Bilingualism

Brain differences

Models of neurocognitive adaptation to bilingualism

{DeLuca et al., 2020} Several models have been proposed which outline different facets of bilingual experience and the related neural and cognitive adaptations to them. We focus here on recent models (Fig. 1) that examine different aspects of the bilingual experience (e.g. duration of L2 use, degree/nature of switching, etc.) and that provide predictions about how the brain adapts to these experiences. The models discussed herein are, in turn, based on earlier models that also covered language-related neuroplasticity, such as the bilingual interactive activation model (van Heuven, Dijkstra, & Grainger, 1998), inhibitory control model (Green & Eckhardt, 1998), and declarative procedural model (Paradis, 2009; Ullman, 2004). Thus, while these earlier models are not directly reviewed, we build on their predictions indirectly. It is also worth noting that the models reviewed here primarily discuss language production as a primary driver of adaptation, but do not explicitly distinguish this from comprehension. Adaptations to experience-based factors involve both the variable enhancement of existing structures and networks with changing environmental demands, and repurposing of existing modules/regions to new processes with continued bilingual language exposure (Hernandez et al., 2019; Rodriguez, Archila-Suerte, Vaughn, Chiarello, & Hernandez, 2018).

The Adaptive Control Hypothesis (ACH) (Abutalebi & Green, 2016; Calabria et al., 2018; Green & Abutalebi, 2013), discusses adaptation to bilingual experience trajectory from a perspective of intensity of engagement with one’s languages. Specifically, the ACH proposes that several cognitive processes are required for successful language control, including goal maintenance, response inhibition, conflict monitoring, interference suppression, salient cue detection, task engagement and disengagement, and opportunistic planning. These processes are variably required, depending on the conversational context. There are three general categories of communicative context. First, the single language context captures the use of only one of the available languages in a specific environment. This context is assumed to only engage goal maintenance and is therefore thought to require only increased input from the inferior frontal gyrus (IFG). Second, the dual language context captures situations in which both languages are used but with different interlocutors. This context is argued to require several control processes, namely goal maintenance, interference control, salient cue detection, selective response inhibition, and task engagement/disengagement. Due to this complexity, it engages a wide network of brain regions (IFG (bilateral), anterior cingulate cortex (ACC), inferior parietal lobule (IPL), caudate, putamen, thalamus, and cerebellum). Third, the dense code-switching context captures situations in which both languages are used with the same interlocutor, with frequent language switches, including those within utterances. This type of language use is less taxing on control processes. It requires only opportunistic planning and relatively less inhibition. It is thought to use a control network of the left IFG and cerebellum. Individuals can engage several (or all) of these interactional contexts and may even shift between them on a regular basis. However, it is argued that increased engagement with a specific context reinforces (by necessity) the cognitive processes required by it and their underlying networks. These network reinforcements, then, are argued to result in increased performance on tasks that tap into the associated cognitive processes and increased plasticity and connectivity in the relevant brain regions associated with them.

The Conditional Routing Model (CRM) (Stocco et al., 2014; Stocco, Lebiere, & Anderson, 2010) proposes a more functional series of adaptations to bilingual language control and processing, with a special focus on the basal ganglia (in particular the striatum). This model is based on the notion of the basal ganglia functioning as a gate of neural signals to prefrontal and other cortical regions (Stocco et al., 2010). It is argued that the basal ganglia (specifically, the caudate and putamen) have the capacity to select rules in response to specific, prerequisite conditions and to override habitual responses encoded in cortico-cortical connections. Bilingualism effectively trains this capacity because the acquisition and use of an additional language means frequent selection and switches between rules and representations. Training the system gives increased efficiency in rule acquisition, selection, and application. Since all these processes are implicated in general executive function processes, bilingualism affects performance in non-linguistic executive control tasks that rely on rule selection, switching and top-down control to maintain a rule behavior in the presence of distracting information.

The Bilingual Anterior to Posterior and Subcortical Shift (BAPSS) framework (Grundy et al., 2017) proposes a series of neural adaptations to prolonged experience with the additional language. The BAPSS framework specifically proposes that early stages of L2 (second language) use require increased reliance on frontal cortical regions (e.g. ACC and IFG), due to increased demands on both language and executive control processes, which are predominantly served by these regions. With prolonged L2 use, neural reliance shifts from frontal regions to subcortical and posterior regions (e.g. basal ganglia, thalamus, occipital lobes) commensurate with more automated and efficient language control and processing. The BAPSS framework also makes predictions regarding neural activity. With respect to electrophysiology, in relation to stimuli probing or requiring switching and inhibitory control prolonged duration of L2 use will correlate with both earlier latencies and increased amplitudes in the N2/P300 components reflecting earlier and more automatic detection of conflicting input, and more confidence in its classification. Decreased error-related negativity (ERN) amplitudes, are also predicted with prolonged experience, indicative of lower resources being allocated at this point, given greater re- sources being allocated temporally earlier (i.e. at the N2). The framework also discusses these neural adaptations to bilingualism as a mechanism for counteracting the anterior shift in neural reliance associated with ageing (Grant, Dennis, & Li, 2014).

Finally, the Dynamic Restructuring Model (DRM) (Pliatsikas, 2019b) proposes a system of stages in neuroanatomical adaptations to the duration of L2 exposure and usage and the varying cognitive demands resulting from it. Three stages of adaptation are proposed. The first stage, initial exposure, confers adaptations geared towards language learning and early control demands. Adaptations to these new language control demands result in increased reliance on cortical structures implicated in cognitive control and short-term memory, including the IFG, ACC, inferior parietal lobule/superior parietal lobule (IPL/SPL), and hippocampus. The second stage, consolidation, occurs with increased L2 experience and is characterized by shifts towards more efficient control of the two languages. This results in increased structural connectivity between regions, impacting on several white matter tracts including the inferior fronto-occipital fasciculus (IFOF), anterior thalamic radiation (ATR) and superior longitudinal fasciculus (SLF), increases in subcortical grey matter (e.g. caudate, putamen, and thalamus), and a ‘renormalization’ or return to baseline levels of cortical grey matter. The final stage, peak efficiency, deals with adaptation towards automation in L2 processing and control. This results in increases in cerebellar grey matter and decreases in frontal connectivity (e.g. within forceps minor), as reliance shifts to posterior regions.

In what follows, we will put forward a unified comprehensive framework, which we refer to as Unifying the Bilingual Experience Trajectories (UBET), on the relationship between the bilingual experience trajectory and neurocognitive adaptations. In this, we are guided by the overlap of the existing models but also make some new, testable predictions. As illustrated in Fig. 3, a variety of neurocognitive adaptations are predicted depending on a range of individual experience-based factors. More specifically, some experience-based factors are predicted to primarily lead to individual differences in control demands while others to individual differences in automaticity/efficiency (Fig. 3). Crucially, though, we propose that these experiences will modulate the effects of one another in terms of how the brain adapts. The specific experience-based factors, their neurocognitive effects, and (modulatory) relationships are discussed in the following section. Regarding the experience-based factors that give rise to these neurocognitive adaptations, there is a range of specific language experiences that might variably translate to different neural adaptations. Herein, we grouped these under four general categories: 1) intensity and diversity of L2 use, 2) language switching, 3) relative proficiency, and 4) duration of bilingual use (Fig. 3).

Intensity and diversity of language use, taken together, refer to the extent and variety of situations in which one is exposed to and uses an additional language (Grosjean, 2016). Diversity of language use here stems from the notion of entropy as operationalized by Gullifer and Titone (2020), specifically the degree of compartmentalization or integration of one’s languages within and across specific contexts. Intensity of use refers to the extent to which an additional language is at least available or used in daily life, overall. Although intensity and diversity of language use are different aspects, they are combined here as they will likely converge with respect to the nature of the demands placed on language control and processing faculties. Specifically, both greater intensity and diversity of language use are predicted to be associated with increased control demands (Beatty-Martínez et al., 2019; Green & Abutalebi, 2013).

Language Switching herein is defined as a general term encompassing the extent to which one switches between- and/or mixes their languages on a regular basis. We discuss language switching using the typology proposed by Muysken (2000), and the control demands associated (Green & Wei, 2014; Hofweber, Marinis, & Treffers-Daller, 2016). The different types of switching are argued to exist on a spectrum including single language use, alternation between languages, insertion, and dense code switching (Hofweber et al., 2016). The associated control demands range from maximal interference suppression in alternation or inter-utterance switches to minimal requirements of interference suppression in dense code-switching (Hofweber et al., 2016; Treffers-Daller, 2009). En- gagement with a specific type of switching (as defined above) will necessitate commensurate neurocognitive adaptations (Adler, Valdés Kroff, & Novick, 2020; Hofweber et al., 2016). Cognitive demands associated with monitoring changing conflict, however, follow the reverse pattern: dense code switching requiring the most monitoring and single language use requiring the least conflict monitoring (Hofweber et al., 2016). Despite requiring potentially differential engagement of control processes, neurocognitive adaptations to language switching are herein proposed to fall under control demands (discussed further in section 2.1).

Relative proficiency refers to the balance of fluency and comfort/confidence in using the languages at one’s disposal. Relative proficiency herein is operationalized as being formed of two subcomponents: language proficiency and dominance. Proficiency herein is defined as the extent to which the representation fits a target (native-like) state, typically as measured by a standardized test. Of the two subcomponents, it is possible that proficiency (past a certain point of attainment) does not necessarily affect control processes and indeed might be considered an outcome measure in its own right (see for discussion DeLuca et al., 2018). However, within early stages of acquisition, previous work has shown higher proficiency levels to relate to increased degrees of neural plasticity (Mamiya et al., 2016; Mårtensson et al., 2012). Although to some degree related to proficiency, here we define language dominance as the comfort with using and/or ease of access to a given representation (Bedore et al., 2012). Dominance is a subject-internal relative measure, whereas proficiency (as defined above) in each language is not necessarily dependent on the other. In terms of neurocognitive adaptation, more balanced relative proficiency should lead to increased efficiency of language control (Stocco et al., 2014; Yamasaki et al., 2018), but this is likely dependent on several of the other experience-based factors (see section 2.3) (Beatty-Martínez et al., 2019; Grundy et al., 2017; Gullifer et al., 2018). While fully balanced relative proficiency is rare, a minimum threshold should likely be met for both maintenance of the linguistic representations and control demands to be continuously necessitated (Iverson & Miller, 2017; Miller & Rothman, 2019).

Finally, duration of L2 use refers to the overall length of time one engages with more than one language. Neurocognitive adaptations to longer duration of experience is predicted to lead to increased efficiency of language control and processing (Grundy et al., 2017; Pliatsikas, 2019b).

2.1. Neurocognitive adaptations

The neurocognitive adaptations to the experience-based factors described above can be grouped into two general categories:

  1. adaptations due to changes in executive control demands, and

  2. adaptations due to changes in efficiency.

As described above, changes in executive control demands arise from various demands on language use and control, such as increased intensity and diversity of language use and increased language switching. A remarkable change in control demands is expected when the L2 is used in novel situations or patterns, e.g. from initial exposure to an L2, or when the degree of intensity of L2 usage dramatically increases or decreases. Cognitive processes relevant to language control become increasingly stressed with higher language control demands, and the brain adapts accordingly. Given previous findings, we predict the following neurocognitive adaptations to occur as a result of changes in control demands. Neural correlates of control demands include increased recruitment of cortical regions involved in control processes (e.g. the IFG, ACC, and IPL). Structurally, these regions are predicted to exhibit greater grey matter volume to handle the additional control demand (Calabria et al., 2018; Pliatsikas, 2019b). Increased control demands have also been shown to manifest as a general reliance on proactive control strategies (Gullifer et al., 2018), which might be evident in the performance in non-verbal behavioral tasks, such as smaller mixing/switching costs in switching tasks and superior suppression of interfering information/stimuli in selective attention tasks. Increased language control demands should reinforce the use of these networks in non-linguistic contexts. Consequently, the increased control demands should likely also manifest as increased stimulus-related theta band power over mid-frontal scalp regions, increased alpha band power over task-irrelevant sensory cortices and suppression of alpha band power over task-relevant regions. Theta band activation has previously been linked with several processes including conflict monitoring and interference suppression (Janssens, De Loof, Boehler, Pourtois, & Verguts, 2018; Nigbur et al., 2011). The increase in alpha activity in task-irrelevant region has been associated with functional inhibition of distracting and efficient allocation of resources (see Van Diepen et al., 2019).

Efficiency arises from becoming accustomed to existing language control demands. In this instance, cognitive economy dictates adaptations that allow for more efficient and automated cognitive processes to handle these demands. These adaptations likely only occur once the language user is maximally effective at handling the cognitive load associated with the control demands of the (overall) language environment. Thus, prolonged duration of L2 use (a scale of years) and increased L2 proficiency will confer adaptations that group under efficiency. We predict the following neurocognitive adaptations to occur as a result of changes in efficiency. Neural correlates of efficiency (as discussed by the BAPSS framework, CRM and DRM) include:

  1. preferential recruitment of subcortical structures (specifically caudate, putamen, and thalamus) (Stocco et al., 2014) and

  2. posterior regions such as the occipital lobes and cerebellum (Grundy et al., 2017; Pliatsikas, 2019b).

Increasing automaticity will be associated with

  1. prior cortical grey matter increases reverting towards near-baseline levels (Grundy et al., 2017; Lövdén, Wenger, Mårtensson, Lindenberger, & Bäckman, 2013; Pliatsikas, 2019b), and

  2. increases then occurring in subcortical and posterior structures (Grundy et al., 2017; Pliatsikas, 2019b).

Regarding ERP components related to attentional/executive control, prolonged L2 exposure should result in earlier latency and increased amplitude for components such as the N2 and P300 (Grundy et al., 2017). Here as well, it is likely that increased L2 exposure during this time frame would shorten the latency by which this transition occurs. Increases in stimulus-related alpha power in task-irrelevant cortical regions would also indicate a shift towards efficiency, specifically in the allocation of attentional resources. As explained above, alpha is argued to be an index of the gating of activity in non-relevant brain regions for specific tasks, including interference suppression and the biasing of attentional states (Foxe, Simpson, & Ahlfors, 1998). Indeed, previous work has shown increased alpha band activation to be linked to inhibitory processes in tasks requiring conflict monitoring and interference suppression (Janssens et al., 2018). Finally, cognitively, increased efficiency is predicted to manifest as a reliance on reactive control processes (Gullifer et al., 2018).

2.2. Relationships between different neurocognitive outcomes

Patterns of neurocognitive adaptation are likely interrelated, although this has not been explicitly discussed in previous models. Here we discuss several predictions of what relationships between brain structure, function, and cognition exist, as modulated by bilingual experience-based factors. We predict that brain function patterns will follow similar patterns of adaptation to structural adaptations with initial exposure to an L2. In other words, the initial stages of contact with the L2 will result in increased functional recruitment of mainly fronto-cortical regions to handle the increased load. We also predict this increase in functional recruitment will be associated with increases in grey matter in these regions. Increased efficiency and automation lead to a reduction in executive control demands, meaning that functional recruitment patterns in fronto-cortical regions will begin to revert towards a baseline level (Grundy et al., 2017). In line with predictions of the DRM, we predict that this reversal will be associated with decreased grey matter within these regions (Pliatsikas, 2019b) and as result with a decreased theta power induced by executive control demands (Fig. 4).

Regarding the relationship between adaptations in stimulus-related oscillatory dynamics and brain structure, an increase in alpha power over sensory regions in situations requiring executive control could indicate a reliance on subcortical structures and the cerebellum for control processes (see Mazaheri, Nieuwenhuis, Van Dijk, & Jensen, 2009 for discussion) (Fig. 4). Thus, we expect an increase in alpha power over cortical regions in tasks requiring interference suppression as task demands become more reliant on subcortical structures (particularly the basal ganglia) (Mazzetti et al., 2019; Talakoub et al., 2016) and reduced cortical requirements to maintain successful language control. Interestingly, a recent study observed a relationship between the inter-individual variability among participants’ abilities to modulate alpha activity over the visual cortex, and volume of the globus pallidus, a basal ganglia structure (Helfrich, Huang, Wilson, & Knight, 2017). Previous research has suggested that alpha activity over the sensory-cortices to be partially controlled by sources in the prefrontal cortex (Helfrich et al., 2017; Mazaheri et al., 2014, 2010; Sauseng, Feldheim, Freunberger, & Hummel, 2011). We speculate that with prolonged bilingual experience, a shift would occur from frontal cortical structures to more subcortical regions and tracts connecting these, modulating alpha activity (Fig. 4).

What do these changes mean for cognitive (behavioral) outcomes pertaining to executive control (reaction times and accuracy)? We argue that those will improve in early stages of bilingualism and then plateau at the level of peak efficiency. Specifically, upon peak performance being reached (in terms of behavioral responses), this performance would be maintained with an increase in alpha over task-relevant and irrelevant cortices, indicating a further shift towards automated and more efficient processing. Regardless, maintenance of peak efficiency will drive the specific trajectory of adaptations. That is, cognitive adaptations are likely to follow neural adaptations-only arriving at what is required to handle the competing systems. After this, the adaptation trajectory shifts to decreasing (neuro)cognitive requirements for maintaining this level of performance. Therefore, it is unlikely that, once peak performance has been reached, changes in (neural) processing speed would manifest any differently on behavioral measures in executive function tasks such as reaction times. However, as has been argued previously, the degree to which task performance is affected likely also depends on the nature of the task used (see for discussion Bialystok, 2016; Valian, 2015). Employing tasks which require a greater degree of attentional control, such as the attentional network task (ANT), or manipulating (typically increasing) the ratio of congruent to incongruent trials (e.g. Costa et al., 2009) is more likely to elicit behavioral effects stemming from bilingual experience (Bialystok, 2017; Zhou & Krott, 2018). Similarly, tasks which isolate the contributory aspects of executive control, such as the AX-continuous performance task (AX-CPT) which is thought to isolate proactive and reactive control, will likely better capture the behavioral correlates of neurocognitive adaptation to specific experience-based factors (see e.g. Gullifer et al., 2018). For tasks better suited to directly measure differences between proactive and reactive control, we predict increased proactive control to be associated with greater frontal activation and theta band power (Fig. 4). Shifts towards more reliance on reactive control for control demands would be associated with a suppression of alpha power over task relevant cortical regions, and recruitment of subcortical/posterior structures (Grundy et al., 2017) (Fig. 4).

2.3. Relationships between experience-based factors making up the bilingual experience trajectory

Crucially, while the above experience-based factors are expected to relate to specific neural outcomes, they do not occur in isolation. We therefore need to consider interactions of the factors and what these interactions mean with regards to neurocognitive adaptations (Sulpizio, Del Maschio, Del Mauro et al., 2020). While it is possible to formulate predictions for the combinations of some of our experience-based factors, others have either not been sufficiently studied or we do not have sufficient theoretical reason to relate them. Furthermore, the socio-linguistic environment in which a bilingual speaker is situated should determine the combination and subsequent neurocognitive effects of these experience-based factors. In what follows, we will describe two likely interactions between our experience-based factors, specifically how they might modulate their respective neurocognitive adaptations. First, we predict that increased intensity and/or diversity of language use will shorten the latency by which adaptations towards efficiency and automation occur with duration of use (Fig. 3). Prolonged L2 immersion as an experience-based factor provides a key example of this. Regarding the role of language use, immersion in non-native language settings can reliably be assumed to at least intensify exposure to the L2 and increase the opportunities for use within a given timeframe (Linck, Kroll, & Sunderman, 2009). As such, the language control system optimizes to handle the additional language control load, in the above cases towards maximizing efficiency of language processing and control. Previous research has found that highly immersed late L2-acquired bilinguals and early L2-acquired bilinguals not in immersion settings exhibit very similar brain structure adaptations towards increased efficiency and automation (as predicted by BAPSS and DRM), relative to functional monolingual control groups. Specifically, in the subcortical structures, immersed bilingual groups showed increased surface displacement in the putamen, thalamus and globus pallidus (Pliatsikas et al., 2017) relative to monolingual controls, a restructuring pattern that highly overlaps with early simultaneous bilinguals not experiencing L2 immersion (Burgaleta et al., 2016). A similar pattern has been found for white matter tracts. A study by Pliatsikas and colleagues found increased white matter myelination in the corpus callosum, IFOF, and SLF for immersed bilinguals (Pliatsikas et al., 2015). Here again, this pattern overlaps with early-acquired bilinguals who were not living in L2 immersion settings (García-Pentón, Perez Fernandez, Iturria-Medina, Gillon-Dowens, & Carreiras, 2014). The tracts and structures reported in the above studies have been implicated in duration-based models (BAPSS and DRM) as being increasingly used as language control becomes more automated. Such effects, then, can be interpreted as increased intensity of L2 use shortening the latency by which adaptations towards increased efficiency/automation of language control occur. Given the current evidence, the UBET framework would predict a shift towards reliance on reactive control processes (e.g. Gullifer et al., 2018) with prolonged L2 immersion.

Neurocognitive adaptations to degree of relative proficiency are likely to be conditioned by both the opportunities for use (e.g. compartmentalization of languages, Beatty-Martínez et al., 2019; Gullifer et al., 2018) and the duration of engagement (Abutalebi & Green, 2016; Grundy et al., 2017; Pliatsikas, 2019b). As specified earlier, L2 proficiency is likely conditioned to both duration of L2 use and intensity/diversity of language use (Fig. 3), particularly past a specific point of attainment. Furthermore, maintenance of the L1 during this time frame will alter the latency and nature of adaptation towards efficiency. Based on existing theory (Calabria et al., 2018; Pliatsikas, 2019b) we argue that prolonged L1 maintenance would maintain the neural adaptations in language efficiency (increased subcortical and cerebellar use). Decreased L1 maintenance would require renewed executive control demands (fronto-cortical recruitment), specifically in situations where the L1 is required or used (Iverson & Miller, 2017).

References

  • DeLuca, V., Segaert, K., Mazaheri, A., Krott, A. 2020. Understanding bilingual brain function and structure changes? U bet! A unified bilingual experience trajectory model. Journal of Neurolinguistics 56: 100930