Variation and change in grammatical gender marking: the case of Dutch ethnolects

1 Introduction

As cities in industrialized societies become ethnically more complex due to labor migration, the range of varieties in the national languages spoken there increases, and ethnolectal varieties are added to the urban social dialect landscape. Ethnolectal varieties emerge in multilingual contexts and are often rooted in processes of second language acquisition. However, they soon acquire a life of their own, are used by speakers with an immigrant ethnic background, and even cross over to speakers without an immigrant ethnic background (for an overview of ethnolects in the Netherlands see Muysken 2010, 2013).

In the study of ethnolects and immigrant varieties, variation in gender inflection has played an important role for those European languages with strong gender systems: German (cf. e.g., the overview in Clyne 2003; Muysken and Rott 2013), French (cf. e.g., Doran 2002, 2004, 2007; Fagyal and Stewart 2011), and Italian (cf. e.g., Chini 2011). Gender inflection is interesting for several reasons. (a) It lies at the interface between lexical, phonological, and morphosyntactic processing; (b) It is a frequent source of developmental problems for second language learners in these languages.

Examples of the phenomena studied here include the use of the article de instead of neuter het, as in (1) and (2), leading to an overgeneralization of the non-neuter, common gender form^[1]:

The same holds for the use of the common gender demonstratives deze ‘this’/die ‘that’ instead of the neuter equivalents dit/dat:

The overuse of the ending –e with indefinite neuter adjectives is quite possibly related to these phenomena:

To help understand these cases, this article first sketches the expression of grammatical gender in Standard Dutch in contrast with Dutch dialects, Turkish, Moroccan Arabic and Berber (Section 2), and then surveys earlier studies on variation in the expression of grammatical gender in ethnic varieties of Dutch (Section 3). Section 4 describes the Roots of ethnolects project from which the data in this paper are drawn, and Section 5 our main findings. In Section 6, these findings are discussed and a few questions for further research are listed in Section 7.

2 The expression of grammatical gender in standard and dialectal Dutch, Turkish, Moroccan Arabic and Berber

3 Variation in the expression of grammatical gender in ethnic varieties of Dutch: a survey of the literature

It has often been noted in the literature that there is overgeneralization of common gender in Dutch ethnolectal varieties. In fact, the acquisition of neuter gender in first language acquisition is also quite late.

Blom et al. (2008) and Blom and Vasić (2011) show that L1 learners of Dutch, Moroccan child L2 learners and Moroccan adult L2 learners of Dutch all overgeneralize common gender articles, also in neuter contexts. In all groups, the reverse is rarely found. While child learners overgeneralize one particular suffix (namely the schwa), adult learners use both adjectival forms, the schwa-adjective and the bare adjective, incorrectly.

Cornips et al. (2006), Hulk and Cornips (2006), Cornips (2008), Brouwer et al. (2008), and Cornips and Hulk (2008) studied the overgeneralization of common gender in the acquisition process of bilingual children in more detail. Their results reveal no significant difference between Moroccan-Dutch and Turkish-Dutch children, indicating that this overuse is not simply due to interference of the original language of the speakers (as we pointed out in Section 2.3 above, Turkish has no gender distinction, while Moroccan-Arabic and Berber do).

An earlier study of bilingual English-Dutch children by Unsworth (2008) confirms these findings, since the English-Dutch children also overgeneralize de for neuter nouns to a large extent. Furthermore, Unsworth’s study shows that the overgeneralization of common gender is not restricted to children from immigrant communities. However, there seems to be a difference between the overgeneralization of children from immigrant communities (e.g., Moroccans and Turks) and the English-Dutch children Unsworth investigated.

While Unsworth’s subjects did not overgeneralize common gender for stylistic purposes, and overgeneralized less frequently over time, Nortier and Dorleijn (2008) quote a Moroccan-Dutch informant who explicitly says that he knows how to use the Dutch gender system correctly, but that he overgeneralizes common gender on purpose when talking with friends.

In 5.2 below we will return to this aspect. Even though the expression of gender has been often discussed, there is very little systematic research on the variation in adolescent speakers and with a research design enabling the study of the origin of the variation, its place in the speakers’ verbal repertoires and its social spread. In our study we hope to have filled this gap.

4 The Roots of Ethnolects project

The present authors were all involved in the Roots (of Ethnolects) project.^[8] The research questions in the project are primarily descriptive, but also directed at specific research issues. The focus is on two components of the grammar: phonology and morphosyntax. In the Roots project spontaneous in-group conversations are recorded, and a set of elicitation tasks administered.

Three questions are geared towards disentangling the origins of the ethnolectal features. One concerns the role of L1. To what extent are ethnolects based on interference from the original language (L1) of the ethnic group in question? In other words, to what extent do ethnolectal features originate in the original mother tongues of both ethnic groups (“substrate effects”)?

To answer this question, in the Roots project, young male Dutch speakers from Moroccan (Moroccan Arabic and/or Rif Berber) and Turkish backgrounds are compared – labeled M and T, respectively. Language background of the parents (first generation speakers of Rif Berber and/or Moroccan Arabic, resp. Turkish) and place of birth (Amsterdam, resp. Nijmegen) are controlled for; and only speakers with strong inter-ethnic network ties are selected. The speakers’ language use is systematically contrasted with the original languages of their community as well as with traditional Dutch.^[9]

A second question focusing on the roots of the ethnolects concerns the extent to which we can reduce features of ethnolects to properties resulting from processes of L2 acquisition. Second language acquisition plays an important role in the emergence of ethnolects, although the ethnolect for the speakers themselves is the first language, maybe next to the community heritage language.

Generally speaking, language acquisition can leave two different types of traces behind: either interference from L1 or general, language independent characteristics of second language acquisition.^[10] Both kinds of acquisition traits are found in ethnolects. However, although ethnolects emerge in the context of second language acquisition, they are not fully determined by it, since they are not mere learner’s varieties. These latter varieties are used to communicate, but their norms are – in contrast with norms of ethnolects – not rooted in the speech community. On the contrary, learner varieties as such are not subject to clear norms, and are therefore unstable and somehow fluctuate “between” varieties.

To determine the role of acquisition, in the Roots project (i) speakers of two age groups (12 and 20 years of age) are compared, and (ii) the data are systematically compared with existing recorded first and second language acquisition data from the relevant ethnic groups. Van Emmerik et al. (2014) investigated the acquisition of nominal gender by low-skilled adult L2 learners of Dutch from Turkish and Moroccan origins and deaf learners of Dutch (late L1 acquisition). They also had a low-skilled control group of Dutch L1 adults. In the study by van Emmerik et al., the Turkish and Moroccan learners are first generation immigrants. All groups had high accuracy scores for common nouns in a writing task (Turkish 92.6%, Moroccan 96.6%). Except for the control speakers, all groups had dramatically lower scores for the neuter nouns (Turkish 35.9%, Moroccan 50.1%). The Moroccans had higher scores, but the difference between the Turkish and Moroccan group was not significant. In the present study, the age comparisons are made to see to what extent socialization in the ethnolect takes place (sociolinguistic fine-tuning), and the data from acquisition are explored to see to what extent the ethnolect shows interlanguage features.

There may be an interaction between local dialect features and ethnolectal features (cf. the references to the Manningham accent in Bradford and to Glaswasian (Alam 2015) in Glasgow in the U.K.), and hence we want to investigate to what extent ethnolects are based on local non-standard varieties. In order to answer this question in the Roots project neighborhoods are studied in two large towns with quite different urban dialects: Amsterdam (specifically the neighborhood called Transvaal) and Nijmegen (various neighborhoods). These towns are chosen because both have been the object of systematic sociolinguistic studies of local urban dialects.

In addition to these three questions about the possible sources of the ethnolect features, we want to find out whether ethnolects are specific for an individual ethnic group or rather reflect a more global non-majority identity. Do features spread across ethnolects? – a question inspired by the Scandinavian research on multi-ethnolects (cf. e.g., Quist 2008).

To establish to what extent ethnolects are specific for an individual ethnic group, or in contrast reflect a more global non-native identity in the Roots project first, specifically Turkish Dutch and specifically Berber Dutch features are compared to features common to the Dutch of both groups; second, informal conversations between Turkish-Dutch and Moroccan-Dutch (out-group contact) per urban setting are recorded; and third, the in-group recordings are systematically compared and contrasted with recordings from Dutch peer group speakers.

We also want to establish, following the work of Hewitt (1986) and Rampton (1995, 1998), whether there is any evidence of crossing or spread of ethnic varieties to peers outside of the ethnic groups involved? If so, do uniquely ethnolectal traits spread to peers outside the ethnic networks because of their covert prestige and/or also because they represent less marked options?

To find out whether there is any evidence of spread of ethnic varieties to peers outside of the ethnic group, in the Roots project in both cities two peer groups of Dutch speakers of comparable social backgrounds are recorded: a group with frequent contacts with Moroccan Arabic and Turkish background speakers of Dutch (D) and a group with few contacts with Moroccan and Turkish background speakers of Dutch (C). The latter group of speakers serves as the control group; statistically, it serves as the reference group for the effects of speaker background.

A final question in our research concerns the issue to what extent speakers of an ethnolect can shift to more standard varieties and to non-ethnic non-standard varieties. In the Roots project systematic comparisons are made of the language behavior of the members of the Turkish-Dutch and Moroccan-Dutch groups in various settings:

1. in-group conversations,

2. in out-group conversations with members of the other minority group and in out-group conversations with peers from traditional backgrounds,

3. an experimental situation with elicitation tasks.

As is evident from the various research questions, the project is oriented towards linguistic and sociolinguistic rather than ethnographic aspects of ethnolectal variation. Attention focuses on the origins, generalizations and restrictions on variation, its spread through the speech community at large and its place in the verbal repertoires (cf. Hinskens 2011; van Meel 2016). Van Meel (2016) presents the outcomes of quantitative sociolinguistic analyses of three phonetic variables, their mutual coherence as well as their coherence with other variable features.

To answer all the questions raised in our research project, we needed a speaker design in which ethnic background, age, city and, for the speakers with a traditional background, the presence or absence of regular contacts with Turkish-Dutch and Moroccan-Dutch are interwoven. Table 3 shows how this is obtained.

What this table does not show is the “within subjects factor” ingroup-outgroup contact, a situational factor which is relevant to questions about spread and style shifting.

As pointed out above, relatively spontaneous conversations were recorded both in in- and out-group contact situations. An early step in the analyses of these data consists of drawing up inventories of non-standard features.

For the present study, data were analyzed for the speakers in what we refer to as the “core corpus”, which is a proper subset of the data from the overall design (Table 4)

Most data were coded by Ariën van Wijngaarden. Statistical analyses included mixed models logistic regression. Researchers interested in the data underlying the analyses presented in this article are encouraged to contact the authors.

5 Findings

In this section we first present the findings for the effects of both the extra-linguistic and the linguistic factor groups. In both respects, grammatical gender (common vs. neuter) and its expression (on determiners and adjectival elements) are central.

The first main finding is that the inflection of the determiners and adnominal elements (in our data mainly adjectives) before common nouns hardly shows any variation; cf. Figure 1.

Figure 1:

Common gender: average usage of standard flexion per background.

As all groups of speakers appear to use almost 100% standard forms for common nouns, we focus on neuter gender marking. We will first discuss the variation in the determiner system.

5.1 Determiners

The data consist of 755 occurrences, realized by 52 speakers, an average of 14.52 occurrences per speaker, with large differences in frequency between speakers (SD = 7.56; minimum is 3, maximum is 40 occurrences). In all, 278 different lexical items occur in the data, with 160 hapax legomena, and only 12 words have a token frequency of 10 or larger; the mean word token frequency is 2.72 (SD 5.00).

About one quarter of the forms used by our speakers are not standard neuter forms but rather common gender forms such as definite article de, proximal demonstrative deze, and distal die. Mixed effects logistic regression analyses (glmer, lme4 package, R) were run with speakers as random effect; words were not included as a random factor, as there are too many empty cells when cross-classified with speakers. The background of both speakers and interlocutors were dichotomized to have sufficient numbers of speakers (C, D versus M, T). The dependent variable was a neuter form (1) or a common form (0). We included four extra-linguistic factors (see Table 5) and three linguistic ones: diminutive yes or no, animacy yes or no, and demonstrative vs. article.^[11] We tested all two- and three-way interactions. The best model found, using the AIC as the index to evaluate models, is given in Table 5. The best model has only one two-way interaction, containing the only linguistic factor that survived, demonstrative die.

Table 5:

Significant effects of the best model for variation in the determiners for neuter heads (Akaike IC 668.5).

Main effects	Estimate	Standard error	p
Extra-linguistic
Speaker background T+M	−1.7432	0.3592	0.000***
Age 18–20 years	1.3170	0.2952	0.000***
City Nijmegen	−0.9068	0.2935	0.002**
interloc.background T+M	−0.5286	0.2356	0.024*
Linguistic
Demonstrative die	0.4241	0.5050	0.405
Interaction effects
Speaker background T+M × demonstr die	−2.1267	0.5662	0.000***

First, as to the speakers’ backgrounds: Table 5 shows that the Turkish Dutch and Moroccan Dutch speakers use significantly fewer standard neuter forms than the speakers from the other groups (see Figure 2 below, columns labeled ‘m’ and ‘t’ versus ‘c’ and ‘d’). Second, the 18–20 year old speakers use more standard neuter forms than the 10–12 year olds (the green and blue bars, respectively, in Figure 2) and this probably points towards an overall acquisition effect. Third, the Nijmegen speakers use significantly fewer neuter forms than their Amsterdam counterparts (lower and upper graph in Figure 2, respectively). The fact that Nijmegen borders on dialect areas with three-gender systems and that considerable numbers of speakers of the relevant dialects live in Nijmegen (Section 2.1 above) may well play a role here. The background of the interlocutor has an effect in that neuter standard forms are used less frequently in interaction with Turkish and Moroccan Dutch interlocutors, also by ‘d’ speakers (Figure 3), is a clear style and audience design effect (Bell 1984).

Figure 2:

Percentage of usage of standard determiner forms for neuter heads broken down for speaker background, age group and city.

Figure 3:

Percentage of usage of standard determiner forms for neuter heads broken down for the background of the interlocutor (‘partner’).

Three internal factor groups were studied: one of them regarding the determiner type, and the other ones a formal and a semantic property of the noun: is it a diminutive or not (diminutives are always treated as grammatically neuter) and does it refer to an animate entity or not. There are no statistically significant main effects of any of the three internal factor groups. This is especially remarkable for the factor group simplex/diminutive, as in Dutch the diminutive is one of the not so many lexical domains^[12] where grammatical gender is formally predictable.

The determiner type is involved in a statistically significant interaction with speaker background; see Figure 4 below. In standard Dutch neuter determiner types are definite article het and proximal demonstrative dit (blue bars) or distal demonstrative dat (green bars). The non-standard variants in our data are common gender forms. The Moroccan and Turkish Dutch show remarkably high non-standard, common gender scores for the distal demonstrative die, such as die boekje, die raam for standard Dutch dat boekje ‘this little book’, dat raam ‘this window’; this variant makes up over 60% of the Moroccan Dutch and almost 70% of the Turkish Dutch distal demonstrative realizations. For the proximal demonstratives and the definite articles we do not find an effect of this type.

Figure 4:

Percentage of usage of standard determiner forms for neuter heads broken down for determiner type.

5.2 Adnominals

For the adnominals (including attributive adjectives), the analyses were confined to indefinites; after all, as was pointed out in Section 2.1, for indefinites with a neuter head the schwa ending is optional in standard Dutch: het / dit klein(e) meisje, ’the / this little girl’.

For the D, M and T speakers our data consist of 277 observations, 73 of which are non-standard forms, such as e.g., een klein-e meisje, ‘a little-infl girl’. These 277 occurrences were realized by 37 speakers (a proper subset of the speakers in the core corpus – for the remaining 14 speakers, our transcriptions did not contain any instances of an adnominal preceding grammatically neuter nouns). This amounts to an average of 7.49 occurrences per speaker, with differences in frequency between speakers (SD = 3.43; minimum is 2, maximum is 15 occurrences). In all, 122 different items occur in this part of the data, with 79 hapax legomena, and only 3 words having a frequency of 10 or more. The mean word frequency of usage is 2.27 (SD 2.96).

All these cases concern overuse of the schwa suffix. Table 6 summarizes the main outcomes of the mixed models logistic regression with speakers as a random effect (glmer, lme4 package, R; words were not included as a random factor, as there are too many empty cells when cross-classified with speakers).

Table 6:

Significant effects for variation in the determiners for neuter heads (Akaike IC 294.1).

Main effects	Weight	Standard error	p
Extra-linguistic
Speaker backgr T	−1.6034	0.3944	0.000**
18–20 year old	0.1053	0.0372	0.005*
Linguistic
Inanimate	−1.5401	0.4413	0.009*

In this model (i.e., the one with the highest Akaike IC, hence the best fit), there are no main effects for the extra-linguistic variable city, nor for the linguistic variables diminutive or non-diminutive head and the presence or absence and the nature of the determiner (not all speakers realize a determiner). Neither are there any significant interaction effects between the various independent variables.

The variation between standard and non-standard inflection is determined only by two of the extra-linguistic parameters and one of the linguistic parameters. The extra-linguistic parameters are speaker background and age group; proportionally, the Moroccan-Dutch and Turkish-Dutch use most non-standard inflected forms, as do the 10–12 year olds. See Figure 5.

Figure 5:

Percentage of usage of standard inflected adnominal forms for indefinite neuter heads broken down for speaker background and age group.

The linguistic factor animacy appears to be a prominent determiner of the variation, with inanimate heads being the main predictor of non-standard schwa-suffixation to the adnominal element. Further study is needed to determine why inanimate neuter nouns attract common gender marking. Is it a case of (re)semanticization (Audring 2006; Dahl 1999)? In nouns referring to animate entities, biological gender seems to outweigh grammatical gender (genus), as in e.g., de meisje; de wijf; de jongetje; de dier (‘the girl’, ‘the woman’, ‘the boy_DIM’, ‘the animal’); in standard Dutch these nouns have neuter gender (and thus select the definite article het). We find this relatively often in the speech of Moroccan Dutch and Turkish Dutch speakers, although our data do not display a statistically significant interaction effect between the +/− animacy semantics of the noun and the speakers’ background.

6 Discussion

What does all this mean? How do the findings relate to our main questions regarding the linguistic origins of ethnolect features, the place of ethnolect in the verbal repertoires and the spread of ethnolectal features to other ethnic groups?

As regards the roots of the variation in grammatical gender marking: this is clearly a second language phenomenon. There is a strict, one-way overuse neuter → common, hardly ever the other way around; this is visible among other things in overuse of the non-neuter definite article de over neuter het and in overuse of the generic inflected adjective rather than the non-inflected article limited to neuter indefinite contexts. This overuse of common gender is much stronger than in the L1 and L2 acquisition data reported in the literature (cf. Section 3). With its full vowel, the emphatic common gender distal demonstrative die may be more accessible to adult learners; what looks like the emergent replacement of the definite article de (also in its non-standard use as neuter definite article) with die may thus be rooted in a second language effect (the parents of the T and M speaker groups, generally first generation migrants).

Especially in adnominal inflection there seems to be a tendency towards a new regularity with biological gender outweighing grammatical gender. This amounts to the development of a much more regular and transparent grammar. What adds to the interpretation of these patterns as acquisition effects is the fact that overall there are significant age effects, manifested in the use of more correct standard forms by the older ethnolect speakers.

For the adnominals, the acquisition effect is probably amplified by general tendencies towards deflection in the Dutch diasystem at large and for the determiners by dialect geography, more in particular properties of the surrounding endogenous dialects (see Section 2.2 above). In the Nijmegen area, which borders on Dutch dialectal and German three-gender areas, language learners may well receive input from both common-neuter and masculine-feminine-neuter systems and the variation Common versus Masculine-Feminine, which is largely located in articles, demonstratives and adnominal inflection as well as in 3sg pronouns, may require so much processing energy and storage space that it weakens the cognitive position of the neuter.

What role does this specific variation have in the speakers’ verbal repertoires? To tell from the effect of the interlocutors’ background on variation in the determiner, there is evidence of style shifting linked to audience design. Moreover, the data indicate that there may be stylization going on with respect to what looks like the overuse by Moroccan Dutch and Turkish Dutch speakers of the common gender distal demonstrative die in contexts where the use of demonstratives seems uncommon in traditional varieties of Dutch. An example is (15), where speakers of traditional Dutch would have said ‘het hele land’.

The traditional Dutch distal demonstrative ‘dat hele land’ would not be appropriate here. Acton and Potts (2014: 23) argue that in the speech of Sarah Palin during TV interviews distal demonstratives serve as “vehicles for establishing and maintaining social bonds”. Independent of the fact that in (16) the ‘wrong’ gender is used, here and in similar cases in our data distal demonstratives do not seem to play that role. Non-standard variation in gender marking, especially the use of the salient emphatic common gender form die in non-emphatic neuter contexts and the omnipresent schwa on adnominal forms, may have acquired a social symbolic meaning for speakers of Dutch ethnolects, turning them into targets of speaker agentivity, possibly expressing a more global non-native identity (cf. Johnstone’s et al. 2006 indexicality concept). This is an issue for future research.

In our data there is no significant difference between the two groups of traditional Dutch speakers, i.e., the ones with strong and the ones with weak interethnic ties. Neither is there any statistical evidence of spread to members of the D group, i.e., traditional Dutch speakers with strong interethnic network ties, although six of the 11 speakers of this type in our sample do incidentally use non-standard (common gender) inflection in neuter contexts. An example is realized by a Nijmegen 12 years old in a conversation with a peer with a Turkish-Dutch background.

But non-standard forms are also used with D interlocutors, as in (17), realized by a Nijmegen 20 years old.

Three of the D speakers who have been observed to use non-standard inflected adnominals are from Amsterdam (one of them 10–12 years of age, two of them 18–20 years old), the other three are from Nijmegen (two of them 10–12 year of age, one of them 18–20 years old). With the exception of one Amsterdam 20 years old and one Nijmegen 12 years old, all speakers produced one single non-standard inflected adnominal form. Of the 10 cases of non-standard inflection^[13] produced by these six D speakers, seven were realized in a conversation with a peer without a migration background (hence another representative of the D set).^[14]

Overall, Turkish Dutch speech seems to display the lowest proportions of standard Dutch forms – which is what we expected since Turkish does not have grammatical gender, unlike Moroccan Arabic and Berber. But the 20 year old Moroccan Dutch deviate from this general pattern. Are they borrowing this phenomenon from Turkish Dutch or expressing and claiming their own ethnolect position? Although it may rather reflect a more global non-native identity in general, it is clearly used by our Moroccan-Dutch and Turkish-Dutch informants to mark their ethnicity.,

7 Issues for future research

The fact that grammatical gender varies in a variety of Dutch spoken by members of communities with many second language learners of Dutch does not come as a big surprise; ethnolects are perhaps ahead of future changes in Dutch. The fact that in endogenous varieties of Dutch the grammatical gender system appears to be increasingly dismantled does not come as a big surprise either. Especially in underived nouns, gender adds opaque, meaningless and/or functionless substantial complexity – “junk, historical debris” in Lass’ (1997: 316) terminology. Nevertheless, crucial questions remain concerning the internal and social trajectories of the dismantlement process.

First of all, the variable presence or absence of the determiner in environments where endogenous varieties of Dutch require a determiner deserves closer analysis (cf. Section 5.2 above). Schönenberger (2014) found that monolingual German and bilingual Turkish-German children (around 4 years old, followed for 20 months), showed a clear difference in determiner omission, but not in ‘article misuse’, i.e., the inappropriate use of a definite/indefinite determiner.

Next, we have not yet analyzed a set of data elicited at a later stage than our initial recordings with similar speaker groups. In case the elicited data show different patterns, the question arises whether this is related to the stylistic difference between conversational and scripted speech.

Then, experimental studies could throw further light on two issues raised by our findings so far. The first one concerns the remarkable finding that animacy plays a statistically significant role in conditioning the use of adnominal schwa with neuters (see Section 5). The second one concerns the social-emblematic status of adnominal schwa as hypothesized by Nortier and Dorleijn (2008); our conversational data show that this variation is clearly related to the speaker background.

The spread of the distal demonstrative die to determiner contexts also merits further exploration, since it is a feature of several languages related to Dutch which have arisen in situations of intensive language contact in the 17th and 18th centuries, such as Afrikaans and Virgin Islands Dutch Creole (also known as “Negerhollands”). In these languages (which have no grammatical gender), die serves as the definite article.

In short, there is still enough work to do concerning the dismantlement of grammatical gender and related morphosyntax in ethnolectal varieties and daughter languages of Dutch.