Following research that has demonstrated that certain life-experiences can shape cognition (e.g., enhanced spatial memory in taxi drivers in London, jugglers, and musicians [1
]), it seems intuitive that language, being one of the most intense and durable human life-experiences, could also enhance domain-general cognitive performance [3
]. Cognitive advantages for bilinguals have indeed been observed in studies comparing performance of bilinguals and monolinguals on a series of cognitive tasks that measure (components of) executive control, most notably inhibition. Building upon the influential model of executive control by Miyake and colleagues [4
] that distinguishes four components of executive functioning; inhibition, switching, monitoring and updating, the dominant view is that the enhanced cognitive performance of bilinguals is due to their continuous inhibition of the nontarget language in a specific context to resolve competition for selection, as both languages in a bilingual brain are always active [5
]. This continuous cognitively effortful task would carry over into non-linguistic cognitive tasks, making bilinguals respond faster to non-verbal cues in especially conflict-monitoring tasks—such as the Simon, Stroop or Flanker task—than monolinguals [6
A seminal paper by Bialystok and colleagues from 2004 demonstrated that there was a bilingual cognitive advantage (faster response on a Simon task) for older bilingual adults compared to their monolingual age peers and younger monolinguals [7
]. In 2007, Bialystok and colleagues reported that bilingual patients diagnosed with probable Alzheimer’s disease received this diagnosis on average four years later than their monolingual peers, whereas they performed on a par on measures of cognitive control and there were no interfering effects of occupational level, gender and immigration status [8
]. These findings collectively sparked a wealth of research on what has become known as the ‘bilingual advantage’ (BA), with varying results. Since 2004, the strength of a BA has decreased from strong to moderate effects for specific populations of bilinguals or no differences between bi- and monolinguals at all [9
]. Recent (critical) reflections on the existence of bilingual advantages (see the discussion article by Paap et al. [12
], and corresponding commentaries in Cortex) have given rise to calls to uncover more about the underlying constructs of language and cognitive control, in order to move our understanding of the differential results regarding a BA forward.
1.1. Cognitive Control
] observes in Cortex
that the research on BAs lacks clear theories on how language management influences cognitive control. He argues that we need information on the source domain (language control) and how this transfers to the target domain (cognitive control). Without being clear on the processes involved in the source and target domain, we cannot even begin to interpret the tranfer process. Currently, our main predictions on how bilingualism may impact cognitive performance is by the joint activation of languages in a bilingual brain, through which bilinguals exert enhanced control on processes of inhibition, monitoring or directing attention. This enhanced training carries over (it is unclear by which mechanism) into general processes of executive control.
However, there are a few problems with this view. First of all, the robustness of the inhibitory control account, whereby response inhibition was put forward as the driving force behind cognitive advantages, has been called into question by research on linguistic interference in (picture) naming. In a Dutch L1 picture naming experiment with a small group of university students, researchers found L2 English interference at the phonological level [14
], suggesting facilitation and interference effects of the non-target language [15
]. This goes against the notion that selection of the appropriate language is solely accounted for by the mechanism to inhibit the non-appropriate language (form), hence propelling researchers to argue that inhibition alone
cannot explain executive control advantages (e.g., [10
Even more, studies demonstrate that the cognitive tasks used in behavioural studies to elicit a BA do not necessarily correlate with each other, revealing (different combinations of) a multitude of cognitive processes measured by these tasks (see [18
]). Tasks that measure executive functions always tap into multiple components of cognitive control, creating ‘task impurity’ (see [19
]). This makes it challenging to relate specific components of executive functions directly to bilingualism. Perhaps precisely because of this complexity and opaqueness, very few studies so far have attempted to grasp the underlying cognitive processes involved in bilingual decision making. This echoes the argument of Hartsuiker that clear theoretical underpinnings of the target domain and the transfer process are lacking.
1.2. What Is Bilingualism?
On the side of the source domain, in turn, we may have to take one step back and consider what it is that we define as bilingualism. Bilingualism is not a static ‘state’. Treating bilingualism as a dichotomous variable; solely as the ability to speak more than one language (the initial operationalisation in the 2004 study), falls short on acknowledging the vast differences between bilingual groups, or indeed individuals. Ihle et al. [20
] tested whether the more languages an individual speaks, the better s/he performs on cognitive tasks, which would logically follow if the number of languages spoken directly relates to a BA. They conclude that, indeed, number of languages spoken contributes to cognitive reserve, yet not in all participants, and depending on other cognitively stimulating activities the participants engaged in, their verbal abilities in general, and basic cognitive processing speed. (See however Kave et al. [21
] who longitudinally followed a group of multilingual elders in Israel and observed that the number of languages spoken reflects better cognitive performance, independent of educational level). A pure ‘knowledge-based’ (do you know
multiple languages, yes/no) operationalisation of bilingualism therefore falls short of explaining differences in research towards a BA.
Perhaps the best way to solve some of the controversy in the BA debate is to stop comparing groups of mono- and bilinguals and rather pay closer attention to the type of bilingualism under investigation, treating this on a continuum based on bilingual language use rather than as a knowledge variable (also see [22
]). This is not a new idea, yet something that has perhaps been overshadowed in recent years by a stronger focus on group-comparisons and dichotomous categorisations. In the 1960s, Cooper [24
] demonstrated that Puerto-Rican speakers of English and Spanish in the USA performed differently on word naming and association tasks depending on how much they used each language in five societal domains; home, religion, neighbourhood, education and work. He advocates for more fine-grained operationalisations of bilingualism according to how the languages are used. Similarly, Grosjean [25
] argues for the inclusion of language modes (how long is a subject in a monolingual or bilingual mode, and how much switching takes place in this bilingual mode), language stability and language function, and a more detailed account of the language history of subjects in bilingualism research.
The interactional nature of bilingualism, which goes beyond the static notions of number of languages and simultaneity of acquisition, is captured in Green and Abutalebi’s [26
] adaptive control hypothesis (ACH). The ACH posits that different interactional contexts, either single-language, dual-language or dense code-switching contexts, place different cognitive demands on an individual (related to e.g., conflict monitoring, interference suppression and goal-orienting). Greater neural efficiency is expected to be observed for bilinguals who frequently reside in a dual-language context, as they show skills in monitoring (language) cues, allowing for more rapid switching. In dense code-switching contexts and single-language contexts, effective communication is less affected by careful background monitoring of language cues, suggesting fewer switching advantages for bilinguals in these contexts.
The differential contextual demands of diverse populations of bilinguals may be explanatory in understanding the presence or absence of bilingual advantages. An illustration of the ACH is a study by MacNamara and Conway [27
] towards bimodal bilinguals (cross-sectionally measured with two groups of college students; one with and one without two years of experience with American sign language). They looked at whether high Bilingual Management Demands (BMDs), operationalised as degree and frequency of switching between languages (cognitive control), and the experience with managing those demands is the mechanism responsible for cognitive enhancement. High BMD experience was associated with better performance on tasks of cognitive control and working memory capacity after two years. They suggest that rapid switching and the coordination of simultaneously comprehending and producing in two languages, which becomes more efficient with experience, enhances cognitive control. Because of the variation within bilingual populations with regard to BMD (type, experience and magnitude), the presence and absence of a bilingual advantage also varies, they argue.
Bilingual advantages may then have more to do with the domain of switching between languages. Indeed, studies have found that the degree with which bilingual switch languages is predictive of a bilingual advantage [28
]. However, language switching, especially when measured in isolation and under strict, artificial conditions (e.g., cued switching), is problematic from the view whereby language use is mostly interactional in nature. The question may be asked whether bilinguals indeed exert language control and inhibit one language when switching to the other, especially in conversational interactions where switching may occur mid-sentence, and the strict ‘boundaries’ that separate languages do not apply. Moreover, as de Bot (2017) aptly notes, switching is not unique to bilingual settings. Monolinguals may switch codes or registers depending on the situation. When switching is such a common phenomenon, both in bi- and monolinguals, can we than really assume that switching costs (even though evidence for lower switch costs for bilinguals is robust), or inhibition, is enough to drive a BA? (see [31
Perhaps it is not merely language control, but rather the environment in which language is used, possibly in addition to other cognitively enriching experiences, by which a cognitive advantage may be observed. Informative in this regard is a study on performance on EF tasks by bilinguals and two monolingual groups in different linguistic environments (French-dominant Quebec and English-French bilingual Ottawa). The language environment seemed to offer a more robust explanation for enhanced EF performance—by English monolinguals in Ottawa, where they are exposed to French in their environment, contrary to French monolinguals in Quebec, who lack this bilingual exposure—than bilingualism itself [32
The influence exerted by the environment on experiments is often overlooked. With an analogy of the temperature of boiling water—which depends on altitude—Bak argues that even when we control for differences in experimental settings, the selection of participants, and different methods of data analysis, the environment in which an experiment is conducted may result in different observations [33
]. As such, he advocates for the importance to compare results conducted in different environments, rather than replicating the same experiment in the same environment. As above, the context in which bilinguals use their languages (in a highly bilingual environment where switching is practised daily as opposed to a monolingual environment) may offer an explanation for some of the conflicting evidence for a BA.
Moreover, Bak argues that the attitudes towards bilingualism or certain languages in different environments (which may be more positive in highly bilingual populations such as in Brussels or more negative when language use is politically coloured) may additionally play a role in the conflicting evidence; believing that being bilingual is an asset rather than a disadvantage. A recent investigation into the different operationalistations of bilingualism in studies towards BAs between 2005 and 2015 [34
] revealed that the degree to which different characteristics of bilingualism are reported differs greatly. Moreover, there is also a lack of sociolinguistic information, which is of particular importance when viewing bilingualism as an interactive life experience. The authors advocate for a better documentation of the social context (usage and status of languages in the population), but also the quality of foreign language instruction, by which in some countries bilingualism is not a life experience, but a learning experience.
In a recent contribution to a special issue of Linguistic Approaches to Bilingualism, Valian [35
] argues that, as experiences accumulate over the lifespan, singling out executive function benefits belonging to certain experiences is increasingly difficult. Benefits of bilingualism may be additive, or additive up to a certain point or are only visible when they occur together with other benefits (e.g., from being physically active, playing a musical instrument, a specific diet, and so forth). The fact that there are no negative results reported for bilinguals (only positive or null results), leads Valian to propose that there is a benefit, but that this benefit competes with other benefits, hence showing positive results in some populations, but null results in others. Because individuals are so diverse in experiences, and tasks measure different aspects of executive functions, she argues that it is more likely that there are also different mechanisms by which executive functions may be enhanced.
When BA’s are so haphazardly observed, we might want to delve deeper into the construct of bilingualism. Bak [36
] speaks of a forest of confounding variables in research on bilingualism. Hence, rather than a factor in isolation, bilingualism may contribute to enhanced cognitive performance when it is viewed together with other experiential factors. The visibility of such an effect then perhaps depends on the presence and strength of these other factors (see [37
]). What we need are large-scale studies whereby bilingualism is present in many forms, and examine not only its static presence, but also consider language usage patterns, to be able to gauge whether there is indeed a benefit attributable to bilingualism, or whether BA’s lie in the combination of experiential factors. This study is one of the first attempts to investigate the nature of multilingualism and possible cognitive effects in a large and diverse group of older adults.
1.3. This Study
Moving away from a static, isolated and knowledge-based view of bilingualism, this paper details the observations of a study among a highly variable group of multilingual older adults in a small geographical area in the northern Netherlands, rich in dialects and languages. Given the contextual nature of bilingualism, it is informative to look at a bilingual population in an area where bilingual practices are widespread, but the degree of bilingual language use and type (two languages, dialect and standard language combinations, etc.) differ from person to person. We answer to the call to uncover more about the nature of bilingual advantages by investigating the different types of multilingualism (language combinations, usage intensities, social context, etc). By doing this in a population that is multilingual to varying degrees and in varying forms/manifestations, comparison to a monolingual population is not needed, and ineffective.
The main aim of this study is twofold:
To uncover more about if, and what aspect(s) of, multilingualism may facilitate enhancement of executive functions and what this tells us about the nature of multilingualism (as a knowledge- or experience-based variable) and the cognitive constructs that are involved;
Whether multilingualism can contribute to enhanced cognitive performance in the presence of other (known) ‘confounding’ factors relating to health, wellbeing, and quality of life.
For sake of continuity based on the persistent ideas of cognitive control in the literature, we assess inhibition, attention-direction and set-shifting processes in a diverse elderly multilingual population. For the first question, we expect that it is not the number of languages or degree of proficiency of individuals (the knowledge-based operationalisation) that enhances cognitive performance, but rather the intensities with which different multilinguals (who may differ in number of languages, proficiency and language combinations) use their languages in different contexts. For this we draw on the premise of the adaptive control hypothesis, in that we expect that a more balanced use of different languages across different social contexts elicits better (faster or more accurate) performance on cognitive tasks related to switching and attention, similar to the observations by MacNamara and Conway [27
]. If this is indeed the case, this may confirm the speculations that there are different BA’s for different populations, and that these populations likely mainly differ in language usage intensities.
However, to truly assess the uniqueness of bilingual populations, and whether populations with enhanced EF performance can effectively be discerned on the basis of their language usage the second question forms the basis of a model in which we insert the linguistic information of our participants, together with demographic, health and lifestyle information. We may be hard-pressed to find that speaking multiple languages is more effective for EF enhancement than other lifestyle factors, such as playing a musical instrument, and perhaps that only by putting these factors together in a model, a significant effect can be observed. Or there may be other, environmental factors, that offer a more ready explanation of enhanced EF performance. The great variety in our multilingual sample in terms of language experiences, usage intensities and other demographic, health and wellbeing characteristics offers a unique insight into how these factors may, or may not, interact with each other and perhaps, under specific circumstances, show an effect of multilingualism, in some type or form, on cognitive performance. At least it offers more insights into why some research does and other does not find bilingual advantages, while at the same time also demonstrating the complexity of studying one isolated factor (bilingualism) in a social context.
The current study analysed the performance of a diverse group of older adults with varying levels of multilingualism on two cognitive tasks relating to (most strongly) inhibition and attention (Flanker) and set-shifting (WCST). The novelty in this analysis lies in the fact that we assessed multilingualism along a continuum and with a dynamic ‘language usage’ operationalisation of multilingualism. In combination with information of factors that are known to enhance cognitive performance from previous studies (such as level of education, and playing a musical instrument) but also other health and lifestyle factors, we built a linear regression model in which we found that degree of contextual second language (L2) usage significantly impacted cognitive performance, but only in one cognitive task and not in another. Purely knowing different languages did not relate to enhanced performance, suggesting that it is not the ability to speak multiple languages, but the use of these different languages that may show small positive effects on cognition. A subsequent multivariate PLS regression model including background variables, language variables from the LEAP-Q, and personality questionnaire factors yielded a two-component solution where proficiency in second and third language and the usage of especially the second and third language across different social domains were predictive of Flanker performance. This underscores the earlier linear regression model mentioned above on the operationalisation of multilingualism, in which language usage variables are more predictive of cognitive advantages than merely ‘knowing’ different languages. We suggest that simply asking a person whether or not they are multilingual is not useful, but probing this more carefully and taking into account proficiency and social language usage patterns does appear to contribute to performance above and beyond the contribution of age, gender, education and income. In the following paragraphs, we critically review our results and answer the two questions that were asked in the beginning of this paper, thereby highlighting possible caveats and misinterpretations of the data, and listing the limitations of our research.
In line with our hypothesis, yet contrary to previous research by [20
], the regression models in Table 6
demonstrated that a traditional knowledge-based operationalisation of multilingualism (as number of languages, early vs. late onset of acquisition, language proficiency and type of language combinations) does not unequivocally lead to enhanced cognitive performance. As individuals differ on many levels, finding an effect of number of languages or degree of cognitive performance would have been strong evidence in favour of a general bilingual advantage. However, given the diverse nature of the study population, plus the uns evidence of a BA reported in the literature, it would have been highly unlikely that in such a heterogeneous population effects for such general observations of multilingualism would be found. As such, our findings are in line with a meta-analysis by Paap et al. [9
] of the positive effects of bilingualism on EF (which mostly emerge from small, underpowered studies), where the authors conclude that there are no systematic differences between bi- and monolinguals when regarding these generic factors (early/late, balanced/unbalanced). Indeed, as becomes evident in the PLS regression, there are other, individually distinct factors that covary with measures of multilingualism, such as certain personality traits and higher levels of education.
It may be that so-called ‘confounding variables’ mask the effect of bilingualism on EF (see [36
]), or, rather more likely when observing the small effect sizes, help in detecting an effect of language. Perhaps, as some authors suggest, it is by virtue of a number of factors which covary with bilingualism that enhanced EF for multilinguals may be observed. Previous research has attested that factors such as level of education or immigrant status (the ‘healthy migrant effect’) enhance bilingual performance, especially when set-off against a lower-educated or non-immigrant monolingual group. Our PLS regression model demonstrated that multilingualism indeed covaries with a number of other experiential factors in explaining enhanced cognitive performance. When consider the first two components of the Flanker PLS regression, a high quality of life, character traits relating to being open to new experiences and degree of switching languages according to the context covary with the degree of usage of the L2 across different social domains. Proficiency in the second and third language co-occur with extravertness and agreeableness. Together they explain almost 10% of the variance in the Flanker model, which is marginal but noteworthy in a model with such a diverse collection of variables. Although we cannot speculate on the size of the social network of the individuals, scoring high on openness to new experiences might well be an indication of the presence and positive values of social relationships and engaging in social networks. Perhaps those people who use their L2 also in different social domains display a more diverse social network, helped established in part by their personality traits and positive outlook on life. Higher quality of life might imply that people are relatively mobile and/or can more easily maintain their social network. Moreover, the high score of education might reflect a better cognitive disposition from the outset, as enhanced executive functions are often found to be related to educational level.
This is related to the second component, where we observe covariance of L2 proficiency level, age of onset of acquisition of the L1 and extravertness. The contrast between the two components explain different mechanisms of enhanced cognitive performance. Individuals may use their experiences with monitoring language cues, which enhances their attentional control system, through using languages across different social domains. Alternatively, their extensive training in developing language proficiency aids the degree of inhibiting the non-target language. Contrastively, the number of languages one speaks has very low predictive power. This underscores the contextual nature of multilingualism; language use and proficiency, in relation to quality of life and personality variables predicts cognitive performance. This lines up with the argument that other experiential factors may in combination with bilingualism collectively contribute to enhanced cognitive performance [37
], but differently in different circumstances.
Of course there are constraints on the extent to which these variables do actually contribute to enhanced cognitive performance. Moreover, we have presumably tapped into a very specific population sample. The descriptive statistics display that the group was on average highly educated, had a relatively high income, experienced high qualities of life and many of them were early bilinguals. This underscores Bak’s [33
] argument that taking note of the environment in which experiments are conducted is vital. As participation was voluntary, we have perhaps attracted mostly those people who already find this type of research topic interesting. (We will come back to the issue of self-selection bias below).
Although our sample consisted of non-immigrant, native-born northern Dutch multilingual older adults, the relative homogeneity of the sample at face-value is distorted when looking more closely at the characteristics of not only health status, wellbeing or personality, but especially degree, type and intensity of multilingualism.
In such a vein, we find differences in the dynamic operationalisation of multilingualism, as degree and intensity of multilingual language usage. What is more, these differences do highlight enhanced performance for some multilinguals on one cognitive task (the Flanker). This result points to the notion that multilingualism is an individually distinct/varying concept, and converges with the observation of those studies that find advantages only for specific populations of bi-/multilinguals or under specific conditions. In our study, especially for the Flanker effect score, those participants who report to use more than one language across different social contexts (with family, friends, neighbours and acquaintances) and especially the L2, and who furthermore switch between languages depending on the social context, show smaller Flanker effect scores. This suggests faster performance on the incongruent trials in comparison to those participants who have a clear usage-preference for one language and/or do not switch in different contexts.
This observation aligns with the research on language balance (for bimodal bilinguals in [27
], Yow et al. [63
] in younger populations and Houtzager [64
] in older populations) and may be explained by the adaptive control hypothesis in that more intense usage of different languages, especially in a dual-language context where bilinguals constantly monitor language cues (and thus focus attention), confers cognitive benefits. It is especially this dual-language context, rather than a dense code-switching context, that incurs benefits in our sample. After all, it is only in interaction with contextual switching that the use of the second language in different social domains leads to faster Flanker performance (in our linear regression model). Without this interaction, the use of the second language across domains yields higher Flanker effect scores, suggesting that there needs to be some element of control/monitoring of attention to language cues that is present in this dual-language mode and which carries over into more general cognitive processes. That the linear model on the WCST error score with the same factors does not reach significance, may give us some insight into the cognitive processes that are at work for this group of multilinguals.
In the WCST, participants have to actively inhibit the old rule in favour of the new one. This switching and inhibition mechanism would, in our understanding of the EF involved in a BA, elicit more accurate performance for those bilinguals who frequently use different languages in different contexts. The absence of an effect here is extra evidence that it is not inhibition per se that drives a potential bilingual advantage, but rather more general attention-orienting behaviour. Bilinguals who use their languages frequently across different social contexts are thus not necessarily better switchers or inhibitors, but may demonstrate an enhanced attention-orienting mechanism.
Evidence for this comes from a study towards the mechanisms underlying the Flanker task, conducted by Ong et al. [65
], using diffusion modeling. With this technique, they calculated whether faster response times on the Flanker items are the result of suppressing conflicting information or enhanced attentional control. They found that the bilingual group of older adults showed shorter non-decision times than the monolingual group on incongruent Flanker items, which suggests an enhanced processing efficiency when faced with distracting information. Both groups performed on a par on the other diffusion modeling measures which relate to inhibition, which led the authors to conclude that a BA on the Flanker task is the result of an enhanced attentional control.
Recent theoretical re-evaluations of the processes at work during bilingual decision-making too suggest that the advantage for bilinguals may be in the more general domain of attention selection [3
]. Underlying the selection of the appropriate language is the constant monitoring of conflict, which may be part of a more general, non-verbal attention or selection system, Bialystok argues ([3
], p. 235). Nonetheless, the WCST also involves focusing attention, yet not in the immediate presence of ‘noise’ or distracting information. Cautious interpretation of the found results is thus warranted as long as we are unclear on what the cognitive tasks actually tap into (also see below).
Moreover, the interaction effect may also be observed as a result of the specific linguistic environment. Recall the study on bilinguals in French-dominant Quebec and English-French bilingual Ottawa, where the linguistic environment proved to be a more robust explanation for enhanced executive control than bilingualism in itself [32
]. As exposure to different languages in our group is also relatively high—the Frisian language and Groninger and Drents dialects are omnipresent in the provinces—it may well be that we observe not an effect of social language switching, but simply an effect that culminates from the ability to use the different languages in the immediate environment. This observation underscores the importance of reporting on the social norms and status of different languages in populations, as pointed out by Surrain and Luk [34
The PLS regression that we computed to answer the question of whether multilingualism could have any explanatory power regarding better cognitive performance iterates the linear model on the operationalisation of multilingualism in that more crude measures such as number of languages do not predict much. Rather, more sensitives measures relating to language usage and proficiency combine with wellbeing and personality traits, and are together informative of cognitive performance. The PLS regression technique limits the degree of ‘gold-digging’ in the data to find patterns that in hindsight can be explained by existing theories, something that Hartsuiker rightfully warns for [13
]. Nonetheless, even with such a diverse sample of variables the overall explanatory power of both models is pretty low. This underscores the importance to establish a clear theory of the relation between language and cognitive performance. Our findings regarding the operationalisations of multilingualism above may be especially enlightening for this purpose when considering how language control modulates cognitive control.
Beyond this finding, however, there are still gaps in the knowledge pertaining to exactly how language engages executive functions. In line with De Bruin and Della-Sala [66
], we advocate that comparing groups of bi- and monolinguals should not form the foundation of such a framework, but rather the characteristics of bilingualism, especially language usage patterns, following the adaptive control hypothesis. However, we are still unsure as to how this task-specific executive control (either inhibition, attention direction, or some other mechanism) transfers to more broader domains of cognitive control, and whether a Flanker task and a WCST aptly measure these control processes (also see [67
]). Nonetheless, the observation that multilinguals who use their languages often and in different contexts demonstrate slightly enhanced performance on the Flanker task, but not the WCST suggests that language use, or the multilingual linguistic environment are contributing factors in enhanced attention-direction. Whether this transfers to other executive functions or cognitive domains cannot be ascertained, given the overall low significance of the linear regression models.
There are a number of other explanations for the found (absence of) effects of multilingualism on EF that warrant a cautious interpretation of the found results. The absence of an effect of the ‘knowledge’ variables associated with multilingualism may have emerged because the majority of the participants report the maximum number of five languages they could list. These are not only the languages and/or dialects they use frequently, but also the languages that they have learned in school but which they do not use as productively. This results in a population that is already highly multilingual in terms of number of languages. Self-reported proficiency may also have biased the data. In addition, it is likely that proficiency in language 2 and 3 is higher in general than in an average population, as those people who responded to the questionnaire may well be ‘advanced’ speakers of multiple languages, resulting in a self-selection bias (also see below).
Given the subjectivity with which we measure proficiency we do not rule out an effect of proficiency found in other studies. However, measuring proficiency is notoriously difficult and interpretation in degree of bi- or multilingualism is inherently flawed [68
]. Can there be a ‘cut off’ point at which someone with a certain level of proficiency is considered more multilingual than others? The absence of an effect of proficiency in this study, we therefore argue, can most solidly be explained by the lack of variation/differentiation in proficiency scores.
Self-reports come with inherent shortcomings. Participants may be more or less optimistic in judging their language abilities, also depending on their personal motivation to participate in the study. A self-selected test population is always biased. The majority of the participants likely participated because they consider themselves multilingual language users and have a positive attitude towards using their languages (which we explicitly asked in the questionnaire). They may be proud speakers of minority languages and/or dialects. By approaching multilingualism from an ‘inclusive’ perspective, meaning that it is more important whether someone speaks/uses multiple languages than their relative degree of proficiency in these languages, some participants may have listed their knowledge of languages in which they have a very basic proficiency and which they sparsely use, whereas others have listed only those languages which they frequently use. These differential interpretations we have tried to keep in check by being as detailed as possible in our questionnaire and asking not only questions on knowledge of languages but also usage patterns. Furthermore, as referenced in the materials section, participants completed the questionnaire in an online environment, which already selects those participants with computer skills. A small minority of the participants completed the questionnaire by means of a face-to-face interview. Because we have tried to match the conditions of the interview procedure as closely as possible with the online questionnaire, and because the ‘interview participant’ sample was so small, we do not expect any qualitative differences in the results to occur because of these different modes of data collection. Nonetheless, the reliability of the data can only be warranted by performing replication studies with different populations.