Abstract
Aims and Objectives:
The current study aimed to investigate whether there was inhibitory processing in code-switching during language comprehension among Chinese–English–Japanese trilinguals and whether n–2 language repetition costs were contingent upon particular stimulus–response configurations.
Methodology:
This experiment employed semantic categorization tasks, with non-cognate animal and non-animal words as experimental stimuli, and the n–2 language repetition costs as a discerning experimental indicator, to explore whether there was inhibitory processing in code-switching during language comprehension among Chinese–English–Japanese trilinguals. Furthermore, we investigated the impact of stimulus–response configuration on inhibitory processing during switching between phonetic and logographic writing systems. This approach allowed for a detailed exploration of the cognitive mechanisms underpinning language switching and the factors influencing inhibitory control in multilingual contexts.
Data and Analysis:
The experimental data collected were analyzed using the R language with the lme4 package. Reaction times were fitted to a linear mixed-effects model, while accuracy was binary coded and fitted to a generalized mixed-effects model with a logistic link function.
Findings:
The study found that: (1) there were n–2 language repetition costs in Chinese–English–Japanese trilinguals’ code-switching during language comprehension and (2) the n–2 language repetition costs in code-switching processing during language comprehension was not affected by specific stimulus–response configurations but was related to the psychological representation of the competing languages.
Originality:
There is a lack of research investigating the switching of three language tasks using n–2 language repetition costs as an index. Limited studies employing n–2 language repetition costs as an index have predominantly focused on the level of language production, with language comprehension yet to be examined. In addition, there exists a dearth of research addressing the inhibitory mechanisms involved in the code-switching process between phonetic and logographic writing systems; even more limited attention has been directed toward the intricacies of switching between three non-cognate languages.
Significance/Implications:
The study reveals the inhibitory processing involved in code-switching between phonetic and logographic writing systems. It extends previous findings concerning the inhibitory processing of code-switching during language production within phonetic languages, providing new insights into the research on code-switching and the n–2 language repetition costs among multilingual individuals.
Introduction
In the era of globalization, the importance of cross-cultural and cross-linguistic communication has been magnified, leading to an increased focus on the phenomenon of multilingualism (Sokolovska, 2023; Zhong, 2023; Zhong & Fan, 2023a). Among the various linguistic phenomena observed in multilingual individuals, code-switching stands out for its ubiquity and complexity (Li, 2007; Zhong & Fan, 2023b). This practice, often seen during informal conversations, involves the dynamic alternation between two or more languages or dialects and has drawn considerable attention from researchers (Grosjean, 2001). Code-switching not only reflects the linguistic agility of multilingual speakers but also serves as a window into understanding the cognitive mechanisms underpinning language processing (Grosjean, 1982). Critical to this discussion is the role of inhibitory control in managing potential interference between languages. As individuals switch from one language to another, they need to inhibit the non-target language to minimize confusion and ensure communication efficacy (Grosjean, 1998; Kroll et al., 2012). The concept of switching costs, defined as the time delay and decreased accuracy observed when transitioning between languages (Costa et al., 2006; Green, 1998), is pivotal in studying these cognitive dynamics. Such costs, which can vary depending on the direction of the switch, provide insights into the asymmetric nature of language switching (Grainger & Beauvillain, 1987; Meuter & Allport, 1999). However, recent studies challenge the conventional understanding of switching costs and their implications for cognitive processing. For instance, a comprehensive meta-analysis by Gade et al. (2021a, 2021b) finds no substantial evidence for switch cost asymmetry in language production, questioning the use of such asymmetry as a definitive marker for inhibitory control mechanisms. Similarly, the work of Philipp, Gade, and Koch (2007) and Declerck et al. (2015) offers valuable insights that necessitate a re-evaluation of how we interpret the cognitive underpinnings of code-switching. Philipp, Gade, and Koch (2007) utilized language-defined response sets (digit names from 1 to 9 in various languages) to investigate inhibitory processes during language switching. Their study revealed a shift cost in switching between languages, with a notable asymmetry favoring the dominant language, suggesting differential functional characteristics of inhibitory processes in language switching. These findings are further complicated by the work of Declerck et al. (2015), who examined highly proficient bilinguals and found substantial n–2 language repetition costs, a marker of inhibition when switching back to a recently abandoned language. This effect was significant for all three languages studied, with a more pronounced effect for the two dominant languages. The insights provided by Gade et al. (2021a, 2021b), Philipp, Gade, and Koch (2007), and Declerck et al. (2015) underscore the necessity for a more nuanced understanding of the cognitive processes involved in multilingual language production, moving beyond simplistic interpretations of switching costs and their asymmetries.
Code-switching can be divided into two major categories, code-switching at the level of language production and code-switching at the level of language comprehension. In the field of code-switching during the language comprehension process, researchers often use the reading paradigm for experimental studies (Dalrymple-Alford, 1985). Dalrymple-Alford (1985) observed that participants exhibit a notably elevated reading speed when engaging with texts exclusively composed in a singular language compared with texts necessitating the switching between two languages. Lexical judgment, a commonly employed experimental paradigm, entails the task of discerning between authentic words and pseudo-words (Von Studnitz & Green, 1997; Thomas & Allport, 2000). It is worth noting that in this paradigm, the magnitude of the switching cost is influenced by the characteristics of the task and the nature of the responses that multilingual speakers need to make (Thomas & Allport, 2000). Specifically, in a certain lexical judgment task, when the demand was to determine whether a string of letters was a real word in one language but a pseudo-word in another, the switching cost was significantly higher than in a general lexical judgment task that required the determination of whether a string of letters was a real word in multiple languages (see Von Studnitz & Green, 2002). Furthermore, in non-word rejection tasks, experimental participants showed a higher sensitivity to the word form legality of letter strings in non-target languages (Altenberg & Cairns, 1983). Cross-linguistic semantic categorization tasks are deemed to be more suitable as experimental paradigms relative to lexical judgment tasks, particularly in the investigation of the processing of switching costs. This preference is grounded in their proximity to the authentic utilization of natural language, thereby enhancing the ecological validity of the study (Von Studnitz & Green, 2002). In this paradigm, semantics rather than word form is the most critical variable. Other investigations have incorporated cross-linguistic priming paradigms (e.g., Grainger & Beauvillain, 1987) to delve into the mechanisms underlying code-switching within the language comprehension process. In the cross-linguistic priming paradigm, the prime and target words belong to the same language or different languages. Prior studies showed that participants’ reaction times were significantly longer when the prime and target words belonged to different languages than when those belonged to the same language.
In code-switching-related research, some scholars have conducted experiments using the n–2 language repetition costs as an indicator. The concept of n–2 repetition costs originated from a task-switching paradigm designed by Mayr and Keele (2000) to measure inhibitory processing. In this paradigm, participants switch between three different tasks. The aftereffects of inhibition are examined by comparing the performance of participants in n–2 repetition sequences (i.e., ABA sequence, where the first and third tasks are the same) with n–2 non-repetition sequences (i.e., CBA sequences, where all three tasks differ). Mayr and Keele (2000) observed a decrement in participant performance on ABA sequences compared with CBA sequences, identifying this phenomenon as “backward inhibition” (Mayr & Keele, 2000, p. 5). In other words, once a task that has just been executed is inhibited, this inhibition will persist. When participants switch back to performing the inhibited task, they will expend more cognitive resources (Mayr & Keele, 2000). Mayr and Keele attributed the increased reaction time and error rate in ABA task sequences to the continued inhibition of the task that had just been switched away from. Subsequently, researchers have used the more theoretically neutral term “n–2 repetition cost” to refer to this phenomenon; it is often used by researchers as an experimental marker of sustained inhibition in switching tasks. In studies that adopted the n–2 repetition costs, there were mainly two types of task sequences, ABA sequences (n–2 task repetition) and CBA sequences (n–2 task non-repetition). Participants showed increased response times and error rates for the final Task A in ABA sequences relative to CBA sequences. This was because when performing Task B in ABA sequences, Task A had just been performed and, due to inertia, the setup for Task A still existed to some extent. To execute Task B smoothly, the setup for Task A was inhibited during the setup for Task B. Due to the inertia effect of this task set (which includes inhibitory components), the final A task in ABA sequences was inhibited (Philipp, Gade, and Koch 2007). The n–2 repetition costs are largely believed to depend on the interference resolution of preceding trials (Schuch & Koch, 2003). Many studies have also reported the existence of the n–2 repetition costs (e.g., Arbuthnott, 2005; Fan & Li, 2013; Philipp & Koch, 2005; Philipp & Koch, 2009; Philipp, Gade, and Koch 2007; Jolicoeur et al., 2007). The n–2 task repetition costs serve as compelling evidence for task inhibition, and substantial n–2 task repetition costs have been found in various tasks, including language switching processes (Koch et al., 2010; Mayr, 2007; Scheil & Kleinsorge, 2023). A crucial component in understanding these dynamics involves the examination of stimulus–response combinations, which refers to the configurations wherein participants’ response to the stimulus type either repeats or switches. In the current study, the stimulus is the word presented to participants, and the response is the key press indicating whether the word belongs to the category of animals or not. Specifically, if the word belongs to the category of animals, press the left “f” key; if not, press the right “j” key. Such response patterns have implications for episodic memory contributions, a factor extensively explored in non-language classification tasks (Grange et al., 2017; Mayr, 2002). Grange et al. (2017) and Mayr (2002) delve into the effects of episodic retrieval on n–2 repetition costs, proposing that episodic memory can modulate these costs, thus offering an alternative explanation to inhibitory accounts. This research underscores the episodic memory’s role in task performance by showing that response patterns, influenced by prior episodic retrieval, can either facilitate or impede task execution. Previous studies also show that the stimulus–response configuration plays a pivotal role in understanding the cognitive costs associated with language switching. Philipp and Koch (2009) observed that n–2 language repetition costs under stimulus–response configuration repetition conditions were less than those under non-repetition conditions. This finding challenges the hypothesis that inhibition of a specific stimulus–response configuration should augment n–2 language repetition costs, thereby suggesting that the conventional explanation of inhibition does not fully account for the observed phenomena in code-switching. The discrepancy observed by Philipp and Koch (2009) invites further investigation into the cognitive processing differences across languages and modalities of language use (production vs. comprehension).
To date, investigations into the inhibitory mechanisms underlying code-switching predominantly derive insights from examinations focused on the interplay between two languages. Utilizing the asymmetry of switching costs as an experimental proxy, extant research findings contribute to our understanding of the inhibitory processing intricacies inherent in code-switching phenomena. There is a lack of research investigating the switching of three language tasks using n–2 language repetition costs as an index. Limited studies employing n–2 language repetition costs as an index have predominantly focused on the level of language production, with language comprehension yet to be examined. In addition, existing deductions pertaining to inhibitory processing in code-switching predominantly emanated from investigations focusing on the switching between phonetic languages. However, there exists a dearth of research addressing the inhibitory mechanisms involved in the code-switching process between phonetic and logographic writing systems; even more limited attention has been directed toward the intricacies of switching between three non-cognate languages. In light of this, the present study employed a semantic categorization task to explore whether there was inhibitory processing in Chinese–English–Japanese (isolating-inflecting-agglutinating language) trilinguals during language comprehension and the effect of stimulus–response configuration on the inhibitory processing during the switching process between phonetic and logographic writing systems. Our investigation utilized non-cognate animal and non-animal words as experimental stimuli, with the n–2 language repetition costs as a discerning experimental indicator. The study sought to elucidate the ramifications of stimulus–response configuration on the processing of code-switching within the context of phonetic and logographic writing systems. This endeavor extends the antecedent research findings concerning the inhibitory processing of code-switching during language production within phonetic languages.
Research questions
This study aimed to investigate whether there was inhibitory processing in code-switching during language comprehension among Chinese–English–Japanese trilinguals, particularly whether the n–2 language repetition costs were related to specific stimulus–response configurations, as indicated by the magnitude of the n–2 language repetition costs. We were interested in whether the language used in n–2 trials was inhibited and whether the specific stimulus–response configuration used in n–2 trials was inhibited. This experiment used a semantic categorization task to examine the impact of specific stimulus–response configurations on the n–2 language repetition costs by measuring participant responses in ABA and CBA language sequences. The specific research questions are as follows:
RQ1. Are there n–2 language repetition costs in code-switching among Chinese–English–Japanese trilinguals during language comprehension?
RQ2. If so, are the n–2 language repetition costs in code-switching among Chinese–English–Japanese trilinguals during language comprehension related to specific stimulus–response configurations?
Research method
Participants
Thirty undergraduate students from Beijing Foreign Studies University participated in the experiment. The participants were aged between 18 and 22 years, with an average age of 20.47 years (SD = 1.20). All participants’ first language (L1) was Chinese, second language (L2) English, and third language (L3) Japanese. Participants’ average learning time for L2 was between 12 and 19 years, with an average of 14.90 years (SD = 1.64). Their learning time for L3 ranged from 1 to 3 years, with an average learning time of 1.90 years (SD = 0.75). Participants were proficient in Chinese, relatively proficient in English, and not proficient in Japanese. Before the experiment, participants filled out a questionnaire on their language learning experiences. The questionnaire design drew inspiration from the work of Li et al. (2020), wherein participants were prompted to assess their proficiency in Chinese, English, and Japanese using a 7-point scale (“7” = “very proficient” and “1” = “not proficient at all”). The average self-assessed proficiency rating for Chinese was 6.95 (SD = 0.14), for English was 4.90 (SD = 0.60), and for Japanese was 2.74 (SD = 0.75). The results of the analysis of variance (ANOVA) for self-assessed proficiency ratings in these three languages showed a significant difference (p < .001). The participants’ self-assessed proficiency levels for Chinese, English, and Japanese are shown in Figure 1. Post hoc multiple comparisons indicated that there were significant differences in average proficiency levels between Chinese and English, English and Japanese, and Chinese and Japanese (ps < .001). All participants exhibited right-handedness and reported no history of either brain damage or psychiatric illness. In addition, participants possessed normal vision or corrected vision and were devoid of color blindness. Prior to the experiment, participants provided written informed consent. Post-experiment, participants received compensation as stipulated.
Participants’ self-assessed proficiency in Chinese, English, and Japanese.
Experimental design
A 3 (language type: Chinese, English, Japanese) × 2 (language sequence: ABA, CBA) × 2 (stimulus–response configuration: n–2 repetition, n–2 non-repetition) within-subjects factorial design was adopted. Language type, language sequence, and stimulus–response configuration were the independent variables, while accuracy and reaction time were the dependent variables.
Experimental materials and tasks
The experimental stimuli comprised non-cognate animal words (fish, bird, dog, pig, duck, cat) and non-cognate non-animal words (moon, wind, rain, snow, hill, river), selected to delineate ABA and CBA language sequences. The choice of stimuli was intended to contrast sequences where the stimulus–response configuration in n–2 trials was identical to that in n trials (stimulus–response configuration repetition) against those where the configuration in n–2 trials differed from n trials (stimulus–response configuration non-repetition). Each set of trials consisted of three trials. The experiment was designed with a total of 288 sets of trials, divided into two main types based on the sequence of languages used: ABA and CBA. Each of these sequence types comprised 144 sets of trials, ensuring a balanced exploration of language switching patterns. Within these sequences, we further categorized the trials based on the stimulus–response configuration’s repetition from n–2 to n trials, resulting in two subcategories: n–2 repetition and n–2 non-repetition, each consisting of 144 sets of trials. This categorization was crucial for analyzing the cognitive processes underlying language switching and repetition effects. To delve into the specifics of language involvement, the final Task A within each set of trials was evenly distributed across three languages: Chinese, English, and Japanese, with each language being the focus of the final Task A within 96 sets of trials. This distribution allowed for a comprehensive examination of how the final language in a trial sequence influences language processing and switching dynamics. Participants were assigned the task of making semantic categorizations of these stimuli, which were presented in three languages: Chinese, English, and Japanese. To control for potential confounding variables, we ensured the items selected for the experiment had comparable usage frequencies across the three languages. This was achieved by referencing established language corpora, namely, the Beijing Language and Culture University Chinese Corpus (BCC) for Chinese, the Corpus of Contemporary American English (COCA) for English, and the Tsukuba Web Corpus (TWC) for Japanese. This selection process aimed to minimize memory-related influences on processing fluency by equating item frequency across languages. Throughout the experiment, we maintained an equal number of stimuli presentations in each language sequence, stimulus–response configuration, and language type to ensure a well-balanced linguistic representation and enable a robust analysis of language switching phenomena.
Experimental procedure
The experiment was implemented and executed utilizing E-Prime 3.0, which concurrently served as the data collection platform. Preceding the formal initiation of the experiment, participants underwent a practice phase. The practice phase allowed for repeated iterations until participants achieved a level of familiarity with the experimental procedures. They would then press the “q” key to start the actual experiment. Each set of trials consisted of three trials. A red fixation point “+” was displayed for 500 ms on the computer screen first, followed by a word in Chinese, English, or Japanese lasting for 1,500 ms, during which participants were required to categorize the attribute of the word. If it was an animal word, they would press the “f” key; if it was a non-animal word, they would press the “j” key. If the participant did not respond within 1,500 ms, the word would disappear. There was a 500 ms interval between each trial (see Figure 2 for the experimental procedure). The computer automatically recorded the reaction time and accuracy of the participants. The formal experiment included 4 sessions of 72 sets of trials each, totaling 288 sets of trials. In each session, the number of occurrences of each language (Chinese, English, Japanese), stimuli (animal and non-animal words), language sequences (ABA and CBA), and stimulus–response configurations (n–2 sequence repetitions and n–2 sequence non-repetitions) were equal. The presentation of the animal and non-animal words as stimuli was unpredictable to the participants. Our programming did not allow for direct stimulus repetition so that we could more accurately assess participants’ inhibition processing in code-switching, without potential interference from previous exposures to the same stimulus. This practice helps maintain the integrity and validity of the current experimental design and enhances the reliability of our findings.
Experimental procedure.
Data analysis
The experimental data were analyzed using the R language (R Core Team, 2023) with the lme4 package (Bates et al., 2015) for linear mixed-effects models for reaction times, and generalized mixed-effects models with a logistic link function for accuracy (see Bates et al., 2023). Our models incorporated three fixed factors: language type (Chinese, English, Japanese), language sequence (ABA, CBA), and stimulus–response configuration (n–2 repetition, n–2 non-repetition), with subject and item as crossed random factors for reaction time and subject as a random factor for accuracy to account for variability among participants and stimuli (see Baayen et al., 2008; Kuznetsova et al., 2017). We integrated language type as a fixed factor into our data analysis in response to the inconsistencies about differential inhibition deployed because of language dominance reported in previous studies, such as those by Philipp, Gade, & Koch (2007), Philipp and Koch (2009), and Declerck et al. (2015). This decision was driven by the intention to delve into the complex interplay between language dominance and code-switching behaviors within a rigorously controlled experimental setting. Our approach was anchored in the premise that language dominance, a pivotal aspect of bilingual language processing, could significantly influence the dynamics of code-switching. By treating language as a fixed factor, we aimed to systematically dissect the influence of language dominance across the participant cohort, thereby enabling a thorough examination that encapsulates the spectrum of individual language proficiencies and dominances.
In selecting the best-fitting models, we followed Barr et al.’s (2013) recommendations to balance model complexity with the explanatory power, considering both random intercepts and slopes. This approach allowed us to capture general differences and the specific impact of our factors of interest across participants and items. We began with a comprehensive model that considers all potential random intercepts and slopes. We then simplified the random structure progressively until convergence was achieved (see Bates et al., 2018), resulting in a linear mixed-effects model with random intercepts, and random slopes by subject for language type, language sequence, and stimulus–response configuration, as well as random intercepts by item. The generalized mixed-effects model similarly included random intercepts and random slopes by subject. The selection process was guided by theoretical expectations and the principle of treatment contrasts for our factors, ensuring that the comparisons made were meaningful and aligned with our research objectives. We validated our model choices through ANOVA, confirming that these models offered the best balance of complexity and fit for our data, capturing the nuanced effects of code-switching on reaction times and accuracy without unnecessary complexity.
Results
In this study, the dependent variables are accuracy and reaction time, where the latter refers to the time interval between stimulus presentation and semantic categorization. The accuracy analysis includes both correct and incorrect trials, while the reaction time analysis only includes correct response trials. Prior to commencing the reaction time analysis, a data exclusion protocol was implemented, which involved the exclusion of data from participants who exhibited response times below the 200-ms threshold and the elimination of trials that fell outside three SDs from the mean reaction time. The excluded trials accounted for 6.46% of all trials. Statistical analyses were conducted on participants’ semantic categorization accuracy and reaction time, with the average accuracy and reaction time shown in Figures 3 and 4, respectively.
Average accuracy rates for semantic categorization.
Average reaction time for semantic categorization.
Response accuracy of semantic categorization
The statistical analysis of the accuracy rate of participants’ semantic categorizations (see Table 1) revealed a significant main effect for language sequence (z = 2.565, p < .05), with a lower accuracy rate under the ABA condition compared with the CBA condition (Figure 3). Other effects, including stimulus–response configuration and language type, were not statistically significant (ps > .05).
Best-fitting model output for accuracy.
Signif. codes: 0 “***” 0.001 “**” 0.01 “*” 0.05 “.” 0.1 “ ” 1.
Formula:
Reaction time of semantic categorization
The statistical analysis of participants’ reaction times for semantic categorizations (see Table 2) showed a significant main effect for language sequence (t = –5.842, p < .001), with participants exhibiting slower reaction times in ABA sequences compared with the CBA sequences (Figure 4). The analysis also revealed a significant main effect for language type (t = 6.237, p < .001), with participants showing significantly slower reaction times for L3 compared with L1. Other effects, including stimulus–response configuration, were not statistically significant (ps > .05).
Best-fitting model output for reaction time.
Signif. codes: 0 “***” 0.001 “**” 0.01 “*” 0.05 “.” 0.1 “ ” 1.
Formula:
Discussion
This study investigated whether there were n–2 language repetition costs during the language comprehension process among Chinese–English–Japanese trilinguals and whether the costs were related to specific stimulus–response configurations.
Presence of n–2 language repetition costs
In the experiment, participants switched between tasks in three languages, and the n–2 language repetition costs were measured to examine the aftereffects of inhibitory processing. The statistical results showed that participants’ accuracy in ABA sequences is lower than their accuracy in CBA sequences, and they also exhibited slower reaction times in ABA sequences compared with the CBA sequences. This indicates that there are n–2 language repetition costs in code-switching among Chinese–English–Japanese trilinguals during language comprehension. This finding is consistent with that of previous studies on inhibitory processing in task switching between three languages and two languages (e.g., Costa & Santesteban, 2004; Jackson et al., 2001; Meuter & Allport, 1999; Philipp, Gade, & Koch 2007; Philipp & Koch, 2009), suggesting that inhibitory processing exists in code-switching processing during language comprehension among non-balanced trilinguals.
From the perspective of inhibitory processing, the n–2 language repetition costs in code-switching can be interpreted as a cognitive mechanism aiming to control the activation of multiple languages to ensure the use of the appropriate language in a specific context. This mechanism is particularly important in bilingual and multilingual speakers, as the multiple language systems in their brains are activated simultaneously. Therefore, the n–2 language repetition costs may be due to the additional inhibition required to overcome the previous inhibition of n–2 language when switching back to that language. Guo et al. (2013) found that multilinguals’ inhibition of non-target languages occurred at the lexical selection stage, not at the language task scheme stage. This implies that during semantic categorization tasks involving words within an ABA sequence, trilinguals may necessitate additional temporal and cognitive resources to suppress stimuli from the n–2 language. Consequently, this heightened demand is associated with diminished accuracy and prolonged reaction times.
The analysis on reaction times indicated a significant decrease in participants’ response times when engaging with L3 as opposed to L1. This observation may be intricately linked to the language proficiency and the frequency of language use of participants. Multilinguals may require more cognitive resources to inhibit their mother tongue when processing non-native languages, leading to increased response times. In addition, the cognitive processing mechanisms in trilinguals are not merely an extension of bilingualism (Schroeder & Marian, 2017). Szubko-Sitarek (2011) found that trilinguals process all three languages simultaneously during word recognition. This implies that when trilinguals engage in semantic categorizations of L3 words, they may require more time to simultaneously suppress semantic information related to their L1 and L2.
The impact of stimulus–response configuration on n–2 language repetition costs
The present study investigated inhibitory processing in language switching among Chinese–English–Japanese trilinguals and explored if n–2 language repetition costs are linked to specific stimulus–response configurations. We specifically aimed to identify the target of inhibition in n–2 trials: whether it is the language used or the stimulus–response configuration. Our research defined specific stimulus–response configurations as the pairings between a stimulus and its corresponding response in the semantic categorization task. Here, the stimulus refers to the word presented to participants, and the response involves pressing a key to indicate whether the word belongs to the category of animals. We used two sets of stimuli: animal-related words like “cat” and non-animal words such as “wind.” Participants were required to press the left “f” key for animal words and the right “j” key for non-animal words. Statistical analysis revealed that the stimulus–response configuration did not significantly influence the n–2 language repetition costs, suggesting these costs are not tied to specific stimulus–response configurations. This conclusion is consistent with Philipp and Koch’s (2009) findings on code-switching among German–English–French trilinguals. They observed that inhibitory processing affected switching between languages but not the repetition of specific stimulus–response configurations. Furthermore, their study noted lower n–2 repetition costs under conditions where the stimulus–response configuration was repeated, contradicting the hypothesis that inhibition should increase these costs.
In the code-switching process of trilinguals, the specific stimulus–response configuration employed in n–2 trials was not inhibited. This raises the question: What was inhibited? Philipp, Gade, and Koch (2007) were among the first to present evidence of language inhibitory processing in trilingual switching, a finding that has sparked considerable debate. Some researchers argue that inhibitory processing may target specific components of the psychological task representation rather than the language itself. Expanding on this, Philipp and Koch (2009) suggested that it might be the language cues that are inhibited, not the language per se. However, this suggestion was inconsistent with Finkbeiner et al.’s (2006) viewpoint regarding the switching costs between two languages. In Finkbeiner et al.’s (2006) study, participants were tasked with naming digits or pictures in their L1 or L2. They found that reaction times for naming pictures were consistent, regardless of the language used in previous digit-naming trials. From this, Finkbeiner et al. (2006) concluded that the switching costs were specific to the target stimulus–response configuration, which in their study, related to digits rather than pictures. However, Philipp and Koch (2009) contested this interpretation. They argued that the transition from naming digits to pictures constitutes a task switch in itself, thus the language used in consecutive trials may not be as influential as suggested. Supporting this, Kleinsorge and Heuer (1999) found that switching tasks could overshadow the effects of switching languages, leading Philipp and Koch to propose that Finkbeiner et al.’s (2006) conclusions might not fully capture the complexities of code-switching. This discussion illustrates the nuanced dynamics of task and language switching in cognitive processes, highlighting the need for further research to disentangle these interconnected factors.
The results of our study align with the viewpoint of Philipp and Koch (2009) that the key element in Finkbeiner et al.’s (2006) findings might be the transition between digit naming and picture naming, which led them to conclude that inhibition is linked to the specific stimulus–response configuration rather than the language itself. In their 2009 experiment, Philipp and Koch required participants to switch among six task settings: digits and colors across three languages (L1, L2, and L3). Similarly, our experiment involved switching among tasks using animal and non-animal words in three languages, suggesting that varying stimulus–response configurations correspond to different task settings. However, both experiments crucially involve task switches, including language switches. Our findings indicate that n–2 language repetition costs are unaffected by whether the task configuration between the n–2 and n trials remains consistent (e.g., Chinese animal, English non-animal, Chinese animal) or changes (e.g., Chinese animal, Japanese non-animal, Chinese non-animal). This implies that the interpretation of inhibitory aftereffects cannot solely rely on the stimulus–response configuration or task setting differences. Instead, our results suggest a broader level of inhibitory processing, targeting the psychological representation of competing languages, potentially focusing on specific aspects such as mental lexicons or individual language elements. The debate over whether different languages share a common mental lexicon or maintain separate lexicons has been ongoing (French & Jacquet, 2004). While this issue remains unresolved, Philipp and Koch (2009) propose that regardless of how language entries are organized in the brain, they are all susceptible to inhibition. This perspective underscores a more comprehensive approach to understanding the mechanisms of language inhibition in multilingual contexts.
The current study examined the language comprehension processes across phonetic and logographic writing systems, focusing on Chinese, English, and Japanese. The findings corroborate those of Philipp and Koch (2009), which explored transitions among phonetic languages (German, English, and French) during language production. Both studies suggest that while the specific stimulus–response configurations used in n–2 trials were not inhibited, the psychological representation of competing languages was affected. This indicates that the inherent characteristics of language scripts do not significantly influence n–2 language repetition costs, a conclusion supported by additional research. For instance, Costa et al. (2006) investigated the impact of language similarity on switching costs and determined that language similarity does not alter the lexical selection mechanisms in bilinguals, further supporting the limited impact of script characteristics on switching costs.
In conclusion, the findings reveal that n–2 language repetition costs are present during the language comprehension processes of Chinese–English–Japanese trilinguals engaged in code-switching. These costs are not determined by specific stimulus–response configurations but are linked to the psychological representation of competing languages. In addition, these costs may be influenced by factors such as the multilinguals’ language proficiency and frequency of language use, suggesting a complex interplay of cognitive and usage-based factors in language switching.
Conclusion
This study utilized a semantic categorization task to investigate the inhibitory processing of code-switching during language comprehension among Chinese–English–Japanese trilinguals and to determine if the n–2 language repetition costs are related to specific stimulus–response configurations. The findings are twofold: (1) There are n–2 language repetition costs in Chinese–English–Japanese trilinguals’ code-switching during language comprehension and (2) the n–2 language repetition costs in code-switching processing during language comprehension are not influenced by specific stimulus–response configurations but are related to the psychological representation of the competing languages.
The findings indicate that the n–2 language repetition costs reflect a more macro-level inhibitory processing. The research results validate, enrich, and extend previous experimental conclusions investigating code-switching between phonetic languages, providing new insights for research into code-switching and the n–2 language repetition costs in multilinguals. Our study has revealed the inhibitory processing involved in code-switching among Chinese–English–Japanese trilinguals and provided a better explanation for the characteristics of n–2 language repetition costs in the context of multilinguals’ switching between different languages. It deepens our understanding of the inhibitory processing mechanisms in the code-switching process, providing valuable insights for developing a universally applicable theory and theoretical models of code-switching inhibitory processing. In addition, it offers guidance for multilingual vocabulary teaching and learning. Subsequent research can further investigate the factors influencing the magnitude of n–2 language repetition costs in the code-switching process.
Footnotes
Acknowledgements
The authors extend sincere gratitude to the editors and reviewers whose insightful suggestions have significantly enriched this manuscript through each revision. Additionally, the authors would like to express appreciation to all participants involved in this research experiment. Their contributions were invaluable to the success of this study.
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) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Double First-Class Disciplines Project of Beijing Foreign Studies University (2022SYLZD008), and Fundamental Research Funds for the Central Universities (2024JX063).
