A bilingual’s two languages may always be active, even when only one is required (Thierry & Wu, 2007). This process might be expected to produce frequent language errors, but unintended cross-language intrusions are rare in spontaneous speech and the laboratory (Gollan et al., 2011). Bilinguals must, therefore, benefit from a control mechanism that allows them to select which language to use at a given moment and in a given context (Abutalebi et al., 2008).

There is growing evidence that the control processes are not fixed but differ depending on context (Jiang et al., 2023, Jiang et al., 2024). In line with the influential Adaptive Control Hypothesis (ACH, Green & Abutalebi, 2013), much research has found that the language control mechanism varies as a function of the social context of the communication (i.e., interactional context) (Blanco-Elorrieta and Pylkkänen, 2017, Rafeekh and Mishra, 2021). Some studies, for example, have revealed that stricter language control is used when switching in response to external cues (e.g., the monolingual interlocutor) than when switching language voluntarily (e.g., when surrounded by other bilinguals speaking the same languages) (de Bruin et al., 2020, Jevtović et al., 2020).

In this study, we tried to broaden the notion of context by examining whether and how bilinguals’ language control system differs depending on their emotional states. In life, bilingual language production often occurs in a state of heightened emotion (Dewaele and Costa, 2013, MacIntyre and Gardner, 1991a, Pavlenko, 2004). For example, language learners often experience negative affect (e.g., fear and anxiety) when they have to speak in a second language (L2) (Cohen & Norst, 1989). Bilinguals may feel stressed when forced to switch languages (Smith et al., 2020). However, there is a lack of direct and reliable empirical evidence regarding whether and how emotional states affect bilingual language control.

Furthermore, according to the ACH, the control system itself adapts to the demands placed on them by the interactional context. This study further explored how this adaptive control mechanism proposed by the ACH works in emotional contexts. One possibility is that it functions as initially proposed in the ACH. Here, we propose another possibility: the control system adaptively triggers compensatory control processes in the face of emotional disruptions according to the communicative/task goals in the interactional contexts. The two possibilities, in turn, would result in different patterns of the interaction between interactional contexts and emotional states in modulating control processes during bilingual language production. Thus, to adjudicate between the two possibilities, we investigated the interaction of the effects of emotional states with (cued versus voluntary) switching contexts. The findings would extend the predictions of ACH by incorporating within-individual variation in emotional states, and thus further unravel complex bilingual language control mechanisms in the real world where diverse types of contexts (e.g., social and personal contexts) appear to affect language processing within the same time frame (Hasson et al., 2018).

Substantial experimental evidence indicates that emotional states influence various cognitive processes, including cognitive control (Figueira et al., 2017). At the same time, it has been repeatedly shown that cognitive control is involved in bilingual language production (Jylkkä et al., 2018, Liu et al., 2016). Thus, bilingual language control can differ depending on emotional states. However, the available supportive evidence is mainly indirect and far from conclusive.

Evidence comes, firstly, from the negative association between foreign language anxiety and L2 speech production performance (Hewitt and Stephenson, 2012, MacIntyre and Gardner, 1989, MacIntyre and Gardner, 1991b, Wilson, 2006). Specifically, in L2 oral examinations, individuals with high foreign language anxiety produce more errors, including first language (L1) lexical intrusions, a form of language control failure (Hewitt and Stephenson, 2012, Wilson, 2006). Moreover, they have deficits in lexical access (i.e., generate fewer items) when tested in L2 verbal fluency tasks (MacIntyre and Gardner, 1989, MacIntyre and Gardner, 1991b), which is probably due to language control deficiency (i.e., high interference from the non-target language) (Bialystok et al., 2008, Ivanova et al., 2016, Sandoval et al., 2010). It should be noted, however, that mainly weak correlations between foreign language anxiety and L2 speech production performance were observed (|r| range = 0.34–0.40). Moreover, these findings do not necessarily require a control deficiency account to explain them. For example, the inferior performance in L2 verbal fluency tasks within anxious individuals can be explained in terms of vocabulary deficit resulting from the impairment of anxiety on vocabulary learning (MacIntyre & Gardner, 1994).

Further evidence supporting the effect of heightened emotional states derives from the findings that bilinguals are more likely to switch between languages when talking about past emotional events or expressing emotion (e.g., when swearing or reprimanding) (Buxbaum, 1949, Dewaele, 2010, Dewaele, 2015, Dewaele and Costa, 2013, Greenson, 1950, Krapf, 1955, Movahedi, 1996, Pavlenko, 2004, Pavlenko, 2005, Resnik, 2018, Rolland et al., 2017, Rozensky & Gomez, 1983, Santiago-Rivera et al., 2009). One explanation is that the heightened emotional state the individuals experienced when coding and expressing emotions interferes with strict inhibitory control of the non-target language, thus leading to unintended language switching (Dewaele, 2010). Evidence of the relation between language switching and heightened emotional states, nevertheless, is primarily restricted to clinical case studies of bilingual patients (Buxbaum, 1949, Greenson, 1950, Krapf, 1955, Movahedi, 1996, Rozensky & Gomez, 1983) and self-report studies (Dewaele, 2010, Dewaele, 2015, Pavlenko, 2004, Pavlenko, 2005, Resnik, 2018), leaving open both the generalizability and specificity of this link.

Williams et al. (2019) examined the associations between emotional states and code-switching frequency by observing 34 pairs of Chinese-American children and parents from bilingual immigrant families during a 5-minute emotion-inducing puzzle box task. The frequency of parents’ code-switching and the valence and intensity of their facial emotion behavior were coded at each 5-s interval. The results revealed that more intense negative facial emotion behavior was associated with increased code-switching at the subsequent 5-s epoch (i.e., parents switched more frequently after showing more negative facial emotion). The association between positive facial emotion and code-switching frequency at the subsequent 5-s epoch did not reach significance. However, more intense positive facial emotion predicted decreased concurrent code-switching frequency (i.e., parents switched less often when expressing more positive facial emotion). Williams et al. (2019) interpreted this finding by proposing that a negative emotional state temporarily disrupts cognitive control engaged in language control, thereby freely permitting entry of items from both languages into speech output and inducing more frequent code-switching. On the other hand, a positive emotional state facilitates cognitive control involved in language control, thus resulting in less frequent code-switching.

Williams et al. (2019) provided the first empirical evidence for the association between emotional states and language switching. However, using observational measures, they did not manipulate bilingual speakers’ emotional states and thus might not adequately reveal the causal connection between emotional states and language switches. Moreover, the switch rate may not be a reliable index for language control efficiency. Put concretely, the relationship between switch rate and language control efficiency may be modulated by the (unintended versus voluntary) types of switching. On the one hand, cross-language intrusions have been found to increase with declines in cognitive control (Gollan et al., 2011). Increases in voluntary language switching, on the other hand, have been reported to be linked to better efficiency in language control, though this link is not consistently observed (de Bruin et al., 2020, Gollan and Ferreira, 2009).

Taken together, the current study investigated the impacts of experimentally-induced emotional states on three prominent markers of top-down language control – namely, switching cost, mixing cost, and the reversed language dominance effect (as discussed below), in order to offer more compelling evidence for whether and how bilingual language control processes vary depending on emotional states.

In the following section, we begin by describing the ACH and different switching contexts. Next, we propose two possibilities for how the adaptive control mechanism operates in emotional contexts and how switching contexts and emotional states interact in modulating control processes.

In the ACH, Green and Abutalebi (2013) distinguish three interactional contexts (single language, dual language, and dense code-switching). In the single-language context, languages are used separately in distinct environments (e.g., one language at work and the other at home), and language switching rarely occurs. In the dual-language context, both languages are used in the same environment, but different languages are used with different addressees (e.g., two languages are used at work but with different monolingual colleagues with different language backgrounds), and bilinguals may switch languages in conversational turns with different addressees. The interference must be resolved in these two contexts to avoid cross-language intrusion errors. Consequently, a set of top-down control processes are implemented in the single-language context to ensure efficient suppression of the nontarget language over extended periods of time. In the dual-language context, a wider range of cognitive processes are triggered as a consequence of the increased demands for interference inhibition, language switching, and constant monitoring of the appropriate language. Finally, in the dense code-switching context, bilinguals share the same languages and may switch languages within a single conversational turn for no apparent external reasons. Bilinguals may use an opportunistic planning approach to use the words and constructions that are most readily available regardless of their language membership. This context is the least demanding, as it comes with limited needs for additional control processes.

The bilingual language control mechanism is most often tested by a cued language switching task, which requires bilinguals to name pictures or digits in response to a cue indicating which language to use (Costa and Santesteban, 2004, Meuter and Allport, 1999). Some recent studies (de Bruin et al., 2018, de Bruin and Xu, 2022, Zhu et al., 2022), however, investigated voluntary language switching where participants named pictures/digits in their language of choice. While the cued language switching is similar to a dual-language context, the voluntary language switching is comparable to a dense code-switching context (Blanco-Elorrieta and Pylkkänen, 2018, de Bruin et al., 2018, de Bruin et al., 2020, Jevtović et al., 2020). In line with the ACH, voluntary language switching has been observed to be at least partly driven by bottom-up processes related to lexical access in the case of opportunistic planning (Blanco-Elorrieta and Pylkkänen, 2017, Gollan et al., 2014, Gollan and Ferreira, 2009, Jevtović et al., 2020, Kleinman and Gollan, 2016, Zhu et al., 2022). Moreover, studies comparing different switching contexts have observed that more (top-down) language control is needed during cued than voluntary context (Blanco-Elorrieta and Pylkkänen, 2017, de Bruin et al., 2018, de Bruin et al., 2020, de Bruin and Xu, 2022, Gollan et al., 2014, Gollan and Ferreira, 2009, Jevtović et al., 2020, Zhu et al., 2022), which aligns with the proposal in the ACH that dual-language context imposes more demands on control processes than dense code-switching context (de Bruin & Xu, 2022). Furthermore, it has been observed that relative to having to stay in one language (most comparable to the single-language context in the ACH), forced mixing is more costly while freely mixing two languages is less effortful (de Bruin et al., 2018, de Bruin and Xu, 2022). This line of evidence is in accord with the proposal in the ACH that single-language context involves lower levels of control than dual-language context, but triggers higher levels of control than dense code-switching context.

It should be noted that the studies discussed above mainly tested habitual code-switchers (Blanco-Elorrieta and Pylkkänen, 2017, de Bruin et al., 2018, de Bruin et al., 2020, de Bruin and Xu, 2022, Jevtović et al., 2020) who may find it relatively cognitive effortless to process the code-switches that are congruent with their usual mode of language use (Green & Abutalebi, 2013). It remains unclear whether a similar pattern of results could emerge for non-habitual code-switchers, for whom switching may be unnatural and effortful, and voluntary switching may require some higher order decision regarding what language to use and when to switch. However, several studies (Jiao et al., 2022, Liu et al., 2021) testing Chinese-English bilinguals from a non-habitual codeswitching community (i.e., universities in Mainland China) have reported (some) benefits in voluntary over cued language switching, thus providing preliminary evidence for the ACH in the case of non-habitual code-switchers (though for this bilingual sample, voluntary switching might engage somewhat different mechanisms, such as executive decision, than those involved in dense code-switching).

Given the differences between cued and voluntary contexts discussed above, two possibilities can be raised for how the adaptive control mechanism operates in emotional contexts, which would be reflected in the interaction between switching contexts and emotional states. An intuitive possibility, the cognitive effort account, is that the control system functions as initially proposed in the ACH (Green & Abutalebi, 2013); thus, voluntary language switching could be less cognitively demanding than cued language switching (de Bruin et al., 2018, Gollan et al., 2014, Jevtović et al., 2020). In this case, emotional effects should be less robust in voluntary than cued switching because the impact of emotional states has been widely observed to be greatest with relatively difficult and demanding tasks (Egidi and Gerrig, 2009, Forgas, 1995, Seibert & Ellis, 1991).

Here, we propose an alternative and more intriguing possibility, namely, the adaptive compensatory control account: the control system may selectively compensate for temporary language control failures when strict language control is required (e.g., during cued switching) but not when less strict control is preferred (e.g., during voluntary switching). On this view, cued switching would be more resistant to the (detrimental) emotional effects than voluntary switching. Specifically, prior research indicates that deviations from required performance (e.g., errors or delayed responses) trigger compensatory adjustments in control processes, which bring behavior more in line with task goals (Botvinick et al., 2004, Green and Abutalebi, 2013). For instance, in the flanker task, the difference between conflicting and non-conflicting trials is reduced after a conflicting trial (Botvinick et al., 2004). Hence, additional control processes may be triggered to compensate for the detrimental effects of (negative) emotional states (if any) when top-down language control failures, such as cross-language intrusions, cause deviation from the task goal.

As discussed above, interactional contexts have been proposed to differ in terms of communicative/task goal (Green & Abutalebi, 2013). Specifically, bilinguals in single- and dual-language contexts establish and maintain task goals such as speaking in one language rather than another and avoiding cross-language intrusions. On the contrary, bilinguals in a dense code-switching context aim to use both languages opportunistically, circumventing the need to strongly suppress non-target languages (Green & Abutalebi, 2013). Thus, increases in top-down language control failures under (negative) emotional states would deviate from the task goal in single- and dual-language contexts. In contrast, given the lack of negative consequences of selecting the “wrong” language (e.g., cross-language intrusions can be used opportunistically) in the dense code-switching context (Green & Abutalebi, 2013), top-down language control failures may not deviate from the task goal in this context. Consequently, compensatory adjustments in control processes would be triggered in single- and dual-language contexts but not in dense code-switching contexts. Accordingly, the detrimental effects of (negative) emotional states, if any, should be more robust in voluntary switching (comparable to a dense code-switching context) than cued switching (comparable to a dual-language context).

The present study aims to examine bilingual language control in emotional contexts. We focus on whether and how emotional states can modulate bilingual language control and how emotional states and switching contexts interact.

The experiment featured a mixed design with task group (voluntary vs. cued) as a between-group factor and emotion (neutral vs. negative vs. positive), trial type (switch trials in mixed-language conditions vs. nonswitch trials in mixed-language conditions vs. blocked trials), and language (L1 vs. L2 trials) as within-group factors. One group of Chinese-English bilinguals from a non-habitual codeswitching community completed a voluntary task where they named pictures in their language of choice, while another matched group performed a cued task where pictures had to be named in Chinese or English in response to a cue. The two tasks were matched for the average switch rate, which has been reported to affect control processes (Jylkkä et al., 2018) and thus may interfere with the influence of switching type. All participants were tested under negative, positive, and neutral states generated by a standard emotion-induction procedure involving music and guided rumination (Chepenik et al., 2007, Guo et al., 2020, Jefferies et al., 2008, Rowe et al., 2007, Spachtholz et al., 2014, van Steenbergen et al., 2010).

Three prominent markers of language control, namely, switching cost, mixing cost, and the reversed language dominance effect (RLDE), were used. The switching cost refers to poorer performance (e.g., slower response speed and reduced naming accuracy) on switch trials (a response in a different language than in the previous trial) than nonswitch trials (a response in the same language as on the previous trial) (Green, 1998). The mixing cost refers to worse performance on non-switch trials in mixed language conditions than trials in blocked single-language conditions in which bilinguals must use one pre-specified language (Ma et al., 2016). Notably, switching into the dominant L1 often incurs a greater switching cost than switching into the nondominant L2 (Costa and Santesteban, 2004, Meuter and Allport, 1999), and the dominant L1 often incurs larger mixing costs than the nondominant L2 (Christoffels et al., 2007, Peeters and Dijkstra, 2018) (especially for unbalanced bilinguals). Typically, when both switching and mixing costs are measured, the asymmetry across languages is present in only one of them (Declerck, 2020). The RLDE refers to slower responses in the dominant L1 than in the nondominant L2 in the mixed language condition (Christoffels et al., 2007, Costa and Santesteban, 2004).

As for the effect of emotional states, based on previous research (Dewaele, 2010, Williams et al., 2019), we hypothesized that negative states should disrupt language control, while positive states should boost the control system. Increased control efficiency would manifest as overall faster responses (Wu & Struys, 2021), smaller switching cost (de Bruin et al., 2018, de Bruin et al., 2020, Gollan and Ferreira, 2009, Weissberger et al., 2012), smaller switching cost asymmetry (Liu et al., 2016), smaller mixing cost (de Bruin et al., 2020, Weissberger et al., 2012) and larger RLDE (see Stasenko et al., 2021 for discussion of aging-related deficit in dominance reversal). Thus, we expected overall slower responses, a larger switching cost, switching cost asymmetry, mixing cost, and smaller RLDE under a negative than neutral state. By contrast, there should be overall faster responses, a smaller switching cost, switching cost asymmetry, mixing cost, and larger RLDE under a positive than neutral state. Whereas the switching cost and its asymmetry represent a reactive type of language control, the RLDE and the mixing cost have been associated with a more proactive type of control (for reviews of measures of reactive and proactive language control, see Bobb and Wodniecka, 2013, Declerck, 2020). Reactive language control resolves cross-language interference after it is detected; however, proactive language control anticipates and prevents potential interference before it occurs (Declerck, 2020). Reactive control is implemented at the local, trial-by-trial level, but proactive control is implemented at the global, non-trial-specific level (Ma et al., 2016). In addition, we used the switch rate as a supplementary index for language control in voluntary tasks. Following previous research (Williams et al., 2019), we expected more frequent switching under the negative than the neutral state but less frequent switching under the positive than the neutral state.

Two possibilities for how the adaptive control system operates in emotional contexts are considered: (1) it operates as initially proposed in the ACH (Cognitive effort account), or (2) it adaptively triggers the compensatory mobilization of top-down control according to the communicative goals in the interactional contexts (Adaptive compensatory control account). Concerning the interplay between switching contexts and emotional states, two different patterns can be expected based on the two accounts. According to the cognitive effort account, voluntary switching should reveal a less robust emotional effect than cued switching, regardless of whether the emotional effect is detrimental or facilitative. The adaptive compensatory control and the cognitive effort accounts contradict their predictions regarding whether the cued or voluntary switching context exhibits greater emotional disruption. Specifically, according to the adaptive compensatory control account, when the emotional effect is detrimental, a compensatory mobilization of top-down control should be triggered in cued switching, where a loss of strict control deviates from the task goal. However, it should not be triggered in voluntary switching, where a loss of strict control does not deviate from the goal-related requirement. Accordingly, relative to the voluntary switching context, cued switching should be more resistant to the detrimental effect of, for example, negative states.

Note that the present study is based on the assumption that voluntary switching is less costly than cued switching, even for non-habitual code-switchers. Accordingly, we expected to find faster responses, smaller switching costs, and mixing costs in voluntary than cued switching context (de Bruin et al., 2018, de Bruin et al., 2020, Jevtović et al., 2020, Jiao et al., 2022) at least in neutral state. However, given that the non-habitual code-switchers may find the voluntary switching unnatural and effortful, we expected smaller benefits in voluntary over cued language switching relative to those previously observed in habitual code-switchers (e.g., Jevtović et al., 2020).