English subjects in the linguistic production of L1 Spanish, L1 Bosnian and L1 Danish speakers: typological similarity and transfer

UNIVERSITY OF VALLADOLID FACULTY OF ARTS ENGLISH DEPARTMENT PhD PROGRAM: ADVANCED ENGLISH STUDIES: LANGUAGES AND CULTURES IN CONTACT PhD DISSERTATION: English subjects in the linguistic production of L1 Spanish, L1 Bosnian and L1 Danish speakers: typological similarity and transfer Submitted by Sonja Mujcinovic in partial fulfillment of the requirements for a PhD degree from the University of Valladolid PhD Supervisor: Raquel Fernández Fuertes 2020 ii colorless green ideas sleep furiously (Chomsky 1957) iii ABSTRACT This study contributes to the analyses of transfer in the case of typologically similar and typologically different language interactions from three different perspectives: L1, modality and time of instruction. To do so the L2 English sentential subjects produced by 26 L1 Spanish, 26 L1 Bosnian and 26 L1 Danish children are analyzed. These L2 English participants are divided into two proficiency groups depending on the time of instruction received (2 or 4 years). Written production data (story-telling) were obtained by means of a wordless picture sequence adapted from the Edmond Narrative Norms Instrument (Schneider et al. 2005) which participants had to narrate. Oral production data were obtained through a semi-guided individual interview which was audio recorded and then transcribed in CHAT (Codes for the Human Analysis of Transcripts) format (CHILDES, MacWhinney 2000). The subjects produced by these participants were classified following three criteria: form (overt vs. null), grammaticality (grammatical vs. ungrammatical) and adequacy (adequate vs. nonadequate). Two formal proposals on sentential subjects are tested against these L2 English data: Holmberg (2005) and Sheehan’s (2006) with regards to [+null subject] languages being superset to [-null subject] languages; and Fernández Fuertes & Liceras (2018) and Liceras & Fernández Fuertes’ (2019) on the so-called lexical specialization approach that accounts for both directionality and effect of cross-linguistic influence. The results show that typological similarity is a conditioning factor in what regards both core grammatical structures and syntax-pragmatics interface related issues. Time of instruction, however, does not have any effects on these children’s L2 English acquisition of sentential subjects. In the case of modality, the written task is proven to be cognitively more demanding. These results offer a new window into the analysis of English L2 subjects in that they not only confirm the vulnerability of interfaces also in the case of under-studied languages, but they also show how Liceras & Fernández Fuertes’ proposal applies to L2 acquisition: cross-linguistic influence from the superset language (i.e. Spanish and Bosnian) results in positive transfer. iv ACKNOWLEDGEMENTS Completing a dissertation is an enduring and lonesome road, but I have never felt alone. I have had the privilege of sharing this roller-coaster ride with my family, friends and colleagues. First of all, I want to thank my supervisor, Raquel Fernández Fuertes, for her guidance and her support over the last decade (or so) and for never letting go of my hand throughout this process. Her great work and influence clearly shows and I am very aware of how lucky and privileged I have been to share this experience with her. As a supervisor, she has proven to be one of the kind. Her knowledge, her continuous feedback and various insights that she has provided throughout all these years have clearly shaped this dissertation. She has taught me so much. I am forever thankful! Her support goes beyond academia. She is a person whose opinion I truly value and trust. All that I can say is: ¡Mil millones de gracias por ti, Raquel! Thank you for trusting me and for believing in me. Esther Álvarez de la Fuente together with the University of Valladolid Language Acquisition Lab (UVALAL) members have been very helpful, too. It is a great pleasure working with then. This team has taught me a lot about what it means to be a “piña.” Let’s keep up with this. I am particularly thankful to Ed, Tamara, and Diana! Iban Mañas, my stats star, has also played a great role in the development of this dissertation. Вы пришли, чтобы остаться (aunque sea de usted). Спасибо, amigo! Many thanks as well to Damir Pavelic for the drawings, Estíbaliz Vázquez Tabera for collecting the data in Canada and Luis Miguel Toquero Pérez for helping me with the transcription. I also want to thank the external examiners for their generous and kind comments and reviews. v During the first years of this dissertation, I have had the opportunity of carrying out a research stay at the University of Copenhagen (KUA) under the guidance of Kasper Boye to whom I am also very thankful and indebted. This stay has not only provided me with useful knowledge on linguistics from many people from the Department of Scandinavian Studies and Linguistics, but it has made possible the data collection that took place both in Denmark and in Bosnia. Morgens Larsen and Nadja Michael (my old elementary school teachers) have been so nice to help me get by with the data compilation and all the related paperwork in Pedersborg skole. Svjetlana Radusin, the director of the Aleksa Santica school in Banja Luka, has also been so was very kind and forthcoming. I thank her and Dijana Lazic for all the help received and for letting me occupy their work area. They have opened the doors to their school, when others had shut them close, and I can only be grateful for that. Once back in Spain, Eduardo Gómez Garzarán, has been my right hand at the Ave María School in Valladolid. The children who participated in this study and their parents deserve my warmest acknowledgements. Without them, this dissertation would not exits. One of the most joyful moments of data compilation took place during the interviews. I am very grateful for this experience. Parts of this dissertation have been presented at various conferences such as GALA (Generative Approaches to Language Acquisition), BICLCE (Biennial International Conference on the Linguistics of Contemporary English), EuroSLA (European Second Language Association), AESLA (Asociacón Española de Lingüística Aplicada), AILA (International Association of Applied Linguistics), CILC (Corpus Linguistics Conference). This investigation is part of the research funded by the Castile and León Regional Government and the ERD (European Regional Development Fund) under Grant Ref. vi VA009P17, and by the Spanish Ministry of Science, Innovation and Universities and ERDF under Grant Ref. PGC2018-097693-B-I00. I am also very grateful to the Department of Investigation, Innovation and Transfer and the English Department at the University of Valladolid for making this financially possible. A space was also provided to me at the University of Granada. I am very grateful to Cristobal Lozano and his ANACOR team for letting me join in. I have, at all times, been treated as any another member. A special thanks goes to Teresa Quesada and Fernando Martin Villena for taking their time to discuss both academic and personal issues and for showing me around Granada. My special emotional support team also needs a special dedication. I am so lucky to have been able to have private tutorials on how to survive a dissertation with David Carvajal and Jon P. Arregui. I am also indebted to my dear friends and family both near and far who have suffered my everlasting cries. Nonetheless, those who have made this more than possible are my parents. They have had to sacrifice a lot for me, in so many ways. Majko, tajo, ovaj doktorat posvećujem vama. Volim vas puno! Hvala vam za sve! Together with my parents is my baby brother and his family. They have given me so much joy and support! Tak for Freya og Falke! I am the proudest tetka in the world. And finally, my partner who has the greatest patience. He has always been my peace and my safe (zen) zone, whenever I needed it. Gracias, habibi! vii INDEX ABSTRACT ............................................................................................................................. ii ACKNOWLEDGEMENTS ..................................................................................................... iv LIST OF TABLES .................................................................................................................. ix LIST OF FIGURES ................................................................................................................ xi CHAPTER 1: INTRODUCTION ............................................................................................ 1 CHAPTER 2: THEORETICAL BACKGROUND ................................................................... 6 2.1 Sentential subjects cross-linguistically ......................................................................... 8 2.2 The Extended Projection Principle: from a principle to a feature-checking mechanism .......................................................................................................................................... 10 2.3 Overview of the Null Subject Parameter .................................................................... 13 2.3.1 The Null Subject Parameter as a cluster of properties: the original formulation 13 2.3.2 On the Agreement Parameter: the initial formulation ......................................... 17 2.3.3 On Extended Projection Principle checking cross-linguistically & the Agreement Parameter ...................................................................................................................... 19 2.3.4 On the Agreement Parameter & pronouns .......................................................... 23 2.3.5 A reformulation of the Null Subject Parameter in terms of phonological overtness ....................................................................................................................... 26 2.4 Summary ..................................................................................................................... 30 CHAPTER 3: SECOND LANGUAGE ACQUISITION ........................................................ 32 3.1 The acquisition of more than one language ................................................................ 32 3.2 L1 transfer in L2 acquisition ...................................................................................... 36 3.3 Age of exposure .......................................................................................................... 43 3.4 Input: quantity & quality ............................................................................................ 47 3.5 Review of L2A studies on the Null Subject Parameter .............................................. 50 3.5.1 Contact between [+null subject] & [-null subject] languages ............................. 51 3.5.2 Contact between two [+null subject] languages .................................................. 56 3.5.3 Contact between two [-null subject] languages ................................................... 57 3.6 Summary ..................................................................................................................... 59 CHAPTER 4: METHODOLOGY .......................................................................................... 62 4.1 Participants ................................................................................................................. 63 4.1.1 Selection criteria .................................................................................................. 63 4.1.2 Language Background Questionnaire ................................................................. 66 4.1.3 Measuring proficiency via MLUw values ........................................................... 68 4.1.4 L1 Spanish groups ............................................................................................... 71 4.1.5 L1 Bosnian groups ............................................................................................... 72 viii 4.1.6 L1 Danish groups ................................................................................................. 73 4.1.7 Control group ....................................................................................................... 74 4.2 Data protocol & fieldwork .......................................................................................... 75 4.3 Experiments & data collection ................................................................................... 77 4.3.1 Oral task ............................................................................................................... 77 4.3.2 Written task .......................................................................................................... 78 4.3.3 Pilot study ............................................................................................................ 84 4.4 Transcription & coding procedure .............................................................................. 85 4.5 Statistical methods for data analyses .......................................................................... 91 4.6 Summary ..................................................................................................................... 92 CHAPTER 5: RESEARCH QUESTIONS & HYPOTHESES ................................................ 94 5.1 Research questions ..................................................................................................... 94 5.2 Hypotheses .................................................................................................................. 95 CHAPTER 6: DATA ANALYSIS & DISCUSSION ............................................................... 99 6.1 Overall approach to the data: main effects and interactions ..................................... 100 6.2 Break-down of the data analysis: pairwise comparisons .......................................... 102 6.2.1 Hypothesis #1: transfer due to typological similarity ........................................ 103 6.2.2 Hypothesis #2: The different availability of subject types between the L1 and the L2 ................................................................................................................................ 109 6.2.3 Hypothesis #3: modality (oral vs. written) ........................................................ 113 6.2.4 Hypothesis #4: time of instruction ..................................................................... 117 6.3 Summary of main findings ....................................................................................... 128 CHAPTER 7: CONCLUSION ............................................................................................ 132 REFERENCES .................................................................................................................... 138 APPENDIX I ....................................................................................................................... 162 APPENDIX II ..................................................................................................................... 163 ix LIST OF TABLES Table 1: Verbal paradigm across the four languages under study ……………………………6 Table 2: Participant groups ………………………………………………………………...65 Table 3: Descriptive statistics for English MLUw values per participant group …………..69 Table 4: Data codification variables ……………………………………………………….86 Table 5: Summary of the participants’ linguistic profile …………………………..……….93 Table 6: Summary of the GLMM fixed effects for grammaticality ……………………….100 Table 7: Summary of the GLMM fixed effects for adequacy …………………….……….101 Table 8: Distribution of subjects per language group: grammaticality ...…………..……...103 Table 9: Comparisons across participant groups: grammaticality ………………..……….104 Table 10: Distribution of subjects per language group: adequacy ………………..……….105 Table 11: Comparisons across participant groups: adequacy ……………………..………106 Table 12: Distribution of subjects per language group: subject types ……………...……...110 Table 13: Comparisons within participant groups: subject types (overt and null) …...…....111 Table 14: Comparisons across participant groups: subject types (overt and null) ….…….111 Table 15: Distribution of subjects per language group: modality and grammaticality …….113 Table 16: Comparisons within participant groups: modality …………………………..…114 Table 17: Comparisons across participant groups: modality and grammaticality ……..…114 Table 18: Distribution of subjects per language group: modality and adequacy …………..115 Table 19: Comparisons within participant groups: modality and adequacy ………….…...116 Table 20: Comparisons across participant groups: modality and adequacy ……………….116 Table 21: Distribution of subjects per language group: time of instruction and grammaticality…………………………………………………………………………….119 5 Taking as a point of departure the revision in chapters 2 and 3, in chapter 5 the research questions are raised and the corresponding hypotheses formulated. Chapter 6 is dedicated to the analysis of the results obtained followed by a discussion. Both the answers to the research questions and the confirmation or rejection of the hypotheses that have been initially posed are provided. The conclusions reached are available in chapter 7, where the contribution and limitations of the present study as well as suggestions for further research are included. 6 CHAPTER 2: THEORETICAL BACKGROUND In this chapter the focus is set on how sentential subjects have been described in linguistic theory and how differences across languages have been formally accounted for in generative linguistics. A comparison is offered between different formal accounts of sentential subjects in English, as the language under investigation in this study, as well as Spanish, Bosnian and Danish, as the first languages (L1s) of the participants from whom data have been collected and analyzed (chapter 4). Typologically, Spanish is a Romance language and Bosnian a Slavic language – more specifically South Slavic (Franks 1995, 2005, 2017; Lindseth 1997; Godjevac 2000; Progovac 2005, among others). Both languages have a rich morphological verbal agreement system, as the paradigm in table 1 shows. In relation to the Null Subject Parameter, this morphological richness has led many researchers to classify them as [+null subject] languages, as they allow their subjects to be null. In contrast, English and Danish are both Germanic languages and they have a poor verbal agreement morphology, as table 1 shows, and are thus classified as [-null subject] languages, requiring their subjects to be overt. ENGLISH SPANISH BOSNIAN DANISH I sing yo cant-o ja pjeva-m jeg synger you sing tu cantas-s ti pjeva-š du synger he/she/it sings el/ella cantaon/ona pjevahan/hun synger we sing nosotros canta-mos mi pjeva-mo vi synger you sing vosotros canta-is vi pjeva-te I synger they sing ellos canta-n oni/one pjeva-ju de synger Table 1: Verbal paradigm across the four languages under study 7 Using the terminology provided by Jaeggli & Safir (1989), Bosnian and Spanish are classified as morphologically uniform languages, as each grammatical person is identified by an independent morphological marker, as reflected in table 1. This type of uniform agreement is associated with the possibility of allowing null subjects. Furthermore, Bosnian has an overtly marked case system with seven different cases (i.e. nominative, accusative, genitive, dative, locative, instrumental and vocative), subjects bearing mainly nominative case 2 . It also has a relatively free word order that “serves to express functional sentence perspective information rather than grammatical relations” (Franks 1995:3). Also in Spanish, subjects bear nominative case and the word order is relatively free (Olarrea 1998). Bearing in mind the above-mentioned classification and the basic distinctions between [+null subject] languages (like Spanish and Bosnian) and [-null subject] languages (like English and Danish), two different approaches to the Null Subject Parameter are reviewed in this chapter. On the one hand, Alexiadou & Anagnostopoulou (1998), among others, consider that in [+null subject] languages verbal agreement affixes and null subjects have the same status and that overt pronouns are pragmatically marked. On the other hand, Holmberg (2005) and Sheehan (2006), among others, propose that the only difference between null and overt subjects in [+null subject] languages lies in whether or not sentential subjects are phonologically articulated or not. The chapter is organized as follows: section 2.1 provides some general information about the nature of the subjects and their availability in the four languages under consideration; section 2.2 provides an account of the Extended Projection Principle and its 2 All Slavic languages allow their subjects to be in dative case with impersonal predicates if they also have experiencer theta-role. In contrast to other Slavic languages, Bosnian does not allow dative subjects with infinitives. In Spanish, the impersonal subjects are null and they are marked with se (Otero 1999). 8 change of status from a principle (as part of the Principles and Parameters approach) to a feature (under more minimalist assumptions); section 2.3 offers an analysis of the Null Subject Parameter and its link to both the Extended Projection Principle as well as the Agreement Parameter cross-linguistically. The last section provides an overview of the chapter and states the main theoretical foundations that will serve as the bases of this dissertation. 2.1 Sentential subjects cross-linguistically The distribution of sentential subjects in the four languages involved in the present dissertation differs in terms of the availability of null subjects across languages, as illustrated in the examples bellow. Across the four sets of examples, the three possible forms are illustrated as follows: in examples 1-4a the subjects are Determiner Phrases (DPs), in 1-4b personal pronouns and in 1-4c the subjects are null. 1) English: a) These green apples are the best [DP subject] b) They had to take care of the boy [pronominal subject] c) *Ø were happy that day [null subject] 2) Spanish: a) Estas manzanas verdes son las mejores [DP subject] b) Ellos tenían que cuidar al niño [pronominal subject] c) Ø estábamos contentos aquel día [null subject] 9 3) Bosnian: a) Ove zelene jabuke su najbolje [DP subject] b) Oni su trebali čuvati dječaka [pronominal subject] c) Ø bili smo sretni taj dan [null subject] 4) Danish: a) Disse grønne æbler er de bedste [DP subject] b) De skulle passe drengen [pronominal subject] c) *Ø var lykkelige den dag [null subject] In spite of the availability of subjects across languages, as per the Extended Projection Principle, the nature of subjects is indeed subject to variation, as displayed in the examples above, and so languages are divided into two groups: [+null subject] and [-null subject] languages. Verbal agreement, as the functional projection of the verbal lexical head that involves a checking relationship between the subject and the verb, has also been argued to play a crucial role in this classification (Pollock 1989; Chomsky 1991, 1995, among others). This link between the nature of the subject and that of verbal agreement has been captured in the Agreement Parameter (Pollock 1989; Chomsky 1995; Alexiadou & Anagnostopoulou 1998; Kato 1999, among others) in that the presence of null subjects is tied to the [+pronominal agreement] nature of verbal inflection (Alexiadou & Anagnostopoulou 1998 and Kato 1999). As shown in the examples above, [+null subject] languages like Spanish, as in 1, and Bosnian, as in 2, are also [+pronominal agreement] languages in contrast to [-null subject] languages like Danish, as in 3, and English, as in 4, which are [-pronominal 10 agreement] languages 3 . These, as well as other grammatical properties of subjects, are discussed in more detail in the subsequent sections. 2.2 The Extended Projection Principle: from a principle to a feature-checking mechanism It is a universal requirement that all sentences must have a subject, as initially captured under the Extended Projection Principle which, in its original formulation (see Chomsky 1981, 1982), as in 5, involved the projection of the specifier of the inflection phrase (i.e. SpecIP) as the canonical position of sentential subjects. 5) Canonical position of subjects The Extended Projection Principle imposes the specifier position to be projected at all syntactic levels, but it does not impose this position to be filled. Further investigation on the Extended Projection Principle has led to the development of other possible positions, including the specifier of the verb phrase under the VP-internal Subject Hypothesis (Koopman & Sportiche 1985, 1991; Radford 1997; Lasnik 2001; Lasnik & Park 2003; 3 This theory is challenged by Asian languages such as Chinese, Korean, Thai and Japanese, among others, because they are [-pronominal agreement] languages, but they license null subjects (Huang 1984 and Neeleman & Szendröi 2005, among others). These languages are classified as topic-drop, since they allow null pronominal arguments (i.e. both subjects and objects can be dropped). Hence, not all [+null subject] languages are also [+pronominal agreement] languages. IP I´ VP (subject position) SpecIP 11 Radford 2004), as well as to a somewhat different formulation in terms of feature checking (Chomsky 1982, 1995; Uriagereka 1996), which will be discussed throughout this chapter. In [-null subject] languages, if the Extended Projection Principle position is not filled with a thematic subject, then, an expletive pronoun has to fill this position, otherwise the Extended Projection Principle would be violated making the sentence ungrammatical. In [+null subject] languages, since null referential subjects are allowed, this position is said to be filled with an empty category (i.e. pro). In the early minimalist interpretation of the Extended Projection Principle, Chomsky (1995), among others, considers that there is a feature in T(ense) that is part of subject licensing (i.e. the EPP feature). The EPP feature is a universal [–interpretable] 4 nominal feature that merges in T and that, therefore, must be valued and deleted before reaching the interfaces. EPP checking is also related to verbal agreement morphology and the status of agreement (i.e. [+/-pronominal agreement]). In [-pronominal agreement] languages, the EPP feature can be checked in two ways (Chomksy 1995): i) merge, as in 6, and ii) move, as in 7: 6) merge XP there came a woman [IP there EPP [I [VP came a woman]]] 7) move X a woman came [IP a woman EPP [I [VP came]]] 4 Following Chomsky (1995) and further taken up by Holmberg (2005:536), person, number and gender features marked on a DP are [+interpretable], restricting the denotation of the DP; whereas these same features marked on lexical verbs, auxiliary verbs and adjectives are [-interpretable], as they do not denotate these categories. 12 These examples are derived from a structure such as came a woman. In example 6, the EPP feature is checked via agreement by merging an expletive (i.e. there). In example 7, a woman moves to SpecIP to check the EPP feature. Merging operations are less costly than moving operations. As for Case features and phi-features, they are checked via agreement. In the merge operation, I carries strong specifier-features which are checked in SpecIP by the subject as a result of the merge operation. In the move operation, subjects carry the Case features which can only be checked if the subject is moved (or raised) to SpecIP (Radford 2004:166). In [+pronominal] agreement languages, the EPP feature and the [D] feature are checked when the verb moves to I, as shown in 8:. 8) [+pronominal] agreement languages llegó una mujer came a woman “a woman came” [ IP [I’ llegói +D +θ [VP una mujer ti]]] IP VP V’ Spec I Spec [ EPP ] [ D ] [ θ ] una mujer [v] llegó I´ V 13 In other words, initially, it was proposed that subjects are licensed via IP functional projections and the different features are checked through movement (i.e. A-movement). Subsequently under minimalist assumptions, the licensing of subjects is said to occur through EPP feature checking (Chomsky 1995, 1999, 2000 and Lasnik 2001, among others), even though the EPP checking is satisfied differently across languages (see section 2.3.2). 2.3 Overview of the Null Subject Parameter Parameters involve clusters of properties and divide languages typologically, thus capturing cross-linguistic variation (Perlmutter 1971; Chomsky & Lasnik 1977; Taralsden 1978, 1980; Jaeggli 1981, 1982, 1984; Rizzi 1982, 1997, 2005; Chomsky 1981; Phinney 1987; Platzack 1987; Liceras 1988, 1989; Jaeggli & Safir 1989; Bel 2001; Belletti 2001, 2004; Holmberg 2005, 2010; Sheehan 2006; Frascarelli 2007; Camacho 2006, 2008, 2011, 2013, 2016; Holmberg & Roberts 2011; Cuza & Camacho 2017; Roberts 2018, among others). The Null Subject Parameter divides languages into [-null subject] languages, like English and Danish which do not allow their subjects to be null, in contrast to [+null subject] languages, like Spanish and Bosnian which allow their subjects to be both null and overt. 2.3.1 The Null Subject Parameter as a cluster of properties: the original formulation The initial approach to the Null Subject Parameter formulated by Perlmutter (1971) was based on the idea that languages can be classified into [+/-null subject] on the basis of the Extended Projection Principle, which requires the projection of subjects, and the presence or absence of verbal agreement. 14 Chomsky (1981) and later Rizzi (1982) observed how null subject languages displayed similarities that have been classified into at least four different clusters of properties. In the examples bellow, Spanish and Bosnian are used to illustrate these clusters: i) possibility of pro (i.e. referential subjects) in subject position of tensed clauses, as in examples 9 for Spanish and 10 for Bosnian presented above and repeated here: 9) Estábamos contentos aquel día Be-PRS.1PL happy that day “we were happy that day” 10) Bili smo sretni taj dan Be-PRS.1PL happy that “we were happy that day” ii) possibility of subjects in post-verbal position, as in examples 11 and 12 for Spanish and 13 and 14 for Bosnian: 11) Juan ha venido John has arrived “John has arrived” 12) ha venido Juan has arrived John “John has arrived” 13) Marko je došao Marko has arrived “Marko has arrived” 14) došao je Marko has arrived Marko “Marko has arrived” iii) possibility of an explicit complementizer when the subject of an embedded clause is moved, as in examples 15 for Spanish and 16 for Bosnian: 15) quiéni pro crees que ti se ha ido a casa who do you think that went home “who do you think Ø went home” 21 26) [+null subject] languages a) Spanish (Romance) Vamos. go-PRS.1PL “We go.” (adapted from Liceras & Fernández Fuertes 2019:5) b) Bosnian (Slavic) Idemo. go-PRS.1PL “We go.” (adapted from Liceras & Fernández Fuertes 2019:5) 27) [-null subject] languages a) English (Germanic) “We go.” (adapted from Liceras & Fernández Fuertes 2019:5) b) Danish (Germanic) Vi går. “We go.” (adapted from Liceras & Fernández Fuertes 2019:5) TP VP T [ vai ] [ mosj ] vai-mosj TP VP T [ idei ] [ moj ] idei-moj VP V go TP [ wei ] wei DP Spec VP V går TP [ vii ] vii DP Spec 22 In languages like English agreement does not bear pronominal affixes and, so, agreement does not have argumental features as Spanish has. Therefore, in [-null subject] languages, subjects must be represented either as free morphemes (such as pronouns) or as overt DPs. This same rationale applies to Danish as a [-null subject] language. In Spanish and Bosnian, [+null subject] languages, agreement does bear pronominal affixes and thus, null subjects are allowed. In Spanish-like-languages, overt pronouns have received two analyses: i) overt pronouns and null pronouns co-occur with verbal agreement affixes; the difference between the two subject types is that overt pronouns have a phonological form whereas null pronouns do not (Holmberg 2005 and Sheehan 2006, among others); and ii) null pronouns and overt pronouns differ in that the later have a pragmatic value which null pronouns do not (Alexiadou & Anagnostopoulou 1998 and Kato 1999, among others). We agree with the first approach in that overt pronouns need not have a pragmatic value. Therefore, in Spanish or Bosnian ([+null subject] languages), when a null subject is produced, the subject is not phonologically articulated, although it is still specified. When an overt subject is produced, it is argued to be both phonologically articulated and specified. In either case, whether subjects are overt or null, they are specified as per EPP requirements. Thus, [+null subject] languages exhibit a double option, as subjects can be phonologically articulated or not, whereas [-null subject] languages, such as English or Danish, require their subjects to always be phonologically articulated. 23 2.3.4 On the Agreement Parameter & pronouns Cardinaletti & Starke (1999) propose a classification of pronouns into strong and deficient forms, the latter being subdivided into weak pronouns and clitics. Deficient forms can be semantically empty, because they do not necessarily occupy a theta-position. Whereas, strong pronouns can be referential without being associated with an antecedent that is prominent in the discourse. Their function is both syntactic and emphatic and they can double both weak pronouns and clitics in a structure. As of semantics, weak pronouns can be expletives, impersonal, non-referential datives, possibly non-human. That is, they can be semantically empty, but they must have a D-antecedent. Strong pronouns are, on the contrary, similar to morphemes and they occupy a theta-position, which means that they cannot be semantically empty. They can be referential without being associated with an antecedent in the discourse. As of distribution, strong pronouns are full nominal projections, whereas weak pronouns lack the highest functional layer and clitics lack both highest functional layers. That is, the deficient forms lack C’ and, therefore, they do not contain case features. Since verbal agreement is necessary for Case to be assigned, deficient forms must occur in local structural configuration with agreement. Cardinaletti & Starke consider pro a deficient pronoun, a superset of weak pronouns (see also Holmberg 2005 and Saab 2009, 2010, 2014, 2016, among others). It is projected in the specifier of AgrP in the same line as weak elements (cf. Rizzi 1986 and Chomsky 1993). As of choice, weak forms are preferred over strong forms. This is captured in the Avoid Pronoun Principle (Chomsky 1981 and Fernández Soriano 1998). Nonetheless, as Camacho (2013) states, weak pronouns have fewer syntactic/functional projections, but this does not necessarily mean that they have less prosodic content or less semantic complexity. “A weak 24 pronoun can remain prosodically weak but show one of the properties of strong pronouns” (Camacho 2013:91). To sum up, under Cardinaletti & Starke’s (1999) proposal, clitics are distributionally, morphologically, semantically and emphatically deficient, if compared to weak pronouns, whereas weak pronouns such as pro are deficient, if compared to strong pronouns. Kato (1999), following Cardinaletti & Starke (1999), classifies pronouns into strong, weak and clitics. Under this classification, Kato considers free pronouns, clitics and pronominal agreement affixes to be weak, supposing that all three bear the [D] feature and appear as independent items in the numeration. Agreement morphemes are independent [D] features that bear both case and theta-features which are merged with tense inflected verbs. In order for these features to be checked, Agreement has to raise to T, and so SpecTP is not projected. In the same line as Alexiadou & Anagnostopoulou (1998), Kato also states that these affixes are merged as VP arguments and have case. She argues that strong pronouns and lexical subjects are in a higher projection where they check nominative case. This idea has previously been taken up to discussion by Roberts (1991), who claims that the licensing of null subjects is governed by agreement in the sense that null subjects are licensed where nominative case can be assigned. Kato (1999) argues that it is not the [+pronominal] inflection as such that licenses null subjects (i.e. pro) in [+null subject] languages. It is rather the agreement morphemes that contain Case and phi-features and are considered independent [D] items which merge with inflected verbs as external arguments. Pronominal agreement affixes are [+strong] or [+interpretable] and, in fact, replace overt subjects in its EPP licensing capacity and receive theta-role. If agreement is [+interpretable] in [+null subject] languages, then pro is redundant, because it is not even projected and, therefore, does not exist. Pronominal agreement is in the 25 numeration and it has the same status as a subject clitic or a weak pronoun, as mentioned above. In addition, more than one weak form can appear in complementary distribution which means that pronominal agreement can co-exist with, say, a weak pronoun. As of strong pronouns, they are available in all languages and can double weak forms. For example, in [-null subject] languages this doubling occurs with the use of a strong and a weak pronoun, as illustrated in 28 below. 28) ME, I want bananas 29) YO quiero bananas As of [+null subject] languages, the doubling of the subjects involves [+pronominal agreement] (i.e. the weak pronoun -o in quier-o) and strong pronouns (i.e. the nominative subject pronoun yo), as in 29. According to Kato, doubling in [+null subject] languages does not involve pro; it involves the agreement affix and the (always strong) pronoun. Agreement affixes enter the numeration as independent items bearing case features, just like free weak pronouns in [-null subject] languages. Strong and weak pronouns also differ in their respective domains; the domain for strong pronouns is C, while that for weak pronouns is XP. In both [+/-null subject] languages, DPs that function as subjects fill the SpecIP position, which is also argued to be the same position occupied by strong pronouns. Both [+/-null subject] languages have the option of strong and weak pronouns in subject position. Following Kato, the subject strong pronoun option seems to be related to pragmatic factors (i.e. contrast) rather than to syntactic factors. That is, strong pronouns are pragmatically marked. To sum up and putting together EPP checking and the nature of pronouns, following Alexiadou & Anagnostopoulou’s (1998) and Kato’s (1999) approach, Spanish agreement markers and English pronouns differ in terms of EPP-feature checking. Spanish agreement 26 affixes are considered pronominal elements with a [+D] feature in the numeration and they are [+interpretable]. In English, overt pronominal elements merge in SpecTP. Thus, English overt pronouns and Spanish null pronouns occupy different positions, although they carry similar syntactic value, whereas Spanish overt pronouns bear both semantic and pragmatic value and occupy the focus position (i.e. Adjunct Phrase). 2.3.5 A reformulation of the Null Subject Parameter in terms of phonological overtness In contrast to Alexiadou & Anagnostopoulou’s (1998) and Kato’s (1999) approach, Holmberg (2005, 2010), Sheehan (2006), Martínez-Sanz (2011) and Liceras & Fernández Fuertes (2019), among others, argue that overt pronouns in [+null subject] languages and overt pronouns in [-null subject] languages occupy the same position (e.g. SpecIP) and are equally interpreted, that is, have a similar value. The difference is that overt pronouns are spelled out, that is, they have a phonological form, whereas null pronouns are not spelled out, that is, they have a syntactic function but no phonological form. Based on his analysis of Finnish, Holmberg (2005) argues that there are at least three types of null subject pronouns: i) null weak pronouns (as argued by Cardinaletti & Starke (1999)) that bear the phi-feature but lack the [D] feature; ii) a deleted DP under recovery conditions; and iii) pro (i.e. a bare noun with no phi-features that is only available in languages with no agreement). To compare both approaches (i.e. Alexiadou & Anagnostopoulou’s, on the one hand, and his own approach, on the other), Holmberg provides two hypotheses which are based on the interpretability of agreement. Hypothesis A (i.e. supported, among others, by Alexiadou & Anagnostopoulou (1998) and Kato (1999)) considers agreement as [+interpretable] in 27 [+null subject] languages and hypothesis B (i.e. supported, among others, by Holmberg (2005) himself and Sheehan (2006)) assumes that null subjects in [+null subject] languages have no phonological form, but they value the [-interpretable] features of agreement. More specifically, these two hypotheses are defined as follows: Hypothesis A: There is no pro at all in null subject constructions. Instead, Agr (the set of phi-features of I) is itself interpretable; Agr is a referential, definite pronoun, albeit a pronoun phonologically expressed as an affix. As such, Agr is also assigned a subject theta-role, possibly by virtue of heading a chain whose foot is in vP, receiving the relevant thetarole (Holmberg 2005:537). Thus, hypothesis A indicates that, if agreement is [+interpretable] then it has to be referential. If it is referential, then it fulfills the Extended Projection Principle; that is, it has to check nominative Case and a subject theta-role. If this is so, then there is no need for pro (see examples 30 and 31 below). Under hypothesis A, if agreement is [+interpretable] and can check the EPP features, then SpecIP is not projected (see example 30). 30) representation of hypothesis A with [+interpretable] agreement features Spanish: abrimos el libro open-PRS.1PL the book “We open the book.” 28 Nevertheless, if agreement is [-interpretable] and cannot check the EPP features, SpecIP is projected and occupied by a pronoun, as in 31. 31) representation of hypothesis A with [-interpretable] agreement features English: we open the book IP VP V´ el libro I DP abri-mos I´ V DP [EPP] IP VP V´ the book Spec wei I DP open I´ V DP ti [EPP] 29 Hypothesis B shows an alternative view. Hypothesis B: The null subject is specified for interpretable phi-features, values the uninterpretable features of AGR, and moves to Spec,IP, just like any other subject. This implies that the nullness is a phonological matter: the null subject is a pronoun that is not pronounced (Holmberg 2005:538). Thus, hypothesis B indicates, contrary to hypothesis A, that agreement morphology is [- interpretable]. The SpecIP position is always occupied by a pronoun checking the EPP features, and, therefore, this position cannot be occupied by another category. This makes the null subject a pronoun that has no phonological form. Under hypothesis B, SpecIP is only available for a pronoun that can check the EPP features (example 32). 32) tree diagram representation of hypothesis B In the case of English, we moves to the SpecIP position to check its features leaving a trace in Spec VP. In Spanish, nosotros “we” (overt subject) is considered a weak pronoun, whereas IP VP V´ el libro a book Spec proj.(PF =nosotros) wei I Spec abri-mosj open I´ V DP [ EPP ] ti 30 pro is an even weaker pronoun, both being coreferential with verbal inflection (-mos). Therefore, in Spanish subjects may have two realizations: the overt subject (i.e. PF realization) and the null subject (i.e. no PF realization). 2.4 Summary To sum up, under Alexiadou’s & Anagnostopoulou’s (1998) and Kato’s (1999) account of the Null Subject Parameter, verbal agreement markers in a [+null subject] language are equivalent to weak pronouns in a [-null subject] language. While, under Holmberg’s (2005, 2009 and 2010) account pro is considered to be phonologically silent, but a syntactically realized head. Following this idea, preverbal null subjects of a finite clause occupy SpecIP and can have two different forms; i) a null pronoun that is specified for phifeatures but lacks the [+D] feature or ii) a fully specified pronoun with a [+D] feature, which has been deleted in the phonology (Holmberg 2005: 559). In this dissertation, Holmberg’s (2005) and Sheehan’s (2006) account of the Null Subject Parameter is adapted, because, as argued by Liceras & Fernández Fuertes (2019), it has the following advantages over Alexiadou’s & Anagnostopoulou’s (1998) and Kato’s (1999) proposal: i) in both [+/-null subject] languages, preverbal subjects occupy the SpecIP position (in contrast to previous accounts where verbal agreement affixes (i.e. -mos) occupy one position and overt pronouns (i.e. we) a different one); ii) it takes into an account the nature of Spanish nominative pronouns as weak pronouns, so that they do not necessary have to have a pragmatic value; and iii) following the Superset/Subset Parameter, English and Danish represent the subset option, since they allow one option (i.e. subjects must be phonologically realized), compared to languages such as Spanish and Bosnian that allow two 37 available and have the same distribution in both languages. Thus, it results in successful acquisition of the L2 properties (i.e. positive transfer). In other words, positive transfer is said to be limited to instances where identical or equivalent language properties are to be processed, whereas negative transfer emerges when differences or conflicting properties are to be processed, which typically result in ungrammatical/non-adequate production. The amount of transfer, both positive and negative, is related to at least two issues: the degree of proficiency the L2 learner has in the L2 and the amount of similarity that the learner is able to identify between the two languages in contact and at the different linguistic domains (Cenoz 2001; Gass & Salliner 2008). It is assumed that, in language contact situations, L2 learners rely heavily on their L1, at least in the initial stages. Ideally, the more these participants are exposed to a language, the more proficient they become, the more native-like their production gets and, consequently, the less L1 transfer occurs (e.g. Ringbom 2007, 2016; Blom & Baayan 2012; Montrul & Ionin 2012; Gathercole 2002, 2016; Unsworth 2016a; Llinàs-Grau & Bel 2019, among others). Since proficiency is also related to exposure and to the use of the language, it seems reasonable to argue that the better knowledge speakers have of a language (i.e. the more proficient they are), the less negative transfer (i.e. errors) there will be in their L2 production (Gathercole 2016:123). In fact, Blom & Baayan (2012) argue that effects of transfer are the highest at an intermediate stage of proficiency, because developmentally L2 speakers are both ready and proficient enough to produce structures influenced by their L1. In other words, as argued by Montrul & Ionin (2012), errors (i.e. negative transfer) are more likely to occur in the beginning and in the intermediate stages and are expected to diminish in the advanced stages. Transfer is, therefore, nuanced in subsequent stages of language learning, but it does not necessarily disappear. This is evident for transfer in the syntactic or 38 morphological domain, but different issues related to the syntax-pragmatics interface continue to be vulnerable in very proficient stages, even if the languages are typologically similar (Park 2004; Tsimpli & Sorace 2006; Rothman 2009; Slabakova & Ivanov 2011; Pladevall Ballester 2012, 2016; Lozano 2018; Mitkovska & Bužarovska 2018, among others). In order to explain and account for non-native-like production at very advanced stages of acquisition, the Interface Hypothesis is put forward (Sorace & Filiaci 2006). It states that “narrow syntactic properties are completely acquirable in a second language, even though they may exhibit significant developmental delays, whereas interface properties involving syntax and another cognitive domain may not be fully acquirable” (Sorace & Filiaci 2006:340). Interface properties are more complex and, therefore, acquired later (if at all), as interfaces integrate syntactic knowledge and other cognitive systems and so require more effort. However, not all interfaces demand the same effort and are equally complex. In fact, studies on the Interface Hypothesis have focused on the connection between the internal interfaces (i.e. between syntax and other linguistic domains such as semantics and morphology), on the one hand, and between syntax and other cognitive modules such as discourse and pragmatics (i.e. the so-called external interfaces), on the other (Tsimpli & Sorace 2006; Sorace & Filiaci 2006; Domínguez 2009; Sorace & Serratrice 2009; Sánchez et al. 2010). The syntax-pragmatics interface seems to be especially problematic both in L1 acquisition and in L2 acquisition until very advanced stages of proficiency (even at nearnative levels). Furthermore, Sorace (2005) has found that interfaces are problematic for learners regardless of the languages in contact. In fact, she has found traces of non-nativelike production (i.e. residual L1 effects) in L1 Italian L2 Spanish speakers’ production, even 39 though the same pragmatic conditions regulate the distribution of overt and null subjects in both languages. Interfaces in child L2 acquisition have also been explored. In this case, the problematic nature of the syntax-pragmatics interface preventing even very proficient L2 speakers to acquire native-like performance is found to be additionally problematic due to cognitive maturity, an effect that is also found in child L1 speakers (Tsimpli & Roussou 1991; Müller & Hulk 2000; Paradis & Navarro 2003; Sorace 2005; Haznedar 2007; Rothman 2009; White 2009, 2011; Cuza & Frank 2011; Zdorenko & Paradis 2011; Müller 2017 among others). Given that L1 transfer characterizes and shapes (at least) the initial stages of L2 acquisition, attention has been placed on comparing across the L1 and the L2. Being transfer an overt manifestation of the L1 (Gass 1996:385), the similarities between the languages in contact have some bearing on the type of transfer that might appear. In the case of typological proximity and typological similarity, while the first one has centered the attention of L2 acquisition research the latter has received much less attention (Rothman 2010, 2011; Rothman & Cabrelli Amaro 2010; Montrul et al. 2011; Liceras & Alba de la Fuente 2015; Westergaard et al. 2017; Cuza et al. 2018, among others). Typological proximity groups languages that belong to the same family and share the same origin; an example is Spanish and French, both derived from Latin and considered Romance languages. However, typological similarity refers to languages that share the same option at a micro-parametric level so that “a typological or formal universal is equally realized in these two typologicallyclose languages” (Liceras & Alba de la Fuente 2015:333). In other words, Spanish and French are typologically proximate languages, because they are both Romance languages, but they are not typologically similar when dealing with subject realization, because Spanish 40 is a [+null subject] language and French is a [-null subject] language. While Spanish and Bosnian are not typologically proximate (i.e. they do not belong to the same family since Spanish is a Romance language while Bosnian is a Slavic language), they are indeed typologically similar languages, since they both share similar features within the realization of subjects (i.e. both are [+null subject] languages). Danish and English are both typologically proximate (i.e. Germanic languages) and typologically similar (i.e. [-null subject] languages). In languages that are typologically similar, the L1 can facilitate the acquisition of the L2, following the Facilitation Hypothesis (Gundel & Tarone 1992). In the same line, if linguistic differences are found between the L1 and the L2, lower L2 learnability can arise (i.e. more transfer which can impede the learning of the L2) (Schepens et al. 2016). What is evident so far is that relatedness of the languages plays a crucial role. In the case of linguistic distance between the learners’ L1 and their L2, the less typologically similar the languages are, the more negative transfer is expected. For instance, Ringbom’s (2007, 2016) study on L1 Swedish L2 English and L1 Finish L2 English proves that, since Swedish is typologically similar to English, while Finish is typologically different, this difference facilitates L2 acquisition in the case of the L1 Swedish group. In particular, as sentential subjects are equally projected and have the same distribution in English and in Swedish, an acceleration in the acquisition of this grammatical property on the part of the L1 Swedish participants is seen, when compared to the L1 Finish participants. Confirmation of the role of typological similarity is seen also in Muñoz et al.’s (2018) analysis of L1 Danish and L1 Spanish primary school learners of L2 English. Albeit the considerable difference in hours of instruction in English (10 and 12 hours for the Danish groups, and 287 and 520 hours for the Spanish groups), the scores obtained by the L1 Danish 41 participants matched the ones obtained by the L1 Spanish learners. This result is interpreted by the authors as an indication that typological similarity between Danish and English (both being [-null subject] languages) is crucial because it is making these L1 Danish speakers behave like the L1 Spanish speakers in spite of their very reduced exposure to English. Westergaard et al. (2017) analyze 2L1 Norwegian-Russian L3 English speakers. Norwegian and English are both typologically proximate (i.e. both are Germanic languages) and typologically similar in what regards many of their morphosyntactic structures. Russian, on the other hand, is typologically distant (i.e. as it is a Slavic language) and typologically different with respect to the morphosyntax in general terms. Analyzing two different aspects, i) adverb-verb word order (no V2), where English patterns with Russian and not with Norwegian and ii) subject-auxiliary inversion (residual V2), where English patterns with Norwegian and not with Russian, it was possible to determine whether Norwegian or Russian cause cross-linguistic influence in L3 English. The results from the adverb-verb word order task showed positive transfer from Russian into English. In the subject-auxiliary inversion task, both the L1 Norwegian and the bilinguals obtained very similar results. The reason for this similarity, they argue, is that this property has already been acquired, and so no effect is shown (neither facilitative from Norwegian, nor non-facilitative from Russian). Finally, Westergaard et al. (2017:33) conclude that “the typological proximity between Norwegian and English was overridden by facilitative CLI [cross-linguistic influence] from Russian, which exhibits structural similarity with English in this condition.” That is, contrary to what was expected, Norwegian does not function as a facilitator in the acquisition of English and, in this way, typological difference between English and Russian supersedes typological similarity between English and Norwegian. 42 These studies evidence that the degree of typological similarity and typological difference clearly plays a prominent role in how speakers manage their two languages and how performance in the L2 is shaped. In an attempt to shed further light on how the grammatical properties of the L1 affect the acquisition of those of the L2, Liceras & Alba de la Fuente (2015) claim that the acquisition of the L2 grammar and, consequently, the type of transfer expected are not affected by the typological proximity between the L1 and the L2 as such but rather by the typological similarity between the two languages where the focus is placed on the microparametric syntactic level. Evidence for their proposal is found in the case of Spanish and French, when it comes to the analysis of sentential subjects. As already mentioned, both are Romance languages and, therefore, typologically proximate, but, at the micro-parametric level, they are different: Spanish is a [+null subject] language while French is a [-null subject] language. Liceras & Alba de la Fuente show that L1 French speakers have an advantage over L1 English speakers of L2 Spanish, because French and Spanish verbal morphology are typologically proximate but not typologically similar. Then, either positive or negative transfer can occur, even if languages are typologically similar/different, because it is these languages’ internal similarities or differences the ones that lead to transfer, when languages are in contact. Cuza et al. (2018), in their study on the acquisition of Differential Object Marking in Spanish, analyze whether typological proximity and typological similarity are a determining factor, when languages such as L1 Mandarin L2 Spanish and 2L1 Spanish/Brazilian Portuguese (where Spanish is the heritage language and Brazilian Portuguese the dominant language) are in contact. Thus, they analyze language development both in the case of L2 speakers and heritage speakers. Their results show that typological proximity is not a 43 conditioning factor neither for the heritage nor for the L2 speakers, whereas typological similarity plays an essential role. Their results go in line with what Liceras & Alba de la Fuente (2015) also conclude: the crucial factor is not typological proximity but rather typological similarity. As presented above, issues such as L2 proficiency and L1/L2 typological similarity have been proven to influence or activate transfer in some cases. Other studies, however, find no evidence of L1 transfer in child L2 acquisition (e.g. Blom et al. 2007; Meisel 2008; Paradis 2005; Paradis et al. 2008; cf. Blom & Unsworth 2010:206-207). They argue that the reason for the absence of L1 transfer is threefold: i) the properties from a specific developmental stage might coincide with the properties attributed to L1 transfer (i.e. the production of null subjects is evident in both [+null subject] and [-null subject] languages at the very initial stages of acquisition in the so-called omission stage where, in the case of [-null subject] languages, child output does not coincide with the adult requirement); ii) L1 transfer is mainly evident in the initial stages of acquisition; and iii) not all properties are equally sensitive to L1 transfer. Issues other than typology can, of course, play a role in how speakers process and acquire the L2 and, in turn, in how transfer is shaped. Some are discussed in the subsequent sections. 3.3 Age of exposure Within language acquisition research, age in general and age of exposure (or age of onset) in particular play a crucial role. In fact, age (including critical period effects) has been the focus of attention in previous studies and has been used to classify bilingual speakers into 44 different groups, as well as to capture differences in cognitive maturity and language development. For instance, as discussed in section 3.1 above, age differences are behind the distinction between simultaneous and sequential bilinguals as captured under the critical period analyses; and, in the case of the latter, between early/child and late/adult L2 bilinguals. Cognitive maturity and language development make child L2 bilinguals pattern with L1 bilinguals in some respects and with adult L2 bilinguals in some others. Child L2 acquisition resembles adult L2 acquisition as both are L2 acquisition processes, even if adult L2 speakers have completely developed their L1 system, while child L2 speakers have not done so yet. What is more, stronger L1 transfer effects are found within child L2 than in the case of simultaneous L1 bilinguals (Unsworth 2013 and Unsworth et al. 2014). Therefore, it seems that child L2 acquisition shares transfer effects with adult L2 acquisition; while it shows a developing grammar as in the case of simultaneous L1 bilingual child acquisition. The difference between early and late L2 acquisition is established according to the specific timing when the first exposure to the L2 takes place. As early as 1979, Krashen highlighted the importance between ultimate attainment and rate (i.e. time) of acquisition. He argued that, even if older L2 learners perform at a higher rate during the first stage when it comes to issues related to morphology and syntax, while younger L2 learners’ initial performance is poorer, it is in fact younger L2 learners who reach a higher level of ultimate attainment. Older learners (i.e. late bilinguals) typically refer to speakers who have started learning the L2 around puberty. They are considered faster, because they use explicit learning mechanisms, which the younger learners have not mastered yet. In fact, previous studies have shown that older L2 children perform better, because they are more experienced and have greater cognitive maturity (Gathercole 2002a, 2002b, 2002c, 2007; Golberg et al. 2008, among others). Early learners have been found to have certain advantages over late learners 45 in natural but not necessarily in institutional contexts. These advantages are not evident in the initial stages but rather in the long term (Singleton & Ryan 2004:223). In the short term, and especially in institutional settings, this advantage has in fact not been corroborated. As Lambelete & Berthele (2015) argue, there are many limitations in L2 acquisition in an institutional setting, such as limited exposure both in terms of time of exposure and type of input. In fact, usually, learners only receive input from the teacher and varied proficiency exposure in their interaction with other learners. Taken Krashen’s (1979) initial ideas as a point of departure, different studies have attempted to capture the distinction between younger and older L2 learners in terms of their (different) linguistic abilities. Based on neurolinguistic evidence, Meisel (2008:59) establishes a tentative age range for optimal acquisition based on previous research related to the age of onset. If the onset of acquisition is before the age of 3, then simultaneous bilingual acquisition takes place. If the onset of acquisition is between the ages of 4 and 8, then it should be considered as child L2 acquisition. And finally, if the onset of acquisition occurs after the age of 10, then it should be considered as being more similar to adult L2 acquisition. Other studies, however, suggest a different time line. Guasti (2002) proposes that the critical period starts at the age of 4. For Schwartz (2004), the age of 7 is claimed to be the limit beyond which native-like attainment is no longer possible. Unsworth (2013) and Unsworth et al. (2014) also analyze age effects in early child acquisition, as proposed by Meisel (2008). In the first study on gender marking, she focuses on the comparison between bilingual children with different linguistic profiles: i) 2L1 English-Dutch children; ii) sequential bilingual children, who were exposed to Dutch between the ages of 1 and 3; and iii) L2 children who were exposed to Dutch between the ages of 4 and 10. She concludes 46 that, when it comes to gender marking, there is no evidence that the critical period ends at the age of 4, contrary to what is stated by Meisel. In the second study, they found errors in the production of the early sequential bilinguals and the L2 children that they attribute to L1 transfer, something that is not reported in the case of the 2L1 EnglishDutch children. In a more recent study conducted by Hartshorne et al. (2018), L2 English speakers are analyzed to determine whether there is a critical period, and, if so, how long it lasts and how it actually might affect the L2 acquisition process and how it is modulated (if at all) by the degree of proficiency. The participants were demographically diverse (i.e. 38 different L1s are analyzed). Their data show that the learners who are exposed to the L2 at the ages of 10 to 12 are able to reach the same level as the 2L1 bilinguals. After that age, there seems to be a decline, but they do not find that final attainment ceases after puberty, as some studies have suggested. Child learners have received quite a lot of attention as to how their acquisition process is to be analyzed and, in particular, whether their production should be compared to that of native speakers, as a baseline. Singleton & Ryan (2004) highlight the importance of comparing between L2 bilinguals themselves as well. They claim that a comparison should be made between early bilinguals and late bilinguals and taking into consideration the specific conditions under which the L2 is acquired. The studies referred to in the preceding paragraphs point to different conceptualizations of the critical period and to how age of onset may constrain L2 attainment. From the above, it can be concluded that there is no agreement where the line between early L2 and late L2 acquisition should exactly be drawn. Perhaps the key in this debate, as in Meisel (2008, 2013), is the conceptualization of the critical period as such. The critical period does not refer to a single age period but rather to sensitive phases in developing grammars. 53 subjects in the L2 but that, certainly, negative transfer from the L1 might not be a determinant factor. Mitkovska & Bužarovska (2018) analyze transfer from L1 Macedonian to L2 English with a special focus on the subject pronoun realization (i.e. both referential and nonreferential). All the participants are prepuberty learners aged between 8 and 15 and they are distributed into four proficiency groups (beginners, elementary, pre-intermediate and upperintermediate). Their data show subject omission cases in the initial stages, but as the proficiency level of the speakers increases the ungrammaticality rate in subject production decreases. In the more advanced stages, omission is mostly found within non-referential subjects. These authors attribute the omission to L1 transfer which is modulated by proficiency. However, the fact that illicit null subjects are found at all proficiency levels under study is an indicator that L1 influence persists in the L2 acquisition process to a higher or lesser degree. From the two previous sets of studies on typologically different languages in contact, the following conclusions can be drawn. If the speakers’ L1 is a [-null subject] language, overt subjects tend to be overproduced in the L2 because of negative transfer and because of the complexity of the interfaces given that the use of explicit subjects is grammatical in the L2 but regulated at the syntax-pragmatics interface. If the speakers’ L1 is a [+null subject] language, illicit null subjects can appear and this production is attributed to L1 transfer. Even if the illicit null subject rate does not seem to be very high, residual cases seem to always appear and be modulated by proficiency. In the case of 2L1 acquisition research, a proposal has been put forward to account for the presence as well as for the directionality and effect of cross-linguistic influence that we would like to adapt to L2 contexts. Following Holmberg’s (2005) and Sheehan’s (2006) 54 proposal, a [+null subject] language like Spanish is considered the superset language when compared to a [-null subject] language like English. This is so because Spanish has two realizations of the subject (the phonologically realized option and the phonologically null option) whereas English has one (the phonologically realized one). This phonological realization is the common option as it is shared by both languages and it is the marked option in Spanish (as null pronouns are less marked than overt pronouns). Based on the so-called lexical specialization approach, Liceras & Fernández Fuertes (2019) propose that the superset language does not receive cross-linguistic influence from the subset language (i.e. no transfer from English into Spanish would occur). Rather, the superset language is the one causing cross-linguistic influence (i.e. transfer from Spanish into English) and with a specific effect: acceleration in the development of the overt subject requirement in 2L1 bilingual English. Their study analyzes and compares the Spanish and the English subject omission and production rates in the naturalistic data of the 2L1 English Spanish bilingual twins from the FerFuLice corpus (age range 1;10-2;11) and the adults who interact with them, as it appears in CHILDES (MacWhinney 2000). They focus on cross-linguistic influence that might arise when English and Spanish are in contact. In the case of Spanish, when it comes to null subjects, the bilinguals’ (≈73%) and the monolingual’s (70%) rates are quite comparable. These results indicate that there is no crosslinguistic influence from English into Spanish, because the bilinguals do not produce less null subjects (i.e. something that could be the result of cross-linguistic influence from English). In fact, the bilinguals’ production is actually higher than that of the monolingual’s in this respect. In the case of English, a higher omission rate could be expected, if the unmarked option of Spanish (i.e. the null subject) is transferred, which will help reinforce subject 55 omission in the so-called omission stage monolinguals also go through. Alternatively, and if Spanish works as a facilitator, as Liceras & Fernández Fuertes argue, the null subject rate could be lower than what is the norm in the early stages in monolingual speech. This would involve that the obligatory overt subject requirement in English is set earlier in bilingual than in monolingual speech. Their data, in fact, go in this last direction since more pronominal subjects are produced by the bilinguals (63%) when compared to the monolinguals (44%). This result confirms their initial hypothesis in that there is transfer from the superset language (i.e. Spanish) into the subset language (i.e. English) and that this transfer has a positive effect in that it makes the bilinguals reach the adult requirement sooner than monolinguals. While in 2L1 acquisition both languages have the same status as L1s, in the L2 acquisition process, the L2 learner seems to depart from his L1 knowledge and so the L1 tends to influence the L2. Nonetheless, if we apply Liceras & Fernández Fuertes’ (2019) proposal and adapt it to the L2 acquisition of sentential subjects, transfer would be expected from the L1 superset language (i.e. Spanish) into the L2 (i.e. English) and would have a positive effect. This will involve that these learners will use one of the options available in Spanish (i.e. the overt subject) as a reinforcement for the only available option in their L2 English, resulting in positive transfer. In fact this is what previous studies seem to suggest in that no transfer from the null subject L1 into English takes place (e.g. Park 2004 and Pladevall Ballester 2012, 2016). Even though many studies have dealt with the two opposite values of the null subject parameter in contact, not many have included languages such as Bosnian. Given the lack of studies based on L1 Bosnian, this dissertation seeks to contribute to fill this gap. 56 3.5.2 Contact between two [+null subject] languages L2 acquisition research on subjects involving the interaction between two [+null subject] languages also discusses the role of L1 transfer and proficiency effects (Bini 1993; Margaza & Bel 2006; Sorace & Filiaci 2006; Bel et al. 2016; Lozano 2018, among others). This language pair, although different from the one under consideration in this dissertation, is relevant for our analysis in that studies on languages with the same option of the parameter predict no L1 transfer effects i) because both languages behave in the same way in their availability of null subjects, as per the Interference Hypothesis, and ii) because both languages have two sets of subjects (rich verbal agreement inflection and overt subject pronouns), as per the lexical specialization approach (Fernández Fuertes & Liceras 2018 and Liceras & Fernández Fuertes 2019). In their studies, Bini (1993), Sorace & Filiaci (2006), Margaza & Bel (2006) and Lozano (2018) find non-native-like production related to the syntax-pragmatics interface. All participants understand very quickly that null subjects are licensed grammatically, but they find difficulties in acquiring the syntax-pragmatics interface conditions that regulate the presence of overt pronominal subjects. Proficiency level affects adequacy in subject production so that, even if syntax is at place from very early stages, pragmatics continues being problematic for advanced speakers, even when the two languages share the same parametric option. In fact, the same pragmatic conditions regulate the distribution of overt and null subjects in the languages under analysis (L1 Spanish L2 Italian in Bini’s study; L1 Italian L2 Spanish in Sorace & Filiaci’s study; and L1 Greek L2 Spanish in Margaza & Bel’s and Lozano’s studies). The pragmatic interface conditions are the ones that govern the felicitous use of both subject types (i.e. null and overt pronominals). The overproduction in this case does not seem to be related to L1 transfer but is rather considered a default 57 (unmarked) option. In all four studies, as proficiency increases, the learners’ production becomes more native-like, but even the most advanced learners show overuse of overt pronominal subjects in contexts where a native would use a null subject. The findings in these studies lead to the conclusion that typological proximity and typological similarity do not seem to be a facilitating factor, especially not when it comes to the syntax-pragmatics interface. As a consequence, L1 positive transfer might not always take place or not fully so. Studies on two [+null subject] languages in contact seem to be mainly conducted on L2 adult speakers. The conclusions that can be drawn are the following: i) proficiency seems to play a crucial role; the more advanced speakers are, the more native-like their production is; and ii) the overproduction of overt subjects is interpreted as related to interface vulnerability and not to typological proximity or typological similarity. 3.5.3 Contact between two [-null subject] languages Comparatively, a very small amount of L2 acquisition research on subjects has been conducted on the interaction between two [-null subject] languages (White 1985; Liceras 1989; Liceras & Alba de la Fuente 2015 and Mujcinovic 2015). White’s (1985) study analyzes the contact between L1 Spanish and L2 English (i.e. [+null subject] and [-null subject] languages respectively), but she compares this data set to data from L1 French L2 English speakers (i.e. both being [-null subject] languages, that is typologically similar). In the case of the L1 French group, subject omission cases are correctly judged as ungrammatical making this group pattern with the control group, as opposed to the L1 Spanish group. Furthermore, in the case of overt subjects evidence of positive transfer in felicitous judgments is found. Therefore, she concludes that typological 58 similarity plays a role in L2 acquisition, because L1 French and L1 Spanish speakers differ in their L2 English judgments (in spite of French and Spanish being typologically proximate languages). Making reference to White (1985) and referring back to Liceras (1989), Liceras & Alba de la Fuente (2015), also analyze French and English as typologically similar languages in the sense that both are under the [-null subject] option of the Null Subject Parameter. French and Spanish, on the other hand, are typologically proximate in the sense that they are both Romance and synthetic languages. Liceras (1989) and Liceras & Alba de la Fuente (2015) compare L1 French L2 Spanish learners and L1 English L2 Spanish learners. Their data show that French verbal morphology seems to have a facilitating role in the acquisition of L2 Spanish when compared to English, since verbal morphology in English is poor. In this case, typological proximity between French and Spanish (i.e. both being synthetic languages and French having verbal agreement markers although not with the same value as the Spanish ones) seems to be a conditioning factor in the acquisition of L2 Spanish. Both studies conclude that “typological proximity may supersede the fact that these two languages [Spanish and French] differ in terms of the microparameters (or properties) associated to the Null Subject Parameter” (Liceras & Alba de la Fuente 2015:353). Thus, they argue that since French is of special character, more sophisticated linguistic analyses must be done in order to get a more refined view on the interaction between typological proximity and typological similarity between the languages involved. Mujcinovic (2015) analyzes production data of L1 Danish L2 English speakers. Her data show a preference for overt subjects over null ones, thus adhering to the L2 overtness requirement. As for the production of null subjects, a very low rate is produced by these learners and, out of this, less than 2% of these null subjects are in fact non-native-like. As for 59 the adequacy of overt subjects, an overproduction of DPs over overt pronominal subjects is found and classified as redundancy errors related to the task being used to elicit the data (between 27.3% and 0.8%). The results show an overall native-like performance, which is attributed to typological similarity and positive transfer. Given the scarcity of works on typologically similar languages and, in particular, on studies that consider two [-null subject] languages in contact, this dissertation also seeks to contribute to fill this gap. 3.6 Summary In this chapter the properties that define L2 acquisition have been reviewed with a focus on i) the effects of L1 transfer; ii) the role played by age of exposure; and iii) the role played by input. L1 transfer seems to be dependent on the typological similarity between the languages in contact (i.e. the L1 and the L2), which contributes to a twofold classification of transfer: positive and negative. A connection is also found between proficiency and transfer: as proficiency increases transfer related errors decrease. In addition, L1 transfer is also related to the interfaces involved, being the syntax-pragmatics interface the most vulnerable in this case. In view of these interactions, and as we are concerned in the present study with a parametric property (i.e. the availability, or lack therefore, of null subjects), we explore the effects of transfer as well as the role played by proficiency and the vulnerability of the syntaxpragmatics interface. 60 As for age-related issues, the critical age is a topic that has received a lot of attention within L2 acquisition research. No real agreement has been reached as to where the exact line has to be drawn between early L2 and late L2 acquisition, because different sensitive periods are appreciated depending on the specific issue under analysis, along with the maturational state and cognitive development of the participants involved, among others. Thus, many opt for considering puberty as the end of the critical period, whereas others question its existence to begin with. For the present study, we rely on Meisel’s (2008) account of sensitive periods rather than using a fixed chronology to determine the beginning and end of child L2 acquisition. To classify L2 bilinguals, Meisel’s (2008) trifold division is followed in combination with what DeKeyser et al. (2010) and Muñoz & Singleton (2011) propose. That is, children that start learning an L2 between the ages of 4 and 8 are considered as early child L2 bilinguals; when this happens between the age of 8 and puberty, they are referred to as late child L2 bilinguals, while those who start learning an L2 after the age of 15-16 are considered to be adult L2 bilinguals. Therefore, this study is concerned with early child L2 bilinguals. As of input, the L2 acquisition process is constraint by both the type of input speakers are exposed to (both in a naturalistic setting as well as in an institutional context), as well as by input quantity, quality and formality. These issues have been addressed in different ways in previous studies concerned with the analysis of sentential subjects in speakers with L1s and L2s presenting different parametric options. L1 transfer is closely connected to typological similarity, because the closer the L1 and the L2 are linguistically speaking, the less negative transfer is expected to be found. However, this statement only holds for morphosyntactic issues. Regardless of 61 typological similarity, the syntax-pragmatics interface seems to be problematic until very advanced stages of proficiency even in the case of typologically similar languages. Since both morphosyntactic and pragmatic properties have to be acquired for a native-like production to occur in L2 acquisition, this makes L2 speakers of typologically similar languages equally vulnerable to discourse conditions. If [-null subject] and [+null subject] languages are in contact, then negative transfer in the form of overproduction of either overt or null subjects is expected. In addition, interface complexity, as mentioned above, is also present, because even near-native speakers are said to experience difficulties to fully acquire the properties that are required in this interplay. An alternative interpretation of transfer is based on the superset-subset theory, the so-called lexical specialization approach (Fernández Fuertes & Liceras 2018 and Liceras & Fernández Fuertes 2019), based on the fact that, if the L1 is the superset language, it will not cause negative but positive transfer reinforcing the only available option in the L2 (i.e. the overt subject in this case). 62 CHAPTER 4: METHODOLOGY The aim of this chapter is to present the research methodology used to elicit the data that constitute the core of the present dissertation. In particular, information pertaining to both the participants as well as the experimental tasks that have been used is provided. As it has been specified in the previous chapters, this study analyzes the nature of sentential subjects as produced by L2 speakers of English with different L1 backgrounds classified according to language typology: [+null subject] languages such as Spanish and Bosnian and [-null subject] languages such as Danish and English. Following this idea, participants are divided into three experimental groups, depending on their L1, and a control group. Therefore, the experimental English data come from L1 Spanish, L1 Danish and L1 Bosnian speakers and they are compared to those of L1 English speakers. For all the experimental groups, participants have only received L2 English instruction in the primary schools they are attending. The control group consists of L1 English participants from Calgary, Canada. In order to collect data on English sentential subjects, both oral and written tasks were conducted. The oral task is a semi-guided interview, while the written task is a picture sequence narration. All the selected participants completed both tasks. The chapter is organized as follows: section 4.1 deals with the description of the participants and the selection criteria that have been applied for each language group. Sections 4.2 and 4.3 provide information about the data protocol, the experiments and the data elicitation process itself. Section 4.4 illustrates the transcribing and coding procedures, followed by a description of the statistical methods used for the data analysis in section 4.5. Finally, in section 4.6 a summary of this chapter is provided. 69 Unsworth 2008; Unsworth & Blom 2010; Hawkins & Filipovic 2012; Lundell & Lindqvist 2012, 2014, among others). Therefore, we have also measured the participant’s proficiency in the L2 in terms of MLUw values. In figure 1 and table 3 below the MLUw values obtained by the participants in the 7 groups are indicated. Figure 1. English mean MLUw values per participant group group Spanishgroup 1 Spanishgroup 2 Bosniangroup 1 Bosniangroup 2 Danishgroup 1 Danishgroup 2 Control group Mean 5.585 6.561 4.243 5.154 6.607 7.006 6.241 SD 0.99 0.75 0.75 0.88 1.38 1.11 1.18 Table 3: Descriptive statistics for English MLUw values per participant group To determine whether there are any significant differences between the groups, a oneway ANOVA was conducted after confirming homogeneity of variance of the data (F(3,87)=0.697, p=.556). The results show that there is an effect for group (F(1,6)=11.63, p<.001). Within each group, the participants who have been instructed for a longer period in their L2 have a higher MLUw. Thus, it can be argued that the participants who have been 70 exposed longer to L2 English have more developed language skills and are, therefore, more proficient in this sense. However, this increase between the participants in groups 1 and those in groups 2 is found to be significant for the L1 Spanish and L1 Bosnian groups (p<.010) but not for the L1 Danish groups (p=.425). For group 1 participants, an across groups comparison shows significant differences between the L1 Bosnian and the L1 Spanish groups (p=.027), the L1 Bosnian and the L1 Danish groups (p<.001) and the L1 Bosnian and the L1 English groups (p<.001). For group 2 participants, an across groups comparison shows significant differences between the L1 Bosnian and the L1 Spanish (p=.016) and the L1 Bosnian and the L1 Danish (p<.001). None of the L2 groups is statistically different from the L1 English group. These results point towards an overall difference in MLUw terms between groups 1 and groups 2, except for the L1 Danish group. In this last case, however, information on data dispersion, as in figure 1, shows that indeed a difference in variability is seen in the L1 Danish group, where group 2 is more homogeneous than group 1. Across language groups, only the L1 Bosnian groups significantly differ from the rest of the L2 groups and from the control group (i.e. the L1 English group). To sum up, the participants were chosen on the basis of their language background, the time of instruction they have received in L2 English and the fact that they have not spent a long period of time in an English-speaking country. Their MLUw values were included as proficiency measures and they show that across groups of exposure (group 1 vs. group 2) differences are found where the participants in groups 1 show a lower MLUw than those in groups 2. More information about each of the groups appear in the subsequent sections. 71 4.1.4 L1 Spanish groups The L2 English data from the L1 Spanish participants were recorded during the month of June 2015. These participants come from a school located in Valladolid (Spain), Colegio Ave María. The compulsory Spanish education system follows a model, where the children attend primary school from the age of 6 to 12 followed by secondary school between the age of 12 to 16. Infant education is not compulsory. This semi-private primary school has implemented the CLIL (Content and Language Integrated Learning) methodology, which involves teaching part of the curriculum by using the second language instead of the students’ L1. In the case of the Colegio Ave María, two subjects in the curriculum are taught in English: Natural and Social Sciences and Arts as part of the CLIL program 11 . During these classes, not only is the content taught in English, but some reference to English grammar is included, too. That is, on the one hand, these participants receive direct instruction in English during the traditional English language classes (as in a school subject) and, on the other hand, they receive indirect instruction in English in the Natural and Social Sciences and Arts classes. The participants in this study were selected according to the time of instruction in English they have had, that is, 2 and 4 years. The ones who have been exposed to English during a period of 2 years were about 9 years old (grade 3) and the ones that have been so during a period of 4 years were about 11 years old (grade 5). The periods dedicated specifically to the teaching of English as an L2 are two per week for both groups, where each period lasts for one hour. Since these were CLIL students, 11 CLIL promotes linguistic competence and at the same time stimulates cognitive flexibility (Coyle et al. 2010:10). Lasagabaster & Sierra (2009) claim that the CLIL approach provides more exposure to real language usage of the target language and thereby strengthens the ability to process the L2. To be more specific, the L2 is used for teaching curricular content under the same conditions as the L1. 72 some of the content subjects were also taught in English. Thus, the overall input in English was on average 6.5 hours per week. The participants that have been learning English for 2 years have received a total of 455 hours of instruction and those who have been learning English for 4 years have received a total of 910 hours of instruction. The communication between students and teachers during the English language class and the English content classes was entirely in English. The rest of the subjects were taught in Spanish which was the language children were exposed to and used outside of the school. 4.1.5 L1 Bosnian groups The L2 English data from the L1 Bosnian participants were recorded during the month of October 2014. These participants come from a public Bosnian primary school called Aleksa Šantića located in Banja Luka (Bosnia and Hercegovina). The curricular program follows the 9-year model, where the children start the primary school at the age of 6. The division is made following three cycles: grades 1 to 3 (preparatory); grades 4 to 6 (classroom instruction) and grades 7 to 9 (subject instruction). Two foreign languages are included in the curriculum, where the first to be learned is English (introduced in grade 3) and the second foreign language is usually German (introduced in grade 6). Two groups (grade 5 (ca. 10 years old) and grade 7 (ca. 12 years old)) were selected according to the time of English instruction they have had, that is, 2 and 4 years. Until grade 4, two periods are dedicated to the first foreign language; in grade 5 it is increased to four periods and in grade 6 they are reduced again to two periods because of the introduction of the second foreign language. Each teaching period lasts for 45 minutes so that the participants received 1.5 hours of English institutional instruction per week during the first 2 years, three 73 hours in the third year and one hour and a half during the fourth year. The participants that have been learning English for 2 years have received a total of 120 hours of instruction and those who have been learning English for 4 years have received a total of 300 hours of instruction. The rest of the subjects were taught in Bosnian which was the language children were exposed to and used outside of the school. 4.1.6 L1 Danish groups The L2 English data from the L1 Danish participants from the Pedersborg skole in Sorø (Denmark) were recorded during the months of November and December 2014. Pedersborg skole is a public Danish primary school from grade 1 (age +/- 6) to grades 9 or 10 (age +/- 16) 12 . This academic period is subdivided as follows: primary education (grades 1 to 6) and lower secondary education (grades 7 to 9/10). Pedersborg skole is one of many Danish schools that no longer uses printed books in the classrooms. All the pupils in the school are provided with an iPad that is exclusively used for educational purposes through different educational platforms that are available for primary school teachers such as CFU (Center for Undervisningsmidler Danmark, ‘Center for Teaching Materials Denmark’). From time to time some books are also used, but they are not the primary source of teaching as is generally the case of Bosnian or Spanish schools. Two foreign languages are included in the curriculum, where the first to be learned is English (from grade 3) and the second foreign language is German or French (from grade 7). The communication between students and teachers during the English language class is 12 In Denmark, grade 10 is an optional course. It is specially designed for students who want to go to a lower secondary independent boarding school or students who are still not ready for secondary school. This is a popular option among students in Denmark. 74 entirely in English, while the rest of the time the participants spend at school, they speak their L1 (i.e. Danish). Two groups (grade 5 and 7) were selected according to the time of instruction they have received in English, that is, 2 and 4 years. The participants that have been instructed in English during a period of 2 years were about 10 years old and the participants that were so for a period of 4 years were about 12 years old. Each teaching period lasts for 45 minutes and so the participants received one and a half hour of institutional instruction in English per week during the first 2 years and 2 hours and 15 minutes during the 3rd and 4th year. The participants that have been learning English for 2 years have received a total of 120 hours of instruction and those who have been learning English for 4 years have received a total of 300 hours of instruction. The rest of the subjects are taught in Danish which is the language children are exposed to and use outside of the school. 4.1.7 Control group The control group consists of L1 English participants from the St. John Paul II school in Calgary (Canada). The participants selected are in grade 6 and are 10-11 years old. The data obtained from the control group (both oral and written) were recorded during the month of December 2016. The reason why the control group involves students in Canada deals with the availability to collect the data. Nowadays, monolinguals are rare, especially within a country such as Canada. Following the data available for 2016, 90.5% of the population in Calgary speak English and 67.8% have English as their L1 13 . Only 1.5% are L1 French speakers and the remaining 13 Census 2016, Statistics Canada: https://www.calgaryeconomicdevelopment.com/research-andreports/demographics-lp/languages/ 75 33.7% are L1 speakers of other non-official languages in Canada, being Tagalog the most spoken one. The speakers who participated in this study were all born in Canada. They are all considered monolinguals (i.e. they live in an L1 environment at home and never speak another language). They have never lived in a setting where other languages are spoken, except for short holiday stays. 4.2 Data protocol & fieldwork Data samples were collected in primary schools situated in Spain, Denmark, Bosnia and Canada. Before collecting the data, the school directors were contacted and the investigator was allowed to get in contact with the teachers and the parents of the specific groups of children that were to be recorded. Since the participants were under the age of consent, the parents were asked to sign a consent form which followed the guidelines established by the University of Valladolid Research Ethics Board, thereby given explicit written consent to the investigator to take data from their children through the use of two linguistic tasks. Then the English teachers were contacted and a brief meeting was held before the investigators were introduced to the participants in order to inform about the research to be conducted. Also, a small presentation on language acquisition and language learning was given to the schoolboard, teachers, parents, and the participants to contextualize the study, to stress the importance of their participation and to give them the opportunity to ask any questions 76 they may have. No mention was made regarding the specific research topic or the structures under analysis in the present investigation as reflected in the two tasks. Warming-up sessions with the participants also took place at the schools following the indications in the literature (e.g. McDaniel et al. 1995; Thornton 1996; Rice et al. 1999; Unsworth 2005, among others). That is, some time was spent with the participants in some of the English classes in the school so to avoid the observers’ paradox (Labov 1972) (i.e. participants do get influenced by the presence of people they are not familiar with and this can influence their production) and for the participants to comfortably and adequately do the experiment with the researcher (Nortier 2008:42). The investigator was introduced to the children as a language researcher, who is interested in understanding their ability to learn English and it was highly emphasized that she was not their teacher and that the participants were not being examined, since this seemed to be of great concern for some of them. They were conscious of the fact that the investigator could speak their L1, but they were told to use only English to perform the different tasks, although they were allowed to ask for vocabulary. For the oral task, the role of the investigator was to make the participants understand the task and thereby make them produce full sentences by avoiding asking yes-no questions and guiding the participant towards more elaborated answers. For the written task, the role of the investigator was to help the participants understand the task and guide the participants to write stories that contained as much information as possible of the actions being performed by the story characters using full sentences. 77 4.3 Experiments & data collection In order to obtain the data, two tasks have been designed to elicit oral and written production data through experimental and semi-spontaneous procedures as described below. 4.3.1 Oral task Semi-spontaneous production data have been elicited via an oral semi-guided interview. The participants have been interviewed individually and voice recorded. The total duration of the interviews selected for the study is 16 hours and 20 minutes, where each individual interview ranged from 10 to 16 minutes maximum. A protocol was designed to ensure uniformity across participants and across groups as well as to encourage a more naturalistic speech, where the topics proposed by the researcher in most cases were related to the participants’ family, hobbies, interests, school, preferences, music, friends, etc. They were also encouraged to talk about any topic of their choice, but most of them only answered the questions asked. The interviewed participants were allowed to ask for vocabulary, which was always provided to them in the most grammatically neutral form (e.g. verbs were provided in infinitive, nouns in singular, etc.). To encourage their part-taking, they were praised throughout the whole session. The setting in which the experiment was conducted was known to the participants, since it was as small room or a classroom in the institution where they studied. It was a quiet and well-lighted place, and no interruptions were made during the recordings. 78 4.3.2 Written task The experimental written production data have been elicited via a wordless picture sequence task adapted from the A1-ball story from the Edmonton Narrative Norms Instrument (ENNI) (Schneider et al. 2005) (see appendix I) 14 . The A1-ball story illustrates how two characters, a male giraffe and a female elephant, play with a ball near a swimming pool. Accidentally, the giraffe drops the ball and it falls into the water. The elephant jumps, fetches the ball and gives it back to the giraffe, who appears to be very happy. The changes that have been made to the original ENNI story are related to the characters and their biological gender. Both Spanish and Bosnian are [+gender] languages, and, so, nouns have grammatical gender. In Spanish this division is twofold (i.e. masculine and feminine), whereas in Bosnian it is trifold (i.e. masculine, feminine and neuter). Following Harris (1991) and Roca (1989, 2005, 2006), and Fernández Fuertes et al. (2016) gender is divided into four different domains: semantics, morphology, phonology and syntax 15 . A brief explanation focusing on the point of interest for this dissertation is provided here. What regards semantics, [+animate] nouns have the features [+male] or [+female] in accordance with their biological gender. Gender is semantically arbitrary both in Spanish and in Bosnian. What regards morphology, nouns include word markers, which are the final suffixes that can contribute to transparent gender identification and can, at the same time, be condensed to phonological features that phonetically distinguish between word markers. 14 The ENNI is a language assessment tool for children developed at the University of Alberta, Canada, and freely available at http://www.rehabresearch.ualberta.ca/enni/. 15 It is not our aim to discuss thoroughly these domains in this dissertation. An exhaustive account is provided by the authors. 85 analyzed because they did not produce one single complete sentence. After 2 years of instruction, participants started producing complete sentences and, accordingly, the production of subjects could be analyzed. In order to see a development in production, the next set of data collected was from participants who have been instructed in English for 4 years. The data collected from this pilot study have not been used in this dissertation, but some amendments were made to the way the tasks were implemented. The second pilot study was run on English heritage speakers from the International School in Valladolid. It was run following exactly the same procedure that have later been used for the data collection for this dissertation. Regarding the written task, the purpose of this pilot study in particular was to help establish the number of times that the picture sequences were to be projected. The participants felt most comfortable with three projections. They stated that it helped them mentally to create a story while the pictures were being projected and, then, they only had to write the story on the sheet of paper provided. Both the participants from the private language school and the heritage speakers did both tasks, the oral one and the written one. 4.4 Transcription & coding procedure Both the oral and the written production tasks have been transcribed in CHAT (Codes for the Human Analysis of Transcripts) format and analyzed using the CLAN (Computerized Language ANalysis) software. Both CHAT and CLAN are the resource tools used in CHILDES (MacWhinney 2000) and TalkBank (MacWhinney 2019). CHAT is a transcription format used for transcribing linguistic data. Data transcribed in CHAT can be analyzed by using a series of inbuilt programs, which form part of CLAN. These programs 86 facilitate different automatic calculations and searches on selected CHAT data (for example the MLU program is used to calculate the MLUm or the MLUw). The use of this specific format was also chosen, because our intention is to contribute these data to TalkBank. Once the data for each participant and for each task were transcribed, sentential subjects were isolated and classified in terms of i) form, ii) grammaticality, iii) S-V agreement, and iv) adequacy in terms of referentiality, as shown in table 4 and as described below. Form DPs proper names overt pronouns null pronouns grammatical ungrammatical Agreement grammatical person & number ungrammatical non-inflected form: omission of 3rd person -s marker non-inflected form: infinitive use of singular forms for plural use of plural forms for singular use of 3rd person for 1st person null lexical verb null auxiliary verb null past tense null copula Adequacy adequate DP used for reference introduction DP used for reference reintroduction null pronoun used for reference maintenance pronoun used for reference maintenance non-adequate DP used for reference maintenance pronoun used reference introduction pronoun used for reference reintroduction Table 4: Data codification variables In terms of form, sentential subjects were codified into five categories as illustrated in examples 36 to 40: 36) *CHI: my mother is a chef [full DPs] (oral L1 Danish-group 2; SODKVIIB.13; 13 years) 87 37) *CHI: Tom jumped down in the pool [proper names] (written L1 Danish-group; SODKVC.02; 10 years) 38) *CHI: he found a ball [overt personal pronouns] (oral L1 Bosnian-group 2; BLBOVII2.01; 12 years) 39) *CHI: we paint and Ø 17 draw [null pronouns in coordinated structures] (oral L1 Spanish-group 2; VAESVA.09; 11 years) 40) *CHI: Ø adopt a baby lion [null ungrammatical personal pronouns] (written L1 Spanish-group 2; VAESVB10; 11 years) Being the focus set on referential sentential subjects, expletive subjects, as in 41, and null subjects in imperative constructions, as in 42, were excluded from the analysis. 41) *CHI: it was raining (oral L1 Danish-group 2; SODKVIIB.09; 13 years) 42) *CHI: Ø help me! (written L1 Danish-group 2; SODKVIIB.01; 11 years) In terms of grammaticality, sentential subjects were codified into two categories: grammatical (as in 36 to 39 above) and ungrammatical (e.g. null subjects that are not found in coordinated structures, as in 40). In this case only grammaticality of the subject was considered so that, if other mistakes appear in the sentence, they were disregarded and, as long as the subject was correct, the example was tagged as grammatical. Other factors that were not related to subject expression and that were not taken into consideration when determining grammaticality include the ones in 43 to 46: 43) *CHI: him have a job [correct/incorrect case assignment] (oral L1 Bosnian-group 2; BLBOVII2.10; 12 years) 44) *CHI: Marry plays with Tom [spelling mistakes] (written L1 Danish-group 1; SODKVIIB.02; 11 years) 45) *CHI: yes, I like Ø [omission of objects] (oral L1 Spanish-group 2; VAESVA.09; 11 years) 46) *CHI: and that baby was daymon [made-up words] (written L1 Bosnian-group 1; BLBOVII2.01; 10 years) 17 Ø is the symbol used to mark the omission of a category (i.e. subjects, auxiliary verbs, main verbs, etc.). 88 In terms of S-V agreement, both grammatical and ungrammatical S-V agreement was coded for. S-V agreement is grammatical when person and number are checked correctly as in adult native grammar. Lack of S-V agreement involves both omission cases and non-finite forms, as in 47 to 50: 47) *CHI: she work in the lab [omission of –s markers] (oral L1 Danish-group 2; SODKVIIB.08; 13 years) 48) *CHI: Tom be decide swimming [use of non-inflected forms] (written L1 Bosnian-group 1; BLBOV2.11, 10 years) 49) *CHI: I Ø elevens years old [omission of copula verbs] (oral L1 Spanish-group 2; VAESVA.03; 11 years) 50) *CHI: I Ø not like English [omission of auxiliary verbs] (oral L1 Bosnian-group 1; BLBOVII2.05; 10 years) In the case of adequacy, sentential subjects were also classified in terms of their pragmatic adequacy by relying on the three formal categories (i.e. DPs, overt pronouns and null pronouns) and their compliance with the adult native grammar rules in a given linguistic context. Thus, each subject form is classified according to its usage (either adequate or nonadequate) in relation to their referent as follows. DPs were adequate if used for referent introduction, that is, when the referent was introduced for the first time, as in 51; for referent reintroduction, when the referent was introduced beforehand, but a DP is needed to restate the referent; or for disambiguation, as in 52. 51) *CHI: one sunny day Mary and Tom go to pool (written L1 Bosnian-group 1; BLBOV2.14; 12 years) 52) Previous context: *CHI: Mary giraffe and Tom elephant was playing with ball *CHI: But the ball was fall in the water and they couldn’t take Example: *CHI: Tom elephant jumped to the water to tooks the ball (written L1 Spanish-group 2; VAESVB.09; 11 years) 89 DPs were inadequate if used for reference maintenance when the same DP is repeated where a pronoun is expected, as in 53. 53) Previous context: *CHI: Mary and Tom was out for a walk and they stop beside a little pool to play with a ball Example: *CHI: Mary and Tom plays with a ball (oral L1 Danish-group 2; SODKVC.01; 13 years) Overt pronouns were adequate if used for reference maintenance, as in 54. 54) Previous context: *CHI: but Tom Elephant jumped to the swimming pool and Ø took the ball Example: *CHI: he gave the ball to Mary Giraffe (written L1 Spanish-group 2; VAESVA.08; 11 years) Overt pronouns were not adequate if used for referent introduction, when the referent has not been previously mentioned, as in 55, 55) Previous context: there is no previous context as this is the first sentence produced by the participant Example: *CHI: they were very good friends. (written Control group; COCA12; 10 years) or when they are used for reference reintroduction that results in ambiguity because the referent cannot be identified. 56) Previous context: *CHI: the last we have done a bit of Guy Forks when he was a catholic terrorist. *CHI: they tried to explode the whole the House of Parliament. *CHI: while the king was inside. *CHI: he meant that the Bibel@s:dan should be read in English and not only Latin. Example: *CHI: so he decided to try to kill him (oral L1 Danish-group 1; SODKVC06; 11 years) 90 The referent for the overt pronoun used in 56 cannot be identified because various 3rd person singular masculine referents have been previously mentioned. In this case it is not clear if he refers to Guy Forks or to the king. Null pronouns were only adequate if used for reference maintenance in coordinated structures, as in 57. If a null pronoun is used in any other context, it was codified as ungrammatical, as in 58. 57) *CHI: I can go for a walk with a dog and Ø play with a dog (oral L1 Danish-group; SODKVC.13; 11 years) 58) *CHI: today Ø are very happy (oral L1 Spanish-group 1; VAESIIIB.12; 8 years) Other examples that have been excluded from the study include cases like the ones in 59-62: 59) *CHI: Mary xxx to get out [incomplete sentences] (oral L1 English group; COCA.15; 12 years) 60) *CHI: ovaj@s:bos sing (.) sings [codeswitching involving subjects] (oral L1 Bosnian-group 1; BLBOV2.15; 11 years) 61) *CHI: who was happy [wh-pronouns in subject position] (oral L1 Bosnian-group 2; BLBOVII2.02; 13 years) 62) *CHI: you are welcome [fixed expressions] (oral L1 Danish-group 2; BLBOVII2.13, 13 years) These examples were excluded either because i) they did not provide sufficient information regarding the form or the referent of the subject; ii) they were not entirely in English (i.e. they included code-switching); iii) they involve other mechanisms that can interact with word-order (i.e. the use of whpronouns in subject position); and iv) they were not instances of the participants’ productive language. 91 4.5 Statistical methods for data analyses The statistical analysis was conducted in the following manner. Different statistical tests have been run using R, version 3.6.2 (R Core Team, 2017). As the results from the Levene’s Test for homogeneity of variance showed lack of homoscedasticity, two GLMMs (General Linear Mixed Models) were fitted using the lme() function in the lmerTest package; one test was fitted for the grammaticality of subjects (p<.001) and another test for the adequacy of the grammatical subjects (p<.001). The omnibus ANOVA tests conducted are based on the GMMLs and they are reported without the effect size, because, due to the way that variance is partitioned in GLMMs, there is no agreement as to how to calculate standard effect sizes for individual model terms such as main effects or interactions (Richardson 2011; Singmann & Kellen 2015 and Rights & Sterba 2019). However, a GLMM that permits the introduction of random effects to create the model was run with a view to refining the sources of variance. Therefore, the main focus is on the fixed effects and their interactions. For grammaticality, the amount of grammatical subjects produced was calculated in percentages before introducing them as response variables in the model. L1 with four levels (i.e. Spanish, Bosnian, Danish and English), time of instruction (i.e. 2 years and 4 years) and modality (i.e. oral and written) were used as fixed effects variables. Participant was introduced as a variable of random effects which accounts for 14% of the overall variance. In order to explore the differences between groups, the F ratio values were obtained from the omnibus ANOVA tests. When significant results for main effects and interaction effects were observed, the follow-up pairwise comparisons were conducted adjusting the p-values with the Bonferroni method. 92 For adequacy the same procedure was followed. The amount of adequate subjects produced was calculated in percentages from the total of overall grammatical subjects before introducing them as response variables in the model. L1 with four levels (i.e. Spanish, Bosnian, Danish and English), time of instruction (i.e. 2 years and 4 years) and modality (i.e. oral and written) were used as fixed effects variables. Participant was introduced as a variable of random effects, although the results did not detect this variable as a source of variability of the response variable. In order to explore the differences between groups, the F ratio values were obtained from the omnibus ANOVA tests. When significant results for main effects and interaction effects were observed, the follow-up pairwise comparisons were conducted adjusting the p-values with the Bonferroni method. Once again, in order to obtain results related to the preference of subject type (i.e. overt, grammatical-null and ungrammatical-null), the overall subject production was calculated in percentages: L1 with four levels (i.e. Spanish, Bosnia, Danish and English), time of instruction (i.e. 2 years and 4 years), modality (i.e. oral and written) and subject type (i.e. overt, grammatical-null y ungrammatical-null) were used as fixed effects variables. Participant was introduced as a variable of random effects, which account for 4% of the overall variance. In order to explore the differences between groups, the F ratio values were obtained from the omnibus ANOVA tests. When significant results for main effects and interaction effects were observed, the follow-up pairwise comparisons were conducted adjusting the p-values with the Bonferroni method. 4.6 Summary This chapter has outlined the methodology followed to implement this study. Information has been provided regarding i) the participants (including the selection criteria 93 used and their linguistic profile), ii) the tasks used to elicit the oral and written production data, iii) the extraction and codification procedures used, and iv) the statistical analyses conducted. The participants that took part in this study are divided into three groups depending on their L1 (i.e. L1 Spanish, L1 Bosnian and L1 Danish). All of these participants are L2 English speakers. They were further divided into two subgroups depending on the time of instruction they have received in L2 English (i.e. 2 years and 4 years). A summary of the linguistic profile of the different participant groups is provided in table 5 below where years/hours of instruction correspond to L2 English instruction at school. Table 5: Summary of the participants’ linguistic profile To elicit data from these participants, two different task modalities were used: oral and written. The overall number of utterances obtained from the participants is 11,196. Nonetheless, since the aim of this dissertation is to account for sentential subjects, only full sentences were analyzed. Therefore, a total amount of 6,051 tokens (i.e. sentences) constitute the corpus of analysis for the present study and were compiled and codified in an Excel spread sheet (v. 2013). The Excel database was later exported to R, version 3.6.2 (R Core Team, 2017), in order to run the statistical analyses. Results obtained and the statistical analyses implemented are discussed in chapter 6. group age years of instruction hours of instruction mean MLUw Spanish-group 1 9-10 2 455 5.585 Spanish-group 2 11-12 4 910 6.561 Bosnian-group 1 10-11 2 120 4.243 Bosnian-group 2 11-12 4 300 5.154 Danish-group 1 10-11 2 120 6.607 Danish-group 2 11-12 4 300 7.006 Control group 10-11 n/a n/a 6.241 94 CHAPTER 5: RESEARCH QUESTIONS & HYPOTHESES In this chapter, the aim of the study in form of research questions and hypotheses is put forward. Section 5.1 deals with the main research questions, whereas section 5.2 deals with the hypotheses that are related to these research questions. Subsequently, the potential outputs are presented in order to further explain and specify the issues that might arise in relation to each specific research question and hypothesis at stake. 5.1 Research questions This dissertation deals with the effects of transfer that may surface in relation to sentential subjects when typologically similar or typologically different languages are in contact in an L2 English context. The aim is to account for how the oral and the written production of the L2 English speakers might be influenced, either positively or negatively, by typological similarity and time of instruction in L2 English. We particularly seek to answer i) whether the availability of both null and overt subjects in the participants’ [+null subject] L1s has an effect on the production of sentential subjects in L2 English and ii) whether the overt subject requirement in the participants’ [-null subject] L1s has an effect on the production of sentential subjects in L2 English. That is, the focus is placed on how typological similarity of a specific linguistic property affects the oral and written production of L2 English sentential subjects. In this line and in the light of the literature previously considered (chapters 2 and 3), the research questions that have guided this research are the following: i) What is the role, if any, played by typological similarity? ii) What is the role, if any, played by the different availability of subject types across languages? 101 (F(3,84)=13.662, p<.001) and time of instruction (F(1,84)=8.478, p=.004). A three-way interaction effect is observed between modality, L1 and time of instruction (F(2,84)=4.182, p=.018). Fixed effects19 estimate SE t df p (Intercept) 54.792 3.667 14.943 168.000 <0.001 modality oral 34.910 5.185 6.732 168.000 <0.001 L1 Danish 25.511 5.185 4.920 168.000 <0.001 L1 Spanish 20.061 5.185 3.869 168.000 0.000 L1 English 30.846 5.185 5.949 168.000 <0.001 time of instruction 4 years 34.555 5.185 6.664 168.000 <0.001 modality oral: L1 Danish - 17.973 7.333 - 2.451 168.000 0.015 modality oral: L1 Spanish - 12.693 7.333 - 1.731 168.000 0.085 modality oral: L1 English - 20.548 7.333 - 2.802 168.000 0.005 modality oral: time of instruction 4 years - 30.595 7.333 - 4.172 168.000 <0.001 L1 Danish: time of instruction 4 years - 39.012 7.333 - 5.320 168.000 <0.001 L1 Spanish: time of instruction 4 years - 54.846 7.333 - 7.479 168.000 <0.001 modality oral: L1 Danish: time of instruction 4 years 37.317 10.371 3.598 168.000 0.000 modality oral: L1 Spanish: time of instruction 4 years 51.158 10.371 4.933 168.000 <0.001 Table 7: Summary of the GLMM fixed effects for adequacy As of adequacy, a summary of the results appears in table 7. From this model, an omnibus ANOVA test was run to detect main and interaction effects. The main effects are significant for modality (F(1,168)=112.005, p<.001) and for L1 (F(3,168)=7.199, p=.000), but not for time of instruction (F(1,168)=1.6473, p=.201). Nonetheless, time of instruction was significant in interaction with modality and L1 (i.e. a significant three-way interaction is observed between modality, L1 and time of instruction; F(2,168)=13.02, p<.001). Apart from showing different interaction effects, the summaries of the GLMMs (table 6 and table 7) also show an effect in the analyses conducted for L1, modality and time of 19 The reference parameters that the GLMM used for the fixed effects are the following: i) for modality: written; ii) for L1: Bosnian; and iii) for time of instruction: 2 years. 102 instruction for both grammaticality and adequacy conditions. These effects are related to the four research questions in the following sense: i) the effect of the L1 indicates that, when analyzing the results related to transfer and typological similarity (research question #1 and hypothesis #1) along with the superset and subset classification of the language pairs and, thus, the availability of null and overt subjects (i.e. research question #2 and hypothesis #2), there is a difference among the L1s in terms of whether they are [+null subject] or [-null subject] languages; ii) the effect of the task modality (research question #3 and hypothesis #3) indicates that there is a difference between the oral and the written data obtained; and iii) the effect of time of instruction (research question #4 and hypothesis #4) indicates that there is a difference between the participants that have been instructed in L2 English for a period of 2 years and those that have been so for a period of 4 years. For a more in-depth analysis of these effects, pairwise comparisons were run and analyzed in relation to the hypotheses formulated. These effects, therefore, yield a positive answer to the four research questions in that these L2 English speakers’ production is shaped by their L1, by the modality of the task used to elicit the data and by the time of instruction these speakers have had in English. For a more in-depth analysis of these effects and to actually be able to address the specific role played by the L1, task modality and time of instruction, pairwise comparisons are described and analyzed in the following section. 6.2 Break-down of the data analysis: pairwise comparisons The three-way interaction between L1, modality and time of instruction is explored next and, in order to do so, it is broken down into different pairwise comparisons of the 103 variables under analysis. Since these variables are captured in the four hypotheses presented in chapter 5, the subsequent analysis is done by addressing each of these hypotheses in the light of the pairwise comparison analysis. In each case, first a brief summary of the hypothesis is provided, followed by the results obtained and, finally, by the corresponding discussion. 6.2.1 Hypothesis #1: transfer due to typological similarity In language contact situations with an L2 being acquired after the L1 and in an institutional context, L2 learners typically rely on their L1. The more similarities the learners are able to identify between the two languages (i.e. their L1 and the L2), the more grammatical and adequate their L2 production will be. Therefore, the first issue examined is the influence of the learners’ L1. Displayed in table 8 is the overall distribution of grammatical (example 63) and ungrammatical subjects (example 64) considering the typological similarity of the learners’ L1 (i.e. [+/-null subject] languages) when compared to that of the L2 (i.e. English as a [-null subject] language). 63) my favorite subject is P E (oral L1 Bosnian-group 1; BLBOV2.11, 10 years) 64) Ø is a good teacher (oral L1 Bosnian-group 2; BLBOVII2.04, 12 years) typology L1 grammatical ungrammatical total % [# of cases] % [# of cases] SD % [# of cases] SD [+null] Spanish 87.07 [1,105] 12.2 12.93 [164] 12.2 100 [1,269] Bosnian 95.59 [1,386] 7.65 4.41 [64] 7.65 100 [1,450] [-null] Danish 97.46 [2,183] 3.82 2.54 [57] 3.82 100 [2,240] English [control] 98.92 [1,100] 1.74 1.08 [12] 1.74 100 [1,112] Table 8: Distribution of subjects per language group: grammaticality 104 Being both Spanish and Bosnian [+null subject] languages, these participants are expected to produce a greater number of ungrammatical subjects in comparison with the participants whose L1 is a [-null subject] language (i.e. Danish), because the L2, English, is a [-null subject] language. The vast majority of the subjects produced by the three L2 groups, as in table 8, are grammatical. The L1 Spanish group produces the highest rate of ungrammatical subjects (12.93%) followed by the L1 Bosnian group (4.41%) and the L1 Danish group (2.54%). Some ungrammatical subjects are also found in the case of the control group although the rate is very low (1.08%). Thus, the [+null subject] language group produces the most ungrammatical subjects, as expected, even though the differences between the L1 Bosnian group and the [-null subject] language groups are not that sizable. language groups grammaticality L1 Spanish vs. L1 Bosnian .007* L1 Spanish vs. L1 Danish <.001* L1 Spanish vs. L1 English <.001* L1 Bosnian vs. L1 Danish .502 L1 Bosnian vs. L1 English .233 L1 Danish vs. L1 English 1.000 Table 9: Comparisons across participant groups: grammaticality From the general ANOVA test in the GLMM model, a pairwise comparison with a Bonferroni adjustment has been made and a summary of the p-values is provided in table 9. In the case of grammaticality, the analysis shows a statistically significant difference between the L1 Spanish and the L1 Bosnian groups (p=.007), the L1 Spanish and the L1 Danish groups (p<.001) and the L1 Spanish and the L1 English groups (p<.001). Initially, these results seem to indicate that typological similarity, at least what regards the L1 Spanish 105 speakers, plays a role in the production rate of grammatical subjects. If typological similarity should be the main effect to influence grammaticality, the L1 Bosnian group should perform similarly to the L1 Spanish group and different from the L1 English group, which is not the case. The production of the L1 Bosnian participants is, in fact, more native-like. In other words, only the L1 Spanish group’s ungrammaticality rate makes this group statistically different from the L1 English control group (p<.001). The L1 Bosnian and the L1 Danish groups produce native-like subjects much in the same proportion as the natives, while the L1 Spanish group does not. Displayed in table 10 is the overall distribution of adequate subjects, as in 65, and non-adequate subjects, as in 66, also considering the typology of the L1 ([+/-null subject] languages): 65) Pervious context: *CHI: we have a dog. *CHI: his name is Sofus. Example: *CHI: he is a beagle (oral L1 Danish-group 2; SODKVIIB.11; 13 years) 66) Pervious context: *CHI: Mary and Tom are playing. Mary is watching Tom. Tom has a ball. Example: *CHI: Tom hit the ball in the water. (written L1 Danish-group 1; SODKVC13; 11 years) Table 10: Distribution of subjects per language group: adequacy typology L1 adequate non-adequate total % [# of cases] % [# of cases] SD % [# of cases] SD [+null] Spanish 90.50 [1,000] 6.65 9.50 [105] 6.65 100 [1,105] Bosnian 91.63 [1,270] 12.46 8.37 [116] 12.46 100 [1,386] [-null] Danish 95.97 [2,095] 3.77 4.03 [88] 3.77 100 [2,183] English [control] 94.82 [1,043] 3.55 5.18 [57] 3.55 100 [1,100] 106 What regards the production of adequate subjects, the same rationale used to address grammaticality applies: the participants whose L1s are [+null subject] languages are expected to produce a greater number of non-adequate subjects in comparison with the participants whose L1 is a [-null subject] language, because the L2 (i.e. English) is a [-null subject] language. In this case, the similarity between the participant groups with [+null subject] L1s is more pronounced. Both the L1 Spanish and the L1 Bosnian groups produce a similar number of non-adequate subjects; the L1 Spanish group produced 9.50% while the L1 Bosnian group 8.37%. De novo, both [-null subject] groups (Danish and English) produce the least non-adequate subjects, as expected. Language groups adequacy L1 Spanish vs. L1 Bosnian .719 L1 Spanish vs. L1 Danish <.001* L1 Spanish vs. L1 English .007* L1 Bosnian vs. L1 Danish <.001* L1 Bosnian vs. L1 English .016* L1 Danish vs. L1 English .589 Table 11: Comparisons across participant groups: adequacy A pairwise comparison with a Bonferroni adjustment in the case of adequacy shows that typological similarity plays a role. In a comparison across groups (see table 11), both the L1 Spanish and the L1 Bosnian groups are statistically different both from the L1 Danish group (the L1 Spanish vs. the L1 Danish group: p<.001; and the L1 Bosnian vs. the L1 Danish group: p<.001) and from the L1 English group (the L1 Spanish vs. the L1 English group: p=.007; and the L1 Bosnian vs. the L1 English group p=.016). The L1 Danish group produces the lowest number of non-adequate subjects (4.03%). Their results are even slightly better 107 than the results produced by the L1 English group (5.18%), but this difference is found to be statistically not significant (p=.589). These results indicate that the production of the L1 Spanish and the L1 Bosnian groups in terms of adequacy differs from that of the L1 English group, while that of the L1 Danish group is native-like. Previous studies on transfer and typology have argued that transfer is affected by typological similarity rather than typological proximity (Rothman 2010, Rothman & Cabrelli Amaro 2010, Montrul et al. 2010, Liceras & Alba de la Fuente 2015, Cuza et al. 2018, among others). These studies have proven that when typologically similar languages are in contact, the L1 can facilitate the acquisition of the L2, an idea captured under the Facilitation Hypothesis (Gundel & Tarone 1992). In the same vein, typological difference can produce lower L2 learnability, a fact also argued by Schepens et al. (2016). If typological similarity is at stake, then the L1 Spanish and L1 the Bosnian groups (i.e. both with [+null subject] L1s), on the one hand, and the L1 Danish and the L1 English groups (i.e. both with [-null subject] L1s), on the other, should pattern similarly. What regards grammaticality, our results show that the L1 Danish and the L1 English groups pattern alike. Therefore, it can be argued that in the case of [-null subject] languages (i.e. L1 Danish), the L1 functions as a facilitator. The results for adequacy show that the L1 Spanish and the L1 Bosnian groups, on the one hand, and the L1 Danish and the L1 English groups, on the other hand, pattern alike. Thus, in the case of adequacy, it can be argued that typological similarity does play a role, because the L1 Danish group shows a more native-like production, which suggests that the L1 has a facilitating effect in the acquisition of English as an L2. In the case of the L1 Spanish and the L1 Bosnian groups, since they significantly produce more nonadequate subjects, it can be argued that the fact that their L1 is [+null subject] is a conditioning factor. Therefore, for the [+null subject] language groups, typological similarity 108 only plays a role in the case of adequacy. In the case of the [-null subject] language group, typological similarity functions as a facilitator in the case of both grammaticality and adequacy. The production of adequate subjects is not only contingent on the acquisition of the syntactic properties that characterize sentential subjects in each language. It also involves the combination and mastery of other linguistic domains which, therefore, places the production of adequate subjects at the interface level. In fact, this has been argued to be behind the problems learners have when mastering sentential subjects, as suggested in different L2 studies (Sorace 2005; Tsimpli & Sorace 2006; Sorace & Filiaci 2006, among others). Therefore, the distinction between grammaticality and adequacy as part of the data classification procedure allows for a more refined analyses of the production of sentential subjects by the L2 English speakers. In particular, the classification based on this distinction i) involves a separation between purely grammatical issues and the interface conditions that interact with these syntactic requirements; and consequently, ii) gives the possibility to determine whether the interface at stake (i.e. syntax-pragmatics) is a conditioning factor in the production of speakers with [+null subject] L1s as well as in that of speakers with [-null subject] L1s; or rather, iii) whether purely syntactic constraints is what explains the speakers’ production without their being affected by pragmatic factors. We take the argumentation above to consider that results for adequacy are, in fact, the ones that capture in a more refined way the sensitivity that the speakers have to the linguistic properties that constrain sentential subjects in English. Therefore, our data lend support to the fact that typological similarity plays a role in the acquisition of L2 English subjects, which results in the confirmation of hypothesis #1. 109 6.2.2 Hypothesis #2: The different availability of subject types between the L1 and the L2 Under this hypothesis, the focus is placed on the number of subject types available in the participants’ L1s when compared to those in the language under analysis (i.e. L2 English). The two [-null subject] language groups under consideration (i.e. L1 Danish and L2 English) represent the subset option (i.e. only one subject type is available, the overt subject), while the availability of two subject types (i.e. null and overt) in the two [+null subject] language groups under consideration (i.e. L1 Spanish and L1 Bosnian) makes them the superset option. The prediction is that transfer will take place from the superset languages with a very specific outcome: to facilitate the production of overt subjects in English as an L2. That is, no overproduction of null subjects is expected (i.e. no negative transfer is expected) from L1 Spanish or L1 Bosnian into L2 English. In the case of Danish, as a one subject type language, no overproduction is expected either given that English is also a one subject type language (i.e. positive transfer is expected). As shown in table 12, the vast majority of subjects produced by all four groups are grammatical and overt, as in 65. This distribution is also illustrated in figure 3. The production of grammatical null subjects, as in 66, is the highest in the L1 English group (10.16%), followed by the L1 Danish group (5.67%). The L1 Spanish and the L1 Bosnian groups produce grammatical null subjects in less than 3% of the cases. The L1 Spanish group produces the greatest number of ungrammatical subjects (12.92%), as in 67, followed by the L1 Bosnian group (4.41%). The L1 Danish and L1 English groups produce ungrammatical subjects in less than 3% of the cases. 65) *CHI: I play with my brother in the garden (oral L1 Spanish-group 1; VAESIIIA01; 9 years) 66) *CHI: we have study and Ø look the book (oral L1 Spanish-group 1; VAESIIIB10; 9 years) 110 67) *CHI: today Ø go to the swimming pool of my house (oral L1 Spanish-group 1; VAESIIIA02; 9 years) grammatical ungrammatical overt null % [# of cases] SD % [# of cases] SD % [# of cases] SD Spanish 84.16 [1,068] 4.69 2.92 [37] 4.69 12.92 [164] 12.20 100 [1,269] Bosnian 93.45 [1,355] 2.29 2.14 [31] 2.29 4.41 [64] 7.65 100 [1,450] Danish 91.79 [2,056] 4.48 5.67 [127] 4.48 2.54 [57] 3.82 100 [2,240] English [control] 88.77 [987] 3.27 10.16 [113] 3.27 1.07 [12] 1.74 100 [1,112] Table 12: Distribution of subjects per language group: subject types Figure 3: Distribution of subjects per language group: subject types A pairwise comparison with a Tukey adjustment and the summary of the p-values provided in table 13 show that within groups significant differences are found for all groups between the overt grammatical subject rates and the null grammatical subject rates (p<.001) and between the overt grammatical subject rates and the null ungrammatical subject rates (p<.001). 117 What regards grammaticality, only the L1 Spanish group performs significantly better in the written task than in the oral task. The production of the rest of the groups is very similar in both tasks, indicating that there is no actual clear difference between their performance in the oral task when compared to that in the written task. However, in the case of adequacy, all the L2 groups perform better in the oral task than in the written task. This difference is statistically significant even in the L1 group. As previously seen in hypothesis #1, in the case of modality the double analysis is terms of grammaticality and adequacy has also proven to be essential for a more refined analysis of the production of sentential subjects by these L2 speakers. As before, differences across participant groups clearly emerge in terms of adequacy when comparing written and oral production, which again points to properties located at interfaces being especially vulnerable. What regards purely grammatical issues, these L2 speakers have obtained very high rates, but when other factors, such as pragmatics, are involved, the acquisition seems to be more problematic in the written production. In the case of the oral data, no such effect is seen. Therefore, in the light of the results for adequacy, the written task seems to be more demanding for all groups (including the control group), and so, hypothesis #3 receives confirmation. 6.2.4 Hypothesis #4: time of instruction As different studies have previously shown, time of instruction in the L2 correlates with better performance. That is, the longer L2 learners have been instructed in the L2, the more native-like their performance becomes. In this dissertation, a better performance is interpreted as a more grammatical and more adequate production of sentential subjects. 118 Under hypothesis #4, participants who have been instructed in L2 English for 4 years (group 2) are expected to outperform, in terms of both grammaticality and adequacy, those participants who have been instructed in L2 English for a period of 2 years (group 1). In order to provide a more refined account of the effect of time instruction, a series of interactions will also be included to address hypothesis #4: the interaction between time of instruction and L1 (to account for the effect of typological similarity between groups 1 and groups 2), the interaction between time of instruction and MLU (to account for proficiency differences between groups 1 and groups 2) and the interaction between time of instruction and modality (to account for cognitive load effects that could affect groups 1 and groups 2). In the case of grammaticality, a pairwise comparison with a Bonferroni adjustment shows that the production of grammatical subjects increases the longer the participants have been instructed in L2 English, as illustrated in table 21. Initially, a global effect (excluding the L1s of the participants) is found between the participants in group 1 and the participants in group 2 (p=.020) and between the participants in group 1 and the participants in the control group (p<.001). No significant difference is found between the participants in group 2 and the participants in the control group (p=.353). This indicates that the production of sentential subjects in group 2 participant is similar to that of the native controls. 119 grammatical ungrammatical total % [# of cases] MLU L1 % [# of cases] SD % [# of cases] SD Spanish #1 82.60 [375] 13.58 17.40 [79] 13.58 100 [454] 5.585 Spanish #2 89.58 [730] 7.53 10.42 [85] 7.53 100 [815] 6.561 Bosnian #1 93.14 [380] 8.98 6.86 [28] 8.98 100 [408] 4.243 Bosnian #2 96.55 [1,006] 5.86 3.45 [36] 5.86 100 [1,042] 5.154 Danish #1 97.24 [951] 5.12 2.76 [27] 5.12 100 [978] 6.607 Danish #2 97.62 [1,232] 1.52 2.38 [30] 1.52 100 [1,262] 7.006 English [control] 98.93 [1,100] 1.74 1.07 [12] 1.74 100 [1,112] 6.241 Table 21: Distribution of subjects per language group: time of instruction and grammaticality If the L1 factor is included in the analysis, the results diverge. Within language groups, the difference between grammatical subject rates and ungrammatical subject rates ranges from a 6.98% difference in the L1 Spanish groups (between 82.60% and 89.58%) and a 3.14% in the L1 Bosnian groups (between 93.14% and 96.55%) to a 0.34% difference in the L1 Danish groups (between 97.24% and 97.62%). A within group comparison, as in table 22, shows a significant interaction in the L1 Spanish groups only (p=.002). That is, a significant increase in the rate of grammatical subjects is produced from group 1 to group 2 in the case of the L1 Spanish participants only. In contrast, for the L1 Bosnian and the L1 Danish speakers, the rate of grammatical subjects is not significantly affected by the time of instruction. language groups group 1 vs group 2 L1 Spanish .002* L1 Bosnian .288 L1 Danish .986 Table 22: Comparisons within participant groups: time of instruction and grammaticality 120 The MLUw is used to see whether there are any significant differences between the groups in terms of proficiency. It is argued that the longer the learners have been exposed to the L2, the more proficient they would get and, consequently, this should be reflected in a higher MLUw value. Thus, the effect observed above where differences across groups appear can also be related to the groups’ MLUw values, since significant MLUw differences are found between the two L1 Spanish groups and the two L1 Bosnian groups (see table 3). In the case of the L1 Danish groups, since no significant difference is found between the two groups’ MLUw values (see table 3), the results for the L1 Danish participants should not significantly differ, and this is indeed what the data in table 22 show. Therefore, there is an interaction between MLUw and time of instruction in that all the participants in group 2 that show higher MLUw values are the ones that show an increase in grammaticality rates (i.e. the L1 Spanish and the L1 Bosnian groups). Likewise, the participants in group 2 that do not differ in MLUw terms from the participants in group 1 consequently do not show an increase in grammaticality rates (i.e. the L1 Danish group). This points to MLUw as a valid indication of proficiency in the case of these L2 speakers. To analyze any possible interactions between L1 and grammaticality, an across group comparison, as in table 23, was conducted showing that the L1 Spanish-group 1 produces the highest number of ungrammatical subjects when compared to the L1 Bosnian-group 1 (p=.005) and the L1 Danish-group 1 (p<.001). The L1 Spanish-group 1 also differs from the L1 English group. Thus, a hierarchy can be established as follows from most grammatical to least grammatical: Danish > Bosnian > Spanish. The same hierarchy can be established for group 2 participants in the amount of ungrammatical subjects they produce. Nonetheless, statistically, this difference across the groups 2 is not significant (p>.005). 121 language groups group 1 group 2 L1 Spanish vs. L1 Bosnian .005* .479 L1 Spanish vs. L1 Danish <.001* .182 L1 Spanish vs. L1 English <.001* .350 L1 Bosnian vs. L1 Danish .214 .934 L1 Bosnian vs. L1 English .306 .099 L1 Danish vs. L1 English .999 .999 Table 23: Comparisons across participant groups: time of instruction and grammaticality The comparison across groups in terms of their adequacy rates does not follow the same pattern as for grammaticality, since the production of adequate subjects does not increase the longer the participants have been instructed in English as an L2, except for the L1 Bosnian groups (table 24). L1 adequate non-adequate total % [# of cases] MLU % [# of cases] SD % [# of cases] SD Spanish #1 92.27 [346] 7.03 7.73 [29] 7.03 100 [375] 5.585 Spanish #2 89.59 [654] 6.34 10.41 [76] 6.34 100 [730] 6.561 Bosnian #1 83.16 [316] 11.73 16.84 [64] 11.73 100 [380] 4.243 Bosnian #2 94.83 [954] 9.28 5.17 [52] 9.28 100 [1,006] 5.154 Danish #1 96.00 [913] 4.68 4.00 [38] 4.68 100 [951] 6.607 Danish #2 95.94 [1,182] 2.49 4.06 [50] 2.49 100 [1,232] 7.006 English [control] 94.82 [1,043] 3.55 5.18 [57] 3.55 100 [1,100] 6.241 Table 24: Distribution of subjects per language group: time of instruction and adequacy In both the L1 Danish and the L1 Spanish groups, group 1 slightly outperforms group 2 while both groups show a very high adequacy rate: in the L1 Danish groups, group 1 produces a 96% of adequate subjects and group 2 a 95.94%; and in the L1 Spanish groups, group 1 has a 92.27% and group 2 89.59%. Only in the L1 Bosnian groups adequacy increases as time of instruction increases and group 1 (83.16%) performs better than group 2 122 (94.83%). Therefore, no effect is found between the production of participants in groups 1 and those in groups 2, except for the L1 Bosnian groups (p<.001) (table 25). language groups group 1 vs group 2 L1 Spanish .230 L1 Bosnian <.001* L1 Danish 1.000 Table 25: Comparisons within participant groups: time of instruction and adequacy If MLUw rates are correlated with adequacy rates, the L1 Spanish and the L1 Bosnian groups should show significant differences from group 1 to group 2 since MLUw differences are found in terms of time of instruction for both language groups (see table 3). The L1 Danish groups, however, should behave quite similarly in terms of adequacy, since no significant differences appear in their MLU rates (table 3). Therefore, again, there is an interaction between MLUw and time of instruction in that the group that shows higher MLUw values also shows an increase in adequacy rates (i.e. the L1 Bosnian groups). Likewise, the participants in group 2 that do not differ in MLUw terms from those in group 1 consequently do not show an increase in adequacy rates (i.e. the L1 Danish). To analyze any possible interactions between L1 and adequacy, comparisons within each group and across languages were also conducted. These comparisons are detailed in table 26 below. What regards participants in group 1, the L1 Bosnian participants differ from the rest of the groups and, what regards participants in group 2, the L1 Spanish participants differ from those in the rest of the language groups. 123 language groups group 1 group 2 L1 Spanish vs. L1Bosnian .002* <.001* L1 Spanish vs. L1 Danish 1.000 .012* L1 Spanish vs. L1 English .773 <.001* L1 Bosnian vs. L1 Danish .001* 1.000 L1 Bosnian vs. L1 English <.001* .999 L1 Danish vs. L1 English .993 .960 Table 26: Comparisons across participant groups: time of instruction and adequacy For all participants in group 2, a significant difference, in terms of adequate subjects produced, is found between the L1 Spanish and the L1 Bosnian groups (p<.001), the L1 Spanish and the L1 Danish groups (p=.012) and the L1 Spanish and the L1 English groups (p<.001). These results point towards a correlation between L1 and time of exposure. So far, our data show that there is an increase in the production of grammatical subjects the longer the L2 participants have been instructed in L2 English. Thus, time of instruction plays a role in the acquisition of grammatical subjects in L2 English for these participants. Nonetheless, this increase in production is only significant in the case of the L1 Spanish group. What regards the production of adequate subjects, the longer the participants have been instructed in L2 English does not necessarily mean that their production improves. This is true for all groups except for the L1 Bosnian group. In their case, the longer they have been instructed the better their performance is (being this difference statistically significant). To shed more light on the possible effect of time of L2 instruction in the speakers’ production, an analysis of the effect of time of instruction considering modality has also been performed. Results for grammaticality are displayed in table 27 and figure 4. 124 L1 oral total % [# of cases] written total % [# of cases] grammatical ungrammatical grammatical ungrammatical % [# of cases] SD % [# of cases] SD % [# of cases] SD % [# of cases] SD Spanish #1 80.66 [292] 17.11 19.34 [70] 17.11 100 [362] 90.22 [83] 8.59 9.78 [9] 8.59 100 [92] Spanish #2 88.57 [581] 7.88 11.43 [75] 7.88 100 [656] 93.71 [149] 10.22 6.29 [10] 10.22 100 [159] Bosnian #1 93.24 [276] 7.97 6.76 [20] 7.97 100 [296] 92.86 [104] 12.84 7.14 [8] 12.84 100 [112] Bosnian #2 95.65 [747] 12.33 4.35 [34] 12.33 100 [781] 99.23 [259] 2.01 0.77 [2] 2.01 100 [261] Danish #1 96.92 [819] 5.98 3.08 [26] 5.98 100 [845] 99.24 [132] 6.93 0.76 [1] 6.93 100 [133] Danish #2 97.55 [1,034] 1.97 2.45 [26] 1.97 100 [1,060] 98.01 [198] 5.65 1.98 [4] 5.65 100 [202] English [control] 98.95 [662] 2.15 1.05 [7] 2.15 100 [669] 98.87 [438] 3.98 1.13 [5] 3.98 100 [443] Table 27: Distribution of subjects per language group: time of instruction, modality and grammaticality These results show once again that the production of grammatical subjects is very high in both tasks and within all groups. In the oral task, the participants in group 2 outperform the participants in group 1; nonetheless these differences are found not to be statistically significant (p>.005). In the written task, all groups 2 outperform groups 1, expect for the L1 Danish-group 1 which outperforms the L1 Danish-group 2. However, none of these differences are found to be statistically significant. Across tasks, the greatest difference is found in the L1 Spanish-group 1 between the oral task (80.66%) and the written task (90.22%), followed by the L1 Spanish-group 2 between the oral task (88.57%) and written task (93.71%). The greatest difference is a 3.58% found between the oral task and the written task in the L1 Danish-group 1. For the rest of the groups the difference between the oral task and the written task is below 3%. In all the groups, except for the L1 Bosnian-group 1 and the L1 English group, the participants produce more grammatical subjects in the written task than in the oral task. 125 Figure 4: Distribution of subjects per language group: time of instruction, modality and grammaticality Statistically, modality has an effect within the L1 Spanish-group 1, as depicted in table 28. That is, in the case of the L1 Spanish-group 1 participants, the distribution of grammatical subjects is different in the oral task when compared to the written task with a more grammatical production in the case of the written task (p=.011). No such difference is found in the other groups. language groups group 1 oral vs written group 2 oral vs written L1 Spanish 0.011* 0.414 L1 Bosnian 0.912 0.055 L1 Danish 0.396 0.782 Table 28: Comparisons within participant groups: time of instruction, modality and grammaticality In other words, modality is a conditioning factor only for the L1 Spanish-group 1 with the production of significantly more grammatical subjects in the written task. Results for adequacy are shown in table 29 and depicted in figure 5. 126 L1 oral total % [# of cases] written total % [# of cases] adequate nonadequate adequate nonadequate % [# of cases] SD % [# of cases] SD % [# of cases] SD % [# of cases] SD Spanish #1 96.92 [283] 3.90 3.08 [9] 3.90 100 [292] 75.90 [63] 22.58 24.10 [20] 22.58 100 [83] Spanish #2 97.76 [568] 3.53 2.24 [13] 3.53 100 [581] 57.72 [86] 18.54 42.28 [63] 18.54 100 [149] Bosnian #1 92.75 [256] 11.22 7.25 [20] 11.22 100 [276] 57.69 [60] 27.42 42.31 [44] 27.42 100 [104] Bosnian #2 97.19 [726] 10.40 2.81 [21] 10.40 100 [747] 88.03 [228] 11.20 11.97 [31] 11.20 100 [259] Danish #1 98.78 [809] 6.20 1.22 [10] 6.20 100 [819] 78.79 [104] 11.19 21.21 [28] 11.19 100 [132] Danish #2 99.42 [1,028] 0.85 0.58 [6] 0.85 100 [1,034] 77.78 [154] 11.99 22.22 [44] 11.99 100 [198] English [control] 100 [662] 0.00 0 [0] 0.00 100 [662] 86.99 [381] 12.08 13.01 [57] 12.08 100 [438] Table 29: Distribution of subjects per language group: time of instruction, modality and adequacy Figure 5: Distribution of subjects per language group: time of instruction, modality and adequacy The results within tasks show that, in the case of the oral tasks, groups 2 outperform groups 1. However, none of these differences are found to be statistically significant. Within the written task, group 2 (57.69%) outperforms group 1 (88.03%) only for the L1 Bosnian participants. For the L1 Spanish participants, group 1 participants (75.90%) outperform those 133 2011). Fewer studies have focused on whether two languages that have the same value of the Null Subject Parameter influence each other, and, if so, how (e.g. in the case of two [+null subject] languages, Bini 1993; Margaza & Bel 2006; Sorace & Filiaci 2006; Bel et al. 2016; Lozano 2018; and, in case of two [-null subject] languages, White 1985; Liceras 1989; Liceras & Alba de la Fuente 2015; Mujcinovic 2015). From a formal point of view, Holmberg (2005) and Sheehan (2006) state that [+null subject] languages are superset to [-null subject] languages, in that superset languages have two possible realizations of the subject (i.e. overt and null), whereas subset languages have only one option (i.e. the phonologically realized one). Based on the so-called lexical specialization approach, Fernández Fuertes & Liceras (2018) and Liceras & Fernández Fuertes (2019) argue that the superset language causes acceleration in the production of overt subjects (i.e. the shared option) in the subset language, accounting both for directionality and effect of crosslinguistic influence. When the two languages in contact only allow overt subjects, an acceleration should also take place, since the same option is reinforced. Viewing the results obtained in the context of the formal proposals and previous acquisition works discussed, the following conclusions are reached in the present investigation. What regards typological similarity, previous studies have indicated that the more similar the languages are, the less negative transfer is expected to occur in language contact situations (e.g. Rothman 2010; Rothman & Cabrelli Amaro 2010; Montrul et al. 2010; Liceras & Alba de la Fuente 2015; Cuza et al. 2018). Our data show that typological similarity is a conditioning factor in the case of both the [+null subject] language groups and the [-null subject] language groups. In particular, the L1 Danish group (i.e. [-null subject] language) produces both grammatical and adequate subjects in a high proportion. Considering that their production is native-like, positive transfer can be argued to occur from 134 these participants’ L1 Danish into L2 English. In other words, L1 Danish functions as a facilitator. In the case of the [+null subject] language groups, both produce a high amount of grammatical subjects. However, the L1 Spanish and the L1 Bosnian participants also produce significantly more non-adequate subjects than the L1 Danish and the L1 English groups. Consequently, it can be argued that their L1 is a conditioning factor, in that their production is less native-like because of negative transfer from their L1s into L2 English. Thus, typological similarity is not a conditioning factor when it comes to core grammatical properties (i.e. the production of grammatical subjects, that is, of overt subjects, is at ceiling for all groups and no significant differences appear). However, typological similarity does play a role when taking into account syntax-pragmatics interface related issues (i.e. the [+null subject] groups produce significantly more non-adequate subjects). The availability of different subject types, when comparing the L1 and the L2, also has a bearing on L1 transfer. Previous studies (e.g. Holmberg 2005; Sheehan 2006; Fernández Fuertes & Liceras 2018 and Liceras & Fernández Fuertes 2019) point to the fact that transfer will take place from the superset language to the subset language facilitating the production of overt subjects (i.e. the shared option of the Null Subject Parameter). Our data show that overt subjects are the favored option for all language groups, regardless of whether the participants’ L1 is a [+null subject] language or a [-null subject] language. The production of null subjects (grammatical or ungrammatical) is scarce. Thus, the availability of two subject types (i.e. in the superset languages, Spanish and Bosnian) has a facilitating effect in the acquisition of a one subject type language (i.e. the subset language, English). In the light of these results, it could be argued that the lexical specialization approach, as proposed by Fernández Fuertes & Liceras (2018) and Liceras & Fernández Fuertes (2019) for 2L1 acquisition, also holds for L2 acquisition. However, as pointed out above, interface 135 conditions might alter this facilitating effect and, thus, explain the difference between grammaticality and acceptability that we find in the data. What regards modality, previous studies have indicated that in the case of child L2 acquisition, and as opposed to adults, written tasks are more demanding due to the fact that more cognitive load is required (e.g. Kellog 1996; Granfeldt 2008; Kuiken & Vedder 2011; Williams 2012). Our data point in this direction, too, and the written task results are in fact worse for all language groups in the case of adequacy. What regards time of instruction, the longer the learners have been instructed the better their performance is expected to be (e.g. Gathercole 2002, 2016; Muñoz 2006; Blom & Baayan 2012; Unsworth 2016; Muñoz et al. 2018). That is, they should produce more grammatical and adequate subjects. Our data, however, did not show any effect for time of instruction, neither considering the amount of years exposed to L2 English instruction nor considering the groups’ MLUw values as a sign of linguistic development. To gain further insight on these results, and since modality was proven to be a conditioning factor, the interaction between time of instruction and modality was explored. Nonetheless, no effect was found. Hence, our data point to time of instruction not being a conditioning factor in the L2 acquisition of sentential subjects for these participant groups. This dissertation offers a series of contributions. In the case of the languages under investigation: i) it deals with the contact between typologically different languages, as in previous studies, but it focuses on under-studied languages (such as Bosnian (also SerbianCroatian-Bosnian or SCB)); ii) in the case of the contact between typologically different languages, it compares under-studied languages (such as Bosnian) with languages that have been long studied (such as Spanish); and iii) it analyzes the contact between typologically similar languages that have also been under-studied (such as Danish). 136 This study also offers a new perspective in the analysis of L2 data in terms of two formal proposals on sentential subjects: the so-called superset-subset approach and the socalled lexical specialization approach. These have been previously used in the analysis of 2L1 bilingual data but have not been tested against L2 data. Therefore, the present study shows how these formal proposals can also account for the L2 English data of the three groups of participants under investigation. Furthermore, it concludes, as in the case of 2L1 acquisition studies, that there is positive transfer from the superset language (i.e. Spanish and Bosnian) to the subset language (i.e. English). The data collected for this study focuses on production and, as opposed to previous works, it targets both oral and written production. It offers, therefore, a more comprehensive approach to this linguistic skill and it allows us to further address the differences between these two linguist modes in the case of child speech. To the best of our knowledge, this has not been addressed before, neither in the case of child L2 acquisition nor in the analysis of sentential subjects for this population. Furthermore, the data compiled for this dissertation will be made available via TalkBank (MacWhinney 2019), so that they can be used by the research community for further analyses and comparisons. A further novelty of this study is the use of the MLUw as an indicator of proficiency in the case of L2 participants. Indeed, the MLUw have been used to determine language development in the case of child 2L1 acquisition (as well as L1 acquisition). However, its potential in the analysis of the production of L2 children has not been explored so far. In fact, as shown in this study, MLUw has been proven to be a valid indicator and a more reliable one than time of instruction. In the present investigation, however, some issues have been left unexplored. These include, among others, the combination of data elicited via a different methodology (e.g. 137 judgment data, processing data) which could allow us to gain further insight into the representational nature of subjects in these participants; or the use of standardized proficiency tests that we were unable to implement in this case. These issues will be taken into consideration in future works. 138 REFERENCES Abrahamsson, N., & Hyltenstam, K. (2009). Age of onset and nativelikeness in a second language: Listener perception versus linguistic scrutiny. Language Learning, 59(2), 249-306. Al-Kasey, T., & Pérez-Leroux, A. T. (1998). Second language acquisition of Spanish null subjects. In Flynn, S., Martohardiono, G., & O’Neil, W. (eds.), The Generative Study of Second Language Acquisition, 161-185. New York: Psychology Press. Alexiadou, A., & Anagnostopoulou, E. (1998). Parametrizing AGR: Word order, Vmovement and EPP-checking. Natural Language & Linguistic Theory, 16(3), 491539. Anderson, J. A., Mak, L., Chahi, A. K., & Bialystok, E. (2017). The language and social background questionnaire: Assessing degree of bilingualism in a diverse population. Behavior Research Methods, 50(1), 250-263. Arnaus Gil, L., & Müller, N. (2018a). French postverbal subjects: A comparison of monolingual, bilingual, trilingual and multilingual French. Languages, 3, 1-28. Arnaus Gil, L., & Müller, N. (2018b). Acceleration and delay in bilingual, trilingual and multilingual German-Romance children: Finite verb placement in German. In Linguistic Approaches to Bilingualism, 1–29. Arnaus Gil, L., Jiménez-Gaspar, A., & Müller, N. (2018). The acquisition of Spanish SER and ESTAR in bilingual and trilingual children: Delay and enhancement. In Cuza A., & Guijarro-Fuentes, P. (eds.), Language Acquisition and Contact in the Iberian Peninsula, 91-124. Berlin, Boston: De Gruyter Mouton. Arnaus Gil, L., Müller, N., Sette, N., & Hüppop, M. (2020). Active biand trilingualism and its influencing factors. International Multilingual Research Journal. Bel, A. (2001). Sujetos nulos y sujetos explícitos en las gramáticas iniciales del castellano y el catalán. Revista Española de Lingüística, 31(2), 537-562. Bel, A., García, E., & Rosado, E. (2016) Reference comprehension and production in bilingual Spanish: the view from null subject languages. In Alba de la Fuente, A., Valenzuela, E., & Sanz-Martínez, C. (eds.), Language Acquisition Beyond Parameters: Studies in Honour of Juana Liceras, 27-70. Amsterdam: John Benjamins. 139 Bel, A., Sagarra, N., Comínguez, J. P., & García-Alcaraz, E. (2016). Transfer and proficiency effects in L2 processing of subject anaphora. Lingua, 184, 134-159. Belletti, A. (2001). Inversion as focalization. In Hulk, A., & Pollock, J-Y. (eds.), Subject Inversion in Romance and the Theory of Universal Grammar, 60-91. Oxford: Oxford University Press. Belletti, A. (2004). Answering with a “cleft”: The role of the null subject parameter and the VP periphery. In Brugé, L., Giusti, G., Munaro, N., Schweikert W., & Turano, G. (eds.), Proceedings of the XXX Incontro di Grammatica Generativa, 63-82. Venice: Cafoscarina. Bhatia, T., Ritchie, W. (eds.). (2006). Handbook of bilingualism. Cambridge, MA: Blackwell. Bialystok, E. (1997). The structure of age: In search of barriers to second language acquisition. Second Language Research, 13(2), 116-137. Biberauer, T., Holmberg, A., Roberts, I., & Sheehan, M. (2009). Parametric variation: Null subjects in minimalist theory. Cambridge: Cambridge University Press. Bini, M. F. (1993). La adquisición del italiano: más allá de las propiedades sintácticas del parámetro pro-drop. In Liceras, J. M. (ed.), La Lingüística y el Análisis de los Sistemas no Nativos, 126-140. Ottawa: Doverhouse Editions. Blom, E., & Baayen, H. R. (2012). The impact of verb form, sentence position, home language, and second language proficiency on subject–verb agreement in child second language Dutch. Applied Psycholinguistics, 34(4), 777-811. Blom E. & Unsworth, S. (eds). (2010). Experimental Methods in Language Acquisition Research. Amsterdam: John Benjamins Burton, S., & Grimshaw, J. (1992). Coordination and VP-internal subjects. Linguistic Inquiry, 305-313. Camacho, J. (2006). Do Subjects have a place in Spanish? In Montreuil, J.P., & Nishida, C. (eds.), New Perspectives in Romance Linguistics, 51-66. Philadelphia, PA: John Benjamins. Camacho, J. (2008). Variation in generative grammar: state of the art. Studies in Hispanic and Lusophone Linguistics, 1, 415-434. Amsterdam: John Benjamins. 140 Camacho, J. (2011). Chinese-type pro in a Romance-type null-subject language. Lingua, 121, 987-1008. Camacho, J. (2013). Null subjects. Cambridge: Cambridge University Press. Camacho, J. (2016). The null subject parameter revisited. In Kato, M. A., & Ordoñez, F., (eds.), The Morphosyntax of Portuguese and Spanish in Latin America, 27-48. Oxford: Oxford University Press. Cardinaletti, A., & Starke, M. (1999). The typology of structural deficiency: A case study of the three classes of pronouns. In van Riemsdijk, H. (ed.), Clitics in the Languages of Europe, 145-235. Berlin: Mouton de Gruyter. Cenoz, J. (2001). The effect of linguistic distance, L2 status and age on cross-linguistic influence in third language acquisition. In Cenoz, J., Hufeisen, B., & Jessner U. (eds.), Crosslinguistic Influence in Third Language Acquisition: Psycholinguistic Perspectives. 8-20. Clevedon: Multilingual Matters. Cenoz, J. (2003). The additive effect of bilingualism on third language acquisition: A review. International Journal of Bilingualism, 7, 71-87. Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1982). Some concepts and consequences of the theory of government and binding. MIT press. Chomsky, N. (1991). Linguistics and cognitive science: Problems and mysteries. In Kasher, A. (ed.), The Chomskyan Turn. 26-53. Oxford: Blackwell. Chomsky, N. (1995). Language and nature. Mind, 104(413), 1-61. Chomsky, N. (1999). Language and freedom. Resonance, 4(3), 86-104. Chomsky, N. (2000), Linguistics and Brain Science. In Marantz, A., Miyashita, Y. & O’Neil, W. (eds.), Image, Language, Brain. 1328. MIT Press. Chomsky, N. & Lasnik, H. (1977). Filters and control. Linguistic Inquiry, 8(3), 425-504. Codó, E., Dans, L., & Wei, M. M. (2008). Interviews and questionnaires. In Wei, L., & Moyer, M.G. (eds.), The Blackwell Guide to Research Methods in Bilingualism and Multilingualism, 158-176. Oxford: Blackwell. Cook, V. (2016). Second language learning and language teaching. New York & London: Routledge. 141 Coyle, D., Hood, P., & Marsh, D. (2010). Content and language integrated learning. Ernst Klett Sprachen. Cummins, J. (1991a). Interdependence of first-and second-language proficiency in bilingual children. In Bialystok, E. (ed.), Language Processing in Bilingual Children, 70-89. Cambridge: Cambridge University Press. Cummins, J. (1991b). The development of bilingual proficiency from home to school: A longitudinal study of Portuguese-speaking children. Journal of Education, 173(2), 8598. Cuza, A. (2013). Crosslinguistic influence at the syntax proper: Interrogative subject–verb inversion in heritage Spanish. International Journal of Bilingualism, 17(1), 71-96. Cuza, A., & Camacho, J. (2017). Pronominal subject expression with inanimate reference in heritage speakers of Cuban Spanish. In Cuza A. (ed.), Cuban Spanish Dialectology: Variation, Contact and Change, 229-246. Georgetown University Press. Cuza, A., & Frank, J. (2011). Transfer effects at the syntax-semantics interface: The case of double-que questions in heritage Spanish. The Heritage Language Journal, 8, 66-89. Cuza, A., Guijarro-Fuentes, P., Pires, A., & Rothman, J. (2013). The syntax-semantics of bare and definite plural subjects in the L2 Spanish of English natives. The International Journal of Bilingualism, 17(5), 632-652. Cuza, A., Jiao, J., & López-Otero, J. (2018). Does typological proximity really matter? Evidence from Mandarin and Brazilian Portuguese-speaking learners of Spanish. Languages, 3(2), 13. Cuza, A., Pérez-Leroux, A. T., & Sánchez, L. (2013). The role of semantic transfer in clitic drop among simultaneous and sequential Chinese-Spanish bilinguals. Studies in Second Language Acquisition, 35(1), 93-125. De Houwer, A. (1995). Bilingual language acquisition. In Fletcher, P., & MacWhinney B. (eds.), The Handbook of Child Language, 219-250. Oxford: Blackwell. DeKeyser, R. (2000). The robustness of critical period effects in second language acquisition. Studies in Second Language Acquisition, 22, 499-533. DeKeyser, R. (2012). Interactions between individual differences, treatments, and structures in SLA. Language Learning, 62, 189-200. 142 DeKeyser, R., Alfi-Shabtay, I., & Ravid, D. (2010). Cross-linguistic evidence for the nature of age effects in second language acquisition. Applied Psycholinguistics, 31(3), 413438. DeThorne, L. S., Johnson, B. W., & Loeb, J. W. (2005). A closer look at MLU: What does it really measure? Clinical Linguistics & Phonetics, 19(8), 635-648. Domínguez, L. (2009). Charting the route of bilingual development: Contributions from heritage speakers’ early acquisition. International Journal of Bilingualism, 13(2), 271287. Döpke, S. (1992). One parent one language: An interactional approach. Amsterdam: John Benjamins. Ellis, N. (1999). Cognitive approaches to SLA. Annual Review of Applied Linguistics, 19, 22-42. Ellis, N., & Collins, L. (2009). Input and second language acquisition: The roles of frequency, form, and function introduction to the special issue. The Modern Language Journal, 93(3), 329-335. Ellis, R. (2014). Principles of instructed second language learning. In Celce-Murcia, M., Brinton, D. M., & Marguerite, A. S. (eds.), Teaching English as a Second of Foreign Language, 31-45. Boston, MA: Heinle. Ellis, R., & Yuan, F. (2004). The effects of planning on fluency, complexity, and accuracy in second language narrative writing. Studies in Second Language Acquisition, 26(1), 59-84. Fernández Fuertes, R., & Liceras, J. M. (2018). Bilingualism as a first language: language dominance and crosslinguistic influence. In Cuza, A., & Guijarro-Fuentes, P. (eds.), Language Acquisition and Contact in the Iberian Peninsula. 159-186. Berlin: De Gruyter. Fernández Fuertes, R., Álvarez de la Fuente, E. & Mujcinovic, S. (2016). The acquisition of grammatical gender in L1 bilingual Spanish. Language Acquisition Beyond Parameters: Studies in Honour of Juana Liceras, 237-279. Amsterdam: John Benjamins.