Master thesis
CROSS-LINGUISTIC INFLUENCE
IN EARLY CHILD BILINGUALISM
Cross-linguistic influence in the acquisition of Dutch grammatical genderby Dutch-Serbian bilingual children
Student: Mirko Cvetković Student number: 10031332 Date: 14 July 2014 Mentor: prof. dr. Aafke C. J. Hulk
University of Amsterdam Faculty of Humanities Master Nederlandse taal en cultuur
2 Abstract
This thesis is based on a study that addresses the topic of cross-linguistic influence in the domain of grammatical gender in Dutch-Serbian bilingual children. Cross-linguistic influence was defined as delay or acceleration of the acquisition of grammatical gender, compared to Dutch monolingual children of the same age. Moreover, it was predicted that the language that transparently marks gender will have positive effects on the acquisition of grammatical gender in the less transparent language. When the bilinguals perform similarly as or better than monolinguals children of the same age, this was seen as a possible effect of cross-linguistic influence, since the bilinguals received less input than monolinguals. Production data of 27 Serbian-Dutch bilingual children aged 6-12 was collected and analyzed. An additional questionnaire was administered to the parents in order to assess children’s sociolinguistic background. The results showed that cross-linguistic influence was not observable in all language groups. In correlation with the parental questionnaire, the sociolinguistic environment of the child when both parents spoke Serbian and at least one of them Dutch was shown to be the best predictor of cross-linguistic influence in this study.
3 Table of contents
1. Introduction ... 4
2. Factors that influence language acquisition in early bilinguals ... 6
3. Cross-linguistic influence in early bilinguals ... 8
4. Grammatical gender in Dutch and Serbian ... 13
4.1 Grammatical gender in Dutch ... 14
4.2 Grammatical gender in Serbian ... 15
2.3 Grammatical gender marking in Dutch and Serbian compared ... 15
5. The acquisition of Dutch grammatical gender ... 17
5.1 Gender assignment ... 17
5.2 Gender agreement ... 19
5.3 Cross-linguistic influence ... 20
6. Hypotheses and predictions ... 22
7. Methodology ... 24 7.1 Participants ... 24 7.2 Task ... 24 7.3 Procedure ... 26 8. Results ... 27 8.1 Results PDT ... 27
8.1.1 Comparison with Dutch monolingual children ... 29
8.1.2 Comparison with bilingual children with language other than Serbian ... 33
8.1.3 Consistency analysis... 38
8.2 Results according to dominance ... 40
9. Discussion ... 43
10. Conclusion... 47
Limitations of this thesis... 47
References ... 47
4 1. Introduction
Until the late eighties of the previous century, research on bilingual children who acquire both (or more) languages at the same time considered that both (or more) languages function as a unitary system in child’s mind. This perspective has since shifted from one to two separate grammar systems (Meisel 1989, De Houwer 1990, Genesee, Nicoladis and Paradis 1995), however the question was if these separate language systems developed in monolingual and bilingual children in a similar (De Houwer 1990) or in a different way. Moreover, there was (and still is) the question of how the two systems in a bilingual child’s mind interact, and how and when is that influence manifested in the child’s linguistic performance. Since the publication of the studies by Hulk and Müller (2000) and Müller and Hulk (2001) on cross-linguistic
influence in bilingual children, the main focus of the research on cross-linguistic influence are the domains in which cross-linguistic influence is expected to happen and what are the factors that can account for individual differences between bilinguals.
Both questions, whether the two systems develop similarly in monolingual and bilingual children and whether cross-linguistic influence in this domain is evidenced, have so far been tested in a substantial amount of research on the acquisition of the Dutch grammatical gender (for an overview see Cornips and Hulk 2008; Hulk and Van der Linden 2011, Unsworth 2013, Ntalli 2013).
Unlike other gendered-languages that often have phonological or morphological cues on the endings of the nouns, and the same endings on other words that agree with the noun which makes them highly frequent in the input, the Dutch gender system is non-transparent. The grammatical gender of the words is assigned by the choice of the definite determiner and has to be acquired on a lexical word-by-word basis (Blom et al. 2008). This path of acquisition is dependent on the frequency in the input from the environment and it therefore takes long to acquire. The monolinguals’ target acquisition of the Dutch grammatical gender takes place after the age of 6 or 7 (among others Van der Velde 2003), whereas monolingual children of other, more transparent gendered-languages, acquire the grammatical gender in the respective language around the age of 3 (e.g. for Spanish, see Pérez-Pereira 1991; for Russian, see Rodina and Westergaard 2012).1 By definition, since the monolingual children get twice as much input than the bilinguals, the latter
in turn need twice as long exposure to the native Dutch input in order to be on the same level as monolinguals.
Aside from the lexical assignment of gender to the nouns, the adjectives that precede the nouns in Dutch are either inflected or bare if the noun is indefinite, depending on the gender of that noun. Here, the syntactic rule [-neuter] for the inflection of the adjective needs to be acquired. Once the monolingual child activates the existence of neuter gender, they are automatically expected to inflect the adjective according to the syntactic rule as [+neuter] or [-neuter]. According to the Domain-by-Age model (Schwartz 2003), early bilingual learners acquire inflectional morphology as monolingual children through the grammar-based, while this is not the case with the late (adult) learners, who rely again on a lexicon-based learning strategy (Blom et al. 2008).
A clear distinction between early and late learners cannot easily be made. In his seminal work on this effect in first language acquisition, Lenneberg’s (1967) proposed the Critical Period hypothesis, according to which the latest age on onset needs to be before puberty. Johnson and Newport (1989) tested this hypothesis in second language acquisition and found a correlation between language performance and age of arrival in
1 Although the more opaque forms of gender that deviate from the transparent gender endings are also acquired later
(around the age of 6), as is the case with the marked ending for neuter gender in Russian (Rodina and Westergaars 2012).
5 the country where the majority language is spoken. They found that after the age of 3, there is a gradual decline in target-like acquisition of the second language. Meisel (2007) also proposes several critical
periods, and Blom et al. (2008) take on this division, as their study showed that the syntactic rule needed to be activated before the age of 4 in bilingual children in order for monolingual-like acquisition to take place. The final classification combines those three: the bilingual child is every child that is exposed to two languages up to the age of 10, with further classification of bilingual children as: simultaneous bilinguals if they were exposed to both languages from birth on, the early successive bilingual children if they were exposed to an additional language between 1 and 4 years and the second language children, whose age of exposure was from 4 to 10 years (Unsworth et al. 2014, Unsworth 2013). This classification will be adopted in the remainder of this thesis.
Next to the question of uneven language development between the languages in bilingual children, the question of linguistic influence has already been put forward. In this thesis, the definition of cross-linguistic influence will be adopted as interaction between two languages that falls under one of the three manifestations, either as delay, transfer or acceleration (Paradis and Genesee 1996). The delay as cross-linguistic influence predicts that bilingual children will, at times, demonstrate a slower rate of acquisition. Conversely, acceleration refers to the notion that bilingual children will, at times, demonstrate a faster rate of acquisition when compared to their monolingual peers. Lastly, the occurrence of transfer is defined as the integration of a grammatical property into one language taken from the other.
On the notion of possible cross-linguistic influence in the domain of Dutch grammatical gender, two studies claimed to have found cross-linguistic influence in the form of acceleration. Both were done with
simultaneous bilingual children whose other L1 has a more transparent gender system. The bilingual children in the respective studies (Hulk and Van der Linden 2011, Ntalli 2013) outperformed monolingual children of the same age, although they received less Dutch input than monolinguals. Other studies (Hulk and Cornips 2008) found no difference between the bilinguals of gendered and non-gendered languages in the acquisition of Dutch grammatical gender.
The main goal of this thesis is to investigate the differences and similarities in the acquisition of gender assignment and gender agreement in Dutch by early bilingual Dutch-Serbian speakers and to compare them with monolinguals. Since Serbian is a gendered language with transparent gender, the possible effects of cross-linguistic influence will be addressed based on the bilingual’s performance.
The thesis is structured in the following way. In Section 2, the factors that determine failure and success in the acquisition of grammatical gender in Dutch by bilinguals are cited. In Section 3, an overview of the literature on cross-linguistic influence is presented. Section 4 gives an description of grammatical gender in Dutch and Serbian, while an overview of previous acquisition studies with Dutch is presented in Section 5. Based on the previous sections, hypotheses and predictions of this thesis are formulated in Section 6. In Section 7, the methods and participants are presented. Section 8 reports the results of the acquisition of Dutch grammatical gender by Dutch-Serbian bilingual children. Section 9 discusses the results, and section 10 concludes the thesis.
6 2. Factors that influence language acquisition in early bilinguals
In order to assess the external factors that might predict failure or success of the acquisition of Dutch definite determiners by bilingual children whose one of the respective languages is Dutch, Cornips and Hulk (2008) have analyzed the studies that were done on the topic of the determiner choice in Dutch (gender assignment) up to that date. These studies included roughly three groups of children. The first group consisted of, as authors name them, the bilinguals of the ‘old’ type, or the children that have been
acquiring another dialect Dutch dialect from birth and started acquiring standard Dutch at the same time or at school. The second group of bilinguals were the ones from the ‘new’ type communities, whose
grand(parents) acquired Dutch in a non-instructed context and/or who were exposed to Dutch by the time they start school (at age of 5). Finally, the third group formed the children from expatriate families where one parent was often Dutch, thus these children were mostly exposed to the one-parent-one-language learning context.
Cornips and Hulk (2008: 268) differentiated four external factors that can be used as predictors in assessing the success of the acquisition of Dutch grammatical gender by different types of bilinguals: (i) the age of onset; (ii) the length of exposure; (iii) the quality of the input; and (iv) the role of the other language. The authors stress that all the factors are more or less directly related to input. It should also be noted that these are the factors that influence bilingual language acquisition in general, and not only with respect to grammatical gender.
With respect to the age of onset, the children from the studies who were not born in the Netherlands and the children from the ‘new’ bilingual communities have a later age of onset and have found to have performed worse on gender assignment in Dutch than children who are born in the Netherlands and have contact with (native) Dutch from birth on with one (mostly expatriate) or both (bidialectal) Dutch speaking parents.
As for the quantity of input, they rightfully note that all bilingual children are exposed to less input in both their languages than their monolingual peers, but that some children still did show target-like development in their assignment of gender to the Dutch nouns. Moreover, they raise the question of how to establish the threshold in the input for bilingual children to be successful in this domain. Unsworth (2007: 456) in her study on gender assignment and gender agreement in Dutch by Dutch-English bilinguals argues that, if the quantity of input is measured as the length and amount of exposure, it will have to be significantly more for the bilingual children: “if, for the sake of argument, we assume that the bilingual children are exposed to Dutch for around half the time, and to English for the other half, then this will mean that they would need at least 12 years of exposure to Dutch in order to be on a par with monolingual children.” For this reason, in a later study the author compared the bilinguals with monolinguals in the following way: 3, 4 and 5 year olds with 2 year olds; 6 year olds with 3 year olds; 7 and 8 year olds with 4 year olds; 9, 10 and 11 year olds with 5 year olds; 12 and 13 year olds with 7 year olds and the bilingual 14 to 17 year olds with a group of monolingual 8 year olds. The results showed that the bilinguals either performed as good as or
outperformed their peers in terms of cumulative length of exposure.
The quality of input was reported as a significant factor only in the communities of the ‘new’ type only (Cornips and Hulk 2008). Since the (grand)parents of the children have learned Dutch in a non-instructed context, their own input could have been less target. Paradis and Navarro (2003) refer to this type of input as contact-modified input, that was evidenced in their study on Cuban/Spanish-English child bilinguals, as their overproduction of subject drop correlated with the frequency of subject drop in their parents modified input (see more on this study in the next Section).
7 Based on the factors that were mentioned so far: the age of onset, the quantity and the quality of input, one cannot put a finger on a single sociolinguistic profile for a bilingual Dutch-Serbian child in the Netherlands. In the first instance, they can be classified as coming from the expatriate families, with the majority of them having one speaking and one Dutch-speaking parent. However, some Serbian-speaking parents also speak Dutch with their children to a more or less extent. A part of them immigrated to the Netherlands as older or younger children with one or both Serbian-speaking parent as a
consequence of the 1990’s Balkan wars, when they came in contact the first time with Dutch at school or later. It is not clear to which extent their Dutch might have also been modified in the input when they use it with their children now. Lastly, there is the group of parents who both speak only Serbian with their child. The effects of the fourth and last external factor2, the role of the other language, was claimed by Cornips
and Hulk (2008) not to have been found between the groups of gendered languages, that could have influenced the acquisition of Dutch gender, and the languages without gender, that could have not influenced it, except in the case of bidialectal children with another language which morphosyntactically overlapped with standard Dutch. In this group, other factors also contributed to other factors for successful assignment of Dutch gender, such as the early age of onset, quantitatively and qualitatively sufficient input. On the other hand, other studies (Hulk and Van der Linden 2011, Ntalli 2013) claimed to have found
positive cross-linguistic influence in the domain of the acquisition of Dutch grammatical gender by bilingual children. In the next section, a more detailed description of recent studies on cross-linguistic influence is given.
2 Note that other factors, such as language aptitude (Paradis 2007) and education and socioeconomic status of the
parents (Unsworth, Hulk and Marinis 2011) are also evidenced to be correlated with the success of bilingual language acquisition. However, as Cornips and Hulk (2008) socioeconomic status is often reflected in the quality of parental input. Language aptitude has not been addressed in this thesis and therefore it will not be further discussed for its purposes.
8 3. Cross-linguistic influence in early bilinguals
This section discusses the main postulates of and conditions for cross-linguistic influence based on the previous studies on this topic. The hypothesis of cross-linguistic influence was first introduced Hulk and Müller (2000) and Müller and Hulk (2001). According to the authors, in order for cross-linguistic influence to take place, two conditions need to be met: (1) the phenomenon in question has to be at the interface between syntax and pragmatics, and (2) there has to be a structural overlap between the two languages at the surface level. The interface between syntax and pragmatics was shown to be often problematic in bilingual language acquisition (see Serratrice 2013 for an overview). The second condition, the structural overlap (or indirect syntactic influence) is considered “the case when language A has a construction which may seem to the child to have more than one structural analysis (one of which is not correct) and that language B reinforces one of these two structural analyse (the wrong one - leading to problems - or the right one - leading to facilitation).” (Hulk and Müller 2000: 228). The authors tested their hypothesis on object drop and root infinitives in spontaneous speech data of two bilingual children (Dutch-French child aged 2;3-3;10 and German-Italian child aged 1;8-???). Both domains are on the interface between syntax and pragmatics, but the latter does not have structural overlap between the languages that the bilingual children spoke. As a consequence, cross-linguistic influence was observed in object drop, which occurred more in the Romance L1 of the bilinguals, compared to their monolingual age-mates. However, there was no cross-linguistic influence present in the root infinitives since the second condition was not fulfilled - there was no structural overlap - and the bilinguals performed similarly to their monolingual age mates. Regarding the second condition, ‘the structural overlap at the surface level’, Döpke (2001:28, in reply to Müller and Hulk 2001) pointed out that her data indicated that structural overlap between languages could lead to variation at any level of structural hierarchy. The condition for structural overlap was later
rephrased that the condition is met if “the surface strings of the two languages A and B are analyzable in terms of the syntactic derivation of one language (which is less complex)” (Schmitz, Patuto and Müller, 2012:229).
Hulk and Müller argue that cross-linguistic is due to language internal processes that are manifested in child’s language production and that it is not affected by other external factors (such as dominance). While partially accepting the authors’ claim of internally-motivated mechanisms, Paradis and Navarro (2003) introduced the possibility that the bilingual children’s output can, next to internal processes, also be attributed to externally controlled mechanisms, such as the specific language input bilingual children are exposed to, or as they call it, the contact-modified input. They compared the frequency of subject drop in production of a Spanish-English bilingual child aged 1;09-2;06 to that of two monolingual Spanish age mates, but also to subject drop in the input of the parents. The bilingual child produced more overt subjects (35%) than the monolingual children (20% and 17%), which can be argued to be cross-linguistic influence that is child-internal, since the domain of subject drop satisfies both condition from Hulk and Müller’s hypothesis. However, the parental input of the bilingual child also included more overt subjects (> 50%) than the parental input of monolingual children (< 50%). Therefore, it cannot conclusively be said whether the cross-linguistic influence is a result of internal mechanisms of the child, or that it was present in the contact-variety Spanish input the child was receiving, or the combination of both.
Hauser-Grüdl et al. (2010) criticized the contact-modified input-driven explanation of cross-linguistic influence, since none of the three predictions that are supposed to be compatible with this approach were borne out in their study. They tested subject use, word order and root infinitives of a bilingual Italian-German child aged 2;3-3;1 (one parent Italian one parent Italian-German) and of a monolingual L1 Italian-German child during the same period. Since the bilingual child was raised in Germany, in order to support the hypothesis
9 of the contact-modified Italian input, then the adult Italians living in Germany should also produce more overt subjects than adult Italians living in Italy. As the authors point out, there was no difference in production of overt subjects between the two adult groups in another study, even though the bilingual child in their study produced more overt subjects than the monolingual child. By the same token, if the bilingual child was raised in Germany, then their German should not show effects of contact-modified input, but the bilingual child preferred VO over OV order, which is common in Italian, but not in German (and thus not exhibited by the monolingual child). Lastly, as the authors point out, if the contact-modified Italian input was to be supported, than the child’s output should mirror the parental input. This prediction was again not borne out in the case of Root Infinitives, since the bilingual child produced them more than the adult speakers.
Although the predictions by Hauser-Grüdl et al. (2010) have logical ground, they are set rather strictly and contradict the generally observed phenomena in language acquisition as such. For instance, not all adult Italians living in Germany are relevant for comparison on subject drop, but only the parents of the specific children that have been tested, similarly to Paradis and Navarro (2007). Moreover, the failure to confirm the third prediction that the child’s output should mirror the parental input at all times, at first seems to support the Poverty of Stimulus hypothesis3 (Chomsky, 1988), since it was evidenced in the overproduction
of Root Infinitives by bilingual child and not in adults. However, since the same hypothesis was often supported for monolingual children in the same domain, i.e. that their output does not at all times 100% reflect the parents input (see White 2012), it is not clear how this is relevant in the case of the bilingual child.
Although Hauser-Grüdl et al. (2010) reported to have supported the internally-based model proposed by Hulk and Müller (2000), they note that this model still cannot explain why effects of cross-linguistic
influence vary individually from child to child. In order to account for the question of individual differences in the child’s output, they make use of Toribio’s (2004) hypothesis according to which the high processing load of two simultaneously active languages in bilinguals makes cross-linguistic influence possible. Recall that the condition for structural overlap was later rephrased as “the surface strings of the two languages A and B are analyzable in terms of the syntactic derivation of one language (which is less complex)” (Schmitz, Patuto and Müller, 2012:229). In other words, the structures that have a less complex syntactic derivation will be more easily transferred to the language with the more complex derivation if the latter demands more cognitive load, i.e. if it is not automatized in the child’s language. In order to assess the cognitive load in the language with the more complex analysis, Hauser-Grüdl et al. (2010) propose the assessment of the child’s fluency in that language: the more fluent the child is in the language with the computationally complex analysis, the least is the effect of cross-linguistic influence. However, the authors fail to account for the definition of fluency in their study. They measured the MLUs of their participant as the participant’s fluency and found correlations between the MLUs and target-deviant object omissions (similar findings are reported in Cantone and Müller 2005, Hauser-Grüdl and Arenciba Guerra 2007).
As we will see, the definition of fluency is not very clear-cut. In bilinguals, it is often referred to in terms of language dominance, since “bilinguals who possess a dominant language are often referred to as
unbalanced bilinguals, whereas balanced bilinguals are described as individuals who exhibit about the same fluency in two languages” (Cantone et al. 2010). Such a formulation, however, can seem circular, since neither dominance nor fluency can be disentangled from each other, and a satisfactory definition of neither
3 According to the Poverty of Stimulus hypothesis, there are patterns in all natural languages that children learn,
although they were not present in the parental input, such as the incorrect rules they apply when their language is developing.
10 of the two is given. In other studies, a more dominant language is considered the language in which
bilingual child is more proficient (see Cantone et al. for an overview on issues on measuring and defining language dominance in bilinguals). The constructs of fluency and proficiency have been used
interchangeably across studies, and they were also measured in different ways, in order to assess the effects of language dominance on cross-linguistic influence in bilingual children, when the conditions structural overlap or/and the interface between syntax and pragmatics were not satisfied.
For example, Yip and Matthews (2000) studied influence of Cantonese patterns in the production of English wh-interrogatives and relative clauses in bilingual Cantonese-English children. Cross-linguistic influence was found in the child who had greater fluency in Cantonese, even when there was no structural overlap
between the languages. Moreover, in line with the three manifestations of cross-linguistic influence (that of delay, acceleration or transfer, Paradis and Genesee 1996), Kupisch (2007) found influence of the stronger language on the weaker language in the acquisition of the determiner by Italian and German-English bilingual children, in that their acquisition was accelerated, compared to the acquisition rate of the children’s monolingual peers. In the case of German-English bilinguals, the acceleration has occurred also in the absence of structural overlap, which left dominance as the only factor.
According to these studies, we can speak of cross-linguistic influence even in cases when there is no structural overlap. Based on other recent studies on cross-linguistic influence, Serratrice (2013) rightfully poses the question whether the two conditions proposed by Hulk and Müller (2000) are actually necessary for cross-linguistic influence to take place. Studies that investigated cross-linguistic influence in absence of interface but with a structural overlap were for the most part focused on derivational morphology. Foroodi-Nejad and Paradis (2009) compared the production of right- and left-headed compounds of Persian-English bilingual children (mean age 4;7) with the compounds of monolingual children in each language. There were differences observed between the monolingual groups: the Persian monolingual children produced more left-headed than right-headed compounds, while the English monolingual children produced more right-headed compounds the majority of the time. Both left-headed and right-headed compounds are possible in Persian, while adult English only allows right-headed compounds (even though monolingual English children produced left-headed compounds 25% of the time). Therefore, the prediction for structural overlap would be unidirectional influence from English to Persian since English allows only one option, but not vice versa. The results showed that bilingual children produced both more right-headed compounds in Persian (46%) than the monolingual children (19%), however, they also produced more left-headed
compounds in English (42%) than their monolingual age mates (26%). This cross-linguistic influence was not compatible with the conditions proposed by Hulk and Müller (2000). Therefore, an alternative explanation was sought in dominance. If dominance could account for cross-linguistic effects, then there would be transfer from the dominant to the non-dominant language such that in their dominant language, bilinguals would perform closely to their monolingual peers. The results could also only partly be explained by language dominance, since only in the case of Persian, the Persian-dominant bilinguals behaved more like their monolingual peers than the English-dominant children. There were no dominance effects for English. Similar results on cross-linguistic influence in the domain of derivational morphology were reported in Nicoladis (2002). She studied production of compounds in French-English bilingual children aged 4-5. Unlike already mentioned Persian, French allows only left-headed compounds. Therefore, apart from the absence of interface, there is no structural overlap between French (left-headed) and English (right-headed) in this respect. Still, the bilingual children produced more ungrammatical left-headed compounds in English than the monolingual children. To account for the results of Nicoladis (2002) and their own study, Foroodi-Nejad and Paradis (2009) propose integration of dominance and structural overlap, in that cross-linguistic
11 influence is modulated by dominance. Therefore, if cross-linguistic influence is expected from language B to language A, then the cross-linguistic structures will be more prominent, if language B is the dominant language.
Let us now compare the prediction of Foroodi-Nejad and Paradis (2009) with that of Hauser-Grüdl et al. (2010). The latter is formulated as: the more fluent the child is in the language with the computationally complex analysis, the least is the effect of cross-linguistic influence. The former state that if cross-linguistic influence is expected from language B to language A, then the cross-linguistic structures will be more prominent, if language B is the dominant language. In their essence, both predictions are fairly similar in the case of unbalanced bilinguals: where Foroodi-Nejad and Paradis argue “if cross-linguistic influence is expected”, Hauser-Grüdl et al. are more precise by saying that the language with less complex analysis will have influence on the language with that is more complex. Moreover, Foroodi-Nejad and Paradis say that the more the child is dominant in the language that has influence (and we assume correspondingly the less dominant the child is in the language being influenced), the cross-linguistic influence is more likely to happen. Hauser-Grüdl et al. predict similar, though using the term fluency: the less the child is fluent in the language that under influence, the more likely the cross-linguistic influence. In balanced bilinguals, or the bilinguals who are more fluent or dominant in the language that can be influenced (the language with more complex analysis), cross-linguistic influence is not expected to take place.
To summarize, so far we have observed that the hypothesis of cross-linguistic influence was at first argued to be a child internal process, independent of external factors if the two conditions were fulfilled: (i) there interface between syntax and pragmatics at play and (ii) there structural overlap on the surface level. Other studies proposed that contact-modified input should also be taken into account when the cross-linguistic influence was observed to have taken place. Finally, in order to account for individual differences, and/or in the case of lack of one of both of the conditions for cross-linguistic influence, different studies have found correlations with the occurrence of cross-linguistic influence and the dominance/fluency/proficiency in the case of unbalanced bilinguals.
In the case of acquisition of grammatical gender of two gendered languages by bilingual children, only the condition for structural overlap can be argued as being fulfilled, although the differentiation between the languages in the sense of the complexity of the syntactic derivation is not an easy one to make, especially in the case of Dutch and Serbian4. Therefore, the mere existence of gender in both languages, with the gender
assignment in one language being more transparent than in the other will be taken as a possible route for cross-linguistic influence from the more transparent to the more opaque gender system to take place in the form of acceleration, i.e. the faster acquisition of the more opaque gender system, compared to
monolingual children. If the bilinguals perform similarly as monolinguals of the same age, then this will also be observed as acceleration, seeing that the bilinguals receive two times less input than monolinguals. If dominance plays a role, than the more dominant the child is in the language with the more transparent gender, the more likely for cross-linguistic influence to happen. If, however, the child is already sufficiently dominant in the language with the more opaque gender system, they will show delay compared to
monolinguals of the same age, since there was no cross-linguistic influence and they were still exposed to the language less than monolinguals. A description of Serbian and Dutch grammatical gender systems with
4 As Dutch is a language that has DPs, it is not completely clear in the literature whether Slavic languages (among
which Serbian) have DPs. Zlatić (1998) takes the stance that noun phrases in Slavic languages are NP and not DP. Therefore, since there are less computational steps in Adj-N than in D-Adj-N strings, it could be argued that Serbian in this case has less complex derivation.
12 respect to transparency and frequency, as well as the repercussions they have for gender acquisition in the respective languages is given in Section 4.
13 4. Grammatical gender in Dutch and Serbian
This section outlines the transparency of Dutch and Serbian gender systems, but before we turn to gender systems, a brief description of the phenomenon of grammatical gender will be given, followed by what is considered transparent and what opaque gender.
With respect to grammatical gender, there are gendered and non-gendered languages. Non-gendered languages have no gender (e.g. English), while gendered language have at least two genders, that can be divided as masculine and feminine (e.g. Romance languages), animate and animate, or otherwise (e.g. Dutch). Other gendered languages have three genders (e.g. Slavic languages and German that also have the neuter gender next to masculine and feminine). Languages can also have more than three genders. In gendered languages, gender is assigned to every noun. However, the way gender is assigned to nouns differs across languages. A chair, for example, does not have any imminent natural gender as being masculine or feminine. Still, it is feminine in Italian (la sedia) and masculine in German (der Stuhl). Both Italian and German mark gender on the definite determiner (la, der), and on the noun, based on the ending (-a in Italian is mostly feminine, -ø in German is often masculine). In Slavic languages, gender is also
assigned to the noun based on the noun ending, however not on the determiner since the most Slavic languages do not have determiners. Even the Slavic languages that are very closely related, such as Serbian and Croatian, make a difference in gender on the same noun if it has a different ending (in Serbian, ‘a/the chair’ is feminine stolica, in Croatian it is masculine stolac).
The noun endings represent morphophonological cues based on which gender of the noun can be inferred from the input. If there is a one-to-one relationship between meaning and form, i.e. if the nouns ending on –a are consistently marked as feminine, than this relationship is transparent (Hengeveld 2011). If there are more different endings (form) for the same gender (meaning), out of which some endings (or the lack thereof) overlap with other genders in that language, then the relationship is less transparent. Moreover, if there are no morphophonological cues and the gender is assigned arbitrary (from synchronic perspective), this gender system will be considered to be the most opaque. In these languages, gender is acquired on a lexical word-by-word basis and takes longer to acquire. In Subsections 4.1 and 4.2 a more detailed analysis of Dutch and respectively Serbian gender systems is given. In Subsection 4.3 a comparison of the two gender systems is given, with respect to transparency.
In gendered languages, other words can also agree with the gender of the noun, such as the determiners, pronouns (with numerals) and/or adjectives. These words are inflected (or not, depending on the gender of the noun), in order to agree with the noun. To illustrate, the construction such as “my beautiful chair” would therefore be l-a mi-a bell-a sedia in Italian and mein-ø schon-er Stuhl in German. In other languages, specifically Slavic, also the participle of the verb in past tense is inflected and agrees with the gender of the noun to which it refers (see Subsection 4.2 for Serbian). The more words there are that agree with the noun through inflection, the more frequent they are in discourse. According to Trudgill (2011: 166), who looks at gender systems from a typological perspective, it is exactly “this level of frequency in discourse that permits the passing on of gender from one generation to another even though it is relatively unimportant”. He also adds that it is “the amazing language abilities of the human infant” (165) that make this passing possible. As we have already broken down those abilities according Schwartz’ (2003) Domain-by-Age model, early bilingual learners acquire inflectional morphology as monolingual children through the grammar-based approach to acquisition, while this is not the case for the late (adult) learners, who have great difficulty acquiring inflectional morphology in their L2. An overview of the acquisition studies on Dutch grammatical gender is given in Section 5.
14 4.1 Grammatical gender in Dutch
Dutch has a two-way gender system (common and neuter) and gender is not assigned on the endings of the underived nouns but through the choice of the definite determiner in the singular. This means that, if a noun is preceded by the definite determiner de, it is common gender (de boom ‘the tree’). If the definite determiner het precedes the noun, it is neuter gender (het huis ‘the house’). The gender distinction is not marked on the plural definite determiner, which is always de, nor on the singular indefinite determiner, which is always een (Table 1).
Table 1. Dutch determiner system
Singular Plural
Common Neuter Common Neuter
Indefinite een boom een huis bomen huizen
Definite de boom het huis de bomen de huizen
The frequency and distribution of the two genders in Dutch is not proportional. According to a dictionary-based estimate, roughly 75% of Dutch nouns are common and only 25% are neuter (Hulk and Cornips 2006: 269), while the relative distribution of the de-words and het-words is roughly 2:1 (Van Berkum 1996). The defnite determiner de far more frequent in the input than the definite determiner het.
Next to definite determiners, Dutch also marks gender on relative (die, dat) and demonstrative determiners (deze, die, dit, dat) and adjectives (Corbett, 1991). The attributive adjectives immediately preceding the noun also need to agree with the noun. There are two adjective forms: the bare form and the inflected form (or the schwa-form). The adjective is bare only in neuter indefinite singular condition. In all other conditions it is inflected. If we include the occurrence of neuter words in discourse, then it is clear that the inflected form of the adjective is far more frequent in the input. Examples (1)-(4) illustrate the Dutch grammatical gender paradigm in the singular:
(1) een grot-e boom
INDEF big-COM tree(COM) (2) de grot-e boom
DEF(COM) big-COM tree(COM) (3) een klein huis
INDEF small(NEU) house(NEU) (4) het klein-e huis
DEF(NEU) small-NEU huis(COM)
Since there is no morphophonological patterning on the end of the words based on which gender can be inferred, Dutch gender system is considered opaque. The only reliable morphophonological cue is suffix –je on diminutives, that is always neuter gender (de stad ‘the city’ but het stadje).
15 4.2 Grammatical gender in Serbian
Serbian distinguishes three major gender categories: masculine, feminine and neuter. Masculine nouns usually end in a consonant (dom ‘home’), feminine nouns end in –a (ruka ‘arm’), and neuter nouns in –o (selo ‘village’) and –e (more ‘sea’). Based on the salience of morphophonological cues, there is in general a one-to-one relation between meaning and form in the case of grammatical gender. However, there are sets of nouns that are less transparent: a small group of masculine nouns ending in –o (sto ‘table’), and a small set of feminine nouns ending with a consonant that are mainly abstract nouns (ljubav ‘love’) and a few concrete nouns (peć ‘oven’) (Table 2).
Table 2. Transparent and opaque noun endings across genders in singular in Serbian
Masculine Feminine Neuter
Transparent -ø dom -a ruka -o selo
-e more
Opaque -o sto -ø stvar
The adjectival endings are often the same as the noun endings, as we can see in examples (5)-(7). The same rule applies to the inflection of other words that agree with the noun: the demonstrative and possessive pronouns, the attributive numbers and the past participle.
(5) topl-i dom warm-MASC home(MASC) (6) mal-a ruk-a small-FEM hand-FEM (7) velik-o sel-o/dugm-e big-NEU village-NEU/button-NEU
Serbian makes use of an abundant repertoire of of bound morphemes for lexical and for morphosyntactic purposes. In the case of nouns, adjectives and pronouns, next to the three genders, singular and plural, they are also marked for seven cases (nominative, genitive, dative, accusative, vocative, instrumental, locative).
2.3 Grammatical gender marking in Dutch and Serbian compared
So far, we have observed the asymmetry between Dutch and Serbian gender systems in at least three aspects. Firstly, Dutch has two genders (common and neuter), where Serbian has three (masculine, feminine and neuter). Secondly, in terms of transparency, the Dutch gender system is more opaque than Serbian because there are no phonological cues on the underived nouns on which a language learner can rely. Thirdly, Serbian marks gender on more words that agree with the noun than Dutch. This difference can be exemplified by the following example in Serbian (8):
(8) Juče je ovde bil-a jedn-a crven-a knjig-a. Yesterday AUX here was-FEM one-FEM red-FEM book-FEM. ‘There was a red book here yesterday.’
16 Je.li t-a knjig-a još tu?
Is that-FEM book-FEM still here? ‘Is that book still here?’
Where Serbian marks gender consistently with the same ending on five words: the noun (knjig-a), the adjective (crven-a), the numeral (jedn-a), the demonstrative pronoun (t-a) and the past participle (bil-a), the Dutch language would mark ik only two times: on the adjective and on the definite determiner. Based on the presence of abundant gender marking in the input, along with the transparency of the
morphophonological cues, it can be hypothesized that grammatical gender in Serbian is acquired faster by monolingual children than by bilinguals. A study that was done primarily on the adjectival agreement with diminutives in Serbian also tested adjectival agreement with masculine and feminine nouns with
transparent endings (Ševa et al. 2007), corroborates the hypothesis that target gender agreement in Sebian (at least masculine and feminine) is acquired early by monolinguals. The adjectives in the study were elicited after the experimenter showed the picture to the child and named the animal on the picture, followed by a question about its characteristic in the form of an adjective. The experiment was tested on 22 children aged 3;0-4;1 (M = 3;7), and children performed at ceiling (100%) for agreeing the adjective with underived masculine familiar nouns, and highly (95,4%) with the underived feminine familiar nouns. The study also tested gender agreement in Russian by monolingual children of the same age and found no significant differences between the two languages. Therefore, a cautious comparison can be made with Russian in terms of neuter gender and non-transparent endings. Gvozdev (1961) found that Russian monolingual children acquired the transparent forms of masculine and feminine gender by the age of 2;4, while the neuter nouns were not acquired before the age of 2;5. The nouns with opaque endings (similarly to Serbian for the feminine gender ending on consonant) were acquired rather late, around the age of 7. Cited age ranges for the acquisition of grammatical gender in Serbian (and Russian) differ from the acquisition of this language phenomenon in Dutch. In Section 5 we look into more detail the studies that tested the acquisition of grammatical gender in Dutch, with both monolingual and bilingual child language users.
17 5. The acquisition of Dutch grammatical gender
In this Section, an overview of the studies that were done so far on the acquisition of Dutch grammatical gender with monolingual and bilingual children will be given, with respect to gender assignment
(determiner choice), gender agreement (adjectival inflection) and cross-linguistic influence.
5.1 Gender assignment
Differently from the languages that mark gender in a transparent way, which is in turn argued to be acquired early by monolingual children in the respective languages (e.g. for Spanish, see Pérez-Pereira 1991; for Russian, see Rodina and Westergaard 2012), Dutch grammatical gender has often been found acquired not before the age of 6 or 7 by monolingual children (see, among others, Van der Velde 2003). This is commonly believed to be the case due to the lack of morphophonological cues on the nouns endings5 in the Dutch input, and the gender needs to be acquired by children on a lexical, word-by-word
basis (Blom et al. 2008).6
Monolingual Dutch children make errors in gender assignment by overgeneralizing the definite determiner for common nouns to neuter nouns (Van der Velde 2003, Blom et al. 2008, Unsworth and Hulk 2010, Blom and Vasić 2011, among others). However, they are not found to overgeneralize in the opposite direction, by producing het with common nouns (Van der Velde 2003). Still, It should be noted that there is a ‘statistically insignificant’ number of monolingual children in the studies who also overgeneralized the determiner for neuter nouns to common nouns. In Blom et al. (2008), the reported age of a moderate het overuse (7-18%) is between 5 and 7 years. Blom and Vasić (2011) also report 2 out of 21 monolingual children aged 5;0-5;11 who frequently overused het with common nouns. The children that overuse het with common nouns are usually reported to already frequently use het with neuter nouns. Finally, Unsworth and Hulk (2010) divided their monolingual participants (4;4-7;6) in two groups based on their vocabulary scores. In the production task, the group with the higher vocabulary scores outperformed the children from the lower group in the neuter condition (66% vs. 41%). In the judgment task, however, the children from the higher group scored still higher on the neuter nouns, but lower on the common nouns than the children from the lower group (87% vs 92%). In other words, in judgment they also overused the definite determiner for neuter nouns to common nouns. As the authors conclude, this result could be a developmental step due to the discovery of the paradigmatic relation between the two definite determiners. At this stage of
development, the children are expected to also show some hesitation with respect to common nouns, by sometimes incorrectly assuming that certain common nouns are neuter.
Bilingual children follow in general the pattern of monolingual children when assigning gender to nouns. As already mentioned, bilinguals can be divided in three groups based on their age of onset: simultaneous bilinguals (2L1, age of onset 0;0-1;0), early successive bilinguals (ESB, age of onset 1;1-4;0) and second language children (L2, age of onset 4;1-10;0). In general, just like monolinguals, the bilinguals greatly overuse the definite determiner for common nouns with neuter nouns (Hulk and Cornips 2006, Cornips et al. 2006, Unsworth 2007, Blom et al. 2008, Blom and Vasić 2011, Hulk and van der Linden 2011, Unsworth
5 Except in the case of diminutives, which have a typical ending –je (or its allomorphic variations) and that are always
neuter. For a study on the acquisition of Dutch diminutives see also Blom and Vasić 2011).
6 On the presence of transparent endings that are abundant in the input, it should be however noted that in a recent
study no (quantity of) input effects were found between the acquisition of an opaque (Dutch) and a transparent (Greek) gender system in English-Dutch and English-Greek bilinguals, when they were cumulatively exposed to both languages in question to the same amount (Unsworth et al. 2014).
18 2013). As for the production of the definite determiner for neuter nouns with common nouns, some studies have reported this finding. For example, in Unsworth (2007), who tested 2L1 and L2 Dutch-English bilingual children (5;3-17;4) on determiner choice, found that the definite article for neuter nouns was produced with common nouns in as much as 25,4%. After a closer inspection, it turned out that children with intermediate or high levels of proficiency exhibit this pattern more than children with lower levels of proficiency (14,1% for intermediate and 11,1% for high compared to 2,1% for low), who mostly used the definite determiner for common nouns over the board. Similar findings were reported by Blom et al. (2008), who tested 64 Moroccan/Arabic-Dutch bilinguals on gender assignment and agreement. The group of more proficient bilinguals (aged 6;2-8;4) overused the definite determiner for neuter gender with common nouns is 10%, while the percentage for the bilinguals who were less proficient (aged 4;2-8;3) was 0%. It might have been the case that the more proficient L2 children have only started to discover the paradigmatic relation of Dutch gender, since they have correctly assigned the definite article for neuter nouns to neuter nouns more than less proficient L2 children (15% vs 3%).
Also older bilingual children aged 10;5-12;1 of Moroccan/Arabic-Dutch and Turkish-Dutch language
backgrounds were also reported to overgeneralize the definite determiner for neuter gender with common nouns more than monolinguals of the same age (23,6% vs. 4,2%). Proficiency of the subjects was not tested and therefore no conclusions based on proficiency can be made.
Still, based on these studies, the following trends can be observable in the assignment of Dutch gender. In the early stages of acquisition, both monolinguals and bilinguals overgeneralize the definite determiner for common gender to neuter nouns. As their proficiency in Dutch develops, once they discover the paradigm for two genders in Dutch language, or the existence of neuter gender, both monolinguals and bilinguals are expected to overuse the definite determiner for neuter gender with common nouns. However, the
bilinguals have been reported to do this more than monolinguals of the same age. Moreover, the bilinguals were found in the respective studies that tested development to show delay in the acquisition of the definite determiner for neuter gender (they discover the paradigm later), compared to monolinguals of the same age.
The observed delay does not mean that the bilinguals performed worse than monolinguals in terms of length of exposure. Therefore, the comparison with the monolinguals of the same age may not be
completely fair since the bilinguals receive less input than their monolingual peers. In order to account for this difference, Unsworth (2013) gathered information about the quantity of the Dutch input from the parents of the simultaneous Dutch-English bilinguals in her study and measured it in terms of cumulative exposure. In the comparison, she matched them with the monolinguals according to the amount of exposure. In other words, the bilingual 3, 4 and 5 year olds were compared with a group of monolingual 2 year olds, the bilingual 6 year olds were compared with the monolingual 3 year olds, the bilingual 7 and 8 year olds with the monolingual 4 year olds, the bilingual 9, 10 and 11 year olds with the monolingual 5 year olds, the bilingual 12 and 13 year olds with the monolingual 7 year olds and the bilingual 14 to 17 year olds with a group of monolingual 8 year olds. There were no differences observed in terms of gender
assignment between the compared groups.
If gender is assigned lexically and can be successfully assigned to the nouns with the sufficient amount of input in the case of simultaneous bilinguals, then there is still the question if the same would be true for ESB and L2 children. Unsworth et al. (2014) tested gender assignment on simple (without adjective) and complex DPs (with adjective) with 2L1, ESB and L2 Dutch-English bilinguals and found no difference in production of target DPs between different ages of onset of the groups, although 2L1 (27%) outperformed ESB (14%) who in turn outperformed L2 (11%) on the target neuter gender. Ntalli (2013) compared
19 different ages of onset of Greek-Dutch bilinguals with respect to Dutch grammatical gender aged 4;4-13;3 and also found no significant differences between the groups in her study. She points out that this might have been due to the small number of children in the ESB group in her study (4), as she adds that 2L1 children outperformed and ESB and 2L1 children in the neuter condition, while in the judgment task 2L1 and ESB children outperformed L2 children. However, this trend does not differ from the two groups of monolinguals reported in Unsworth and Hulk 2010. As ESB and L2 children have later ages of onsets, they are even less exposed to Dutch input than 2L1s and are in turn expected to be less proficient. By comparing the findings in production and judgment tasks in Ntalli’s study, it might be the case that ESB children are discovering the paradigm, while the input was still not sufficient enough for L2 children to discover the paradigm.
All studies mentioned so far on the gender assignment in Dutch that tested development across the bilinguals’ age groups found (at least some) development across the groups. The discovery of the gender paradigm in Dutch is shown to be correlated with the proficiency of the children, which has in turn been correlated with the amount of Dutch input the bilinguals received. Contrary to expectations, the age of onset did not show any significant effects across the groups in the respective studies.
5.2 Gender agreement
The grammatical gender in Dutch is, next to the determiner that precedes the noun, also marked on the attributive adjective that is placed between the determiner and the noun. However, as the choice of the determiner is acquired lexically, the acquisition of the adjective takes a different route, as it is acquired morphosyntactically. In other words, the child needs to have access to the feature [±neuter] in their grammar (Blom et al. 2008). Based on Meisel’s (2007) critical period, Blom et al. argue that the rule for adjectives needs to be activated before the age of 4 in order for it to be adequately acquired. In their study on the acquisition of grammatical gender by L1 Dutch children and L2 Moroccan/Arabic-Dutch children and adults, although some development was observable in the target assignment between the two L2 groups with different proficiency, the children seemed to fossilize with respect to gender agreement (target in 94% and 98% respectively with common gender and only 9% with neuter gender). This pattern, however, was not observed in adults, who according to the authors produced non-target adjectival endings with common nouns significantly more than children (71-77%). The authors have further conducted a consistency analysis in order to see how many nouns that were consistently assigned with the same gender (irrespective of it being incorrect or correct), were also used with the correct inflection of the adjective based on the gender that is consistently assigned to the noun by the child. Unlike L2 adults who were inconsistent with neuter gender, the L2 children were highly consistent with common gender, and sometimes with neuter gender, if they consistently produced any neuter nouns at all. L2 children with lower proficiency produced 0
consistent neuter nouns, while the group with higher proficiency produced 2% of inconsistent neuter nouns and 1% of consistent neuter nouns. Since monolingual Dutch children showed a paralleled development of gender assignment and gender agreement while it remained unparalleled in L2 children, Blom et al. (2008) conclude that gender agreement the L2 children in their study seemed to fossilize.
However, it seems rather difficult to claim that gender agreement of L2 children in Blom et al. (2008) has fossilized based on such a low level of consistently assigned neuter gender to the nouns by the child (irrespective of the nouns’ actual gender). This could be an implication that these children have just started to discover the Dutch gender paradigm, and that it is occurring late due the insufficient input they received, and not because they have passed the critical period for accessing the grammatical feature as they still
20 have access to the [±neuter] feature since their consistency pattern more closely resembled that of L1 children than that of L2 adults.
5.3 Cross-linguistic influence
In the studies on the acquisition of Dutch grammatical gender, the possibility of cross-linguistic influence has often been addressed. In order for the cross-linguistic influence to take place, the other language that the child is acquiring apart from Dutch also needs to have a gender system. If that gender system is more transparent than the already opaque Dutch gender system, then it can be hypothesized that the mere existence of a gender system in the child’s grammars could influence and aid the target acquisition of the Dutch grammatical gender. In their study on the acquisition of definite determiners in Dutch by bilingual children of various language backgrounds, Cornips and Hulk (2008) found no difference in target acquisition of Dutch determiners between the groups of children with gendered languages (e.g. feminine and
masculine in Moroccan/Arabic) and non-gendered languages (e.g. Turkish). As the authors argued, there was not enough structural overlap between the languages in question on the one hand, and Dutch on the other. Similar results were reported with older children (10;5-12;11) in Cornips et al. (2006). The study included 12 Moroccan/Arabic-Dutch bilinguals, 13 Turkish-Dutch bilinguals and 5 Dutch monolinguals. Both groups performed similarly with respect to gender assignment (approx. 70% for target common nouns and 40% for target neuter nouns) and with respect to gender agreement (approx. 86% and 99% target inflected adjectives and 27% for target bare adjectives). Monolinguals of the same age were more correct in gender assignment with neuter nouns (approx. 70%) and in the target use of the bare adjective (approx. 50%). The authors conclude that no cross-linguistic influence was observed.
However, in another study on Dutch-Spanish and Dutch-French bilinguals, Hulk and Van der Linden (2010) found acceleration effects in the acquisition of Dutch gender, compared to their monolingual age mates. Spanish and French, as other Romance languages, have transparent gender marking, and, although they have two genders, like Dutch, their gender systems do not overlap: Romance languages distinguish
masculine and feminine, whereas Dutch distinguishes common and neuter. Hulk and Van der Linden found that where Dutch 4-year-old monolinguals in Blom et al. (2008) overgeneralized the definite determiner for common gender to neuter nouns in 56% of the cases, the Dutch-French children of the same age
overgeneralized only 32,5%. The same holds for Dutch-Spanish children who overgeneralized 57%, compared to 88% of 3-year-old monolingual children in Blom et al.’s study. The authors attributed these results to the possibility that the transparency of the French and Spanish gender systems could have increased the awareness of gender marking in general, and led to earlier gender acquisition in Dutch. Both Moroccan-Arabic and Romance languages distinguish two genders: masculine and feminine. How do we account for the difference in the presence of cross-linguistic effects in the two bilingual groups
observed in previous studies? Cornips et al. (2006) suggest that there were probably no cross-linguistic effects observed in Moroccan-Arabic-Dutch bilingual children due to the quality of input the children got from their (grand-)parents who have learned Dutch in non-instructed contexts and are more prone to “errors” in speech. More importantly, this input comes from both parents. In contrast, the majority of Spanish-Dutch and French-Dutch bilinguals comes from middle-class families where one parent is Dutch and the other parent speaks the native speaker of the other language in most cases. The bilingual Greek-Dutch children in Ntalli (2013) are from a similar socio-linguistic environment as the subjects in the study by Hulk and Van der Linden (2011) and are evidenced to have shown acceleration effects in their acquisition of the Dutch grammatical gender, compared to monolingual children of the same age.
21 Ntalli (2013) conducted a series of tests on gender assignment and gender agreement in Dutch and Greek with 33 bilingual Greek-Dutch children aged 4;4 to 13;3. In contrast with Dutch, Romance languages and Moroccan-Arabic, Greek differentiates three genders: masculine, feminine and neuter. Gender of the nouns is transparent as there are phonological cues on the nouns that make gender assignment easier and faster for the speaker. Simultaneous bilingual children were divided in three group, based on age. The youngest group was aged 4;4-7;0, the middle group 7;3-9;6 and the oldest group 11;3-13;3. In the Picture Description Task (PDT), that tested primarily the adjectival agreement (but also gender assignment through the
determiner choice) the bilinguals outperformed monolinguals in both common (100%) and neuter condition (65%), compared to their age-matched monolinguals (98.63% and 44.30% in Unsworth 2013) in the middle group, as reported in Unsworth (2013). This pattern was also present in the oldest group: bilinguals scored 100% in common and 78% in neuter condition, outperforming the monolinguals reported in Cornips et al. (2006). The same PDT task will be used for the purposes of this thesis and the comparison with that of Ntalli (2013), who proposed that the subjects in her study did better on the neuter gender than the monolinguals due to the fact that neuter gender is the default gender in Greek.
To summarize, it has been observed that gender assignment in Dutch is acquired on a lexical word-by-word basis and that it is therefore dependent on the quantity and the quality of input, while the adjectival agreement with the gender of the nouns takes a syntactic path since it is automatically applied to the nouns after the paradigm of the Dutch gender has been acquired by the monolingual child. Monolingual children firstly overuse the determiner for common gender with neuter nouns and accordingly consistently inflect the adjective, until they discover the paradigm of the Dutch two-gender-system (between the ages of 5 and 7, during which period they overuse the definite determiner for neuter gender with common nouns to some extent). The bilingual children follow in general this pattern, but since they receive less Dutch input than monolinguals, their discovery of the paradigm is often delayed, compared to
monolinguals of the same age. However, in terms of length of exposure, the bilinguals have been shown to perform as good as monolinguals. On the other hand, in the studies that claimed to have found cross-linguistic influence in this domain, the results showed that bilinguals outperformed the monolinguals of the same age on neuter gender. Based on these findings, the hypothesis and predictions of the current study will be formulated in Section 6.
22 6. Hypotheses and predictions
The acquisition of Dutch grammatical gender by early bilinguals is the topic under investigation in the current study as well as the effects of cross-linguistic influence in the process of acquisition. For this purposes, the other language of the bilinguals is also a gendered language that was not previously investigated in this respect, namely Serbian. As became clear from the description of the two gender systems and their acquisition, Dutch system is more opaque as there are no phonological or other cues based on which gender can be assigned to underived nouns. On the other hand, the grammatical gender in Serbian is more transparent and overtly marked on words. This difference is also manifested in the
different periods by which monolingual children acquire gender in the respective languages. In Dutch, grammatical gender is not acquired adequately before the age of 6 since children greatly overgeneralize the definite determiner for common gender with otherwise neuter nouns. On the other hand, the
acquisition of grammatical gender in Serbian is not expected to be acquired later than the age of 3, at least for the transparent endings of the nouns.
The bilingual children follow in a similar path like monolinguals in the acquisition of Dutch grammatical gender, but since they receive less Dutch input than monolinguals, they are delayed in acquisition, compared to monolinguals of the same age. However, in terms of length of exposure, the bilinguals have been shown to perform as good as monolinguals. On the other hand, previous studies on cross-linguistic effects in the acquisition of Dutch gender with bilinguals found acceleration effects with Spanish-Dutch, French-Dutch (Hulk and Van der Linden 2010) and Greek-Dutch (Ndalli 2013) bilingual children, compared to monolingual children of the same age, as the former group performed better on the neuter gender even though they received less input than the monolinguals. The bilingual children from these studies come from middle-class families with educated parents, where one parent is most often Dutch while the other parent is the native speaker of the other language and the child. The bilingual Serbian-Dutch children in the Netherlands belong to a comparable socio-linguistic environment. It seems therefore plausible to
hypothesize that there will be cross-linguistic influence from Serbian to Dutch in Dutch-Serbian bilinguals. The Serbian and Dutch gender systems also partly overlap; even though both languages have neuter gender, the number and order of genders is asymmetrical: where Dutch has common gender, Serbian differentiates masculine and feminine. In this respect, Serbian is more comparable to Greek, since both languages have three genders that are transparently marked. It should however be stressed that Serbian does not use determiners, which are in both Dutch and Greek marked for gender. In this study, the DPs in respective languages are not expected to influence the acquisition of grammatical gender, but rather that of definiteness, which is beyond the scope of this thesis. Based on the mentioned findings, we formulate our first hypothesis:
(i) Dutch-Serbian bilingual children will assign the target definite determiner for neuter gender faster than Dutch monolinguals of the same age and cross-linguistic influence will be found. The mere existence of grammatical gender in both languages, with the gender assignment in Serbian being more transparent than in Dutch will be taken as a possible route for cross-linguistic influence from the Serbian to Dutch gender system to take place in the form of acceleration, i.e. the faster assignment of target definite determiner to common nouns, compared to monolingual children of the same age. If the bilinguals perform similarly as monolinguals of the same age, then this will also be observed as
acceleration, seeing that the bilinguals receive two times less input than monolinguals.
In order to investigate this hypothesis, production data of Dutch-Serbian bilingual children with respect to gender assignment and gender agreement will be analyzed and compared with the data from monolingual
23 children of the same age that were reported in the previous studies (Cornips et al. 2006, Blom et al. 2008, Unsworth 2013). The data will be also compared with the results of bilinguals of the same age who are acquiring Dutch together with a non-gendered (Unsworth 2013) and a gendered (Ntalli 2013) language. If our hypothesis is supported, then the Serbian bilingual children should perform better than Dutch-English children, and similarly to Greek-Dutch children on the neuter gender.
In case the prediction is borne out and the influence does occur, one more factor will be addressed. It is not clear from the previous literature whether dominance plays a role in the likelihood of cross-linguistic influence. If dominance plays a role, then the more dominant the child is in Serbian, the more likely for cross-linguistic influence to happen. If, however, the child is already sufficiently dominant in Dutch, they are expected to show delay compared to monolinguals of the same age, since there was no cross-linguistic influence and they were still exposed to the language less than monolinguals. However, Ntalli (2013) did not observe dominance effects between the two groups of bilingual subjects in her study with different levels of dominance in Greek. Therefore, our second hypothesis will be formulated as a null-hypothesis:
(ii) The observed cross-linguistic effects in Dutch-Serbian bilingual children will not correlate with their level of dominance in Serbian.
In order to examine this prediction, participants will be divided in two groups based on the relevant answers from parental questionnaire. One group will consist of children with suspected higher dominance in Serbian, while the second group will include bilinguals with lower levels of dominance. Their
performance will be compared. If both groups perform similarly on neuter nouns, then our second hypothesis will be supported.
