2.1. The frequency issue: quantity versus quality

Combining languages in one utterance can be motivated by different factors. It can help the speaker to fill lexical gaps, demonstrate group identity or manage bilingual interaction. But this is not what CS refers to in cognitive research. CS is regarded as a complex process having a pivotal impact on the executive control system. However, a common problem in the study of the relationship between CS and the executive system seems to be that the CS behaviour of the participants is underspecified. In the reviewed studies, we observed a broad and inconsistent labelling of CS types (e.g., Dense CS or Congruent Lexicalization or Opportunistic Planning for cases of rapid clause internal switching, Insertion, Alternation). However, due to the heterogeneous understanding of CS and, related thereto, the divergence of assessment of the CS types (if assessed), it is problematic to conjecture whether the CS types are comparable across the studies. Since the processing of any kind of linguistic input or the process of language production unfolds by coordinating multiple cognitive sources, (linked to distinct cognitive functions), the exact switch patterns, namely when, what and how individuals are switching, are critical. On the other hand, the switched utterances must be processed and comprehended by the interlocutor, inducing different processing mechanisms compared to those of the speaker (for a more detailed argumentation, see Section 2.3.).

In the reviewed studies, CS types are supposed to induce different aspects of cognitive control. One aspect of cognitive control is associated with the level of inhibition employed. The Inhibitory Control Continuum Hypothesis (Treffers-Daller et al., Reference Treffers-Daller, Bullock and Toribio2009) implies that the languages are supposed to be most separate in alternation and least separate in Dense CS. During inserting items from the other language both languages are supposed to be active (Treffers-Dalleret al., Reference Treffers-Daller, Bullock and Toribio2009). Treffers-Daller et al. (Reference Treffers-Daller, Bullock and Toribio2009) further suggested that inhibition enables language separation and that in order to alternate between the languages the level of inhibition must be very high. Moreover, Insertion employs a lower level of inhibition compared to Alternation whereas Dense CS engages minimal levels of inhibition.

Hofweber et al. (Reference Hofweber, Marinis and Treffers-Daller2020b) investigated the effects of experimentally induced language modes (single language versus CS modes) on bilinguals’ EF performance (Flanker task) and referred to the terms Coupled control code-switching (Green and Wei, Reference Green and Wei2014) and dual mechanism framework (see Braver, Reference Braver2012). Green and Wei (Reference Green and Wei2014) suggested that the CS types Insertion and Alternation can be realized by Coupled control under which the matrix language temporarily cedes control to the other language and allows for the intended Insertion or Alternation before control is returned back (see also Green, Reference Green2018). The dual mechanism framework, on the other hand, postulates that “proactive control reflects the sustained and anticipatory maintenance of goal-relevant information” and “reactive control reflects transient stimulus-driven goal reactivation” (p.106). Based on these assumptions, Hofweber et al. (Reference Hofweber, Marinis and Treffers-Daller2020b) stated that their results confirmed the predicted interactions between CS and EFs, namely that Coupled control code-switching (Alternation, Insertion) was predicted to train reactive control modes and Dense CS was predicted to train proactive control modes. They explained the proactive control mode in Dense CS by stating that “the prolonged linguistic co-activation in this CS type requires a constant readiness to use inhibition.” (p.923). Moreover, Hofweber et al. translated the training effect of CS into bilingual advantages. On the contrary to these results, Jylkkä et al. (Reference Jylkkä, Laine and Lehtonen2021) reported that more frequent everyday language switching, irrespective of type, was associated with worse performance in a cued naming task and a Simon and Flanker task. The researchers argued that participants with lower executive functioning had the tendency to switch more often and concluded that the hypothesis that training language switching can train EFs is questionable since everyday language switching is an automatized process that does not require top-down control (Jylkkä et al., Reference Jylkkä, Laine and Lehtonen2021; for a review see also Lehtonen et. al., Reference Lehtonen, Fyndanis and Jylkkä2023).

By referring to only a very few recent studies in the previous paragraph (details about other studies are displayed in Table 1) we wanted to emphasize how different the argumentation regarding the relation between CS and EFs (here inhibition) can be. Furthermore, it is not easy for the reader to disentangle the interdependence of the specific variables due to overgeneralized descriptions of bilingual language use. Hofweber et. al. (Reference Hofweber, Marinis and Treffers-Daller2020b) stated that, given their results, Alternation and Insertion train reactive control modes and Dense CS trains proactive control modes. The participants’ CS behaviour was assessed according to how they judged themselves when “encountering” switched utterances. Therefore, it remains unclear whether these control modes are trained when encountering code-switches or when producing them. Similarly, it is not conceivable whether the constant readiness to use inhibition in the Dense CS type is valid when encountering or producing switches. A further important question that arises here is what is causing the “training” effect. Is it the Alternation between the languages per se or does the content of the switched elements make a difference. Hence, in the following we want to explain the importance of the content of switches in relation to EFs.

The results of the reviewed studies, which are displayed in Table 1, are showing that the CS type might play a decisive role in how cognitive processes are affected. However, in the reviewed literature, the CS type is mostly categorized based on the frequency of switches by applying a self-assessment questionnaire (see Table 1). If it comes to the methodology, it gets even more problematic. The frequency of switches is evaluated often only by a questionnaire where participants are told to rate the switches they produce or encounter in different situations (see de Bruin, Reference de Bruin2019, for more details about assessment questionnaires on language use). Thus, a series of questions arise when self-assessed switch frequency is taken as the criterion for classifying CS behaviour.

• - What is counted as a switch by the participant?

• - Does the participant perceive switching in the same way as the experimenter?

• - Is the participant able to make a distinction between borrowings or buzzwords and switches?

• - How can the participant determine an average number of the switches produced per day?

• - How is the participant able to rate how common it is for her/him to switch in specific contexts? (for a CS Assessment Questionnaire, see The Bilingual Code-Switching Profile – Olson, Reference Olson2022; see also Section 2.3.)

• - What about the switches at the level of (morpho-)syntax? How can these be counted? (for switches at the level of (morpho-)syntax, see examples 1–3 in the Introduction)

• - How can the participant differentiate between intra-sentential (e.g., Insertion) and inter-sentential (e.g., Alternation) CS? This is important since different modes of language control might be involved (e.g., Green & Abutalebi, Reference Green and Abutalebi2013; Green & Wei, Reference Green and Wei2014).

In comparison to a conceptualization of Dense CS as a frequent switch between the languages, Hofweber et al. (Reference Hofweber, Marinis and Treffers-Daller2020b) elucidated the term Dense CS in a more qualitative manner. They stated that “Dense code-switching describes cases of language mixing in which languages remain co-activated at both the grammatical and lexical levels. This means that speakers select lexical items and structural features based on their socio-pragmatic and structural appropriateness, not based on language membership. Switching takes place more frequently in Dense CS than in Alternation, but switch points are hard to identify. Contrary to Insertion, it is impossible to identify a Matrix language as speakers combine structural rules from both languages. Dense CS primarily occurs between closely related languages.” (p. 910).

Although this qualitative explanation of Dense CS is much more differentiated than in most of the other studies, we want to draw attention to a further point that seems to be neglected in most of the reviewed studies, namely the context where bilinguals might also switch between their languages in a very “dense” (in the sense of frequent) manner. Dense switching can also occur at schools/universities, where a foreign language is the medium of instruction. Other contexts where switching might be “dense” are international conferences and meetings as well as interpreting and translating. In a context where switching is very dense but challenging (e.g., interpreters, bilingual academic discussions, bilingual schools, etc.), the engagement of a high focus of attention would be expected in order to activate the specific item(s) which the speaker consciously selects as the most fitting ones in that specific utterance. On the other hand, no high focus of attention would be required when specific linguistic items are switched frequently in everyday communication, like communication between friends or family members, and are therefore mostly already active, if not to say, automatized. That means that even if the frequency (quantity) of the switched items is similar in two persons that are categorized as “Dense Code-Switcher”, the switched elements and content (or “quality” as the counterpart of quantity) should also be considered.

Cognitive mechanisms during the course of conscious selection of the “to be switched element” or the conscious selection of the “most fitting expression” in the other language do entail different control mechanisms compared to the utterance of items that “come first” or “automatically”. In other words, we put forward the assumption that the CS types Insertion and Alternation could both differ in terms of cognitive control related to the context and content they are used in. Both CS types are commonly produced in bilingual personal conversations as well as in bilingual professional contexts. We think that the main difference is that “automatically” produced switches in everyday communication do not tap into the EFs like monitoring, conflict resolution, attention and inhibition to the same extent as switches that are deliberately performed at a specific moment. Switches that are produced consciously do require a higher level of monitoring in order to find the most accurate equivalent in the other language. For instance, interpreters are required to densely alternate between the languages by using the most accurate expressions. In case more than one appropriate item in the mental lexicon is activated, the most fitting one has to be selected, which entails the resolution of a conflict as a cognitive function. All other closely related expressions have to be inhibited in a very rapid manner. It is conceivable that this intentional switching process requires a high focus of attention and monitoring. On the other hand, when switching in everyday communication, the need to focus attention to the switched elements and to monitor the utterance is supposed to be lower as in the first situation since the content and the time of the switch are not considered as inducing a conflict at that moment. That means that the most active items, irrespective of the language, can be straightforwardly uttered (this also applies to Dense CS). A further point, that we would like to underline, is that the CS type does not describe the type of how a bilingual is permanently using both languages. Rather we suppose that it is mostly the description of the momentary switching process. Hence, we concur with the assumption of Green and Abutalebi (Reference Green, Abutalebi, Bayram and Kras2023) that the states of different executive control processes are adjusting to the CS type, the CS context and the languages per se, even within moments, depending on the needs of the bilingual speaker or listener.

In short, it can be said that Dense CS is far more than switching frequently between two languages and cannot be reduced to a quantitative phenomenon. The bilingual context and content (i.e., academic discussions versus everyday conversations) and, very importantly, the intention of the bilingual speaker (deliberate versus free switching) seem to be crucial factors regulating the underlying cognitive functions. These functions are also susceptive to the level of complexity of the content (for the adjustment of attentional states to changes in content and context, see Green and Abutalebi, Reference Green, Abutalebi, Bayram and Kras2023). In studies investigating the interrelationship between CS and cognitive functions, not only the “quantity” but also the “quality”, as described above, should therefore be considered when assessing the CS type.

2.2. The “proficiency and organization of the mental lexicon” issue

In communities where two languages are used frequently, it is expected that in daily life switching between the languages is voluntary and free. One may say that, in conversations where switching is common and occurs naturally, this switching becomes a generalized ability facilitated by frequent use. Consequently, CS has often been considered fluent and smooth (Poplack, Reference Poplack1980) and described as the “most comfortable way” to speak (Dorleijn, Reference Dorleijn2017, p. 12). However, experimental studies measuring response times or switch costs during involuntary switches often report longer response times and larger switch costs (e.g., Burkholder, Reference Burkholder2018; de Bruin et al., Reference de Bruin, Samuel and Duñabeitia2018; Jevtović et al., Reference Jevtović, Duñabeitia and de Bruin2020; among many others) compared to voluntary switches. Moreover, Gross and Kaushanskaya (Reference Gross and Kaushanskaya2015) found that voluntary language switching in children can be less costly even when naming items in the non-dominant language, depending on the frequency of use. This suggests that reaction time is influenced not only by whether a switch is voluntary or involuntary but also by language proficiency, which is a critical factor affecting cognitive costs during the switching process and should be evaluated in connection with the interactional context.

A further factor modulating EFs during lexical selection is the organization of languages in the mental lexicon. In the remainder of the section, we examine the effects of age of acquisition and language proficiency on CS, as these factors influence the organization of languages in the mental lexicon. To gain a broader perspective, we include relevant electrophysiological studies on the processing of code-switches.

In one of the first electrophysiological studies investigating neural responses to switched items during online sentence processing in proficient speakers and language learners, Moreno et al. (Reference Moreno, Federmeier and Kutas2002) found that CS can be less costly in the non-dominant language. They interpreted switching as an unexpected event processed similarly to low-cloze words, suggesting that the processing of a CS may be less costly than an unexpected within-language item. In contrast, van der Meij et al. (Reference Van Der Meij, Cuetos, Carreiras and Barber2011) suggested that proficiency modulates the processes triggered by language switches. Their EEG study indicated that higher L2 proficiency leads to greater involvement of L2 grammar in the processing of language switching in the group of more proficient learners compared to those with lower levels of L2 competence.

Contrary to these findings, Proverbio et al. (Reference Proverbio, Leoni and Zani2004) did not relate the difference in L1/L2 switching to proficiency differences. According to the findings in their EEG study with simultaneous interpreters and monolingual controls during a semantic processing task, Proverbio et al. (Reference Proverbio, Leoni and Zani2004) concluded that L2 neurofunctional organization depends on age of acquisition. They suggested that when L2 is learned later in life, it is more likely based on the translation of pre-existing lexical knowledge, whereas early acquisition results in overlapping semantic and conceptual systems. Similarly, Gosselin and Sabourin (Reference Gosselin and Sabourin2021) pointed out that CS can lead to processing costs for some but not for all bilinguals, likely due to differences in participants’ experience (cf. Beatty-Martínez & Dussias, Reference Beatty-Martínez and Dussias2017; Beatty-Martínez et al., Reference Beatty-Martínez, Valdés Kroff and Dussias2018). Verreyt et al. (Reference Verreyt, Woumans, Vandelanotte, Szmalec and Duyck2016) also suggested that language switching experience is crucial for executive control in adults and that language usage and proficiency differentially affect bilingual cognition.

In the first phases of child’s language acquisition, input is not only restricted to the passively perceived language. The way how children use their language(s) has also an influence on how children process any new language (see Lieven et al., Reference Lieven, Salomo and Tomasello2009). The acknowledgement of bilingual speakers’ social and contextual environments is critical, since it determines the repeated use of specific languages and structures in different social contexts (Grosjean, Reference Grosjean2015). Korenar et al. (Reference Korenar, Treffers-Daller and Pliatsikas2023) suggested that the repeated use of specific language forms and specific languages might be stored in memory as more salient in that context. To exemplify this supposition, Korenar et al. gave the example of a tourist guide who intensely uses English with the context of guided tours but rarely anywhere else. The authors argue that this guide will strongly associate English with specific vocabulary about architecture and history, and with talking to tourists. It can be inferred that interactional contexts and how the languages are used in these contexts (characteristic practices in bilingual communities) can therefore shape how and when children switch between the languages they know (for a review see Gaskins et al., Reference Gaskins, Quick, Verschik and Backus2022). Accordingly, it can be assumed that often recurring switched forms or structures, which the child encounters in her/his environment might be stored as a switch. To give an example here: a typical Turkish–German switch occurs in a structure V_infinitive in German + V_finite in Turkish, such as in “malen yapıyorum” (“to draw/German I am doing/Turkish”). This form can be used with all verbs that express activities “doing sth.” and is very common in the Turkish Community living in Europe (for a Turkish–German Code-Switching Corpus, see Çetinoğlu, Reference Çetinoğlu2016¸ p. 4216). We consider this kind of insertion as noteworthy, since a structure where two full verbs are combined with the finite verb following the non-finite verb does not exist as a construction in either of these two languages. In other words, the subsequent use of two verbs, namely the infinitive form of an action verb (i.e., malen/to draw – German) followed by the finite verb “to do” (yapmak/etmek – Turkish) is not possible in both languages albeit for different reasons. Therefore, the only explanation of the widespread usage of this structure can be that it is acquired and stored as a “switch-construction”. Therefore, we assume that when uttering these kinds of frequently produced switches, only a low level of monitoring and attention is engaged. In their recent review Lehtonen et al. (Reference Lehtonen, Fyndanis and Jylkkä2023), for instance, indicated that with increasing CS practice, bilingual language behaviours might become more automatic and less executively demanding since the cognitive system adapts through automatization. The authors further asserted that the engagement of domain-general executive resources diminishes over time, leading to an establishment of automatic task schemas in long-term memory that are not challenging EFs (Lehtonen et al., Reference Lehtonen, Fyndanis and Jylkkä2023).

Overall, it can be said that the interaction of age of onset of bilingual language acquisition, purpose of language use, linguistic structures, content and bilinguals’ sociolinguistic context are all components constituting the intricate and interwoven linkages shaping the mental lexicon. Hence, we presume that the bilingual mental lexicon is not organized according to language proficiency and not according to the language systems, but rather according to how the languages are used and to what extent the switches are automatized. As aforementioned, the automatized production of frequently used switches might indicate that these are actually stored as a switch or a “switch template” (in the sense of the construction, like “malen yapıyorum” as explained above). In other words, we assume that often recurring switches in a bilingual community cannot be stored separately. This would overstrain cognitive resources. There is also the fact that most switches are individual. Hence, we hypothesize that often-used switches are stored as “switch templates” which can be modified according to the need of the bilingual speaker.

On the other hand, a bilingual’s frequent and effortless switching abilities in daily communication might differ when this same person is involved in a bilingual context (same person, same languages) where the subject of conversation is not common for this person. A bilingual person might be very fluently switching when talking to family members or friends about common issues of daily activities, whereas when trying to explain a “special circumstance”, CS would probably not be as effortless or automatic compared to “routine contents”. Since accessing or activating less frequently used items takes longer compared to often used ones, it is not likely that a switch can be produced effortlessly and is therefore supposed to include conscious switching between the languages. The effortless/automatic production of frequently produced switches can be counted as a further argument that supports the idea that these are stored as “switch templates” in the mental lexicon. It should be noted that we do not suggest here that automatized code-switches are produced unconsciously. We think that, irrespective of containing a switch or not, each uttered lexical item is monitored before articulation which, of course, also requires attention. However, the level of the executive control functions is adapting to the degree of ease/difficulty of the lexical selection process.

In sum, future studies investigating the relation between CS, proficiency, processing costs and cognitive functions should also assess in detail to what extent the switched elements might be automatized, since these might be stored as such and would not tax EFs in the same level compared to intentionally produced switches.

2.3. The speaker–listener issue

Besides the issues mentioned above, CS must be regarded from the speaker’s and listener’s perspective. When planning an utterance, it is the speaker’s individual choice how to make use of the available linguistic knowledge and accordingly when and which linguistic item(s) to switch. In other words, it is an individual planning process to accomplish a linguistic goal in the most expedient way. On the other hand, it is unclear up to now, in how far the listener is able to anticipate when a switch might occur, even in a context where switching is part of the way of communication. In current models of cognitive control, it is assumed that different CS types have different effects on aspects of the executive system. However, due to different language processes during production and comprehension, it is obvious that the effect on the executive system should also differ. Therefore, in studying the influence of CS on cognitive functions it is not sufficient to only consider the type of CS. These should further be subdivided according to whether a switch is being produced or encountered (see also Section 2.1). This crucial point has been mostly neglected so far. As can be seen in Table 1, in most of the studies, a self-assessment questionnaire is used in order to determine the switching type of the participants. These questionnaires are mainly based on determining the history of language switching, the use of language switching, the ease of language switching and the attitudes towards language switching (e.g., The Bilingual Code-Switching Profile, Olson (Reference Olson2022)). These kinds of surveys are mostly assessing CS production and do this in a very limited manner only. However, it is not evident whether the EF tasks used in the reviewed studies (Flanker, Stroop, Simon, ANT, etc., for the details see Table 1) tap into the associated cognitive constructs.

In Section 2.1, we suggested that the CS types could differ in terms of cognitive control related to the context and content. In this section, we aim to extend this assumption by exemplifying how the executive control processes might differ related to whether a switch is produced or perceived. When producing a switch, the intention and the speech planning process of the bilingual speaker are crucial. In other words, when the speaker in a bilingual context intends to make use of the other language, only when there is a need for it (Insertion or Alternation), it is likely that the activation threshold of the language in use is lower, and the other language has to be inhibited. As proposed by Green and Wei (Reference Green and Wei2014), the matrix language temporarily cedes control to the other language and allows the intended Insertion or Alternation before control is returned back. In such a situation, it is probable that the EFs inhibition, conflict resolution, monitoring and attention are highly entailed in order to monitor where a switch is needed and how to switch appropriately. However, for the listener, the same EFs might at least not be engaged or not to the same extent. Moreover, we would suppose that the listener’s cognitive flexibility functions should be more entailed since the listener does not know when and what type of switch to expect. The listener is therefore in a state where rapid adaptation is required. We, therefore, argue that the speaker’s and listener’s major EFs engaged in a specific situation might be different or at least not engaged to the same extent.

Hence, we suggest that the CS behaviour should be assessed in an ecologically valid manner, such as natural conversations, question and answer sessions, or descriptions, where production and comprehension of switches are clearly differentiated. Furthermore, future studies should consider the specific cognitive functions relevant to their study context to identify appropriate tasks for assessing each language modality (comprehension versus production). It is important to use EF tasks with robust psychometric properties to accurately assess domain-general cognitive abilities within the context of CS. This is crucial since otherwise the EFs tested in a study might not tap into the cognitive constructs of language production or comprehension, respectively.

2.4. The (socio)linguistic issue

In almost all reported studies, CS behaviour is taken to be a stable measurable factor across different contexts. The respective values are correlated with results from cognitive experiments. Given our knowledge from linguistic and sociolinguistic studies on CS, we see the need to capture linguistic behaviour in a far more differentiated way in order to allow for valid conclusions on the interrelation between CS and cognitive functioning.

We already touched on this briefly at the beginning. Previous linguistic studies looked at the patterns of mixed language under structural and functional aspects, resulting in a fine-grained description and explanation. The following insights are relevant.