantage and could effectively suppress cross-language interference during language switching in the initial period of language learning. However, a potential limitation of this study is that only the low-IC group received training, and there were no control conditions for the low-IC group. Nonetheless, we ensured that the low-IC group indeed benefited from the inhibition training for the following reasons. First, the posttest results of the low-IC group were compared with their pretest results, not with the posttest results of the high-IC group. Second, the low-IC group indeed strengthened their IC ability after the inhibition training (see Figure 3), which was indexed by similar congruent and incongruent switch costs. Finally, if switching-task experience enabled the low-IC group to perform better, they would not only have exhibited changes in the LPC, but also in the N2 component. In fact, the low-IC group did not show any significant difference in the N2 between the pretest and posttest, but only exhibited significant difference in the LPC. Such benefits in the LPC patterns cannot be attributed to practice effect, because the language-switching words used in the pretest were different from those in the posttest. Overall, it is very likely that it is the enhanced inhibitory control ability after training that improved the low-IC group’s language-switching efficiency. Besides, another limitation is that the two participant groups’ working memory was not matched. The interference of working memory should be controlled in future work. Nevertheless, the cued language-switching task adopted by the current study is very simple, and thus may not cause a heavy load on the working memory.

To sum up, the present study provides direct evidence that inhibitory control benefits bilinguals in suppressing non-target lexical interference during language switching for new-language learners. More importantly, the low-IC group could effectively suppress non-target lemmas after intensive inhibition training, and exhibited symmetrical switch costs similar to the high-IC group in the posttest, suggesting that, to an extent, inhibition training could facilitate language switching.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author biographies

Huanhuan Liu got a PhD in Psychology from the School of Psychology at Beijing Normal University, and now works at Research Center of Brain and Cognitive neuroscience at Liaoning Normal University. Her research interests focus on language switching, task switching and inhibition.

Susan Dunlap works at a Children’s Learning Institute at University of Texas Health Science Center. She interests in morphosyntactic processing, bilingual processing and L2 vocabulary learning.

Lijuan Liang works as a Lecturer at the School of English and Education at Guangdong University of Foreign Studies. Her research focuses on phonological-orthographic processing during second language acquisition.

Baoguo Chen is a Professor from the School of Psychology at Beijing Normal University. He mainly conducts research on language switching, L2 vocabulary learning, L2 semantic and syntactic processing.

References

Allport

Styles

E. A.

Hsieh

(1994). Shifting intentional set: Exploring the dynamic control of tasks. In Umiltà

Moscovitch

(Eds.), Attention and performance XV: Conscious and non-conscious information processing (pp. 421–452). Cambridge, MA: MIT Press.

Allport

D. A.

Wylie

G. R.

(1999). Task-switching: positive and negative priming of task-set. In Humphreys

G. W.

Duncan

Treisman

A. M.

, (Eds.), Attention, space and action: Studies in cognitive neuroscience (pp. 273–296). Oxford: Oxford University Press.

Benikos

Johnstone

S. J.

Roodenrys

S. J.

(2013). Short-term training in the Go/Nogo task: Behavioural and neural changes depend on task demands. International Journal of Psychophysiology, 87, 301–312.

Berkman

E. T.

Kahn

L. E.

Merchant

J. S.

(2014). Training-induced changes in inhibitory control network activity. The Journal of Neuroscience, 34(1), 149–157.

Bialystok

(2006). Effect of bilingualism and computer video game experience on the Simon task. Canadian Journal of Experimental Psychology, 60, 68–79.

Christoffels

I. K.

Firk

Schiller

N. O.

(2007). Bilingual language control: An event-related brain potential study. Brain Research, 1147, 192–208.

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

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: Learning, Memory & Cognition, 32, 1057–1074.

De Bruin

Roelofs

Dijkstra

FitzPatrick

(2014). Domain-general inhibition areas of the brain are involved in language switching: FMRI evidence from trilingual speakers. NeuroImage, 90, 348–359.

10.

Declerck

Koch

Philipp

A. M.

(2014). The minimum requirements of language control: Evidence from sequential predictability effects in language switching. Journal of Experimental Psychology: Learning, Memory, and Cognition. doi.org/10.1037/xlm0000021

11.

Gratton

Coles

M. G. H.

Donchin

(1983). A new method for off-line removal of ocular artifact. Electroencephalography and Clinical Neurophysiology, 55, 468–484.

12.

Green

D. W.

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

13.

Guo

Liu

(2013). An ERP study of inhibition of non-target languages in trilingual word production. Brain & Language, 127, 12–20.

14.

Hilchey

M. D.

Klein

R. M.

(2011). Are there bilingual advantages on nonlinguistic Interference tasks? Implications for the plasticity of executive control processes. Psychonomic Bulletin & Review, 18, 625–658.

15.

Jackson

G. M.

Swainson

Cunnington

Jackson

S. R.

(2001). ERP correlates of executive control during repeated language switching. Bilingualism: Language and Cognition, 4, 169–178.

16.

Kapa

L. L.

Colombo

(2014). Executive function predicts artificial language learning. Journal of Memory and Language, 76, 237–252.

17.

Linck

J. A.

Schwieter

J. W.

Sunderman

(2012). Inhibitory control predicts language switching performance in trilingual speech production. Bilingualism: Language and Cognition, 15, 651–662.

18.

Liu

Fan

Rossi

Yao

Chen

(2016). The effect of cognitive flexibility on task switching and language switching. International Journal of Bilingualism, 20(5), 563–579.

19.

Liu

Rossi

Zhou

Chen

(2014). Electrophysiological evidence for domain-general inhibitory control during bilingual language switching. PLoSONE, 9(10), e110887.

20.

Martin

C. D.

Strijkers

Santesteban

Escera

Hartsuiker

R. J.

Costa

(2013). The impact of early bilingualism on controlling a language learned late: An ERP study. Frontiers in Psychology, 4, 1–15.

21.

Misra

Guo

Bobb

S. C.

Kroll

J. F.

(2012). When bilinguals choose a single word to speak: Electrophysiological evidence for inhibition of the native language. Journal of Memory and Language, 67, 224–237.

22.

Novick

J. M.

Hussey

Teubner-Rhodes

Harbison

J. I.

Bunting

M. F.

(2014). Clearing the garden-path: Improving sentence processing through cognitive control training. Language, Cognition, and Neuroscience, 29, 186–217.

23.

Perani

Abutalebi

Paulesu

Brambati

Scifo

Cappa

S. F.

Fazio

(2003). The role of age of acquisition and language usage in early, high-proficient bilinguals: An fMRI study during verbal fluency. Human brain mapping, 19(3), 170–182.

24.

Philipp

A. M.

Gade

Koch

(2007). Inhibitory processes in language switching: Evidence from switching language-defined response sets. European Journal of Cognitive Psychology, 19, 395–416.

25.

Snodgrass

J. G.

Vanderwart

(1980). A standardized set of 260 pictures: Norms for naming agreement, familiarity, and visual complexity. Journal of Experimental Psychology: Human Learning & Memory, 6, 174–215.

26.

Verhoef

Roelofs

Chwilla

D. J.

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

27.

Verhoef

K. M. W.

Roelofs

Chwilla

D. J.

(2010). Electrophysiological evidence for endogenous control of attention in switching between languages in overt picture naming. Journal of Cognitive Neuroscience, 22, 1832–1843.

28.

Waldron

E. J.

Hernandez

A. E.

(2013). The role of age of acquisition on past tense generation in Spanish–English bilinguals: An fMRI study. Brain and Language, 125(1), 28–37.

29.

Wang

Xue

Chen

Xue

Dong

(2007). Neural bases of asymmetric language switching in second-language learners: An ER-fMRI study. NeuroImage, 35, 862–870.

30.

Yoshida

Tran

D. N.

Benitez

Kuwabara

(2011). Inhibition and adjective learning in bilingual and monolingual children. Frontiers in Psychology, 2, 210.

31.

Zhang

Yang

(2003). The determiners of picture naming latency. Acta Psychologica Sinica, 35, 447–454.