Aicha Is More Dutch but Less Dynamic than Ahmed: The Gendered Nature of Race in the Netherlands

ABSTRACT

In this article we rely on accent evaluation to test the ‘intersectional invisibility hypothesis’¹ that social cognition about men (but not women) is overrepresented in group-level beliefs.² As a case in point, we investigate the evaluation of male and female Moroccan accents to gain insight into impression formation of Muslims in the Netherlands, and to find out whether stereotypical qualities associated with Moroccan-Dutch people, such as aggressive, macho, and criminal, are in fact associated with Moroccan-Dutch men. Two matched-guise experiments featuring regional and ethnic accents of Dutch (one with male speakers, one with female speakers) confirm the intersectional invisibility hypothesis, but the inclusion of traditional and modern prestige measures in accent evaluation research results in arguably richer stereotype and prejudice accounts, and in this sense, the present investigation adds nuance and shade to Gloria Wekker’s (2016) pessimistic account of racism in Dutch society. Male Moroccan-Dutch speech is strongly deprecated and is always deemed inferior to indigenous speech; at the same time it is also found to be the most dynamically prestigious of all accents. Female Moroccan-Dutch speech does not engender extreme reactions, but it is not regarded as dynamic either and a long way from being accepted as indigenous speech.

KEYWORDS:

• Intersectionality
• gendered race
• stereotyping
• accent attitudes
• conservative prestige
• modern prestige

Introduction

In 2014, on the evening of the municipal elections in the Netherlands, the Dutch right-wing politician Geert Wilders shocked the nation when, upon hearing the election results, he asked a gathering of party members ‘Do you want more or fewer Moroccans’?³ This reprehensible query is not just a low point in racial prejudice in the Netherlands, it also lays bare a little-noted correlation between race and gender. In a post-hoc justification of his rant, Wilders declared to have targeted not all the Dutch of Moroccan descent but only (criminal) Moroccan-Dutch men. The idea that it is especially men who are the carriers of negative Moroccan-Dutch stereotypes is also confirmed in linguistic accent evaluation research. Grondelaers and Speelman extracted free response reactions to language labels such as ‘Dutch with a Moroccan accent’, and found that participants frequently used qualifications that specifically target Dutch men of Moroccan descent, such as asocial, hard, fast, macho, aggressive, or criminal.⁴

A socio-psychological explanation which was proposed in 2008 for this correlation between race and gender is the so-called ‘intersectional invisibility hypothesis’⁵: societal standards of heterocentrism, ethnocentrism and especially androcentrism (the idea that a typical person is male) foster conceptualizations of the prototypical person as a heterosexual White male. As a result, ‘social cognition about men (and not women) may be overrepresented in group-level beliefs’.⁶ Crucially, a multitude of studies suggest that while Blacks are more readily assigned a male gender, East Asians are more often assumed to be women.⁷ By contrast, the correlation between gender and Arab/Muslim ethnicity has been given little or no attention by researchers. A notable exception is Ghavami and Peplau, who found that stereotypes of Middle Eastern Americans (the top three being terrorists, dark-skinned, oppress women) were more in line with stereotypes of Middle Eastern American men (bearded, dark-skinned, terrorists) than with stereotypes of Middle Eastern American women (quiet, covered [i.e. veiled], religious).⁸

In this article, we will rely on linguistic markers of ethnic group membership to test the hypothesis that stereotypes of Moroccan-Dutch people in general are based on stereotypical perceptions of Moroccan-Dutch men. We will present two speaker evaluation experiments (also known as matched-guise experiments),⁹ in which respondents had to listen to unlabelled audio clips representing a range of accent varieties and evaluate a number of aspects of each speaker’s personality (this speaker is nice, smart, educated, hip, …).

Compared to the unrestricted elicitation of stereotypes mentioned in the previous paragraphs, our speech-based extraction technique has two main advantages. The first advantage is that by building on descriptors pertaining to the personality of the speakers rather than to the quality of their speech, we can keep respondents unaware of the experiment’s actual objective. As a consequence, ‘deeper’, less conscious and more meaningful, evaluations will be collected, which cannot be obtained through direct elicitation. A crucial finding from previous studies with this experimental set-up showed that non-standard varieties which are publicly deprecated in terms of traditional prestige traits (highly educated, professionally competent, serious, …) may subconsciously be upgraded in terms of modern prestige traits hip, trendy, urban, streetwise, or cool. A prerequisite for the experiment to access these ‘deeper’ evaluations is that respondents should be unaware that they are evaluating language varieties. For instance, by keeping his adolescent respondents unaware of his linguistic goals, Kristiansen found evidence for the dynamic upgrading of Københavensk (the working class dialect of Copenhagen) in Denmark.¹⁰ Since then, similar studies have confirmed the correlation between non-standard or emergent varieties and modern prestige evaluations in Flanders,¹¹ Ireland,¹² and the Netherlands.¹³ It is reasonable to expect that any potential positive evaluation of the Moroccan accent will also emerge on this modern prestige dimension.

The second advantage of our reliance on speech markers is that we can test variable degrees of group membership by manipulating the broadness of the accents in our experiments. Label stimuli are categorical and static indicators of group membership: one is, or one is not a member of, for instance, the group labelled ‘Middle Eastern American’. The strength of a regional or ethnic accent, by contrast, often reflects the extent to which speakers profile themselves as group members. To be sure, accent is not a commodity a speaker can fully manipulate: getting completely rid of a regional or ethnic speech flavour is a challenge for most untrained speakers. Manipulating the broadness of a regional or ethnic accent in specific social contexts, however, is a feasible and effective tool to implement the degree to which one is a member of a particular social group. Crucially, a stronger accent may be indicative in this light of how strongly a person identifies with the group represented by this accent.

Since Giles, it is well known that as accents get lighter, they are more likely to be deemed prestigious.¹⁴ Grondelaers, Van Hout and Van Gent discovered that in the Netherlands this correlation is conditioned by the prestige associated with the accent itself.¹⁵ The accent of the Randstad, the affluent urbanized area in the west of the country which includes the cities of Amsterdam, Rotterdam and The Hague, is deemed a prestigious accent, and its prestige is unaffected by accent strength. Accents from the more rural areas in the northeast (Groningen) and southeast (Limburg) areas of the country are deemed less prestigious, but the strength of these accents influences the degree to which they are regarded as less prestigious, both in terms of traditional prestige (Superiority) as well as modern prestige (Dynamism). In light of these findings, we expect that it will be the milder versions of the Moroccan accent that engender the less negative speech and speaker evaluations.

In the next section, we will first review the available research on the evaluation of the Moroccan accent in the Netherlands. This will be followed by a presentation of our research questions and hypotheses. Section 3 will explain our research methods and discuss the results of two experiments, while section 4 reports our findings in relation to the research questions and hypotheses. We conclude with an exploration of some theoretical and practical ramifications of our research results in the final section.

The Perception of Moroccan-Dutch People and the Moroccan Accent in the Netherlands

In 2021, 2.4% of the 17,475,415 inhabitants of the Netherlands were people with a first- or second-generation Moroccan migration background;¹⁶ third-generation migration background is not tracked in the same way. In the 1960s and 1970s, Moroccan immigrants were attracted to the Netherlands as temporary labourers (so-called gastarbeiders). As a result of the government policy of speedy integration and the fact that until the late 1990s, Dutch society was comparatively open-minded and multicultural,¹⁷ with a general aversion to (extreme) right-wing politics,¹⁸ the socio-political context was initially conducive to the acceptance of ethnic minorities.

However, the perception of (Muslim) people with a migration background changed considerably in the initial years of the twenty-first century. The 9/11 attacks and terrorist actions in Europe negatively influenced the perception of Muslims in the West.¹⁹ At the same time, populist right-wing politics became more influential in the Netherlands, initially with Pim Fortuyn, the first Dutch politician who was outspoken about the danger of ‘Islamization’ of Dutch society.²⁰ The assassinations of Fortuyn in 2002 and Islam-critical filmmaker Theo van Gogh in 2004 further fanned the flames, increasing the nation-wide support for (populist) right-wing politicians such as Rita Verdonk and Geert Wilders.²¹ As a result, Muslims became ‘highly stigmatised’ in Dutch society.²²

While these socio-political developments suggest a (very) negative assessment of the Moroccan-Dutch identity, there are also developments which may contribute to a more positive view of Muslims in the Netherlands. Since the new millennium, education levels among minorities have been rising relatively fast,²³ which has had a positive impact on their socio-economic status.²⁴ In addition, second-generation immigrants are less inclined to move to neighbourhoods with high migrant concentrations,²⁵ and intermarriage rates are decreasing,²⁶ while unmarried cohabitation among young Muslims is increasing.²⁷

Moreover, the Netherlands has seen Muslims rise to positions of high social status. Examples include people such as Ahmed Aboutaleb, mayor of Rotterdam (the second-largest city in the Netherlands), or Ahmed Marcouch, mayor of Arnhem (twelfth-largest city in the Netherlands), but also highly-regarded literary authors such as Hafid Bouazza or Abdelkader Benali. Their Moroccan background does not preclude them from being accepted as prestigious members of Dutch society. Dutch Muslims have also acquired modern prestige in the popular arts: cases in point are singers (Hind), stand-up comedians (Najib Amhali), athletes (Ibrahim Afellay), and actors (Touriya Haoud). The most notorious rise to glory of a Dutch Muslim is Boef, an Algerian-French-Dutch rapper who epitomizes the growing appreciation for Arab badass cool. This appreciation is also evident in the appropriation of Moroccan speech features by Surinamese rappers to project a tough and streetwise image, an association that the Surinamese accent does not have.²⁸

While the available evidence with respect to the social integration and evaluation of Moroccan-Dutch people seems to be inconclusive, the linguistic evidence for the most part shows negative evaluations. In free response experiments carried out in 2011 and 2017, Grondelaers and Speelman asked a sample of highly educated Dutch people to list the three adjectives that first came to mind when confronted with labels of regional and ethnic varieties of Dutch.²⁹ As explained in the introduction, adjectives pertaining to a Moroccan-Dutch individual were invariably negative; this was also the case for keywords which referred to the Moroccan accent (for instance unintelligible, ugly and unclear). Even in studies which carefully hide their research aims (in order to collect deeper attitudes), migrant accents are harshly judged, in and beyond the Netherlands. For example, Torstensson examined native speaker reactions to the L2-speech of Turkish-Danish adolescents in Sweden, a country which self-identifies as the most open to immigrants in all of Western Europe. Even so, Torstensson found that the speech of immigrants with a Muslim background was evaluated more negatively than the speech of immigrants with a Western Christian background.³⁰ Speech samples that were incorrectly ascribed to a speaker with a Muslim background were judged as harshly as genuine Muslim migrant speech.

Two speaker evaluation experiments which specifically targeted the evaluation of the Moroccan accent produced conflicting results. In Grondelaers, Van Gent and Van Hout,³¹ 212 male and female listeners were asked to rate eight short clips of spontaneous speech produced by indigenous³² and Moroccan-Dutch men on twelve traits that were chosen to ascertain five aspects (called dimensions) of accent evaluation; these aspects included Superiority and Dynamism, as mentioned before, but also Integrity, Accent Norm, and Aesthetic Evaluation.³³ Subsequent statistical analysis (through dimension reduction),³⁴ however, showed that the traits that had been selected revealed no more than two underlying dimensions of evaluation: Status (Superiority + Accent Norm) and Attractiveness (Integrity + Dynamism). On both the Status and Attractiveness dimensions, the Moroccan accent was systematically graded below the average. Crucially, the dimension that was most likely to generate favourable evaluations, viz. Dynamism, did not emerge as a separate entity.

In an earlier account of the experiment reported in the present article, Grondelaers and Van Gent addressed some concerns in the experimental setup that may have obscured or blocked Dynamism-dimensions in the experiment.³⁵ They introduced the following improvements to the design: the inclusion of three levels of accent strength in the speech clips, and a selection of new traits for Dynamism specifically accommodated to the evaluation of ethnic accents. They found that while mildly-accented Moroccan speech was deemed less inferior than its strongly-accented counterpart (though not as superior as indigenous speech), any type of Moroccan-accented speech was found to be the most dynamic of the speech varieties featured in the experiment.

In order to investigate to what extent these findings could be extended to female Moroccan-Dutch speakers, we set up a new experiment designed to answer three research questions:

1. To what extent does the evaluation of a female Moroccan accent match that of a male Moroccan accent? More specifically, how plausible is it that for Moroccan-Dutch women we will find the same complementary relation between low Superiority ratings and high Dynamism evaluations that characterizes the evaluation of Moroccan-Dutch men? What little evidence there is on the evaluation of female Muslims, notably Ghavami et al., reveals that the stereotypes most frequently listed for Middle Eastern Women in the United States are quiet, religious, covered (i.e. veiled), oppressed, and conservative³⁶: these traits are unlikely to give rise to the kind of dynamic and streetwise prestige we found for the male Moroccan speakers.

2. Can accent evaluation evidence be used in support of the intersectional invisibility hypothesis? Is the evaluation of Moroccan-Dutch as a minority group based on the stereotyping of male or female Moroccan-Dutch people? Building once more on the data in Ghavami et al.,³⁷ we predict a strong alignment between general Moroccan-Dutch stereotypes and the evaluation of Moroccan-Dutch men.

3. What do our data suggest with regard to how prejudices against male and female Muslims in the Netherlands are evolving? What do they reveal about key concepts such as ‘identity’ and ‘integration’, and how do these correlate with gender?

Evaluating Male and Female Regionally and Ethnically Accented Speech

Method

This study uses the experimental technique of speaker evaluation or matched-guise experiments pioneered by Lambert et al. in 1960, in which participants evaluated unlabelled speech clips on a number of traits that measure specific aspects of evaluation.³⁸ We carried out a new experiment with a design that was near-identical to the setup of the experiment reported in Grondelaers and Van Gent,³⁹ albeit that in the present experiment, female speech was used. For logistical reasons it was not possible to include both genders in a single experiment: the total length of an integrated experiment would have greatly exceeded what could be expected from unpaid respondents, but we also needed to adjust some traits to the extraction of gendered stereotypes.

Speech Clips

Both the experiment with male and female speakers included eighteen speech clips, all of which consisted of an audio recording of one out of nine short sentences taken from the corpus Sprekend Nederland.⁴⁰ These single-sentence clips were only three to eight seconds long but they were edited to contain two identical deliveries of each sentence, with a two-second pause in between. Each audio clip started with a spoken (fictitious) first name introduction of the speaker (‘you will now listen to Name’), delivered in a regionally neutral Dutch accent.

Both experiments contained three groups of speakers, namely six indigenous persons from the Randstad area,⁴¹ six indigenous persons from the southernmost province of Limburg, traditionally associated with rural but friendly stereotypes and a low-prestige accent, and six persons with a second- or third-generation Moroccan background. In each of these three groups, two clips represented speech with no discernible accent, two clips featured a speaker with a mild accent, and in two clips a strong accent could be heard.

Speech clips for the Moroccan-Dutch speakers were recorded specifically for these experiments: they were read by second- or third-generation Moroccan-Dutch men and women who were recruited in the streets of Gouda (South-Holland), the town with the largest Moroccan-Dutch community in the Netherlands. Because these Moroccan-Dutch speakers were recruited in the Randstad area with the high-prestige accent, we can be fairly certain that any potential negative judgment exclusively depends on their ethnicity. Audio clips produced by the indigenous Dutch speakers were previously used in Grondelaers et al. (2019)⁴²; they had been sourced from Sprekend Nederland as well.⁴³ An initial selection of potential speech clips was entered in two pre-experiments (one for the male, one for the female speakers) in which expert listeners, bachelor students of linguistics, were asked to determine the ethnicity and regional origin of the speakers, as well as the strength of their accent. Clips for the final experiment were selected on the basis of correct regional identification and on perceived accent strength (as measured on a seven-point strength scale); clips were chosen to maximize the difference between the broader and milder accents within each accent group.

All accent-free clips were produced by speakers who had received the lowest accent strength ratings in the pre-experiment for Grondelaers et al.,⁴⁴ and who, consequently, were poorly identifiable in terms of regional origin. In order to give accent-free clips an ethnically specific label (Moroccan vs. indigenous), all clips were preceded by a short statement in which a neutral voice introduced the speaker with a fictitious first name (‘you will now listen to Ahmed/Willem’). All introductions in the experiments were delivered by the same male speaker. Dutch names were taken from lists of popular baby names from the past fifteen years. Moroccan names were selected from a similar list approved by the Moroccan national government. Two remarks are in order here. In the absence of Dutch first names that unambiguously index the regional origin of the speaker, Randstad and Limburg speakers could not be distinguished in the accentless condition. For this reason, they were analysed as one group of speakers. In addition, it is crucial to understand that the samples of accent-free speech presented as spoken by a Moroccan-Dutch individual, were in fact spoken by an indigenous (i.e. ethnically Dutch) individual. This we did to ensure that a listener’s assessment could never be affected by potential traces of an ethnic accent.

Traits

Clips were rated on fifteen traits formatted as Likert statements and complemented with seven-point scales. Since Grondelaers et al.,⁴⁵ which included five Dynamism scales, had been unable to extract a separate Dynamism dimension from their ratings, we validated potential traits in a preliminary experiment; 251 respondents were asked to list the first three adjectives which sprang to mind in response to twelve photos from four types of environments, namely restaurants, workplaces, music concerts, and couches. For each type, there were pictures representing one of three subcategories associated with the three target evaluative dimensions, namely Superiority, Dynamism, and Personal Integrity. In the restaurant category, for example, a picture of the Ritz hotel dining room represented Superiority, a flashy but not exceedingly expensive-looking sushi bar represented Dynamism, and a cosy Dutch pancake house represented Integrity.

This preliminary experiment yielded a wealth of keywords in response to visuals associated with the three targeted dimensions of evaluation, viz. Superiority, Dynamism, and Integrity. For the actual experiment, we selected traits for each dimension that met the following conditions: first, they had to be frequent returns for a specific dimension, while being infrequent for the others; secondly, and crucially, they had to be applicable to humans (e.g. ‘tasty’ may be an appropriate response to a restaurant, but not to a person).

The five traits used to measure Superiority were used for both genders: deze persoon is chic (this person is posh), heeft goede cijfers gehaald (received good grades), kan goed leiding geven (has good leadership skills), heeft al veel geld verdiend (has made a lot of money), and heeft veel professionele ervaring (has a lot of professional experience).

To measure Integrity, four traits were used for both genders: deze persoon is aardig (this person is friendly), behulpzaam (helpful), eerlijk (honest) and heeft een warme persoonlijkheid (has a warm personality). The fifth trait zou een goede vader zijn (would be a good father) was reserved for male speakers, while zou een goede moeder zijn (would be a good mother) was used for female speakers.

In order to capture the streetwise form of Dynamism that we expected to correlate with the Moroccan accent, some Dynamism traits were sourced from Grondelaers and Speelman,⁴⁶ a free response experiment in which respondents produced keywords in response to language varieties represented as labels (see above). Dynamism was measured with three traits that were used for both genders: deze persoon is modern (this person is modern), hip (hip) and zelfverzekerd (has a lot of confidence). Two Dynamism traits were adapted according to gender: deze persoon is cool (this person is cool) and stoer (tough/macho) were used in the experiment with male speakers to capture masculine stereotypes of Dynamism. For female speakers, deze persoon is trendy (this person is trendy) and modebewust (has a good sense of fashion) were used to capture more feminine Dynamism stereotypes.

Respondents were also asked to identify the regional origin of the speaker by selecting from a dropdown menu one of six potential cities of origin. These six Dutch cities were of similar size (between 120,000 and 240,000 inhabitants): Haarlem and Leiden, both located in the Randstad, Eindhoven and Maastricht, both located in the south, and Groningen and Enschede, both situated in the north (east) of the country. This selection deliberately excluded the four largest cities of the country (Amsterdam, Rotterdam, The Hague and Utrecht), which are often associated with (negative) urban stereotypes.

Procedure

The experiment was implemented and presented on the LimeSurvey-platform, which allowed us to customize the randomization of clips to make sure that no two Moroccan-Dutch speakers were played consecutively (in an effort to conceal our research aims). Student assistants recruited online respondents via their private networks as well as on university campuses in two targeted regions, namely the Randstad and Limburg.

Respondents were asked to sign a declaration of consent, and provided some demographic data about themselves (gender, age, education, language and region of origin), which we used to check whether they were part of our target groups. Respondents who failed to meet the criteria were not allowed to continue, the rest was first presented with the practice clip, followed by the clips of the actual experiment in a randomized order. At the conclusion of the experiment, a debriefing question was included to check whether respondents had guessed our research objectives.

Respondents were required to have Netherlandic Dutch (or a local dialect of Dutch) as their native language, to be between eighteen and thirty years of age, to be studying or to have studied at a university (college), and to have grown up in either the western (Randstad) or the southern (Limburg or Brabant) parts of the Netherlands. Results from respondents who took more than two hours to complete the experiment (three respondents in the experiment with male speakers, seven in the experiment with female speakers), or who had guessed our research objectives (four in the male experiment, none in the female experiment) were excluded from our analysis.

Data from 107 respondents were included in the analysis of the male experiment. 58 respondents were from the south, 49 were from the Randstad. 60 respondents identified as female, 47 as male, and gender was balanced across the regions. Data from 91 respondents were included in the analysis of the female experiment. 56 respondents were from the south; 39 were from the Randstad. 51 respondents identified as female, 44 as male.

Results

We first carried out a principal component analysis (with Varimax rotation and selection criterion Eigenvalue > 1) to reduce the dimensionality in the ratings from 15 individual traits to a smaller number of underlying evaluative dimensions. For both experiments, the initial analysis revealed a factor solution with the three envisaged underlying dimensions Superiority, Dynamism and Integrity. In the male speaker experiment, confident and modern mapped onto both Dynamism and Superiority. After removal of these traits, the resulting solution with thirteen traits accounted for 73.7% of variance in the data. In the female speaker experiment, only confident mapped onto both Dynamism and Superiority, and was subsequently removed. The resulting solution with fourteen out of fifteen traits accounted for 72.7% of variance in the female speaker data. Crucially, both factor solutions were found to be stable over splits of the material.

The component matrices in Table 1 show how the fifteen traits mapped onto the three dimensions the software distinguished in the data, shown separately for the male and female experiment. Any score above .4 (shaded in grey) means that that trait statistically constitutes part of that dimension. The labels Superiority, Dynamism and Integrity were not given by the software but inferred by us based on which traits loaded on that dimension (i.e. had scores > .4).

Table 1. Rotated component matrices.

Variable	Male speakers			Female speakers
	Superiority	Dynamism	Integrity	Superiority	Dynamism	Integrity
Posh	.829	.148	−.024	.688	.376	.021
Good grades	.747	−.036	.380	.670	.212	.342
Management skills	.692	.304	.319	.742	.285	.209
Making money	.864	.167	.103	.850	.180	.096
Professional experience	.835	.124	.195	.842	.133	.216
Tough	.051	.885	.070	n/a	n/a	n/a
Cool	.173	.858	.184	n/a	n/a	n/a
Hip	.255	.817	.151	.185	.885	.149
Fashionable	n/a	n/a	n/a	.231	.887	.110
Trendy	n/a	n/a	n/a	.243	.867	.118
Modern	n/a	n/a	n/a	.313	.814	.145
Confident	n/a	n/a	n/a	n/a	n/a	n/a
Nice	.113	.125	.859	.089	.146	.860
Helpful	.132	.085	.854	.139	.084	.843
Warm	.085	.173	.831	.106	.065	.855
Good mother	.250	.154	.759	.213	.062	.758
Honest	.191	.039	.816	.187	.179	.720

For each clip, we computed a score for each of the three dimensions by taking the average of the traits which constituted part of that dimension (the grey-shaded traits in Table 1). Next, a statistical analysis was used to determine the significance and the size of the effects.⁴⁷ Importantly, neither the gender nor the regional background of the respondents had any impact on the evaluations in the male and female experiments.

Superiority

Statistical analyses showed that the Superiority ratings for both male and female speakers were affected by the speaker group as well as the strength of their accent.⁴⁸ Tables 2 and 3 show the average Superiority scores per speaker group and accent strength for the two experiments.⁴⁹ Combinations of speaker group and accent strength (i.e. guises) are sorted from highest to lowest rated. Guises which share a cell (such as ‘No or mild accent’ for the Moroccan accent in Table 2) did not significantly differ in their scores. Note that the no accent-versions of the Randstad and Limburg accents always share a cell, because they are indistinguishable. The grand mean, which is a reference-point to determine if evaluations are positive or negative, is computed by averaging over all the scores per gender.

Table 2. Superiority scores for male speakers in each guise.

Superiority	Urban West	Rural South	Moroccan
4.50	Any accent	No accent
4.05	Grand mean
3.99		Mild accent
3.86			No or mild accent
3.62		Broad accent
3.08			Broad accent

Table 3. Superiority scores for female speakers in each guise.

Superiority	Urban West	Rural South	Moroccan
4.51	No or mild accent	No accent	No accent
4.30	Broad accent
4.22		Mild accent
4.10	Grand mean
3.51			Mild accent
3.43		Broad accent	Broad accent

The present experiment corroborates some findings from earlier research:⁵⁰ it confirms the uncontested superiority of the Randstad accent, which is hardly affected by accent strength. Both the mild and especially the broader versions of the Limburg and Moroccan samples are harshly judged. By far the most noticeable difference between the male and female evaluations, is the fact that the unaccented male Moroccan guise leads to significant negative judgements: even without any discernible ethnic or regional accent, this guise is rejected purely on the basis of the Arabic name mentioned in the introduction. For the unaccented female Moroccan guise, this is not the case: identification of a female speaker as Aicha or Kadidja for example does not render her any less superior for the respondents than the unaccented indigenous speakers.

Dynamism

Statistical analyses likewise showed that both the speaker group and the strength of their accent affected the Dynamism ratings for both male and female speakers.⁵¹ Tables 4 and 5 show the average Dynamism scores per speaker group and per accent strength for the two experiments.⁵²

Table 4. Dynamism scores for male speakers in each guise.

Superiority	Urban West	Rural South	Moroccan
4.07			Any accent
3.81	No or mild accent	No or mild accent
3.79	Grand mean
3.59	Broad accent
3.07		Broad accent

Table 5. Dynamism scores for female speakers in each guise.

Dynamism	Urban West	Rural South	Moroccan
4.76	Broad accent
4.54	Mild accent
4.32	No accent	No accent	Any accent
4.23	Grand mean
3.90		Mild accent
3.36		Broad accent

Compared to the Superiority scores for the men, the male Dynamism scores reveal completely mirrored preferences: any Moroccan flavour is deemed significantly more dynamic than the indigenous accents, which for the most part are similarly evaluated, except for the (predictably) more extreme negative judgements of the broad Limburg guise. For the female speakers, Dynamism scores distinguish between a dynamic indigenous accent (Randstad) and a non-dynamic indigenous accent (Limburg). The female Moroccan accent is relatively neutral in terms of Dynamism, regardless of the accent strength.

Interpretation of the Results

The findings in the previous section allow us to answer and reflect on the three research questions put forward at the end of Section 2.

The evaluation of a female Moroccan accent in the Netherlands diverges markedly from the evaluation of the male Moroccan accent (Research Question 1). More specifically, the complementary relation between low Superiority ratings and high Dynamism evaluations that characterizes the evaluation of the male Moroccan accent is absent for Moroccan-Dutch women. To our knowledge, there are no research data available on the stereotyping of Moroccan-Dutch women in the Netherlands, but the stereotypes of American Middle Eastern women as recorded by Ghavami and Peplau (i.e. quiet, religious, covered, oppressed, and conservative)⁵³ are antithetical to a dynamic and streetwise type of prestige, which requires independence and assertiveness. If the public opinion on Muslims in the Netherlands is in any way similar to those American stereotypes, their comparatively low dynamism ratings are unsurprising.

The most startling difference between male and female Moroccan-Dutch evaluations, however, is the fact that a first name such as Ahmed or Mohammed, but not Aicha or Kadidja, generates rejection, even in the absence of any discernible accent. This asymmetry leads us to conclude that perfect linguistic integration in the form of accent-free Dutch suffices for Moroccan-Dutch women but not for Moroccan-Dutch men.

As far as the second research question is concerned, our data demonstrate that the evaluation of the Moroccan-Dutch minority group is based almost exclusively on the stereotyping of Moroccan-Dutch men. This observation is in line with Ghavami and Peplau’s findings on the convergence of Middle Eastern and Middle Eastern male stereotypes in the US.⁵⁴

Finally, our findings allow us to formulate a number of predictions about how prejudices against Moroccan-Dutch people in the Netherlands are developing (Research Question 3). A key concept in this respect is the tension between ‘integration’ and ‘identity’, and the extent to which these variables correlate with gender. If we conceive of Superiority as a category entrenched in demographic variables – birth, upbringing, education – which an individual can only partially control, and of Dynamism (whatever its incarnation, yuppie cool or urban toughness) as a conscious and deliberate stylization an individual can choose to adopt, then we can (re)interpret the evaluation of Moroccan-Dutch men and women as, respectively, ‘compensatory identity creation’ and ‘unsuccessful integration’. Moroccan-Dutch men may be rejected on traits they cannot control (like first name, skin colour) or traits they can only partially control (like ethnic accent), but they are arguably successful in carving out for themselves the sort of cool, urban and streetwise identity which increasingly matters in Late Modern societies.⁵⁵ Moroccan-Dutch women, by contrast, are ‘Dutch’ if they can completely suppress their Moroccan accent: any amount of ethnic flavouring immediately leads to a sort of depreciation which is only partially compensated by an ability (or desire?) to be dynamically prestigious in terms of urban cool. While Moroccan-Dutch women score above average in terms of Dynamism, and certainly higher than their Limburgian compatriots, their hipness ratings are not of a kind to compensate for their depreciation in terms of Superiority (as is the case for the male speakers).

Conclusion

In this article, we have presented the results of two virtually identical experiments to gauge the evaluation of male and female versions of the Moroccan accent in Dutch, which we regard as a powerful indicator of Moroccan-Dutch group membership. By studying Dutch attitudes towards Moroccan accents of men and women separately, we attempted to find evidence for the ‘intersectional invisibility hypothesis’⁵⁶ that stereotypes about men (and not women) are overrepresented in group-level beliefs.⁵⁷

Our findings strongly corroborate this intersectional invisibility hypothesis: the group-level image of Moroccan-Dutch people in the Netherlands almost completely overlaps with the image of Moroccan-Dutch men. Yet, our findings also demonstrate that linguistic research can offer additional insights into key socio-psychological issues such as stereotyping and prejudice: experimental investigations of accent evaluation allow us access to ‘deeper’ and more private evaluations beyond the public stereotypes reflected in public discourses. More particularly, positive dynamism evaluations of low prestige accents and groups are not impressions that respondents share easily: evaluations pertaining to the urban, streetwise or cool prestige of stigmatized varieties require sophisticated extraction tools, but they are essential for understanding group stereotyping and prejudice in all its dimensions. In this light, the passionate denial of discrimination that, according to Wekker (2016), is postulated in the Netherlands as a cover-up for active and aggressive racism, is more than just a tension between public ideology and private practice: our analysis has revealed the plainest kind of racism – the fact that a ‘wrong’ name suffices for stigmatization – but it has also uncovered a pathway for change, viz. a new prestige dimension along which a stigmatized language variety can acquire status. In an additional experiment, which was not discussed here, the modern appeal of the male Moroccan accent was found to be highly suited to ‘hip’ professions such as advertising or deejaying for adolescent radio stations. Such associations may suggest a more promising future for male Moroccans and the male Moroccan accent than can be surmised from their categorical rejection in terms of superiority.

On a methodological note, it is impossible to define notions such as ‘integration’ and ‘identity’, or determine how prejudices will develop in the future without recourse to the full dimensionality of impression formation. Not only the different dimensions of prestige – conservative vs. modern – but also gender represent key determinants of the image of Moroccan-Dutch people in the Netherlands. Our findings remind us that the Moroccan accent and the Moroccan-Dutch identity in the Netherlands exist at the intersection of ethnicity and gender. In spite of the fact that there are male, female, and non-binary Dutch people of Moroccan descent, it is the men who carry the burden of racial prejudice. Still, the conclusion that can be drawn from our data is decidedly less negative than Wekker’s (2016): while evaluations of Dutch Moroccans along traditional prestige dimensions are negative, at whatever level of (un)consciousness they are elicited, the growing diversity, informalization, and digitalization of modern societies, including the Netherlands, have spawned modern success dimensions along which stigmatized groups and their accents can acquire prestige. This does not obliterate the ubiquitous and ongoing racism and xenophobia in any way, but it does open up a potential domain for reconsideration, acceptance, and eventually also rapprochement.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Additional information

Notes on contributors

Stefan Grondelaers

Stefan Grondelaers is a senior researcher and lecturer at the Meertens Instituut, Amsterdam, and the Radboud Universiteit, Nijmegen. Grondelaers has made use of laboratory experiments and corpus analysis in multivariate studies of syntactic variation. Since joining the research group Variation and Distance at the Radboud University in 2007, Grondelaers has focused on the attitudinal and ideological correlates of variation and change processes in Belgian, Netherlandic and Surinamese Dutch, but he continues to be interested in the psycholinguistic and computational modelling of syntactic variation.

Paul van Gent

Paul van Gent is a master student in linguistics at the Radboud Universiteit, Nijmegen. Van Gent divides his time between his studies, freelance work and volunteer work for environmental and social issues. Together with Stefan Grondelaers, he has carried out a series of experiments that measure the evaluation of different language varieties in Netherlandic and Belgian Dutch. Van Gent’s research interests lie in the wider field of sociolinguistics as well as variation and change.

Notes

1. Purdie-Vaughns and Eibach, “Intersectional Invisibility”.

2. Phills et al., “Intersecting Race and Gender Stereotypes”.

3. We are indebted to the editors and reviewers for their valuable comments.

4. Grondelaers and Speelman, “A Quantative Analysis of Qualitative Free Response Data”, as well as Grondelaers and Speelman, “Groot-Nederlands en ‘taalverloedering’”.

5. See note 1 above.

6. Phills et al., “Intersecting Race and Gender Stereotypes,” 1173.

7. For an overview see Phills et al., “Intersecting Race and Gender Stereotypes,” 1173.

8. Ghavami and Peplau, “An Intersectional Analysis of Gender and Ethnic Stereotypes,” 118, Table 2.

9. Lambert et al., “Evaluative Reactions to Spoken Languages”.

10. Kristiansen, “The Macro-Level Social Meanings”.

11. Grondelaers and Speelman, “Can Speaker Evaluation Return Private Attitudes”; Grondelaers and Marzo, “Why Does the Shtyle Spread?”; Rosseel, “New Approaches”.

12. Ó Murchadha, “Authenticity, Authority and Prestige”.

13. Grondelaers, Van Gent and Van Hout, “On the Inevitability of Social Meaning and Ideology”.

14. Giles, “Evaluative Reactions to Accents”.

15. Grondelaers, Van Hout and Van Gent, “Re-Evaluating the Prestige of Regional Accents”.

16. Centraal Bureau voor de Statistiek, 2021.

17. Billiet, Eisinga and Scheepers, “Ethnocentrism in the Low Countries”.

18. De Witte and Klandermans, “Political Racism in Flanders and the Netherlands”.

19. Bleich, “Where Do Muslims Stand”.

20. Buijs, “Muslims in the Netherlands,” 423.

21. Vervoort and Dagevos, “The Social Integration of Ethnic Minorities”.

22. Nortier, “The Moroccan Community in the Netherlands,” 202.

23. Van Kempen and Bolt, 2009.

24. Vervoort and Dagevos, “The Social Integration of Ethnic Minorities”.

25. See note 23 above.

26. Vervoort and Dagevos, “The Social Integration of Ethnic Minorities”.

27. Crul and Doomernik, “The Turkish and Moroccan Second Generation”.

28. Grondelaers and Speelman, “A Quantative Analysis of Qualitative Free Response Data”.

29. Grondelaers and Speelman, “Groot-Nederlands en ‘taalverloedering’”.

30. Torstensson, “Judging the Immigrant”.

31. Grondelaers, Van Gent and Van Hout, “Is Moroccan-Flavoured Standard Dutch Standard or Not?”.

32. By indigenous Dutch we mean people of the same ethnic group as the earliest known inhabitants of the Netherlands.

33. The trait Integrity measures the likeability of the speaker through traits such as ‘friendly’ and ‘cordial’. The trait Accent Norm measures how much the speaker matches the societal norm for speaking in a certain context; and an example trait is ‘The way this man speaks is appropriate when delivering a speech for the entire country’. Aesthetic Evaluation does not pertain to the speaker, but to the speech itself and is measured on one trait: ‘This man’s speech is beautiful’.

34. This technique is used to verify whether the conceptual grouping of traits by researchers prior to an experiment matches the degree to which respondents react similarly to traits that measure the same construct during the actual experiment. This was clearly not the case in Grondelaers, Van Gent and Van Hout, “Is Morrocan-Flavoured Standard Dutch Standard or Not?”.

35. Grondelaers and Van Gent, “How ‘Deep’ is Dynamism?”.

36. Ghavami, Negin and Peplau, 118.

37. Ibid.

38. See note 9 above.

39. Grondelaers and Van Gent, “How ‘Deep’ is Dynamism?”.

40. Van Leeuwen et al., “Sprekend Nederland,” Table 1, 25.

41. See note 15 above.

42. Ibid.

43. See note 40 above, 25.

44. See note 15 above.

45. See note 31 above.

46. See note 28 above.

47. Repeated-measures ANOVA.

48. Superiority ratings for male speakers were affected by Speaker group (F(1.82, 192.58) = 81.05, p < .001, ηp2 = .43) and Accent strength (F(1.72, 181.81) = 92.98, p < .001, ηp2 = .47), as well as the interaction between the two (F(3.20, 339.54) = 30.08, p < .001, ηp2 = .21). Superiority ratings for female speakers were likewise affected by Speaker group (F(1.83, 170.19) = 65.22, p < .001, ηp2 = .41), Accent strength (F(1.76, 163.69) = 97.44, p < .001, ηp2 = .51), as well as by the interaction between the two (F(3.24, 301.69) = 33.67, p < .001, ηp2 = .27).

49. Note that directly comparing the Superiority scores for the two experiments (i.e. the two genders) is problematic; the same goes for the Dynamism scores. The two experiments were carried out separately, with listener-judges returning their evaluations in the context of a single gender of speakers. We cannot be sure, therefore, that differences between the two experiments are indicative of actual evaluative differences towards the two genders. It is therefore not possible to state that a mildly-accented indigenous Dutch man from the Urban West is superior (4.51) to the same extent as a Moroccan-Dutch woman without an accent (4.50).This is why we limit our interpretation to the relative position these accents take compared to the grand mean of each experiment.

50. Notably Grondelaers, Van Hout, and Steegs, 2009; and Grondelaers, Van Hout, and Van Gent, “Re-Evaluating the Prestige of Regional Accents”.

51. Dynamism ratings for male speakers were affected by Speaker group (F(1.51, 159.76) = 25.84, p < .001, ηp2 = .20) and Accent strength (F(1.88, 199.33) = 28.46, p < .001, ηp2 = .21), as well as by the interaction between both (F(3.65, 387.20) = 30.31, p < .001, ηp2 = .22). Dynamism ratings for female speakers were likewise affected by Speaker group (F(1.68, 156.14) = 43.54, p < .001, ηp2 = .32) and Accent strength (F(1.87, 174.01) = 12.21, p < .001, ηp2 = .12), as well as by the interaction between the two (F(3.39, 313.21) = 38.64, p < .001, ηp2 = .29).

52. Comparing the Superiority scores with the Dynamism scores within a single experiment is problematic for the same reason we gave in note 48. This is why we limit our interpretation to the relative position these accents take compared to the grand mean.

53. Ghavami and Peplau, “An Intersectional Analysis of Gender and Ethnic Stereotypes”.

54. Ibid., 118.

55. See Grondelaers, Van Hout and Van Gent, “Destandardization is not Destandardization” for further discussion.

56. See note 1 above.

57. See note 6 above.