Mass-Count and Article Choice in children with High Functioning Autism (HFA) and with Specific Language Impairment (SLI)

(1)

Mass-Count and Article Choice in children with High Functioning

Autism (HFA) and with Specific Language Impairment (SLI)

Master’s thesis, July 7th 2014

Name: Ava Creemers

Student number: 6063706

E-mail: avacreemers@gmail.com

Program: Linguistics (Research Master)

Department: Department of Language and Literature Faculty: Faculty of Humanities

University: University of Amsterdam

Supervisors: dr. J.C. (Jeannette) Schaeffer dr. J.E. (Judith) Rispens

(2)

CONTENTS

ACKNOWLEDGEMENTS iii

ABSTRACT iv

1. Introduction 1

2. HFA and SLI 3

2.1 Overlapping aetiologies? 5

2.2 Grammar and pragmatics 9

3. The two properties of the DP 10

3.1 The mass-count distinction 10

3.1.1 Previous research 15

3.2 The choice for a definite/indefinite article 16

3.2.1 The Concept of Non-Shared Assumptions 17

3.2.2 Scalar implicatures 19

3.2.3 Previous studies on article choice in SLI and HFA 21

3.3 Summary 23

4. Hypotheses and predictions 24

5. Participants 26

6. Experiment 1: Mass-count Quantity Judgement task 27

6.1 Methods and procedure 27

6.2 Materials and scoring 28

6.3 Results and discussion 29

6.3.1 Overall results 29

6.3.2 Individual analysis 30

6.3.3 Interim summary 33

7. Experiment 2: Article choice Elicited Production task 34

7.3.1 The indefinite conditions 38

7.3.2 The definite condition 40

7.3.4 Explaining een-overuse 43

8. Experiment 3: Article choice Felicity Judgement task 49

8.3.1 The indefinite conditions 54

8.3.2 The definite condition 55

8.3.3 Error analysis 56

9. Discussion and comparison of the subgroups 62

10. Conclusions 68

REFERENCES 71

APPENDIX A – TEST ITEMS IN THE MASS-COUNT QUANTITY JUDGEMENT TASK 76 APPENDIX B – RESULTS MASS-COUNT QUANTITY JUDGEMENT TASK 76 APPENDIX C – TEST ITEMS IN THE DEFINITENESS ELICITED PRODUCTION TASK 77 APPENDIX D – RESULTS DEFINITENESS ELICITED PRODUCTION TASK 79 APPENDIX E – TEST ITEMS IN THE DEFINITENESS FELICITY JUDGEMENT TASK 79

(3)

ACKNOWLEDGEMENTS

First and foremost I would like to thank Jeannette Schaeffer and Judith Rispens for supervising this master’s thesis – their support, helpful comments and critical questions have proven to be very useful in the course of writing this thesis.

I am very grateful to be given the opportunity to write my thesis within the project of Jeannette Schaeffer, which enabled me to work with a lot of data of different groups of participants. I would like to thank the student assistants who worked on Jeannette’s project, and who collected the data in the period before I started writing this thesis: Doatske de Haan, Sybren Spit, Kim Schoof, and Leanne Matimba. A special thank you goes out to Merel van Witteloostuijn for coordinating the project and for answering many e-mails with my questions regarding experiments and participants. I appreciate the effort of everyone who participated in the ‘leesclub’ connected to the project which was organised by Jeannette and Merel: Jorik Geutjes, Sybren Spit, Ana Žagar, Irene Rademaker, and Doatske de Haan.

I would like to thank Angeliek van Hout for her comprehensive explanations on Optimality Theory and scalar implicatures, and Jan Don and Hedde Zeijlstra for sharing their ideas on implicatures with me. I would also like to thank Jeannette Schaeffer and Jori Stam for reading and commenting on earlier versions of this thesis. I am furthermore indebted to Judith Rispens and Erik Meesters for helping me with the statistical parts of this thesis.

Finally, this research would not have been possible without all participants. I thank all of the children, their parents, and the staff at the schools for participating in this study.

(4)

ABSTRACT

This study investigates the relationship between HFA and SLI by looking at the nominal domain, which allows us to test both a grammatical and a pragmatic property: the mass-count distinction in so-called flexible nouns – being a purely grammatical phenomenon – and the choice between a definite and indefinite article – being driven by pragmatic principles. This was evaluated in 28 children with High Functioning Autism (HFA, mean age = 10;4), age and gender matched to 28 children with Specific Language Impairment (SLI, mean age = 10;2), and 28 Typically Developing controls (TD, mean age = 10;4). Mass-count is investigated using a Quantity Judgement task; article choice is investigated using an Elicited Production task and a Felicity Judgement task.

For the mass-count distinction, the results show that the SLI group as a whole performs significantly worse than the TD group, while the HFA group does not differ from the TD group. For article choice production, the results show that both the HFA and the SLI group perform worse than the TD group: both groups produce een-overuse (but no the-overuse), which is suggested to follow from the inability to always draw a scalar implicature. In the Felicity Judgement task the HFA group has an overly liberal interpretation of the indefinite article, which could also be caused by not drawing a scalar implicature. The results thus show that indeed mainly pragmatic knowledge is affected in autism, but the results do not support the claim that only grammatical knowledge is affected in SLI.

The results also provide evidence for the hypothesis that grammar and pragmatics are distinct domains: the data show a double dissociation, as there is an SLI subgroup that has impaired grammatical skills, but intact pragmatic skills; and a HFA subgroup that has impaired pragmatic skills, but intact grammatical skills.

Finally, the results do not provide evidence in favour of overlapping aetiologies between HFA and SLI. Although the data do provide evidence for a language impaired subgroup of children with HFA (of 25%), this subgroup is not similar to the SLI subgroup, when we compare accuracy scores on a nonword repetition (NWR) task between these groups. Also, the pragmatically impaired subgroup within SLI did not behave like the HFA subgroup: although the SLI group did not show a grammatical impairment in the nominal domain, the group did have very low scores on NWR and the CELF (testing a range of language abilities). It is concluded that the resemblance between the SLI and HFA subgroups is therefore only superficial, and does not show that the underlying causes are the same.

(5)

1. Introduction

An important current debate in different domains of science (psychology, neurology, cognitive science, genetics, and linguistics) concerns the relationship between Autism Spectrum Disorder (ASD) and Specific Language Impairment (SLI). The traditional view is that ASD and SLI are two unrelated disorders with distinct aetiologies: individuals diagnosed with SLI are argued to have a core deficit in their grammar, while the language problems that children with ASD encounter are argued to be secondary consequences of their pragmatic deficit. Similarly, pragmatic problems in individuals with SLI are argued to follow from their grammatical deficit, while pragmatic problems in individuals with ASD are argued to be their primary deficit. More recently, researchers have questioned this traditional view and have proposed an aetiological overlap between the two conditions, based on genetic linkage studies and on studies that show an above chance level of comorbidity between ASD and SLI. It has been shown that individuals with SLI do not only have an impairment specific to language, but that they also have problems with pragmatics that cannot be explained as secondary consequences (Bishop, 2000; Friedmann & Novogrodsky, 2011; Rapin & Allen, 1983). Additionally, it has been shown that the language development of individuals with ASD is not only delayed, but also deviant when compared to typically developing (TD) children (Kjelgaard & Tager-Flusberg, 2001; Perovic, Modyanova & Wexler, 2013).

Although recent literature seems to predominantly argue in favour of similar aetiologies for ASD and SLI, there is no consensus. Several studies have shown that there is a lack of substantive evidence in support of the claim that ASD and SLI have overlapping aetiologies (Lindgren, Folstein, Tomblin & Tager‐Flusberg, 2009; Taylor, Maybery & Whitehouse, 2012; Whitehouse, Barry, & Bishop, 2007, 2008;Williams, Botting & Boucher, 2008). Furthermore, there is no conclusive evidence on whether these the fact that there are children with SLI who experience pragmatic difficulties and children with HFA who experience grammatical difficulties is indicative of a common underlying genetic cause, or whether they indicate a superficial similarity. Moreover, most of the studies that compare SLI and ASD focus on cognitive and non-linguistic characteristics (i.e. working memory, Theory of Mind, emotion recognition, executive functioning and central coherence), and not so much on the differences or similarities in the linguistic development of individuals with ASD and SLI. When a study did take into account language abilities of children with ASD, they mostly focused on grammatical abilities, and not on pragmatic abilities.

(6)

The present study therefore investigates the performance of children with Specific Language Impairment (SLI) and children with High Functioning Autism (HFA) from a linguistic perspective. This study looks at the nominal domain, which allows us to test both grammatical and pragmatic skills: the mass-count distinction – being a purely grammatical phenomenon – and the choice between a definite and indefinite article – being driven by pragmatic principles.

The results of this study show, on the one hand, that there are similarities between the SLI and HFA groups: there is a small subgroup of children with SLI who perform well on mass-count and badly on article choice, and there is a subgroup of children with HFA who perform well on article choice, and badly on mass-count. However, it is argued that the resemblance between these SLI and HFA subgroups is only superficial: the groups do differ on the CELF (testing various aspects of language), and on a nonword repetition task (NWR, testing verbal working memory). Although the results of this study provide evidence for a language-impaired subgroup of HFA, this subgroup is not comparable to SLI, and similarly, the pragmatically impaired subgroup of SLI is not comparable to HFA. Therefore, the results do not provide evidence for overlapping aetiologies between the two conditions.

On the other hand, there appears to be a HFA subgroup that is only pragmatically, and not grammatically impaired (who perform badly on article choice but well on mass-count), and, vice versa, an SLI subgroup that is only grammatically, but not pragmatically impaired (who perform badly on mass-count but well on article choice). This double dissociation supports the claim that grammar and pragmatics can be impaired independently of each other, and hence provides evidence for the hypothesis that grammar and pragmatics are two distinct components of language.

This thesis is structured as follows. In section 2 I provide an overview of the relevant studies that have investigated the relationship between SLI and ASD. In section 3 I describe the two properties of the DP that will be investigated in this thesis: the mass-count distinction (§3.1) and article choice (§3.2), as well as an overview of previous acquisitional studies that investigated these two phenomena in TD children, children with HFA and children with SLI. In section 4 I will lay out my hypotheses and predictions, followed by an overview of the participants in the present study: a group of children with HFA, age and gender matched to a group of children with SLI and a group of TD children. Section 6 is concerned with the Quantity Judgement task that investigates the mass-count distinction; section 7 with the Elicited Production task that investigates article choice and section 8 with

(7)

the Felicity Judgement task that also investigates article choice. In section 9 an extensive discussion of the subgroups of the three experiments is given and section 10 concludes.

2. HFA and SLI

The two groups of participants that the present study is concerned with are children with High Functioning Autism, and children with Specific Language Impairment. High Functioning Autism (HFA) is a subgroup of the Autism Spectrum Disorders (ASDs); the spectrum illustrates the varying severity of the various forms of autism, with Autistic Disorder at the more severe end of the spectrum, through Asperger’s Disorder, Pervasive Developmental Disorder-Not Otherwise Specified (PDD-NOS), and HFA at the least severe end of the spectrum (Taylor, Maybery & Whitehouse, 2012). Individuals with HFA are higher functioning than other individuals with autism: they attend regular schools and all acquire functional use of verbal language. Autism is a developmental disorder, which is characterized by impairments in social interactions – such as being aware of other people’s feelings – and verbal and nonverbal communication, which must appear within the first three years of life (Kazdin, 2000).

Specific Language Impairment (SLI) is a developmental language disorder. There is a general consensus that it represents an impairment in language that is disproportionately greater than the impairments in other non-linguistic domains. Specific language impairment is defined as a combination of normal intelligence (an IQ equal to or greater than 85) and impaired performance on a battery of standardised language assessments (a composite language measure of more than 1.25 SD below the mean) (Leonard, 1998; Tomblin et al., 1997). Potential causes such as hearing loss, low general ability, or physical impairment of articulators are excluded. Autism is also excluded when making a diagnosis of SLI: the textbook example of a child with SLI is of a child with normal social interaction and nonverbal communication, but with specific difficulties in mastering structural aspects of language, especially morphosyntax (Bishop, 2010).

This latter point shows that traditionally, ASD and SLI are regarded as distinct conditions with separate aetiologies: individuals with SLI are assumed to have a core deficit in their grammar, while the grammatical problems that children with ASD encounter are argued to be secondary consequences to their communicative problems. In the same line of reasoning, if individuals with SLI have pragmatic problems as well, these are argued to follow from their grammatical problems, while pragmatic or more general social interaction

(8)

problems in individuals with ASD are argued to be their primary deficit. Indeed, many studies have shown that individuals with SLI are characterized by a primary deficit of the grammatical system: they have shown to be morphosyntactically impaired in the comprehension and expression of language. This characterization seems to hold especially for the so-called “Grammatical SLI” or G-SLI subgroup of SLI: it is shown that this subgroup is even more characterized by a persistent and disproportionate impairment in the grammatical comprehension and expression of language, when compared to those individuals with SLI who do not have G-SLI. Van der Lely & Ullman (2001) for instance show that the grammatical computations that underlie the regular past tense formation are impaired in the G-SLI grammar, Van der Lely et al. (2011) show that children with G-SLI are unable to consistently compute the syntactic dependencies between a moved element and its trace in Wh-questions, and Van der Lely & Stollwerck (1997) argue that children with SLI have an underspecified syntactic representation with respect to co-indexation between constituents (characterized by Binding Theory).

In addition, the social and pragmatic skills of children with SLI are usually assumed to be intact. Van der Lely (1997) for example investigated language abilities on inflectional morphology and syntax, and pragmatic abilities (such as pragmatic awareness, pragmatic inference, and verbal reasoning) of a 10-year-old boy with SLI, and found a clear dissociation between the boy’s severely impaired modular language abilities, and his intact pragmatic abilities.

Individuals with ASD often have problems in the area of language: between 20% (Walenski et al. 2006) and 50% (Rapin & Dunn, 2003) never acquire any functional use of verbal language. It is said that those who do acquire verbal language have difficulties in the pragmatic use of the language, whereas the grammatical aspects of their language are not disproportionally impaired. The pragmatic impairments in children with ASD range from difficulties with the interpretation of nonverbal communicative gestures (e.g. Mundy, Sigman & Kasari, 1990), conversational discourse (Paul & Cohen, 1984), pragmatic aspects of prosody (Shriberg et al., 2001), and the interpretation of non-literal language such as idioms, metaphors, jokes, and lies (MacKay & Shaw, 2004). These pragmatic impairments are widespread and are found in both children and adults, with diagnoses ranging from Autistic Disorder to High Functioning Autism (HFA), the subgroup of ASD that is investigated in the present study. Volden (2004) for example replicated the study by Paul & Cohen (1984) for conversational discourse and found that also high functioning school-age children with autism

(9)

Recently, this neat divide between the two disorders has been questioned for several reasons. Researchers have proposed that the two conditions overlap and that they show a closer link than was previously assumed. I will address the arguments for this position in the next section.

2.1 Overlapping aetiologies?

There are two main lines of evidence in favour of overlap between the two conditions: (i) an above chance level of comorbidity between the two conditions, and (ii) shared genetic risk factors. The latter line of evidence comes from genetic linkage studies, which suggest that autism and SLI may involve one or more shared genes (Fombonne, Bolton, Prior, Jordan & Rutter, 1997; Folstein & Mankoski; 2000; Tomblin & Zhang, 1999; Tomblin, Hafeman & O-Brien, 2003; Vernes et al., 2008; a.o.). For the present purpose, I will focus on comorbidity in this section, and not on the non-linguistic evidence from genetic linkage studies.

With respect to comorbidity, it has been shown – in contrast to the traditional view – that children and adolescents with SLI are not only grammatically impaired, but also show pragmatic impairments reminiscent of autism1

. Many different names have been given to these pragmatically impaired subgroups of SLI: Rapin & Allen (1983) coin the term ‘Semantic Pragmatic Deficit’; Bishop (1997) proposed the term ‘Pragmatic Language Impairment’ (PLI), and Friedmann & Novogrodsky (2011) propose ‘PraSLI’ (pragmatic SLI). Bishop (2000) argues against the idea that pragmatic difficulties in SLI are secondary to the oral language difficulties, and shows that it is hard to explain all of these pragmatic problems in SLI as secondary consequences.Friedmann & Novogrodsky (2011) show that the subgroup of children with SLI they termed ‘PraSLI’ has difficulties with pragmatics, but not in other domains of language.

1_{Also, recent research has shown that children diagnosed with SLI have several cognitive impairments. This has}

challenged the claim that SLI is specific to language. Lum, Conti-Ramsden, Page & Ullman (2012) for example

examined measures of working, declarative and procedural memory in 51 children with SLI and 51 matched typically developing (TD) children (mean age 10), and found that when compared to the TD children, the children with SLI were impaired at procedural memory and verbal working memory. Lum, Conti-Ramsden & Lindell (2007) presented 26 adolescents with SLI and 14 controls with a rapid serial visual presentation task. The SLI group was found to be significantly less accurate in the processing of rapid stimulus sequences. Henry, Messer & Nash (2012) assessed 41 children with SLI and 88 TD children on ten separate measures of executive functioning (higher ordering thinking and reasoning skills). They found that children with SLI had significantly lower performance than typical children on 6 of the 10 EF tasks once age and nonverbal IQ had been controlled. Finally, Evans, Saffran & Robe-Torres (2012) examined 35 children with SLI (aged 6;5–14;4) and 78 TD children (aged 5;7–12;10) and found that the children with SLI had difficulties in implicitly computing the probabilities of adjacent sound sequences (statistical learning), while the TD children had not.

(10)

In addition, the claim that children with ASD are only pragmatically impaired has been challenged. It is known that language impairment is a central feature of autistic disorder, which led researchers in the 1970’s to focus on language as the central feature of autism (Bishop, 2003). Over the years this view changed and the grammatical problems in autism were seen as secondary consequences of the limited communication skills. Yet, recently it is again suggested that the grammatical skills of a subgroup of children with autism might not only be delayed, but really deviant. Perovic, Modyanova & Wexler (2013) for example compared the knowledge of binding principles in children with autism with language impairment (ALI), children with autism with normal language (ALN) and TD children. They found that the interpretation of reflexives the ALI subgroup was significantly worse than that of both the TD control group and the ALN subgroup. The comprehension of personal pronouns was delayed in all groups. Because the interpretation of reflexives (Binding Principe A) relies on purely syntactic constraints, the authors conclude that the grammatical development in a subgroup of children with autism is not only delayed but really deviant: ‘The observed pattern is in contrast to the traditional view that there is a pragmatic, but not a syntactic deficit in autism’.

With these findings, Perovic et al. (2013) provide additional evidence for the claim by Kjelgaard & Tager-Flusberg (2001) that there is a significant heterogeneity in the language skills of children with autism. Kjelgaard & Tager-Flusberg investigated language functioning in a group of 89 children diagnosed with autism, using a broad range of standardised measures of language (phonological, lexical, and higher order semantic and syntactic measures). Their main finding is that language functioning in autism is very variable: at one end, there are children with autism whose language skills are within normal range, while for other children the language skills are significantly below age expectations and meet criteria for SLI. Semantic and syntactic knowledge was especially impaired in the language impaired ASD group, while vocabulary tends to be relatively less impaired. They conclude that their findings provide evidence for a theoretically significant overlap between SLI and ASD, and that a subgroup of children with autism may have a parallel or overlapping SLI disorder. In a follow-up study by Roberts, Rice & Tager-Flusberg (2004), it was found that indeed children with ASD and language impairment made tense marking errors that were similar to those of children with SLI.

The aforementioned observations led researchers to postulate that there exists an area of overlap between ASD and SLI, which means that the boundaries between SLI and ASD

(11)

form a continuum, which would explain the phenotypic and aetiological overlap between the two disorders. Bishop & Norbury (2002) provide a first attempt to investigate the proposed overlap between SLI and ASD. They applied a number of standard instruments for the diagnosis of autism to a group of children diagnosed with Pragmatic Language Impairment (PLI) and SLI, and found that some children with PLI showed autistic features, but that also a subset of the children did not show autistic features. Moreover, they found that in the group of children with SLI, some children presented some or all of the features typical of autism. A similar but more extensive study was done by Conti-Ramsden, Simkin & Botting (2006). They examined the presence of ASD in 76 14-year-olds with a confirmed history of SLI using a variety of instruments, and found that a proportion of the children with SLI also met criteria for a diagnosis of autism (3.9%), and a considerable proportion met criteria for autism in at least on of the two instruments used (26.4%). They conclude that their study points to a clear overlap between SLI and autism.

Although there is an accumulating amount of studies that argue in favour of an aetiological overlap between ASD and SLI, there are also some recent studies that provide evidence that the aetiological distinction between ASD and SLI should be maintained. One of the studies that argues for the maintenance of the aetiological distinction between SLI and ASD is the literature overview by Taylor, Maybery & Whitehouse (2012). They review existing literature to find out whether individuals with SLI exhibit cognitive characteristics reminiscent of autism: Theory of Mind (the ability of individuals to attribute mental states to oneself and others), emotion recognition, executive functioning (i.e. planning, inhibition, impulse control, cognitive and behavioural flexibility) and central coherence (the ability to understand context or to ‘see the big picture’). Taylor et al. conclude that the available research reveals that findings are inconsistent and frequently conflicting, and that therefore there is little evidence supporting phenotypic overlap between children with ASD and SLI.

Another study is provided by Lindgren, Folstein, Tomblin & Tager‐Flusberg (2009), who examine IQ, and language and reading abilities of 52 children with ASD and 36 children with SLI and their first-degree relatives (siblings and parents). The ASD group was divided into a group with language impairment (ALI) and a group without language impairment (ALN). The authors found that the children with ALI and SLI performed similarly on most measures while the children with ALN scored higher. When comparing the relatives, it was found that the ALI relatives scored higher than the SLI relatives on all the language and reading measures. The authors conclude that since the children with ALI and SLI performed

(12)

similarly while the ALI relatives performed better than the SLI relatives, the language deficits in these populations cannot be linked to an identical genetic aetiology. A similar conclusion was reached by Whitehouse, Barry, & Bishop (2007) who showed that while structural language impairments are part of the heritable communication deficit in SLI, this is not the case in autism.

Williams, Botting & Boucher (2008) review evidence on the behavioural, neurobiological, and etiological level, and also arrive at the conclusion that there is little evidence that structural language impairments in ASD-LI (ASD-Language Impaired) can be explained in terms of shared underlying cause as SLI: although children with ASD-LI made errors in using inflectional endings on verbs, the types of errors differed from those seen in SLI. The authors conclude that ‘conceptualizing ASD-LI and SLI as (partially) overlapping disorders and, hence, concentrating efforts on finding similarities between the two, may result in us missing more about each disorder than we explain’ (2008: 960). Similarly, Whitehouse, Barry, & Bishop (2008) found that the pattern of errors on a nonword repetition task differed between children with SLI and children with ASD. These studies suggest that although the phenotype of ASD may lead to language impairment, the resemblance with specific language impairment is only superficial.

We see that the evidence for aetiological overlap between ASD and SLI is inconsistent: on the one hand we see similar profiles in terms of grammatical difficulties and pragmatic difficulties, but on the other hand we see qualitative differences between the errors made and relatively spared language abilities in relatives of children with ASD. Many of the studies that investigated the relationship between ASD and SLI have focused on cognitive abilities, rather than linguistic abilities. The studies that have investigated linguistic skills have only focused on grammatical skills, but not on pragmatic skills. We believe it is important to study both grammar and pragmatics in order to be able to see whether there are true subgroups within SLI that behave like ASD, and whether there are true subgroups within ASD that behave like SLI. For that reason, the present study investigates the relationship between HFA and SLI from a linguistic perspective, distinguishing between grammar and pragmatics. The domain chosen to investigate these two properties is the nominal domain, in which we distinguish a grammatical property in the form of the mass-count distinction, and a pragmatic property in the form of article choice. Before the mass-count phenomenon and the choice for a definite or indefinite article is elaborated on in Section 3, I will first briefly spend a word on what is exactly meant with ‘grammar’ and ‘pragmatics’.

(13)

2.2 Grammar and pragmatics

In investigating the relationship between ASD and SLI, it is important to distinguish between the two domains of language that are traditionally argued to be impaired in one of the two conditions: pragmatics in ASD and grammar in SLI. When referred to ‘grammar’ in this thesis, we mean the structural aspect of language, i.e. morphosyntax. When referred to pragmatics, we mean the skills used to communicate with others, or in other words: the use of language in social contexts.

In investigating both pragmatics and grammar in children with SLI and HFA in the present study, we will also be able to see whether these two domains of language – grammar and pragmatics – really are distinct specific domains. An on-going debate in linguistics is whether language consists of autonomous domain-specific modules that operate independently, or whether language is non-modular in the sense that language results from non-linguistic and domain-general influences. The latter position is argued for by connectionist theories (e.g. Rumelhart & McClelland, 1982), which model language processing in a non-modular and holistic way. According to these theories, linguistic elements are connected with each other in an associative neural network, which is not specific to language. The former (domain-specific) position is argued for by for example Fodor (1983). He states that language consists of specific modules, and that a module is a domain-specific and informationally encapsulated brain system responsible for handling a particular type of mental representation.

Evidence in favour of a domain-specific view would come from the finding of a so-called double dissociation, a method borrowed from neuropsychology (Teuber, 1955). This method is used to show that two related mental processes function independently from each other: patient 1 has function A intact and function B impaired, while patient 2 shows the opposite pattern; function A is impaired and function B is intact. Such a double dissociation has been claimed to show that the brain is organized into specialized modules, which can be differentially damaged. With regard to grammar and pragmatics, this would mean that we have to find one group of individuals that has pragmatics intact, but grammar impaired, and another group of individuals that has grammar intact, but pragmatics impaired. More specifically: evidence for a double dissociation would come from a subgroup of HFA that performs well on mass-count, but badly on article choice, and a subgroup of SLI that performs well on article choice, but badly on mass-count.

(14)

(morphosyntactic) and pragmatic abilities in individuals with HFA and ASD: the mass-count distinction and the choice for a definite or indefinite article.

3. The two properties of the DP

The testing domain of the present study is the nominal domain, or more specifically the Determiner Phrase (DP). The two properties of the DP that will be investigated in this study are the distinction between mass and count nouns, and the choice for an indefinite or definite article. I will show, on the basis of the existing literature, that the former property is grammatical in nature, while the latter is pragmatic. In §3.1 I discuss the mass-count distinction and I give a brief overview of the literature. In §3.2 I discuss article choice.

3.1 The mass-count distinction

The mass-count distinction is the grammatical distinction between nouns that can be counted (e.g. dog(s), car(s), house(s)) and nouns that cannot be counted (e.g. dough, sand, water). In English and Dutch, count nouns can appear in both singular and plural forms, singular count nouns may be preceded by an indefinite article, and count nouns can be combined with cardinal numerals (e.g. one, two). Mass nouns cannot be pluralized, cannot be preceded by an indefinite article, and cannot be combined with cardinal numerals2

. Also, quantifying expressions are sensitive to mass, count, and plurality properties of the nouns they combine with. Quantifiers implicating plurality, such as many, several, and different in English, and een paar ‘a few’, and verschillende ‘different’ in Dutch, can only combine with plural count nouns, but not with singular count nouns or mass nouns. Quantifiers such as each and every, and elke ‘each/every’ and iedere ‘each/every’ in Dutch can only combine with singular count nouns. Moreover, a small set of quantifying expression can only be combined with mass nouns: much, a bit and (a) little in English and een (klein) beetje ‘a (little) bit’ in Dutch.

Yet, the mass-count distinction is not as clear-cut as presented above, as some mass nouns can, under certain conditions and in certain contexts, be used in count syntax and some count nouns can be used in mass syntax. Examples of mass-to-count-shifts are given in (1); examples of count-to-mass-shifts in (2).

2_{To count or quantize a mass noun denotation we need a classifier phrase (like piece of or truckload of) or a}

measure phrase (like litres of or tons of). Dutch slightly differs from English in this respect, as in English a pseudo-partitive construction is used (two bowls of rice), while in Dutch a structure is used that does not include

(15)

(1) a. territoriale water-en territorial-INFL water-PL ‘territorial waters’ b. three wines c. twee biertjes two beer-DIM ‘two beers’

(2) a. Er zit hond in de soep

There is dog in the soup (De Belder, 2011)

b. Johnny is very choosy about his food. He will eat book, but he won’t touch shelf. (example from Gleason 1965, cited by Doetjes 2013)

c. There’s not enough table for everyone to sit at (Allan, 1980: 547)

In (1a) the Dutch mass noun water occurs in a plural form. This alters the meaning such that the noun receives a count interpretation. In (1b) we see that some mass nouns can, under certain conditions, be used with a numeral without a measure term. This type of construction is common in English as well as in Dutch for drinks such as beer and wine, when the speaker is referring to glasses or bottles of this drink.3

Finally, in (1c) we see another construction with a numeral, but here the noun takes the diminutive suffix. The diminutive suffix in Dutch only occurs with nouns that have a count interpretation (De Belder, 2011). In (2) it is shown that basically any count noun can easily be used with a mass interpretation, if provided the right context: even nouns that are predominantly used as count nouns such as dog, book, shelf and table can be used in mass contexts.4

The possibility of these mass-to-count and count-to-mass-shifts is one of the main pieces of evidence in favour of a syntactic or contextual analysis of the mass-count distinction. Previous approaches can roughly be classified as ontological approaches (Cheng, 1973; Quine, 1960), and as a conceptual-semantic approaches (Bloom, 1994; Chierchia, 1998). Ontological theories assume that the mass-count distinction is rooted in pre-linguistic ontological knowledge and is a distinction operating between real-world entities5

. The ontological view thus assumes that properties in the real, physical world determine whether a noun can be counted or not. According to the conceptual-semantic view, the distinction

3_{Note that Dutch differs from English at this point, because in English a mass noun combined with a numeral}

behaves syntactically as a count noun too, taking plural morphology (three wines). In Dutch the mass noun, when combined with a numeral, does not always take plural morphology (Doetjes, 2013): drie wijn.

4_{Pelletier (1975:6) has illustrated this idea by the concept of the Universal Grinder. Mass-to-count shifts have}

been illustrated by a the idea of a Universal Sorter by Bunt (1985).

5_{The ontological view is mainly based on two conceptual properties of the mass-count distinction: cumulative}

reference and divisity of reference. The conceptual-semantic phenomenon is mainly based on the idea of individuation.

(16)

between count and mass is not primarily connected to the world outside, but to the way language users perceive and conceptualize that world. In his semantic approach, Chierchia (1998) claims that the difference between mass and count nouns comes from the lexical entries: count nouns are associated with individual atoms, whereas mass nouns do not single out a set of atoms, but a whole sub lattice (p. 68). Mass nouns are thus inherently plural: they come out of the lexicon already pluralized6_.

These ontological and conceptual-semantic approaches fail to account for mass-to-count-shifts as in (1), count-to-mass-shifts as in (2), and ‘mass-count flexible terms’ as described by Barner & Snedeker (2005), and illustrated in (3).7

(3) Mass-count flexible nouns

a string / many strings some string a stone / many stones some stone a chocolate / many chocolates some chocolate a paper / many papers some paper

In (3), it is illustrated that nouns such as string, stone, chocolate, and paper can be use as either count nouns (left column) or as mass nouns (right column), depending on the context in which they occur. Following Barner & Snedeker (2005), we will refer to these nouns as flexible nouns. Barner & Snedeker investigated the effect of shifting mass-count syntax when particular terms and referential contexts are held constant (e.g. a string /some string), and showed that both children and adults based quantity judgments on number when the terms were used with count syntax, but on total amount of stuff (volume) when used with mass syntax. This provides evidence that for flexible nouns it is the linguistic context rather than world knowledge, which determines speakers’ judgments of what is mass and what is count.

The main problem for ontological and conceptual-semantic approaches is how mass-count flexible terms like string and stone are represented8

. In contrast, contextual approaches

6_{According to Chierchia, this explains the distributional properties of mass nouns: they cannot be pluralized}

because they are already plural; and they cannot be directly preceded by numerals because mass nouns do not correspond to sets of atoms. As we need to individuate (in sets of atoms) to be able to count, mass nouns do not provide a suitable counting criterion.

7_{Chierchia does acknowledge the observation that there are so-called flexible nouns, and for these he proposes a}

type shifting operation called MASS, which turns a count noun into a mass noun by neutralizing the singular-plural distinction. How this would work for mass nouns that get a count interpretation (e.g. a beer) is not clear. It is also not clear from Chierchia’s proposal where the type shifting operation takes place.

8_{These approaches do face many other problems, which will not be fully addressed here due to space}

limitations: for instance problems with divisity of reference (Barner & Snedeker, 2005; Gillon, 1996; Grimm, 2012). Also, if count and mass really were properties of real-world entities, then one would expect that all existing languages make the same choices with regard to mass-count for the same entities. This is not the case as

(17)

(Barner & Snedeker, 2005; De Belder, 2011; Borer, 2005) are perfectly able to account for these flexible terms, because flexible nouns show that the syntactic context of a noun is necessary to decide whether the noun is used as a mass or as a count noun. Borer (2005)9 offers a contextual categorization analysis of the mass-count distinction and argues that the mass-count distinction is not grammatically marked on the nouns themselves, but rather on the syntactic context within which they embed. This syntactic context is formed by plural inflection or the indefinite article in English, which divide stuff (mass) into countable items, or, in other words, assign a division value. Borer further assumes that the default interpretation in the mass-count distinction is mass: all nouns are ‘born’ unspecified for any properties, and as a default, unless more structure is provided in the form of an indefinite article or a plural suffix (in the case of English and Dutch), they will be interpreted as mass. The proposed structures of respectively count and mass nouns are given in (4) (Borer 2005:109-110).

(4) a. Syntactic structure of three cats b. Syntactic structure of much salt

DP DP

D #P D #P

<e>d

three much

<e># CLmax <e># LàN

salt cat.<div>

/cats/ <e>DIV LàN

cat

In (4a) we see the syntactic structures of two DPs, the former interpreted as a count noun and the latter interpreted as a mass noun. In both structures we see at the lowest point the lexical item which is, at the start of the derivation, categoriless (‘L’), and which is turned into a noun (N) because of the structural material heading the lexical item (LàN). In (4a) the Classifier Phrase (CLmax

), headed by CL°, gives rise to count interpretation: the plural inflection is the realization of an abstract feature which assigns range to the open value <e>DIV that heads the Classifier Phrase. In classifier languages, classifiers are situated here, and in English the plural morphology situated here, as they are both the realization of the abstract feature <div>. In (4a) this feature requires the movement of the L-head cat and causes it to be spelled out as

9_{Borer points at the complementary distribution that exists between classifiers in classifier languages on the one}

hand, and plural inflection and the indefinite article in Romance and Germanic languages on the other hand, and argues that plural inflection and the indefinite article in English are simply instantiations of the classifier function in classifier languages as Chinese.

(18)

/cats/. Dominating the Classifier Phrase is the Quantity Phrase #P, which is responsible for the assignment of quantity to stuff or to divisions of it. The structure in (4b) shows that the Classifier Phrase may be missing from the structure, and that the absence of CLmax

(the absence of dividing structure) within a DP structure gives rise to a mass interpretation. In this structure we also see that the default (mass) is associated with ‘less’ structure than the non-default interpretation (count).

Barner & Snedeker (2005) also view the mass-count distinction as a grammatical phenomenon driven by syntax, although they argue that so-called object-mass nouns such as furniture are lexically retrieved with a lexical feature in their denotation. In the present paper we will not focus on object-mass nouns, but only on flexible nouns. One can easily see how these flexible nouns fall into a contextual approach that assumes that it is the structure within which a particular noun embeds that determines the interpretation of the noun as either mass or count10

. The syntactic structures of many strings and much string according to the contextual categorization analysis are given in (5).

(5) a. Syntactic structure of many strings b. Syntactic structure of much string

DP DP

D #P D #P

<e>d

many much

<e># CLmax <e># LàN

string string.<div>

/strings/ <e>DIV LàN

string

De Belder (2011) applies the analysis by Borer to Dutch, and proposes a more extended structure in which she proposes a feature [Size]11

that is realized by the diminutive affix in Dutch. De Belder shows that in Dutch, the absence of the diminutive affix gives rise to kind readings, while its presence makes the unit reading the most salient one. It goes beyond the

10_{One could question why not all nouns are equally flexible between a mass and a count reading (Chierchia,}

2010). Borer argues that the oddity of some nouns to be used as a mass or as a count noun is not due to a grammatical restriction, but rather to a conceptual restriction.

11_{This feature contributes the property of being bounded in space to the structure of the NP. The two count}

readings (unit and kind) and the mass reading follow from the interaction between the two features [Num] and [Size]. The mass reading is similar as to what Borer proposed: this is characterized by the absence of functional structure and thus lacks both Num° and Size°. Kind readings allow for plural marking, but do not allow diminutive affixes. Therefore, De Belder proposes that kind readings are derived by merging Num° but not Size° with the lexical item. Finally, unit readings allow for both plural marking and diminutive affixes. The co-occurrence of these affixes consequently leads to the conclusion that unit readings are derived from a structure

(19)

scope of the present thesis to fully explain De Belder’s arguments, but what is important is that also in Dutch the syntactic context in which (flexible) nouns occur determines whether the noun gets a count or a mass interpretation. For that reason, we consider the mass-count distinction – especially in flexible nouns – as a purely grammatical phenomenon, dependent on the morphosyntactic context.

3.1.1 Previous research

Previous research on the mass-count distinction in English showed that TD children distinguish mass from count around age 4 (Barner & Snedeker, 2005). It has also been shown that this is language specific, as Hebrew-speaking children begin to show some sensitivity between the age of 7 and 8, but are not adult-like even at age 17 (Hacohen & Schaeffer, 2013). For Dutch TD children, Van Witteloostuijn & Schaeffer (2014) show that sensitivity appears around age 6. They show that the number of syntactic cues in the input in Dutch is higher than in Hebrew, but lower than in English, and that therefore the amount of mass-count distinguishing cues in the input matters.

For flexible nouns, Barner & Snedeker show that adults almost always based their judgments on number when the flexible noun was presented in count syntax (97%) but hardly ever when the noun was used in mass syntax (3%). The distinction was also apparent in the child data, with number-based judgements of 95% for flexible nouns presented in count syntax and 25% for flexible nouns presented in mass syntax.12_{The results show that both} English adults and children aged 4;0-4;5 (mean age 4;2) interpret count uses as quantifying over individuals, and mass uses as quantifying over mass or volume. Van Witteloostuijn & Schaeffer (2014) find that Dutch TD children from age 6 onwards show a clear distinction between judgements based on volume in the flexible mass and judgements based on number in the flexible count condition, while children aged 4;1-5;10 perform randomly13

.

As for children with autism or SLI, the mass-count distinction is a quite uninvestigated topic. The only study investigating the mass-count distinction in children with SLI is the study by Froud and van der Lely (2008), who investigated the mass-count distinction in 17 English speaking children with G-SLI (aged 8;0 to 15;6), two groups of younger TD children

12_{As observed by Hacohen & Schaeffer, the relatively high rate of number-based judgments in the flexible mass}

condition in the children is not addressed.

13_{Hacohen & Schaeffer have also investigated flexible count and mass nouns in Hebew (e.g. niyarot 'papers'}

and niyar 'paper'), but in the flexible mass condition a substantial percentage of judgments were not based on volume. As an explanation, Hacohen & Schaeffer propose that the acquisition data reflect a process of change in Hebrew towards a language type that does not grammatically distinguish between mass and count nouns.

(20)

(mean age 6;2 and 7;4), and a group of chronologically age-matched controls. Using a production task, children were presented with novel nouns with a simple CVC structure (e.g. dap) together with potential syntactic and semantic cues. Syntactic cues could signal mass (some dap) or count readings (a dap), as well as the semantic cues (object referents in case of count nouns, substance referents in case of mass nouns). It was found that among the TD children, the integration of syntactic and semantic cues matured over time, while the children with G-SLI performed unlike any of the control groups as they were not able to discriminate between mass and count novel nouns. They for instance pluralized nouns in mass syntax, showing that the children with G-SLI had only limited use of syntactic cues to distinguish between novel count or mass nouns.

To date, there are no studies examining the acquisition of the mass-count distinction in children with autism. The present study is therefore the first to investigate this phenomenon for children with HFA.

3.2 The choice for a definite/indefinite article

In the preceding section the first property in the nominal domain – the mass-count distinction – was discussed, which is a grammatical property. The second property of the DP that will be investigated in this project is the choice for an indefinite or definite article, which is a pragmatic rather than a grammatical property. Definite NPs are typically associated with established discourse referents that are familiar to speaker and hearer, while indefinite NPs are used to introduce new referents in the discourse. The mini-discourse in (6) illustrates this.

(6) a. Indefinite b. Definite

Marie zag vorige week een film. De film was niet erg interessant. ‘Mary saw a movie last week’ ‘The movie was not very interesting’

In (6), the indefinite NP a movie is used to introduce a new entity into the discourse context, and subsequently, when that entity is introduced, it is further referred to with the definite article de in (6) (Abbott, 2006)14

.

Acquisitional literature on article choice shows that pre-school TD children have difficulties with the felicity of articles, and that the different properties of definite and

14_{The present study only distinguishes between definites and two different types of indefinites: referential and}

non-referential. Partitive indefinites will not be taken into consideration, as well as other types of definites, such as bridging definites (in which definiteness is established via shared world knowledge as in a wedding – the

(21)

indefinite articles follow different developmental paths. Firstly, many studies find that children overuse the definite article. Maratsos (1974) for example tested the choice for a definite/indefinite article in children aged 3;0–3;6 (the 3-year-olds), and 4;0–5;0 (divided into the less able 4-year-olds and the more able 4-year-olds) using a verbal elicitation task without visual support. He found overuse of the in the indefinite condition of 42% for the 3-year-old group and the ‘less able 4-year-olds’. Similar results were found by Karmiloff-Smith (1979) for French-speaking children aged up till 8 (overuse of 39–63%), and by Schaeffer & Matthewson (2005) for English-speaking children aged between 2;1-3;10 (overuse of 25%) in indefinite referential contexts15

.

Secondly, a study by Van Hout, Harrigan & Villiers (2010) finds that children have difficulties comprehending the indefinite article: pre-school TD children are overly liberal in their interpretation of a, accepting it for both old and new referents.

Finally, some studies which not only investigated TD children but also children with SLI, found a-overuse in production. These studies will be briefly discussed in §3.2.3, but no satisfying explanation has been given for the a-overuse. We will first focus on the approach by Schaeffer & Matthewson (2005) that explains the overuse of the definite article by making use of the notion ‘common ground’ in §3.2.1, and on the approach by Van Hout et al. (2010) that explains the comprehension of the indefinite article at chance level in §3.2.2.

3.2.1 The Concept of Non-Shared Assumptions

This first approach is based on the idea that adult article systems in English and Dutch rely on the state of the common ground between speaker and hearer (Heim, 1982; Stalnaker 1974, 1978). Beliefs that are shared by all interlocutors in a discourse are said to be in the common ground. Heim (1982) argues that the definite article places a requirement on the content of the common ground at the time of utterance, or in other words: a noun phrase headed by a definite article requires that the existence of the referent to which the noun phrase refers, is part of beliefs between speaker and hearer at the time of utterance. When the definite article is used when the reference is not in the common ground, this is infelicitous. For indefinite articles, the existence of the entity corresponding to the noun phrase may be assumed by the speaker but not by the hearer, or it may be assumed by neither the speaker nor the hearer. This is illustrated in the schema in Table 1, inspired by Schaeffer & Matthewson (2005) and taken

15_{The the-overuse found by Van Hout et al. (2010) for English-speaking children aged between 3;7–5;3 (overuse}

of 26%) will be discussed in a later section, as we will show that actually the evidence for the-overuse in this study is not convincing.

(22)

from Schaeffer, Van Witteloostuijn & De Haan (to appear).

TABLE 1. The English and Dutch adult language article system

In the definite condition (A), the entity referred to is believed to ‘exist’ by both speaker and hearer, and is therefore part of the common ground. The entity in the referential indefinite condition (B) is believed to ‘exist’ by the speaker only, and is not part of the common ground. Crucially, this approach distinguishes a second indefinite condition: the non-referential indefinite condition (C), in which there is no existing entity corresponding to the noun phrase, and which is thus believed by neither the speaker nor the hearer and is also not part of the common ground. An example of the indefinite referential condition is given in (7); an example of the non-referential indefinite condition is given in (8) (taken from Schaeffer & Matthewson, 2005).

(7) I saw a movie last night.

(8) I haven’t read a book for weeks.

Schaeffer & Matthewson (2005) investigated the production of articles in the language of English children (aged between 2 and 4 years) in all three conditions given in table 1. They found an overuse of the definite article the in the B condition of 25% (in contrast to adults who only overgenerated the definite article in 2% of the cases), but crucially, the children hardly ever overgenerated the in C contexts (5%) and a in A contexts (2%). Schaeffer & Matthewson (2005) explain these findings by claiming that children lack a pragmatic principle that states that speaker and hearer assumptions are always independent, and which they call the Concept of Non-Shared Assumptions (CNSA). If a child attributes her own (speaker) beliefs to the hearer, she will not distinguish environment B (believed by speaker only) from environment A (believed by speaker and hearer). Therefore we indeed expect that in those cases environment B becomes environment A, and we expect article mistakes in the B condition. The lack of the CNSA thus causes English-speaking children to sometimes not distinguish common ground (A) contexts from speaker beliefs-only (B) contexts, causing

the-A: Definite Referential

Referent assumed to exist by speaker and hearer

part of common ground

the / de/het

B: Indefinite Referential

Referent assumed to exist by speaker only

not part of common ground

a/an / een

C: Indefinite Non-referential

Referent assumed to exist by neither speaker nor hearer

not part of common ground

(23)

overuse in the B-condition16 .

This approach does not take comprehension of articles into account. One of the few studies investigating not only the production, but also the comprehension of articles by children is Van Hout et al. (2010), which will be discussed in the following subsection. 3.2.2 Scalar implicatures

Using a referent-selection paradigm, Van Hout, Harrigan & Villiers (2010) investigate the production and interpretation of articles by English preschool children aged 3;7–5;3. Test items were set up in stories; the stories contained one main character that moved from picture to picture for each item. For comprehension, the child was told that the experimenter was making some stories for children, but that there were mistakes in the stories, and that the experimenter needed help fixing them: nothing was wrong with the words, but something was wrong with each of the pictures. The child had to move one object (attached with Velcro) in each picture to match the story that the experimenter was telling. Examples are given in (9) and (10).

(9) Definite condition

Stacey has an orange sweater. She knows that orange is John’s favourite colour, so she lets him wear the orange sweater! (Target: same orange sweater).

(10) Indefinite condition

John sees his teacher with a piece of cake. He asks her if he can have a piece of cake. (Target: a different piece of cake).

Van Hout et al. found that both the child and the adult control group correctly interpreted the as referring to the determined referent previously mentioned in the discourse (87% and 96%). For indefinite NPs however, the results were different: the adults correctly chose a new referent 87% of the time, whereas the children chose a new referent only 41% of the time, and the already mentioned referent 59% of the time. The comprehension of the indefinite article by the children is not different from chance, showing that they were overly liberal in their interpretation of indefinites.

Van Hout et al. base their explanation of the comprehension of the indefinite article at

16_{In a way, Maratsos explanation was similar: he claims that in order to use the definite article correctly, a child}

should have so called non-egocentric abilities to introduce referents properly to one’s listener in conversation – use of a definite reference implies that the speaker not only has a particular referent in mind, but that he also expects the definite reference to call up in his listener the same particular referent (1974: 447). Otherwise, the speaker would have used the indefinite article.

(24)

chance level on the idea that preschool children are often insensitive to implicatures when they interpret scalar terms (Noveck, 2001; Papafragou, 2006; Papafragou & Musolino, 2001; Pouscoulous, Noveck, Politzer & Bastide, 2007). Scalar terms are terms like some, might and or, and scalar implicatures arise in examples like (11).

(11) a. Some professors are famous. b. Not all professors are famous.

The utterance in (11a) induces the scalar implicature in (11b), based on Grice’s maxim of Quantity (1975), reformulated by Levinson (1987:401) and called the Q-principle as given in (12).

(12) Q-principle

a. Speaker's maxim: Do not provide a statement that is informationally weaker than your knowledge of the world allows (…)

b. Recipient's corollary: Take it that the speaker made the strongest statement consistent with what he knows, and therefore that: if the speaker asserted A(W), and <S, W> form a Horn scale (such that A(S) H-A(W)), then one can infer K (A(S)), i.e. that the speaker knows that the stronger statement would be false.

If we stick to the hearer’s perspective in (12b), we see how a hearer induces the scalar implicature in (11a) from the utterance in (11b): the hearer knows that the speaker always makes the strongest statement consistent with what he knows; therefore the speaker’s use of some typically indicates that s/he had reasons not to use a more informative term, i.e. all. The sentence therefore gives rise to the implicature that not all professors are famous. A scalar implicature thus is an inference that goes beyond the explicit meaning of an utterance, and is due to pragmatic factors.

Hawkins (1991) and Horn (2006) extend this to the article system, and argue that adults draw a scalar implicature when they interpret indefinite NPs. They argue that the and a provide a contrastive set, in which the is the logically stronger and most informative member of the pair. Indefinite interpretations are then analysed as implicatures that result from not using the definite article in corresponding expressions (Hawkins, 1991:417). When hearing an indefinite, the hearer would reason as follows: had the speaker had a familiar referent in mind (had she intended a determined referent meaning), she would have used the, because that would have been the strongest statement consistent with what she knows. Since the speaker did not use the but a, the stronger statement must be false, and she must have intended a

(25)

non-determined referent (Van Hout et al., 2010).

Following Horn’s (2006) scalar implicature approach to the interpretation of indefinites, Van Hout et al. (2010) claim that children lack the ability to draw a scalar implicature. If children fail to draw a scalar implicature when they interpret indefinite NPs, we indeed expect that children will arbitrarily choose between a determined referent meaning and a non-determined referent meaning when they hear an indefinite, which is supported by Van Hout et al.’s findings17

.

3.2.3 Previous studies on article choice in SLI and HFA

There are only few studies that examine the acquisition of definite versus indefinite articles in children with SLI. Polite, Leonard & Roberts (2011) investigate the production of articles in a group of 12 preschool children with SLI (aged 4;5-7;0, mean age 5;1), a group of MLU-matched TD children (aged 2;10-3;9, mean age 3;3) and a group of age-MLU-matched TD children (mean age: 5;1). A definite condition (e.g. Examiner: A cat and a dog were playing in the backyard. One of them barked. Guess which. Child: The dog), and an indefinite condition were distinguished, of which the indefinite condition tested ‘world knowledge’ (e.g. Examiner: Stephen wants to call home. What can he use to call home with? Child: A telephone). Polite et al. found that both SLI and TD-MLU matched groups very frequently produce the indefinite article a in place of the appropriate definite article the, whereas the three groups did not differ in the indefinite condition. Note that this result is different from the the-overuse that was frequently found in TD children; the results of this study show a-overuse rather than the-overuse.

Chondrogianni & Marinis (2014) find similar results. They investigate the production of definite and indefinite articles in 24 English-speaking children with SLI (mean age: 7;5),

17_{The absence of the CNSA does not correctly predict the errors that Van Hout et al. found for early child}

comprehension: they found that for the interpretation of the indefinite article (context B), children in only 41% correctly chose the new referent, whereas they chose in 59% the incorrect already mentioned referent (context A). The CNSA would in comprehension also state that speaker/hearer assumptions are independent; in other words: the adult hearer would according to the CNSA know that the speaker might know more than the hearer himself knows. When the speaker uses an indefinite in a discourse, then the hearer knows that there are two options: it could be an indefinite Non-referential or an Indefinite Referential. Absence of the CNSA in comprehension would cause the hearer to believe that speaker/hearer knowledge are not always independent. The hearer will therefore (sometimes) incorrectly assume that the speaker has exactly the same assumptions as the hearer does. When the speaker uses an indefinite in a discourse, this always means that the hearer does not know the reference. Lack of CNSA may cause the hearer to assume that the speaker does also not know the reference; thereby assuming that the indefinite is a C-condition, while it could also have been the B-condition. Following this reasoning, we predict that absence of the CNSA would predict overgeneralisation of the C-condition in the C-condition, while Van Hout et al. found overgeneralisation of the A-C-condition in the B-condition.