First language primacy in multilingualism and emotion research: a view from Africa

ABSTRACT

The interaction between multilingualism and emotions has so far mostly been researched in Northern contexts. The findings obtained suggest that the first language is typically preferred for emotional and mental activities (i.e. ‘first language primacy’). In this article, we extend this area of investigation to Africa. We use the Bilingualism and Emotions Questionnaire to probe first language primacy in South Africa, employing a novel analytical approach – multiple correspondence analysis – that allows us to examine the relationship between particular languages and their order of acquisition. Our findings, in part, do reflect favouring of the L1, but with two important qualifications. First, English is also widely used for emotional expression in non-intimate domains. Second, L1 isiXhosa speakers prefer English for mental calculations, and isiXhosa is generally not associated with positive emotional characteristics. We explain these trends in relation to the multilingual context of South Africa, which, like many postcolonial countries, uses a former colonial language extensively in formal institutional spaces.

KEYWORDS:

• Multilingualism
• emotions
• first language primacy
• BEQ

Introduction

Language plays an important role in the expression and construal of emotions (Besnier 1990; Martin 2000; Ochs and Schieffelin 1989), while emotions also affect people’s language learning trajectories and affinities to language (Kramsch 2006; Richards 2022; Swain 2013). Recently, the effects of multilingualism on how emotions are linguistically expressed, perceived, and experienced have been investigated (Dewaele 2011; Pavlenko 2005, 2006a, 2006b). The findings suggest that multilingualism may interact with emotions in various ways. The focus of the present paper is one particular finding: namely, the first language (L1) learned by a multilingual is often the preferred language for emotional expression and for inner speech (e.g. Pavlenko 2004; Dewaele 2004b). This phenomenon is known as ‘L1 primacy’.

Research on L1 primacy has, however, mostly been conducted in Western contexts and with European languages. Typically, multilingual participants resident in Africa are not included in such studies, and African languages rarely feature in the constellation of languages investigated in this paradigm. This is an important shortcoming, considering that multilingualism in Africa differs in some important respects from multilingualism in Europe and North America (see Banda 2009; Myers-Scotton 1993; Oostendorp 2012). In the latter regions, multilingualism is often treated as an individual characteristic that develops either out of choice (so-called ‘elective multilingualism’) or through migration. Multilingualism in Africa, on the other hand, often entails the use of the former colonial languages together with local African languages in multilingual settings by multilingual individuals (Banda 2009). Further, multilingualism is typically not elective but arises out of necessity. Another difference is that in African settings, additional languages are often learned informally, and thus multilingual language users might not display ‘standard’ writing practices in these additional languages but rather forms of ‘grassroots literacy’ (Blommaert 2008). These differences in both the development and everyday manifestation of multilingualism across African contexts and more commonly studied Western contexts may entail differences in the relationships between individual languages and the functions that they fulfil.

Presently, our knowledge about L1 primacy is confined to how this phenomenon manifests in contexts in the Global North.¹ In both linguistics and other fields, the over-reliance on Northern contexts and subjects has been criticized, and researchers have been urged to expand their participant bases and to be more careful in making claims about universality (Bylund et al. 2023; Castro Torres and Alburez-Gutierrez 2022; Kidd and Garcia 2022). Furthermore, non-Western, or Southern contexts are increasingly recognized as spaces that provide not only raw data but also privileged theoretical insights, especially into issues such as diversity (Comaroff and Comaroff 2012; Connell 2017). This article investigates L1 primacy in South Africa to increase the ecological validity of the research paradigm and pursue some critical questions raised by researchers located in the Global North.

Pavlenko (2004, 192) notes that L1 primacy may not be ‘a phenomenon that exists across the board but rather a reflection of [a] romantic ideology … associated with European languages’ (Pavlenko 2004, 192). In addition to questioning the universality of L1 primacy, Pavlenko (2004, 192) asks ‘whether speakers of non-Western languages, or those who grew up speaking a dialect rather than a standard variety, feel the same way’. This latter question is of particular interest in settings such as South Africa, in which pronounced inequalities in language status are present across domains (e.g. business, education) and may be reflected in speakers’ own language perceptions and use.

Whether L1 primacy holds in non-Western contexts, and whether it is dependent on the L1 in question, has not yet been empirically investigated. This article presents such an investigation, conducted among a sample of multilingual participants in the Western Cape province of South Africa. The scope of the investigation encompasses participants’ language preferences for the expression of emotions, mental calculations, and inner speech, as well as their evaluation of their languages in terms of emotional characteristics. Specifically, the research questions the article addresses are.

• Which language(s), both in terms of the language itself (e.g. English, isiXhosa) and order of acquisition, are preferred for expressing emotions?

• Which language(s) are preferred for mental calculations and inner speech?

• How do participants evaluate their languages in terms of their emotional characteristics (i.e. ‘useful’, ‘colourful’, ‘rich’, ‘poetic’, ‘emotional’, and ‘cold’)?

We investigate these questions using an existing research instrument – the Bilingualism and Emotions Questionnaire – and an analytic approach, multiple correspondence analysis, that preserves the complexity of our data (see e.g. Ortega 2020).²

The following sections provide an overview of the South African linguistic context and discuss relevant literature on emotions and multilingualism. The methodology, results, and discussion follow.

Multilingualism in South Africa

South Africa is commonly regarded as one of the most multilingual countries in the world in terms of legislation, with 12 languages declared as official (in alphabetical order, Afrikaans, English, isiNdebele, isiXhosa, isiZulu, South African Sign Language, Sepedi, Setswana, Sesotho, Siswati, Tshivenda, and Xitsonga), and several other languages receiving protection under the constitution. In addition, as, migration from other parts of Africa increased after 1994, languages such as Shona (mostly spoken in Zimbabwe) and Lingala (mostly spoken in the Congo) are also increasingly used as home languages in South Africa (Mayoma and Williams 2021; Siziba and Hill 2018). As the rich multilingual situation suggests, multilingualism is a fruitful research area in South Africa (Bylund 2014; Coetzee-Van Rooy 2014; Heugh 2013a; Makalela 2015; Makukule and Brookes 2021).

In the Western Cape province, where the data for the current study were collected, three languages are official: Afrikaans, isiXhosa, and English. Although Afrikaans and isiXhosa have more L1 speakers, English is dominant in the public space (Posel and Zeller 2016; Rudwick 2021). Indeed, while official government communication must be conducted in all three of these official languages, a hierarchy of languages exists, with English being the most dominant, followed by Afrikaans. In education, mother-tongue schooling is available from Grades 1–3, but beyond this point only English and Afrikaans are extensively used as media of instruction. Thus, L1 isiXhosa speakers typically learn in English from Grade 4 onwards. IsiXhosa’s lower status is also reflected in its reduced use on public signage, where even when it is used, it is often riddled with mistakes and mistranslations (Dowling 2012).

The primacy of English in South African public life has generated some concern about a language shift to English in South Africa (e.g. Anthonissen 2013; Dyers 2008). However, several quantitative studies indicate that the mother tongue is typically maintained alongside English (Berghoff 2024; Coetzee-Van Rooy 2014; Posel, Hunter, and Rudwick 2022). The South African situation, then, is one in which the colonial languages exist and are widely used alongside the indigenous languages regardless of distinctions in language status. As such, it is an interesting context for investigating L1 primacy.

Emotions and multilingualism

Although emotions in linguistic expression have been investigated under various paradigms, such as language socialization (Ochs and Schieffelin 1989), systemic functional linguistics (Martin 2000), and from cross-cultural and cross-linguistic perspectives (Wierzbicka 1999), it has been less than 20 years since multilingualism has been put on the agenda as a serious area of investigation (see the 2004 special issue in Journal of Multilingual and Multicultural Development as the first demarcation of the topic). Research in this area varies both methodologically – ranging from experimental studies (Harris, Aycicegi, and Gleason 2003; Pavlenko 2008) to first-person-centred approaches (Besemeres 2004; Kramsch 2006; Prior 2015) – and theoretically, with some scholars opting to use not ‘emotion’ but the broader term ‘affect’, which includes emotions as well as moods, dispositions, and attitudes associated with specific people or situations (Ochs and Schieffelin 1989, 7).

Variation aside, this research shares a focus on speakers who use more than one language in their everyday lives and the influence of the use of multiple languages on how they express, experience, and construe emotions. Here, we focus primarily on sociolinguistic studies on multilingualism and emotions, and particularly studies that focus on L1 primacy, as this is the tradition on which this article builds.

Like the present study, most of the studies discussed here use the Bilingualism and Emotions Questionnaire (BEQ) as a data collection instrument (Dewaele 2004b). The original BEQ consists of 34 questions related to bilingualism and emotions, some of which are Likert-scale items and some of which are open-ended. It collects demographic information, language background information (e.g. age of acquisition, self-rated proficiency), questions regarding language choice for several emotion-related activities and mental activities, questions regarding the perceived emotional force of swearwords and taboo words in participants’ various language, and questions regarding participants’ perceptions of their languages’ characteristics (Dewaele 2004b). Data collected using this instrument have yielded several influential publications (e.g. Pavlenko 2005, with over 1,300 citations). The BEQ has been used since 2001 to investigate specific emotions or, more generally, to gather information on how linguistic trajectories and other background variables interact with how multilingual speakers express their emotions and perceive the emotionality of their languages. The original database consisted of 1,309 responses to the questionnaire. Although the sample featured 75 different L1s, African languages were absent (languages with fewer than 10 speakers were not explicitly named).

Many studies using the BEQ have found that multilingual speakers’ L1 is perceived as more emotional than their later-acquired languages and is the preferred language in which to express emotions. This L1 primacy was observed by, for example, Dewaele (2004a), although he also states that rather than being universal, the extent of L1 primacy is likely to be influenced ‘by several independent variables, all related to the individual’s linguistic history’ (Dewaele 2004a, 94). In a subsequent study (Dewaele 2011), an L1 preference for swearing and communicating emotions was observed even in individuals who described themselves as ‘maximally proficient’ in the L2. Notably, however, follow-up interviews revealed that participants who had been extensively socialized in the L2 culture and had less contact with the L1 felt that their L2 had taken on the more emotional role due to the contexts in which they used it (Dewaele 2011, 45). This observation is particularly relevant for the current study, conducted in a setting in which people with different cultures live side by side and where socialization into an L2 culture does not necessarily entail distance from the L1 culture.

The BEQ has also been used to explore specific emotions. Dewaele (2006) investigated the effects of L2 socialization on language preferences for the expression of anger and found that a language other than the L1 ‘can become the preferred language for anger expression, once emotion repertoires and scripts have been acquired in the process of language socialization’ (Dewaele 2006, 143). Similar findings were obtained with respect to context of acquisition and self-rated proficiency, as participants who had learned a language in an instructed environment were less likely to use that language for expressing anger (Dewaele 2006, 143), and those with high self-rated L2 proficiency were more likely to use the L2 for expressing anger (Dewaele 2006, 144). Once again, the findings from this study identify interesting aspects for consideration in the South African context. English is very often learned in instructed settings, but its pervasiveness raises the question of whether L2 speakers of the language would use it to express anger.

In more recent work with BEQ data, Dewaele focused on inner speech (i.e. the language individuals use for talking to themselves, whether out loud or silently) and found that multilinguals preferred the L1 for inner speech and for emotional inner speech, while later-learned languages were used much less frequently for these tasks. Subsequently learned languages were also used more frequently for non-emotional inner speech.

Studies using the BEQ suggest that there is a strong L1 preference for the expression of emotions. However, these studies are relatively homogenous in that they were all conducted in Northern settings. Research conducted in Africa using a different methodological approach hints that L1 primacy may not hold to the same extent in other kinds of multilingual contexts. For example, Spernes and Ruto-Korir (2021) investigated language choice for emotional expression among multilingual Grade 8 pupils in rural Kenya. The pupils spoke English, Swahili, and Nandi (an indigenous Kenyan language), with Swahili and English but not Nandi being used in education. The researchers observed that learners preferred to use Swahili and English for emotional expression, with Nandi being little used. They attribute this finding to the exclusion of Nandi from educational settings.

Spernes and Ruto-Korir (2021) is an important first examination of language and emotion in a multilingual African context. However, the authors themselves acknowledge some shortcomings of the study, specifically that they could not definitively determine the pupils’ L1 and that their data collection procedure may have led to the participants favouring some languages over others. The present study builds on this first African contribution while avoiding such limitations by employing the BEQ, a now-standard method for gathering data on multilingualism and emotion.

Method

Participants

170 participants (43.5% male; average age 35 years; range 15–78 years) provided complete responses to the questionnaire. 73.6% of the participants had a university degree (31.2% undergraduate only and 42.4% postgraduate). All participants reported knowing at least two languages, and 88.2% claimed to be functionally bilingual, that is, they reported using more than one language on a daily basis. Further, 52.9% of participants reported knowing three languages, 28.8% four languages, and 10.6% five languages.

Participants reported having some proficiency in 48 different languages.³ The languages most commonly spoken were English (spoken by 100% of participants), Afrikaans (74.12%), isiXhosa (21.76%), isiZulu (13.53%), French (18.24%), German (12.94%), and Shona (6.47%). Afrikaans (30%) and English (37.65%) were the languages most commonly selected as L1 (note that participants could select more than one language as their L1), followed by isiXhosa (12.94%). The order of acquisition of the most commonly used languages is presented in Table 1. Here, ‘Other African’ is an umbrella category for African languages other than isiXhosa; ‘Other European’ comprises languages spoken extensively in (Western) Europe that more than 5% of participants reported knowledge of; and ‘Other’ comprises languages not included in the previous categories.

Table 1. Order of Acquisition by Language Groups.

	Language Group
L	Afrikaans	English	isiXhosa	Other African	Other European	Other	Total
1	51	64	22	17	12	4	170
	30.00	37.65	12.94	10.00	7.06	2.35	100.00
2	50	90	4	11	13	2	170
	29.41	52.94	2.35	6.47	7.65	1.18	100.00
3	12	13	7	20	31	7	90
	13.33	14.44	7.78	22.22	34.44	7.78	100.00
4	12	1	2	13	19	2	49
	24.49	2.04	4.08	26.53	38.78	4.08	100.00
5	1	2	2	5	7	0	17
	5.88	11.76	11.76	29.41	41.18	0.00	100.00
Total	126	170	37	66	82	15	496
	25.40	34.27	7.46	13.31	16.53	3.02	100.00

Note: Columns are languages. Column totals reflect the number of participants (out of the total of 170) who report knowledge of a particular language. Rows (L1, L2, L3, L4, L5) indicate the number of participants reporting that language as L1, L2, etc. Percentages are provided under the raw frequency numbers.

Age of acquisition data was collected in a broad fashion: participants indicated, for each of their additional languages, whether they had learned it between ages 0 and 6, ages 7 and 12, ages 13 and 18, or age 19 and above. Both English and Afrikaans, when acquired as additional languages, were most commonly acquired from age 0–6 (47.2% and 46.7%, respectively), with the second most common age of onset reported being age 7–12 (38.7% and 36%, respectively). For isiXhosa, 40% of additional-language speakers reported an age of acquisition of 19 years and above, with the remaining participants being equally divided (20% each) between the other age categories.

We also collected self-rated proficiency data for each of participants’ languages. Speakers of English as an additional language rated their proficiency as 13.5/15, on average (SD = 2.21). For Afrikaans, this figure was 9.79/15 (SD = 2.92); and for isiXhosa, it was 7.73/15 (SD = 3.97).

Materials

Data were collected using a slightly modified BEQ (for information on the original and the use of web questionnaires in general, see Wilson and Dewaele 2010). The modification entailed adding an additional response option – ‘I use more than one language in this situation’ – to some of the questions in the original BEQ version. This addition was motivated by the context of the study, in which people commonly use more than one language in their everyday lives. The survey was created and hosted on the SurveyMonkey website and was available for six months. The survey link was distributed via email and Facebook to potential participants. The first part of the survey collected demographic information, while the second part asked questions about participants’ linguistic background. The third part of the questionnaire comprised Likert-scale questions that elicited information about participants’ language use and language choices in various situations, with particular emphasis on emotional language use. The fourth and final part of the questionnaire comprised questions about emotional perceptions and language choice and asked participants to indicate, for a set of characteristics (‘useful’, ‘colourful’, ‘rich’, ‘poetic’, ‘emotional’, and ‘cold’), whether each of their languages possessed these characteristics.

Analysis

We used multiple correspondence analysis (MCA) to explore the associations between participants’ languages, their order of acquisition, and the numerous language-related variables included in the questionnaire. MCA is a statistical technique that allows for the graphical representation of the structure of multivariate categorical data (Yelland 2010, 1). Such data would typically be represented in a frequency table. However, when the number of variables (i.e. rows and columns) is large, the frequency table format makes it difficult to identify patterns in the relationships between and within rows and columns. In such cases, the plots created via MCA facilitate the representation and interpretation of the data (Durbach 2014, 14).

MCA represents variables or categories of variables as coordinate points on a correspondence plot. The maximum number of dimensions of such a correspondence plot is calculated by subtracting 1 from the number of rows or columns (whichever is highest) in the frequency table of the data. Visualizing this maximum number of dimensions is, however, impractical and does not facilitate the interpretation of the data. As such, a process of dimensionality reduction is undertaken so that the data can be represented in fewer dimensions, typically two.

We analysed only the data relevant to our research questions, namely that on language use (which included scores for feelings, anger, mental calculations, inner speech, and memory recall) and language perception (which included rating scores of languages as useful, colourful, rich, poetic, emotional, and cold). These data were simultaneously cross-tabulated against language group and order of acquisition, with language group and order of acquisition (e.g. L1 English, L2 isiXhosa) represented as rows and the various language features or characteristics represented as columns.

We then generated several MCA plots to represent the associations in these high-dimensional tables between language group and order of acquisition on the one hand and the different measures of language use and language perception on the other. A dimensionality reduction process was employed to allow for the illustration of these many associations in two dimensions.

In the resulting two-dimensional plots, weighted row and column averages (termed ‘profiles’) are represented as points, with the distance between row points indicating the extent of their similarity. Here, rows with similar profiles are positioned in greater relative proximity than rows with dissimilar profiles, with the same being true for columns. Further, the distance between a point and the origin of the plot (i.e. the 0,0 point) is referred to as the point’s ‘inertia’, where a higher inertia value reflects a greater difference of the point’s profile from the average distribution of the data (Greenacre 2017, 38). Thus, a particular category is more distinct from the average when it lies further from the plot’s origin.

The two axes (horizontal and vertical) of these plots each explain some proportion of the variance in the data. The sum of these two values indicates the total amount of variance that is captured in the plot and reflects the extent to which the plot is able to reflect the data structure.

While the aim of the study was not to conduct statistical testing of particular hypotheses, we also report on the results of chi-square tests run to follow up on patterns related to L1 primacy that were identified in the MCA plots. In interpreting the chi-square results, categories with residuals (i.e. discrepancies between the actual values and the expected values) over 2 are understood to be strong contributors to the significant chi-square value (Sharpe 2015). Where chi-square results are not significant but there are clear trends in the plots, we report response percentages.

R (version 4.2; R Core Team 2022) was used to analyse the data and produce the plots. We used the ca (version 0.71.1; Nenadic and Greenacre 2007) and ggplot2 (version 3.3.6; Wickham 2016) packages. The tables containing the raw data and the accompanying R scripts are available at https://osf.io/g392j/?view_only = feedec9a096f4af49360ba0bd7efb95f.

Results

Language use and perception patterns are illustrated in four MCA plots: one for the expression of feelings, one for the expression of anger, one that reflects mental activities (mental calculations, inner speech, and memory recall), and one that reflects participants’ perceptions of their languages’ characteristics.

Figure 1 illustrates the associations between each language group and language use patterns when communicating feelings in a variety of situations (when alone, with friends, with family, and in written correspondence), where participants indicated whether they did or did not use each of their languages in each situation. Cumulatively, the plot captures 92% of the variance in the data; indicating that the plot accurately illustrates the data’s structure. The horizontal dimension shows the contrast between use and non-use of languages for the given situations, with non-use on the left side and use on the right side. As shown by the greater spread of the points along this axis, it is the contrast between use and non-use that explains most of the variance in the data. In the vertical dimension, there is a contrast between English, which appears in the upper half of the dimension, and all other languages, which appear in the lower half. In addition, written correspondence (letters) contrasts with other situations.

Figure 1. Language Use Patterns for Feelings. Language point labels are in green and follow the format of L:LanguageName_OrderofAcquistion, such that ‘L:isiXhosa_1’ is the point for L1 isiXhosa. In the language point labels, ‘OE’ and ‘OA’ denote ‘Other European’ and ‘Other African’, respectively. Each language use situation appears on the plot as two points: one reflecting use (e.g. ‘letters_emails:1’) and one reflecting non-use (e.g. ‘letters_emails_0’).

Points representing languages and language use patterns for expressing feelings distributed across four quadrants. Points representing non-use of languages for expressing feelings are in the left horizontal dimension. Points representing use of languages for expressing feelings are in the right horizontal dimension. All L1s fall in the right horizontal dimension, as well as L2 English and L3 English, and are thus associated with use for expressing feelings. All other languages fall in the left horizontal dimension and are thus associated with non-use for expressing feelings.

Overall, there seems to be a clear preference for using the L1 to communicate feelings in any situation, as can be seen by the appearance of the L1 points on the right of the plot. This general preference for the L1 is reflected in the fact that a chi-square test conducted on the L1 data is not statistically significant (X2 (12, N = 170) = 18.85, p = .09). It is only L2 and L3 English, as well as L2 isiXhosa, that are also associated with use for communicating feelings. Indeed, a chi-square test conducted on the L2 data is statistically significant (X2 (12, N = 170) = 21.25, p = .04), suggesting differences in the use patterns of the different L2s. An examination of the residuals indicates that this difference is driven by preferences for the expression of feelings in writing, where English is strongly preferred (residual = 3.77). The residuals for L2 isiXhosa are not large, likely because only four respondents had isiXhosa as an L2, and only three of these individuals responded to the relevant question. However, 2–3 of these respondents indicated that they used isiXhosa to communicate feelings in all situations except written correspondence.

All other languages appear on the left of the plot and are associated with non-use. The clustering of language points here indicates similar profiles for all participants’ additional languages, suggesting that for languages other than English, it is primarily language status (L1 vs. non-L1) that matters here rather than the particular language in question.

In terms of the column profiles, language use with family and when alone follow similar patterns, as can be seen by their proximity on the plot, while language use for written correspondence follows a different pattern: as discussed above, English is generally preferred here, regardless of its order of acquisition.

Figure 2 illustrates the associations between the language groups and their use for expressing anger in various situations, with the added situation of communicating with strangers. Cumulatively, the horizontal and vertical dimensions of the plot explain 89% of the variance in the data, with the horizontal dimension again contributing most in this regard.

Figure 2. Language Use Patterns for Anger.

Points representing languages and language use patterns for expressing anger distributed across four quadrants. Points representing non-use of languages for expressing anger are in the left horizontal dimension. Points representing use of languages for expressing anger are in the right horizontal dimension. All L1s fall in the right horizontal dimension, as well as L2 English, L3 English, and L4 English, as well as L2 isiXhosa, and are thus associated with use for expressing anger. All other languages fall in the left horizontal dimension and are thus associated with non-use for expressing anger.

As in Figure 1, the horizontal dimension shows the contrast between the use versus non-use of specific languages for expressing anger, with non-use on the left and use on the right. The arrangement of language groups in Figure 2 is largely identical to that in Figure 1, where it is generally the L1 that is used to communicate anger, with English also being used regardless of its order of acquisition (i.e. L1, L2, L3, L4). L2 isiXhosa is again the only non-L1 that is associated with use versus non-use to communicate anger.

In the vertical dimension, we see differentiation between language use with family and when alone, on the one hand, and language use in written communication and with strangers. In the latter situations, which are more public in nature, English is primarily used, whereas the other L1s are used in the intimate domains. The significant result of a chi-square test conducted on the L1 data (X2 (16, N = 170) = 37.37, p = .002) supports the interpretation that the L1s differ with respect to their use for the expression of anger. Specifically, of all the L1s, the preference for English when expressing anger in writing is especially strong (residual = 3.84), and isiXhosa and other African languages are dispreferred for this use (residuals = −2.33, and – 2.37, respectively).

Figure 3 illustrates the associations between the language groups and their patterns of use for mental operations: mental calculations, inner speech, and memory recall. Cumulatively, the two dimensions explain 96.7% of the variance in the data. The horizontal dimension, which accounts for most of the variance, again shows the contrast between language non-use (left) versus use (right) for the various activities.

Figure 3. Language Use Patterns for Mental Use.

Points representing languages and use for mental activities (inner speech, memory recall, and mental mathematics) distributed across four quadrants. Points representing non-use of languages for mental activities are in the left horizontal dimension. Points representing use of languages for mental activities are in the right horizontal dimension. All L1s fall in the right horizontal dimension, as well as L2 English and L3 English, and are thus associated with use for mental activities. All other languages fall in the left horizontal dimension and are thus associated with non-use for mental activities.

This distribution of points along this dimension shows that the L1 is generally associated with use for these activities, while the L2, L3, and L4 are not strongly associated with use in these cases. Exceptions arise in the cases of L1 isiXhosa and other African-language L1s, which are positioned far away from the mental mathematics point in the vertical dimension, reflecting that although these L1s are used for inner speech and memory recall, they are less used for mental calculation. While a chi-square test on the entirety of the L1 data is not statistically significant (X2 (8, N = 170) = 13.68, p = .09), this difference in L1 use for mental mathematics is reflected in the percentages: 45% and 6% of L1 isiXhosa and other L1 African-language speakers, respectively, reported using their L1 for this task. These percentages are markedly lower than those for the other L1s (Afrikaans 82%, English 100%, and Other European 75%).

Lastly, once again, English seems to be used for these activities even when it is an L2 or L3, with L2 English especially being associated with use for mental mathematics: 65% of L2 English speakers use English in this domain, but other L2s are hardly used (Afrikaans 4%, isiXhosa 0%, Other African 0%, and Other European 33%).

Figure 4 illustrates the characteristics attributed to each of the language groups. Cumulatively, the plot captures 84.1% of the variance in the data. The horizontal dimension contrasts the absence versus presence of language characteristics, with the exception of ‘cold’, which is the only negative characteristic. All the L1s except isiXhosa were associated with the more positive language characteristics (rich, emotional, colourful, etc.). A chi-square test conducted on the L1 data was not statistically significant (X2 (20, N = 170) = 10.7, p = .9), but in terms of percentages, isiXhosa scored lowest on all positive characteristics, with particularly few respondents reporting that this language was ‘colourful’ (27%) and ‘poetic’ (36%). English, in contrast, was always associated strongly with positive characteristics, regardless of whether it was the L1 or a later acquired language (note the positions of L:English_1, L:English_2, and L:English_3). Indeed, in terms of percentages, English scored highest of all L2s and L3s on all the positive characteristics. Other European L4s were also positively perceived.

Figure 4. Perceptions of Language Characteristics.

Points representing languages and language characteristics (‘useful’, ‘colourful’, ‘rich’, ‘poetic’, ‘emotional’, and ‘cold’) distributed across four quadrants. Points representing absence of positive language characteristics (i.e., all characteristics except ‘cold’) are in the left horizontal dimension. Points representing presence of positive language characteristics are in the right horizontal dimension. All L1s except isiXhosa fall on the right horizontal dimension and are thus associated with the presence of positive characteristics. All non-L1s, as well as L1 isiXhosa, fall in the left horizontal dimension, and are thus associated with the absence of positive characteristics.

The second dimension, although it captures only a small amount of variation in the data, contrasts usefulness (in the upper dimension) with non-usefulness (in the lower dimension). English is concentrated around the ‘useful’ point.

Discussion and conclusion

The impetus for this paper was to investigate Pavlenko’s (2004) question about whether the idea of L1 primacy (i.e. that the L1 is always the preferred language and carries the strongest emotional weight) holds outside of Northern settings. In doing so, the paper went beyond previous research, which has focused on order of acquisition without considering specific L1s, L2s, and so forth. Tailoring our approach to suit our multilingual setting, in which indigenous languages exist alongside societally dominant English, we employed a data analysis method that allowed for the investigation of order of acquisition in combination with specific languages. Also relevant here is the fact that our questionnaire allowed participants to report use of more than one language in a specific situation, in keeping with the fluid multilingual contexts found in South Africa.

The results reveal some patterns that warrant further discussion. These patterns speak to the debate regarding the universality of L1 primacy while also shedding new light on the emotional aspects of South African multilinguals’ linguistic repertoires.

Beginning with the communication of feelings and anger: here, we observed a preference for use of the L1, regardless of what language it was. The exceptional situation was the communication of feelings and anger in written correspondence (letters, emails) and the expression of anger towards strangers. For these uses, English was the preferred language, regardless of order of acquisition. This finding makes sense in light of the more public-facing nature of these activities, where it perhaps cannot be assumed that the recipient of communication shares the same L1 as the speaker. In such situations, English may serve as a lingua franca.

We again observe a general L1 preference for the mental activities of inner speech and memory recall, with all L1s being associated with use for these activities. Mental mathematics, a less personal mental activity than inner speech and memory recall, patterned distinctly, where L1 isiXhosa and other African-language L1s tended not to be used for mental calculations, and L2 and L3 English were associated with use for this purpose. This pattern may relate to the preference for English in the teaching of mathematics even to learners with another mother tongue (e.g. Cekiso, Meyiwa, and Mashige 2019). More generally, English terms for numbers and numerical concepts have been widely integrated into the isiXhosa lexicon, such that many speakers use these terms rather than the ‘pure’ isiXhosa equivalents, which tend to be much more complex (e.g. the isiXhosa equivalent of ‘seventeen’ is ishumi elinesixhenxe, literally ‘ten that has seven’). Some language and education researchers have also claimed that mathematical curricula are currently not ‘responsive to the character of African languages’ (Lepota 2023, 26). As such, L1 African language speakers’ use of L2 and L3 English to perform mental calculations makes sense.

Lastly, we examined respondents’ emotional evaluations of their various languages in terms of the characteristics of ‘useful’, ‘rich’, ‘emotional’, ‘poetic’, ‘colourful’, and ‘cold’. Here, all L1s except isiXhosa were associated with the positive characteristics. Languages that were also positively perceived even when they were additional languages were English and other European languages. English in particular was more strongly associated with the characteristic of ‘usefulness’ than the other languages.

L1 isiXhosa speakers’ non-positive evaluation of their mother tongue stands out in the data and may be attributable to the language’s position on the hierarchies of language status and power in South Africa. Notably, Afrikaans, the other well represented language in the Western Cape, was associated with all the positive characteristics. Afrikaans, like isiXhosa, is less societally prominent than English; however, it is more powerful than isiXhosa, in that Afrikaans remains a medium of instruction throughout secondary schooling and even to some extent in tertiary education, and Afrikaans media is strong and well-funded (Steyn 2016). The relative accessibility and prominence of Afrikaans in education, literature, and entertainment may contribute to its perception as ‘rich’, ‘poetic’, and ‘colourful’ despite the even greater prominence of English.

We note further that although L1 isiXhosa was not associated with the positive characteristics, other African-language L1s were. This finding is somewhat difficult to interpret because the ‘Other African Language’ category encompasses several languages – some indigenous to South Africa, some not – all of which may be differently perceived. One possible explanation for this finding may be that non-indigenous languages, and L1 speakers’ perceptions of them, are to some extent unaffected by the language hierarchies in South Africa.

While our results partly support previous findings in multilingualism and emotions research, they also contribute a new perspective on the topic. Like the findings of Spernes and Ruto-Korir (2021) and other studies using qualitative autobiographical data (Bristowe et al. 2014; Mashazi and Oostendorp 2022), our results suggest that the relationship between language and emotion might differ across Global North and African multilingual contexts. Specifically, our findings prompt a rethinking of the widely accepted ideology of L1 primacy and suggest that more can be discovered when examining the characteristics of particular languages rather than focusing solely on order of acquisition, which has previously been the trend in this area of study.

The findings also suggest that more research is needed on the effect of language of education, and specifically literacy in the L1, on perceptions of language use for and in emotional contexts. Spernes and Ruto-Korir (2021) attribute their participants’ non-use of Nandi for the expression of emotions to this language’s exclusion from the school system. In other words, they propose that learners do not use their indigenous language to tell emotion-related stories because their expressive abilities in this language are comparatively underdeveloped. Indeed, many African language speakers have only limited literacy in their L1 (see for example Williams and Cooke 2002); regarding the South African context specifically, Heugh (2013b) reports a decline in the number of students who study African languages at university and a neglect of teacher training in reading and writing in the home language. Thus, in our data, language preferences for written correspondence and mental calculations – activities for which L2 English was favoured over L1 isiXhosa and other African-language L1s – may have less to do with emotionality and choice and more to do with the individual’s schooling experience and/or their ability to express themselves in their L1 in writing. In fact, our background data suggests that very few of the L1 isiXhosa speakers had any exposure to isiXhosa in educational settings (either as medium of instruction or language subject).

Finally, our findings also speak to the ongoing debate regarding language shift to English in South Africa. Like other studies examining South African multilinguals’ language use and preferences (Berghoff 2024; Coetzee-Van Rooy 2014; Posel, Hunter, and Rudwick 2022), our results do not support the idea that English is displacing the indigenous South African languages. Rather, despite preferences for English in certain situations, we observe a prioritization of the L1 for more personal uses, such as inner speech and the communication of emotions and anger to loved ones. Notably, this preference for the L1 persists even though our participants report high proficiency in English as a second language, which aligns with Dewaele’s (2011) finding of an L1 preference for emotional expression even among participants who reported maximal proficiency in the L2.

The study is subject to a few limitations. Firstly, our participants were drawn from one province in South Africa – the Western Cape – and were, on average, highly educated. Multilingual individuals from different regions and of different socio-economic backgrounds may have different language use preferences and different perceptions of the emotionality of their various languages. In addition, our sample of L1 isiXhosa speakers is relatively small. Further research might seek to replicate our findings with a larger sample of native speakers of this language.

Overall, the present study hints that the language(s) of the heart may be considerably more complex than what current literature suggests and raises the possibility that the type of multilingual context in which languages are acquired may be as important as multilingualism per se in the construal and expression of emotions. The patterns we have identified will hopefully invite more research on language and emotion in Southern contexts. Future research might, like the present study, pursue approaches to the analysis of large-scale data that aim to preserve its complexity. Another avenue for exploration would be to develop new tools for collecting data on multilingualism and emotions that are grounded in the realities of language learning and use in non-Western settings. Here, it may be particularly useful to think in terms of repertoires rather than named languages (e.g. Wei and García 2022). Such endeavours can contribute to building ‘a richer and more complete understanding’ of our species (Apicella, Norenzayan, and Henrich 2020, 326) and the languages we use to make sense of our worlds.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by National Research Foundation.

Notes on contributors

Marcelyn Oostendorp

Marcelyn Oostendorp is an Associate Professor in the Department of General Linguistics at Stellenbosch University, South Africa. Her research is primarily concerned with multilingual and multimodal forms of meaning-making in contexts such as education, the media, and the workplace. Her research has appeared in outlets such as Applied Linguistics, Text and Talk, Critical Discourse Studies, and Social Semiotics.

Tanya Little

Tanya Little completed her MA in General linguistics at the Department of General Linguistics at Stellenbosch University.

Robyn Berghoff

Robyn Berghoff (Stellenbosch University) is a psycholinguist interested in language acquisition as well as bilingual language representation and processing. She is currently leading a project investigating morphosyntactic representation and processing in second language learners of isiXhosa. Her work has appeared in outlets such as Applied Linguistics, Bilingualism: Language and Cognition, Second Language Research and Studies in Second Language Acquisition.

Notes

1 In much of the social sciences and humanities, the distinction between Western and non-Western contexts is not used, and reference is instead made to the ‘Global North’ and ‘Global South’. According to Dados and Connell (2012, 13), ‘Global South functions as more than a metaphor for underdevelopment. It references an entire history of colonialism, neo-imperialism, and differential economic and social change through which large inequalities in living standards, life expectancy, and access to resources are maintained’. These authors also acknowledge that the Global South is entangled with the Global North, in that pockets of the North can be found in the South and vice versa.

2 Correspondence analysis, while not widely used in multilingualism research, is frequently employed in other subdomains of linguistics, such as corpus linguistics (e.g., Deshors 2017) and cognitive linguistics (e.g., Janda 2019).

3 The 48 languages listed by participants were, in alphabetical order: Afrikaans, Arabic, Catalan, Chichewa, Croatian, Dutch, English, Fang, French, German, Greek, Gujarati, Hebrew, Hindi, Italian, Japanese, Kikongo, Kimbala, Kurdish, Latin, Lingala, Malay, Mandarin Chinese, isiNdebele, isiXhosa, isiZulu, Ngemba, Norwegian, Norwegian Sign Language, Sepedi, Persian, Polish, Portuguese, Russian, Sesotho, Setswana, Shanghainese, Shona, Siswati, Spanish, Swahili, Swedish, Tamil, Tshangana, Tshivenda, Upper Ngemba, Yoruba, and Xitsonga. Note that here we list languages as they were listed by participants without attempting to distinguish between languages, dialects, and varieties.