2. Previous attitudinal research on non-standard varieties and foreign accents

A longstanding tradition of language attitudinal research has demonstrated that non-standard varieties and foreign accents are typically evaluated differently, and often more negatively, than norm-conforming speech. Standard language use is typically associated with ‘Superiority’ (also often called ‘Competence’), scoring high on features such as intelligence, status and prestige, whereas non-standard varieties are mostly downgraded on such traits (Zahn & Hopper Reference Zahn and Hopper1985, Garrett Reference Garrett2010). In addition, foreign accents are mostly judged more negatively than native accents: while native speakers of non-standard varieties often do score well on ‘Solidarity’ (or ‘Social attractiveness’), characterised by features such as warmth or likeability, foreign-accented speech is often downgraded on both Superiority and Solidarity (Gluszek & Dovidio Reference Gluszek and Dovidio2010, Fuertes et al. Reference Fuertes, Gottdiener, Martin, Gilbert and Giles2012, Dragojevic et al. Reference Dragojevic, Giles, Beck and Tatum2017).

These associations have mostly been laid bare in so-called speaker evaluation experiments (SEEs), which often make use of the classic matched-guise paradigm (Lambert et al. Reference Lambert, Hodgson, Gardner and Fillenbaum1960), in which listener-judges evaluate audio clips of different varieties or accents on a number of features pertaining to the personality of the speaker. The term ‘matched guise’ refers to the fact that the different speech fragments (or ‘guises’) are produced by a single (bilingual or bidialectal) speaker, ensuring that no other features (such as voice quality) influence the ratings. Additionally, the features on which the speakers are rated are selected to evaluate their personality and not their speech, in order to keep respondents ignorant of the experimental goal. While most SEEs typically present the participants with decontextualised speech fragments, the matched-guise design (or variants of it, such as the ‘verbal guise’ design, in which more than one speaker produces the speech clips) has also been deployed successfully in more contextualised experimental designs, in order to pinpoint potentially adverse consequences of a negative bias vis-à-vis non-conforming varieties or accents in specific settings. SEEs evoking recruitment contexts, for instance, have demonstrated that job candidates with regional and foreign accents are generally considered less hireable than standard and native-accented candidates (Spence et al. Reference Spence, Hornsey, Stephenson and Imuta2024).

Apart from employment or housing (Heylen & Van den Broeck Reference Heylen and Van den Broeck2016, Ghekiere et al. Reference Ghekiere, Martiniello and Verhaeghe2023), education stands out as another context in which negative attitudes towards linguistic varieties or accents may have palpable consequences for their speakers. A substantial body of research conducted on higher education in the United States has revealed an aversion of students to foreign-accented university instructors. Rubin & Smith (Reference Rubin and Smith1990) conducted a matched-guise experiment in which US undergraduates rated one version of a lecture on a humanities or science topic, delivered by a speaker with either a moderate or strong Chinese accent. These speech clips were paired with photographs of either a white or an Asian woman. Rubin & Smith found that level of accent did not affect perceived accentedness, nor did it affect comprehension of the lecture or the perception of the instructor’s teaching ability. The latter, however, was predicted by the degree to which students believed the speech samples were accented: the higher the level of perceived accentedness, the poorer the judgements of teaching ability. In a further exploration of non-language factors affecting undergraduates’ judgements of non-native English-speaking instructors, Rubin (Reference Rubin1992) had students rate speech clips produced in Standard American English by a native speaker, accompanied by images of either a white or an Asian instructor. Interestingly, Rubin found that the mere visual cue of an Asian ethnicity led to perceptions of the accent as foreign (while it was in fact native), and to diminished listening comprehension.

These findings indicate that negative perceptions of an instructor’s language proficiency and teaching ability can be influenced not only by a foreign accent, but also by ethnic stereotypes triggered by cues unrelated to language. Studies deploying other than experimental methods, such as corpus linguistic and discourse analytical methodologies, are generally in line with the experimental findings mentioned earlier. In a mixed-methods investigation, Subtirelu (Reference Subtirelu2015) compared students’ scores and comments of mathematics instructors with US and Chinese or Korean last names on RateMyProfessors.com (RMP). He found that instructors with Chinese or Korean last names were rated significantly lower on clarity and helpfulness. RMP users moreover frequently commented on the language proficiency of the Asian instructors, but made no comments on the language of instructors with US names, and only gave extremely positive evaluations to instructors with US names. Even though Subtirelu’s analysis of the comments also highlighted the recurring use of formulae that countered the presupposition that a foreign accent is a problem (e.g. ‘Even though X has a foreign accent …’), he nevertheless interpreted his findings as indicative of the dominance of an ‘ideology of nativeness’ (Shuck Reference Shuck2006). This ideology naturalises native speech, categorises non-native speakers as national and racial outsiders, and forges iconic associations between race, language, and incomprehensibility. This interpretation contrasts with Bresnahan et al.’s (Reference Bresnahan, Ohashi, Nebashi, Ying Liu and Morinaga Shearman2002) findings, who in their SEE compared students’ evaluations of ‘intelligible’ and ‘unintelligible’ foreign accents with those of native accents. The results revealed that when intelligibility was high, foreign-accented teaching assistants were judged to be as attractive and competent as teaching assistants who were native speakers of American English.

The research tradition on language attitudes sketched above has predominantly focused on anglophone contexts, particularly the United States. For languages other than English, attitudes towards non-standard varieties and non-native accents are generally less well-researched when it comes to specific contexts. The available studies on accent bias in employment, for instance, focus solely on English and (to a much lesser extent) on German (Spence et al. Reference Spence, Hornsey, Stephenson and Imuta2024). This scarcity of contextualised experimental attitude research is also evident for Flanders, where the data for the present study were collected. While a considerable number of experimental studies have been conducted on attitudes towards standard vs. non-standard Dutch in Flanders (Impe & Speelman Reference Impe and Speelman2007, Ghyselen & De Vogelaer Reference Ghyselen, De Vogelaer, Kristiansen and Grondelaers2013, Grondelaers & Speelman Reference Grondelaers, Speelman, Kristiansen and Grondelaers2013, Rosseel et al. Reference Rosseel, Speelman and Geeraerts2019b), most of these studies have focused on decontextualised speech, with only a few exceptions (e.g. Vandekerckhove & Cuvelier Reference Vandekerckhove, Cuvelier, Pol Cuvelier, Meeuwis and Teck2007, who used speakers set in shopping contexts and in a movie theatre, and Zenner et al. Reference Zenner, Grondelaers, Rosseel, Speelman, Esselinckx and Rombouts2021, who investigate the context of speech therapy). Experimental studies on attitudes vis-à-vis foreign-accented Dutch in Flanders, and especially contextualised ones, are even scarcer (but see Rosseel Reference Rosseel2021, Van Puyvelde et al. Reference Van Puyvelde, Sarah Van Hoof and Plevoets2023).

Higher education is a context that particularly calls for attention, given the increasing internationalisation of academic institutions in Flanders. Flemish universities actively aim to recruit foreign staff, and currently 30.4% of all Flemish university staff (i.e. professors, teaching assistants, PhD students, and postdoctoral researchers) have neither Belgian nor Dutch nationality, which implies that for most of them Dutch is presumably not their native language. Foreign professors are required to master Dutch at CEFR level B2 within five years after obtaining their position at a Flemish university. Moreover, all teaching staff are subjected to end-of-term evaluations by their students, which may comprise questions about the instructor’s language use, and which may impact their career. It is therefore highly relevant to investigate Flemish students’ attitudes towards non-conforming Dutch produced by university instructors. Before we do so, we will further describe the sociolinguistic context in which our research is set in the next section.

3. Context: language attitudes in Flanders

The official linguistic norm in Flanders is Belgian Standard Dutch, the strictest form of which is devoid of any regional accent and can almost exclusively be heard on radio and television, and more particularly in the news and current affairs shows. Linguists therefore often refer to this variety with the synonyms ‘VRT Dutch’ (VRT being the abbreviated name of the Flemish public broadcasting corporation; Grondelaers et al. Reference Grondelaers, van Hout and van Gent2016) or ‘Newscast Dutch’ (Plevoets Reference Plevoets2008). In the postwar decades, VRT Dutch was vigorously propagated as the unequivocal linguistic standard for oral communication within the Flanders region, as a result of which it has been argued that Flanders is today still characterised by a powerful standard language ideology (Jaspers & Van Hoof Reference Jaspers and Van Hoof2013). The perceived superiority of VRT Dutch has been confirmed by speaker evaluation experiments, which have laid bare fairly consistent associations of its speakers with competence, intelligence, and professionalism (Impe & Speelman Reference Impe and Speelman2007, Vandekerckhove & Cuvelier Reference Vandekerckhove, Cuvelier, Pol Cuvelier, Meeuwis and Teck2007, Ghyselen Reference Ghyselen2009). Even though this variety is ideologically powerful and considered the norm not only in broadcasting but also in education, linguists have also characterised it as ‘virtual’, since it is rarely spoken outside broadcast contexts. Grondelaers & van Hout (Reference Grondelaers and van Hout2011) have argued that the closest non-virtual variety to the VRT norm is one exemplified by teachers, which they have therefore dubbed ‘Teacher Dutch’ (see also Delarue & Ghyselen Reference Delarue and Ghyselen2016). Indeed, teachers’ speech is usually (though certainly not always: see below) fairly standard when it comes to its grammatical features, but it typically deviates from the strict VRT norm through the presence of an easily identifiable regional accent.

In general, hybrid colloquial styles are far more common in oral communication than Standard Dutch. These colloquial styles encompass a range of linguistic practices often subsumed under the term tussentaal (literally ‘in-between-language’) or ‘Colloquial Belgian Dutch’. This variety is readily distinguishable from Belgian Standard Dutch, not only because of the presence of stronger regional accents, but also due to morphological and syntactic deviations from the prescribed standard. Colloquial Dutch is widely used, mainly in informal, private, and unofficial domains, but is also often observed to permeate in more formal, public, and mediatised domains, such as education (Delarue Reference Delarue2016, Van Lancker Reference Van Lancker2017) and entertainment genres on radio and television (Van Hoof Reference Van Hoof2018). In public discourse Colloquial Dutch often faces disapproval, its perceived ‘omnipresence’ causing regular outcry about the poor quality of Flemish language use. Nevertheless, SEEs have demonstrated that Colloquial Dutch carries associations of Solidarity and Dynamism: its speakers are perceived as likeable, cool, entertaining, and trendy (Impe & Speelman Reference Impe and Speelman2007, Grondelaers & Speelman Reference Grondelaers, Speelman, Kristiansen and Grondelaers2013, Rosseel Reference Rosseel2017).

Whereas associations of Standard Dutch with Superiority and Colloquial Dutch with Solidarity and Dynamism traits have been demonstrated repeatedly by experimental studies that used fairly decontextualised stimuli, it remains an open question whether the same associations can be found in more contextualised experimental settings. Indications that these associations may indeed be different are provided by Zenner et al. (Reference Zenner, Grondelaers, Rosseel, Speelman, Esselinckx and Rombouts2021), who conducted an SEE in which speech therapy students and a control group evaluated a speech therapist producing both therapeutic discourse and more relational, informal discourse in either Standard Dutch or Colloquial Dutch. Whereas the authors had expected an upgrading of Standard Dutch in the therapeutic context and an upgrading of Colloquial Dutch in the relational context, they found that in both groups of participants, the standard guise did not elicit significantly higher scores than the colloquial guise, and the standard-speaking speech therapist was downgraded in the relational context on what the authors termed ‘modern competence’ (comprising features such as ‘enthusiastic’ and ‘funny’). The authors interpreted these results as revealing ‘some ideological relaxation’ of the standard language expectation that the speech therapy session is generally assumed to entail (Zenner et al. Reference Zenner, Grondelaers, Rosseel, Speelman, Esselinckx and Rombouts2021: 13).

For the context of education, one might hypothesise that, in line with the prevailing standard language ideology and similar to the speech therapy context, students will expect their instructors to use Standard Dutch and will penalise non-conforming speech. However, a recent study that deployed a contextualised speaker evaluation design, aiming to enhance the ecological validity of the experimental data, failed to confirm this expectation. Lybaert et al. (Reference Lybaert, Van Hoof and Deygers2022) investigated the influence of language variety (Standard vs. Colloquial Dutch), ethnicity (Flemish vs. Moroccan name), and religious affiliation (wearing a hijab vs. not wearing one) on students’ evaluation of a female university instructor, by means of a between-subjects SEE, in which students evaluated one of six guises on several evaluative dimensions. The guises consisted of a clip of a lecture produced by a native speaker (either in the standard or the colloquial version), combined with a photograph of a woman (with or without hijab) and a name (Flemish or Moroccan). Whereas the standard guises were mostly (but not consistently) rated higher on comprehensibility than the colloquial guises, Standard Dutch was not consistently rated more positively on professionalism (i.e. features such as having a successful career, intelligence, credibility, and reliability) than Colloquial Dutch. Moreover, language variety did not have any impact on the instructor’s perceived authority, or on social attractiveness. These findings challenge the assumption that the use of Standard Dutch is an absolute prerequisite for university instructors in the higher education context. They suggest that, while decontextualised experiments may readily elicit stereotypes in line with a standard language ideology, such beliefs may have a more limited impact on judgements of an instructor’s professional competences in a contextualised experimental design. As for the impact of ethnicity, no general downgrading of the minority identities was observed. Instead, the veiled standard-speaking Maghrebi instructor was rated more positively on professionalism and received the highest scores for the perceived standardness of her speech. The authors hypothesised that the ‘bonus’ for the veiled standard speaker may have been the result of a positive breach of participants’ low expectations about this stigmatised minority identity. Finally, the study failed to replicate Rubin’s (Reference Rubin1992) finding that ethnic cues may lead to the perception of a foreign accent: the participants of Lybaert et al. (Reference Lybaert, Van Hoof and Deygers2022) did not indicate hearing a foreign accent in any of the guises.

Neither a non-conforming ethnic and religious identity nor the use of a non-conforming language variety thus had a negative impact on the evaluation of an instructor’s teaching ability in Lybaert et al. (Reference Lybaert, Van Hoof and Deygers2022). However, it remains an open question whether and how a foreign accent influences the perception of such a speaker. The present research therefore partly replicates the methodology employed by Lybaert et al. (Reference Lybaert, Van Hoof and Deygers2022), but incorporates accent as an independent variable, in addition to language variety and name, as a cue to ethnic identity.

4. Methodology

This study is a contextualised SEE aiming to examine the combined effect of language variety, accent and name on students’ evaluation of university instructors. The experiment followed a 2 (Variety: standard vs. colloquial) × 2 (Accent: Flemish vs. Eastern European) × 2 (Name: Flemish vs. Eastern European) between-subjects design in which eight groups of students and recent graduates each evaluated a single lecture delivered by a lecturer presented to them in different guises. This section describes the verbal and visual input (Section 4.1), procedure and measurements (Section 4.2), participants (Section 4.3) and analysis (Section 4.4).

4.1 Materials

4.1.1 Auditory input

For the auditory input, we constructed an excerpt of a lecture on Pavlov’s dog and the principles of classical conditioning. We deliberately opted for a topic that we believed would be comprehensible to students with diverse educational backgrounds and varied interests. The script for the lecture was derived from an authentic class delivered to first-year medical students at a Flemish university. We created four versions of the script:

1. (1) Standard Dutch with a Flemish accent

2. (2) Colloquial Dutch with a Flemish accent

3. (3) Standard Dutch with an Eastern European accent

4. (4) Colloquial Dutch with an Eastern European accent

These four versions were syntactically identical, some small differences caused by the spontaneous rendering of the text left aside. All the speech samples consisted of approximately 420 words and had a duration of approximately 2 minutes and 30 seconds. They were rendered by a female actress who is a bilingual speaker of Dutch and Russian. Originally from Uzbekistan, she relocated to Antwerp, Belgium at the age of 14. We selected a Russian speaker because Russian is one of the most widely spoken languages in Europe, and Russians are in the top 15 of non-Belgians with European citizenship in Belgium.Footnote ¹ Additionally, there has been an increase in the population of Eastern European migrants in Flanders, a trend largely influenced by the ongoing war in Ukraine.Footnote ²

The difference between the standard and colloquial versions lay in the rendering of a number of morphological and phonological variables: in the Standard Dutch versions the speaker consistently used the Standard Dutch variants of those variables; in the colloquial versions she systematically used non-standard variants, which are commonly associated with Colloquial Dutch (we refer to the online supplementary materials for a transcript of the scripts and an overview of the variants included in the stimuli; see Geeraerts & Van de Velde Reference Geeraerts, Van de Velde, Hinskens and Taeldeman2013 for a discussion of Colloquial Dutch variants). The speaker delivered the Flemish accented versions with a mild regional accent from the central, Brabant–Antwerp region. The decision to incorporate a regional accent aligns with prior research findings indicating that Flemish teachers often have accents that disclose their regional backgrounds, rather than conforming strictly to the accent norms commonly displayed by television and radio news presenters (Grondelaers & van Hout Reference Grondelaers and van Hout2011). The foreign-accented versions were delivered in the speaker’s native Russian accent. We will refer to this accent as ‘Eastern European’ in the remainder of this paper, because this is how most participants perceived it (see Section 4.2).

4.1.2 Visual input

Names were used to operationalise instructor ethnicity. To represent a Flemish identity, we selected the first name ‘An’, which is a very common first name given to women in Flanders, and combined it with the surname ‘Mertens’, which ranks as the fifth most prevalent surname. For the Eastern European identity, we used the Eastern European equivalent ‘Anna’ as the first name, and chose the surname ‘Mirzayeva’ from a list of common Uzbek last names.Footnote ³

To gauge the likelihood of participants associating these names with either Flemish or Eastern European origins, we conducted a pretest: 38 participants indicated on a scale from 1 to 7 how likely it was to them that people with these names have Eastern European or Flemish parents. The analysis of the results revealed a strong association between the name ‘An Mertens’ and Flemish ethnicity (M = 6.4, Md = 7), and a considerably lower likelihood of the name being perceived as indicative of Eastern European heritage (M = 2.3; Md = 2). Conversely, the name ‘Anna Mirzayeva’ was primarily associated with an Eastern European background (M = 5.7; Md = 6), rather than being considered of Flemish parentage (M = 2.8; Md = 3). We combined the two names with a photograph of a female lecturer. From the website Shutterstock we selected a photo of a woman who could plausibly be associated with both a Flemish and an Eastern European background. We used the same photograph in all scenarios to eliminate any potential influence of factors such as attractiveness, charisma, or age (Riniolo et al. Reference Riniolo, Johnson, Sherman and Misso2006, Baert & Decuypere Reference Baert and Decuypere2014). A pretest with the same pool of participants as those who rated the names revealed that the woman in the picture was considered about equally likely to have Eastern European (M = 4; Md = 4) or Flemish parents (M = 4.2; Md = 4).

We thus created two fictional female university instructors representing the tested identities (depicted in Figure 1). These two identities were combined with the four audio recordings in a total of eight conditions.

Figure 1. Identities tested.

4.2 Procedure and measurements

The experiment was conducted using the online survey platform Qualtrics. The participants were randomly assigned to one of the eight conditions. In the introductory section of the online survey, we informed the participants that the survey was part of a study focusing on teaching quality. The speaker was introduced as an instructor (‘docent’) who gave a lecture to undergraduate students in medicine, and participants were told that the accompanying PowerPoint presentation she used during her lecture would also be displayed. The participants were instructed to pay careful attention as they would subsequently be presented with questions concerning the instructor and the lecture. Prior to listening to the speech sample, participants were also required to provide their informed consent. Throughout the lecture, a photograph of the instructor, accompanied by her name, was projected on the left side of the screen, while the PowerPoint presentation, displaying the key points of the lecture, was shown on the right side (see Figure 2).

Figure 2. Projection of photograph, name, and PowerPoint presentation (Condition 1).

Immediately after seeing the recorded lecture, the participants were directed to the questionnaire. Participants had to respond to 23 Likert statements on a 7-point scale ranging from ‘completely disagree’ to ‘completely agree’ (see Table 5 for an overview of all the statements, and the online supplementary materials for the original Dutch statements). The Likert statements included were retrieved from Lybaert et al. (Reference Lybaert, Van Hoof and Deygers2022), with some modifications. In the latter study, the statements represented the evaluative dimensions Social attractiveness, Professionalism, Comprehensibility, Authority, Physical attractiveness, and Standardness. Since none of the statements explicitly addressed the language variety of the instructor, most of them can be regarded as indirect measures of language attitudes. Some statements assessing comprehensibility (i.e. ‘I think this instructor speaks clearly’, ‘I have understood the explanation well’, ‘I think this instructor explains the matter clearly’) can be considered semi-direct measures, as they could be interpreted as reflecting the instructor’s speech style. We also included a control question to make sure that participants were actively engaging with the statements during their responses. In the control question, participants were instructed to indicate the response option ‘rather agree’.

On the following web page, we verified awareness of the research purpose by means of an open question asking what the study was about. The subsequent page included 12 Likert items focusing on the instructor’s language and appearance (see Table 5) and another control question. The items related to language use can be categorised as direct measures of language attitudes. Additionally, participants were prompted to indicate their perception of the instructor’s country of birth (either ‘Belgium’ or ‘other’). If they believed the instructor was born outside Belgium, they were asked to specify the country. Furthermore, participants were requested to identify the instructor’s native language and indicate the Flemish province where they believed she resided. Afterwards, we included a second open question on the purpose of our study and a manipulation check to verify if our participants remembered the name of the lecturer correctly.

Finally, the questionnaire included a section gathering demographic information about the participants, i.e. gender, place of origin, nationality, and home language(s). The questionnaire was presented to the participants across consecutive web pages, and once they advanced to a new page, they were unable to revisit the previous page.

4.3 Participants

The participants consisted of students and recent graduates who had completed their studies in 2018 or more recently at Flemish universities or university colleges. They were recruited by marketing agency Bilendi. A total of 393 participants successfully completed the survey (see Table 1). These participants were paid for their participation. Additionally, 562 participants started the survey but were excluded because they graduated prior to 2018 (43%) or because they answered incorrectly to one of the control questions (23%). Participants who failed to complete the entire questionnaire (34%) were also excluded from the dataset.

Table 1. Eight conditions

The geographical distribution of our participants spanned the whole of Flanders, with 19% residing in West Flanders, 23% in East Flanders, 28% in Antwerp, 12% in Limburg, and 19% in Flemish Brabant. 37% of the participants identified as ‘female’, 63% as ‘male’, 0.3% indicated ‘other’, and 0.3% chose not to disclose their gender identity. 93% held Belgian citizenship; 4.3% had a dual nationality, which included Belgian citizenship, while 2.5% did not have Belgian nationality. 86% exclusively spoke Dutch at home, 11% used both Dutch and another language in their home environment, and 3% had a different primary home language.