UNIVERSITY OF AMSTERDAM Graduate School of Humanities July 2013 On the Acquisition of Grammatical Gender – When ‘Late Phenomenon’ meets ‘Early Phenomenon’ An Empirical Study with Dutch-Greek Bilingual Children and its Relevance for Investigations of Bilingual SLI by Evelyn Egger Thesis submitted in partial fulfillment of the requirements for the degree of Master of Arts in Clinical Linguistics Supervisors: Dr. J. de Jong Prof. Dr. A.C.J. Hulk ABSTRACT The present study investigates the acquisition of grammatical gender in a group of Greek-Dutch bilingual children aged 4;4 – 13;3 years. The subjects were tested on determiner-noun and determiner-adjectivenoun agreement in both Dutch and Greek and were grouped according to their scores on a vocabulary test in the respective languages. Results for the Dutch tests revealed a tendency for the Greek-Dutch bilinguals to overgeneralize the neuter determiner het to common nouns which was taken as evidence for increased gender awareness. Given that the English-Dutch bilinguals in Unsworth (2013a) and the ItalianDutch bilinguals in Algoe (2011) did not show overgeneralization of het, it is argued that the way gender is marked in a bilingual’s ‘other language’ affects the acquisition of grammatical gender in Dutch. In contrast, performance patterns on the Greek tests were in line with previous findings from Mastropavlou (2006) for Greek monolinguals and with Unsworth, Argyri, Cornips, Hulk, Sorace & Tsimpli (2011b) for English-Greek bilinguals suggesting no cross-linguistic influence from Dutch to Greek. Based on previous findings from Greek and Dutch language-impaired monolinguals, it is suggested that the positive cross-linguistic influence from Greek on the acquisition of Dutch gender may also extend to languageimpaired bilingual populations. Keywords: grammatical gender, agreement, bilingual acquisition, cross-linguistic influence MA Thesis 2013 Evelyn Egger Table of Contents 1 Introduction ..................................................................................................................................................4 2 Gender ........................................................................................................................................................10 2.1 Gender Assignment vs. Gender Agreement ........................................................................................10 2.2 Gender in Dutch ..................................................................................................................................13 2.2.1 Descriptive Notes .........................................................................................................................13 2.2.2 Acquisition of Grammatical Gender in Dutch .............................................................................15 2.3 Gender in Greek ..................................................................................................................................21 2.3.1 Descriptive notes ..........................................................................................................................21 2.3.2 Acquisition of Grammatical Gender in Greek .............................................................................24 3 The Study ...................................................................................................................................................25 3.1 Background .........................................................................................................................................25 3.2 Method ................................................................................................................................................30 3.2.1 Participants ...................................................................................................................................30 3.2.2 Materials ......................................................................................................................................31 3.2.3 Procedure .....................................................................................................................................36 3.2.4 Coding and analysis .....................................................................................................................36 3.3 Results .................................................................................................................................................38 3.3.1 Background measures ..................................................................................................................38 3.3.2 Dutch Production Tasks ...............................................................................................................40 3.3.3 Dutch Knowledge Task (GJT) .....................................................................................................45 3.3.4 Comparisons ................................................................................................................................46 3.3.4 Greek Production Tasks ...............................................................................................................51 3.3.5 Greek Knowledge Task (GJT) .....................................................................................................57 3.3.6 Comparisons ................................................................................................................................58 4 Discussion ..................................................................................................................................................61 4.1 General Discussion .............................................................................................................................61 4.2 Relevance & implications for the study of bilinguals SLI ..................................................................66 5 Conclusions ................................................................................................................................................69 References .....................................................................................................................................................70 APPENDICES ..............................................................................................................................................75 3 MA Thesis 2013 Evelyn Egger 1 Introduction Grammatical gender is an intriguing phenomena present in many of the world’s languages. Research on the acquisition of grammatical gender systems indicates that it is a particularly vulnerable area for second language learners (e.g. Montrul, Foote, & Perpiñán, 2008; Dewaele & Veronique, 2001), with many showing signs of fossilization in the form of persistent difficulties with gender marking (e.g. Francheschina, 2005). For children growing up bilingually from a young age, the general finding is that they lag behind monolinguals in the acquisition of gender, which is usually attributed to factors related to language input or exposure (e.g. Hulk & Cornips, 2006a; Unsworth, Argyri, Cornips, Hulk, Sorace & Tsimpli, 2011b). Results for children growing up bilingually from birth are more mixed with some researchers finding that simultaneous bilinguals show a similar developmental pattern than monolinguals albeit with a slight delay (e.g. Kupisch, Müller & Cantone, 2002), while others report a positive influence of the dual-language exposure in the form of an accelerated rate of acquisition (e.g. Cornips & Hulk, 2006). Many explanations have been put forward to account for the differences and similarities between monolingual and bilingual acquisition of gender, the most frequently discussed factors being age of onset, length of exposure, input quality and the nature of the L1 (see Cornips & Hulk, 2008 and Unsworth, Hulk & Marinis, 2011 for a discussion of the relevant factors and related findings). Grammatical gender has also been found to be an affected area in atypical language development in many languages (see Keij, Cornips, van Hout & van Emmerik, 2012; Orgassa, 2009 for Dutch; Silveira, 2011 for Portugese; Clahsen, 1989 & 1991 for German). Such findings complicate the picture of bilingual language acquisition because the linguistic profiles of these two populations (bilinguals and children with SLI) may overlap in a given language, which could lead to confounds and even potential misdiagnoses of bilingual children as having SLI and of bilingual children with SLI as simply demonstrating the familiar ‘bilingual delay’ (Paradis, 2005). To avoid such confounds, it is important to carefully investigate the developmental paths of various monolingual and bilingual populations with and without SLI and compare the groups against each other in order to be able to draw a boundary between typical second language acquisition and language pathology. Although grammatical gender has been identified as an area of persistent difficulty for both second language learners (including bilinguals) and language-impaired children, it is clear that the linguistic profiles of the various populations are subject to cross-linguistic variation. One obvious reason for the attested variation are structural differences between the languages under investigation and in the present case more specifically, the manifestation and marking of grammatical gender. Grammatical gender is a means of nominal classification present in many languages of the world. According to Corbett (1991) and many other linguists in the field, a defining characteristic of grammatical gender systems is 4 MA Thesis 2013 Evelyn Egger agreement, i.e. the overt marking of gender on other elements outside the head noun itself. Thus, a basic distinction that pertains to grammatical gender is made between gender assignment and gender agreement. Gender assignment refers to the categorization of all nouns of a language into a small number of gender categories, the most common distinctions being masculine-feminine, masculine-feminineneuter and common-neuter, although systems based on other criteria (e.g. animacy) exist as well. Gender assignment of nouns is usually determined by a combination of both semantic and formal (morphological or phonological) rules. This is because semantic criteria such as biological sex (masculine vs. feminine) cannot be applied to inanimate entities, which is why many languages pose an additional set of morphological or phonological rules for the gender assignment of such nouns (for example in German, all nouns ending in –heit are feminine). As a consequence, gender assignment in a given language may appear more or less arbitrary, depending on the combination of semantic and formal assignment rules, as well as the amount of exceptions to these rules. Thus, grammatical gender is said to be an inherent property of nouns and as such, it is assumed to be part of the lexical entries of nouns that has to be learned (at least partly) on an item-by-item basis. With regard to the defining property of grammatical gender, i.e. agreement, there is a general consensus that it is a purely syntactic process, which in the generative framework (e.g. Chomsky, 1995), is typically described in terms of the matching or sharing of features between related elements. In other words, agreement is the systematic co-variance of related forms, i.e. the head noun (also called agreement trigger or controller) and all other elements that lie within the scope of agreement. Just as languages differ in terms of gender assignment rules, so do they with regard to the scope of agreement marking, i.e. the number and types of elements that have to be overtly marked for gender. For example, Modern English has a very restricted gender system, which is primarily based on natural gender and only surfaces on pronominals. Thus, English only shows (natural) gender agreement between singular pronouns and antecedents as illustrated in (1) and (2) 1. (1) Yesterday I saw Mary [FEM] . She [FEM] was taking her [FEM] dog for a walk. (2) Yesterday I saw George [MASC] . He [MASC] was taking his [MASC] dog for a walk. In contrast to English, German also shows gender agreement in the wider nominal domain, namely on determiners and attributive adjectives, as exemplified in (3) and (4). Although gender marking is more extensive in German than in English, the high number of syncretic forms (e.g. the definite article for plural nouns of all three genders is die; the indefinite article for both neuter and masculine nouns in the nominative singular is ein, etc.) makes the German gender system appear rather opaque. 1 Note that the pronominal gender system of Modern English differs from the more proper instances of grammatical gender systems (e.g. in Romance languages) in that gender is no longer an inflectional category, i.e. overtly marked by means of affixes, but it is said to be ‘covert’. 5 MA Thesis 2013 (3) Evelyn Egger Ein kleiner A-MASC little-MASC Hund sass vor der Tür. dog-MASC sat in front of the door. Katze sass vor der Tür. cat-FEM sat in front of the door. ‘A little dog sat in front of the door.’ (4) Eine kleine A-FEM little-FEM ‘A little cat sat in front of the door.’ Grammatical gender in Italian is considerably more transparent compared to the German system. In Italian, gender is not only expressed on determiners and attributive adjectives (ex. (5)), but also on predicative adjectives (ex. (6)) and in most cases, also on the head noun itself with masculine singular nouns typically ending in –o (ex. (6)) and the majority of feminine singular nouns in –a (ex. (5)). Moreover, Italian exhibits gender concord in a variety of constructions involving past participles, such as the present perfect of verbs that take essere (‘to be’) as an auxiliary (ex. (7), (8)) and past participles that are preceded by a direct object pronoun (ex. (9), (10)), among others. (5) La bambin-a piccolo-a canta un canzon-e. The-FEM girl-FEM little-FEM sings a-MASC song-MASC. ‘The little girl sings a song.’ (6) Il libr-o è ross-o. The-MASC book-MASC is red-MASC. ‘The book is red.’ (7) Massimo/lui è partit-o. Massimo/he-MASC has left-MASC ‘Massimo/He has left.’ (8) Maria/lei è Maria/she-FEM has partit-a. left-FEM. ‘Maria/She has left.’ (9) Pietro ha ricevut-o la letter-a. Peter received-PP.MASC the-FEM letter-FEM has-3.SG ‘Peter has received the letter.’ 6 MA Thesis 2013 (10) Evelyn Egger L’ It-FEM ∅ ha ricevut-a. has-3.SG received-PP. FEM ‘He has received it.’ As can be seen from the examples above, grammatical gender marking is subject to considerable cross-linguistic variation in terms of scope and transparency. Of the various kinds of agreement, the focus of the present paper will lie on nominal agreement 2 and more specifically, on determiner-noun and adjective-noun agreement. We assume the basic structure of a noun phrase (or DP) to be as in (11), although the sequential ordering of the constituents is subject to cross-linguistic variation (e.g. attributive adjectives in Italian typically occur in post-nominal positions, while in German they occur in prenominal positions). (11) Given the cross-linguistic differences in gender assignment and agreement marking, it is reasonable to expect that the acquisition of grammatical gender is subject to high levels of variability across languages and populations. While most research on the acquisition of grammatical gender has focused on determiner-noun agreement, studies on adjectival agreement are rather scarce. A general finding that pertains to both monolinguals and bilinguals is that transparent and extensive marking of gender facilitates acquisition. For example, many Romance languages exhibit a high degree of transparency with regard to gender marking (e.g. Spanish and Italian) which leads to early acquisition of gender compared to other morpho-syntactic phenomena in these languages (e.g. Kupisch, Müller & Cantone, 2002; Eichler, Jansen & Müller, 2012). In contrast, in languages with more restricted and opaque gender marking, both monolingual and bilingual children have been observed to show a 2 As opposed to, for example, agreement in the verbal domain (e.g. subject-verb agreement). 7 MA Thesis 2013 Evelyn Egger protracted course of acquisition (e.g. Blom, Polisenska & Weerman, 2008 for Dutch, Gathercole & Thomas, 2005 for Welsh; Rodina & Westergaard, 2013 for Norwegian). Moreover, differences between monolingual and bilingual acquisition of grammatical gender in a given language can be more or less pronounced. Some researchers have found highly similar developmental paths between monolinguals and bilinguals, while others observe both quantitative and qualitative differences between the two populations (e.g. Hulk & Cornips, 2006b). Another interesting observation is that not all genders of a language seem to be acquired with the same ease or difficulty. For example, in many Romance languages, masculine gender is thought to be the unmarked form that functions as a default during early stages of acquisition. This is reflected in a tendency to overgeneralize masculine gender to feminine nouns and consequently, lower accuracy rates for feminine than masculine gender (e.g. Nicoladis & Marchak, 2011 for French; Canta, 2012 for Italian). With regard to differences between the various types of agreement, a number of studies have found evidence that determiner-noun agreement poses less difficulty than adjective-noun agreement (e.g. Nicoladis & Marchak, 2011; Canta, 2012; Montrul & Potowski, 2007). A number of explanations have been put forward to account for this difference between determiner-noun and adjectivenoun agreement. One obvious factor is frequency, that is, determiners are among the most frequently used lexical items in most languages. Moreover, although gender marking in the form of agreement is a defining characteristic of gender languages, not all languages exhibit gender marking on nouns and in those languages where gender is overtly marked on nouns, it is not always consistent and usually a fair number of exceptions exist. This has led to the assumption that children use the gender of the determiner as a cue for the gender of nouns (Lew-Williams & Fernald, 2007), which would explain the higher accuracy rate on determiner-noun agreement compared to adjective-noun agreement. Others have proposed an account in terms of syntactic complexity, namely that adjectival agreement involves an additional syntactic derivation compared to determiner-noun agreement and that similarly, the derivation of post-nominal adjectives requires even more syntactic operations (i.e. movement) and is thus, more complex than the derivation required for the agreement of pre-nominal adjectives (Roulet-Amiot & Jakubowicz, 2006). Although studies on the acquisition of gender agreement have offered important insights into both cross-linguistic variation and the developmental pattern of various populations, more research is needed to come to a better understanding of the issues involved and to extend the findings to less-well studied populations. The purpose of the present paper is twofold. The first part discusses gender acquisition in typically developing populations and investigates the acquisition of gender agreement in a group of Dutch-Greek bilinguals. The motivation for this is the fact that there is a stark contrast between the developmental patterns that have been attested in the two languages. While research on the acquisition of 8 MA Thesis 2013 Evelyn Egger the Dutch gender system has revealed persistent difficulties with gender assignment and agreement in all populations (e.g. Blom, Polisenska & Weerman, 2008; Cornips, van der Hoeck & Verwer, 2006; Hulk & Cornips, 2006b), studies investigating gender acquisition in Greek have found quite the opposite, at least with regard to monolingual and early bilingual populations (e.g. Stephany, 1995, Mastropavlou, 2006, Tsimpli, 2003). Given that the nature of the L1 or ‘other language’ has been found to influence bilingual acquisition (Cornips & Hulk, 2006; Eichler, Jansen & Müller, 2012), it is interesting to see how the Dutch-Greek bilinguals compare to other bilingual populations. With regard to gender acquisition in Dutch, the first research question thus asks whether bilingual children whose other language has a highly transparent system and shows extensive gender marking (like Italian or Greek) have an advantage over bilinguals whose other language shows opaque or no gender marking (e.g. English). To this end, the performance and error patterns of the subjects under investigation will be compared to those found in other studies on gender acquisition that have used the same kind of tasks and material (i.e. Unsworth, 2013a; Algoe, 2011) to see whether the transparency of the gender system in the other language somehow facilitates the discovery of the Dutch gender paradigm. The second and third research questions concern the relationship between determiner-noun agreement and adjectival agreement. Thus, the present study seeks to replicate the finding of an asymmetry between determiner-noun and adjective-noun agreement (with the latter yielding higher error rates) and parallel improvement for the two types of agreement with increasing proficiency in both Dutch and Greek. As a fourth objective, we will examine the error types that Dutch-Greek children make with adjectival agreement in Greek and compare the error patterns to those found in other studies (i.e. Canta, 2012 for Italian). The fifth research question to be addressed in the paper is whether the Dutch-Greek bilinguals show similar performance patterns as previously attested in the two languages, namely the finding of an asymmetry between common and neuter nouns in Dutch (i.e. high accuracy rates for common and low accuracy rates for neuter nouns and overgeneralization of de to neuter contexts, but not the other way round) and between neuter, masculine and feminine in Greek (with neuter functioning as a default and yielding high accuracy rates and masculine being the most problematic of the three genders). Finally, the last research question concerns the children’s relative performance on the production and judgment tasks for the various gender categories (i.e. whether they perform better on judgment than on production.) The second aim of the paper is to discuss the results of the typically developing (TD) GreekDutch bilingual children of the present study in terms of their relevance to research on bilingual SLI. The specific language combination of Dutch and Greek might be particularly interesting because of certain differences between the two gender systems and their acquisition. Another advantage of studying DutchGreek bilinguals is that both monolingual and bilingual acquisition by children with and without language 9 MA Thesis 2013 Evelyn Egger impairment is relatively well-documented for both languages. Based on the present findings, we will consider the possibility of a positive influence of bilingualism on the acquisition of Dutch gender by children with SLI provided that the gender system and marking of the ‘other language’ (e.g. Greek) is extensive and transparent. The paper is structured as follows: Section 2.1 discusses the notion of grammatical gender and gender agreement in some more detail. In addition, a short description of the Greek and Dutch gender system will be provided (2.2.1 & 2.3.1) and previous findings on gender acquisition in the two languages that are relevant for the present study will be summarized (2.2.2 & 2.3.2). Section 3.1 will present a short review of research on gender agreement in a variety of bilingual populations and introduces the specific research questions of the present study. The experimental method will be outlined in section 3.2, followed by the presentation of the data in 3.3 and the discussion of the results in section 4.1. The findings will then be interpreted with regard to their relevance to investigations of bilingual SLI (4.2). Finally, the paper will conclude with a summary (section 5). 2 Gender 2.1 Gender Assignment vs. Gender Agreement Many languages use a grammatical gender system to classify their nouns into a small set of categories (also called noun classes or genders). The grammatical category of gender has to be distinguished from biological (or natural) gender, which is a semantic property that is restricted to animate nouns (e.g. girl vs. boy, niece vs. nephew, chicken/hen vs. rooster). Similarly, grammatical gender should not be confused with declensional class, which defines the forms a particular noun has, although there is often a close correspondence between the two (Comrie, 1999). Following Corbett (1991), gender assignment is determined by two major types of principles or rules, semantic rules that are based on a noun’s meaning and formal rules that are based on a noun’s form. Although all gender systems have a semantic core, a purely semantic division may not be applicable to all nouns and thus, in addition to semantic rules, most languages also have formal rules that determine a noun’s gender according to its phonological or morphological properties 3. For instance, many Romance languages use a two-way gender 3 Corbett (1991) distinguishes between phonological rules that refer to a single form of the noun, while morphological rules are usually based on a whole inflectional paradigm, although the distinction between the two is not always clear-cut. An example of a 10 MA Thesis 2013 Evelyn Egger system with the categories masculine and feminine, where all inanimate nouns are assigned a gender on the basis of formal rules, e.g. in Spanish nouns ending in –o are typically masculine and most nouns ending in –a are feminine. Gender assignment of nouns that is guided by formal rules often leaves the impression that the classification is somewhat arbitrary in that there is no semantic motivation for the choice of a particular gender, but rather it is determined by the grammar alone. Moreover, in languages that use both semantic and formal criteria for gender assignment, there might be a conflict between the semantic and the formal rules. To illustrate, on semantic criteria the German word for ‘girl’ (Mädchen) should be assigned the feminine gender (because it coincides with a girl’s natural gender), but the formal rule 4 prevails in this case and classifies it as a neuter noun resulting in a mismatch between the grammatical and natural gender of the noun. As already mentioned in the introduction, languages differ with regard to the marking of gender (i.e. the scope of agreement). Since gender is not always overtly marked on nouns, agreement (also termed gender concord) is often taken as a defining characteristic of grammatical gender systems or in the words of Hockett (1958: 231) “genders are classes of nouns reflected in the behavior of associated words”. Adopting a generative framework for descriptive purposes, agreement or gender concord in the nominal domain can be described as follows: All elements involved in gender agreement carry a gender feature [Gender: X]. The gender specification of nouns (i.e. gender assignment) is typically determined by a set of language-specific rules (i.e. the semantic and formal assignment rules discussed above) and thus, it is assumed that the gender feature of nouns carries a fixed value. For example, the German word Blume (‘flower’) is lexically specified as a feminine noun (according to the gender assignment rules of German) and consequently, it carries a valued gender feature in the form of [Gender: Feminine]. In contrast, the gender feature of all other elements that lie within the scope of gender agreement (agreement targets) is lexically unvalued, i.e. [Gender: ∅]. The syntactic operation of agreement is triggered by the head noun and involves the matching or sharing of features with the agreement targets (determiners, adjectives, etc.), as illustrated in (12) below. In other words, during the syntactic computation of agreement, the unvalued gender features of targets receive their value by the head noun carrying the valued counterpart feature (Pesetsky & Torrego, 2007). system that is mainly based on phonological rules is French, while German is described as a morphological system, which usually make use of all three kinds of rules, morphological, phonological and semantic (Corbett, 1991). 4 In this case, the rule for the formation of diminutives, which always assigns neuter gender. 11 MA Thesis 2013 Evelyn Egger Gender assignment refers to the inherent gender feature of nouns, which is commonly assumed to be stored in the lexicon in form of a fixed value, while gender agreement involves the syntactic operation whereby the lexically unvalued gender feature of agreement targets receive their value from the head noun (Kibort & Corbett, 2008). Thus, the crucial difference between gender assignment and gender agreement is that the former is a property of the lexicon, while the latter is a purely syntactic operation. Acquisition of grammatical gender involves thus both learning the inherent gender feature of nouns at the lexical level and the ability to establish gender agreement or concord between a noun and its agreement targets at the syntactic level. This means that in practice it is very difficult to distinguish between the two. Because gender agreement requires access to the inherent gender feature assigned to a noun, agreement errors in production may result either from incorrect gender assignment to nouns (i.e. an error of the lexicon, e.g. fr: *la livre verte ‘the-FEM green-FEM book’) or alternatively, they may be due to faulty application of agreement rules (i.e. an error of syntax, e.g. fr.: *le livre verte ‘the-MASC green-FEM book’). Since not all gender languages mark their nouns overtly and unambiguously for gender, the determiner is often taken as evidence for lexical assignment of gender, despite the fact that gender errors with determiners could also be interpreted in terms of agreement (Montrul, Foote & Perpiñán, 2008). The reason for taking the determiner as indicative for gender assignment is based on the finding that monolingual children appear to use the gender of the determiner as a cue to gender assignment of nouns during acquisition (Caroll, 1989; Lew-Williams & Fernald, 2008). In the following sections, we will now turn to descriptions of the Dutch and Greek gender systems and summarize previous findings of studies on the acquisition of gender in the two languages. 12 MA Thesis 2013 Evelyn Egger 2.2 Gender in Dutch 2.2.1 Descriptive Notes Dutch has a two-way gender system that classifies its nouns into neuter and common gender 5. There is no overt gender marking on nouns, the diminutive suffix –(t)je being an exception since it always assigns neuter gender in the singular (and thus, takes the neuter determiner het) regardless of the grammatical gender of the root noun. With regard to the gender assignment rules of nouns, there seem to be some morphological and semantic regularities, however, these are too limited and come with too many exceptions to serve as reliable cues to a noun’s gender (Donaldson, 1987). For this reason, gender assignment of nouns in Dutch is said to be largely arbitrary (Deutsch & Wijnen, 1985). In contrast to nouns, gender is marked on a number of agreeing elements both inside and outside the noun phrase such as definite determiners, demonstratives, attributive adjectives, relative pronouns, first person plural possessives and wh-phrases. Since the focus of the present paper is on determiner-noun and adjectivenoun agreement, only the Dutch article system and the inflectional paradigm for attributive adjectives will be illustrated here. The definite determiner for common nouns is de, while neuter nouns take the definite determiner het. However, there are no gender distinctions on indefinite determiners, which is always een and in the plural, where the definite determiner is always de. An overview of the Dutch article system is provided in Table 1. Table 1. Overview of the Dutch article system Definite Indefinite Common Neuter Common Neuter Singular de het een een Plural de de - - The inflectional paradigm for Dutch attributive adjectives comprises only two forms, a schwa ending and a bare form. The bare form is required with singular, indefinite neuter nouns, while in all other contexts gender-marking is neutralized and the adjective is inflected with a schwa for either gender. The 5 Like most Germanic languages, Dutch traditionally had a three-way gender system (masculine, feminine, neuter). However in most varieties and dialects spoken in the Netherlands, masculine and feminine nouns have collapsed into a common gender that retains the masculine inflections and pronouns. 13 MA Thesis 2013 Evelyn Egger inflectional paradigm for attributive adjectives is presented in Table 2 below and examples for both genders in the definite and indefinite singular are given in (13) - (16) 6. Table 2. Inflectional paradigm for attributive adjectives in Dutch Definite Indefinite Common Neuter Common Neuter Singular -e -e -e Plural -e -e -e -∅ (13) Een klein-e hond A small -e dog-COMMON ‘A small dog’ (14) De klein-e hond The small dog-COMMON ‘The small dog’ (15) Een A klein-∅ huis small house-NEUTER ‘A small house’ (16) Het klein-e huis The small house-NEUTER ‘The small house’ As can be seen from this short description, the Dutch gender system is highly opaque with gender-marking being limited to a handful of agreeing elements and a fair amount of syncretic forms that neutralize the gender distinction in many contexts. The lack of reliable and consistent cues to Dutch gender has consequences for its acquisition, to which we will turn in the next section. 6 Since gender-marking on adjectives is neutralized in the plural, it will not be considered here. 14 MA Thesis 2013 Evelyn Egger 2.2.2 Acquisition of Grammatical Gender in Dutch Monolingual Acquisition The acquisition of grammatical gender in Dutch has been shown to be particularly difficult, even for typically developing monolinguals, as indicated by the delayed mastery of the gender system (at around age 6 to 7) compared to other morpho-syntactic phenomena (e.g. Blom, Polišenská & Weerman, 2008). With regard to the acquisition of determiners, four developmental stages have been identified, which are presented in Table 3. Table 3. Developmental stages in the acquisition of Dutch determiners (taken from Cornips & Hulk, 2006 with minor changes) Stage 1: only bare nouns Stage 2: schwa-element + noun; the schwa-element can be interpreted as the indefinite article een (before the age of 2) Stage 3: a) definite article de with both common and neuter nouns b) first appearance of het, but massive overgeneralization of the definite determiner de Stage 4: target grammar (not before the age of 6) As can be seen from the table, children go through a stage where they overgeneralize the definite determiner de to neuter nouns (e.g. *de huis instead of het huis ‘the house’), however this tendency is unidirectional in that there is no overgeneralization of het to common nouns (e.g. *het boom instead of de boom ‘the tree’). The attested overgeneralization pattern is often explained in terms of input-frequencies or distributions. More specifically, it is argued that common nouns outnumber neuter nouns by a ratio of approximately 3:1 in token frequency and about 2:1 in type frequency (Van Berkum, 1996 as cited in Blom, Polisenska & Weerman, 2008). Moreover, the plural definite determiner for both genders is de, which further adds to the distributional asymmetry between the two determiner forms in the input. Given the lack of reliable semantic and morphological cues to gender assignment of nouns (especially neuter nouns), it is assumed that de functions as a default form (Tsimpli & Hulk, to appear) and that the gender of neuter nouns is acquired on an item-by-item basis (Unsworth, 2008), which is supported by the finding that production of de with common nouns is target-like from early on, while correct use of het increases only gradually (Blom et al, 2008). Moreover, it has been argued that certain properties of the Dutch gender system such as the lack of overt gender-marking on head nouns, the absence of gender distinctions 15 MA Thesis 2013 Evelyn Egger in indefinite and plural contexts and the fact that het also functions as a pronoun and nominalizer in certain constructions (Roodenburg & Hulk, 2008) considerably complicate the discovery of reliable and salient cues to a noun’s gender which in turn, contributes to the late acquisition of gender in Dutch (Unsworth, Argyri, Cornips, Hulk, Sorace, & Tsimpli, 2011a). As a consequence, it has been suggested that young children acquiring Dutch initially are not aware of the presence of gender as an abstract grammatical feature, which would explain the use of de as a default for any noun during early development (Stage 3a) (Cornips & Hulk, 2008). At some point, however, children do become aware of the presence of grammatical gender distinctions (Stage 3b). In other words, they realize that de and het are related to gender, but they still have to learn the correct gender specification for the various nouns 7. It is important to note here, that the unidirectional overgeneralization pattern and the late mastery of the gender system in Dutch are somewhat unusual when looked at from a cross-linguistic perspective (Cornips & Hulk, 2008). This is because in other languages that employ grammatical gender distinctions, children do not show such unidirectional overgeneralization patterns and the phenomenon does not seem to pose a significant problem to L1 children and is thus, generally acquired early (around age 3 to 4, depending on the semantic and morphological transparency of the system) (see Eichler, Jansen & Mueller, 2012 for a short review of gender acquisition in Spanish, Italian, French and German; Tsimpli, 2003 for Greek). Noting that most of the studies on the acquisition of Dutch gender have used production data only, Unsworth & Hulk (2010a) compared accuracy on gender assignment (i.e. determiner-noun agreement) in a production and a judgment task. Participants were monolingual children aged 4;3 to 7;6 and were divided into two groups, one with average vocabulary scores (group I) and one with above average vocabulary scores (group II). The results for the production showed that children with higher vocabulary scores (group II) performed significantly better than children with average vocabulary scores (group I) on neuter nouns, but no such difference was found for common nouns since both groups performed near ceiling (>95% correct). In contrast, there were no significant differences between the performances of the two groups on the judgment task. A comparison between the production and judgment data revealed that while group I performed equally well in both tasks, group II had lower overall scores in judgment than production. For neuter nouns, group I performed significantly better on the judgment task than the production task, but this difference was not significant for group II. For common nouns, the reverse was found, namely a significant difference between performance on the production and judgment tasks for group II, but not group I. Interestingly, the observed significant difference between judgment and production of common nouns observed for group II goes in the opposite 7 In fact, given that de functions as a default, it has been argued that gender specification has to be learned only for neuter nouns (Blom et al., 2008). 16 MA Thesis 2013 Evelyn Egger direction, that is judgment is worse than production. To account for these findings, it was argued that children in group II are at a more advanced level in their development of gender than group I and that unlike group I, they are passed the stage where they use de as a default and thus, are aware of the paradigmatic relationship between de and het which leads them to occasionally assign neuter gender to common nouns. As far as adjectival inflection is concerned, studies have shown that L1 children overgeneralize the inflected form to contexts that require the bare adjective (e.g. *een kleine huis instead of een klein huis ‘a small house’), while errors in the other direction are infrequent (Unsworth, 2013a). As already mentioned, agreement errors of this type have two potential sources, they may either be the result of wrong gender assignment to the head noun (i.e. the child thinks that huis is common and thus, chooses ‘correctly’ the inflected form) or alternatively, they may arise due to an error in the syntactic computation of agreement (i.e. the child knows that huis is neuter, but does not know the adjectival agreement rules). Studies that investigate this issue indicate that most of these agreement errors result from wrong gender assignment of the head noun rather than an inability to establish gender concord, which means that once gender assignment is taken into account adjectival agreement is found to be target-like from early on (Polisenska, 2010). Furthermore, it is argued that the close relation between accuracy scores on definite determiners and adjectival inflection constitutes evidence for the activation of an abstract gender feature in the child’s grammar (Blom et al., 2008; Polisenska, 2010). Bilingual acquisition Given that the acquisition of grammatical gender has been found to pose significant difficulties for Dutch monolinguals, it is reasonable to assume that bilingual children encounter even larger problems in this area. This prediction is confirmed by a series of studies on various bilingual populations, who show an even more protracted course of gender acquisition compared to monolinguals (Blom et al., 2008; Unsworth, 2007 & 2008) , with some even showing signs of fossilization (e.g. Hulk & Cornips, 2006a & 2006b). In other words, some of these studies indicate that the developmental trajectory of bilingual children sometimes differs both quantitatively and qualitatively from what has been reported for monolingual children. Both 2L1 (i.e. simultaneous bilinguals) and L2 children have been found to show the same overgeneralization pattern as monolinguals, namely the use of de in contexts that require the neuter definite determiner, but unlike monolinguals they produce such errors well beyond the age of 6. For example, Hulk & Cornips (2006b) investigated the acquisition of Dutch gender as indicated by the correct use of definite determiners by a group of bilingual children (with age of onset <4 years) that were born in ethnic communities in the Netherlands and had a variety of language backgrounds. The subjects were 17 MA Thesis 2013 Evelyn Egger divided into three age groups (young: 3;0-3;10, middle: 4;11-5;2 and old: 9;3-10;5) and performance was compared to a group of Dutch monolingual children matched for chronological age. Results revealed that although both monolinguals and bilinguals showed a clear development in the correct use of de across age groups, the bilinguals were delayed compared to the monolinguals (accuracy rates per age group for monolinguals and bilinguals respectively: young group 59% vs. 26%; middle group 90% vs. 46% and old group 92% vs. 76%). However, the same comparisons for the correct use of het showed a clear development for monolinguals (from 7% in the young group to 77% in the middle and 90% in the old group) but hardly any development for the bilinguals (from 7% in the young group to 15% in the middle and 32% in the old group; with no significant differences between subsequent age groups but only between the young and old one). Moreover, the lower performance of the bilinguals could not be ascribed to cross-linguistic influence since children whose L1 employed grammatical gender did not significantly differ from those whose L1 did not show gender distinctions of nouns (see Cornips et al., 2006 for similar findings). Finally, it was argued that in addition to the quantitative difference between monolinguals and bilinguals in the form of a delay in the acquisition of the correct use of het, the results seemed to indicate also a qualitative difference between the grammars of the two populations in that the bilinguals appear to fossilize in a stage where they continue to overgeneralize de. Similar results were reported by Blom et al. (2008) for Moroccan child L2 learners of Dutch whose age of onset was between 4 and 7. The child L2 learners were divided into two groups based on their level of proficiency in Dutch. Both child L2 groups showed the same overgeneralization pattern (de instead of het) as the monolinguals, but there were no statistically significant developments between the two for either common or neuter nouns. However, a comparison with production data obtained in other studies that included older bilingual groups suggested that bilingual children do show a gradual development in the correct use of het, i.e. that they are delayed but do not fossilize (see Cornips, van der Hoeck & Verwer, 2006 for similar conclusions with regard to determiner-noun agreement). Unsworth (2008) tested English-speaking L2 learners of Dutch (children, preteens and adults) and found the same overgeneralization pattern of de to neuter nouns. Moreover, although there were many L2 learners who did not show target-like use of the definite determiner het even after lengthy exposure to Dutch, she did find a handful of learners who used het with neuter nouns in a native-like fashion. Thus, it was suggested that acquisition of grammatical gender in Dutch is, in principal, possible for English-speaking learners of Dutch (both children and adults), but that lengthy and extensive exposure may be a prerequisite to reach native-levels in the correct use of het (Unsworth, 2008). The observed differences between monolinguals and bilinguals are often accounted for in terms of input-related factors, namely that the bilinguals receive quantitatively and sometimes even qualitatively less input than the monolinguals (Cornips & Hulk, 2008; Unsworth, 2008). This is due to the assumption 18 MA Thesis 2013 Evelyn Egger that input differences affect vocabulary acquisition and that the acquisition of gender-marking on determiners involves the interface between the lexicon and morpho-syntax (Hulk & Cornips, 2006a). The difference in amount of input the monolingual and bilingual children are exposed to is particularly crucial in the case of Dutch, since there is very little evidence for neuter gender even in the input of monolingual children which is reflected in the lengthy developmental path. Thus, it is possible that some bilinguals simply do not receive a sufficient amount of input to fully acquire the neuter gender in Dutch. Moreover, it is likely that the input of bilingual children raised in ethnic minorities or communities in the Netherlands is also qualitatively different, i.e. the Dutch variety spoken by the members of the ethnic community may differ from standard Dutch (Hulk & Cornips, 2006a). In contrast to the above mentioned studies, Cornips & Hulk (2006) found evidence for acceleration in the acquisition of Dutch gender by bilingual children from dialectal communities compared to monolingual children. It was argued that the high degree of morpho-syntactic overlap between the dialectal and standard Dutch determiner system allows for cross-linguistic influence in the form of facilitation or positive transfer. This structural overlap was not present in the languages of the bilinguals from the ethnic communities in Hulk & Cornips’ (2006a) study which might explain the different findings for these two types of bilinguals (i.e. ethnic communities and dialectal communities) (Hulk, 2007). As far as the acquisition of gender marking on adjectives is concerned, bilingual children have been found to show the same error profile as monolinguals, namely the overgeneralization of the schwaending where the bare adjective is required (e.g. Blom et al., 2008; Cornips et al., 2006). Cornips et al. (2006) compared monolinguals’ and bilinguals’ performance on determiner-noun (Det-N) and adjectivenoun (Adj-N) agreement and found that both groups were more accurate on Det-N agreement. It was argued that the difference in the accuracy rates for the two types of agreement can be accounted for in terms of differences in the structural configurations between the two types of agreement 8. Collating the data obtained by Blom et al. (2008), Hulk & Cornips (2006a) and Cornips et al. (2006) suggested that there is a difference between monolingual and bilingual children with regard to the developmental pattern across gender domains. While monolingual children showed a parallel development of Det-N and Adj-N agreement, bilinguals showed a dissimilar development in the two gender domains, i.e. there was development in Det-N agreement but not in Adj-N agreement (Blom et al., 2008). The results were interpreted as evidence for the activation of an abstract gender feature in the monolinguals’ underlying grammar, whereas this might not be the case for bilinguals. 8 “The determiner, being the head, needs to agree syntactically more than the adjective and the relative pronoun, bringing about two morphological forms according to the gender of the noun.” (Cornips et al., 2006: 11). 19 MA Thesis 2013 Evelyn Egger Unsworth (2013a) further investigated the acquisition of adjectival inflection in Dutch and its relationship with Det-N agreement in a group of English-Dutch bilinguals aged between three and 17 years. In contrast to Blom et al. (2008), results revealed a close relationship between accuracy rates for Det-N and Adj-N agreement for both monolinguals and bilinguals indicating that bilinguals too make use of an abstract gender feature. Moreover, it was found that current amount of exposure and cumulative length of exposure were both significant predictors for gender-marking on determiners, but only indirectly for gender-marking on adjectives where the relationship is mediated by the accuracy rate on determiners. In other words, amount of exposure (both current and cumulative) affect gender assignment (Det-N agreement) more than gender agreement (adjectival inflection). This is to be expected if we assume that gender assignment (Det-N) agreement has to be learned on an item-by-item basis, whereas gendermarking on adjectives involves the application of rules that make use of abstract grammatical features (Unsworth, 2013a). Finally, Unsworth (2013b) investigated the source of adjectival agreement errors in the same group of bilinguals and found that once gender assignment was taken into account accuracy scores on Adj-N agreement increased, which means that a considerable amount of agreement errors on adjectives result from wrong gender assignment to the head noun, rather than an inability to acquire the adjectival agreement rule. Thus, it seems that bilinguals’ persistent difficulties are caused by the lexical aspect of the Dutch gender system, rather than the morpho-syntactic (Unsworth, 2013b). Unsworth (2013a) also investigated the relationship between judgment and production of determiner-noun agreement and found that the performance patterns of the English-Dutch bilinguals were very similar to the patterns previously attested Dutch monolinguals in Unsworth & Hulk (2010a). More specifically, with common nouns children were more likely perform target-like on production, but not on judgment, while for neuter nouns the reverse was found, namely that children were more likely to be on target on judgment rather than on production. According to Unsworth & Hulk (2010a), the better performance on production than on judgment for common nouns indicates a developmental stage where children have become aware of the paradigmatic relationship between de and het which leads to occasional errors in the judgment of common nouns, while in production, they are more likely to continue to use de as a default. 20 MA Thesis 2013 Evelyn Egger 2.3 Gender in Greek 2.3.1 Descriptive notes Modern Greek employs a tripartite gender system with the values ‘masculine’, ‘feminine’ and ‘neuter’. With regard to the gender assignment rules, semantics plays an important role in determining the gender of animate nouns in Greek and consequently, the natural gender of nouns denoting humans usually coincides with their grammatical gender such as andras masc (man) and gyneka fem (woman) or ɣios masc (son) and kori fem (daughter). However, there are some exceptions to this rule, e.g. koritsi neut (girl) and as noted by Ralli (2002), it is not the case that all remaining nouns are simply neuter which means that in general, the gender values of inanimate nouns are semantically unpredictable. In contrast to Dutch, Greek is a highly inflecting language and nouns typically carry an inflectional suffix 9. However, these suffixes are ‘portmanteau’ morphemes in that they combine information about number, gender and case. In addition to gender distinctions, Greek nouns can also be classified according to their inflectional paradigms. Crucially though, there is no one-to-one correspondence between the three gender values and the inflectional paradigms they display, leading to a considerable overlap between the two classification systems. To illustrate, the inflectional paradigm of nouns ending in –os contains both masculine (e.g. tichos “wall”) and feminine nouns (e.g. isoδos “entrance”), but at the same time each of the three gender classes shows considerable inflectional variation (e.g. masculine and neuter each have over ten different inflectional paradigms 10), which means that noun endings in Greek do not provide unambiguous cues to gender assignment. Nevertheless, it has been argued that noun suffixes in Greek have a predictive value for gender, with some being more predictive (and consequently less ambiguous) than others. This claim is supported by the empirical finding that (most) noun endings constitute strong cues for gender assignment by native speakers (Mastropavlou & Tsimpli, 2011). This means that as to formal rules to gender assignment, there are a number of morphological and phonological regularities as indicated by the relatively strong correlation between gender and inflectional class (Ralli, 2002), but at the same time there are many exceptions to those rules, rendering the various endings more or less transparent, i.e. predictive of gender. Table 1 gives an overview of the most common noun endings for each gender class in the nominative singular. 9 With the exception of some borrowings and loan words. However, only a handful of these inflectional paradigms are productive, the others being obsolete or derived from Ancient Greek. 10 21 MA Thesis 2013 Evelyn Egger Table 4. Overview of the most common Greek noun suffixes in the nominative singular. Masculine Feminine Neuter most common -ós/ -os -á/ -a -o -i common -ís/ -is -í/ -i -ma rather common -ás/ -as -es -ús/ -us rare -os -os -u -on -as Grammatical gender in Greek is determined by a combination of semantic and morphological rules, but nevertheless, it has been argued that Greek nouns are not always intrinsically specified for a particular gender. According to Ralli (2002), for a large number of Greek noun stems, gender constitutes an intrinsic property in the sense that the gender value of those stems is neither related to morphological nor semantic information, such that for those noun stems, the gender feature is fully specified, as in, for example, kip masc -os ‘garden’. On the other hand, stems of nouns whose gender value is determined by semantic features and/or morphological processes are underspecified for their gender value. For example, noun stems that can take different endings such as aðerf-, may surface either as a masculine noun, e.g. aðerf-os masc (brother) or as a feminine noun, e.g. aðerf-i fem (sister) and similarly, gat- which can take all three genders depending on the semantic context (i.e. gat-a fem , gat-os masc and gat-i neut (female/male) ‘cat’). In addition, there is a group of nouns denoting human professions which are persistently underspecified in that gender assignment is not resolved by means of semantic or morphological information of the noun itself, e.g. jatr- from jatr-os masc/fem (male/female) ‘doctor’. For those nouns it is argued that gender is determined at the phrasal level through agreement rather than at the lexical level through semantic or formal assignment rules (see Ralli, 2002 for a more extensive discussion). The crucial point here is that in contrast to Dutch, where gender assignment can be said to be a property of the lexicon only, gender assignment of certain nouns in Greek involves also morphological rules and in some cases even syntactic processes. Grammatical gender in Greek is realized on all declinable nominal elements, including the noun itself, both predicative and attributive adjectives, determiners, numerals, quantifiers and pronouns. Again we will restrict our discussion to the Greek determiner system and adjectival inflectional paradigm. Unlike Dutch, Greek shows gender distinctions not only on definite determiners in the singular, but also in the plural and on indefinite determiners in the singular. Definite determiners in the singular take the 22 MA Thesis 2013 Evelyn Egger forms o, i and to for masculine, feminine and neuter nouns respectively. Table 5 gives an overview of the Greek determiner system. Table 5. Overview of the Greek determiner system in the Nominative. Definite Indefinite Masculine Feminine Neuter Masculine Feminine Neuter Singular o i to enas mia ena Plural i i ta - - - As already mentioned Greek adjectives have to agree with the head noun in gender, number and case. Similar to nouns, adjectives in Greek can be divided into different classes depending on their inflectional paradigm. Although there are a number of different adjective classes, most adjectives in Greek take the suffix –os for masculine, -i for feminine and –o for neuter nouns in the nominative singular as illustrated in Table 6 below. Moreover, there are a handful of adjectives (mainly colour terms that were borrowed from other languages), which are indeclinable, e.g. mov MASC/FEM/NEUT ‘purple’. Table 6. Main inflectional paradigm of regular adjectives in Greek in the nominative case singular and plural. Masculine Feminine Neuter Singular -os -i -o Plural -us -es -a Examples (17) - (19) illustrate agreement in the noun phrase in Greek. Note that in contrast to most Romance languages, attributive adjectives in Greek occur in pre-nominal position. (17) O kal-os anthrop-os The- MASC good- MASC man- MASC omorf-i gynaik-a ‘The good man’ (18) I The- FEM beautiful- FEM woman- FEM ‘The beautiful woman’ 23 MA Thesis 2013 (19) Evelyn Egger To megal-o vivli-o The- NEUT big- NEUT book- NEUT ‘The big book’ To summarize, in Greek grammatical gender is marked consistently on all nominal elements, including the noun itself. Moreover, there is a close relationship between gender and inflectional morphology which means that Greek suffixes provide reliable cues to gender assignment of nouns. In other words, the abundance of gender-marking and the high predictive values of inflectional suffixes in Greek lead to a relatively high degree of transparency of the grammatical gender system. 2.3.2 Acquisition of Grammatical Gender in Greek Monolingual Acquisition Studies on the monolingual acquisition of grammatical gender in Greek have shown that the system is mastered early, that is at around age 3;6 to 4;0 (Tsimpli, 2003). In fact, gender has been found to be one of the first grammatical features to be acquired in Greek (Stephany, 1995). Mastropavlou (2006) tested children aged 3;0 to 3;7 on gender marking and agreement in the nominal domain and found that neuter nouns were the least problematic suggesting that neuter is the unmarked form and consequently, bears a default function (see Stephany, 1995 for similar results). Moreover, incorrect gender-marking on nouns and errors in determiner-noun agreement occurred only in masculine contexts (2% - 10%), but again performance was very high (>90% correct). Children were also tested on determiner-adjective-noun agreement, which yielded slightly more errors than the Det-N task, but the difference was not statistically significant. Comparing the three genders, again masculine nouns turned out to be the most problematic, while errors with feminine nouns were very rare and with neuter non-existent. Finally children were also tested on pseudo-nouns and results indicated that morpho-phonological cues on the noun’s suffix are used by 5-year-olds, but not by 3-year-olds, i.e. they are not used in the early stages of gender production in gender agreement contexts. Bilingual Acquisition Unsworth et al. (2011b) investigated the acquisition of grammatical gender in Greek-English (and DutchEnglish) bilingual children on two production tasks 11. The children varied in their age of onset to Greek (i.e. simultaneous bilinguals from birth vs. early successive bilinguals before the age of 4;0 vs. child L2 11 The two production tasks were the Story and the PDT tasks that were also used in the present study. 24 MA Thesis 2013 Evelyn Egger learners after the age of 4;0) and the amount of input they were exposed to. Results showed that children performed best on neuter nouns, followed by feminine and masculine and that they tend to overgeneralize the neuter definite determiner to with masculine and feminine nouns. Moreover, for masculine and feminine nouns, children did better with determiner-noun agreement in simple DPs (det-N) than in complex DPs (det-adj-N). A regression analysis suggested that for masculine and feminine nouns, amount of exposure and vocabulary score were significant predictor variables, while for neuter nouns, vocabulary score was the only significant predictor variable. Group comparisons further revealed that there were no significant differences between the simultaneous bilinguals and the Greek monolinguals who were both at ceiling with all three genders. In contrast, the performance of the early successive and the L2 children differed significantly from the simultaneous bilinguals (and the monolinguals) for masculine and feminine nouns. For neuter nouns, the only significant between-group differences were between the L2 children and all the other groups, but no differences were observed between the simultaneous bilinguals, the early successive bilinguals and the monolinguals. Having discussed the gender systems of the two languages under study here as well as the related findings from research on the monolingual and bilingual acquisition thereof, the next section (3.1) will introduce the specific research questions to be answered in the present study. 3 The Study 3.1 Background In light of the above mentioned findings on the acquisition of grammatical gender in Dutch and in Greek, a number of questions emerge. First, studies on the bilingual acquisition of Dutch gender have yielded controversial findings, with some populations showing signs of fossilization with regard to Det-N agreement (Hulk & Cornips, 2006b) or alternatively, Adj-N agreement (Blom et al., 2008), other studies reporting simply a ‘bilingual delay’ in the two domains (e.g. Unsworth, 2008 & 2013b), while yet others have found evidence for acceleration effects in bi-dialectal bilinguals (Cornips & Hulk, 2006). To account for these differences, researchers have proposed a number of factors that determine the rate of success or failure in the bilingual acquisition of Dutch gender (Cornips & Hulk, 2008). One of these factors relates to the nature of the L1, and more specifically, the amount of overlap between the two gender systems of a bilingual. Thus, for the bi-dialectal bilinguals in Cornips & Hulk’s (2006) study, it is argued that the two 25 MA Thesis 2013 Evelyn Egger gender systems present a sufficient amount of overlap to allow for positive transfer from the bilinguals’ L1 (in this case the Dutch dialect), which would lead to an acceleration in the acquisition of neuter gender in Dutch. In contrast, English-Dutch bilinguals have no gender instantiated in their other L1 (i.e. English) and thus, no (positive) influence from English on the acquisition of Dutch gender is expected. Moreover, studies that investigated bilinguals from a variety of L1 backgrounds have found no significant differences in the performance on Dutch neuter nouns between children who have gender instantiated in their other L1 from those whose L1 lacks grammatical gender distinctions (Cornips, van der Hoeck & Verwer, 2006; Hulk & Cornips, 2006b). Similarly, the French-Dutch children in Hulk (2007) seem to perform equally poorly on Dutch gender as the English-Dutch bilinguals in Unsworth (2008). Thus, it seems that the presence or absence of grammatical gender in the other L1 of a bilingual is not a decisive factor for the acquisition of Dutch gender. However, it is possible that what matters is not so much the presence/absence of grammatical gender, but rather the way in which grammatical gender is marked in those languages that do employ grammatical gender. For example, Eichler, Jansen & Kupisch (2012) investigated the acquisition of determiner-noun agreement in a group of bilingual children acquiring German together with a Romance language (Italian, Spanish or French) or two Romance languages (i.e. Italian-French bilinguals). Results indicated that accuracy on gender assignment (as indicated by correctly produced determiner-noun combinations) is highly influenced by the way gender is marked in the language that is being acquired. Thus, similar to what has been found for monolinguals, the bilinguals had more difficulties with acquiring the German gender system than the French system, while French caused only slightly more problems than Italian and Spanish, which were the easiest to acquire. It was argued that this ranking could be explained in terms of transparency of the various gender systems, with German being the most opaque, followed by French, while Italian and Spanish present the most transparent systems of the four languages (Eichler et al., 2012). Given that the way how gender is marked affects its acquisition, it is possible that this factor may also play a role in the bilingual acquisition of Dutch gender. The first research question to be addressed is thus: 1. What are the effects of having a transparent gender system with extensive gender marking in the L1 (other language) on the bilingual acquisition of Dutch gender? In order to address this question, the results of the Greek-Dutch bilinguals will be compared to those obtained by Unsworth (2013b) for English-Dutch bilinguals, as well as to data from Italian-Dutch bilinguals gathered by Algoe (2011), who both used the same experimental tasks. If the transparency of the Greek gender system does indeed have an effect on the acquisition of Dutch gender, the bilinguals of the present study should show a different performance pattern than the English-Dutch bilinguals since 26 MA Thesis 2013 Evelyn Egger English does not have grammatical gender instantiated and their performance should be comparable to the Dutch-Italian bilinguals given that similar to Greek, the gender system of Italian is argued to have a high degree of transparency (i.e. consistent and largely unambiguous gender marking on nominal elements, including the noun itself). However, the Greek and Italian gender systems also differ in several ways, which could lead to possible qualitative and quantitative differences in the acquisitional path between the two populations. For example, in Greek gender is not only intertwined with number as in Italian, but also case. This factor has been argued to account for the later mastery of the German gender system compared to the French system in Eichler et al. (2012). On the other hand, Italian has only two gender values, while Greek makes a three-way gender distinction which could lead to enhanced gender awareness in Greek children compared to Italian children. Other factors that may influence the acquisition of gender are the amount of variation of nominal endings (high for Greek vs. little for Italian), the nature of the determiner system (allomorphy of masculine determiners in Italian vs. no allomorphy in Greek) and the position of the attributive adjectives (mostly post-nominal in Italian vs. prenominal in Greek). The second question that remains open concerns the relationship between the two agreement types, namely Det-N agreement and (Det-)Adj-N agreement. As to adjectival agreement in Dutch, Polisenska (2010) found evidence that Dutch monolinguals’ persistent difficulties with gender arise from the lexical aspect of the acquisition task, namely from gender assignment, rather than agreement. In fact, when gender assignment was taken into account, monolinguals were found to be target-like on adjectival agreement from much earlier on than has previously been suggested. Findings for bilinguals, however, are more controversial. While Blom et al. (2008) investigated a group of Moroccan children aged 4;2-8;4 (mean age of onset 4;0) and found dissimilar developments for the two types of agreement when collating her data with other studies (with Det-Adj-N showing signs of fossilization), Unsworth (2013a) found evidence for parallel development of determiner and adjectival agreement in simultaneous English-Dutch bilinguals, which was taken as evidence for the activation of an abstract gender feature in the children’s grammar. Moreover, the data from the Dutch-Italian bilinguals in Algoe (2011) (same subjects as Canta, 2012) also indicate development with increasing proficiency for both agreement types and no observable difference in accuracy rates between Det-N and Det-Adj-N agreement. In addition, the results in Algoe (2011) suggested that proficiency in Dutch (as indicated by vocabulary score) better predicts accuracy with neuter gender in bilingual children than chronological age. The second research question thus concerns the relationship between the two types of agreement and their development with increasing proficiency in Dutch, more specifically: 2. Do the Dutch-Greek bilinguals show (parallel) development for both types of agreement with increasing proficiency in Dutch (as indicated by vocabulary scores)? 27 MA Thesis 2013 Evelyn Egger The relationship between determiner-noun and adjectival agreement is further investigated in a cross-linguistic comparison with findings from Dutch-Italian bilinguals. To elaborate, Canta (2012) investigated gender agreement in Italian in a group of Dutch-Italian bilinguals aged 3;3-7;11 and found lower accuracy scores for adjectival agreement than for determiner-Noun agreement (which was above 90% in both groups). Thus, Det-N agreement seems to be acquired earlier. Moreover, in terms of error types, for the younger group (<5;6) the most frequent error type was Det-N-*Adj for both genders (e.g. la FEM casa -FEM * nuovo -MASC , ‘the new house’) the same holds for the older children (>5;6) with regard to masculine nouns but not for feminine nouns, where the most frequent error type was *Det-N-*Adj (e.g. *il -MASC pecora -FEM *nero -MASC . This indicates that for both groups, adjectival agreement errors with masculine nouns are due to a failure to maintain gender concord in ‘syntactically more complex’ environments. For feminine nouns, the two groups showed a different error pattern with older children’s most frequent error being of the type *Det-N-*Adj (suggesting an underlying error in gender assignment), while for the younger ones it is Det-N-*Adj. Finally, accuracy on masculine nouns does not improve from the younger to the older group (probably a ceiling effect) while a clear improvement can be seen for feminine nouns (from 55% to 84%). The following two questions are thus stated with regard to the Greek data obtained from the Greek-Dutch bilinguals: 3. Do the Dutch-Greek bilinguals show different accuracy rates for Det-N agreement and DetAdj-N agreement in Greek? 4. What kind of agreement errors do the Dutch-Greek bilinguals commit in Greek? Assuming that language-internal factors such as the nature of the two gender systems of a bilingual child somehow has an impact on the acquisitional path of grammatical gender in the two languages (see Cornips & Hulk, 2008 for a detailed discussion of the proposed factors for Dutch), comparing performance patterns across bilingual populations can offer important insights into the specific characteristics of the various bilingual groups and further indicate the relationship between the various factors. For example, in Greek, the order of acquisition of the three genders is neuter, which is the unmarked form and thus, assumes a default function, followed by feminine and finally, masculine gender, which seems to be the most problematic. Mastropavlou (2006) has observed this pattern in Greek monolinguals and similarly, Unsworth et al.’s (2012b) results showed that early successive English-Greek bilinguals (age of onset between 1-4 years) and L2 children (age of onset between 4-10 years) performed significantly better with feminine than neuter nouns and better with neuter than both feminine and 28 MA Thesis 2013 Evelyn Egger masculine 12. As a consequence, the neut>fem>masc pattern seems to be a robust finding and as such it is expected that the Greek-Dutch bilinguals show the same pattern, if not, this would suggest a possible influence of the Dutch gender system on the acquisition of Greek gender. For Dutch, the overgeneralization of de to neuter nouns and the –e ending to contexts that require a bare adjective has frequently been attested in monolingual and early bilingual 13 populations. However, a number of studies have also reported overgeneralization of het for some bilinguals, which suggests that this particular tendency may indicate a qualitative difference between monolingual and bilingual acquisition or alternatively they may point to differences between different kinds of bilingual groups. To shed further light on these issues, the following questions are asked: 5. Do the Dutch-Greek bilinguals show the same overgeneralization and performance patterns as previously attested in monolingual and early bilingual populations, namely a. neut>fem>masc for Greek? b. and overgeneralization of de for Dutch? The last question to be addressed in the present study concerns the relationship between performance on production and knowledge tasks. The general assumption is that knowledge or comprehension precedes production and as a result acquirers and learners typically do better on knowledge or judgment than on production tasks. For Dutch, however, it has been found that while this general advantage of knowledge over production holds for neuter nouns, it seems to be reversed for common nouns, i.e. production is better than judgment. Moreover, this pattern has been observed in monolinguals (Unsworth & Hulk, 2010a) as well as (English-Dutch) bilinguals (Unsworth, 2013a) and has been related to a developmental stage in the acquisition of Dutch gender. The last question thus asks whether this pattern also extends to other bilingual populations, namely Dutch-Greek bilinguals, more specifically: 6. Do the Greek-Dutch bilinguals show an asymmetry between the performance on the judgment tasks and the performance on the production tasks in the two languages and is this asymmetry the same for the various genders (i.e. common and neuter in Dutch and masculine, feminine and neuter in Greek)? 12 In contrast, no significant differences between the three genders were reported for the simultaneous bilinguals and the monolinguals due to their ceiling-performance. 13 That is simultaneous bilinguals (2L1) and early successive bilinguals with age of onset before 4;0. 29 MA Thesis 2013 Evelyn Egger Having presented the various research questions, we will next turn to a discussion of the research method, including background information about the participants, descriptions of the experimental material and some notes on the coding and analysis of the responses. 3.2 Method 3.2.1 Participants Children were recruited from a Greek Saturday School in Amstelveen, where they attend two and a half hours of Greek language classes 14 every weekend. Information about the subjects’ language background, exposure patterns and other input-related variables were gathered by means of a detailed parental questionnaire (Unsworth, 2008). Participants consisted of 33 Dutch-Greek bilingual children aged 4;4 to 13;3 years. 19 children were born in the Netherlands and 14 were born in Greece. From the 14 children that were born in Greece, 13 lived there for an extended time period (mean: 3;10 years, range: 0;4-7;0) and one child lived in Germany for six and a half years before the family moved to the Netherlands. 21 were simultaneous bilinguals (i.e. exposed to both languages from birth) and the remaining 12 were early successive bilinguals with Greek as their L1 and an average age of onset of exposure to Dutch of 4;0 years (range: 2;0-6;6). 19 of the children receive both Dutch and Greek input at home, i.e. most of them are raised in a one-parent-one-language situation, although some parents report speaking both languages to the children. For 12 of the children (mainly the successive bilinguals) Greek is spoken by both parents at home and two of the children are exposed to Greek and English at home. For 18 of the children the language that is spoken between the parents is Greek, while for 11 it is Dutch and for the remaining 4 it is either English and Greek or English only. 14 of the children were reported to speak only Dutch with their siblings, 9 speak only Greek and 11 speak both Greek and Dutch with their siblings. All the children attended Dutch primary schools or daycare/preschool at the time of testing, except for one child who attended an international school where English is the language of instruction and Dutch is taught as a foreign language. A detailed overview of the participants and information regarding a number of input variables that were extracted from the questionnaires is provided in Appendix A. 14 The Greek School hosts three educational levels, kindergarten, primary school and high-school. 30 MA Thesis 2013 Evelyn Egger 3.2.2 Materials 3.2.2.1 Background Measures In addition to the experimental tasks described below, children were administered a vocabulary test in each language. For Dutch, the PPVT-III-NL (L.M. Dunn, Dunn & Schlichting, 2005) was used, which is a standardized test measuring receptive vocabulary. For Greek, children were tested on the vocabulary section of the Diagnostic Verbal IQ test (henceforth DVIQ), which is a measure of productive vocabulary knowledge developed by Stavrakaki & Tsimpli (2000). Unlike the PPVT-III-NL, the DVIQ is not fully standardized up to the age ranges that were included in the present study and therefore, children were grouped and compared based on their raw scores 15, while children’s Dutch vocabulary scores were adjusted for age. The results of the two vocabulary measures were taken as a general indicator of children’s overall proficiency level in the two languages. 3.2.2.2 Experimental Tasks Dutch Tests In order to assess children’s knowledge of grammatical gender-marking on definite determiners and attributive adjectives in Dutch, two elicited production tasks and one grammaticality judgment task were administered. The tasks that were used were taken from Unsworth et al. (2011b) and were part of a larger project on early child bilingualism. There are two versions of the tests, one for children below 6 years, which contains 12 target nouns (6 common and 6 neuter) and an version for children above 6 years that included 18 target nouns (9 common and 9 neuter, see Appendix B for a complete list of the 18 target nouns). Moreover, the tests were preceded by a number of practice items and were interspersed with fillers that tested a different phenomenon and thus, will not be mentioned further. In the first task, the Picture Description Task (henceforth PDT), children are presented with two pictures (e.g. a yellow robot and a blue robot). Then one of the pictures is highlighted and children are asked to name it using the prompt: “Look! Here we see two pictures. This is a…“ (“Kijk! Hier zie je twee plaatjes. Dit is een…”), on which the child is required to response with an adjective-noun string, which in the present example would be “yellow robot” (“gele robot”). Following the child’s response to the first item, the second item gets highlighted and the experimenter continues by saying “And this is a…” (“En dit is een…”) and the child is expected to answer with “blue robot” (“blauwe robot”). After that, a third item (e.g. a ball) appears next to one of the two objects (e.g. the yellow robot) and the experimenter says: 15 The vocabulary section of the DVIQ consists of 27 items, so the maximum score is 27. 31 MA Thesis 2013 Evelyn Egger “The ball is in front of…” (“De bal ligt voor…”), on which the child answers with “the yellow robot” (“de gele robot”). Finally, a fourth item appears next to the other object (here the blue robot) followed by the prompt: “And the finger points to…” (“En de finger wijst naar…”) and the child is required to answer with “the blue robot” (“de blauwe robot”) (see also Figure 1 below). Thus, each noun is elicited four times, twice in an adjective-noun string in the indefinite condition and twice in a determiner-adjectivenoun string in the definite condition. An illustration of a neuter test item is provided in Figure 1. Exp: Dit is een… Child: wit konijn This is a… Exp: De bal ligt voor… The ball lies before… white rabbit Child: het grijze konijn the grey rabbit Exp: En dit is een… And this is a… Child: grijs konijn grey rabbit Exp: En de hond staat naast…Child: het witte konijn And the dog stands next to… the white rabbit Figure 1: Example item from the PDT task. The target noun konijn takes het and thus, requires the bare adjective in the indefinite condition (upper row), but the adjective has to be inflected with a schwa in the definite condition (second row). The second elicited production task used in the study was a story task (STORY), where children were asked to help the experimenter tell a story that is accompanied by pictures. The experimenter starts to tell a story, e.g. about a boy who is having a birthday party and the presents he received, a bicycle a robot and a guitar and the children are asked to name the presents as they appear on the screen. The story then continues with another character appearing on the screen followed by a question such as “What does Leo want to play with?” and the child is expected to answer “the guitar”, thus producing a determiner- 32 MA Thesis 2013 Evelyn Egger noun string in a definite context for each of the 12 or 18 target nouns. Figure 2 below illustrates two screens of one of the three stories that were used to elicit the determiner-noun strings in definite DPs. Exp: Wat heeft Leo gekregen? Een… What did Leo receive? A… C: robot [etc] Exp: Waar wil Paul mee spelen? robot [etc] What does Paul want to play with? C: de robot the robot Figure 2. Example of two screens from the STORY task, which was used to elicit each of the target nouns (here robot) together with a definite determiner. The nouns are first elicited in an indefinite context (left picture) and then in a definite context (right picture). In addition to the two elicited production tasks, children were also administered a grammaticality judgment task (GJT) that took the form of a forced choice task, where children have to decide between congruent and incongruent determiner-noun combinations. The items are first introduced by the experimenter using an indefinite determiner, such as “Here we see a window, a sheep and a key.” (“Hier zien we een raam, een schaap en een sleutel”) 16. Following this, each item is presented individually and two puppets appear, with one puppet naming the item with a congruent determiner-noun combination (e.g. het- NEUT raam- NEUT ‘the window’) and the other one with an incongruent one (e.g. *de- COM raamNEUT ‘the window’). Children are then asked which puppet said it correctly (“Jij moet zeggen wie het goed heft”). Correct responses were counterbalanced across the two puppets. Thus, each of the 18 (or 12) target nouns was elicited multiple times and for each noun the following responses were included for subsequent analysis: • 1x in a determiner-noun string (STORY) • 2x with an indefinite determiner in an adjective-noun string (PDT) • 2x in a definite determiner-adjective-noun string (PDT) • 1x as a determiner-noun string in a forced choice task (GJT) 16 Note that introducing the items in this way does not give away any gender cues since indefinite determiners in Dutch do not show any gender distinction. 33 MA Thesis 2013 Evelyn Egger Greek Tests The set-up of the Greek tests was very similar to the Dutch one, namely two production tasks to elicit the determiner-noun and determiner-adjective-nouns strings and a grammaticality judgment task were children had to choose between a congruent and an incongruent determiner-noun combination. The Greek tests were also taken from Unsworth et al. (2011b), but in contrast to the Dutch tests, there was only one version for all children, regardless of their age. The number of target nouns that were tested was 18 (6 masculine, 6 feminine and 6 neuter) and for each gender, two different endings were used, -os and –as for masculine, -a and –i for feminine and –o and –i for neuter (see Appendix C for a complete list of the Greek target nouns). Moreover, each of the three tasks was preceded by a number of practice items. In the Det-N task, the 18 target nouns are divided into 6 sets that contain one noun of each gender. Children are presented with a set showing three target nouns and are first asked to name them. Next, the experimenter asks a series of questions about the depicted objects (e.g. ‘What is brown?’), thereby eliciting determiner-noun strings in definite contexts. An example of a test item is provided in Figure 3. Exp: Ti ine prasino? What is green? Child: o vatrachos the- MASC frog- MASC Exp: Ti echi mavri ura? What has a black tail? Child: i gata the- FEM cat- FEM Figure 3. Example of an experimental set of the Greek Det-N task that was used to elicit determiner-noun strings. Each set consists of three target nouns, one for each gender. The second production task (Det-Adj-N task) that was used was highly similar to the Det-N task. Children are again presented with an experimental set that contains three different nouns (one of each gender), but this time, there are two instances of each target noun in the picture. Crucially, the two instances of a given target noun differ in some aspect, e.g. in terms of size or colour. Children are asked to 34 MA Thesis 2013 Evelyn Egger name the three target nouns first to ensure that they are familiar with the items. Then the experimenter asks again a series of questions, such as ”What is red?” and the children are required to answer with an determiner-adjective-noun string (e.g. “The big fish”). Thus, each target noun is produced twice in this task (e.g. the brown dog and the white dog). Figure 4 illustrates an experimental set of the Det-Adj-N task. Exp: Ti ine roz? Child: I megali bala What is pink? The- FEM big- FEM ball- FEM Exp: Ti ine kokkino? What is red? Child: To mikro milo The- NEUT small- NEUT apple- NEUT Figure 4. Example of an experimental set of the Greek Det-Adj-N task that was used to elicit determiner-adjective-noun strings. Each set depicts two instances of one masculine, one feminine and one neuter noun, thereby eliciting each target noun twice in a determiner-adjective-noun string. Finally, the children were administered a grammaticality judgment task, where they had to choose between congruent and incongruent determiner-noun combinations. The task was presented orally and children had to tap the experimenter’s left or right hand depending on whether they judged the first or the second determiner-noun combination as correct. The gender of the determiner in the incongruent pairs was counterbalanced. In sum, for each target noun four responses were elicited: • 1x in a determiner-noun string in definite DPs • 2x in a determiner-adjective-noun string in definite DPs • 1x as a determiner-noun string in a forced choice task 35 MA Thesis 2013 Evelyn Egger 3.2.3 Procedure The tests were carried out in two sessions, one in Dutch (approx. 45 min.) and one in Greek (approx. 30 min.). Each child was tested individually by a (near-) native speaker of Dutch/Greek at the Greek Saturday School or at home. The Dutch session started with the vocabulary task followed by the PDT task, the GJT and finally, the STORY task. The Greek tests were administered in the order: vocabulary (DVIQ), Det-N task, Det-Adj-N task and GJT. The Dutch gender tasks were run on a computer in the form of PowerPoint presentations, while for the Greek production tasks picture booklets were used. The parental questionnaire was either completed in an interview session with one of the experimenters or by the parents themselves at home. 3.2.4 Coding and analysis Children’s responses were recorded and entered into lists after each session. For the analysis of the Dutch production tasks, the following coding schemes were adopted: Determiner-Noun agreement i. Correct responses: correct determiner and correct target noun (het neut huis neut ‘the house’) ii. Incorrect responses: incorrect determiner with correct target noun (*de com huis neut ‘the house) iii. Excluded responses Use of indefinite determiners (een huis neut ‘a house’) Determiner omission (huis ‘house’) No answer Other (use of non-target nouns) Adjective-Noun agreement i. Correct responses: correct adjectival inflection and correct target noun (een wit konijn neut /het witte konijn neut ’a white/the white rabbit’) ii. Incorrect responses: incorrect adjectival inflection and correct target noun (een *witte huis neut / het *klein huis neut ‘a white/the white house’) iii. Excluded responses Adjective omission No answer Other (use of non-target nouns) 36 MA Thesis 2013 Evelyn Egger For the analysis of the Greek production tasks, similar coding schemes were used. However, given that Greek has a rich morphology, a couple of adjustments and decisions were made. For example, in Dutch there is no overt gender marking on root nouns and thus, the choice of the determiner was taken as indicative of the gender a child has assigned to a given target noun. In Greek, nouns do carry nominal suffixes and are inflected not only for gender, but also for number and case. Thus, for Greek too the choice of the determiner was taken to be indicative of the gender assigned to a particular noun. However, case errors on determiners and/or nouns were scored as correct as long as the noun phrase (DP) was marked for the correct gender. The use of diminutives was also scored as correct since they are generally very frequently used in Greek. Moreover, cases were children produced target nouns, but in some nontarget (but still grammatically correct) form were scored as correct. This is the case for a number of multigendered root nouns (see section 2.3.1), for example if the target noun was i- FEM gata -FEM ‘the cat’, but the child produced it in its neuter form to- NEUT gati- NEUT , it was scored as correct. Determiner-Noun Agreement i. Correct responses: correct (definite/indefinite) determiner and correct target noun (i fem papia fem ‘the duck’, ena neut vivlio neut ‘a book’) ii. Incorrect responses: incorrect (definite/indefinite) determiner and correct target noun (*to neut skilos masc /skilo masc-ACC , *ena neut gata fem ) iii. Excluded responses Determiner omission No answer Other (use of non-target nouns and target-nouns in non-existing forms) (i *krana/ i *vrisa target: i fem vrisi fem ‘the tap’, *to *karxari target: o masc karxarias masc ) For the Det-Adj-N task, the coding scheme was slightly different. Here, responses that contained case errors on target nouns were coded separately (*Det-*Adj-*N) 17. Determiner-Adjective-Noun Agreement i. Correct responses: correct determiner, correct adjective and correct target noun ii. Incorrect responses *Det-*Adj-N (to neut mikro neut bala fem ‘the small ball’) *Det-*Adj-*N (to neut mikro neut elefanda masc-ACC ‘the small elephant’) *Det-Adj-N (o masc megalo neut vivlio neut ‘the big book’) Det-*Adj-N (o masc kitrino neut ilios masc ‘the yellow sun’) 17 Note that here, the * in front of the noun indicates a case error rather than a non-target noun, since responses containing nontarget nouns were excluded from the analysis, i.e. coded as ‘other’. 37 MA Thesis 2013 iii. Evelyn Egger Excluded responses Determiner omission Adjective omission Use of indeclinable adjective No answer Other (use of non-target nouns or target nouns in non-existing forms) 3.3 Results 3.3.1 Background measures Children were first administered a vocabulary test that was used as an indicator of the children’s overall language proficiency. Children were then grouped according to their vocabulary scores to see whether and how the bilinguals ‘develop’ with increasing language proficiency. There are several reasons why the children were grouped according to their vocabulary scores in the two languages and not some other variable like age. First, there is a vast amount of variability in the amount of input these bilinguals receive in the two languages both currently and in the past. For example, some children speak Greek at home, while for others the only source of Greek input is the Saturday school. Similarly, for some children the only place where they are exposed to Dutch is school, while others speak almost exclusively Dutch both in school and at home. These differences in amount of input and exposure are likely to be attenuated over the years, which in turn, could skew the results. Second, Unsworth et al. (2011b) found that vocabulary size was the second best predictor variable for bilingual children’s performance on grammatical gender in both Greek and Dutch (the best being amount of input/exposure). Moreover, the vocabulary groupings facilitate subsequent comparisons with data obtained by Algoe (2011), who used the same experimental materials and also found that vocabulary scores were a good indicator of bilinguals’ accuracy rates on the gender tasks. Although the groupings based on vocabulary scores may inform us about the relationship between acquisition of gender and language proficiency, it leaves out other factors such as amount of exposure that have been shown to affect gender acquisition in the two languages (Unsworth et al., 2011b). The Dutch vocabulary test that was used was the PPVT-III-NL, which is a measure of receptive vocabulary and is fully standardized. The average score (standard score) on the PPVT-III-NL is 100, with 85 and 115 representing the cut-off scores that lie within one standard deviation (SD) from the mean. Thus, the experimental subjects were grouped according to their vocabulary scores (WBQ), namely group A: WBQ <85, group B: WBQ 85-100, group C: WBQ 101-115 and group D: WBQ >115. Table 7 gives an overview of the groupings for the analyses of the Dutch data. Scores across groups range from 79 to 132 and the mean score is 99. It should be noted that groups A and C are comparable in age, as are groups 38 MA Thesis 2013 Evelyn Egger B and D. Moreover, there are no big differences between the groups in terms of the age ranges that are encompassed in each. Table 7. Overview of Dutch groupings based on WBQ scores. Group A WBQ Number of Average age (mean) participants (range) <85 6 9;1 (80.5) B 85-100 (5;2-11;8) 14 (93.5) C 101-115 (5;8-12;11) 8 (107.5) D >115 (123) 8;5 9;5 (4;4-13;3) 5 8;4 (5;11-11;5) To measure children’s general proficiency in Greek, the vocabulary section of the DVIQ was used. The test assesses productive vocabulary knowledge and consists of 27 items. As in Dutch, for the Greek analyses too, participants were divided into groups according to their vocabulary scores (DVIQ scores). Since the Greek vocabulary test is not fully standardized, raw scores were used, which ranged from 4 to 21 with an average of 13. It should be noted that the relatively low overall scores are the result of a very strict scoring procedure, where only the target answers that were provided by the test developers were accepted and synonymous expressions and answers were scored as incorrect. Given that there are no standard scores for the DVIQ, participants were evenly distributed across three groups, that is group A: DVIQ <11, group B: DVIQ 11-16 and group C: DVIQ >16. An overview of the groupings for the analyses of the Greek data is provided in Table 8. Again the three groups include similar age ranges, however, group A is on average more than two years younger than group B, and group B is slightly older than group C. 39 MA Thesis 2013 Evelyn Egger Table 8. Overview of Greek groupings based on DVIQ scores. Group A DVIQ Number of Average age (mean) participants (range) <11 11 7;5 (8) B 11-16 (4;4-11;4) 11 (13) C >16 (18.5) 9;10 (5;2-12;8) 11 9;0 (5;11-12;8) Thus, for the analyses of both data sets, children were grouped according to their proficiency level. For Dutch, group A represents the children that scored below average on the standardized test and are thus the least proficient. Groups B and C are within the normal range of standard scores for monolinguals and group D are above average and therefore the most advanced in terms of overall proficiency. For Greek, there was one group less, i.e. three groups, with group A again representing the least advanced and group C the most proficient. However, since the Greek vocabulary test was not fully standardized, the raw scores that were used as a measure of overall proficiency do not take into account the children’s chronological age. 3.3.2 Dutch Production Tasks STORY Task (Det-N) The story task was used to elicit definite determiners together with the 18 target nouns. Table 9 below, summarizes the group results. Accuracy on determiner-noun agreement was calculated in terms of the average number/percentage of nouns in definite noun phrases (DPs) produced with the target determiner 18. Excluded responses are given as the average number for each type of invalid responses divided by the total number of responses. Results show that children perform better on common nouns than on neuter nouns. All groups score high on common nouns (above 96%) with the exception of group D who only scores 89% correct. The lower accuracy rate for common nouns in group D is caused by two children (twins) who show a tendency to overgeneralize the common definite determiner de to neuter nouns. As to the correct use of the neuter determiner het, there seems to be a gradual improvement from group B to C and from C to D, reaching 85% correct responses in the last group. Moreover, group A 18 i.e. The number of nouns produced with the target definite determiner (de for common nouns or het for neuter nouns) divided by the total number of nouns of the same gender produced with either of these determiners. 40 MA Thesis 2013 Evelyn Egger scores almost 10% higher than group B on neuter nouns, which is unexpected. All groups show overgeneralization of de to neuter nouns, however there is a gradual decrease of this error type from group B to C and from C to D, with group A again lying between groups B and C. Another thing to note is the high number of excluded responses for groups A, B, and C. In all three groups, the total number of excluded items is higher for neuter than for common nouns. There are only three instances where children used an indefinite determiner with common nouns, while indefinite determiners with neuter nouns occurred more frequently (12 times in total). Group B omitted determiners quite frequently and more so than group C, while there were no determiner omissions for groups A and D. Group B also has the highest percentage of ‘no answer’ and ‘other’ responses. Table 9. Group results for the Story Task in both percentages and absolute numbers for the various types of responses. Group Correct Incorrect Indefinite Determiner responses responses determiner omission No answer Other Total excluded Responses A Common n=6 Neuter B Common n=14 Neuter C Common n=8 Neuter D Common n=5 Neuter 96.2% 3.8% 1.9% 1.9% 51/53 2/53 1/54 1/54 62.2% 37.8% 11.1% 3.7% 1.9% 16.7% 28/45 17/45 6/54 2/54 1/54 9/54 96.4% 3.6% 0.8% 7.9% 2.4% 11.1% 108/112 4/112 1/126 10/126 3/126 14/126 52.9% 47.1% 2.4% 5.6% 7.1% 2.4% 17.5% 55/104 49/104 3/126 7/126 9/126 3/126 22/126 100% 2.8% 1.4% 1.4% 5.6% 68/68 2/72 1/72 1/72 4/72 79.7% 20.3% 4.2% 4.2% 1.4% 1.4% 11.2% 51/64 13/64 3/72 3/72 1/72 1/72 8/72 88.9% 11.1% 40/45 5/45 84.4% 15.6% 38/45 7/45 Given the relatively high numbers of excluded responses, it is interesting to see the proportions of these three types of responses per gender and group in order to have a more complete picture of the performance patterns. The percentages of correct, incorrect and excluded responses for each group and 41 MA Thesis 2013 Evelyn Egger gender condition are thus further illustrated in Figure 5. Group B has the highest number of excluded and incorrect responses, followed by group A and C, while group D has no excluded responses. Group D produced more incorrect responses with common nouns than group A and B, while group C made no mistakes in this gender condition. Figure 5 also shows that when invalid responses are taken into account, group D still scores nearly 15% higher on neuter nouns than group C and group C around 25% higher than group B. Moreover, group B scores around 10% lower on neuter nouns than group A, while the number of excluded responses is similar for the two groups. Overall, both Table 9 and Figure 5 indicate that group B performed worse on this task than group A which suggests that at least for the less proficient learners, accuracy on determiner-noun agreement in Dutch was affected by some other factor(s) apart from vocabulary score. 100% 80% 60% 40% excluded responses 20% incorrect responses correct responses A B C Neuter Common Neuter Common Neuter common Neuter Common 0% D Figure 5. Proportions of correct, incorrect and excluded responses per group and gender condition (in percentages). PDT Task (Det-Adj-N) The picture description task (PDT) was used to elicit adjectives with definite determiners and adjectives in indefinite contexts. The group results are presented in Table 10. Accuracy scores for determiner-noun agreement were calculated in the same way as in the Story task (i.e. by dividing the number of nouns produced with the target definite determiner by the total number of nouns of the same gender produced with either de or het). Similarly, accuracy scores for adjective-noun agreement were calculated by dividing the number of responses containing target adjectival inflection (i.e. in indefinite DPs, bare adjectives with neuter nouns and in definite DPs inflected adjectives with nouns of either gender) by the total number of responses containing adjectives either inflected or bare. With regard to the use of definite determiners, all groups perform better on common than on neuter nouns, except group D, who scores slightly higher on het than on de. Group A and C perform target-like with the definite determiner de 42 MA Thesis 2013 Evelyn Egger (>90%), while groups B and D seem to use het instead of de quite often (B: 11.2%, D: 16.7%). In fact there are four children who overgeneralize het to de in at least 25% of the cases and two more who produce this kind of error in about 10% of the elicited responses. The use of the neuter definite determiner het seems to improve with increasing proficiency (as indicated by the vocabulary scores), with group A scoring 44%, group B 52%, group C 78.8% and group D reaching 84.5% correct. However, the relatively high score in group D might be influenced by two children who show a tendency to overgeneralize het. When these two children are excluded from the group average, performance drops to 72.3%, which is slightly lower than group C. However, this also leaves group D with only three participants, two of which score 100% and one who uses het with neuter nouns in only 17% of the cases. The use of the inflected adjective in definite contexts does not seem to cause significant problems for any of the four groups, B, C, and D score above 95% correct with both neuter and common nouns and group A also performs above 93%. Moreover, 13 out of the 14 errors with adjectives in definite contexts in group A were produced by one child, as were the 5 errors in group C. As to the use of adjectives in indefinite contexts, all groups perform better with common nouns than with neuter nouns. The use of the inflected adjective with common nouns is target-like in groups B, C and D (>97%). The relatively lower scores of group A (88.2%) are caused by one child who consistently produces bare adjectives in indefinite contexts with both common and neuter nouns. The same child is also partly responsible for the higher scores of group A in the indefinite neuter condition, compared to group B. Leaving group A aside, the use of bare adjectives in indefinite contexts clearly improves from group B to C and from C to D, with the most advanced group reaching 93% correct. 43 MA Thesis 2013 Evelyn Egger Table 10. Group results for the Picture Description Task (PDT). Group Indefinite -e A common n=6 neuter B common n=14 neuter C common n=8 neuter D common n=5 neuter Definite -∅ de het -e -∅ 88.2% 11.8% 100% 0% 93.1% 6.9% 90/102 12/102 102/102 0/102 95/102 7/102 36.3% 63.7% 56.0% 44.0% 93.1% 6.9% 37/102 65/102 56/100 44/100 95/102 7/102 97.8% 2.2% 88.8% 11.2% 100% 0% 221/226 5/226 198/223 25/223 226/226 0/226 50.4% 49.6% 48.0% 52.0% 98.2% 1.8% 114/226 112/226 107/223 116/223 222/226 4/226 100% 0% 97.0% 3.0% 100% 0% 138/138 0/138 130/134 4/134 138/138 0/138 21.0% 79.0% 21.2% 78.8% 96.4% 3.6% 29/138 109/138 29/137 108/137 132/137 5/137 100% 0% 83.3% 16.7% 95.8% 4.2% 72/72 0/72 60/72 12/72 69/72 3/72 6.9% 93.1% 15.5% 84.5% 97.2% 2.8% 5/72 67/72 11/71 60/71 69/71 2/71 Overall, there were less invalid responses for the PDT than the Story task. Moreover, definite determiners yielded most omissions and ‘no answer’ responses. An overview of the number and types of excluded responses per group and condition is provided in Table 11. There were much less determiner omissions in the PDT task compared to the Story task, and children did not use indefinite determiners instead of definite determiners in the definite condition in PDT. 44 MA Thesis 2013 Evelyn Egger Table 11. Overview of numbers and percentages of excluded responses from the PDT. Group Indefinite - Definite - Definite - Total adjectives determiners adjectives excluded omission A common n=6 neuter B common n=14 neuter C no answer neuter no answer responses 2.0% 2.0% 2/102 2/102 0.9% 0.9% 4.0% 2/228 3/228 2/228 2/228 9/228 0.9% 1.3% 0.9% 0.9% 4.0% 2/228 3/228 2/228 2/228 9/228 0.7% 2.2% 2.9% 1/138 3/138 4/138 neuter n=5 omission 1.3% n=8 common no answer 0.9% common D omission 0.7% 0.7% 1.5% 1/138 1/138 2/138 1.4% 1.4% 2.8% 1/72 1/72 2/72 3.3.3 Dutch Knowledge Task (GJT) The knowledge task (GJT) took the form of a forced choice task, where children had to choose between congruent and incongruent determiner-noun strings. Accuracy scores were calculated by dividing the number of nouns for which the child chose the correct determiner-noun combination by the total number of nouns of the same gender. The group results are summarized in Table 12. The picture is surprising since children seem to perform better on neuter than on common nouns. Similar to the Story task, the high scores on neuter nouns and low scores on common nouns for group D can again be ascribed to two children (the twins already mentioned above), who show the same tendency to overgeneralize het to common nouns also in the knowledge task. Group C is the only one who performs better on common than neuter nouns and reaches high scores for both (>91%). The results of group A and B are more difficult to interpret. In group B there seems to be a handful of children who show chance performance, but at the same time four of the children in this group score 100% correct in both conditions. The picture for group A is however, different. In this group there are two children, who score above 90% in both 45 MA Thesis 2013 Evelyn Egger conditions and one who seems to show chance performance, while the remaining three score considerably higher with neuter than with common nouns. Table 12. Group results (percentages correct) for the knowledge task (GJT). Group Common Neuter Both A 64.7% 80.4% 72.6% n=6 33/51 41/51 74/102 B 67.5% 78.1% 72.8% n=14 77/114 89/114 166/228 C 97.1% 91.3% 94.2% n=8 67/69 63/69 130/138 D 69.4% 94.4% 81.9% n=5 25/36 34/36 59/72 3.3.4 Comparisons Comparison with Unsworth (2013a) and Algoe (2011) In order to compare the performance patterns of the Dutch-Greek bilinguals with those of the EnglishDutch bilinguals of Unsworth’s (2013a) study, the group results of the latter are replicated in Table 13 19. As for adjective-noun agreement in indefinite contexts, both the English-Dutch and the Greek-Dutch bilinguals perform better with common than neuter nouns and both populations show gradual improvement with neuter nouns (with increasing age for the English-Dutch and vocabulary for the GreekDutch 20), while accuracy scores for common nouns are very high for all groups (>93% except for the 6year-olds in Unsworth and group A in the present study who score around 88%). With regard to definite determiners, production is again better with common than with neuter nouns for both the English-Dutch and the Greek-Dutch bilinguals and both show (more or less) gradual improvement for neuter nouns with increasing age/vocabulary, while accuracy scores for common nouns are generally high. However, the English-Dutch bilinguals appear to commit very few errors where they use het instead of de (max. 3.5% for the 6-year-olds), whereas the Greek-Dutch bilinguals of the present study produced such errors much more frequently (11% for group B, 3% for group C and 17% for group D, see Table 10). There are at least 19 It should be kept in mind though, that the comparison is not entirely straight-forward, since the bilinguals in Unsworth’s study were grouped according to their chronological age, while the participants of the present study were divided on the basis of their vocabulary scores. 20 At least from group B to C and C to D. 46 MA Thesis 2013 Evelyn Egger 6 Greek-Dutch bilinguals who show a tendency to overgeneralize het to common nouns (i.e. in more than 14% of the cases). When it comes to judging the grammaticality of determiner-noun combinations, there seems to be a difference between Unsworth’s bilinguals and the children of the present study. As can be seen in Table 13, the English-Dutch bilinguals perform again better with common than with neuter nouns, while this is not necessarily the case for the Greek-Dutch bilinguals (see Table 12). For the bilinguals of the present study all groups, seem to perform better with neuter than with common nouns, except group C who is near target-like in both gender conditions 21. Table 13. Group results of the English-Dutch bilinguals in Unsworth (2013a) for the production of adjectives in indefinite contexts (PDT) and determiners in definite contexts (PDT) as well as accuracy scores for the GJT with congruent and incongruent determiner-noun combinations. Indef. + Adjective Def. + Determiner- Production Common Neuter Common Neuter 4 year olds 93.6% 20.2% 96.7% 10.2% 5 year olds 96.6% 27.0% 99.6% 6 year olds 88.5% 52.6% 7 year olds 98.0% 8 year olds Def. + Determiner- Judgment Common neuter 21.2% 77.8% 56.1% 96.5% 42.1% 85.6% 72.6% 50.0% 98.1% 69.3% 92.1% 83.3% 94.8% 60.7% 97.7% 65.1% 96.3% 83.0% 9 year olds 100% 81.1% 98.6% 87.0% 100% 93.1% 10 year olds 99.5% 80.3% 98.7% 74.4% 96.0% 91.9% 11 year olds 97.0% 58.3% 96.8% 55.6% 90.7% 83.3% 12 & 13 year olds 100% 74.2% 98.0% 80.8% 95.7% 90.9% 14 to 17 year olds 100% 52.5% 96.6% 64.2% 85.9% 85.9% In sum, the two bilingual populations show largely similar performance patterns across tasks and conditions. However, the there are also differences between the two population in that the Greek-Dutch bilinguals show a tendency to overgeneralize het to contexts that require the common definite determiner. This is crucial since the overgeneralization of het is an error that monolinguals hardly ever make 22 and has not been consistently found in bilingual populations either 23. Another difference between the two populations is the Greek-Dutch bilinguals’ better performance with neuter than with common nouns in the GJT task. This finding is highly unusual given that (as far as we know) no study so far has reported better 21 It should be noted that the group averages for this task might be misleading, since there were a handful of children who showed chance-level performance and some had a (more or less clear) tendency to overgeneralize het. Moreover, in all groups there were children who performed target-like in both gender conditions. 22 In monolinguals overgeneralization of het is found only occasionally in very advanced stages of gender acquisition. 23 E.g. Blom et al. (2008) report overgeneralization of het for all tested groups (monolinguals, L2 children and L2 adults), while other studies have not found such a tendency (e.g. Unsworth, 2013a; Unsworth et al., 2011b; Hulk & Cornips, 2006a). 47 MA Thesis 2013 Evelyn Egger performance in neuter than in common conditions in any task. These differences between the EnglishDutch and the Greek-Dutch bilinguals suggest that the nature of the gender system of the ‘other L1’ has an influence on the acquisition of Dutch gender. Unfortunately, a more direct comparison between the two experimental populations is difficult due to the different grouping criteria that were employed 24. A comparison with the Italian-Dutch bilinguals in Algoe (2011) is provided in Table 14 25. Again the general performance pattern of the two populations seems to be highly similar. Both perform better on common than on neuter nouns in both agreement types (i.e. determiner-noun and adjective-noun agreement) and both show higher accuracy with neuter nouns as a function of increasing vocabulary scores 26. Unlike the Greek-Dutch bilinguals however, the Italian-Dutch children in Algoe did hardly produce errors of the type het for de (only 3% in group C), similar to the English-Dutch bilinguals in Unsworth (2013a) suggesting a qualitative difference between the Italian-Dutch and the Greek-Dutch bilinguals as well. Finally, comparing the performance with adjectival inflection in the definite condition between the two kinds of bilinguals, another interesting difference emerges. The Italian-Dutch bilinguals perform at ceiling with common nouns while performance on neuter nouns seems to decrease from group A to B, B to C and C to D. In contrast, all Greek-Dutch bilinguals perform (near) target-like in both gender conditions (>93%) 27. Comparing the actual accuracy scores between the two populations, it appears that with regard to neuter nouns, the Greek-Dutch children of the present study did consistently better than the Italian-Dutch bilinguals in Algoe (2011) (except group D on determiner-noun agreement) and the Italian-Dutch bilinguals generally did better on common nouns than the Greek-Dutch children, although these differences on common nouns were much smaller due to the high performance levels. 24 The age distribution of the Greek-Dutch bilinguals is quite uneven and grouping them according to chronological age would leave many groups with only one or two participants. Moreover, the group averages for the vocabulary measure (PPVT-III-NL) of the English-Dutch children in Unsworth lie between 103-117, while there seems to be more variation in the Greek-Dutch bilinguals, which means that only group C and some children of group B and D of the present study would match the vocabulary scores of Unsworth’s participants. 25 Here it should be kept in mind that although the two populations are grouped on the same criteria, i.e. vocabulary score, they differ considerably in terms of age. For the Italian-Dutch bilinguals the average ages are 5;2 for group A, 5;10 for group B, 6;10 for group C and 5;7 for group D, while the average age of the participants of the present study was considerably older, i.e. 9;1 for group A, 8;5 for group B, 9;5 for group C and 8;4 for group D. 26 With the exception of groups B who in both studies score lower than the respective groups A. 27 The results of the Story task are not discussed here, since the picture is very similar to the results for the determiner-noun combinations in the PDT task. Children in both studies perform better on common than on neuter and overall slightly better with in the Story task than in the PDT with this type of agreement. Moreover, only the Greek-Dutch bilinguals show a tendency to overgeneralize het to contexts that require de (albeit to a lesser degree than in the PDT task). 48 MA Thesis 2013 Evelyn Egger Table 14. Comparison of percentages correct for the PDT between the Dutch-Greek bilinguals of the present study and the Dutch-Italian bilinguals in Algoe’s (2012) study. Group A Indefinite + Adj. common neuter B common neuter C common neuter D common neuter Definite determiner Definite + Adj. Greek Italian Greek Italian Greek Italian 88.2% 100% 100% 100% 93.1% 100% 90/102 12/12 102/102 12/12 95/102 12/12 63.7% 33.3% 44.0% 16.6% 93.1% 91.6% 65/102 4/12 44/100 2/12 95/102 11/12 97.8% 100% 88.8% 100% 100% 100% 221/226 90/90 198/223 90/90 226/226 90/90 49.6% 24% 52.0% 28% 98.2% 87% 112/226 25/90 116/223 28/90 222/226 76/90 100% 96% 97.0% 97% 100% 98% 138/138 137/144 130/134 139/144 138/138 141/144 79.0% 40% 78.8% 35% 96.4% 80% 109/138 56/144 108/137 50/144 132/137 114/144 100% 100% 83.3% 100% 95.8% 100% 72/72 18/18 60/72 18/18 69/72 18/18 93.1% 78% 84.5% 100% 97.2% 50% 67/72 14/18 60/71 18/18 69/71 9/18 Comparison determiner-noun agreement vs. adjective-noun agreement Accuracy scores for determiner-noun agreement and adjective-noun agreement are plotted against each other in Figure 6. For adjectives, only the responses in the indefinite condition were taken into consideration and accuracy scores were again calculated by dividing the number of responses with target adjectival inflection by the total number of responses for nouns of the same gender that contained an adjective (regardless of whether it is inflected or not). For the definite determiners, the plots in Figure 6 represent the average percentage correct of both production tasks for de and het respectively. For common nouns, determiner-noun agreement is target-like in all groups and hence there is no notable improvement across groups, but interestingly, as noted above, there is a drop in accuracy between group C and D (from 98.4% to 84.4%). For adjectives with common nouns, there is improvement from group A to B, who reaches target-like performance as do groups C and D (>97%). Group A performs better on determinernoun agreement than on adjectival agreement, while the reverse is true for group D. Groups B and C do 49 MA Thesis 2013 Evelyn Egger not show a difference between accuracy scores in determiner-noun and adjective-noun agreement. As to neuter nouns, group A and D seem to perform better on adjectival agreement than on determiner-noun agreement, but this difference does not hold for groups B and C who show similar accuracy rates for the two types of agreement. Finally, accuracy scores for neuter nouns seem to improve more or less in parallel across groups. In other words, results point to a close relationship between accuracy rates on determiner-noun and adjectival agreement for neuter nouns (at least for groups B, C and D). 100.0% % correct 80.0% 60.0% 40.0% 20.0% 0.0% A B C D de 98.1% 93.0% 98.4% 84.4% het 48.1% 48.6% 79.9% 83.1% -e 83.3% 97.6% 100.0% 100.0% -Ø 65.7% 49.2% 77.4% 91.7% Figure 6. Accuracy scores for determiner-noun and adjective-noun agreement for both gender conditions. Comparison production tasks vs. knowledge tasks The last comparison concerns the performance on determiner-noun agreement in the production and judgment tasks (since the GJT only tested determiner-noun combinations). Table 15 gives the average percentages of correct determiner-noun agreement of the two production tasks and the average scores on the judgment task. For common nouns, it seems that all groups do better with production than with judgment and the difference is particularly big for groups A and B (over 30%) as well as D (over 20%), while for group C the difference is negligible. The reverse pattern holds for neuter nouns, here judgment appears to be better than production, however the discrepancy also gets smaller with increasing vocabulary scores, from over 30% for group A to 25% in group B and about 10% in groups C and D. Thus, the findings are in line with Unsworth & Hulk (2010a) and Unsworth (2013a) who have reported the same kind of asymmetries between the two genders and tasks. However, for the monolinguals in Unsworth & Hulk (2010a) this asymmetry was only found for the more advanced group (group II), but not for the less advanced children (group I) and was argued to reflect a developmental stage in the acquisition of Dutch gender where children become aware of the paradigmatic relationship between de 50 MA Thesis 2013 Evelyn Egger and het which leads them to occasionally assign neuter gender to common nouns in the judgment task. In contrast, in the present study all groups showed this performance pattern (even the less advanced groups) suggesting that the Greek-Dutch bilinguals reach this developmental stage much earlier than monolinguals and/or remain longer in this particular stage. Table 15. Average percentage correct for determiner-noun agreement in the two production tasks and the average scores for the judgment task. Group de Production het Judgment Production Judgment A 98.1% 63.0% 48.1% 80.6% B 93.0% 59.9% 48.6% 72.2% C 98.4% 97.2% 79.9% 91.0% D 84.4% 63.3% 83.1 93.3% 3.3.4 Greek Production Tasks Det-N Task The Det-N task was used to elicit determiner-noun strings for the 18 target nouns. We will first consider all types of responses for each gender seperately. The group results for neuter nouns are presented in Table 16. Children did not produce any ungrammatical determiner-noun strings for neuter nouns and all groups produced diminutives in this condition. However, they hardly used the indefinite determiner regardless of whether they produced a root noun or a diminutive. To note is the high number of excluded responses, which reaches over 35% in group A, 7% in group B and almost 14% in group C. Moreover, all but two of the excluded responses are determiner omissions, which are particularly frequent in group A (which is also the youngest group), where they constitute almost 35% of the total responses and group C omitted determiners more frequently than group B. The determiner omissions in group A were produced by six children, in group B all omissions are due to one child and in group C there were three children that omitted determiners in this task. 51 MA Thesis 2013 Evelyn Egger Table 16. Group results in numbers and percentages for the various types of responses for neuter nouns in the Det-N task. Correct responses definite A B C def + dim Total indef indef + dim Excluded responses omiss no answer Total other 48% 10.7% 2.7% 1.3% 62.7% 34.7% 2.7% 37.3% 36/75 8/75 2/75 1/75 47/75 26/75 2/75 28/75 89.0% 4.1% 93.1% 6.9% 6.9% 65/73 3/73 68/73 5/73 5/73 73.8% 8.8% 3.6% 86.2% 13.8% 13.8% 59/80 7/80 3/80 69/80 11/80 11/80 Unlike neuter nouns, masculine nouns yielded a relatively high number of incorrect responses. The numbers and percentages for the various kinds of responses for masculine nouns are presented in Table 17. From the total number of responses, group A only gave 5% correct responses, while group B provided the correct answer in 73% of the cases and group C in 87%. In fact only two children in group A produced masculine determiners at all and two children in group B did not use any masculine determiners. Moreover, group A produced over 40% incorrect determiner-noun combinations, while group B produced errors in 16% of the cases and group C made very few errors with masculine nouns, namely less than 4%. All but three of the errors involved the substitution of the masculine determiner by the neuter determiner (n=36, 2 indefinite and 34 definite) and from the total 39 incorrect responses, 24 (62%) contained nouns where children had dropped the final –s, e.g. to neut elefanda instead of o masc elefandas masc ‘the elephant’, to neut skilo instead of o masc skilos masc ‘the dog’, to neut /i fem kuva instead of o masc kuvas masc ‘the bucket’, etc. This type of error is interesting because on the one hand, the dropping of the final –s with masculine nouns usually coincides with the form it would take in the accusative singular (e.g. ton masc elefanda masc ) 28 and on the other hand, -s dropping with the (typically) masculine endings –os and –as yields endings that are highly predictive of neuter gender, i.e. –o and –a (where –a is also a possible ending for feminine nouns which would explain the single occurrence of i fem kuva ‘the bucket’). As with neuter nouns, children also produced many invalid responses with masculine nouns (group A: 52%, group B: 11% and group C: 9%). The majority of excluded responses were determiner omissions, which were highest for group A (44%) and considerably lower for groups B and C (7-8%). 28 Note that only masculine nouns have two different forms for nominative and accusative singular, while the vast majority of feminine nouns and all neuter nouns take the same form in the nominative and accusative singular. 52 MA Thesis 2013 Evelyn Egger Table 17. Group results in numbers and percentages for the various types of responses for masculine nouns in the Det-N task. A B C Correct Incorrect Total definite def. neut indef. neut def. fem 4.8% 35.5% 3.2% 4.8% 3/62 22/62 2/62 3/62 73.0% Excluded Total omission no answer other 43.6% 43.5% 8.1% 51.6% 27/62 27/62 5/62 32/62 15.9% 15.9% 7.9% 1.6% 1.6% 11.1% 46/63 10/63 10/63 5/63 1/63 1/63 7/63 87.0% 3.7% 3.7% 7.4% 1.9% 9.3% 47/54 2/54 2/54 4/54 1/54 5/54 Children did better with feminine nouns than with masculine nouns. The group results for feminine nouns are presented in Table 18. Considering again all types of responses, group A gave correct answers in 18% of the cases, group B in 84% and group C in 95%. There was only one instance where a (feminine) diminutive was used instead of a root noun and three cases where children used an indefinite determiner. Only three children in group A produced feminine determiners at all, while two children in group B did not use any feminine determiners. The number of incorrect responses for feminine nouns was lower than for masculine nouns, namely 18% for group A, 5% for group B and group C did not produce any incorrect responses. When children gave incorrect responses, they always used a neuter determiner instead of a feminine determiner (1 indefinite and 13 definite). The number of excluded responses was high also for feminine nouns, reaching 64% in group A, 11% in group B and 5% in group C and the majority of these kinds of responses involved the omission of the determiner (group A: 48%, group B: 11% and group C: 5%). Table 18. Group results in numbers and percentages for the various types of responses for feminine nouns in the Det-N task. Correct def A B C Total def+dim indef Incorrect def indef neut neut Total Excluded omission Total no other answer 16.4% 1.6% 18.0% 16.4% 1.6% 18.0% 47.5% 13.1% 3.3% 64.0% 10/61 1/61 11/61 29/61 8/61 2/61 11/61 10/61 1/61 39/61 83.9% 83.9% 4.8% 4.8% 11.3% 11.3% 52/62 52/62 3/62 7/62 7/62 3/62 90.6% 1.6% 3.1% 95.3% 4.7% 4.7% 58/64 2/64 3/64 3/64 1/64 61/64 53 MA Thesis 2013 Evelyn Egger The group results for the three gender conditions are summarized in Table 19. The average percentage correct was calculated by dividing the total number of correct responses by the number of responses containing nouns of the same gender produced with any determiner. Considering only the valid responses, all groups scored 100% correct with neuter nouns suggesting that children might use neuter as a default gender. All groups score higher on feminine than on masculine nouns and accuracy improves from group A to B and group B to C for both genders. Whereas group C performs target-like on all three genders, group B is target-like on neuter and feminine, while scores for masculine are lower, but still relatively high (82%). Group A scores very low with masculine nouns (10%) and performs considerably better with feminine nouns (50%). Seven of the eleven children in group A only produced neuter determiners, one produced only neuter and a few feminine determiners and one child did not use any determiners at all. In group B, one child only produced neuter determiners and one only neuter and feminine determiners. The number of excluded responses was very high for group A, regardless of gender condition (between 37% and 64%), while for groups B and C the percentages were much lower (between 5% and 14%). Overall there were more invalid responses for masculine nouns than for neuter nouns and more for feminine nouns than neuter and masculine nouns. Table 19. Summary of the group results for the three gender conditions of the Det-N task. Neuter Correct A B C Masculine Incorrect Excluded Feminine Correct Incorrect Excluded Correct Incorrect Excluded 100% 37.3% 10% 90% 51.6% 50.0% 50.0% 64.0% 47/47 28/75 3/30 27/30 32/62 11/22 11/22 39/61 100% 6.9% 82.1% 17.9% 11.1% 94.5% 5.5% 11.3% 68/68 5/73 46/56 10/56 7/63 52/55 3/55 7/62 100% 13.8% 95.9% 4.1% 9.3% 100% 4.7% 69/69 11/80 47/49 2/49 5/54 61/61 3/64 Det-Adj-N Task The Det-Adj-N task was administered to elicit two determiner-adjective-noun strings for each of the 18 target nouns. As with the Det-N task, we will first consider all types of responses for each gender seperately. Table 20 presents the group results for neuter nouns. Children made very few errors with neuter nouns (group A: 2%, group B: 3% and group C: 0%) and five of the seven errors were of the type *D-*Adj-N, that is the determiner and adjective were agreeing in gender, but they were not agreeing with the gender of the target noun (e.g. i fem megali fem tilefono neut ‘the big telephone’, i fem kitrini fem psari neut ‘the 54 MA Thesis 2013 Evelyn Egger yellow fish’). Children in group C performed better than group B and group B did better than A and all groups had more than 90% correct from the total number of responses. The number of excluded items was much lower than in the Det-N task and decreases from group A (8%) to group B (5%) and from B to group C (1%). Most invalid responses of groups A and B were due to the omission of the determiner or adjective and there were four ‘no answer’ responses in group A. Table 20. Group results in number and percentages for the various kinds of responses for neuter nouns in the Det-Adj-N task. Correct Incorrect D-Adj-N A B C *D*Adj Total Excluded *D*Adj*N *D Total Det. Adj. Indecl. No omiss omiss Adj. answer other 90.1% 1.3% 0.7% 2.0% 4.6% 0.7% 2.6% 7.9% 136/151 2/151 1/151 3/151 7/151 1/151 4/151 12/151 92.5% 2.1% 0.7% 2.7% 3.4% 1.4% 4.8% 135/146 3/146 1/146 4/146 5/146 2/146 7/146 99.3% 0.7% 0.7% 149/150 1/150 1/150 Children had considerably more problems in establishing gender agreement with masculine nouns. The group results are given in Table 21. Group A gave correct responses in 8% of the cases, group B in 70% and group C in 88% of the total number of responses. The number of incorrect responses was highest for group A (70%), followed by group B with 24% and group C gave 8% incorrect responses. The major error type in all groups was *Det-*Adj-*N where the determiner and adjective are agreeing in gender, but there is no agreement between the determiner and the noun, which was not produced in its target form (e.g. to neut aspro neut skilo instead of o masc aspros masc skilos masc ‘the white dog’, to neut megalo neut kuva instead of o masc megalos masc kuvas masc ‘the big bucket’). Moreover, group A (and to a less extent also group B) produced errors of the type *Det-*Adj-N (to neut mikro neut vatrachos masc ‘the small frog’, to neut megalo neut skilos masc ‘the big dog’), while errors that involved a mismatch between the gender of the determiner and the gender of the adjective were very infrequent (2.4% for group A and 1.5% for group B). The number of invalid responses reaches 22% in group A and drops to 6% in group B and 4% in group C. Most excluded responses were due to determiner or adjective omission (24/41, 59%), but there were also some cases were children used indeclinable adjectives or non-target nouns (e.g. karchari instead of karcharias ‘shark’ or kuvi instead of kuvas ‘bucket’, which were coded as ‘other’) and a few cases were children did not provide any response. 55 MA Thesis 2013 Evelyn Egger Table 21. Group results in number and percentages for the various kinds of responses for masculine nouns in the Det-Adj-N task. A B C Correct Incorrect D-Adj-N *D*Adj Total *D*Adj*N *D *Adj Excluded Total Det. Adj. Indecl. No omiss omiss Adj. answer other 8% 33.6% 33.6% 1.6% 0.8% 69.6% 10.4% 3.2% 1.6% 3.2% 4.0% 22.4% 10/125 42/125 42/125 2/125 1/125 87/125 13/125 4/125 2/125 4/125 5/125 28/125 70.2% 4.6% 17.6% 1.5% 23.7% 1.5% 1.5% 3.1% 6.1% 92/131 6/131 23/131 2/131 31/131 2/131 2/131 4/131 8/131 88.1% 7.6% 7.6% 0.8% 1.7% 1.7% 4.2% 104/118 9/118 9/118 1/118 2/118 2/118 5/118 Overall, children did again better with feminine than with masculine nouns. The group results are presented in Table 22. The number of correct responses increases from group A (23%) to group B (80%), reaching 98% in group C. Group A gave incorrect responses in 55% of the cases and group B in 8%, while group C did not produce any incorrect responses. The most common error type was *Det-*Adj-N (e.g. to neut megalo neut zoni fem ‘the big belt’, to neut megalo neut gata fem ‘the big cat’) and there were only two instances where the determiner did not agree with the gender of the noun (and adjective). The number of excluded responses was 22% for group A, 12% for group B and 2% for group C. Determiner omissions were quite frequent in group A (10%), while group B only produced a handful of responses were either the determiner or adjective were missing (2% and 3%, respectively). In all groups there were children who used indeclinable adjectives in their responses and all groups produced other kinds of responses that did not fit into the adopted coding scheme. Finally, there were only two instances where children did not provide any answer, which were both in group A. Table 22. Group results in number and percentages for the various kinds of responses for feminine nouns in the Det-Adj-N task. A B C Correct Incorrect D-Adj-N *D*Adj Total *D Excluded Total Det. Adj. Indecl. No omiss omiss Adj. answer other 23.3% 53.3% 1.7% 55.0% 10.0% 1.7% 1.7% 8.3% 21.7% 28/120 64/120 2/120 66/120 12/120 2/120 2/120 10/120 26/120 79.8% 8.4% 8.4% 1.7% 2.5% 5.0% 2.5% 11.8% 95/119 10/119 10/119 2/119 3/119 6/119 3/119 14/119 97.7% 1.6% 0.8% 2.3% 125/128 2/128 1/128 3/128 56 MA Thesis 2013 Evelyn Egger A summary of the group results is provided in Table 23. As in the Det-N task, the average percentage correct for each gender condition was calculated by dividing the number of correct responses by the total number of nouns of the same gender that were produced with a determiner and an adjective that were marked for (any) gender. When only the valid responses are taken into account, all children seem to perform target-like on neuter nouns, scoring over 97% correct and thus, supporting the claim that neuter has a default function. All groups perform better on feminine than on masculine and better on neuter than both masculine and feminine. For both masculine and feminine, accuracy improves from group a to B and from group B to C. Group A scores very low on both masculine (10%) and feminine (30%), while group B is almost target-like with feminine (>90%) and scores much higher on masculine than group A (75% compared to 10%). Group C scores 100%, but still makes occasional errors with masculine nouns (92% correct). Table 23. Summary of the group results for the three gender conditions of the Det-Adj-N task. Neuter A B C Masculine Feminine Correct Incorrect Excluded Correct Incorrect Excluded Correct Incorrect Excluded 97.8% 2.2% 7.9% 10.3% 89.7% 22.4% 29.8% 70.2% 21.7% 136/139 3/139 12/151 10/97 87/97 28/125 28/94 66/94 26/120 97.1% 2.9% 4.8% 74.8% 25.2% 6.1% 90.5% 9.5% 11.8% 135/139 4/139 7/146 92/123 31/123 8/131 95/105 10/105 14/119 100% 0.7% 92.0% 8.0% 4.2% 100% 2.3% 149/149 1/150 104/113 9/113 5/118 125/125 3/128 3.3.5 Greek Knowledge Task (GJT) In the knowledge task (GJT), children had to choose between congruent and incongruent determiner-noun combinations. Accuracy scores were calculated by dividing the number of correct responses by the total number of nouns of the same gender. Two children showed a response bias and were excluded from the analysis (both were from group A and among the youngest participants, i.e. < 6;0). The group results are presented in Table 24 below. Performance on each of the three genders improves with increasing proficiency (as indicated by the vocabulary scores). Overall children perform better on feminine than on masculine nouns and better on neuter than both masculine and feminine nouns. The lower scores for group A are at least partly due to three children who perform at chance level. Exclusion of these three children results in scores above 90% in all three genders for group A. In group B there is also one 57 MA Thesis 2013 Evelyn Egger participant (aged 5;2) who seemed to choose between the congruent and incongruent determiner-noun strings in a random fashion. Generally, scores are high for all groups, with the most advanced group scoring 100% in all three conditions. Table 24. Group results (percentages correct) for the knowledge task (GJT). Group Average Neuter Masculine Feminine A 79.6% 88.9% 72.2% 77.8% B 90.4% 92.4% 86.4% 92.4% C 100% 100% 100% 100% 3.3.6 Comparisons Comparison Det-N agreement vs. Det-Adj-N agreement Accuracy scores for the Det-N and the Det-Adj-N task are presented in Figure 7. For neuter nouns, all children perform at ceiling in both tasks, although groups A and B did make some errors in the Det-Adj-N task, but not in the Det-N task. For masculine nouns, group A scores equally low in both tasks, while groups B and C do slightly worse in the Det-Adj-N task (the difference amounts to 7% for group B and 4% for group C). For feminine nouns, there is quite a big difference in accuracy scores between the two tasks for group A (20%), whereby children did again better on the Det-N than on the Det-Adj-N task. For group B this difference is very small (5%), but again children did better in the Det-N task, while group C performs at ceiling in both tasks. Overall, it seems that type of agreement has not a lot of influence on childrens’ accuracy scores (with the exception of group A, who does considerably worse with feminine nouns in the Det-Adj-N compared to the Det-N task). Nevertheless, the fact that children do make more errors in the Det-Adj-N task than in the Det-N task is inconsistent with the hypothesis that children make use of a rule-based system for syntactic agreement already from the earliest stages of gender acquisition as has been suggested by Unsworth (2013b). Alternatively, it could be argued that the application of these agreement rules is compromised in syntactically more complex contexts which would explain the higher error rates in determiner-adjective-noun agreement than in determiner-noun agreement. 58 MA Thesis 2013 Evelyn Egger 100% % correct 80% 60% 40% 20% 0% Det-N Det-Adj-N Neuter Det-N Det-Adj-N Masculine Det-N Det-Adj-N Feminine A 100% 97.80% 10% 10.30% 50% 29.80% B 100% 97.10% 82.10% 74.80% 94.50% 90.50% C 100% 100% 95.90% 92% 100% 100% Figure 7. Accuracy scores for the Det-N and the Det-Adj-N task in percentages. Comparison with Canta (2012) As mentioned above, Canta (2012) tested a group of Italian-Dutch bilinguals on an Italian version of the PDT task. The group results are summarized in Table 25. The young group consisted of 17 children aged 3;3-5;6 and the old group comprised 10 children aged 5;6-7;11 29. For the Italian-Dutch bilinguals the most common error type involved a mismatch between the gender of the adjective and the gender of the determiner and noun (i.e. *Adj). With feminine nouns, children also committed a number of errors where both determiner and adjective were produced in the incorrect gender and for the older bilinguals this was the most frequent error type (10%). A comparison of the error types of the Italian-Dutch bilinguals tested in Italian with the error profile of the Greek-Dutch bilinguals tested in Greek (see Table 26) reveals some interesting differences. Unlike in Italian, in Greek children hardly produce any errors that involve either an incorrect determiner or an incorrect adjective, that is, they almost always make the adjective agree with the gender of the determiner. In Italian, errors that involve only one incorrect element within the DP are much more frequent (5% - 6% for adjectives in the masculine condition and 6% - 19% for adjectives in the feminine condition). In Greek, children produced this kind of error only in three instances with masculine nouns. In contrast, the Greek-Dutch bilinguals produced many errors of the type *Det *Adj (including *Det *Adj *N for masculine). For group A more than 50% of the valid responses for feminine nouns contained this kind of error and incorrect responses of group B were all of this type (8%). For 29 It should be kept in mind that the Greek-Dutch bilinguals of the present study were grouped according to a different criteria (i.e. vocabulary scores) and were considerably older than the Italian-Dutch bilinguals in Canta (2012). 59 MA Thesis 2013 Evelyn Egger masculine nouns, 67% of all responses of group A contained this error and which decreased to 22% in group B and to 8% in group C. Table 25. Group results of the Italian-Dutch bilinguals in Canta (2011) for the Italian version of the PDT task. For masculine nouns the il/l’ distinction was not taken into account. Masculine Group Correct Feminine Incorrect *Det Excluded *Adj Correct *Det Incorrect *Det Excluded *Adj *Det *Adj Young 92.8% 1.5% *Adj 5.7% 30.7% 180/194 3/194 11/194 Old 94.1% 0.3% 5.3% 86/280 0.3% 68.1% 32/608 18.8% 10% 109/160 5/160 30/160 16/160 5.3% 542/576 2/576 30/576 2/576 3.1% 83% 1.3% 5.9% 9.8% 439/529 7/529 31/529 52/529 35.5% 88/248 2.8% 15/544 Table 26. Number and percentages of error types in the Det-Adj-N task per group and gender condition. Errors of the type *Det*Adj*N are collapsed with the errors of the type *Det *Adj. Neuter A B C Masculine Feminine *D*Adj *D *D*Adj *D *Adj *D*Adj *D 1.3% 0.7% 67.2% 1.6% 0.8% 53.3% 1.7% 2/151 1/151 84/125 2/125 1/125 64/120 2/120 2.7% 22.1% 1.5% 8.4% 4/146 29/131 2/131 10/119 7.6% 9/118 Comparison production tasks vs. knowledge task To compare children’s performance with determiner-noun agreement on the production tasks and the knowledge task, accuracy scores are summarized in Table 25 below. For masculine and feminine nouns, all groups perform better in the judgment than in the production tasks. The difference is particularly large for group A where it reaches 63% for masculine and 41% for feminine nouns. Group C is target-like on determiner-noun agreement in all conditions (>98%), with the exception of the production of masculine nouns which still yields occasional errors even in the most advanced group (92% correct). Group B also demonstrates few problems with determiner-noun agreement and scores over 86% correct in all conditions, except with the production of masculine nouns where accuracy scores are 77%. For neuter 60 MA Thesis 2013 Evelyn Egger nouns, the picture looks slightly different. While group C performs at ceiling in both kinds of tasks (>99%), groups A and B seem to do better on production than on judgment of the target determiner-noun combinations, although this difference is admittedly small with 6% for group A and 4% for group B. Neuter Production Masculine Judgment Production Feminine Judgment Production Judgment A 95.1% 88.9% 9.4% 72.2% 36.7% 77.8% B 96.3% 92.4% 76.9% 86.4% 87.2% 92.4% C 99.7% 100% 92% 100% 98.9% 100% 4 Discussion The present study aimed at investigating the acquisition of grammatical gender in Greek-Dutch bilingual children in the two respective languages. Having presented the relevant data, in the next section the results will be interpreted in light of the research questions introduced in section 3.1 30. Finally, in section 4.2, we will try to put the present findings into a broader context and discuss their relevance for investigations of bilingual SLI. 4.1 General Discussion The first research question asked whether having a transparent gender system with extensive gender marking in the L1 (in the present case Greek) has an effect on the bilingual acquisition of Dutch gender. The first observation that can be made is that the Greek-Dutch bilinguals show improvement on gender agreement (both determiner-noun and adjective-noun) as a function of general language proficiency (as indicated by their vocabulary scores). Thus, the bilinguals of the present study do not seem to fossilize in the acquisition of Dutch gender, i.e. given they receive sufficient amount of exposure (and time), they are perfectly able to acquire the Dutch gender system in a target-like fashion. Unfortunately, a direct comparison with the English-Dutch bilinguals from Unsworth’s (2013a) study who used the exact same tests is a bit difficult due to different grouping criteria (age in Unsworth vs. vocabulary score in the present study). Nevertheless, results indicate that the two populations show some crucial differences in 30 Throughout the discussion of the results, it should be kept in mind that we did not test for statistical significance of the various differences. 61 MA Thesis 2013 Evelyn Egger their performance patterns. First, Greek-Dutch bilinguals seem to overgeneralize het to common nouns much more frequently than both the English-Dutch and the Italian-Dutch bilinguals. Second, the GreekDutch bilinguals appear to perform better on neuter than on common nouns in the grammaticality judgment task, while the English-Dutch bilinguals show the reverse pattern. It is likely that the better performance on neuter nouns in the GJT by the Greek-Dutch bilinguals is the result of a tendency to overgeneralize het, i.e. this tendency to use het instead of de is even stronger when children are asked to judge congruent and incongruent determiner-noun combinations. Given the lack of consistent and transparent gender cues in Dutch, which has been claimed to be one of the main factors contributing to its late acquisition, we take the overgeneralization of het to common nouns as a sign of gender awareness, i.e. children are aware that there is a grammatical gender distinction. Moreover, the overgeneralization of the neuter determiner in Dutch can be interpreted as an instance of cross-linguistic influence from Greek to Dutch, given that the Italian-Dutch bilinguals in Canta (2012) did not show such a tendency and given that neuter is the default gender in Greek (Tsimpli & Hulk, to appear). When comparing the results of the present study with those obtained by Algoe (2011) for Italian-Dutch bilinguals, a similar observation can be made. The Italian-Dutch bilinguals hardly ever overgeneralize het to common nouns, while the GreekDutch produce this error quite frequently. Moreover, a direct comparison between the two data sets reveals an interesting pattern. On the one hand, the Greek-Dutch bilinguals score consistently higher on neuter nouns than the Italian-Dutch bilinguals, but on the other hand, they seem to make more errors with common nouns compared to the Italian-Dutch bilinguals. This indicates a qualitative difference in the acquisition of gender between the two bilingual populations. It is possible that a 2-way gender system is not enough to enhance gender awareness. However, it is not clear whether this is simply an age effect (given that the bilinguals of the present study were considerably older than the bilinguals in Algoe’s (2011) study) or whether this difference is caused by the nature of the gender system in the L1 or differences related to input factors (i.e exposure patterns). A more systematic approach is needed where the two populations are matched for age and/or input to determine the source of these differences. Nevertheless, given that overgeneralization of het is an error profile that (almost) never occurs in monolinguals and is only rarely reported in studies on the bilingual acquisition of Dutch gender, the present findings strongly suggest that the way gender is marked in a bilinguals’ (other) L1 has an effect on the acquisition of Dutch gender. The second research questions concerned the relationship between determiner-noun and adjective-noun agreement. A comparison between the accuracy scores on the two types of agreement suggests that not only do the Greek-Dutch bilinguals show more or less parallel development for determiner-noun and adjective-noun agreement, but some of the groups also reach similar accuracy scores for the two types of agreement (i.e. groups B and C). For common nouns, group A performs better on 62 MA Thesis 2013 Evelyn Egger determiner-noun agreement than on adjectival agreement, while for group D the pattern is reversed showing better performance on adjectival agreement than on determiner-noun agreement. This suggests that during the initial stages of gender acquisition in Dutch (group A in the present study), children have more problems with the syntactic aspects of gender agreement than with the lexical part, while at later stages (group D in the present study) rules for syntactic agreement are in place, but children are still learning which nouns take het and which take de. Alternatively, it could be argued that adjectival agreement is more complex than determiner-noun agreement (see Silveira, 2011) since it requires more syntactic computations and that this difference in complexity affects performance during initial stages of acquisition (group A), but not at more advanced levels (group D). For neuter nouns, both group A and D show higher accuracy scores for adjective-noun agreement than for determiner-noun agreement. Thus, for group B and C (and possibly also D) it can be argued that the results provide evidence that these children make use of abstract gender representations (i.e. features), while this is not clear for group A since there is improvement from group A to B for the use of het and adjectives with common nouns, but at the same time, there is a drop in accuracy rates for de and bare adjectives. Generally speaking, group A poses a bit of a problem since they performed better on neuter nouns than group B on adjectival agreement in the PDT and on determiner-noun agreement in the Story task, but not in the PDT. As such it is not clear whether group A is indeed at a less advanced stage in the acquisition of Dutch gender than group B as suggested by their vocabulary scores. The relatively high accuracy scores of group A indicate that for this group, some other factor than vocabulary score may be a better indicator of their performance with neuter nouns. A closer look at the input variables for the children in group A (see Appendix A) reveals that for 3 of the 6 children in group A, Dutch seems to be the dominant language (since it is the language they speak with their siblings) and they scored low on the Dutch vocabulary test (but relatively high on the gender tasks) in spite of having had long and extensive exposure to Dutch. As to the Greek results, a comparison between accuracy scores on the Det-N task and the DetAdj-N task reveals that children did better with determiner-noun agreement than with determineradjective-noun agreement. However, these differences were very small, with the exception of group A, who scored considerably higher on the Det-N than on the Det-Adj-N task for feminine nouns (50% vs. 30%). The results for masculine and neuter nouns are thus in line with findings from Mastropavlou (2006) for Greek monolinguals, who also showed no notable difference in accuracy scores between the determiner-noun and adjectival agreement. On the other hand, the higher scores for determiner-noun agreement in the feminine condition by group A suggest that children have more problems with adjectival 63 MA Thesis 2013 Evelyn Egger agreement than with determiner-noun agreement 31. Similar findings have been reported for L2 learners of Greek (Tsimpli, 2003), who performed significantly better on determiner-noun agreement than on adjectival agreement. There are several explanations for children’s better performance on the Det-N task. One the one hand, it can be argued that determiner-adjective-noun agreement requires more syntactic operations than determiner-noun agreement. On the other hand, it is also the case that determiners are generally more frequent in the input and show more phonological regularities than adjectives (Silveira, 2011). Nevertheless, the parallel improvement of the two types of agreement across groups indicates that children make use of gender cues to establish abstract representations of the gender paradigm in Greek. In addition, the finding that accuracy on the two types of agreement seems to improve in parallel as a function of increasing vocabulary scores suggests that vocabulary score is a good indicator of performance on gender agreement in Greek. Turning to the question about the error types the Greek-Dutch bilinguals commit in determineradjective-noun agreement, results showed that children almost always made the determiner agree with the adjective, but not necessarily the determiner with the noun. Moreover, when children failed to establish correct gender agreement, errors of the type *Det *Adj (including *Det *Adj *N for masculine nouns) were by far the most frequent type regardless of group and gender condition which is in line with findings from Mastropavlou (2006) for Greek monolinguals with and without SLI. In contrast, children hardly ever produced determiner-adjective-noun strings where only one element was marked for the wrong gender (i.e. error types *Det and *Adj). This pattern could be explained by claiming that when children do not know the gender of a noun, they automatically fall back on the neuter default. The use of a default gender has been argued to occur when no agreement can be established and appears to involve less or different syntactic operations (Hulk & Tsimpli, to appear). In contrast, the Italian-Dutch bilinguals in Canta (2011) made very few errors that involved both an incorrect determiner and an incorrect adjective. Overall, the Italian-Dutch bilinguals made less errors than the Greek-Dutch bilinguals suggesting that they are at a more advanced stage in the acquisition of gender, where they know the gender of the respective nouns, however, they make occasional errors in maintaining gender agreement on adjectives. It is also possible that the different error patterns of the two populations are due to structural differences of the respective languages. In Italian, adjectives typically occur in post-nominal positions, while in Greek they are always prenominal. Thus, adjectives in Italian are often not adjacent to determiners which appears to cause children more difficulties to maintain gender concord. Interestingly, Mastropavlou (2006) found that Greek monolingual children with SLI also occasionally produce determiner-adjective-noun strings that contain a mismatch between the adjective on the one hand and the 31 It is likely that the lack of such a relatively large difference between the two tasks for masculine nouns is due to the fact that many children in group A did not seem to have established masculine gender in their grammar yet, since they did not produce any masculine determiners or adjectives in the masculine form at all. 64 MA Thesis 2013 Evelyn Egger determiner and the noun on the other (i.e. *Adj). Crucially though, all the instances of incorrect gender marking on the adjective were produced by one child and all involved an incorrect word order, in that the child put the adjectives in the post-nominal position (i.e. determiner-noun-adjective instead of determineradjective-noun). Moreover, Roulet-Amiot & Jakubowicz (2006) also found that French typically developing and language-impaired monolinguals produced more agreement errors with DPs that contained post-nominal adjectives compared to phrases containing prenominal adjectives. Thus, the present results strongly suggest that the difference in types of agreement errors between the Italian-Dutch bilinguals in Canta (2012) and the Greek-Dutch bilinguals of the present study are due to structural differences between Italian and Greek. The fifth research question asked whether the bilinguals of the present study show the same performance patterns across gender conditions as previously attested in the respective monolinguals and other bilingual populations. For Greek, the answer is an unequivocal yes, children in all groups are at ceiling with neuter nouns and perform worse with masculine nouns, while accuracy rates for feminine nouns are inbetween (i.e. neut>fem>masc). Moreover, when children make errors in gender agreement with feminine and on masculine nouns, they always fall back on neuter (i.e. they almost never use feminine instead of masculine or masculine instead feminine), which is in line with previous findings (e.g. Mastropavlou, 2006; Unsworth et al., 2011b) and supports the claim that neuter is the default gender in Greek (Tsimpli & Hulk, to appear). As to the Dutch tests, here the bilinguals of the present study seem to perform differently than monolinguals and other bilinguals in that they showed a relatively strong tendency to overgeneralize het to common nouns, which is an error that has hardly been found in other bilingual populations tested so far. As already argued above, this qualitative difference between the Greek-Dutch bilinguals on the one hand and the Italian-Dutch bilinguals in Canta (2012) and the EnglishDutch bilinguals in Unsworth (2013a) on the other, is taken as evidence for cross-linguistic influence from Greek to Dutch. The last research question concerned the relative performance on the judgment and production tasks for the two genders. For Dutch, results showed that there is an asymmetry on children’s performance on the two tasks and for the two gender conditions. For neuter nouns, all groups performed better on judgment than on production, while the reverse was found for common nouns (i.e. production was better than judgment). This two-fold asymmetry is in line with findings for monolinguals (Unsworth & Hulk, 2010a) and English-Dutch bilinguals (Unsworth, 2013a), however, it is not clear why this pattern holds for both the advanced and less advanced groups alike. On the assumption that knowledge (i.e. judgment) generally precedes production, it could be argued that the children’s higher accuracy rate for production with common nouns reflects the use de as a default, rather than explicit knowledge of which nouns take the common determiner. In other words, since production is assumed to be more difficult than 65 MA Thesis 2013 Evelyn Egger judgment, children are more likely to fall back on the default-use of de in production than in judgment. Similarly, in the Greek tests, children also did better on production than on judgment for the ‘default’ gender (i.e. neuter) 32, but accuracy was lower on production than on judgment for masculine and feminine nouns. Thus, it seems that better performance on production than on judgment is a characteristic of ‘default genders’ (i.e. common in Dutch and neuter in Greek), but not of the ‘more difficult’ gender classes in a given language (i.e. neuter in Dutch and masculine and feminine in Greek). In sum, the findings of the present study provide evidence for cross-linguistic influence from Greek to Dutch, as indicated by the qualitative differences between the Greek-Dutch bilinguals on the one hand, and the Italian-Dutch (Canta, 2012) and English-Dutch bilinguals (Unsworth, 2013a) on the other. In contrast, performance patterns in the Greek tests were highly similar to those that have been attested for monolinguals (Mastropavlou, 2006) and English-Greek bilinguals (Unsworth et al., 2011) suggesting that there is no cross-linguistic from Dutch to Greek. Having discussed the results of the present study in light of previous findings for the respective monolingual children and other bilingual populations, the next section will try to point out the relevance of the present findings for language-impaired bilingual children. 4.2 Relevance & implications for the study of bilinguals SLI Mastropavlou (2006) investigated gender agreement in a group of Greek monolingual children with SLI aged 4;2 to 5;9. Results showed that the language-impaired children did not have significant problems with determiner-noun agreement (accuracy rates were above 92% for all three genders). However, similar to the bilinguals of the present study, the children with SLI committed more agreement errors with masculine and feminine nouns in the Det-Adj-N task than in the Det-N task (7.3% vs. 17.5% for masculine and 0% vs. 8.3% for feminine). Moreover, there was no statistically significant difference in accuracy scores between the language-impaired children and the monolingual TD controls that were matched on language development (LD controls) 33 for the Det-N task. In contrast, results for the Det-AdjN task indicated that the LD controls performed slightly better than the children with SLI (93% vs. 83% for masculine, 98% vs. 93% for feminine and 100% vs. 100% for neuter), although the differences did not reach significance. Overall, children with SLI did not seem to have serious difficulties neither with gender marking nor with gender agreement and performed high on all gender tasks. 32 33 Although the difference between judgment and production was much smaller than for the common nouns of the Dutch tests. Language development was determined by children’s scores on the morpho-syntactic sections of the DVIQ. 66 MA Thesis 2013 Evelyn Egger Keij, Cornips, van Hout, Hulk & van Emmerik (2012) investigated the acquisition of Dutch definite determiners for both production and knowledge in a group of monolingual children with SLI, a monolingual control group 34 and a group of bilinguals of the same age range as the TD controls. Moreover, each of the three groups was divided into two age levels (level I and level II) in order to examine the developmental paths of the three groups. For the production task, all groups performed better with common nouns than with neuter nouns and the monolingual controls performed significantly better than both the bilinguals and the SLI group, but there was no significant difference between the bilingual group and the SLI group. However, there was a qualitative difference between the bilinguals and the SLI group in that the bilinguals performed better at the higher age level for neuter nouns, whereas there was no overall development for the SLI group whose scores did not improve across the two age levels for either gender. As to the knowledge task (grammaticality judgment task), the monolinguals performed again better than the bilinguals and the children with SLI, but no difference was found between the bilinguals and the SLI group. The monolingual controls performed better on common than on neuter nouns and there was improvement in accuracy scores across the two age levels. In contrast, there was no effect of gender for the bilinguals and the younger group (level I) performed better than the older group (level II) (83% vs. 62% for common and 82% vs. 65% for neuter nouns). Finally, for the SLI group no significant effects were found for either gender or age level. The children with SLI at the younger age level reached similar accuracy scores for neuter and common nouns, while at level II they did better with common than with neuter nouns. For common nouns, accuracy rates improved from level I to level II, but dropped for neuter nouns. A comparison between performance on the production and the knowledge task revealed that the bilinguals showed the same pattern as previously attested in Unsworth & Hulk (2010a) and Unsworth (2013a) as well as the bilinguals of the present study, namely better production than judgment for common nouns and better judgment than production for neuter nouns. The monolingual control group did not show this pattern, instead performance was better on the knowledge task than the production task for both genders (although the differences were very small, i.e. between 1% and 6%). The younger SLI group (level I) patterned similar to the bilingual group (i.e. production>judgment for common and judgment>production for neuter) while the older SLI group did better on knowledge than on production for both genders as the monolingual controls. However, the analysis of variance revealed different effects for the bilingual and the SLI group suggesting not only a qualitative but also a quantitative difference between the two groups. To account for the findings, it was argued that the three groups are at different stages in the discovery of the Dutch gender paradigm. While the older monolinguals have discovered the Dutch gender paradigm, the bilinguals at level II are still in the process 34 The monolingual control group was two years younger, since previous research has indicated that the language-impaired children typically show a delay of around two years in terms of language development compared to their typically-developing monolingual peers (Keij et al., 2012). 67 MA Thesis 2013 Evelyn Egger of acquiring the paradigm, although they show increased awareness of Dutch grammatical gender compared to the younger bilinguals (level I). The children with SLI are still at an earlier stage, where they are aware of the gender distinction in Dutch, but unlike the bilinguals, they do not show any development in production or knowledge. Orgassa (2009) investigated the acquisition of determiner-noun and adjectival agreement in Dutch in Turkish-Dutch bilinguals with and without SLI and Dutch monolinguals with and without SLI. Turkish does not have grammatical gender instantiated, so cross-linguistic influence could be ruled out. Statistical analysis revealed both SLI effects (TD monolinguals > monolinguals with SLI and TD bilinguals > bilinguals with SLI) and L2 effects (TD monolinguals > TD bilinguals and monolinguals with SLI >bilinguals with SLI). In fact, it was questioned whether the two SLI groups had an abstract representation of gender at all since they produced only very few neuter definite determiners both with root noun and diminutives (especially the bilinguals with SLI). The bilingual group with SLI performed worse than all other groups which was taken as evidence for a cumulative effect of SLI and bilingualism. Similar results were obtained for adjectival inflection. In contrast to the above mentioned studies, some researchers have found that bilingual children with SLI can reach accuracy levels that are as high as those of their monolingual peers with SLI (e.g. Paradis, 2007). Unsworth & Hulk (2010b) argue that the different findings can be related to four factors, age of onset, societal context, cognitive maturity and the nature of the languages involved. For example, the subjects in Orgassa (2009) were all sequential bilinguals and the societal context was not necessarily ideal for successful dual-language development. Moreover, gender is absent in Turkish which might further hinder the discovery of gender distinctions in Dutch. Thus, more research is needed in order to determine the factors that facilitate positive cross-linguistic influence for both typically-developing and language-impaired bilinguals. Coming back to the findings of the present study, it was argued that Greek has a positive influence on the acquisition of Dutch gender in bilinguals. In addition, unlike language-impaired Dutch monolinguals, Greek monolinguals with SLI do not seem to have significant problems with gender agreement. Thus, it is possible that the positive cross-linguistic influence observed for the bilinguals of the present study also extends to Greek-Dutch bilinguals that have been diagnosed with SLI. Future research is needed to address this question and as well as systematic comparisons between the languageimpaired and the typically-developing monolinguals and bilinguals to examine whether there is an overlap between the error profiles of the various groups that could lead to potential misclassifications of bilinguals with and without SLI. 68 MA Thesis 2013 Evelyn Egger 5 Conclusions The present paper aimed at investigating the acquisition of grammatical gender in a group of Greek-Dutch bilingual children. A short overview on the notions of gender assignment and gender agreement was provided and the gender systems of the two languages under study were presented together with the relevant findings for the respective monolinguals and bilinguals. Six research questions were formulated that concerned the comparison of the Greek-Dutch bilinguals’ performance patterns across tasks and conditions as well as comparisons with other bilingual populations. On the basis of the findings it was argued that the nature of the gender system of the ‘other language’ and more specifically the way in which gender is marked has an effect on the bilingual acquisition of Dutch gender. The fact that Greek has a three-way gender system and the consistent and extensive marking of gender on all nominal elements, including the noun itself together with the abundance of reliable morphological cues leads to an enhanced awareness of grammatical gender in Greek children at early stages of their language development. This is reflected in the performance of the Greek-Dutch bilingual children of the present study and more specifically in their increased tendency to overgeneralize het to common nouns, which is an error pattern that has hardly been attested in other bilingual populations and is largely absent in Dutch monolinguals (at least during the early stages of gender acquisition). As a consequence, it was suggested that the facilitative effect of the Greek gender system on the acquisition of Dutch gender may extend to language-impaired bilingual populations, given that Greek monolingual children with SLI do not show significant problems with gender agreement in Greek. Although the findings of the present study point to a positive cross-linguistic influence from Greek to Dutch, the results should be interpreted with caution due to several limitations of the study. First, we did not test for statistical significance for the various differences between groups, tasks and conditions. Moreover, the children were matched on a relatively general measure of language proficiency (i.e. vocabulary score), but other variables that have been shown to be significant predictors of performance on the gender tasks such as amount of exposure were not taken into account. Nevertheless, the present study serves as a starting point for subsequent investigations of the acquisition of Dutch gender in Greek-Dutch bilingual children that can be extended to language-impaired populations as well. 69 MA Thesis 2013 Evelyn Egger References Algoe, S. (2011). Verwerving van het bepaald lidwoord en het adiectief in het Nederlands bij tweetalige kinderen. Research Report, University of Amsterdam. Blom, E., Polišenskà, D., & Weerman, F. (2008). Articles, adjectives and age of onset: The acquisition of Dutch grammatical gender. Second Language Research, 24, 297-332. Canta, E. (2012). “Il mucca nero”: a study on gender acquisition in Italian-Dutch bilingual children. MA Thesis, University of Turin. Carroll, S. E. (1989). Second-language acquisition and the computational paradigm. Language Learning, 39, 535-594. Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: The MIT Press. Clahsen, H. (1989). The grammatical characterization of developmental dysphasia. Linguistics, 27, 897920. Clahsen, H. (1991). Child Language and Developmental Dysphasia: Linguistic studies of the acquisition of German. Amsterdam: John Benjamins. Comrie, B. (1999). Grammatical gender systems: A linguist’s assessment. Journal of Psycholinguistic Research, 28, 457-466. Corbett, G.G. (1991). Gender. Cambridge: CUP. Cornips, L. & Hulk, A.C.J. (2006). External and internal factors in bilingual and bidialectal language development: Grammatical gender of the Dutch definite determiner. In: C. Lefebvre, L. White & C. Jourdan (Eds.), L2 Acquisition and Creole Genesis. Dialogues, pp. 355-378. Amsterdam: Benjamins. Cornips, L., & Hulk, A. (2008). Factors of success and failure in the acquisition of grammatical gender in Dutch. Second Language Research, 28, 267-296. Cornips,L., van der Hoeck, M., & Verwer, R. (2006). The acquisition of grammatical gender in bilingual child acquisition of Dutch (by older Moroccan and Turkish children): The definite determiner, attributive adjective and relative pronoun. In: B. Los & J. van de Weijer (Eds.), Linguistics in the Netherlands 2006, 40-51. Amsterdam: Benjamins. Deutsch, W., & Wijnen, F. (1985). The article’s noun and the noun’s article: Explorations into the representation and access of linguistic gender in Dutch. Linguistics, 23, 793-810. 70 MA Thesis 2013 Evelyn Egger Dewaele, J.-M., & Veronique, D. (2001). Gender assignment and gender agreement in advanced French interlanguage: a cross-sectional study. Bilingualism: Language and Cognition, 4, 275-297. Donaldson, B. C. (1987). Dutch reference grammar. Leiden: Martinus Nijhoff. Dunn, L. M., Dunn, L. M., & Schlichting, L. (2005). Peabody picture vocabulary test-III-NL. Amsterdam, The Netherlands: Pearson. Eichler, N., Jansen, V., & Müller, N. (2012). Gender acquisition in bilingual children: French-German, Italian-German, Spanish-German and Italian-French. To appear in: International Journal of Bilingualism. Franceschina, F. (2005). Fossilised second language grammars. The acquisition of grammatical gender. Amsterdam: John Benjamins. Gathercole, V. C. M., & Thomas, E. M. (2005). Minority language survival: Input factors influencing the acquisition of Welsh. In J. Cohen, K. McAlister, K. Rolstad & J. MacSwan (Eds.), Proceedings of the 4th international symposium on bilingualism. Somerville, MA: Cascadilla Press. Hockett, C.F. (1958). A course in Modern Linguistics. New York: Macmillan. Hulk, A.C.J. (2007). Deviance in early child bilingualism. In: J. Camacho, N. Flores-Ferrán, L. Sánchez, V. Déprez & M.J. Cabrera (Eds.), Romance Linguistics 2006. Selected papers from the 26th Linguistic Symposium on Romance Languages (LSRL). 177-198. Amsterdam: Benjamins. Hulk, A.C.J. & Cornips, L. (2006a). Neuter gender and interface vulnerability in child L2/2L1 Dutch. In: S. Unsworth, T. Parodi, A. Sorace & M. Young-Scholten (Eds.), Paths of Development in L1 and L2 acquisition, 107-134. Amsterdam: Benjamins. Hulk, A.C.J. & Cornips, L. (2006b). Between 2L1- and child L2 acquisition: an experimental study of bilingual Dutch. In: C. Lleó (Ed.), Interfaces in multilingualism: acquisition and representation. Hamburg Studies on Multilingualism, volume 4, pp. 115-138. Amsterdam: Benjamins. Keij, B., Cornips, L., van Hout, R., Hulk, A.C.J., & van Emmerik, J. (2012). Knowing versus producing: the acquisition of grammatical gender and the definite determiner in Dutch by L1-TD, L1-SLI, and eL2 children. Linguistic Approaches to Bilingualism, 2, 379-403 Kupisch, T., Müller, N., & Cantone, K. F. (2002). Gender in monolingual and bilingual first language acquisition. Lingue e Linguaggio, 1, 107-150. Kibort & Corbett (2008). Grammatical features. http://www.grammaticalfeatures.net 71 MA Thesis 2013 Evelyn Egger Lee-Williams, C., & Fernald, A. (2007). Young children learning Spanish make rapid use of grammatical gender in spoken word recognition. Psychological Science, 18, 193–98. Mastropavlou, M. (2006). The role of phonological salience and feature interpretability in the grammar of typically developing and language impaired children. Ph.D. dissertation, Aristotle University of Thessaloniki. Mastropavlou, M., & Tsimpli, I. (2011). The role of suffixes in grammatical gender assignment in Modern Greek: A psycholinguistic study. Journal of Greek Linguistics, 11, 27-55. Montrul, S., Foote, R., & Perpiñán, S. (2008). Gender agreement in adult second language speakers and heritage speakers: The effects of age and context of acquisition. Language Learning, 58, 503-553. Montrul, S., & Potowski, K. (2007). Command of gender agreement in school-aged Spanish-English bilingual children. International Journal of Bilingualism, 11, 301-328. Nicoladis, E., & Marchak, K. (2011). La Carte Blanc or laCarte Blanche? Bilingual children’s acquisition of French adjective agreement. Language Learning, 61, 734-758. Orgassa, A. (2009). Specific language impairment in a bilingual context. Ph.D. dissertation, University of Amsterdam. Paradis, J. (2005). Grammatical morphology in children learning English as a second language: Implications of similarities with Specific Language Impairment. Language, Speech and Hearing Services in the Schools, 36, 172–187. Pesetsky, D., & Torrego, E. (2007). The syntax of valuation and the interpretability of features. In: S. Karimi, V. Samiian & W. Wilkins (Eds.), Phrasal and clausal architecture, 262-294. Amsterdam: Benjamins. Polišenskà, D. (2010). Dutch children's acquisition of verbal and adjectival inflection. Ph.D. dissertation, University of Amsterdam. Ralli, A. (2002). The role of morphology in gender determination: Evidence from Modern Greek. Linguistics, 40, 519-551. Rodina, Y., & Westergaard, M. (2013). The acquisition of gender and declension class in a nontransparent system: Monolinguals and bilinguals. Studia Linguistica, 67, 47-67. Roodenburg J., & Hulk, A.C.J. (2008). Puzzles on grammatical gender. Lingue e linguaggio, 7, 67-92. Roulet-Amiot, L., & Jakubowicz, C. (2006). Production and perception of gender agreement in French SLI. Advances in Speech-Language Pathology, 8, 335-346. 72 MA Thesis 2013 Evelyn Egger Silveira, M. (2011). Specific language impairment revisited: evidencefromapsycholinguistic investigation of grammatical gender abilities in Brazilian Portuguese-speaking children. Ph.D. dissertation, University College London. Stavrakaki, S., & Tsimpli, I. (2000). Diagnostic verbal IQ test for school and preschool children: Standardization, statistical analysis, and psychometric properties. In Proceedings of the 8th conference of the panhellenic association of speech and language therapists. Athens: Ellinika Grammata. Stephany, U. (1995). The acquisition of Greek. Köln, Germany: Institut für Sprachwissenschaft Universität zu Köln. Tsimpli, I. (2003). Clitics and articles in L2 Greek. In J. M. Liceras, H. Zobl & H. Goodluck (eds.), Proceedings of the 6th Generative Approaches to Second Language Acquisition Conference (GASLA 2002), pp. 331−339. Somerville, MA: Cascadilla Proceedings Project. Tsimpli & Hulk (to appear). Grammatical gender and the notion of default: insights from language acquisition. Lingua. Unsworth, S. (2007). Age and input in early child bilingualism: The acquisition of grammatical gender in Dutch. In A. Belikova & L. Meroni (Eds.), Galana 2 -- Proceedings of the Conference on Generative Approaches to Language Acquisition North America 2, 448-458. Somerville, MA: Cascadilla Press. Unsworth, S. (2008). Age and input in the acquisition of grammatical gender in Dutch. Second Language Research, 24, 365-396. Unsworth, S. (2013a). Assessing the role of current and CUMULATIVE exposure in simultaneous bilingual acquisition: The case of Dutch gender. Bilingualism: Language and Cognition, 16, 86110. Unsworth, S. (2013b). Assessing age of onset effects in (early) child L2 acquisition. Language Acquisition, 20, 74-92. Unsworth, S., Argyri, F., Cornips, L., Hulk, A.C.J., Sorace, A. & Tsimpli, I. (2011a). Bilingual acquisition of Greek voice morphology and Dutch gender: What do they have in common?. In: N. Danis, K. Mesh & H. Sung (Eds.), Proceedings of the 35th annual Boston University Conference on Language Development, pp. 590-602. Cascadilla Press. 73 MA Thesis 2013 Evelyn Egger Unsworth, S., Argyri, F., Cornips, L., Hulk, A.C.J., Sorace, A. & Tsimpli, I. (2011b). On the role of age of onset input in early child bilingualism in Greek and Dutch. In: M. Pirvulescu, M.C. Cuervo, A.T. Pérez-Leroux, J. Steele & N. Strik (Eds.), Selected proceedings of the 4th conference on Generative Approaches to Language Acquisition North America (GALANA 2010), pp. 249-265. Cascadilla Press. Unsworth, S. & Hulk, A.C.J. (2010a). L1 acquisition of neuter gender in Dutch: Production and judgment. Proceedings of Generative Approaches to Language Acquisition 2009, 50-51. Unsworth, S. & Hulk, A.C.J. (2010b). Bilingualism as a kind of therapy? Commentary on Keynote article by J. Paradis. Applied Psycholinguistics, 31, 297-303. Unsworth, S., Hulk, A.C.J. & Marinis, T. (2011). Internal and external factors in child second language acquisition. Linguistic Approaches to Bilingualism, 1, 207-212. 74 APPENDICES APPENDIX A TABLE WITH INPUT VARIABLES EXTRACTED FROM QUESTIONNAIRES. Age of onset Greek Age of onset Dutch Length of Exposure Greek Length of Exposure Dutch Years living in Greece Years living in NL L1 Mother L1 Father Language between parents Language between siblings GR Farsi (NL) GR GR GR *GR (50%50%) GR NL NL GR GR GR NL GR/NL GR GR NL NL GR GR GR:20% NL:80% - GR GR/NL GR GR PL (ENG) GR ENG GR NL GR GR GR GR GR GR GR GR GR GR GR GR GR GR GR GR GR Nr Age DVIQ WBQ type of Bilingual 44 11;8 18 74 SUC Birth 4;0 11;8 7;8 - 12;8 GR 51 1 43 35 9;6 5;2 9;2 7;9 11 11 15 21 80 81 81 83 SIM SUC SUC SUC Birth Birth Birth Birth Birth 2;3 2;0 6;1 9;6 5;2 9;2 7;9 9;6 2;11 7;2 1;8 2;3 1;0 6;1 9;6 2;11 8;2 1;8 GR GR GR GR 47 11;4 9 84 SIM Birth Birth 11;4 11;4 - 11;4 NL 24 3 7;6 5;11 17 19 86 86 SIM SUC Birth Birth Birth 4;6 7;6 5;11 7;6 1;5 6;3 4;7 1;3 1;4 NL GR 59 25 11;4 7;3 16 8 90 91 SIM SIM Birth Birth Birth Birth 11;4 7;3 11;4 7;3 - 11;4 7;3 5 36 57 20 7 70 6;1 8;4 11;1 6;8 5;10 11;9 10 17 17 19 12 13 92 93 93 95 96 96 SIM SUC SUC SUC SUC SUC Birth Birth Birth Birth Birth Birth Birth 6;6 4;0 3;9 4;2 4;0 6;1 8;4 11;1 6;8 5;10 11;9 6;1 1;10 7;1 2;11 1;8 7;9 4;10 6;7 (GER) 3;0 3;9 4;2 - 1;3 1;9 8;1 2;11 1;8 11;9 GR/NL GR/NL GR:20% NL:80% GR/ENG GR GR/NL GR NL 75 *GR (50%50%) GR GR *GR NL NL GR NL NL GR/NL NL 42 13 40 67 8;11 5;8 8;8 12;11 6 7 12 12 97 97 99 99 SIM SIM SIM SIM Birth Birth Birth Birth Birth Birth Birth Birth 8;11 5;8 8;8 12;11 8;11 5;8 8;8 12;11 - 8;11 5;8 8;8 12;11 NL NL GR/NL NL 55 68 34 41 75 38 76 12;3 13;3 7;5 9;0 8;8 8;3 4;4 20 14 20 18 15 10 5 102 104 105 108 109 110 111 SUC SIM SIM SUC SIM SIM SIM Birth Birth Birth Birth Birth Birth Birth 4;0 Birth Birth 2;6 Birth Birth Birth 12;3 13;3 7;5 9;0 8;8 8;3 4;4 8;3 13;3 7;5 6;6 8;8 8;3 4;4 3;5 7;0 0;4 3;7 8;8 13;3 0;9 8;8 5;11 8;3 0;9 GR GR GR GR GR GR GR GR GR GR (ENG) GR NL NL GR (ENG) GR/NL GR/NL NL GR/NL NL NL NL 54 12;2 11 111 SIM Birth Birth 12;2 12;2 - 12;2 GR GR/NL NL GR NL NL NL *NL (NL:75% GR:25%) GR GR NL 9 5;11 9 119 SIM Birth Birth 5;11 5;11 - 5;11 NL NL NL 29 7;0 9 119 SIM Birth Birth 7;0 7;0 - 7;0 ENG GR/NL 58 10 61 11;3 5;11 11;5 6 8 19 120 126 132 SIM SIM SIM Birth Birth Birth Birth Birth Birth 11;3 5;11 11;5 11;3 5;11 11;5 - 11;3 5;11 11;5 GR *GR (50/50) NL GR GR PL (ENG) NL GR NL NL NL NL NL NL - The * in the columns L1 mother/father indicates that the parent is speaking to the child (at least some of the time) in a language that is not his/her mother tongue. 76 APPENDIX B DUCTH TESTS –LIST OF TARGET NOUNS (n=18) COMMON baby boom fiets telefoon sleutel klok gitaar helikopter robot baby tree bicycle telephone key clock guitar helicopter robot NEUTER huis bad raam konijn schaap vliegtuig hert net eiland house bath tub window rabbit sheep aeroplane deer net island 77 APPENDIX C GREEK TESTS – LIST OF TARGET NOUNS (n=18) MASCULINE skilos vatrachos ilios elefantas kuvas karcharias dog frog sun elephant bucket shark FEMININE papia bala gata zoni tileorasi vrisi duck ball cat belt television tap NEUTER vivlio milo tilefono molivi puli psari book apple telephone pencil bird fish 78
© Copyright 2026 Paperzz