1. Introduction
The relationship between bilingualism and cognitive control has been the focus of an increasing number of studies over the past decades, particularly since the seminal article by Peal and Lambert (Reference Peal and Lambert1962), who reported that a sample of 10-year-old bilingual children living in Canada exhibited advantages in both verbal and nonverbal intelligence tests compared with their monolingual peers. According to Peal and Lambert (Reference Peal and Lambert1962), bilingual children showed superior performance in mental flexibility, concept formation and a broader range of cognitive abilities, and these advantages were likely related to their bilingual status. Bilingualism, defined as the use of a more dominant language (L1) and a less dominant one (L2), is thought to involve cognitive control mechanisms that enable speakers to manage both languages. The L1 and L2 are automatically and simultaneously activated to some degree, even when the bilingual individual is immersed in a context that requires the use of only one language (e.g., Hernandez et al., Reference Hernandez, Li and MacWhinney2005; Hoshino & Thierry, Reference Hoshino and Thierry2011; Jacobs et al., Reference Jacobs, Fricke and Kroll2016). This constant coactivation appears to occur at different linguistic levels (e.g., Costa, Reference Costa, Kroll and de Groot2005; Spivey & Marian, Reference Spivey and Marian1999; Van Heuven et al., Reference Van Heuven, Dijkstra and Grainger1998). At the lexical level, a mental representation (e.g., the image of a dog) is assumed to activate at least two lexical items (dog and perro) that compete for selection (e.g., Abutalebi & Green, Reference Abutalebi and Green2008; Kroll et al., Reference Kroll, Bobb, Misra and Guo2008). To select the intended language (e.g., English) while controlling interference from the non-target language (e.g., Spanish), bilinguals are hypothesized to recruit domain-general rather than language-specific cognitive control mechanisms (e.g., Abutalebi & Green, Reference Abutalebi and Green2008; Bialystok et al., Reference Bialystok, Craik and Luk2008; Kroll et al., Reference Kroll, Dussias, Bogulski, Valdes Kroff and Ross2012). These findings suggest that the experience of managing two or more languages can yield cognitive benefits; specifically, language selection, context monitoring, inhibition of the non-target language and switching between languages appear to enhance performance on nonverbal cognitive tasks.
Cognitive control is central to executive functions, which encompass a set of domain-general mechanisms that guide human cognition and behavior, particularly self-control and self-regulation (Braver, Reference Braver2012; Miyake & Friedman, Reference Miyake and Friedman2012). Behavioral and neurocognitive studies have examined the relationship between cognitive control and bilingualism across various linguistic levels, using diverse methodological approaches and age groups. Some studies have reported positive effects (e.g., Grundy & Timmer, Reference Grundy and Timmer2017; see also van den Noort et al., Reference Van den Noort, Vermeire, Bosch and Hugdahl2019, for a review), whereas others have found null results (e.g., Donnelly et al., Reference Donnelly, Brooks and Homer2019; Lehtonen et al., Reference Lehtonen, Soveri, Laine, Järvenpää, de Bruin and Antfolk2018; Nichols et al., Reference Nichols, Wild, Stojanoski, Battista and Owen2020). Researchers have also investigated cognitive control in adult L2 learners at the lexical (Bartolotti et al., Reference Bartolotti, Marian, Schroeder and Shook2011; Grant et al., Reference Grant, Fang and Li2015; Linck & Weiss, Reference Linck and Weiss2015), grammatical (Kappa & Colombo, Reference Kappa and Colombo2014) and phonological levels (e.g., Darcy et al., Reference Darcy, Mora and Daidone2016), with some studies finding no relationship between cognitive control and L2 learning (e.g., Linck & Weiss, Reference Linck and Weiss2015; Stone & Pili-Moss, Reference Stone and Pili-Moss2016).
A number of theoretical models have been proposed to explain how bilinguals engage cognitive control processes to select, monitor and inhibit languages as needed. The Inhibitory Control Model (ICM; Green, Reference Green1998) posits that inhibition is the primary mechanism responsible for resolving competition between lexical candidates from both languages. When a speaker intends to use one language, the non-target language is temporarily inhibited to prevent interference. The magnitude of inhibition required to suppress the unwanted lexical item depends on the level of activation of that language; thus, greater inhibition is needed to suppress the dominant language than the non-dominant one. Furthermore, this inhibition is reactive; that is, the non-target lexical candidate is inhibited only after it has been activated. Another inhibitory model, the Bilingual Interactive Activation model (BIA; Dijkstra & Van Heuven, Reference Dijkstra, Van Heuven, Grainger and Jacobs1998; Grainger & Dijkstra, Reference Grainger, Dijkstra and Harris1992), proposes that as word representations in one language become activated, this activation spreads to a language node that, in turn, inhibits word representations of the non-target language. In a later version, the BIA+ model (Dijkstra & Van Heuven, Reference Dijkstra and Van Heuven2002), inhibitory control is managed by a task/decision system that relies on top-down control mechanisms. These mechanisms integrate the speaker’s goals with contextual factors (e.g., lexical or syntactic cues) that guide word selection.
Both the ICM and BIA+ models explain bilinguals’ more efficient performance, when compared to monolinguals with comparable characteristics, in terms of reactive control processes, which are responsible for resolving competition between the target and non-target languages. Reactive control, which involves detecting and resolving interference after it occurs (Braver, Reference Braver2012), offers the advantage of minimizing working memory demands because it operates temporarily. However, it also has the disadvantage that task goals must be reactivated when needed, and the mechanisms responsible for resolving conflict must signal when inhibitory control is required.
Recent studies have argued that inhibitory control perspectives alone do not fully explain how bilinguals manage their two languages and that monitoring and goal maintenance may better account for their more efficient performance in attentional tasks (Colzato et al., Reference Colzato, Bajo, van den Wildenberg, Paolieri, Nieuwenhuis, La Heij and Hommel2008; Costa et al., Reference Costa, Hernández, Costa-Faidella and Sebastián-Gallés2009). Thus, single-process accounts appear insufficient to explain the potential impact of bilingualism on cognitive performance (Morales et al., Reference Morales, Calvo and Bialystok2013). In line with these findings, Green and Abutalebi (Reference Green and Abutalebi2013) proposed a revised version of the Inhibitory Control Model (Green, Reference Green1998), known as the Adaptive Control Hypothesis (ACH). This framework incorporates proactive control, defined as the speaker’s ability to maintain task goals and select the appropriate response before competition arises (Braver, Reference Braver2012). According to the ACH, proactive and reactive control mechanisms enable bilinguals to adapt to the changing demands of different interactional contexts. The model distinguishes three such contexts: single-language, dual-language and dense code-switching. In single-language contexts, bilinguals use each language in distinct situations and with specific interlocutors. In dual-language contexts, both languages coexist within the same environment but are typically used with different partners. In dense code-switching contexts, speakers frequently alternate between languages, sometimes within a single conversation or even a single sentence. Consequently, whether linguistic environments promote separate or integrated use of the L1 and L2 may differentially affect the cognitive demands placed on executive functions.
Bilinguals’ efficient performance on tasks involving conflict resolution may be better understood through the dynamic interaction between monitoring/proactive and inhibitory/reactive control mechanisms (Bialystok et al., Reference Bialystok, Craik and Luk2012; Costa et al., Reference Costa, Hernández, Costa-Faidella and Sebastián-Gallés2009; Morales et al., Reference Morales, Calvo and Bialystok2013, Reference Morales, Yudes, Gomez-Ariza and Bajo2015). Reactive control through inhibition accounts for smaller differences in reaction times between congruent and incongruent trials, also known as the interference effect, which would reflect bilinguals’ ability to ignore or suppress irrelevant information (e.g., in Stroop, Simon and Flanker tasks). Proactive control through monitoring and goal maintenance may explain bilinguals’ more efficient performance when tasks require adjustment to both congruent and incongruent trials within a mixed block. Proactive strategies also account for bilinguals’ differential performance in tasks involving sustained goal maintenance, such as the Attention Network Test (ANT; e.g., Costa et al., Reference Costa, Hernández and Sebastián-Gallés2008), task switching (e.g., Prior & Gollan, Reference Prior and Gollan2011) and visual search (e.g., Hernandez et al., Reference Hernandez, Costa and Humphreys2012). In such cases, when task success depends on encoding relevant information to anticipate responses, participants rely on proactive control. Proactive control allows behavior to be adapted in advance to achieve a goal; however, it is highly demanding of working memory resources, as the goal must remain active throughout task execution (Morales et al., Reference Morales, Calvo and Bialystok2013). In summary, successful language selection and inhibition of the non-target language depend on a combination of both proactive and reactive control strategies (Bialystok et al., Reference Bialystok, Craik and Luk2012; Costa et al., Reference Costa, Hernández, Costa-Faidella and Sebastián-Gallés2009; Luque & Morgan-Short, Reference Luque and Morgan-Short2021). Importantly, traditional cognitive tasks used to compare monolinguals and bilinguals, such as the Stroop (Stroop, Reference Stroop1935), Simon (Simon, Reference Simon1969) and Flanker (Eriksen & Eriksen, Reference Eriksen and Eriksen1974) tasks, primarily assess reactive control. Therefore, the present study employed a more fine-grained measure encompassing both proactive and reactive control: the Automated Continuous Performance Task (CPT; Rosvold et al., Reference Rosvold, Mirsky, Sarason, Bransome and Beck1956). The extent to which these control mechanisms are engaged, however, appears to depend on both individual and contextual characteristics of the bilingual experience.
Studies on the relationship between bilingualism and cognition increasingly emphasize the heterogeneity of bilingual populations and the importance of combining relatively static variables, such as L2 age of acquisition (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018; Kousaie et al., Reference Kousaie, Chai, Sander and Klein2017; Tao et al., Reference Tao, Marzecová, Taft, Asanowicz and Wodniecka2011), with more dynamic ones, such as L2 proficiency (Luque & Morgan-Short, Reference Luque and Morgan-Short2021) and the linguistic characteristics of the contexts in which bilinguals use and are exposed to two or more languages. These contextual characteristics include the extent to which the L1 and L2 are used separately or in an integrated manner, as well as the tendency to code-switch (Beatty-Martinez et al., Reference Beatty-Martinez, Navarro-Torres, Dussias and Kroll2019; Hartanto & Yang, Reference Hartanto and Yang2016; Hofweber et al., Reference Hofweber, Marinis and Treffers-Daller2016, Reference Hofweber, Marinis and Treffers-Daller2020). Moreover, the compartmentalized or integrated use of the L1 and L2 across different contexts, such as home, work or social environments, has recently been quantified through a continuous measure of linguistic diversity known as entropy. Highly compartmentalized contexts are typically more predictable and therefore exhibit low entropy, whereas linguistically integrated contexts are less predictable and display high entropy (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018).
Recent studies have examined the modulating roles of L2 age of acquisition, L2 proficiency and language context in the recruitment of proactive and reactive control, often using the AX-CPT task. Gullifer et al. (Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018) investigated the effects of L2 age of acquisition and social linguistic diversity on the engagement of proactive and reactive control. They found that an earlier L2 age of acquisition was associated with lower reliance on proactive control, whereas greater linguistic diversity across social contexts predicted increased dependence on proactive control. Luque and Morgan-Short (Reference Luque and Morgan-Short2021) explored the relationship between cognitive control and L2 proficiency in adult L2 learners using both the Flanker and AX-CPT tasks. Their results indicated that the role of cognitive control shifts as learners become more proficient: among early and intermediate learners, reactive control played a more prominent role, whereas proactive control became more distinct as L2 proficiency and language experience increased (see also Grant et al., Reference Grant, Fang and Li2015; Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018).
Finally, Beatty-Martinez et al. (Reference Beatty-Martinez, Navarro-Torres, Dussias and Kroll2019) compared three groups of highly proficient Spanish–English bilinguals who lived in contexts differing in L1 and L2 use and exposure. They found that, regardless of proficiency, bilinguals exhibited distinct patterns of language use shaped by the specific characteristics of their interactional contexts.
Building on previous findings suggesting that bilingual language experience modulates cognitive control engagement (e.g., Bice & Kroll, Reference Bice and Kroll2015; Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018; Morales et al., Reference Morales, Calvo and Bialystok2013; Zirnstein et al., Reference Zirnstein, van Hell and Kroll2018), the present study sought to further examine how contextual patterns of language use shape the recruitment of proactive and reactive control in bilingual speakers. Specifically, we focused on Mexican bilingual young adults who differ in the degree to which their two languages – Spanish and English – are used in separate versus integrated linguistic contexts.
2. Current study
The general aim of the present study was to examine whether the separate or integrated use of Spanish and English among Mexican bilingual young adults influences their recruitment of cognitive control mechanisms – specifically, proactive and reactive control (Braver, Reference Braver2012). Participants were immersed in one of two linguistic contexts: a separate context, in which Spanish and English are typically used in distinct settings and/or with different conversational partners; or an integrated context, in which both languages are used more fluidly, often within the same situations and with the same interlocutors. The sample consisted of bilingual young adults, most of whom had acquired English after the age of five (late bilinguals), with proficiency levels ranging from low to high intermediate. All participants were immersed in their L1, Spanish.
The diverse linguistic experiences that characterize the contexts in which bilingual speakers are immersed may differentially affect their engagement of executive control mechanisms when performing nonverbal cognitive tasks. To investigate this, the present study employed an adaptation of the AX Continuous Performance Task (AX-CPT) used by Gullifer et al. (Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018) to examine the use of proactive and reactive control processes (Braver, Reference Braver2012). In this task, participants respond ‘yes’ when detecting an AX sequence and ‘no’ to all other combinations (AY, BX, BY). AX trials occur 70% of the time, and each of the other trial types occurs 10% of the time. The dependent measures are error rates and reaction times across the four conditions (AX, AY, BX, BY). Proactive control (monitoring) is indexed by a higher percentage of AY errors compared with BX and BY errors, with little difference between the latter two. Reactive control (inhibition) is reflected in similar error rates and reaction times for AY and BY trials, while performance on BX trials depends on successful inhibitory processes. The AX-CPT has been effectively used in previous studies examining the relationship between bilingualism and the reliance on proactive and reactive control (Bice & Kroll, Reference Bice and Kroll2015; Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018; Morales et al., Reference Morales, Calvo and Bialystok2013; Zirnstein et al., Reference Zirnstein, van Hell and Kroll2018). Prior studies comparing bilingual groups differing in language use and exposure have yielded mixed results, particularly among young adults. The present study therefore aimed to further investigate cognitive control recruitment among bilinguals living in Mexico, who are immersed in contexts that differ subtly in the use and exposure to their L1 (Spanish) and L2 (English).
This study pursued two main objectives: 1) To assess the recruitment of proactive (monitoring) and reactive (inhibitory) control mechanisms in bilingual young adults who use and are exposed to their L1 and L2 in either a separate or an integrated manner; 2) To examine the modulating roles of L2 age of acquisition and L2 proficiency in the recruitment of proactive and reactive control across the two contexts.
We hypothesized that participants immersed in the separate context would rely more on proactive control strategies due to their ability to anticipate which language to use in a given situation or with specific interlocutors. In contrast, bilinguals from the integrated context were expected to exhibit a more balanced use of reactive and proactive control strategies, reflecting the coexistence of their L1 and L2 in shared settings. We further predicted that linguistic diversity within each context would correspond to different patterns of proactive and reactive control recruitment (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018).
Regarding L2 proficiency, we expected differences based on proficiency level. Luque and Morgan-Short (Reference Luque and Morgan-Short2021) found that beginning and intermediate adult L2 learners rely more on reactive control, whereas more proficient learners tend to depend more on proactive control. Given that our participants’ proficiency ranged from pre-intermediate to upper intermediate, we expected a predominance of reactive control strategies but also anticipated an interaction between L2 proficiency and linguistic diversity, such that participants in the integrated context would show a more balanced engagement of both control types.
Concerning L2 age of acquisition, Gullifer et al. (Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018) found that an earlier age of acquisition is associated with lower reliance on proactive control. Their sample comprised highly proficient French–English bilinguals (L2 age of acquisition range: 0–13 years, M = 7.5). In our sample, L2 age of acquisition varied widely (range: 1–23 years, M = 7.59), with proficiency ranging from low to high intermediate.
To address our research questions, participants completed an AX variant of the Continuous Performance Task (modified from Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018) to assess proactive and reactive control recruitment. The variables of interest were accuracy and reaction times in each of the four trial types (AX, AY, BX and BY).
3. Method
The procedure described in this study was carried out in accordance with the Norma Oficial Mexicana [Official Mexican Standards], and both experiments were approved by the Comité de Ética de la Facultad de Psicología de la Universidad Nacional Autónoma de México [Ethics Committee of the School of Psychology of the National Autonomous University of Mexico], approval number EP/PMDPSIC/0219/2021.
3.1. Participants
Fifty participants between the ages of 19 and 34 (M = 24.6, SD = 4.51), with a mean age of English acquisition of 7.59 (SD = 4.48), were recruited through non-random sampling by convenience, by sharing posts through social media. They were placed in one of two groups: separate context or integrated context. Twenty-five participants had been born and lived in Mexico City and surrounding areas (19 in Mexico City, 4 in Estado de Mexico and 2 in Querétaro). They reported the use of Spanish and English in different situations and with different interlocutors and, therefore, were included in the separate group. The other twenty-five participants had been born and lived in the north of Mexico, close to the border with the United States (10 in Baja California, 5 in Chihuahua, 2 in Coahuila, 2 in Sonora and 6 in Nuevo León). They reported the use of Spanish and English in many of the same contexts and with the same interlocutors and were included in the integrated group. Contexts of relative use and exposure to Spanish and English encompassed family, friends, radio, TV and self-instruction. Data regarding participants’ birthplace and residence and their use of both languages across different contexts were collected through the Language Experience and Proficiency Questionnaire (Marian et al., Reference Marian, Blumenfeld and Kaushanskaya2007).
When participants reported knowledge of a third language (L3), they were admitted only if they did not exceed the pre-intermediate level in that language and /or if it was a language that they did not practice at the time of recruitment, to confirm that English was the participants’ L2. The list of L3 included: French (13), German (8), Japanese (3), Italian (7), Chinese (2) and Portuguese (1). To be accepted into the study, participants had to report normal hearing, normal or corrected-to-normal vision, no history of neurological disorders and a minimum proficiency of low intermediate in both English and Spanish. Lastly, participants were recruited only if they confirmed that they alternated between Spanish and English (code-switching). Although code-switching direction and frequency were not the focus of the present study, participants also completed the Bilingual Switching Questionnaire (BSWQ; Rodriguez-Fornells et al., Reference Rodriguez-Fornells, Krämer, Lorenzo-Seva, Festman and Münte2012) as part of a larger study, and the data and analyses related to that instrument can be found in Raisman-Carlovich et al. (Reference Raisman-Carlovich, Carrasco-Ortiz and Arias-Trejo2024). Bilinguals who completed the study were either invited to participate in a raffle of Amazon gift cards of MNX $250.00 each (USD $13.55 approximately) or were paid MNX $150.00 (USD $8.15 approximately). Participants were sent an email with a link to a digital Google format that started with an informed consent. Once they checked the box confirming that they accepted to be part of the study, they were redirected to the questionnaire and tests described in the following paragraphs.
Participants were first asked to complete an adaptation of the Language Experience and Proficiency Questionnaire (LEAP-Q; Marian et al., Reference Marian, Blumenfeld and Kaushanskaya2007) in Spanish that includes information on participants’ age, gender, L2 age of acquisition, self-rated measures of language proficiency, relative use of and exposure to Spanish and English, and immersion experiences in a country where English was the dominant language, among others. Then they were administered two standardized vocabulary tests, the LexTALE-Lexical test for advanced learners of English (Lemhöfer & Broersma, Reference Lemhöfer and Broersma2012) and the Lextale-Esp-Lexical test for advanced learners of Spanish (Izura et al., Reference Izura, Pérez, Agudelo, Ferré, García-Orza, Lapuente and Sánchez-Casas2014). These are non-timed, visual lexical decision tasks in which bilinguals are presented with words/pseudowords and must respond with ‘yes’ or ‘no’ depending on whether the word exists or not in the language being assessed (e.g., Spanish, English). The purpose of administering these tests was to confirm participants’ proficiency levels in Spanish and English. Variables of interest collected from the LEAP-Q, the LexTALE in English and the Lextale-Esp are shown in Table 1.
Table 1. Mean and standard deviations for participants’ age, years of formal education, English age of acquisition, percentage of exposure to Spanish and English, length of immersion experiences in a country where English was spoken and proficiency scores in Spanish and English, by context of language use
Note. Values in parentheses represent standard deviations. Asterisks indicate significant between-group differences at p < .05. N = 50 (25 per context). Years of formal education are counted from the first year of elementary school. (*) Immersion in a country where English is spoken was statistically significant, W = 422, p = .020 (Mann–Whitney–Wilcoxon nonparametric test). Proficiency was computed using scores from the LexTALE-Esp (Spanish) and LexTALE (English) tests. All other data were self-reported.
Bilinguals from the separate context were overall younger when they started to acquire English, but the statistical difference between groups was non-significant (t(48) = 1.089, p = .281), and relative exposure to Spanish and English was very similar for both contexts. Bilinguals from the integrated context reported having spent on average significantly longer periods of time in countries where English was the dominant language, but data were highly variable, with some of them reporting experiences of up to a year and others of just a few days or no immersion experience at all. Except for the length of immersion experiences in English-speaking countries, no significant between-group differences were found.
3.2. Experimental procedure
Participants who completed the questionnaire and the two vocabulary tests described in the Participants section received an email with a link to the behavioral experimental platform Gorilla Experiment Builder (Anwyl-Irvine et al., Reference Anwyl-Irvine, Massonnié, Flitton, Kirkham and Evershed2019). Once they consented to participate, they were given access to the AX-CPT task.
The AX-CPT task was administered to evaluate participants’ specific use of proactive and reactive control strategies and to determine whether a more separate or more integrated use of L1-Spanish and L2-English would impact participants’ cognitive performance differently. In the task, participants view a continuous string of letters (see Figure 1). The instructions are to respond ‘yes’ only when they see the probe letter X preceded by the cue letter A. All other combinations of cues and probes must be given a ‘no’ response. The task includes 100 trials and, critically, the AX condition, or target sequence, appears in 70% of the trials, thus creating an impulse to answer ‘yes’ whenever participants see an A or an X. The other three sequences _AY, BX and BY_ appear in 10% of the trials, accounting for 30% of the total number of trials (see Figure 1). The AY and BX conditions are taken as indicators of bilinguals’ relative reliance on proactive and reactive control strategies. On one hand, when an AY sequence appears, the participant sees the A-cue first and then the non-X probe. A high reliance on proactive control would prompt the participant to give a ‘yes’ response upon seeing the A, although the probe is not an X. In the BX condition, on the other hand, participants see a non-A cue, and then the X probe. Therefore, they need to use proactive control (monitoring and contextual cue processing) to give a ‘no’ answer, although the sequence has an X probe. Finally, the BY sequence is the baseline sequence and, according to the literature, an index of processing speed (Luque & Morgan-Short, Reference Luque and Morgan-Short2021). In sum, proactive control assessment is based on participants’ performance on BX trials and reactive control, on their performance on AY trials, and performance is measured through indexes of accuracy and reaction times in the different sequences.
Figure 1. AX variant of the Continuous Performance task. Note. Representation of the AX variant of the continuous performance task (adapted from Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018). Each sequence starts with a cue in black and a probe in red. There are four possible types of trials: AX, AY, BX and BY. Letters Y and B are used as placeholders for any non-A cue and non-X probe (e.g., N, F, R, G). Each trial lasts 3000 ms with a 1000-ms interval between trials.
Before the experimental block, there is a practice block that comprises 10 trials that include the four sequences of interest. The task takes approximately 15 minutes to complete.
3.3. Analyses and results
We computed indexes of accuracy and reaction times for each of the four conditions: AX, AY, BX and BY. The AX condition was not critical for our purposes, but was included in the descriptive analyses to confirm that there were no statistically significant differences between contexts. Responses under 100 ms and over 1000 ms were removed from the analyses (Luque & Morgan-Short, Reference Luque and Morgan-Short2021; Morales et al., Reference Morales, Calvo and Bialystok2013) resulting in the elimination of 0.03% of the data. Table 2 shows indexes of correct answers across the four conditions (AX, AY, BX and BY) for both groups.
Table 2. Mean and standard deviations for indexes of accuracy in the four conditions of the AX-CPT task, per context
Note. Values in parentheses represent standard deviations. Indexes of accuracy for each type of trial _AX, AY, BX and BY_ per context.
Overall accuracy rates were high across conditions (Moverall = 96.6%; M AX = 98.7%; M AY = 91.3%; M BX = 97.7%; M BY = 98.8%). Participants in both groups made fewer errors in the AX and BY conditions than in the AY and BX conditions. These results are consistent with previous findings (e.g., Fricke et al., Reference Fricke, Zirnstein, Navarro-Torres and Kroll2018; Luque & Morgan-Short, Reference Luque and Morgan-Short2021; Morales et al., Reference Morales, Calvo and Bialystok2013). Visual inspection of the reaction times revealed that participants from both contexts responded faster in the BX and BY conditions than in the AX and AY conditions, also in line with earlier studies (Luque & Morgan-Short, Reference Luque and Morgan-Short2021). When comparing reaction times across contexts, participants in the integrated context were overall faster across the four conditions (see Table 3).
Table 3. Mean and standard deviations for reaction times in the four conditions of the AX-CPT task, per context
Note. Values in parentheses represent standard deviations. Reaction times per condition are in milliseconds.
To address the study’s research questions, we conducted correlational analyses and mixed-effects models. The entropy measure was used to quantify linguistic variability within the separate and integrated contexts and to examine its association with participants’ performance on the AX-CPT task, specifically, accuracy and reaction times in the AY (reactive control), BX (proactive control) and BY (processing speed) sequences (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018).
Entropy scores range from 0 to 1.00 for participants who report using two languages and from 0 to 1.56 for those who report using an additional third language (L3). A score close to 0 indicates that the speaker primarily uses a single language, whereas higher scores reflect a greater likelihood of using multiple languages. Thus, lower entropy values indicate a more separate or compartmentalized use of two or more languages, while higher values indicate a more integrated use of the languages (Wagner et al., Reference Wagner, Byers-Heinlein, Schimke, Meade, Trofimovich and Kroll2023). For instance, an entropy score of 0.38 for a bilingual participant would suggest that one language (L1 or L2) is used more frequently, whereas a score of 0.51 would represent a more balanced use of both languages. Measures of social context and overall exposure to languages (L1, L2 and, when applicable, L3) were extracted from self-reported data obtained through the LEAP-Q and converted into entropy scores (see Gullifer & Titone, Reference Gullifer and Titone2020, for details on the computation of entropy scores). These two measures were selected because our hypotheses focused on how differences in the recruitment of proactive and reactive control vary according to linguistic variability in the separate and integrated contexts. We expected that a more compartmentalized versus more integrated use and exposure to Spanish and English across settings and activities would differentially affect participants’ performance, particularly when language use was more integrated in social interactions.
Social data were collected using a Likert scale ranging from 1 (‘never’) to 5 (‘always’) (e.g., ‘Please rate to what extent you are currently exposed to English in the following contexts’). Following Gullifer and Titone (Reference Gullifer and Titone2020), scores were baseline-adjusted by subtracting 1, such that 0 represented ‘never’ and 4 represented ‘always’. For instance, if a participant reported language use with friends as L1 = 3, L2 = 2 and L3 = 0, the resulting proportions for that context would be L1: 3/5, L2: 2/5 and L3: 0/5. Overall exposure data were elicited using percentage estimates. Participants were asked: ‘Please list what percentage of the time you are currently and on average exposed to each language’. Percentages added up to 100 across all reported languages. All responses were first converted into proportions and then transformed into entropy scores using the language Entropy package in R (Gullifer & Titone, Reference Gullifer and Titone2018).
We first examined correlations between social and overall linguistic exposure and performance on AY, BX and BY trials. For bilinguals in the separate context, we observed a significant negative correlation between reaction times in BY trials (processing speed) and overall exposure (r = −.39, p = .05), indicating that greater overall linguistic diversity was associated with faster responses. For the integrated context, greater linguistic diversity in social contexts correlated positively with accuracy in BY trials (r = .42, p = .03), suggesting that increased integration of language use related to higher processing efficiency.
Next, we tested correlations between L2 proficiency and the AY, BX and BY measures (Luque & Morgan-Short, Reference Luque and Morgan-Short2021). L2 proficiency was calculated as a composite score derived from participants’ performance on the LexTALE (English) and their mean number of words produced in the English block of the Category Fluency task (Luque & Morgan-Short, Reference Luque and Morgan-Short2021; Raisman-Carlovich et al., Reference Raisman-Carlovich, Carrasco-Ortiz and Arias-Trejo2024). Both measures were standardized (z-scored) and averaged to yield the final proficiency composite, which provided a more comprehensive index of participants’ receptive and productive L2 performance. A significant positive correlation emerged between the L2 proficiency composite and accuracy in BX trials (r = .49, p = .01) for bilinguals in the separate context, indicating that higher L2 proficiency was associated with greater accuracy in trials requiring proactive control. Finally, no evidence was found that L2 age of acquisition was related to either proactive or reactive control in either group.
To further examine the modulating roles of linguistic diversity in social language use, linguistic diversity in overall language use and English proficiency in the recruitment of proactive and reactive cognitive control mechanisms across the separate and integrated contexts, we conducted a series of mixed-effects models. The fixed effects included English proficiency, social language entropy, overall language entropy, Condition (AY, BX, BY) and Context (0 = separate, 1 = integrated). English proficiency, social language entropy and overall language entropy were z-scored to facilitate interpretation, whereas Condition and Context were dummy-coded. The AY sequence served as the reference level for Condition, allowing comparisons with the BX and BY sequences, respectively (Gullifer et al., Reference Gullifer, Chai, Whitford, Pivneva, Baum, Klein and Titone2018). Random effects consisted of by-participant intercepts and slopes for Condition. Model comparisons were conducted using chi-squared log-likelihood ratio tests with maximum likelihood estimation to determine the optimal random-effects structure. The final model was specified as follows: RT ~ Context × Condition × Social entropy + (1 + Condition | Participant). Table 4 presents the fixed and random effects, along with the corresponding statistical estimates.
Table 4. ANOVA results on the final linear mixed-effects model for RT per condition of the AX-CPT and context
Note. Sum Sq = Sum of Squares; Mean Sq = Mean Squares; Num DF = numerator degrees of freedom; Den DF = denominator degrees of freedom; F = F statistic. Statistically significant effects are marked as follows: *p < .05; **p < .01; ***p < .001.
As this was our final model, we checked assumptions of linearity of data, normality of residuals and homogeneity of variance and found no violations of any of these assumptions. We additionally tested the model for multicollinearity using the variance inflation factor (VIF) and none of the factors displayed values above 1.5, indicating that there was no multicollinearity in the final model.
We conducted post-hoc comparisons using the emmeans function (Lenth, Reference Lenth2024) in R (R Core Team, 2023). Tukey’s HSD test for multiple comparisons revealed significant differences between conditions. Overall, participants showed enhanced proactive control and processing speed compared to reactive control. In the separate context, response times in the BX condition were significantly faster than in the AY condition (SE = 0.1085, t = 6.728, p < .0001, df = 54.5) and response times in the BY condition were also significantly faster than in the AY condition (SE = 0.0984, t = 7.011, p < .0001, df = 54.4). A similar pattern emerged in the integrated context, with significantly faster reaction times in the BX condition than in the AY condition (SE = 01095, t = 6.512, p < .0001, df = 54.4) and in the BY condition compared to the AY condition (SE = 0.0994, t = 6.477, p < .0001, df = 54.4).
The model revealed a significant interaction between Context, Condition and Social language entropy. For the integrated context (see Figure 2), a higher entropy score predicted slower reaction times for the BX and BY conditions, indicating that as linguistic variability in social situations increases, bilinguals tend to take longer to adjust to proactive control strategies and to exhibit overall slower processing speed.
Figure 2. Social entropy as predictor of RT per condition in the integrated context. Note. Log RT = logged reaction times in each condition (AY, BX, BY) of the AX-CPT. Social entropy scores closer to 0 typically represent compartmentalized contexts. As scores increase, the likelihood of language variability increases too.
The opposite pattern was found in the separate context (see Figure 3). As social entropy increases, reaction times in the three conditions of interest (AY, BX, BY) become faster. For these bilinguals, greater linguistic variability in social situations appears to enhance the recruitment of both proactive and reactive control as well as processing speed.
Figure 3. Social entropy as predictor of RT per condition in the separate context. Note. Log RT = logged reaction times in each condition (AY, BX, BY) of the AX-CPT. Social entropy scores closer to 0 typically represent compartmentalized contexts. As scores increase, the likelihood of language variability increases too.
4. Discussion
In the present study, we examined the engagement of proactive and reactive control mechanisms in two groups of Mexican young adult Spanish–English bilinguals who differed in their contexts of language use and exposure. In the separate context, bilinguals reported using Spanish and English in distinct situations and with different conversational partners, whereas in the integrated context, bilinguals frequently alternated between both languages within the same situations and often with the same interlocutors. We expected to find differences between the two groups as a function of the separate or more integrated use of the L1 and L2, based on previous findings suggesting that bilinguals’ engagement of proactive and reactive cognitive control mechanisms is shaped by the linguistic characteristics of their context (Beatty-Martinez et al., Reference Beatty-Martinez, Navarro-Torres, Dussias and Kroll2019). Additionally, we aimed to investigate the modulating roles of L2 proficiency and L2 age of acquisition in the recruitment of proactive and reactive control, as reported in studies employing the AX-CPT task, a fine-grained measure of cognitive control (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018; Luque & Morgan-Short, Reference Luque and Morgan-Short2021).
Our participants generally demonstrated enhanced performance, indexed by faster reaction times, in conditions requiring proactive control (BX) and processing speed (BY), relative to those requiring reactive control (AY), regardless of the context. Notably, for bilinguals in the separate context, greater linguistic variability in social situations predicted faster responses in trials involving proactive and reactive control as well as processing speed. In contrast, for participants in the integrated context, higher linguistic variability in social contexts predicted slower responses in trials engaging proactive control and processing speed, though not in those related to reactive control.
Descriptive analyses of reaction times and accuracy rates across the four sequences (AX, AY, BX and BY) indicated that both groups performed similarly, with participants from the integrated context responding slightly faster across all conditions. Overall, participants were slower in AY trials (reactive control) than in BX (proactive control) and BY (processing speed) trials, although accuracy rates across the four conditions did not differ significantly. Nonetheless, bilinguals in the separate context made more errors in AY trials than in the other three conditions, whereas bilinguals in the integrated context showed comparable accuracy across all conditions.
The present findings align with previous research suggesting that proactive control processes may play a more prominent role once bilinguals attain higher proficiency levels and accumulate greater experience in the L2 (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018; Luque & Morgan-Short, Reference Luque and Morgan-Short2021). In terms of proficiency, our participants ranged from low to high-intermediate. Regarding experience with English, approximately half lived in a context that facilitated frequent use and exposure to both languages across various situations (integrated context), and 46% (23 participants) reported immersion experiences in countries where English was the dominant language. Thus, participants’ overall proficiency levels and L2 experience may help explain the greater engagement of proactive over reactive control across both groups.
Although our primary focus was on the AY, BX and BY sequences, we also analyzed reaction times and accuracy indices for the AX condition and found no significant between-group differences. The AX condition serves as an indicator of adequate monitoring of contextual cues, as it involves the sequence that correctly predicts the target. Finally, it is noteworthy that our results point to a possible contribution of processing speed, given that accuracy and reaction times in BY trials were comparable to those in BX trials for both groups.
We also examined whether L2 age of acquisition and L2 proficiency modulated the recruitment of proactive and reactive control, as reported in previous studies (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018; Luque & Morgan-Short, Reference Luque and Morgan-Short2021). No associations were found between L2 age of acquisition and proactive or reactive control for either group. However, among bilinguals in the separate context, higher English proficiency predicted greater accuracy in trials requiring proactive control, again supporting the notion that proactive control mechanisms become increasingly relevant as bilinguals reach higher levels of L2 proficiency (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018; Luque & Morgan-Short, Reference Luque and Morgan-Short2021). This association was not observed in the integrated context.
When performing mixed-effects models, linguistic diversity in social contexts emerged as a significant modulator of the cognitive control strategies our participants engaged in. In a separate context, higher entropy in social situations predicted faster reaction times in proactive and reactive control, as well as in processing speed. In contrast, in the integrated context, higher entropy across social domains predicted slower reaction times in proactive control and processing speed. These findings partially support the Adaptive Control Hypothesis (Green & Abutalebi, Reference Green and Abutalebi2013), which posits that both proactive and reactive control are implicated in bilingual language performance, and that contextual characteristics differentially shape the recruitment of these control processes.
The principal distinction between our two contexts lies in the degree to which Spanish and English are used separately or in an integrated manner. Bilinguals in the separate context primarily use English in academic settings, such as language classes, reading discipline-related materials and consuming media, while Spanish dominates their everyday interactions with peers. Conversely, bilinguals in the integrated context are exposed to and use both languages across many situations, largely due to their geographical environment, frequently alternating between Spanish and English in social interactions. The variability in language use across social spheres may significantly influence bilinguals’ cognitive performance beyond the effects of L2 age of acquisition and proficiency (Green & Abutalebi, Reference Green and Abutalebi2013; Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018).
The entropy measure allowed us to quantify linguistic diversity in social contexts to examine how a more separated or integrated use of Spanish and English contributes to adaptive changes in cognitive control engagement (Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018). Compartmentalized environments typically exhibit low entropy, with predictable language choices, whereas integrated contexts display higher linguistic entropy, making language selection more variable and less predictable. In settings where both languages are active and thus compete for selection, the demands on goal maintenance, conflict monitoring and interference suppression are likely to increase. However, contexts may differ in how bilinguals manage interference (Beatty-Martinez et al., Reference Beatty-Martinez, Navarro-Torres, Dussias and Kroll2019). For bilinguals in the separate context, efficient conflict resolution may rely on response inhibition; moreover, greater linguistic variability across social situations was associated with faster reaction times, reflecting adaptive responses linked to reactive control. In contrast, bilinguals in the integrated context may engage proactive control to manage conflict, possibly by limiting the activation of competing linguistic information through context monitoring and opportunistic planning, that is, by flexibly using available linguistic resources to achieve successful communication.
Researchers have distinguished theoretical accounts that emphasize control processes aimed at maintaining task goals and monitoring for conflict from those that highlight the need to inhibit competing representations from the L1 and L2 (Green & Abutalebi, Reference Green and Abutalebi2013; Hilchey & Klein, Reference Hilchey and Klein2011). Moreover, language representations are assumed to be in a cooperative relationship in integrated contexts, whereas they are hypothesized to be in competition in separate contexts (Green & Abutalebi, Reference Green and Abutalebi2013). Thus, bilinguals adapt to their interactional environments to minimize the cognitive cost of not doing so. In this sense, when participants from the separate context are immersed in social situations where both languages are present and language choice becomes less predictable, they may be prompted to monitor the context for linguistic cues signaling language selection and alternation, as well as to inhibit the language that is not currently relevant. This process could enhance their engagement of proactive and reactive control strategies, as well as processing speed. In contrast, bilinguals from the integrated context experience daily linguistic environments characterized by the constant co-activation of Spanish and English. When engaged in social situations, accuracy, rather than speed, may be prioritized. For these participants, better regulation of context processing and opportunistic planning may occur at the expense of longer response times (Morales et al., Reference Morales, Calvo and Bialystok2013).
Findings from the present study support the need for models that integrate both types of cognitive control mechanisms (e.g., the Adaptive Control Hypothesis; Green & Abutalebi, Reference Green and Abutalebi2013) rather than those focused exclusively on inhibitory control. Another important implication is that bilinguals living in the same country, who share an L1 and speak the same languages, may nonetheless exhibit distinct patterns of adaptive responses (Beatty-Martinez et al., Reference Beatty-Martinez, Navarro-Torres, Dussias and Kroll2019). Subtle differences in our participants’ use and exposure to Spanish and English were sufficient to produce differences in cognitive control patterns when performing a fine-grained task like the AX-CPT. The use of such complex tasks or the combination of multiple measures is particularly relevant for young adults, who may otherwise reach ceiling effects on simpler tasks. Incorporating diverse measures of cognitive control could thus help elucidate how proactive and reactive mechanisms function among bilinguals in compartmentalized versus integrated contexts, and how their engagement may adapt to varying degrees of linguistic variability.
Finally, we cannot rule out the possibility that between-group differences in the recruitment of cognitive control mechanisms were also influenced by participants’ cultural identity, as suggested by previous research demonstrating the impact of biculturalism on cognitive control (West et al., Reference West, Zhang, Yampolsky and Sasaki2017; Ye et al., Reference Ye, Mo and Wu2016). Bilinguals from northern Mexico (integrated context) are more likely to be exposed to both Mexican and American cultures from an early age and to identify as bicultural bilinguals, unlike young adults from Mexico City (separate context).
5. Conclusions
Interpreting results from the AX-CPT task in terms of ‘advantages’ is controversial: lower error rates and faster reaction times do not necessarily indicate superior cognitive performance (Beatty-Martínez et al., Reference Beatty-Martinez, Navarro-Torres, Dussias and Kroll2019; Gullifer et al., Reference Gullifer, Kousaie, Philippe, Pivneva, White, Titone and Sheldon2018). Instead, it is the interplay between proactive and reactive control mechanisms that enables bilinguals to adapt to the demands of their linguistic context. The complementary contributions of proactive/monitoring and reactive/inhibitory mechanisms appear to underlie successful cognitive performance (Braver, Reference Braver2012; Burgess & Braver, Reference Burgess and Braver2010).
The patterns of proactive and reactive control observed in this study reveal differences in how bilinguals manage the coactivation of Spanish and English across social spheres, supporting the Adaptive Control Hypothesis (Green & Abutalebi, Reference Green and Abutalebi2013). Bilinguals in the separate context appear to achieve conflict resolution primarily through response inhibition, whereas bilinguals in the integrated context seem to limit the activation of competing representations via careful context monitoring and opportunistic planning.
These findings underscore the need to characterize bilinguals more thoroughly by considering both static variables, such as L2 age of acquisition, and dynamic variables, such as L2 proficiency and patterns of language use and exposure. In the present study, between-group differences emerged when examining the interaction between participants’ recruitment of proactive and reactive control strategies and the linguistic variability of their social contexts.
The study has several limitations. A larger sample including a combination of early- and late-L2 age of acquisition participants might have clarified the potential role of age of acquisition in the recruitment of proactive and reactive control, as most participants in this study were late bilinguals, and no correlations were observed between L2 age of acquisition and AX-CPT performance. Additionally, a more comprehensive assessment of L2 proficiency might have revealed further interactions between proficiency and cognitive control, although we did compute a composite score combining the Category Fluency task (average number of words produced; Raisman-Carlovich et al., Reference Raisman-Carlovich, Carrasco-Ortiz and Arias-Trejo2024) and LexTALE scores.
Data availability statement
The authors confirm that the data supporting the findings of this study are available within the article and at https://osf.io/4xzgj/.
Acknowledgements
We are grateful to the study participants and to the members of the Psycholinguistics Lab (Laboratorio de Psicolingüística) at the Universidad Nacional Autónoma de México for their support and feedback. We also extend our special thanks to Andrea Takahesu-Tabori for her invaluable advice on the design of the AX-CPT task.
Funding statement
The authors disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This work was supported by the CONACYT (Consejo Nacional de Ciencia y Tecnología de México) National grant 2018 awarded to Alejandra Raisman-Carlovich (grant number 726489, CVU 178455).
Competing interests
The authors report no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Open access
