The psychological and linguistic profiles of self-reported code-switchers

Abstract

Aims and objectives/purpose/research questions:

The study considers the relationships between multilinguals’ linguistic history, current linguistic practices, personality dimensions and self-reported frequency of code-switching (CS) in interactions with four types of interlocutors. It looks at variation in CS according to interlocutor type and the link with all the independent variables.

Design/methodology/approach:

Data was collected through an online questionnaire with closed questions on use of CS with friends, family members, colleagues and strangers.

Data and analysis:

A total of 298 adult multilingual participants of various nationalities provided scores on Likert scales. Statistical analyses were carried out to establish relationships between independent variables and CS.

Findings/conclusions:

The amount of self-reported CS differs significantly according to the type of interlocutor. A high level of multilingualism and early onset of bilingualism were linked with significantly more CS. High levels of Openmindedness and low levels of Flexibility were linked with significantly more frequent self-reported CS with certain types of interlocutors. Sex, age and education had no – or very little – effect on self-reported frequency of CS.

Originality:

The study confirms earlier research that considerable intra-individual variation exists in self-reported CS with various interlocutors. It shows that not just the linguistic but also the psychological profiles of participants determine the amount of self-reported CS.

Significance/implications:

Both linguistic and psychological variables need to be taken into account in the study of sociolinguistic phenomena.

Limitations:

Reporting frequency of CS with different categories of interlocutors requires a certain degree of metalinguistic awareness and some standard error cannot be excluded in self-reports.

Keywords

code-switching individual differences personality traits multilingualism intra-individual variation inter-individual variation

Introduction

Malala Yousafzai describes the moving scene when her father Ziauddin – with whom she always uses Urdu – arrived at her hospital bed after she had been shot by a Taliban fighter: ‘“My daughter, you are my brave daughter, my beautiful daughter”, he said over and over, kissing my forehead and cheeks and nose. He didn’t know why he was speaking to me in English. I think somehow I knew he was there even though my eyes were closed. My father said later “I can’t explain it. I felt she responded”’ (Yousafzai & Lamb, 2013, p. 208). Although father and daughter knew English, they did not use this language between themselves, and the code-switch was remarkable enough for those witnessing the scene to notice and to tell Malala later. Apparently, Ziauddin later reflected on his unusual code-switching (CS) without being able to explain it. The phenomenon is not unknown, as sudden bursts of emotion are linked to unexpected CS among multilinguals (Dewaele, 2010). Ziauddin’s CS to English is an occurrence of free variation (i.e. non-systematic): an individual deviating from the habitual language choice with a known interlocutor in a given situation. We can assume that he engaged in CS with other interlocutors, including foreign politicians, in his activities as campaigner. One possible focus could be intra-individual variation in Ziauddin’s CS: are there interlocutors or situations where he engages more or less frequently in CS? A slightly different perspective would be to compare Ziauddin’s amount of CS with other multilinguals’ frequency of CS: researchers focusing on inter-individual variation try to uncover the reasons why some multilinguals use CS more frequently than others.

Rodriguez-Fornells, Krämer, Lorenzo-Seva, Festman and Münte (2012) observed that while many researchers have worked on CS, defined as ‘changes from one language to another in the course of conversation’ (Li, 2007, p. 14) in the last decennia, very little attention has been paid to individual differences. Reflecting on this observation, Dewaele and Li (2014a) wondered whether this lack of systematic research into individual differences in CS could be linked to the type of design used by researchers. Data on CS can be collected from a relatively large number of participants through experimental manipulation in psycholinguistic labs (Rodriguez-Fornells et al., 2012), or through self-report in questionnaires (Dewaele, 2010; Dewaele and Li, 2014a, 2014b; Rodriguez-Fornells et al., 2012) but it is harder to collect CS in authentic interactions from a large number of participants.

Researchers have argued that authentic data have most ecological validity (Pavlenko, 2014). While this is undoubtedly true, transcriptions of conversations in which CS occurs are very time-consuming, which can explain why studies based on authentic speech data tend to have small numbers of participants (Gardner-Chloros, 2009). However, to carry out solid statistical analyses researchers need a relatively large number of participants. The other problem with data collection is linked to the nature of CS itself: it emerges spontaneously in conversations between bi- and multilinguals, and one bilingual who might usually be happy to code-switch might refrain from doing so if the circumstances are not right (and this could include the unfamiliar presence of a microphone). In other words, researchers cannot ‘force’ bilinguals to use CS and are hence at the mercy of circumstances over which they have little control. Metaphorically, one could compare the search for CS as the sifting through the speech stream for gold nuggets. It requires determination and stamina.

Dewaele and Li (2014a) also wondered whether the lack of interest in individual differences in CS could also be linked to the confusion over the very notion of individual differences: ‘For instance, does it refer to the same individual’s different usage in different contexts, i.e. intra-speaker variation, or is it differences between speakers in producing the same code-switching pattern, i.e. inter-speaker variation, or both?’ (p. 226).

A final point is that the range of potential independent variables linked to CS is so vast, and spanning so many disciplines, that it may overwhelm researchers interested in studying individual differences. Applied linguists typically select variables that they are familiar with, such as sociobiographical variables, fluency in various languages and history of language learning and use. But other factors, such as personality traits, which have spawned a huge amount of research among psychologists, remain under-explored in CS research (Dewaele and Li, 2014a, 2014b).

The paper is structured in seven parts. We first review the existing research on individual variation in CS, focusing on sociolinguistic studies. We then introduce the five multicultural personality traits, which we hypothesize to have a prima facie link with self-reported CS. Our research instruments include an online questionnaire specially designed for the present study and the Multicultural Personality Questionnaire-Short Form (MPQ-SF) (Van Der Zee, Van Oudenhoven, Ponterotto, & Fietzer, 2013). We will describe the design of our empirical study, and we will present our four broad research questions and hypotheses. Subsequently, we will present the statistical analyses and discuss the findings in relation with the previous literature. The findings will be summed up in the concluding section.

Existing studies of individual variation in code-switching

Ritchie and Bhatia (2013) categorized the various factors affecting CS into four broad categories, which are participants’ individual characteristics, situational factors, sociopsychological factors and linguistic or pragmatic factors. These factors are highly interconnected and intertwined so that interdisciplinary research is of great importance.

The first category encompasses individual speakers’ and their interlocutor’s characteristics, including gender, age and social class. A classic study in this category is Li (1995), who examined a number of CS patterns – inter-speaker, inter-sentential, inter-clausal and content words – in three groups of Chinese-English bilinguals: students, British-born Chinese youth and long-term immigrants from Hong Kong, all living in Newcastle upon Tyne (UK). He found that variation existed across these groups according to bilingual experience and social network contacts the speakers had, as well as topic, setting and interlocutor of the interaction. A more recent study on CS and participants’ individual characteristics is Rodriguez-Fornells et al. (2012), who looked at self-reported CS from 582 Catalan-Spanish university students in Barcelona. They found that if Spanish (L1) had been acquired later, an increase in CS to Catalan was observed. The reverse pattern emerged for age of acquisition of Catalan. A higher level of proficiency in Spanish or Catalan was correlated with fewer code-switches to the other language. Predominant use of Spanish was linked to increased CS to this language and a decrease in CS to Catalan. Individual differences in multilinguals’ language profiles were also linked to self-reported differences in CS and attitudes towards CS in Dewaele and Li (2014a, 2014b). The study based on feedback from 2116 multilinguals from around the world showed that having grown up in a bilingual family is linked to more self-reported CS and more positive attitudes towards CS later in life. It confirmed earlier findings about the importance of speakers’ social networks, their interactional settings and social acceptance of languages to understand their CS patterns (Li, 1995; Muysken, 2000). Dewaele and Li (2014a, 2014b) found that the degree of multilingualism and ethnic diversity in the wider social environment, in which multilinguals grew up, was equally linked with more positive attitudes and more frequent CS. High levels of individual multilingualism (knowing more languages and knowing them really well) was linked to higher frequencies of CS but was only loosely related to attitudes towards CS. The degree of ethnic diversity in the work environment was also linked to more positive attitudes towards CS. Level of education, age and gender of the speaker have also been linked to attitudes towards CS and self-reported frequency of use of CS (Dewaele and Li, 2014b). Participants with the lowest and highest levels of education appreciated CS most and those with higher levels of education reported more CS with colleagues and family members. Participants in their teens and twenties appreciated CS less than older participants. The effect of age was different according to the interlocutor: older participants reported using CS more with family members and strangers and less with friends and colleagues compared to younger participants. Female participants had more positive attitudes towards CS than their male peers and reported significantly more CS in interactions with friends, family and colleagues.

Ritchie and Bhatia’s second category, situational factors, has been linked to Bell’s notion of Audience Design (Bell, 2001), in which speakers are assumed to modify their style to accommodate their interlocutor/s. This mostly takes place unconsciously in response to addressees, topics or settings and accounts for the most minor changes in style. Many studies have highlighted the strong influence of interlocutor (Dewaele, 2010; Li, 1995), showing that CS is most common amongst friends and only rarely used in conversations with strangers. Moreover, formality of the situation plays a role in CS patterns. Thus, in private contexts, such as amongst peer groups and family (Muysken, 2000), CS behaviour is assumed to differ from behaviour in public settings with strangers. Formality might also affect the amount of CS between the same interlocutors. Dewaele (2001) found that interactions between researcher and students contained much more CS to the students’ L1 in an informal classroom setting, while the CS disappeared during conversations with the same students in a formal L2 exam situation.

Ritchie and Bhatia’s third category, socio-psychological factors in CS, contains abundant research from bilingual environments, such as Canada, Belgium, Switzerland, Catalonia and the Basque country, to name a few. Indeed, prestige, language status and power relations, as well as linguistic norms and environment, are very influential on language choice and CS behaviour (Gardner-Chloros, 2009). Choosing a language, or switching to another language, can show group membership or neutrality. It can also show the stance of an individual or of sub-groups in political attitudes or ideologies. In addition, Dewaele and Li (2014b) found that positive attitudes towards CS itself were linked with higher levels of frequency of self-reported CS.

Ritchie and Bhatia’s final category, linguistic and pragmatic considerations in CS, focuses on intra-individual variation, in other words, decisions participants are faced with when wanting to communicate specific illocutionary intentions with multilingual interlocutors. One study that looked into this is Dewaele (2010), who measured the effect of type of interlocutor and pragmatic considerations linked to conversational topic on self-reported frequency of CS among 1453 multilinguals who filled out the BEQ (Bilingualism and Emotions Questionnaire; Dewaele & Pavlenko, 2001–2003) and among 20 multilinguals who talked about their language preferences with different interlocutors in different circumstances. Statistical analysis of the BEQ data showed that talking about personal or emotional topics was linked to significantly higher levels of self-reported CS than when speaking about neutral topics (Dewaele, 2010). Dewaele argued that when strong emotions need to be verbalized quickly to serve their cathartic function: there is no time to think about appropriate emotion words in the foreign language. Moreover, the script in which the words would figure might be incomplete and some grammatical information could be missing, adding to a growing sense of frustration. At that point, CS might seem like an acceptable option. The ‘bottleneck of the weaker language’ (p. 215) can be ignored and the strong emotions get verbalized in a language in which the speaker can channel his/her feelings quickly and in a satisfactory manner. Dewaele (2010) speculated that the emotional state of the speaker affects his/her executive control over language choice. The strong emotion may temporarily lift the inhibition of certain languages, leading to unplanned CS. In other words, ‘strong emotional arousal can force the speaker from monolingual into bilingual language mode with more CS’ (p. 219). The analysis of the interviews showed that CS is strongly linked to the perceived emotionality of the languages and that CS is deployed strategically. The typical direction of the CS in situations where strong emotions had to be expressed was from the foreign language to the L1. Some participants, often of Arab or Asian origin, reported CS in the opposite direction, especially for the expression of anger and swearing. They said that for them, CS to English (a foreign language) allowed them to escape the social constraint that weighs on them in their home environments, where anger cannot be displayed as openly as in English and where swearing carries a strong social stigma.

Dewaele also found that the higher frequency of CS with known rather than less known, or unknown, interlocutors was linked to a conscious, strategic choice of the speaker. In dealing with an unknown interlocutor the most logical option is to stick to the language in which the interaction was started. Conversations with known interlocutors allow CS to the shared languages. Resnik (2012) found similar effects of the type of interlocutor on self-reported CS in emotional interactions among 178 German, Chinese and Japanese foreign language users of English.

The strategic use of CS also emerged in Dewaele and Costa’s (2013) investigation on self-reported CS patterns between 182 multilingual clients and their therapists. Clients were found to use or initiate significantly more CS than their therapists, typically when the emotional tone was raised. CS was used when discussing episodes of trauma and shame, creating proximity or distance according to their need. It allowed clients to express themselves more fully to the therapist, adding depth and nuance to the therapy.

One factor that has hitherto not yet been examined systematically in CS research is personality. The only studies to date are Dewaele and Li (2014a, 2014b).

Dewaele and Li Wei (2014a) found that, in addition to participants’ linguistic history and current use of languages, certain personality traits, such as the degree of Extraversion and Cognitive Empathy (but not Tolerance of Ambiguity) were linked to significantly higher levels of self-reported CS (see also Dewaele & Li Wei, 2012, 2013). Dewaele and Li Wei (2014a) argue that risk-loving extraverts might enjoy CS as a form of verbal acrobatics, and that their more gregarious nature might encourage them to discover whether they share another language with an interlocutor, leading to more CS in order to underline the common linguistic roots. The positive relationship between Cognitive Empathy and self-reported CS in interactions with friends was interpreted as a willingness to accommodate to the friend’s linguistic choice in order to help the flow of the interaction.

Dewaele and Li (2014b) found that high levels of Tolerance of Ambiguity, Cognitive Empathy and Emotional Stability (but not Extraversion) are linked with significantly more positive attitudes towards CS.

In the present study, we hypothesize that certain personality traits contribute to a higher self-reported frequency of CS, possibly in interaction with particular social variables such as context and language background. In the following section, we discuss the personality traits that we intend to investigate.

Personality traits

Personality traits are hierarchically organized with five broad, independent dimensions at the summit (Pervin & Cervone, 2013). Personality questionnaires allow researchers to establish profiles of participants. In the present study, we used the Multicultural Personality Questionnaire (MPQ; Van der Zee et al., 2013), which was developed to assess five traits that are key to cultural adaptability and psychological well-being in a foreign environment. The multicultural personality traits correlate with the Big Five personality traits (Openness, Conscientiousness, Extraversion, Agreeableness and Neuroticism), which makes them valid constructs for general research. These five traits have ‘demonstrated incremental validity over broad personality measures such as the Big Five in predicting criteria such as students’ international orientation’ (Van der Zee et al., 2013, p. 118). Van der Zee and van Oudenhoven (2000) showed correlations of the MPQ traits with the Big Five traits: Openmindedness is strongly related to the Big Five’s Openness, Emotional Stability is negatively linked to Neuroticism, Social Initiative scores strongly correlate with Extraversion, while Cultural Empathy is linked to Conscientiousness, as well as to Extraversion and Agreeableness. Flexibility scores correlate negatively with Conscientiousness and positively with Extraversion.

We used the recent 40-item version of the questionnaire, the MPQ-SF (Van der Zee et al., 2013). It utilizes a five-point Likert scale. The MPQ-SF yields separate scores on each of the five dimensions: (1) Cultural Empathy, which ‘refers to empathizing with the feelings, thoughts, and behaviours of culturally diverse individuals’ (p. 118); (2) Flexibility, which ‘refers to interpreting novel situations as a positive challenge and adapting to these situations accordingly’ (p. 118); (3) Social Initiative, which ‘refers to actively approaching social situations and demonstrating initiative in these interactions’ (p. 118); (4) Openmindedness, which ‘reflects an open and unprejudiced attitude toward cultural differences’; and (5) Emotional Stability, which ‘reflects an ability to stay calm under novel and stressful conditions’ (p. 118).

Research questions and hypotheses

The present study aims to address the following research questions.

Does the amount of self-reported CS vary according to the interlocutor type? We expect most frequent CS with friends, followed by family, colleagues and least of all with strangers.

Are prior and current linguistic practices linked to the amount of self-reported CS? We expect participants who know more languages, who have advanced knowledge of several languages and who grew up with two or more languages before age 5 to use CS most frequently.

Are sociobiographical variables linked to the amount of self-reported CS? We expect female participants to code-switch more with known interlocutors. Age and education level could also be linked to self-reported CS.

Are personality traits linked to the amount of self-reported CS? We expect participants who score high on the five MPQ dimensions to code-switch more.

Method

Participants

A total of 298 participants (70 males, 205 females¹) from 58 different nations were included in the analysis. The majority of participants were British (19.5%), closely followed by Germans (19.1%), Spaniards (8.4%), French (5.4%), Swedes (4.0%), US-Americans (3.7%) and the remaining 38.9 % share another 52 nationalities (34 participants state having dual citizenships). The mean age was 34 years (Standard Deviation (SD) = 12.9) with a range from 16 to 80 years. Three quarters of the participants were female, which is not surprising in foreign language and personality research (Wilson and Dewaele, 2010). Concerning education level, the sample of this study is highly educated, with 35 participants having a high school degree, 91 holding a Bachelor’s degree, 112 having a Master’s degree and 58 a PhD. Two-hundred-and-forty participants (80%) lived or had lived abroad for an average of 10.2 years (SD = 9.8; ranging from 0.25 to 67 years). The sample consists of 19% bilinguals, 29% trilinguals, 25% quadrilinguals, 19% pentalinguals and 8% knowing six to ten languages. Forty-two per cent of the participants had grown up with a L2, or had learnt a L2 before the age of 5.

Participants’ information on self-perceived proficiency in their various languages was used to develop a global measure of multilingualism (Dewaele and Stavans, 2014). The ‘total proficiency score’ is the sum of self-perceived proficiency scores collected on five-point Likert scales for oral proficiency and written proficiency in up to five languages (maximal possible score 10 × 5 = 50). Such a measure is much more fine-grained than the sum of languages known. It allows us to distinguish, for example, a quadrilingual with limited knowledge of two languages from a trilingual with advanced knowledge of three languages (Dewaele and Stavans, 2014). The trilingual might know fewer languages, but knows them better. In other words, we avoid the relatively vague labels such as ‘bilingual, trilingual, quadrilingual, pentalingual’, where every language is included, despite the fact that knowledge in some can be very limited and that the decision to include a language in the count can vary between individuals. In the present sample total proficiency scores range from 7 to 48 with a mean of 27 (SD = 8.0). Participants with scores that were more than 1 SD below the mean were categorized as ‘Low Proficiency’ (n = 57), those with scores that were more than 1 SD above the mean were categorized as ‘High Proficiency’ (n = 55), while the remaining participants were categorized as ‘Medium Proficiency’ (n = 182). Linking this global language measure with CS makes more sense than proficiency measures for particular languages (L1, L2, L3…).

Instruments

Data were collected through the snowballing technique, that is, a non-probability sampling process. The anonymous online CS questionnaire was an open-access survey, advertised through several listservs, targeted emails to teachers and students, and informal contacts, asking them to forward the link to colleagues and friends. Because participants left occasional questions blank, totals for specific variables can vary. Participants also filled out a short sociobiographical questionnaire with questions about gender, age, education, language history and present language use. The research design and questionnaires received ethical clearance from the research institution.

Wilson and Dewaele (2010) have argued that the increased use of online questionnaires in applied linguistic research is a positive development. The main advantage is that researchers can collect large amounts of data automatically at a fraction of the cost and time of ‘pen and paper’ equivalents. Reaching participants through social media and the web allows researchers to cast their net much wider than in traditional research (often based on researchers’ students) and compose population samples that are more diverse in terms of gender, age, race, socio-economic status, cultural background and geographical location than the ‘pen and paper’ samples (Wilson and Dewaele, 2010).

The fact that the samples are not representative of the general population is not a handicap, although it needs to be kept in mind when interpreting the results. In multilingualism research participants must meet specific linguistic criteria, have high levels of metalinguistic and metacognitive awareness, and must be able and willing to engage with the questions on language preferences and use (Wilson and Dewaele, 2010).

We have argued that multilinguals are aware of the amount of CS they use in specific interactions, and they typically have no reason to lie about the frequency with which they use CS, as they perceive the phenomenon as something positive (Dewaele and Li, 2014b).

In addition to the background questionnaire, participants filled out the items of the MPQ-SF. The form and wording of the MPQ SF-40 were developed following strict guidelines; its questions are asked in the third person using mainly adjectives and enquiring about concrete behaviour. This protocol was used to make the process of filling out the questionnaire as easy and unambiguous as possible to prevent bias in results. Participants reply on a five-point Likert-scale (ranging from 1 = not at all applicable to 5 = totally applicable). Questions address five dimensions of multicultural personality in intermingled order. Scores are calculated by adding responses for respective items (some of which need to be reversed) so that total scores can range from 40 to 160. Cronbach’s alpha analyses revealed that the five scales have sufficient internal consistency with four dimensions hovering around .80 and only one of .68 (see Table 1).

Table 1.

Range, means, standard deviations and Cronbach’s alpha for the Multicultural Personality Questionnaire-Short Form (MPQ-SF).

Variable	Range	M	SD	Cronbach’s alpha
Cultural Empathy	15–40	32.4	4.2	.81
Flexibility	8–37	24.1	5.3	.80
Social Initiative	13–39	28.6	5.3	.83
Openmindedness	19–40	29.8	4.1	.68
Emotional Stability	10–39	25.7	5.3	.78

We used the same procedure as described for total proficiency to create three groups for each personality dimension: the ‘medium’ group, with scores ranging from minus 1 SD to plus 1 SD around the mean, the ‘low’ group with scores of 1 SD below the mean, and the ‘high’ group with scores of 1 SD above the mean.

Data about CS practices of the participants were elicited through the following general closed question: ‘Do you switch between languages within a conversation with certain people?’. It then specified four situations: ‘when speaking with some friends’, ‘with some family members’, ‘with some colleagues or clients at work/school’ and ‘with strangers’. Participants were asked to choose a response on a five-point Likert scale, that is, ‘never’, ‘rarely’, ‘sometimes’, ‘frequently’, ‘always’. We are therefore more concerned with the frequency of CS with different interlocutors in different contexts rather than with the structural patterns of switching. We cannot exclude a social desirability bias, as some participants may have exaggerated or played down their amount of CS depending on attitudes prevalent in their community. However, we feel that this bias could have gone in opposite directions for members of different groups, cancelling each other out. The categorization of interlocutors is obviously an abstraction. Participants pointed out that they used more CS with specific friends, colleagues or family members. However, we feel that the categories have sufficient validity because they reflect typical behaviour within particular language domains (Grosjean, 2010). Dewaele and Li (2014a) pointed out that statements in personality questionnaires are even more decontextualized as they attempt to elicit feedback on habitual behaviour in certain situations.

A series of one-sample Kolmogorov–Smirnov tests showed that the values for self-reported frequency of CS with the four types of interlocutors are not normally distributed (K-S Z-values vary between 3.1 and 3.7, all significant at p < .05). However, a look at the distribution shows curves with a slight skew towards scores on the high end of the distribution for friends and family, and a skew towards the low end for strangers and colleagues. Given the results of the normality tests, we exercised caution with the data analysis, testing the relationships reported below with both parametric and non-parametric approaches. Since there was no difference between the two approaches in the significance tests of the independent variables, and since one-way analyses of variance (ANOVAs) and t-tests tolerate moderate violations to their normality assumption rather well (skewed distributions) and allow for more detailed post hoc tests, we will present the parametric statistics (Rosenkrantz, 2008). However, because one-way repeated measures ANOVAs can only be performed on continuous data, we have opted for the non-parametric equivalent, Friedman’s ANOVA test.

Results

A Friedman’s ANOVA test for related samples revealed that the effect of type of interlocutor is highly significant on the amount of self-reported CS (N = 294,Chi² = 215.9, df = 3, p < .0001; (Figure 1). This result confirms that multilinguals report CS most with their friends followed by members of their family, colleagues and finally with strangers.

Figure 1.

Mean values (and SD) for self-reported code-switching (CS) with four types of interlocutors.

Participants’ linguistic history and current practice is linked to self-reported CS. The number of languages known by participants had a highly significant effect on self-reported frequency of CS with the various types of interlocutors (Table 2). Participants knowing more languages reported more frequent CS with all interlocutors (Figure 2).

Table 2.

The effect of linguistic history on code-switching (analyses of variance).

	Friends	Family	Colleagues	Strangers
Number of languages known (df = 4, 293)	F = 3.9	F = 2.8	F = 4.5	F = 1.8
	p < .004	p < .028	p < .002	p = ns
	Eta² = .051	Eta² = .037	Eta² = .059
Total Proficiency (df = 2, 293)	F = 11.4	F = 7.1	F = 9.3	F = 2.3
	p < .0001	p < .001	p < .0001	p < .099
	Eta² = .073	Eta² = .046	Eta² = .060	Eta² = .016

Figure 2.

Number of languages known and self-reported frequency of code-switching (CS).

Tukey HSD post hoc analyses showed significant differences between the bilinguals and the pentalinguals and sextalinguals in interactions with friends (p < .003 and p < .046, respectively). Only the bilinguals and the sextalinguals differed significantly in CS in interactions with family members (p < .018). Significant differences emerged again between the bilinguals and the pentalinguals and sextalinguals in interactions with colleagues (p < .031 and p < .001, respectively). No significant differences emerged in interactions with strangers.

The same pattern emerged for Total Proficiency, which had a significant effect on self-reported frequency of CS with all interlocutors (Table 2). The High Total Proficiency group reported more frequent CS with all interlocutors (Figure 3).

Figure 3.

Total proficiency and self-reported frequency of code-switching (CS).

Tukey HSD post hoc analyses showed significant differences between the low and the medium group, the medium and high group and the low and the high group in interactions with friends (p < .031, p < .003 and p < .0001, respectively).

The pattern was slightly different in family interactions, where no difference emerged between the low and medium group, but where differences existed between the low and high group and the medium and high group (p < .001 and p < .005, respectively).

The pattern for interactions with colleagues resembled that of interactions with friends: significant differences between the low and the medium group, the medium and high group and the low and the high group in interactions with friends (p < .01, p < .05 and p < .0001, respectively).

An independent t-test showed that the 123 participants who had grown up with a L2 before the age of 5 reported significantly more CS use with family members than the 173 participants who came into contact with a L2 at a later age (Table 3). Weak (non-significant) differences also emerged in interactions with friends and strangers, with early L2 acquirers reporting slightly more CS. An independent t-test showed no significant differences in self-reported CS use between the 240 participants who had lived abroad and the 58 participants who had not (Table 3).

Table 3.

The effect of early contact (<5 yrs) with a L2 and having lived abroad on code-switching (independent t-test).

	Early contact with a L2		Lived abroad
	t	p	t	p
Friends	1.8	.075	−0.7	ns
Family	5.6	.0001	0.3	ns
Colleagues	−0.5	ns	−0.3	ns
Strangers	−1.7	.094	−0.4	ns

The next research question deals with the effect of sociobiographical variables such as age, gender and education level. A Pearson correlation analysis showed that age is not significantly linked with self-reported frequency of CS in interactions with various interlocutors (Table 4).

Table 4.

The relationship between age and self-reported frequency of code-switching (Pearson correlation).

Interlocutor	N	Pearson’s r	p
Friends	297	−.063	ns
Family	293	.045	ns
Colleagues	296	−.052	ns
Strangers	295	.052	ns

An independent t-test showed no significant differences between the 70 male and the 205 female participants in frequency of self-reported CS in interactions with friends, family and colleagues (Table 5).

Table 5.

The effect of gender on self-reported code-switching (independent t-test).

	t	p
Friends	−1.4	ns
Family	−0.5	ns
Colleagues	0.8	ns
Strangers	0.9	ns

ANOVAs revealed only a significant effect of education on self-reported CS in interactions with strangers. Participants with higher levels of education reported more CS with strangers (df = 3, 291, F = 3.3, p < .022, eta² = .033).

The final research question focuses on the effect of the personality traits on self-reported CS. The ANOVAs revealed a weak effect of Cultural Empathy on self-reported CS in interactions with family members and colleagues (Table 6). Two significant effects emerged for Flexibility, which was linked with self-reported CS in interactions with family members and colleagues: those with lower levels of Flexibility reported more use of CS (Figure 4). Tukey HSD post hoc analyses showed significant differences between the low and the medium group in interactions with family members and colleagues (p < .032 and p < .033, respectively). Social initiative had a weak, non-significant effect on self-reported CS in interactions with family members. Openmindedness was significantly linked with self-reported CS in interactions with friends and family members. Tukey HSD post hoc analyses showed significant differences between the low and the high group in interactions with friends (p < .018), and between the low and the high group, and the medium and high group in interactions with family members (p < .001 and p < .03, respectively). Emotional Stability turned out to have no significant effect at all (Table 6).

Table 6.

The effect of personality traits on code-switching (analyses of variance).

Cultural Empathy (df = 4, 293)	Friends	Family	Colleagues	Strangers
	F = 1.3	F = 2.5	F = 2.8	F = 1.5
	p = ns	p = .08	p = .06	p = ns
		Eta² = .017	Eta² = .019
Flexibility (df = 2, 291)	F = 1.4	F = 3.3	F = 3.3	F = 1.4
	p = ns	p < .037	p < .039	p = ns
		Eta² = .022	Eta² = .022
Social Initiative (df = 2, 291)	F = 2.3	F = 1.2	F = 0.8	F = 0.4
	p = .09	p = ns	p = ns	p = ns
	Eta² = .013
Openmindedness (df = 2, 293)	F = 3.7	F = 6.8	F = 1.4	F = 2.8
	p < .025	p < .001	p = ns	p = .06
	Eta² = .025	Eta² = .045		Eta² = .019
Emotional Stability (df = 2, 293)	F = 1.5	F = 0.9	F = 0.1	F = 0.7
Emotional Stability (df = 2, 293)	p = ns	p = ns	p = ns	p = ns

Figure 4.

The effects of Flexibility (FL) and Openmindedness on self-reported frequency of code-switching (CS).

Discussion

The analyses of the data confirmed our first hypothesis: self-reported frequency of CS varies according to the type of interlocutor. CS was most frequent in interactions with friends, followed by interactions with family members and colleagues, and it was least frequent in interactions with strangers.

Our second hypothesis was also confirmed: parts of participants’ linguistic history – number of languages known, total proficiency and the age at which they came into contact with a L2 – but not the fact of having lived abroad – had a significant effect on self-reported CS. The effect sizes vary from small to medium, explaining between 3.7% and 7.3% of variance (Cohen, 1992).

Our third hypothesis was largely rejected: participants’ age did not correlate with frequency of self-reported CS. No gender differences emerged in self-reported frequency of CS. Level of education was only significant in interactions with strangers, where highly educated participants reported more CS.

Our final hypothesis was partly confirmed: the degree of Cultural Empathy had a weak positive effect on levels of self-reported CS, although only with family and colleagues. Flexibility had a significant effect on frequency of self-reported CS with family and colleagues. Surprisingly, it was the participants with lower levels of Flexibility who reported using more CS. However, the effect size was small (explaining 2.2% of variance). Social Initiative had a weak positive effect on self-reported CS with friends. Openmindedness turned out to have the strongest effects (explaining between 2.5% and 4.5% of variance): it had a significant positive link with self-reported CS in interactions with family and friends. Emotional Stability was not linked to frequency of self-reported CS.

The finding that multilinguals adapt their frequency of CS to the type of interlocutor confirms earlier research (Dewaele, 2010; Dewaele and Li, 2014a). Multilinguals are less likely to engage in CS with interlocutors they do not know, probably because they need to establish which languages they have in common with their interlocutors. CS becomes more likely when the interlocutor is known to the speaker, and participants are aware of which languages are shared. The highest amount of self-reported CS was reported with friends and family members. Frequency of self-reported CS with colleagues is slightly lower. This result is slightly different from the rank order established by Dewaele and Li (2014a) where self-reported CS with family members was lower than with colleagues. One possible reason for this is the linguistic heterogeneity of the different groups of interlocutors, and conventions on language preferences and CS.

The finding that early bilingualism or early L2 acquisition is linked to more CS later in life with family members and, to some extent, with friends and strangers, confirms the importance of multilingual practices in childhood. These practices might be linked to the development of high levels of multilingualism, meaning knowledge of several languages and a high level total proficiency in different languages. These two variables are in turn linked to more self-reported CS with family members, friends and colleagues, and their effect size varies from small to medium. Since CS is a defining form of bi- and multilingual behaviour (Grosjean, 2010; Li, 2013), it comes as no surprise that the more multilingual a person is, the greater the use of CS. It also shows that CS is linked to high levels of proficiency in different languages. This confirms findings reported by Dewaele (2010) and Dewaele and Li (2014a) about the positive relationship between high degrees of multilingualism, high self-perceived competence in foreign languages and self-reported frequency of CS. CS is clearly ‘not an indication of a deficit in the LX but on the contrary a characteristic of participants who feel proficient in their LXs’ (Dewaele, 2010, p. 201). The finding that those having lived – or are living – abroad had no effect on self-reported CS is surprising because the database on which Dewaele and Li (2014a) based their findings showed that the 1446 participants who had lived abroad reported significantly more CS with friends and family members than the 600 participants who had not lived abroad. Maybe the proportion of those who had, and had not lived abroad is too skewed (80% versus 20% of participants) in the present study to draw any meaningful conclusion. More information would have been needed about the linguistic environments of our participants, preferably longitudinal. Living in a multilingual neighbourhood in the home country might indeed be more conducive to CS than spending some time in a closed community abroad.

Very few sociobiographical variables were linked with self-reported CS. Age had no effect on self-reported CS, which confirms the findings of Dewaele (2010). However, Dewaele and Li (2014a) did find that older participants used significantly more CS with family members and strangers but also significantly less CS with friends and colleagues.

The absence of a gender effect in self-reported use of CS in interactions with friends, family and colleagues is unexpected considering previous research. Female participants reported using CS significantly more with friends, but not with colleagues (Dewaele, 2010). Dewaele and Li (2014a) found that female multilinguals reported more CS with friends, family and colleagues.

The lack of a significant effect of education level on self-reported CS in interactions with strangers in the present study also sets it apart from previous research: Dewaele and Li (2014a) found that more highly educated participants reported using more CS with colleagues and family members.

One of the most interesting findings of the present study is the link between four personality traits and self-reported CS. The weak positive link between Cultural Empathy and self-reported CS in interactions with family and colleagues cannot be over-interpreted, but it suggests a certain similarity with the finding of Dewaele and Li (2014a) between Cognitive Empathy and self-reported CS. It suggests that multilinguals who are better able to empathize with the feelings, thoughts and behaviours of culturally diverse interlocutors tend to CS more. As Dewaele and Li (2014a) pointed out, the ability to see the interlocutor’s perspective can convince speakers to resort to CS in order to increase the flow of the interaction. The absence of a link between Social Initiative and self-reported CS differs from the finding of a positive relationship between Extraversion and self-reported CS in Dewaele and Li (2014a), where it was interpreted as an indication that more gregarious multilingual extraverts might resort to CS when discovering common languages with an interlocutor in order to underline the common ground with the interlocutor and converge towards them.

The most puzzling finding was the negative relationship between CS and Flexibility. We hypothesized that high Flexibility participants, that is, those who interpret novel situations as a positive challenge and who are able to adapt to these situations, would report more CS, as this is a form of linguistic adaptation. Our results suggest the opposite, namely more CS among those who are less flexible. One possible explanation is that the high-Flexibility participants accommodate the linguistic choice of the interlocutor and stick to it; in other words, they do not switch back to their own original language choice. The finding of a positive effect of Openmindedness on self-reported CS is easier to interpret. Participants who scored high on this dimension displayed an open and unprejudiced attitude towards cultural differences, which meant that they were happy to switch back and forth between languages they shared with an interlocutor. In other words, they did not try to impose their own language preference, but did not accommodate entirely to the choice of the interlocutor either. The CS allowed communication to be more efficient and to the point (Dewaele & Li, 2014a), allowing the speaker to find ‘le mot juste’ (Grosjean, 2010, p. 53). The final personality trait to be considered was Emotional Stability. It turned out that this dimension was unrelated to the amount of self-reported CS. In other words, the ability to stay calm under novel and stressful conditions does not affect the frequency of self-reported CS.

We are aware of the advantages and disadvantages of our research design.

The use of an online questionnaire allowed us to tap into a large, culturally and socially diverse pool of participants, strengthening the ecological validity of the database (Wilson & Dewaele, 2010). We looked for general relationships between independent and self-reported dependent variables. In other words, we looked for generic patterns in the data provided by the multilinguals who participated in this study, independent of particular sociolinguistic and sociohistorical contexts.

An obvious limitation is that participants’ self-reports concerning CS are not the same as actual CS by participants, as social desirability bias cannot be excluded and the amount of meta-linguistic awareness about CS may vary. Indeed, there are instances of CS that multilinguals may be unaware of. However, we feel that multilinguals are capable of making generalizations about their CS with specific categories of interlocutors, and that our data have sufficient validity to investigate variation. Dewaele and Li (2014a) argued that a research design based on frequencies of actual CS would be dogged by a different and even more serious set of ethical problems in having to wire up participants and having to transcribe masses of data recorded in a variety of settings.

Some of our participants may have had a negative attitude towards CS and could have underestimated the frequency of their CS as a consequence. There are three arguments to counter this criticism: firstly, with nearly 300 participants, the sample is large enough to iron out biased information from some participants; secondly, participants could not gain anything by choosing what they thought was the socially desirable option as the questionnaire was anonymous; and thirdly, we found our participants had a generally positive attitude towards CS.

A final limitation is the lack of information on the amount of CS used by friends, family and co-workers, including past levels of CS with these interlocutors. For some participants, the opportunity to code-switch might only exist with specific language combinations with family members or colleagues. A low score on self-reported CS with a type of interlocutor can be the result of a conscious choice or simply a necessity (because the interlocutors do not share all the languages of the speaker; Dewaele & Li, 2014a).

Conclusion

We started this paper referring to Ziauddin Yousafzai’s highly emotional and totally unexpected code-switch to English when arriving at his daughter Malala’s hospital bed. Bystanders had been struck by this particular switch to English because Ziauddin had diverged from his usual CS practices. This means that multilinguals have a keen sense of what is ‘normal CS practice’, and what is ‘abnormal’. The aim of the present study was to tap into multilinguals’ awareness of their own habitual practices. We found patterns of systematic inter- and intra-individual variation. Intra-individual variation in self-reported CS linked to the type of interlocutor turned out to be highly significant: our participants reported significantly more CS in interactions with friends, followed by family members, colleagues and significantly less in interactions with strangers.

Inter-individual variation in self-reported CS linked to participants’ linguistic and psychological profiles was also significant. Participants reporting more CS had significantly higher levels of multilingualism and earlier onset of bilingualism (acquisition of a L2 before age 5, many languages known, high levels of total proficiency in the languages).

The present study is the first to investigate the link between the five personality dimensions and CS. Weak (non-significant) relationships emerged between two personality traits, Cultural Empathy and Social Initiative and self-reported CS while significant links appeared between Openmindedness, Flexibility and frequency of self-reported CS with certain types of interlocutors.

Sociobiographical variables (gender, age, education) had no, or very weak links with self-reported frequency of CS.

To sum up, the frequency of self-reported CS is not just linked to participants’ linguistic profile as most sociolinguists have found, but also to their personality.

Footnotes

Acknowledgements

We would like to thank the reviewers for their excellent comments on a previous version of the article and our participants for having agreed to fill out our questionnaire.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author biographies

Jean-Marc Dewaele is Professor of Applied Linguistics and Multilingualism at Birkbeck, University of London. He investigates individual differences in psycholinguistic, sociolinguistic, pragmatic, psychological and emotional aspects of Second Language Acquisition and Multilingualism. He is Vice-President of the International Association of Multilingualism, Convenor of the AILA Research Network Multilingualism and former President of the European Second Language Association. He is General Editor of the International Journal of Bilingual Education and Bilingualism.

Inga Zeckel has recently completed her Masters in Bilingualism and Multilingualism in the Department of Applied Linguistics and Communication at Birkbeck, University of London. The present paper is based on a part of her MA research project.

References

Bell

(2001). Back in style: Reworking audience design. In Eckert

Rickford

J. R.

(Eds.), Style and sociolinguistic variation (pp. 139–169). Cambridge, UK: Cambridge University Press.

Cohen

(1992). Statistical power analysis for the behavioral sciences. New York, NY: John Wiley.

Dewaele

J.-M.

(2001). Activation or inhibition? The interaction of L1, L2 and L3 on the language mode continuum. In Jessner

Hufeisen

Cenoz

(Eds.), Cross-linguistic influence in third language acquisition: Psycholinguistic perspectives (pp. 69–89). Clevedon, UK: Multilingual Matters.

Dewaele

J.-M.

(2010). Emotions in multiple languages. Basingstoke, UK: Palgrave Macmillan.

Dewaele

J.-M.

Costa

(2013). Multilingual clients’ experience of psychotherapy. Language and Psychoanalysis, 2, 31–50.

Dewaele

J.-M.

(2012). Multilingualism, empathy and multicompetence. International Journal of Multilingualism, 9, 352–366.

Dewaele

J.-M.

(2013). Is multilingualism linked to a higher tolerance of ambiguity? Bilingualism: Language and Cognition, 16, 231–240.

Dewaele

J.-M.

(2014a). Intra- and inter-individual variation in self-reported code-switching patterns of adult multilinguals. International Journal of Multilingualism, 11, 225–246.

Dewaele

J.-M.

(2014b). Attitudes towards code-switching among adult mono- and multilingual language users. Journal of Multilingual and Multicultural Development, 35, 235–251.

10.

Dewaele

J.-M.

Pavlenko

(2001–2003). Web questionnaire ‘Bilingualism and Emotions’. London, UK: University of London.

11.

Dewaele

J.-M.

Stavans

(2014). The effect of immigration, acculturation and multicompetence on personality profiles of Israeli multilinguals. The International Journal of Bilingualism, 18, 203–221.

12.

Gardner-Chloros

(2009). Code-switching. Cambridge, UK: Cambridge University Press.

13.

Grosjean

(2010). Bilingual. Life and reality. Cambridge, MA: Harvard University Press.

14.

(1995). Variations in patterns of language choice and code-switching by three groups of Chinese-English speakers in Newcastle upon Tyne. Multilingua, 14, 297–323.

15.

(2007). Dimensions of bilingualism. In Li

(Ed.), The bilingualism reader (pp. 3–22; 2nd edition). London, UK: Routledge.

16.

(2013). Code-switching. In Bayley

Cameron

Lucas

(Eds.), The Oxford handbook of sociolinguistics (pp. 360–378). New York, NY: Oxford University Press.

17.

Muysken

(2000). Bilingual speech. A typology of code-mixing. Cambridge, UK: Cambridge University Press.

18.

Pavlenko

(2014). The bilingual mind and what it tells us about language and thought. Cambridge, UK: Cambridge University Press.

19.

Pervin

L. A.

Cervone

(2013). Personality: Theory and research (12th edition). Hoboken, NJ: John Wiley & Sons.

20.

Resnik

P. A.

(2012). Differences in feeling – feeling the difference: Language, gender and emotion in bilingualism (Unpublished doctoral dissertation). Karl-Franzens-Universität, Graz, Austria.

21.

Ritchie

W. C.

Bhatia

T. K.

(2013). Social and psychological factors in language mixing. In Bhatia

T. K.

Ritchie

W. C.

(Eds.), The handbook of bilingualism (pp. 375–390, 2nd edition). Chichester, UK: John Wiley & Sons.

22.

Rodriguez-Fornells

Krämer

U. M.

Lorenzo-Seva

Festman

Münte

T. F.

(2012). Self-assessment of individual differences in language switching. Frontiers in Psychology, 2, 388.

23.

Rosenkrantz

W. A.

(2008). Introduction to probability and statistics for science, engineering, and finance. Boca Raton, FL: Chapman & Hall-CRC.

24.

Van der Zee

K. I.

Van Oudenhoven

J. P.

(2000). The multicultural personality questionnaire: A multidimensional instrument of multicultural effectiveness. European Journal of Personality, 14, 291–309.

25.

Van der Zee

K. I.

Van Oudenhoven

J. P.

Ponterotto

J. G.

Fietzer

A. W

. (2013). Multicultural personality questionnaire: Development of a short form. Journal of Personality Assessment, 95, 118–124.

26.

Wilson

Dewaele

J.-M.

(2010). The use of web questionnaires in second language acquisition and bilingualism research. Second Language Research, 26, 103–123.

27.

Yousafzai

Lamb

(2013). I am Malala: The girl who stood up for education and was shot by the Taliban. New York, NY: Little, Brown and Company.