Anaphora Resolution in Children and Adults An Experimental Study of Mature Speakers and Learners of Basque Maialen Iraola Azpiroz L anguage D evelopment 34 Anaphora Resolution in Children and Adults Language Development Herausgegeben von Cristina Flores (Braga), Tanja Kupisch (Konstanz), Jürgen M. Meisel (Hamburg/ Calgary), Esther Rinke (Frankfurt am Main) Band 34 Maialen Iraola Azpiroz Anaphora Resolution in Children and Adults An Experimental Study of Mature Speakers and Learners of Basque Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http: / / dnb.dnb.de abrufbar. © 2015 · Narr Francke Attempto Verlag GmbH + Co. KG Dischingerweg 5 · D-72070 Tübingen Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Gedruckt auf chlorfrei gebleichtem und säurefreiem Werkdruckpapier. Internet: www.narr.de E-Mail: info@narr.de Printed in Germany ISSN 2197-6384 ISBN 978-3-8233-6929-5 Attona Migueli vii Table of Contents List of figures ........................................................................................................ix List of tables ..........................................................................................................xi Abbreviations..................................................................................................... xiii Abstract..................................................................................................................xv Acknowledgements ......................................................................................... xvii Introduction............................................................................................................ 1 1 Pronominal anaphora ..................................................................................... 5 1.1 Anaphora ................................................................................................. 6 1.1.1 Clause order ................................................................................. 8 1.1.2 Directionality.............................................................................. 12 1.1.3 Third-person (personal and demonstrative) pronouns ....... 14 1.1.4 Null vs. overt subject pronouns .............................................. 18 1.2 A discourse-based account: Accessibility Theory............................ 22 1.3 A generativist account: Binding Theory............................................ 24 1.3.1 Personal pronouns vs. demonstratives .................................. 28 1.4 A processing account: The Position of Antecedent Strategy.......... 30 2 Pronouns in child language......................................................................... 35 2.1 Bilingual acquisition ............................................................................. 35 2.2 The syntax-discourse interface ........................................................... 38 2.2.1 Binding conditions in monolingual children ........................ 39 2.2.2 Pragmatically appropriate use of subject pronouns in discourse contexts...................................................................... 43 2.3 The Interface Hypothesis ..................................................................... 46 2.3.1 Representational account ......................................................... 49 2.3.2 Processing account .................................................................... 50 2.3.3 Extralinguistic factors: Input and age of onset effects ......... 50 3 The interpretation of null and overt subject pronouns in Basque ...... 55 3.1 Third-person morphology ................................................................... 55 3.2 Third-person pronouns........................................................................ 58 3.2.1 Intrasentential antecedent choice ............................................ 59 3.2.2 Frequency ................................................................................... 63 3.3 Aims and research questions .............................................................. 65 3.4 Participants ............................................................................................ 68 viii 3.5.2 Experiment 2: bera...................................................................... 85 3.5.3 Discussion................................................................................... 94 3.5.4 - Experiment 3: har(e)k ...................................................................... 99 - 3.5.5 - Experiment 4: berak ....................................................................... 111 - 3.5.6 - Discussion ...................................................................................... 119 - 3.6 - Acceptability Judgement Task ................................................................ 122 - 3.6.1 - Experiment 5: hura ........................................................................ 122 - 3.6.2 - Experiment 6: bera ......................................................................... 126 - 3.6.3 - Discussion ........................................................................................ 128 - 4 - Reference of null and overt subject pronouns in Basque .......................... 131 - 4.1 - Adults’ interpretation of pronouns ........................................................ 132 - 4.1.1 - NSP vs. hura in forward anaphora ............................................. 132 - 4.1.2 - NSP vs. bera in forward anaphora .............................................. 134 - 4.1.3 - Clause order and anaphora directionality ................................ 136 - 4.1.4 - NSP/ hura/ bera .............................................................................. 138 - 4.2 - Children’s interpretation of pronouns................................................... 145 - 4.2.1 - NSP vs. hura in forward anaphora ............................................. 145 - 4.2.2 - NSP vs. bera in forward anaphora .............................................. 147 - 4.2.3 - Clause order and anaphora directionality ................................ 148 - 4.2.4 - NSP/ hura/ bera .............................................................................. 149 - 5 - Conclusions ......................................................................................................... 159 - References .................................................................................................................. 163 - Appendix A. Figures ................................................................................................ 181 - 3.5 Picture Selection Task........................................................................... 71 3.5.1 Experiment 1: hura..................................................................... 72 ix List of figures Figure 1. Syntactic tree of example (4b) .............................................................. 10 Figure 2. Syntactic tree of example (4d) .............................................................. 11 Figure 3. Accessibility hierarchy for discourse referents (adapted from Ariel 1990) ................................................................................................... 23 Figure 4. Direction of anaphora/ branching directionality parameter ........... 41 Figure 5. Geographic origin of the participants (adapted from ...................... 70 Figure 6. An example of the picture pairs presented in the Picture Selection ................................................................................................................ 76 Figure 7. Preference for the subject antecedent in NSPs and hura ‘that-ABS’ by group and experimental condition ......................................... 78 Figure 8. Adults’ preference for the subject antecedent in forward anaphora by pronoun (Ø/ hura ) and clause order.......................................... 79 Figure 9. Adults’ preference for the subject antecedent in backward anaphora by pronoun (Ø/ hura ) and clause order.......................................... 79 Figure 10. cL1 group’s preference for the subject antecedent by pronoun (Ø/ hura ) and anaphora type ............................................................. 80 Figure 11. cL2 group’s preference for the subject antecedent by anaphora type and clause order........................................................................ 81 Figure 12. Preference for the subject antecedent in forward anaphora by group and pronoun type (Ø/ hura )............................................ 82 Figure 13. Preference for the subject antecedent in backward anaphora by group and pronoun type (Ø/ hura )............................................ 82 Figure 14. Preference for the subject antecedent in forward anaphora by group and clause order ............................................................... 84 Figure 15. Preference for the subject antecedent in backward anaphora by group and clause order ............................................................... 84 Figure 16. Preference for the subject antecedent in NSPs and bera ‘(s)he (him/ herself)-ABS’ by group and experimental condition ........ 87 Figure 17. Adults’ preference for the subject antecedent in forward anaphora by pronoun type (Ø/ bera ) and clause order.................................. 88 Figure 18. Adults’ preference for the subject antecedent in backward anaphora by pronoun type (Ø/ bera ) and clause order ............... 88 Figure 19. cL1 group’s preference for the subject antecedent in forward anaphora by pronoun type (Ø/ bera ) and clause order .................. 89 Figure 20. cL1 group’s preference for the subject antecedent in backward anaphora by pronoun type (Ø/ bera ) and clause order ............... 90 Figure 21. cL2 group’s preference for the subject antecedent by anaphora type and clause order........................................................................ 90 Figure 22. Preference for the subject antecedent in forward anaphora by group and pronoun type (Ø/ bera )............................................. 91 x Figure 23. Preference for the subject antecedent in backward anaphora by group and pronoun type (Ø/ bera )............................................. 92 Figure 24. Preference for coreference with the subject in forward anaphora by group and clause order ............................................................... 93 Figure 25. Preference for coreference with the subject in backward anaphora by group and clause order ............................................................... 93 Figure 26. Visual stimulus provided together with the auditory stimuli ................................................................................................. 102 Figure 27. Visual stimulus provided together with the auditory stimuli (46c-f) ..................................................................................... 103 Figure 28. Adults’ antecedent choices for NSPs and har(e)k in forward and backward anaphora ................................................................... 105 Figure 29. cL1 group’s antecedent choices for NSPs and har(e)k in forward and backward anaphora............................................................... 106 Figure 30. cL2 group’s antecedent choices for NSPs and OSP har(e)k in forward and backward anaphora............................................................... 107 Figure 31. Between-group pairwise comparisons by experimental condition ............................................................................................................. 108 Figure 32. Coreference for NSPs and har(e)k ‘that-ERG’ by group and experimental condition ............................................................................. 108 Figure 33. Coreference for NSPs and har(e)k ‘that-ERG’ by group ............... 110 Figure 34. Adults’ antecedent choices for NSPs and berak in forward and backward anaphora .......................................................................................... 113 Figure 35. cL1 group’s antecedent choices for NSPs and berak in forward and backward anaphora................................................................................... 114 Figure 36. cL2 group’s antecedent choices for NSPs and berak in forward and backward anaphora................................................................................... 115 Figure 37. Between-group pairwise comparisons by experimental condition in three-antecedent contexts .......................................................... 116 Figure 38. Coreference for NSPs and berak ‘(s)he (him/ herself)-ERG’ by group and experimental condition............................................................ 116 Figure 39. Coreference for NSPs and berak ‘(s)he (him/ herself)-ERG’ by group.............................................................................................................. 118 Figure 40. Overt pronoun hura preference in [-TS] and [+TS] contexts by group.............................................................................................................. 125 Figure 41. Overt pronoun bera preference in [-TS] and [+TS] contexts by group.............................................................................................................. 127 Figure 42. The distribution of Basque pronouns with regard to topic (non)continuity .................................................................................................. 140 xi List of tables Table 1. Personal pronouns and the present-tense agreement system in Basque ................................................................................................................... 57 Table 2. Pronoun frequency in different texts produced by Basque teenagers ............................................................................................................... 63 Table 3. Location and antecedent of Basque demonstratives .......................... 64 Table 4. Linguistic profiles of the participants in the six experiments ........... 69 Table 5. Crosslinguistic patterns of 6to 8-year-olds and adults for OSPs in discourse contexts ......................................................................................... 154 xiii Abbreviations ABS Absolutive AUX Auxiliary BT Binding Theory cL1 child L1 cL2 child L2 CP Complementiser Phrase DAT Dative DET Determiner DP Determiner Phrase ERG Ergative FUT Future IP Inflectional Phrase IPF Imperfect NEG Negation NP Noun Phrase NSP Null Subject Pronoun OSP Overt Subject Pronoun OVS Object - Verb - Subject PAS Position of Antecedent Strategy PF Perfect PRS Present SG Singular SOV Subject - Object - Verb SVO Subject - Verb - Object [-TS] Topic continuity [+TS] Topic shift VP Verb Phrase xv Abstract This book focuses on the comprehension of null and overt subject pronouns in intrasentential anaphora contexts in Basque, a language which employs overt referential devices that fall outside of the scope of what are traditionally considered third-person pronouns, namely the demonstrative hura ‘that’ and the quasipronoun bera ‘(s)he (him/ herself)’. Data on the referential properties and the discourse features of null and overt pronouns obtained from adult native speakers in two experimental off-line tasks set a baseline for comparison with a) the insights reported in descriptive grammars and b) developmental data from 6to 8-year-old child L1 and L2 groups. As a first result, the adult data corroborate previous findings highlighting the division of labour, i.e. the different antecedent biases of null and overt pronouns. In a Picture Selection Task, null pronouns were predominantly coreferent with the subject. In contrast, bera and especially hura were interpreted as referring to the extrasentential referent. The division of labour, which remained almost the same regardless of the number of potential antecedents, was not based on syntactic prominence in Basque, contrary to previous studies in other null subject languages. The results were replicated in an Acceptability Judgement Task, where adults selected null pronouns in topic-continuity contexts, whereas hura and (to a lesser degree) bera were selected in topic-shift contexts. Second, the experimental evidence supports Basque grammarians’ description of hura as an obviative (disjoint) pronoun, but does not completely support the proposed proximate/ intensive and logophoric properties of bera. The results show that bera seems to be more equivalent to the thirdperson pronouns of other languages than hura. Finally, the data reveal similar response patterns in children and adults for null pronouns, but divergent patterns for overt pronouns. Both child L1 and L2 groups showed a general preference for coreference with the subject regardless of pronoun type in the Picture Selection Task, and they preferred null to overt pronouns in the Acceptability Judgement Task. Thus, the interpretable [+topic shift] feature that is mapped onto the overt pronoun in the adult grammar has not been fully acquired by child L1 and L2 participants, consistent with crosslinguistic studies. Children’s delay in internalizing the properties of overt pronouns and their general preference for the null pronoun seem to be related to the low frequency of overt pronouns in the input to which children are exposed. Thus, Basque child data do not exhibit the paradox between the linguistic default (null pronoun) and the learner default (overt pronoun) observed in other null-subject languages, since the null pronoun performs both functions in Basque. xvii Acknowledgements This book is a revised version of my Ph.D. thesis, which I completed under the joint supervision (cotutelle) of the University of the Basque Country (UPV/ EHU) and the University of Konstanz (Germany) and presented at the University of the Basque Country in April 2014. A number of people have played vital roles in the development of this book. First of all, I want to thank my supervisors Maria José Ezeizabarrena and Georg A. Kaiser for their help and feedback during the process of putting this work together. I am particularly thankful to Prof. Dr. Barış Kabak from the University of Würzburg for inspiring me to keep on working on bilingualism and language acquisition through his interesting classes. Prof. Dr. Bettina Braun from the University of Konstanz also deserves my profound thanks for answering my endless questions regarding the statistical analysis of my data. I would also like to express my gratitude to the editors of the series Language Development, Prof. Dr. Jürgen M. Meisel, Dr. Cristina Maria Moreira Flores, Prof. Dr. Tanja Kupisch and Prof. Dr. Esther Rinke, for carefully poring through the text and for their detailed comments and suggestions on an earlier version of the manuscript, which have been crucial for the work’s improvement. Special thanks also to Claire Bacher for her thorough proofreading. I am very grateful to the Basque schools Laskorain ikastola and Paz de Ziganda ikastola, which kindly allowed me to collect my data. I would also like to acknowledge the invaluable help of the pupils and the teachers who took part in the experiments and especially that of Maribi Mateo for making everything easier for me. Prof. Antonella Sorace is also owed a debt of thanks for generously lending us her materials to adapt them to Basque. I wish to thank Georg Höhn, Mikel Santesteban, Kepa Erdozia, Oxel Uribe-Etxeberria and Mikel Iraola for their contribution to the preparation of some materials, and of course Maria José Ezeizabarrena, who was of immense help in the data collection process with the cL2 participants. Dr. Fatemeh Nemati deserves my deepest gratitude for her constant support despite being thousands of kilometres away. A special mention goes out to my family for their comforting words and for showing me the way forward when I felt stuck, especially my mum, whose presence made my data collection easier. I also wish to thank my friends back home, Alazne, Eli and Maite, for their friendship and for lending a hand during the running of the experiments. Last but not least, I want to thank Flo for his endless patience and for believing in me during this long process. This research was supported by the Basque Government (grant PFPI 2009-2012). Tolosa, 31 December 2014 1 Introduction One area in which syntactic aspects of language interact with discourse and pragmatic considerations is anaphora resolution, the process of resolving a pronoun in accordance with an earlier or a later item in discourse, the antecedent. Anaphoric dependencies are one of the most important linguistic mechanisms employed to maintain discourse coherence. The use of pronouns to refer to previously mentioned referents or to anticipate arguments mentioned later is a common linguistic device that serves to avoid redundancy or repetition. In so-called ‘null-subject languages’, the existence of two separate forms to express pronouns (both overt and null) has led researchers to believe that different types of pronouns have different preferences in terms of determining their antecedents. The expression or omission of the subject, albeit grammatically optional, relies heavily on discourse-pragmatic considerations, such as whether a new referent is introduced in the discourse or not. Since the 1990s, it has been argued that linguistic phenomena at the syntax-discourse interface level are somewhat more complex and costly (in terms of processing) than purely syntactic structures, as multiple types of information must be integrated (for a discussion, see Sorace 2011) and hence are principal loci of developmental delays. Although the syntactic knowledge necessary for dependencies between reflexives and their antecedents seems to be attained by children without any difficulty at an early age, the same children exhibit delays with respect to the interpretation of discourse dependencies between non-reflexive pronouns and their antecedents. This phenomenon, known as the Delay of Principle B Effect, has been extensively discussed in the literature on the acquisition of binding conditions (i.e. English monolingual children allow coreference between pronouns and local antecedents, as in *Peter i likes him i ; Avrutin & Wexler 1992, Chien & Wexler 1990). The claim that the syntax-discourse interface is an area that is particularly vulnerable to instability has become increasingly prominent in contemporary research, especially since the formulation of the Interface Hypothesis (Sorace & Filiaci 2006). Evidence for the validity of the ‘syntax before discourse’ hypothesis has emerged from a substantial body of research on anaphora resolution in various bilingual developing grammars in both children and adults acquiring a null-subject language together with a non-nullsubject language (see, among others, Serratrice, Sorace & Paoli 2004, Tsimpli, Sorace, Heycock & Filiaci 2004). Early studies on bilingual child production and (near-native and attrited) adult comprehension pointed towards a unidirectional non-target-like pattern consisting of an overuse of overt pronouns due to the overextension of their scope to topic-continuity contexts. 2 Interestingly, these bilingual populations showed native-like behaviour with respect to null pronouns. Thus, it was concluded that the overt pronoun becomes the default option for learners when they fail to integrate syntactic knowledge with peripheral systems such as discourse-pragmatics. However, the pattern established in earlier studies has been challenged by more recent child comprehension data from several Romance languages: monolingual and simultaneous bilingual (English-Spanish, Spanish-Italian and Spanish- Catalan) children of 6-8 years of age have been reported to display a bidirectional non-adult-like extension of the scope of both null and overt pronouns (Barquin & Costa 2011 and Shin & Cairns 2012 for Spanish; Sorace, Serratrice, Filiaci & Baldo 2009 for Italian). In recent decades, the field of bilingual acquisition in childhood has experienced a significant shift in the focus of study from the simultaneous acquisition of two languages from birth to child second language development. The successive acquisition of a second language beyond age 3 has become an emerging field of study, and the number of studies on children whose exposure to the second language occurs between the ages of 4 and 7 has steadily grown. At this point, the question of whether successive early L2 learners will follow the pattern of monolingual learners or instead that of adult L2 learners has already been clarified in some grammatical domains, such as inflectional morphology, where child L2 learners seem to behave like adult L2 learners (e.g. Meisel 2008, Schlyter 2011). However, in anaphoric dependencies, a linguistic phenomenon situated at the interface between internal and external modules (i.e. the syntax-discourse interface), there has been practically no investigation concerning the performance of child L2 learners, as research has mainly focused on the mastery of simultaneous bilinguals in comparison to monolingual children with regard to the syntactic and discourse properties of subject pronouns. However, preliminary results from English child L2 learners of Spanish have revealed residual optionality in the discourse constraints in line with the response patterns of adult advanced learners (Pladevall 2010). The study presented here is intended to contribute to the ongoing debate on anaphora resolution in null-subject languages by presenting comprehension data on intrasentential anaphoric dependencies in Basque, a language that lacks true third-person pronouns. The aim of this work is threefold. Firstly, it investigates whether null pronouns and overt referential devices — namely, the demonstrative hura ‘that’ and the quasipronoun bera ‘(s)he (him/ herself)’ (de Rijk 2008: 114) — differ in their antecedent choices in Basque in a manner consistent with the crosslinguistic patterns. Secondly, it examines whether the linguistic descriptions established in Basque traditional grammars of hura ‘that’ as a obviative/ neutral pronoun vs. bera ‘(s)he (him/ herself)’ as a proximate/ intensive pronoun (Ortiz de Urbina 1989: 147) and their equivalence to third-person pronouns in other languages (de Rijk 2008: 115, 209) can be supported by experimental evidence. Finally, it com- 3 pares the interpretation of null and overt pronouns by three groups of Basque-Spanish bilinguals: native adults (L1 adults), native children (the cL1 group) and early successive bilingual children (the cL2 group). This last comparison allows us to investigate the effect of extralinguistic factors (such as biological age, age of onset of acquisition and amount of exposure) on the interpretation of null and overt pronouns in Basque. Data from child L2 learners will provide more insights on the behaviour of early sequential bilinguals in the acquisition of syntactic and discourse subject properties. To this end, the results of six off-line tasks are presented. Using two different methodologies, the interpretations of Basque null and overt pronouns in subject position are analysed from various perspectives. In a Picture Selection Task, the referential properties of hura and bera in comparison to null pronouns were tested in forward and backward anaphora conditions with different clause orders (main-subordinate and subordinate-main) in onereferent (Experiments 1 and 2) and two-referent sentences (Experiments 3 and 4). Another methodology was used for Experiments 5 and 6, namely an Acceptability Judgement Task in which participants had to choose between null and overt pronouns in topic-continuity and topic-shift contexts. The acceptability status of null and overt pronouns in this task, based on speakers’ preferences, allows examination of whether Basque learners have any default form when confronted with a choice between null and overt pronouns. The structure of the book is as follows. The first sections of Chapter 1 concentrate on defining pronominal anaphora and discussing the different factors that may affect antecedent assignment, such as: i) clause order, ii) anaphora type and iii) the overt realisation of the subject in null-subject languages. The troublesome crosslinguistic category of the third-person personal pronoun is also discussed; this category may not be universal, since the so-called ‘two-person’ languages have no true third-person pronouns (Bhat 2004). Instead, demonstratives fulfill this role, blurring the dividing line between the two anaphoric forms. The subsequent sections focus on reviewing the three major theories of referring expressions that attempt to capture the restrictions governing anaphoric dependencies from different angles: Accessibility Theory (Ariel 1990), Binding Theory (Chomsky 1981) and the Position of Antecedent Strategy (Carminati 2002). Chapter 2 starts by pointing out a distinction between different types of early bilingual language acquisition, such as simultaneous bilingual acquisition (2L1) and child L2 acquisition (cL2). The following sections give a thorough and up-to-date discussion of the background literature on the instability at the syntax-discourse interface, providing the results from child studies in the generative (Binding Theory) model (Principle B). Results obtained from child studies on Principle C, which directly sets syntactic constraints on the interpretation of backward anaphora structures, are also reported. In addition, empirical research on children’s interpretations of null and overt 4 subject pronouns is discussed in relation to the Interface Hypothesis (Sorace & Filiaci 2006) and the possible explanations proposed in the literature to account for the delays observed at the syntax-discourse interface. In Chapter 3, comprehension data from the two tasks focusing on the interpretation of null and overt subject pronouns are reported. First, a description of the criteria for using third-person pronouns and a discussion of their frequency in prior non-experimental empirical work are provided. The linguistic profiles of the participants are then described, including information about differences in the age of onset of acquisition between the cL1 and the cL2 groups. Subsequently, the materials and the methods used are introduced and the results from the six experiments are presented. Chapter 4 offers an analysis of the data presented in Chapter 3 in light of the prior research presented in Chapter 2. The discussion is centred on the research questions formulated for the present study: a) whether a division of labour is found between null and overt pronouns in Basque in line with the pattern observed crosslinguistically, b) whether the experimental evidence supports the observations of usage found in the descriptive grammars of Basque and c) to what extent children’s performance is similar to that of adults and homogeneous between linguistic profiles (cL1 and cL2 groups). Finally, Chapter 5 presents conclusions, emphasising the findings of the study and providing suggestions for future research. 5 1 Pronominal anaphora On issues related to pronominal elements and the establishment of coreference, one of the central questions is how a pronoun is interpreted — that is, how the pronoun selects its referent. The process of finding an antecedent for a pronoun is relevant because a pronoun itself does not carry referential content (apart from morphological feature specifications such as gender, number or case). However, this does not mean that a pronoun can pick any antecedent available in the mental representation or in the immediate context of a given discourse. Ever since the publication of Lectures on Government and Binding (Chomsky 1981), anaphoric dependencies have been one of the most common topics of research in the generative framework. The beginning of the research paradigm of principles and parameters was based on (among other theories) Binding Theory, which addresses anaphoric relations and the syntactic restrictions found between coreferents. However, it was soon discovered that anaphoric dependencies could not be addressed exclusively from a syntactic perspective, since in certain coreference relations, syntactic considerations seem to be irrelevant (see Section 1.3). Thus, more recently, the problem of resolving a pronoun in accordance with an earlier or later linguistic item in the discourse has been examined from a discourse-based and processing point of view. Nevertheless, the underlying process involved in resolving what a pronoun or a noun phrase refers to is still unclear. Hence, it is still a challenging and active area of research. The structure of this introductory chapter is as follows. Section 1.1 starts off with the possible relationships between an anaphoric noun phrase and its antecedent and continues with the potential effects of clause order (mainsubordinate vs. subordinate-main) and anaphora directionality (forward vs. backward anaphora) on pronominal-antecedent dependencies. A discussion of each of these components present in anaphoric relations is followed by an overview of the findings reported in the literature, especially with regard to the interpretation of pronouns. Section 1.1.3 is devoted to the description of third-person (personal and demonstrative) pronouns and their different antecedent biases. In addition, in Section 1.1.4, special attention is focused on the different interpretative behaviour exhibited by null and overt pronouns. The remainder of this introductory chapter reviews different theoretical streams in the analysis of anaphoric relations that will be relevant for the interpretation of null and overt subject pronouns in the context of intrasentential anaphora: Accessibility Theory (Ariel 1990, 2001), Binding Theory (Chomsky 1981) and the Position of Antecedent Strategy (Carminati 2002). The first account argues for the selection of antecedents in relation to the activation status of the mental representations (Ariel 1990), while the second 6 approach views anaphoric dependencies as constrained by the syntactic configuration (Chomsky 1981). The third approach is a processing account that predicts different antecedent biases for null and overt pronouns based on syntactic prominence (Carminati 2002). These three accounts enable us to incorporate both syntactic and extra-syntactic (discourse) information in order to develop a better understanding of the establishment of dependencies linked to pronominal interpretation. 1.1 Anaphora A number of definitions of the term ‘anaphora’ exist in the literature, but the same concept underlies all of them: reference to something mentioned or implied in previous discourse (Green 1989). This characteristic element of tracking is related to the word’s etymology. The Ancient Greek word anaphora (ανα ϕ oρα), made up of the separate words ανα (‘back, upstream, back in an upward direction’) and ϕ oρα (‘the act of carrying’), denotes the act of carrying (something) back upstream (Peral & Ferrández 2003). More precisely, Huang (2000: 1, referring to Huang 1994, Lust 1986, Wasow 1986) states that anaphora is a relation between two linguistic forms “wherein the interpretation of one (called an anaphor) is in some way determined by the interpretation of the other (called an antecedent)”. Here, Huang alludes to the referential dependency of the anaphor (usually the pro-form) towards the antecedent (the entity to which it refers), since the anaphor does not convey any semantic meaning on its own. In other words, the anaphor repeats the reference or the sense that the antecedent has already established, such that the two linguistic expressions end up having the same referential value. At this point, a clarification concerning the term ‘anaphor’ is needed. Although the term has traditionally been used for any NP (reflexive or not) that has an antecedent in the preceding discourse, there is another distinct sense of anaphor. Within the generative theory, anaphor is presented as an NP with the features [+anaphor, -pronominal], as opposed to pronouns with the features [-anaphor, +pronominal] (Chomsky 1982); the restrictions of this theory of anaphoric relations will be discussed in Section 1.3. Anaphora is a very common linguistic device in natural language because, as Blackwell (2003: 2) notes, “anaphoric reference enables speakers to avoid redundancy or repetition by the use of a semantically, lexically, and phonologically attenuated linguistic expression (the anaphor), in place of the full lexical expression (the antecedent)”. According to Huang (2000), anaphora can be syntactically divided into two main categories: NP-anaphora and VP-anaphora; the former can be encoded by gaps (or empty categories), pronouns, reflexives, names and descriptions (Huang 2000: 2). The example in (1) shows the most common type of NP-anaphora, the so-called ‘pronominal anaphora’ in which the coindexed NPs exhibit sameness of reference (or coreference). 7 (1) After the baroness i had visited the lord, she ij left the house. (adapted from Büring 2005: 1) However, the pronoun she in (1) can also be contra-indexed with its antecedent the baroness, thus becoming disjoint in reference (non-coreference). In such a case, it would refer to an extrasentential referent. This process of assigning an antecedent to a pronoun is known as anaphora resolution. There are three logically possible coreference relations between two NPs 1 in a given sentence (Reinhart 1983: 29): (2) a. Obligatory (stipulated) coreference: Zelda i bores herself i/ *j . b. Obligatory (non-stipulated) non-coreference: She i adores Zelda *i/ j ’s teachers. c. Optional (free) coreference: Zelda adores her ij teachers. The three-valued system of coindexing illustrated in (2) operates along these lines: (i) NPs such as the R-expression (the ‘referring’ or referential expression) Zelda and the reflexive pronoun herself can be positively coindexed (2a), (ii) negatively coindexed (2b) or (iii) neutrally indexed (i.e. neither positively nor negatively coindexed) (2c). In (2a), the reflexive herself must be interpreted anaphorically as referring to the subject antecedent Zelda due to a syntactic constraint on the distribution of reflexives (see Principle A in Section 1.3). In (2b), the R-expression Zelda must be disjoint in reference or non-coreferent with the preceding pronoun she (see the following section). Finally, (2c) presents the optional (free) coindexing of a pronoun and a nominal antecedent. Büring (2005: 2) adopts the following definition of antecedent: “A is the antecedent of B iff (if and only if) (i) A precedes B, and (ii) A and B corefer”. This holds for forward anaphora 2 sentences like (1) in which the pronoun follows the antecedent (Guasti 2004: 300). There are, however, cases in which the ordering of the antecedent and anaphor can be switched, resulting in backward anaphora (or cataphora, as it is also known) in which the pronoun shows reference to something later in the text: 1 In generative syntax, ever since Abney’s (1987) proposal, the structure of NPs has been extended to DPs. Throughout this work, we refer to NPs and DPs interchangably, depending on the terminology used by the cited researchers. 2 In the generative tradition, pronouns in examples like (1) are regarded as forward anaphora because one has to move forwards (starting from the antecedent) when the pronoun follows the antecedent and backwards when the former precedes the latter. However, scholars in other traditions working on cohesion (such as Halliday & Hasan (1976)) use anaphora for cases in which the antecedent precedes the pronoun and cataphora when the antecedent follows it. 8 (3) After she had visited the lord, the baroness left the house. Examples of both forward (1) and backward anaphora (3) can be regarded as anaphora more generally because the term is sometimes used generically to cover both types of anaphoric relations (Dixon 2010: 247). 1.1.1 Clause order A main clause can be interpreted independently of its subordinate clause; however, the interpretation of a subordinate clause depends on the meaning of the main clause. Consequently, it is plausible to ask whether the antecedent biases of pronouns are likely to be affected by different clause orders during the processing of complex sentences consisting of a subordinate (dependent) and a main (independent) clause. In fact, studies on adult English that do not directly address pronoun interpretation have demonstrated that main and subordinate clauses are processed differently (Bever & Townsend 1979, Clark & Clark 1968), but these differences do not always point in the same direction. For example, better verbatim memory performance has been observed in sentence-initial subordinate clauses (especially for adversatives such as though-clauses) than in final subordinates in lexical probe latency tasks, although this tendency is statistically non-significant (Bever & Townsend 1979). However, the same authors report better semantic comprehension for initial main clauses in semantic latency probe tasks. Clause-effect differences have been attributed to the encoding stages of processes: the meaning of a subordinate clause must be kept in memory until the main clause is processed — unlike main-subordinate clauses, where the superficial form will be lost and short-term memory is available for processing subordinate clauses. This suggests that a higher processing load is required in subordinate clauses, especially when they precede main clauses, because they are not semantically complete, and the complete semantic level of processing depends on the content of the main clauses. According to Bever and Townsend’s (1979) results, the on-line accessibility of superficial form in initial clauses decreases in this order, with though-clauses more easily held in memory than the rest: though-clause>while-clause>when-clause>sinceclause>if-clause. Bever and Townsend’s (1979) finding that subordinate clauses are retained longer in short-term memory is in line with Jarvella and Herman’s (1972) results from an imitation task and Garnham, Oakhill and Cain’s (1998) conclusions on the processing of VP-ellipsis. Although Bever and Townsend’s study does not directly deal with pronoun interpretation, they mention that in contexts such as *He i ate supper before John i left town, syntactic rules (i.e. disallowing coreference) precede listeners’ usual strategy of representing asserted main clauses in semantic form as quickly as possible (Bever & Townsend 1979: 211). Attention to constraints on coreferential pronouns depending on clause order had already 9 been emphasised by, among others, Lakoff (1968), Lasnik (1976) and Reinhart (1976). Although coreference between an embedded DP and the following matrix pronoun is possible, as seen in (4c), a pronoun in an initial main clause that is coreferent with a subsequent embedded DP is not grammatically allowed (4b). Note that anaphora directionality changes from forward (4c) to backward anaphora (4b). In contrast, pronouns that occur in subordinate clauses are not subject to coreference constraints ((4a) and (4d)). (4) a. Sarah i listens to music [when she i reads poetry]. b. *She i listens to music [when Sarah i reads poetry]. c. [When Sarah i listens to music], she i reads poetry. d. [When she i listens to music], Sarah i reads poetry. (examples from Lust 2006: 214) In backward anaphora in which the main clause precedes the subordinate clause, as in (4b), the structure-dependent notion of c-command comes into play, blocking intrasentential coreference with the subsequent subject antecedent. Hence, the only possible interpretation remaining is the disjoint reference interpretation. C-command, first proposed by Reinhart (1976, 1983), is defined in terms of dominance: (5) A node A c-commands a node B if and only if (i) A does not dominate B (ii) B does not dominate A (iii) The first branching node dominating A also dominates B On this basis, the following set of c-command relations is possible in the structure in (6): (6) A B C D E 10 D mutually c-commands E (D c-commands E, E c-commands D) B mutually c-commands C (B c-commands C, C c-commands B) B c-commands D B c-commands E (taken from Frank, Hagstrom & Vijay-Shanker 2002: 111) From this, it follows that anaphoric dependencies are explained in terms of binding in the generative tradition, which explicates when a pro-form may be bound or free (i.e. not bound) (Haegeman 1994: 212): (7) A binds B if and only if a) A c-commands B b) and A and B are coindexed (i.e. A and B refer to the same entity) Returning to example (4b), the reason why she and Sarah cannot be coreferent is that she c-commands Sarah. That is, following the criteria mentioned in (5), she does not dominate Sarah, Sarah does not dominate she, and the first branching node (IP) dominating the DP she dominates the DP Sarah (see Figure 1. R-expressions like Sarah cannot be c-commanded by any antecedent as formulated in Principle C (see Section 1.3 for more details), explaining why she and Sarah cannot be coreferent. (8) Figure 1. Syntactic tree of example (4b) 11 In contrast, in the sentence in (4d), where the subordinate clause precedes the main clause, the embedded DP she does not c-command the matrix subject Sarah: the first branching node dominating she, the IP, does not dominate the DP Sarah. Thus, as Sara is not c-commanded by any antecedent, it can be coreferential with the preceding subject pronoun she, as shown in Figure 2. (9) Figure 2. Syntactic tree of example (4d) 12 1.1.2 Directionality Section 1.1.1 introduced the distinction between types of anaphora in terms of anaphora directionality, i.e. whether pronouns occur after their antecedent (forward anaphora) or instead precede them (backward anaphora). Forward anaphora structures appear to be crosslinguistically more common than backward anaphora sentences — this being the reason for defining forward pronominalisation as the ‘unmarked’ word order (Blackwell 2003: 17). However, backward anaphora sentences are of particular interest, since they have generally been considered to be a good way to prove that it is not the linear ordering of the antecedent and the pronoun that determines whether or not coreference is allowed. Rather, the structural constraints (ccommand) of the sentences themselves account for the unavailability of coreference, as explained in (4b). Such syntactic rules confirm one of the fundamental characteristics of human language: structure dependence, a principle of the Universal Grammar (the set of innate constraints on the formal properties of all human languages and a core component of the Language Acquisition Device, the innate linguistic capacity with which humans are endowed) that is therefore present in child language from birth (Chomsky 1980). In this regard, studies analysing the processing of anaphoric dependencies in adults have devoted special attention to backward anaphora constructions in which the innateness of structure-dependent rules like Principle C can be tested. As demonstrated by evidence from off-line and on-line studies, this type of anaphora is particularly challenging for the processor because the potential antecedents are introduced after the pronoun. For example, Gordon and Hendrick (1997) carried out an interesting offline study investigating the acceptability judgements of native English speakers with regard to different kinds of anaphors in diverse syntactic environments. Positive results were obtained in sentences testing reflexive and non-reflexive pronouns. English participants accepted sentences like (10a) but rejected sentences like (10b), in which coreference is not viable because the pronoun cannot be bound (i.e. it has to be referentially free in its domain under Principle B; see Section 1.3). However, results were not highly conclusive for sentences testing the accuracy of Principle C, which depends on prohibiting coreference in (4b) but allowing it in (4d). Adult English native speakers’ judgements in backward anaphora contexts were not determined by whether coreference was licit or illicit; they simply showed a general preference for non-coreference. In (10c), where the pronoun precedes the name but does not c-command it, the participants seldom accepted coreference (only 28% of the time). Acceptance rates were also marginal in namename contexts without a c-command relation involved, especially when the antecedent was the subject NP of the clause, as in (10d). 13 (10) a. John thought Susan i injured herself i . b. Joan i respects her *i/ j . c. His i roommates met John i at the restaurant. d. John i ’s roommates met John i at the restaurant. Based on the frequency of acceptance, Gordon and Hendrick (1997) came up with the following ordering of ease of coreference in the sequence of referring expressions shown in (11): coreference is more readily achieved in Name-Pronoun sequences than in Name-Name sequences, which in turn allow coreference more easily than Pronoun-Name sequences do. (11) Name-Pronoun > Name-Name > Pronoun-Name More recent research on the interpretation of pronouns in adults has concentrated on the on-line or real-time processing of pronouns. Studies analysing anaphora resolution in backward anaphora sentences have found that the parser immediately starts establishing an anaphoric dependency. For example, van Gompel and Liversedge (2003) used eye-tracking to examine reading times of complex sentences involving licit backward anaphora constructions with gender match (12a) in comparison to gender mismatch (12b). (12) a. When he was fed up, the boy visited the girl very often. b. When he was fed up, the girl visited the boy very often. Their main finding was that the processor first establishes coreference between a cataphoric pronoun and an antecedent in a specific upcoming position — namely, the subject of the following main clause. Thus, the appearance of a gender-mismatching NP in that position yields an identifiable gender mismatch effect, leading to longer reading times. Processing difficulty occurs in such a context because structural syntactic strategies favour coreference, but morphological information does not allow it. The gender mismatch effect was also observed by Kazanina, Lau, Yoshida, Lieberman and Pilliphs (2007) using a self-paced reading task: reading times were slower in gender mismatch conditions. Interestingly, though, this effect was not found in contexts in which Principle C was violated, as the parser concentrates only on potential antecedents that are not subject to structural constraints on coreference (see Cowart & Cairns 1987 for similar results). The reason that structural constraints are applied at an early stage is assumed to be due to parsing efficiency (limiting the set of possible representations) rather than architectural priority. The authors conclude that the grammatically constrained active search mechanism operative in the processing of backward anaphora is comparable to the mechanism available in the processing of wh-dependencies, another type of a long-distance dependency (Crain & Fodor 1985, Stowe 1986). Analogous results for coreference 14 restrictions in backward anaphora structures have also been observed beyond English in languages such as Japanese (Aoshima, Yoshida & Phillips 2009) and Russian (Kazanina & Phillips 2010) using on-line techniques. Evidence from off-line studies has also indicated the parser’s tendency to resolve the dependency as soon as possible in the matrix subject position of a backward anaphora sentence. In an interpretation task conducted by Sorace and Filiaci (2006), both Italian natives and near-natives selected the subject of the matrix clause (la mamma) as the antecedent of the null pronoun in (13) in 85% of the cases. (13) Mentre___ i si mette il cappotto, la mamma i dà un bacio alla figlia. ‘While she i is wearing her coat, the mother i kisses her daughter.’ 1.1.3 Third-person (personal and demonstrative) pronouns The third-person pronoun has been variously described as “that of which is spoken” or (in Semitic grammatical traditions) either “that which is hidden” or “that which is absent” (Dixon 2010: 489). This final case represents the oftcited expression popularised by Benveniste: “celui qui est absent”, the nonpersonne (Benveniste 1966: 228). Other linguists including Lyons (1977: 638) have regarded it as a ‘negatively defined’ category with respect to the first and second person — that is, “some person or thing which is neither speaker nor addressee”. Third-person pronouns vary a great deal crosslinguistically, as can be seen in the types of features they encode. German, French and Spanish thirdperson pronouns (er, il and él ‘he’, respectively), for instance, encode gender, number and case — third-person masculine nominative singular pronoun — , whereas in languages like Finnish, the third-person pronoun hän is marked for number and case but not gender. In addition, in English there is a thirdperson pronoun to refer to inanimate referents (it), but the Spanish thirdperson subject pronouns él/ ella ‘he/ she’ must necessarily refer to people. 3 Another point where languages differ in their personal pronominal inventories concerns whether phonologically null pronominal forms are allowed or not in a given grammar. The grammars of English and German, for example, generally allow no phonologically null forms in subject position, except for diary-drop and topic-drop cases (e.g. ‘Saw a film yesterday’ and Bin gleich da ‘I’ll be right back’). However, omission of pleonastic (nonreferential) pronouns is sometimes required in German (Morgen wird (*es) getanzt ‘Tomorrow we’ll dance’). Languages like Italian and Spanish, in con- 3 No explicit pronouns are used when referring to inanimate subjects: Compré un sillón. (*Él) tiene tapizado de cuero. ‘I bought an armchair. It has leather upholstery’ (Fernández Soriano 1999: 1220). The Spanish neuter subject pronoun ello shows limited use, and when it is used, it refers to abstract concepts or to whole ideas, never to concrete things: Yo sé de ello. ‘I know about it (a concept)’ (Hill & Bradford 2000: 125). 15 trast, have both phonologically null referential and non-referential subject pronouns (Vengo ‘I’m coming’; es posible, ‘it’s possible’) (Harbert 1995). This grammatical possibility of argument omission has been assumed to be linked to the rich inflection displayed in the verbal system (Rizzi 1982, 1986), although argument omission is also possible in languages without any verbal inflection, such as Chinese (Huang 1984). A great deal of evidence suggests that the category ‘third-person personal pronoun’ may not be a characteristic of all languages of the world, but only of some. More precisely, several languages show a ‘1/ 2 pronominal system’ that lacks genuine third-person pronouns; their functional roles are often taken over by demonstratives, classifiers or other grammatical forms (Bhat 2004). Such languages, among them Turkish and Basque (Bhat 2004: 141), are known as two-person languages. The relationship between thirdperson pronouns and demonstratives in two-person languages is limited to one of the sets of demonstrative pronouns (or encompasses all of them; e.g. Lak, an East-Caucasian language) or is derivationally related to them (e.g. Khasi, an Austroasiatic language). This affinity between demonstratives and pronouns is also seen in Romance languages such as French, Spanish and Italian, where third-person pronouns developed from demonstrative pronouns. However, in languages like Spanish, third-person pronouns properly form part of the pronominal system from a synchronic point of view, as they differ from demonstratives; hence, Spanish is a three-person language 4 (Bhat 2004: 14, 289). The fact that some languages lack formally distinct third-person pronouns in the pronominal system, with demonstratives fulfilling part of their functional role, makes it important to precisely define the term ‘demonstrative’. Deictic expressions that “point” at an object, location or person (apart from the speech participants) are traditionally divided into three semantic categories: person, place and time (Bühler 1934). Demonstratives are place or spatial deictic expressions that denote the relative distance of the speech situation vis-à-vis the deictic centre (Diessel 1999: 36). The pragmatic functions of third-person pronouns and demonstratives are closely related, since both referential items can take their reference in two ways: (a) from the extralinguistic context, referring to an object present in the environment (as when someone says Look at her! while pointing at a woman passing by), or 4 Bhat has analysed the relationship that exists between third-person pronouns and demonstratives crosslinguistically (Bhat 2004: 277-291). His conclusions are that (i) Spanish third-person pronouns and demonstratives are not related to each other, (ii) French and German demonstratives and third-person pronouns are related by gender markers and (iii) in English, the third-person pronoun is related to all demonstratives, which is in line with Cardinaletti and Starke’s (1999b: 283) argument that the plural pronoun them is a demonstrative. However, this conclusion is questionable from a synchronic point of view, since the difference between English pronouns and demonstratives is clear (Kibrik 2011: 125). Italian was not included in Bhat’s sample. 16 (b) from the discourse, referring back to what has been said earlier (Dixon 2010: 489). The first type of reference (a) is sometimes regarded as exophoric (Diessel 1999, Halliday & Hasan 1976) or deictic, which is the prime function of demonstratives — to orient the hearer outside of the discourse in the surrounding situation (often accompanied by a pointing gesture). The second usage, organising the information flow in the ongoing discourse by keeping track of prior participants, is primarily fulfilled by third-person pronouns and is referred to as anaphoric. Although these differences in the usages of third-person pronouns and demonstratives may lead to complementary distribution, it is perfectly possible for a deictic term to be used anaphorically (Lyons 1977: 676). The same is also true for the deictic role of pronouns, as in It was him (accompanied by a pointing gesture), which is similar to It was this. However, such deictic functions in pronouns are secondary. Despite similarities in the functions of third-person pronouns and demonstratives, which may sometimes result in overlap, recent studies have shown that languages using two lexical anaphoric forms (personal vs. demonstrative pronouns) have distinct referential properties. Both off-line and on-line evidence (eye-tracking and sentence completion tasks) from Finish has suggested that the interpretation of the personal pronoun hän ‘s/ he’ is most strongly influenced by syntactic role, whereas the demonstrative pronoun tama ‘this’ is sensitive to word order and information structure (Kaiser & Trueswell 2008). In intersentential anaphora, hän is interpreted as referring back to preceding subjects, regardless of word order (SVO, OVS). The demonstrative tama, in contrast, is assigned to postverbal non-topical referents (i.e. O in SVO and S in OVS order). Based on these findings, Kaiser and Trueswell (2008) conclude that salience cannot be described by a singlefactor concept onto which pronouns and demonstratives can be mapped. Rather, anaphora resolution requires a model with multiple constraints in which referential forms can show different degrees of sensitivity to several factors, as captured in the Form-Specific Multiple-Constraints approach (Kaiser & Trueswell 2008). Interestingly, studies on pronouns (ta ‘s/ he’) and demonstratives (see ‘this’) in Estonian, a Finno-Ugric language with flexible word order that is closely related to Finnish, have revealed the same patterns (Kaiser & Vihman 2006). Personal pronouns show a primary sensitivity to the syntactic role of the antecedent, in contrast to demonstrative pronouns, which do not merely refer to entities with lower-ranked syntactic roles but rather to non-topic antecedents. Demonstratives’ preference for a non-topical antecedent has also been found in other languages including Czech (Sturgeon 2008), Dutch (Comrie 2000, Kaiser 2011, van Kampen 1997), Russian (Kibrik 1996, Krasavina & Chiarcos 2007) and Swedish (Mörnsjö 2002). Evidence for the validity of the Form-Specific Multiple-Constraints approach to intersentential anaphora resolution can also be found in non-null- 17 subject languages such as German, where different referential properties have been observed between personal pronouns (er, sie ‘(s)he’) and demonstratives (der (m.), die (f.)). By means of an off-line judgement task and eyetracking experiments, sentences such as (14) in both SVO and OVS word order were tested by Wilson (2009, see also Wilson, Keller & Sorace 2009) in native German speakers, advanced L2 German learners and German attrited speakers living in the UK. The results of native speakers showed that the topicality of the antecedent appears to be the main factor affecting demonstratives, which were consistently interpreted as referring to a non-topical (i.e. postverbal) antecedent regardless of their syntactic role (subject or object). On the other hand, both topicality and the syntactic role were relevant for personal pronouns in native speakers, who showed no clear preference in the OVS condition but a preference for the subject in the SVO condition. (14) Der Kellner i erkennt den Detektiv j , als das Bier umgekippt wird. Er i / Der j ist offensichtlich sehr fleißig. ‘The waiter i recognises the detective j as the beer is tipped over. He i / DEM j is apparently very hard working.’ The pattern of results between German natives and both L2 and attrited speakers was similar for pronouns. However, variability was observed in the antecedent choice of demonstratives, with L2 learners and attrited speakers showing no clear preference or only a slight preference for the subject. 5 As Sorace points out (2011), there seems to be a division of labour between German personal and demonstrative pronouns, similar to what Carminati (2002) proposes for Italian null and overt pronouns; this will be discussed in Section 1.4. 5 These observations are consistent with previous off-line and on-line findings on German natives’ different referential preferences of the two anaphoric forms (corpus studies by Bosch, Katz & Umbach 2007 and Bosch, Rozario & Zhao 2003 and self-paced reading and completion tasks by Bosch, Katz & Umbach 2007 and Bosch & Umbach 2007). See also Diessel (1999) for more information on German demonstratives. 18 1.1.4 Null vs. overt subject pronouns Thus far, the characteristics of overt tracking devices in subject position have been provided, but the grammar of certain languages allows the omission of subject pronouns in finite clauses (represented by pro in the traditional generative grammar; Chomsky 1982). This occurs in pro-drop or null-subject languages like Italian or Spanish, where overt and null pronouns alternate. If a language has two forms in its pronominal system, they are likely to have different, specialised functions that are reflected in both how pronouns are used by the speakers and how they are interpreted by the perceivers. Thus, despite subject realisation being grammatically optional, discoursepragmatic constraints account for the alternation of null and overt subject pronouns. As shown in the Spanish examples in (15), the use of a phonologically realised lexical or pronominal subject is required in contexts of emphasis (15a) or contrastive focus constructions (15b) in which the emphasised element (intonational stress) contrasts with another constituent of the same category. Overt subject realisation is also needed in contexts when new information 6 is provided, such as answers to questions (informative focus), as in (15c). In addition, overt pronouns are also used to signal switch reference when a new referent is introduced, as in (15d), since they are specified for the interpretable 7 feature [+topic shift] (Sorace 2000). By contrast, overt subjects are pragmatically odd (redundant) when there is no topic shift, but the topic is maintained in the discourse (15e). In such cases, null pronouns are appropriately specified as [-topic shift]. (15) a. Tú lo sabías. ‘You knew it.’ b. LUIS/ ÉL/ *___ es un amigo, y no toda esa panda de hipócritas. ‘LUIS/ HE/ *he is a friend, unlike that whole bunch of hypocrites.’ (examples from Brucart 1987: 216) 6 Brucart (1987: 219) accounts for these uses of overt pronouns in a null-subject language like Spanish via his Principle of Pronominal Lexicalization, which, in his view, is part of a more general requirement to express any new information lexically: any pronominal that contributes new information must be expressed phonetically. 7 In the Minimalist version of the Principles and Parameters framework (Chomsky 1995), a distinction is made within the category of formal features between semantically interpretable features and uninterpretable features (such as Case and Agreement on verbs). 19 c. Quién vino? Él/ Mario/ *___ vino. ‘Who came? He/ Mario/ *___ came.’ d. Hoy no fui a trabajar. Pepe/ él / *___ pensó que estaba enferma. ‘Today I did not go to work. Pepe/ he thought I was sick.’ e. Pepe no vino hoy a trabajar. *Pepe/ ? él/ ___ estará enfermo. ‘Pepe did not come to work today. He must be sick.’ (examples from Montrul 2004: 176) Chomsky’s Avoid Pronoun Principle makes explicit predictions for the use of null and overt pronouns. According to this principle, in languages where an overt/ null pronominal alternation is permitted, a null pronoun is preferred over an overt pronoun whenever possible (Chomsky 1981: 65). Thus, the null pronoun is preferred where a local coreferential or bound interpretation is intended. Consequently, in contexts in which a coreferential or bound zero anaphor may occur, the use of an overt pronoun is assumed to solicit disjoint reference. According to Haegeman, this principle follows from a postulated general principle of economy, as “the omission of the subject pronoun requires less effort than the overt expression of the pronoun, and therefore, subject pronouns will only be present when the added effort of overtly expressing them has some yield” (Haegeman 1994: 217). A more recent version of Chomsky’s (1981) proposal is Cardinaletti and Starke’s (1999a) Minimise Structure Principle — a version of the economy of representation (Chomsky 1995) — whereby more deficient forms have less structure. According to this principle, the most deficient pronoun among the universal tripartite distinction of pronouns in (16) must be chosen if possible (Cardinaletti & Starke 1999a: 198). This tripartition is the result of the study of Germanic pronouns and the parallel study of some “exceptional” Romance pronouns that turned out to be neither clitic nor strong but ‘weak’: (16) personal pronouns strong non-strong: (deficient) weak clitic The division in (16) represents a competition amongst pronomimals ruled by the Choice Principle (Cardinaletti & Starke 1999a) that accounts for the choice ordering for strong and deficient forms: clitics are chosen over weak pronouns, and weak pronouns are in turn chosen over strong pronouns. 20 Underlying the classification in (16) are hierarchically organised deficiencies across a variety of linguistic domains such as phonology, morphology, syntax and semantics: clitic (severe deficiency) < weak (weak deficiency) < strong (no deficiency). In addition, these deficiencies are interrelated — that is, a weak form is semantically dependent (dependent on the presence of an antecedent), morphologically deficient (morphologically reduced) and phonologically deficient (e.g. it cannot bear contrastive stress). In Cardinaletti and Starke’s account, the null argument in matrix subject position represented by pro is a weak pronoun, whereas the overt subject pronoun instead falls under the category of strong pronouns. The reference assignment of these pronominal forms in terms of semantic deficiency is formulated under semantic asymmetry (Cardinaletti & Starke 1999a: 154): (17) Semantic asymmetry #1 Deficient personal pronouns must have an antecedent prominent in the discourse. Thus, pro, being a deficient element in Cardinaletti and Starke’s classification, “must associate to an antecedent prominent in the discourse (i.e. discourse-internal co-reference: ‘sphericity’, ‘old information’, etc., but also impossibility in most contrastive contexts, or with ostension)” (Cardinaletti & Starke 1999a: 192). This is not the case with a strong pronoun like the overt pronoun, which need not be associated with a prominent antecedent. We will come back to Cardinaletti and Starke’s account in Section 1.3.1 when discussing a number of properties that make personal pronouns different from demonstratives. Another proposal seeking to explain the interpretive differences in pronoun (non-)realisation causes is that of Luján (1985, 1986), whose focus is on stress. The example in (18a) displays null and overt pronouns’ distinct interpretive behaviour in a preposed adverbial construction, a context in which they may both appear. Whereas the null pronoun is naturally understood as coreferent with Juan, the use of the overt pronoun in the same context induces disjoint reading (i.e non-coreference) despite the syntactically possible coreference between the R-expression Juan and the preceding overt pronoun. In other words, backward pronominalisation with a lexical pronoun is impossible in contexts in which a null pronoun may be coreferent with a following antecedent. Surprisingly, this contrastive behaviour between null and overt pronouns does not occur when the adverbial clauses are postponed (18b). In such a context, both null and lexical pronouns may be interpreted as coreferential, even though, as Luján observes, “in these examples the null pronouns would lend themselves to the coreferential reading” (Luján 1985: 427) and the disjoint interpretation would be more natural with the overt pronoun. 21 (18) a. Cuando ___ i / él *i/ j trabaja Juan i no bebe. When ___/ he work( PRS .3 SG ) John NEG drink( PRS .3 SG ) ‘When he works, Juan doesn’t drink.’ b. Juan i no bebe, cuando ___ i / él ij trabaja. John NEG drink ( PRS .3 SG ) when ___/ he work ‘Juan doesn’t drink when he works.’ (19) a. When HE *i/ j / he i works, John i doesn’t drink. b. John i doesn’t drink when he i / HE i works. Interestingly, Luján observes that the distinct interpretive behaviours of the null and overt pronouns in the preposed adverbial construction displayed in (18a,b) are parallel to those found in English neutral (unstressed) and stressed pronouns in (19a,b). The neutral and stressed forms differ in interpretation only in preposed adverbials (19a), not in postposed adverbials (19b). According to Luján, these contrastive interpretations between preposed and postposed adverbial clauses can be explained in terms of the universal precedence constraint, according to which “stressed pronouns cannot precede their antecedent where they alternate with unstressed pronouns” (Luján 1986: 249). Luján’s correlation between Spanish null and overt pronouns and their English counterparts has been criticised by Carminati (2002: 320), who finds evidence of Italian anaphoric dependencies and proposes the alternative correspondence shown in (20): (20) Italian English Subject antecedent Ø pronoun [neutral] Non-subject antecedent pronoun [neutral] pronoun [neutral] Carminati (2002) states that unstressed English pronouns may correspond to both null and neutral (unstressed) overt Italian pronouns. When a pronoun refers to a subject antecedent, a neutral overt pronoun is used in English and a null pronoun in Italian, as seen in (20). On the other hand, when a pronoun retrieves a non-subject antecedent, a neutral overt pronoun is used in both languages. The counter-examples of (21) in English would require a neutral overt pronoun for both reference-continuance and switch-reference contexts. (21) Quando Mario i ha telefonato a Maria j ___ i / lei j era appena tornato a casa. ‘When Mario i telephoned Maria j , he i / she j had just returned home.’ (example from Carminati 2002: 316) 22 To sum up the main points of Section 1.1, we have seen that in order to clarify the question of what makes a good antecedent for a subject pronoun, several factors need to be taken into account. The effect of clause order in relation to anaphora directionality is evident in cases of unavailability of coreference between a matrix subject pronoun and a following lexical subject (R-expression). The type of overt anaphoric form also has an effect on antecedent choice, since although anaphoric demonstratives may functionally be similar to personal pronouns or may even fulfill their function (in twoperson languages (Bhat 2004)) they differ in their antecedent preferences, as proposed by the Form-Specific Multiple-Constraints approach (Kaiser & Trueswell 2008). Finally, we have discussed the fact that the alternation of null and overt pronouns in null-subject languages is not optional. For example, when a change in reference is needed, an overt subject should be used, since overt pronouns in null-subject languages are specified for [+topic shift] (Sorace 2000), whereas a null subject is required when there is no switch reference. The following sections summarise the different theoretical models of anaphora resolution. The selection and interpretation of different kinds of referring expressions are discussed in Section 1.2 and are followed by an explanation of syntactic restrictions in pronoun-antecedent relations in Section 1.3. The last section of Chapter 1 deals with the opposite antecedent preferences of null and overt pronouns in null-subject languages. 1.2 A discourse-based account: Accessibility Theory In discourse-based theories, the choice of referring expression is closely related to the notion of salience, which is in turn determined by relevant linguistic structural features such as syntactic role and word order, as well as by semantic features like topicality and animacy. Existing approaches to the relationship between discourse salience and anaphoric expressions differ in terms of what they view as relevant factors that establish entities as more or less salient, but they uniformly assume that more salient entities are referred to with the most reduced form — in most cases, a null subject or a pronoun, depending on the language. Among the theories of reference that argue that the cognitive status of referents can be characterised in terms of a graded scale, Ariel’s (1990, 2001) Accessibility Theory has enabled good predictions for the distribution of different types of referring expressions based on text analyses in English and Hebrew. Accessibility Theory, regarded as an extension of Chafe’s (1974) and Givón’s work (1983), assumes that the choice of a particular referring expression is determined by the degree of availability (i.e. accessibility) of mental representations. The speaker signals the addressee with regard to how accessible the mental entity represented by the referring expression is to him. Thus, there is a correlation between the degree of accessibility in memory and the morphological encoding of a referent. 23 A simplified version of Ariel’s hierarchy of referring expressions known as accessibility markers (depicted in (22)) shows that pronouns are more accessible than demonstratives within the hierarchy. (22) High accessibility Agreement markers (pro, reflexives) Cliticized pronoun Unstressed pronoun Stressed pronoun Stressed pronoun + gesture Proximal demonstrative (+NP) Distal demonstrative (+NP) Definite description Name Low accessibility Figure 3. Accessibility hierarchy for discourse referents (adapted from Ariel 1990) Ariel argues that the form-function correlations in the accessibility hierarchy are not arbitrary. Three relate to the coding principles for determining the association of a particular referring expression with a degree of accessibility: informativity (the amount of lexical information), rigidity (the selection of a unique referent) and attenuation (the amount of phonological material). As these principles overlap a great deal, it is expected that the more informative, rigid and unreduced a referring expression is (proper names or full DPs), the better it is at referring to a less accessible referent. On the other hand, the less informative, flexible and reduced a referring expression is (clitics, agreement markers and the null pronoun pro), the better it is at retrieving a highly accessible referent. Thus, the null pronoun will refer back to a more accessible referent than the overt pronoun, since null pronouns are higher in the hierarchy in (22). The degree of accessibility illustrated in (22) is measured by multiple factors, including (a) distance between the first mention vs. second mention (recency being linked to a high degree of accessibility), (b) competition (the larger the number of potential antecedent candidates, the lower the degree of accessibility), (c) salience (determined by syntactic role, order of mention, high vs. low physical salience, etc.) and (d) unity (the lower the number of syntactic or text boundaries (clause or paragraphs), the closer the relationship between the antecedent and the referring expression). The last factor in Ariel’s theory, unity, predicts that higher accessibility markers should be used when the relation between the antecedent and the anaphor is close. It is important to note that Ariel’s proposal does not provide details of how the four factors that determine a discourse entity’s degree of accessibility inter- 24 act with each other, e.g. whether competition outweighs salience, or vice versa. The reason for this is that the degree of accessibility of a given entity is measured after analysing all the factors. To summarise Section 1.2, discourse approaches such as Ariel (1990) assume a converse matching of antecedent salience (in terms of the degree of accessibility in memory) and anaphora complexity (morphological encoding of a referent): maximally salient antecedents are referred to by minimally complex anaphora, and vice versa. Factors such as distance and competition have an effect on the degree of accessibility: the shorter the distance and the fewer the number of potential antecedents, the more accessible the referring expression. Accessibility Theory makes good predictions with regard to the type of relationship (close or distant) between an antecedent and an anaphor. Reflexive-antecedent relations precede pronoun-antecedent ones on the scale; these in turn show a closer relation than name-antecedent relations. However, discourse-based accounts are not sufficient by themselves to account for all anaphoric dependencies. Structural approaches (such as the Binding Theory discussed in the following section) will shed some light on the syntactic conditions at work in the interpretation of sentences with anaphoric elements. 1.3 A generativist account: Binding Theory The notions of reference and anaphoric relations have been extensively discussed in the generative approach to language acquisition, as will be discussed in Chapter 2. The distribution of anaphoric forms became a central issue in the 1980s, as it was believed that the syntactically constrained pattern of anaphora would reflect the deeper properties of Universal Grammar. In other words, anaphora was regarded as ‘the window into the mind’: the key point in furthering our understanding of the nature of the human brain (Chomsky 1982: 23). Likewise, studies of anaphora helped answer the question of how children know more about language than they could have learned by observation and imitation of adult usage (called the ‘logical problem of language acquisition’, or ‘Plato’s problem’ in Chomsky’s terms). Among the accounts that impose purely structural restrictions on the representation of anaphoric relations, Chomsky’s (1981) influential Binding Theory (BT) — a subtheory of Government and Binding Theory (Chomsky 1981, 1986) — laid the groundwork for ample discussion on the (un)availability of coreference. BT is the set of conditions or properties that deals with anaphoric relations and the restrictions found between coreferents, including definite and reflexive pronouns and the traces left by movement (Goodluck 1991: 72). These constraints are considered to be operative in every language, and it is expected that they will be revealed quite early in the process of language development, once all structures required to express 25 binding relations are productive in the child (see Section 2.2.1 for the relevant literature). In BT, the morphological form (23) and the syntactic relation between the NPs (24) determine whether coreference is possible or not between two NPs. According to (mostly) morphosyntactic criteria, there are three types of NPs (Büring 2005: 3): (23) a. Reflexives and reciprocals (‘anaphors’): himself, herself, each other, one another, etc. b. Non-reflexive pronouns (‘pronominals’): he, she, him, her, etc. c. Full NPs including names (‘R-expressions’): the baroness, this, Peter, etc. The syntactic approach to indexing relationships between nominal expressions comprises three principles, as illustrated in (24), each of which regulates the distribution and interpretation of one specific type of NP mentioned in (23): (24) a. Principle A: An anaphor (e.g. a reflexive or reciprocal) must be bound in its domain. b. Principle B: A pronominal is free (not bound) in its domain. c. Principle C: An R-expression (e.g. a name) must be free (everywhere). (adapted from Chomsky 1981: 188) The constraints in (24) applying to the NPs in (23) are determined by the notion of binding introduced in Section 1.1.1. In contrast to R-expressions, anaphors and pronouns lack inherent reference. More precisely, anaphors need an antecedent for their interpretation, and pronouns inherently specify certain properties of the referent (gender, number or case), but for a complete determination of the referent, contextual information is needed. In contrast, given that R-expressions have independent reference, they do not need an antecedent; in fact, they do not tolerate binding from another element, as described in Principle C. Although BT was proposed to govern the assignment of reference to different types of nominal expressions crosslinguistically, it was soon discovered that it can only explain one type of anaphoric relation known as variable binding (bound variable anaphors). There are also coreference relations not represented at any syntactic level that are not directly licensed or prohibited by syntactic constraints. The syntactic binding embodied in the three principles in (24) represents one kind of anaphoric relation, but there is another kind of anaphora that does not involve binding: cases in which two expressions corefer or refer to the same individual despite not being interpreted as bound variables (Grodzinsky & Reinhart 1993, Reinhart 1983). In this case, 26 coreference comes into play; this includes situations like (25), where words written in capital letters indicate stress. (25) a. When Aladdin looks in the mirror, he doesn’t see Jasmine. Aladdin i sees HIM i . b. Pinocchio i ’s father saved him ij . (examples from Guasti 2004: 281, 282) Although Principle B blocks coindexation between the pronoun HIM and the R-expression Aladdin in (25a), they still corefer (a case of accidental coreference). In (25b), Pinocchio can be interpreted as the antecedent of the non-ccommanded pronoun him because no structural conditions are operative. In such a case, Rule I*, taken from Guasti and Chierchia (1999/ 2000) and inspired by Reinhart’s work (Grodzinsky and Reinhart 1993, Reinhart 1983), is operative: (26) Rule I* If a pronoun is not (semantically) bound by an NP A, it is generally interpreted as non-coreferential with A, unless it appears in an Evansstyle 8 context. (taken from Guasti 2004: 282) According to Rule I*, the default option is non-coreference when a pronoun is unbound. A coreferential interpretation is only allowed when an indistinguishable reading cannot be obtained by means of variable binding. This implies that in a context like (27a), a coreferential interpretation between the non-reflexive pronoun and Goofy would be excluded unless accidental coreference was intended, since a similar interpretation can be obtained through variable binding, as in (27b): (27) a. Goofy i admires him j/ ? i. b. Goofy i admires himself i . (examples from Guasti 2004: 283) Under Rule I*, variable binding should always be used when available because this mechanism is a more efficient and economical way to obtain an anaphoric relation than coreference. Why is variable binding preferred? One answer developed by Fox (1998) is based on the notion of semantic processing, according to which variable binding is less costly because it entails closing an open expression immediately, whereas corefence entails a longer 8 Evans-style sentences, commonly known as accidental coreference, refer to contexts such as Goofy i admires him i (Guasti 2004: 281) in which the pronoun him and the nominal expression refer to the same individual even though Principle B blocks coindexation. 27 search for an appropriate antecedent. In fact, research on the differences between pronouns and reflexives in terms of the on-line processing load observed in English adults points in this direction. For example, in a set of cross-modal priming studies, Piñango, Burkhardt, Brun and Avrutin (2001) found significantly higher reaction times for the interpretation of coreferring pronouns (28b) than for bound variable pronouns (28a). (28) a. Everyone i thinks that students like him i b. The teacher i thinks that students like him ij These differences in reaction times seem to stem from the distinct mechanisms used to resolve the anaphoric dependency. In (28a), the syntactic mechanism alone (i.e. variable binding) determines the only possible interpretation, whereas the interpretation of the referring pronoun him in (28b) is ambiguous, and coreference with a referent not mentioned in the sentence requires accessing information beyond syntax (i.e. discourse). Thus, whenever syntactic dependencies are attainable, they have priority over discourse dependencies, as the interpretation of (28a) is less costly than (28b) (for a discussion, see Burkhardt 2005). The discourse processing complexity hypothesis proposed by Piñango et al. (2001) has been supported by evidence from eye-tracking experiments conducted by Sekerina, Stromswold and Hestvik (2004) and by results on binding principles in children (see Chapter 2). Within the syntactically-oriented framework of Government and Binding, BT was an essential foundation for proving the validity of linguistic assumptions, but after the field’s progression to the Minimalist Program (Chomsky 1993), the incorporation of BT into the more recent generative approach has turned out to be problematic. In the Minimalist framework, a strictly derivational (non-representational) approach is taken with regard to the computational system, whereas the traditional BT is based on syntactic constraints at the representational level (Hicks 2009: 309). Nevertheless, a minimalised formulation of the classic binding conditions was ultimately found to be technically viable in terms of interpretative principles (Chomsky 1993: 43) (with D as an undefined local domain) by resorting to the copy 9 theory of movement (Hornstein, Nunes & Grohmann 2005). However, for the current study, we base our theorisation on the classic BT because in contexts in which matrix subject pronouns are followed by a subordinate clause, only Principle C can be tested when complex sentences are used. 9 Under the copy theory, a trace is a copy of the moved element that is deleted in the phonological component (in the case of overt movement) but is available for interpretation at the Logical Form (LF). If traces are copies, reconstruction effects may be captured at LF without the need to postulate non-interface levels of representation. The binding conditions can then be explained in terms of LF only, without Deep and Surface Structures (Hornstein, Nunes & Grohmann 2005). 28 Principle B is not operative in such complex sentences, since the pronouns in question have no local antecedents. 1.3.1 Personal pronouns vs. demonstratives Cardinaletti and Starke’s (1999a: 284) tripartite pronominal classification introduced in Section 1.1.4 involves a competition amongst pronominals, not between pronouns and antecedents. Because pronouns are not involved in ccommanded contexts with respect to their antecedents, the binding properties of strong and weak pronouns are not directly addressed in their analysis. However, one aspect of the classification scheme describes the properties of demonstratives as opposed to those of personal pronouns (Cardinaletti & Starke 1999b: 274-289). As has been noted in Section 1.1.3, the relationship between personal pronouns and demonstratives is close, especially in certain languages that lack real third-person pronouns and in which alternative referential devices like demonstratives play the pronominal role (the language-internal function). Thus, it is not surprising that the two anaphoric forms can be difficult to tell apart. However, Cardinaletti and Starke (1999b: 284) list a number of properties (at least five) that make demonstratives different from personal pronouns: (29) a. Demonstratives always have a special morphological marker (in English th-ese, th-ose, in German d-en, d-ie) never found on personal pronouns. b. Demonstratives may refer to non-human entities in contexts requiring strong forms (personal pronouns cannot). c. Demonstratives must be disjoint from any c-commanding antecedent (Principle C), while personal pronouns must be disjoint only from local antecedents (Principle B). d. Demonstratives, contrary to personal pronouns, cannot overrule their disjointness requirement through accidental coreference; pronouns may violate the local disjointness requirement (as in John i saw only HIM i ), which does not occur with demonstratives (*John i saw THIS ONE i in the mirror). e. Demonstrative systems typically make spatial distinctions of the near/ far type, while pronouns never seem to (compare the personal pronoun in French le/ lui ‘him’ vs. celui-ci ‘this one’, celui-là ‘that one’). As noted in point (29c), being disjoint from any c-commanding antecedent (as introduced in Section 1.3), demonstratives are usually subject to Principle C, but a “small set of trouble makers respects principle B on a par with personal pronouns” (Cardinaletti & Starke 1999b: 284). This is the case for the English them. The example in (30) illustrates the puzzling nature of them, 29 which differs from pronouns with respect to its humanness (i.e. it is able to refer to a [-human] referent in a context requiring a strong form because of its demonstrative morphology). At the same time, them differs from demonstratives with respect to binding conditions (behaving under Principle B). (30) a. I didn’t buy the cars i , because I liked neither them i nor their owners. b. *I didn’t buy the cars i , because I liked neither those i nor their owners. (examples from Cardinaletti & Starke 1999b: 283) Cardinaletti and Starke resolve the interpretative differences between (30a) and (30b) by arguing that demonstratives vary in their binding conditions across languages and that sensitivity to Principle C (29c) is not an inherent property of demonstratives, as opposed to the rest of the properties noted in (29). Based on the generalisation in (31), Principle B is assigned to the first available pro-form in a particular language, where personal pronouns are preferred to demonstrative pronouns, and Principle C to the rest. Thus, in languages like English, the first pro-form may be a demonstrative and still behave under Principle B, as there is no English third-person plural pronoun corresponding to the same set of φ-features (person and number in such a case). (31) Repartition of Pro-Form Binding a. B>C (i.e. Assign Principle B to the first pro-form, C to the rest) b. Personal Pronouns > Simple Demonstratives French le/ lui ‘him’ ce/ ça/ cela ‘this/ that’ English them those Italian lui/ lei ‘(s)he’ questo/ quello ‘this/ that’ (adapted from Cardinaletti & Starke 1999b: 288) To sum up, Section 1.3 has devoted special attention to the ambiguity in anaphoric relations whereby a distinction between bound variables and coreferential pronouns should be made, an aspect missing from the standard BT. Variable binding driven by syntactic constraints should be used whenever a coreferential reading is intended. In an anaphoric relation involving coreference, syntactic binding is not operative, and the pronoun is generally interpreted as non-coreferential (according to Rule I*) unless accidental coreference was intended. We have also seen that demonstratives show variability in their binding conditions crosslinguistically, whereas personal pronouns show adherence to Principle B. This stems from the fact that in languages lacking a third-person personal pronoun, the first available pro-form is a 30 demonstrative. Hence, demonstratives may behave under Principle B if they are the first pro-form in a given language and under Principle C if not. 1.4 A processing account: The Position of Antecedent Strategy One recent pragmatically motivated processing strategy that explicitly addresses the interpretive differences between null and overt pronouns is the Position of Antecedent Hypothesis proposed by Carminati (2002, 2005), later referred to as the Position of Antecedent Strategy (PAS). 10 This strategy, which was initially tested in Italian monolingual adults, explains the distinct behaviour of the two pronominal forms in terms of their antecedent choices in different syntactic positions and the correlation between form and function (Ariel 1990) (see Section 1.2): (32) The Position of Antecedent Hypothesis for the Italian null and overt pronouns in intra-sentential anaphora: the null pronoun prefers an antecedent which is in the Spec IP position, while the overt pronoun prefers an antecedent which is not in the Spec IP position (Carminati 2002: 57). Carminati stipulates that null pronouns show a strong bias towards the most prominent antecedent located in the Spec(ifier) I(nflectional) P(hrase) position (generally a subject). Their overt counterparts, in contrast, show more flexibility in their antecedent preferences, but tend to select a less prominent antecedent (i.e. a non-subject antecedent, usually an object). A violation of the bias expressed in (32) does not result in ungrammaticality, but rather in a pragmatic anomaly, since the PAS is operative at the syntax-discourse interface. Through a series of on-line and off-line experiments using self-paced reading tasks and questionnaires, the antecedent biases of null and overt pronouns in relation to a variety of antecedents 11 were tested in monolingual Italian adults. In the majority of cases, the linguistic items consisted of complex sentences with subordinate-clause order. 12 Overall, the findings sup- 10 Although Carminati originally named her proposal the Position of Antecedent Hypothesis, later studies that replicated her findings (including those of Filiaci et al. (2014) and Sorace & Filiaci (2006)) refer to it as the Position of Antecedent Strategy. This renaming is likely related to the fact that Carminati sufficiently demonstrated the validity of the hypothesis. 11 Carminati (2002) tested referential nominative and dative subjects, expletive and quantified subjects, all of which are assumed to occupy Spec IP at s-structure. 12 Carminati decided to test primarily subordinate-main clauses based on the idea that in this clause order, the surface representation of the subordinate clause must be kept in 31 ported the PAS, which predicts a division of labour, i.e. asymmetrical antecedent preferences between the two pronominal forms. In addition, the results revealed that overriding the overt pronoun’s bias seemed less costly than overriding the null pronoun’s bias, as demonstrated by the shorter reading times in the former case. In a two-referent forward anaphora sentence such as (33a), in which the pragmatic content does not help to disambiguate to whom the pronoun refers, the PAS applies differently to null and overt pronouns. Null pronouns (___) in such a context were interpreted as referring to the subject antecedent (81%), whereas overt pronouns were assigned to the object (83%). However, this preference for a non-subject antecedent with overt pronouns was disregarded by Italian adults in one-referent sentences such as (33b), where the overt lui ‘he’ was generally interpreted as coreferential with the subject of the preceding clause (86%). (33) a. Marta i scriveva frequentemente a Piera j quando ___ i / lei j/ ? i era negli Stati Uniti. ‘Marta i wrote frequently to Piera j when ___ i / she j/ ? i was in the United States.’ b. Gregorio i ha detto che ___ i / lui i/ ? j sará presente al matrimonio di Maria. ‘Gregorio i said that ___ i / he i/ ? j will be present at the wedding of Maria.’ (adapted from Carminati 2002: 78, 91) The coreferential reading with the overt pronoun in (33b) was not expected by Carminati. If the PAS held true in one-referent sentences to the degree it did in two-referent sentences (33a), a disjoint reference interpretation would be expected based on the overt pronoun’s preference for a non-subject antecedent. However, the overt pronoun preferred to take as its antecedent the subject from the preceding linguistic context rather than a referent outside the sentence, even when that meant violating its antecedent biases. As Carminati postulated, the overt pronoun’s preference for the subject could lie in the fact that “while processing isolated sentences, the processor may prefer to find an antecedent in the sentence itself rather than resorting to an extrasentential referent” (Carminati 2002: 89). Nevertheless, a statistically significant difference was found between the coreference rates of null and overt pronouns in (33b), (96% and 85%, respectively), which was interpreted by the author as a residue of the division of labour between the two pronominal forms. short-term memory until the main clause has been processed (Bever & Townsend 1979). 32 On the basis of these findings, Carminati concluded that there is a clear asymmetry in the antecedent assignment of null and overt pronouns. The preference of the null pronoun for the subject of the preceding clause is consistent, whereas the overt pronoun shows more flexibility (as illustrated by its variability in antecedent choice depending on the number of referents in the preceding clause). More precisely, the bias of the overt pronoun for a non-subject antecedent is more robust in ambiguous (two referents of the same gender) sentences (33a) than in unambiguous (two sex-differentiated referents or only one referent (33b)) ones for which a violation of the biases would not automatically lead to miscommunication. In order to confirm the universal applicability of the PAS, Carminati (2002) conducted a brief crosslinguistic survey of (one or two) speakers of several null-subject languages, 13 based on the presumption that if a language has two pronominal forms in its system, they should have distinct functions. From these crosslinguistic investigations, Carminati concluded that the null pronoun displays a special affinity for a subject antecedent across languages. In contrast, the behaviour of the overt pronoun across languages is more variable: it ranges from a situation in which the pronoun prefers to retrieve the direct object to cases in which it shows no particular preference for a subject or object antecedent. Positive results from other ‘agreement null-subject languages’ such as Romanian (Diaconescu & Goodluck 2004, Geber 2006) and Catalan (Mayol 2009) as well as from ‘discourse null-subject languages’ such as Korean (Ok Kweon 2011) and Japenese (Okuma 2012) in studies using off-line questionnaire tasks have shown that antecedent differences between null and overt pronouns hold beyond Italian. The validity of the PAS was also tested in Spanish by Alonso-Ovalle, Fernández-Solera, Frazier and Clifton (2002) in intraand intersentential anaphora contexts in a study based on questionnaires. Although the authors inferred that the PAS makes valid predictions in Spanish, a careful examination of the data shows that the antecedent preferences for the overt pronoun are at chance level, remaining ambiguous in intersentential two-referent contexts. This milder effect of the PAS could be due to the language change (Italian versus Spanish) or the type of anaphora tested (intrasentential versus intersentential). However, recent on-line comprehension evidence published by Filiaci, Sorace and Carreiras (2014) from a study involving monolingual Italian and Spanish adults supports the idea that the variability of the overt pronoun is not limited to an intralinguistic pattern. In fact, there are interlinguistic differences between null-subject languages. In a self-paced reading task, participants were asked to read intrasentential anaphoric sentences from Carminati’s materials (similar to (34)) 13 The informants were Spanish (2), Russian (1), Polish (1), Hebrew (1), Korean (2) and Japanese (1). 33 in which the temporarily ambiguous antecedent was semantically disambiguated by the plausibility of the sentence: (34) Dopo che Giovanni i ha criticato Bruno j cosí ingiustamente, __ j / lui j si è sentito offeso. Después de que Bernardo i criticó a Carlos j tan injustamente, __ j / él j se sintió muy ofendido. ‘After John i had criticised Bruno j so unjustly, he j felt offended. ’ Italian sentences containing an overt pronoun were read more quickly when they were associated with a change in subject reference (namely, to the object), but Spanish overt pronouns did not show any such behaviour, although there was a tendency towards a topic-continuity reading. However, these opposite patterns for the two overt pronouns were not observed in null pronouns. The results from both languages revealed faster reading times and lower error rates for null pronouns retrieving a subject antecedent than an object antecedent. These different biases of Italian and Spanish overt pronouns were explained in terms of Cardinaletti and Starke’s (1999a) typology of pronouns (introduced in Section 1.1.4). The Spanish overt strong pronoun was interpreted as being closer to the properties of weak pronouns like the Italian weak pronoun egli, as opposed to the strong pronoun lui. Interestingly, Filiaci et al.’s (2014) results in Spanish adult speakers from Spain contrast with those of Mexican-Spanish adult speakers in an off-line experiment. Shin and Cairns (2012) conducted a judgement task involving brief stories in which adult preference patterns were collected after participants listened to two sentences (one containing a null pronoun and the other an overt pronoun) in either topic-shift or topic-continuity contexts. Mexican- Spanish participants showed a preference for overt pronouns in intersentential topic-shift contexts in 83% of the cases. The different anaphoric contexts under study (intrasentential vs. intersentential anaphora) as well as some dialectal divergence (Spain vs. Mexico) may account for the apparently contradictory results for Spanish anaphoric pronouns. To summarise Section 1.4, processing accounts such as the Position of Antecedent Strategy (PAS) propose a division of labour between null and overt pronouns in terms of syntactic prominence. Despite showing variation in several aspects, 14 studies on the validity of the PAS have thus far concluded that there is a crosslinguistic validity of the null subject bias towards the subject. However, intralinguistic differences (depending on the number of referents) and crosslinguistic variance persist in terms of the biases that guide the resolution of overt pronouns, consistent with Carminati’s (2002) 14 Note that these studies display variation in anaphora type (forward vs. backward), context (intrasentential vs. intersentential), clause type (adverbial complex sentences (33a), VP-complement clauses (33b)) and methodology (ranging from on-line selfpaced reading tasks to off-line judgement tasks). 34 prediction. Interestingly, Schwartz (2011) casts doubts on the stable and consistent bias of the null pronoun for the subject, drawing attention to the high rates (50-60%) of non-subject antecedents for null pronouns in several studies (e.g. Belletti et al. 2007, Sorace & Filiaci 2006, Tsimpli et al. 2004). What the different accounts discussed in Chapter 1 have in common with respect to the distribution of null and overt pronouns is that more reduced forms (i.e. null pronouns) tend to refer to more accessible antecedents (the more topical entity, which usually correlates with the syntactically more prominent element, the subject). Conversely, less reduced forms (i.e. overt pronouns) correlate with antecedents that are less accessible/ syntactically prominent. We have also seen that based on a principle of economy, the most deficient/ weaker pronoun must be chosen if possible according to the typology of structural deficiency proposed by Cardinaletti and Starke (1999a). Thus, the deficient element, the null pronoun, is the default form, and strong pronouns such as overt pronouns are the ‘marked’ form. 35 2 Pronouns in child language The present chapter concentrates on reviewing studies on children’s interpretation of (non-reflexive) pronouns, a process that requires the integration of different levels of information (such as syntax and discourse pragmatics). Section 2.1 starts off by describing the various types of multiple acquisitions, focusing in particular on research on child L2 learners that has shed light on the extent to which these early learners of an L2 replicate the developmental path seen in L1 children. Next, Section 2.2 deals with the delays observed at the syntax-discourse interface by reporting a) results from monolingual children on the acquisition of binding conditions (Principles B and C, the structural requirements governing certain anaphoric relations), and b) findings on the discourse conditions for the distribution of null and overt subject pronouns in monolingual and bilingual children. Finally, Section 2.3 describes the idea that linguistic phenomena at the interface syntax-discourse are particularly vulnerable, as captured in the Interface Hypothesis (Sorace & Filiaci 2006), and summarises the distinct approaches taken by researchers seeking to account for the difficulty of bilinguals (and, to a lesser degree, monolinguals) in mastering linguistic structures at this particular interface. 2.1 Bilingual acquisition Typically, a child exposed to (any) language in normal circumstances (from birth) will succeed in acquiring a language in his or her first years of life. Nevertheless, there are several different types of language acquisition. If the acquisition process involves just one language, then it is defined as monolingual first language acquisition. As Klein (1986: 3) defines the term, “first language acquisition is thus primary in at least two ways: in terms of sequence (first) and in terms of (mostly life-long) importance”. If the acquisition entails more than one language from birth, then the process is known as simultaneous bilingual first language acquisition. However, if one language is learned from birth and another at a later stage in childhood, this is known as early sequential or successive acquisition (here referred to as child L2 acquisition), whereby the development of the L2 grammar begins once the majority of the L1 grammar is already in place. Finally, late sequential bilingualism, which occurs after the age of 8, is regarded as adult second language acquisition. These various types of multiple acquisitions have been classified by Meisel (2008: 59) in terms of the age of onset of acquisition: 36 • (bilingual) first language acquisition ((2)L1) (age of onset before or at the age of 3 to 4), • child L2 acquisition (cL2) (age of onset ranging between 3-4 and approximately 7), • adult L2 acquisition (aL2) (age of onset around 8 or later) A number of studies in bilingual first language acquisition carried out over recent decades have emphasised the early differentiation of the two languages by bilingual children (see, among others, De Houwer 1990, Genesee, Nicoladis & Paradis 1995, Meisel 1989). In addition, the development of each emerging grammar in the bilingual child is very similar to that of corresponding monolingual L1 grammars, despite temporary effects of crosslinguistic influence in some cases 15 (Döpke 2000, Hulk & van der Linden 1996, Yip & Matthews 2000). What bilingual first language acquisition and monolingual language acquisition have in common is the success, rapidity and uniformity of the process, properties that cannot be attributed to successive L2 acquisition. The outcome of second language acquisition is not always successful, and there is a great deal of variation across individuals and within learners themselves with regard to the speed at which learners acquire a second language. Initially, the fundamental differences between first and second language acquisition were primarily attributed to an age-related decline in languageacquiring capacity in relation to maturational changes; this idea was formulated in the Critical Period Hypothesis, originally proposed by Penfield and Roberts (1959) and later developed by Lenneberg (1967). By this point, however, it has been established that differences between L1 and L2 acquisition cannot be explained in terms of age-related maturational changes alone; several other factors — such as the more mature cognitive skills of learners, their socio-psychological development and dependence on previously acquired linguistic knowledge — also come into play at later ages of onset (Meisel 2013). The identification of the developmental period, the age range in which crucial changes take place in the course of successive language acquisition, has been widely discussed in the literature. Lenneberg (1967) identified puberty as the crucial period, but based on the current knowledge of the topic, the age of onset of 6-7 seems to be the range in which significant changes occur; ultimate attainment in second language acquisition seems to be possible only if the age of onset occurs in childhood, probably before age 8 (Abrahamsson & Hyltenstam 2009), despite noticeable earlier maturationalrelated changes around age 4 in certain aspects of grammar such as morpho- 15 For studies finding no evidence of crosslinguistic effects, see, for example, Meisel (1994) for French-German bilingual children and Paradis and Genesee (1996) for French-English bilingual children. 37 syntax (Meisel 2009, 2011), or even around 12 months for domains such as phonology (Hyltenstam & Abrahamsson 2003). Note that this is not to say that language as a whole becomes affected by maturational changes, but rather that some domains of grammars do. Moreover, it seems more appropriate to talk about multiple sensitive phases rather than a single critical period, as different components of grammar (like phonology and morphology) do not follow the same development timeline (see Eubank & Gregg 1999, Lee & Schachter 1997 and Seliger 1978). Despite cL2 learners being cognitively less mature than adult L2 learners, they are assumed to have greater success in achieving the ultimate attainment of L2 properties, and it has therefore been argued that they are guided by Universal Grammar (Felix 1991, Schwartz 1992). However, one issue that was hotly debated in the literature in the 1990s (similar to the issues debated in adult L2 acquisition) was parameter setting — that is, whether Universal Grammar is fully or partially operative at the early stages of cL2 language development, and the significance of the L1 influence (Clahsen 1991, Grondin & White 1996, Lakshmanan 1991). In recent years, in contrast, much of the research has focused on the status of child L2 acquisition itself and in comparison to adult L2 acquisition and monolingual and bilingual L1 acquisition (see, among others, Haznedar & Gavruseva 2008, Meisel 2007, 2008, Schwartz 2004, Unsworth 2005a,b). Despite Hyltenstam and Abrahamsson’s (2003: 575) statement that “At least up to AOs 6 or 7, all learners will automatically reach levels that allow them to pass as native speakers — provided that there is sufficient input and that the learning circumstances are not deficient”, early L2 learners do not always show native-like performance in their L2 since “to ‘pass as native speakers’ does not necessarily imply that learners have acquired native grammatical competences” (Meisel 2013: 78). In the domain of lexical argument structure, for example, differences have been found between the performance of cL1 and cL2 learners. Whong-Barr and Schwartz (2002) compare the acquisition of goal (to) and benefactive (for) double objects by L1 English children and L1 Korean and Japanese learners of English using an oral grammaticality judgement task. All the groups performed similarly in goal double objects by showing an overgeneralisation, but only the Japanese children overaccepted illicit benefactive double objects. The reason that the Korean children rejected illicit for-datives was attributed to transfer from their L1 grammar, where benefactive double objects are licensed by a bound morpheme. With regard to the domain of phonology where the popular belief is that children (unlike adults) attain native-like pronunciation in the L2, early learners sometimes do not succeed in eliminating a foreign accent; in the acoustic analysis of consonant sounds, child L2 learners may differ from their monolingual peers (see, e.g. Harada 2006 on cL2 Japanese and Yavas 2002 on cL2 English). 38 In contrast to the relatively limited number of studies on the phonological and lexical acquisition of cL2 learners, a substantial body of research has been produced on the acquisition of syntax-morphology, where there has been a long-standing debate on the similarities and differences between cL2 acquisition, child L1 acquisition and adult L2 acquisition in the domain of inflectional morphology and syntax. For instance, Schwartz (2004) suggests that child L2 learners at age 4 resemble adult L2 learners in the domain of syntax, but show more similarities to L1 acquisition in the domain of inflectional morphology. Similarities in inflectional errors between Dutch cL1 and cL2 were also observed by Blom (2008). However, Meisel (2008) finds some evidence of errors in the domain of inflectional morphology in cL2 French learners, which leads him to conclude that cL2 acquisition resembles adult L2 acquisition at the age of 4, exactly the opposite view as the one expressed by Schwartz (2004). Studies such as Schlyter (2011) on the acquisition of tense and aspect marking by Swedish cL2 learners of French and Meisel (2009) on the analysis of grammatical gender in German cL2 learners of French have also concluded that it is precisely in the morphological domain that child L2 learners resemble adult L2 learners and differ from child L1 learners. As argued by Unsworth (2005b) for the acquisition of object scrambling in Dutch, in a syntactic phenomenon driven by semantic factors, the pattern of results shown by child L1 learners differs from that of cL2 learners of Dutch. Unsworth (2005b) found that child and adult L2 learners go through the same developmental stages, characterised by an initial stage with L1 transfer in direct object scrambling over negation, thus distinguishing their learning pattern from the developmental path of child L1 learners. The results obtained from a comparison between cL2 acquisition and L1 acquisition in constructions at the syntax-discourse interface (which are highly relevant to the current study) will be discussed in Section 2.2.2.2. 2.2 The syntax-discourse interface Delays in the acquisition of linguistic structures that involve integration of information across grammatical and other cognitive modules (such as syntax and discourse-pragmatics) have been confirmed by contemporary empirical research. Children’s difficulty in integrating different levels of linguistic knowledge was first pointed out in studies on the acquisition of binding conditions in monolingual children (see Section 2.2.1), but it has become an even more prominent issue following the formulation of the Interface Hypothesis (see Section 2.3). The verification of this hypothesis has generated a large amount of research on the distribution and interpretation of null and overt subject pronouns in areas of bilingual first language acquisition, second language acquisition and attrition (see Section 2.2.2 for details). 39 2.2.1 Binding conditions in monolingual children 2.2.1.1 Principle B As suggested in Section 1.3, linguistic phenomena at the syntax-discourse interface level have been argued to be somehow more complex and costly (in terms of processing) than purely syntactic structures, as multiple types of information have to be integrated. While very young children have knowledge of syntactic properties — for example, with respect to the interpretation of reflexives, which is believed to be uniquely driven by syntactic constraints (namely, Principle A Every bear i is touching herself i/ *j (Chien & Wexler 1990)) — they reveal delays or difficulties with regard to the pragmatic knowledge required for the correct interpretation of non-reflexive pronouns (Principle B). Although such principles are believed to hold universally and be part of the innate language faculty, English-speaking 5and 6-year-olds often interpret an object pronoun as though it were an anaphor, accepting coreference with a local subject NP, as in *The boy i touched him i (Chien & Wexler 1990, Grimshaw & Rosen 1990, Jakubowicz 1984, Solan 1987; see Guasti 2004 for a review). This linguistic phenomenon, known as the Delay of Principle B Effect or as the Pronoun Interpretation Problem (PIP), 16 has been widely discussed by Ken Wexler and his colleagues (e.g. Avrutin & Wexler 1992, Chien & Wexler 1990). The asymmetry in the results between reflexives (Principle A) and non-reflexives (Principle B) seems to stem from the fact that the former can only enter into an anaphoric relation through a syntactic dependency (variable binding), whereas such a requirement is not imposed for non-reflexive pronouns, which are resolved via coreference or variable binding (see Section 1.3). However, crosslinguistic research has challenged the non-adult-like performance of English-speaking children. Children acquiring Romance 17 languages or Greek do not accept an interpretation in which the pronoun is anaphorically linked to a local antecedent, in contrast to English learners (see Baauw et al. 1997 for Spanish and Italian, and McKee 1992 and Padilla 1990 for Italian and Spanish respectively, Escobar & Gavarró 1999 for Cata- 16 Baauw and Cuetos (2003) find the name ‘Delay of Principle B Effect’ inadequate and prefer to refer to the phenomenon as the Pronoun Interpretation Problem for two reasons: (i) agrammatics also exhibit problems with Principle B even though they are not in the process of acquiring a language and thus delay is not adequate as a descriptor, and (ii) children do not show a delay with Principle B itself, as demonstrated by the positive results obtained in the interpretation of quantifiers. 17 See Cardinaletti and Starke’s (1995) syntactic account of the crosslinguistic differences between English and Romance languages with regard to Principle B. The morphosyntactic status of pronouns in the two groups of languages — full pronouns (English) vs. clitics/ Principle B demonstratives (Romance languages and Greek) — seems to be the primary reason for these differences. Thus, non-adult responses by children appear only in languages in which pronominal forms are homophonous with weak forms, such as in English. 40 lan, Hamann 2002, Hamann et al. 1997 for French, Sanoudaki 2003 and Varlokosta 2001 for Greek). There are divergent views on why children show delays in the interpretation of non-reflexive pronouns. Some researchers believe that children possess the necessary syntactic knowledge but experience pragmatic deficiencies in the course of language development that makes the sentenceexternal interpretation difficult to access. Chien and Wexler (1990), for example, propose that although English-speaking children have knowledge of binding principles, they lack the pragmatic rule P (contraindexed NPs are non-coreferential unless the context explicitly forces coreference), a rule very similar in outcome to Rule I* (introduced in Section 1.3). Alternatively, methodological artefacts have been proposed to explain children’s nonadult-like interpretations. For instance, Grimshaw and Rosen (1990) claim that children show a preference for accepting sentences in grammaticality judgement tasks (p. 196), and that for pragmatic reasons, children prefer to assign a discourse antecedent to a pronoun (p. 201). Along these lines, Baauw, Zuckerman, Ruigendijk and Avrutin (2011) also speculate on whether or not experimental design could mask the results concerning Principle B with pronouns. The authors propose that the Truth Value Judgement Task (a single picture for which the child has to judge whether an accompanying sentence correctly describes the picture) imposes a higher processing load than the Picture Selection Task (three pictures, of which the one that best depicts the test sentence must be chosen). In the Truth Value Judgement Task, unlike the Picture Selection Task, participants are forced to establish and evaluate the two possible readings (variable binding/ coreference; see Section 1.3), which is difficult due to processing reasons (Grodzinsky & Reinhart 1993, Reuland 1998). 2.2.1.2 Principle C The initial results testing children’s knowledge of Principle C were not positive with regard to structures, ruling out backward anaphora sentences such as (4b) *She i listens to music when Sarah i reads poetry but allowing (4d) When she i listens to music, Sarah i reads poetry. Early on, children’s grammar was not believed to be constrained by structure-dependent conditions like Principle C; it was instead thought to be determined by a linear constraint that would block a pronoun from being anaphorically linked to its antecedent in backward anaphora contexts (see Solan 1983, Tavakolian 1978). However, later studies such as those conducted by Crain and McKee (1985) showed that 3to 5-year-old English-speaking preschool children correctly rejected coreference between the c-commanding pronoun and the following R-expression in sentences such as *He i washes Goofy i 88% of the time. Since then, other studies (see Lust, Eisele & Mazuka 1992 for a review; see also Grimshaw & Rosen 1990, McDaniel, Cairns & Hsu 1990) have challenged the initial hypothesis about the linear constraint, instead arguing that children’s grammar is struc- 41 ture-dependent. The overall finding is that the likelihood of rejecting sentences that violate Principle C is higher than that of accepting them. External reasons such as memory limitations, nervousness, etc., seem to account for children’s errors (O’Grady 1997: 224). Adherence to Principle C has also been observed in children acquiring a null-subject language like Italian. Guasti and Chierchia (1999/ 2000) tested Italian children (average age of 4; 5) in a Truth Value Judgement Task — in contrast to most previous studies, using a quantified NP rather than a referential one. Children rejected the coreferential reading 89% of the time in contexts in which Principle C was operative, as in (35); however, in contexts of backward anaphora in which coreference was syntactically licit, they accepted the coreferential reading in 94% of cases. (35) ___andava sul cavallo a dondolo, mentre un musicista suonava la tromba. ‘He was riding the rocking horse while a musician was playing the trumpet.’ Before concluding this section, an interesting proposal involving backward anaphora put forth by Lust and colleagues in the 1980s, known as the Principal Branching Direction Parameter (Lust 1981, 1983, Lust & Mangione 1983), should be mentioned. According to this proposal, in language acquisition, the direction of anaphora (forward vs. backward anaphora) is believed to be constrained by the branching directionality of the particular language (Lust & Mangione 1983: 147). In other words, the constraint on anaphora predicts that forward anaphora will be preferred by children acquiring a right-branching language such as English. In contrast, children acquiring left-branching languages such as Korean and Japanese are not expected to show evidence of a predisposition for forward anaphora and may instead exhibit a predisposition for backward anaphora. (36) English: C’ > [C° [IP]] Right-branching Forward anaphora Sarah listens to music when she reads poetry Japanese: C’ > [[IP] C°] Left-branching Backward anaphora When she listens to music, Sarah reads poetry Figure 4. Direction of anaphora/ branching directionality parameter 42 These predictions were verified in both cases (for English, see Lust 1983, 1986; for Japanese, see Lust, Wakayama, Hiraide, Snyder & Bergmann 1982; for Chinese, see Lust & Chien 1984) using imitation tasks. However, a study by O’Grady, Suzuki-Wei and Cho (1986) on monolingual Korean children in which the same methodology was applied showed that the preference for forward pronominalisation exists even in left-branching languages (see O’Grady, Cho & Sato 1994 for similar results in Japanese). O’Grady et al.’s (1986) findings are in line with previous claims that forward patterns of pronominalisation are universally preferred (see, among others, C.S Chomsky 1969, Goodluck 1978, 1991 and Tavakolian 1977). Why should the acquisition device prefer forward patterns of pronominalisation? One possibility is that the processing mechanisms used for sentence comprehension seek to interpret pronouns as soon as possible (Reinhart 1986: 140). The existence of universal tendencies favouring forward anaphora has also been supported by Bley-Vroman and Chaudron (1990), who have harshly criticised the Principal Branching Direction after analysing Flynn’s findings on the application of the Principal Branching Directionality to L2 (see Flynn 1983, 1986, 1987 and Flynn & Espinal 1985). According to Bley-Vroman and Chaudron, the link between the branching direction parameter and anaphora directionality formulated under the Principal Branching Direction has no foundation in linguistic theory (Binding Theory 18 ). In addition, they claim that a preference for forward pronominalisation in English-acquiring children has been empirically supported, but a preference for backward anaphora in leftbranching languages has not been proven by the data. 19 Based on these contradictory results, further research involving more languages should seek to clarify whether the preference for forward pronominalisation is manifested in the early development of all languages, regardless of branching direction. 18 Reinhart (1986) also argues that Lust’s Principal Branching Direction Parameter is not related to the Binding Theory. 19 Bley-Vroman and Chaudron (1990) mention that the preference for backward anaphora sentences is only based on studies conducted by Lust and her colleagues for which the results cannot be deemed conclusive (Lust & Chien 1984 for Chinese, Lust & Mangione 1983 for Japanese), since the constructions under investigation largely involved coordinate conjunctions, which are not the same as pronominal anaphora. 43 2.2.2 Pragmatically appropriate use of subject pronouns in discourse contexts 2.2.2.1 Monolingual and simultaneous bilingual children In line with production data from simultaneous bilingual children (see Hacohen & Schaeffer 2007 on Hebrew-English bilinguals, Haznedar 2010 on Turkish-English bilinguals, Paradis & Navarro 2003 20 on Spanish-English bilinguals, Pinto 2006 on Dutch-Italian bilinguals, Schmitz 2003 on Italian- German bilinguals and Serratrice, Sorace & Paoli 2004 on Italian-English bilinguals), initial studies on the comprehension of null and overt subject pronouns in simultaneous bilingual children have also pointed to the unidirectional target-deviant use of overt subject pronouns. More precisely, bilinguals showed an overgeneralisation of overt subject pronouns in pragmatically infelicitous (topic-continuity) contexts in comparison to their monolingual peers in studies focusing on the mapping between a particular pronominal form (null vs. overt) and the felicitous pragmatic considerations for contextual appropriateness. However, evidence of the non-native-like use of null subject pronouns was not found. For example, Serratrice (2007) tested the referential property preferences of null and overt subject pronouns in forward and backward anaphora contexts by English-Italian bilinguals (age range: 6; 11-8; 4; mean age: 8; 2), age-matched monolinguals and native Italian adults. The results from a Picture Verification Task indicated no differences among the three groups in the antecedent assignment of null pronouns, but significant differences were observed with respect to overt subject pronouns in both forward and backward anaphora contexts. In forward anaphora sentences such as Il portiere saluta il postino, mentre lui apre la porta ‘The porter greets the postman while he opens the door’, overt pronouns were accepted as coreferential with the subject antecedent (il portiere) significantly more often by bilinguals than by monolinguals. The pattern shown by bilinguals was regarded as an instance of crosslinguistic priming. The routine of processing overt pronominal subjects as coreferential with a subject antecedent in English increased the likelihood of processing overt pronouns in the same way in Italian. For backward anaphora sentences such as Mentre lui versa il vino nel bicchiere, il cliente paga il conto al cameriere ‘While he pours wine in the glass, the client pays the bill to the waiter’, both monolingual and bilingual children accepted the subject as the antecedent of the overt pronoun significantly more often than adults, who predominantly chose the extrasentential referent as antecedent. Serratrice (2007) hypothesises that the infrequent choice of the extralinguistic referent in children could be related to the fact that a third plausible antecedent may exceed their processing capacities. 20 In this case, however, the child’s Spanish input included a higher number of overt subject pronouns than that exhibited by Spanish monolingual peers, and hence the possibility that the child was simply mirroring the parental input cannot be excluded. 44 Acceptance of pragmatically infelicitous overt pronouns in topiccontinuity contexts has also been observed in 8-year-old English-dominant English-Greek bilinguals (Argyri & Sorace 2007). However, as there was only one significant difference between English-dominant bilinguals and adults, these results do not provide the conclusive evidence of crosslinguistic influence 21 from English in the choice of Greek subject pronouns as the authors claim. More recently, child studies investigating the acceptability of null and overt pronouns in different discourse (topic-continuity [-TS] and topic-shift [+TS]) contexts in intrasentential anaphora condition have reported a bidirectional target-deviancy: overacceptance of overt pronouns in [-TS] contexts (resulting in redundancy), and an underacceptance of overt pronouns in [+TS] contexts (causing ambiguity). Sorace, Serratrice, Filiaci and Baldo (2009), for instance, observed that younger monolinguals (aged 6-7) and both Italian-English and Italian-Spanish bilinguals accepted significantly more overt pronouns referring to topic antecedents than both older monolingual (aged 8-10) peers and adults. More importantly, bilingual children also accepted some infelicitous null subject pronouns in [+TS] contexts, regardless of age and language combination, in contrast to prior production studies in children (e.g. Hacohen & Schaeffer 2007, Serratrice et al. 2004) in which the non-target-like performance of children was limited to overt pronouns. Using exactly the same materials as those of Sorace et al. (2009), Barquin and Costa (2011) tested two Spanish-Catalan bilingual groups aged 6-7 and 8-10 for comparison to age-matched Spanish monolingual children. Despite generally increasing preferences for null pronouns in [-TS] contexts and overt pronouns in [+TS] contexts with age, both younger (6-7) and older (8- 10) Spanish-Catalan bilinguals accepted significantly fewer Spanish overt pronouns in both [+TS] discourse contexts 22 than Spanish monolingual children. In other words, bilinguals preferred more null pronouns than monolinguals, regardless of the discourse context. Nevertheless, a careful analysis of individual data reveals that Spanish-Catalan bilingual children had significantly stronger and more consistent 23 preferences than Spanish monolingual children in both contexts. 21 However, crosslinguistic influence was observed in another syntax-discourse interface structure, subject placement in wide-focus contexts, and also in purely syntactic structures such as what-embedded interrogatives. 22 Likewise, data from the semi-spontaneous production of spoken and written Catalan narratives by Catalan-Spanish bilinguals aged 9 to 16 have shown that Catalan overt pronouns have no clear preference in selecting antecedents in both intrasentential and intersentential anaphora contexts (Bel, Perera & Salas 2010). The authors suggest that the behaviour of overt pronouns might be language-specific. 23 Preference for a particular pronoun in 75% of the items. 45 Redundant overt pronouns in [-TS] contexts and ambiguous null pronouns in [+TS] contexts 24 have also been reported in 6and 7-year-old Mexican-Spanish monolingual children in intersentential anaphora contexts. Shin and Cairns (2012) presented brief stories to 6to 15-year-olds and made the following observations: the ability to reject pragmatically inappropriate null pronouns in [+TS] contexts increases with age and reaches stability at an earlier stage than the ability to reject overt pronouns in [-TS] contexts. Whereas the preference for overt pronouns in [+TS] contexts develops earlier (around age 8) — when children familiarise themselves with perspectivetaking in complex linguistic tasks — the preference for null pronouns in [- TS] contexts is still non-adult-like at the age of 14. These results are in line with the findings of Sorace et al. (2009) in Italian monolingual children, for whom the sensitivity to interpret overt pronouns as signalling topic shift emerges earlier (by age 6) than the sensitivity to interpret null pronouns as topic continuity (by age 8). 2.2.2.2 Successive bilingual children In comparison to the large number of studies on syntax-morphology in cL2 acquisition mentioned in Section 2.1, research on the development of the syntactic and discourse properties of subject pronouns in successive bilingual children in comparison to cL1 learners, which is particularly relevant for the current study, has not been very extensive. One exception is a study conducted by Pladevall (2010), who explored the L2 acquisition of both syntactic and stylistic subject properties by British child L2 learners of Spanish. Specifically, the subject properties at the syntax-discourse interface — such as topic vs. focus contexts and co-reference in backwards anaphora contexts in null/ overt subjects, as well as lexical verb class and the presence of a [+strong] uninterpretable functional focus feature (Belletti 2000, 2003) — were analysed in preverbal/ postverbal subject contexts. Children in three different age groups (5-, 10and 17-year-olds) whose age of first exposure was at the age of 4 performed grammaticality and preference-judgement tasks. The results showed that the syntactic properties of Spanish are fully acquired at the age of 17. With respect to the acquisition of discourse properties, delays were observed in all age groups, and even 17-year-olds still showed minor residual indeterminacy, accepting some redundant overt subjects both in main clauses (such as ¿Qué decidisteis hacer ayer por la tarde? Finalmente nosotros decidimos ir de compras a Madrid ‘What did you decide to do yesterday afternoon? In the end, we decided to go shopping in Madrid’) and subordinate clauses (such as Cuando ellos trabajan, mis padres no vienen a dormir (ellos = mis padres) ‘When they work, my parents do not come home to 24 Bidirectional target-deviancy (pragmatically odd null and overt pronouns) has also been found in the production of pronouns by L2 speakers (Montrul & Rodríguez- Louro 2006, Rothman 2009 for L2 Spanish). 46 sleep (they = my parents)’). Pladevall (2010) concludes that child L2 acquisition is clearly different from child L1 acquisition in terms of purely syntactic aspects, but similar with respect to discourse properties in the 5and 10year-old groups, where processing limitations with respect to discourse properties affected both native and non-native children. In addition, similar to findings from English L2 adult learners of Spanish (Hertel 2003, Lozano 2006, Pérez-Leroux & Glass 1997), the results from 17-year-olds reveal that some optionality remains in properties involving discourse. 2.3 The Interface Hypothesis The observation that the formal features licensing null subjects are acquired without any effort at an early stage, whereas the discursive features are enduringly unsettled has gained relevance in recent years with the formulation of the Interface Hypothesis (Sorace & Filiaci 2006). According to this hypothesis, interface properties involving syntax and another cognitive domain may not yet be acquired at the time when narrow syntactic properties are completely acquired. This hypothesis first appeared in Sorace and Filiaci’s (2006) study on antecedent preferences for null and overt preverbal pronouns in intrasentential forward and backward anaphora contexts among English-speaking near-native speakers of Italian. The authors’ results showed that the L2 learners accepted the subject as the antecedent for the Italian overt pronouns (especially in backward anaphora) more often than Italian native speakers did. However, no significant differences were observed between the groups with respect to null pronouns, suggesting that the syntactic conditions for licensing null subjects had already been acquired. Similar asymmetric results pointing to the unidirectionality of deficits had already been observed in precisely the same linguistic phenomenon in the very early stages of L1 attrition (Tsimpli et al. 2004), as well as in bilingual first language acquisition (Serratrice, Sorace & Paoli 2004). 25 The conclusion reached from these studies is that linguistic structures at the syntaxdiscourse interface (such as anaphora resolution) are prone to residual optionality and long-term effects in endstate L2 speakers, emerging optionality in L1 attrition and protracted indeterminacy in bilingual L1 acquisition (see Sorace 2011, 2012 for a discussion). Note that optionality in the production or comprehension of properties located at the level of the syntax-discourse interface results in pragmatic anomaly and not ungrammaticality, as it 25 The overuse of overt pronouns for topic antecedents is found to varying degrees in different bilingual populations, as pointed out by Schwartz (2011): 23%-60% in early 2L1ers (Argyri & Sorace 2007, Serratrice 2007, Sorace et al. 2009), 30-50% in nearnatives (Belletti et al. 2007, Sorace & Filiaci 2006) and 22% in early attriters (Tsimpli et al. 2004). 47 would if violations at the syntax-semantics level occurred (Sorace & Keller 2005, Sorace 2006). The overextension of overt pronouns to null subject contexts results in the overt pronoun being the learner-default option at the mapping of syntax and discourse, which paradoxically contrasts with the linguistic default, i.e. the null pronoun (Tsimpli 2011), the weaker element (Cardinaletti & Starke 1999). Authors such as Schwartz (2011), however, have cast doubt on the default status of the overt pronoun for compensating inefficiency in syntax ‐ discourse mapping: if the overt pronoun were the default for bilinguals, they would be expected to use it at monolingual levels in [+TS] contexts, which is not the case. 26 In a similar vein, as O’Grady (2011) observes, a locality processing account also leads to the selection of topical antecedents for overt pronouns. More recently, researchers in favour of the Interface Hypothesis have claimed that the source of optionality and instability lies at particular interfaces, especially at grammar-external interfaces (Sorace & Serratrice 2009), which require the integration of syntax and contextual discourse-pragmatic information. This interface, often regarded as problematic, contrasts with linguistic phenomena at grammar-internal interfaces such as syntaxsemantics (formal features internal to grammatical representations; see Slabakova 2006, Tsimpli & Sorace 2006), which pose fewer difficulties than the external ones. White (2011), however, argues that the distinction between internal and external interfaces may be premature, and that we must be cautious in assuming that all linguistic interfaces are problematic or unproblematic to the same extent or that different linguistic structures at the same interface will exhibit the same behaviour. Similarly, other authors including Montrul (2011) and Slabakova and Ivanov (2011) have expressed doubts as to whether the instability is limited to certain interfaces. Despite much evidence pointing to persistent difficulty at the syntaxdiscourse interface, several results have contradicted the predictions of the Interface Hypothesis, calling its validity into question (see Domínguez 2013 for a discussion). For instance, although some studies (Rothman 2007, 2009) have shown that intermediate and advanced learners of Spanish have successfully acquired the syntactic constraints of null subject licensing but have not yet acquired the discursive properties that constrain subject expression (thus supporting the ‘syntax before discourse’ phenomenon), others have found evidence of problems in purely syntactic constructions (e.g. Argyri & Sorace 2007). Interestingly, researchers such as Lozano (2009) have concluded on the basis of corpus data that interface deficits are selective, with only a 26 Schwartz (2011) provides percentages from the study conducted by Sorace et al. (2009): bilinguals’ rates of overt pronouns hover around 62-70%, whereas monolinguals’ are around 84%. 48 subset of features being affected 27 in English-speaking learners of Spanish. In addition, in contrast to claims that the syntax-discourse interface is a principle locus of fossilisation in adult grammars (e.g. Valenzuela 2006), several studies have also revealed that optionality at the interface syntax-discourse is not permanent and that delays can be overcome in advanced-level studies (Domínguez & Arche 2008, Ivanov 2009, Iverson, Kempchinsky & Rothman 2008, Slabakova & Ivanov 2011). A number of studies of heritage speakers of Spanish 28 have also contradicted the predictions of the Interface Hypothesis, with heritage speakers exhibiting performance similar to that of their monolingual counterparts in various linguistic constructions at the syntaxdiscourse interface (see Leal Méndez, Rothman & Slabakova 2014 on clitic left dislocation and focus fronting and Leal Méndez, Rothman & Slabakova 2014 on clitic right dislocation). Finally, the results from a very recent study testing the Interface Hypothesis at the L3 syntax-discourse interface have also challenged its predictions, with mixed results being observed. Slabakova and García Mayo (2013) tested Basque-Spanish bilinguals dominant in Basque, Basque-Spanish bilinguals dominant in Spanish, Spanish L2 English learners and native English speakers in three English syntax-discourse interface constructions: topicalisation, focus fronting and left dislocation. Disproving the authors’ prediction that the participants would not perform accurately in the three constructions, participants only exhibited difficulties in topicalisation, where their performance indicated the influence of their L1 or L2. In constructions such as focus fronting and left dislocation, in which the L1 or L2 behaved similarly to English, the participants performed well. Several proposals seeking to explain learners’ difficulties at the syntaxdiscourse interface have appeared in the literature. The initial attempts to account for the exact source of delays in bilinguals pointed to crosslinguistic influence at the level of representation, where interpretable features such as [topic] and [focus] that are involved in the syntax-discourse interface remain underspecified due to crosslinguistic influence. An alternative explanation for the vulnerability at the syntax-discourse interface comes from the processing deficit account, which postulates that there is a processing cost when knowledge is integrated from different domains. More recently, extralinguistic factors such as the quality and quantity of input have also been proposed as possible causes for deficits at the syntax-discourse interface. 27 Deficits are selective, as they do not equally affect all phi features in the pronominal paradigm, but rather a subset of them (i.e. third-person animate). 28 Sorace is reluctant to extend the Interface Hypothesis to domains that were not originally addressed (heritage speakers and intermediate L2 learners; see Sorace 2011), but other researchers find the exclusion of such domains to be unnecessarily restrictive (Lardiere 2011, Montrul & Polinsky 2011, White 2011). 49 2.3.1 Representational account According to the representational account, the reason why bilinguals extend the scope of overt pronouns to [-TS] contexts is due to the underspecification of the interpretable feature [+topic shift] mapped onto the overt pronoun in the monolingual grammar (Tsimpli et al. 2004). Here, the assumption is that the most economical language (English), in which overt subjects are the norm, influences the language with a more complex interface system (Italian, Spanish), in which null subjects are allowed alongside overt subjects (Sorace 2011). The claim that crosslinguistic influence is prone to occur in some linguistic domains, particularly at the interface between syntax and discourse, has been prominent in the literature for some time. A convincing hypothesis in this regard was put forward by Hulk and Müller (Hulk & Müller 2000, Müller & Hulk 2001), according to which crosslinguistic interference is likely to occur at the syntax-discourse interface, provided that there is a structural overlap at the surface level between the two languages. This hypothesis was tested in object omission in a bilingual French-Dutch child and a German- Italian child in comparison to monolinguals before the instantiation of the Csystem (which connects internal grammar and other cognitive domains). The results showed that the bilingual children dropped the object more often due to the influence of the Germanic topic-drop language (Dutch/ German) on the Romance non-topic-drop language (French/ Italian). Hulk and Müller’s proposal concerning crosslinguistic influence in bilingual first language acquisition was later extended to the domain of anaphoric dependencies by Serratrice, Sorace and Paoli (2004), who tested older children after the instantiation of the C-system (MLUw 3.0-4.0). A longitudinal study of an Italian-English bilingual child revealed an overproduction of overt subjects in comparison to monolingual peers, but not an overproduction of null subjects. Since that study, compelling evidence has emerged for crosslinguistic influence 29 in the domain of subject realisation from a variety of language combinations consisting of a null-subject and a non-nullsubject language like English (see Hacohen & Schaeffer 2007 and Haznedar 2010, among others), as mentioned in Section 2.2.2.1. Nevertheless, the fact that late adult bilingual speakers of two null-subject languages also exhibit an inappropriate overuse of overt pronouns has revealed the limitations of the underspecification account (see Bini 1993 for L1 Spanish-L2 Italian, García-Alcaraz & Bel for L1 Moroccan Arabic-L2 Spanish, Guido Mendes & 29 However, not all studies on bilingual performance have found crosslinguistic influence. This was the case, for example, in the balanced bilingual twins studied longitudinally by Liceras et al. (2008, 2012), who showed monolingual-like development in the production of overt subjects in English and Spanish. Evidence of interlinguistic influence was not found in Portuguese-German bilinguals (Hinzelin 2003), nor in Inuktitut- English bilinguals (Zwazinger et al. 2005). 50 Iribarren 2007 for L1 Spanish-L2 European Portuguese, Lozano 2006 for L1 Greek-L2 Spanish, Margaza & Bel 2006 for L1 Greek-L2 Spanish). Note that similar findings have also been observed in bilingual children acquiring two null-subject languages like Italian and Spanish (Sorace et al. 2009), as discussed in Section 2.2.2.1. 2.3.2 Processing account Sorace and colleagues have also argued that the non-target-like performance of bilinguals could arise from a processing load at the interfaces that may be related to the executive function (Sorace & Serratrice 2009, Sorace 2011). The processing costs involved at this particular interface level would also explain why monolingual adult speakers from several null-subject languages do not exhibit ceiling performance in selecting the pragmatically most appropriate antecedent for a particular pronoun, which sometimes leads to inaccurate production or interpretation (pragmatic anomaly), as mentioned in Section 1.4. According to the proposed processing account, bilinguals and (to a lesser degree) monolinguals adopt the overt pronoun as the default form due to their difficulty in accessing and integrating two levels of representation (such as syntax and discourse) in real time (Sorace 2011). Differences at the level of processing between bilinguals and monolinguals would thus be one reason for bilinguals being less efficient than monolinguals. Although several studies have argued that there are differences between native and non-native language processing, the debate is far from resolved. Some studies have concluded that L2 processing is qualitatively different from L1 processing, since adult L2 learners process sentences more slowly than native speakers (see, among others, Hahne & Friederici 2001, Marinis, Roberts, Felser & Clahsen 2005). Clahsen and colleagues have gone further to state that non-native speakers follow lexical-semantic cues during sentence processing and at the same time under-use deep structural information (i.e. shallow processing) in the processing of temporarily ambiguous sentences (Felser, Marinis & Clahsen 2003, Felser, Roberts, Gross & Marinis 2003, Papadopoulou & Clahsen 2003), thereby differing from native speakers (Clahsen & Felser 2006). However, there is also evidence pointing to no fundamental differences between L1 and L2 processing (e.g. Frenck-Mestre & Pynte 1997, Juffs 2006). 2.3.3 Extralinguistic factors: Input and age of onset effects In addition to the limits of processing resources, extralinguistic factors such as the input received by bilingual speakers (in terms of both quantity and quality) have also been suggested as a possible source for the overgeneralisation of the overt pronoun (Sorace & Serratrice 2009, Sorace 2011). Quality of input in terms of parental input has been observed to affect children’s 51 performance. For instance, the English-Spanish bilingual child analysed by Paradis and Navarro (2003) showed a higher rate of inappropriate subjects than monolingual children. However, it was later concluded that the parents themselves used more overt subjects, which could have had an effect on the child’s output. With regard to the effect of quantity of input, in the study conducted by Sorace et al. (2009), bilingual children receiving more English input performed less accurately, accepting more redundant pronominal subjects in topic-continuity contexts than bilingual children living in Italy. Similarly, the overextension of overt subject pronouns was also observed in Greek-English bilinguals, but only in English-dominant bilinguals (Argyri & Sorace 2007). It has also been suggested that the quantity of input also plays a role in the development of reference selection in terms of the acquisition of the discourse features of pronouns, especially in relation to the low frequency of overt pronouns. In this regard, Shin and Cairns (2012) note the difficulty in determining whether the delay observed in the selection of overt versus null third-person subject pronouns among the Spanish monolingual children in their study (see Section 2.2.2.1) was due to the infrequency of the forms in adult null-subject languages (Cameron 1992: 233) or the inherent complexity of third-person pronouns. There is no doubt that in order to acquire a language, children need to be exposed to that particular language, but the exact quantity of input (and more precisely, the minimal amount of language input children must receive) is an issue that is far from settled. Although bilinguals will probably receive qualitatively less input in both of their languages than their monolingual counterparts (Paradis & Genesee 1996: 20), several studies have shown that simultaneous bilingual children do not differ in the achievement of developmental milestones from monolingual children (see Genesee & Nicoladis 2007 for an overview). The results of these studies reveal that the relationship between input quantity and language development in a bilingual context is not linear, but the exact role that the quantity and quality of linguistic input play in children's language development is in dispute. The impact of input factors on bilingual development has been observed in a variety of domains (see Unsworth 2014 for a review). For example, in the domain of lexicon, in a series of studies on bilingual English-Spanish children in Miami testing the mass/ count distinction in English, Gathercole (2002a, 2002b, 2002c) observed that-trace effects in English and Spanish and grammatical gender in Spanish. The analysis revealed that younger bilingual children who received less input in the target language (both at home and/ or the school environment) were less accurate in their performance in comparison to their monolingual counterparts. However, these differences were not noticeable at the age of 10, inducing Gathercole to conclude that input effects may be more evident at early stages. In fact, the trivial nature of the effect of the amount of exposure in early years has also been observed in Basque-Spanish bilinguals. Barreña, Ezeizabarrena and García (2008) con- 52 clude that children as young as 23 months who grow up with less exposure to Basque (<60%) have a more restricted lexicon (around 160 fewer words) than children from Basque-dominant environments (60-90%), a difference which persists later on. Apart from the effect of the quantity of input in simultaneous bilingual children in comparison to their monolingual peers mentioned above, another extralinguistic factor, namely age of onset, has turned out be of great importance in recent years, helping disentangle differences between simultaneous and successive learners. However, it is sometimes not clear which factor is more crucial in explaining learner differences, since input and age effects are sometimes difficult to differentiate. More specifically, as Unsworth, Argyri, Cornips, Hulk, Sorace and Tsimpli (2014) note, input quantity interacts with and is affected by several other factors, including input quality, parental education, socio-economic status and age of onset. The authors analysed the acquisition of grammatical gender in Greek and Dutch by bilingual children whose other language was English in order to differentiate input effects from age effects. Three groups of children were tested: simultaneous bilingual children (exposure to both languages from birth), L2 children, whose exposure to Greek or Dutch started between the ages of 4 and 10, and early successive bilinguals, who were exposed to English from birth and Greek or Dutch after age 1 but before age 4. The findings indicated that in general, the amount of input has more impact on the acquisition of grammatical gender in both languages than the age of onset. The amount of exposure played a role, with bilingual Greek children exhibiting fewer difficulties in acquiring gender in comparison to bilingual Dutch children, since the cues for gender are abundant and transparent in Greek. However, age of onset effects could be perceived in simultaneous (from birth) and successive (with age of onset from age 1 to 4) Greek bilinguals, but not in Dutch bilinguals. Age effects have also been observed in the domain of morpho-syntax in successive bilinguals of French with an age of onset between age 3 and 6 and Swedish as the L1; these children differed significantly from monolingual and simultaneous bilingual children, producing errors similar to those exhibited by L2 adults (Granfeldt, Schlyter & Kihlstedt 2007). Little is known about the successive bilingual acquisition of linguistic properties involving the interaction of syntax and discourse, as noted in Section 2.2.2.2, and even less is known about the potential effect of the age of onset in this domain. However, the study conducted by Pladevall (2010) discussed in 2.2.2 suggests that the age of onset may not be that crucial for subject properties at the discourse-syntax interface, as cL2 learners (with age of onset at 4) do not differ qualitatively from simultaneous bilinguals or monolinguals reviewed in 2.2.2.1 in terms of their performance at the syntax-discourse interface; hence the acquisition of discourse properties is prone to delays in both monolingual and bilingual (simultaneous and successive) development. 53 To summarise Chapter 2, we have seen that there are different types of language acquisition, among which cL2 acquisition has gained increasing relevance in recent research. The successive learners of an L2 seem to exhibit performance parallel to cL1 learners in some domains and to adult L2 learners in others. The limited research carried out thus far on structures at the syntax-discourse interface in null-subject languages suggests that cL2 learners resemble adult L2 learners (Pladevall 2010), showing delays in the acquisition of discourse properties in line with results from simultaneous bilingual children (Serratrice 2007, Sorace et al. 2009) and monolingual children (Shin & Cairns 2012), where a preference for overt subject pronouns in [+TS] contexts reaches adult standards earlier than a preference for null subject pronouns in [-TS] contexts. We should note that linguistic structures at the syntax-discourse interface are loci of childhood delays even in monolingual children acquiring a non-null-subject language like English (the Delay of Principle B Effect). Although there have been several proposals seeking to address the instability at this interface, there has not yet been an ultimate explanation; as Sorace (2011a) notes, various factors may interact in a cumulative way without necessarily cancelling each other out, which could result in a complex explanation for linguistic phenomena at the interface level. In a similar vein, Unsworth et al. (2014) note that just one factor (such as age of onset) cannot explain all the differences between simultaneous and successive bilingual children with regard to a particular linguistic structure, since this factor will interact with and may be constrained by language-specific properties related to the phenomenon under study and the acquisition pattern shown by monolingual children. 55 3 The interpretation of null and overt subject pronouns in Basque This chapter covers data on the interpretation of null and overt subject pronouns. In Sections 3.1 and 3.2, relevant information on third-person morphology and pronouns in Basque is presented, followed by a review of previous empirical work on the frequency of third-person pronouns. The objectives of the present study and the research questions with their corresponding predictions are explained in Section 3.3. A brief description of the linguistic profiles of the three groups of Basque-Spanish bilingual participants in the study is then given in Section 3.4. Finally, Sections 3.5 and 3.6 describe the details of the experimental study, which used two different methodologies (a Picture Selection Task and an Acceptability Judgement Task) and report the results obtained from six experiments on the groups’ interpretations of null and overt pronominal subjects in one-referent and two-referent sentences. 3.1 Third-person morphology Basque is a morphologically ergative language that treats the subjects of transitive verbs differently from the subjects of intransitive verbs in both the nominal and verbal domains. It is a language with a very rich person marking on the verb, agreeing with up to three arguments: the subject and the direct/ indirect objects (if any) in both synthetic and analytic (or ‘periphrastic’, in the Basque grammatical tradition) verb forms. The majority of verbs are analytic, 30 consisting of a participle form and an auxiliary. Monovalent intransitive verbs (such as etorri ‘to come’ in (37a)) take the intransitive auxiliary izan ‘to be’, while bivalent transitive verbs like egin ‘to do’ (in (37b)) take the transitive auxiliary *edun ‘to have’. Thus, the choice of auxiliaries in Basque seems to be largely dependent on the valency 31 of the predicate. In the nominal domain, ergativity is shown, in that the subject of an intransitive verb and the object of a transitive verb are equally marked, unlike the subject of a transitive verb. More precisely, the subject of an intransitive predicate and the (direct) object of a transitive predicate receive the absolutive case-marking (the zero case marking; see (37a,b)), whereas the subject of a 30 There are currently only about a dozen synthetic verbs consisting of a single word (e.g. nator ‘I come’) (Hualde, Oyharçabal & Ortiz de Urbina 2003: 195). 31 There is no definite one-to-one correspondence between morphological and syntactic transitivity. A class of verbs known as ‘unergatives’ (e.g. dimititu ‘to resign’) takes the transitive auxiliary *edun ‘to have’ despite typically prohibiting objects (Etxepare 2003: 388). 56 transitive predicate takes the ergative case -(e)k (as in (37b)). Despite being an ergative language morphologically, there is no trace of syntactic ergativity 32 in Basque. (37) a. Miren-Ø etorr-i da. Miren- ABS come- PF AUX ‘Miren has come.’ b. Miren-ek afari-a-Ø egin-Ø du. Miren- ERG dinner- DET - ABS do- PF AUX ‘Miren has prepared dinner.’ Basque verbal morphology is uniform in Jaeggli and Safir’s (1989) terms: seven different inflected forms (four singular and three plural) are found in the paradigm, and argument markers are different for all persons, as seen in Table 1. The first and second persons have a lexical representation as pronouns, unlike the third person, ‘the nonperson’. The pronominal system thus consists of only the first and second persons, since Basque lacks genuine third-person pronouns (Eguzkitza 1986: 29, Laka 1996, Trask 2003: 151). Hence, it has been regarded as a two-person language (Bhat 2004). In addition, person marking is overt for both the first and second person through affixes on the verbal inflection, whereas the subject-agreement marking on the verb is considered to be zero 33 with respect to the third person. A partial homophony exists between the verb morphology (agreement affixes 34 ) and the corresponding personal pronouns, which is consistent with the idea that Basque belongs to the [+pronominal] Agr in the sense of Kato (1999). In this respect, strong pronouns in Basque may be considered cases of doubling of the subject agreement itself (Ezeizabarrena 2003). 32 This is why Basque has been referred to as an ergaccusative language — a syntactically accusative language with ergative morphology (Oyharçabal 1992). 33 For a discussion of the possibility of the existence of a zero or an overt third-person morpheme, see Trask (1981) and Albizu (2002). 34 Gómez and Sainz (1995) claim that Basque agreement affixes are historically derived from pronouns. 57 Table 1. Personal pronouns and the present-tense agreement system in Basque (adapted from Ezeizabarrena 2003: 88) Pronouns Absolutive Ergative agreement (prefix) agreement (suffix) 1s ni ‘I’ n-aiz ‘I am’ du-t ‘I have (it)’ 2s (intimate) hi ‘you’ 35 h-aiz ‘you are’ du-k/ du-n 36 ‘you have (it)’ 2s (unmarked) zu ‘you’ z-ara ‘you are’ du-zu ‘you have (it)’ 3s ___ ø-da ‘(s)he/ it is’ du-ø ‘(s)he/ it is has (it)’ 1pl gu ‘we’ g-ara ‘we are’ du-gu ‘we have (it)’ 2pl zuek ‘you’ z-are-te ‘you are’ du-zue ‘you have (it)’ 3pl ___ ø-dira ‘they are’ dut-e ‘they have (it)’ The omission of subjects in finite clauses in Basque was at first attributed to the language’s rich (and/ or regular) person marking (Jaeggli & Safir 1989, Rizzi 1982, 1986). However, the permissibility of empty categories in nonfinite clauses as well (Goenaga 1984, Ortiz de Urbina 1989) was later explained in terms of topic D-linking by Elordieta (2001). See Duguine (2010) for a unified account of the licensing of null arguments in both types of clauses. In addition to dropping subjects, Basque grammar also permits object omission, hence its categorisation as a multi-drop language (de Rijk 2008: 205) or a three-way pro-drop language (Laka 1996, Ortiz de Urbina 1989). Thus, in Basque, the ergative, dative and absolutive free pronominal arguments of finite verbs are frequently omitted in non-contrastive, non-emphatic contexts, since the missing information is recovered from the rich inflectional morphology, i.e. the agreement markers on the finite verb. On this basis, Basque can be regarded as a bound-pronoun language in which the inflected forms in ditransitive verbs (such as ekarri ‘bring’ in (38)) can contain up to three bound pronouns relating to the subject (I), the direct object (thirdperson singular it) and the indirect object (you) (Saltarelli 1988: 238-45). (38) pro (ABS) pro (DAT) pro (ERG) ekarr-i d-i-zu-t. bring- PF PRS root-3 SG . ABS -2 SG . DAT - 1 SG . ERG ‘I brought it to you.’ (adapted from Saltarelli 1988: 97) 35 The second person singular intimate, also regarded as familiar, hi ‘you’ is of very restricted use in the context of family or close friendship and has fallen out of active use in urban areas where the unmarked form zu ‘you’ has taken its place (de Rijk 2008). 36 In the allocutive forms addressing the intimate pronoun hi ‘you (familiar)’, the finite verb takes a different suffix depending on whether the addressee is male or female. 58 3.2 Third-person pronouns Third-person reference through a null pronoun is only possible if the pronoun in question is neither in focus nor a new topic and shows no other case endings than those on the conjugated verb form — absolutive, ergative or dative (de Rijk 2008: 795). Whenever these conditions are not fulfilled, demonstrative pronouns are used in Basque, as the language lacks distinct forms for third-person pronouns (Bhat 2004: 135). The three-term 37 demonstrative system consists of the proximal hau/ hauek ‘this/ these’ (proximate to the speaker), the medial hori/ horiek ‘that/ those’ (not far from the speaker, proximate to the addressee) and the distal hura/ haiek ‘that/ those (over yonder/ over there)’ (remote from both) (Saltarelli 1988: 213, Trask 2003: 123). Thus, like Spanish, Basque has a demonstrative system based on spatial proximity. Saltarelli (1988: 207) observes that any of the demonstrative forms can be used for third-person reference. However, it seems that hura ‘that’ is the least deictic and hence the most anaphoric demonstrative (Garzia 1996); it is also defined as the default form by Rebuschi (2003: 842). Note that in languages lacking genuine third-person pronouns, demonstratives may be more likely to have an anaphoric function than in languages with thirdperson pronouns (Dixon 2010: 489). The demonstrative hura is required in the presence of emphatic topic pronouns and when the pronoun is in focus position, i.e. in the immediately preverbal position (see (39a)) (Arregi 2003, Eguzkitza 1986, Irurtzun 2006, Ortiz de Urbina 1989). An overt form may also occur in a postverbal position, mainly in story-telling contexts (39b) (Larrañaga 2008) or when the clause-initial position is occupied by an emphasised constituent (39c). (39) a. Hari eman-Ø dio liburu-a-Ø. To him give- PF AUX book- DET - ABS ‘He has given the book to him.’ b. eta etor-tzen da hura-Ø. and come- IPF AUX that- ABS ‘and he comes.’ c. Hemen dator hura-Ø. Here comes that- ABS ‘Here he comes.’ In addition to the demonstrative hura, there is also another overt form that acts as a third-person pronoun. As Trask (2003: 151) states, “Western varieties […] have acquired third singular bera and third-plural berak, 38 these con- 37 According to Irigoyen (1981), the three-deixis system is derived from a previous twodeixis system through an internal reconstruction. 38 Bizkaian has a variant eurak for berak ‘they’. 59 sist of ber- ‘self’ and the article -a (historically the distal demonstrative)”. Note, however, that other Basque grammarians consider the prefix ber- 39 to have the adjectival meaning of ‘same’ (Laka 1996, de Rijk 2008). In fact, this adjective function is visible in expressions such as andre bera ‘the same lady’. The quasipronoun status that de Rijk (2008: 114) assigns to bera ‘(s)he (him/ herself)’ is argued to stem from the combination of the prefix ber with the null pronoun (Ø + bera = bera), “thereby creating the semblance even of being a personal pronoun” (de Rijk 2008: 794). Thus, bera 40 competes with the pronominal uses of hura 41 in western 42 varieties of Basque (de Rijk 2008: 795) such as Gipuzkoan, the dialect under investigation. Both overt forms are marked for person, case and number (third-person absolutive singular) but lack the gender feature, as there is no gender distinction in the Basque (pro)nominal system. 3.2.1 Intrasentential antecedent choice Descriptive grammars present both hura and bera as convenient counterparts to the third-person pronouns of other languages (Rebuschi 1995: 314, de Rijk 43 2008). However, the words differ from each other in terms of antecedent assignment in intrasentential anaphora contexts, namely in the preceding clause of a complex sentence, as in (40). The general and common criterion, as Saltarelli (1988: 97) explains, is that “third person pronouns (hura, haiek) take a special form (bera, beraiek) when referring to persons or objects previously mentioned in the discourse”. More specifically, Laka (1996) states that bera, not hura, must be used if the antecedent and the pronoun are in the same sentence. 39 The root ber- ‘re-’ can be related to the prefix found in hogei ‘twenty’ → berr-ogei ‘fourty’, or even in egin ‘make’ → berr-egin ‘remake’, conveying the idea of ‘twice’, or ‘again the same’ (Martínez-Areta 2013: 317). 40 The form bera literally means ‘the same’ (Rebuschi 1995: 314) or ‘idem’ (Martínez-Areta 2013: 317). 41 Hura’s variant in the Bizkaian dialect is (h)a. The archaic form (h)a ‘that’ is assumed to be the origin of the article -a, which is common to all dialects (de Rijk 2008: 209). 42 Note that de Rijk (2008) refers to the dialects spoken in France as ‘northern’, as opposed to the ‘southern’ ones spoken in Spain. However, we will adopt the ‘western’ (dialects spoken in Spain)/ ‘eastern’ (dialects spoken in France) distinction, following Hualde’s (2003: 4) observation that it is geographically more appropriate to refer to them in these terms. 43 As de Rijk notes, “[…] this anaphoric noun bera occurs with considerable frequency, for it serves as a convenient counterpart to 3 rd person pronouns of other languages” (de Rijk 2008: 115). With respect to hura, he claims that “the three demonstratives serve as equivalents for third-person pronouns found in many other languages” (de Rijk 2008: 209). 60 (40) a. Nere senarr-a-Ø i etxera-tzen d-enean bera-k i/ ? j My husband- DET - ABS come home- IPF AUX - COMP he- ERG garbi-tzen ditu ontzi-a-k. wash- IPF AUX dish- DET - ABS . PL ‘When my husband i comes home, he i/ ? j washes the dishes.’ b. Nere senarr-a-Ø i etxera-tzen d-enean har-k *i/ j My husband- DET - ABS come home- IPF AUX - COMP he- ERG garbi-tzen ditu ontzi-a-k. wash- IPF AUX dish- DET - ABS . PL ‘When my husband i comes home, he *i/ j washes the dishes.’ (adapted from Saltarelli 1988: 97) Berak (the ergative form of bera) is free in reference. It can behave as an anaphoric pronoun, as in (40a), be coreferential with the subject of the preceding clause or refer to an extrasentential referent. In contrast, hark (the ergative form of hura) in (40b) functions as an obviative pronoun, avoiding intrasentential coreference. Based on this diverging usage, Ortiz de Urbina (1989: 147) defines bera ‘(s)he (him/ herself)’ as a proximate/ intensive pronoun, and hura ‘that’ as an obviative/ neutral 44 pronoun. The distributive differences between the two overt pronouns hura and bera have been explained by Abaitua (1994) in terms of logophoricity, which can be described in the strict sense as an element whose reference is determined in relation to the speech, thought or point of view represented in the surrounding discourse (Sells 1987). Thus, the [+log] pronoun bera is linked to the SOURCE (Sells 1987: 455-461), referring to the individual from whence the communicative act starts (Miren in (41)). In contrast, the [-log] pronoun hura selects disjoint reference. Note that the disjoint reading for bera is not discarded (i.e. it is not marked as ungrammatical), but it is called into question, suggesting that the disjoint reading is a marked interpretation. (41) Miren-ek i esan-Ø du bera-Ø i/ ? j / hura-Ø *i/ j joan-go Miren- ERG say- PF AUX she herself- ABS / she- ABS go - FUT d-ela. AUX - COMP ‘Miren has said that she herself/ she is coming.’ (adapted from Saltarelli 1988: 97) 44 Obviative pronouns have been regarded as ‘fourth person’, reserving ‘third person’ for proximate pronouns (Anderson & Keenan 1985: 262). 61 In a similar vein, Garzia (1996) claims that the use of bera in an indirect speech context like (41) is obligatory for continuity of reference with a referent in the text. Hura would instead be used to mark switch reference. Abaitua (1994) also observes that the distributive properties of bera in comparison to the null pronoun are almost the same in intrasentential anaphora contexts, as seen in (42a), except for backward anaphora contexts like (42b) where coreference between bera and Pello is unnatural, since bera is subject to precedence: 45 (42) a. Pello-k i irabaz-i zuen, ___ i / bera-k i erantzun-a-Ø Pello- ERG win- PF AUX ___/ he- ERG answer- DET - ABS baitzekien. since.know ‘Pello i won because ___ i / he i knew the answer.’ b. ___ i / Bera-k ? i erantzun-a-Ø baitzekien, Pello-k i ___/ he- ERG answer- DET - ABS since.know Pello- ERG irabazi zuen. win AUX ‘Because ___ i / he ? i knew the answer, Pello won.’ (Abaitua 1994: 864-865) Although different approaches have been developed to describe the use of hura and bera, the debate still continues on the usage and scope of each form. This is especially the case with regard to bera, which represents extensions of the intensive (or emphatic) 46 (Trask 2003: 151) third-person pronouns that diverge from those seen in Table 1. Although according to the Linschmann- Aresti Law, 47 the distribution of the possessive (genitive) forms bere ‘his (proximate)’ and haren ‘his (neutral)’ is presumed to be complementary, Martínez-Areta points out (2013: 319) that it is difficult to make predictions 45 Precedence does not play a role in VP-complement clauses (as opposed to adjunct clauses like (42)), as both the null pronoun and bera have the same referential properties (examples from Abaitua 1994: 865): Pellok i esan zuen ___ i / berak i erantzuna bazekiela. ‘Pello i said that ___ i / he i knew the answer.’ ___ i / Berak i erantzuna bazekiela esan zuen Pellok i . ‘That he i knew the anwer said Pello i .’ 46 Emphatic or intensive pronouns are typically used as topicalised noun phrases or when the pronoun is in focus (preverbal position) (Trask 2003: 151). 47 According to this law, which was proposed by the two scholars independently, the emphatic genitive form will be chosen if the person corresponding to that genitive is indexed — i.e. he or she has a corresponding (non-allocutive) morpheme — in the finite verb of the corresponding clause, whether main or embedded (Martínez-Areta 2013: 312): ni Joni i bere i/ *j semeaz mintzatu natzaio ‘I talked to John about his son’ (Rebuschi 1995: 319). This law is evident to varying degrees in the literature from the sixteenth to nineteenth centuries in texts in every dialect. 62 regarding the distribution of bera and hura in cases other than the genitive. As he also notes, “In authors and varieties where bera, berak, etc., has not encroached on contexts which originally belonged to hura, hark, etc. (i.e. the purely pronominal ones), bera has a certain anaphoric implication” (Martínez-Areta 2013: 318). However, in the attested history of western dialects, particularly Bizkaian and Gipuzkoan, “bera becomes progressively less marked, pushing hura […] into 3rd grade deictic meanings and becoming a fully-fledged 3rd p. pronoun” 48 (Martínez-Areta 2013: 320). As de Rijk (2008: 795) observes, “in colloquial Guipuzcoan […] one sees a tendency to equate bera with the Spanish pronouns él and ella”. Some authors hold a view contrary to Martínez-Areta’s (2013) claim that hura is the purely pronominal form. Eguzkitza (1986: 204), for instance, refers to bera as the real pronoun obeying Principle B — that is, “free within [its] governing category (and in the case at hand discourse bound)” (see Section 1.3 for more information on Binding Theory). Hura, in contrast, which “Basque grammarians have traditionally conceived of as a pronoun” is a deictic term under Principle C (Eguzkitza 1986: 204). In other words, it seems to be a lexical item that behaves like a noun and must therefore be free everywhere. This explains why hura cannot be coreferential with the subject of the preceding clause Jon in (43a), in contrast to bera, which is free in terms of its governing category (43b). According to Eguzkitza (1986: 28), these interpretative differences between hura and bera are a consequence of their belonging to two different pronoun sets. (43) a. Jonek i dio hura-Ø *i/ j azkarra de-la Jon- ERG says that- ABS smart AUX - COMP Jon-ek i dio bera-Ø ij azkarra de-la. b. Jon- ERG says he- ABS smart AUX - COMP ‘Jon says that he is smart.’ In conclusion, Section 3.2.1 has shown that the category of the third-person personal pronoun is troublesome in Basque due to the lack of true thirdperson pronouns. The detailed analysis presented above has highlighted the descriptive complexity of the two lexical third-person forms resulting from differences in dialectal and historical-synchronical usage. Based on the claim that within the three-term demonstrative system, the distal demonstrative hura is the most anaphoric (Garzia 1996), the present study focuses only on this pronoun. The experiments presented in Sections 3.5 and 3.6 are intended to clarify the intrasentential referential properties of hura, bera and the null pronoun in one-referent and two-referent sentences. Hura is expected to be 48 Euskaltzaindia (the Royal Academy of Basque) and some scholars have noted the “abuse” in the use of bera ‘(s)he (him/ herself)’ instead of hura ‘that’ in both written contexts and in the speech of many western-central speakers over recent decades (Euskaltzandia-I 1991: 73, Petrirena 1999). 63 disjoint in reference, referring to a non-topic antecedent, similar to anaphoric demonstratives from languages such as German and Finnish (Bosch, Katz & Umbach 2007, Kaiser & Trueswell 2008). With regard to bera, the overextension in its scope in western dialects suggests that the anaphoric (coreferential) properties of bera may have been lost. As for the null pronoun, it is expected to refer to an established topic at the centre of attention. 3.2.2 Frequency Empirical work on pronouns in Basque has been conducted by Garcia- Azkoaga (2003), who analysed cohesive devices in different text genres produced by adults and teenagers of 11, 13 and 15 years of age. Pronouns referring to the third person 49 were highly frequent in narratives and much less common in explicative texts and opinion letters in the texts produced by adults, as seen in Table 2. The primary function of these pronouns was to identify the main characters or thematic subjects of the story; this finding is in agreement with the results of Karmiloff-Smith (1981), Hickmann (1980, 1984, 1987) and De Weck (1991). With regard to teenagers, a development towards a target-like use of third-person pronouns was observed in narratives but did not appear in explicative texts or opinion letters. In narratives, the teenagers’ use of pronouns was still non-adult-like at the age of 11 (36% vs. 51% in the adults), but teenagers resembled adults at the age of 13 in terms of the decrease in their use of proper nouns. The opposite pattern was observed in the analysis of the opinion letters, where the youngest group used third-person pronouns the most often. This tendency decreased in the middle group, but increased again at age 15. As for explicative texts, there was a decrease in the use of third-person pronouns with age, but this usage still differed from adults’ performance. Table 2. Pronoun frequency in different texts produced by Basque teenagers and adults (adapted from Garcia-Azkoaga 2003: 261) Participants 11-year-olds 13-year-olds 15-year-olds Adults Text genre N 50 E O N E O N E O N E O Third personal pronouns 36 10 36 50 36 3 58 66 18 51 32 38 Unfortunately, as it was not her research topic, Garcia-Azkoaga (2003) does not report the exact percentages of each type of third-person pronouns (hura vs. bera) and their corresponding antecedents. However, another corpus 49 Pronouns (bera/ berak ‘(s)he (him/ herself)/ they (themselves)’) and the three-term demonstratives (hau/ hauek ‘this/ these’, hori/ horiek ‘that/ those’, hura/ haiek ‘that/ those (over yonder/ over there)’) were included. 50 N=narratives, E=explicative texts, O= opinion letters. 64 study by Aduriz, Ceberio and Díaz de Ilarraza (2006) using the Eus3LB real corpus (consisting of 54,000 words) addresses such matters, at least with regard to demonstratives. The authors annotated the number of each demonstrative type reaching over 700 anaphoric elements; the results are compiled in Table 3. Table 3. Location and antecedent of Basque demonstratives (adapted from Aduriz et al. 2006: 103) Anaphora type Location of the antecedent Antecedent type Demonstratives Total Forward Backward Intrasen tential In the previous clause Others NP Sentence Hau ‘this’ 177 86% 14% 59% 32% 9% 73% 27% Hori ‘that’ 251 99% 1% 50% 47% 3% 33% 67% Hura ‘that’ 321 98% 2% 64% 33% 3% 100% 0% Total 749 95% 5% 58% 37% 5% 71% 29% As shown in Table 3, hura ‘that’ achieves the highest percentage (321 occurrences). This is not surprising; as mentioned in the previous section, it is the default form among the demonstratives (Rebuschi 2003). This distal demonstrative usually occurs in forward anaphora contexts (98%) and typically takes an antecedent in the same sentence (64% of the time). In addition, similar to hau ‘this’, hura’s antecedent is always a noun phrase, in contrast to the medial hori ‘that’, which usually takes a sentence as antecedent. Reviews of previous (non-experimental) research on the frequency of third-person pronouns in Basque (Aduriz et al. 2006, Garcia-Azkoaga 2003) reveal a glaring omission in the area of the interpretation of null and overt pronouns from an experimental point of view. Additionally, the low percentage of overt pronouns in written data stems from the fact that arguments are usually omitted in Basque, since the verbal inflection can mark up to three arguments on the verb. The limited usage of overt pronouns may also be a result of the frequent use of synonyms or associations to refer to an earlier established discourse referent. This preference for using NPs over pronouns has been attributed to the Basque language’s lack of grammatical gender (Garcia-Azkoaga & Idiazabal 2004). 65 3.3 Aims and research questions In order to delve deeper into the development of discourse management of pronouns in null-subject languages, the current study seeks to expand the scope of the field by presenting data from Basque, a language that is typologically distant from the Romance languages studied to date. The aims of the present research are threefold: To analyse whether: a) null and overt subject pronouns differ in their antecedent choices in Basque; b) the description of hura ‘that’ as obviative/ neutral and bera ‘(s)he (him/ herself)’ as proximate/ intensive (Ortiz de Urbina 1989: 147) and their presumed equivalence to third-person pronouns from other languages presented in Basque grammars (de Rijk 2008) is backed up by experimental data; c) the interpretations of null and overt pronouns differ between adult and child language, and also between the two groups of children (cL1 and cL2). The literature on syntactic and discourse constraints on the anaphoric dependencies of null and overt pronouns leads to the following research questions and the corresponding predictions: RQ1: Do Basque null and overt subject pronouns (hura/ bera) differ in terms of their antecedent assignments...? In intrasentential anaphoric dependencies, Italian null and overt subject pronouns differ in their biases towards antecedents in different syntactic positions (Carminati 2002). Under the Position of Antecedent Strategy (PAS; see Section 1.4), null pronouns are biased towards an antecedent in a structurally prominent position (the SpecIP). In contrast, overt pronouns show more variability but select an antecedent lower in the phrase structure, a non-subject antecedent. As the PAS applies crosslinguistically (slight differences between languages with respect to the bias of overt pronouns notwithstanding), a division of labour between null and overt pronouns is expected in Basque. 66 a. …regardless of clause order? Although the effect of clause order has been verified in studies beyond anaphora resolution (Bever & Townsend 1979, Garnham et al. 1998), Carminati (2002) did not observe differences in the biases of null and overt pronouns with respect to the order of the main and subordinate clauses in forward anaphora contexts. Consequently, clause order is not expected to have an effect on the antecedent choices of null and overt pronouns. b. …regardless of anaphora type? Principle C of the BT rules out coreference between matrix subject pronouns followed by embedded R-expressions in backward anaphora conditions, but coreference is not excluded in cases of forward anaphora (see Section 1.1.1). Thus, differences are expected between forward and backward anaphora with regard to the antecedent assignment of pronouns. In addition, bera may show different referential properties depending on anaphora directionality, as (according to Abaitua (1994)) it is subject to precedence (see (42)). c. …regardless of predicate type? The division of labour between null and overt pronouns has been observed to be more marked in two-referent sentences than in onereferent sentences (Carminati 2002: 96, see 1.4). A similar pattern is expected for Basque. d. …depending on discourse context? Null pronouns tend to refer to topical antecedents crosslinguistically (Ariel 1990), whereas overt pronouns express switch reference (or [+topic shift] (Sorace 2000)), as mentioned in Section 1.1.4. A similar distribution is predicted for Basque null and overt pronouns. RQ2: Do the two Basque third-person pronouns, the demonstrative hura ‘that’ and the pronoun bera ‘(s)he (him/ herself)’, differ in their referential properties? Differences have been found in the antecedent choice of personal and demonstrative pronouns in intersentential anaphora contexts (Form- Specific Multiple-Constraints approach; Kaiser & Trueswell 2008; see Section 1.1.3). Whereas the former is driven primarily by syntactic constraints (coreference with the preceding subject), the latter is discourse- 67 constrained (non-topical antecedent). In line with this approach and the distinction between hura as obviative/ neutral vs. bera as proximate/ intensive (Ortiz de Urbina 1989: 147), hura is predicted to refer to a non-topical antecedent, whereas bera may tend to corefer with the subject (topic) in intrasentential anaphora contexts (see Section 3.2.1). RQ3: Do 6to 8-year-old (cL1 and cL2) participants perform differently from adults in the interpretation of null and overt subject pronouns? Monolingual and simultaneous bilingual children at this age typically show a non-adult-like use of null and overt pronouns: they overextend both the use of null pronouns (to topic-shift contexts), resulting in ambiguity, and the use of overt pronouns (to topic-continuity contexts), resulting in redundancy (Barquin & Costa 2011, Shin & Cairns 2012, Sorace et al. 2009; see Section 2.2.2), but redundancy would seem to be more problematic. Such overextensions are a product of the developmental delay in the mapping between syntax and discourse. Thus, the antecedent choice of pronouns is expected to differ between Basque children and adults. Moreover, as extralinguistic factors such as the amount of exposure to the language may affect language development (Sorace et al. 2009), and prior studies on successive bilinguals at the syntax-discourse interface have observed a delay in discourse properties (Pladevall 2010), children with less exposure to the target language as well as a later age of onset (cL2) are expected to exhibit a more target-deviant performance. The answers to these three research questions will be provided on the basis of data from the six off-line experiments presented in Sections 3.5 and 3.6. For data collection, two different methodologies were employed: a Picture Selection Task and an Acceptability Judgement Task (the Basque version of the experiment conducted by Sorace, Serratrice, Filiaci & Baldo (2009)). Participants in the experiments were three groups of Basque-Spanish bilinguals: adults, cL1 and cL2. The experimental data from the adults will provide a) a baseline for the interpretation of null and overt pronouns in Basque (RQ1) and b) a description of possible differences in the referential properties of hura and bera (RQ2). The comparison of results between adults and 6to 8year-olds will clarify whether interface properties manifest developmental delays, as argued in the literature on children of that age (RQ3). Finally, the comparison between children with earlier/ more exposure to the language (cL1) and those with later/ less exposure (cL2) will shed light on the effect of the interplay between extralinguistic factors such as age of onset of acquisition (the age at which children are first exposed to the target language) and the amount of exposure to the language (the quantity of input to which children are exposed) for the acquisition of referential properties of pronouns. In 68 addition, data from cL2 learners will help us to obtain a more detailed understanding of whether properties at the syntax-discourse interface are problematic, as observed in prior studies in simultaneous bilingual children. 3.4 Participants A total of 64 Basque-Spanish bilingual adults who acquired Basque as their L1 and had a daily use of both languages participated in the different tasks as a control group. In addition, two groups of 6to 8-year-old Basque- Spanish bilingual children took part in the study: a group of Basque-Spanish bilingual children, native speakers of Basque (the cL1 group), and a group of native Spanish speakers who were early learners of Basque as L2 (the cL2 group). A total of 149 children out of 168 participants in the cL1 group and 115 children out of 141 in the cL2 group were ultimately included in the data analysis (for more details, see the sections on coding and data analysis for each experiment). Table 4 summarises the number of participants included in the data analysis for each experiment. 69 Table 4. Linguistic profiles of the participants in the six experiments Participants Language used at home Sociolinguistic environment Exp1 hura ‘that-ABS’ Exp2 bera ‘(s)he (him/ herself)- ABS’ Exp3 har(e)k‘ that’ -ERG. Exp4 berak ‘(s)he (him/ herself)’- ERG. Exp5 hura ‘that’ Exp6 bera ‘(s)he (him/ herself)’ Adults Basque- Spanish Mainly Basquespeaking N=10 mean: 25; 6 (17; 5-59; 10) N=10 mean: 27; 5 (24; 3-48; 1) N=10 mean: 30; 4 (18; 8-65; 2) N=10 mean: 36; 8 (17; 6-63; 1) N=14 mean: 19; 0 (18; 0-20; 0) N=10 mean: 28; 3 (17; 2-57; 5) cL1 Basque> (Spanish) Mainly Basquespeaking N=21 mean: 6; 5 (6; 1-7; 0) N=30 mean: 6; 9 (6; 5-7; 6) N=26 mean: 6; 4 (6; 1-7; 0) N=30 mean: 6; 8 (6; 4-7; 3) N=19 mean: 6; 6 (6; 3-7; 2) N=23 mean: 6; 4 (6; 1-7; 0) cL2 Spanish Spanishspeaking N=22 mean: 6; 8 (6; 4-7; 2) N=25 mean 6; 10 (6; 8-7; 5) N=12 mean: 7; 5 (7; 2-8; 0) N=11 mean: 6; 3 (6; 1-6; 8) N=23 mean: 7; 9 (6; 2-8; 0) N=22 mean: 6; 5 (5; 11-7; 0) 70 Figure 5. Geographic origin of the participants (adapted from http: / / en.wikipedia.org/ wiki/ Basque_Country_(greater_region), accessed 19 December, 2013) The majority of children tested were in their first year of primary school (6-7 years), but some were in the second year (7-8 years). Both children’s groups were enrolled in the educational Model D, in which Basque is the main language of instruction; they attended schools located in different geographic areas but with an equal socio-economic status. This age range was selected for investigation because studies on pronoun interpretation (reviewed in Section 2.2.1) have shown that up to the age of 6, children do not exhibit an adult-like performance with regard to binding conditions, especially the condition applying to non-reflexive pronouns — namely, Principle B (Guasti 2004: 271). In addition, the data collected on children of this age can be compared to prior studies on pronoun interpretation (reviewed in Section 2.2.2). The children in the cL1 group were raised in Basque-speaking families and came from a Basque-dominant sociolinguistic environment in the province of Gipuzkoa within the Basque Autonomous Community (see Figure 5), where Basque is the official language, along with Spanish. The children were speakers of the Gipuzkoan dialect, the central 51 dialect among the western varieties of Basque. In Gipuzkoa, the percentage of bilinguals 52 in the population is 49.1%, 53 with an additional 15.9% passive bilinguals. The rest of the population is Spanish monolingual (35%) (Basque Government 2008). In contrast, the cL2 group, the children of native Spanish-speaking parents, came from an almost exclusively Spanish-speaking city, Villava in Navarre, where the percentage of Spanish monolingual speakers is 83.5%, with 9.9% active and 6.6% passive bilinguals (Government of Navarre 2008). The early successive L2 Basque acquisition of these children through the educational system started at age 3 in preschool, but their exposure to Basque was 51 According to Zuazo’s classification (1998). 52 The rate of bilinguals between the ages of 5-14 is above 80% (Basque Government 2008). 53 These rates correspond to the time at which the data collection was carried out. cL1 group and native adults cL2 group 71 basically limited to school hours. Villava, which is close to the capital city Pamplona, is part of the mixed zone 54 within the Autonomous Community of Navarre, where Basque has no official status. Most parents in this mixed zone are in favour of enrolling their children in the mixed educational models in which both Basque and Spanish are vehicular languages (38%), followed by the model with Basque as the main language of instruction (27.6%) 55 (Government of Navarre 2008). The creation of the standardised language, Euskara Batua ‘unified Basque’ (based on the varieties of Gipuzkoa and Lapurdi), took place in the 1960s. Euskara Batua is now the variety used in the vast majority of Basque educational settings, in educational materials and in teaching Basque as a second language (Cenoz 2009: 72). However, as Cenoz notes, many teachers use the local varieties orally, although textbooks are written in the standard language. Parental permission was requested in accordance with the schools’ requirements. The experimenter talked to the teachers of the cL1 group to identify children whose parents were both native speakers of Basque. In the case of the cL2 group, questionnaires were distributed to the parents to determine how much input each child received in each language and to select children whose exposure to Basque was restricted to school hours. None of the participants (neither adults nor children) took part in more than one experiment in order to avoid excessive familiarity with the methodology that might lead to the development of strategies for better performance. 3.5 Picture Selection Task This section describes the comprehension data from four experiments on null subject pronouns (henceforth, NSPs) in comparison to the overt subject pronouns (henceforth, OSPs) hura ‘that’ and bera ‘(s)he (him/ herself)’ in onereferent (Sections 3.5.1 and 3.5.2) and two-referent (Section 3.5.4 and 3.5.5) sentences. The task was designed to analyse the referential properties of NSPs and OSPs in different clause orders, anaphora types and predicates (RQ1a,b,c) and to determine whether there are any differences between the two OSPs (hura and bera) in accordance with the properties usually assigned to them crosslinguistically (RQ2). At the same time, these experimental data provide details on whether the antecedent assignment of pronouns by 54 Under the Basque Language Law (1986), Navarre is divided into three specific areas: the north part of the region is known as the ‘Basque-speaking area’, the central part is the so-called ‘mixed zone’ and the ‘non-Basque-speaking area’ is in the south. Basque is only recognised as an official language in the Basque-speaking area. 55 The rest of the population is in favour of the model in which Spanish is the vehicular language (18.2%) or the British model in which English and Spanish are the vehicular languages (13.6%). 72 Basque children differs from that of their adult counterparts, and whether there are any differences between cL1 and cL2 groups (RQ3). 3.5.1 Experiment 1: hura 3.5.1.1 Participants Twenty-five children in the cL1 group (age range 6; 1-7; 0, mean age 6; 5), 26 in the cL2 group (age range 6; 4-7; 2, mean age 6; 8) and 10 L1-Basque adults (age range 17; 5-59; 10, mean age 25; 6) participated in the experiment. Eight participants (4 from each child group) were excluded from the final analysis because they did not complete the entire task or did not respond correctly to the half of the filler items (4 out of 8). 3.5.1.2 Materials and design The experiment was based on a 2 x 2 x 2 within-subject design with two levels of the variable pronoun type (NSPs vs. OSPs), two levels of the variable clause order (main-subordinate vs. subordinate-main) and two levels of the variable anaphora type (forward vs. backward). In addition, there was a between-subject variable, group, with three levels: adults, cL1 and cL2. The variable of clause order (main-subordinate vs. subordinate-main clause) was included in the experiment because previous studies have shown that different clause orders are not processed in the same way (Bever & Townsend 1979, Garnham et al. 1998), as described in Section 1.1.1. Carminati (2002: 342-343) also argues that for the sake of completeness, the PAS should be tested in main-subordinate clause order, since she primarily concentrated on subordinate-main clause order. However, it should be noted that in the few cases in which Carminati tested main-subordinate clause order (as in (33a)), the distinct biases of null and overt pronouns were also observed. The type of anaphora was also introduced as a variable, since it has been claimed in the literature that forward and backward anaphora pose different processing demands, as discussed in Section 1.1.2. In addition, it would be interesting to determine whether the PAS also holds in the same way in backward anaphora (as Carminati only tested its validity in forward anaphora), or whether it is overridden by other strategies in which the parser assigns coreference with the first referent encountered (van Gompel & Liversedge 2003, Sorace & Filliaci 2006; see Section 1.1.2). It is important to note that in the linguistic items tested here, coreference between the matrix or embedded subject pronouns and their potential subject antecedent is syntactically licit. Thus, no violation of the binding condition Principle B can occur, as the subject pronouns tested are ‘free’ in their governing category. However, the binding condition Principle C is operative in a particular linguistic stimulus, i.e. in backward anaphora sentences with main-subordinate clause 73 order where the R-expression must necessarily be free. Previous off-line studies on forward vs. backward anaphoric dependencies in intrasentential anaphora contexts (e.g. Belletti et al. 2007, Sorace & Filiaci 2006,) have only tested forward anaphora sentences with main-subordinate order and backward anaphora sentences with subordinate-main order, where the binding condition (Principle C) is not operative. For the present study, however, the crossing of the variables clause order (main-subordinate vs. subordinatemain) and anaphora type (forward vs. backward) with all the possible combinations (4 conditions) was taken into account. Much previous research on anaphora resolution has focused on the references of subject pronouns of transitive predicates in complex sentences in SVO nominative-accusative languages, where the subject and the object, both NPs, are morphologically zero-marked. However, there are a number of good reasons to test first one-referent sentences in an ergative-absolutive language like Basque. In this way, the possible effect of different case marking morphology for the subject (ergative -k) and the object (absolutive Ø) can be avoided (see Sauerland & Gibson 1998 for the processing difficulty in case-mismatching contexts in relative clauses). At the same time, in onereferent sentences, there is no intervening NP (a new discourse referent) between the potential subject antecedent and the subject pronoun of the second clause (see Gibson 1998 on the substantial incremental cost of processing an NP that refers to a new discourse referent (the object) in nested syntactic dependencies). Focusing on one-referent sentences also allowed us to check whether the syntactic distinction of subordinated clauses (an adjunct in the Picture Selection Task vs. an argument in the Acceptability Judgement Task) affected the biases of the pronouns. The experiment consisted of 40 linguistic items: 32 experimental items for 8 experimental conditions (4 items per condition) and 8 filler items, which were added to determine whether the participants could distinguish the two characters involved in the task. The experimental items were equally divided into subordinate-main clause order and main-subordinate clause order contexts; in addition, half of the items contained an OSP and the other half, an NSP. The linguistic auditory stimuli consisted of complex sentences made up of a main clause and an adverbial (adjunct) clause with the temporal link when with the meaning of subsequence or ‘right after’. The adverbial modifier when was chosen over others such as while or after because when is one of the first adverbial subordinators that researchers have found in spontaneous child speech (Diessel 2004). The subordinate clause or the main clause introduced only one male individual in the subject position. The NSP or the OSP in either the subordinate clause or the main clause contained a statement that was not pragmatically biased to refer to the subject antecedent in the preceding or subsequent clause. 74 The eight experimental conditions included an unaccusative verb pair such as igo (lit. ‘go up’, but ‘get on’ in the context of (44)) and haserretu ‘get angry’ in forward and backward anaphora, as shown in (44). In (44a,e) and (44b,f), the subordinate clause precedes the main clause and the subject pronoun changes from null to overt, respectively. In (44c,g) and (44d,h), the subordinate clause follows the main clause, differing in pronoun type. The subjects of both clauses were always matched for number (singular) and case (absolutive). The sentences contained analytic verb forms in thirdperson perfective present 56 (present perfect), consisting of the non-finite perfective participle 57 and the intransitive auxiliary izan ‘to be’. The different types of referring expressions (NP, Ø, hura) in each clause are described within square brackets. (44) FORWARD ANAPHORA a. Subordinate clause [NP] + main clause [..Ø..] Mikel-Ø igo-Ø de-nean ____ haserretu egin-Ø da. Mikel- ABS get on- PF has-when ____ get angry do- PF has ‘When Mikel got on (the bus), (he) got angry.’ b. Subordinate clause [NP] + main clause [hura] Mikel-Ø igo-Ø de-nean hura-Ø haserretu egin-Ø da. Mikel- ABS get on- PF has-when he 58 - ABS get angry do- PF has ‘When Mikel got on (the bus), he got angry.’ c. Main clause [NP] + subordinate clause [..Ø..] Mikel-Ø haserretu egin-Ø da ____ igo-Ø de-nean. Mikel- ABS get angry do- PF has ____ get on- PF has-when ‘Mikel got angry when (he) got on (the bus).’ d.Main clause [NP] + subordinate clause [hura] Mikel-Ø haserretu egin-Ø da hura-Ø igo-Ø de-nean. Mikel- ABS get angry do- PF has he- ABS get on- PF has-when ‘Mikel got angry when he got on (the bus).’ 56 Perfective present is used for past actions with present relevance and also as a hodiernal past (Hualde, Oyharçabal & Ortiz de Urbina 2003: 257). 57 In order to control the effect of focus on the pronoun and/ or its potential antecedent, the verb emphatic marker egin ‘do’ was inserted into the main periphrastic verb, resulting in the focalization of the verb. This insertion was maintained across items and conditions in the Picture Selection Task. 58 We have translated hura as ‘he’ in this context, although hura literally means ‘that’. 75 BACKWARD ANAPHORA e. Subordinate clause [NP] + main clause [..Ø..] ____Igo-Ø de-nean Mikel-Ø haserretu egin-Ø da. ____Get on- PF has-when Mikel- ABS get angry do- PF has ‘When (he) got on (the bus), Mikel got angry.’ f. Subordinate clause [NP] + main clause [hura] Hura-Ø igo-Ø de-nean Mikel-Ø haserretu egin-Ø da. He- ABS get on- PF has-when Mikel- ABS get angry do- PF has ‘When he got on (the bus), Mikel got angry.’ g. Main clause [NP] + subordinate clause [..Ø..] ____Haserretu egin-Ø da Mikel-Ø igo-Ø de-nean. ____Get angry do- PF has Mikel- ABS get on- PF has-when ‘(He) got angry when Mikel got on (the bus).’ h. Main clause [NP] + subordinate clause [hura] Hura-Ø haserretu egin-Ø da Mikel-Ø igo-Ø de-nean. He- ABS get angry do- PF has Mikel- ABS get on- PF has-when ‘He got angry when Mikel got on (the bus).’ Each experimental sentence in (44), recorded by a native speaker, was aurally presented together with a pair of pictures of the same size (see Figure 6). In the various pictures, two male characters Mikel and Julen are involved in eight intransitive actions such as falling, sitting or getting angry. Pictures A and B in Figure 6 are a sample item for the experimental condition in (44a) or (44b) Mikel igo denean____/ hura haserretu egin da ‘When Mikel got on (the bus),___/ he got angry’. In picture A (left) below, Mikel (the boy with straight dark hair) performs both the action described in the main clause and the action described in the subordinate clause. Thus, Mikel gets on the bus and gets angry, and Julen (the boy with curly hair and a striped pullover) is just standing there. In picture B (right), Mikel performs the action described in the first clause (getting on the bus), whereas Julen carries out the action of the second clause (getting angry). 76 Picture A Picture B Figure 6. An example of the picture pairs presented in the Picture Selection Task Two alternative readings of the anaphoric relation described in (44a,b) are possible in this setting: a) a coreferential reading, i.e. intrasentential coreference between the pronoun and the preceding subject, Mikel, as displayed in Picture A, or b) a disjoint reading, i.e. the pronoun’s antecedent is an extrasentential referent (not verbally mentioned but visually present), as shown in Picture B. Pseudorandomised picture pairs were balanced for conditions, characters and the location of the character on the screen (left vs. right). 3.5.1.3 Procedure The participant and the experimenter sat together in front of a laptop computer in a quiet room. Before the actual experiment began, the experimenter introduced the two male characters (Mikel and Julen), showing several pictures of each to ensure that the participant was familiarised with the characters involved in the experiment. Afterwards, there was a short training session of two trials. In each trial, participants were simultaneously presented with a sentence presented auditorily and two pictures presented on the screen depicting two possible interpretations of the auditory stimuli. Participants were instructed to point to the picture that they thought best matched the auditorily presented stimulus. During the training session, two characters who were different from those used in the actual experiment performed transitive actions in a coordinated sentence. The experimental session then began. Each slide was seen by the child once or twice if required. Participants were individually tested at school on two different days during school hours. Each session lasted around 10 minutes. Like the children, the adults were tested individually. 77 3.5.1.4 Coding and data analysis The results were collected in an Excel spreadsheet by the experimenter and later transferred to the statistical programme R. Answers were analysed using a mixed-effects binomial regression model, as this is considered suitable for the analysis of categorical data (Baayen 2008, Jaeger 2008). First, the best model for each participant group was identified in order to determine whether there were any interactions. Afterwards, the main effects were calculated using effect coding; with the alternative technique of dummy coding, the main effects are not ‘real’ main effects as calculated in an ANOVA model, which result from higher or lower patterns in all experimental conditions. For between-group comparisons, the same procedure was followed. Clause order (main-subordinate vs. subordinate-main), anaphora type (forward vs. backward anaphora) and pronoun type (null vs. overt) were defined as fixed factors, and participants and items as crossed random factors. Crossed random adjustments were made for intercepts for participants and items. For the between-group comparisons (child groups vs. the adult group and the cL1 group vs. the cL2 group), the between-subject variable group was included. 3.5.1.5 Results 59 The distribution of the mean percentage 60 of preference for intrasentential coreference with the subject antecedent for each group with sentences containing NSPs and OSPs across conditions is shown in Figure 7. The columns represent the pattern of responses for each group of participants, and the eight experimental conditions under study are displayed along the x-axis in the order presented in (44). Note that the mean rate of disjoint reference (i.e. the extrasentential referent as antecedent) is simply the complementary rate presented in Figure 7. Overall, adults made a distinction between antecedent assignments depending on pronoun type, whereas antecedent choice in child groups did not depend heavily on pronoun type. 59 Part of the content of this section has been published in Iraola and Ezeizabarrena (2011) and Iraola and Ezeizabarrena (2012). 60 Percentages are rounded off to the nearest whole number in all figures in this chapter. The descriptive statistics shown in the figures are not exactly the same as the estimates derived from the statistical model, as the model adjusts for differences across subjects and items. 78 Figure 7. Preference for the subject antecedent in NSPs and hura ‘that-ABS’ by group and experimental condition First, separate statistical analyses of each participant group will be reported, starting with the adults; intergroup comparisons will then be presented. The results from the adults revealed significant main effects of pronoun type (β = 2.17, z = 7.22, p < .001) and clause order (β = -0.63, z = -2.10, p = .004). These main effects showed that, overall, preferences for the subject antecedent were significantly lower in the OSP condition than in the NSP condition, and that coreference rates were significantly lower in the main-subordinate clause order condition than in the subordinate-main condition. In addition, there was a three-way significant interaction between clause order, pronoun type and anaphora type (β = 2.59, z = 1.20, p = .046). More specifically, in the forward anaphora condition, the interaction between pronoun and clause order was non-significant, indicating that clause order did not have an effect on the referential properties of pronouns (see Figure 8). Coreference rates between matrix NSPs (91%) and the preceding lexical subject were significantly higher in sentences like (44a) than those involving matrix OSPs (10%) like (44b) (β = -5.97, z = -7.68, p < .001). At the same time, coreference rates were significantly higher between embedded NSPs and the preceding lexical subject (94%) in sentences like (44c) than those of embedded OSPs (3%) (44d) (β = -6.37, z = -7.47, p < .001). Moreover, coreference rates were significantly higher for the matrix OSPs than embedded OSPs (10% vs. 3%; β = 2.07, z = 2.08, p < .05). 79 Figure 8. Adults’ preference for the subject antecedent in forward anaphora by pronoun (Ø/ hura) and clause order In backward anaphora contexts (see Figure 9), the interaction between pronoun and clause order was significant (β = -0.69, z = -0.89, p < .05). In the subordinate-main condition, the embedded NSPs (80%) were significantly more frequently interpreted as referring to the following lexical subject in sentences like (44e) than the embedded OSPs were (14%) (44f) (β = -3.35, z = -7.48, p < .001). In the main-subordinate condition, the same pattern was observed between the matrix NSPs (44g) and OSPs (44h) (34% vs. 4%, respectively; β = -2.64, z = -4.05, p < .001). Moreover, coreference rates for the embedded NSPs were significantly higher than those of matrix NSPs (80% vs. 34%; β = 2.27, z = 5.89, p < .001), and the same held true for the embedded and matrix OSPs (14% vs. 4%; β = 1.48, z = 2.08, p < .05). Figure 9. Adults’ preference for the subject antecedent in backward anaphora by pronoun (Ø/ hura) and clause order 80 With regard to the results of the cL1 group, overall, there was a significant main effect of pronoun type, suggesting that coreference rates were significantly lower for OSPs than for NSPs (β = 0.28, z = 4.67, p < .001). There was also a highly significant main effect of anaphora type (β = -0.31, z = -5.13, p < .001), implying that coreference with the subject was preferred more in forward than in backward anaphora conditions. Only the interaction between pronoun and anaphora type approached significance (β = -0.45, z= - 1.87, p = .061), as displayed in Figure 10. Regardless of pronoun type, the cL1 group showed a general preference for coreference with the subject. However, the marginally significant interaction between pronoun and anaphora revealed that the rates for OSPs (61%) were significantly lower in comparison to those for NSPs (76%) in both forward (β= -0.80, z= -4.53, p < .001) and backward anaphora conditions (52% vs. 60%, respectively; β = - 0.36, z = -2.18, p = .029). In addition, coreference rates with the lexical subject were significantly higher for NSPs in forward (76%) than in backward anaphora structures (60%) (β = 0.84, z = 4.78, p < .001). Higher coreference rates for OSPs were also observed in forward anaphora (61%) in comparison to backward anaphora (52%) (β = 0.43, z = 2.56, p < .05). Figure 10. cL1 group’s preference for the subject antecedent by pronoun (Ø/ hura) and anaphora type For the cL2 group, there was not a main effect of pronoun type, revealing that, in general, preferences were not determined by whether the pronoun was null or overt. However, there was a significant main effect of anaphora type (β = -0.21, z = -2.28, p < .05), indicating that, overall, coreference with the subject was preferred more in forward than in backward anaphora. Only the interaction between clause order and anaphora type was significant (β = -1.20, z = -2.99, p = .003). In the main-subordination condition (see Figure 11), coreference rates were significantly higher in forward (87%) than in backward anaphora (72%) (β= 0.95, z= 4.64, p< .001). However, in the sub- 81 ordinate-main condition, coreference rates were equal in forward and backward anaphora (79%). Figure 11. cL2 group’s preference for the subject antecedent by anaphora type and clause order The comparison between adults and the cL1 group indicated main effects of pronoun type (β = 0.74, z = 6.31, p < .001) and anaphora type (β = -0.37, z = - 3.19, p < .01), suggesting that coreference rates were generally higher for NSPs than for OSPs. This pattern was more visible in forward than in backward anaphora. In addition, there was a main effect of group (β = -0.53, z = -3.87, p < .001), revealing that antecedent choices differed significantly between adults and the cL1 group: the cL1 group showed a greater preference for the subject antecedent than adults. There was also a significant three-way interaction between pronoun, anaphora and group (β = 3.36, z = 2.93, p= .004). More precisely, in forward anaphora with NSPs, both adults and the cL1 group showed a preference for the subject. However, this preference differed significantly between adults and the cL1 group; the former overwhelmingly selected the subject as the antecedent (93% vs. 76%; β = - 1.44, z = -3.32, p < .001), as seen in Figure 12. Interpretations also differed significantly between the adult and cL1 group in OSP contexts: adults chose disjoint reference, i.e. the non-coreferential interpretation (only 6% coreference with the subject), whereas the children again showed a preference for the subject antecedent (62%) (β = 3.91, z = 6.45, p < .001), although these rates were lower than those for NSPs. 82 Figure 12. Preference for the subject antecedent in forward anaphora by group and pronoun type (Ø/ hura) Figure 13 presents data from the backward anaphora condition, where adults and the cL1 group showed similar coreference rates for NSPs (57% and 60%, respectively). However, intergroup differences were significant for OSPs (β = 2.78, z = 5.42, p < .001). Adults strongly preferred the extrasentential referent as antecedent (only 9% preference for the subject), while the cL1 group did not show a clear pattern (52%). Figure 13. Preference for the subject antecedent in backward anaphora by group and pronoun type (Ø/ hura) The comparison of adults vs. the cL2 group revealed significant main effects of pronoun and anaphora, suggesting that coreference rates were significantly higher for NSPs than for OSPs (β = 0.71, z = 4.49, p < .001), and this pattern was more evident in forward than in backward anaphora (β = -0.32, z = 2.14, p = .03). The significant main effect of group suggested that antecedent preferences differed significantly between the cL2 group and adults; the 83 subject antecedent was chosen more often by the former than by the latter (β = -1.01, z = -8.81, p < .001). In addition, there was a significant interaction between pronoun, anaphora and group (β = 3.69, z = 3.18, p < .01). As shown in Figure 12 above, in the forward anaphora condition, both adults and the cL2 group chose coreference with the subject for NSPs (93% and 84%, respectively), but coreference rates were significantly higher for the former than for the latter (β= 0.90, z = 2.55, p = .011). For OSPs, in contrast, asymmetric results were found between the groups: adults opted for the extrasentential antecedent (only 6% coreference), while the cL2 group again chose coreference with the subject (81% of the time; β = 4.85, z = 9.06, p < .001). In the backward anaphora condition in Figure 13, coreference rates were lower in adults than in the cL2 group for NSPs (57% vs. 77%, respectively), with the difference being statistically significant (β = 1.17, z = 2.88, p < .001). With respect to OSPs, the opposite patterns were found between groups (β = 3.68, z = 9.29, p < .001): adults selected the extrasentential antecedent (only 9% coreference with the subject), and the cL2 group once again showed a preference for the subject (75%). When comparing child groups (cL1 vs. cL2), there was a significant main effect of anaphora, i.e. coreference rates were generally higher in forward than in backward anaphora (β = -0.27, z = -3.84, p < .01). The significant main effect of group suggested that coreference rates were higher for the cL2 group in comparison to the cL1 group (β = -0.46, z = -4.14, p < .001). No interaction involving pronoun type was significant, but a marginally significant interaction was found among clause order, anaphora and group (β = 0.02, z= -1.85, p = .06). Further analyses indicated that, as shown in Figure 14 in forward anaphora contexts, the interaction between clause order and group approached significance (β = -0.52, z = -1.93, p = .05). The cL2 group showed significantly higher coreference rates than the cL1 group in both subordinate-main (79% vs. 68%; β =1.10, z = 3.99, p < .001) and mainsubordinate conditions (86% vs. 69%; β = 0.56, z = 2.67, p < .001). 84 Figure 14. Preference for the subject antecedent in forward anaphora by group and clause order In backward anaphora sentences (Figure 15), a marginally significant interaction between clause order and group was found (β = 0.44, z =1.78, p = .07). As in forward anaphora sentences, the cL2 group chose the subject as the antecedent significantly more often than the cL1 group in both subordinatemain (79% vs. 55%; β = 0.83, z = 2.29, p < .05) and main-subordinate conditions (72% vs. 56%; β = 1.18, z = 4.99, p < .001). Figure 15. Preference for the subject antecedent in backward anaphora by group and clause order To summarise, the results of Experiment 1 revealed that pronoun type (null/ hura) had an effect on adults’ referent choices: they showed a preference for the intrasentential subject antecedent in almost all NSP conditions except in the context of backward anaphora with main-subordinate clause order, while they opted for the extrasentential antecedent in all experimental 85 conditions with the OSP hura. Both cL1 and cL2 groups instead chose the intrasentential lexical subject as the antecedent of both types of pronouns in all experimental conditions. Thus, the antecedent choice in children did not depend to any great extent on pronoun type. However, two things should be noted here: a) both child groups appeared to show a slightly stronger preference for the subject antecedent in the NSP condition than in the OSP condition, and b) the corresponding coreference rates for each pronoun were higher in forward than in backward anaphora contexts. More specifically, in the cL1 group, coreference rates were 76% for NSPs and 61% for OSPs in forward anaphora sentences; these dropped to 60% and 52%, respectively, in backward anaphora sentences (see Figure 10). The same tendency was also observed in the cL2 group, but only in the general results (see Figure 7); pronoun type had no effect on antecedent choices, as illustrated by the nonsignificant interactions where this variable was involved. In forward anaphora, coreference rates in cL2 reached 84% for NSPs and 81% for OSPs; these fell to 77% for NSPs and 75% for OSPs in backward anaphora, regardless of clause order (see Figure 7). 3.5.2 Experiment 2: bera 3.5.2.1 Participants Thirty-three children in the cL1 group (age range 6; 5-7; 6, mean age 6; 9), 25 children in the cL2 group (age range 6; 8-7; 5, mean age 6; 10) and 10 L1- Basque adults (age range 24; 3-48; 1, mean age 27; 5) took part in the experiment. Four participants (3 from cL1 and 1 from cL2) were excluded from the data analysis in accordance with the exclusion criteria described in Section 3.5.1.1. 3.5.2.2 Materials and design The same experimental items as those from Experiment 1 were used in Experiment 2, but this time NSPs were contrasted with the OSP bera ‘(s)he (him/ herself)-ABS’. (45) FORWARD ANAPHORA a. Subordinate clause [NP] + main clause [..Ø..] Mikel-Ø igo-Ø de-nean ____ haserretu egin-Ø da. Mikel- ABS get on- PF has-when ____ get angry do- PF has ‘When Mikel got on (the bus), (he) got angry.’ 86 b. Subordinate clause [NP] + main clause [bera] Mikel-Ø igo-Ø de-nean bera-Ø haserretu egin-Ø da. Mikel- ABS get on- PF has-when he- ABS get angry do- PF has ‘When Mikel got on (the bus), he got angry.’ c. Main clause [NP] + subordinate clause [..Ø..] Mikel-Ø haserretu egin-Ø da ____ igo-Ø de-nean. Mikel- ABS get angry do- PF has ____ get on- PF has-when ‘Mikel got angry when (he) got on (the bus).’ d.Main clause [NP] + subordinate clause [bera] Mikel-Ø haserretu egin-Ø da bera-Ø igo-Ø de-nean. Mikel- ABS get angry do- PF has he- ABS get on- PF has-when ‘Mikel got angry when he got on (the bus).’ BACKWARD ANAPHORA e. Subordinate clause [NP] + main clause [..Ø..] ____Igo-Ø de-nean Mikel-Ø haserretu egin-Ø da. ____Get on- PF has-when Mikel- ABS get angry do- PF has ‘When (he) got on (the bus), Mikel got angry.’ f. Subordinate clause [NP] + main clause [bera] Bera-Ø igo-Ø de-nean Mikel-Ø haserretu egin-Ø da. He- ABS get on- PF has-when Mikel- ABS get angry do- PF has ‘When he got on (the bus), Mikel got angry.’ g. Main clause [NP] + subordinate clause [..Ø..] ____Haserretu egin-Ø da Mikel-Ø igo-Ø de-nean. ____Get angry do- PF has Mikel- ABS get on- PF has-when ‘(He) got angry when Mikel got on (the bus).’ h. Main clause [NP] + subordinate clause [bera] Bera-Ø haserretu egin-Ø da Mikel-Ø igo-Ø de-nean. He- ABS get angry do- PF has Mikel- ABS get on- PF has-when ‘He got angry when Mikel got on (the bus).’ 87 3.5.2.3 Results The mean rates of preference for coreference with the subject per group are presented in Figure 16 below. Figure 16. Preference for the subject antecedent in NSPs and bera ‘(s)he (him/ herself)-ABS’ by group and experimental condition Starting with the results from the adults, there were main effects of pronoun (β = 1.46, z = 5.80, p < .001) and anaphora (β = -1.21, z = -4.82, p < .001), suggesting that coreference rates were higher in NSP contexts than in OSP contexts, and also that the subject was more frequently chosen as the antecedent in forward anaphora conditions than in backward conditions. Further analyses showed that there was a significant interaction between pronoun type, clause order and anaphora type (β = 2.16, z = 2.13, p = .034). More specifically, in the forward anaphora condition, the interaction between pronoun type and clause order was non-significant (see Figure 17), indicating that clause order did not affect the antecedent choices of the pronouns. In the subordinate-main condition (45a), the matrix NSPs were overwhelmingly coreferential with the preceding subject (90%), whereas the matrix OSPs were interpreted as disjoint in reference in sentences like (45b) (29%) (β = -3.27, z = -7.07, p < .001). The same pattern was found for embedded NSPs and OSPs in the main-subordinate condition ((45c) and (45d) respectively), with coreference rates significantly higher in the former (93%) than in the latter (39%) (β = -4.33, z = -6.66, p < .001). 88 Figure 17. Adults’ preference for the subject antecedent in forward anaphora by pronoun type (Ø/ bera) and clause order In the backward anaphora condition (see Figure 18), the interaction between clause order and pronoun type was significant (β = -1.85, z = -2.58, p = .009), suggesting that clause order had an effect on the antecedent choice of NSPs and OSPs. Whereas the embedded NSPs were usually interpreted as coreferent with the following lexical subject (69%), as in (45e)), the matrix NSPs were interpreted as referring to the extrasentential antecedent in sentences like (45g) (19% coreference with the subject). This difference was statistically significant (β = 2.29, z = 6.09, p < .001). In contrast, the antecedent assignment for OSPs remained relatively constant, referring to the extrasentential referent in both the subordinate-main condition (45f) (only 10% coreference with the subject) and the main-subordinate condition (45g) (6%). In addition, significant differences were observed between the rates of NSPs and OSPs in subordinate-main (69% vs. 10%; β = -3.03, z = -6.76, p < .001) and mainsubordinate contexts (19% vs. 6%; β = -1.30, z = -2.31, p < .05). Figure 18. Adults’ preference for the subject antecedent in backward anaphora by pronoun type (Ø/ bera) and clause order 89 With regard to the cL1 group, there was a main effect of pronoun (β = 0.28, z = 3.04, p < .01), suggesting that, overall, coreference rates were lower in OSP contexts than in NSP contexts. The significant main effect of anaphora indicated that coreference rates were higher in forward than in backward anaphora (β = -0.56, z = -5.86, p < .001). There was also a marginally significant three-way interaction between pronoun type, clause order and anaphora type (β = -0.76, z =-1.72, p = .08). In forward anaphora sentences (see Figure 19), the interaction between pronoun type and clause order was significant (β = -0.81, z = -2.44, p < .05). More precisely, in the forward anaphora with the subordinate-main clause order condition, the antecedent most frequently chosen by the cL1 group was the subject, both in NSP (85%) and in OSP contexts (65%). These differences in the coreference rates for NSPs and OSPs were significant (β = -1.18, z= -5.01, p < .001). In the mainsubordinate condition, cL1 again exhibited a preference for the subject, showing marginally significant differences between the coreference rates for NSPs (82%) and those for OSPs (75%) (β = -0.47, z = -1.96, p = .05). However, the coreference rates for OSPs differed significantly between subordinatemain and main-subordinate conditions (65% vs. 75%; β = -0.56, z = -2.61, p = .009). Figure 19. cL1 group’s preference for the subject antecedent in forward anaphora by pronoun type (Ø/ bera) and clause order In the backward anaphora condition (Figure 20), coreference rates for both NSPs and OSPs were lower than in forward anaphora sentences, and they approached chance level in the main-subordinate condition. The interaction between clause order and pronoun type was non-significant. Unlike in forward anaphora sentences, the coreference rates for NSPs and OSPs did not differ significantly between the two types of clause order. However, coreference rates were higher for both NSPs and OSPs in subordinate-main clause order (65% and 58%, respectively) than in main-subordinate clause order 90 (55% and 50%, respectively). These differences in rates were statistically significant for both NSPs (β = 0.45, z = 2.26, p < .05) and OSPs (β = 0.48, z = 2.17, p < .05). Figure 20. cL1 group’s preference for the subject antecedent in backward anaphora by pronoun type (Ø/ bera) and clause order Data analysis of the cL2 group’s results revealed a significant main effect of anaphora (β = -0.55, z = -3.65, p < .001), indicating that coreference rates were significantly higher in forward than in backward anaphora. Only the interaction between clause order and anaphora type was significant (β = - 1.12, z = -3.33, p < .001), as illustrated in Figure 21. Further analyses showed that coreference rates were significantly higher in forward than in backward anaphora structures in both subordinate-main (69% vs. 55%; β = 0.65, z = 2.71, p < .01) and main-subordinate conditions (73% vs. 42%; β = 1.54, z = 3.38, p < .001). Figure 21. cL2 group’s preference for the subject antecedent by anaphora type and clause order 91 With regard to the comparison between adults and the cL1 group, there were significant main effects of pronoun (β = 0.56, z = 4.42, p < .001) and anaphora (β = -0.68, z = -5.00, p < .001). These effects indicated that coreference with the subject antecedent was lower for OSPs than NSPs, and that the subject antecedent was chosen more frequently in forward than in backward anaphora. In addition, there was a significant main effect of group (β = -0.63, z = -3.52, p < .001), suggesting that the cL1 group had an overall greater preference for the subject than adults. The interaction between pronoun, anaphora type and group was significant (β = 2.09, z = 2.49, p = .013). Further analyses showed that in the forward anaphora condition (Figure 22), the interaction between pronoun type and group was significant (β = 2.82, z = 6.86, p < .001). A strong preference for the subject was observed in NSPs for both adults (92%) and cL1 (83%). However, significant differences were found in the interpretation of OSPs: adults opted for the extrasentential antecedent (32% subject coreference), while cL1 again displayed a preference for the subject (70%), with the difference between groups being statistically significant (β = 3.46, z = 4.97, p < .001). Figure 22. Preference for the subject antecedent in forward anaphora by group and pronoun type (Ø/ bera) For the backward anaphora condition (Figure 23), the interaction between pronoun and group was significant (β = -2.00 z = -4.53, p < .001). Adults more frequently selected the extrasentential referent as the antecedent of NSPs (44% coreference with the subject), whereas the cL1 group showed a preference for the subject antecedent (60%) (β = 0.77, z = 2.19, p < .05). The differences between groups were also significant in OSPs: cL1 opted for coreference in 54% of the cases, whereas adults predominantly selected the extrasentential antecedent (only 8% coreference) (β = 3.46, z = 4.97, p < .001). 92 Figure 23. Preference for the subject antecedent in backward anaphora by group and pronoun type (Ø/ bera) Concerning the comparison between adults and cL2, there were main effects of pronoun type (β= 0.59, z= 4.21, p < .001) and anaphora (β = -0.68, z = - 4.84, p < .001), indicating that coreference rates were generally lower in OSP contexts than in NSP contexts, and coreference with the subject was also chosen less frequently in backward anaphora in comparison to forward anaphora. There was also a significant main effect of group (β = -0.41, z = - 2.65, p < .01), suggesting that the interpretations between the two groups differed significantly: the cL2 group chose coreference more often than adults. In addition, the interaction between pronoun type, anaphora and group was significant (β = 1.68, z = 1.99, p < .05). Specifically, in the forward anaphora condition (see Figure 22), the interaction between pronoun type and group was significant (β = 2.80, z = 6.95, p < .001), which revealed a smaller pronoun type effect for the cL2 group than for the adults. The cL2 group chose the subject as the antecedent of NSPs significantly less often (78%) than adults (92%) (β = -1.22, z = -3.22, p < .01). In OSP contexts, however, the intergroup differences were more pronounced: adults generally preferred the extrasentential referent as the antecedent (32% coreference), whereas the cL2 group selected coreference with the subject (63%) (β = 1.53, z = 3.97, p < .001). In the backward anaphora condition (Figure 23), the interaction between pronoun type and group was significant (β =1.87, z = 4.94, p < .001). Between-group differences were non-significant in NSP contexts (44% and 53% coreference for adults and cL2, respectively), but differences with respect to OSPs were significant despite the groups’ common preference for the extrasentential antecedent: 8% coreference for adults and 44% for cL2 (β = 1.87, z = 4.94, p < .001). The statistical analysis between the cL1 and cL2 child groups found a significant main effect of anaphora type (β = 1.11, z = -4.06, p < .001), with higher coreference rates in forward than in backward anaphora. In addition, the interaction between clause order, anaphora type and group approached 93 significance (β = -0.92, z = -1.96, p = .05). More precisely, in the forward anaphora condition shown in Figure 24, the interaction between clause order and group was non-significant (β = -0.02, z = -0.09, p = .93), indicating that the effect of clause order did not differ between the groups. There was a general preference for coreference with the subject exhibited by the cL1 and cL2 groups, both in the subordinate-main condition (75% and 59%, respectively) and the main-subordinate condition (78% and 73%, respectively). These differences between the child groups were not found to be significant. Figure 24. Preference for coreference with the subject in forward anaphora by group and clause order In the backward anaphora condition (Figure 25), the interaction between clause order and group was non-significant (β = 0.23, z = 1.05, p = 0.29). The cL1 group showed (non-significantly) higher coreference rates than the cL2 group in subordinate-main clause order (61% and 55%, respectively), as well as in the reverse order (53% and 42%, respectively). Figure 25. Preference for coreference with the subject in backward anaphora by group and clause order 94 To summarise the results of Experiment 2, adults made a distinction regarding antecedent assignment based on the pronoun type: null vs. overt bera. In general, they overwhelmingly interpreted NSPs as coreferent with the lexical subject of the previous and the following clause. For the OSP bera condition, adults predominantly selected the disjoint reading, preferring an extrasentential referent as the antecedent. In contrast, the antecedent choices in the cL1 and cL2 groups did not depend on pronoun type to a great extent. Both child groups chose the subject referent as the antecedent of both NSPs and OSPs in both experimental conditions, but their judgements were at or near chance level for backward anaphora. 3.5.3 Discussion 3.5.3.1 Pronoun type in forward anaphora Experiments 1 and 2 were designed to test how NSPs are interpreted in Basque in comparison to the OSP hura ‘that-ABS’ and bera ‘(s)he (him/ herself)-ABS’, as well as to determine whether the results were consistent with the division of labour formulated under the PAS (Carminati 2002). Three variables — namely, pronoun realisation, clause order and anaphora type — were manipulated to investigate their influence on anaphora resolution and whether or not 6to 8-year-olds would interpret pronouns in an adult-like fashion. Pronoun type (null/ hura and null/ bera) did not greatly affect the interpretations of the cL1 and cL2 groups, but this factor did have an effect on the adults’ results. Both child groups and the adults interpreted NSPs as coreferential with the only intrasentential referent, the subject. However, antecedent choices differed significantly in OSP contexts between the age groups: children selected the subject as the antecedent (especially with hura), whereas adults chose the extrasentential referent. The results of Experiment 1 will be discussed first, and then those of Experiment 2. This section only discusses the results of forward anaphora; for backward anaphora, see Section 3.5.3.2. In forward anaphora contexts, regardless of the position of the NSP in the complex sentence (whether it was in the subordinate (44a) or main clause (44c)), NSPs predominantly coreferred with the preceding lexical subject in adults (93%). In contrast, the adults chose the extrasentential referent as the antecedent of the OSP hura, also regardless of whether it appeared in the subordinate (44b) or main clause (44d) (only 6% coreference; see Figure 8). Thus, clause order had no effect on the adults’ results, as the rates cited above represent the mean values of NSPs and OSPs obtained in subordinatemain and main-subordinate conditions. The cL1 group also preferred the preceding intrasentential lexical subject as the antecedent of both embedded and matrix NSPs in the forward anaphora condition (76%). A similar pattern was observed for the embedded and matrix OSP hura, which was interpreted as coreferential with the preceding 95 lexical subject in 61% of cases (see Figure 10). However, the differences in the coreference rates for NSPs and OSPs were significant, which suggests a residue of the division of labour between the two types of pronouns. With regard to the cL2 group, pronoun type did not affect their results; there was an overwhelming preference for the subject antecedent regardless of pronoun type (NSP/ OSP). A comparison of the results for the adults and the child groups revealed that the patterns of responses between groups were similar in NSP contexts (see Figure 12). Both cL1 and cL2 groups exhibited a preference for the preceding lexical subject (76% and 84%) similar to that of the adults (93%), but to a significantly lower degree. With regard to the OSP hura, the preferences between groups differed significantly: both children’s groups again showed a preference for coreference with the subject (62% and 81%, respectively), whereas adults predominantly chose the extrasentential antecedent (only 6% coreference). When the results of the two child groups were compared, no significant differences were observed with respect to the referential properties of NSPs and OSPs, as illustrated by the non-significant interactions involving the variables pronoun type and group. However, the significant interaction between anaphora, clause order and group revealed that in forward anaphora contexts, the cL2 group showed higher coreference rates than the cL1 group, both in subordinate-main (79% vs. 68%) and main-subordinate conditions (86% vs. 69%; see Figure 14). The patterns of results obtained for NSPs resemble previous crosslinguistic findings in null-subject languages testing one-referent sentences (Alonso- Ovalle et al. 2002, Carminati 2002). For OSPs, the Basque data are in line with the unstable and variable preferences of OSPs predicted in the PAS, with the results showing evidence of this variability in two different domains: from a developmental (children vs. adults) perspective and from a crosslinguistic (Italian vs. Basque adults) perspective. We will first attempt to explain the differences between adults and children. In general, Basque adults interpreted the demonstrative hura ‘that’ as non-coreferential with the only intrasentential antecedent (i.e. the subject), referring instead to an extrasentential antecedent. In contrast, both child groups considered it to be anaphoric, a property typically associated with pronouns and not usually with demonstratives, which are place or spatial deictics (Diessel 1999). This raises a question about the description of hura: does the habitual description of the use of hura support the disjoint interpretation preferred by adults? We first examine the possible explanations for the adults’ rejection of coreference with hura. In Section 1.3, we explained the two possible ways of establishing an anaphoric relation between a pronoun and a preceding referential NP: either through variable binding or coreference (Grodzinsky & Reinhart 1993, Reinhart 1983). In forward anaphora sentences with subordinate-main clause order 96 (as in (44b), repeated here for convenience), hura ‘that-ABS’ is not syntactically bound to the preceding lexical subject Mikel: Mikel i igo denean hura j haserretu egin da. ‘When Mikel got on (the bus), he got angry.’ In such a case, Rule I* (cited in Section 1.3) is operative, and Basque adults behave accordingly: Rule I* If a pronoun is not (semantically) bound by an NP A, it is generally interpreted as non-coreferential with A, unless it appears in an Evans-style context. (taken from Guasti 2004: 282) Because the demonstrative hura ‘that’ is unbound in (44b), it should be interpreted as non-coreferential (disjoint in reference), since that is the default reading. For forward anaphora sentences with main-subordinate clause order such as (44d), repeated below, a different explanation is needed, because in this context, hura ‘that’ is syntactically bound. Despite fulfilling the role of a third-person pronoun when an overt referent is required, hura behaves like an ordinary demonstrative with respect to binding principles. Under Principle C, hura must be disjoint from any c-commanding antecedent, i.e. it must be free everywhere (Eguzkitza 1986: 204; see Section 3.2.1). This explains why Basque adults’ disjoint interpretation is the only grammatical interpretation of hura: Mikel i haserretu egin da hura *i/ j igo denean. ‘Mikel got angry when he got on (the bus).’ As mentioned in Section 1.3.1, demonstratives, unlike personal pronouns, cannot overrule their disjointness requirement through accidental coreference, where Principle B is violated by some grammatical process (Cardinaletti & Starke 1999b: 284). Thus, cL1 and cL2’s coreferential reading of hura is ungrammatical; see Section 4.2.4 for an explanation of children’s target-deviant performance. Turning to the crosslinguistic variability between Italian (Carminati 2002) and Basque results, Basque adults’ preference for the extrasentential referent (90%) contrasts with Italian adults’ preference for coreference with the subject (86%) (see example (33b) in Section 1.4). Note, however, that Basque adults’ disjoint reading of the OSP hura is in accordance with the overall prediction of the PAS that OSPs will choose a non-subject antecedent (in the case at hand, the extrasentential referent); surprisingly, this preference is 97 more robust in Basque than in Italian (94% in Basque adults vs. 14% in Italian adults). Differences in the properties of the pronoun lui ‘he’ in Italian and the Basque demonstrative pronoun hura seem to offer an explanation for the opposite patterns observed in the two null-subject languages. This will be explored in more detail in Section 4.1.4. As far as the results of Experiment 2 are concerned, as in Experiment 1, in the forward anaphora condition, the strong bias of the PAS in assigning the subject as the antecedent to NSPs was found in adults regardless of whether the NSP occurred in the subordinate (45a) or matrix clause (45c) (91%; see Figure 17). For the OSP bera, the extrasentential referent was the antecedent preferred by adults in both embedded (45b) and matrix OSPs (45d) (only 32% subject coreference). The cL1 group chose the preceding lexical subject as the antecedent of NSPs (84%) regardless of whether the NSP was the embedded or the matrix subject. For bera, the cL1 group again selected coreference for embedded NSPs (65%) and to a significantly higher degree for matrix NSPs (75%). Similar to the outcome from Experiment 1, pronoun type had no effect on the results of the cL2 group. In intergroup comparisons between children and adults, both cL1 and cL2 groups preferred the subject as the antecedent for NSPs (83% and 78% of the time), a rate similar to that of adults (92%), as shown in Figure 22. The differences in the rates of NSPs between cL2 and adults were significant. For the OSP bera, both children’s groups again chose coreference with the preceding lexical subject, differing significantly from the adults, who selected the extrasentential referent as the antecedent of bera (only 32% coreference with the subject). A comparison of the two child groups showed that (similar to Experiment 1) only the interaction between clause order, anaphora and group approached significance. In forward anaphora sentences, the cL1 group’s coreference rates were (non-significantly) higher than those of cL2, both in subordinate-main (75%/ 59%) and main-subordinate conditions (78%/ 73%). Adults’ disjoint reading of bera (66%) in forward anaphora sentences such as (45b), shown below, reveals certain unanticipated aspects of the properties of bera. Here, the behaviour of bera diverges from the explanation given in descriptive grammars (i.e. that it corefers with the nearest previously mentioned referent; Saltarelli 1988; see Section 3.2.1). Note, however, that the permissibility of intrasentential coreference with the subject is still greater for bera (34%) than for hura (Experiment 1) in the same context (only 6% coreference). This suggests that the properties of the pronoun bera are closer to the Italian third-person pronoun, as will be discussed in Section 4.1.4. Mikel i igo denean bera j>i haserretu egin da. ‘When Mikel got on (the bus), he got angry.’ 98 Basque children’s coreferential interpretation of bera is at odds with the interpretation of adults. The use of OSPs in null-subject languages in topiccontinuity contexts is regarded as redundant. However, it is important to note that the children’s tendency to corefer bera with an intrasentential referent cannot be considered pragmatically incorrect, due to its discourse-bound properties (Eguzkitza 1986). For children, the intrasentential preceding subject is a much more accessible referent for the OSP in terms of distance (Ariel 1990) than the extrasentential referent, which was not linguistically mentioned and was only visually present. Adults instead looked for an appropriate antecedent for overt vs. null pronouns and mapped the form of a referring expression to the degree of salience, which is in line with Accessibility Theory (Ariel 1990, 2001; see Section 1.2). The most underspecified forms (such as NSPs) were interpreted as referring to the most salient (in terms of topichood or subjecthood) entity, the intrasentential antecedent (Mikel in ((44b) and (45b)). The OSP, as a more complex form in terms of morphological encoding, referred to a less salient referent, i.e. the extrasentential referent. 3.5.3.2 Clause order and anaphora type One of the questions under investigation was whether clause order (mainsubordinate clause vs. subordinate-main clause) would affect the interpretations of the participants (RQ1a). The data from forward anaphora sentences discussed in the previous section revealed that the position of the (null/ overt) pronoun in the complex sentence — whether it occurred in the main or subordinate clause — did not affect the results of adults and children. In the cL2 group, the significant interactions obtained between clause order and anaphora in Experiments 1 and 2 indicated that coreference rates were significantly higher in forward anaphora than backward anaphora (see Figures 11 and 21) in both both clause orders. In backward anaphora sentences, clause order in relation with anaphora type had an effect on a particular context in the data obtained from adults. This can be seen in the significant interactions observed between clause order and pronoun type (see Figures 9 and Figure 18). In matrix NSPs, coreference rates decreased to 34% in Experiment 1 and to 19% in Experiment 2 in structures like (44g) and (45g), repeated below: ___ i Haserretu egin da Mikel *i/ j igo denean. ‘(He) got angry when Mikel got on (the bus).’ In such a context, as predicted in (RQ1b), the adults discarded coreference between the c-commanding NSP and the following lexical subject, since Rexpressions like Mikel have to be free everywhere (i.e. they cannot be ccommanded by a pronoun). Thus, the NSP’s PAS-based bias towards the subject weakened due to the grammatical constraint known as Principle C 99 (Chomsky 1981; see Section 3.1). With matrix OSPs in backward anaphora contexts ((44h) and (45h)), adults exhibited an even clearer rejection of coreference than that shown with NSPs (only 4%/ 7% coreference with the subject). See Section 4.1.3 for possible explanations for these patterns. The cL1 and cL2 groups’ interpretations in backward anaphora differed greatly from adults, as can be seen in the absence of a significant interaction between clause order and pronoun type, signalling no differences in the antecedent choice. Preference for the following lexical subject was observed for both matrix NSPs and OSPs in backward anaphora contexts in Experiment 1 (60%/ 53% for cL1 and 72%/ 73% for cL2), and rates were often at chance level for matrix NSPs and OSPs in Experiment 2 (55%/ 50% for cL1 and 48%/ 36% for cL2). In the backward anaphora condition with embedded NSPs, where coreference is syntactically allowed (as in ((44e) and (45e)), the adults chose the following lexical subject as the antecedent of NSPs in Experiments 1 and 2 (80% and 69% of the time, respectively), a pattern similar to that exhibited by the children (60%/ 65% for cL1 and 82%/ 59% for cL2). For embedded OSPs, as in (44f) and (45f), adults generally selected the extrasentential antecedent (only 14%/ 10% coreference with the subject), differing significantly from the slight preference for the subject shown by children (51%/ 58% for cL1 and 77%/ 52% for cL2). See Section 4.1.3 for a discussion of these patterns. It should be noted that whereas adults showed sensitivity towards the type of pronoun in their responses, children relied more heavily on anaphora directionality than on pronoun type. This is particularly evident in the general response patterns of the cL1 group, where coreference rates for the two pronoun types were more similar within each type of anaphora, with rates ranging between 60% and 85% in forward anaphora contexts and close to chance level in backward anaphora contexts (50%-65%; see Figures 7 and 16). The effect of anaphora type in children is discussed in greater detail in Section 4.2.3. 3.5.4 Experiment 3: har(e)k As discussed with regard to RQ1c in Section 3.3, one of the aims of the present study is to explore whether the division of labour between NSPs and OSPs is more prevalent in two-referent sentences. In such sentences, the distinct biases of the PAS for NSPs and OSPs are applied in a more restrictive way, as there are more potential antecedents (Carminati 2002). Consequently, Experiments 3 and 4 investigate the references of null and overt pronouns in two-referent sentences using the same methodology as in Experiments 1 and 2. This allows us to determine whether the referentiality of these pronouns remains the same as in one-referent sentences (Experiments 1 and 2) or changes depending on contextual factors such as the number of referents (one or two potential intrasentential antecedents). 100 3.5.4.1 Participants Thirty-three children in the cL1 group (age range 6; 1-7; 0, mean age 6; 4), 15 children in the cL2 group (age range 7; 2-8; 0, mean age 7; 5) and 10 L1-Basque adults (age range 18; 8-65; 2, mean age 30; 4) participated in the experiment. However, 10 participants (7 from the cL1 group and 3 from the cL2 group) were excluded from the data analysis, either because they did not complete the task or because they selected the logically impossible option more than 10 times out of 42 items (see the next section for more details). 3.5.4.2 Materials The complex sentences consisted of two clauses containing morphologically transitive predicates with the transitive auxiliary *edun ‘to have’. Thus, the subjects of the two clauses were always matched for number (singular) and case (ergative). However, only one of the clauses had two arguments (subject and object) — for example, in clauses with the verb bultzatu ‘push’, the auxiliary shows agreement with both the subject (ergative) and the object (absolutive), as seen in (46). The other clause in (46) with the unergative verb txistu egin ‘whistle (do whistle)’ has just one argument, the ergative subject. Constructions such as txistu egin are comparable to complex predicates composed of the verb egin ‘do’ and a (typically) bare nominal indicating the kind of action being performed (Etxepare 2003: 394). Note that the ergative form of hura is hark in Standard Basque; however, in the dialect under study, Gipuzkoan, har(e)k is a more common form. Listed below are the eight experimental conditions for the verb pair bultzatu-txistu egin ‘push-whistle’: (46) FORWARD ANAPHORA a. Subordinate clause [NP] + main clause [..Ø..] Mikel-ek Julen-Ø bultza-tu du-enean ____txistu egin-Ø du. Mikel- ERG Julen- ABS push- PF has-when____whistle do- PF has ‘When Mikel pushed Julen, (he) whistled.’ b. Subordinate clause [NP] + main clause [har(e)k] Mikel-ek Julen-Ø bultza-tu du-enean har(e)-k txistu egin-Ø Mikel- ERG Julen- ABS push- PF has-when he- ERG whistle do- PF du. has ‘When Mikel pushed Julen, he whistled.’ c. Main clause [NP] + subordinate clause [..Ø..] Mikel-ek txistu egin-Ø du ____Julen-Ø bultza-tu du-enean. Mikel- ERG whistle do- PF has____Julen- ABS push- PF has-when ‘Mikel whistled when (he) pushed Julen.’ 101 d.Main clause [NP] + subordinate clause [har(e)k] Mikel-ek txistu egin-Ø du har(e)-k Julen-Ø bultza-tu Mikel- ERG whistle do- PF has he- ERG Julen- ABS push- PF du-enean. has-when ‘Mikel whistled when he pushed Julen.’ BACKWARD ANAPHORA e. Subordinate clause [NP] + main clause [..Ø..] ____Julen-Ø bultza-tu du-enean Mikel-ek txistu egin-Ø ____ Julen- ABS push- PF has-when Mikel- ERG whistle do- PF du. has ‘When (he) pushed Julen, Mikel whistled.’ f. Subordinate clause [NP] + main clause [har(e)k] Har(e)-k Julen-Ø bultza-tu duenean Mikel-ek txistu He- ERG Julen- ABS push- PF has-when Mikel- ERG whistle egin-Ø du. do- PF has ‘When he pushed Julen, Mikel whistled.’ g. Main clause [NP] + subordinate clause [..Ø..] ____Txistu egin-Ø du Mikel-ek Julen-Ø bultza-tu ____Whistle do- PF has Mikel- ERG Julen- ABS push- PF du-enean. has-when ‘(He) whistled when Mikel pushed Julen.’ h. Main clause [NP] + subordinate clause [har(e)k] Har(e)-k txistu egin-Ø du Mikel-ek Julen-Ø bultza-tu He- ERG whistle do- PF has Mikel- ERG Julen- ABS push- PF du-enean. has-when ‘He whistled when Mikel pushed Julen.’ Each experimental sentence in (46) was presented auditorily together with three drawings of three male characters (Mikel, Julen and Gorka). Note, however, that the linguistic items in (46) differ in terms of the number of appropriate interpretations for NSPs and OSPs. Figure 26 corresponds to the linguistic items (46a,b) Mikelek Julen bultzatu duenean___/ har(e)k txistu egin du 102 ‘When Mikel pushed Julen,___/ he whistled’ and also to (46g,h) ___/ Har(e)k txistu egin du Mikelek Julen bultzatu duenean ‘He whistled when Mikel pushed Julen’. In Picture B, Mikel (the boy with dark hair) performs both the action described in the main clause and that described in the subordinate clause (46a,b), namely pushing and whistling. In Picture C, Mikel performs the action described in the first clause (pushing) (46a,b), whereas Julen (the boy with curly hair in a striped pullover) carries out the action of the second clause (whistling). In Picture A, Mikel pushes Julen, and it is Gorka (the character wearing boots) who is whistling. Figure 26. Visual stimulus provided together with the auditory stimuli (46a,b) and (46g,h) The target referent could be present visually and/ or auditorily, and there were three possible interpretations of the relationship between the pronoun and the antecedent. Selecting Picture A meant that the participant opted for disjoint reference, taking the extrasentential referent (not verbally mentioned but visually present) Gorka as antecedent. Picture B represented intrasentential coreference between the pronoun and the lexical subject of the preceding clause, Mikel. Finally, choosing Picture C implied intrasentential coreference between the pronoun and the object of the preceding clause, Julen. The other four experimental items (46c-f) were also presented simultaneously with three pictures, but there were only two possible readings of the sentences; one of the pictures was logically impossible. The visual items corresponding to the linguistic items in (46c,d) Mikelek txistu egin du ___ / har(e)k Julen bultzatu duenean ‘Mikel whistled when ___/ he pushed Julen’ and (46e,f) ___/ Har(e)k Julen bultzatu duenean Mikelek txistu egin du ‘When ___/ he pushed Julen, Mikel whistled’ are shown in Figure 27. 103 Figure 27. Visual stimulus provided together with the auditory stimuli (46c-f) On the basis of the characters’ identities and the predicates, only Picture A or B could match the sentence. Picture C was logically impossible, as Gorka was whistling in the picture; in the experimental sentence, it was specifically stated that it was Mikel who was whistling. The two remaining possible interpretations involved either intrasentential coreference, referring to the preceding or following lexical subject Mikel (Picture A), or the disjoint reading, referring to the extrasentential antecedent Gorka (Picture B). 3.5.4.3 Coding and data analysis The data were analysed in two different tests. First, the experimental items with three possible readings — as in (46a,b,g,h) — were analysed in a multinomial logistic regression for each group of participants, with antecedent choice as a three-level dependent variable (subject, object and other, i.e. the extrasentential referent). For between-group comparisons, pairwise comparisons were made for each experimental condition. Second, the experimental items with two possible interpretations — as in (46c-f) — were analysed using a mixed-effects binomial regression model, similar to the models employed in Experiments 1 and 2. Logically impossible responses were coded as missing data (NA: not available in R) and were not included in the statistical analysis. Although Experiments 3 and 4 had a 2(pronoun type) x 2(anaphora type) x 2(clause order) design like Experiments 1 and 2, the analysis of the data in two tests meant that one of the levels of the variable clause order matched a specific level of the variable anaphora type. More precisely, forward anaphora items from the multinomial analysis had subordinate-main clause order, and backward anaphora items exhibited mainsubordinate clause order. In contrast, forward anaphora items from the binomial analysis (46c-f) had main-subordinate clause order, whereas backward anaphora items had subordinate-main clause order. Thus, the data analysis was ultimately carried out based on a 2 x 2 design in which anaphora type (forward vs. backward anaphora) and pronoun type (NSP vs. OSP) were defined as fixed factors. When reporting results in the next section, we 104 will refer only to the type of anaphora. Participants and items were defined as crossed random factors. 3.5.4.4 Results In the following paragraphs, the results of the multinomial regression for each group of participants with respect to the constructions found in (46a,b,g,h) will be presented; this will be followed by between-group pairwise comparisons (see Figure 31). The results obtained from the items in (46c-f) will then be reported using a binomial regression analysis for withinand between-group analyses. We begin with the results of the multinomial regression for each group, starting with the adults. There was a significant interaction between anaphora and pronoun (β = 3.41, t = 3.92, p < .001), indicating that the effect of anaphora type was larger for NSPs than for OSPs. Specifically, in the forward anaphora condition with matrix NSPs (as in (46a)), the preceding lexical subject was chosen significantly more often (76%) than either the object (14%) (β = -1.71, t = -5.23, p < .001) or the extrasentential referent (10%) (β = - 2.03, t = -5.40, p < .001; see Figure 28). In contrast, in backward anaphora contexts with matrix NSPs (46g), the following lexical subject (18%) was chosen as the antecedent significantly less often than the extrasentential referent (73%) (β = 1.42, t = 4.77, p < .001); the object (10%) was also chosen significantly less often than the extrasentential referent (73%) (β = 1.98, t = 5.25, p < .001). With regard to the antecedent choice for OSPs, adults selected the subject and the object significantly less often than the extrasentential referent in both forward anaphora conditions (as in (46b)); β = 4.92, t = 7.73, p < .001 and β = 3.20, z = 4.63, p < .001, respectively) and backward anaphora conditions (as in (46h)); β = 1.52, t = 2.57, p < .05 and β = 2.34, t = 2.17, p < .05, respectively). 105 Figure 28. Adults’ antecedent choices for NSPs and har(e)k in forward and backward anaphora A multinomial regression analysis of the cL1 group’s data revealed that the interaction between pronoun and anaphora was non-significant, indicating that antecedent choices did not differ on the basis of anaphora directionality. In the forward anaphora condition with matrix NSPs, the preceding lexical subject was significantly preferred (57%) over the extralinguistic antecedent (20%) (β = 1.06, t = 5.88, p < .001; see Figure 29). The subject was also more frequently selected as the antecedent for matrix OSPs (52%) than the object (22%) (β = -1.04, t = -3.77, p < .001) and the extrasentential referent (26%) (β = -1.78, t = -7.19, p < .001). In the backward anaphora condition, the most frequently chosen antecedent for matrix NSPs was the following lexical subject (50% of the time), with a significant difference in selection rate in comparison to the extrasentential antecedent (26%) (β = 0.63, t = 3.82, p < .001). For matrix OSPs, the subject was again the most preferred antecedent choice (59%), with rates differing significantly from both those of the object (18%) (β = -1.06, t = -3.92, p < .001) and the extrasentential referent (22%) (β = -1.59, t = -6.69, p < .001). 106 Figure 29. cL1 group’s antecedent choices for NSPs and har(e)k in forward and backward anaphora With respect to the antecedent assignments in the cL2 group displayed in Figure 30, the interaction between pronoun and anaphora turned out to be non-significant. In forward anaphora sentences with matrix NSPs, the preceding lexical subject was chosen significantly more often (47% of cases) than either the object (38%) (β = 0.88, t = 2.85, p < .01) or the extrasentential referent (16%) (β = 1.09, t = 3.68, p < .001). In addition, the object was chosen more often (38%) than the extrasentential referent (16%), and this difference was significant (β = -0.88, t = -2.85, p < .01). A similar pattern was also found in the antecedent choice of matrix OSPs in forward anaphora contexts: the subject was clearly preferred as the antecedent choice (64%) over the object (21%) (β = -1.99, t = -4.96, p < .001) and the extrasentential referent (16%) (β = -2.50, t = -6.03, p < .001). In backward anaphora sentences with matrix NSPs, the subject was chosen significantly more often (47%) than the extrasentential referent (25%) (β = 0.63, t = 2.49, p < .05). With respect to matrix OSPs, the most-frequently chosen antecedent was again the subject (53% of the time); this differed significantly from the rates corresponding to the object (25%) (β = -0.87, t = -2.33, p < .05) and the extrasentential referent (22%) (β = -1.52, t = -4.19, p< .001). 107 Figure 30. cL2 group’s antecedent choices for NSPs and OSP har(e)k in forward and backward anaphora In order to compare the responses between groups of participants, pairwise comparisons were performed for each type of experimental condition, as shown in Figure 31. The results indicated that the differences between adults and children were significant under all experimental conditions except for NSPs in forward anaphora contexts, where the rates of selecting the extrasentential referent did not differ between adults and children. In addition, the rates of selecting the object in the forward anaphora condition with NSPs and OSPs did not differ significantly between the cL2 group and adults. Interestingly, there were almost no statistically significant differences in the overall patterns of interpretation between cL1 and cL2. Significant differences between the child groups were only found with respect to the selection of the object in the forward anaphora condition with NSPs and the selection of the subject in forward anaphora contexts with OSPs. See the footnote for the statistical analyses 61 for each experimental condition. 61 Reading from left to right on the graph: forward anaphora with NSPs: subject adults vs. cL1 (β = -0.90, z = -2.72, p < .01), adults vs. cL2 (β = -1.30, z = -3.85, p < .001), object adults vs. cL1 (β = 0.77, z = 2.24, p < .05), cL1 vs. cL2 (β = 1.11, z = 3.64, p < .001). Forward anaphora with OSPs: subject adults vs. cL1 (β = 2.09, z = 3.16, p < .01), adults vs. cL2 (β = 4.69, z = 1.06, p < .001), cL1 vs. cL2 (β = 1.97, z = 3.97, p < .001), object adults vs. cL1 (β = 1.50, z = 3.50, p < .001), other adults vs. cL1 (β = -3.56, z = -7.04, p < .001), adults vs. cL2 (β = -4.45, z = -6.95, p < .001). Backward anaphora with NSPs, subject adults vs. cL1 (β = 1.90, z = 3.56, p < .001), adults vs. cL2 (β = 1.60, z = 3.34, p < .001), object adults vs. cL1 (β = 1.18, z = 2.11, p < .05) adults vs. cL2 (β = 1.30, z = 2.72, p < .01), other adults vs. cL1 (β = -3.04, z = -3.97, p < .001), adults vs. cL2 (β = -2.39, z = -4.63, p < .001). Backward anaphora with OSPs: subject adults vs. cL1 (β = 4.12, z = 4.84, p < .001), adults vs. cL2 (β = 3.07, z = 5.56, p < .001), other adults vs. cL1 (β = -5.01, z = -6.58, p < .001), adults vs. cL2 (β = -4.93, z = -6.27, p < .001). 108 Figure 31. Between-group pairwise comparisons by experimental condition in three-antecedent contexts Turning to the results of the other four experimental conditions (46c-f), Figure 32 displays the percentages of coreference between the pronoun type and the only intrasentential antecedent (the subject) used in the binomial regression analysis. Because coreference/ disjoint rates were complementary answers in the task, the mean number of times the disjoint reference was selected is simply the reverse of the number reported in Figure 32. See Figure 1 in Appendix A for the mean percentage of missing data (the logically impossible option), for which the cL2 group exhibited higher rates than the cL1 group. Figure 32. Coreference for NSPs and har(e)k ‘that-ERG’ by group and experimental condition 109 We start with the results from each group of participants and then proceed to the analysis between groups. An analysis of the adult data revealed significant main effects of pronoun type (β = 2.87, z = 9.29, p < .001) and anaphora type (β = -0.54, z = -2.42, p < .01), indicating that coreference rates were significantly higher for NSPs than for OSPs; this pattern was more visible in forward than in backward anaphora. The interaction between anaphora type and pronoun was non-significant, suggesting that the interpretations of adults did not differ on the basis of anaphora directionality. Further analyses showed that in the forward anaphora condition (such as (46c)), the embedded NSPs were overwhelmingly coreferential with the subject (91%); this rate differed significantly from that of the embedded OSPs (46d), which were primarily interpreted as referring to the extrasentential antecedent (only 3% coreference) (β = -6.43, z = -7.12, p < .001). The same pattern was observed in the backward anaphora condition: the embedded NSPs took the following lexical subject as the antecedent (74%) in sentences like (46e), whereas the embedded OSPs referred to the extrasentential referent in virtually all cases (only 3% coreference) in structures like (46f) (β = -5.08, z = -6.10, p < .001). The results from the cL1 group revealed a main effect of anaphora (β = - 0.26, z = -2.00, p < .05), with coreference rates generally higher in forward than in backward anaphora. In addition, there was a significant interaction between pronoun type and anaphora type (β = -1.00, z = -3.14, p < .01), with the effect of pronoun type larger for forward than backward anaphora. That is, in the forward anaphora condition, the embedded NSPs were significantly more often interpreted as coreferent with the following lexical subject than the embedded OSPs (78% vs. 65%, respectively; β = -0.69, z = -2.97, p < .001), as displayed in Figure 32. In the backward anaphora condition, coreference was the most frequently preferred option for both types of pronouns, and these rates did not differ significantly between embedded NSPs (58%) and OSPs (64%). For the cL2 group (see Figure 32), there was a main effect of pronoun type (β = -0.28, z = -2.19, p < .05) and a significant interaction between anaphora and pronoun (β = -1.01, z = -2.16, p < .05), suggesting that the effect of pronoun type differed between forward and backward anaphora. While coreference rates did not differ significantly between NSPs and OSPs in forward anaphora sentences, the coreference rates for the following lexical subject were significantly higher for OSPs (72%) than for NSPs (53%; β = 1.07, z = 3.18, p < .01) in backward anaphora sentences. Between-group comparisons of the adults and cL1 revealed main effects of pronoun type (β = 0.63, z = 4.75, p < .001) and group (β = -0.58, z = -4.39, p < .001), as well as a significant interaction between pronoun type and group (β = 5.22, z = 6.24, p < .001), indicating that pronoun type had a larger effect for adults than children. Whereas adults chose the subject as the antecedent of NSPs significantly more often (83%) than cL1 (68%) (β = -0.91, z = -2.33, p 110 < .05; see Figure 33), they selected the subject as the antecedent of OSPs far less often (3%) than cL1 (64%; β = 5.13, z = 6.11, p < .001). Figure 33. Coreference for NSPs and har(e)k ‘that-ERG’ by group A comparative data analysis between the adults and cL2 showed significant main effects of pronoun type (β = 0.72, z = 5.51, p < .001) and group (β = - 0.42, z = -4.43, p < .001). There was also a significant interaction between pronoun type and group (β = 5.87, z = 6.89, p < .001), revealing that the pattern of performance between adults and cL2 varied since pronoun type had a larger pronoun type effect for adults. As displayed in Figure 33, adults chose the subject as the antecedent of NSPs significantly more often (83%) than cL2 (59%) (β = -1.54, z = -3.74, p < .001) and selected the subject as the antecedent of OSPs significantly less often than cL2 (3% vs. 68%) (β = 6.09, z = 5.16, p < .001). The data analysis comparing the child groups (cL1 vs. cL2) revealed a significant interaction between pronoun type and group (β = 0.80, z = 1.98, p < .01), indicating that the effect of pronoun type in the response patterns was greater in cL2 than cL1. As shown in Figure 33, cL2 chose the coreferential reading of NSPs significantly less often (59%) than cL1 (68%) (β = -0.67, z = - 1.97, p < .05), but the results did not differ significantly between cL1 and cL2 with respect to OSPs (64% and 68%, respectively). To summarize the results of Experiment 3, data from the two analyses (multinomial and binomial) revealed that in forward anaphora with NSPs, the preferences of adults and children converged, with both age groups choosing the preceding lexical subject as the antecedent in both threeantecedent (46a) and two-antecedent (46c) contexts, although adults selected this option to a larger extent. In backward anaphora with NSPs, all groups of participants opted for the subject in two-antecedent contexts (46e), but antecedent choices differed in three-antecedent contexts (46g): adults exhibited a preference for the extrasentential referent, while the child groups selected the following lexical subject as the antecedent. With OSPs, the data from the 111 multinomial and binomial tests showed that adults had a clear preference for the extrasentential antecedent in both forward and backward anaphora, whereas both child groups opted for the subject antecedent. 3.5.5 Experiment 4: berak 3.5.5.1 Participants Thirty-four children in the cL1 group (age range 6; 4-7; 3, mean age 6; 8), 21 in the cL2 group (age range 6; 1-6; 8, mean age 6; 3) and 10 L1-Basque adults (age range 17; 6-63; 1, mean age 36; 8) took part in the experiment and were included in the data analysis. Fourteen participants (4 from the cL1 group and 10 from the cL2 group) were excluded from the analysis following the exclusion criteria described in Section 3.5.4.1. 3.5.5.2 Materials The materials were the same as in Experiment 3, except that the OSP harek was substituted for berak ‘(s)he (him/ herself)-ERG’. Examples of items with three possible interpretations are displayed in (47a,b,g,h), and of items with two possible antecedents in (47c-f). (47) FORWARD ANAPHORA a. Subordinate clause [NP] + main clause [..Ø..] Mikel-ek Julen-Ø bultza-tu du-enean ____txistu egin-Ø du. Mikel- ERG Julen- ABS push- PF has-when____whistle do- PF has ‘When Mikel pushed Julen, (he) whistled.’ b. Subordinate clause [NP] + main clause [berak] Mikel-ek Julen-Ø bultza-tu du-enean bera-k txistu egin-Ø Mikel- ERG Julen- ABS push- PF has-when he- ERG whistle do- PF du. has ‘When Mikel pushed Julen, he whistled.’ c. Main clause [NP] + subordinate clause [..Ø..] Mikel-ek txistu egin-Ø du ____Julen-Ø bultza-tu du-enean. Mikel- ERG whistle do- PF has____Julen- ABS push- PF has-when ‘Mikel whistled when (he) pushed Julen.’ 112 d.Main clause [NP] + subordinate clause [berak] Mikel-ek txistu egin-Ø du bera-k Julen-Ø bultza-tu Mikel- ERG whistle do- PF has he- ERG Julen- ABS push- PF du-enean. has-when ‘Mikel whistled when he pushed Julen.’ BACKWARD ANAPHORA e. Subordinate clause [NP] + main clause [..Ø..] ____Julen-Ø bultza-tu du-enean Mikel-ek txistu egin-Ø ____ Julen- ABS push- PF has-when Mikel- ERG whistle do- PF du. has ‘When (he) pushed Julen, Mikel whistled.’ f. Subordinate clause [NP] + main clause [berak] Bera-k Julen-Ø bultza-tu duenean Mikel-ek txistu He- ERG Julen- ABS push- PF has-when Mikel- ERG whistle egin-Ø du. do- PF has ‘When he pushed Julen, Mikel whistled.’ g. Main clause [NP] + subordinate clause [..Ø..] ____Txistu egin-Ø du Mikel-ek Julen-Ø bultza-tu ____whistle do- PF has Mikel- ERG Julen- ABS push- PF du-enean. has-when ‘(He) whistled when Mikel pushed Julen.’ h. Main clause [NP] + subordinate clause [berak] Bera-k txistu egin-Ø du Mikel-ek Julen-Ø bultza-tu He- ERG whistle do- PF has Mikel- ERG Julen- ABS push- PF du-enean. haswhen ‘He whistled when Mikel pushed Julen.’ 3.5.5.3 Coding and data analysis The coding and data analysis procedure was the same as in Experiment 3. 113 3.5.5.4 Results Starting with the results of the multinomial regression analysis of the adult data, there was a significant interaction between anaphora and pronoun (β = 2.86, t = 4.21, p < .001), suggesting that anaphora type had a greater effect for NSPs than for OSPs. Further analyses revealed that in forward anaphora sentences (see Figure 34), the preceding lexical subject was chosen as the antecedent of matrix NSPs significantly more often (85%) than either the object (4%) (β = -3.12, t = -5.29, p < .001) or the extrasentential referent (11%) (β = -2.02, t = -5.70, p < 0.001) in structures like (47a). In contrast, the extrasentential referent was chosen as the antecedent of matrix NSPs significantly more often (83%) than either the subject (13%) (β = 1.79, t = 5.50, p < .001) or the object (4%) (β = -1.29, t = -1.99, p < .05) in backward anaphora sentences such as (47g). In addition, the object was selected significantly less often (4%) than the subject (13%) (β = 3.09, t = 5.24, p < .001). For matrix OSPs, the extrasentential referent was chosen significantly more often (61%) than the subject (18%) in both forward anaphora (β = 3.28, t = 7.02, p < .001) and backward anaphora contexts as in (47b) and (47h) respectively. Figure 34. Adults’ antecedent choices for NSPs and berak in forward and backward anaphora With respect to cL1, the interaction between pronoun and anaphora was non-significant, implying that pronoun type had no effect on anaphora directionality. Further analyses revealed that the preceding lexical subject was the preferred antecedent of matrix NSPs in forward anaphora sentences (56%), beating out the object (28%) (β = -0.72, t = -4.76, p < .001) and the extrasentential referent (16%) (β = -1.24, t = -6.83, p < .001), as seen in Figure 35. In addition, the object was selected significantly more often (28%) than the extrasentential referent (16%) (β = -0.53, t = -2.60, p < 0.01). Coreference with the subject was also the most-frequently chosen interpretation for ma- 114 trix OSPs (44%); the difference between this figure and the selection rate for the extrasentential referent (27%) was significant (β = 0.73, t = 3.06, p < .01). In backward anaphora sentences with matrix NSPs, the subject was again selected more often (50%) than either the object (28%) (β = -0.58, t = -3.82, p < .001) or the extrasentential referent (22%) (β = -0.82, t = -4.95, p < .001). The antecedent choices for OSPs were almost evenly distributed, but the extrasentential referent was preferred as the antecedent (38%) over both the subject (35%) (β = 0.90, t = 0.22, p < .001) and the object (27%) (β = 0.59, t = 2.43, p < .05). Figure 35. cL1 group’s antecedent choices for NSPs and berak in forward and backward anaphora For cL2, the interaction between anaphora and pronoun was non-significant, with the effect of pronoun type not varying between forward and backward anaphora. As Figure 36 illustrates, NSPs in forward anaphora sentences were ambiguously interpreted as coreferent with the subject (48%) or the object (43%); these options were chosen significantly more often than the extrasentential referent (9%) (β = -1.66, t = -4.30, p < .001 and β = -1.56, t = - 4.01, p < .001, respectively). The OSP berak in the forward anaphora condition could either be interpreted as coreferent with the preceding lexical subject (41%) or the object (40%); these rates did not differ significantly from that of the extrasentential referent (19%). In backward anaphora, a similar pattern was found, whereby NSPs were ambiguously interpreted as coreferent with either the following lexical subject (48%) or the object (44%), rates that differed significantly from those corresponding to the extrasentential referent (8%) (β = -1.79, t = -4.39, p < .001 and β = -1.72, t = -4.18, p < .001, respectively). For OSPs in backward anaphora contexts, the following lexical subject was the preferred option (50%), but the object (32%) was chosen significantly more often than the extrasentential referent (18%) (β = 1.16, t = 2.24, p < .05). 115 Figure 36. cL2 group’s antecedent choices for NSPs and berak in forward and backward anaphora Intergroup comparisons demonstrated that the interpretations of adults and cL1 differed significantly in almost all antecedent choices (see Figure 37). There were only two conditions in which the antecedent preferences of adults and cL1 did not differ: in forward anaphora sentences with NSPs, the rates of the extrasentential referent did not differ significantly, and the same was true for selection rates of the object in forward anaphora sentences with OSPs. The comparison between adults and cL2 revealed that responses differed significantly in all cases except for the rates of the extrasentential referent in forward anaphora contexts with NSPs. In comparing cL1 and cL2, the data analysis revealed that the two child groups did not differ significantly in their interpretations, except for the selection rates for the object in forward and backward anaphora sentences with NSPs. See the footnote for the analysis of the pairwise comparisons. 62 62 Forward anaphora with NSPs: subject adults vs. cL1 (β = -1.67, z = 3.10, p < .001), adults vs. cL2 (β = -1.82, z = 4.82, p < .001), object adults vs. cL1 (β = 2.31, z = 3.86, p < .001), adults vs. cL2 (β = 2.97, z = -4.74, p < .001), cL1 vs. cL2 (β = 1.01, z = 3.07, p < .01). Forward anaphora with OSPs: subject adults vs. cL1 (β = 1.4, z = 2.60, p < .01), adults vs. cL2 (β = 1.10, z = 3.06, p < .01), object adults vs. cL2 (β = 1.01, z = 2.50, p < .05), other adults vs. cL1 (β = -2.62, z = -3.01, p < .01), adults vs. cL2 (β = -2.09, z = -4.35, p < .001). Backward anaphora with NSPs: subject adults vs. cL1 (β = 2.06, z = 3.34, p < .001), adults vs. L2 (β = 1.67, z = 3.07, p < .01), object adults vs. cL1 (β = 2.82, z = 3.43, p < .001), adults vs. cL2 (β = 3.53, z = 4.37, p < .001), cL1 vs. cL2 (β = 0.76, z = 2.40, p < .05), other adults vs. cL1 (β = -4.45, z = -4.72, p < .001), adults vs. cL2 (β = -6.64, z = -4.99, p <.001). Backward anaphora with OSPs: subject adults vs. cL1 (β = 2.37, z = 2.89, p < .01), adults vs. cL2 (β = 1.10, z = 3.06, p < 0.01), other adults vs. cL1 (β = -6.53, z = -3.38, p < .001), adults vs. cL2 (β = -6.21, z = -5.03, p < .001). 116 Figure 37. Between-group pairwise comparisons by experimental condition in three-antecedent contexts For the responses to the other four experimental conditions (47c-f), a binomial regression analysis was carried out, as described for Experiment 3. See Figure 2 in Appendix A for the rates of missing data (the logically impossible option). The percentages corresponding to coreference with the subject of the preceding clause (forward anaphora) or the following clause (backward anaphora) are shown in Figure 38. Figure 38. Coreference for NSPs and berak ‘(s)he (him/ herself)-ERG’ by group and experimental condition Analysis of the adult data revealed main effects of pronoun type (β = -2.53, z = -8.12, p < .001) and anaphora type (β = 1.18, z = 4.00, p < .001), indicating that coreference rates were higher for NSPs than for OSPs and that the rates were higher in forward than in backward anaphora. However, there was not a significant interaction between the variables; this suggests that the effect of 117 pronoun type did not vary depending on the anaphora type. As shown in Figure 38, in forward anaphora sentences, adults chose the preceding lexical subject as the antecedent of embedded NSPs (as in (47c)) significantly more often (90%) than for embedded OSPs (47d) (53%) (β = -2.78, z = -5.43, p < .001), a pattern similar to the one observed in the backward anaphora condition (β = -2.66, z = -6.13, p < .001). Analysis of the cL1 data showed main effects of pronoun type (β = 0.56, z = 5.87, p < .001) and anaphora type (β = -0.47, z = -4.94, p < .001), revealing that coreference rates were higher for NSPs than for OSPs and that coreference was preferred more often in forward than backward anaphora. However, no significant interaction between the two variables was observed; this indicates that the effect of pronoun type was similar in forward and backward anaphora. In forward anaphora sentences (see Figure 38), there was a general preference for intrasentential coreference, an effect that was significantly more pronounced for NSPs (83%) than for OSPs (66%) (β = -1.21, z = - 2.06, p < .05), similar to the pattern observed in backward anaphora sentences (β = -0.98, z = -3.77, p < .001). For cL2, there were neither main effects nor significant interactions. Overall, there was a preference for coreference with the subject regardless of pronoun type, as is evident from the non-significant differences found between NSPs and OSPs in both forward and backward anaphora contexts (see Figure 38). The statistical analysis comparing adult and cL1 data revealed main effects of pronoun type (β = 0.79, z = 9.41, p < .001) and anaphora type (β = - 0.49, z = -5.99, p < .001), as well as a significant interaction between pronoun type and group (β = 1.53, z = 3.18, p < .01). NSPs were most often coreferent with the subject in both adults (83%) and cL1 (77%); this difference was not statistically significant. OSPs, in contrast, were generally considered disjoint in reference by adults (41% coreference) and coreferent with the subject by cL1 (60%); rates did not differ significantly between the groups, as shown in Figure 39. 118 Figure 39. Coreference for NSPs and berak ‘(s)he (him/ herself)-ERG’ by group The comparison between adults and cL2 revealed a main effect of pronoun type (β = -1.29, z = -6.66, p < .001) and anaphora type (β = 0.61, z = 3.19, p < .01). In addition, there was a significant interaction between pronoun and group (β = 2.03, z = 3.70, p < .001), indicating a larger effect of pronoun type for adults than children. Adults and cL2 exhibited similar patterns for NSPs, preferring coreference (83% and 77%, respectively, as seen in Figure 39), but differed significantly for OSPs: adults opted for the disjoint reading (41% subject coreference), while the cL2 group preferred the corefential reading (70%) (β = 1.59, z = 2.83, p < .01). In comparing the two groups of children (cL1 vs. cL2), a main effect of pronoun type was found (β = 0.40, z = 1.88, p < .05), with coreference rates higher for NSPs than for OSPs. In addition, a significant interaction was observed between pronoun type and group (β = 0.96, z = 2.87, p < .01), indicating that pronoun type had more effect on cL1 than on cL2, but there were no significant differences in the selection rates of NSPs and OSPs between the groups, as seen in Figure 39. To summarize, the results of Experiment 4 revealed that in forward anaphora contexts, the adults showed a clear preference for the subject as the antecedent of NSPs in both twoand three-antecedent contexts. In backward anaphora with two potential antecedents, the adults again opted for the following lexical subject as the antecedent of NSPs, but in threeantecedent contexts, they chose the extrasentential referent. The overt counterpart, berak, showed more variability in terms of its antecedent among adults: the extrasentential referent was the preferred antecedent out of three possible antecedents, but no preference for any particular referent was observed in contexts with two possible antecedents. The cL1 group showed a general preference for the subject as the antecedent of NSPs and OSPs. These rates were very similar regardless of anaphora directionality in threeantecedent choices; however, cL1 coreference rates for the subject were 119 higher for NSPs than for OSPs in both forward and backward anaphora in two-antecedent contexts. For cL2, NSPs and OSPs could be ambiguously interpreted as coreferent with either the subject or the object out of three potential antecedents, and these children showed a general preference for the subject in two-antecedent contexts. 3.5.6 Discussion 3.5.6.1 Pronoun type in forward anaphora Experiments 3 and 4 revealed that the adults exhibited a dichotomous or binary pattern of response depending on the pronoun type (null vs. har(e)k/ null vs. berak). In general, the antecedent of NSPs was the subject (except for one particular case; see Section 3.5.6.2), whilst that of OSPs was the extrasentential referent. The least preferred option for adults was the object, with a mean percentage of 8% across the two experiments. In contrast, the children’s responses were more evenly distributed among the three options for both NSPs and OSPs. In Experiment 3, the selection rates for the subject were around 50%, and those for the object and the extrasentential were around 20%. In Experiment 4, the selection rates of the extrasentential and especially the object antecedent were higher in the child groups, while those of the favoured subject option dropped slightly. We first discuss the results from forward anaphora contexts; those from backward anaphora will be addressed in Section 3.5.6.2. Adult results from Experiments 3 and 4 showed that in the forward anaphora condition, NSPs were usually coreferent with the preceding lexical subject in three-antecedent contexts ((46a) and (47a)) (76%/ 85%) and to an even greater degree in two-antecedent contexts ((46c) and (47c)) (91%/ 90%). With regard to the interpretation of OSPs, Basque adults predominantly chose the extrasentential referent as the antecedent of har(e)k ‘that-ERG’ in three-antecedent contexts (46b) (90%) and to a greater degree in twoantecedent contexts (46d) (only 3% subject coreference). The interpretation of berak ‘(s)he (him/ herself)-ERG’ varied depending on the number of potential antecedents. Whereas adults interpreted berak as disjoint in reference (61%) in contexts with three possible antecedents (47b), they showed no clear pattern in two-antecedent contexts (47d), where the responses were almost equally distributed between the subject (53%) and the extrasentential referent (47%). Thus, the number of possible interpretations had more effect on the antecedent bias of berak than that of har(e)k. The preceding lexical subject was also favoured as the antecedent of NSPs by both child groups in Experiments 3 and 4 in three-antecedent contexts (57%/ 56% for cL1; 47%/ 48% for cL2), but to a significantly lower degree than adults. The rates for the subject were even higher in twoantecedent contexts in these experiments (78%/ 83% for cL1; 65%/ 70% cL2). For the OSP har(e)k, the preceding subject was again the preferred option for 120 both child groups out of both three (52% for cL1; 64% for cL2) and two potential antecedents (65% for cL1; 64% for cL2). Similar to the pattern for har(e)k, the subject was the cL1 group’s favoured antecedent for the OSP berak in three-antecedent contexts (44%) and to an even greater degree in two-antecedent contexts (66%). For the cL2 group, preferences in the antecedent choice for berak were equally distributed between the preceding lexical subject (41%) and the object (40%), and this group showed a preference for the preceding lexical subject in two-antecedent contexts (73%). The adult data from Experiments 3 and 4 on referential preferences for NSPs and OSPs in two-referent sentences revealed that, in line with the PAS (Carminati 2002), asymmetric results were found between NSPs and OSPs - the former coreferring with the subject and the latter with the extrasentential (only visually present) referent. However, these distinct functions were more evident with the demonstrative har(e)k than with the pronoun berak. This difference in the biases of NSPs and OSPs was not evident in child data, since these groups prioritized the intrasentential subject antecedent rather than resorting to an extralinguistic referent. 3.5.6.2 Clause order and anaphora type In contrast to Experiments 1 and 2, the effect of clause order (RQ1a) in tworeferent sentences (Experiments 3 and 4) is not directly applicable, since one level of the variable clause order matched a specific level of anaphora type, resulting in different types of items depending on the possible interpretations. The results of Experiments 3 and 4 in forward anaphora contexts discussed in the previous section revealed that the referential properties of NSPs and the OSP har(e)k did not change in any group of participants regardless of whether the pronouns were in the main (in three-antecedent contexts) or the subordinate clause (two-antecedent contexts)(Compare Figures 31 and 32 in Experiment 3.) However, the adults’ preferences for berak did change depending on the clause order: the extrasentential referent was the antecedent of the matrix berak (see Figure 37), whereas the embedded berak did not exhibit a clear preference for an antecedent (see Figure 38). See 4.1.4 for more details on this point. In backward anaphora contexts, as discussed in the results for onereferent sentences (see Section 3.5.3.2), clause order had an effect on a particular context in which the usual bias of NSPs for the preceding or following lexical subject (as predicted in the PAS) was not expected to occur. More precisely, in backward anaphora contexts with main-subordinate order in three-antecedent contexts ((46g) and (47g)), coreference between an Rexpression and a c-commanding pronoun antecedent is blocked by a syntactic constraint (RQ1b) (Principle C; see Section 1.3). The prediction of noncoreference in such a context is confirmed by the adults’ general preference for the disjoint interpretation of both NSPs (74% in Experiment 3 and 83% in Experiment 4) and OSPs (95% in Experiment 3 and 90% in Experiment 4). In 121 contrast, the child groups did not respect this restriction in that context, opting as usual for the subject as the antecedent of both NSPs (50%/ 47% for cL1 and cL2, respectively, in Experiment 3; 50%/ 48% in Experiment 4) and OSPs (59%/ 53% for cL1 and cL2, respectively, in Experiment 3; 35%/ 50% in Experiment 4). The target-like interpretation was only chosen <22%-26%> of the time by cL1 and <8%-38%> by cL2 in Experiments 3 and 4. In backward anaphora contexts, where embedded (null/ overt) pronouns can syntactically corefer with the following lexical subject, adults opted for coreference with the subject for NSPs ((46e) and (47e)) (74% in Experiment 3 and 75% in Experiment 4) and for the extrasentential antecedent for OSPs ((46f), (47f)) (only 3% subject coreference in Experiment 3 and 26% in Experiment 4), a pattern similar to the one shown in the forward anaphora condition. The cL1 and cL2 groups, in contrast, exhibited a preference for the following lexical subject for both NSPs (58%/ 71%, respectively) and OSPs (53%/ 54%). 122 3.6 Acceptability Judgement Task Like the Picture Selection Task (Experiments 1-4), the Acceptability Judgement Task (Experiments 5 and 6) deals with the interpretation of the OSP hura ‘that’ or bera ‘(s)he (him/ herself)’ in comparison to NSPs. Here, the same linguistic phenomenon was analysed using a different approach. In the Picture Selection Task, participants’ referential preferences in sentences containing OSPs and NSPs were analysed, whereas in Experiments 5 and 6, participants were given the opportunity to express the acceptability status of NSPs vs. OSPs in different discourse contexts: topic-continuity (coreference) contexts vs. topic-shift (non-coreference) contexts. Experiments 5 and 6 allow us to determine whether NSPs and OSPs show different biases in terms of their antecedents (RQ1), and if so, whether the discourse context has an effect on this bias (RQ1d), as crosslinguistic studies have suggested (RQ2). In addition, the existence of developmental differences can be addressed by comparing the results of adults and children, and the effect of the relationship between non-linguistic factors (like age of onset of acquisition and the amount of exposure to the language) can be investigated using data from cL1 and cL2 (RQ3). Experiments 5 and 6 provide answers to the questions of a) whether the bidirectional targetdeviant patterns observed for both NSPs and OSPs in previous studies (Barquin & Costa 2011 for Spanish, Sorace et al. 2009 for Italian, using the same methodology and materials as in the present study) discussed in Section 2.2.2.1 can be extended to bilinguals of two null-subject languages like Basque and Spanish, and b) whether the overt pronoun becomes the default strategy when bilinguals fail to integrate the correct syntax-discourse mapping, as observed in the literature on bilingual performance. 3.6.1 Experiment 5 63 : hura 3.6.1.1 Participants Twenty children in the cL1 group (age range 6; 3-7; 2, mean 6; 6), 28 children in the cL2 group (age range 6; 2-8; 0, mean 7; 9) and a control group of 14 L1- Basque adults (age range 18; 0-20; 0 mean 19; 0) participated in the experiment. Six participants (1 from cL1 and 5 from cL2 groups) were excluded from the analysis because they did not meet the 50% threshold for correct responses on the filler items. The adult controls were recruited among university students at the University of the Basque Country (Vitoria-Gasteiz, Spain). 63 The results in this section and Section 3.6.2 stem from joint work with Mikel Santesteban and Maria José Ezeizabarrena (Iraola, Santesteban & Ezeizabarrena 2014). 123 3.6.1.2 Materials and design The experiment was based on a within-subject variable, context (topic-shift vs. topic-continuity), and a between-subject variable, group, with three levels: adults, cL1 and cL2. The task included 36 linguistic items in total: 16 experimental items for two conditions (8 items per condition) plus 10 filler items. The fillers had the same structure as the experimental items, but one of the sentence pairs uttered by one of the characters was ungrammatical (e.g. incorrect subject-verb agreement, incorrect verb-auxiliary selection, etc.). The aim of these items was to ensure that participants had understood the task and were focused on the structure of the sentences that were being presented to them. The materials and the methodology used by Sorace et al. (2009) for Italian were adapted to Basque. Each item consisted of a short video clip with four characters (Mickey Mouse, Minnie Mouse, Donald Duck and Daisy). One of the characters in the background uttered the experimental item containing an NSP ((48a), (49a)), and another character repeated it with the OSP (hura ‘that’) ((48b), (49b)). In the experimental items, one character performed a one-referent action consisting of an unaccusative or unergative verb; this was commented upon either by the character himself ([-topic shift] [-TS] context) or by a second character who witnessed the action but was not involved in it ([+topic shift] [+TS]) condition). The focus then shifted to a pair of characters in the background who reported one after the other what had happened, starting with a main clause such as “X said that…” and continuing with a clause containing either a NSP or a OSP. Note that although the translation of the complex sentence in English is formed by a main clause followed by a subordinate clause, in Basque the subordinate clause (VPcomplement) is in the preverbal 64 position of the main clause, since the subordinate clause follows the subject of the main clause. Thus, the experimental sentences in Basque show a subordinate-main clause order. Two example experimental items with an unaccusative verb are given in ((48)) and (49)). (48) Topic-continuity context [-TS] (Minnie and Daisy in the foreground; Mickey and Donald in the background) Minnie falls and says: Erori egin-Ø naiz! Fall do- PF AUX ‘I’ve fallen! ’ 64 Finite complement clauses may appear in either preverbal or postverbal position. The preverbal position is more frequent when the finite complement is focalized (Artiagoitia 2003: 651-652). 124 a. Donald: Minnie-k i ___ i erori-Ø de-la esan-Ø du. Minnie- ERG ___ fall- PF AUX - COMP say- PF AUX ‘Minnie i has said that ___ i has fallen.’ b. Mickey: Minnie-k i hura-Ø i erori-Ø de-la esan-Ø du. Minnie- ERG she- ABS fall- PF AUX - COMP say- PF AUX ‘Minnie i has said that she i has fallen.’ (49) Topic-sfift context [+TS] (Minnie and Daisy in the foreground; Mickey and Donald in the background) Daisy falls and Minnie says: Daisy-Ø erori egin-Ø da! Daisy- ABS fall do- PF AUX ‘I’ve fallen! ’ a. Donald: Minnie-k i ___ j erori-Ø de-la esan-Ø du. Minnie- ERG ___ fall- PF AUX - COMP say- PF AUX ‘Minnie i has said that ___ j has fallen.’ b. Mickey: Minnie-k i hura-Ø j erori-Ø de-la esan-Ø du. Minnie- ERG she- ABS fall- PF AUX - COMP say- PF AUX ‘Minnie i has said that she j has fallen.’ As the subject of an unaccusative verb, the demonstrative hura in the embedded clause in (48b) and (49b) takes (zero-marked) absolutive case, but other experimental items contained an unergative verb, in which case hura was marked in the ergative case (with the suffix -k, har(e)k ‘that-ERG’). In Basque, because there is no morphological gender, the demonstrative hura could ambiguously refer to either of the two foreground characters (these were always of the same gender: e.g. Minnie and Daisy or Mickey and Donald). 3.6.1.3 Procedure The materials (embedded in a PowerPoint presentation) were presented on a laptop, and participants’ responses were recorded on an answer sheet by the interviewer. Participants were told that the characters were learning Basque and were then instructed to decide which one of the characters in the background spoke “better” Basque. The experimental session was preceded by a short training session of two trials that was repeated until participants understood the task. The experimental session then started. Trials were pseudorandomised for each participant so that no more than three experimental trials and no more than two trials of the same experimental condition were presented in a row. 3.6.1.4 Coding and data analysis Similar to the Picture Selection Task, a mixed-effects binomial regression model was carried out. The model consisted of the dependent variable pro- 125 noun type (participants’ preferences for NSPs vs. OSPs), the variable context ([-TS] vs. [+TS]) as a fixed effect and subjects and items as random effects. Responses for which participants chose a character who had not reported on the action were coded as missing data (e.g. participants chose Mickey when the characters reporting on the action had been Daisy or Minnie; a total of 8 responses over both experiments (Experiments 5 and 6)). 3.6.1.5 Results Figure 40 reports the mean percentage of the OSP hura preference in [-TS] and [+TS] contexts. The mean number of times NSPs were selected in each discourse context is the opposite of what is displayed in Figure 40. Figure 40. Overt pronoun hura preference in [-TS] and [+TS] contexts by group A significant main effect of context was observed in the three groups of participants (adults, β = -4.31, z = -9.06, p < .001; cL1, β = -0.78, z = -2.75, p < .01; cL2, β = 0.85, z = -1.96, p = .05), with participants preferring the use of hura more in [+TS] than in [-TS] contexts. A comparison between adults and cL1 showed main effects of context (β= -4.01, z= -9.82, p < .001) and group (β = -2.66, z = -5.72, p < .001), indicating that OSPs were generally chosen less often in [-TS] than in [-TS] contexts, and that OSP choices were less common in the cL1 group than in adults. The significant interaction between context and group (β = 3.87, z = 7.21, p < .001) revealed the existence of different patterns of OSP preferences among adults and cL1 in [-TS] and [+TS] contexts. More precisely, in [-TS] contexts, adults showed a weaker preference for OSPs than the cL1 group (10% vs. 27%; β = 2.33, z = 2.32, p < .05), but in [+TS] contexts, the effect was the reverse, with adults exhibiting a greater preference for OSPs than the cL1 group (89% vs. 41%; β = -4.20, z = -4.33, p < .001). 126 The comparison between adults and cL2 showed main effects of context (β = -4.76, z = -8.98, p < .001) and group (β = -3.02, z = -5.81, p < .001), demonstrating that participants in general chose fewer OSPs in the [-TS] than in the [+TS] condition, and that the effect of context was greater for adults than for children. In addition, there was a significant interaction between context and group (β = 3.96, z = 6.27, p < .001), indicating a pattern of results similar to those reported for the comparison between adults and cL1. In [-TS] contexts, adults showed a marginally significant lower preference for OSPs than cL2 (10% vs. 22%; β = 1.90, z = 1.93, p = .05), but in [+TS] contexts, the opposite was true, with adults exhibiting a greater preference for OSPs than cL2 (89% vs. 36%; β = -4.36, z = -4.82, p < .001). Finally, the comparison between cL1 and cL2 revealed only a significant main effect of context (β = -0.80, z = -2.12, p < .05), suggesting that children in general accepted fewer OSPs in [-TS] contexts (around 25%) than in [+TS] contexts (around 39%). The interaction between context and group was nonsignificant, revealing the existence of similar patterns of responses across the two children’s groups. 3.6.2 Experiment 6: bera 3.6.2.1 Participants Twenty-three children in the cL1 group (age range 6; 1-7; 0, mean 6; 4), 25 in the cL2 group (age range 5; 11-7; 0, mean 6; 5) and a control group of 10 L1- Basque adults (age range 17; 2-57; 5, mean 28; 3) were included in the data analysis. Three participants from the cL2 group were excluded following the exclusion criteria described in Section 3.6.1.1. 3.6.2.2 Materials and Design The materials were the same as those in Experiment 5, but except that the OSP hura was substituted for the OSP bera ‘(s)he (him/ herself)’. (50) Topic-continuity context [-TS] (Minnie and Daisy in the foreground; Mickey and Donald in the background) Minnie falls and says: Erori egin-Ø naiz! Fall do- PF AUX ‘I’ve fallen! ’ a. Donald: Minnie-k i ___ i erori-Ø de-la esan-Ø du. Minnie- ERG ___ fall- PF AUX - COMP say- PF AUX ‘Minnie i has said that ___ i has fallen.’ b. Mickey: Minnie-k i bera-Ø i erori-Ø de-la esan-Ø du. Minnie- ERG she- ABS fall- PF AUX - COMP say- PF AUX ‘Minnie i has said that she i has fallen.’ 127 (51) Topic-sfift context [+TS] (Minnie and Daisy in the foreground; Mickey and Donald in the background) Daisy falls and Minnie says: Daisy-Ø erori egin-Ø da! Daisy- ABS fall do- PF AUX ‘I’ve fallen! ’ a. Donald: Minnie-k i ___ j erori-Ø de-la esan-Ø du. Minnie- ERG ___ fall- PF AUX - COMP say- PF AUX ‘Minnie i has said that ___ j has fallen.’ b. Mickey: Minnie-k i bera-Ø j erori-Ø de-la esan-Ø du. Minnie- ERG she- ABS fall- PF AUX - COMP say- PF AUX ‘Minnie i has said that she j has fallen.’ 3.6.2.3 Results Figure 41 reports the mean percentage of the preference for OSP bera ‘(s)he (him/ herself) in [+TS] conditions. Figure 41. Overt pronoun bera preference in [-TS] and [+TS] contexts by group There was a main effect of context in adults (β = -2.85, z = 0.41, p < .001), implying that OSPs were chosen significantly less often in [-TS] contexts (16%) than in [+TS] contexts (74%). In contrast, the variable context had no effect for cL1 and cL2, with participants not showing any clear preference for NSPs or OSPs on the basis of discourse context. The comparison of the results of adults and cL1 revealed main effects of context (β = -2.99, z= -6.11, p < .001) and group (β = -1.95, z = -4.73, p < .001), indicating that participants were less likely to select OSPs in [-TS] than in [+TS] contexts, and that cL1 children had a less of a preference for OSPs than 128 adults. The interaction between context and group was significant (β = 3.44, z = 7.16, p < .001), indicating that the patterns of results varied between adults and cL1. More precisely, cL1 chose significantly more OSPs than adults in [-TS] contexts (42% vs. 16%, respectively; β = 1.64, z = 3.14, p < .01) and significantly fewer OSPs in [+TS] contexts (33% vs. 74%, respectively; β = -2.08, z = -4.29, p < .001). The comparison between adults and cL2 revealed main effects of context (β = -2.78, z = -6.83, p < .001) and group (β = -1.16, z = -3.25, p < .01), suggesting that participants had an overall weaker preference for OSPs in [-TS] than in [+TS] conditions, and that the cL2 group showed less acceptance of OSPs than adults. The significant interaction between context and group (β = 3.10, z = 6.81, p < .001) indicated that the cL2 group, like cL1, chose significantly more OSPs than adults in [-TS] contexts (54% vs. 16%, respectively; β = 1.94, z = 4.96, p < .001) and selected significantly fewer OSPs in [+TS] contexts (46% vs. 74%, respectively; β = -1.14, z = -3.38, p < .001). With regard to cL1 and cL2, there was only a main effect of group (β = 0.67, z = 2.43, p < .05), indicating that cL2 chose more OSPs than cL1 in general. There was neither a significant main effect of context nor a significant interaction between context and group (β = 0.44, z = 1.66, p = .09 and β = - 0.10, z = -0.34, p = .73, respectively), suggesting that context did not play a role in the OSP choices of cL1 and cL2 groups. 3.6.3 Discussion The performance patterns of adults in Experiments 5 and 6 showed that context had a clear effect on their OSP choices: adult participants selected fewer OSPs (especially hura) in [-TS] contexts than in [+TS] contexts. For cL1 and cL2 groups, context only affected their preferences for pronoun resolution in Experiment 5, where more hura choices were made in the [+TS] condition than in the [-TS] condition. Note, however, that both children’s groups generally preferred NSPs more than 60% of the time regardless of context. In Experiment 6, neither child group showed a clear preference for a particular pronoun type in the different discourse contexts. This is particularly evident in the cL2 group, where the preferences were at chance level. The cL1 group, in contrast, again showed a general preference for NSPs (58% in [-TS] contexts and 67% in [+TS] contexts); the difference in preference between the two pronouns was not significant. Intergroup comparisons between the adults and child groups in Experiments 5 and 6 indicated significant differences in both discourse contexts: children selected more OSPs than adults in the [-TS] condition, and the opposite pattern was observed in the [+TS] condition. With regard to the patterns of responses between the two children’s groups, no significant differences were observed between cL1 and cL2 in Experiments 5 and 6. The bidirectional pragmatic deviance consisting of the misinterpretation of both 129 NSPs and OSPs observed in child Basque is consistent with findings from child Italian and Spanish (Barquin & Costa 2011, Shin & Cairns 2012, Sorace et al. 2009). This topic will be discussed further in Section 4.2.4. 131 4 Reference of null and overt subject pronouns in Basque Based on the interpretations of NSPs and OSPs in complex sentences by Basque adults and children, the present study sought to answer three questions: a) whether NSPs and OSPs would differ in their antecedent choices in Basque, as predicted by the PAS for Italian (Carminati 2002; see Section 1.4), b) whether the experimental data gathered would be compatible with the descriptions provided by Basque grammarians for the two third-person OSPs hura and bera (see Section 3.2.1) and c) to what extent the interpretations of children in the age range of 6 to 8 would resemble those of native adults, and whether there would be differences between cL1 and cL2. On the basis of a Picture Selection Task (Experiments 1-4), the referential properties of the OSPs hura ‘that’ and bera ‘(s)he (him/ herself)’ and NSPs were tested in one-referent and two-referent sentences. Two variables were taken into account: a) the order of the complex sentences (the subordinate clause following or preceding the main clause), and b) the direction of anaphora (forward vs. backward). Additionally, a second methodology — namely an Acceptability Judgement Task (Experiments 5 and 6) — was used to test the discourse properties [+TS] of the OSPs hura and bera and NSPs in the forward anaphora condition. The structure of this chapter is as follows. Sections 4.1 and 4.2 present adult and child interpretations of pronouns in a parallel distribution of subsections. Adult data for NSPs vs. OSPs (hura/ bera) in forward anaphora contexts are reported in Sections 4.1.1 and 4.1.2, and child data in Sections 4.2.1 and 4.2.2. The response patterns of adults and children in one-referent sentences (Experiments 1-2 and Experiments 5-6) are presented in Sections 4.1.1.1 and 4.1.2.1 and those of two-referent sentences (Experiments 3 and 4) in Sections 4.1.1.2 and 4.1.2.2 . The effects of clause order and anaphora directionality are then discussed in Sections 4.1.3 and 4.2.3. The properties of Basque pronouns in adult language in comparison to other (mostly Romance) null-subject languages are commented upon in Section 4.1.4. Finally, the results of children are compared to those of adults and to prior studies on the acquisition of pronouns in Section 4.2.4. 132 4.1 Adults’ interpretation of pronouns 4.1.1 NSP vs. hura in forward anaphora 4.1.1.1 One-referent sentences (Experiments 1 and 5) Experiment 1 tested the effect of a pronoun’s realisation (null/ demonstrative hura) on its referential properties: coreferent with the lexical (proper noun) subject of the immediate (preceding or following) clause, or disjoint in reference from the lexical subject (i.e. the extrasentential referent as antecedent). In bi-clausal forward anaphora sentences with NSPs — as in (44a) Mikel igo denean___ haserretu egin da ‘When Mikel got on (the bus), (he) got angry’ and (44c) Mikel haserretu egin da___igo denean ‘Mikel got angry when (he) got on (the bus)’ — adults overwhelmingly selected the picture depicting the same character doing both actions (93%). This choice was interpreted as a coreferential reading between the preceding lexical subject (Mikel) and the matrix (44a) or embedded NSP (44c), respectively. Because the mean rate includes the results from both subordinate-main and main-subordinate clause orders, this indicates a very strong preference for the coreferential interpretation of NSPs, regardless of the clause type in which the NSP appears. In contrast, adults clearly rejected intrasentential coreference with the subject for the OSP hura ‘that-ABS’ (only 6% coreference), regardless of whether the OSP appeared in the main or the subordinate clause. The noncoreferential (disjoint) interpretation of the matrix OSP hura in sentences like (44b) Mikel i igo denean hura j haserretu egin da ‘When Mikel got on (the bus), he got angry’ is consistent with Rule I* (Guasti & Chierchia 1999/ 2000), as explained in Section 3.5.3.1. In such a context, the pronoun hura is not (semantically) bound, and it is therefore generally interpreted as referring to an extrasentential referent. On the other hand, Principle C as typically applied to demonstratives (see Section 1.3.1) prohibits hura in the embedded clause in (44d) Mikel i haserretu egin da hura j igo denean ‘Mikel i got angry when he j got on (the bus)’ from being bound by the c-commanding lexical argument Mikel. The divergent methodology used in Experiment 5, namely the Acceptability Judgement Task, confirmed the differences in the biases of NSPs and the OSP hura in Experiment 1 from a new perspective, as discussed in Section 3.6.2. The effect of (topic-shift [+TS] and topic-continuity [-TS]) discourse contexts on the acceptability of NSPs and the OSP hura was tested in indirect speech contexts such as (48a,b) Minniek___/ hura erori dela esan du ‘Minnie has said that ___/ she has fallen’. The experimental setting forced the participants to choose the best option out of two sentences containing either a NSP (48a) or an OSP (48b). In [-TS] contexts in which the character was commenting on the action and falling at the same time, adults overwhelmingly preferred NSPs (90%). The opposite pattern was found in [+TS] 133 contexts in which the act of reporting the action and the act of falling were assigned to different characters. Under these conditions, the OSP hura was predominantly chosen (89%). Adults’ preference for hura in [+TS] contexts as signalling switch reference — namely, having a character as the antecedent who differs from the preceding lexical subject — is in line with Sorace’s claim that overt pronouns are specified for [+topic shift] (see Section 1.1.4). Likewise, as Haegeman (1994) points out for null-subject languages, “subject pronouns will only be present when the added effort of overtly expressing them has some yield” (Haegeman 1994: 217). This principle of economy, also visible in the Avoid Pronoun Principle (Chomsky 1981), links the OSP to the disjoint reading, as this is the marked option. In addition, adults’ preference for the extrasentential referent as the antecedent of hura in linguistic items such as (49b) Minniek hura erori dela esan du ‘Minnie has said that she has fallen’ is consistent with the impossibility of a coreferential reading between the embedded demonstrative hura and the preceding lexical subject (Principle C). 4.1.1.2 Two-referent sentences (Experiment 3) Using basically the same methodology as Experiment 1, Experiment 3 included a modification of the linguistic stimuli with respect to the types of predicates tested. Instead of complex sentences made up of two one-referent clauses (Experiment 1), Experiment 3 employed complex sentences composed of two (morphologically) transitive clauses, one of which contained two arguments (the subject and the object), whereas the other had only a single argument (the subject). As in Experiment 1, variables such as clause order and anaphora type were taken into account in the process of designing the materials. However, the linguistic items were split into two types depending on the combination of the levels of each variable; this led to differences in the number of potential antecedents. In the condition of forward anaphora with subordinate-main clause order and backward anaphora with main-subordinate clause order, there were three possible antecedents for NSPs and the demonstrative OSP ‘har(e)k-ERG’: the subject, the object and the (visually present) extrasentential referent (three-antecedent contexts). However, there were only two possible antecedents (the subject and the extrasentential referent) in forward anaphora with main-subordinate clause order and backward anaphora with subordinate-main clause order (twoantecedent contexts). The aim of Experiment 3 was to ascertain whether the increased number of plausible antecedents in the linguistic context (from one to two) would alter the biases of NSPs and OSPs (RQ1c). At the same time, it sought to determine whether the different antecedent preferences of NSPs and OSPs would be regulated in terms of a primarily syntactic notion of prominence, as predicted by the PAS. Specifically, the experiment investigated whether the OSP would corefer with an argument occupying a position lower than the Spec IP in the preceding clause. 134 Adults generally interpreted NSPs as coreferent with the preceding lexical subject in three-antecedent contexts (76%), 65 as in (46a) Mikelek Julen bultzatu duenean___txistu egin du ‘When Mikel pushed Julen, (he) whistled’. Preferences for the preceding lexical subject were even higher in twoantecedent contexts (91%) such as (46c) Mikelek txistu egin du___Julen bultzatu duenean ‘Mikel whistled when (he) pushed Julen’. However, the antecedent choices for the OSP har(e)k ‘that-ERG’ revealed a different picture. Adults preferred the extrasentential referent as the antecedent for har(e)k out of three potential antecedents (90%), 66 as in (46b) Mikelek Julen bultzatu duenean ha(r)ek txistu egin du ‘When Mikel pushed Julen, he whistled’, and even to a higher degree out of two possible antecedents (97%), as in (46d) Mikelek txistu egin du ha(r)ek Julen bultzatu duenean ‘Mikel whistled when he pushed Julen’. Thus, the results of two-referent sentences (Experiment 3) provide additional evidence of the non-intrasentential coreferential properties of the demonstrative hura in adult Basque grammar, supporting the findings from one-referent sentences (Experiment 1). The adults’ preference for the nonlinguistically present referent in the case of the OSP h(a)rek seems to be induced by the rejection of any intrasentential (subject or object) antecedent, rather than the unavailability of coreference with a subject antecedent, as the data from one-referent sentences (Experiment 1) suggested. In other words, the results from three-antecedent contexts reveal that adults reject coreference with the syntactically most prominent antecedent, the subject (only 10% acceptance), as well as with a syntactically lower referent such as the object (only 5% acceptance). Thus, the PAS does not seem to play a role in resolving the interpretation of the demonstrative har(e)k, as the pattern observed was similar to that found in one-referent sentences (Experiment 1) . 4.1.2 NSP vs. bera in forward anaphora 4.1.2.1 One-referent sentences (Experiments 2 and 6) Using the same methodology (a Picture Selection Task) as in Experiment 1, Experiment 2 tested the interpretation of NSPs in comparison to the OSP bera ‘(s)he (him/ herself)-ABS’. In forward anaphora sentences with NSPs — as in (45a) Mikel igo denean___ haserretu egin da ‘When Mikel got on (the bus), (he) got angry’ and (45c) Mikel haserretu egin da___igo denean ‘Mikel got angry when (he) got on (the bus)’ — coreference with the lexical subject of the previous clause (Mikel) was the preferred option for adults (91%). Thus, the type of clause in which the NSP occurred did not affect the results. In contrast, the extrasentential antecedent was preferred for the OSP bera (only 32% coreference with the subject), regardless of its position in the 65 The mean rate for the object was 14%; for the extrasentential referent, 10%. 66 The mean rate for the object was 5%; for the extrasentential referent, 5%. 135 complex sentence (subordinate or main clause). The non-coreferential reading of bera in sentences such as (45b) Mikel igo denean bera haserretu egin da ‘When Mikel got on (the bus), he got angry’or (45d) Mikel haserretu egin da bera igo denean ‘Mikel got angry when he got on’ can be understood on the basis of Accessibility Theory (Ariel 1990). According to this theory, the form of a linguistic expression used to refer to a discourse entity reflects the entity’s degree of salience, as explained in Section 1.2. Here, the assumption is that in contexts of unity, the closer the relationship between the clauses in which the referring expression and the antecedent occur, the more likely it is that maximally salient antecedents like subjects will be referred to by minimally complex anaphora (i.e. NSPs). Less salient antecedents (in our case, the extrasentential referent) are referred to by morphologically more complex forms, OSPs. The same idea has also been captured by the Reversing- Mapping Hypothesis (Bittner, Kuehnast & Gagarina 2008) and the Semantic Asymmetry formulated by Cardinaletti and Starke (1999a: 154), under which deficient personal pronouns like null pronouns (as opposed to strong pronouns) must have an antecedent prominent in the discourse (see Section 1.1.4) The divergent patterns of results between NSPs and the OSP bera found in Experiment 2 were replicated in Experiment 6 using an Acceptability Judgement Task. In [-TS] contexts, adults selected NSPs (84%), whereas they preferred bera in [+TS] contexts (74%). These results indicate that the coreferential reading (structurally possible for non-reflexive pronouns under BT) in items such as (51b) Minniek bera erori dela esan du ‘Minnie has said that she has fallen’ was favoured by adults in only 36% of cases. Thus, the data do not support the observation of Basque scholars that bera functions as a logophoric pronoun that is coreferential with the subject of the reporting clause — that is, a pronoun that refers back to the speaker of the quoted utterance (Abaitua 1994, Garzia 1996; see Section 3.2.1) in the case at hand. Note, however, that these authors contrast the use of bera to hura in reported speech contexts, whereas the experimental sentences in Experiment 6 compare bera to NSPs. 4.1.2.2 Two-referent sentences (Experiment 4) Similar to Experiment 3, Experiment 4 analysed the interpretation of NSPs in comparison to that of the OSP berak ‘(s)he (him/ herself)-ERG’ in a Picture Selection Task. Adults generally interpreted NSPs as coreferent with the preceding lexical subject in three-antecedent contexts (85%), 67 as in (47a) Mikelek Julen bultzatu duenean___txistu egin du ‘When Mikel pushed Julen, (he) whistled’. Coreference with the subject reached higher levels in twoantecedent contexts (90%), as in (47c) Mikelek txistu egin du___Julen bultzatu duenean ‘Mikel whistled when (he) pushed Julen’. 67 The mean rate for the object was 4%; for the extrasentential referent, 11%. 136 On the other hand, berak was interpreted as referring to the extrasentential referent (61%) 68 in three-antecedent contexts such as (47b) Mikelek Julen bultzatu duenean berak txistu egin du ‘When Mikel pushed Julen, he whistled’. In two-antecedent contexts such as (47d) Mikelek txistu egin du berak Julen bultzatu duenean ‘Mikel whistled, when he pushed Julen’, the adults’ preferences were at chance level, with selections equally divided between the subject (53%) and the extrasentential referent (47%). See Section 4.1.4 for possible explanations for these patterns. 4.1.3 Clause order and anaphora directionality Half of the linguistic items in the Picture Selection Task (Experiments 1-4) had subordinate-main clause order, and the other half had main-subordinate order. The same distribution was also ensured for forward and backward anaphora sentences. This allowed us to test the possible effects of these variables in the response patterns. The hypothesis (RQ1a) proposed in Section 3.3 predicted that clause order (main-subordinate vs. subordinate-main) would not affect participants’ preferences for antecedent selection in forward anaphora contexts, based on Carminati’s results (2002). In addition, the variable anaphora type (forward vs. backward) was not expected to have an effect on the responses except in one certain context — namely, backward anaphora sentences with main-subordinate clause order (see RQ1b in Section 3.3). The results of the forward anaphora condition in one-referent sentences discussed in Sections 4.1.1 and 4.1.2 indicated no effect of clause order in the response patterns of adults. Interestingly, the bias of NSPs for the subject and the bias of OSPs for the extrasentential referent remained the same regardless of whether the pronouns occurred in adjunct clauses (Experiments 1 and 2) or VP-complement clauses (Experiments 5 and 6). This suggests that the syntactic distinction (adjunct-argument) did not affect the biases of the pronouns. A direct analysis of the effect of clause order in two-referent forward anaphora contexts (Experiments 3 and 4) was not possible, since a specific level of the variable clause order was linked to a particular level of the variable anaphora type. However, a comparison of the results from the different types of items (three or two potential antecedents) indicates that antecedent choices remained constant for the OSP ha(r)ek despite changes in clause order (see Figures 31 and 32). More precisely, in the forward anaphora condition with subordinate-main clause order, the rates for the extrasentential referent for har(e)k prevailed (90%) in three-antecedent contexts. In the twoantecedent condition, slightly higher preferences for the sentence-external antecedent were observed in the main-subordinate clause order (98%). In the 68 The mean rate for the subject was 18%; for the object, 21%. 137 case of berak, clause order and also the number of potential antecedents available affected the pronoun’s interpretation. The matrix berak (subordinate-main clause order) was interpreted as referring to the extrasentential referent (61%), whereas the embedded berak was ambiguously interpreted as either coreferent with the subject (53%) or with the extrasentential referent (53%) (see Figures 37 and 38). See Section 4.1.4 for possible explanations. With respect to backward anaphora contexts, clause order had an effect on a particular context in both one-referent and two-referent sentences (Experiments 1-4). In the backward anaphora condition with main-subordinate clause order, the referent choices of NSPs diverged: in forward anaphora, the overall coreference rate for the subject was >76%, whereas for backward anaphora, the overall rate for the extrasentential referent was >66%. Note that in such a context, coreference between a c-commanding antecedent (either a NSP or an OSP) and a following lexical subject (an R-expression in the terms laid out by Principle C) is structurally prohibited. In line with the findings of Kazanina et al. (2007), where the parser did not consider antecedents in positions that would violate the binding constraint, Basque adults rejected coreference as expected (RQ1b). Note, however, that the unavailability of coreference between an OSP and the following c-commanded lexical subject was more evident in the empirical results (coreference with the subject <4-10%>) than that between a NSP and the following c-commanded lexical subject (coreference with the subject <14-34%>). The adults’ selection of the ungrammatical coreferential reading of NSPs may be a result of participants interpreting the sentences as having a gap in a different position than the one intended by the experimenter; that is, they may have understood that the lexical subject of the main clause Mikel was located in the postverbal position, as in ‘haserretu egin 69 da Mikel, ___ igo denean ‘Mikel got angry when (he) got on (the bus)’ (literally, ‘got angry Mikel when he got on the bus’) and ‘txistu egin du Mikelek, ___ Julen bultzatu duenean ‘Mikel whistled, when (he) pushed Julen (literally, ‘whistled Mikel when he pushed Julen.’ With regard to the results of backward anaphora with subordinate-main clause order, coreference relations are syntactically allowed between the embedded NSP and the following matrix lexical subject, as in (44e) and (45e) ___igo denean Mikel haserretu egin da ‘When (he) got on (the bus), Mikel got angry’. In such a case, similar to the pattern shown in forward anaphora, adults generally selected coreferential reading with the following subjects (80% in Experiment 1 and 69% in Experiment 2). In two-referent sentences such as (46e) and (47e) ___ Julen bultzatu duenean Mikelek txistu egin du ‘When he pushed Julen, Mikel whistled’ (74% in Experiment 3 and 75% in Experiment 4). Off-line tasks such as the present studies and those conducted by 69 Notice that the subject in Basque can appear either in the preverbal position (as in Jon etorri egin da ‘John has come’) or in the postverbal position (as in etorri egin da Jon ‘has come John’), where the verb phrase is in focus position. 138 Sorace and Filiaci (2006) and Belletti et al. (2007) have revealed that referential dependencies are established as soon as a licit antecedent becomes available, in keeping with the results from on-line studies (van Gompel & Liversedge 2003). In the same backward anaphora context, coreference between the embedded OSP and the upcoming matrix lexical subject was not observed, as in (44f) and (45f) Bera/ Hura igo denean, Mikel haserretu egin da ‘When he got on (the bus), Mikel got angry’ or (46f) and (47f) Ha(r)ek/ Berak Julen bultzatu duenean Mikelek txistu egin du ‘When he pushed Julen, Mikel whistled’. Despite Mikel being a syntactically legitimate antecedent, Basque adults predominantly selected the picture in which the non-linguistically mentioned but visually present character (the extrasentential referent) performed the action corresponding to the first clause (<3%-14%> coreference with the subject for hura; <10%-28%> for bera across experiments). This is in line with the patterns of responses of native Italian adults (Belletti et al. 2007, Sorace & Filiaci 2006). The disjoint interpretation of the embedded OSP seems to be affected by the universal precedence constraint formulated by Luján (1986: 249; see Section 1.1.4), according to which lexical pronouns in preposed adverbial clauses cannot precede their antecedents where they alternate with null pronouns. The unavailability of coreference in the case of bera in such a context is consistent with Abaitua’s (1994) claim that bera cannot precede its antecedent (see (42) in Section 3.2.1). However, anaphora directionality did not affect its referential properties (RQ1b). Overall, the differences in processing demands with respect to forward and backward anaphora observed in other studies (Belletti et al. 2007, Sorace & Filiaci 2006) are in line with the less consistent antecedent choices of Basque adults in backward anaphora constructions. The inclusion of factors such as clause order and anaphora type in the present study has revealed that the validity of the PAS, which has primarily been tested in forward (subordinate-main) sentences, can be extended to both forward (mainsubordinate) and backward (subordinate-main) conditions. 4.1.4 NSP/ hura / bera The results from the Picture Selection Task (Experiments 1-4) revealed that in general, coreference with the lexical subject of the preceding clause was predominant for NSPs, consistent with the results of previous studies (Alonso-Ovalle et al. 2002 for Spanish, Carminati 2002 for Italian, Filiaci et al. 2014 for Italian and Spanish, Geber 2006 for Romanian, Mayol 2009 for Catalan, Ok Kweon 2011 for Korean). The bias of the NSP towards the subject was especially strong (a) in the forward anaphora condition (76-94%) in comparison to the backward anaphora condition (69-80%) and (b) in contexts of two potential antecedents (the subject and the extrasentential referent) (>90% regardless of intransitive or transitive predicates) in comparison 139 to contexts with three potential antecedents (the subject, the object and the extrasentential referent) (76-85%). The disjoint reading of any intrasentential antecedent generally prevailed for the two OSPs, but especially for hura/ ha(r)ek, where the rates for the extrasentential referent reached over 90% across tasks. In the case of bera/ berak, these rates were only 60-65%. Consequently, the opposite referential properties of NSPs and OSPs were more evident in the experiments with the demonstrative hura/ ha(r)ek (Experiments 1 and 3) than those with the pronoun bera/ berak (Experiments 2 and 4). The patterns of results obtained from adults in the Acceptability Judgement Task (Experiments 5 and 6) confirmed those of Experiments 1 and 2, as NSPs were preferred in [-TS] contexts, whereas bera and especially hura were chosen in [+TS] contexts. It is worth noting that even Basque adults did not select the pragmatically appropriate antecedent for both NSPs and OSPs in all cases, which coincides with previous findings in studies on adult Italian speakers (Carminati 2002, Sorace & Filiaci 2006, Tsimpli et al. 2004,). The results of the six experiments validate the claim that distinct referential forms (the demonstrative hura vs. the pronoun bera) carry their own specific constraints on selecting an antecedent, as predicted by the Form- Specific Multiple-Constraints approach (Kaiser & Trueswell 2008; see Section 1.1.3). In line with this approach, as well as with the crosslinguistic tendency for demonstratives to look for a non-topic antecedent (e.g. Comrie 2000, van Kampen 1997 for Dutch; Bosch et al. 2007, Diessel 1999, Zifonun et al. 1997 for German), the results support the expected correlation between the [topic shift] feature and the demonstrative hura. No specific predictions regarding the discourse feature of bera had been made, since the lexical properties of bera have not yet been completely defined in descriptive grammars, which detail the pronoun’s mixed behaviour. Despite Sorace’s (2000) claim that OSPs in NS languages like Italian are specified for the interpretable feature [+TS], for bera this feature is only evident in 70% of cases in one-referent sentences and in 60% of cases in two-referent sentences. Thus, in the topic (non-)continuity continuum shown below (52), bera is located in the middle, sometimes coreferring with a recently 70 mentioned referent. In such cases, bera exhibits a functional overlap with the NSP. In other cases, bera may fulfill the function of the demonstrative hura, not accepting coreference. Therefore, bera would seem to not be necessarily specified for a certain discourse feature. The discourse features of Basque NSPs (specified for [-TS]) and OSPs (specified for [+TS]) are in line with the literature on the pragmatically felicitous options for marking the different contexts (e.g. Cardinaletti 1997 for Italian, Philippaki-Warborton 1987 and Tsimpli 1990 for Greek). 70 I.e. three or four words before. 140 (52) Topic continuity Topic shift NSPs OSP pronoun bera OSP (demonstrative) ‘(s)he (him/ herself)’ hura ‘that’ Figure 42. The distribution of Basque pronouns with regard to topic (non )continuity In a null-subject language like Basque, OSPs are the ‘marked’ option (in the linguistic sense), as opposed to NSPs, which are unmarked (Cardinaletti & Starke 1999a) — but which overt form is more marked? In this regard, the OSP hura seems to be more marked in intrasentential anaphora, not allowing coreference. Bera is more flexible, allowing several interpretations consistent with Laka’s (1996) claim that bera (but not hura) can have an intrasentential antecedent, as discussed in Section 3.2.1. The adults’ preference for the extrasentential referent as the first choice out of three competitors in the case of berak in forward anaphora contexts may be surprising a) based on bera’s discourse-bound properties (referring to a referent that has been recently mentioned in the discourse (Eguzkitza 1986)) and b) considering that there are two potential intrasentential antecedents in the preceding clause. Descriptive grammars usually provide examples of one-referent sentences to explain the criteria for the use of bera (more anaphoric) and hura (obviative, i.e. disjoint in reference) in intrasentential contexts (see Section 3.1.2). The examples of complex sentences with two referents provided in descriptive grammars usually involve intersentential anaphora, as in (53): (53) Mikelek i eskutitza zabaldu zuen, baina Koldok j ez zuen irakurri nahi izan. Hari i ez zitzaion ondo iruditu. ‘Mikel i opened the letter but Koldo j did not want to read it. He i didn’t seem to like that.’ (example from Laka 1996) In contexts involving more than one previously mentioned referent, as is the case for the complex (coordinated) sentence in (53), descriptive grammars present hura as a disambiguator (i.e. making the referent uniquely identifiable in an ambiguous intrasentential context with two previously mentioned potential referents). Bera is ruled out in such contexts, and hura becomes the default form (Laka 1996). The function of anaphoric demonstratives like hura seems to be limited to cases in which reference tracking is somewhat problematic, or, as Himmelmann (1996: 227) puts it: “demonstratives are used for tracking only if other tracking devices fail”. From these observations, one could claim that the OSP berak cannot play the role of disambiguator. Evidence for this can also be derived from the almost equal distribution of selec- 141 tion rates for the subject and the object in the adult data (18% and 21%, respectively). Thus, the adults’ preference for the extrasentential referent may have been the result of a process of elimination, i.e. rejection of the two worst alternatives — namely, the two intrasentential (subject and object) antecedents. A more explicit referring expression such as a lexical (proper noun) subject might have seemed more natural to the participants as a reference back to the object Julen: Mikelek Julen bultzatu duenean Julenek txistu egin du ‘When Mikel pushed Julen, Julen whistled’. In fact, Sorace et al. (2009) note that in giving participants only the option of choosing between a null or an overt pronoun, the task captures certain options allowed by the grammar but not all of them. Other studies such as Barquin and Costa (2011) have also emphasised that the lack of the lexical subject option could have an effect on the results obtained; for some participants, the most felicitous way of marking topic shift would be to use a proper noun. 71 The use of full DPs rather than pronouns in ambiguous contexts has also been observed in recent findings in which speakers are less likely to use a pronoun (relative to repeated noun phrases) when a feature-sharing (e.g. same-gender) competitor is present in the visual context (Fukumura et al. 2010), even in non-gender marking languages such as Finnish (Fukumura & Hyönä 2011). These results suggest that the competitor can become part of the discourse representation even though it has not been mentioned in the discourse. In two-antecedent contexts (the subject and the extrasentential referent) such as (47d) Mikelek txistu egin du berak Julen bultzatu duenean ‘Mikel whistled when he pushed Julen’, the absence of an intervening element (a new discourse NP, in this case the object; see Gibson 1998) between the pronoun and its antecedent seems to facilitate the coreferential reading. This can be seen in the chance rates between the subject antecedent (53%) and the extrasentential referent (47%). The two possible interpretations of bera are also visible in (54), where bera exhibits optional coreference (in Reinhart’s (1983) terms) either marking topic continuity (coreferring with Peru) or switch reference (referring to an extrasentential antecedent). (54) Peruk i esan digu berak ij egingo duela ‘Peru i has told us that he ij will do it.’ (example from Euskaltzaindia 1993: 75) The Basque adult data confirm the crosslinguistic variability in the antecedent choice of OSPs, as postulated by Carminati (2002). Whereas Basque 71 Only 40% of Spanish monolinguals and 45% of Spanish-Catalan bilinguals showed strong preferences (75% preference) for OSPs in topic-shift contexts (Barquin & Costa 2011). 142 adults chose the extrasentential antecedent for the two OSPs (over 70%) in the two tasks, Italian adults selected coreference with the preceding subject (86%) (Carminati 2002), and Spanish adults exhibited chance level selection or a slight preference for the preceding subject (Alonso-Ovalle et al. 2002; see Section 1.4). This variability across studies may be a result of the different methods used: for example, a questionnaire (Carminati 2002) instead of a Picture Selection Task and an Acceptability Judgement Task. In addition, in the Italian study, the one-referent sentences consisted of VP-complement clauses (33b), repeated here as (55a), whereas the present study included both temporal adjunct clauses (Experiments 1 and 2) and VP-complement clauses (Experiments 5 and 6). (55) a. Gregorio ha detto che __ i / lui i/ ? j sará presente al matrimonio di Maria. ‘Gregorio i has said that __ i / he i/ ? j will be present at the wedding of Maria.’ b. Mikel i haserretu egin da __ i / hura j / bera j/ ? i igo denean. ‘Mikel got angry when he got (on the bus).’ c. Minniek i __ i / hura j / bera j/ ? i erori dela esan du. ‘Minnie i has said that she has fallen.’ On the basis that complements bear a much closer relationship to the main predicate than adjuncts do (Tallerman 2005), this could have resulted in a closer relationship between the DP Gregorio and the OSP lui than the DP Mikel and hura/ bera in the adjunct clause (55b). However, note that the interpretation of hura/ bera in a VP-complement clause such as (55c) was also dominantly disjoint in reference, as the OSPs primarily took the extrasentential antecedent, despite the permissibility of coreference in some cases. Evidence thus far has shown that the biases of NSPs and OSPs seem to be based on syntactic prominence in two-referent sentences, where the OSP selects an antecedent that is relatively low in the phrase structure (e.g. an object; Carminati 2002, Filiaci et al. for Italian, Mayol 2009 for Catalan, Ok Kweon 2012 for Korean). In contrast to the uniform crosslinguistic pattern found to date, Basque adults only chose the object in 5% of the responses in Experiment 3 and 21% in Experiment 4. The distinct functions of NSPs and OSPs in Basque are thus seemingly not based on syntactic prominence as predicted by the PAS: visual context (the extrasentential referent) had a stronger effect on the experimental results than syntactically lower arguments like the object. Carminati’s observation that the bias of the Italian OSP for a non-subject antecedent is weaker in one-referent than in two-referent sentences (and hence the claim that the division of labour between NSPs and OSPs is more marked in the latter context due to the several potential antecedents) is not supported by the data presented here (see RQ1c in Section 3.3). The bias of 143 the OSP hura towards an extrasentential referent remained constant throughout all the experiments, regardless of the number of referents in the sentence. In contrast, the bias of bera was affected by the number of arguments, its preference becoming less clear in two-referent sentences. Thus, the larger the number of potential antecedent candidates, the lower the degree of accessibility (competition; Ariel 1990) — but this was only the case for bera, not hura. As anticipated in Section 3.5.3.1, the evidence points to language-internal specificities such as properties of the pronoun lui ‘he’ in Italian vs. the Basque demonstrative pronoun hura ‘that’ in explaining the different interpretations observed in the two null-subject languages. This also clarifies the dissimilar outcomes when coreference is established in intrasentential anaphora contexts: pragmatic anomaly in terms of redundancy occurs in the case of Italian, as opposed to ungrammaticality in the case of Basque. The intraand crosslinguistic differences in the interpretation of personal and demonstrative pronouns have been explained in the generative literature in terms of binding conditions, as reviewed in Section 1.3.1. Under BT (Chomsky 1981), pronouns following Principle B must be free in their governing category, implying that they cannot have their antecedent in the finite clause in which they are located (their governing category). This principle does not constrain the possibility of intrasentential coreference with the subject in complex sentences like those in the present study, since the possible intrasentential antecedents were not in the finite clause containing the overt pronoun, i.e. in the preceding or following clause. Thus, intrasentential coreference with the subject is possible for Italian personal pronouns (Cardinaletti & Starke 1999b), but not for the demonstrative hura — not even accidental coreference, the coreference between a pronoun and its antecedent in local environments: John i only saw HIM i (Cardinaletti & Starke 1999b: 284). 72 In this regard, the distribution of the coreferential vs. non-coreferential readings of the two overt Basque pronouns, the pronoun bera (32%/ 68%) and the demonstrative hura (6%/ 94%), in the results from adults is compatible with the more flexible nature of personal pronouns as opposed to demonstratives. These differences indicate that hura and bera must belong to two different categories in the pronominal distribution proposed in (56), consistent with Eguzkitza (1986: 28): 72 I am aware that the items tested in the experiments are not comparable to the example given by Cardinaletti & Starke (1999b: 284). 144 (56) Repartition of Pro-Form Binding a. B>C (i.e. Assign Principle B to the first pro-form, C to the rest) b. Personal Pronouns > Simple Demonstratives French le/ lui ‘him’ ce/ ça/ cela ‘this/ that’ English them those Italian lui/ lei ‘(s)he’ questo/ quello ‘this/ that’ Basque bera ‘(s)he (him/ herself)’ hura ‘that’ Greek aftos/ afti ‘(s)he’ (adapted from Cardinaletti & Starke 1999b: 288) Various reasons may be behind the non-coreferential interpretation of the OPSs hura and bera, as clearly illustrated in the distribution of pronominal forms in (56). In opposition to de Rijk’s (2008: 115, 209) claim that both bera and hura are equivalent to the third-person pronouns of other languages, only bera seems to be a true counterpart, a member of the first pro-form category in which pronouns are preferred to demonstratives. Under the Repartition of Pro-form Binding approach, the heterogeneous first pro-form category includes various kinds of pronominals such as (real) personal pronouns in Romance languages, Basque (quasi)pronouns — consistent with Eguzkitza’s (1986: 204) description of bera as the real pronoun obeying Principle B — and Greek (Principle B) demonstratives like aftos/ afti (Sanoudaki 2003). Thus, the demonstrative hura’s strong bias towards the extrasentential referent shown by adults corresponds to the obviative interpretation proposed by Ortiz de Urbina (1989: 147). However, the results from bera are inconsistent with the description of this pronoun as proximate (Ortiz de Urbina 1989) and logophoric in indirect speech contexts (Abaitua 1994, Garzia 1996). As the present study does not deal with the analysis of language evolution, we cannot hypothesise that bera may be in an ongoing process of grammaticalisation in which ultimately its only remaining function may be the third-person pronoun, as Heine and Kuteva (2006: 247) predict. According to these authors, Basque is grammaticalising the identity pronoun ber ‘self, same’ (corresponding largely to what König and Siemund (2000) would call an intensifier) into a third-person pronoun. Heine and Kuteva (2006) describe grammaticalisation in terms of desemantisation, whereby the specific intensifier (‘self’) or identity semantics (‘the same’) fades away with the effect that the third-person reference is the only semantic function that is left. In contrast to traditional grammars’ description of the demonstrative hura ‘that’ as the neutral or the purely pronominal form (Martínez-Areta 2013: 318, Ortiz de Urbina 1989: 147), the current experimental data show that bera is more of an ordinary third-person pronoun than a quasipronoun (de Rijk 2008: 114). This is also consistent with Martínez-Areta’s (2013: 320) observation that in western(-central) varieties, bera has become progressively less 145 marked, pushing hura into third-grade deictic meanings and itself becoming a fully-fledged third-person pronoun. 4.2 Children’s interpretation of pronouns The developmental data gathered from 6to 8-year-old children allowed us to test whether the referential and discourse properties of NSPs and OSPs in child language are adult-like at this age, and whether extralinguistic factors such as biological age, age of onset of acquisition and amount of exposure to the language had any effect on children’s knowledge of the properties of NSPs and OSPs. To that end, the six experiments discussed in the previous section were run on two (cL1 and cL2) child groups similar in age. In general, both child groups showed a preference for intrasentential coreference with the preceding or following subject regardless of pronoun type in the Picture Selection Task, and a general preference for NSPs regardless of discourse contexts in the Acceptability Judgement Task. In the following subsections, the results obtained from the two child groups in the forward anaphora condition in each task are compared to the adult data (discussed in Sections 4.1.1 and 4.1.2 above). The effects of clause order and anaphora type are addressed in Section 4.2.3. Finally, Section 4.2.4 discusses Basque children’s a) reluctance to use non-discourse referents for pronouns, as observed in studies on the acquisition of binding conditions, and b) target-deviant patterns of both NSPs and OSPs in different discourse contexts in comparison to prior findings. 4.2.1 NSP vs. hura in forward anaphora 4.2.1.1 One-referent sentences (Experiments 1 and 5) In Experiment 1, the cL1 group chose the preceding lexical subject as the antecedent of both matrix and embedded NSPs (76%) in complex sentences with two intransitive predicates. Although the coreference rates were significantly lower for the OSP hura ‘that-ABS’ than for NSPs, the cL1 group again tended to prefer a coreferential interpretation for hura (62%) regardless of the clause in which the OSP occurred. For the cL2 group, the variable pronoun type had no effect on the results; there was a generally strong tendency for coreference with the subject regardless of pronoun type (null or overt, <78-85%>). A comparison of the adult results with those of the child groups showed that in forward anaphora, all children exhibited a preference for the subject antecedent in the NSP condition (76%/ 84% for cL1 and cL2), differing significantly from adults’ overwhelming preference (93%). With respect to the OSP hura, children differed significantly from adults, again preferring the preceding lexical subject antecedent (62%/ 81% for cL1 and cL2), in contrast 146 to adults’ preference for the extrasentential antecedent (only 6% coreference). When the results of the two child groups were compared, the significant three-way interaction between anaphora, clause order and group revealed a stronger tendency to prefer coreference with the subject for the cL2 group (79-86%) than for the cL1 group (55-69%), regardless of clause order and anaphora type. In the Acceptability Judgement Task (Experiment 5), although children exhibited preferences similar to those of adults, accepting more NSPs in [- TS] than in [+TS] contexts, the patterns of responses differed significantly, as children generally preferred NSPs. In [-TS] contexts, cL1 and cL2 chose significantly more OSPs (27% and 22%, respectively) than adults (10%). In [+TS] contexts, in contrast to adults’ preference for the OSP hura (89%), cL1 and cL2 children chose hura only 41% and 36% of the time, respectively. A comparative analysis of the data from cL1 and cL2 groups did not yield significant differences, which suggests that the response patterns of the two child groups were similar. 4.2.1.2 Two-referent sentences (Experiment 3) In three-antecedent contexts in complex sentences containing two transitive clauses, the patterns of responses of children and adults were similar for NSPs, but the preference for the preceding lexical subject was significantly higher for adults (76%) than for cL1 and cL2 (57%/ 47%, 73 respectively). With respect to the OSP har(e)k ‘that-ERG’, the extrasentential antecedent was overwhelmingly chosen by adults (90%) out of three possible antecedents. This option was only favoured by cL1 in 26% of cases and by cL2 in 16% 74 of cases. The response patterns of the two child groups were very similar in general for both NSPs and OSPs. With regard to two-antecedent contexts, both cL1 and cL2 groups chose coreference with the preceding lexical subject for NSPs significantly less often (68% and 59%, respectively) than adults (83%). The differences in the coreference rates of the child groups were also significant. For the OSP ha(r)ek, both cL1 and cL2 groups preferred coreference with the preceding lexical subject (64% and 68%, respectively), as opposed to adults, who preferred the extrasentential antecedent (only 3% subject coreference). 73 The mean rate for the object was 23% for the cL1 group and 28% for the cL2 group; the mean rates for the extrasentential referent were 20% and 16%, respectively. 74 The mean rate for the subject was 52% for the cL1 group and 64% for the cL2 group; the mean rates for the object were 22% and 21%, respectively. 147 4.2.2 NSP vs. bera in forward anaphora 4.2.2.1 One-referent sentences (Experiments 2 and 6) In the Picture Selection Task (Experiment 2), coreference with the preceding lexical subject was predominantly chosen (84%) by the cL1 group for both matrix and embedded NSPs. For the OSP bera ‘(s)he (him/ herself)-ABS’, the cL1 group chose the subject as the antecedent in subordinate-main clause order sentences (65%) and to a significantly higher degree in mainsubordinate sentences (75%). As in Experiment 1, pronoun type had no effect on the results of the cL2 group, who showed a preference for coreference regardless of pronoun type (<63%-83%>). The comparison of results between age groups revealed that both cL1 and cL2 groups preferred the subject as the antecedent for NSPs (83% and 78%, respectively), similar to adults (92%). The difference in the strength of this preference between cL2 and adults was significant. For the OSP bera, both cL1 and cL2 again chose coreference with the preceding lexical subject (70% and 63%), differing significantly from adults’ preference for the extrasentential antecedent (only 32% coreference). The comparison between child groups showed that, similar to Experiment 1, only the interaction between clause order, anaphora and group approached significance, suggesting that the cL1 group’s coreference rates were not significantly higher than those of cL2 in either forward (75%/ 59%) or backward (61%/ 55%) anaphora. In the Acceptability Judgement Task (Experiment 6), both child groups differed significantly from adults’ choice patterns for NSPs and the OSP bera in [+TS] contexts since discourse context had not effect for children. In the [- TS] condition, the cL1 and cL2 groups chose significantly more OSPs (42% and 54%, respectively) than adults (16%), but they selected fewer OSPs (33% and 46%, respectively) than adults (74%) in the [+TS] condition. 4.2.2.2 Two-referent sentences (Experiment 4) Similar to the results of Experiment 3, in three-antecedent contexts, the preceding embedded lexical subject was the preferred antecedent for NSPs for the cL1 group (56%), 75 whereas the responses were more equally divided between the preceding embedded lexical subject (43%) and the object (48%) in cL2; this contrasts with the adults’ overwhelming preference for the subject (85%). For the OSP berak ‘(s)he (him/ herself)-ERG’, coreference with the preceding subject was again selected by the cL1 group (44%); 76 in the cL2 group, similar to the pattern shown for NSPs, preferences were equally distributed between the preceding lexical subject (41%) and the object (40%). The children’s results differed significantly from those of the adults, whose 75 The mean rate for the object was 28%; for the extrasentential referent, 16%. 76 The mean rate for the object was 29%; for the extrasentential referent, 27%. 148 preferred option was the extrasentential referent (61%). A comparison of the results from the two child groups revealed that the response patterns were generally very similar. In two-antecedent contexts, the children’s and adults’ patterns of responses were similar for NSPs, with coreference rates being slightly higher for the latter (83%) than the former (77% for cL1 and 71% for cL2). With respect to the OSP berak, coreference rates were lower in adults (41%) than in cL1 (60%) or cL2 (70%); the difference between the rates of the adults and those of cL2 was significant. The two child groups did not differ significantly in their responses in two-antecedent contexts. Notably, the increase in referents (from two-referent contexts to three-antecedent contexts) did not alter the children’s preference for the subject. 4.2.3 Clause order and anaphora directionality As discussed in previous sections, the child data from Experiments 1 and 2 indicated the predominance of the subject as the antecedent for both NSPs and OSPs in forward anaphora contexts, regardless of clause type. Similarly, children showed a preference for coreference with the following subject in both matrix and embedded NSPs and OSPs in backward anaphora contexts. However, the coreference rates for both NSPs and OSPs were higher in forward anaphora contexts than in backward contexts for the cL1 group (<74%- 85%> for NSPs and <60%-75%> for OSPs in forward anaphora vs. <60%- 65%> and <50%-58%> in backward anaphora). The same tendency of higher coreference rates in forward than in backward anaphora (<69%-87%> vs. <42%-79%>) was also observed for cL2 (see Figures 11 and 21), despite the fact that the variable pronoun had no effect. As indicated previously in 4.1.3, the effects of clause order and anaphora type could not be analysed in two-referent sentences (Experiments 3 and 4). However, like in forward anaphora, both cL1 and cL2 groups showed a slight preference for the following subject as the antecedent for NSPs and OSPs in two-antecedent contexts (<53%-72%> for both kinds of pronouns), with rates often at chance level. Similarly, in three-antecedent contexts, where coreference is not syntactically allowed (Principle C), they opted for coreference between the NSP and the c-commanded lexical argument, both in NSPs (<50%-69%> for cL1 and <47%-72%> for cL2) and OSPs (< 35%- 53%> for cL1 and <36%-73%> for cL2). Basque children’s non-target-like performance in backward anaphora contexts contrasts with Italian children’s target-like performance (Italian children rejected coreference in this context; Guasti & Chierchia 1999/ 2000), as mentioned in Section 2.2.1.2. However, it should be noted that the linguistic items used in the current study contained a referential NP such as Mikel, thus differing from the items used in the Italian study, in which a quantified NP (un musicista) was used (see example (35) in Section 2.2.1.2). 149 Based on the Principle Branching Direction (Lust 1981; see Section 2.2.1.2) — according to which in early language, the direction of pronominalisation is in agreement with the branching direction of the particular language being acquired — one might hypothesise that children acquiring a leftbranching language like Basque would exhibit more adult-like performance in backward anaphora contexts than in forward anaphora contexts. However, a comparison of adult and child data with regard to differences in the preferences observed in forward and backward anaphora does not lead to any clear conclusions; the differences between the adults and the child groups are sometimes greater in forward anaphora and other times in backward anaphora (compare for example Figures 12 and 13, and Figures 22 and 23). As mentioned above, the similar percentages obtained for NSPs and OSPs within each type of anaphora (forward vs. backward) in Experiments 1 and 2, especially in the cL1 group, seem to suggest that anaphora type plays a more important role in assigning an antecedent than pronoun type; see the general results of Figures 7 and 16, where coreference rates for NSPs and OSPs are higher in forward <60%-85%> than in backward anaphora <50%- 65%>. These data and the less consistent responses of children in backward anaphora (where preferences are often close to chance) suggest that the children’s results may be motivated by the Forward Pronominalisation Principle (O’Grady 1997: 217), according to which forward patterns of pronominalisation are preferred to backward anaphora by the acquisition device. Basque acquisition data are compatible with the more costly processing of backward anaphora, since this processing requires retention of the pronoun in memory before its potential referents are encountered, as suggested by Reinhart (1986: 140). The processing demands that backward anaphora in particular entail were also evident in the response patterns of adults, as discussed in Section 4.1.3. 4.2.4 NSP/ hura / bera In general, the almost complementary distribution of NSPs and OSPs observed in adults was largely missing from 6to 8-year-olds’ grammar, as children’s interpretations of OSPs and NSPs in the Picture Selection Task reflected similar choice patterns. However, coreference rates were occasionally higher for NSPs than for OSPs, as seen with cL1 in Experiment 1. The patterns of results exhibited by children and adults for NSPs were similar but quantitatively different: the subject antecedent was predominantly chosen by adults, but to a significantly lower degree by children. For OSPs, both cL1 and cL2 showed a(n overextended) preference for the coreferential reading of both hura and bera in comparison to adults, who preferred the extrasentential referent as the antecedent of the OSPs. The visual context appeared to be a compelling source of information for adults, despite its low 150 accessibility (only being visually present). In contrast, the sentence-external referent was not a viable competitor for children, who showed a preference for finding a discourse antecedent. In the case of hura/ har(e)k, the children’s coreferential reading is target-deviant (ungrammatical due to its noncoreferential specific properties, as discussed in Section 4.1.4), and the cL2 group exhibited an even more target-deviant pattern than the cL1 group in Experiment 1. However, cL1 and cL2 participants’ tendency to corefer the OSP bera/ berak with a subject antecedent, diverging from the Basque adults’ interpretation, cannot be considered pragmatically incorrect or odd in terms of redundancy. Remember that in adult Basque, it is possible to interpret the overt pronoun bera as coreferent with an antecedent recently mentioned in discourse (see Section 3.2.1). In the Acceptability Judgement Task (Experiments 5 and 6), children and adults differed significantly: the cL1 and cL2 groups had not yet acquired the discourse specificities of the two Basque OSPs hura [+TS] and bera [+TS>- TS]. The general rejection of both OSP choices regardless of the [+TS] context (due to the children’s preference for NSPs, especially in Experiment 5) suggests children’s unfamiliarity with the particular (non-deictic and nonintensive) anaphoric use of hura and bera, respectively. The between-group comparisons from the six experiments discussed in Sections 4.2.1 and 4.2.2 thus confirm that biological age (adults vs. children) had an effect on the results: the data indicate that the knowledge of some properties of these pronouns are still underdeveloped at the age of 6-8. The prediction that differences between children and adults would primarily be observed in the interpretations of OSPs (RQ3) (based on Carminati’s (2002) observation that OSPs show more flexibility in selecting an antecedent) has been fulfilled, but the results also revealed some unexpected differences in NSPs. With regard to the referential properties of OSPs, the children’s targetdeviant interpretation of the overt pronouns hura and bera is compatible with the misinterpretation/ overproduction of overt pronouns observed in (child and attrited/ near-native) bilingual speakers in comparison to monolinguals (Serratrice 2007, Tsimpli et al. 2004, Tsimpli & Sorace 2006). However, the prediction that the cL2 group would show a more target-deviant performance — based on the idea that for these successive bilinguals, Basque is their weaker language due to the later age of onset and limited amount of exposure — has not been upheld (RQ3). Only minimal differences were observed between cL1 and cL2 groups. More precisely, the cL2 group only showed different (higher) coreference rates than those of the cL1 group for hura (Experiment 1). The potential influence of Spanish could be expected in terms of a transfer of the use of Spanish overt anaphoric pronouns into Basque. However, this possibility seems to be quite remote. The low occur- 151 rence of Spanish overt pronouns in the speech of children 77 and the fact that the distribution of overt anaphoric pronouns in Basque (demonstratives and bera-like pronouns) does not correspond to that of Spanish weaken the explanation of the cL2 group’s patterns by means of interlinguistic interference. 78 In addition, as even the participants in the adult control group were Basque-Spanish bilinguals, we cannot discard the possibility that Spanish has some influence on the use of Basque anaphoric pronouns. In this regard, it should be noted that no comparison between monolingual and bilingual Basque speakers is possible due to the fact that no Basque monolingual adults or children participated in the studies. As discussed in 2.3.3, the effects of extralinguistic factors such as the age of onset and amount of exposure are not easy to untangle, since they interact with each other; this was certainly the case for the two groups of children in the present study. However, based on the observation that both cL1 and cL2 groups showed target-deviant performance, exhibiting similar response patterns for NSPs and OSPs across experiments, and the fact that the minimal differences between groups seem to be quantitative rather than qualitative, the effect of the age of onset of acquisition of the cL1 (from birth) and the cL2 group (since age 3) does not seem to be a major determining factor. With regard to the quantity of input (amount of exposure) to Basque, in the case of the cL2 group, this variable was controlled, since their exposure to Basque was restricted to school hours (6 hours per day). For the cL1 group, it is difficult to know the amount of exposure they had to Basque outside of the school environment. Thus, the cL1 group was not completely homogenous, as their exposure might have ranged from 60% to 90%. However, a unified explanation is required for the data from cL1 and cL2. Among the explanatory factors discussed in 2.3 with regard to the instability at the syntax-discourse interface, the amount of exposure in relation to the frequency of third-person pronouns seems to play an important role in the data obtained, as will become clear from the following paragraphs. However, methodological limitations should not be disregarded when accounting for some differences between cL1 and cL2 groups. In the experiments with tworeferent sentences (Experiments 3 and 4) in which participants were asked to choose one out of three pictures, some children showed some difficulties. In fact, the number of children selecting an infelicitous (logically impossible) 77 Bel (2003) did not find any occurrence of Spanish third-person singular pronouns before age 3 in the production of the children she analysed, and only one form of a third-person pronoun was observed in Catalan. 78 Ezeizabarrena (personal communication) observes that Basque-dominant children inappropriately use the Spanish demonstrative aquél ‘that’ as anaphoric in contexts such as Aquél me dijo que… ‘He told me that…’ where Spanish native speakers would use a null or an overt pronoun instead, __/ él me dijo que. These examples may be regarded as instances of children transferring the anaphoric properties of the Basque demonstrative hura to the Spanish demonstrative aquél. 152 option (as mentioned in 3.5.4 and 3.5.5) should not be passed over, where the mean rate of missing data was higher in cL2 than in cL1 (<16%-18%> for cL2 vs. <7%-10%> for cL1, see Figures 1 and 2 in Appendix A). The differences in the rates indicate that the task was more demanding for cL2 than for cL1. Moreover, ten cL2 participants were excluded in Experiment 3 because they chose the impossible option more than 10 times out of 42 items. These data suggest that the task may have also turned out to be more difficult for the cL2 group. The child data appear to be incompatible with the form-function correlation proposed in the Accessibility Theory (Ariel 1990; see Section 1.2). However, the subject that was the most salient antecedent in terms of syntactic role and topichood, as well as the most accessible referent in terms of distance (most recently mentioned), generally determined the antecedent choices of the children. One possible explanation for Basque children’s general preference for coreference with the subject could be that children operate under the Parallel Function Strategy (Sheldon 1974), 79 according to which a pronoun with two possible antecedents will choose as its antecedent the element with the same grammatical function — in our case, the subject of the preceding clause. However, as the linguistic items in the present study only analysed pronouns in subject position and not in object position (see Smyth & Chambers 1996 and Chambers & Smyth 1998 for parallelism effects), we cannot conclude whether the children’s preference for the subject is due to parallelism in the syntactic role or due to a subject bias. As mentioned in Section 2.2.1, the generative acquisition literature has shown that children rarely take a referent unmentioned in the experimental item as the antecedent for a pronoun (Grimshaw & Rosen 1990: 200-202). Furthermore, when there is no syntactically licit intrasentential antecedent for a pronoun, children may disregard syntactic constraints in order to provide one. Children’s preference for local coreference (permitting coreferential readings with counterindexed NPs) has been explained in the acquisition literature in terms of the Delay of Principle B Effect (see Section 2.2.1). This well-established pattern seems to arise from children’s knowledge that pronouns have to be anchored in discourse, such that when an anchor is not immediately obvious, they may look for it in the wrong domain, with pragmatics overriding syntax (Wexler 1999). In this regard, evidence of children’s overwhelming preference for a sentence-internal referent and reluctance to assign unmentioned antecedents has been found in studies using a variety of methodologies: grammaticality judgement tasks (Sigurjonsdottir, Hyams & Chien 1988, Wexler & Chien 1985, acceptance of non-coreference around 79 Sheldon observes that comprehension of subject relative clauses in English children is facilitated when the grammatical function of the relativized NP is the same in the matrix clause and the embedded clause (the subject in both cases). 153 5%), picture selection and verification tasks (Sekerina et al. 2004, Serratrice 2007) and, to a lesser extent, act-out tasks (Solan 1983). The linguistic items in the present study were not designed to test whether 6to 8-year-olds have acquired Principle B, since the pronouns of the linguistic items were free within their clause. However, the results of Basque child participants are compatible with previous findings in terms of their preference for sentenceinternal antecedents, despite discrepancies in the design of the study and in the linguistic structures tested. It should be emphasised that all of the sentences tested in Experiments 1 and 2 had only one potential referent in the linguistic context (the preceding or following subject), and this might have increased children’s preference for a sentence-internal antecedent for the overt pronouns: the extrasentential referent, only visually available, was not highly accessible to them. Recall that while processing isolated one-referent sentences, Italian native adults also preferred to take an antecedent in the sentence itself rather than resorting to an extrasentential referent in onereferent sentences (Carminati 2002: 96). However, the increase of referents (from twoto three-antecedent contexts) did not alter Basque children’s preference for the subject. As O’ Grady (1997: 239) suggests, it is quite justifiable for children to prefer local coreference — this is known as the Locality Requirement (to reduce memory load; see Solan 1983: 119) — and topical entities (due to the Prominence Requirement; Givón 1984: 138, Kuno 1987: 159). Evidently, the children in the studies assigned the pronoun to an intrasentential antecedent at the cost of violating hura’s non-intrasentential coreferential properties. As Serratrice (2007) notes, this may indicate that the construction of an extralinguistic context in which this referent can be a plausible antecedent exceeds children’s processing capacities. With regard to the discourse properties of the NSPs and OSPs tested in the Acceptability Judgement Task, Basque children’s bidirectional targetdeviant patterns in both [-TS] and [+TS] contexts are compatible with the findings from same-age children acquiring Romance languages, as shown in Table 5. Using exactly the same methodology, in [-TS] contexts in which the pragmatically felicitous pronoun choice would be the null pronoun, all child data demonstrate an overacceptance of OSPs (<25%-54%> in Sorace et al. (2009) for Italian, <40%-48%> in Barquin & Costa 2011 for Spanish and <22%-54%> in the present study). Note that studies carried out in intersentential anaphora contexts with a different methodology, such as Shin and Cairns (2012) for Spanish, have also reported redundant OSPs in [-TS] contexts at ages 6-7. However, the non-target-like performance of these children is not completely asymmetric, in contrast to previous comprehension data (e.g. Serratrice 2007), in which only a unidirectional deviance pattern (a misinterpretation of OSPs but not of NSPs) was observed. Pragmatically inappropriate NSPs were also chosen in [+TS] contexts across studies (<25%- 45%> for Italian, which increases to <51%-62%> for Spanish and to <54%- 154 67%> for Basque); this may prevent the successful identification of the pronoun antecedent. Table 5. Crosslinguistic patterns of 6to 8-year-olds and adults for OSPs in discourse contexts Studies Discourse contexts Italian OSP lui/ lei [-TS] [+TS] Sorace et al. (2009) monolingual children 30% 75% Italy English-Italian bilingual children 35% 70% UK English-Italian bilingual children 54% 65% Spanish-Italian bilingual children 25% 55% monolingual adults 10% 80% Spanish OSP él/ ella Barquin and Costa (2011) monolingual children 48% 49% Spanish-Catalan bilingual children 40% 38% monolingual adults 25% 61% bilingual adults 11% 63% Shin and Cairns (2012) monolingual children 41% 51% monolingual adults 27% 83% Basque OSP Iraola (2014) hura/ bera hura/ bera cL1 27%/ 42% 41%/ 33% cL2 22%/ 54% 36%/ 46% bilingual adults 10%/ 16% 89%/ 74% Studies testing children at different ages have shown uneven development in sensitivity to [+TS] discourse contexts: sensitivity to the [+TS] feature of OSPs emerges earlier in monolingual children (by age 6 for Italian and by age 8 for Mexican-Spanish and Castilian-Spanish) than a preference for NSPs in [-TS] contexts (by age 8 for Italian; not visible at age 8-10 for Castilian- Spanish, nor at age 14-15 for Mexican-Spanish). Likewise, proportionally more difficulties were observed in avoiding OSPs in [-TS] contexts in English-Italian and Spanish-Italian bilingual children (Sorace et al. 2009). However, bilingual children in general show persistent difficulties in both dis- 155 course contexts (Barquin & Costa 2011, Sorace et al. 2009); this effect is also observable in the two groups of Basque bilingual children, who showed a general preference for NSPs over OSPs regardless of the discourse context in Experiment 5. The Basque study does not provide a developmental picture of differential sensitivities to discourse conditions, since only data from children aged 6 to 8 were collected, but the preferences between adults and children diverged more in [+TS] contexts (48% difference for cL1 and 53% for cL2) than in [-TS] contexts (17% difference for cL1 and 12% for cL2) in Experiment 5. A similar pattern was also observed in cL1 in Experiment 6 (37% difference in [+TS] contexts vs. 22% in [-TS] contexts); the differences were greater in [-TS] contexts in cL2 (34%) than in [+TS] contexts (24%). Thus, in contrast to previous results, where redundancy (i.e. overacceptance of OPS in [+TS] contexts) was more problematic, ambiguity (i.e. overacceptance of NSPs in [+TS] contexts) presents a more significant problem in the Basque data. The high rates of NSPs in a switch-referent context are especially evident in the case of hura: 64% preference for cL2 and 59% for cL1, which is consistent with Spanish-Catalan bilinguals’ patterns in the same context (62% preference; Barquin & Costa 2011). The Basque children’s general preference for NSPs may have been affected by the difficulty of taking another person’s perspective — a third-person perspective when reporting what had occurred in the [+TS] context — an effect observed in other child studies (Shin & Cairns 2012). Interestingly, this preference for the NSP is the reverse of previous studies, where the overt pronoun form has been regarded as the default option for both bilingual and monolingual speakers, either because their resources are developing (in the case of monolinguals) or because of their two linguistic systems (in the case of bilinguals’ preferences; Sorace 2011; see 2.3.2). As Tsimpli (2011) observes, there is thus a paradox in the overall uniform developmental findings from different bilingual populations involving the discrepancy between the notion of linguistic default and that of learner default. Despite the fact that in null-subject languages, the null pronoun is considered to be the weaker/ non-marked element (see Cardinaletti & Starke 1999a), the learner treats the strong element (the overt pronoun) as the default, similar to the primacy of strong pronouns in monolingual acquisition data observed in the acquisition of English pronouns (Cardinaletti & Starke 1995), 80 as mentioned in Section 2.2.1. Tsimpli (2011) hypothesises that the use of the overt pronoun as the default may be a result of its multiple interpretations. For example, the overt pronoun can be coreferent with a nontopical antecedent when unstressed, it can corefer with the preceding subject topic when emphasised or it can be used as deictic. In this regard, two char- 80 Cardinaletti and Starke (1995) explain the Delay of Principle B in English children as the inability to differentiate between <deficient; strong> ambiguous pronouns; as a result, children use the strong form as the default. 156 acteristics of the Basque language should be taken into account for the data obtained. As mentioned in Section 3.1, multiple (subject and object) argument drop is a frequently employed option in Basque, as the verbal inflection can mark up to three arguments, and the frequency of the OSPs hura and bera in the anaphoric use for establishing reference seems to be low, with synonyms or associations more frequently observed in narratives (Garcia-Azkoaga & Idiazabal 2004). A small corpus-based study using four readings (Ikastolen elkartea 2002) typically included in the school materials of the child participants has shown that hura and bera are truly scarce in (at least written) narratives. 81 Out of a corpus of 428 utterances, of which 293 contained a third-person singular inflected verb, the NS was the dominant option (40.27%, n=118). Proper noun or determiner phrases were the most frequent overt subjects (33.55%, n=161), while pronouns were much less frequent (n=14). Among them, there was only one occurrence of hura and three of (the ergative form of) bera (berak). The null pronoun’s high usage rate suggests that its scope is much wider than that of the overt pronoun, coreferring with both topic (39.20%, n=118) and non-topic antecedents (17.06, n=52). The present data thus suggest that Tsimpli’s paradox does not apply to child Basque, since it is the null pronoun that functions as both the linguisticand the learner-default option. Consequently, the learner default may vary crosslinguistically even across null-subject languages (the Basque null vs. the Italian overt pronoun as default options), depending on the specificities of the language. With respect to the frequency of overt third-person subject pronouns in adult Basque speech, the only available data that can be brought to bear on this question is the frequency of overt subjects in a story narrated by a Basque native adult, which was part of an experiment conducted by Ezeizabarrena (2012). This adult produced 46 overt subjects out of 109 finite clauses (42.2%), but there was not a single occurrence of hura or bera fulfilling the role of third-person pronouns. It is therefore possible that the (in)frequency of overt third-person subject personal pronouns in adult null-subject languages (input) plays a significant role in the development of these pronouns in child language, as hinted at by Shin and Cairns (2012). Similarly, Sorace et al. (2009) also claim that for the demanding task of the efficient coordination of the multiple factors involved in the choice of pronominal forms, in particular sensitivity to redundancy, many years of exposure are needed. Additionally, one cannot ignore the fact that null-subject languages may present an extra challenge for children in terms of the acquisition of the pragmatic conditions governing the use of each referring form. In the case of Basque, for instance, the distal demonstrative hura competes not only with null pro- 81 I am very grateful to Prof. Antonella Sorace for suggesting the idea of counting the number of overt pronouns in children’s short stories. 157 nouns but also with the pronoun bera, as well as with other overt forms such as the proximate and medial demonstratives hau ‘this’ and hori ‘that’. Although the comparative effect of the age of onset of acquisition between cL1 and successive bilinguals of Basque was not highly significant in the present study, it is worth noting that other Basque studies on different linguistic phenomena (also in the external domain, such as discourse cohesion) have reported significant differences. For instance, Manterola and Almgrem (2013) have observed that 5-year-old cL2 children have acquired a good knowledge of communicative skills in storytelling, comparable to those of cL1; in some cases, the cL2 group may even exhibit a performance closer to that of adults. Similarly, in linguistic structures involving an interface between linguistic subcomponents (such as syntax-morphology, where the subject ergative-marking is located), significant differences have also been observed between cL1 and cL2; even at age 8, cL2 still omit the ergative case suffix -k (Ezeizabarrena 2012). The discrepancies in the results of cL1 and cL2 depending on linguistic phenomena suggest that there is a need for further research to determine the linguistic domains in which cL2 acquisition is similar to L1 acquisition. 159 5 Conclusions The adult results of the six experiments conducted using two different methodologies provide new empirical evidence of the division of labour between NSPs and OSPs in Basque, in line with Carminati (2002). In a Picture Selection Task in which participants were asked to express their preferences for the referential properties of NSPs and OSPs, NSPs were predominantly interpreted as coreferent with the subject by adults, regardless of the number of potential antecedents in the sentence. With regard to OSPs, the demonstrative hura was interpreted as disjoint in reference, pointing to an extrasentential referent in contexts involving both two and three possible antecedents. The requirement of the disjoint reading for hura/ har(e)k across experiments (in both one-referent and two-referent sentences) was explained by its non-intrasentential coreferential properties. The pronoun bera was linked to an extrasentential referent, but to a lesser degree than in the case of hura, and the rates for the extrasentential antecedent were higher in onereferent sentences than in two-referent ones. Adults’ preference for the nonlinguistically mentioned referent in the case of bera was explained on the basis of Accessibility Theory (Ariel 1990, 2001), according to which the less accessible a referent is (non-linguistically mentioned and only visually present), the more information adult speakers will encode in the chosen referring expression (overt pronouns). The adults’ choice for the extrasentential (non-linguistic) antecedent of berak out of three potential antecedents may have been the result of the elimination of the two other possible (subject and object) antecedents. The results of the Acceptability Judgement Task confirmed those of the Picture Selection Task, as the adult results indicated a clear correlation between the [+TS] feature and both OSPs (hura slightly more than bera), and between the [-TS] feature and NSPs. However, the division of labour between NSPs and OSPs was not evident in children, for whom pronoun type did not determine antecedent choices. In the Picture Selection Task, both child groups showed a general preference for intrasentential coreference with the subject for both OSPs and NSPs. With regard to the response patterns in the Acceptability Judgement Task, NSPs were generally the preferred antecedent for both child groups, regardless of discourse contexts with hura (Experiment 5) and children showed no clear preference with bera (Experiment 6). Bidirectional targetdeviant patterns were observed, resulting in redundant OSPs in [-TS] contexts as well as ambiguous NSPs in [+TS] contexts, as reported in other child studies (Barquin & Costa 2011, Shin & Cairns 2012, Sorace et al. 2009). The variables clause order and anaphora type included in the Picture Selection Task revealed that the relative order of main and subordinate clauses 160 did not affect either adult or child participants in forward anaphora conditions. However, in backward anaphora sentences with main-subordinate clause order, coreference is not grammatically possible, as the adult data confirmed whereas for children, coreference with the subject was still the preferred option, similar to the tendency shown in forward anaphora contexts. Interestingly, children’s interpretations seem to be more dependent on anaphora type than pronoun type, as is evident from their similar preferences for coreference with the subject within each type of anaphora (Experiments 1 and 2). Both the chance-level performance of children in backward anaphora sentences and adults’ less consistent responses in such conditions suggest that anaphora resolution in backward anaphora contexts is more demanding. RQ1 explored whether NSPs and OSPs differ in terms of their antecedent choices in Basque; this distinction is supported by the experimental data from adults. The referential and discourse features of NSPs in adult data resemble those of previous research in several null-subject languages. Nevertheless, differences were observed in the interpretative properties of OSPs when compared to the crosslinguistic patterns and in the nature of the division of labour between NSPs and OSPs, which was not based on syntactic prominence in the present study. As opposed to the crosslinguistic pattern where the preceding object was the most preferred antecedent for the OSP out of two (or three) potential antecedents, the extrasentential referent was the favoured choice for ha(r)ek and less clearly for berak. The crosslinguistic differences found between the Basque demonstrative hura and OSPs in other null-subject languages were explained via the degree of pronominality: hura as a typical demonstrative is subject to Principle C (Eguzkitza 1986) and hence cannot have an intrasentential (subject or object) antecedent, whereas the third-person personal pronouns in other null-subject languages operating under Principle B may (though not necessarily) have an intrasentential antecedent, as is the case for the pronoun bera. With regard to RQ2, which asked whether experimental data are compatible with the descriptions provided by Basque grammarians, the rigidity of hura across tasks supports the description of Ortiz de Urbina (1989) of hura as an obviative (disjoint) pronoun. However, the proximate/ intensive description of bera (Ortiz de Urbina 1989) and its logophoric properties in reported speech contexts (Abaitua 1994, Garzia 1996) are not completely borne out by the data from the Acceptability Judgement Task. Despite Basque descriptive grammars’ explanation of both forms as equivalent to third-person pronouns in other languages (de Rijk 2008), bera seems to be a closer parallel than hura. At least, this is the case with regard to the binding conditions, since bera behaves as a proper personal pronoun. The flexibility (broader coreferential possibilities) of bera as opposed to the rigidity of hura supports the versatile functions of the former. These differences in the referential 161 properties of hura and bera are consistent with the idea that different types of referring expressions can be differentially sensitive to a number of constraints, as predicted by the Form-Specific Multiple-Constraints approach (Kaiser & Trueswell 2008). Finally, with respect to RQ3, which investigated whether or not children and adults would exhibit similarities in the interpretation of NSPs and OSPs, the child data revealed that the referential properties and the discourse features related to pronouns are susceptible to development, and are still not adult-like in children aged 6 to 8. That is, the interpretable [+TS] feature mapped onto the demonstrative hura and the pronoun bera in the adult grammar are still underspecified in the grammars of cL1 and cL2, consistent with previous results from simultaneous and successive bilingual children acquiring Romance languages (Barquin & Costa 2011, Pladevall 2010, Shin & Cairns 2012, Sorace et al. 2009). The expected differences between children with an age of onset of 3 years (cL2) and less exposure to Basque and children with a younger age of onset (cL1) and more exposure have not been confirmed. The comparison of results between the child groups reveals similar preferences for the antecedents of pronouns, the minimal differences being more quantitative than qualitative. Therefore, the non-linguistic factor of age of onset does not seem to have had any impact on the results. In contrast, the amount of exposure plays a crucial role in the acquisition of the discourse features of pronouns, especially in relation to the infrequency of pronouns in the input to which children are exposed. Interestingly, the Basque data do not exhibit the paradox of discrepancy between the linguistic default (the null pronoun in Cardinaletti & Starke 1999a) vs. the learner default (the overt pronoun) observed in other nullsubject languages like Italian (Tsimpli 2011), since the null pronoun performs both functions in Basque. Although Basque cL1 and cL2 learners showed delays at the syntax-discourse interface, it should be noted that the nature of the target language property under study also plays a significant role in the response patterns obtained. Multiple argument drop, the lack of real third-person pronouns and the varied subject pronouns may explain Basque children’s preference for null pronouns. In addition to these interesting observations, the current study has also raised certain questions regarding the methodology used. Furthermore, the existence of crosslinguistic influence effects in the linguistic performance of bilinguals is an aspect that requires further attention. This study reported experimental data from off-line tasks. In future studies, on-line evidence could uncover differences between referent preferences in off-line and online anaphora resolution (see Sekerina et al. 2004), especially since discrepancies between implicit and explicit knowledge have been observed. In order to delve deeper into the issue of crosslinguistic influence at the interface 162 level, it would be necessary to test L1 Basque-Spanish and L2 Basque- Spanish bilingual children in both Basque and Spanish. For such research, the collection of data from L2 Basque-Spanish adults would be necessary as a control group for cL2. 163 References Abaitua J. (1994). Dependencias locales y anáforas vacías en euskara. International Journal of Basque Linguistics and Philology (ASJU) XXVIII: 837-670. Abney, S. (1987). The English noun phrase in its sentential aspect. Unpublished doctoral dissertation. MIT, Massachusetts. Abrahamsson, N. & Hyltenstam, K. (2009). Age of acquisition and nativelikeness in a second language - listener perception vs. linguistic scrutiny. Language Learning 59: 249-306. Aduriz, I., Ceberio, K. & Díaz de Ilarraza, A. (2006). Pronominal anafora in Basque: annotation of a real corpus. Actas del XXII congreso de la Sociedad Española para el Procesamiento del Lenguaje Natural (SPLN). Zaragoza, Spain. Albizu, P. (2002). Basque verbal morphology: redefining cases. In X. Artiagoitia, P. Goenaga & J. Lakarra (eds.), Erramu Boneta: Festschrift for Rudolf P. G. de Rijk, pp. 1- 20. Bilbao: Universidad del País Vasco. Alonso-Ovalle, L., Fernandez-Solera, S., Frazier, L. & Clifton, C. (2002). Null versus overt pronouns and the topic-focus articulation in Spanish. Rivista Linguistica/ Italian Journal of Linguistics 14: 151-170. Anderson, S. R. & Keenan, E. L. (1985). Deixis. In T. Shopen (ed.), Language typology and syntactic description III: grammatical categories and the lexicon, pp. 259-308. Cambridge: Cambridge University Press. Aoshima S, Yoshida, M. & Phillips, C. (2009). Incremental processing of coreference and binding in Japanese. Syntax 12: 93-134. Argyri, E. & Sorace, A. (2007). Crosslinguistic influence and language dominance in older bilingual children. Bilingualism: Language and Cognition 10: 77-99. Ariel, M. (1990). Accessing noun phrase antecedents. London: Routledge. Ariel, M. (2001). Accessibility theory: An overview. In T. J.M. Sanders, J. Schilperoord & W. Spooren (eds.), Text representation: linguistic and psycholinguistic aspects, pp. 29- 87. Amsterdam: John Benjamins. Arregi, K. (2003). Focus in Basque movements. Unpublished doctoral dissertation. MIT, Massachusetts. Artiagoitia, X. (2003). Complementation (noun clauses). In J.I. Hualde & J. Ortiz de Urbina (eds.), A grammar of Basque, pp. 634-710. Berlin/ New York: Mouton de Gruyter. Avrutin, S. & Wexler, K. (1992). Development of principle B in Russian: coindexation at LF and coreference. Language Acquisition 2: 259-306. Baauw, S. & Cuetos, F. (2003). The interpretation of pronouns in Spanish language acquisition and breakdown: Evidence for the “Principle B Delay” as a non-unitary phenomenon. Language Acquisition 11: 219-275. Baauw, S., Escobar, M. & Philip, W. (1997). A delay of principle B effect in Spanish speaking children: The role of lexical feature acquisition. In A. Sorace, C. Heycock, & R. Shillcock (eds.), Proceedings of the GALA 97 Conference on Language Acquisition, pp. 16-21. Edinburgh: University of Edinburgh. Baauw, S., Zuckerman, S., Ruigendijk, M.E. & Avrutin, S. (2011). Principle B delays as a processing problem: evidence from task effects. In A. Grimm, A. Muller, C. 164 Hamann & M.E. Ruigendijk (eds.), Production-comprehension asymmetries in child language, pp. 247-272. Berlin: De Gruyter Mouton. Baayen, R. H. (2008). Analyzing linguistic data. A practical introduction to statistics using R. Cambridge: Cambridge University Press. Barquin, E. & Costa, A. (2011). Anaphora resolution in bilingual children: beyond cross-linguistic influence. Paper presented at International Symposium of Bilingualism (ISB8), Oslo, Norway. Barreña, A., Ezeizabarrena, M.J. & García, I. (2008). Influence of the linguistic environment on the development of the lexicon and grammar of Basque bilingual children. In C. Pérez-Vidal, M. Juan-Garau & A. Bel (eds.), A portrait of the young in the new multilingual Spain. Issues in the acquisition of two or more languages in multilingual environments, pp. 146-169. Londres: Multilingual Matters. Basque Government. (2008). Fourth sociolinguistic survey 2006: Basque Autonomous Community, Northern Basque Country, Navarre, Basque Country. Basque Government: Vitoria-Gasteiz. Basque Language Law (1986). Foral law on the Basque language. No 18/ 86, Parliament of Navarre. Official Gazette of Navarre. Bel, A. (2003).The syntax of subjects in the acquisition of Spanish and Catalan. Probus 15: 1-23. Bel, A., Perera, J. & Salas, N. (2010). Anaphoric devices in written and spoken narrative discourse: Data from Catalan. Written Language and Literacy 13: 236-259. Belletti, A. (2000). ‘Inversion’ as focalization. In A. Hulk and J.-Y. Pollock (eds.), Subject inversion in Romance and the theory of Universal Grammar, pp. 60-90. Oxford: Oxford University Press. Belletti, A. (2003). Aspects of the lower IP area. In L. Rizzi (ed.), The structure of CP and IP. The cartography of syntactic structures, Vol. 2, pp. 16-51. Oxford: Oxford University Press. Belletti, A., Bennati, E. & Sorace, A. (2007). Theoretical and developmental issues in the syntax of subjects: evidence from near-native Italian. Natural Language and Linguistic Theory 25: 657-689. Benveniste, E. (1966). Problèmes de linguistique générale I. Paris: Gallimard Bever,T.G. & Townsend, D.J. (1979). Perceptual mechanisms and formal properties of main and subordinate clauses. In W.E. Cooper & E.C.T Walker (eds.), Sentence processing, pp. 159-226. Hillsdale, N.J: Lawrence Earlbaum Associates. Bhat, D. N. S. (2004). Pronouns. Oxford: Oxford University Press Bini, M. (1993). La adquisición del italiano: más allá de las propiedades sintácticas del parámetro pro-drop. In J. Liceras (ed.), La lingüística y el análisis de los sistemas no nativos, pp. 126-139. Ottawa: Dovehouse. Bittner, D., Kuehnast, M. & Gagarina, N. (2008). Resolution of intersentential pronouns in German, Russian, and Bulgarian. Paper presented at at the Intersentential Pronominal Reference in L1 Acquisition Symposium, XI Congress of the International Association for the Study of Child Language (IASCL), Edinburgh, UK. Blackwell, S. (2003). Implicatures in discourse: The case of Spanish NP anaphora. Amsterdam: John Benjamins. Bley-Vroman, R. & Chaudron, C. (1990). Review essay: second language processing of subordinate clauses and anaphora first language and universal influences: a review of Flynn's research. Language Learning 40: 245-285. 165 Blom, E. (2008). Testing the domain-by-age model: inflection and placement of Dutch verbs. In B. Haznedar & E. Gavruseva (eds.), Current trends in child second language acquisition: a generative perspective, pp. 271-300. Amsterdam/ Philadelphia: Benjamins. Bosch, P., Katz, G. & Umbach, C. (2007). The non-subject bias of German demonstrative pronouns. In M. Schwarz-Friesel, M. Consten & M. Knees (eds.), Anaphors in text: Cognitive, formal and applied approaches to anaphoric reference, pp. 145-164. Amsterdam: John Benjamins. Bosch, P., Rozario, T. & Zhao, Y. (2003). Demonstrative pronouns and personal pronouns. German der vs. er. Proceedings of the EACL2003, Workshop on the Computational Treatment of Anaphora, pp. 61-68. Budapest, Hungary. Available online: http: / / aclweb.org/ anthology-new/ W/ W03/ W03-2609.pdf. Bosch, P. & Umbach, C. (2007). Reference determination for demonstrative pronouns. In D. Bittner & N. Gargarina (eds.), Intersentential pronominal reference in child and adult language: Proceedings of the Conference on Intersentential Pronominal Reference in Child and Adult Language (ZAS Papers in Linguistics 48), pp. 39-51. Berlin: ZAS. Available online: http: / / aclweb.org/ anthology-new/ W/ W03/ W03-2609.pdf. Brucart, J. M. (1987). La elisión sintáctica en español. Bellaterra: Publicacions de la Universitat Autónoma de Barcelona. Bühler, K. (1934). Sprachtheorie. Die Darstellungsfunktion der Sprache. Jena: Verlag von Gustav Fischer. Büring, D. (2005). Binding theory. Cambridge: Cambridge University Press. Burkhardt, P. (2005). The syntax-discourse interface. Representing and interpreting dependency. Amsterdam: John Benjamins. Cameron, R. (1992). Pronominal and null subject variation in Spanish: Constraints, dialects, and functional compensation. Unpublished doctoral dissertation. University of Pennsylvania, USA. Cardinaletti, A. (1997). Subjects and clause structure. In L. Haegeman (ed.), New comparative syntax, pp. 33-63. London: Longman Linguistics Library. Cardinaletti, A. & Starke, M. (1995). The tripartition of pronouns and its acquisition: Principle B puzzles are ambiguity problems. In J. Beckman (ed.), Proceedings of the 25th North East Linguistic Society (NELS), Volume Two: Papers from the Workshop on Language Acquisition & Language Change, pp. 1-12. Amherst, MA: GLSA. Cardinaletti, A. & Starke, M. (1999a). The typology of structural deficiency: A case study of the three classes of pronouns. In H. van Riemsdijk (ed.), Clitics in the languages of Europe, pp. 145-233. Berlin/ New York: Mouton de Gruyter. Cardinaletti, A. & Starke, M. (1999b). Responses and demonstratives. In H. van Riemsdijk (ed.) Clitics in the languages of Europe, pp. 280-289. Berlin/ New York: Mouton de Gruyter. Carminati, M. N. (2002). The processing of Italian subject pronouns. Unpublished doctoral dissertation. University of Massachusetts at Amherst, USA. Carminati, M. N. (2005). Processing reflexes of feature hierarch (person>number>gender) and implications for linguistic theory. Lingua 115: 259-285. Cenoz, J. (2009). Towards multilingual education: Basque educational research from an international perspective. Bristol: Multilingual Matters. Chafe, W. (1974). Language and consciousness. Language 50: 111-133. Chambers, G. C. & Smyth, R. (1998). Structural parallelism and discourse coherence: A test of Centering Theory. Journal of Memory and Language 39: 593-608. 166 Chien, Y.-C. & Wexler, K. (1990). Children’s knowledge of locality conditions in binding as evidence for the modularity of syntax and pragmatics. Language Acquisition 1: 225-295. Chomsky, C. S. (1969). The acquisition of syntax in children from 5 to 10. Cambridge, MA: MIT Press. Chomsky, N. (1980). Rules and representations. Oxford: Basil Blackwell. Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1982). Some concepts and consequences of Government and Binding theory. Cambridge, MA: MIT Press. Chomsky, N. (1986). Barriers. Cambridge, MA: MIT Press. Chomsky, N. (1993). A minimalist program for linguistic theory. In K. Hale & S. J. Keyser (eds.), The view from Building 20, pp. 1-52. Cambridge, MA: MIT Press. Chomsky, N. (1995). The minimalist program. Cambridge, MA: MIT Press. Clahsen, H. (1991). Constraints on parameter setting. A grammatical analysis of some acquisition stages in German child language. Language Acquisition 1: 361-391. Clark, H. H. & Clark, E.V. (1968). Semantic distinctions and memory for complex sentences. Quarterly Journal of Experimental Psychology 20: 129-138. Comrie, B. (2000). Pragmatic binding: demonstratives as anaphors in Dutch. In M. L. Juge and J. L. Moxley (eds.), Proceedings of the Twenty-Third Annual Meeting of the Berkeley Linguistics Society, pp. 50-61. Berkeley, CA: Berkeley Linguistics Society. Cowart, W. & Cairns, H. S. (1987). Evidence for an anaphoric mechanism within syntactic processing: some reference relations defy semantic and pragmatic constraints. Memory and Cognition 15: 318-331. Crain, S. & Fodor, J. D. (1985). How can grammars help parsers. In D. Dowty, D. Kartunnen & A. M. Zwicky (eds.), Natural language parsing: Psycholinguistic, computational, and theoretical perspectives, pp. 94-128. Cambridge: Cambridge University Press. Crain, S. & McKee, C. (1985). The acquisition of structural restrictions on anaphora. In S. Berman, J-W. Choe & J. McDonough (eds.), Proceedings of the North East Linguistic Society (NELS) 16, pp. 94-110. Amherst, MA: GLSA. De Houwer, A. (1990). The acquisition of two languages from birth: A case study. Cambridge: Cambridge University Press. De Weck, G. (1991). La cohésion dans les textes d’enfants. Paris: Delachaux & Niestlé. Diaconescu, R. & Goodluck, H. (2004). The pronoun attraction effect for d(iscourse)linked phrases: Evidence from speakers of a null subject language. Journal of Psycholinguistic Research 33: 303-319. Diessel, H. (1999). Demonstratives: Form, function, and grammaticalization. Amsterdam: John Benjamins. Diessel, H. (2004). The acquisition of complex sentences. Cambridge: Cambridge University Press. Dixon, R.M.W. (2010). Basic linguistic theory volume 2: Grammatical topics. Oxford: Oxford University. Döpke, S. (2000). Generation of and retraction from cross-linguistically motivated structures in bilingual first language acquisition. Bilingualism: Language and Cognition 3: 209-26. Domínguez, L. (2013). Understanding interfaces: second language acquisition and native language attrition of Spanish subject realization and word order variation. Amsterdam: John Benjamins. 167 Domínguez, L. & Arche, M. J. (2008). Optionality in L2 grammars: the acquisition of SV/ VS contrast in Spanish. In H. Chan, H. Jacob & E. Kapia (eds.), Proceedings of the 32nd Annual Boston University Conference on Language Development, pp. 96-107. Somerville, MA: Cascadilla. Duguine, M. (2010). Argumentu isilak eta hizkuntz aldakortasuna [Silent arguments and linguistic variation]. Bilbao, Spain: Udako Euskal Unibertsitatea. Eguzkitza, A. (1986). Topics in the syntax of Basque and Romance. Doctoral dissertation. UCLA, USA. Elordieta, A. (2001). The role of verbal agreement in licensing null arguments. In B. Fernández & P. Albizu (eds.), Kasu eta komunztaduraren gainean/ On case and agreement, pp. 113-132. Bilbao, Spain: Universidad del País Vasco. Escobar, L. & Gavarró, A. (1999). The acquisition of Catalan clitics and its implications for complex verb structure. Report GGT-99-3, Grup de Gramàtica Teòrica. Universitat Autònoma de Barcelona. Etxepare, R. (2003). Syntax: Valency and the auxiliary system: an outline. In J.I. Hualde & J. Ortiz de Urbina (eds.), A grammar of Basque, pp. 363-425. Berlin/ New York: Mouton de Gruyter. Eubank, L. & Gregg, K. R. (1999). Critical periods and (second) language acquisition: Divide et impera. In D. Birdsong (ed.), Second language acquisition and the critical period hypothesis, pp. 65-99. Mahwah, NJ: Lawrence Erlbaum Associates. Euskaltzaindia (Royal Academy of the Basque Language) (1991). Euskal gramatika. Lehen urratsak I (EGLU I). Bilbo: Euskaltzaindia. Ezeizabarrena, M.J. (2003). Null subjects and optional infinitives in Basque. In N. Müller (ed.), (In)vulnerable domains in multilingualism, (Hamburg series on multilingualism), pp. 83-106. Amsterdam: Benjamins. Ezeizabarrena, M.J. (2012). The (in)consistent ergative marking in early Basque: L1 vs. child L2. Lingua 122: 303-317. Felix, S. (1991). The accessibility of Universal Grammar in second language acquisition. In L. Eubank (ed.), Point Counterpoint: Universal Grammar in the second language, pp. 89-104. Amsterdam: John Benjamins. Felser, C., Roberts, L., Marinis, T. & Gross, R. (2003). The processing of ambiguous sentences by first and second language learners of English. Applied Psycholinguistics 24: 453-489. Fernández Soriano, O. (1999). El pronombre personal. Formas y distribuciones. Pronombres átonos y tónicos [The personal pronoun. Forms and distributions. Stressed and unstressed pronouns]. In I. Bosque & V. Demonte (eds.), Gramática descriptiva de la lengua española, pp. 1209-1273. Madrid, Spain: Espasa. Filiaci, F., Sorace, A. & Carreiras, M. (2014). Anaphoric biases of null and overt subjects in Italian and Spanish: a cross-linguistic comparison. Language and Cognitive Processes 29: 825-843 Flynn, S. (1983). A study of the effects of principal branching direction in second language acquisition: The generalization of a parameter of universal grammar from first to second language acquisition. Unpublished doctoral dissertation. Cornell University, New York. Flynn, S. (1986). Parameter-setting in L2 acquisition: Production vs. comprehension. Studies in Second Language Acquisition 8: 135-164. 168 Flynn, S. (1987). A parameter-setting model of L2 acquisition: Experimental studies in anaphora. Dordrecht, The Netherlands: D. Reidel. Flynn, S. & Espinal, I. (1985). Head-initialhead-final parameter in adult Chinese L2 acquisition of English. Second Language Research 1: 93-117. Fox, D. (1998). Locality in variable binding. In P. Barbosa, D. Fox., P. Hagstrom, M. McGinnis, D. Pesetsky (eds.), Is the best good enough? , pp. 129-156. Cambridge, MA: MIT Press and MITWPL. Frank, R., Hagstrom, P. & Vijay-Shanker, K. (2002). Roots, constituents and c- Command. In A. Alexiadou (ed.), Theoretical approaches to universals, pp. 109-137. Amsterdam: John Benjamins. Frenck-Mestre, C. & Pynte, J. (1997). Syntactic ambiguity resolution while reading in second and native languages. The Quarterly Journal of Experimental Psychology 50: 119-148. Fukumura, K. & Hyönä, J. (2011). Two of the same gender compete: evidence from reference production. Paper presented at the 24th Annual Conference on Human Sentence Processing, Stanford, California, USA. Fukumura, K., van Gompel, R.P.G. & Pickering, M. J. (2010). The use of visual context during the production of referring expressions. Quarterly Journal of Experimental Psychology 63: 1700-1715. García-Alcaraz, E. & Bel, A. (2011). Selección y distribución de los pronombres en el español L2 de los hablantes de árabe. Revista de Lingüística y Lenguas Aplicadas 6: 165-179. Garcia-Azkoaga, I. M. (2003). Kohesio anaforikoa hiru testu generotan. Adinaren araberako garapenaren azterketa [Anaphoric cohesion in three text genres. A study on the age development]. Bilbo: Euskal Herriko Unibertsitatea. Garcia-Azkoaga, I. M. & Idiazabal, I. (2004). Los mecanismos de cohesión nominal en las narraciones de los alumnos bilingües. In A.M Lorenzo, F. Ramallo & X. P. Rodríguez-Yánes (eds.), Bilingual sozialization and bilingual language acquisition: Proceedings II Simposio Internacional Bilingüismo, pp. 571-580. Vigo: Servizo de Publicacións da Universidade de Vigo. Garnham, A., Oakhill, J. & Cain, K. (1998). Selective retention of information about the superficial for of text: ellipses with antecedents in main and subordinate clauses. Quarterly Journal of Experimental Psychology 51: 19-39. Garzia, J. (1996). Hura, bera eta abarren adar gehiago [On hura, bera etc.] SENEZ. 18. EIZIE: 81-112. Gathercole, V. C. M. (2002a). Command of the mass/ count distinction in bilingual and monolingual children: An English morphosyntactic distinction. In D. Kimbrough Oller, & Rebecca E. Eilers (eds.), Language and literacy in bilingual children, pp. 175-206. Clevedon: Multilingual Matters. Gathercole, V. C. M. (2002b). Grammatical gender in bilingual and monolingual children: A Spanish morphosyntactic distinction. In D. Kimbrough Oller, & Rebecca E. Eilers (eds.), Language and literacy in bilingual children, pp. 207-219. Clevedon: Multilingual Matters. Gathercole, V. C. M. (2002c). Monolingual and bilingual acquisition: Learning different treatments of thattrace phenomena in English and Spanish. In D. Kimbrough Oller, & Rebecca E. Eilers (eds.), Language and literacy in bilingual children, pp. 220- 254. Clevedon: Multilingual Matters. 169 Geber, D. (2006). Processing subject pronouns in relation to non-canonical (quirky) constructions. Ottawa Papers in Linguistics 34: 47-61. Genesee, F. & Nicoladis, E. (2007). Bilingual first language acquisition. In Erika Hoff, & Peggy McCardle (eds.), Handbook of language development, pp. 324-342. Oxford: Blackwell. Genesee, F., Nicoladis, E. & Paradis, J. (1995). Language differentiation in early bilingual development. Journal of Child Language 22: 611-631. Gibson, E. (1998). Linguistic complexity: locality of syntactic dependencies. Cognition 68: 1-76. Givón, T. (1983). Topic continuity in discourse: An introduction. In T. Givón (ed.), Topic continuity in discourse. A quantitative cross-language study, pp. 11-42. Amsterdam/ Philadelphia: John Benjamins. Givón, T. (1984). Syntax: A functional-typological introduction. Volume I. Amsterdam: John Benjamins. Goenaga, P. (1984). Euskal sintaxia: Konplementazioa eta nominalizazioa. [Basque syntax: Complementation and nominalization]. Unpublished doctoral dissertation. Universidad del País Vasco, Bilbao, Spain. Gómez, R. & Sainz, K. (1995). On the origin of the finite forms of the Basque verb. In J. I. Hualde, J. A. Lakarra & R. L. Trask (eds.), Towards a history of the Basque language, pp. 235-274. Amsterdam: John Benjamins. Goodluck, H. (1978). Linguistic principles in children’s grammar of complement interpretation. Unpublished doctoral dissertation. University of Massachusetts at Amherst, USA. Goodluck, H. (1991). Language acquisition: A linguistic introduction. Malden, MA: Blackwell Publishing Ltd. Gordon, P. C. & Hendrick, R. (1997). Intuitive knowledge of linguistic co-reference. Cognition 62: 325-370. Government of Navarre (2008). The situation of Basque in Navarre. Granfeldt, J., Schlyter, S. & Kihlstedt, M. (2007). French as cL2, 2L1 and L1 in preschool children. PERLES Lund University 24: 7-42. Green, G. M. (1989). Pragmatics and natural language understanding. Hillsdale, N.J.: L. Erlbaum Associates. Grimshaw, J. & Rosen, S. R. (1990). Knowledge and obedience: The developmental status of the binding theory. Linguistic Inquiry 21: 87-222. Grodzinsky Y. & Reinhart, T. (1993). The innateness of binding and of coreference. Linguistic Inquiry 24: 69-101. Grondin, N. & White, L. (1996). Functional categories in child L2 acquisition of French. Language Acquisition 5: 1-34. Guasti, M.T. (2004). Language acquisition: The growth of grammar. Cambridge, MA: MIT Press. Guasti, M. T. & Chierchia, G. (1999/ 2000). Reconstruction in child grammar. Language Acquisition 8: 129-170. Guido Mendes, C. & Iribarren, I. C. (2007). Fixaç-o do parâmetro do sujeito nulo na aquisiç-o do português europeu por hispanofalantes. In Associaç-o Portuguesa de Linguística (eds.), XXII encontro nacional da associaç-o portuguesa de linguística: Textos seleccionados, pp. 483-497. Lisbon: Associaç-o Portuguesa de Lingüística. 170 Hacohen A. & Schaeffer, J. (2007). Subject realization in early Hebrew/ English bilingual acquisition: The role of crosslinguistic influence. Bilingualism: Language and Cognition 10: 333-344. Haegeman, L. (1994). Introduction to Government and Binding theory. Oxford: Blackwell, 2nd edn. Hahne, A. & Friederici, A. D. (2001). Processing a second language: Late learners' comprehension mechanisms as revealed by event-related brain potentials. Bilingualism: Language and Cognition 4: 123-141. Halliday, M. A. K. & Hasan, R. (1976). Cohesion in English. London: Longman. Hamann, C. (2002). From syntax to discourse. Pronominal clitics, null subjects and infinitives in child language. Boston: Kluwer Academic Publishers. Hamann, C., Kowalski, O. & Philip, W. (1997). The French ‘Delay of Principle B Effect’. In E. Hughes, M. Hughes, & A. Greenhill (eds.), Proceedings of the 21st Boston University Conference on Language Development, pp. 205-219. Somerville, MA: Cascadilla Press. Harada, T. (2006). The acquisition of single and geminate consonants by Englishspeaking children in a Japanese immersion program. Studies in Second Language Acquisition 28: 601-632. Harbert, W. (1995). Binding theory, control and pro. In G. Webelbuth (ed.), Government and binding theory and minimalist program, pp. 177-240. Oxford: Blackwell. Haznedar, B. (2010). Transfer at the syntax-pragmatics interface: Pronominal subjects in bilingual Turkish. Second Language Research 26: 355-378. Haznedar, B. & Gavruseva, E. (eds.) (2008). Current trends in child second language acquisition: a generative perspective. John Benjamins: Amsterdam. Heine, B. & Kuteva, T. (2006). The changing languages of Europe. Oxford: Oxford University Press. Hertel, T. (2003). Lexical and discourse factors in the second language acquisition of Spanish word order. Second Language Research 19: 273-304. Hickmann, M. (1980). Creating referents in discourse. A developmental analysis of linguistic cohesion. In J., Kreiman & A.E., Ojeda (eds.), Bibliografia 317 Papers from the parasession on pronouns and anaphora, pp. 192-203. Chicago: Chicago Linguistic Society. Hickmann, M. (1984). Fonction et contexte dans le développement du langage. Langage et communication à l’âge préscolaire, Presses Universitaires de Rennes 2: 27-57. Hickmann, M. (1987). The pragmatics of reference in child language. Some issues in developmental theory. In M. Hickmann (ed.), Social and functional approaches to language and thought. pp. 165-184. Orlando: Academic Press. Hicks, G. (2009). The derivation of anaphoric relations. Amsterdam: John Benjamins. Hill, S. & Bradford, W. (2000). Bilingual grammar of English-Spanish syntax: A manual with exercises and key (Revised ed.). Lanham, MA: University Press of America. Himmelmann, N. (1996). Demonstratives in narrative discourse: A taxonomy of universal uses. In Barbara Fox (ed.), Studies in anaphora, pp. 205-254. Amsterdam: John Benjamins. Hinzelin, M. O. (2003). The acquisition of subjects in bilingual children: Pronoun use in Portuguese-German children. In N. Müller (ed.). (In)vulnerable domains in multilingualism, (Hamburg Series on multilingualism), pp. 107-138. Amsterdam: Benjamins. Hornstein, N., Nunes, J. & Grohmann, K. (2005). Understanding minimalism. Cambridge: Cambridge University Press. 171 Hualde, J.I. (2003). Introduction. In J. I. Hualde & J. Ortiz de Urbina (eds.), A grammar of Basque. pp. 1-13. Berlin: Mouton de Gruyter. Hualde, J.I., Oyharçabal, B. & Ortiz de Urbina, J. (2003). Verbs. In J. I. Hualde & J. Ortiz de Urbina (eds.), A grammar of Basque, pp. 195-246. Berlin: Mouton de Gruyter. Huang, C.-T. J. (1984). On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531-574. Huang, Y. (1994). The syntax and pragmatics of anaphora. Cambridge: Cambridge University Press. Huang, Y. (2000). Anaphora: A cross-linguistic study. Oxford: Oxford University Press. Hulk, A. C. J. & Müller, N. (2000). Bilingual first language acquisition at the interface between syntax and pragmatics. Bilingualism: Language and Cognition 3: 227-244. Hulk, A. C. J. & van der Linden, E. (1996). Language mixing in a French-Dutch bilingual child. In E. Kellerman, B. Weltens & T. Bongaerts (eds.), EUROSLA 6. A selection of papers (special issue), pp. 89-103. Amsterdam: VU, Uitvegerij. Hyltenstam, K. & Abrahamsson, N. (2003). Maturational constraints in SLA. In C. J. Doughty & M. H. Long (eds.), The Handbook of Second Language Acquisition, pp. 539- 588. Oxford: Blackwell. Ikastolen elkartea (2002). Irakurgaiak 1 [Readings 1]. Donostia: Elkar. Iraola, M. (2014). Anaphora resolution in Basque: An experimental study on the interpretation of null and overt subject pronouns by children and adults. Unpublished doctoral dissertation. University of the Basque Country and University of Konstanz. Iraola, M. & Ezeizabarrena, M.J. (2011). Anaphora resolution in Basque: Null vs. overt subject hura. In N. Danis, K. Mesh & H. Sung (eds.) BUCLD 35 Online Proceedings Supplement. Available online: http: / / www.bu.edu/ bucld/ files/ 2011/ 05/ 35 Iraola.pdf Iraola, M. & Ezeizabarrena, M. J. (2012). La resolución anafórica en la adquisición del euskera. Actas del XXIX Congreso Internacional AESLA, pp. 65-76. Salamanca: Universidad de Salamanca. Iraola, M., Santesteban, M. & Ezeizabarrena, M. J. (2014). The discursive feature specification of the pronoun hura in child (L1 and L2) and adult Basque. Selected Proceedings of the Romance Turn V. Workshop on the Acquisition of Romance Languages, pp. 134-157. Newcastle: Cambridge Scholars Publishing. Irigoyen, I. (1981). Haur ola zirola. Elementos deícticos en la lengua vasca. Iker 1, pp. 365-402. Bilbo: Euskaltzaindia. Irurtzun, A. (2006). Focus and clause structuration in the minimalist program. In C. Boeckx (ed.), Minimalist essays, pp. 68-96. Amsterdam: John Benjamins. Ivanov, I. (2009). Topicality and Clitic Doubling in L2 Bulgarian: A Test Case for the Interface Hypothesis. In M. Bowles, T. Ionin, S. Montrul & A. Tremblay (eds.), Proceedings of the 10th Generative Approaches to Second Language Acquisition Conference (GASLA 2009), pp. 17-24. Somerville, MA: Cascadilla Proceedings Project. Iverson, M., Kempchinsky, P. & Rothman, J. (2008). Interface vulnerability and knowledge of the subjunctive/ indicative distinction with negated epistemic predicates in L2 Spanish. In L. Roberts, F. Myles & A. David (eds.), EUROSLA Yearbook 8, pp. 135-163. Amsterdam: John Benjamins. 172 Jaeger, F. 2008. Categorical data analysis: Away from transformations (transformations or not) and towards logit mixed models. Journal of Memory and Language 59: 434-446. Jaeggli, O. & Safir, K. (1989). The null subject parameter and parametric theory. In O. Jaeggli & K. Safir (eds.), The null subject parameter, pp.1-44. Dordrecht: Kluwer Academic Publishers. Jakubowicz, C. (1984). On markedness and binding principles. In C. Jones, & P. Sells (eds.), Proceedings of the 14th Annual Meeting of the North East Linguistic Society (NELS), pp. 154-182. Amherst, MA: GLSA. Jarvella, R. J. & Herman, S. J. (1972). Clause structure of sentences and speech processing. Perception and Psychophysics 11: 381-384. Juffs, A. (2006). Grammar and parsing and a transition theory. Applied Psycholinguistics 27: 66-69. Kaiser, E. (2011). Salience and contrast effects in reference resolution: The interpretation of Dutch pronouns and demonstratives. Language and Cognitive Processes 26: 1587-1624. Kaiser, E. & Trueswell, J. C. (2008). Interpreting pronouns and demonstratives in Finnish: Evidence for a form-specific approach to reference resolution. Language and Cognitive Processes 23: 709-748. Kaiser, E. & Vihman, V. (2006). On the referential properties of Estonian pronouns and demonstratives. Paper presented at the 22nd Scandinavian Conference of Linguistics, Aalborg, University, Denmark. Karmiloff-Smith, A. (1981). The grammatical marking of thematic structure in the development of language production. In W. Deutsch (ed.), The child’s construction of language, pp. 121-147. New York: Academic Kato, M. A. (1999). Strong pronouns, weak pronominals and the null subject parameter. Probus 11: 1-37. Kazanina, N. & Phillips, C. (2009). Differential effects of constraints on the processing of Russian cataphora. Quarterly Journal of Experimental Psychology 63: 371-400. Kazanina, N., Lau, E., Lieberman, M., Yoshida, M. & Phillips, C. (2007). The effect of syntactic constraints on the processing of backwards anaphora. Journal of Memory and Language 56: 384-409. Kibrik, A. A. (1996). Anaphora in Russian narrative prose: A cognitive account. In B. Fox (ed.), Studies in anaphora, pp. 255-304. Amsterdam/ Philadelphia: John Benjamins. Kibrik, A. A. (2011). Reference in discourse. Oxford: Oxford University Press. Klein, W. (1986). Second language acquisition. Cambridge: Cambridge University Press. König, E. & Siemund, P. (2000). Intensifiers and reflexives: A typological perspective. In Z. Frajzyngier & T. Curl (eds.), Reflexives: Forms and functions, pp. 41-74. Amsterdam: Benjamins. Krasavina, O. & Chiarcos, C. (2007). SPoCoS: Potsdam Coreference Scheme. In Proceedings of the First Linguistic Annotation Workshop (LAW). Held in conjunction with ACL-2007, pp. 156-163. Prague. Kuno, S. (1987). Functional syntax: anaphora, discourse, and empathy. Chicago: University of Chicago Press. Laka, I. (1996). A brief grammar of euskara, the Basque language, Ms., University of the Basque Country, Vitoria-Gasteiz. Available online: http: / / www.ehu.es/ grammar. 173 Lakoff, R. (1968). Abstract syntax and Latin complementation. Cambridge, MA: MIT Press. Lakshmanan, U. (1991). Morphological uniformity and null subjects in child second language acquisition. In Lynn Eubank (ed.), Point and Counterpoint: Universal Grammar in the Second Language, pp. 389-411. Amsterdam: John Benjamins. Lardiere, D. (2011). Who is the interface hypothesis about? Linguistic Approaches to Bilingualism 1: 48-53. Larrañaga, P. (2008). The subjects of unaccusative in bilingual Basque/ Spanish bilingual. In P. Guijarro-Fuentes, M. P. Larrañaga & J. Clibbens (eds.), First language acquisition of syntax and morphology, pp. 183-208. Amsterdam: John Benjamins. Lasnik, H. (1976). Remarks on coreference. Linguistic Analysis 2: 1-22. Leal Méndez, T., Rothman, J. & Slabakova, R. (2014). A rare structure at the syntaxdiscourse interface: Heritage and Spanish-dominant Native Speakers weigh in. Language Acquisition 21: 411-429. Leal Méndez, T., Rothman, J. & Slabakova, R. (2014). Discourse-sensitive Cliticdoubled Dislocations in Heritage Spanish. Lingua: 1-13. (doi: 10.1016/ j.lingua.2014.01.002). Lee, D. & Schachter, J. (1997). Sensitive period effects in binding theory. Language Acquisition 6: 333-362. Lenneberg, E. (1967). Biological foundations of language. New York: Wiley. Liceras, J. M., Fernández Fuertes, R., Alba de la Fuente, A., Boudreau G. & Acevedo. E, (2012). Interliguistic influence in simultaneous bilingualism: Core syntax phenomena and lexical transparency. In S. Ferré, Ph. Prévost, L. Tuller & R. Zebib (eds.), Selected Proceedings of the Romance Turn IV Workshop on the Acquisition of Romance Languages, pp. 214-239. Newcastle: Cambridge Scholar Publishing. Liceras, J. M., Fernández Fuertes, R. & Pérez-Tattam, R. (2008). Null and overt subjects in the developing grammars (L1 English/ L1 Spanish) of two bilingual twins. In C. Pérez, A. Bel & M. Juan (eds.), A portrait of the young in the new multilingual Spain, pp. 111-134. Clevedon, UK: Multilingual Matters. Lozano, C. (2006). The development of the syntax-information structure interface: Greek learners of Spanish. In V. Torrens & L. Escobar (eds.), The acquisition of syntax in Romance languages, pp. 371-399. Amsterdam: John Benjamins. Lozano, C. (2009). Selective deficits at the syntax-discourse interface: Evidence from the CEDEL2 corpus. In N. Snape, Y.I. Leung & M. Sharwood-Smith (eds), Representational Deficits in SLA, pp. 127-166. Amsterdam: John Benjamins. Luján, M. (1985). Overt pronouns in Spanish. Chicago Linguistic Society Proceedings 21: 424-438. Luján, M. (1986). Stress and binding of pronouns. Chicago Linguistic Society Proceedings 22: 248-262. Lust, B. (1981). Constraints on anaphora in child language: A prediction for a universal. In S. L. Tavalokian (ed.), Language acquisition and linguistic theory, pp. 74-96. Cambridge, MA: MIT Press. Lust, B. (1983). On the Notion Principal Branching Direction: A Parameter of Universal Grammar. In Y. Otsu, H. Van Riemsdijk, K. Inoue, A. Kamio & N. Kawasaki (eds.), Studies in generative grammar and language acquisition, pp. 137-151. Japan: Tokyo Gakugei University. 174 Lust, B. (1986). Studies in the acquisition of anaphora. Volume 1: Applying the constraints. Dordrecht, [Holland]; Boston : Hingham, MA, U.S.A: D. Reidel Pub. Co. Lust, B. (2006). Child language: Acquisition and growth. Cambridge: Cambridge University Press. Lust, B. & Chien, Y-C. (1984). The structure of coordination in first language acquisition of Chinese. Cognition 17: 49-83. Lust, B., Eisele, J. & Mazuka, R. (1992). The binding theory module: Evidence from first language acquisition for principle C. Language 68: 333-358. Lust, B. & Mangione, L. (1983). The principal branching direction parameter in first language acquisition of anaphora. In P. Sells & C. Jones (eds.), Proceedings of the 13th Annual Meeting of the Northeastern Linguistic Society (NELS), pp. 145-160. Amherst, MA: University of Massachusetts. Lust, B., Wakayama, T., Hiraide, H., Snyder, W. & Bergmann, M. (1983). Comparative studies on the first language acquisition of Japanese and English - Language Universal and Language-specific Constraints. In S. Hattori & K. Inoue (eds.), Proceedings of the XIIIth International Congress of Linguists, pp. 1131-1135. Tokyo, Japan. Lyons, J. (1977). Deixis and anaphora. In T. Myers (ed.), The development of conversation and discourse. Edinburgh: University of Edinburgh. Manterola, I. & Almgrem, M. (2013). Las grandes líneas de progresión de los aprendizajes lingüísticos en la escolarización en euskera. In J. Dolz & I. Idiazabal (eds.), Enseñar lenguas en contextos multilingües, pp. 217-240. Bilbao: Universidad del País Vasco. Margaza P. & Bel, A. (2006). Null subjects at the syntax-pragmatics interface: Evidence from Spanish interlanguage of Greek. Proceeding of 8 th Generative Approaches to Second Language Acquisition Conference (GASLA 2006), pp. 88-97. Somerville, MA: Cascadilla Proceedings Project. Marinis, T., Roberts, L., Felser, C. & Clahsen, H. (2005). Gaps in second language sentence processing. Studies in Second Language Acquisition 27: 53-78. Martínez-Areta, M. (2013). Demonstratives and personal pronouns. In M. Martínez- Areta (ed.), Basque and Proto-Basque Language-Internal and Typological Approaches to Linguistic Reconstruction (Mikroglottka Minority Language Studies), pp. 283-321. Frankfurt am Main: Peter Lang GmbH. Mayol, L. (2009). Pronouns in Catalan: information, discourse and strategy. Unpublished doctoral dissertation. University of Pennsylvania, USA. McDaniel, D., Cairns, H. S. & Hsu, J. R. (1990). Binding principles in the grammars of young children. Language Acquisition 1: 121-138. McKee, C. (1992). A comparison of pronouns and anaphors in Italian and English acquisition. Language Acquisition 2: 21-54. Meisel, J. M. (1989). Early differentiation of languages in bilingual children. In K. Hyltenstam & L. Obler (eds.), Bilingualism across the lifespan: Aspects of acquisition, maturity and loss, pp. 13-40. Cambridge: Cambridge University Press. Meisel, J. M. (1994). Getting FAT: Finiteness, agreement and tense in early grammars. In J. M. Meisel (ed.), Bilingual first language acquisition: French and German grammatical development, pp. 89-129. Amsterdam, Netherlands: Benjamins. Meisel, J. M. (2007). Age of onset in successive acquisition of bilingualism: Effects on grammatical development. Working Papers in Multilingualism, 80, University of Hamburg. 175 Meisel J. M. (2008). Child Second Language Acquisition or successive First Language acquisition? ”. In B. Haznedar & E. Gavruseva (eds.), Current trends in child second language acquisition. A generative perspective, pp. 55-80. Amsterdam - Philadelphia: John Benjamins Publishing Company. Meisel, J. M. (2009). Second language acquisition in early childhood. Zeitschrift für Sprachwissenschaft 28: 5-34. Meisel, J. M. (2011). First and second language acquisition: Parallels and differences. Cambridge: Cambridge University Press. Meisel, J. M. (2013). Sensitive Phases in Successive Language Acquisition: The Critical Period Hypothesis Revisited. In C. Boeckx & K. Grohmann (eds.), Handbook of Biolinguistics, pp. 69-85. Cambridge: Cambridge University Press. Mörnsjö, M. (2002). V1 Declaratives in spoken Swedish. Unpublihed doctoral dissertation, Lund University, Sweden. Montrul, S. (2004). The acquisition of Spanish. Morphosyntactic development in monolingual and bilingual L1 acquisition and in adult L2 acquisition. [Series on Language Acquisition and Language Disorders]. Amsterdam: John Benjamins. Montrul, S. (2011). Multiple interfaces and incomplete acquisition. Lingua 121: 591- 604. Montrul, S. & Polinsky M. (2011). Why not heritage speakers? . Linguistic Approaches to Bilingualism 1: 58-62. Montrul, S. & Rodríguez-Louro, C. (2006). Beyond the syntax of the null subject parameter: a look at the discourse-pragmatic distribution of null an overt subjects by L2 learners of Spanish. In V. Torrens & L. Escobar (eds.), The acquisition of syntax in Romance Languages, pp. 401-418. Amsterdam: John Benjamins. Müller, N. & Hulk, A. (2001). Crosslinguistic influence in bilingual language acquisition: Italian and French as recipient languages. Bilingualism: Language and Cognition 4: 1-21. O’Grady, W. (1997). Syntactic development. Chicago: University of Chicago Press. O’Grady, W. (2011). Interfaces and processing. Linguistic Approaches to Bilingualism 1: 63-66. O’Grady, W., Cho, S. W. & Sato, Y. (1994). Anaphora and branching direction in Japanese. Journal of Child Language 21: 473-487. O’Grady, W., Suzuki-Wei, Y. & Cho, S. W. (1986). Directionality preferences in the interpretation of anaphora: Data from Korean and Japanese. Journal of Child Language 13: 409-420. Ok Kweon, S. (2011). Processing null and overt pronoun subject in ambiguous sentences in Korean. International Journal of Linguistics 3: 1-12. Okuma, T. (2012). L2 acquisition of discourse constraints on the use of Japanese pronouns. In A. Biller, E. Chung & A. Kimball (Eds.) The 36th Annual Boston University Conference on Language Development (BUCLD36) Online Proceedings Supplement. Available online: http: / / www.bu.edu/ bucld/ files/ 2012/ 07/ Okuma-36.pdf Ortiz de Urbina, J. (1989). Parameters in the grammar of Basque. Dordrecht: Foris. Oyharçabal, B. (1992). Structural case and inherent case marking: ergaccusativity in Basque. In J. Lakarra & J. Ortiz de Urbina (eds.), Syntactic theory and Basque syntax, pp. 309-342. Diputación Foral de Guipúzcoa, San Sebastian. Padilla, J. (1990). On the definition of binding domains in Spanish. Dordrecht: Kluwer. 176 Papadopoulou, D. & Clahsen, H. (2003). Parsing strategies in L1 and L2 sentence processing: A study of relative clause attachment in Greek. Studies in Second Language Acquisition 25: 501-528. Paradis, J. & Genesee, F. (1996). Syntactic acquisition: autonomous or interdependent? Studies in Second Language Acquisition 18: 1-25. Paradis, J. & Navarro, S. (2003). Subject realization and crosslinguistic interference in the bilingual acquisition of Spanish and English: what is the role of input? Journal of Child Language 30: 371-393. Penfield, W. & Roberts, L. (1959). Speech and Brain Mechanisms. Princeton, NJ: Princeton University Press. Peral, J. & Ferrández, A. (2003). Translation of pronominal anaphora between English and Spanish: Discrepancies and evaluation. Journal of Artificial Intelligence Research 18: 117-147. Pérez-Leroux, A.T. & Glass, W. R. (1999). Null anaphora in Spanish second language acquisition: probabilistic versus generative approaches. Second Language Research 15: 220-249. Petrirena, P. (1999). Euskara mintegia 98-99, Egunkaria. Philippaki-Warburton, I. (1987). The theory of empty categories and the pro-drop parameter in Modern Greek. Journal of Linguistics 23: 289-318. Pinto, M. (2006). Subject pronouns in bilinguals: interference or maturation. In L. Escobar & V. Torrens (eds.), The acquisition of syntax in Romance languages, pp. 331- 352. Amsterdam: Benjamins. Piñango, M., Burkhardt, P., Brun, P. & Avrutin, S. (2001). The architecture of the sentence processing system: The case of pronominal interpretation. Paper presented at Cognitive Science Annual Meeting, Edinburgh, UK. Pladevall, E. (2010). Child L2 development of syntactic and discourse properties of Spanish subjects. Bilingualism: Language and Cognition 13: 185-216. Rebuschi, G. (1995). Weak and strong genitive pronouns in northern Basque: A Diachronic perspective. In J. I. Hualde, J. A. Lakarra & R. L. Trask (eds.), Towards a history of the Basque language, pp. 313-356. Amsterdam: Benjamins. Rebuschi, G. (2003). Basque from a typological, dialectological and diachronic point of view. In Thorsten Roelcke (ed.), Variation typology. Variationstypologie, pp. 837-865. Berlin & New York: W. de Gruyter. Reinhart, T. (1976). The syntactic domain of anaphora. Unpublished doctoral dissertation. MIT, Cambridge, MA. Reinhart, T. (1983). Anaphora and semantic interpretation. London: Croom Helm. Reinhart, T. (1986). Center and periphery in the grammar of anaphora. In B. Lust (ed.), Studies in the acquisition of anaphora, pp. 123-150. Dordrecht: D. Reidel. Reuland, E. J. (1998). Structural conditions on chains and binding. In P.N. Tamanji & K. Kusumoto (eds.), Proceedings of the 28 th North Eastern Linguistic Society (NELS), pp. 341-356. Amherst, MA: GLSA. Rijk, R. P. G. de (2008). Standard Basque. A Progressive grammar. Cambridge, MA: MIT Press Rizzi, L. (1982). Issues in Italian syntax. Dordrecht: Foris. Rizzi, L. (1986). Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501- 557. Rothman, J. (2007). Pragmatic Solutions for Syntactic Problems: Understanding Some L2 Syntactic Errors in Terms of Pragmatic Deficits. In S., Baauw, F., Dirjkoningen & 177 M., Pinto (eds.). Romance languages and linguistic theory 2005, pp. 299-320. Amsterdam: John Benjamins. Rothman, J. (2009). Pragmatic deficits with syntactic consequences? : L2 pronominal subjects and the syntax-pragmatics interface. Journal of Pragmatics 41: 951-973. Saltarelli, M. (1988). Basque. London: Croom Helm. Sanoudaki, I. (2003). Greek 'Strong' pronouns and the Delay of Principle B Effect. Reading Working Papers in Linguistics 7: 103-124. Sauerland, U. & Gibson, E. (1998). Case matching in relative clause attachment. Paper presented at the 11th Annual CUNY Sentence Processing Conference, Rutgers University, New Brunswick, NJ. Schlyter, S. (2011). Tense and Aspect in early French development in aL2, 2L1 and cL2 learners. In Esther, R. & Tanja, K. (eds.), The development of grammar : language acquisition and diachronic change. Volume in honour of Jürgen M. Meisel [Hamburg Studies on Multilingualism, 11], pp. 47-74. John Benjamins Publishing. Schmitz, K. (2003). (In) vulnerable subjects? . Manuscript, University of Hamburg. Schwartz, B. D. (1992). Testing between UG-based and problem-solving models of L2A: Developmental sequence data. Language Acquisition 2: 1-19. Schwartz, B. D. (2004). On child development of syntax and morphology. Lingue e Linguaggio 3: 97-132. Schwartz, B. D. (2011). Parsing up the Interface Hypothesis. Linguistic Approaches to Bilingualism 1: 84-88. Sekerina, I., Stromswold, K. & Hestvik, A. (2004). How do adults and children process referentially ambiguous pronouns? . Journal of Child Language 31: 123-152. Seliger, W. (1978). Implications of a multiple critical period hypothesis for second language learning. In C. Ritchie (ed.), Second Language Acquisition: Issues and Implications, pp. 11-19. New York: Academic Press. Sells, P. (1987). Aspects of logophoricity. Linguistic Inquiry 18: 445-479. Serratrice, L. (2007). Cross-linguistic influence in the interpretation of anaphoric and cataphoric pronouns in English-Italian bilingual children. Bilingualism: Language and Cognition 10: 225-238. Serratrice, L., Sorace, A. & Paoli, S. (2004). Crosslinguistic influence at the syntaxpragmatics interface: Subjects and objects in English-Italian bilingual and monolingual acquisition. Bilingualism: Language and Cognition 7: 183-205. Sheldon, A. (1974). The role of parallel function in the acquisition of relative clauses in English. Journal of Verbal Learning and Verbal Behaviour 13: 272-281. Shin, N. L. & Cairns, H. S. (2012). The development of NP selection in school-age children: reference and Spanish subject pronouns. Language Acquisition 19: 3-38. Sigurjonsdottir, S., Hyams, N. & Chien, Y-C. (1988). The acquisition of reflexives and pronouns by Icelandic Children. Papers and Reports on Child Language Development 27: 97- 106. Slabakova, R. (2006). Learnability in the L2 acquisition of semantics: a bidirectional study of a semantic parameter. Second Language Research 22: 498-523. Slabakova, R. & García Mayo, M.P. (2013). The L3 syntax-discourse interface. Bilingualism: Language and Cognition (DOI: http: / / dx.doi.org/ 10.1017/ S1366728913000369). Slabakova, R. & Ivanov, I. (2011). A more careful look at the syntax-discourse interface. Lingua 121: 637-651. 178 Smyth, R. & Chambers, C. (1996). Parallelism effects on pronoun resolution in discourse contexts. In S. Botley, J. Glass, T. McEnery & A. Wislon (eds.), Approaches to Discourse Anaphora: Proceedings of the Discourse Anaphora and Anaphor Resolution Colloquium (DAARC'96), pp. 268-280. Lancaster University, UK. Solan, L. (1983). Pronominal reference: Child language and the theory of grammar. Dordrecht: D. Reidel. Solan, L. (1987). Parameter setting and the development of pronouns and reflexives. In T. Roeper & E. Williams (eds.), Parameter setting, pp. 189-210. Dordrecht, The Netherlands: Reidel. Sorace, A. (2000). Differential effects of attrition in the L1 syntax of L2 near-native speakers. In In S. C. Howell, S. A. Fish, and T. Keith-Lucas (eds.), Proceedings of the 24th Boston University Conference on Language Development (BUCLD24), pp. 719-725. Somerville, MA: Cascadilla Press. Sorace, A. (2006). Gradience and optionality in mature and developing grammars. In G. Fanselow, C. Fery, M. Schlesewsky & R. Vogel (eds.), Gradience in Grammars: Generative Perspectives, pp. 106-123. Oxford: Oxford University Press. Sorace, A. (2011). Pinning down the concept of 'interface' in bilingualism. Linguistic Approaches to Bilingualism 1: 1-33. Sorace, A. (2012). Pinning down the concept of 'interface' in bilingualism: a reply to peer commentaries. Linguistic Approaches to Bilingualism 2: 209-216. Sorace, A. & Filiaci, F. (2006). Anaphora resolution in near-native speakers of Italian. Second Language Research 22: 339-368. Sorace, A. & Keller, F. (2005). Gradience in linguistic data. Lingua 115: 1497-1524. Sorace, A. & Serratrice, L. (2009). Internal and external interfaces in bilingual language development: Beyond structural overlap. International Journal of Bilingualism 13: 195-210. Sorace, A., Serratrice, L., Filiaci, F., & Baldo, M. (2009). Discourse conditions of subject pronoun realization: Testing the linguistic intuitions of older bilingual children. Lingua 119: 460-477. Stowe, L. A. (1986). Parsing WH-constructions: evidence for on-line gap location. Language and Cognitive Processes 1: 227-245. Sturgeon. A. (2008). Topic and demonstrative pronouns in Czech. In G. Zybatow, L. Szuchsich, U. Junghans & R. Meyer (eds.), In Proceedings of the 5 th European Conference on Formal Description of Slavic Languages, pp. 514-525. Frankfurt and Main: Peter Lang Publishers. Tallerman, M. (2005). Understanding Syntax, 2nd edition. London: Hodder Arnold. Tavakolian, S. (1977). Structural principles in the acquisition of complex sentences. Unpublished doctoral dissertation. University of Massachusetts at Amherst, USA. Tavakolian, S. (1978). Children’s comprehension of pronominal subjects and missing subjects in complicated sentences. In H. Goodluck & L. Solan (eds.), Papers in the structure and development of child language. University of Massachusetts Occasional Papers in Linguistics 4, pp. 145-152. Amherst, MA: GLSA. Trask, R. L. (1981). Basque verbal morphology. Iker 1: 285-304. Bilbao: Euskaltzaindia. Trask, R. L. (2003). The noun phrase: nouns, determiners and modifiers, pronouns and names. In Hualde, J.I. & J. Ortiz de Urbina (eds.), A grammar of Basque, pp. 119- 318. Berlin/ New York: Mouton de Gruyter. Tsimpli, I. M. (1990). The clause structure and word-order of Modern Greek. University College London Working Papers in Linguistics 2: 226-258. 179 Tsimpli, I. M. (2011). External interfaces and the notion of ‘default. Linguistic Approaches to Bilingualism 1: 101-103. Tsimpli, I. M., & Sorace, A. (2006). Differentiating interfaces: L2 performance in syntax-semantics and syntax-discourse phenomena. In D. Bamman, T. Magnitskaia & C. Zaller (eds.), Proceedings of the 30th Annual Boston University Conference on Language Development (BUCLD30), pp. 653-664. Somerville, MA: Cascadilla Press. Tsimpli, I. M., Sorace, A., Heycock, C. & Filiaci, F. (2004). First language attrition and syntactic subjects: A study of Greek and Italian near-native speakers of English. International Journal of Bilingualism 8: 257-277. Unsworth, S. (2005a). Child L1, child L2 and adult L2 acquisition: Differences and similarities. In A. Brugos, L. Micciulla and C. E. Smith (eds.), Proceedings of the 28th Annual Boston University Conference on Language Development (BUCLD28), pp. 633- 644. Somerville, MA: Cascadilla Press. Unsworth, S. (2005b). Child L2, adult L2, child L1: Differences and similarities. A study on the acquisition of direct object scrambling in Dutch. Unpublished doctoral dissertation, Utrecht University. Unsworth, S. (2014). Comparing the role of input in bilingual acquisition across domains. In T. Grüter & J. Paradis (eds.), Input and experience in bilingual development, pp. 181-201. Amsterdam: John Benjamins. Unsworth, S., Argyri, F., Cornips, L., Hulk, A., Sorace, A. & Tsimpli, I. (2014). On the role of age of onset and input in early child bilingualism in Greek and Dutch. Applied Psycholinguistics 35: 765-805. Valenzuela, E. (2006). L2 end state grammars and incomplete acquisition of Spanish CLLD constructions. In R. Slabakova, S. Montrul & P. Prévost (eds.), Inquiries in linguistic development: Special volume in honour of Lydia White, pp. 283-304. Amsterdam: John Benjamins. van Gompel, R. P. G. & Liversedge, S. P. (2003). The influence of morphological information on cataphoric pronoun assignment. Journal of Experimental Psychology: Learning, Memory and Cognition 29: 128-139. van Kampen, J. (1997). First steps in wh-movement. Delf: Eburon. Varlokosta, S. (2001). On the acquisition of pronominal and reflexive reference in child Greek. In M. Almgren, A. Barreña, M.-J. Ezeizabarrena, I. Idiazabal & B. MacWhinney (eds.), Research on child language acquisition, pp. 1383-1400. Somerville MA: Cascadilla Press. Wasow, T. (1986). Reflections on anaphora. In B. Lust (ed.), Studies in the acquisition of anaphora, vol. I, Defining the Constraints, pp. 107-123. Dordrecht: Reidel. Wexler, K. (1999). Maturation and growth of grammar. In W.C. Ritchie & T.K. Bhatia (eds.), Handbook of child language acquisition, pp. 55-109. San Diego: Academic Press. Wexler, K. & Chien, Y-C. (1985). The development of lexical anaphors and pronouns. Papers and Reports on Child Language Development 24: 138-149. White, L. (2011). Second language acquisition at the interfaces. Lingua 121: 577-590. White, L. (2011). The interface hypothesis How far does it extend? . Linguistic Approaches to Bilingualism 1: 108-110. Whong-Barr, M. & Schwartz, B. D. (2002). Morphological and Syntactic Transfer in Child L2 Acquisition of the English Dative Alternation, Studies in Second Language Acquisition 24: 579-616. 180 Wilson, F. (2009). Processing at the syntax-discourse interface in second language acquisition. Unpublished doctoral dissertation. University of Edinburgh, UK. Wilson, F., Keller, F. & Sorace, A. (2009). The antecedent preferences of anaphoric demonstratives and personal pronouns in L2 German. In J. Chandlee, M. Franchini, S. Lord & G.-M. Rheiner (eds.), Proceedings of the 33 rd Boston University Conference on Language Development (BUCLD33), pp. 634-645. Somerville, MA: Cascadilla Press. Yavas, M. (2002). Voice onset time patterns in bilingual phonological development. In F. Windsor, L. Kelly, & N. Hewlett (eds.), Investigations in clinical phonetics and linguistics, pp. 341-349. Mahwah, NJ: Erlbaum. Yip, V. & Matthews, S. (2000). Syntactic transfer in a Cantonese-English bilingual child. Bilingualism: Language and Cognition 3: 293-308. Zifonun, G., Hoffmann, L. & Strecker, B. (1997). Grammatik der Deutschen Sprache [Grammar of the German language]. Berlin: De Gruyter. Zuazo, K. (1998). Euskalkiak, gaur [Basque dialects today]. Fontes linguae vasconum: Studia et documenta 30: 191-234. Zwazinger, E., Allen, S. H., & Genesee, F. (2005). Crosslinguistic influence in bilingual acquisition: Subject omission in learners of Inuktitut and English. Journal of Child Language 32: 893-909. 181 Appendix A. Figures Figure 1. Participants’ antecedent choices by experimental condition in twoantecedent items in Experiment 3 Figure 2. Participants’ antecedent choices by experimental condition in twoantecedent contexts in Experiment 4 Narr Francke Attempto Verlag GmbH+Co. KG • Dischingerweg 5 • D-72070 Tübingen Tel. +49 (07071) 9797-0 • Fax +49 (07071) 97 97-11 • info@narr.de • www.narr.de JETZT BES TELLEN! JETZT BES TELLEN! VERSATZ 190 MM/ 30 MM Christine Möller Young L2 learners’ narrative discourse Coherence and cohesion Multilingualism and Language Teaching 3 2015, XIV, 280 Seiten €[D] 68,00 ISBN 978-3-8233-6903-5 How do second language learners’ text/ discourse abilities develop? The present monograph contributes to this largely unanswered question by investigating narrative texts produced by elementary school students in an English immersion program in Germany. On the basis of a psycholinguistic model of discourse production, the texts are analyzed with respect to their coherence and cohesion. This work focuses on the comprehension of null and overt subject pronouns in intrasentential anaphora contexts in Basque, a language which employs overt referential devices that fall out of the scope of what traditionally counts as third person pronouns, namely the demonstrative hura 'that' and the quasipronoun bera '(s)he (him/ herself)'. Data from native adults obtained from two experimental off-line tasks on the referential properties and the discourse features of null and overt pronouns set a baseline for comparison with a) the insights reported in descriptive grammars and with b) developmental data from 6-8-year-old child L1 and child L2. L anguage D evelopment 34