226 62 3MB
English Pages 273 [274] Year 2016
Pedro Guijarro-Fuentes, Maria Juan-Garau, Pilar Larrañaga (Eds.) Acquisition of Romance Languages
Studies on Language Acquisition
Edited by Peter Jordens
Volume 52
Acquisition of Romance Languages
Old Acquisition Challenges and New Explanations from a Generative Perspective
Edited by Pedro Guijarro-Fuentes Maria Juan-Garau Pilar Larrañaga
ISBN 978-1-61451-802-0 e-ISBN (PDF) 978-1-61451-357-5 e-ISBN (EPUB) 978-1-5015-0038-1 ISSN 1861-4248 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. 6 2016 Walter de Gruyter Inc., Boston/Berlin Typesetting: RoyalStandard, Hong Kong Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com
Table of contents Contributors
vii
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga Introduction: The acquisition of Romance languages from a generative 1 perspective: New challenges and approaches
1
Heather Goad Superset and subset grammars in second language acquisition: The role of 17 sonority in the representation of /s/+consonant clusters
2
Harald Clahsen and João Veríssimo Dual morphology in the mental lexicon: Experimental evidence from the 41 acquisition and processing of Greek and Portuguese
3
Bernadette Plunkett Microparametric variation and the acquisition of quantitative de 65 in French
4
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco Semantic features and L1 transfer in the L2 learning of differential object 91 marking: The view from Romanian and Persian
5
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti Elicited production of who-questions by school-aged Italian-speaking 121 children
6
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes Mood interpretation in Spanish: Towards an encompassing view of L1 and 141 L2 interface variability
7
Tim Diaubalick and Pedro Guijarro-Fuentes How do German speakers acquire the Tense-Aspect System in Spanish as a 171 second language? An empirical study
vi
8
9
Table of contents
Pedro Guijarro-Fuentes, M. Carmen Parafita Couto, Rocio Perez-Tattam and Myrthe Wildeboer The non-native acquisition of Spanish (un)interpretable features by Dutch 199 L1 learners Margherita Pivi, Giorgia del Puppo, and Anna Cardinaletti The elicited oral production of Italian restrictive relative clauses and cleft sentences in typically developing children and children with 231 developmental dyslexia
Index
263
Contributors Eds.: Pedro Guijarro-Fuentes1, Maria Juan-Garau1 and Pilar Larrañaga2 Affiliation: 1University of the Balearic Islands, 2Hamburg University Email: [email protected]; [email protected]; [email protected]; [email protected] Aoife Ahern Universidad Complutense de Madrid [email protected] José Amenós-Pons Universidad Complutense de Madrid [email protected] Larisa Avram University of Bucharest [email protected] Anna Cardinaletti Università Ca’ Foscari Venezia [email protected] Cristina Ciovârnache University of Bucharest [email protected] Harald Clahsen Universität Potsdam [email protected] Tim Diaubalick Wuppertal University and University of the Balearic Islands [email protected]
Pilar Larrañaga Hamburg University [email protected]; [email protected] M. Carmen Parafita Couto Leiden University m.parafi[email protected] Rocio Perez-Tattam Swansea University [email protected] Margherita Pivi Università Ca’ Foscari Venezia [email protected] Bernadette Plunkett University of York [email protected] Giorgia del Puppo Università Ca’ Foscari Venezia [email protected] Anca Sevcenco University of Bucharest [email protected]
Heather Goad McGill University [email protected]
João Veríssimo Universität Potsdam [email protected]
Pedro Guijarro-Fuentes University of the Balearic Islands [email protected]
Myrthe Wildeboer Leiden University [email protected]
Maria Juan-Garau University of the Balearic Islands [email protected]
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga
Introduction The acquisition of Romance languages from a generative perspective: New challenges and approaches The present volume includes nine contributions that deal with a wide range of topics in language development, encompassing first as well as second language acquisition studies conducted from a generative perspective. The participants come from various language backgrounds, but the target languages are all ascribed to the Romance family. All the chapters have used an experimental approach involving different types of tests and cross-sectional populations. The various topics dealt with include genuine grammatical issues such as quantificational expressions, the representation of consonant clusters in the interlanguage of speakers of different languages, typical morphological issues such as the subjunctive in Spanish and, finally, issues related to the mental lexicon. These specific topics are discussed in the chapters that follow in relation to some general issues, to which we turn next, related to the participants’ variation in performance, the complexity of the features to be acquired, and cross-linguistic influence. Amongst the issues raised by the contributors to this volume – and by many other researchers in the past decade – is the observation that variation in performance, sometimes reflecting variable input (see Plunkett, this volume, on the distribution of quantitative de in French), is found, to a higher or lesser degree, in first language acquisition (FLA), bilingual first language acquisition (BFLA), and second language acquisition (SLA). In the early 1990s (see, e.g., Meisel 1992), the generative debate on FLA and BFLA for core grammar centred around the issue of whether the errors and variation observed were a reflection of missing functional categories that were implemented in the course of language acquisition. The role of input in the implementation of syntactic structures and morphological paradigms was deemed insignificant for children acquiring one or more languages from birth. The debate has shifted significantly from that time as the role of input has gained relevance in generative language acquisition forums. In this vein, Slabakova, Leal and Liskin-Gasparro (2014) claim that Pedro Guijarro-Fuentes and Maria Juan-Garau, University of the Balearic Islands Pilar Larrañaga, Hamburg University
2
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga
the role of input has always been – and continues to be – crucial for setting parameter values and add that its importance is particularly underscored at present in the areas of synchronic variation and construction frequency. This emphasis is partly due to the increasing understanding of the way in which fine-grained syntactic structures interact with semantic and pragmatic interfaces. In this respect, the seminal articles by Hulk and Müller (2000) and Müller and Hulk (2001), devoted to cross-linguistic influence in BFLA, are particularly enlightening. The greater variation in performance observed in SLA – as opposed to FLA and BFLA – that has been known for decades is now being discussed in the light of the knowledge accrued on the fine mechanisms acting on interfaces as well as on the extralinguistic factors conditioning the aforementioned variation. As regards the former, the complexity of the structures to be acquired seems to play a role in the way a second language is learned, as noted by some of the researchers in this volume (see, e.g., Ahern, Amenós-Pons, and Guijarro-Fuentes on mood interpretation; Avram, Ciovârnache, and Sevcenco on differential object marking). That is, the more complex a structure, the more variation to be found amongst the learners at a given proficiency level. For less complex structures, one may expect high convergence in the judgement of the learners tested even at low levels of proficiency. Amongst the topics considered less problematic for FLA, one may note such domains of morphology as verb agreement or domains of syntax such as the word order within phrases. These two areas have been deemed robust in the research carried out in the past decades by a great number of researchers in various languages ranging from Germanic languages (Meisel 1992), over to Romance languages, to non-Indo-European languages such as Basque (see, e.g., Meisel 1994, 2011). The reason that has been adduced in the past is that they involve prototypically syntactic issues (see, e.g., Meisel 2011). Much recent investigation has noted, furthermore, that the aforementioned areas have very little interface import, and are therefore deemed easier to learn even for L2 learners (Meisel 2011: 144). However, if one looks at the target-like use of very specific parts of verb morphology, for example the use of the subjunctive, much previous research – including Borgonovo, Bruhn de Garavito, and Prévost (2008) and Ahern, AmenósPons, and Guijarro-Fuentes (this volume) – shows that its use poses many problems to learners not just with low but also high proficiency profiles. This means that one should be careful when postulating that a certain extensive domain is easy to learn when the observation is based on one small area of that domain. The subjunctive is a subdomain that is considered difficult for L2 learners both because of its morphological richness and of the intricate usage
Introduction: The acquisition of Romance languages
3
rules that are at the interface with LF (i.e., the syntax-semantics interface). That is, whether a sentence is only possible in the subjunctive, the indicative or both depends to a lesser extent on syntactic factors related to the interpretation of the sentence at stake. This is the case when negative orders are being given and the subjunctive is mandatory. The rules governing the use of the subjunctive are nowadays a relatively well-understood matter. Consequently, a wide range of very good descriptions of the grammatical phenomenon – for example, in the volume by Bosque and Demonte (1999) – exists, but its teaching at schools and universities is still far from appropriate, as Borgonovo, Bruhn de Garavito, and Prévost (2008) point out. They note that the use of the subjunctive with relative clauses with specific/non-specific alternation is usually taught by presenting one single example such as Peter is looking for a secretary who can speak Polish. The explanation that learners of Spanish get in Spanish classes is that the subjunctive is required if the secretary is not known and the indicative if the secretary is known. Borgonovo and colleagues (2008) further add that learners may not really grasp the specificity rule that is at work in the choice of the subjunctive and the indicative. Allegedly, this may be an under-researched source of variation in L2 performance found even at very high levels of proficiency. L2 language learners are also seen to face difficulties with regard to the acquisition of interpretable and uninterpretable features. Guijarro-Fuentes, Parafita Couto, Perez-Tattam and Wildeboer (this volume) investigate such features by Dutch learners of L2 Spanish. It appears from their findings that the acquisition of uninterpretable gender and number features can benefit from explicit instruction, while implicit instruction may not be sufficient for the interpretable constraints on adjective position to be acquired. The processing and acquisition of morphologically complex words by L1 learners is also a difficult area that is still poorly understood, as Clahsen and Veríssimo (this volume) remark. These scholars propose a dual-mechanism model where complex words can be processed and learnt either associatively (i.e., through stored full-form representations) or through the application of rules that decompose inflected word forms into morphological constituents. Subject and object asymmetries constitute yet another difficult area that is well known across languages, as extraction from an object or subject sentence shows differences. Production and understanding of questions and extractions where the object is involved have been considered more problematic, since the participants tested are seen to avoid the “target” structure more often than when the subject is involved. Del Puppo, Pivi and Cardinaletti (this volume) tested this issue with Italian speakers (children between the ages of six and nine and a
4
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga
control group of adults). One added difficulty Del Puppo and associates introduced was a postverbal subject that made disambiguation between subject and object impossible. They found significant differences in the sense that all subjects adhered to the structure Wh V DP when the subject was the questioned entity. Interestingly, the children resort to rephrasing and other structures significantly more often with object questions. More complex structures such as object extractions, which are problematic not only for adult speakers, are even more so for school-age children. In this sense, it would be interesting to determine at what point in their development children attain adult-like competence in object questions. Higher complexity is also known to affect normally developing children and those who suffer from a condition that impinges on language – whether totally or partially – in different ways. Pivi, Del Puppo, and Cardinaletti (this volume) investigated two populations, one of them diagnosed as dyslexic and the other one with suspected dyslexia. They used cleft sentences and subject and object restrictive relative sentences. As expected, all subjects performed worse with object sentences, but the children suffering from dyslexia did even worse than the control group consisting of children with the same age. In other words, object restrictive relative sentences impose a higher processing load and are more prone to errors in populations that are considered normal, but the error rate increases significantly with populations suffering from a condition such as dyslexia. Interestingly, Pivi and colleagues conclude that the group classified as “suspected dyslexic” could suffer from an even more severe condition, since the performance of this group is significantly below the performance of the children investigated in the dyslexic group. It remains for future research to ascertain whether highly complex tasks can be used to discriminate between normally developing children and children suffering from conditions that affect language processing and production. Not only structural complexity is a source of errors and later implementation of grammatical features into the language repertoire, grammatical phenomena that are located at the interface of two modules are also amongst the areas that pose more difficulties to second language learners. This is the case of the preterite in Spanish. Diaubalick and Guijarro-Fuentes (this volume) test its acquisition by native speakers of German acquiring Spanish as an L2. The authors conclude that aspect, being an interpretable feature, is difficult to acquire as the high amount of interpretation errors encountered even at advanced levels of proficiency shows. In short, complexity must be understood as a multifactorial construct involving complex syntactic structures, movements, interface phenomena, a combination thereof or their sum. All of which makes language acquisition more
Introduction: The acquisition of Romance languages
5
difficult. It remains to be investigated in the future whether a complexity ranking can be established either on internal language-specific grounds or as a consequence of observations made by scholars investigating acquisition processes. Ideally, more complex structures should be implemented later in FLA and should cause more problems to learners acquiring a second language. Finally, complexity that causes problems in L2 acquisition may be located at the LF, that is, at the layer in which interpretation of the meaning takes place. Another important issue tackled in the present volume is cross-linguistic influence or transfer, as it has traditionally been called (Kellerman 1995). This is hardly surprising as the influence the mother tongue or other languages known by a speaker exert on a subsequent language being learned has been a topic of extensive research in the past, some of which is summarized by Meisel (2011). Ever since the term transfer was coined by Weinreich (1953), a great number of studies have tried to determine which parts of the L1 grammar are active or play a role in the process of SLA. In broad terms, this scholar made the observation that languages influence each other by virtue of being in contact, whether in society or in the same individual. Beyond doubt, the L1 is the initial state of the L2 and may represent the source of interference, as noted by Weinreich (1953). A much debated topic in L2 acquisition, which goes back to the seminal paper by Bley-Vroman (1990), is whether the L1 is a burden in the task of learning a second or third language, given that a wide range of language patterns in the learners’ interlanguage can be directly accounted for by the structures present in their L1. In that sense, Sprouse (2006) claims that not only the parameters are transferred but also the entire lexicon with the entire phonological, syntactic and semantic specification. Following this idea, Ahern, AmenósPons, and Guijarro-Fuentes (this volume) investigated whether the L1 had any bearing in the learning of the subjunctive in Spanish. The authors chose French and English native speakers in order to test whether significant differences arose in the acquisition of the subjunctive by French and English native speakers. Note that French has subjunctive mood while English lacks it. Their results show that both French and English native speakers performed equally well in most tasks. More importantly, a high number of errors remained even among participants with very high proficiency profiles. The authors claim that the complexity of the phenomenon at stake – involving morphology as well as LF – is responsible for these weak results. That is, complexity overrides any hints about the use of the subjunctive, since French native speakers perform in much the same way as the English native speakers in the L2 despite the existence of the subjunctive in French. The transferred grammar can also be a problem for interlanguage patterns in phonology. The paper by Goad (this volume), however,
6
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga
shows that the evidence for this may sometimes require a little probing. Although the branching onset representation present in Ibero-Romance languages for obstruent+liquid clusters may be redeployed for /s/+consonant in L2 Germanic, seemingly putting learners at an advantage, it is not, in fact, the appropriate representation. Use of the inappropriate branching onset representation is reflected in the patterns observed in learners’ outputs: /s/+consonant clusters with a steep sonority profile are preferred over those with a shallow profile, which does not reflect the typological options that these clusters display across languages. The debate about whether the L1 boosts or hinders L2 acquisition is very old and may not have a simple answer. One should rather ask whether all domains of linguistic knowledge are equally prone to cause interferences, either positive or negative, in the course of L2 – and BFLA – acquisition. Some domains of language involving phonology are known to be very sensitive to interferences (Trujillo 2001). Other purely syntactic domains seem to be more robust and cause few or no difficulties to learners of any linguistic background (Ezeizabarrena 1996). Hence, one of the issues intensively discussed in recent literature is which specific domains of linguistic knowledge may be more prone to transfer in the course of SLA (Meisel 2011). This line of thought is not new, but it can be worth pursuing in more detail in view of more refined theoretical and empirical approaches such as Hulk and Müller’s (2000) and Müller and Hulk’s (2001), which incorporate better defined hypotheses to explain the acquisition of grammatical structures in very specific contexts including bilingual settings. Another topic that has been intensely discussed is whether the genetic distance between the learner’s L1 and L2 plays a role in the interferences caused. Much research in the past has been able to show that languages that do not belong to the same language family such as Basque and Spanish interfere with each other less for the acquisition of verb morphology by bilingual children than language combinations that are genetically close (Ezeizabarrena 1996). This is the more so if the L1 and L2 are acquired simultaneously. In this sense, a number of studies (e.g., Larrañaga and Guijarro-Fuentes 2013, and references therein) conducted with Basque and Spanish simultaneous bilinguals show that genetically unrelated languages suffer from less interference than language combinations of the same language family such as Spanish or French. Goad (this volume) provides support for this from Ibero-Romance versus East Asian speaking learners’ acquisition of Germanic /s/+consonant clusters. Learners from the former group redeploy their branching onset representation for /s/+consonant while East Asian learners, who have no representation for initial clusters to redeploy, must start from scratch. Although the former population would seem
Introduction: The acquisition of Romance languages
7
to be at an advantage, a comparison of /s/+consonant clusters in the two populations of learners shows that only the East Asian learners appropriately prefer clusters with a shallow sonority profile, consistent with cross-linguistic patterns of /s/+consonant cluster well-formedness. The topic of cross-linguistic influence and transfer is conquering new domains of research. A prominent development in this respect over the past two decades concerns the investigation, from a clearly multilingual perspective (e.g., Ecke 2015), of the effects of knowing two or more languages on the acquisition of additional languages. For instance, recent studies are investigating the way in which the L1 and the L2 exert influence on the lexicon among other aspects of the L3 being acquired (e.g., Cenoz 2003, 2013). In fact, third language acquisition has emerged as an independent field of enquiry given that the interaction of three languages in the learning process entails factors and effects unparalleled in SLA, as Sanz, Park and Lado (2015) point out. These authors further note that in L3 acquisition attention has been devoted to exploring the interplay between the three languages at work in the learner’s mind and to identifying the source of transfer (e.g., Ringbom 2007; Rothman 2011, 2015; Sánchez 2015), focusing mostly on the lexicon, while morphosyntax has started to draw attention lately. Similarly, Jarvis (2015) reviews different studies that have explored the effects of bilingualism on the learning of a third language in general and in content and language integrated learning (CLIL) contexts in particular. Research in language acquisition faces a large number of methodological challenges. We will only address a limited number of the methodology-related problems present in this field. As early as the late seventies, Bialystok and Swain (1978) described the adequacy of some research methods depending on the research question being investigated. Methodological issues have been discussed ever since in SLA research in particular, and in psycholinguistic research and applied linguistics in general. Despite the care with which researchers conceive their tasks – grammaticality judgement tasks (GJT), acceptability judgement tasks (AJT), and many others – some of the risks that they are exposed to in trying to uncover language acquisition processes are still present. Borgonovo, Bruhn de Garavito and Prévost (2008) make the observation that the order in which the test sentence and the context are presented may be of relevance in the interpretation of an AJT, a task type that is usually used in studies investigating mood. The observations made by Borgonovo and colleagues (see also Borgonovo, Bruhn de Garavito and Prévost 2015) are of central interest to the present volume as all the authors contributing have reported studies similar to the ones that these scholars have conducted themselves, but often with opposite results. They note that more investigation is needed to ascertain whether the
8
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga
diverging results are due to methodological constraints, that is, whether authors attain different results because they use different methodologies. In this sense, Borgonovo and associates’ rationale and thoughts are worth considering in more detail. They raise the issue that the acceptability/appropriateness of the judgement tasks that current studies use to test whether sentences are considered acceptable or not may suffer from methodological drawbacks that may negatively influence the results obtained. A relevant observation is that AJTs should include the item to be tested with a context that helps the participants to judge whether they consider it acceptable or not. Borgonovo, Bruhn de Garavito and Prévost (2008) believe that the order in which the tested item and the context are presented may be of relevance since the order sentence-context is more related to interpretation and the order context-sentence to production. In order to test whether the order of presentation of the items had an impact on the results, Borgonovo and colleagues tested mood in Spanish with French L1 native speakers by administering an AJT with sentence-context order and a Sentence Combination Felicity Task (SCFT). The results clearly show that acceptance or rejection of subjunctives or indicatives is much sharper with SCFT than AJT. The SCFT results prove significant for the two moods in all tested conditions and, more importantly, even in the conditions that did not show significant scores with AJT. These findings clearly demonstrate that subtle differences in methodology when designing and testing grammatical phenomena may lead to completely different results. The authors of the study claim that scholars ought to consider a number of issues when designing new testing materials or using well-known tools, in the sense that these instruments really measure what is meant to be measured. It may be the case that the order of presentation of testing items and context does not matter in all knowledge domains, but grammatical domains that are located at the interface of LF with another domain seem to be especially sensitive to order of presentation. In general terms, the observations made and the results attained by Borgonovo and colleagues suggest that any scholar wishing to research any grammatical domain should carefully choose the measuring tools to be used depending on the types of subjects that need to be investigated. This claim is not new in FLA, BFLA and SLA research (see, e.g., Klee 1992). The sharp results obtained by using the SCFT contrast with the blurred results with AJT. This, however, does not mean that all subsequent research from now on should make use of this data collection technique. However, researchers who do not obtain clear results at high performance levels should entertain the possibility of administering a further test to their participants in order to check whether the results are possibly the reflection of a methodological artefact. Although time-consuming and not so easy to implement, a post-hoc task could help to elucidate some of the questions that have arisen and to which
Introduction: The acquisition of Romance languages
9
scholars do not have appropriate answers despite the rigorous choice of topic, tasks and language material, as Toth and Guijarro-Fuentes (2013) have recently proposed. In the remainder of this chapter, the papers of the present volume will be briefly sketched. To begin with, Goad seeks to uncover knowledge of the representation of /s/+consonant clusters in the interlanguage of native speakers of English exposed to the constraints of Brazilian Portuguese through Portugueseaccented English and native speakers of Ibero-Romance learning Germanic. Germanic languages allow word-initial /s/+consonant clusters, whereas IberoRomance (Spanish and Brazilian Portuguese) lack them. The paper first considers the acquisition of /s/+consonant clusters in language pairs when the L1 grammar is a subset of the L2: Ibero-Romance learners of Germanic, along with East Asian learners of English as a point of comparison. Despite similarities found in their native languages concerning the illicit status of /s/+consonant, East Asian speakers performed better on such clusters than Ibero-Romance speakers. This difference is attributed to the presence or absence of true branching onsets in the L1 grammar, combined with the ability to redeploy the representation for branching onsets for /s/+consonant clusters which, in turn, is proposed to stem from the relative stridency of /s/ in the L1 grammar. The second half of the paper examines whether native speakers of English are sensitive to the illicit status of /s/+consonant clusters in Brazilian Portuguese, where such clusters are repaired through epenthesis. It is experimentally shown that nearly half of the English speakers tested can determine that /s/+consonant is illicit in Brazilian Portuguese through exposure to such clusters being repaired in Portuguese-accented English. Further, a subset of these speakers can appropriately conclude that /s/+sonorant clusters are ill-formed, after exposure only to data indicating the ill-formedness of /s/+stop. Goad concludes that the patterns observed in interlanguage grammars comply with principles of Universal Grammar, at least in the domain of consonant cluster acquisition. The paper by Clahsen and Veríssimo is devoted to the mental lexicon, in particular the acquisition and processing of inflected verb forms. The authors set out with the observation shared by many scholars that the organization of the mental lexicon and the processing of verbal forms are still poorly understood. The literature they present illustrates conflicting views as to how inflected verb forms are processed and stored. The authors discuss a range of studies that have investigated how verb morphology is acquired and processed focussing mainly on Greek and Portuguese. Clahsen and Veríssimo conclude that theoretical frameworks that ban morphology from the lexicon are less adequate to explain mental processes regarding the storing processing and learning of the latter. They claim that dual approaches which implement morphosyntactic information
10
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga
alongside with associative links between sound/orthography and meaning account for lexical processing better. Plunkett analyzes the distribution of quantitative DE in Belgian French in child and adult populations on the basis of the York corpus available in CHILDES. Quantitative de is mandatory with countable and mass nouns in negated sentences and utterances as well as after some quantificational expressions such as beaucoup/peu, according to standard grammars of French. Plunkett makes the observation that not all varieties of French adhere to this grammatical rule. Most notably the variety spoken in Belgium shows variation that Plunkett investigates. Her data demonstrate that variation is not only present in child but also in adult French spoken in Belgium. More importantly, not all contexts are equally affected by variation. Given these preliminary results, Plunkett claims that the distribution of the data can only be explained if one assumes that part of the human cognitive endowment is a sensitivity to syntactic alternatives in language production and the child’s capability of positing competing grammars to capture these (as in Adger 2006). Avram, Ciovârnache and Sevcenco conduct a cross-sectional experimental study in order to investigate whether L1 transfer could be attested in a pair of languages that exhibit differential object marking (DOM), Persian and Romanian. These languages share a number of properties but differ, crucially, in the way DOM is marked. Whilst Romanian is sensitive to animacy, this feature does not play a major role in Persian. In other words, Romanian marks the direct objects that are animate, whilst Persian marks direct objects regardless of whether they are animate or not, with the only exception of indefinites that show animacy bias. Moreover, referential stability (e.g. definites vs. indefinites) also plays a role in DOM, but languages differ as to which representatives in the referential scale are being considered as stable (CRS) or not stable (RNS). The authors hypothesize that if L1 transfer occurs, it should occur in different directions for native speakers of Romanian and Persian, who are learning Persian and Romanian, respectively. The authors conducted a study with two AJTs administered to L2 speakers of Romanian and L2 speakers of Persian. The results show that no transfer from the L1 can be assumed since the hypotheses postulated by the authors are not borne out. Instead, the authors claim that the access to semantic universals such as animacy appears to be at work and fully explains their results. Del Puppo, Pivi and Cardinaletti investigate subject and object who-questions in four groups of children aged between six and nine years of age and a control group of adults. The authors conducted an elicited production experiment based on Guasti, Branchini and Arosio (2012) in order to test whether subject-object
Introduction: The acquisition of Romance languages
11
asymmetries can be found with this type of questions, especially if the subject appears in post-verbal position and cannot be disambiguated by verbal agreement. When the asked entity is the subject, the Italian-speaking children resorted to high amounts of target-like Wh V DP and cleft interrogatives. On the contrary, if the object was being asked, children used these and other strategies to build the questions. More specifically, they employed questions with left-dislocated and null subjects, questions embedded under a matrix, declarative verb, passive questions and a residue of other strategies. Overall, the results show that children avoid target-like object questions more often than adults by resorting to the aforementioned strategies. Most notably, the children produce more false starts and rephrasings than the adults and the more so with object questions. The authors attribute these findings to the fact that object extraction is more difficult if a post-verbal subject is found in the sentence, as is often the case in Italian. Ahern, Amenós-Pons, and Guijarro-Fuentes investigate the acquisition of indicative vs. subjunctive mood alternation in Spanish concessive clauses by French- and English-speaking learners. Mood choice in concessive environments is extremely complex in Spanish due to the fact that it is ruled by semantics and pragmatics. The choice of the participants was motivated by the fact that French has a subjunctive mood that English lacks. The authors raise two issues, namely whether the L1 determines or influences the course of acquisition of the subjunctive and whether proficiency has an effect on subjunctive choice and interpretation. They make use of a multiple-choice test with one answer as distractor. The results are distributed in three interpretation classes encompassing the irrealis, the presuppositional and other interpretations. No differences in performance between the French and the English group were found. Moreover, proficiency appeared to have no impact in the interpretation of the subjunctive. More importantly, native speakers did not always choose the expected response, which led to a follow-on study using a metalinguistic reflection task with a group of native speakers. The results indicate that integrating linguistic and discourse-pragmatic information in an L2 is very difficult, regardless of L1 and proficiency level. In contrast, for L1 speakers, variability is brought about by a richer and more diverse array of contextual assumptions accessible for online interpretative processes, as compared to the L2 speakers. The paper by Diaubalick and Guijarro-Fuentes investigates the acquisition of past forms in Spanish by 71 native speakers of German at three different proficiency levels to whom a completion task and a GJT were administered. German and Spanish tense marking differ in a number of ways, but most notably in the way telicity and perfectivity are implemented. Whilst Spanish incorporates these features morphologically, German resorts to a lexical strategy by adding all sorts
12
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga
of adverbials to convey these meanings. The issues that the authors investigate pertain to language development as well as to the strategies used by the different proficiency groups tested and, ultimately, to the acquisition of interpretable features. The authors show that learners develop closer to Spanish as proficiency increases, which is uncontroversial, but raise the issue as to why learners still make a considerable amount of judgement errors at high levels of proficiency close to C1/C2 of the Common European Framework of Reference for Languages (CEFR). They conclude that the uninterpretable features linked to the AspP head are not yet acquired, as the Interpretability Hypothesis predicts. Guijarro-Fuentes, Parafita Couto, Perez-Tattam and Wildeboer investigate the acquisition of interpretable and uninterpretable features by Dutch learners of Spanish at different proficiency levels. They aim at testing whether interpretable and uninterpretable features are acquired just as easily by Dutch learners of Spanish. Dutch and Spanish differ in the position of the adjective, which is linked to the FocP in the DP in Spanish that does not exist in Dutch. Furthermore, Guijarro-Fuentes and colleagues seek to find out which SLA hypothesis out of the three currently being discussed explains their data best. They discuss the following theoretical frameworks and their implications and predictions, namely the Interpretability Hypothesis, the Interface Hypothesis and the Feature Reassembly Hypothesis. The first two theoretical frameworks are discarded as their predictions are not consistent with the data analyzed. Hence, their results are not conclusive, but they argue that the Feature Reassembly Hypothesis provides a better explanation for their data, although not all predictions are borne out. Due to the variability of results in the pronominal position, the authors claim that the dichotomy interpretable vs. uninterpretable does not account for the findings. They suggest that the learners’ exposure to the adjectives provided may be the reason why some learners perform better in relation to certain adjectives. Pivi, Del Puppo and Cardinaletti investigate normally developing children and children with diagnosed dyslexia as well as a group of children with suspected dyslexia as regards the production of cleft sentences and restrictive relative clauses (subject and object) in Italian. The authors claim that cleft sentences should pose fewer problems for processing than relative clauses. They elicited the spoken data using a preference production task (slightly modifying the task elaborated by Friedmann and Szterman (2006) and Novogrodsky and Friedmann (2006) for Hebrew), a correction task, and a delayed repetition task. As expected, subject and object relative clauses were neither produced equally frequently nor were they always target-like. Moreover, adults used a number of alternative structures to produce an object relative clause. All participants were seen to produce high
Introduction: The acquisition of Romance languages
13
numbers of subject relative clauses. As concerns the dyslexic population, not surprisingly, the accuracy with subject relative clauses was higher than with object relative clauses; most notably, they produced a high number of object relatives with resumptive DPs and repeated them at a significantly lower percentage. As for the production of cleft sentences, all groups, with the exception of young suspected dyslexics, uttered a consistent amount of subject clefs. The most frequent alternative strategy was the production of main SVO sentences with focalized subjects. When object clefts were targeted, SVO sentences were the preferred answering strategy for both TD and non TD children; however, children could repeat object clefts, except for the suspected dyslexics, who attained lower percentage of accuracy in the delayed repetition task. The authors conclude that the suspected dyslexics probably suffer from a more severe impairment than dyslexia, since their results strongly differ from those obtained by the dyslexic group.
Concluding remarks The set of papers included in this volume investigate timely issues focussing on phenomena tightly linked to interfaces. The study on interfaces has become increasingly popular amongst scholars working in FLA and BFLA in the last decade. A variety of investigation methods used in other fields such as SLA have been adapted to capture some phenomena that cannot be easily studied in naturalistic settings more accurately. Many of the papers in this volume have adapted existing methods or have proposed innovative ways to investigate phenomena that are not well understood at present. As a number of authors note, results do not always match the hypotheses or are contradictory with the ones found in the literature. It is not clear how many of the discrepancies are due to the methodologies used. Future research should refine existing methods and evaluate them rigorously in order to be able to use the best possible measuring tools.
Acknowledgements First of all, we would like to thank Peter Jordens and his editorial team for their kind acceptance to publish this edited volume and for their continuous and valuable support and guidance throughout the publication process. We also would like to thank all the contributors who made this research initiative possible
14
Pedro Guijarro-Fuentes, Maria Juan-Garau and Pilar Larrañaga
for their valuable scholarly work. We also remain indebted to the reviewers for all the time they took to read and provide comments on the papers included. In no particular order, we would like to thank: Teresa Guasti, Natascha Müller, Ana Lucia Santos, Stavroula Stavrakaki, José Camacho, Cristóbal Lozano, Sandrine Zufferey, Mike Iverson, Rocio Perez-Tattam, David Stringer, David Adger, Laurent Dekydtspotter, Themis Karaminis, Miquel Simonet, Lucrecia Rallo Fabra, Jacee Cho, Llorenç Comajoan Colomé, Brechje van Osch, and Petra Sleeman. We are deeply appreciative of their professionalism and their commitment to ensuring the high degree of quality reflected in this edited volume. Any remaining errors are solely ours.
References Adger, David. 2006. Combinatorial variability. Journal of Linguistics 42(3). 503–530. Bley-Vroman, Robert. 1990. The logical problem of foreign language learning. Linguistic Analysis 20. 3–49. Borgonovo, Claudia, Joyce Bruhn de Garavito & Philippe Prévost. 2008. Methodological issues in the L2 acquisition of a syntax/semantics phenomenon: How to Assess L2 knowledge of mood in Spanish relative clauses. In Joyce Bruhn de Garavito and Elena Valenzuela (eds.), Selected Proceedings of the 10th Hispanic Linguistics Symposium, 13–24. Somerville, MA: Cascadilla. Borgonovo, Claudia, Joyce Bruhn de Garavito & Philippe Prévost. 2015. Mood selection in relative clauses. Studies in Second Language Acquisition 37(1). 33–69. Bosque, Ignacio & Violeta Demonte. 1999. Gramática descriptiva del español. Madrid: Espasa. Bialystok, Ellen & Merrill Swain. 1978. Methodological approaches in second language learning. McGill Journal of Education 13(2). 137–144. Cenoz, Jasone. 2003. The additive effect of bilingualism on third language acquisition: A review. International Journal of Bilingualism 7. 71–87. Cenoz, Jasone. 2013. The influence of bilingualism on third language acquisition: Focus on multilingualism. Language Teaching 46. 71–86. Ecke, Peter. 2015. Parasitic vocabulary acquisition, cross-linguistic influence, and lexical retrieval in multilinguals. Bilingualism: Language and Cognition 18(2). 145–162. Ezeizabarrena, María José. 1996. Adquisición de la morfología verbal en euskera y castellano por niños bilingües. Bilbao: Servicio de Publicaciones de la Universidad del País Vasco. Friedmann, Naama & Ronit Szterman. 2006. Syntactic movement in orally-trained children with hearing impairment. Journal of Deaf Studies and Deaf Education 11. 56–75. Guasti, Maria Teresa, Chiara Branchini & Fabrizio Arosio. 2012. Interference in the production of Italian subject and object wh-questions. Applied Psycholinguistics 33. 185–223. Hulk, Aafke & Natascha Müller. 2000. Bilingual first language acquisition at the interface between syntax and pragmatics. Bilingualism: Language and Cognition 3(3). 227–244. Jarvis, Scott. 2015. Influences of previously learned languages on the learning and use of additional languages. In Maria Juan-Garau and Joana Salazar-Noguera (eds.), Content-based language learning in multilingual educational environments. Dordrecht: Springer.
Introduction: The acquisition of Romance languages
15
Kellerman, Eric. 1995. Crosslinguistic influence: Transfer to nowhere? Annual Review of Applied Linguistics 15. 125–150. Klee, Thomas. 1992. Developmental and diagnostic characteristics of quantitative measures of children’s language production. Topics in Language Disorders 12. 28–41. Krashen, Stephen. 1982. Principles and practices of second language acquisition. Oxford: Pergamon Press. Larrañaga, Pilar & Pedro Guijarro-Fuentes (eds). 2013. Basque bilingualism: A tale of many stories (Special Issue). International Journal of Bilingualism 17(3). 247–399. Meisel, Jürgen M. (ed.). 1992. The acquisition of verb placement: Functional categories and V2 phenomena in language acquisition. Dordrecht: Springer. Meisel, Jürgen M. (ed.). 1994. La adquisición del vasco y del castellano en niños bilingües. Madrid & Frankfurt: Vervuert Verlagsgesellschaft. Meisel, Jürgen M. 2011. First and second language acquisition. Cambridge: Cambridge University Press. Müller, Natascha & Aafke Hulk. 2001. Crosslinguistic influence in bilingual language acquisition: Italian and French as recipient languages. Bilingualism: Language and Cognition 4. 1–21. Novogrodsky, Rama & Naama Friedmann. 2006. The production of relative clauses in syntactic SLI: A window to the nature of the impairment. Advances in Speech-Language pathology 8. 364–375. Ringbom, Hakan. 2007. Cross-linguistic similarity in foreign language learning. Clevedon: Multilingual Matters. Rothman, Jason. 2011. L3 syntactic transfer selectivity and typological determinacy: The typological primacy model. Second Language Research 27. 107–127. Rothman, Jason. 2015. Linguistic and cognitive motivations for the Typological Primacy Model (TPM) of third language (L3) transfer: Timing of acquisition and proficiency considered. Bilingualism: Language and Cognition 18(2). 179–190. Sánchez, Laura. 2015. L2 activation and blending in third language acquisition: Evidence of crosslinguistic influence from the L2 in a longitudinal study on the acquisition of L3 English. Bilingualism: Language and Cognition 18(2). 252–269. Sanz, Cristina, Hae In Park & Beatriz Lado. 2015. A functional approach to cross-linguistic influence in ab initio L3 acquisition. Bilingualism: Language and Cognition 18(2). 236–251. Slabakova, Roumyana, Tania L. Leal & Judith Liskin-Gasparro. 2014. We have moved on: Current concepts and positions in generative SLA. Applied Linguistics 35(5). 601–606. Sprouse, Rex A. 2006. Full transfer and relexification: Second language acquisition and creole genesis. In Christine Jourdain, Claire Lefebvre and Lydia White (eds.), Dialogues: Processes in L2 acquisition and creole genesis. Amsterdam: John Benjamins. Toth, Paul D. & Pedro Guijarro-Fuentes. 2013. The impact of instruction on second-language implicit knowledge: Evidence against encapsulation. Applied Psycholinguistics 34(6). 1163– 1193. Trujillo, Cristina. 2001. Die Rolle der Sonorität im phonologischen Fremdspracherwerb. Der Erwerb des deutschen Silbenreims von Muttersprachlern des Spanischen. Frankfurt: Peter Lang. Weinreich, Uriel. 1953. Languages in contact. The Hague: Mouton.
Heather Goad
1 Superset and subset grammars in second language acquisition: The role of sonority in the representation of /s/+consonant clusters Abstract: This chapter explores the shapes of interlanguage grammars in the syllable structure domain from the perspective of the evidence available to learners indicating the well- or ill-formedness of /s/+consonant clusters in the language being learnt and how this interfaces with cross-linguistically motivated constraints on representation. The chapter begins with the sonority profile of word-initial /s/+consonant clusters, and concludes that the constraints they respect mirror those that hold of coda+onset clusters, rather than branching onsets. This, in turn, is used to motivate different representations for each: /s/+consonant clusters are heterosyllabic, where /s/ is the coda of an emptyheaded syllable, while obstruent-initial clusters form branching onsets. Drawing on data from several studies in the literature, the chapter then turns to examine the acquisition of word-initial /s/+consonant clusters when the native language grammar is a subset of the second language as concerns cluster profile: East Asian and Ibero-Romance learners of Germanic. It is argued that cross-language differences in the developmental patterns observed – that is, whether or not they respect /s/+consonant cluster well-formedness on the sonority dimension – stem from the presence or absence of true branching onsets in the native language grammar, including the (un)suitability of redeploying this structure for /s/+consonant clusters, which is proposed to be linked to the relative stridency of /s/ in the native language. The chapter then turns to examine the acquisition of /s/+consonant cluster ill-formedness when the first language grammar is a superset of the second language for word-initial cluster profile: English-speaking naïve learners of Brazilian Portuguese. It is experimentally shown that second language learners can use indirect positive evidence to learn that /s/+consonant clusters are ill-formed in the language being learnt: evidence from errors in the learner’s first language made by native speakers of the learner’s second language.
Heather Goad, McGill University
18
Heather Goad
1 Introduction The acquisition of a second language can be aided or hindered by the first language grammar that is already in place. The kinds of evidence available to learners indicating the (un)suitability of the transferred grammar will depend on the particulars of the first and second language grammars and on the acquisition context. In this chapter, I explore the shapes of interlanguage grammars from the perspective of the evidence available to learners and how this interfaces with cross-linguistically motivated constraints on representation. These issues are examined from a single empirical perspective: the acquisition of the well- or ill-formedness of /s/+consonant clusters. I begin by examining patterns in the development of /s/+consonant clusters in language pairs where the first language is a subset of the second language on this dimension. That is, the first language is more restrictive: it is contained within the second language as concerns the grammar of left-edge clusters and, accordingly, positive evidence will be available indicating that such clusters are licit in the language being learnt. Drawing on data from several studies in the literature, it will be seen that the first language grammar to a large extent impacts the developmental patterns that learners display in their outputs. Specifically, learners of Germanic languages who are native speakers of Spanish follow a different path in development from those who are native speakers of East Asian languages (specifically, Mandarin, Cantonese, Japanese, and Korean). Surprisingly, native speakers of Brazilian Portuguese pattern more like the East Asian group, in spite of the genetic link that holds between Spanish and Portuguese. These cross-language differences will be proposed to stem from aspects of the native language grammar: the presence or absence of true branching onsets, including the (un)suitability of redeploying this structure for /s/+consonant clusters, which will be proposed to be linked to the relative stridency of /s/. I then turn to examine patterns in the development of /s/+consonant clusters in a language pair where learners must determine that /s/+consonant clusters are illicit in the language being learnt, as the first language is a superset of the second language on this dimension. That is, the first language is more permissive: it contains the second language as far as the grammar of left-edge clusters is concerned. Drawing on experimental results from Schwartz and Goad (to appear), where English-speaking naïve learners of Brazilian Portuguese are exposed to evidence indicating the ill-formedness of /s/+consonant clusters in this language, it will be suggested that, rather than resorting to some sort of negative evidence, in certain acquisition contexts, second language learners may be able to discover aspects of the second language grammar by relying on indirect positive evidence:
Superset and subset grammars in second language acquisition
19
evidence from errors in the learner’s first language made by native speakers of the learner’s second language. An important theme that will reappear throughout the chapter is the role that sonority plays in determining the well- or ill-formedness of different types of /s/+consonant clusters. This will be seen to impact the grammars that learners build when acquiring both superset Germanic and subset Ibero-Romance. As interlanguage grammars will be shown to respect the constraints that shape cluster well-formedness in end-state languages, it will be concluded that interlanguage grammars are possible grammars in the sense that they are constrained by Universal Grammar (contra Clahsen and Muysken 1986, 1989; Bley-Vroman 1990).
2 Word-initial clusters I begin by examining the shapes of word-initial clusters across both cluster type and language. The discussion will focus on the two most common types of initial clusters, namely /s/-initial clusters (e.g., spy, sly) and obstruent-initial clusters (e.g., ply, cry), where obstruent henceforth refers to obstruents other than /s/. /s/-initial clusters defy many of the constraints that hold of obstruentinitial clusters; most notably, obstruent-initial clusters must rise in sonority in the vast majority of languages, while /s/-initial clusters may show a sonority plateau.1 In non-linear approaches to syllable structure, this observation, among others, has been used to motivate different representations for these two types of clusters: obstruent-initial clusters form left-headed branching onsets (e.g., Kaye, Lowenstamm, and Vergnaud 1990), while /s/ is located outside of the onset constituent in /s/-initial clusters, as shown in (1). (1)
Preliminary representations: a. Obstruent-initial clusters:
b.
/s/-initial clusters (minimal representation):
(1b) is the minimal representation assumed by researchers who adopt an articulated view of the syllable. Probing this literature more extensively, /s/ has been 1 I adopt the following sonority hierarchy: obstruents (including /s/) < nasals < laterals < rhotics < vocoids.
20
Heather Goad
analyzed as extraprosodic (not linked to higher structure) (e.g., Steriade 1982), as organized directly by the syllable node (e.g., van der Hulst 1984) or prosodic word (e.g., Goldsmith 1990), or as the coda of an empty-headed syllable (e.g., Kaye 1992). The representation in (1b), which is contained in each of these proposals, is sufficient for the moment. Sonority is the most commonly used criterion for determining cluster representation. Given that /s/-initial clusters can rise in sonority (e.g., sly), some researchers have analyzed these particular /s/+consonant clusters as branching onsets, in contrast to those that show a sonority plateau (e.g., spy) (e.g., Hall 1992; Fikkert 1994). This view is challenged by the fact that, unlike branching onsets, the well-formedness of word-initial /s/+consonant decreases as the sonority of the consonant following /s/ increases (Goad 2011, 2012), as shown in Table 1.2 Indeed, while obstruent+liquid is the optimal branching onset (Clements 1990), /s/+liquid is highly disfavoured. Table 1: Sonority profile of word-initial /s/+consonant clusters across languages
/s/+stop /s/+nasal /s/+lateral /s/+rhotic
Spanish, Brazilian Portuguese
French, Picard
Greek, Romansch
Dutch, Italian
English, Russian
* * * *
Z * * *
Z Z * *
Z Z Z *
Z Z Z Z
/s/+consonant cluster well-formedness instead parallels the phonotactic constraints that hold of word-internal coda+onset clusters. Specifically, when obstruents are permitted in coda, a flat sonority profile is preferred across the syllable boundary (e.g., [kt], [ks]); as the sonority profile steepens (e.g., [kn], [kl], [kr]), the cluster strives to be analyzed as a branching onset. However, if a branching onset representation is never available for /s/+consonant clusters (Goad 2011), then rising sonority /s/-initial clusters will simply be ruled out, as we see progressively for Dutch (*[sr]), Greek (*[sr, sl]), and French (*[sr, sl, sn]), for example. In view of these observations, I consider that the same representation holds for both flat and rising sonority /s/+consonant clusters, while other rising sonority clusters (i.e., obstruent+sonorant) form branching onsets, shown earlier in (1a) and repeated in (2a). In place of (1b), however, I adopt the position of Kaye (1992) that /s/ in initial /s/+consonant clusters is the coda (rhymal dependent) 2 Note that /s/+nasal clusters are only marginally well-formed in Greek. Information on Picard /s/+consonant clusters was provided by Julie Auger (p.c.).
Superset and subset grammars in second language acquisition
21
of an empty-headed syllable, as shown in (2b) (representation slightly simplified). This representation directly captures the phonotactic relations displayed in Table 1 (see Goad 2012). More concretely, I assume that the typology in Table 1 reflects a set of constraints, which are part of Universal Grammar, that govern coda+onset well-formedness of both word-initial and word-internal /s/+consonant clusters. (2)
Revised representations: a. Obstruent+sonorant clusters:
b.
/s/+consonant clusters:
The proposal that obstruent+sonorant and /s/+consonant clusters have different representations, along the lines of (2), is additionally supported by the observation that some languages contain word-initial clusters of one type but not the other; see Table 2. Table 2: Typology of word-initial clusters
East Asian (Mandarin, Cantonese, Japanese, Korean) Ibero-Romance (Spanish, Brazilian Portuguese) Acoma [Keresan], Blackfoot [Algonquian] Germanic (English, German, Swedish)
Obstruent+sonorant clusters
/s/+consonant clusters
* Z * Z
* * Z Z
The goals of this chapter are to explore the predictions of the different representations in (2) for second language acquisition, focusing principally on /s/+ consonant clusters. We consider the challenges for learning both a superset language (East Asian and Ibero-Romance learners of Germanic) and a subset language (English learners of Brazilian Portuguese). Table 2 shows that East Asian and Ibero-Romance are subsets of Germanic as far as /s/+consonant clusters are concerned. However, East Asian is additionally a subset of Germanic as far as obstruent+sonorant clusters are concerned. Given that obstruent+sonorant and /s/+consonant clusters have different representations, the presence or absence of obstruent+sonorant clusters should have no impact on the development of /s/+consonant. As we will see, this is not necessarily the case. /s/+consonant clusters take a surprising shape in the acquisition of Germanic for learners coming from Ibero-Romance (notably Spanish) backgrounds, surprising from the perspective of the typology of options
22
Heather Goad
on the sonority dimension shown in Table 1. In view of this, we additionally strive to explain the shapes of interlanguage grammars, when these seemingly diverge from the expectations of Universal Grammar.
3 Predictions We begin with the predictions for learners who are acquiring a superset grammar. In this type of acquisition context, positive evidence will be available: through detecting the presence of appropriate clusters in the ambient data, learners should conclude that a given structure is well formed in the language being learnt and acquisition should ultimately be successful. However, as we have seen, not all left edge clusters are of the same type. In view of the differing roles that sonority plays in determining the well-formedness of obstruent+sonorant versus /s/+consonant clusters and, consequently, the different representations that hold for each, specific predictions are in (3): (3)
Predictions for learning a superset grammar: 1. First language East Asian – Second language Germanic: Learners whose first language has no left-edge clusters should not use the same representation for obstruent+sonorant and /s/+consonant clusters in the second language (following from (2)); 2. First language Ibero-Romance – Second language Germanic: Learners whose first language has obstruent+sonorant clusters only should not use this representation for /s/+consonant clusters in the second language (following from (2)); 3. First language East Asian/Ibero-Romance – Second language Germanic: Learners whose first language lacks /s/+consonant clusters should acquire such clusters of shallower sonority profiles before those of steeper profiles (following from Table 1); at any given point in development, this should be reflected in fewer repairs applying to /s/+consonant clusters of shallower sonority profiles than to those of steeper profiles.
When learning a subset grammar, positive evidence is not always available or accessible, suggesting that learners must often rely on some sort of negative evidence to acquire the grammar. Given that second language learners have been shown to successfully learn syllable structure constraints in the absence of positive evidence (Trapman and Kager 2009), Schwartz and Goad (to appear)
Superset and subset grammars in second language acquisition
23
propose that another sort of evidence may be available, in at least some acquisition contexts: indirect positive evidence. Indirect positive evidence is evidence from errors in the learner’s first language made by native speakers of the learner’s second language. To illustrate, consider the following: A native speaker of English, who is in the process of acquiring Brazilian Portuguese, may hear a native speaker of Brazilian Portuguese, who is in the process of acquiring English, produce English words beginning with /s/+consonant with a prothetic vowel (star → [i]star). From this, the English speaker could conclude that the phonotactics of Brazilian Portuguese do not allow word-initial /s/+consonant clusters. Although the evidence that the English speaker used to arrive at this conclusion was positive, it was indirect, because the evidence for the structure of one language was available through errors made in another language. Assuming that indirect positive evidence is practicable for learning a subset grammar, specific predictions are provided in (4): (4)
Predictions for learning a subset grammar: First language English – Second language Brazilian Portuguese: 1. Learners exposed to the ill-formedness of /s/+consonant through indirect positive evidence will realize that /s/+consonant is ill-formed in the language being learnt; 2. Learners who recognize that /s/+consonant is ill-formed in the language being learnt will not overgeneralize this to obstruent+sonorant (following from (2)); 3. Learners exposed only to the ill-formedness of /s/+stop through indirect positive evidence will conclude that /s/+sonorant is also ill-formed (following from Table 1).
We first consider, in section 4, the acquisition of superset grammars, using data drawn from the literature. We then turn, in section 5, to an experiment that was designed to examine the potential utility of indirect positive evidence for learning a subset grammar.
4 Learning a superset grammar 4.1 East Asian learners of Germanic We begin with the results of four studies that systematically examine the acquisition of /s/+consonant and obstruent+sonorant clusters in English by native speakers of one or more of the following languages: Mandarin, Cantonese, Japanese, and Korean, all of which lack both cluster types (see Table 2).
24
Heather Goad
Concerning the acquisition of /s/+consonant, Enochson’s (2012) comparative study observed that sonority distance is negatively correlated with target-like production of such clusters in the English outputs of Mandarin, Cantonese, and Japanese speakers. Specifically, learners’ performance was best on /s/+stop, followed by /s/+nasal, followed by /s/+lateral, followed by /s/+glide.3 Enochson’s findings for Japanese-speaking learners of English mirror the earlier results of Martz (2007) who found that accuracy on /s/+stop was higher than on /s/+nasal which, in turn, was higher than on /s/+lateral. And taken together, the results of Enochson and Martz mirror those of Kim (2000) and Kwon (2006) on Koreanspeaking learners of English: Kim found that performance on /s/+stop was better than on both /s/+nasal and /s/+lateral, and Kwon found that performance on /s/+nasal was better than on /s/+lateral. The results of all four studies are consistent with Prediction 3 in (3), which states that learners whose native language lacks /s/+consonant clusters should acquire such clusters of shallower sonority profiles before those of steeper profiles, as per the typology in Table 1, which, at any given point in development, should be reflected in fewer repairs applying to /s/+consonant clusters of shallower sonority profiles than to those of steeper profiles. Turning to Prediction 1 in (3), that learners whose first language has no leftedge clusters should not use the same representation for obstruent+sonorant and /s/+consonant clusters in the second language, Enochson (2012) reports that there was no correlation between sonority distance and target-like production of obstruent+sonorant clusters, in contrast to her findings for /s/+consonant: learners’ performance was equally good on stop+glide, stop+liquid and fricative+ liquid, in support of Prediction 1. Martz (2007), Kim (2000), and Kwon (2006) compare performance on analogous /s/+consonant and obstruent+sonorant clusters, that is, on clusters with the same sonority slope and voicing profile; specifically, /s/+lateral is compared with voiceless fricative+liquid in Martz, and with voiceless stop+liquid in Kim and Kwon. Martz finds better performance on obstruent+ sonorant than on /sl/; Kwon finds better performance on /sl/ than on obstruent +sonorant (nears significance); and Kim finds better performance on /sl/ for less advanced speakers and on obstruent+sonorant for more advanced speakers. The results from all three studies are consistent with Prediction 1, which expects no
3 We did not include glides in our earlier discussion of cluster type because, across languages, glides in consonant+glide+vowel strings can be syllabified in three different ways, only one of which bears on the constraints that hold of left-edge clusters: they can be located in the onset (either following coda /s/ or as part of a branching onset); they can be in the nucleus, forming a (light) diphthong with the following vowel; or they can be secondary articulations on a preceding onset consonant. Each of these representations respects different constraints.
Superset and subset grammars in second language acquisition
25
correlation to be observed between learners’ performance on /s/+consonant and obstruent+sonorant clusters, even when other factors (sonority slope, voicing) are controlled.4 In sum, the results from all four studies reviewed here are consistent with the particular role that sonority plays in determining the well-formedness of /s/+consonant clusters, in contrast to obstruent+sonorant clusters. We could conclude from this that interlanguage grammars are possible grammars, that is, grammars that are constrained by Universal Grammar. However, when we consider the results from studies that have examined the acquisition of /s/+ consonant clusters by learners whose native language grammars contain obstruent+sonorant clusters, we arrive at a different conclusion: for the majority of learners, order of acquisition of /s/+consonant does not reflect the typology of options seen in Table 1. We examine these studies in the next section.
4.2 Ibero-Romance learners of Germanic Table 3 summarizes the results from several studies on the acquisition of Germanic languages by native speakers of Spanish and Brazilian Portuguese.5 In the Spanish group of studies, performance on /s/+consonant clusters where the consonant is of higher sonority is better than performance on such clusters when the consonant is of lower sonority.6 This is inconsistent with Prediction 3 in (3), that learners whose first language lacks /s/+consonant clusters should 4 Further support for Prediction 1 in (3) comes from Detey and Nespoulous’s (2008) study on Japanese-speaking learners of French. In a syllable counting task, participants treated /s/+ consonant and obstruent+sonorant clusters differently: they displayed more epenthesis for /s/+consonant and earlier acquisition of obstruent+sonorant which, together, suggest different representations for these two cluster types. 5 Results for German [ ʃv], Swedish [sv], and English [sw] have been removed for several reasons. One, the status of glides in syllable structure is variable (see note 3). Two, both Spanish and Brazilian Portuguese have [sw] (where [w] is part of a light diphthong), in contrast to other /s/+consonant clusters. Three, some learners may treat German and Swedish [v] as a low sonority fricative while others treat it as a high sonority glide, leading to results that are difficult to interpret. 6 Note that [sl] stands out in the Abrahamsson study in patterning with stops rather than with other sonorants. Abrahamsson suggests that this may be due to low number of tokens for [sl] in his study. An alternative is that this is due to the ambiguous articulatory properties of [l], which makes it both approximant-like and stop-like. Like oral approximants, air can flow freely out of the oral cavity, on one or both sides of the tongue, but because the tongue tip/blade also makes contact with the dento-alveolar region, the air is obstructed centrally in a similar manner to [t/d]. The ambiguous articulatory status of [l] may play a role in the results from Escartín Ortiz as well: after a vowel, [sl] patterns with /s/+stop but after a pause, it patterns with /s/+nasal.
26
Heather Goad
Table 3: Ibero-Romance learners of /s/+consonant clusters in Germanic. First language
Second language
Spanish
German English English Swedish
Brazilian Portuguese
Performance
Study Tropf (1987) Carlisle (1988) Carlisle (1991) Abrahamsson (1999)
English English English English
ʃl, ʃr, ʃn, ʃm > ʃp, ʃt sl > sm, sn sl > st sm, sn > sp, st, sk > sl (marginally significant) sm, sn > sp, st, sk sl > sn > st sm, sn, sl > sp, st, sk sl > sn > st
English
sp, st, sk > sl
Major (1996)
English English
sp, st, sk > sm, sn, sl sm, sn, sl > sp, st, sk (not significant) sl, sn > st
Rebello and Baptista (2006) Rauber (2006)
English
Escartín Ortiz (2005) Carlisle (2006) Rauber (2006) Carlisle and Cutillas Espinosa (2010)
Cardoso and Liakin (2009)
acquire such clusters of shallower sonority profiles before those of steeper profiles. It appears, then, that these learners have misanalysed /s/+consonant clusters as branching onsets rather than as coda+onset clusters. This, in turn, is inconsistent with Prediction 2 in (3), that learners whose first language has obstruent+sonorant clusters should not use this representation for /s/+consonant clusters in the second language. In the Brazilian Portuguese group of studies, a mixed profile of behaviour is observed: in Major (1996) and Rebello and Baptista (2006), performance on /s/+consonant reflects what one would expect if learners have correctly identified these clusters as coda+onset, rather than as branching onsets. The results of Cardoso and Liakin (2009) follow the pattern observed for Spanish and the Rauber (2006) study trends in this direction as well. We strive to explain these cross-language and cross-study differences in the next section.
4.3 In pursuit of an explanation We first address why the Spanish speakers behave differently from the speakers of East Asian languages in their acquisition of Germanic /s/+consonant clusters. It appears that the East Asian speakers, whose performance mirrors crosslinguistic patterns on /s/+consonant cluster well-formedness (Table 1), start from scratch, whereas the Spanish speakers, whose performance on /s/+consonant
Superset and subset grammars in second language acquisition
27
mirrors the patterns expected for obstruent+sonorant (steeper sonority rises preferred for both cluster types), redeploy their branching onset representation for /s/+consonant. Although this would lead to the observed differences between these two populations of learners, there are two problems that must be addressed before this analysis can be accepted. One, we cannot explain the different patterns observed between Spanish and Brazilian Portuguese in Table 3, two closely related languages that both have obstruent+sonorant clusters. Two, we arrive at the conclusion that the interlanguage grammars built by Spanish-speaking learners are not constrained by Universal Grammar in that, for many of these learners, the only well-formed /s/+consonant clusters would be /s/+sonorant in shape. In view of these problems, we consider an alternative explanation, one which stems from the phonetic properties and phonological behaviour of /s/ in dialects of Spanish and Brazilian Portuguese. In order to pursue this explanation, we must begin with some discussion of why it is /s/ (or some other type of sibilant) that displays unorthodox syllabification behaviour across languages. Recent research has proposed that segments are ordered to maximize their perceptibility (e.g., Wright 2004). At the left edge of syllables and/or words, obstruents are optimally followed by sonorants, as their place and voicing are reliably identified only once they are released into the following segment. In light of this, the relatively free distribution of /s/ in relation to what follows may appear puzzling. However, strident fricatives, unlike other obstruents, have robust internal cues for place and manner, which ensures that they can be accurately identified, even in non-optimal contexts, as when followed by stops. Since phonological constraints require all languages that have /s/+consonant clusters to permit /s/+stop (section 2), it is essential that the sibilant that appears in such clusters meet a minimum threshold for stridency. (Note that stridency is a gradient phenomenon; see Brannen 2011.) In languages where /s/ does not meet this threshold, word-initial /s/+consonant clusters may be absent altogether, as in Spanish and Brazilian Portuguese, or the language may choose to substitute another sibilant in place of /s/ in such clusters, as in German and Acoma. Each of these languages will be discussed further below. In order to achieve the characteristic stridency of /s/, a narrow constriction must be maintained at the constriction site (Shadle 1991). Apico-alveolar articulation is most conducive to realizing strident anterior coronals; laminal tongue posture, as in the production of /θ/, involves greater constriction width than /s/, and is thus incompatible with the minimum threshold for stridency required. We first examine coronal fricatives in German and Acoma from this perspective. When compared to English /s/, German /s/ involves greater constriction width, which results in a lowered spectral mean (Fuchs and Toda 2010). Considered alongside Narayanan, Alwan, and Haker (1995), who report greater
28
Heather Goad
constriction width for English /θ/ than for English /s/, and Jongman, Wayland, and Wong (2000), who observe lower spectral means for English /θ/ than for /s/, Fuchs and Toda arrive at the conclusion that German /s/ is quite [θ]-like. A less strident /s/ is practicable in German because there is no contrast between /s/ and /θ/ in this language. There is, however, a consequence for the shape of /s/+consonant clusters: German selects /ʃ/ over /s/ in such clusters, because /s/ is not strident enough, given the phonological requirement for /s/+stop to be well-formed in all languages with /s/+consonant (Goad 2012). Turning to Acoma, Miller (1965) observes that /s/ in this language is dental and that /s/ is followed by a “theta offglide” (Miller 1965: 13). Both of these suggest that /s/ is non-strident in Acoma, akin to German /s/. As Acoma is a language where /s/+consonant clusters are confined to /s/+stop (like French; see Table 1), if /s/ were to surface as such in /s/+consonant clusters, its perceptibility would be compromised. /s/+consonant clusters are instead realized as posterior [ʂ] or [ ʃ ] (contextually determined), similar to what is observed in German (Goad 2012). With this background, we now turn to consider the phonetic and phonological behaviour of /s/ in dialects of Spanish and Portuguese. We will conclude that /s/ is non-strident in dialects of both languages, which will serve as a possible explanation for why /s/+consonant clusters have the profile of branching onsets in the interlanguage grammars of Ibero-Romance learners of Germanic languages. In his comparison of Andalusian and Castilian Spanish, Romero (1995) experimentally confirms the observations of earlier researchers (e.g. Zamora Vicente 1967) that Andalusian and Castilian have different articulations for /s/. Castilian contrasts strident apico-alveolar /s/ and non-strident lamino-dental /θ/, whereas Andalusian has only one fricative in this region: laminal /s/ with a variable constriction location. Latin American dialects are like Andalusian in that they have similarly neutralized the contrast between /s/ and /θ/ (e.g., Lipski 1994). Romero experimentally demonstrates that the laminal tongue posture in Andalusian involves a reduction of gestural magnitude which, as discussed earlier, is inconsistent with the gestural requirements for stridency identified by Shadle (1991). In other words, Andalusian /s/ is low in stridency, and this likely holds for (some) Latin American dialects as well. Turning to the phonological behaviour of /s/ in Andalusian versus Castilian, Romero proposes that gestural reduction of Andalusian /s/ was the triggering condition for lenition (/s/ → [h] or Ø). Lenition of coda /s/ is particularly widespread throughout the Spanish-speaking world: outside of Andalusia, it occurs in the Caribbean and in coastal regions of Latin America (e.g., Alcina Franch and Blecua 1975; Lipski 1994). While Romero (1995) roots lenition in the articulatory properties of /s/, from a perceptual perspective as well, the process was
Superset and subset grammars in second language acquisition
29
likely facilitated by the weak stridency of Andalusian /s/, in that the inception of this process would not be perceptually costly. Conversely, lenition that specifically targets /s/ would be surprising if /s/ were high in stridency, as the process would be easily detected by listeners. We can thus conclude that, at least for some dialects of Spanish, the stridency of /s/ and the presence/absence of lenition are likely linked. Turning to Portuguese, a similar profile is found as concerns the phonetics of /s/. Conservative dialects, mostly in the north of Portugal, maintain two anterior sibilants, lamino-dental /s̪/ and apico-alveolar /s/ (Mateus and d’Andrade 2000). Most (southern) dialects have only one anterior fricative, which is described by Mateus and d’Andrade as dento-alveolar and by Emiliano (2009) as laminoalveolar. As for Brazilian Portuguese, Mateus and d’Andrade state that it converges with the southern dialects found in Portugal in this respect. Reference to dentality and laminality is consistent with a weakly strident /s/ in both of these varieties of Portuguese, as in Andalusian and possibly Latin American Spanish. In some dialects of Brazilian Portuguese, lenition (particularly deletion) of /s/ also occurs (e.g., Guy 1981 on the dialect spoken in Rio de Janeiro; Hora, Pedrosa, and Cardoso 2010 on the dialect spoken in Paraíba) but it appears that it is not as widely attested (or at least as extensively studied) as in Spanish; for instance, it is not mentioned at all in Mateus and d’Andrade’s (2000) phonological grammar. To summarize, phonetic and phonological evidence points toward nonstrident /s/ in dialects of both Spanish and Brazilian Portuguese. With the information available, it is not possible to determine if /s/ is non-strident in the dialects of Spanish spoken by the learners in Table 3 as well as in the dialects of Brazilian Portuguese spoken by the learners in Cardoso and Liakin (2009)7 and possibly Rauber (2006). However, if /s/ were non-strident in these dialects, then it would be non-strident in the (early) interlanguage grammars that these speakers build. /s/-initial clusters in these learners’ Germanic outputs would then be akin to /f/- or /θ/-initial clusters in other languages. That is, they would be represented as branching onsets, using a structure already available from the native language grammar. Under this analysis, these interlanguage grammars are possible grammars, and the performance displayed by these speakers in their second language, where /s/+consonant clusters with a steeper sonority profile are preferred over those with a shallower profile, is as expected.
7 Walcir Cardoso has informed me that the Brazilian Portuguese speakers in the Cardoso and Liakin (2009) study were from Belém (Pará), a dialect where /s/ lenition (to [h] and Ø) occur, mostly in word-final position, although the process is stigmatised.
30
Heather Goad
To conclude this section, we have argued that patterns in the acquisition of /s/+consonant clusters in English by learners whose first language has no leftedge clusters (East Asian) respect the phonotactic and representational differences observed for /s/+consonant versus obstruent+sonorant clusters. The interlanguage grammars built are thus constrained by Universal Grammar. Patterns in the acquisition of /s/+consonant clusters in Germanic by learners whose first language has left-edge obstruent+sonorant clusters (Ibero-Romance) do not necessarily respect the phonotactic and representational differences observed for /s/+consonant versus obstruent+sonorant. It was argued that this does not mean that these interlanguage grammars are not possible grammars. Instead, the source of parallels observed between /s/+consonant and obstruent+sonorant may lie in the low stridency of /s/ in the native language grammar: /s/+consonant clusters can therefore use the representation for branching onsets available from this grammar.
5 Learning a subset grammar In the preceding section, we were concerned with language pairs where the first language is a subset of the second language for initial /s/+consonant clusters. In such situations, direct positive evidence – discovering that a structure is well-formed from exposure to the ambient input – is available for learners to determine that /s/+consonant clusters are licit in the language being learnt. In this section, we turn to the inverse situation, a language pair, English–Brazilian Portuguese, where the first language is a superset of the second language: as discussed above, Brazilian Portuguese lacks initial /s/+consonant clusters. In these types of learning situations, direct positive evidence is often not available, and even when it is, it may not be very accessible to learners (see Schwartz and Goad to appear). Consider, for example, the potential utility of morphophonemic alternations. Although in Brazilian Portuguese there are alternations revealing that initial /s/+consonant is ill-formed (stems that begin with /s/+consonant undergo [i] epenthesis when they are not preceded by a vowel-final prefix (e.g., [isperar] ‘to hope, wait’ versus [pro-sperar] ‘to flourish, be successful’)), for learners to make use of this evidence, they must understand the morphological composition of words, which requires significant cross-form comparison. Alternatively, learners could rely on loanword adaptations (Trapman and Kager 2009) (e.g., English stand and snob are realized with initial [i] in Brazilian Portuguese (Cardoso and Liakin 2009)). However, if the learner of Brazilian Portuguese does not know the source language or the number of adaptations is too great to be able to recover
Superset and subset grammars in second language acquisition
31
the source word, this type of evidence will not be accessible. A similar argument can be made for learners relying on a comparison of cognates (e.g., Brazilian Portuguese especial corresponds to English special). They can be a useful source of positive evidence, but only if the learner’s native language (or another language that s/he knows) is genetically related to the second language.8 Must we conclude from this that learners often have to rely on negative evidence when learning a subset grammar? Perhaps not. In some acquisition contexts, learners can use indirect positive evidence where, as exemplified in section 3, the evidence for the structure of the grammar of one language is accessible through errors made in another language (Schwartz and Goad to appear). In the following section, I report on an experiment that was designed to test whether indirect positive evidence could be used by second language learners to inform them about aspects of the language they are acquiring.
5.1 Experiment The experiment examined whether English speakers, who are naïve to the structure of Brazilian Portuguese, can infer information about the phonotactics of Brazilian Portuguese after being exposed to indirect positive evidence in the form of Brazilian Portuguese-accented English.9 Our focus was on clusters in word-initial position: (i) /s/+consonant clusters, which are ill-formed and repaired through [i] epenthesis in Brazilian Portuguese and in Brazilian Portugueseaccented English; and (ii) obstruent+sonorant clusters, which are well-formed in Brazilian Portuguese. The predictions that we tested were provided earlier in (4); they are repeated in (5) for convenience: (5)
Predictions for learning a subset grammar (repeated from (4)): First language English – Second Language Brazilian Portuguese: 1. Learners exposed to the ill-formedness of /s/+consonant through indirect positive evidence will realise that /s/+consonant is ill-formed in the language being learnt; 2. Learners who recognize that /s/+consonant is ill-formed in the language being learnt will not overgeneralize this to obstruent+sonorant (following from (2)); 3. Learners exposed only to the ill-formedness of /s/+stop through indirect positive evidence will conclude that /s/+sonorant is also ill-formed (following from Table 1).
8 Thanks to Bernadette Plunkett for mentioning the potential benefit of cognates. 9 The experiment reported on here appears in expanded form in Schwartz and Goad (to appear).
32
Heather Goad
5.1.1 Methodology The participants were 32 native speakers of Canadian English (aged 19–33), with no higher than low-intermediate proficiency in another language. In addition, they had no knowledge of Portuguese or of Spanish, which has the same restrictions on /s/+consonant clusters (Table 1) and also repairs such clusters through epenthesis ([e] instead of [i]). Participants listened to one of two 7 minute dialogues between two native speakers of Brazilian Portuguese speaking Brazilian Portuguese-accented English; they were told that the native language of the speakers in the dialogue was Samoan to minimize the possibility that they would identify the native language as Portuguese (or Spanish). There were 48 tokens of /s/+consonant clusters, all of which appeared with prothetic [i]; place and manner of the consonant following /s/ were controlled. Each dialogue also contained 30 tokens of obstruent+ sonorant clusters (including a range of place, manner and voicing values for the obstruent). Participants were assigned to one of two conditions. Those in the NoSonorant condition listened to a dialogue in which all /s/+consonant clusters were /s/+stop in shape (16 tokens each of /sp, st, sk/); those in the Sonorant condition listened to a thematically near-identical dialogue in which /s/+consonant clusters were both /s/+stop and /s/+sonorant in shape (8 tokens each of /sp, st, sk, sm, sn, sl/). To test the predictions provided in (5), participants were orthographicallypresented with 120 novel words that began with /s/+consonant (n = 36), obstruent+sonorant (n = 18) or singleton consonant (n = 66). The /s/+consonant-initial words contained an equal number of stops and sonorants in second position, regardless of which dialogue participants were exposed to. Participants were asked to pronounce each word with the same accent as they had heard in the dialogue. 5.1.2 Results Prediction 1 in (5) states that learners exposed to the ill-formedness of /s/+ consonant through indirect positive evidence will realise that /s/+consonant is ill-formed in the language being learnt. To test this prediction, the data from the 32 participants were examined for whether or not they repaired /s/+ consonant-initial words; see Figure 1.10 (In all figures, the expected pattern is plotted in black; the unexpected pattern, in grey.) 10 All possible repairs to /s/+consonant clusters (i.e., prothesis [skesu] → [iskesu], anaptyxis [sikesu], deletion [sesu] or [kesu], and metathesis [seksu]) are collapsed together. The vast majority of participants used only prothesis (of all repairs employed, 95% involved prothesis). As performance decreases, however, the proportion of repairs of other types increases (see Schwartz and Goad to appear).
Superset and subset grammars in second language acquisition
33
Figure 1 shows that participants fell into three groups (determined by a binomial test). The behaviour of the 11 participants at the left end of the figure, henceforth the learners, was significantly higher than chance (ps ≤ 0.001); these individuals were able to use the indirect positive evidence available to determine that /s/+consonant clusters are ill-formed in the native language of the dialogue speakers. The behaviour of the 19 participants at the right end of the figure, henceforth the non-learners, was significantly lower than chance (ps ≤ 0.002); these individuals were not able to use the indirect positive evidence available and, instead, treated /s/+consonant clusters as well-formed in the dialogue speakers’ native language. The performance of the remaining two chance performers in the middle was neither significantly higher nor lower than chance. We can conclude from Figure 1 that Prediction 1 is supported, but only for a subset of participants: the 11 learners.
Figure 1: Responses to all /s/+consonant-initial words
Prediction 2 in (5) states that learners who recognise that /s/+consonant is ill-formed in the language being learnt will not overgeneralize this to obstruent+ sonorant clusters. Figure 2 provides the results for obstruent+sonorant clusters for the 11 learners. As can be seen, Prediction 2 is strongly supported: obstruent+ sonorant clusters were not repaired, as appropriate. This is not surprising: there is direct positive evidence in the data to which participants were exposed that obstruent+sonorant clusters are well-formed in the native language of the dia-
34
Heather Goad
logue speakers.11 Even if participants failed to notice this – on grounds that it may be more difficult to notice the presence of something expected (obstruent+ sonorant as well-formed), from the perspective of the participants’ native language grammar, than the presence of something unexpected (/s/+consonant as undergoing repair) – since different representations hold for these two types of clusters, any repair that participants apply to /s/+consonant should not be generalized to obstruent+sonorant.
Figure 2: Responses of learners to obstruent+sonorant-initial words
We turn finally to Prediction 3 in (5), which states that learners exposed only to the ill-formedness of /s/+stop through indirect positive evidence will conclude that /s/+sonorant is also ill-formed. This prediction follows from the typological options displayed in Table 1 and, thus, if interlanguage grammars are constrained by Universal Grammar, we should find support for it. To test this prediction, participants were randomly assigned to one of two conditions, as mentioned earlier (section 5.1.1). See (6): 11 What is surprising is that some participants overgeneralized the repair to words beginning with singleton /s/, which are also well-formed in the dialogue speakers’ grammar; see Schwartz and Goad (to appear) for discussion.
Superset and subset grammars in second language acquisition
(6)
a.
b.
35
NoSonorant condition: 16 participants received evidence for the illicit status of initial /s/+ consonant clusters only from /s/+stop: in the dialogue, all /s/+ consonant-initial target words were of the shape /s/+stop and all were preceded by prothetic [i]; Sonorant condition: 16 participants received evidence for the illicit status of /s/+consonant from both /s/+stop and /s/+sonorant: in the dialogue, /s/+consonantinitial target words were of the shape /s/+stop and /s/+sonorant and all were preceded by prothetic [i].
Of the 11 learners in Figure 1, eight had been assigned to the critical NoSonorant condition. Figure 3 shows that all eight appropriately generalized the pattern they learnt regarding the ill-formedness of initial /s/+stop to /s/+sonorant, strongly supporting Prediction 3 in (5). A paired t-test for the learners confirms that their behaviour on these two types of clusters was not significantly different (t = 1.8825, df = 7, p = 0.1018).
Figure 3: Responses of learners in NoSonorant condition (sT = /s/+stop; sR = /s/+sonorant)
In sum, the results of this experiment have demonstrated that: (i) indirect positive evidence can be used to uncover the phonotactic constraints operative in a second language, and (ii) constraints on the well-formedness of different types of left-edge clusters drive the shapes of interlanguage grammars. As appropriate, the learners did not overgeneralize the repair pattern they had learnt for initial
36
Heather Goad
/s/+consonant to obstruent+sonorant clusters. They did, however, appropriately generalize the repair pattern to /s/+sonorant, when they had only been exposed to evidence that /s/+stop is ill-formed in the language being learnt.
6 Conclusion This chapter has examined: (i) acquisition of the well-formedness of /s/+consonant clusters in language pairs where the first language is a subset of the second language grammar: East Asian and Ibero-Romance learners of Germanic; and (ii) acquisition of the ill-formedness of /s/+consonant clusters in a language pair where the first language is a superset of the second language grammar: English-speaking learners of Brazilian Portuguese. In case (i), learners differed in their treatment of /s/+consonant clusters: for some, relative difficulty reflected cross-linguistic patterns in cluster wellformedness on the sonority dimension, where accuracy on /s/+consonant clusters with a sonority plateau was better than on clusters with a sonority rise; for others, relative difficulty showed the inverse pattern: performance on clusters with a steep sonority profile was better than on clusters with a shallow or flat profile. This difference was argued to be due, in part, to the presence or absence of branching onsets in the native language grammar: East Asian learners of Germanic, who lack branching onsets in their native language, preferred /s/+ consonant clusters with a shallower profile, while some Ibero-Romance learners, whose native language grammar contains branching onsets, preferred /s/+ consonant clusters with a steeper profile. However, Ibero-Romance learners did not behave uniformly: Spanish learners preferred /s/+consonant clusters that mirrored branching onsets, while Brazilian Portuguese learners were divided in whether they patterned with Spanish learners or with East Asian learners. This difference was proposed to be due to the relative stridency of /s/ in the native language grammar. In dialects of Spanish and Brazilian Portuguese with low stridency /s/, the branching onset representation could be redeployed for /s/+ consonant in the second language, whereas in dialects with high stridency /s/, this representation would not be suitable. To test the validity of this explanation for the differing performance of IberoRomance speakers, future research should compare, for example, the behaviour of speakers of Andalusian Spanish and speakers of (northern varieties of) Castilian Spanish. Andalusian speakers, who show lenition of /s/, should have weakly strident /s/ in their native and interlanguage grammars. They should redeploy the branching onset representation from their native language to use
Superset and subset grammars in second language acquisition
37
for /s/+consonant clusters in the interlanguage, and show better performance on /s/+consonant clusters with a steep sonority profile over those with a shallow profile. Castilian speakers, who do not show lenition of /s/, should have highly strident /s/ in their native and interlanguage grammars. They should not see their branching onset representation as suitable for /s/+consonant clusters in the second language and should instead acquire this structure from scratch, showing better performance on /s/+consonant clusters with a sonority plateau over those that rise in sonority. Turning to case (ii), English-speaking naïve learners of Brazilian Portuguese, learners were exposed to data from a subset grammar in the cluster domain. In naturalistic situations like this, direct positive evidence may be lacking or inaccessible and we would thus expect the acquisition process to be protracted. However, it was shown that learners in an experimental setting can use indirect positive evidence to infer aspects of the grammar of a subset language: evidence from errors in the learner’s first language made by native speakers of the learner’s second language. The group of participants who were able to use the available indirect positive evidence – epenthesis errors in Brazilian Portugueseaccented English – appropriately did not overgeneralize the pattern of repair they had learnt for /s/+consonant clusters to obstruent+sonorant clusters. Further, when exposed only to epenthesis before /s/+stop, they correctly generalized the repair pattern to /s/+sonorant, indicating that they did not simply use analogy to perform the task. Although the experiment demonstrated the potential efficacy of indirect positive evidence, a glance back at Figure 1 shows a striking pattern of results: those participants who successfully used the available indirect positive evidence showed near ceiling performance, while most of those who were unsuccessful showed near floor performance. Future research must strive to determine the factors that enable some participants to so easily identify errors in accented speech and recognise that these reflect some mismatch between these speakers’ first language grammar and their second language, while others fail to make this connection. More generally, the potential benefits of indirect positive evidence in naturalistic learning remain to be explored, including comparisons of the developmental trajectories and time course of learners who do and do not have access to this type of evidence when learning a subset grammar. The final conclusion to draw from the patterns of behaviour observed in the acquisition of both the well- and ill-formedness of /s/+consonant clusters is that they appear to reflect natural language grammars, supporting the position that interlanguage grammars are constrained by Universal Grammar.
38
Heather Goad
Acknowledgement I would like to thank the audience at Romance Turn VI, where an earlier version of this work was presented, for questions and comments. The written version has also benefitted from the comments of two anonymous reviewers and the editors. Thanks as well to Hye-Young Bang, Walcir Cardoso, Guilherme Garcia, Lauren Garfinkle, Natália Guzzo, Kris Onishi, and especially Misha Schwartz, who co-authored the study reported on in section 5. This work was supported by grants from the Social Sciences and Humanities Research Council of Canada and from the Fonds de recherche du Québec – Société et culture.
References Abrahamsson, Niclas. 1999. Vowel epenthesis of /sC(C)/ onsets in by Spanish/Swedish interphonology: A longitudinal case study. Language Learning 49(3). 473–508. Alcina Franch, Juan & José Manuel Blecua. 1975. Gramática española. Barcelona: Ariel. Bley-Vroman, Robert. 1990. The logical problem of foreign language learning. Linguistic Analysis 20(1–2). 3–49. Brannen, Kathleen. 2011. The perception and production of interdental fricatives in second language acquisition. Montreal, Quebec: McGill University dissertation. Cardoso, Walcir & Denis Liakin. 2009. When input frequency patterns fail to drive learning: The acquisition of sC onset clusters. In Michael A. Watkins, Andréia Rauber & Barbara O. Baptista (eds.), Recent Research in Second Language Phonetics/Phonology, 174–202. Newcastle Upon Tyne: Cambridge Scholars Publishing. Carlisle, Robert. 1988. The effects of markedness on epenthesis in Spanish/English interlanguage phonology. Issues and Developments in English and Applied Linguistics 3. 15–23. Carlisle, Robert. 1991. The influence of syllable structure universals on the variability of interlanguage phonology. In Angela Della Volpe (ed.), The Seventeenth LACUS Forum 1990, pp. 135–145. Lake Bluff, IL: Linguistic Association of Canada and the United States. Carlisle, Robert. 2006. The sonority cycle and the acquisition of complex onsets. In Barbara O. Baptista & Michael A. Watkins (eds.), English with a Latin beat: Studies in Portuguese/ Spanish English interphonology, pp. 105–138. Amsterdam: John Benjamins. Carlisle, Robert & Juan Antonio Cutillas Espinosa. 2010. The production of /.sC/ onsets in a markedness relationship: A longitudinal study. In Katarzyna Dziubalska-Kołaczyk, Magdalena Wrembel & Małgorzata Kul (eds.), Proceedings of the 6th International Symposium on the Acquisition of Second Language Speech, New Sounds 2010, pp. 67–72. http://www. wa.amu.edu.pl/newsounds/files/proceedings/proceedings_quotable_version.pdf. Clahsen, Harald & Pieter Muysken. 1986. The availability of Universal Grammar to adult and child learners: A study of the acquisition of German word order. Second Language Research 2(2). 93–119. Clahsen, Harald & Pieter Muysken. 1989. The UG paradox in L2 acquisition. Second Language Research 5(1). 1–29.
Superset and subset grammars in second language acquisition
39
Clements, G. Nick. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary Beckman (eds.), Papers in laboratory phonology I: Between the grammar and physics of speech, 283–333. Cambridge: Cambridge University Press. Detey, Sylvain & Jean-Luc Nespoulous. 2008. Can orthography influence second language syllabic segmentation? Japanese epenthetic vowels and French consonantal clusters. Lingua 118(1). 66–81. Emiliano, António. 2009. Fonética do português europeu: Descrição e transcrição. Lisbon: Guimarães. Enochson, Kelly. 2012. L2 production of English onset sC and CC clusters. Proceedings of the International Symposium on the Acquisition of Second Language Speech. Concordia Working Papers in Applied Linguistics 5, pp. 171–184. Department of Education, Concordia University, Montreal, Quebec. Escartín Ortiz, Claudia. 2005. The development of sC onset clusters in Spanish English. Montreal, Quebec: Concordia University MA thesis. Fikkert, Paula. 1994. On the acquisition of prosodic structure. The Hague: HAG. Fuchs, Susanne & Martine Toda. 2010. Do differences in male versus female /s/ reflect biological or sociophonetic factors? In Susanne Fuchs, Martine Toda & Marzena Zygis (eds.), Turbulent sounds. An interdisciplinary guide, 281–302. Berlin: Mouton de Gruyter. Goad, Heather. 2011. The representation of sC clusters. In Marc van Oostendorp, Colin Ewen, Elizabeth Hume & Keren Rice (eds.), The Blackwell companion to phonology, 898–923. Oxford: Wiley-Blackwell. Goad, Heather. 2012. sC clusters are (almost always) coda-initial. The Linguistic Review 29(3). 335–373. Goldsmith, John. 1990. Autosegmental and metrical phonology. Oxford: Blackwell. Guy, Gregory R. 1981. Linguistic variation in Brazilian Portuguese: Aspects of the phonology, syntax, and language history. Philadelphia: University of Pennsylvania dissertation. Hall, Tracy Alan. 1992. Syllable structure and syllable-related processes in German. Tübingen: Max Niemeyer. Hora, Dermeval da, Juliene Lopes Ribeiro Pedrosa & Walcir Cardoso. 2010. Status da consoante pós-vocálica no português brasileiro: Coda ou onset com núcleo não preenchido foneticamente? Letras de Hoje 45(1). 71–79. Hulst, Harry van der. 1984. Syllable structure and stress in Dutch. Dordrecht: Foris. Jongman, Allard, Ratree Wayland & Serena Wong. 2000. Acoustic characteristics of English fricatives. Journal of the Acoustical Society of America 108(3). 1252–1263. Kaye, Jonathan. 1992. Do you believe in magic? The story of s+C sequences. SOAS Working Papers in Linguistics and Phonetics 2. 293–313. Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud. 1990. Constituent structure and government in phonology. Phonology 7(2). 193–231. Kim, Hyouk-Keun. 2000. The interlanguage phonology of Korean learners of English: A computational implementation of optimality theoretic constraints. Washington, DC: Georgetown University dissertation. Kwon, Bo-Young. 2006. Korean speakers’ production of English consonant clusters: Articulatory and perceptual accounts. East Lansing, MI: Michigan State University dissertation. Lipski, John. 1994. Latin American Spanish. London: Longman. Major, Roy. 1996. Markedness in second language acquisition of consonant clusters. In Robert Bayley & Dennis R. Preston (eds.), Variation and second language acquisition, pp. 75–96. Amsterdam: John Benjamins.
40
Heather Goad
Mateus, Maria Helena & Ernesto d’Andrade. 2000. The phonology of Portuguese. Oxford: Oxford University Press. Martz, Chris. 2007. Production of onset consonant clusters/sequences by adult Japanese learners of English. Bloomington, IN: Indiana University dissertation. Miller, Wick. 1965. Acoma grammar and texts. Berkeley and Los Angeles: University of California Press. Narayanan, Shrikanth, Abeer Alwan & Katherine Haker. 1995. An articulatory study of fricative consonants using magnetic resonance imaging. Journal of the Acoustical Society of America 98(3). 1325–1347. Rauber, Andréia S. 2006. Production of English initial /s/-clusters by speakers of Brazilian Portuguese and Argentine Spanish. In Barbara O. Baptista & Michael A. Watkins (eds.), English with a Latin beat: Studies in Portuguese/Spanish English interphonology, 155– 167. Amsterdam: John Benjamins. Rebello, Jeanne T. & Barbara O. Baptista. 2006. The influence of voicing on the production of initial /s/-clusters by Brazilian learners. In Barbara O. Baptista & Michael A. Watkins (eds.), English with a Latin beat: Studies in Portuguese/Spanish English interphonology, 139–154. Amsterdam: John Benjamins. Romero, Joaquin. 1995. An articulatory view of historical /s/-aspiration in Spanish. Rivista di Linguistica 7(1). 191–208. Schwartz, Misha & Heather Goad. to appear. Indirect positive evidence in the acquisition of a subset grammar. Language Acquisition. DOI: 10.1080/10489223.2016.1187616 Shadle, Christine H. 1991. Source parameters for the fricative consonants /s, ʃ, ç, x/. Journal of the Acoustical Society of America 89(4B). 1893. Steriade, Donca. 1982. Greek prosodies and the nature of syllabification. Cambridge, Mass: MIT dissertation. Trapman, Mirjam & René Kager. 2009. The acquisition of subset and superset phonotactic knowledge in a second language. Language Acquisition 16(3). 178–221. Tropf, Herbert. 1987. Sonority as a variability factor in second language phonology. In Allan James & Jonathan Leather (eds.), Sound patterns in second language acquisition, 173– 191. Dordrecht: Foris. Wright, Richard. 2004. A review of perceptual cues and cue robustness. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.), Phonetically based phonology, 34–57. Cambridge: Cambridge University Press. Zamora Vicente, Alonso. 1967 [1960]. Dialectología española, 2nd edn. Madrid: Editorial Gredos.
Harald Clahsen and João Veríssimo
2 Dual Morphology in the mental lexicon: Experimental evidence from the acquisition and processing of Greek and Portuguese Abstract: This chapter discusses the role of morphological and morphosyntactic information in learning and processing inflected word forms, a controversial issue in language acquisition and language processing research. We present results from i) the development of verb inflection (specifically perfective vs. imperfective past tense) in Greek child language and ii) word recognition and elicited production experiments testing inflected verb forms from different conjugational classes in Portuguese. We argue that purely analogy-based models of the mental lexicon are insufficient and suggest instead that a dual architecture of morpho-lexical representation, incorporating both morphological rules and lexical associations, provides a better account for the dissociations found in studies of morphological acquisition, generalization and processing.
1 Introduction In psycholinguistic models of language processing, the mental lexicon is conceived of as a repository to permit efficient representation and processing of words and their component parts; see, for example, Indefrey and Levelt (2004) for language production and Longworth and Marslen-Wilson (2011) for language comprehension. Entries in the mental lexicon are thought to consist of arbitrary mappings between form-level (phonological and orthographic) and meaninglevel (semantic) information. The question, however, of whether the mental lexicon also encodes morphological and morphosyntactic information is controversial, and the mechanisms involved in learning and processing inflected and derived words are not fully understood. Broadly speaking, two approaches can be distinguished in psycholinguistic and neurocognitive research on morphologically complex words. One view holds that there are no distinct operations or representations for encoding morphological information and that inflected or derived Harald Clahsen and João Veríssimo, Universität Potsdam
42
Harald Clahsen and João Veríssimo
words can in principle be handled in the same way as simplex word forms that do not have any internal morphological structure (e.g., Baayen, Milin, Filipović Đurđević, Hendrix, and Marelli 2011; Bybee and Hopper 2001; Feldman and Weber 2012; Gonnerman, Seidenberg, and Andersen 2007; Plaut 2011; Westermann and Ruh 2012). This approach has often been implemented into connectionist models (see McClelland and Patterson 2002, for a review) in which all morphologically complex words are learned, stored and processed within a single associative system that directly maps orthographic or phonological codes onto semantic codes, without explicitly encoding morphological and syntactic structure. These models have eliminated morphology – and more generally grammar – as a target for the acquisition and as a resource for the processing and the mental representation of language. An alternative approach is represented by a family of dual-mechanism models which hold that morphologically complex word forms can be processed and learned both associatively, that is, through stored full-form representations and by rules that decompose or parse inflected word forms into morphological constituents (see Clahsen 2006, for a review). This approach is also in line with much linguistic research on morphology that distinguishes between productive morphological rules or operations and exceptions that are supposed to be lexically listed; see, for example, Kiparsky (1982), Aronoff (1976) and much related work. Within psycholinguistics, the perhaps most popular dual-route account is Pinker’s (1999) words-and-rules model, which incorporates one very basic aspect of linguistic morphology, namely the distinction between concatenation and suppletion. The model distinguishes between regular forms such as walked which are decomposed or grammatically parsed into a stem X and a suffix –ed and irregular unanalysed forms such as held, which are recognized by direct lexical look-up. As shown by a large number of psycholinguistic studies, this binary distinction works well to account for results from acquisition, processing and disorders of relative simple systems such as the English past tense (see Pinker and Ullman 2002, for review). However, as we will see below, extra layers of complexity are needed for more complex phenomena. While much previous work on this theoretical controversy has examined a narrow set of phenomena, for example, regular versus irregular inflection, in a comparatively small number of languages (with a strong focus on English), there are also cross-linguistic psycholinguistic studies from a range of typologically different languages that have contributed to this debate. In this chapter, we present results from two sets of experimental studies covering different morphological phenomena other than the English past tense, (i) perfective past tense formation in Greek, and (ii) conjugational classes and verb stem formation processes in Portuguese (and other Romance languages). It will be shown that
Dual Morphology in the mental lexicon
43
single-mechanism associative models are insufficient to account for the experimental results from these studies and that dual-mechanism models provide a much better fit. Our general conclusions from these studies will be that morphological structure plays a crucial role in language acquisition and processing and that morphology helps to better understand results from psycholinguistic experiments.
2 Acquisition of the perfective past tense in Greek The first set of studies concerns the acquisition of inflectional morphology in child language development. Our focus is on children’s generalization processes within this domain. We specifically report results on the development of the perfective past tense in Greek child language focusing on a large-scale developmental study (Stavrakaki and Clahsen 2009) and discuss a subsequent study (Karaminis and Thomas 2010) which supposedly reproduces the human data in a purely associative single-mechanism model. Here we will argue that this model is not particularly successful and fails to accurately simulate generalization processes of the perfective past tense in Greek.
2.1 Linguistic properties of the perfective past-tense in Greek Tense and aspect marking are closely linked in Greek. Within the past tense, Greek distinguishes between perfective and imperfective aspect, with the former being used when an action or an event was completed, and the latter when it is seen as repeated, in progress or habitual. Unlike the imperfective past which does not have any distinct affix, the so-called sigmatic perfective past tense form involves a segmentable suffix, –s– (for the Greek letter sigma), as well as stem changes. In addition, there is a lexically restricted class of so-called nonsigmatic perfective past-tense forms that involve unsystematic and even suppletive stem changes and no distinct perfective past-tense affix. Furthermore, both types of perfective past tense forms may comprise a prefix (e–) which is not morphologically determined but is, instead, attached when the verb stem is monosyllabic and starts with a consonant. Finally, perfective past tense forms contain person and number endings, for example, –a for the 1st singular; see (1) and (2) for examples:
44 (1)
Harald Clahsen and João Veríssimo
Sigmatic forms: a. e-vap-s-a Prefix-paint-PFV.PST-1SG ‘I painted’ (from: vaf-o ‘I paint’) b.
(2)
e-di-s-a Prefix-dress-PFV.PST-1SG ‘I dressed’ (from: din-o ‘I dress’)
Non-sigmatic forms: a. id-a saw-1SG ‘I saw’ (from vlep-o ‘I see’) b.
e-gir-a Prefix-bent-1SG ‘I bent’ (from: gern-o ‘I bend’)
The sigmatic and the non-sigmatic perfective past tense also differ in terms of frequency. There are many more verbs that require a sigmatic past tense form than that require a non-sigmatic one, with a ratio of 14 to 1 in a large corpus of Greek examined by Stavrakaki and Clahsen (2009). From a dual-morphology perspective, the –s suffixation is the grammatical way of forming a perfective past tense in Greek, by using an affixation rule that operates over a variable (e.g., [Verb stem] → [Verb stem + -s]+past,+perf ), whereas the non-sigmatic stems need to be learned and lexically stored as irregular forms. According to this account, children are expected to apply the sigmatic –s as a default, except for existing verbs for which they have learned a nonsigmatic form that blocks the –s rule, or for novel verbs that are phonologically similar to existing non-sigmatic ones. From the perspective of a single-mechanism account that generalizes on the basis of frequency and phonological similarity (Karaminis and Thomas 2010), it can also be expected that, due to their higher frequency, the sigmatic perfective past tense is overapplied to non-sigmatic verbs. However, one might also expect all verbs, sigmatic and non-sigmatic, to show effects of similarity and be more easily applied and generalized in their tight phonological contexts.
2.2 Elicited production of the perfective past tense in Greek-speaking children and adults The empirical findings we focus on come from Stavrakaki and Clahsen (2009). The study included two experiments, an elicited production and an acceptability judgment task. Here we focus on the results of the elicited production task,
Dual Morphology in the mental lexicon
45
which was performed with 78 children in six age groups (age range: 3;5 to 8;5) and 10 adult native speakers. The materials included 50 verb stimuli in five conditions, all of which were presented in sentence contexts: (i) 10 existing verbs that required sigmatic forms; (ii) 10 existing verbs that required non-sigmatic forms; (iii) 10 novel verbs that rhymed with existing verbs that take sigmatic forms; (iv) 10 novel verbs that rhymed with existing verbs that take non-sigmatic forms; and (v) 10 novel verbs that did not rhyme with any existing Greek verb. The experimental procedure for the elicited production task was as follows. The experimenter first introduced an existing or novel verb by pointing to the upper picture of a pair of pictures, depicting, for example, a girl untying a dog, by saying Here, the girl is untying the dog. Thereafter, the experimenter pointed to the lower picture of the display where the action had been completed and the dog is shown to be untied, by asking And what did the girl do here? Participants were expected to say The girl untied the dog, using a perfective past tense form of the targeted verb. The same design was used for novel verbs. Consider first the results on existing verbs. Figure 1 presents accuracy scores for sigmatic and non-sigmatic perfective past-tense formation, and Figure 2 presents a breakdown of errors. Figure 1 shows that accuracy scores for sigmatic forms were high (>80%), except for the youngest age group of children, the 3- to 4year-olds. By contrast, accuracy scores for non-sigmatic forms were lower than those for sigmatic forms throughout, gradually increasing with age, although even the oldest age group of children, the 8- to 9-year-olds, did not yet reach adult level. Figure 2 shows that children of all age groups produced –s overregularizations, that is, overapplications of sigmatic forms for verbs that require
Figure 1: Percentages of existing (sigmatic/non-sigmatic) verb forms that were correct
46
Harald Clahsen and João Veríssimo
Figure 2: Percentages of errors of each type. “Sigmatic Overapplications” and “Non-sigmatic Overapplications”: overapplications of sigmatic/non-sigmatic forms to existing non-sigmatic and sigmatic verbs, respectively; “Incorrect Stem/Verb Sigmatic” and “Incorrect Stem/Verb Non-sigmatic”: other stem or verb errors on non-sigmatic and sigmatic existing verbs
non-sigmatic forms (shown in straight black), whereas the reverse, that is, overapplications of non-sigmatic forms for verbs that require sigmatic forms (shown in straight grey) were practically non-existent. Children also produced stem and verb errors, instead of correct existing non-sigmatic (shown in dashed black) and sigmatic forms (shown in dashed grey). The vast majority of these stem/ verb errors involved producing the unmarked present tense stem of a given verb or the perfective past tense form of a different verb (which was often semantically related to the target verb). Such stem/verb errors were less common than –s overregularizations (except in the youngest group) and occurred in comparable rates for both existing sigmatic and non-sigmatic verbs. In addition, there is a clear age effect for stem/verb errors, which unlike –s overregularizations, substantially decrease after age 5. Concerning novel verbs, Figure 3 presents percentages of sigmatic and nonsigmatic past tense forms produced for (i) novel verbs that rhyme with existing sigmatic, (ii) existing non-sigmatic, and for (iii) non-rhyming verbs. The three
Dual Morphology in the mental lexicon
47
Figure 3: Percentages of novel verbs produced with sigmatic or non-sigmatic perfective past tense forms
black lines show that sigmatic forms generalize to all kinds of novel verbs at rates of 65% to 95% (with slightly lower proportions only in the youngest group), and even to those verbs that only rhyme with existing non-sigmatic ones. In contrast to that, as indicated by the three grey lines, non-sigmatic forms are rarely applied to novel verbs, and if so, most commonly to novel verbs that rhyme with existing non-sigmatic ones. Importantly, rates of generalization of sigmatic –s are comparable for novel verbs that rhyme with sigmatic forms and for verbs that do not rhyme with any form in the language. Thus, while sigmatic forms generalize beyond (and independently of) similarity, non-sigmatic forms only generalize by analogy, for both children and adults.
2.3 Discussion A dual-morphology account that distinguishes between a grammatical rule for sigmatic perfective past-tense formation and exemplar-based learning and storage of non-sigmatic forms provides a straightforward account for the reported find-
48
Harald Clahsen and João Veríssimo
ings from both existing and novel verbs. Sigmatic forms generalize by default, on the basis of a rule that contains a variable, which can be freely applied to any verbal stem, including cases in which retrieval of a non-sigmatic form fails or when such access is not possible (as in the case of non-rhymes). Non-sigmatic forms, on the other hand, are stored in lexical memory and therefore can be generalized, through analogy, to novel verbs that are similar to existing nonsigmatic verbs. Developmental changes in children’s performance can also be explained in these terms. Children have to learn irregular forms, in this case non-sigmatic stems, which requires repeated exposure over time; hence, as children get older, they become more accurate at inflecting non-sigmatic verbs. In addition, children typically overapply unmarked default patterns, in this case –s affixation and present-tense stems, to irregular (non-sigmatic) forms, and the proportions of these overregularizations decrease over time as children gradually learn the correct non-sigmatic forms. Note, however, that children seem to manipulate stems and –s affixation independently. Stavrakaki and Clahsen (2009: 125) observed that –s overregularizations were not only combined with unmarked (present tense) stem forms, but also occurred on different marked stems of a given verb and even on non-existing stem forms, indicating that stem formation and affixation are controlled and learned separately from each other. There was one additional child/adult difference, namely that in contexts in which a perfective past-tense marking was required (and adults and older children did indeed produce these forms), the youngest group of children, the 3- to 4-year-olds, sometimes produced sentences with verb forms consisting of present-tense stems without the perfective past-tense affix. These kinds of errors are reminiscent of a phenomenon familiar from early child language, the omission of affixes encoding syntactic finiteness, that is, tense and agreement. Children learning, for example, Germanic languages have been found to omit finiteness markings at this age (e.g., the past-tense –ed in English or the 2nd sg. –st in German) and to produce bare stems or infinitives instead (Clahsen, Penke, and Parodi 1993; Poeppel and Wexler 1993). Their syntactic representations are thought to be underspecified for tense and/or agreement at this age allowing them to produce sentences without finite verb forms. This may also apply to Greek-speaking children, leading to omissions of the perfective pasttense affix and to overusage of unmarked stems. Leaving this syntactic type of omission error aside, children’s morphological system for perfective past tense inflection does not seem to differ from the adult one in the age range tested, except that children’s inventory of irregular (non-sigmatic) forms takes time to develop.
Dual Morphology in the mental lexicon
49
As an alternative to a dual-morphology account of the reported findings, Karaminis and Thomas (2010) argued that human performance on the Greek perfective past tense as well as its development during childhood can also be derived from a single-mechanism connectionist model similar to those that have been proposed for the English past tense (McClelland and Patterson 2002). Karaminis and Thomas (2010) employed a three-layered feed-forward connectionist network which they claim is capable of reproducing the human data with respect to the developmental pattern of accuracy levels and error types. A detailed comparison to the human data reveals, however, that this model is not particularly successful. To see this, Figure 4 presents a comparison of the output of Karaminis and Thomas’ simulation model on verbs that require non-sigmatic perfective past tense forms to the corresponding results from the child and adult participant groups tested in Stavrakaki and Clahsen’s (2009) elicited production experiment. The black line depicts proportions of correct non-sigmatic perfective past tense markings, the green line overapplications of sigmatic forms, and the grey one stem/verb errors.1 The dashed yellow box depicts a window in the training time (from epoch 101 onwards), identified by Karaminis and Thomas as matching with the corresponding age-related developmental curve reported by Stavrakaki and Clahsen for Greek children.
Figure 4: Comparison of output of Karaminis and Thomas’ (2010) simulation model to human data from Stavrakaki and Clahsen (2009) for verbs that require non-sigmatic perfective past tense forms in Greek
1 In addition, Karaminis and Thomas (2010) included in their graph overapplications of the 3rd sing. to 2nd sing. contexts (represented by the red line), a syntactic type of error that has been reported for early stages of Greek child language (Varlokosta, Arhonti, Thomaidis and Joffe 2008). As we are focusing on the Greek perfective past tense, these types of agreement errors will not be discussed here.
50
Harald Clahsen and João Veríssimo
One finding the model does indeed capture is the gradual increase of accuracy scores for non-sigmatic verbs. Error patterns are not, however, accurately reproduced. Comparing the grey and the green lines in both panels reveals that the model produces more stem/verb errors than sigmatic (-s) overregularizations, whereas the opposite is true for the child groups (except for the youngest one) in Stavrakaki and Clahsen’s (2009) study. Hence, the model underestimates the frequency of -s overregularizations. Furthermore, as shown in the left panel of Figure 4, the model stops producing –s overapplications (i.e., the green line goes down to zero), when accuracy rates for non-sigmatic forms are around 70%, which corresponds to the level of accuracy reached by 6– to 7–year-old children. The human data shown in the right panel, however, show that children at this age still overapply the sigmatic –s (at a rate of 18.1%), and that is still the case even for the two older groups (7 to 8: 18.6%; 8 to 9: 7.5%). Hence, the model also underestimates the period of time during child development in which –s overregularizations are produced. In what concerns generalization to novel verbs, Stavrakaki and Clahsen (2009) found that children and adults generalize the sigmatic –s to both rhymes and non-rhymes, whereas non-sigmatic patterns only generalize to novel verbs that rhyme with existing non-sigmatic forms. It is not clear whether the simulation model is able to reproduce this contrast. Karaminis and Thomas (2010) did test their model on novel items (created by changing the phonemes of the first syllable of every form in the training set), but the authors do not report separate results for different types of phonological similarity, that is, for rhyming and non-rhyming verbs. Furthermore, the generalization performance of the simulation was only assessed at the end of training, but not at different points during the training phase, making it impossible to compare the developmental trajectory of generalization in children to the output of the model. The generalization behaviour at the end of training can, however, be compared with the adult data obtained by Stravakaki and Clahsen (2009), and suggests that the network model underestimates the application of the sigmatic rule. Whereas the simulation generalized -s to the set of novel items in 71% of the cases, Greek adults applied the sigmatic past tense to 92% of novel verbs that rhymed with sigmatic forms and 91% of non-rhyming verbs. Taken together, the outcome from the connectionist simulation is quite disappointing. The model only captured one aspect of the human data, namely the gradual increase of non-sigmatic correctness scores. By contrast, the distribution and the development of the (over)generalization patterns were not accurately reproduced. It is of course possible that future single-mechanism models will do better than Karaminis and Thomas’ (2010) model. Yet, it is hard to imagine how a single-mechanism model should at the same time simulate the default
Dual Morphology in the mental lexicon
51
behavior of the sigmatic and the similarity-based generalizations of the nonsigmatic forms.
3 Inflectional classes and conjugational stems The second set of studies examines data from experimental studies conducted with adults, on morphological generalization, as well as of computational simulations of these findings. The specific phenomenon we examine is inflectional classes and conjugational stems, which are found in many languages including – within the Indo-European language family – all the Romance languages. Most previous experimental research on morphology focuses on traditional inflectional and derivational morphemes. Priming experiments, for example, have examined morphemic decomposition, that is, how words are broken into morphemic units (e.g., a regular past tense form such as walked into stem plus –ed). What psycholinguists have not yet considered is that there are many properties of inflected words that are not morphemic in nature. One case in point comes from languages that have inflectional classes, for example, verbal stem formation in Romance languages.
3.1 Linguistic properties of verbal stems in Romance languages Most verbal stems in Romance languages are combinatorial, and consist of a verb root, together with a theme vowel (e.g., in Italian: cant-a ‘sing’), which places the verb in one of three conjugation classes, –a for 1st conjugation, –e for 2nd and –i for 3rd conjugation. Beyond being class markers, theme vowels specify no other features; such constituents are semantically and syntactically empty and convey only purely morphological information (Aronoff 1994). In addition, conjugation classes in Romance languages display a striking discrepancy in productivity. The 1st conjugation welcomes new verbs and qualifies as a default process (Marcus, Brinkmann, Clahsen, Wiese, and Pinker 1995), while the 2nd and 3rd conjugations constitute a static set of lexical items. Finally, non-allomorphic verbal stems in the Romance languages are perfectly regular and plausibly analysed as the result of a concatenation operation. From a linguistic perspective, inflectional classes and conjugational stems represent abstract morphological categories which cannot be reduced to phonological, semantic or syntactic properties. They are therefore better suited to
52
Harald Clahsen and João Veríssimo
examine whether pure morphology is required for understanding how inflected words are processed and mentally represented than the commonly studied inflectional morphemes. Broadly speaking, three accounts have been given for these kinds of stem forms. Under rule-based generative approaches (e.g., De Boer 1981, for Italian; Villalva 2000, 2003, for Portuguese), all conjugational stems are thought to be products of operations over variables and have representations with internal morphological structure (i.e., [root + theme vowel]). Alternatively, conjugational stems may be stored together with inflectional affixes as wholes (e.g., Eddington 2002). Finally, from a dual-mechanism perspective, the 1st conjugation may invoke a rule-based stem-formation operation that applies to a grammatical category (i.e., verb root), while 2nd and 3rd conjugation stems are stored as exceptions (e.g., Say and Clahsen 2002). These accounts make different predictions for processing and with respect to their generalization properties. For example, if inflected verb forms such as cantare ‘to sing’ in Italian or falar ‘to speak’ in Portuguese are processed according to their morphological structure, we would expect that they are not only morphemically decomposed but that, corresponding to their structure ([[cant]a], [[fal]a]), the stems canta- and fala- are further segmented down to their roots, [cant] and [fal]. Thus, experimental designs that are sensitive to the internal morphological structure of an inflected word should yield decomposition effects for such forms that are independent of the inflectional morphemes. The question will then be whether the three inflectional classes of Romance verbs produce the same stem-decomposition effects. Furthermore, if 1st conjugation stems are generated by rule, while 2nd and 3rd conjugation stems are stored forms, we expect that while 1st conjugation stems generalize by default, 2nd and 3rd conjugation stems generalize only rarely, in a way that might be sensitive to phonological similarity. In what follows we discuss lexical priming and elicited production studies which have tested these predictions experimentally.
3.2 Priming patterns for conjugational stems To test for potential differences between the different conjugational stems in on-line language comprehension, Veríssimo and Clahsen (2009) conducted a morphological priming experiment with adult native speakers of European Portuguese. In morphological priming experiments, participants are presented with a morphologically complex prime word before a different word form with the same stem or root as a target word, for example, walked as a prime for the target walk. For the target word, participants have to perform a lexical decision task, or they have to read the target word aloud. A robust finding from priming
Dual Morphology in the mental lexicon
53
experiments is that response latencies to target words are shorter when they are preceded by morphologically related than by unrelated prime words. Strong morphological priming effects have been attributed to direct stem/root access. For example, a word form such as walked – presented as a prime – is recognized as a combinatorial form ([walk][ed]), isolating the base stem, which then directly facilitates recognition of the target word walk. Veríssimo and Clahsen (2009) used a cross-modal priming design with visually presented target words immediately preceded by auditorily presented prime words. Here we will focus on two experimental conditions tested by Veríssimo and Clahsen (2009) which included 1st conjugation forms and 3rd conjugation forms as visually presented targets that displayed no phonological changes in the root (see (3) for an example stimulus set). Visual targets were first person singular present tense forms, that is, root-based forms without a theme vowel (e.g., limit-o or adquir-o), and auditory primes were either the same form (Identity condition), the infinitive (stem-based) form of the same verb (Test condition), or the infinitive form of an unrelated verb (Control condition): (3)
Example stimulus set: Cross-modal priming Verb Type 1st conjugation
Primes Identity: Test: Control:
limito limitar desejar ‘to wish’
3rd conjugation
Identity: Test: Control:
adquiro adquirir investir ‘to invest’
Targets limito ‘(I) limit’ adquiro ‘(I) acquire’
Crucially, the target forms only shared the root, rather than the stem, with the primes. Primes and targets were matched across conditions in lemma frequency, word-form frequency, and number of phonemes, letters, syllables, and orthographic neighbors. The results are illustrated in Figure 5, which presents mean lexical decision response latencies (in milliseconds) on the target words (e.g., limito or adquiro) for the three prime types and the two conjugation classes. As can be seen from Figure 5, the two conjugational classes produced different priming effects. First conjugation forms such as limitar yielded the same amount of facilitation on the recognition of the target word (e.g., limito) as an identity prime. Third conjugation forms, on the other hand, produced significantly less facilitation than the repetition-priming condition and a smaller priming effect (relative to the control condition) than 1st conjugation forms. Hence, 1st conjugation stems yielded a
54
Harald Clahsen and João Veríssimo
full root priming effect, whereas 3rd conjugation stems showed a partial priming effect. These findings provide supporting evidence from Portuguese for the dualmechanism account of stem representation originally proposed by Say and Clahsen (2002) for Italian. The full priming effect obtained for 1st conjugation stems indicates that the central lexical entry of these verbs is the basic root, which is activated to the same degree by a stem-based prime and a root-based target. The partial priming effect for 3rd conjugation stems, on the other hand, indicates that the lexical entries of these verbs contain both the basic root (e.g., adquir-) and a separate form for the infinitive stem (e.g., adquiri-). One way of representing these verbs would be in terms of a structured lexical entry (e.g., Wunderlich 1996), such that the stem constitutes a subnode to the root, akin to what has been proposed for verbs that have marked allomorphic stems, for example, pedir – pido ‘to ask – I ask’ in Spanish (Linares, Rodriguez-Fornells, and Clahsen 2006). Thus, a prime-target pair such as adquirir → adquiro activates different subentries of the same lemma, the stem in the case of adquirir and the root in adquiro, and hence the indirect (reduced) repetition-priming effect for 3rd conjugation verbs.
Figure 5: Mean target response latencies (in ms) for 1st and 3rd conjugation verb forms in Portuguese preceded by “Identity”, “Test”, and “Control” prime words (from Veríssimo and Clahsen 2009).
Dual Morphology in the mental lexicon
55
3.3 Generalization properties of conjugational stems The properties of conjugational stems also represent a very clear-cut and unconfounded test case to examine the nature of morphological generalization processes, that is, the question of how novel verbs are morphologically incorporated into a language. The three representational accounts mentioned above provide different answers to this question. If all conjugational stems are the product of morphological rules or operations, as suggested by generative approaches, one would expect all conjugations to be generalized in the same manner, that is, independently of the phonological properties of novel roots. If, on the other hand, conjugational stems do not have internal morphological structure, then all three Romance conjugation classes should be generalized to novel roots in a way that is sensitive to phonological similarity. Finally, a dual-mechanism account of stem-formation predicts unlimited productivity for 1st conjugation stems and similarity-based generalizations for 2nd and 3rd conjugation stems. Say and Clahsen (2002) tested these predictions with an elicited production task in Italian, by constructing nonce forms in the first person present tense indicative (which does not display a theme vowel) and asking participants to complete sentences with a participle form, thereby choosing a theme vowel and assigning each verb to a conjugation class. The nonce forms rhymed almost exclusively with either regular or irregular forms of verbs of the 2nd or the 3rd conjugations. In addition, a “No Similarity” condition contained nonce forms that did not rhyme with any Italian verbs, but were phonologically legal. The results showed that both regular and irregular participles belonging to the 2nd and the 3rd conjugations were frequently produced in their expected high similarity contexts (~20% to 40% of responses), but almost never outside their corresponding conditions. In contrast, 1st conjugation participles were the preferred response in every condition, but especially so in the condition where nonce forms did not resemble Italian verbs (which elicited 90% of 1st conjugation responses). Say and Clahsen (2002) argued that the results reflected the operation of two distinct mechanisms: an analogical mechanism, which generalizes both regular and irregular verbs of the non-default classes and is based on associations between stored stems of the 2nd and 3rd conjugations and their roots; and a default stem formation operation for the generalization of the 1st conjugation. Following Say and Clahsen’s (2002) study, a number of computational implementations have been proposed which were claimed to simulate the human data in a single-system model that makes use of phonological similarity to generalize, without invoking the different representational substrates posited by a dual-
56
Harald Clahsen and João Veríssimo
mechanism account. One computational single-system model is Eddington’s (2002) implementation of the Analogical Modelling of Language (AML) framework (Skousen, 1992; Skousen, Lonsdale and Parkinson 2002). The second one is Colombo, Stoianov, Pasini, and Zorzi’s (2006) connectionist network, and the third one Albright’s (2002) Minimal Generalization Learner (MGL). In the following, we will discuss these models of conjugation class assignment for Romance languages and point out their limitations.
3.3.1 A memory-based learning model of conjugation class assignment Eddington’s (2002) AML model generalizes conjugation classes on the basis of a database of classified lexical forms. The probability of a particular output is obtained by first searching lexical memory for the entries that are most similar with a given input form. The members of the database are then grouped into subcontexts which contain the forms that share phonological constituents with the input. Those entries that share more features with the form to be produced and that display cohesive change patterns have a larger influence on the probability that a particular output is chosen. These properties make the AML model highly sensitive to frequency and similarity effects. Eddington (2002) argued that an AML model trained on an Italian database and presented with the same nonce forms that were employed in Say and Clahsen’s (2002) study accurately simulates participants’ responses. Eddington’s (2002) model did indeed simulate some of the experimental effects reported in Say and Clahsen’s (2002) experiment. Firstly, the model showed frequency and similarity effects for the 2nd and the 3rd conjugations, as did the participants. Secondly, and more importantly, the model also displayed a clear preference for generalizing the 1st conjugation to verbs in the “No Similarity” condition, approximating the proportion of 1st conjugation responses given by human participants (82% in the model vs. 90% for the human data). This preference for the 1st conjugation in Eddington’s (2002) model arises naturally from the higher number of types and tokens belonging to the 1st conjugation, as well as from its more heterogeneous phonological distribution. In other words, the model appeared to behave as if it was performing an operation over a context-free variable, without having the required symbol-manipulation machinery for executing such an operation. A closer inspection of Eddington’s (2002) results, however, reveals that the output of this simulation model differed from participant’s responses in several important respects, indicating that an analogical (AML-type) mechanism cannot fully account for the generalization properties of conjugational stems. First, the AML model displayed much larger frequency effects than those presented
Dual Morphology in the mental lexicon
57
in Say and Clahsen’s (2002) study. Whereas participants’ responses revealed comparatively small differences between nonce verbs rhyming with many or with few real verbs (45% vs. 31% for forms similar to the 2nd conj.; 41% vs. 30% for 3rd conj.), the AML simulation showed a very steep increase in the proportions of 2nd and 3rd responses, as the number and frequency of rhyming verbs is increased (54% vs. 20% for 2nd conj.; 44% vs. 7% for 3rd conj.). Second, the human participants produced a majority of 1st conjugation responses (or similar percentages of 1st conj. and non-default responses), even in cases of very high similarity to the competing conjugations. In contrast, the AML model dispreferred 1st conjugation responses in conditions of great similarity to existing non-default or irregular forms. More specifically, for nonce forms rhyming with 2nd conjugation and 3rd conjugation irregulars, the model produced only 38% and 29% of 1st conj. responses, respectively, whereas the human participants clearly preferred 1st conjugation responses, at rates of 59% and 71%, for the same novel verbs. Likewise, for novel verbs that rhymed with many high-frequency 2nd conjugation regular participles, the AML model produced a majority of 2nd, rather than 1st conjugation participles (54% vs. 27%), whereas the human data shows comparable percentages for both types of responses (45% vs. 43%). We conclude that Eddington’s (2002) memory-based model is not capable of generalizing in the same way as human participants in that the model (i) is oversensitive to phonological similarity and lexical frequency, and (ii) underestimates responses belonging to the default class.
3.3.2 A connectionist model of conjugation class assignment Colombo et al. (2006) proposed a connectionist network of massively interconnected units for conjugation class assignment in Romance languages. In such networks, connection weights are trained to activate phonological outputs from a vector of activations representing phonological inputs (possibly mediated by a layer of hidden units, allowing the network to learn non-linear mappings; MacWhinney and Leinbach 1991). Colombo et al. (2006) trained connectionist networks on the task of producing Italian participles when given four different inflectional variants as inputs. Test forms were novel verbs, designed to be similar to existing 1st, 2nd, and 3rd conjugation verbs. Colombo et al. argued that the responses produced by the connectionist models were comparable to those from human participants tested on the same verbs. However, like Eddington’s (2002) model, Colombo et al.’s (2006) connectionist network is unable to accurately simulate the generalization patterns of conjugational stems in Italian. It is true that like the human participants,
58
Harald Clahsen and João Veríssimo
Colombo et al.’s (2006) model outputs a majority of 1st conjugation responses (90%) for novel forms that rhyme with 1st conjugation verbs. However, in the case of novel inputs presented in the first singular present tense (which is precisely the form that does not contain a theme vowel in the input representation), proportions of 1st conjugation participles produced in response to forms similar to the other classes were very small (25% for forms similar to 2nd conj.; 10% for 3rd conj.), an effect that indicates an oversensitivity of the network to phonological similarity.2 More importantly, Colombo et al.’s (2006) model produced a large number of errors and of “unclassifiable” responses when presented with novel verbs. For ambiguous (1sg pres.) inputs, the percentage of illegal strings or other unexpected responses constituted the majority of the outputs given to 2nd and 3rd conjugation pseudoverbs (75% and 59%). This very poor generalization performance is likely to be a consequence of the high proportion of 1st conjugation forms in the training set. That is, although the higher type frequency of 1st conjugation verbs enabled the network to output an overall majority of 1st conjugation responses, the results suggest that this could only be achieved at the expense of an unacceptable level of performance for the other classes. Despite their differences, three characteristics are common to both Eddington’s (2002) AML model and Colombo et al.’s (2006) connectionist model. Firstly, in both models, generalization is strictly analogical and does not make use of variable-based operations over grammatical categories. Secondly, the models employ a single generalization mechanism, that is, all three conjugation classes (and indeed, any inflectional form) are handled through the same operations. Thirdly, in these approaches, stems do not constitute separate morphological types. Instead, the representations that are manipulated and produced by the computational algorithms are full phonological forms and are not associated with any kind of internal morphological structure. We have seen that these kinds of model, which eschew rules, variables and morphological structure, provide only a limited and partial account of stem generalization processes in human participants.
3.3.3 A similarity-based simulation model of conjugation class assignment In contrast to the previously discussed approaches, Albright (2002) proposed a computational model of conjugation assignment for Italian, the Minimal Generalization Learner (MGL), which generalizes on the basis of phonological similarity, 2 These percentages were recalculated from the results presented in Colombo et al.’s (2006) Table 7, so that they would include only responses given to the ambiguous 1sg pres. forms, which provide no obvious information about conjugation membership.
Dual Morphology in the mental lexicon
59
but does so by extracting rules that incorporate both variables and restricted phonological contexts. That is, whilst this model employs the same similaritydriven mechanism to generalize all conjugation classes, it can also learn and employ context-free generalizations. More specifically, during a learning phase, the MGL algorithm takes pairs of morphologically related words as input and, for each pair, posits a rule of the form “change A into B, in the presence of C and D”, where C and D are terms describing phonological environments (which may or may not include a variable). By comparing different rules with the same structural change, the model then posits “generalized” rules that can encompass increasingly general phonological contexts; in the case of morphological transformations that display some degree of phonological heterogeneity, even a context-free rule can be discovered. As a result of this discovery procedure, the extracted grammar contains many redundant rules, which cover the phonological space in both general and specific ways. Competition between rules is resolved by evaluating their reliability, a measure of a rule’s predictive “success” (i.e., the number of forms a rule can derive correctly divided by the number of forms that the rule can be applied to). In an acceptability judgement study with Italian native speakers, Albright (2002) found that the acceptability ratings of novel infinitives belonging to all three verbal conjugations correlated with the corresponding MGL rule reliabilities for these classes; see also Bonami, Boyé, Giraudo and Voga (2008) for French. That is, besides showing effects of similarity in the generalization of 2nd and 3rd conjugations (as had also been the case in Say and Clahsen’s (2002) elicited production experiment), Albright’s (2002) study also revealed similarity effects for the 1st conjugation. This effect was taken as support for the MGL model, and in particular, as evidence against the generalization of default stems by a context-free operation. We have recently put the MGL account to the test, by conducting an elicited production experiment in Portuguese (Veríssimo and Clahsen 2014). In the following, we will report some results from this study.
3.4 Stem-generalization processes in Portuguese verbs The human data in Veríssimo and Clahsen’s (2014) study comes from an elicited production experiment in which participants were presented with ambiguous 1st person singular present tense indicative forms of novel verbs and had to produce infinitives (by necessarily assigning the root to one of the three verbal conjugations). Following Albright (2002), and unlike previous generalization studies (which adopted intuitive notions of similarity), novel forms were constructed on the basis of the reliability scores of an MGL model applied to the Portuguese verb lexicon. A correlational design was employed, such that the
60
Harald Clahsen and João Veríssimo
novel items spanned a wide range of reliability values for the three conjugations. Furthermore, and unlike Albright (2002), effects of similarity to each class were always tested while statistically controlling for possible blocking effects due to similarity to the competing conjugations. The results of the elicited production experiment revealed a clear contrast between the generalization patterns of the 1st conjugation, on the one hand, and of the 2nd and 3rd conjugations, on the other. Whilst proportions of 2nd and 3rd conjugation responses were solely determined by their corresponding reliability values, the MGL reliabilities for the 1st conjugation did not predict 1st conjugation responses. Instead, proportions of 1st conjugation responses were predicted by the degree of phonological similarity to the 2nd and 3rd conjugations (the higher the reliabilities for these classes, the lower the proportions of 1st conjugation infinitives). These findings are in line with what Say and Clahsen (2002) found for Italian indicating that the 1st conjugation is a default class that is generalized by a context-free operation. In a second study, Veríssimo and Clahsen (2014) directly contrasted two distinct computational implementations of the human data from the elicited production experiment: Albright’s MGL model and an alternative simulation model that we devised, labelled DGL (Default Generalization Learner), a minimally different computational implementation in which the reliability evaluation metric makes a principled distinction between context-free and similarity-based generalizations. As mentioned above, the MGL algorithm learns a number of specific and general rules, including (for Portuguese) a stem-formation 1st conjugation context-free rule. The DGL model employed the same algorithm for the discovery of phonological environments and rule learning, but differed crucially in the ranking of rules, in that the DGL model was prewired to assign the maximal reliability score to the first context-free rule that is derived during the rule-learning process. This modification represents an implementation of the notion of a default, which by definition, can be applied to an unbounded number of cases, so that in terms of a reliability metric of rule evaluation, a default should have maximal reliability or confidence. Consequently, 1st conjugation stems are never derived by any specific rules in the DGL model; instead, only the context-free rule applies. By contrast, 2nd and 3rd conjugation stems generalize by rules that apply to phonologically restricted contexts. In this way, the DGL model is an implementation of a dual-mechanism proposal, predicting that context-free and context-sensitive generalizations align with conjugation membership. In order to compare these two computational models of conjugation class assignment in Portuguese, MGL vs. DGL, the expected proportions of 1st, 2nd, and 3rd conjugation responses were derived from both models, for all the items tested in the elicited production experiment, and these predictions were directly
Dual Morphology in the mental lexicon
61
compared to the participants’ responses. Table 1 displays the models’ predictions, as well as percentages of each type of participant response (1st, 2nd or 3rd conj. infinitive) for three different types of novel items defined by whether they were more similar to (i.e., had higher reliability for) the 1st, 2nd or 3rd conjugations. Table 1: Percentages of infinitives belonging to each conjugation (1st, 2nd and 3rd conj.), for human participants, and the MGL and DGL models, for each type of novel item (more similar to 1st, 2nd or 3rd conjugations). Source Type of novel item
Response
Participants
DGL
MGL
Similar to 1st conj.
1st conj. 2nd conj. 3rd conj. 1st conj. 2nd conj. 3rd conj. 1st conj. 2nd conj. 3rd conj.
80 7 13 51 44 5 65 3 31
79 6 15 51 43 6 56 3 41
74 7 19 35 57 8 42 4 54
Similar to 2nd conj.
Similar to 3rd conj.
As illustrated in Table 1, the MGL model consistently underestimated 1st conjugation responses and overestimated proportions of 2nd and 3rd conjugation responses, especially for novel verbs that were highly similar to these classes, whereas the DGL model made predictions that were statistically indistinguishable from human responses. For example, for novel verbs that were more similar to the 2nd conjugation, participants still produced a majority of 1st conjugation responses, and this difference was almost perfectly approximated by the DGL model, whereas the MGL model predicted the inverse pattern: a majority of 2nd, rather than 1st conjugation responses (see Table 1). In sum, by designing an experiment with items created from the predictions of a computational model, and by comparing two specific implementations that differed in minimal respects, we were able to make evident the shortcomings of similarity-driven models, as well as demonstrate the need to distinguish between two mechanisms of stem generalization.
4 Conclusions A number of psycholinguistic studies on different languages suggest that two mechanisms are involved in the representation and processing of morphology. On the one hand, irregular forms as well as allomorphic stems and combina-
62
Harald Clahsen and João Veríssimo
torial stems of restricted classes are generalized by a similarity-driven mechanism that is sensitive to phonological overlap. On the other hand, regular forms and combinatorial stems of the default classes are generalized by variable-based operations, productively extending to all members of a grammatical category. At a more general level, the results we have discussed in this chapter indicate that purely similarity-driven algorithms are insufficient to account for the specific dissociations that are repeatedly found in studies of morphological acquisition, generalization and processing. Instead, we have shown that morphological structure plays a crucial role in language processing, and argued for approaches that postulate a dual architecture of the language faculty, incorporating both symbolic operations and lexical associations (e.g., Clahsen 1999; Chomsky 1980; Pinker and Ullman 2002). Furthermore, by adopting notions and concepts from theoretical morphology, such as the distinction between combinatorial stems and stem entries, we can achieve some understanding of findings obtained in psycholinguistic experiments.
References Albright, Adam. 2002. Islands of reliability for regular morphology: Evidence from Italian. Language 78. 684–709. Aronoff, Mark. 1976. Word formation in generative grammar. Cambridge, MA: The MIT Press. Aronoff, Mark. 1994. Morphology by itself: Stems and inflectional classes. Cambridge, MA: The MIT Press. Baayen, R. Harald, Petar Milin, Dusica Filipović Đurđević, Peter Hendrix & Marco Marelli. 2011. An amorphous model for morphological processing in visual comprehension based on naïve discriminative learning. Psychological Review 118. 438–482. Bonami, Olivier, Gilles Boyé, Hélène Giraudo & Madeleine Voga. 2008. Quels verbs sont réguliers en français? In Jacques Durand, Benoît Habert & Bernard. Laks (eds.), Congrès Mondial de Linguistique Française – CMLF’08, 1511–1523. Paris: Institut de Linguistique Française. Bybee, Joan. L. & Hopper, Paul J. (eds.). 2001. Frequency and the emergence of linguistic structure. Amsterdam, Philadelphia, PA: John Bemjamins. Chomsky, Noam. 1980. Rules and representations. New York, NY: Columbia University Press. Clahsen, Harald. 1999. Lexical entries and rules of language: A multidisciplinary study of German inflection. Behavioral and Brain Sciences 22. 991–1013. Clahsen, Harald. 2006. Dual-mechanism morphology. In Keith Brown (ed.), Encyclopedia of language and linguistics (Vol. 4), 1–5. Oxford: Elsevier. Clahsen, Harald, Martina Penke & Teresa Parodi. 1993. Functional categories in early child German. Language Acquisition 3. 395–429. Colombo, Lucia, Ivilin Stoianov, Margeritha Pasini & Marco Zorzi. 2006. The role of phonology in the inflection of Italian verbs: A connectionist investigation. The Mental Lexicon 1. 147– 181.
Dual Morphology in the mental lexicon
63
De Boer, Minne Gerben. 1981. The inflection of the Italian verb: A generative account. Journal of Italian Linguistics 2. 55–93. Eddington, David. 2002. Dissociation in Italian conjugations: A single-route account. Brain and Language 81. 291–302. Feldman, Laurie Beth & Katherine Weber. 2012. Morphological processing: A comparison of graded and categorical accounts. In James S. Adelman (ed.), Visual word recognition. (Vol. 2: Meaning and context, individuals and development), 3–23. Hove: Psychology Press. Gonnerman, Laura M., Mark S. Seidenberg & Elaine S. Andersen. 2007. Graded semantic and phonological similarity effects in priming: Evidence for a distributed connectionist approach to morphology. Journal of Experimental Psychology: General 136. 323–345. Indefrey, Peter & Levelt, Willem J. M. 2004. The spatial and temporal signature of word production components. Cognition 92. 101–144. Karaminis, Themis N. & Michael S. C. Thomas. 2010. A cross-linguistic model of the acquisition of inflectional morphology in English and Modern Greek. In Stellan Ohlsson & Richard Catrambone (eds.), Proceedings of 32nd Annual Conference of the Cognitive Science Society. Portland, Oregon. Url: http://mindmodeling.org/cogsci2010/papers/0267/index.html Kiparsky, Paul. 1982. From cyclic phonology to lexical phonology. In Harry van der Hulst & Norval Smith (Eds.), The structure of phonological representations (Part 1), 131–175. Dordrecht: Foris. Linares, Rafael Enrique, Antoni Rodriguez-Fornells & Harald Clahsen. 2006. Stem allomorphy in the Spanish mental lexicon: Evidence from behavioral and ERP experiments. Brain and Language 97. 110–120. Longworth, Catherine & Marslen-Wilson, William. 2011. Language comprehension: A neurocognitive perspective. In Jackie Guendouzi, Filip Loncke & Mandy J. Williams (eds.), The handbook of psycholinguistic and cognitive processes: Perspectives in communication disorders, 227–246. New York: Taylor & Francis. MacWhinney, Brian. & Jared Leinbach. 1991. Implementations are not conceptualizations: Revising the verb learning model. Cognition 40. 121–157. Marcus, Gary. F., Ursula Brinkmann, Harald Clahsen, Richard Wiese & Steven Pinker. 1995. German inflection: The exception that proves the rule. Cognitive Psychology 29. 189–256. McClelland, James L. & Karalyn Patterson. 2002. Rules or connections in past-tense inflections: What does the evidence rule out? Trends in Cognitive Sciences 6. 465–472. Pinker, Steven. 1999. Words and rules: The ingredients of language. New York, NY: Basic Books. Pinker, Steven & Michael T. Ullman. 2002. The past and future of the past tense. Trends in Cognitive Sciences 6. 456–463. Plaut, David C. 2011. Connectionist perspectives on lexical representation. In M. Gareth Gaskell & Pienie Zwitserlood (eds.), Lexical representation: A multidisciplinary approach, 149– 170. Berlin: Mouton de Gruyter. Poeppel, David & Kenneth Wexler. 1993. The full competence hypothesis of clause structure in early German. Language 69. 1–33. Say, Tessa & Harald Clahsen. 2002. Words, rules and stems in the Italian mental lexicon. In Sieb Nooteboom, Fred Weerman & Frank Wijnen (eds.), Storage and computation in the language faculty, 93–129. Dordrecht: Kluwer Academic Press. Skousen, Royal. 1992. Analogy and structure. Dordrecht: Kluwer Academic Press.
64
Harald Clahsen and João Veríssimo
Skousen, Royal, Deryle Lonsdale & Dilworth B. Parkinson. 2002. Analogical modeling: An exemplar-based approach to language. Amsterdam: John Benjamins. Stavrakaki, Stavroula & Harald Clahsen. 2009. The perfective past tense in Greek child language. Journal of Child Language 36. 113–142. Varlokosta, Spyridoula, Anastasia Arhonti, Loretta Thomaidis & Victoria Joffe. 2008. Past tense formation in Williams syndrome: Evidence from Greek. In Anna Gavarro & M. João Freitas (eds.), Proceedings of GALA 2007, 483–491. Cambridge: Cambridge Scholars Publishing. Veríssimo, João & Harald Clahsen. 2009. Morphological priming by itself: A study of Portuguese conjugations. Cognition 112. 187–194. Veríssimo, João & Harald Clahsen. 2014. Variables and similarity in linguistic generalization: Evidence from inflectional classes in Portuguese. Journal of Memory and Language 76. 61–79. Villalva, Alina. 2000. Estruturas morfológicas: Unidades e hierarquias nas palavras do português [Morphological structures: Units and hierarchies in Portuguese words]. Lisboa: Fundação Calouste Gulbenkian, Fundação para a Ciência e a Tecnologia. Villalva, Alina. 2003. Estrutura morfológica básica [Basic morphological structure]. In Maria H. Mira Mateus, Ana M. Brito, Inês Duarte & Isabel Hub Faria (eds.), Gramática da língua portuguesa, 6th edn., 917–938. Lisboa: Editorial Caminho. Westermann, Gert & Nicolas Ruh. 2012. A neuroconstructivist model of past tense development and processing. Psychological Review 119. 649–667. Wunderlich, Dieter. 1996. Minimalist morphology: the role of paradigms. In Geert Booij & Jaap van Marle (eds.), Yearbook of Morphology 1995, 93–114. Dordrecht: Kluwer Academic Press.
Bernadette Plunkett
3 Microparametric variation and the acquisition of quantitative de in French Abstract: This chapter examines the distribution of quantitative de in French, focusing on the variable patterns found in three Belgians from a corpus of spoken French introduced in Plunkett (2002). The use of quantitative de is variable in all varieties of French but I show that the distribution found in these speakers does not follow the prescriptive pattern of variability. The present chapter explains the critical features of two approaches to the acquisition of variable syntax found in the literature. The ultimate goal of this chapter being to determine which is ultimately best able to explain the patterns of de use that are found. I propose a syntactic analysis for the standard pattern of variation surrounding quantitative de. I argue that the two main contexts in which this de is found both involve a triggering element (negation and a quantifier respectively) that licenses a polarity-like null determiner in the complement to quantitative de. The syntactic analysis provided explains why Francophone children might struggle with quantitative de, because uncovering the factors accounting for the pattern requires a sophisticated knowledge of French syntax. However, even though her knowledge of syntax is sophisticated, the Belgian child examined cannot follow the normative pattern, since the input she receives does not conform to it. I lay out in some detail the patterns found in the child’s input and show that her pattern is not quite the same. Following quantitative de both adults and child realise the heads of DPs variably and the patterns are sensitive to the same types of factor but these seem to be weighted differently. The detailed discussion leads to the conclusion that only an approach to the acquisition of variation which has factor weight built into it can hope to explain these types of syntactic variation. For this reason an approach employing microparametric cues alone (such as Westergaard 2009) will ultimately be insufficient to explain the acquisition of all types of variable syntax and I conclude that one like that of Adger (2006) invoking Combinatorial Variability holds more promise.
Bernadette Plunkett, University of York
66
Bernadette Plunkett
1 Introduction The grammatical analysis of French ‘some’ des and the related du, de la, de l’ is made difficult by the fact that in certain circumstances they must be replaced by a bare de. From the perspective of acquisition, learning when to use bare de as opposed to one of the forms of ‘some’ requires a considerable amount of syntactic and semantic sophistication. However, for some francophone children the acquisition task is made even more complex by the fact that the input they receive contains a further layer of variation; these children are being exposed to a variety in which, where the standard language requires bare de, adult speakers appear to use one of the above-mentioned forms of ‘some’ in variation with it. In recent years syntacticians have become interested in explaining microparametric variation and acquisitionists have begun to explore how language development can be explained given variable input. In this chapter, I will do two things. First, I will investigate how the expression of ‘some’ in French can be captured syntactically. Second, I will consider how well two differing approaches to the acquisition of variable input can cope with explaining the acquisition of ‘some’ in a variety of (Belgian) French in which extreme variation in its exponence is manifest. I will conclude that an approach such as the one invoking Combinatorial Variability (Adger 2006) is best suited to explaining the patterns found. I begin in Section Two by delineating two broad positions concerning how syntactic variation can be acquired. In section Three, I lay out the accepted facts concerning de/des variation in ‘standard’ French and propose a syntactic account of these. Section Four examines a variety of Belgian French in which the usage is far more variable than in the standard language. In Section Five I return to discussion of the theoretical approaches to the acquisition of variation and before concluding, evaluate their ability to explain the patterns found in the non-standard variety discussed.
2 Acquisition models and syntactic variation Aside from work on creole genesis, it is only in the last twenty years or so that generative syntacticians have begun to pay attention to syntactic variation. The first forays into this area arose from work investigating how language change comes about. Perhaps the most influential researcher in that area is Kroch. He proposes that while diachronic change is taking place and structures are in variation in a language, speakers are employing alternative grammars (e.g., Kroch 1989). What changes over the time period of syntactic change is the likeli-
Microparametric variation and the acquisition of quantitative de in French
67
hood with which one grammar will be chosen over the other. This Grammar Competition (GC) view is perfectly consistent with parametric approaches to language acquisition. Within each grammar that the child acquires, parameters can be set in the normal way. As work in GC developed however, a consensus emerged that the competition is not between two entire speaker grammars but between (possibly multiple) small bits of grammatical structure. For example, consistent with early minimalist approaches to syntax (e.g., Chomsky 1995) in which within a given grammar, functional heads should be associated with a particular (invariant) feature value (strong or weak), competition can arise between two grammars, one containing Complementiser A, bearing a strong V-feature and one with Complementiser B, bearing a weak V-feature. This can then explain the apparent presence in a single speaker’s grammar of a characteristic such as ‘optional’ V-movement to C. The claim of GC is that speakers are actually employing two grammars, one with movement and one without. However, parametric approaches to language acquisition have been increasingly challenged in the last ten or fifteen years. One such challenge comes from researchers such as Westergaard (see e.g., Westergaard 2009) who point out that not all languages choose a single setting for parameters such as V2. Others, such as the variety of Norwegian that she discusses in detail, apply V2 (or not) clause type by clause type. Lightfoot & Westergaard (2010) explain that a child seeking to set the V2 (macro-)parameter looks for evidence in the input of the ministructure [C V . . . . However, in order to acquire the target V2 properties of her language, a child acquiring Tromsø Norwegian must look in the input not for such general cues as that but for evidence of [Pol V . . . , or [Int V . . . , or [Decl V . . . , where Pol(arity), Int(errogative), Decl(arative) can be thought of as different types of Force (or C) head, along the lines of Rizzi (1997). Westergaard (2009) thus provides a cogent argument for a Micro-cues approach to syntactic acquisition. On that approach, the child is looking for cues within small domains, usually with distinct illocutionary force. The child does not evaluate probabilities over the data she hears globally but within each type of clause (or other distinct domain). As Westergaard details, within a given domain, although the domain itself may make up a small percentage of the overall language input, a cue will usually be very robustly represented within it. Ignoring some complications (mentioned in Plunkett 2011) we can deduce that as long as a child is looking within the right kinds of domain, setting micro-parameters may actually be easier than setting macro-parameters.1 1 It is implicit in Westergaard (2009) that all children default to the inspection of microdomains, rather than these being inspected only if no clear generalization can be reached from macro-domains. This could be seen as in general keeping with the often-expressed view that children are conservative learners.
68
Bernadette Plunkett
The GC and Micro-Cue models of how syntactic variation can arise are by no means the only available accounts of variable syntactic acquisition (see e.g. Yang 2004) but as we will see in section Five, they work very differently and I will ultimately argue that while insights from both must be retained for full coverage, only an implementation of the former can explain the pattern of ‘some’ variation in French, to which we now turn.
3 De/des variation in standard French In contemporary standard French there are two broad types of context in which a bare de ‘of’ (hereafter DE ) is required in place of either a pure indefinite article une/une, des or what appears to be a combination of the preposition de ‘of’ and a definite article du, de la, de l’, (all glossed hereafter as SOME ). The first context requiring DE is negative sentences and the second is after certain adverbs of quantification such as beaucoup ‘many’. These two contexts will be described in turn, after which I will outline a syntactic analysis of the pattern.2
3.1 De/des variation in negative clauses As seen in (1) and (2), plural indefinites are introduced by SOME in affirmative clauses but by DE in the corresponding negatives sentences. (1)
Il y a des/*de he there have SOME /DE ‘There are (some) bananas’
bananes. bananas
(2)
Il n’ y a pas *des/de bananes. he neg there have not SOME /DE bananas ‘There aren’t any bananas’/ ‘There are no bananas’
The same variation occurs when the nominal concerned is singular, un/une ‘a’ is replaced by DE. 2 Traditionally, DE was also said to be required in a third context, preceding prenominal adjectives. However, Grevisse (1980) already noted that in normal then-current usage DE was never used when the following nominal was singular (even in writing) and used only in writing and ‘careful’ speech when the nominal was plural. For the sake of brevity this context will not be examined in detail in this chapter.
Microparametric variation and the acquisition of quantitative de in French
69
However, in negative copular clauses, the variation disappears and only SOME is licit, as shown in (3). (3)
Ce ne sont pas des/*de it neg are not SOME /DE ‘These aren’t bananas’
bananes. bananas
In strict prescriptive terms there is supposed to be a distinction between what Grevisse calls absolute and relative negation (1980: §667.3). However, he recognizes that the distinctions are subtle and rarely observed, with DE usually being preferred.3
3.2 De/des variation in amount contexts In positive clauses, DE is said to be required following a small number of quantificational adverbs, as in (4). (4)
Il y a beaucoup he there has many ‘There are many bananas’
*des/de SOME /DE
bananes. bananas
The adverbs concerned include un peu ‘a bit’, assez ‘enough’, plein ‘full’ and trop ‘too much’. However, other quantificational adverbs such as bien ‘many’ and encore ‘more’, do not allow DE, as seen in (5a), and other quantificational elements like plusieurs ‘several’ and certains ‘certain’ are used more like determiners (5b). (5)
a.
Il y a bien/encore des/*de SOME /DE he there have well again ‘There are lots of /more bananas’
bananes. bananas
b.
Il y a certains/plusieurs bananes He there has certain several bananas ‘There are certain/several bananas that . . .’
qui . . . that . . .
3 Grevisse (1980) provides the following examples to illustrate the absolute and relative respectively: (i) Il n’ a d’ argent que pour ses plaisirs he neg has DE money that for his pleasures ‘He has no money, except when it comes to indulging in his pleasures’ (ii) Il n’ a de l’ argent que pour ces plaisirs he neg has SOME money that for his pleasures ‘He has money only to indulge in his pleasures’
70
Bernadette Plunkett
The variation exhibited between (4) and (5) disappears in partitive contexts. Imagine a context for (6) in which a large amount of fruit was laid out on a table, on offer to the public, with more bananas than any other fruit. The speaker, referring to the bananas that are on the table, instructs their hearer to take a proportion of them. (6) Prenez encore / beaucoup / plusieurs des bananes, j’ en Take-2 pl again / many / several of the bananas I of them ai des masses. have SOME masses ‘Take more/lots/several of the bananas, I have masses of them’ DE is not possible in such contexts and although des in des bananes is homophonous with SOME, it is glossed here as ‘of the’ i.e. de+les to capture the partitive interpretation.4 In true partitive constructions, it is always possible to replace the definite article by a demonstrative, so des could be replaced by de ces ‘of these’ here.5
3.3 Towards a syntactic Account Kupferman (1999) argues that the bare de which appears in negative and amount contexts is one and the same and I will adopt his term ‘quantitative de’ for what I have glossed DE. As the discussion above shows, the semantic distinctions involved in where quantitative de can or should be used in place of SOME are rather fine-grained and it is easy to see why acquiring these distinctions would take some time. Indeed, pre-school aged children are known to sometime fail to use it. An adequate account of de in its partitive and quantitative uses is complex (see Ihsane 2005 for some review). Before sketching an analysis, for ease of reference, the patterns just described are summarized in Table 1. 4 Although introduced by homophonous strings, there is evidence that the structures of these two types of nominal are distinct (Milner, 1978). 5 Grevisse (1980) uses the term partitive article both to cover such cases as seen in (6) and cases which I treat as SOME, such as de l’ in cases like (i). He describes the partitive article as being used before uncountable nouns to indicate that only a part of the kind designated by the noun is being considered. I prefer to restrict the term partitive to cases such as those discussed in the main text. (i) Il nous faut de l’ eau It us needs SOME water ‘We need (some) water’
71
Microparametric variation and the acquisition of quantitative de in French
Table 1: De variants in Polarity and Amount contexts in Standard French Context type Polarity
Amount
Affirmative de le de la de l’ des un(e)
(SOME) (SOME) (SOME) (SOME) (SOME)
de
(DE)
Negative
de
(DE)
de
(DE)
Exception 1: Negative copular
Exception 2: Partitives
de le de la de l’ des un(e)
de le de la de l’ des
(of the) (of the) (of the) (of the)
de le de la de l’ des
(of the) (of the) (of the) (of the)
n/a
(SOME) (SOME) (SOME) (SOME) (SOME)
Let us assume that de is some type of functional head such as a preposition, which, as in Figure 1, may take either a standard definite DP complement or one headed by a null article.6
Figure 1: Types of DP complement to de
We may assume that a feature-checking relationship takes place between de and the head of its complement, whether the complement is definite or not. Let us assume that there are two distinct des with different features, one dequant being compatible only with the null determiner and the other depart only with definite determiners. The definiteness of a determiner is semantically interpretable, so a feature-checking relationship between P and D would require an uninterpretable feature on de, as suggested in the figure above. If this approach is taken, the variation between quantitative and partitive de is grammatically conditioned and the two realisations of D can be clearly distinguished semantically. When overt D is used the interpretation is partitive and when null D is used it is not. However, although the null article is matched with the correct de as in Figure 1, this does not suffice for the null article in the quantitative case to be licit. The null D behaves like a polarity item which requires a c-commanding
6 The discussion in Ihsane (2005) suggests that the null D in this case is indefinite.
72
Bernadette Plunkett
licensor. I will assume that licensing is accomplished via c-command by either a negative or quantitative operator, as in Figure 2.7
Figure 2: Null determiner licensing in Standard French
These two mechanisms suffice to explain the appearance of DE in place of SOME. In addition, however, we need to explain why DE is ruled out in copular clauses. Suppose that the copula is an unaccusative light verb (as frequently assumed) and which, as shown in the vP in Figure 2, has a single argument merged in its specifier which is a DP headed by ce ‘it’. Now suppose that when ce raises to check the EPP feature on T, its complement undergoes obligatory Extraposition. This is roughly the analysis proposed in Higgins (1979) for it-Extraposition structures in English. If the PP complement to ce undergoes Extraposition, even if the copular vP was originally dominated by negation, as shown in Figure 2, then wherever the PP lands it will not be in a position in which either de or null D could be c-commanded by Negation.
Figure 3: Lack of null D licensing in copular contexts 7 I will assume following Adger (2003) that as long as elements undergoing Agree are in a c-command relationship, whether the uninterpretable feature c-commands the interpretable or vice-versa is immaterial.
Microparametric variation and the acquisition of quantitative de in French
73
Although the grammatical patterns relating to quantitative de in French are not straightforward, if the account just sketched is on the right track there is no need to invoke grammar competition to explain de variation in standard French. We can now turn to the question of how closely the prescriptive pattern in Table 1 is matched in spoken French.
4 De in Spoken French An examination of several francophone corpora on the Child Language Data Exchange System (CHILDES) (MacWhinney 2000) shows that the pattern in Table 1 is robust amongst adult speakers of French. For the most part, I will focus discussion on one of those corpora, The York Corpus, of Child French (hereafter the York Corpus), which I collected some years ago.8 The clause types treated as exceptional in Table 1 are treated as distinct by all speakers examined, so we find contrasts such as (7) with different negative clause types and (8) contrasting partitive and non-partitive clauses.9 (7) a. *SAR: (il_n’) y a plus de bébé.10 there has more DE baby ‘There’s no more baby’
PARA03:492
b. *SAR: c(e) (n’) est plus des bébés lá [$SF: non plus]. not more it (not) is more SOME babies there too ‘They’re not babies there any more either.’ PARA03:497 8 The York Corpus consists of approximately 50 hours of spoken French more or less evenly divided between three sub-corpora collected in different francophone regions of the world: France, Belgium and Canada. The data was collected in fortnightly 30-minute sessions over a period of two years. The corpus contains speech from three pre-school aged children, one from each region and adults interacting with them. In each sub-corpus the majority of the adult speech comes from one or two adults who participate in primary care-giving. French was the first language of all main speakers in the corpus, though some speakers in the Canadian sub-corpus were bilingual. The total corpus contains over 90,000 utterances. For more details see Plunkett (2002). Examples from this corpus are followed by a corpus reference, LIEA, PARA or MONA (Liège, Paris and Montreal) and an utterance number. CHILDES corpora do not have fixed line numbering, so these numbers were derived from a special copy of the corpus; no line number appears against examples for non-York corpora. Items assumed to be present in the grammar but not pronounced are shown here in parentheses. 9 One exception to the contrast in (7) was found but it was in a passage read from a book and not part of the spontaneous speech of a participant. 10 The York Corpus, was transcribed using a Codes for the Human Analysis of Transcripts (Chat) format, developed within CHILDES and I retain this format here. This mode of transcription was enhanced to take account of the peculiarities of French, as described in Plunkett
74
Bernadette Plunkett
(8) a. *MAX: ça, c’ est un bout de la tête. that it is an end of the head ‘That is a piece of the head’ b. *CAT: ça, c’ est un bout de singe. that it is an end DE monkey ‘That’s a monkey piece’
MONA016:28
MONA016:29
The use of DE in non-copular negative clauses is less categorical however. There are exceptions in each sub-part of the York Corpus, for example. Alongside cases like (9a) we find cases like (9b). Such cases are numerous in the Belgian sub-corpus, and an examination of another Belgian corpus on CHILDES, the Rondal corpus reveals many similar examples (Rondal 1985a, 1985b). (9) a. *MOT: il (ne) te plaît pas parce+qu’ il est tout he you please not because he is all marron et qu’ il (n’) a pas de couleurs? brown and that he has not DE colours ‘You don’t like it because it’s all brown and isn’t coloured?’ PARA09:299 b. *MOT: on n’ a jamais mangé [$SF:de] spaghettis one neg has never eaten SOME spaghetti avec des pommes. with SOME apples ‘We’ve never eaten spaghetti with apples’
LIEA04:933
In amount contexts, the majority of francophone adults examined from CHILDES corpora follow the standard pattern closely; speakers from France and Canada consistently employ DE. However, once again, the use of DE is not categorical for the Belgians, alongside cases like (10) we again find many non-standard cases in the two Belgian corpora, as in (11). (10) *MOT: mais il y a plein de sable dans le seau. the bucket but he there has full DE sand in ‘But there’s loads of sand in the bucket’ LIEA005:48 (2002). Dialectal or other variation is shown by enclosing the non-standard items within angled brackets, followed by a code here $SF for Standard French and the words of the expected, standard form. Where an utterance was ambiguous between a partitive and a non-partitive interpretation this coding was not used. Nevertheless, when searching the corpus, I found some utterances containing des etc. which, despite being incompatible with a partitive interpretation, were not coded as non-standard. This may indicate the non-salience or this feature for the Belgian coder.
Microparametric variation and the acquisition of quantitative de in French
75
(11) a. *MOT: non parce+qu’ ils auront beaucoup [$SF: de] SOME no because they have-FUT-3PL a lot peine. hurt ‘No, because they’ll have a lot of trouble’
LIEA08:340
b. *MOT: des oeufs de pâques avec beaucoup des couleurs. some eggs of Easter with many SOME colours ‘Easter eggs with lots of colours’ RONDAL09a The amount and pattern of variation in the Belgian corpora on CHILDES suggest that the target grammar to which the relevant children were exposed is more complex than the one in standard French; in particular, in the input received, many phrases such as beaucoup des N were in principle ambiguous between a quantitative and a partitive reading.11 Before looking at how a child copes with such input, we turn to more precise description of the patterns found in the adult data of the Belgian part of the York Corpus.
4.1 Variable de in Adult Belgian French An overall picture of what adults do in the principal contexts can be found in Table 2.12 Figures in all remaining tables relate to the York Corpus. 11 I have examined another small corpus of Belgian French collected near Brussels, the CAT corpus. In that corpus, kindly provided to me by Cecile De Cat the number of relevant contexts is small but the speakers there appeared to follow the ‘standard’ pattern of variation. My Belgian sub-corpus was collected in the Liège region; I do not know the origin of the mother in the Rondal corpus, but Rondal’s corpus followed his own son, and Rondal was based at the University of Liège for many years. Whether the differences between the speakers discussed in the text and those in the Cat corpus are geographical or due to sociolinguistic factors is however, beyond the scope of this chapter. 12 This and all following tables exclude negative copular contexts since these are not variable. A number of other examples had to be excluded. Cases such as (i) where plus is used as a quantifier rather than a negative marker could not be classified as belonging to one context or the other. and those such as (ii) because bare de would equally appear in the positive counterpart. (i) *MAM: tu ne mets pas plus d’ eau que ça dans la bouteille? you neg put not more DE water that that in the bottle ‘Aren’t you putting more water than that in the bottle?’ LIEGE031:18 (ii) *MAM: mais ne change pas de sujet de conversation! but neg change not DE subject of conversation ‘Now, don’t change the subject!’
LIEGE030:531
76
Bernadette Plunkett
Table 2: Belgian Adult use of de in variable contexts Context type
DE
SOME
Total
Negative
69 (50%)
70 (50%)
139
Amount
82 (75%)
28 (25%)
110
This table contains data from three adults, who all exhibit variation. The numbers here are sufficiently large to be submitted to statistical tests. However, the overall pattern hides a number of intricacies and as the data contexts are further subdivided statistical tests are not always appropriate, due to small numbers of tokens. For this reason, I did not attempt to run statistics by speaker since the tokens are not evenly divided and the goal here is not to provide an exhaustive account of the variation.13 I aim here first, to provide an idea of how variable the input is for the Belgian child in the York Corpus and second, to furnish proportions for comparison with the child data, which we will examine shortly. Wherever a statistical test was run, rates have been compared using a Fisher’s Exact test; this is a test used to replace a Chi Square test of significance where numbers in any cell would be less than five. For comparability the same test was used throughout. The comparison between negative and amount contexts in Table 2 is significant; all statistical results are reported in the Appendix. Negative contexts are heavily weighted in favour of those marked by pas ‘not’. Only a handful include jamais ‘never’ and the remainder involved plus ‘no longer’ (literally ‘more’).14 As seen in the next table, the variability with pas appears to be more extreme than in the other negative contexts, where the rates look more like those found in amount contexts, however the number of non-pas tokens was small and the difference could not be shown to be statistically significant. Table 3: Belgian Adult de variation within negative contexts Context type
DE
SOME
Total
pas
54 (49%)
64 (51%)
118
plus/jamais
15 (71%)
6 (29%)
21
13 For example, no attempt has been made to ascertain whether the overt presence or absence of the negative element ne has an affect on the variation. 14 Negative tokens including que ‘only’ were not included in the counts since as seen in (i), these do not trigger SOME in the standard language. (i) Elle n’ achète que du/*de fromage She neg buys that SOME cheese ‘She buys only cheese’
Microparametric variation and the acquisition of quantitative de in French
77
In the case of sentences negated with pas the large majority of ‘standard’ cases are those in which the object was a singular noun, as in (12) and in fact DE was used in 88% of such cases. The large majority of non-standard’ cases, on the other hand, involved a plural concrete noun, as in (15) and only 14% of these took DE. (12) *MAM: il (n’) a pas de bateau? he (ne) has not de boat ‘Doesn’t he have a boat?’
LIEA14:709
(13) *MAM: les animaux n’ ont pas [$SF: de] cheveux. hairs the animals ne have not SOME ‘Animals don’t have hair’ LIEA10: 880 In the presence of a mass noun, as in the examples in (10), the pattern was different again, DE being used in only 38% of cases. All these differences are shown in Table 4. The differences between singular and plural, singular and mass and plural and mass are all significant. Table 4: Effect of nominal type in Belgian Adult pas sentences Nominal type
DE
SOME
Total
singular
34 (85%)
7 (15%)
41
plural
5 (14%)
30 (86%)
35
mass
15 (38%)
25 (62%)
40
It should be noted here that the difference between pas and other negative contexts seen in Table 3 cannot be attributed to the type of nouns occurring in them. Although the numbers are too small to merit statistical analysis and can only be suggestive, in non-pas negative sentences, in all three types of nominal context DE predominates, suggesting that nominal Number is not an important factor in the variation seen in those sentences. We will see in the next section that the child recorded in my Belgian subcorpus also uses DE in the two negative sentence types differently. But first let us look more carefully at what adults do in amount contexts. Examples with at least 11 different quantifiers which in the standard language trigger de were found in the Belgian data: un peu ‘a little’, beaucoup ‘many’, plein ‘full/lots’, assez ‘enough’, trop ‘too many’, combien ‘how many’, moins ‘less’, tant ‘so
78
Bernadette Plunkett
many’, pas mal ‘a fair number’, suffisamment, ‘a sufficient number’. However, all except the first three of these were used categorically with standard de.15 Table 5 shows the figures for each of the variable sub-cases. The differences in the rates with each quantifier are not statistically significant.16 Table 5: Adult Belgian de variation in amount context Nominal type
DE
SOME
Total
un peu
12 (50%)
12 (50%)
24
beaucoup
29 (71%)
12 (29%)
41
plein
5 (56%)
4 (44%)
9
Total
46 (62%)
28 (38%)
74
There is no evidence that whether the complement is a mass noun or a count noun is relevant to variation in amount contexts. 92% of un peu tokens are mass. Six of nine plein tokens are count and two of these take DE but this number is too small to be revealing. With beaucoup the standard form precedes 69% of mass nouns and 74% of count nouns. Tables 2 and 3 exclude cases where negative and quantified elements appear together. There were 11 of these in the Belgian part of the York corpus; all appeared with DE, which may suggest that the effect of the two context types is cumulative. To summarize, we have seen that speakers of the variety represented here, from Liège in Belgium, use SOME in a number of contexts where the standard language demands DE and that the rate at which the standard form is employed depends both on whether the clause is positive or negative and on the individual lexical item which is the trigger for a change to bare de in the standard language. However, in many negative cases, the variation is further affected by the semantics of the nominal complement. It should be noted at this juncture that in order to master this type of variation the child has to be capable of manipulating complex and subtle grammatical differences. We now turn to the child data.
15 The numbers for the remaining quantifiers can only be indicative of invariance. Most have fewer than 5 exemplars, though there are 10 cases with assez and 6 with combien. Ultimately experimental work would be required to verify the pattern of variability with these quantifiers. 16 Moreover, assuming that there genuinely was no variation with the other quantifiers, if they were removed from the figures in Table Three, it would no longer be possible to show that the difference between negative and amount contexts was statistically significant.
Microparametric variation and the acquisition of quantitative de in French
79
4.2 Variable de in the speech of a francophone Belgian child Let us first consider negative contexts. Overall, the child in this corpus, Leah, used DE in 59% of relevant negative clauses (41/69), not statistically distinct from the 50% rate in the adult data.17 In Table 6 the data is split across two periods, the boundary being marked by the appearance of plus contexts.18 Table 6: Leah’s de variation in negative contexts Age
pas DE
pas SOME
plus DE
plus SOME
2;8.22–2;11.18
3 (38%)
5 (62%)
0
0
3;0.05–4;3.21
32 (62%)
20 (38%)
6 (67%)
3 (33%)
Total
35 (58%)
25 (42%)
6 (67%)
3 (33%)
In the adult data we saw that number had an effect on the variation in negative contexts. Table 7 provides a comparison between Leah’s usage in this regard and the adults’. Data up till age 3 is omitted here, since negative plus was not used in that period. Table 7: Effect of plurality in negative de variation in Belgian French pas + sing N
pas + plural N
plus + sing N
plus + plural N
DE
SOME
DE
SOME
DE
SOME
DE
SOME
Leah 3;0.05–
23 (92%)
2 (8%)
8 (40%)
12 (60%)
4
0
2
1
Adults
34 (83%)
7 (17%)
5 (18%)
30 (82%)
3
0
5
3
The child rates look remarkably similar to those of the adults, except in plural pas contexts, where the difference between child and adult is significant. Let us turn to the quantitative contexts. Here we see a difference from the adult data seen in Table 5. Overall, Leah uses the standard DE in 42% of variable amount contexts, compared with 62% in the adult data. The breakdown by age is shown in Table 8.19 17 Since the child data in the Rondal corpus is transcribed phonetically only and there was a lot of phonetic variability, it was not possible to measure the variation in that child corpus. 18 Five cases containing both pas and a quantifying element are discounted; these invariably involved DE. Including these would raise the overall rate of pas DE from 58% to 62%. 19 In addition, Leah used two tokens with plein ‘full’, the third variable quantifier exhibiting variability in the adult data; one contained standard DE while the other did not.
80
Bernadette Plunkett
Table 8: Leah’s de variation in the principle quantified contexts un peu DE
un peu SOME
beaucoup DE
beaucoup SOME
Leah 2;8.22–2;11.18
3
5
2
0
Leah 3;0.05–4;3.21
3 (19%)
13 (81%)
13 (54%)
11 (46%)
Total
6 (25%)
18 (75%)
15 (58%)
11 (42%)
In the period up to three years of age, very few quantified contexts were produced so although different rates could represent linguistic development, the differences in the two age periods may not be meaningful. In the adult negative data, the most obvious factor influencing the variation in quantitative contexts appeared to be the mass-count distinction but it did not seem to be relevant in amount contexts. It is worth checking here however, whether it seems to be relevant for the child data. As in the adult data, the large majority of sentences involving un peu include a mass noun (there are only 3 exceptions). The amount contexts which allow comparison with the adult data are in Table 9. None of these contrasts is significant. However, the fact that the child rates for the two quantifiers when followed by a mass nominal differ so greatly suggests that the choice of quantifier itself is far more important than any effect of the mass-count distinction, in amount contexts. Table 9: The mass-count distinction and variable de in Belgian French quantified contexts un peu + mass
beaucoup + mass
beaucoup + plural
DE
SOME
DE
SOME
DE
SOME
Leah 2;8.22–4;3.21
5 (36%)
9 (64%)
10 (83%)
2 (17%)
10 (53%)
9 (47%)
Adults
11 (50%)
11 (50%)
15 (68%)
7 (32%)
14 (74%)
5 (26%)
Having seen that both the child studied in this corpus and the adults with whom she interacts exhibit considerable variation in the use of DE within different grammatical contexts, we turn next to an examination of how best to account theoretically for the acquisition of such a pattern.
Microparametric variation and the acquisition of quantitative de in French
81
5 Accounting for the acquisition of Belgian De variation The grammatical account of quantitative de sketched in Section 3 will not suffice to explain the Belgian usage presented above. The difficulty that arises in this grammar is that in some circumstances de+ definite article has a partitive interpretation and in others a type of indefinite interpretation. De+le/la is thus ambiguous in non-copular negative contexts and in amount contexts, exactly as it is in positive contexts. To account for the pattern, let us first suppose that grammar competition is at work. Let us assume that the speakers examined all had access to two grammars. Due to factors as yet unknown, speakers choose to use Grammar A x% of the time and Grammar B y% of the time. We might posit that Grammar A is the standard grammar as sketched in Section 3 and that Grammar B is one lacking a null polarity article. In that grammar, quantitative de is simply followed by a relevant overt article. I will now like to examine how a specific implementation of Grammar Competition using Combinatorial Variability (CV) Adger (2006) can be used to explain the data in the previous section. Before this can be done, it is necessary to explain how the CV system works.
5.1 Combinatorial Variability Adger (2006) discusses subject verb agreement and the problem associated with (apparently) non-linguistically conditioned variability. He argues that variable realisation of agreement could be explained and indeed particular rates of use predicted without the need to specify probabilities or weightings either within grammars or alongside competing grammars. Adger demonstrates how input heard by children may be compatible with two (or more) distinct featural analyses of a single target string. Assuming that licit analyses are implemented with equal frequency, where a pair of lexical items are in variation if there are two routes to one item and one route to the other, the former is predicted to be chosen twice as often as the latter. One demonstration provided by Adger relates to the verbal paradigm of Acadian French as discussed in King (2005) where for some person and number combinations two forms are in competition. In Standard French, with a regular verb like parler ‘to speak’, liaison contexts aside, all singular verb forms and the 3pl form are identical and correspond to the verb stem /parl/. Only the 1pl and
82
Bernadette Plunkett
2pl forms exhibit distinct verbal suffixes, with the forms /parlõ/ and /parle/ respectively. In Acadian French, the singular paradigm is the same as in standard, as is the 2pl form, but the remaining plural forms are not; 1pl is realized variably by either /parl/ or /parlõ/ and so is 3pl. An abbreviated version of the proposed analysis is as follows. Gender aside, Adger analyses the subject pronouns using three binary-valued features [+/– singular], [+/– participant] and if the latter has a + value [+/– author]. These features are transparent from the meaning, so we assume that the child can determine these with relative ease. However, she must then match them up with the three input verb forms, /parl/ /parle/ and /parlõ/ by trying to determine which features are instantiated on the verb in French. The assumption is that the verb forms bear uninterpretable features which cannot delete until they have first been valued under Agree with the pronouns. In order for the system to work, the child must determine which uninterpretable feature(s) each verb form carries. Adger proposes an acquisition procedure whereby the child assumes first that verb forms carry a single uninterpretable feature. The six available single-featured sets for the three relevant features are in (14). (14)
[usingular : –]
[usingular : +]
[uparticipant : –]
[uparticipant : +]
[uauthor : –]
[uauthor : +]
All singular pronouns sometimes appear with the verb form [parl], as do all nonparticipant (i.e. 3rd person) pronouns and authored (i.e. 1st person) pronouns. The bolded specifications in (14) are thus each mapped onto morphemes with the phonological representation /parl/. When a relevant pronoun is merged in an agree relationship with one of these verbal morphemes, the uninterpretable feature on the verb will be successfully checked and the derivation can proceed. However, there is no single verb form which can occur with all plural pronouns, nor is there one which can appear with all four [uparticipant : +] pronouns, or both [uauthor : –] (i.e. 2nd person) pronouns. Adger proposes an evaluation metric part of which rejects positing two distinct morphemes with the same featural specification (Reject Optionality) so the child will not posit [uauthor : –]/parl/ and [uauthor : –]/parle/. The child is then unable to match a single feature to either/parle/ or /parlõ/. The next step is to entertain morphemes with two features that might match the so far unanalyzed verb forms. This results in (15a) and (15c) both successfully matching /parlõ/ and (15b) matching /parle/. On this basis the child posits three new morphemes, two with the phonetic form [õ] and one with the form [e].
Microparametric variation and the acquisition of quantitative de in French
(15)
a.
[usingular : –, uparticipant : –]
(i.e. 3 pl)
b.
[usingular : –, uauthor : –]
(i.e. 2 pl)
c.
[usingular : –, uauthor : +]
(i.e. 1 pl)
83
The (a) and(c) sets both match /parlõ/ while the b) set matches /parle/, so three more morphemes are posited. However, importantly, (15a) and (15c) also match /parl/ by virtue of their respective [uparticipant : –] and [uauthor : +] specifications. This results in both 3rd person plural and 1st person plural being compatible with two different verb forms, which are in competition with each other. Adger (2006) argues that it is because of the competition here that the agreement paradigm is unstable and undergoing change.20 Let us examine now whether the data described in Section 4 is compatible with an analysis invoking CV.
5.2 Combinatorial Variability applied to quantitative de in Belgian French To remind the reader of the pattern we need to explain, the exponence in this variety is laid out in Table 10. To employ CV to explain this data assume, as sketched in Section 3, that the DP complement of de is headed either by le/la/les ‘the’ or by a null D ([DØ]) and that where the latter is used it must be licensed by either negation or a quantifier as described. Assume further that, in contrast, definite articles do not require any special licensing. We might have assumed in Section 3 that the licensor checked a feature on dequant rather than on [DØ]. However, it is preferable to assume that it is [DØ] which must be licensed, since if it was dequant which required a licensor it would be difficult to explain the fact that DE (i.e. de+[DØ]) is also licit (at least in written language) when a null-headed DP contains a pre-nominal modifier.21 It is reasonable then to treat expressions involving quantitative de as involving two separate licensing mechanisms, one between de and its complement and
20 He contrasts this with a situation in which distinct features/feature combinations share a phonetic form, which he argues results in stable variability. 21 Thus, even though the pre-nominal adjective triggering contexts have not been discussed in my data, I will make the latter assumption. We can then assume that when pre-nominal adjectives trigger bare de, the adjective is somehow able to license [Dø]. I leave the characterization of such a mechanism to future research.
84
Bernadette Plunkett
Table 10: De variants in Polarity and Amount contexts in Belgian French Context type
Affirmative
Negative
Polarity
de le de la de l’ des un(e)
(SOME) (SOME) (SOME) (SOME) (SOME)
de de le de la de l’ des un(e)
(DE) (SOME) (SOME) (SOME) (SOME) (SOME)
Amount
de de le de la de l’ des un(e)
(DE) (SOME) (SOME) (SOME) (SOME) (SOME)
de de le de la de l’ des un(e)
(DE) (SOME) (SOME) (SOME) (SOME) (SOME)
Exception 1: Negative copular
Exception 2: Partitives
de le de la de l’ des un(e)
de le de la de l’ des
(of the) (of the) (of the) (of the)
de le de la de l’ des
(of the) (of the) (of the) (of the)
(SOME) (SOME) (SOME) (SOME) (SOME)
n/a
another between [DØ] and a c-commanding negative/quantificational element, as previously proposed. Assuming that we capture the use of DE in the standard language as described, what differences could we posit to account for how the variation between DE and SOME is determined in the Belgian variety discussed? Let us consider again the feature-checking relationship between [DØ] and the determiner heading its complement. Ihsane (2005) argues that [DØ] heads an indefinite DP. Note however, that dequant is barred from appearing together with indefinite determiners such as un/une or des, even in Belgian French, as shown by (16). (16) a. *Tu n’ as pas d’un chapeau / de des chapeaux you neg have not DE a hat DE some hats b. Tu n’ as pas un chapeau / de chapeau(x) you neg have not a hat DE hat(s) ‘You haven’t got a hat/any hats’ Given this, we could not simply assume that Belgian contained a dequant which did not need to check against a feature on the D head of its complement. It seems necessary to assume instead that some feature other than pure indefiniteness is being checked by dequant in all varieties. Now suppose that whatever the distinguishing feature is, in standard French only [DØ] can match it but in Belgian French it can (also) be matched
Microparametric variation and the acquisition of quantitative de in French
85
by le/la/les. That is, in addition to the standard grammar A, Belgian French has a Grammar B containing three items homophonous with the definite articles le/las/les in Grammar A but which do not bear the feature [+definite]. It is not entirely clear that when used to introduce mass nouns le/la/les should be considered definite, so it is not unreasonable that Belgian French would come to attribute a different feature specifications to them than the ones they are given in the standard language. In the Belgian variety, the determiners le/la/les might simply be unspecified for definiteness and could therefore check against dequant as in Figure 4.
Figure 4: Quantitative de licensing in Belgian French
In the CV approach, because of Reject Optionality the [DØ] could not bear exactly the same features as any of the non-definite le/la/les. Assume then, that unlike [DØ] these items do not carry the uninterpretable polarity / quantity feature which requires a licensor. This featural difference is enough to ensure that [DØ] cannot appear in affirmative non-quantificational contexts, where no licensor for it can be found. Just like standard speakers, the Belgians have [DØ] in their grammar. Under CV assumptions, all other things being equal, we predict that in a triggering context, following quantitative de the choice of [DØ] or le/la/les would be at chance level. That is approximately what we have seen in the adult usage in several of the contexts examined. That is not the case in all contexts, however. It may be that in Belgian French [DØ] is being reinterpreted as carrying a singular feature, so that there is a second +sg null D in competition with a standard French null D which is compatible with singular, plural and mass nouns. Space constraints do not permit a more detailed analysis here and in any case a more precise picture of the exact variation in this and other French varieties would be needed before a full blown analysis could be achieved. However, it is reasonably clear that the system allows children exposed to this variety to attribute features in a slightly different way from the adults such that number and mass will have slightly different effects in the child grammar than in the adult. The above outline is surely enough to demonstrate that employing CV to treat the problem is capable of some success.
86
Bernadette Plunkett
Let us contrast this with what a Micro-Cues approach could say about the data.
5.3 Micro-Cues applied to quantitative de in Belgian French It is not clear what the micro-cue approach could say to explain the Belgian variability in de use. If, for example, de was used categorically in negative contexts, when the Noun was singular, then (17) might be proposed. (17) [NegP pas etc [tNeg] [VP [ tDP] [tV] de [DP Ø
NP
] ] ...
[–def] [+sg,+count]
But the usage is not categorial and when the complement of de contains a mass noun or plural count noun the rates of bare de are even lower and this cannot be accounted for by the simple absence of a relevant de +[DØ] cue. Similar problems occur with the quantified contexts. Without some way of building rates of use into the cues system, this approach would seem to have little to say on the matter. For similar reasons, Micro-Cues do not fare much better in accounting for the child patterns. If the child applies the variation in a slightly different way to that seen in the adult input (say for example by omitting the [+count] feature from the cue in (17)) this would imply that the child had not correctly identified the cue and that the cue could change generation by generation. Such a view is consistent with the approach but again unless the application of the revised cue is categorical the actual pattern will not be captured. To conclude this section, Combinatorial Variability seems better suited to explaining de/des variation in French than an approach invoking Micro-Cues.
6 Conclusion: Distinct categories of variation It seems then that we must distinguish at least two types of (stable) syntactic variation. The first type, exemplified by the micro-variation in V2 in the Tromsø variety of Norwegian is well explained by the Micro-cue approach. Micro-cues might be invoked in Standard French as well, to explain the licensing of null determiners as illustrated in Figure Two. The second type of variation, exemplified by the variety of Belgian French discussed here, appears to be explainable only by an approach which is able to capture statistical weighting. Thus, a treatment incorporating micro-cues, is not on its own able to account for the pattern of variation examined in this chapter.
Microparametric variation and the acquisition of quantitative de in French
87
To review briefly, the child acquiring standard French requires a sophisticated knowledge of syntax before they can hope to analyze the pattern of how quantitative de is used in the target grammar. However, the two major contexts in the spoken standard language have in common a situation in which a triggering element c-commands a polarity-like null determiner. The distinction between this de and others, though complex, is derivable from the interpretation. Moreover, the variable licensing of [DØ] in copular and non-copular contexts is likely to follow from other syntactic facts not specific to de. In the variety of Belgian French discussed, the pattern is more complex and within exactly the same type of context, both adults and the child examined use null and overt D in variation with each other. The amount of child data available from this variety is insufficient to determine with certainty whether the pattern of variation is truly stable but it seems clear that an approach to acquisition that cannot explain rates of usage is not sufficient to explain the pattern. An instantiation of Grammar Competition invoking Combinatorial Variability has been shown to have the right characteristics to explain the patterns of variation found and provides promise of providing a successful account.
References Adger, David. 2003. Core Syntax, Oxford: OUP. Adger, David. 2006. Combinatorial variability. Journal of Linguistics, vol. 42, (3) 503–530. Chomsky, Noam. 1995. The Minimalist Program. MIT Press. Grevisse, Maurice, & André Goose. 2011. Le Bon Usage: Grammaire, langue française, DeBoeckDuculot, Paris and Louvain-la-Neuve, 15th edition. Higgins, Roger. 1979. The Pseudocleft Construction in English, Outstanding dissertations in Linguistics, Garland Press. Hudson Kam, Carla, L. & Elissa L. Newport. 2005. Regularizing Unpredictable Variation: The Roles of Adult and Child Learners in Language Formation and Change, Language Learning and Development 1:2, 151–195. Ihsane, Tabea. 2005. On the Structure of French du/des ‘of the’ constituents. Generative Grammar in Geneva, Vol. 4:195–225. King, Ruth. 2005. Morphosyntactic variation and theory: subject verb agreement in Acadian French in L. Cornipps and K. Corrigan (eds), Syntax and Variation: reconciling the biological and the social. Amsterdam: John Benjamins, 198–229. Kroch, Anthony. 1989. Reflexes of grammar in patterns of language change. Journal of Language Variation and Change 1, 199–244. Kupferman, Lucien. 1999. Réflexions sur la partition: les groupes nominaux partitifs et la relativisation. Langue Française 122, 30–51. MacWhinney, Brian. 2000. The CHILDES Project: Tools for Analyzing Talk, (3rd edition) Hillsdale, NJ: Lawrence Erlbaum Associates.
88
Bernadette Plunkett
Lightfoot, David. 2010. Language Acquisition and Language change. Cognitive Science 1. 677– 684. Lightfoot, David & Marit Westergaard. 2007. Language acquisition and language change: interrelationships. Language and Linguistic Compass, 1:396–416. Mikkelsen, Line. 2005. Copular Clauses: Specification, Predication and Equation, Amsterdam: John Benjamins. Milner, J. C. 1978. De la syntaxe à l’interprétation, Editions de Seuil, Paris. Moro, Andrea. 1997. The Raising of Predicates: Predicative Noun Phrases and the Theory of Clause Structure, Cambridge University Press. Plunkett, Bernadette. 2002. Null Subjects in child French interrogatives: a view from the York Corpus, in Pusch, C. D. / Raible, W. (eds.): Romance Corpus Linguistics: Corpora and Spoken Language Tübingen: Narr, 441–452. Plunkett, Bernadette. 2011. Review of Westergaard ‘The Acquisition of Word Order: Micro-cues, Information Structure and Economy’, Journal of Linguistics 47, 38–45. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haegeman (ed.), Elements of grammar: Handbook of generative syntax, 281–337. Dordrecht: Kluwer. Roeper, Thomas. 1999. Universal bilingualism, Bilingualism: Language and Cognition, Volume 2:03, 169–186. Rondal, Jean. A. 1985a. Adult-child interaction and the process of language understanding. New York, Praeger. Rondal, Jean, Jean-François Bachelet & F. Perée, 1985b. Analyse du langage et des interactions verbales adulte-enfant. Bulletin d’Audiophonologie 5 (6): 507–535. Safir, Kenneth. 1982. Syntactic chains and the definiteness effect, Ph.D dissertation, MIT. Westergaard, Marit. 2009. The Acquisition of Word Order: Micro-cues, Information Structure and Economy. Amsterdam: John Benjamins. Yang, Charles. 2010. Three factors in language variation. Lingua 120. 1160–1177.
Appendix Statistics Table 2: Belgian adults The difference in DE use between the two context types, negative and amount is significant by Fisher’s Exact test; two-tailed p = 0.03. Table 4: Belgian adults The difference in DE use between contexts containing a singular versus a plural noun is highly significant by Fisher’s Exact test; two-tailed p < 0.0001. The difference in DE use between contexts containing a singular versus a mass noun is highly significant by Fisher’s Exact test; two-tailed p < 0.0001. The difference in DE use between contexts containing a mass versus a plural noun is significant by Fisher’s Exact test; two-tailed p = 0.0354.
Microparametric variation and the acquisition of quantitative de in French
89
Table 7: Belgian child vs adults In plural pas contexts, the difference between child and adult is significant; twotailed p = 0.0477. There is also a significant difference between plural pas and plural plus contexts in the adult data; p = 0.0103. The difference in DE use in the child data between pas contexts containing a singular and pas contexts containing a plural noun is highly significant; two-tailed p = 0.0003. If all negative contexts are considered, it is also highly significant p = 0.0001. Table 8: Leah’s de variation in the principle quantified contexts Within the two types (un peu vs beaucoup), the developmental differences are not statistically significant but the differences between the types are; P = 0.0468 in the latter period and 0.0246, overall.
Larisa Avram1, Cristina Ciovârnache and Anca Sevcenco
4 Semantic features and L1 transfer in the L2 learning of differential object marking: The view from Romanian and Persian Abstract: This chapter investigates the implementation of animacy and referential stability in differential object marking (DOM) systems in two L2 learning contexts: L1 Persian – L2 Romanian and L1 Romanian – L2 Persian. The results of two acceptability judgment tasks indicate that adult L2 learners have access to semantic universals. The integration of animacy and of referential stability in DOM systems is not affected by L1 transfer in either of the two learning contexts. The comparison of the two semantic features in the interlanguage of intermediate L2 learners reveals an asymmetry: the role of animacy is target-like at a stage when full integration of referential stability into the system is delayed. We argue that there is a stage when L2 DOM systems are sensitive to referential stability but the intricate relationship between types of referential stability and differential object marking are not part of the system yet. Our findings are in line with results reported in previous studies which showed that the L2 learning of DOM is delayed. But the delay is not caused by L1 transfer.
1 Introduction Two central issues in adult second language (L2) research have been access to Universal Grammar (UG) and the availability of L1 transfer (see, e.g. Clahsen and Muysken 1986; White 1989; Tsimpli and Roussou 1991; Schwartz and Sprouse 1996; Epstein, Flynn, and Martohardjono 1998, among many others). Compared to the impressive number of studies on syntactic parameters, only relatively few studies focused on the L2 learning of semantic parameters. Those showed that semantic universals are accessible to adult L2 learners (L2ers)
1 Work on this study was financed by research project PN-II-ID-PCE-2011-3-0959 for the first author Larisa Avram, Cristina Ciovârnache and Anca Sevcenco, University of Bucharest
92
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
(Ionin, Ko, and Wexler 2004; Slabakova 2006; Ko et al. 2008; Slabakova and Montrul 2009; Ko, Ionin, and Wexler 2010). Access to semantic universals, however, does not exclude the availability of cross-linguistic interference effects in certain L1/L2 pairings (Ionin, Zubizaretta, and Bautista-Maldonado 2008) or delays in the learning process. Some semantic factors may be more vulnerable than others. Guijarro-Fuentes (2012), in a study on the adult L2 learning of differential object marking (DOM) in Spanish, shows that animacy is less problematic than specificity. Ko et al. (2008) offer evidence that partitivity effects will cause errors for a longer period of time than specificity effects in L2 article choice. Typological similarity, the way in which semantic features interfere with syntax and/or pragmatics, acquisition setting, amount of input, and L2 proficiency can determine variation in the implementation of semantic universals. This is why extending the investigation to other language pairings might contribute to our understanding of the way in which cross-linguistic variation may interfere with semantic universals as well as of the (possible) vulnerability of semantic features in L2 learning. In the present study we look into animacy and specificity in the use of the differential object marker pe in L2 Romanian by adult speakers of L1 Persian. We compare these results to the L2 learning of the marker rā in L2 Persian by adult speakers of L1 Romanian. The main question which we address is to what extent differences between L1 and L2 can affect the implementation of semantic universals in the learning of DOM systems. Both the Romanian and the Persian DOM systems have been analysed as constrained by specificity. But animacy is a central DOM trigger only in Romanian. This L1/L2 pairing offers an ideal context for the bidirectional investigation of the possible availability of L1 transfer of semantic features. The results can also shed light on the issue of adult L2ers’ access to semantic universals. The remainder of the chapter is organized as follows. Section 2 presents the background on the role of semantic features in DOM systems. We focus on animacy and referential stability. In Section 3 we provide a description of the main semantic constraints on DOM in Romanian and in Persian. Starting from the comparison of the two systems, we discuss the predictions for the L2 learning of the Persian differential marker rā by L1 Romanian speakers and for the L2 learning of the Romanian marker pe by L1 Persian speakers. The present study, based on the results of two acceptability judgment tasks, is presented in Section 4. The implications of the main findings are discussed in Section 5. The conclusions are summarized in Section 6.
Semantic features and L1 transfer in the L2 learning of differential object marking
93
2 Background: semantic features in differential object marking Differential object marking (DOM) is a cover term for the correspondence between semantic features and overt case marking and/or designated syntactic positions. This semantically triggered object split is part of UG (Torrego 1998; Carnie 2005; Rodríguez-Mondoñedo 2007, 2008). What differs is the way in which the split is achieved in each language. Among the features identified as relevant to DOM systems are the semantic features of the object DP. Two of these features are animacy and definiteness and/or specificity2 (Bossong 1985, 1991, 1998; Aissen 2003; von Heusinger, Klein, and de Swart 2008). The generalization in most functional-typological studies is that a direct object which is higher on the animacy and on the definiteness/ specificity scales (1) is more likely to be differentially marked (Aissen 2003): (1)
a.
animacy scale human > animate > inanimate
b.
definiteness/specificity scale personal pronoun > proper noun > definite DP > indefinite specific DP > non-specific indefinite DP
According to Farkas (2002), the semantic parameter which explains the differences and the similarities between the various types of definite and indefinite DPs, i.e. the underlying feature of the specificity scale, is the parameter of determined reference. The various distinctions in the domain of determiner phrase semantics which are relevant to specificity (itself a ‘notoriously non-specific’ notion, according to Farkas 2002) build on the conditions on how values are assigned to variables introduced by DPs. Within this framework, determined reference is concerned with the degree to which the condition contributed by a DP restricts the choice of value for the variable introduced by that DP at the point of discourse update. Value assignment may be more or less ‘fixed’.
2 Animacy and specificity are only two of the factors which can affect DOM. In several languages the split is constrained by other factors (Bossong 1998; von Heusinger, Klein, and de Swart 2008). The list includes topicality, the aspectual value of the verb constellation, mood, affectedness, high degree of individuation (see, e.g. Naes 2004). In this study we will be focusing on animacy and specificity, given the standard view on their role in the Romanian and the Persian systems.
94
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
Following this line, Farkas and von Heusinger (2003) propose that the semantic dimension which underlies Aissen’s (2003) definiteness/specificity scale in (1) is referential stability, i.e. prominence on the scale is measured in terms of relative referential stability. On such a view, proper nouns and definite pronouns have determined reference in virtue of the equative condition which they impose on the variable which they introduce. In spite of the differences between these two types of DP, they are both unconditionally referentially stable (URS), i.e. their value remains unchanged throughout the discourse in virtue of their inherent properties. Lexically headed DPs contribute a discourse referent and a predicative condition which requires that the value of the variable be selected from the set denoted by the noun phrase. They achieve determined reference, i.e. their referential stability depends on various factors, among which the context of occurrence. Definite lexically headed DPs are conditionally referentially stable (CRS), i.e. the variable which they introduce is required to be stable, but their referential stability depends on some property of the context. Indefinites and partitives are referentially non-stable (RNS). The value assigned to the variable which they introduce can vary across discourse updates; they are not required to have determined reference. Partitives, however, behave more like definite DPs than like narrow scope indefinites. They have a hybrid nature. On a par with indefinite DPs, they do not require determined reference. But, on a par with definite DPs, the choice of referent is, in their case, restricted to a subset of the value of a discourse referent. Non-stability, then, can be restricted and non-restricted. In the present study we adopt the view that the underlying feature of the specificity scale is referential stability (Farkas and von Heusinger 2003). We assume that the way in which languages choose to partition the scale in terms of DOM trigger strength can differ from one language to another; but trigger strength itself will always observe the direction on the scale: from referentially stable to referentially non-stable DPs. Summing up, two core features of DPs involved in DOM are animacy and referential stability. In terms of learnability, it follows that the learner must identify the semantic trigger(s) of the split, i.e. the semantic features which constrain the system, and the differential marking strategy in the target language. In the present study we focus on the role of these semantic properties of DPs in L2 DOM systems. We are fully aware that in narrowing down the analysis we are investigating only some small pieces of a larger puzzle. By leaving other features aside we do not mean to sweep them under the rug. But we think that it is more realistic and therefore more promising to look into the role of smaller pieces before looking at the whole puzzle. This is why in this study we investi-
Semantic features and L1 transfer in the L2 learning of differential object marking
95
gate only the role in L2 learning of the semantic features of DPs involved in DOM systems. For the languages in the present study, especially Romanian, animacy and specificity have been extensively studied in the theoretical literature. This allows a better understanding of the role of these features than of the other factors involved in DOM, such as the aspectual properties of the verb phrase or discourse information structure. Approaching DOM from a semantic perspective does not mean we ignore the syntactic dimension of the phenomenon or the importance of the syntax-semantics or syntax-discourse information structure interfaces in any study of DOM in L2. But we believe that one cannot investigate the acquisition of all these factors together before looking at each of them in turn.
3 DOM in Persian and in Romanian 3.1 DOM in Romanian In Romanian, the differential object marker is pe, which derives diachronically from a locative preposition. Generally, only animate direct objects can be differentially marked (Farkas 1978; Farkas and von Heusinger 2003; Mardale 2007, 2008, among many others). Marking of inanimate DPs is very rare and it is always associated with an upgrading effect. Referential stability determines obligatory and optional contexts of use. DOM is obligatory with URS direct objects which are animate. This is shown in (2), where the use of pe is obligatory with animate proper names: (2)
Ion a vizitat *(pe) Maria/ Ion has visited *(PE) Maria/ ‘Ion has visited Maria/Paris.’
(*pe) PE
Paris. Paris
Pe is also obligatory with definite pronouns. In this case, the presence of an Accusative clitic is mandatory and the role of animacy is significantly weakened: (3) *(L-) am desenat *(pe) ăla him have.I drawn PE that ‘I have drawn the one over there.’
de of
acolo. there
With CRS direct objects, the use of pe is optional if they are animate and illicit if they are inanimate:
96
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
(4)
Am desenat (pe) copilul / (*pe) banca bench:DET have.I drawn PE child:DET / PE ‘I have drawn the child/the bench in the park.’
din from
parc. park
The optionality3 of pe with definite DPs, compared to its obligatory use with proper names and definite pronouns shows that definiteness of the object DP does not act as a DOM trigger in Romanian. The difference between proper names and pronouns, on the one hand, and definite common nouns, on the other, is one of referential stability: only the former are unconditionally stable. DOM is obligatory with unconditionally stable DPs. This has been considered proof of sensitivity of the system to referential stability (Farkas and von Heusinger 2003; Tigău 2011). That definiteness cannot be analysed as a DOM trigger is further supported by the fact that indefinite DPs behave like CRS objects. They are optionally marked when they are animate: (5)
(O) caut (pe) o fată. her look.I PE a girl ‘I am looking for a girl.’
The marker, in this case, induces a specific interpretation of the marked indefinite (Farkas 1978; Dobrovie-Sorin 1990). In (6a), the indefinite object DP is ambiguous between a specific and a non-specific reading. In (6b), the presence of pe cancels the non-specific reading (Dobrovie-Sorin 1994): (6)
a.
Caut look.I
o a
secretară. secretary
b.
Caut pe o secretară. look.I PE a secretary ‘I am looking for a secretary.’ (Dobrovie-Sorin 1994, her examples 48a and 61a)
According to Farkas and von Heusinger (2003), partitives are stronger DOM triggers than indefinite DPs but weaker than definite DPs. This places them in the area of optional DOM. This analysis predicts some gradability: pe marking 3 By optionality we mean that there is nothing in the syntactic structure of the DP or in its semantic make up which triggers obligatory marking. We do not mean that there is no interpretative difference between a sentence with a marked and one with an unmarked DP or that the two are freely interchangeable.
Semantic features and L1 transfer in the L2 learning of differential object marking
97
is favoured with definite DPs, dispreferred with indefinite DPs and truly optional with partitives. (7)
(Le) cunosc (pe) două them know.I PE two ‘I know two of these girls.’
dintre of
aceste these
fete. girls
Further evidence that referentiality plays a role in DOM comes from the impossibility of pe marking with bare plurals (which do not have determined reference) (8a) or with incorporated indefinite DPs (8b) (Mardale 2008; Tigău 2011): (8)
a.
Am cunoscut (*pe) lingvişti have.I met PE linguists ‘I met linguists at the conference.’
la at
b.
Caut (*pe) profesor de germană. look.I PE teacher of German ‘I am looking for a teacher of German.’
conferinţă. conference
The presence of an Accusative clitic forces the use of pe with post-verbal objects (in accordance with Kayne’s generalization, Avram 20144) in clitic doubling constructions, i.e. in the presence of the clitic, DOM is no longer optional irrespective of the referential stability of the DP (9). The presence of the clitic also weakens the animacy constraint, allowing the marking of inanimate DPs (10): (9) O caut *(pe) o studentă. her-CL look.I PE a student ‘I am looking for a student.’ (10) Am citit -o numai *(pe) una dintre cărţile recomandate. have.I read her-CL only PE one of books:DET recommended ‘I have read only one of the recommended books.’ Farkas and von Heusinger (2003) argue that clitic doubling, unlike pe marking, is sensitive to topicality; they suggest that given the differences between the two structures, pe marking and clitic doubling should be separated. From a different perspective, Hill (2013) also argues that clitic doubling and DOM are not intrinsically related in Romanian. Similarly, Avram and Coene (2009) focus on the 4 See, however, Tigău (2014) for a different point of view.
98
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
different role of the clitic and of the differential object marker. They analyse the former as a topic which acts as an anchor for new assertions, i.e. as a d-linked topic. In this case animacy is not involved. Pe marking signals high ranking, prominence of the argument in event structure. In this case animacy and affectedness are involved. In the present study, we adopt the standard view in the literature on Romanian DOM that clitic doubling and pe marking should be separated.
3.2 DOM in Persian The Persian morpheme rā5 has been traditionally analysed as a marker of definiteness (Lambton 1963; Khanlari 1973; Mahootian 1997). Direct objects which are proper nouns, definite pronouns, lexically headed definite DPs and partitives require rā (Lazard 1992), irrespective of animacy. (11) Man to Ali Tehrān barādar-e to/ ān ketāb *(rā) didam. saw I you Ali Tehran brother-EZ6 you/ that book RĀ ‘I saw you/Ali/Tehran/your brother/that book.’ With indefinite direct objects DOM is optional7. The use of rā with indefinite DPs has been invoked as an argument that the Persian system is not constrained by definiteness but by specificity (Windfuhr 1979; Karimi 1990, 1996; Cagri 2007; Ghomeshi 1997). Unmarked indefinite objects are interpreted as non-specific (12a) but the presence of rā forces a specific reading (12b): (12)
a.
Ali ketābi āvard. Ali book brought ‘Ali brought a book.’
b.
Ali ketābi *(rā) āvard. Ali book RĀ brought ‘Ali brought a book.’ (a certain book)
5 ro/o in the spoken language. 6 The ezāfe (EZ) particle links the head noun to its modifiers. 7 Traditional prescriptive grammars of Persian ban the use of rā with indefinites. According to Najafi (1992: 204) the use of rā with indefinite objects is incorrect. But he adds that the norm is often disregarded both in the spoken and in the written language.
Semantic features and L1 transfer in the L2 learning of differential object marking
99
Windfuhr (1979) focuses on the role of referentiality, noting that the occurrence of rā with indefinites in situations like the one in (13) indicates its function as referential: (13)
kasi rā person-INDEF RĀ ‘I saw someone.’
didam saw.I
The use of rā with partitives is obligatory with both animate and inanimate DPs (14): (14)
a.
Yeki az dānešğuyān one-INDEF of students ‘I saw one of the students.’
*(rā) RĀ
b.
Do tā az ketābhā-ye two classifier of books-EZ ‘I read two of the history books.’
didam. saw.I
tārix history
*(rā) RĀ
xāndam. read.I
As a rule, as seen in the examples above, animacy is irrelevant to the Persian DOM system. However, when low specificity is involved, the system is no longer fully indifferent to this semantic feature. There is a tendency to use rā with indefinite objects if they are [+ animate] (Lazard 1992). Ghomeshi (1997:140) provides the following examples in support of this claim. According to her, in (15) below ‘b sounds much better’: (15) a. ?hame-ye mo’allem-â ye shâgerd-i mo’ arefi kard-and all+EZ teacher+pl one student+indef introduce did+3pl ‘Every teacher introduced a student.’ b. hame-ye mo’allem-â ye shâgerd-i-ro mo’ arefi kard-and All+EZ teacher+pl one Student+indef+râ introduce did+3pl ‘Every teacher introduced a student.’ (examples, glosses included, from Ghomeshi 1997:140) In discussing the clitic doubling of rā-marked direct objects, Ganjavi (2007) points out that animacy also plays a role in the felicitous use of clitic doubling constructions in Persian8. She shows that proper nouns, pronouns and definite 8 There is some disagreement in the literature concerning the acceptability of clitic doubling constructions in Persian (see, for example, Ghomeshi 1997:157).
100
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
DPs, i.e. referentially stable DPs, can all be doubled (16a); but doubling is not possible when the direct object is [– animate] (16b). (16)
a.
Unhā un mard-a-rā they that man:DEF-RĀ ‘They saw-him that man.’
didaneš. saw.they
b. *borj-e ifel-o didameš. tower-EZ Eiffel-RĀ saw.I ‘I saw-it the Eiffel Tower.’ (from Ganjavi 2007: 188)
3.3 Comparing the two systems: predictions for L2 learning The DOM systems of Persian and Romanian differ both with respect to the role of animacy and with respect to trigger strength on the referential stability scale. Romanian marks only animate direct objects. Inanimate ones can be marginally marked, with an upgrading effect. In Persian, marking is generally insensitive to animacy, with the exception of indefinites, where one notices an animacy bias. Both systems are sensitive to referential stability. But in Persian DOM is obligatory whenever the referent choice of the variable introduced by the DP is restricted to a set. In Romanian, marking is also affected by type of referential stability; it is obligatory only when the choice of referent is, in fact, an instance of ‘no choice’. The comparison of trigger strength on the specificity scale in the two DOM systems is summarized in Table 1: Table 1: Referential stability in the DOM systems of Romanian and Persian Language
URS
Romanian
obligatory
Persian
CRS
obligatory
RNS
NS
optional optional
The task of the L2er includes the identification of how the two semantic features, animacy and referential stability, are implemented in the DOM system of the target language. For the L2 learning of syntactic parameters, an impressive number of studies assume both access to UG and the availability of L1 transfer. According to the Full Transfer/Full Access hypothesis (FTFAH) (Schwartz and Sprouse 1996) the L1 system initially transfers to L2 but the learning process is constrained by UG at all stages. Extending the FTFAH to semantic parameters in
Semantic features and L1 transfer in the L2 learning of differential object marking
101
L2 learning, one can state that there is access to semantic universals at all stages. At the same time, the way in which semantic features are assembled in L1 may serve as part of the initial state in L2 learning. The majority of previous studies on semantic parameters argue that there is access to semantic universals in adult L2 learning. But their findings differ with respect to the availability of L1 transfer, which is argued to depend on various factors, among which the L1/L2 pairing. Some studies provide data which show absence of L1 transfer. Ko et al. (2008), for example, in a study on the acquisition of definiteness and specificity in relation to article use in L2 English, offer evidence that the implementation of semantic universals is not subject to crosslinguistic variation “when L1 transfer effects are controlled forˮ (p. 122), i.e. when the L1/L2 pairings under investigation are similar with respect to the property under investigation. Ionin, Zubizaretta, and Bautista-Maldonado (2008) investigated the learning of article choice in two different contexts: L1 Russian – L2 English and L1 (Mexican) Spanish – L2 English. Importantly, in the first pair the L1 lacked articles but in the second pair both L1 and L2 were article languages. Their data show that the Spanish learners of English behaved native-live, unlike the Russian learners. The authors interpret the results as the effect of L1 transfer of article semantics from Spanish into English. The majority of previous studies on DOM in adult L2 learning focused on L2 Spanish9 in an L1 English environment, i.e. where the L1 lacked an overt differential marker (Guijarro-Fuentes and Marinis 2007; Bowles and Montrul 2008; Montrul and Bowles 2009; Guijarro-Fuentes 2012, Martoccio 2012, among many others). They all provide evidence that DOM is problematic to L2 learners whose L1 is English. Errors of both comprehension and production (such as omission of DOM with animate objects, overgeneralization to inanimate objects, errors of commission) persist even after explicit teaching of the constraints. Several factors have been shown to interfere. One such factor is the nature of the features involved, which are not equally vulnerable. Guijarro-Fuentes and Marinis (2007) tested two intermediate and one advanced group of L2 learners of Spanish. They investigated the learning of four factors which constrain DOM: definiteness/ specificity, animacy, subject theta-role, and situation-type aspect (with focus on telicity). Animacy was the least problematic. Still, only the advanced group showed sensitivity to animacy. But even their responses were different from those of native speakers. Instruction has also been shown to affect DOM use. Bowles and Montrul (2008), Montrul and Bowles (2009), Martoccio (2012) provide 9 We know of only two studies on other language pairings: Bohnacker and Mohammadi (2012) for L2 Persian in an L1 Balochi context (the participants were children) and Ciovârnache and Avram (2013) for L2 Persian in an L1 Romanian context.
102
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
evidence that explicit instruction and feedback improve knowledge of DOM. They lead to an increase in awareness of the structure. Instructed L2 learners are better at rejecting ungrammatical differential object marking than the uninstructed ones. But their responses did not differ with respect to accepting correct DOM structures (Bowles and Montrul 2008). That level of proficiency also matters is shown in Killam (2011). According to his results, adult English learners of L2 Spanish mark objects in accordance with the definiteness/specificity constraint. The frequency of use of the marker, though, is affected by proficiency level. Only the advanced L2 learners in his study showed rates similar to those of natives. With low intermediate and low advanced learners the frequency of use was significantly below the one found with native speakers. Another factor which might be significant is language pairing. The studies mentioned above all investigated DOM in L2 Spanish in an L1 English context. The only exception is that of Guijarro-Fuentes and Marinis (2009), where the L2 learning of the Spanish differential marker a is studied in two contexts: English – Spanish bilinguals (who studied Spanish in a formal context) and Catalan – Spanish sequential bilinguals. In the Catalan – Spanish pairing both languages had an overt DOM marker. These sequential bilinguals performed “slightly betterˮ overall. The authors notice that in the L1 English – L2 Spanish pairing, the L1 “exerts some kind of influenceˮ (p. 90)10. These results indicate that the availability of L1 transfer in the L2 learning of semantically constrained phenomena may also depend on the L1/L2 pairing. The present study extends the investigation to the L2 learning of the semantic features which constrain DOM in an L1 Romanian – L2 Persian and an L1 Persian – L2 Romanian context. Both languages have an overt differential object marker. Animacy constrains the use of the marker only in Romanian. If the L1 system represents the initial stage in L2 learning, as stated by the FTFAH, the Romanian L2er of Persian will start with a DOM system within which marking is allowed only with animate DPs. If implementation of animacy is affected by cross-linguistic influence, one can predict that the L2er will erroneously omit the marking of inanimate objects. The Persian L2er of Romanian, on the other hand, would start with a less restrictive system. If there is L1 transfer from Persian, the L2er will erroneously accept marking with inanimate DPs.
10 The authors, however, cautiously mention that it is difficult to analyse the observed differences exclusively in terms of cross-linguistic influence. Other variables can be associated with the participants’ response patterns. This is because the Catalan participants were sequential bilinguals; they had acquired the language naturalistically. The English participants had learned Spanish in a formal context.
Semantic features and L1 transfer in the L2 learning of differential object marking
103
Referential stability constrains DOM in both languages. But in Romanian marking is obligatory with a smaller number of DP types. If, as predicted by the FTFAH, there is L1 transfer during the initial stage, the Romanian L2ers of Persian will start with a system within which marking is obligatory exclusively with URS objects. Potential L1 transfer effects would include the erroneous use of unmarked lexically headed definite DPs and partitives, with which marking is optional in their L1. The Persian L2er of Romanian would start with a DOM system where object marking is obligatory with a larger number of DPs. The effects of L1 transfer could be reflected in the L2ers’ analysis of URS, CRS and restrictive RNS objects as equally strong DOM triggers. Marking would be obligatory with all these DPs. Importantly, this strategy would not result in ungrammatical structures, since marking with all these DPs is allowed in L2. But, it is optional, not obligatory. We believe that the investigation of these language pairings can contribute to our understanding of the interplay between innate representations and L1 transfer in the L2 learning of semantically constrained object marking.
4 The present study 4.1 DOM in L2 Romanian The main purpose of the present study is to investigate the availability of L1 transfer effects on the implementation of animacy and referential stability in DOM systems. The first experiment focused on the learning of DOM in L2 Romanian by native speakers of Persian.
4.1.1 Participants 15 intermediate L2 learners of Romanian with L1 Persian (age range 21–48) and a group of 15 native speakers of Romanian (age range 21–54) took part in this study. The comparison group were all native speakers of Romanian from Bucharest. The Iranian participants were all speakers of L1 Persian. They were all living in Romania at testing time. Their background was heterogeneous. In order to ensure some linguistic proficiency homogeneity, three factors were taken into account: length of formal instruction, duration of stay in Romania, and the participants’ responses to the distractor sentences in the task (see Section 4.1.2.). The 15 L2ers had all studied Romanian in a formal context for a period of 2 to 3 years and had lived in Romania for a period of 3 to 8 years
104
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
at testing time. The ones who were included in the study gave 15 to 2011 correct responses to the distractors in the task12.
4.1.2 Materials In order to test the L2 learners’ knowledge of the semantic conditions under which pe marks direct objects in Romanian we used an acceptability judgement task. This included a total of 16 test sentences, balanced for animacy, ranging over 4 conditions with 4 test sentences per condition. The conditions are rooted in the framework of Farkas and von Heusinger (2003). The first condition targeted DOM with proper nouns, which is obligatory in Romanian. The other three conditions targeted optional contexts of use: DOM with lexically headed definite DPs in condition 2, with partitives in condition 3, and with indefinites in condition 4. We only tested pe marking in the absence of a clitic. In this respect we followed previous studies of DOM in Romanian which argued that clitic doubling and pe marking should be separated (Farkas and von Heusinger 2003; Ciovârnache and Avram 2012, 2013; Hill 2013). This explains why we did not include test sentences with definite pronouns in the task. In Romanian, differentially marked pronouns are licit only in the presence of a clitic, i.e. they can only appear in clitic doubling constructions. Samples of sentences used in the test are given in Appendix 1. The test also included 24 distractors balanced for grammaticality. They targeted areas of Romanian morpho-syntax which are possibly vulnerable in an L1 Persian – L2 Romanian context (e.g. tenses, prepositions, determiners, word order). The task was solved at home by the participants and sent to one of the researchers via e-mail. The participants were asked to rate the sentences for acceptability. For the sentences which they rated as unacceptable they were also required to provide the acceptable counterpart. Each sentence in which the use of the differential object marker was evaluated as acceptable was given 1 point. This allowed us to measure the marking preference of the participants with each type of DP. Responses which focused on other elements in the test sentence were coded as non-target. For example, the sentence in (18) in condition 1 was corrected by several L2 learners. But instead of using the obligatory
11 We included in the study only those participants with an intermediate proficiency level in order to make the comparison with the L2 learners of Persian in the second study possible. This explains the relatively high number of correct responses which was required as a threshold. 12 We tested 21 L2ers of Romanian, but only 15 met all the criteria and were retained in the present study.
105
Semantic features and L1 transfer in the L2 learning of differential object marking
marker in order to correct the sentence, some of them changed the syntactic frame of the main verb (as in 19): (18) *Am întîlnit chiar ieri Maria have.I met even yesterday Maria ‘I met Maria in the street just yesterday.’
pe on
stradă. street
(19)
pe on
stradă. street
Mam întîlnit ST refl:1 SG have.I met ‘I met Maria in the street.’
cu with
Maria Maria
In Romanian, a întîlni ‘meet’ can be used both with a direct object (in which case, if the object is an animate proper name, pe marking is obligatory) and with a prepositional object (meet with someone). Some participants opted for the prepositional object. Such responses were considered ‘non-target’. The responses of the L2ers were compared to those of the control group of native speakers.
4.1.3 Results Table 2 below summarizes the percentages of accepted marked DPs in each condition by both groups. The total number of items for each condition was 4 per participant, i.e. the overall maximum number of target responses per condition was 60. Remember that DOM was obligatory only in condition 1 but optional in the other three conditions. Importantly, marking was acceptable only with animate DPs across conditions, i.e. for each condition only a maximum of 50% (n = 30) of the test sentences were acceptable. This explains the percentages in Table 2. Table 2: Referential stability and DOM in L2 Romanian: results Participant group
Condition 1
Condition 2
Condition 3
Condition 4
L2ers
42.86% (24/5613)
38.33% (23/60)
35% (21/60)
18.33% (11/60)
Comparison group
54.24% (32/59
40% (24/60)
40% (24/60)
5% (3/60)
13 In condition 1 the L2ers gave only 56 target responses and the control group only 59 out of the maximum of 60. The other responses were non-target (see the discussion in Section 4.1.2.).
106
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
The 54.24% result in condition 1 with the native speakers of Romanian must be correlated with the fact that, unexpectedly, four of the participants accepted pe with an inanimate proper name: (20)
*[. . .]
au vizitat doar o dată have:they visited only one time ‘They have visited Berlin only once.’
pe PE
Berlinul. Berlin:DET
We first compared the overall responses of the L2ers to those sentences which contained referentially stable direct objects (i.e. conditions 1 and 2) to the overall responses to sentences with referentially non-stable direct objects (i.e. conditions 3 and 4). The comparison revealed that the L2ers accepted pe more often with referentially stable direct objects (M = 3.2, SD = 0.861) than with referentially non-stable ones (M = 2.2, SD = 1.265) (t(14) = 3.089, p < .01), revealing sensitivity to the underlying semantic feature of the scale. A repeated measures ANOVA on the L2ers’ responses with condition (i.e. direct object type) as within-subjects factor showed a main effect of condition (F(3,42) = 5.76, p = .002). Pairwise comparisons with Bonferroni correction showed no significant difference between URS and CRS objects (p = 1.00) or between URS and partitives (p = 1.00), i.e. the L2ers treated proper names, lexically headed definite DPs and partitives as equally strong DOM triggers. The difference between proper names and indefinite DPs was significant (p = .01). A mixed ANOVA with the factor group (L2ers, native speakers) as betweensubjects variable and condition (i.e. object type) as within-subjects variable showed a main effect of condition (F(3,34) = 39.36, p < .01), no main effect of group (p > .05), and a significant interaction between group and object type (F(3,34) = 6.25, p < .01). Independent t-tests with Bonferroni correction revealed no significant difference between the two groups in conditions 2, 3 and 4 (p > .01). The difference between the responses of the two groups to the test sentences in condition 1 was the only one which reached significance (t(26) = –4.14, p < .01). The comparison group accepted pe with proper names significantly more often than the L2ers. The difference between URS and CRS objects was significant (p < .01) only with the comparison group. The results for the effects of animacy are summarized in Table 3. Percentages of accepted marked animate and inanimate direct objects are reported. The total number of animate/inanimate items for each category was 8 per participant, i.e. across conditions the total number of possible target responses for animate DPs and for inanimate DPs was 120 each. DOM was obligatory only in condition 1 and optional in the other three conditions, which explains the percentages of accepted marked animate DPs.
Semantic features and L1 transfer in the L2 learning of differential object marking
107
Table 3: Animacy and DOM in L2 Romanian: results Participant group
Animate DP (correct: pe)
Inanimate DP (correct: no pe)
L2ers Comparison group
56.9% (66/11614) 65.5% (78/119)
12% (14/116) 4.2% (5/119)
The picture with respect to animacy is straightforward. The L2ers did not accept marked animate (M = 4.33, SD = 0.303) and marked inanimate DPs (M = 0.93, SD = 1.162) at equal rates. The difference between the overall rate of accepted marked animate and inanimate objects was significant (t(26) = –7.42, p < .001) reflecting sensitivity to the animacy constraint in the L2 system. The comparison of the two groups’ acceptance rates of pe with inanimate objects across conditions reveals no significant difference (t(22) = 1.36, p > .05). The L2ers did not accept pe with inanimate objects significantly more often than the native speakers. These findings are further supported by the analysis of response patterns. The L2ers accepted the (correct) test sentences with an unmarked inanimate object 100% across conditions. The number of test sentences with erroneously marked inanimate objects which were accepted was very low: 3 in conditions 1, 2 and 4 each (i.e. 10%) and 5 (16.7%) in condition 3.
4.1.4 Interim summary Our results show that the intermediate L2ers of Romanian did not accept DOM randomly. They applied the animacy constraint to object marking in L2. The L2ers preferentially accepted marked animate direct objects, in accordance with the target system. On the specificity scale, they distinguished between referentially stable and referentially non-stable DPs, i.e. their marking system was constrained by referential stability. In this respect, their response pattern did not differ from that of the native speakers. But the L2ers treated proper names, lexically headed definite DPs and partitives as equally strong triggers. Their responses differed from those of the native speakers. The latter treated proper names as stronger DOM triggers, in accordance with the properties of the target system. The L2ers’ partition of DPs on the specificity scale is summarized in (21a), the one of the native speakers in (21b):
14 The second number indicates the total number of target responses.
108 (21)
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
a.
L2ers of Romanian: proper nouns, lexically headed definite DPs, partitives > indefinite DPs
b.
native speakers of Romanian: proper nouns > lexically headed definite DPs, partitives > indefinite DPs
The only difference between the responses of the two groups which reached significance was the one in condition 1. Since this was the only condition which targeted obligatory object marking, the difference shows that with intermediate learners the L2 DOM system is not target-like. The intermediate L2ers are sensitive to referential stability and animacy, but the frequency of use is different from the one found with the natives (in this respect our results are similar to those reported for DOM in L2 Spanish in Killam 2011). This difference also casts doubt on the statistical similarity between the two groups in conditions 2 and 3. The L2ers’ acceptance rate could reflect the optionality of the target system but it could also reflect a (still) unstable DOM system. The native participants optionally marked lexically headed definite DPs and partitives, in accordance with DOM constraints in Romanian. But the L2ers may have marked these objects optionally because their L2 DOM system was not target-like.
4.2 DOM in L2 Persian The second task investigated the learning of DOM in L2 Persian by adult native speakers of Romanian.
4.2.1 Participants The experiment involved 14 intermediate Romanian learners of L2 Persian (age range 20–30) and a comparison group of 11 native speakers of Persian (age range 24–45). The native controls were all native speakers of Persian from Iran. They were all living in Romania at testing time. The L2ers were 3rd year students of Persian at University of Bucharest. They had studied Persian in a formal setting for 19 months at testing time (10 hours/week). In order to ensure the homogeneity of the group, in the absence of a standardized test, we took into account the participants’ results in all the proficiency language exams they had taken in the department before testing time. The minimum average grade for allowing the participant to remain in the group was 7 on a 1 to 10 scale. This first classification was then corrected on the basis of the responses to the distractor
Semantic features and L1 transfer in the L2 learning of differential object marking
109
sentences in the test (see Section 4.2.2.). Only those participants who gave at least 70% correct answers remained in the group. The use of rā had not been explicitly taught before the test.
4.2.2 Materials In order to investigate the L2ers’ knowledge of the semantic conditions under which rā marks direct objects in Persian we used an acceptability judgement task (the same as the one used in Ciovârnache and Avram 2013). It included a total of 32 test sentences, balanced for animacy and ranging over 4 conditions with 4 test sentences per condition. The first three conditions targeted obligatory contexts of use for rā: condition 1 – proper nouns and definite pronouns, condition 2 – lexically headed definite DPs, condition 3 – partitives. Condition 4 targeted a context in which the use of rā was optional: indefinite DPs. Samples of test sentences for each condition are given in Appendix 2. The test also included 32 distractors (which targeted the use of tense and aspect: perfective vs. imperfective forms, the progressive, the use of the subjunctive, relative clauses, conditional clauses, constructions with ezāfe, prepositions, partitives, the indefinite marker, enclitic pronouns, plural forms, degrees of comparison). The participants were asked to rate the sentences as acceptable or unacceptable. For the sentences rated as unacceptable they were also required to provide the correct version. The task was solved in class by the L2ers, in the presence of one of the researchers. The native speakers solved it at home and sent it to one of the researchers via e-mail.
4.2.3 Results The L2ers’ acceptance of rā with referentially stable direct objects (i.e. with definite pronouns, proper names and definite DPs) (M = 11.28, SD = 3.930) was higher than with referentially non-stable ones (partitives and indefinites) (M = 7, SD = 3.762): t(14) = 4.229, p < .01). The percentages of accepted marked DPs in each condition are summarized in Table 5. They show that both the L2ers and the native speakers preferentially accepted differential marking with direct objects which were more referentially stable.
110
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
Table 5: Referential stability and DOM in L2 Persian: results Participant group
Condition 1
Condition 2
Condition 3
Condition 4
L2ers n = 14
71.43% (80/112)
66.07% (74/112)
54.46% (61/112)
30.35% (34/112)
Comparison group n = 11
97.72% (86/88)
97.72% (86/88)
95.45% (84/88)
55.68% (49/88)
A repeated measures ANOVA with object type as within-subjects factor showed a main effect of condition with the intermediate L2ers (F(3,39) = 12.88, p = .000). Pairwise comparisons with Bonferroni correction revealed no significant difference between URS and CRS direct objects (p = 1.00). The difference between partitives and definite descriptions did not reach significance either (p = .29). The difference between indefinites and any other DP type was significant (p < .01). The correlation between the responses to test sentences with a proper noun and those with a lexically headed definite DP was strong (r = .65, p = .01). A strong correlation was also found between the responses to test sentences with a proper noun and a partitive (r = .72, p < .01). The comparison between Persian L2ers and the control group of native speakers of Persian shows that the overall rate of appropriate responses is significantly lower with the L2ers in all four conditions as shown by independent t-tests (proper nouns: t (23) = –3.80, p < .05, definite DPs: t(23) = –4.15, p < .05, partitives: t(23) = –5.41, p < .05, indefinite DPs: t(23) = –3.28, p < .05). There was no significant difference between the L2 learners’ acceptance rate of rā with animate (M = 8.5, SD = 2.90) and with inanimate (M = 9.28, SD = 3.85) direct objects overall (t(13) = 1.10, p = .29). There was a strong correlation between marked animate and marked inanimate DPs overall (r = .82; p < .01). The correlation between marked animate and marked inanimate objects is in between moderate to strong in each of the four conditions: C1: r = .51, p = .06; C2: r = .86, p < .01; C3: r = .56, p < .05; C4: r = .51, p = .06. The data show that the L2ers accept DOM with both animate and inanimate DPs, in accordance with the Persian system. This conclusion is further supported by the results of error analysis. The L2ers made two types of errors: (i) they accepted sentences which erroneously lacked rā; (ii) they rejected sentences which correctly contained rā. Taking rejection as a stronger indicator of the properties of the interlanguage, we took a closer look at the rejection errors (n = 24). The results are summarized in Table 6. They indicate lack of animacy bias. With indefinites, the higher rejection rate of DOM with inanimate indefinites is in line with the animacy bias in the target language.
Semantic features and L1 transfer in the L2 learning of differential object marking
111
Table 6: DOM and animacy in L2 Persian: error analysis Condition
reject rā with inanimates
reject rā with animates
URS CRS Partitives Indefinites
12.5% (3/24) 16.67% (4/24) 29.17% (7/24) 83.33% (20/24)
12.5% (3/24) 20.8% (5/24) 29.17% (7/24) 62.5% (15/24)
But, in spite of the fact that there is no animacy bias, in accordance with the DOM system of Persian, the comparison with the control group reveals that the overall rate of appropriate responses is significantly lower with the L2ers. The difference between the L2ers’ marked animate objects and those of the comparison group of native speakers is significant (t(22) = –4.57, p = .000). The difference between the marked inanimate objects by the two groups is also significant (t(16) = –3.65, p = .002). Table 7 summarizes the acceptance rates of the two groups with respect to animacy. Remember that the use of rā was obligatory in conditions 1–3, but optional in condition 4. Table 7: Animacy and DOM in L2 Persian: results Participant group
Animate DP
Inanimate DP
L2ers Comparison group
53.12% (119/224) 90.34% (159/176)
58.03% (130/224) 82.95% (146/176)
4.2.4 Interim summary The overall rate of the L2ers’ appropriate answers is significantly lower than the one of the comparison group, showing that their knowledge of the system is not native-like. But, despite this low rate, the responses of the intermediate L2ers reveal that differential marking is not random in their interlanguage either with respect to animacy or referential stability. The L2ers of Persian treated animate and inanimate DPs alike, in accordance with the target DOM system (see also Ciovârnache and Avram 2013 for similar results with less proficient L2ers of Persian). URS objects, CRS objects and partitives were treated as equally strong triggers. In this respect, their responses were in accordance with the response pattern of the comparison group.
112
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
5 Discussion The main purpose of the present study was to investigate the availability of L1 transfer effects in the implementation of animacy and referential stability in DOM systems in L2 Romanian and L2 Persian. With respect to animacy, the availability of L1 transfer was predicted to have the effects summarized in Table 8: Table 8: DOM and L1 transfer of animacy: predictions
If + L1 transfer
L1 Persian – L2 Romanian
L1 Romanian- L2 Persian
L2ers will erroneously accept a significant number of marked inanimate DPs
L2ers will erroneously reject a significant number of marked inanimate DPs
These predictions were not borne out by the data. The intermediate 2Lers of Romanian correctly treated animacy as a strong trigger of DOM. They preferentially accepted pe with animate direct objects and correctly rejected marked inanimate DPs. The fact that the L2ers occasionally extended DOM to inanimate objects cannot be interpreted as an effect of L1 transfer. The number of these errors is extremely low and the difference between the accepted marked inanimates by the two participant groups does not reach significance. Besides, some native speakers also overextended the use of pe to inanimate proper names. The use of pe with inanimates has been reported for spoken Romanian in various studies of DOM, with an upgrading effect. The no L1 transfer interpretation of the results is further supported by L1 acquisition data: such errors are attested in the L1 acquisition of DOM in Romanian (Ticio and Avram 2014). The intermediate L2ers of Persian accepted both marked animate and marked inanimate direct objects. Their errors did not reflect any animacy bias. Summing up on animacy, the results reveal lack of L1 transfer effects with intermediate L2ers in both contexts, with no directionality effect. Animacy is not a vulnerable feature in L2 DOM systems, as also shown for DOM in L2 Spanish (Guijarro-Fuentes and Marinis 2007; Guijarro-Fuentes 2012). Referential stability is a DOM trigger in both Persian and Romanian. Trigger strength, however, is reflected differently in the two systems. The availability of L1 transfer was predicted to have the effects summarized in Table 9:
Semantic features and L1 transfer in the L2 learning of differential object marking
113
Table 9: DOM and L1 transfer of referential stability: predictions
If + L1 transfer
L1 Persian – L2 Romanian
L1 Romanian- L2 Persian
L2ers will treat URS, CRS and restrictive RNS direct objects as equally strong DOM triggers; this could result in a higher number of marked CRS and restrictive RNS direct objects
L2ers will treat only URS direct objects as strong DOM triggers; this could result in a lower number of marked CRS and restrictive RNS objects
The responses of the intermediate L2ers were not random. DPs higher on the specificity scale were significantly more frequently accepted with the differential object marker in both learning contexts under investigation. This reflects sensitivity to referential stability, the underlying feature of the scale. The comparison of the results in the two tasks reveals that the intermediate L2ers partition trigger strength on the specificity scale identically, irrespective of the properties of the DOM system in the target L2: URS, CRS and restrictive RNS direct objects are equally strong DOM triggers in their interlanguage. As one anonymous reviewer points out, the data show that intermediate L2 learners distinguish only between restricted and unrestricted referential stability, irrespective of the way in which DOM partitions the referentiality scale in the target language. The absence of L1 effects is more obvious for the L1 Romanian – L2 Persian context, where transfer predicts a significant difference between marked proper names, on the one hand, and marked definite DPs and partitives, on the other hand. This, however, was not borne out by the data. Further evidence that the intermediate L2ers partition trigger strength on the specificity scale identically is provided by the similar gradience of the results across conditions in the two studies: (i) L2 Romanian: 42.86% > 38.33% > 35% > 18%; (ii) L2 Persian: 71.43% > 66.07% > 54.46% > 30.35%. But, at least at first sight, our data could also suggest availability of transfer in the L1 Persian – L2 Romanian context. In this case, transfer of trigger strength on the scale from Persian into Romanian would involve the obligatory marking of proper nouns, lexically headed definite DPs and partitives. This should have entailed (i) a higher number of acceptance responses in conditions 2 and 3 with the L2ers than with the control group; and (ii) a similar acceptance rate across conditions 1, 2 and 3 with the L2ers. The second prediction is indeed borne out by the data. But the fact that we found the same pattern in the L1 Romanian – L2 Persian context weakens this piece of evidence. Moreover, the first prediction was not borne out by the data: there was no significant difference between the two groups in conditions 2 and 3. Interestingly, contrary to the predictions that
114
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
are found in the theoretical literature, the Romanian native speakers in our study treated partitives and definite descriptions as equally strong DOM triggers. The same picture was found with the L2ers. The only difference between the two groups was found in condition 1, the only context where marking is obligatory in both L1 and L2. The acceptance rate was lower with the L2ers, a result which challenges an L1 transfer explanation. This difference may have been determined by the ‘strong overt marking’ in this condition by the Romanian natives15. Indeed, some of the native speakers accepted pe marking even with inanimate proper names. Alternatively, the difference may indicate that, in spite of sensitivity to referential stability, the intermediate L2 system is not target-like yet. We suggest that our data are compatible with a no L1 transfer interpretation. As mentioned above, the L2ers, irrespective of their L1/L2, treat URS, CRS and restrictive RNS objects as equally strong DOM triggers. Prominence on the scale cuts DOM triggers in only two classes (22): (22)
URS, CRS, restrictive RNS > non-restrictive RNS
This could represent an intermediate stage in the L2 implementation of referential stability in DOM systems. The fact that it corresponds to the Persian system may be merely accidental. In the interlanguage of intermediate learners, a DP is a (strong) DOM trigger whenever the choice of the value assigned to the variable which the DP introduces is restricted to a set. In this system, prominence on the scale is not yet determined by relative referential stability, i.e. it is not sensitive to referential stability types. Referential stability constrains object marking. But the L2ers have not yet fully identified in what way different referential stability types correlate with object marking in the target language. This is reflected in optionality of object marking and lower frequency of use than in the target language. The optionality reveals an incomplete DOM system. It is only at a later stage that the L2ers will (probably) fully integrate types of referential stability/non-stability into the target DOM system. According to Papp (2000)16 ‘incomplete optionality’ can be found even with highly advanced L2ers. It differs from optionality in divergent end-state grammars (which involves some sort of fossilization), in that it ‘provides the potential for further progress, even at ultimate attainment’ (p. 183). The acquisition of how referential stability constrains DOM can be delayed. Importantly, the delay is not the result of L1 transfer of DOM trigger strength on the specificity scale. Our findings are similar, in this respect, to those reported 15 We thank one anonymous reviewer for suggesting this possible explanation. 16 We thank an anonymous reviewer for recommending this paper to us.
Semantic features and L1 transfer in the L2 learning of differential object marking
115
in previous studies on the L2 learning of DOM. In particular, the delay in the implementation of specificity in DOM systems has also been reported for adult L2 Spanish (see, e.g. Killam 2011; Guijaro-Fuentes 2012).
6 Conclusions In this chapter we addressed the issue of possible cross-linguistic interference effects in the implementation of semantic features in the L2 learning of DOM. We focused on animacy and referential stability in two learning contexts: L1 Persian – L2 Romanian and L1 Persian – L2 Romanian. Our results showed that adult L2ers of intermediate proficiency were sensitive to the role of animacy and of referential stability in DOM systems, reflecting access to semantic universals. In accordance with several previous studies, our data also provided evidence that semantic factors are not equally vulnerable in L2 DOM systems. Our findings revealed lack of L1 transfer with respect to the role of animacy in DOM, regardless of learning context. The role of animacy in DOM is learned early. An interesting finding was that the implementation of referential stability in DOM systems involves a similar route across the L2 learning contexts which we investigated. The intermediate L2ers go through a stage when the system is sensitive to referential stability but blind to type of referential stability. URS, CRS, and partitives are treated as equally strong DOM triggers, irrespective of the properties of DOM in L1/L2. This stage is also characterized by optional marking of objects. We suggested that this might be a developmental stage in the learning of the role of referential stability in DOM. Such an interpretation is, however, in need of confirmation from further cross-linguistic studies on referential stability in L2 DOM systems.
Appendix 1 DOM in L2 Romanian. Sample of test sentences Condition 1: URS direct objects (proper nouns) *Am întîlnit ieri Maria pe stradă. have.I met yesterday Maria on street ‘I met Maria in the street yesterday.’
116
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
Condition 2: CRS direct objects Am găsit pe băiatul vecinilor. have.I found PE boy:DET neighbour:PL .GEN ‘I have found the neighbours’ son.’ Condition 3: RNS direct objects Au fotografiat pe una dintre studente have.they photographed PE one of students ‘They photographed one of the students.’ Condition 4: non-restrictive NS direct objects *Ieri am cumpărat pe un apartament. E chiar în centru. yesterday have.I bought PE a flat is right in centre ‘Yesterday I bought a flat. It’s right in the centre of the town.’
Appendix 2 DOM in L2 Persian. Sample of test sentences Condition 1: URS direct objects (proper nouns and definite pronouns) Do sāl dar Irān zendegi kardam vali mota’sefāne Esfahān two year in Iran life did but unfortunately Esfahān rā faqat yek bār didam. RĀ only once saw.I ‘I lived in Iran for two years, but unfortunately I visited Isfahan only once.’ Condition 2: CRS direct objects *Nāme-ye u diruz daryāft kardam. letter-EZ he yesterday receive did.I ‘I received his letter yesterday.’ Condition 3: RNS direct objects *Čand tā az medādhā-ye qermez gom kardam. some classifier of pencils-EZ red lost did.I ‘I lost some of the red pencils.’
Semantic features and L1 transfer in the L2 learning of differential object marking
117
Condition 4: non-restrictive NS direct objects Hafte-ye gozašte āpartemān-i rā xaridam. Ān dar markaz-e šahr ast. week-EZ last apartment:INDEF RĀ bought.I it in centre-EZ city is ‘I bought an apartment last week. It is in the centre of the city.’
References Aissen, Judith. 2003. Differential object marking: Iconicity and economy. Natural Language and Linguistic Theory 21(3). 435−483. Avram, Larisa. 2014. Differential object marking: the view from language acquisition. Paper presented at The 16th Annual Conference of the English Department (ACED 16), University of Bucharest, 6−8 June. Avram, Larisa & Martine Coene. 2009. Null objects and Accusative clitics. Bucharest Working Papers in Linguistics XI (1). 233−252. Bohnacker, Ute & Somayeh Mohammadi. 2012. Acquiring Persian object marking: Balochi learners of L2 Persian. Orientalia Suecana LXI. 59−89. Bossong, Georg. 1985. Empirische Universalienforschung. Differentielle Objektmarkierung in den neuiranischen Sprachen. Tübingen: Narr. Bossong, Georg. 1991. Differential object marking in Romance and beyond. In Dieter Wanner & D. A. Kibbee (eds.), New analyses in Romance linguistics. Selected papers from the XVIII Linguistic Symposium on Romance Languages 1988, 143−170. Amsterdam: John Benjamins. Bossong, Georg. 1998. Le marquage de l’expérient dans les langues d’Europe. In Jack Feuillet (ed.), Actance et valence dans les langues de l’Europe, 259−294. Berlin: Mouton de Gruyter. Bowles, Melissa & Silvina Montrul. 2008. The role of explicit instruction in the L2 acquisition of the a-personal. In Mayella Almazan, Joyce Bruhn de Garavito & Elena Valenzuela (eds.), Selected papers from the 8th Hispanic Linguistics Symposium, 25−35. Somerville, MA: Cascadilla Press. Cagri, Ilhan. 2007. Persian accusative case: A lexicalist approach. http://www.linguistik.uni-kiel. de/icil2/58Cagri.pdf. (accessed 15 December 2012) Carnie, Andrew. 2005. A Phase-geometric approach to multiple marking systems. In Martha McGinnis & Norvin Richards (eds.), Perspectives on phases. MIT Working Papers in Linguistics 49. 87−102. Ciovârnache, Cristina & Larisa Avram. 2012. Differential object marking in Persian and Romanian: Predictions for L2 learning. Paper presented at Workshop on Specificity, University of Bucharest, 12−13 December. Ciovârnache, Cristina & Larisa Avram. 2013. Specificity and animacy in the acquisition of DOM in Persian as L2. Revue roumaine de linguistique LVIII (4). 419−438. Clahsen, Harald & Pieter Muysken. 1986. The availability of universal grammar to adult and child learners – a study of the acquisition of German word order. Second Language Research 2(2). 93−119. Dobrovie-Sorin, Carmen. 1990. Clitic doubling, wh-movement, and quantification in Romanian. Linguistic Inquiry 21(3). 351−397.
118
Larisa Avram, Cristina Ciovârnache and Anca Sevcenco
Dobrovie-Sorin, Carmen. 1994. The syntax of Romanian. Comparative studies in Romance. Berlin: Mouton. Epstein, Samuel, Suzanne Flynn & Gita Martohardjono. 1998. The strong continuity hypothesis: Some evidence concerning functional categories in adult L2 acquisition. In Suzanne Flynn, Gita Martohardjono & Wayne O’Neil (eds.), The generative study of second language acquisition, 61−77. Mahwak, NJ: L. Erlbaum. Farkas, Donka. 1978. Direct and indirect object reduplication in Rumanian. In Donka F. Farkas, Wesley M. Jacobsen & Karol W. Todrys (eds.), Papers from the fourteenth regional meeting Chicago Linguistic Society, April 13−14, 1978. 88−97. Chicago: Chicago Linguistic Society. Farkas, Donka. 2002. Specificity distinctions. Journal of Semantics 19 (1). 1−31. Farkas, Donka & Klaus von Heusinger. 2003. Stability of reference and object marking in Romanian. Paper presented at Workshop on Direct Reference and Specificity, ESSLLI, Vienna, August. Ganjavi, Shadi. 2007. Direct objects in Persian. University of Southern California dissertation. Ghomeshi, Jila. 1997. Topics in Persian VPs. Lingua 120 (1). 133−147. Guijarro-Fuentes, Pedro. 2012. The acquisition of interpretable features in L2 Spanish: Personal a. Bilingualism: Language and Cognition 15(4). 701−720. Guijarro-Fuentes, Pedro & Theo Marinis. 2007. Acquiring the syntax/semantics interface in L2 Spanish: the personal preposition a. Eurosla Yearbook 2007 (7): 67−87. Guijarro-Fuentes, Pedro & Theo Marinis. 2009. The acquisition of the personal preposition a by Catalan-Spanish and English-Spanish bilinguals. In Joseph Collentine, Maryellen García, Barbara Lafford & Francisco Marcos Marín (eds.), Selected Proceedings of the 11th Hispanic Linguistics Symposium, 81−92. Somerville, MA: Cascadilla Proceedings Project. von Heusinger, Klaus, Udo Klein & Pieter de Swart. 2008. Variation in differential object marking. Paper presented at Workshop on case variation, Stuttgart, June. Hill, Virginia. 2013. The direct object marker in Romanian: A historical perspective. Journal of Australian Linguistics. 33 (2): 140−151. Ionin, Tania, Heejeong Ko & Ken Wexler. 2004. Article semantics in L2-acquisition: the role of specificity. Language Acquisition 12(1). 3−69. Ionin, Tania, Maria Luisa Zubizarreta & Salvador Bautista-Maldonado. 2008. Sources of linguistic knowledge in the second language acquisition of English articles. Lingua 118 (4). 554−576. Karimi, Simin. 1990. Obliqueness, specificity, and discourse functions: rā in Persian. Linguistic Analysis 20 (3−4). 139−191. Karimi, Simin. 1996. Case and specificity: Persian rā revisited. Linguistic Analysis 26 (3−4). 174−194. Khanlari, P. 1973. Dastur-e zaban-e farsi [Persian Grammar]. Tehran: Entesharat-e Boniad-e Farhang-e Iran. Killam, Jason. 2011. An interlanguage analysis of differential object marking in L2 Spanish. Indiana University dissertation. Ko, Heejeong, Tania Ionin & Ken Wexler. 2010. The role of presuppositionality in the second language acquisition of English articles. Linguistic Inquiry 41(2). 213−254. Ko, Heejeong, Alex Perovic, Tania Ionin & Ken Wexler. 2008. Semantic universals and variation in L2 article choice. In Roumyana Slabakova, Jason Rothman, Paula Kempchinsky & Elena Gavruseva (eds.), Proceedings of the 9th Generative Approaches to Second Language Acquisition Conference (GASLA 2007), 118−129. Somerville, MA: Cascadilla Proceedings. Lambton, Ann K.S. 1963. Persian grammar. Cambridge: Cambridge University Press. Lazard, Gilbert. 1992. A grammar of contemporary Persian. Costa Mesa, CA: Mazda Publishers.
Semantic features and L1 transfer in the L2 learning of differential object marking
119
Mahootian, Shahrzad. 1997. Persian. London: Routledge. Mardale, Alexandru. 2007. Les prépositions fonctionnelles du roumain: étude comparative. Diderot Paris (Paris 7) and University of Bucharest dissertation. Mardale, Alexandru. 2008. Microvariation within differential object marking: Data from Romance. Revue roumaine de linguistique LIII (4). 449−467. Martoccio Alyssa Marie. 2012. The acquisition of differential object marking in L2 Spanish learners. University of Illinois at Urbana-Champaign dissertation. Montrul, Silvina & Melissa Bowles. 2009. Negative evidence in instructed heritage language acquisition: A preliminary study of differential object marking. In Melissa Bowles, Rebecca Foote, Silvia Perpiñán & Rakesh Bhatt (eds.), Selected Proceedings of the 2007 Second Language Research Forum, 252−262. Sommerville, MA: Cascadilla Proceedings Project. Naes, Åshild. 2004. What markedness marks: the markedness problem with direct objects. Lingua 114(9−10). 1186–1212. Najafi, A. 1992. Ghalat nanevisim [Let’s not make mistakes]. Tehran:Markaz-e Nashr-e Daneshgahi. Papp, Szilvia. 2000. Stable and developmental optionality in native and non-native Hungarian grammars. Second Language Research 16(2). 173–200. Rodríguez-Mondoñedo, Miguel. 2007. The syntax of objects: Agree and differential object marking. University of Connecticut dissertation. Rodríguez-Mondoñedo, Miguel. 2008. The acquisition of differential object marking in Spanish. Probus 20. 111–145. Schwartz, Bonnie & Rex A. Sprouse. 1996. L2 cognitive states and the Full Access/Full Transfer model. Second Language Research 12(1). 40–72. Slabakova, Roumyana. 2006. Learnability in the second language acquisition of semantics: a bidirectional study of a semantic parameter. Second Language Research 22 (4). 498–523. Slabakova, Roumyana & Silvina Montrul. 2009. Genericity and aspect in L2 acquisition. Language Acquisition 11(3). 165–196. Ticio, Emma & Larisa Avram. 2014. The acquisition of DOM in Spanish and Romanian: Semantic features or semantic scales? Paper presented at the symposium Variation in the L1 acquisition of DOM? 13th International Congress for the Study of Child Language (IASCL 2014), Amsterdam, 13–18 August. Tigău, Alina. 2011. Syntax and semantics of the direct object in Romance and Germanic languages. Bucharest: Editura Universităţii din Bucureşti. Tigău, Alina. 2014. Specificity effects with clitic doubling and pe marking. Bucharest Working Papers in Linguistics XVI (1). 43−61. Torrego, Esther. 1998. The dependencies of objects. Cambridge, MA: MIT Press. Tsimpli, Ianthi Maria & Anna Roussou. 1991. Parameter-resetting in L2? UCL Working Papers in Linguistics 3. 149−169. White, Lydia. 1989. Universal grammar and second language acquisition. Amsterdam: John Benjamins. Windfuhr, G. 1979. Persian grammar: History and state of its study. The Hague, Paris & New York: Mouton.
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
5 Elicited production of who-questions by school-aged Italian-speaking children Abstract: This chapter reports the results of an elicited production experiment run with Italian-speaking children (6;3–10;4 year olds) and a control group of adults. Participants were induced to produce potentially ambiguous Who V DP questions, i.e. questions where a singular verb agrees with either the whelement (subject-extracted questions) or a singular postverbal subject (objectextracted questions). With respect to adults, children employ a wider range of interrogative structures in addition to Who V DP ones, especially in the object condition. This is similar to the findings by Guasti, Branchini, and Arosio’s (2012) study of the elicited production of unambiguous wh-questions in younger children (aged 3;11–5;11). We describe similarities and differences found across the two studies, and discuss the nature of the differences emerged between subject and object interrogative sentences and between children and adults. Guasti et al.’s analysis in terms of strength of agreement and interference is adopted to analyze the productions by the children we test. Due to their older age, our children produce two additional types of interrogative structures, namely passive and embedded interrogatives, not attested in Guasti et al.’s results.
1 Introduction 1.1 Wh V DP interrogative sentences in Italian In Italian wh-questions, the distribution of subjects is restricted: a DP subject cannot occur between the wh-phrase and the verb, nor can it invert with the verb, as shown in the following object-extracted interrogatives: (1) *Chi il bambino insegue? who the child chases ‘Whom does the child chase?’
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti, Università Ca’ Foscari Venezia
122
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
(2) *Chi sta il bambino inseguendo? who is the child chasing ‘Whom is the child chasing?’ To form a grammatical object wh-question, a lexical DP subject must be placed either in postverbal position1 (3), or in left-dislocated position (4): (3)
Chi sta inseguendo il bambino? who is chasing the child ‘Whom is the child chasing?’
(4)
Il bambino, chi sta inseguendo? the child who is chasing ‘The child, whom is he chasing?’
Moreover, since Italian is a null subject language, a null subject is licit when a non-subject constituent is extracted (5): (5)
Chi sta inseguendo? who is chasing ‘Whom is he chasing?’
Given that the interrogative pronoun who is singular, its role is potentially ambiguous between an object and a subject interpretation in sentences where the verb and the postverbal DP are also singular, as in sentence (3). On the other hand, (4) and (5) cannot be interpreted as questioning the subject: a dislocated object constituent is obligatorily resumed by a clitic pronoun in Italian, as in (6); as for null arguments, Italian does not license object-drop.2 (6)
Il bambino, chi lo sta the child who him is ‘The child, who is chasing him?’
inseguendo? chasing
When the subject occurs postverbally, subject-verb agreement can be a cue for disambiguation; in both (7) and (8), the postverbal DP is plural; (7) contains a plural verb agreeing with the postverbal DP; thus, the sentence can only 1 As regards the location of postverbal subjects in Italian, we follow Cardinaletti (2001, 2002, 2007) according to whom the subject occupies specvP and is destressed in situ. 2 See Rizzi (1986) for a description of the few cases where null objects are permitted in Italian.
Elicited production of who-questions by school-aged Italian-speaking children
123
be interpreted as questioning the object; the opposite is true for the subjectextracted question in (8), which contains a singular verb agreeing with the interrogative constituent who. (7)
Chi stanno inseguendo i bambini? who are chasing the children ‘Whom are the children chasing?’
(8)
Chi sta inseguendo i bambini? who is chasing the children ‘Who is chasing the children?’
1.2 Acquisition data on Italian wh-questions Wh-questions are attested in Italian spontaneous child speech before the age of three (Guasti 1996, p. 263). Guasti (2000) elicited various types of adult-like interrogative sentences in 3 and 4 y.o. children, including cleft questions (9), object questions with postverbal subjects (10), and object questions with leftdislocated subjects (11). (9)
Chi è che aiuta la mamma? (3;1) who is that helps the mum ‘Who is it that helps the mum?’
(10)
Cosa può fare il cowboy? (3;1) what can do the cowboy ‘What can the cowboy do?’
(11)
Luigino, dove non può Luigino where not can ‘Luigino, where can’t he go?’
andare? (4;7) go
As for the comprehension modality, by administering a picture-matching task to 352 Italian-speaking children ranging in age from 3 to 11 years old, De Vincenzi et al. (1999) showed that children comprehend subject who-questions disambiguated by subject-verb agreement like the one in (8) far better than their counterparts involving object extraction, (7); such asymmetry is particularly remarkable from the age of 4 y.o. until the age of 9 y.o.
124
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
In a recent elicitation study, Guasti, Branchini, and Arosio (2012) analyzed preschool children’s production of argument who- and which-questions disambiguated by subject-verb agreement, analogue to those tested by De Vincenzi et al.: children perform more accurately with subject who-questions as compared to object who-questions. Moreover, the authors report production of a wider set of adequate answering strategies alternative to the target sentences in the object condition: besides producing the targeted Wh V DP questions, (12), children often drop the subject DP (13) or place it in a left peripheral position (14) much more frequently than adults do, while hardly ever resorting to nontarget-like questions in the subject condition: (12)
Chi sporcano gli elefanti? who dirty-3PL the elephants ‘Whom are the elephants dirtying?’
(13)
Chi sporcano? who dirty ‘Whom are they dirtying?’
(14)
Gli elefanti, chi sporcano? the elephants who dirty ‘The elephants, whom are they dirtying?’
The aim of this study is to widen the findings reported by Guasti et al. (2012) in two respects: first, we investigate the elicited production of who-questions by older, school-aged Italian-speaking children, in order to explore similarities and differences manifested in development as compared to 3 and 4 y.o. children. Secondly, we test the elicited production of potentially ambiguous whoquestions, i.e. questions that cannot be disambiguated by subject-verb agreement, as shown by the two possible readings of (15), reported in (15a) and (15b), in order to check what is the factor behind the difficulty of object Wh V DP who-questions with respect to subject who-questions, whether marked plural verb morphology or the postverbal position of the subject (note that 3rd person singular is an unmarked form in Italian, with no dedicated morphology)3:
3 We also aimed at determining whether prosodic properties may distinguish between the two interpretations of (15) in Italian, and at investigating children’s ability to realize a so-called marginalization intonation which characterizes postverbal DPs in wh-questions (Cardinaletti 2001, 2002). In order to clarify the issue, a prosodic analysis of adults’ and children’s questions collected during the experiment has been conducted (Del Puppo 2016). The only study facing
Elicited production of who-questions by school-aged Italian-speaking children
(15) #Chi who
sta is
lavando washing
il the
125
bambino? child
a.
‘Who is washing the child?’
b.
‘Whom is the child washing?’
If the complexity of Wh V DP object questions with respect to Wh V DP subject questions is due to the postverbal position of the subject in the former, the same difficulty should be expected in ambiguous and non-ambiguous object who-questions, despite the presence or absence of verb morphology. Altogether, despite the older age of our participants and the different experimental materials employed in our experiment with respect to Guasti et al.’s study, the pattern of responses collected in the two experiments is very similar. Interestingly, some additional structures, namely passive and embedded interrogatives, emerge in our study, which are probably due to the older age of the children. Participants, the experimental task and the coding criteria are presented in the following section. Results are reported and discussed in section 3 and 4. Section 5 addresses the relevant conclusions.
2 Methods In this section, we present the participants in our task, the experimental design, and the coding criteria.
2.1 Participants 113 typically developing children aged 6;3 to 10;4 took part in the production experiment. All children were native speakers of Italian, living and attending primary school in Venice. Eleven adults from Venice and its surroundings volunteered as control participants:
the topic we are aware of is based on Dutch (Read, Kraak, and Boves 1980), and suggests that distinct intonational properties alone do not determine the interpretation of who-questions in absence of morphological or syntactic contrasts, though they can influence it. This raises the issue of the reliability of what has been said by our participants when an ambiguous string of words has been uttered (also see section 2.3).
126
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
Table 1: Participants across age groups Age Groups
No. of Participants
Mean Age
SD (months)
6-Year-Old 7-Year-Old 8-Year-Old 9-Year-Old Adults
17 32 27 37 11
6;7 7;4 8;5 9;6 23;6
2 3 3 4 44
2.2 Design and materials The experimental design is based on Guasti et al. (2012). Participants were induced to ask who-questions to a puppet, named Poldo, which was present in the experimental setting. The experimental stimuli were shown in a PowerPoint presentation: both children and adults saw a set of pictures where either the agent or the patient of the event was hidden, depending on whether the targeted interrogative questioned the subject or the object constituent (Figure 1 and Figure 2, respectively). Simultaneously, participants listened to a prerecorded voice that described what was happening in the depicted event; the hidden, mysterious character was referred to as “someone”.
PUPPET: Qui qualcuno sta pettinando un bambino. E forse Poldo sa chi. Chiedilo a lui. ‘Here, someone is combing a child. Maybe Poldo knows who. Ask him who’.
Figure 1
Elicited production of who-questions by school-aged Italian-speaking children
(16)
a.
TARGET QUESTION:
127
Chi sta pettinando/pettina il bambino? ‘Who is combing/combs the child?’
PUPPET: Qui un bambino sta pettinando qualcuno. E forse Poldo sa chi. Chiedilo a lui. ‘Here, a child is combing someone. Maybe Poldo knows whom. Ask him whom.’
Figure 2
b.
TARGET QUESTION:
Chi sta pettinando/pettina il bambino? ‘Whom is the child combing?’ ‘Whom does the child comb?’
In order to find out who was hidden behind the circles/ellipsis, participants had to ask a question to the puppet Poldo4: participants were told that Poldo was a reindeer coming from Scandinavia; he didn’t speak Italian, but wanted to learn the language; when participants asked him a question, he looked for the answer in his (complete) pictures and responded to the question trying to give the correct answer; participants were then shown the complete images and had to correct Poldo if he was wrong. Six transitive, reversible verbs were employed: inseguire ‘chase’, lavare ‘wash’, pettinare ‘comb’, baciare ‘kiss’, accarezzare ‘caress’, salutare ‘greet’. The relevant DPs were all singular in number. Each verb was presented twice, once to elicit a subject question and once to elicit an object question, so as to collect six minimal pairs of superficially identical, potentially ambiguous interrogative sentences, as shown in (16a) and (16b). On the whole, participants were exposed to 12 stimuli eliciting who-questions; such stimuli were administered together with 24 stimuli eliciting restrictive relative clauses and 6 filler stimuli.
4 The procedure we employed slightly differ from the one used by Guasti et al. (2012): in Guasti et al.’s experiment, adults were expected to ask questions to an imaginary person, and not to a puppet. Moreover, the prerecorded lead-ins that participants heard were slightly different. The following one was aimed at eliciting an object question in Guasti et al.’s experiment (cf. Belletti & Guasti 2015: 212): i. Look here. There are two bears that tie someone. He knows who. Ask him who.
128
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
2.3 Coding In addition to Wh V DP questions, exemplified in (16a) and (16b), children employed other types of correct interrogative sentences; the main typologies are cleft questions (17), questions with left-dislocated subjects (18), object questions with subject drop (19), questions embedded under matrix, declarative verbs like sapere (know) (20), passive questions (21), and other types of appropriate sentences, like the one in (22): (17)
Chi who ‘Who ‘Who
è che sta pettinando il is that is combing the is it that is combing the child?’ is it that the child is combing?’
bambino? child
(18)
Il signore, chi sta salutando? the man who is greeting ‘The man, whom is he greeting?’
(19)
Chi pettina? who combs ‘Whom is (he) combing?’
(20)
Sai chi sta pettinando il bambino? know-2SG who is combing the child ‘Do you know who is combing the child?’ ‘Do you know whom is the child combing?’
(21)
Chi viene accarezzato dal bambino? who comes caressed by-the child ‘Who is being caressed by the child?’
(22)
Chi c’è qua dietro che il who there-is here behind that the ‘Who is there that daddy is greeting?’
papà daddy
sta is
salutando? greeting
Other types of responses, mostly occurring in the object condition, are combinations of the categories just mentioned; sentences like the ones from (23) to (25) have been counted as embedded questions; however, they include, respectively, a cleft structure with postverbal DP, a cleft structure with preverbal subject and a null subject:
Elicited production of who-questions by school-aged Italian-speaking children
(23)
(24)
(25)
Potresti could-2SG ‘Could you ‘Could you
dirmi chi è tell-me who is please tell me who please tell me who
129
che saluta il signore? that greets the man is it that greets the man?’ is it that the man greets?’
Sai chi è che il bambino sta lavando? know-2SG who is that the child is washing ‘Do you know who is it that the child is washing?’ Mi puoi dire chi sta baciando? to me could-2SG tell who is kissing ‘Could you tell me whom is (he) combing?’
Sentences like the cleft one in (26) have been classified as object questions with left-dislocated subject: (26)
Il signore, chi è che sta salutando? the man who is that is greeting ‘The man, who is it that he is greeting?’
Furthermore, interrogative sentences like the ones in (27) and (28) have been counted, respectively, under the “passive” category and the “argument drop” category, despite the fact that they are cleft structures: (27)
Chi è che viene baciato dal bambino? who is that comes kissed by-the child ‘Who is it that is being kissed by the child?’
(28)
Chi who ‘Who ‘Who
è che sta lavando? is that is washing is it that is washing?’ is it that (he) is washing?’
Finally, passive questions like the one instantiated in (29) occurred in adult productions, in the subject condition: (29)
Da chi viene inseguito il bambino? by whom comes chased the child ‘By whom is the child being chased?’
130
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
Sometimes, children gave responses that were classified as incorrect. These include production of undifferentiated forms like the one in (30), subject questions with object-drop, object questions produced instead of subject questions, (31), and other types of responses, such as (32) to (35). (30)
Chi è? who is ‘Who is it?’
(31)
Il bambino, chi sta pettinando? the child who is combing ‘The child, whom is he combing?’
(32)
Chi è che il bambino lo ba, bacia who is that the child it kis kisses ‘Who is it that the child kis, kisses that person?’
(33)
Che cosa sta facendo il what is doing the ‘What is the child doing?’
(34)
Chi è che il bambino pettina qualcuno? who is that the child combs someone ‘Whom is it that the child combs someone?’
(35)
Chi sta lavando i bambini? who is washing the children ‘Who is washing the children?’
quella that
persona? person
bambino? child
A reviewer raised the question, relevant to coding, as to how to distinguish subject from object questions when they are potentially ambiguous. What led us to claim that a question with the order Wh V DP is a subject or an object question is the fact that it was elicited in a subject or object condition. The reviewer is worried that if children made a mistake, it is not possible to detect it. Guasti et al.’s (2012) study reports some subject questions being uttered when object questions were targeted and vice versa, although no percentages are provided. Indeed, we only have evidence that our participants occasionally produced object questions (mainly via subject dislocation) instead of subject ones, as in (31) above. Since our participants are older in age and our results are comparable to those reported by Guasti et al. (2012), we believe that it is fair to
Elicited production of who-questions by school-aged Italian-speaking children
131
analyze subject and object questions produced in the relevant conditions as such, though tolerating some margin of error.5
3 Results Overall, children’s correct responses amount to 94% of the collected corpus, both for the subject and the object condition, and with no difference between age groups. The percentages of correct responses are given in Table 2. Only one occurrence of incorrect response was collected among adults. Table 2: Percentages of correct Who-questions produced across participants (SD in percentage points) CORRECT QUESTIONS
6 Y.O. 7 Y.O. 8 Y.O. 9 Y.O. Adults
SUBJECT
OBJECT
95 (10) 93 (14) 93 (16) 95 (12) 100 (0)
95 (10) 94 (12) 94 (15) 95 (9) 98 (5)
Out of the total amount of correct responses, Wh V DP questions like the ones instantiated in (16a) and (16b) represent the predominant typology of questions employed by adults, while children used a wider range of response types; this is shown in Table 3 and Table 4. The group of 9 y.o. children preferred the use of subject cleft questions to non cleft ones. Table 3: Percentages of correct typologies of questions produced across participants in the subject condition
6 Y.O. 7 Y.O. 8 Y.O. 9 Y.O. Adults
WH V DP
CLEFT
EMBEDDED
PASSIVE
OTHER
51 (47) 47 (45) 42 (43) 40 (40) 68 (43)
40 (43) 25 (37) 32 (40) 45 (40) 9 (17)
3 (3) 20 (13) 12 (3) 5 (17) 0
0 0 2 (3) 0 23 (33)
1 1 6 (18) 5 (16) 0
5 See Schouwenaars, van Hout, and Hendriks (2014), who report a low percentage of agreement errors in object which-questions (less than 4%) in Dutch-speaking children aged 6;7 to 7;10 in an experimental paradigm similar to Guasti et al. and ours.
132
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
Table 4: Percentages of correct typologies of questions produced across participants in the object condition
6 Y.O. 7 Y.O. 8 Y.O. 9 Y.O. Adults
WH V DP
SUBJ-TOPIC
CLEFT
EMBEDDED
PASSIVE
OTHER
SUBJ DROP
40 (35) 28.5 (32) 30 (35) 35 (28) 86 (15)
25 (32) 18.5 (27) 21 (30) 16 (22) 7 (12)
21 (28) 13 (28) 16 (25) 26 (30) 0
2 (3) 18 (17) 10 (10) 5 (18) 0
0 2 (10) 1 (3) 5 (15) 3 (10)
2 (3) 2 (5) 5 (15) 3 (12) 2 (10)
5 (8) 11 (22) 11 (22) 5 (18) 0
Wh V DP questions were generally preferred by the adults with respect to the children. We performed a repeated-measure logistic regression analysis (Dixon 2008; Jaeger 2008) with software R (R Core Team 2013) by setting subjects and items as random factors in a mixed logit model (Baayen 2008). We set the probability of producing a Wh V DP question rather than another type of response as our dependent variable and age group as fixed factor. As a result, we found out that Wh V DP questions were produced more often by adults than children as a whole, in both conditions (subject questions: Wald Z = 2.372, p = 0.01; object questions: Wald Z = 5.284, p < 0.001). Moreover, by setting as our independent variable the targeted type of sentence, we found out that children preferred to use Wh V DP questions in the subject condition as compared to the object condition (on average, across groups, 45% questions in the subject condition vs. 33% in the object condition; Wald Z = 6.293, p < 0.001) and with no significant differences detected across age groups. The same is true as regards cleft questions (subject vs. object cleft questions: Wald Z = 8.95, p < 0.001). When Wh V DP questions and cleft questions are taken together, we observe that both children and adults produce comparable amounts of such questions in the subject condition, while fewer of such questions are found in child production in the object condition (Wald Z = –3.607, p < 0.001). With respect to adults, whose productions display less variation, children employed a larger variety of interrogative structures as regards object-extracted questions: these concern above all the use of cleft questions, questions embedded under a declarative verb, the omission of the subject constituent, and other adequate strategies of answers. On the other hand, adults produced some object questions with leftdislocated subjects, a strategy employed by children as well. Most object cleft questions produced by children contained postverbal subjects, as in (17); however, preverbal subjects are allowed in Italian cleft questions, as (36) shows: (36)
Chi è che il bambino sta pettinando? who is that the child is combing ‘Who is it that the child is combing?’
Elicited production of who-questions by school-aged Italian-speaking children
133
Matrix object cleft questions with preverbal subject occurred 16% of times in children’s productions, out of the total amount of the matrix cleft questions collected in the object condition (21/131). The same is true for embedded cleft questions, like the one given in (24): children produced 6/15 embedded cleft questions with preverbal subjects. As a whole, as Table 5 shows, children produced a larger amount of interrogatives with lexicalized preverbal or leftdislocated subject as compared to adults, namely 25% vs. 7% (Wald Z = 2.082, p < 0.05); such finding will be of particular interest for the analysis of the results given in next section.6 Table 5: Percentages of questions with preverbal/left dislocated subject in the object condition PREVERBAL SUBJECT 6 Y.O. 7 Y.O. 8 Y.O. 9 Y.O. Adults
32 (32) 21 (27) 26 (30) 23 (27) 7 (12)
Finally, children sometimes committed false starts or rephrased their sentences, as exemplified in (37): (37) Chi è che, chi è che sta, il bambino chi sta baciando? who is that who is that is the child who is kissing ‘Who is it that, who is it that is, the child, whom is he kissing?’ Interestingly, this concerns 16% of all object questions but only 1% of subject questions. Adults’ false starts and sentence rephrasing concerns 3% of subject questions and 6% of object questions; as a whole, adults rephrased their sentences less frequently than children (Wald Z = –2.221, p < 0.05).
4 Discussion When induced to produce who-questions, Italian-speaking children aged 6 to 10 utter a high amount of accurate, simple Wh V DP and cleft interrogatives when the subject constituent is questioned. When an object constituent is extracted, a larger variety of answering strategies are employed, namely questions with left6 The reviewer who asked about coding (see section 2.3) also asked to limit the analysis to only non-ambiguous questions, by removing all questions with the order Wh V DP, in order to check whether there is any developmental change. Data reported in Table 5 do not display any significant change, although the highest percentage of preverbal or dislocated subjects concerns the youngest children.
134
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
dislocated and null subjects, questions embedded under a matrix, declarative verb, passive questions and other types of questions. With respect to children, adults present less variation as far as the object condition is concerned: they do not use cleft questions, embedded questions or other types of interrogatives, nor do they omit the subject; however, they allow the subject to be left-dislocated. When all questions containing a preverbal or a left-dislocated subject are considered, one observes that children choose them more frequently than adults. Furthermore, even though children do not produce less accurate object questions than subject questions, they produce false starts by rephrasing their sentences more often when an object interrogative element is extracted and more frequently than adults. All things considered, some kind of subject-object asymmetry seems to be present in child data, with object interrogatives with postverbal subjects sometimes being avoided in favour of structures containing preverbal subjects. It is remarkable that no significant differences among age groups emerge in our study. As is noted by Guasti et al. (2012), though, this unexpected finding somehow mirrors the pattern found in comprehension of unambiguous questions by De Vincenzi et al. (1999), who notice a significant change only from age 10 on, with object interrogatives with postverbal subject being understood far better. Our findings are very similar to Guasti et al.’s study in some respects: the authors report production of a wider set of answering strategies alternative to the target sentences in the object condition by preschool-aged children: in addition to Wh V DP questions, young children drop the DP subject or place it in a left peripheral position more frequently than adults do, while hardly ever resorting to non-target-like questions in the subject condition. Important differences are also detected: with respect to younger children, we observe an increase in the rate of accuracy in older children’s answers, a decrease in object questions with subject omission, a greater amount of cleft questions, and the emergence of embedded and passive questions, probably due to the older age of our participants. We also observe a difference in the adult pattern: our adults produce more passive questions in the subject condition, while the adults participating in Guasti et al.’s experiment favour passive questions in the object condition. Guasti et al. (2012) interpret young children’s subject-object asymmetry, which manifests itself in terms of a higher accuracy rate in subject questions and a larger variety of interrogative structures in the object condition, as the consequence of a greater level of difficulty posed by object-constituent extraction when a postverbal subject is present in the sentence, as is allowed by Italian syntax. Note that such asymmetry is not expected under Friedmann, Belletti, and Rizzi (2009) well-known account according to which children have difficulties in comprehending and producing structures where a moved element
Elicited production of who-questions by school-aged Italian-speaking children
135
and a subject intervening between the first and the last merged position of that element share a lexical restriction, as in object-extracted restrictive relative clauses with DP subjects, (38), and which-questions, (39): (38) Tocca il bambino che il signore saluta Touch the child that the man greets (39) Quali bambini saluta il signore ? Which children does the man greet ? In who-questions, the interrogative element is not lexically restricted; therefore, children should not find who-object questions particularly problematic to compute. According to Guasti et al., the subject-object asymmetry found in production in Italian-speaking children’s questions is better accounted for by taking the distinction between subject-verb (SV) and verb-subject (VS) agreement into consideration, and by conceiving the object questions produced alternatively to the target ones by young children, i.e. questions with null and dislocated subjects, as means to avoid a configuration containing a postverbal subject. The account is based on the generalization, discussed in Guasti and Rizzi (2002), that SV agreement is more robust than VS agreement crosslinguistically: in languages that possess the relevant morphology, when a DP subject occurs in a position higher than the verb, the morphological expression of agreement is compulsory; when not, languages may not express morphological agreement between the verb and the postverbal DP, and agreement is more prone to variation, that is, it is “weak”. Guasti et al. (2012) implement this theoretical notion of robustness of agreement by applying Franck et al.’s (2006) syntactic analysis of attraction to children’s performance in wh-questions. According to Franck et al. (2006), agreement consists of two sub-processes, AGREE and Spec-head. Through AGREE, number and person features of the subject in its thematic position are copied onto an “AgrS” node; then, the verb is assumed to move to AgrS to receive its morphological specification (Figure 3).
Figure 3: AGREE (Franck et al. 2006:180)
136
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
Figure 4: Spec-head agreement (Franck et al. 2006:181)
In languages displaying SV order, the subject moves out of VP to Spec AgrS, giving rise to a local Spec-head relationship (Figure 4). Crucially, Franck et al. assume that the sharing of featural values established by AGREE gets further checked in the local Spec-head configuration. Thus, rephrasing Guasti and Rizzi (2002), Franck et al. propose that the morphological manifestation of agreement is more stable when AGREE is associated with movement of the subject to Spec AgrS, because the relevant features are checked twice. In such cases, agreement manifests itself as SV agreement; superficial VS agreement, on the other hand, is realized solely by AGREE. Guasti et al. (2012) make use of such account to explain why object questions containing a postverbal subject DP may be particularly challenging: when an object constituent is extracted in a question, the object copy interferes in the AGREE relation between the subject in its thematic position and AgrS; if agreement is weak, i.e. the subject DP is postverbal, the object copy may transfer its features into AgrS without the possibility for a Spec-head agreement relation between the subject and the verb to “repair” the error; this gives rise to an attraction error (Figure 5), like the one instantiated in (40), where it is the object interrogative constituent that ultimately agrees with the verb, and not the postverbal subject:
Figure 5: Attraction (Guasti et al. 2012: 205)
Elicited production of who-questions by school-aged Italian-speaking children
(40)
137
Chi sporca gli elefanti? ‘Who is dirtying the elephants?’
Target sentence:
Chi sporcano gli elefanti? ‘Whom are the elephants dirtying?’
Crucially, Guasti et al. (2012) point out that interference is the source of difficulty for young children (and, to a lesser extent, for adults too). To avoid VS agreement configurations in wh-questions and, ultimately, attraction errors, object questions with null or dislocated subjects would be employed, because they allow for double feature checking, involving both AGREE and Spec-head relationships.7 Indeed, following Cardinaletti (2004, 2007), in object questions with null subjects (41) and left-dislocated subjects (42), the argumental subject pro occurs preverbally, and agreement is therefore strong: (41)
chi [AgrSP pros who they ‘Who do they wash?’
(42)
[TopicP
[FocusP
lavanov [vP ts tv ]]] wash
i bambini [FocusP the children ‘Who do the children wash?’
chi [AgrSP who
pros they
lavanov [vP ts tv ]]]] wash
This account can explain the pattern of responses found in young children’s productions of wh-questions disambiguated by subject-verb agreement, and it would be tempting to apply the same line of reasoning to our data as well. Indeed, a subject-object asymmetry is found in our data in both target and cleft wh-questions. In addition to object questions with null subjects or left dislocated subjects, cleft questions, embedded cleft questions and other types of questions with preverbal subjects, like the ones given in (34), (22) and (24), could be seen as attempts made to place the subject preverbally8,9. Furthermore, more false 7 The same approach has been successfully used by Volpato (2010) and Volpato and Vernice (2014) to account for the different production of object relative clauses with preverbal and postverbal subjects. 8 One may argue that the nature of subject omission in object questions, adopted by children but disallowed by adults, is pragmatic in nature: children would choose to omit the DP subject constituent because it is underinformative, i.e. it can be recovered by the experimental context (see Serratrice 2005), while adults would be more committed to the hearer. For this reason, we decided to exclude object questions with null subjects from the count; even excluding them, our children produce a higher amount of structures with preverbal subjects than adults (see Table 5). 9 For a discussion on the syntactic structure of interrogative clefts, see Belletti (2012; 2015). The type of interference taken into account in our study can apply to interrogative clefts containing preverbal subjects as well.
138
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
starts and sentence rephrasing occur when questioning the object constituent in our children’s productions, suggesting that extracting the object might be more difficult than extracting the subject also in questions where agreement does not play any role in disambiguating the syntactic functions of the interrogative element and the postverbal DP. Only one finding is in contrast with Guasti et al.’s results and expectations: the adults they tested produced passive questions in the object condition, which are interpreted as the strategy adopted by adults in order to overcome interference of the object copy, and, consequently, attraction. 3 and 4 y.o. children do not make use of this strategy because they are too young, but the authors expect older children to produce passive questions like adults. However, our school-aged children only employ 3% passive questions in the object condition. Moreover, recall that who-questions in our study were elicited together with argument relative clauses: the very same children produced a substantial amount of passive relatives instead of the targeted gap object relatives,10 which means that the passive structure is perfectly available to them as an avoidance strategy. Yet, these children only rarely resorted to the passive when induced to produce object questions. Secondly, the adults taking part in our experiment show an asymmetry going in the opposite direction: more passive questions are produced in the subject condition. These facts suggest that passive is not to be considered as an avoidance strategy of attraction in who object questions.11
5 Conclusions A production experiment run with Italian-speaking children (aged 6;3–10;4) and a control group of adults aiming at eliciting potentially ambiguous who V DP questions, i.e. questions superficially ambiguous between a subject and an object reading, reveals some differences between the subject and the object experimental conditions. Wh V DP object questions are more often replaced by 10 See Pivi, Del Puppo, and Cardinaletti (this volume), who report the data produced by the very same group of participants, “G4”, and deal with the issue concerning passive relatives used instead of gap object relatives. 11 The lower rate of passives employed in who-questions with respect to those employed in relative clauses is expected under Friedmann, Belletti, and Rizzi (2009) Relativized Minimality account. Consistently, in Guasti et al.’s study, passivization is used more frequently by adults when the target is an object question introduced by the operator which NP, which is lexically restricted, rather than who. It could be maintained that other avoidance strategies different from turning who object questions into who subject questions are favoured by children, when trying to avoid attraction phenomena: omitting or dislocating the subject is preferred.
Elicited production of who-questions by school-aged Italian-speaking children
139
other types of interrogative structures than subject questions in children. These alternative strategies share the property of containing a preverbal subject which is in Spec-head configuration with the verb rather than a postverbal subject. This phenomenon is not present in adults’ production in the same percentages (children 25% vs. adults 7%) As a whole, school-aged Italian-speaking children seem to still have some difficulties with interrogatives containing postverbal subjects. Results are compared with the findings by Guasti, Branchini, and Arosio (2012): the authors find similar results in preschool-aged children by eliciting wh-questions disambiguated by subject-verb agreement. In order to explain the subject-object asymmetry emerged in children and the divergent performance displayed by adults, an explanation in terms of strength of agreement and interference, in line with the one proposed by Guasti et al., could be applied to potentially ambiguous interrogatives as well, and still, even though to a lesser extent, to older children’s productions.
References Belletti, Adriana. 2012. Revisiting the CP of Clefts. In Ede Zimmermann & Günther Grewendorf (eds.), Discourse and Grammar: from Sentence Types to Lexical Categories, 91–114. Berlin: Mouton de Gruyter. Belletti, Adriana. 2015. The Focus Map of Clefts: Extraposition and Predication. In Ur Shlonsky (ed.), Beyond Functional Sequence. The Cartography of Syntactic Structures, (10). New York: Oxford University Press. Belletti, Adriana & Maria Teresa Guasti. 2015. The Acquisition of Italian. Morphosyntax and its interfaces in different modes of acquisition. In Roumyana Slabakova & Lydia White (ed.), Language Acquisition and Language Disorders 57. Amsterdam; Philadelphia: Benjamins Publishing Company. Cardinaletti, Anna. 2001. A second thought on Emarginazione: Destressing vs. “Right Dislocation”. In Guglielmo Cinque & Giampaolo Salvi (eds.), Current Studies in Italian Syntax. Essays Offered to Lorenzo Renzi, 117–135. Amsterdam: Elsevier. Cardinaletti, Anna. 2002. Against optional and zero clitics. Right Dislocation vs. Marginalization. Studia Linguistica 56(1). 29–57. Cardinaletti, Anna. 2004. Toward a cartography of subject positions. In Luigi Rizzi (ed.), The Structure of CP and IP. The Cartography of Syntactic Structures, (2), 115–165. New York: Oxford University Press. Cardinaletti, Anna. 2007. Subjects and wh-questions. Some new generalizations. In José Camacho, Nydia Flores-Ferrán, Liliana Sanchez, Viviane Déprez & Maria José Cabrera (eds.), Romance Linguistics 2006: Selected papers from the 36th Linguistic Symposium on Romance Languages (LSRL), 57–79. Amsterdam/Philadelphia: John Benjamins Publishing Company. De Vincenzi, Marica, Lisa Arduino, Laura Ciccarelli & Remo Jobb. 1999. Parsing Strategies in Children Comprehension of Interrogative Sentences. Proceedings of ECCS 1999, European Conference on Cognitive Science. Siena. 301–308.
140
Giorgia del Puppo, Margherita Pivi and Anna Cardinaletti
Del Puppo, Giorgia. 2016. On the acquisition of Focus: elicited production of cleft sentences and wh-questions by school-aged, Italian-speaking children. Doctoral dissertation, Ca’ Foscari University of Venice. Dixon, Peter. 2008. Measures of accuracy in repeated-measures designs. Journal of Memory and Language 59. 447–496. Franck, Julie, Glenda Lassi, Ulrich H. Frauenfelder & Luigi Rizzi. 2006. Agreement and movement: a syntactic analysis of attraction. Cognition 101. 173–216. Friedmann, Naama, Adriana Belletti & Luigi Rizzi. 2009. Relativized relatives: Types of intervention in the acquisition of A-bar dependencies. Lingua 119. 67–88. Guasti, Maria Teresa. 1996a. Acquisition of Italian interrogatives. In Harald Clahsen (ed.), Generative Perspectives on Language Acquisition, 241–269. Amsterdam: John Benjamins. Guasti, Maria Teresa. 2000. An excursion into interrogatives in Early English and Italian. In MarcAriel Friedeman & Luigi Rizzi (ed.), The Acquisition of Syntax, 105–128. Harlow: Longman. Guasti, Maria Teresa, Chiara Branchini & Fabrizio Arosio. 2012. Interference in the production of Italian subject and object wh-questions, Applied Psycholinguistics 33. 185–223. Guasti, Maria Teresa & Luigi Rizzi. 2002. Agreement and tense as distinct syntactic positions: Evidence from acquisition. In Guglielmo Cinque (ed.), The structure of DP and IP. The cartography of syntactic structures 1. New York: Oxford University Press. Jaeger, Tim Florian. 2008. Categorical data analysis: Away from ANOVAs (transformation or not) and towards logit mixed models. Journal of Memory and Language 59. 434–446. R Core Team. 2013. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/. Read, Charles, A. Kraak & Loe Boves. 1980. The interpretation of ambiguous who-questions in Dutch: the effect of intonation. In Wim Zonneveldt and Fred Weerman (eds.), Linguistics in the Netherlands, 389–410. Dordrecht: Foris. Rizzi, Luigi. 1986. Null Objects in Italian and the Theory of pro. Linguistic Inquiry 17. 501–557. Serratrice, Ludovica. 2005. The role of discourse pragmatics in the acquisition of subjects in Italian. Applied Psycholinguistics 26. 437–462. Schouwenaars, Atty, Angeliek van Hout & Petra Hendriks. 2014. Word Order Overrules Number Agreement: Dutch Children’s Interpretation and Production of Which-questions. In ChiaYing Chu, Caitlin E. Coughlin, Beatriz Lopez Prego, Utako Minai & Annie Tremblay (eds), Selected Proceedings of the 5th Conference on Generative Approaches to Language Acquisition North America 2012, 60–81. Somerville, MA: Cascadilla Proceedings Project. Volpato, Francesca. 2010. The acquisition of relative clauses and phi-features in hearing and hearing-impaired populations. Ca’ Foscari University of Venice dissertation. Volpato, Francesca & Mirta Vernice. 2014. The production of relative clauses by Italian hearing and hearing-impaired children. Lingua 139. 39–67.
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
6 Mood interpretation in Spanish: towards an encompassing view of L1 and L2 interface variability* Abstract: In this chapter we contribute data related to the ability of advanced adult L1 French and L1 English learners of L2 Spanish, as well as a native European Spanish group, to interpret effects on sentence meaning of indicative/ subjunctive mood contrasts in concessive clauses. This is measured through two different tasks. Firstly, a 10-item multiple-choice interpretation task completed by L1 French Speakers (n = 49) and L1 English speakers (n = 40) at B2 and C1 CEFR proficiency levels, and by an L1 Spanish control group (n = 35). Secondly, a task administered only to L1 Spanish speakers (n = 61), consisting of comparing and describing the interpretation of two alternative versions of a group of concessive utterances (adapted from the previous task), varying only in the verbal mood (indicative vs. subjunctive). Two versions of the second task were used, each with 7 different items (5 concessives and 2 distractors). Our results show that integrating linguistic and discourse-pragmatic information is problematic in L2 Spanish, regardless of the L1 and proficiency level. It is also problematic for native speakers; in the items tested variation was found in the L1 speakers’ responses. However, our results suggest that the causes of variability found in the L1 are different from those causing it in L2 speakers. In the case of natives, our data suggests that this variability is brought about by a richer and more diverse array of contextual assumptions accessible for online interpretative processes, as compared to the L2 speakers. Conversely, in L2, it is mainly due to incomplete acquisition of mood as a grammatical feature.
* The research presented in this chapter has been developed within the project Semántica Procedimental y Contenido Explícito III (SPYCE III), financed by the Spanish Ministry for Economy and Competitivity, ref. FFI2012-31785. We thank the three anonymous reviewers who collaborated with the volume editors for their invaluable suggestions; all disclaimers hold. Aoife Ahern and José Amenós-Pons, Universidad Complutense de Madrid Pedro Guijarro-Fuentes, University of the Balearic Islands
142
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
1 Introduction Within the generative framework (Chomsky 2007), grammar consists of the computational system or narrow syntax, the output of which feeds into phonetic form (PF) and logical form (LF) at the interfaces between these levels. With this framework in mind and within generative grammar, a number of hypotheses have emerged, such as the Interface Hypothesis (IH henceforth), which claims that features at the syntax-semantics/pragmatic interface (“interpretable” features) are less accessible than purely narrow syntactic features (“non-interpretable” features). Particularly, recent research into first and second language acquisition has concentrated on how different modules (i.e., syntax, semantics, pragmatics, morphology, phonology) of the grammar intersect during different language acquisition scenarios. Results from some previous studies, which have focused on the acquisition of overt vs. null pronominal subjects, copula verbs and object expression by different learners acquiring quite a wide range of languages, have shown that structures involving interfaces are more difficult to acquire, and more vulnerable in several types of processes, than structures involving only narrow syntax. These studies cover acquisition processes including those of second languages, simultaneous bilingualism and impaired acquisition (e.g., Hopp 2004; Lozano 2006a, 2006b; Montrul and Rodriguez-Louro 2006; Serratrice, Sorace and Paoli 2004; Sorace 2004; Sorace and Filiaci 2006; Tsimpli, Sorace, Heycock and Filiaci 2004 to name a few); where difficulty can be interpreted as failure of acquisition and/or incomplete acquisition (Montrul 2004; Montrul and Polinsky 2011). One way of explaining this failure is by suggesting that at the syntax/discourse or pragmatic level the learner may be exposed to apparently interchangeable sentences in the input, but with subtle contextual differences. Delays in acquisition or permanent optionality may occur due to ‘weaker’ input for discourse related properties, or to the learners’ processing difficulties in integrating information pertaining to different domains. All this may yield the construction of interlanguage (IL) grammars that may diverge significantly, in interface areas, from native grammars. In contrast to the studies above, some others have revealed that native-like grammars can be attained, even in areas that rely heavily on context for interpretation and involve the interfaces (Borgonovo, Bruhn de Garavito and Prévost 2005, 2006; Borgonovo, Bruhn de Garavito, Guijarro-Fuentes, Prévost and Valenzuela 2006; Bruhn de Garavito and Valenzuela 2008, among others). These studies concluded that interface phenomena are acquirable despite their processing complexity, and L2 learners are capable of overcoming the difficulties posed by heavy reliance on context in the interpretation of different kinds of grammatical elements.
Mood interpretation in Spanish
143
In sum, in the quest for understanding of the nature of persistent variation of certain forms in second language grammars, questions continue to be raised of whether or not such selective difficulties are attributable to parametric differences between the first and second language, and whether or not they are permanent and destined to fossilization because of their processing complexity (Clahsen and Felser 2006, among others). Interface phenomena may be more vulnerable and inherently more difficult to acquire because of issues related to processing, rather than to grammatical representation: full processing is assumed to be available to very advanced proficient learners, and native and very advanced non-natives will follow the path of least processing effort. On the other hand, it can be assumed that metalinguistic awareness enables learners to acquire the ability to interpret syntax-semantic-pragmatic interface phenomena. Thus, one of the hypotheses brought up in the present chapter is that competence mismatches between native and non-native grammars may be attributed to differences in the degree of metalinguistic awareness among speakers in each group. To shed more light on these issues, this chapter focuses on the L2 acquisition in adult learners of Spanish, compared to a group of native speakers, of the distributional and interpretational features of the Spanish indicative/subjunctive mood contrast, which involves a multiple (i.e., the syntax-morphology-semanticspragmatics) interface. Furthermore, the chapter reports data related both to L2 learners’ interpretation of mood contrasts in Spanish, and to L1 speakers’ metalinguistic awareness of the effects of mood alternation and of the role of context in its interpretation. The chapter is organized as follows. In Section 2, we present a brief description of mood alternation in concessive clauses in Spanish, French and English; we also develop our background assumptions on the semantics and pragmatics of mood interpretation. In Section 3, after a brief review of previous studies on L2 mood acquisition, we introduce our research questions and hypotheses. In Section 4, we describe the study design and present our results. Finally, section 5 contains the discussion and conclusions.
2 Theoretical considerations 2.1 Mood in Spanish The indicative/subjunctive mood contrast in Spanish can occur in both main and subordinate clauses. However, the presence of the subjunctive in main clauses is strictly limited to the expression of exhortative and desiderative illo-
144
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
cutionary force or, optionally, found in clauses introduced by possibility connectives such as quizá or tal vez (maybe, perhaps). On the other hand, indicative and subjunctive contrast in a wide range of effects in subordinate clauses. In argument clauses, a number of high-frequency predicates select for the subjunctive, such as predicates of volition, desire, causation and obligation, negative epistemics like dudar and negar (doubt, refuse), and factive-emotive predicates like alegrar, sorprender (be glad that, surprise). A few other predicates, such as those known as verba dicendi, predicates of speaking like decir, gritar, indicar (say, shout, indicate) allow alternation of indicative and subjunctive in their argument clauses, expressing thereby that the argument clause is assertive, with indicative, or that it expresses directive force, with subjunctive. As for mood in adverbials, the subjunctive may be selected by the subordinating element, or else alternate with the indicative, in which case each mood leads to certain effects on the interpretation that vary depending on the type of adverbial, as well as other factors, including sentence and discourse context, as will be explained below. Thus, conjunctions such as para que, a menos que, antes de que (so that, unless, before) select for the subjunctive; whereas conjunctions like dado que, puesto que and porque (given that, since, because) select the indicative. What can be observed as a basic generalisation from the predicates (with the exception of factive-emotives) and conjunctions mentioned so far is that those which select for subjunctive can be characterised semantically in that they introduce propositions referring to potential states of affairs, traditionally considered in terms of irrealis modality. Factive-emotive is the label given to predicates, like those exemplified above, that describe an emotional reaction to a state of affairs, thereby creating a factive environment in which the argument clause is interpreted, that is, their semantics can be characterised in that they introduce presupposed propositions (Kiparsky and Kiparsky 1971): (1)
Me asombra que lo haya Me amazes that it has-SUBJ I’m amazed s/he managed it.
conseguido. managed.
In this case, the predicate asombrar denotes a reaction to the situation described in the argument clause, and this situation is presented (by virtue of being reacted to) as a state of affairs that already forms part of the common ground. This very brief overview of indicative/subjunctive mood uses in Spanish shows what have often been considered to be the two basic kinds of semantic triggers of the subjunctive: on one hand, expressions that entail the potentiality
Mood interpretation in Spanish
145
or possibility of the proposition that they introduce; and on the other hand, as in the case of factive-emotive predicates, presuppositional environments (Ahern and Leonetti 2004; Ahern 2004, 2006). Still, there is no overall one-to-one relationship between the type of interpretation that the subjunctive receives, and the type of linguistic environment; in this sense, mood integrates semantic, syntactic and pragmatic features. As will be seen in detail, certain types of linguistic environment allow both kinds of interpretation (potential and presuppositional), while discourse or extralinguistic contextual conditions determine which is more adequate. In the present study, we focus on one of those environments, namely, although (aunque) concessive clauses, in which the subordinate adjunct clause in subjunctive can obtain either of these two types of readings. Thus, interpreting mood in concessives is clearly an interface matter, in which semantic, syntactic and pragmatic factors must be taken into account.
2.2 The pragmatics of mood interpretation Given the apparently diverse nature of semantic and syntactic properties that are involved in mood selection and in mood alternation in Spanish, it has often been concluded that the subjunctive must be understood as a grammatical element that is not amenable to any unitary semantic or syntactic account (Fábregas 2014). However, it has also been proposed that by adopting a pragmatic perspective which provides a systematic model of the integration of contextual assumptions, from both discourse and extralinguistic information, it is possible to develop an analysis of the semantics of the subjunctive that is valid for its diverse range of uses (Quer 2001; Ahern 2004, 2006). This is the perspective that will be taken here. We will consider the semantics of the subjunctive from the point of view of Sperber and Wilson’s (1986/1995) Relevance Theory (RT). In the model of ostensiveinferential verbal communication sustained in RT, inferential processes, triggered by the identification of the speaker’s communicative intention, determine to a great degree not only the implicit but also the explicit content conveyed in any given utterance. This perspective has led to the development of a distinction between two types of linguistic encoding: conceptual and procedural (Blakemore 1986; Wilson and Sperber 1993; Escandell, Leonetti and Ahern 2011). As opposed to expressions which refer to concepts, procedural expressions encode instructions that guide the inferential processes of utterance interpretation, constraining the identification of meaning either at the explicit level, as is the case for connectives such as so (Blakemore 1986; Escandell and Leonetti 2011), or in the
146
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
identification of what is being implicitly communicated. In opposition to expressions that encode conceptual meaning, which can undergo diverse forms of modification – syntactic, semantic and pragmatic – and adaptation in their interpretation, procedural expressions encode computations over conceptual representations, and these computational instructions are not modified but remain consistent across interpretations. Therefore, procedural expressions serve the function of guiding the inferential processes of utterance interpretation, in which, according to RT, the explicitly communicated meaning of an utterance is identified through both decoding and inference. Escandell-Vidal and Leonetti (2011: 84) claim that “a strong connection can be established between the ‘lexical/functional’ distinction in grammatical theory and the conceptual/procedural distinction”. Lexical categories can encode conceptual and/or procedural meaning, but functional categories are always procedural. The term ‘procedural’ can be understood, within this framework, “as a shorthand term to refer to functional categories that encode interpretive instructions” (Escandell-Vidal and Leonetti 2011: 85). Tense and mood, as functional categories, encode procedural meaning. Tense imposes restrictions on the determination of temporal reference, as well as on the inferential processing of conceptual representations of states and events (Nicolle 1997, Moeschler 1998, De Saussure 2003). Mood encodes meaning that constrains inferences on speaker attitude, intention and mental state information (Rouchota 1994; Jary 2009; Ahern 2004, 2006, 2010). Finally, it is also essential to take into account the RT notion of context, due to its decisive role in utterance interpretation at every level, from the process of decoding through to every kind of inference. Within RT, context is a cognitive construct, consisting of the entire set of assumptions available to the individual at any given moment, and thereby including information from a variety of sources including discourse, i.e., utterances used previously in the communicative situation, but also memory, sensory perception, scripts and schemata, even imagination. Thus, contextual assumptions are accessed in order to assign referents to referential expressions, enrich or loosen the denotation of expressions with very general or specific meanings, respectively, and disambiguate the meaning of any polysemic or homonymous terms.
2.3 Mood alternation in concessive clauses in Spanish Spanish concessive clauses, most frequently introduced by the conjunction aunque (although), admit both indicative and subjunctive mood. Of course, each mood is associated with certain effects in the interpretation: the indicative identifies the content as adding to the common ground, while the subjunctive can be
Mood interpretation in Spanish
147
understood either as marking the content as hypothetical, or as given information. Thus, in the classic example: (2)
Aunque llueve, salgo de paseo Although rains-IND, go-out of walk Although it rains, I go out for a walk.
The use of the indicative is only appropriate if the proposition “it’s raining” has not already been introduced in the discourse context – in other words, if this proposition can be understood as adding new information to the common ground. Otherwise the utterance would be pragmatically odd. If the subjunctive is used, however, there are two possible implications that can arise: (3)
Aunque llueva, salgo de paseo Although rains-SUBJ, go-out of walk Although it rains, I go out for a walk.
If the proposition ‘it rains’ does not form part of the common ground, i.e., it has not already been asserted or is not perceivable to the speaker and addressee and therefore identifiable as given information, the subjunctive leads to a possibility reading, similar to “it might rain”. If, on the contrary, the rain is already part of the common ground, the subjunctive has the effect of marking the concessive clause content as given, and thereby, foregrounding or highlighting the information expressed in the main clause. Concessive clauses, therefore, constitute a case in which mood alternation correlates with as the speaker’s propositional attitude with regard to the informative value of the proposition: indicative or subjunctive is chosen depending on whether the propositional content is presented as new or given information, or as hypothetical. For both of the latter two options, the subjunctive is used, and whether the specific interpretation is that of given information or hypothetical is determined by the context as just shown.
2.4 Verbal mood in French and English Since the L2 interpretation task was used to test native French and English speakers acquiring Spanish, a few words regarding mood in these two languages are provided here in order to offer a rough sketch of what may be considered as the starting point for these learners’ acquisition of the Spanish verbal mood system.
148
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
French, like Spanish, has morphological mood, and the systems of both languages are analogous in many respects: potential and presuppositional readings of the subjunctive are found in French (Rhis 2013), as well as in Spanish. Mood alternation is possible in French, in syntactic environments that partially coincide with Spanish (e.g. after certain verbs expressing psychological perception, and after negated epistemic verbs). However, unlike in Spanish, in contemporary French the choice of subjunctive tenses is reduced to present and perfect forms. Imperfect and pluperfect subjunctive are commonly used in Spanish, but not in French. Additionally, environments accepting mood alternation are more restricted in French than in Spanish. In concessive clauses, it has frequently been claimed that in French, the use of indicative or subjunctive forms depends mainly on the connective being used (Riegel et al., 1994): connectives like bien que or quoique are usually followed by subjunctive forms, while meme si is always followed by indicative forms. Traditional translation rules in contrastive grammar books recommend translating aunque + indicative as bien que + subjunctive, and aunque + subjunctive as même si + indicative (Bedel 1997; Gerboin and Leroy 1991). This rule is based on two partially inaccurate assumptions: (1) bien que always introduces factual content, while même si presents hypothetical content; (2) in Spanish, the indicative expresses factual content, while the subjunctive always communicates hypothetical content. In fact, although the subjunctive is indeed more frequent after bien que, the indicative is also found, in oral as well as in written use (Ballestero 2010). It has also been proven that, in practice, professional translators do not systematically follow the rule recommended by contrastive grammarians (Ballestero 2010: 337– 356). Thus, there is no direct mapping between Spanish and French in terms of the conditions for mood alternation. When learning Spanish, L1 French speakers will have to learn not only which mood is allowed by each connective in the L2, but also what type(s) of reading each mood obtains. In English, on the other hand, the subjunctive is not encoded in its own specific morphological forms, but rather it is distinguished from the present indicative in the contrast between the inflected form, as indicative, and the base form of the verb, as subjunctive, also known as the bare infinitive. This contrast, however, is only distinguishable in limited contexts (e.g. between the inflected 3rd person singular form and the base form) and the uses of the subjunctive are very limited in English. They occur only in formal registers (and especially in American English, being limited in British English to archaic, legalistic style), in argument clauses under predicates of recommendation, resolution, demand, surprise, etc. (Quirk et al. 1972: 66), or in formulaic expressions, such as If I were you / So be it! / Suffice it to say that. . .
Mood interpretation in Spanish
149
However, in rare cases, limited, as mentioned above, to formal registers of American English (Quirk et al. 1972: 639), clauses of concession in English may also contain a verb in the subjunctive mood to express non-factual meaning, such as in the rather forced-sounding example given by these authors, Though he be the president himself, he shall hear us. Thus, it may be taken into consideration that the subjunctive exists in English, but only in a marginal set of uses limited to particular, sparsely-used registers and language varieties. Any morpho-phonological distinctions between subjunctive and indicative forms are partial, and over time, the nuances that had been expressed by the subjunctive have come to be transmitted by way of modal auxiliaries (Burrow and TurvillePetre 2005: 50). In this sense, the degree to which any previous knowledge of the subjunctive in L1 English might impact upon the acquisition of Spanish subjunctive by L1 English-speaking learners is probably small. The English equivalents of the semantic functions of Spanish mood are much more frequently covered by modal auxiliaries, which cannot be discussed here due to space limitations.
3 Interpretation of mood and language acquisition 3.1 Previous studies In an attempt to refine research questions and hypotheses in SLA research, particular attention has been paid recently to interfaces between syntax and other modules of grammar (Borgonovo and Prévost 2003; Borgonovo, Bruhn de Garavito and Prévost 2005, 2006; Iverson, Kempchinsky and Rothman 2008, to name a few). It has been suggested that the contrast between indicative and subjunctive pose a problem in different learning scenarios, including child L1 acquisition (PérezLeroux 1998) and adult L2 acquisition. In line with the present study, Ahern, Amenós-Pons and Guijarro-Fuentes (2014) discuss the results of a cross-sectional study comparing English and French L2 learners of Spanish (40 and 48 learners respectively) at different proficiency levels testing their interpretation of mood choice in if-conditional constructions (conditional sentences containing both regular and irregular indicative and subjunctive morphology forms). Regardless of the learners’ different L1 backgrounds, these learner groups performed similarly in the written linguistic task. The results lead the authors to argue that feature re-assembly in L2 Spanish is not trouble-free, and simple feature accounts (i.e., Interpretability Hypothesis
150
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
(Hawkins and Hattori 2006; Tsimpli and Dimitrakopoulou 2007)) would advocate, even though those interpretation features exist in the learners’ L1 grammars. In the light of previous findings, we consider that there is a need to investigate different interface phenomena using a precise analysis of what is entailed by the various interface properties. Our study contributes to this endeavour by considering a linguistic phenomenon involving the multiple syntax/morphology/ semantics/pragmatics interface, the Spanish mood distinction in English and French adult learners of Spanish of different proficiency levels.
3.2 Research questions and hypotheses We focus on although concessive clauses because they constitute a particularly complex instance of mood alternation, where discourse and contextual hypotheses play a major role: the pragmatic enrichment of the linguistic expression may lead to remarkably divergent interpretations of the subjunctive form. Firstly, we offer results of research into the L2 acquisition of Spanish mood selection in concessive clauses by adult French and English L1 speakers. These L2 Spanish learners, as well as an L1 peninsular Spanish control group, completed a written multiple-choice interpretation task including items with concessive clauses in indicative and in subjunctive, as explained in detail below. Later, we go on to report a follow-on study involving a metalinguistic judgement task with L1 Spanish speakers, in which the same items with concessive clauses from the previous task were included. Comparing L1 French and L1 English learners will provide not only empirical conclusions, but also information of theoretical import, as to how (and if) two groups of learners, whose starting point is typologically different, manage to acquire features that are either present or absent in their native language. The research questions we intend to address here are: Do French and English L2 learners of Spanish differ in their interpretation of native-like and non-native-like uses of Spanish although-concessive constructions? Are there group and individual differences among the L2 learners with two different proficiency levels (i.e., advanced vs. upper intermediate) in the conditions underlying the mood selection interpretations?
As sketched out in previous sections, the interpretation of mood alternation in Spanish involves factors pertaining to morphological, syntactic, semantic and pragmatic domains. Research into its acquisition, therefore, involves the analysis of the processes of development of these varied aspects of knowledge of the L2.
Mood interpretation in Spanish
151
From the point of view of the acquisition of L2 syntax, we address the question of whether interpretable features that are not present in the L1 are acquirable. If evidence can be provided that L1 English learners of Spanish fully develop the ability to interpret mood, it would favour the hypothesis that interpretable features can be acquired, since indicative-subjunctive mood alternation is very marginal in the grammar of English. Thus, we address this issue by contrasting results obtained by the English L1 group with those of the L1 French group, and compare the behaviour of each L1 group in finding the most adequate interpretation of mood in concessive clauses. On the other hand, this study also addresses questions related to the Interface Hypothesis (IH) (Sorace 2011; Montrul 2011). Despite the fact that the IH was formulated in relation to simultaneous bilingual acquisition and attrition processes, its claims are also relevant here inasmuch as the hypothesis may have implications for L2 learners, such as that the complexity of interface phenomena may cause fossilization or interlanguage variability in adult L2 acquisition. The interpretation of mood alternation in concessive clauses requires L2 learners to take into account both linguistic and extralinguistic factors, and would therefore be expected to constitute a locus for variability in both the English and the French L1 groups. In a similar vein, according to the Shallow Structure Hypothesis (Clahsen and Felser 2006), L2 learners tend to focus their attention and processing efforts on lexico-semantic information, rather than on information that is encoded through morphosyntactic means. This hypothesis is similar to the ideas put forth by authors such as VanPatten (2007) regarding competition between the requirement of processing meaning versus form in L2 learners due to the limitations of the developing interlanguage. For the present study, an implication of these hypotheses would be that in interpreting mood alternation, both of the learner groups would face considerable challenges due to the limited saliency of mood in terms of its morpho-phonological form; and also to the cognitive demands of the task, since it requires not only decoding the forms but also integrating the decoded information obtained into inferential processes. Thus, in addressing the issues considered in both the IH and the SSH by involving a task that requires interpreting a grammatically encoded mark affecting the interpretation of both explicit and implicitly communicated content, the present study may provide evidence of how L2 learners differ in their approach to processing the utterances as opposed to the L1 control group. In relation to the performance of the non-native groups in the multiplechoice interpretation task, general difficulty was expected in identifying effects of mood alternation. However, an advantage for L1 French speakers was hypothesized, thanks to similarities between their overall mood system and that of
152
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
Spanish. Therefore, an L1 influence was expected in the ability to associate mood with certain interpretative effects in concessives: presuppositional and irrealis (subjunctive) versus assertive (indicative) readings. The second task discussed herein involves probing L1 Spanish speakers’ metalinguistic representations of mood alternation, in direct relation to the items that made up the aforementioned task confronted by the L2 learner groups. Thus, in this task, further data is obtained in relation to the question of differentiating the processing of mood alternation by L1 speakers, as opposed to L2 learners of Spanish. Our hypotheses in relation to this second task include the possibility that L1 Spanish speakers will be able to verbalise their metalinguistic awareness of how mood alternation in concessive clauses affects what is communicated implicitly. This is addressed by analysing the introspections of the participants in relation to the effects of mood alternation in a given set of utterances. The introspective statements elicited in this task are intended to offer evidence of L1 speakers’ metalinguistic awareness of the semantics of mood. In addition, they should shed light on factors underlying the variability in mood interpretation which, as discussed below, in some cases was equal to, or even more prevalent than, the variability shown by the L2 groups in their interpretations of mood alternation effects in concessive clauses.
4 Methodology In the present chapter we report on the results of two tasks: firstly, two groups of L2 Spanish learners, as well as a control group of European Spanish L1 participants, were asked to complete an interpretation task containing twenty randomised if-conditional and ten although-concessive multiple choice items (henceforth this task will be referred to as interpretation task). The if-conditional items were used as distractors; they also constituted a separate piece of research which will not be dealt with in this chapter, due to lack of space. The items were based on contextualised utterances adapted from a press corpus (Corpus de Referencia del Español Actual). For every item, three answer options were given, each describing a different type of interpretation (irrealis, presuppositional or other/distractor). The items were pre-tested in a small scale pilot study with 10 native speakers of peninsular Spanish. The linguistic task was preceded by an ethno-linguistic questionnaire made up of written questions related to the subjects’ language learning experiences and about the approximate length of time they had studied Spanish. No strict time limitation was imposed on the participants, but they were advised not to
Mood interpretation in Spanish
153
take more than forty minutes to complete the task; the amount of time used was voluntarily reported by each participant. Online and paper versions of the material were used (subjects chose the most suitable format). The second task consisted of a written, online metalinguistic description task, administered to a group of L1 peninsular Spanish-speakers (henceforth, metalinguistic task). This task, like the previous one, was preceded by an ethnolinguistic questionnaire including questions regarding age group (under 30 / 30–60 / over 60), gender and language background. It was administered electronically, by way of an online Internet platform (Adobe FormsCentral). The linguistic task was made up of items taken from the interpretation task just described, which had been completed by groups of L2 Spanish learners: brief texts including sentences with concessive and conditional clauses (used as distractors) followed by 3 choices paraphrasing different interpretations. As for the previous task, there was no strict time limit although it was recommended to spend no more than 30 minutes. A total of 14 items were included across two versions of the task, so that each version required participants to read 7 items and respond to 3 open questions about each of these items. The responses given were later classified as described in section 4.3 below, in terms of types of interpretations of mood alternation in concessive clauses. Thus, the methodology in the second task is (unlike in the first task) partially qualitative, since the results were firstly ascribed to a set of categories and, afterwards, quantified within each category.
4.1 Participants The linguistic task and questionnaire were completed by adult learners of L2 Spanish at a formal, relatively homogenous, non-immersion setting: the Instituto Cervantes1, in France or in one of the following countries: the UK, Ireland, the USA or Australia. Both hard copies on paper, and online versions of the material were used (the participants’ institutions chose the most suitable format). A control group of native speakers (n = 35) of peninsular (European) Spanish also completed the same task as the groups of learners. Upon reception of the completed tasks, respective groups of L1 French (n = 48) and of L1 English (n = 40) questionnaires were formed; all the subjects 1 The language courses offered at Instituto Cervantes centers around the world are primarily built on an extremely thorough, single, CEFR-based curriculum, which is subsequently adapted to different areas, following cross-linguistic and cross-cultural criteria. Students are required to pass a placement test before enrolling; exams based on the institutional curriculum are performed at the end of each course.
154
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
had been independently assessed for language proficiency and placed by the Cervantes Institute at CEFR B2 level (L1 French, n = 16 / L1 English, n = 26) or C1/C1+ level (L1 French, n = 32 / L1 English, n = 14). As for the metalinguistic task, the participants consisted of a total of 61 monolingual peninsular Spanish speakers, 32 of which completed one of the two versions, test 1, while 29 completed the second version, test 2.
4.2 Linguistic tasks 4.2.1 Interpretation task The interpretation task administered to the L2 learner groups consisted of a set of 30 randomized, multiple choice items; 20 of the items focused on the interpretation of if-conditionals, and 10 on although-concessive clauses. In this chapter, we focus only in the latter group. Each item contained a brief text, which made up a contextualization, plus the conditional or concessive clause integrated into that text. The text fragments were followed by 3 very brief explanations or paraphrases related to the conditional or concessive clause meaning, from which the participants were required to choose the best option. The following is one example of these items: 25. (En un foro de Internet) Creo que estoy enamorado, aunque ella me ignora y no me hace caso. ¿Qué hago? El autor de este mensaje. . . a) cree que hay una posibilidad de que su amada le ignore ( ) b) explica que la amada no le presta atención ( ) c) expresa que está muy preocupado ( ) 25. (In an Internet forum) I think I’m in love, although she ignores me and pays me no attention. What should I do? The author of this message. . . a) thinks it’s possible that the one he loves ignores him ( ) b) explains that the person he loves pays him no attention ( ) c) expresses that he is very worried ( ) Table 1 shows the distribution of the concessive items in the linguistic task, according to the tense, mood and interpretation of the subjunctive clauses, including an example of each item type.
Mood interpretation in Spanish
155
Table 1: Concessive items (interpretation task) Items 21 and 22. Aunque + Present Subjunctive (irrealis reading expected) No podemos tener un perro en casa, aunque Juan me lo pida de rodillas. We can’t have a dog in the house, even if Juan begs me on his knees. Items 24 and 25. Aunque + Present Indicative Creo que estoy enamorado, aunque ella me ignora. I think I’m in love, although she ignores me. Items 27 and 28. Aunque + Imperfect Indicative Iba acumulando deudas, aunque él manifestaba que todo se iba a solucionar. He went on accumulating debts, although he claimed there would be a solution for everything. Items 23 and 29. Aunque + Perfect Subjunctive (presuppositional reading expected) Aunque haya nacido en Nueva York, mi hijo tendrá un padrino muy flamenco. Although he was born in New York, my son will have a very “flamenco” godfather. Items 26 and 30. Aunque + Imperfect Subjunctive (irrealis reading expected) La junta electoral ordenó la admisión de todos los candidatos, aunque a algunos les faltara la documentación. The electoral committee ordered the admittance of all the candidates, although some were missing documents.
4.2.2 Metalinguistic task For the task that was administered only to L1 European Spanish speakers, a subset of the items from the interpretation task just described was used: the items with concessive clauses that had been used in the tasks completed by the L2 learner groups, in addition to the distractors, made up of conditional items. As shown below in Figure 1, in the metalinguistic task, the participants were asked to compare two different versions of each utterance, varying only in the verbal mood of the concessive or conditional clause: the first with indicative and the second with the subjunctive mood. Each item was preceded by the question – translated from Spanish here for convenience – “What are the differences between (a) and (b)?” followed by two versions, (a) and (b), of the item. Next, two further, more specific questions were asked: 2) Do (a) and (b) have the same meaning? Explain what each means. 3) Briefly describe a situation in which you could use each of the options (a) and (b).
156
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
Figure 1: An item from the metalinguistic task
The task was administered in two versions, each with a different randomized set of 5 concessive and 2 conditional items, the responses of which were combined for analysis of the data.
4.3 Data analysis and results 4.3.1 Analysis of data from the interpretation task For each L1 and proficiency level, individual results in the interpretation task were analyzed. Item results were considered separately, then arranged into groups based on shared tense-mood combinations and clause types (see Table 1 above for the distribution of concessive items). The mean and standard deviation for each L1 group were also obtained and compared. T-tests (L1 French and L1 English), contingency tests, chi-square tests and one-way ANOVAs between groups (L1 French, L1 English and Control) were performed. However, due to space limitations, only the most relevant results will be analyzed and discussed herein.
Mood interpretation in Spanish
157
Figure 2: Results by item and L1
Figure 3: Relation between proficiency level and results
Figure 2 shows the results per item and L1. Figure 3 displays the percentage of learners who achieved higher and lower success in completing the task within each proficiency group (B2 and C1/C1+)2, measured in the number of correct answers. This classification, following Classical Test Theory (Allen and Yen, 2002), was developed by ordering all of the participants from those obtaining
2 As mentioned in 3.1, the participants had completed general proficiency tests at the Cervantes Institutes where they studied and were assessed to be at either B2, C1 or C1+ levels. Figure 2 shows the distribution of individuals in the upper and lower groups of results in the mood interpretation tasks in terms of their placements across the general proficiency levels.
158
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
Figure 4: Answer options chosen, expressed in percentages (items 21–25)
Figure 5: Answer options chosen, expressed in percentages (items 26–30)
highest to lowest scores, dividing them into 3 equally-sized groups and using the upper third as “higher success group”, and the lower third as “lower success group” for each of the L1 groups. Finally, Figures 4 and 5 exhibit the answer options (a, b or c) chosen by the members of each L1 group, for each item. In Figure 2, the overall success rate of the learners (assessed through the amount of correct answers) is shown, grouped by L1, regardless of general proficiency level. In the groups of items, the difference between the three groups (L1 French, L1 English and control), in relation to their L1, is only significant in item 28 (in the Pearson chi-square test, p = 0.023, n = 125). In that item, the L1 French group and the control group clearly outperform the L1 English group. However, this is an isolated case: it cannot be said, overall, that one group significantly outperforms the other. Therefore, no systematic difference was found, in relation to their L1, among the learners. Neither does the control group outperform the L2 groups. In fact, the opposite is true for items 21, 25 and 30: in those items, there is a significant difference (p < 0.05, n = 125) between the relatively low performances of the control group and those of the L2 group having obtained the best results (L1 English for items 21 and 30, L1 French for item 25).
Mood interpretation in Spanish
159
According to Figure 3, the degree of success in the task does not correlate with overall proficiency level. For both L1 groups, there is a higher percentage of B2-level learners in the lower success group. Still, the results are not clearcut for any of the L1 groups: overall proficiency level shows no systematic correlation with task performance. This is why, in Figure 2, we have chosen to represent the learners of each L1 group as a single entity. In Figure 2, it can be observed that variability – i.e., choice of an unexpected interpretation option – is found in all groups, including the control group. Figure 4 and Figure 5 complement the data from Figure 2, as far as the differences between the L2 groups and the native speaker group are concerned. In item 21, where the controls are outperformed by the learners, the difference between the options chosen in the groups is statistically significant (in the chi-square test, p = 0.013, n = 125). The item reads as follows: 21. (Conversación entre una madre y sus hijos que quieren una mascota) Niños: “Juan necesita que le cojamos un cachorrito, los tiene para regalar ya. . .” Madre: “No podemos tener un perro en casa, aunque Juan me lo pida de rodillas.” Por la respuesta de la madre, entendemos que. . . a) Juan está pidiéndolo de rodillas b) no tiene importancia si Juan lo pide o no lo pide c) realmente Juan lo va a pedir 21. (Conversation between a mother and their children about a pet they want) Children: “Juan needs us to take a little puppy, he has them ready to give away now. . .” Mother: “We can’t have a dog in the house, even if Juan begs me on his knees.” According to the mother. . . a) Juan is begging on his knees right now b) it does not matter whether Juan begs on his knees or not c) Juan is actually going to beg on his knees In item 21, the expected answer was b), which is an irrealis reading of the subjunctive. Distractor choices in the L1 English group (10%) evenly distributed between options a) and c); in the L1 French group, there was a very slight preference for option c) (16%), although some learners (12%) also chose a). In contrast, within the control group, c) was the option chosen by 24% of its members, while a) was unanimously rejected. Interestingly, unlike the L2 learners, the native speakers hesitated only between options b) and c). Conversely, they rejected option a), an interpretation that cannot be reconciled with the semantic properties of a subjunctive form.
160
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
In fact, in item 21, the subjunctive might have either presuppositional or irrealis interpretations, depending on the contextual assumptions taken into account in the interpretation, but only the latter was clearly included in the answer options (option b). One of the distractors (option a) refers to something that is real at speech time; this is not consistent with the subjunctive form. The other option (option c), represents an event that is not real at speech time, although it will take place in the future. Option (c) is also close to an irrealis reading of the subjunctive, if it is seen as an inferential development of the modal meaning of subjunctive, from possible to potential future, and this may have influenced the choices of the native speakers. Still, unlike the L2 learners, no native speaker chose option (a), in which it seems that the subjunctive could refer to something really taking place at speech time. The difference in the choices of the three groups approaches significance (p = 0.066) as well in item 25, repeated here for convenience: 25. (En un foro de Internet) Creo que estoy enamorado, aunque ella me ignora y no me hace caso. ¿Qué hago? El autor de este mensaje. . . a) cree que hay una posibilidad de que su amada le ignore b) explica que la amada no le presta atención c) expresa que está muy preocupado 25. (In an Internet forum) I think I’m in love, although she ignores me and pays me no attention. What should I do? The author of this message. . . a) thinks it’s possible that the one he loves ignores him b) explains that the person he loves pays him no attention c) expresses that he is very worried An assertive reading, linked to the present indicative, was expected in item 25; such reading is given in option (b). However, taken as a whole, the utterance can be understood as expressing an inference about the speaker’s attitude (the speaker is worried by the indifference of the other person), regardless of the mood being used. Such inference (expressed in option (c)) has clearly influenced the answers in the control group: 17% of its members chose that option, compared to 5% in the L1 English group and 2% in the French group. Thus, variability is found in all groups, including the control group. This is not surprising in the light of the fact that interface issues affect the interpretation processes for the utterances in these items, as explained above. Still, as it has just been shown, some differences between natives and non-natives emerge
Mood interpretation in Spanish
161
when the answers chosen by each group are taken into account. More specifically, for instance, the results found for items 21 and 25 may be due to natives integrating more pragmatic cues in utterance interpretation than non-natives. Overall, the results obtained in the interpretation task, as shown in Figure 4, suggest that variability among the L1 Spanish speakers was related to choice of context: the distractors chosen by the native informants corresponded to interpretations that were considered less salient by most participants, but they were nevertheless possible if specific contextual hypotheses were retrieved. The metalinguistic task, described in the next section, was designed and applied in order to enquire into the type of contextual hypotheses that are taken into account in the interpretation of the items by native speakers.
4.3.2 Analysis of data from the metalinguistic task As explained above in 4.2.2., ten concessive items and four conditional items (used as distractors) were split into two separate versions of the metalinguistic task (Test 1 and Test 2), in order to shorten test-taking time and avoid potential inaccuracy due to monotony or fatigue. Each version was completed by a different group of informants with similar profiles. For the purpose of analysing the responses, the explanations given by the informants of the meaning of the utterances with subjunctive were classified into three macro categories: presuppositional, irrealis and other. Each category subsumed somewhat different (but related) lines of explanation: answers alluding to discourse saliency, general knowledge or argumentative strength were considered as presuppositional; those mentioning hypothetical or unreal situations were put into the irrealis class; answers unrelated to the previous categories, unclear descriptions or lack of answers were categorized as “other”. The responses given within each category were quantified and analyzed with Wilcoxon and chi-square tests. The results obtained in the metalinguistic task are summarized in Figures 6 and 7 below:
Figure 6: Test 1 (quantified answers)
162
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
Figure 7: Test 2 (quantified answers)
For items 1, 4, 5 and 10, irrealis explanations were expected; for the remaining items, presuppositional accounts were predicted. Figures 6 and 7 clearly show that most speakers were able to identify and successfully substantiate irrealis readings of the subjunctive. Conversely, a significant amount of native informants systematically failed to describe uses of the subjunctive that could be categorized as presuppositional. The distribution of the answers for every item was statistically significant (in the chi-square test, p < 0.05, n = 123–125). Item 4 is the only irrealis case where a significant number of informants did not give the anticipated answer. This was clearly related to the fact that the indicative in item 4 was not consistent with certain words found in the utterance (quantifiers like cualquier otro, ‘any other’, are usual subjunctive triggers); instead of describing the type of interpretation, eight informants pointed out that only the subjunctive could be considered appropriate; additionally, and perhaps for the same reason, six informants chose not to respond to the item. Thus, L1 speakers tend to converge on the notion of irrealis (often verbalized as hypothesis, possibility or irreality). When this is contextually inappropriate, variability increases significantly. At the same time, the notion of presupposition is almost never verbalised as such: instead, L1 speakers claim that “the information is less important”, it pertains to “general knowledge” or it “contradicts something previously said by someone else”, all of which are, indeed, pragmatic effects linked to semantic presupposition. Irrealis readings of the subjunctive are preferred overall, even in cases (such as items 3 and 9) where a presuppositional interpretation could be easily accounted for based on the contextualization provided within each of these items. This tendency is illustrated in Figure 8. Figure 8 reveals that the percentage of each account type is almost identical for both tests, in spite of the items being different (three items had preferably irrealis readings of subjunctive in test 1, but only one in test 2): in both cases, the descriptions provided of the subjunctive versions of the utterances are classified as irrealis in more than 50% of the answers (53% in test 1, 55% in test 2).
Mood interpretation in Spanish
163
Figure 8: Types of interpretation (metalinguistic task)
The relatively high overall percentage of interpretations that do not fit into the irrealis or the presuppositional categories (30% in test 1, 29% in test 2) suggests, again, that not all L1 speakers managed to clearly explain the effect of mood alternation. However, while there is a great deal of variability in the pragmatic nuances described, grammatical judgments are steady, since the utterances (with the exception of the indicative use in item 4) were considered grammatical by almost all participants. This highlights an important fact: failure to metalinguistically describe an interpretation does not necessarily involve failure in conceptualising that interpretation of the utterance. For L1 speakers, grammatical knowledge is effective, but usually unconscious; while metalinguistic knowledge is necessarily conscious and, generally, much more tentative. An ability to use and adequately interpret the subjunctive does not necessarily involve any metalinguistic capacity. At the same time, the results of the metalinguistic task corroborate the idea that natives tend to access a rich array of contextual hypotheses when interpreting subjunctive forms in concessive environments. In other words, these results reinforce the impression derived from the control group’s responses to the interpretation task described in 4.2.1, i.e., that the variability shown in the responses of the native Spanish speakers is primarily related to the diverging pragmatic enrichment paths our informants followed.
5 General discussion and conclusions Two complementary tasks have been described and analysed in the previous sections. For the first task, two research questions were formulated: (1) Do French and English L2 learners of Spanish differ in their interpretations? (2) Are there
164
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
group and individual differences among the L2 learners with two different proficiency levels in the conditions underlying the mood selection interpretations? In relation to these questions, it has been observed that neither the L1, whether French or English, nor the overall level of proficiency in L2 Spanish, have a clear impact on the performances of the learners. In spite of the typological similarities between French and Spanish, the difficulties encountered by L1 French speakers were similar to those of the English-speaking group; we hold – and the results of the metalinguistic task presented here support this view – that this is because interpreting mood in Spanish is challenging, regardless of the L1, due to the fact that it involves integrating pragmatic (information from the communicative and discourse context), morphological (ability to detect mood and tense features) and syntactic processing. Our results suggest that the main source of difficulty for non-natives is not directly related to the tense or mood being used in each utterance; nor can it be attributed to specific interpretations of the indicative or the subjunctive mood. In every utterance, there was one distractor that was not appropriate for contextual reasons, and another one that was not consistent with the contribution of the verb form; the latter were sometimes chosen by the informants, despite being ungrammatical. Thus, what our groups of non-native speakers seem to have taken into account when choosing an interpretation are lexical and pragmatic cues, rather than the morpho-syntactic information encoded in the verb form. As said in Section 3.1, we hypothesized an L1 influence, with putative advantages of similarity in the representation of morpho-syntactic properties between the L1 and L2; this has not been confirmed. However, it should be recalled that the possibility of mood alternation in concessive environments is severely restricted in French. Thus, what our study shows is that having a morphological mood system in the L1 similar to that of the L2 does not help the learners to interpret mood features, in those environments where the L1 and the L2 diverge. Contrastingly, the possibility that an advantage exists in other types of environment remains open. Our findings in this respect are compatible with Lardiere’s (2009) view that, regardless of the presence of similar formal features in the L1 for the French-speakers, differences in the way in which such features are assembled and the conditions for their expression entail that learners must reconfigure sets of formal and semantic properties in order to acquire them in the L2. Context, understood as an individual cognitive construct (Sperber and Wilson 1986/1995), also plays a vital role in the interpretations of the native speakers: in the interpretation task, performance in the control group was not significantly higher than that of the L2 groups; in some specific items, natives chose the
Mood interpretation in Spanish
165
expected responses with even less frequency than the non-native groups. Unlike L2 learners, natives did not usually choose ungrammatical distractors, but options that could be possible if specific contextual hypotheses were accessed, even though the contextualisation of the items did not bring them up. Two of the items in the first task, at least, suggested that native speaker variability (unlike non-native variability) was mainly due to diverging pragmatic enrichment processes. This hypothesis was confirmed in the metalinguistic task, in which native speakers gave considerably different accounts of the interpretive variations linked to mood alternation in the pairs of utterances; such variation proved the rich array of contextual assumptions that our informants were able to contemplate. However, a strong preference for irrealis interpretations was perceived overall. As shown in Section 2.1, the subjunctive mood is frequently related to the expression of potentiality or possibility. Such frequency may be due to, or be the origin of, a default interpretation of this mood: native Spanish users could have a tendency to first access irrealis interpretations, and try other possibilities only if strong contextual assumptions are contradictory with these readings. A possible explanation for this is that L1 Spanish speakers may possess a folk theory of the subjunctive mood, consisting in an awareness of its connection to irrealis modality, which is what the participants verbalised in the task. Still, our results are not conclusive on this issue, and the seeming irrealis bias could be an effect of the task design, and actually just reflect the speakers’ relative inability to clearly verbalise other types of interpretation. Indeed, many instances of such a phenomenon have been found in our data. Compared to L2 users, native speakers have proven to be more efficient in using automatized grammatical knowledge: they are significantly faster in solving the tasks, and their grammatical judgments are steady. Nevertheless, their metalinguistic explanations of the effects of one or another mood tend to be vague and imprecise. This confirms the idea developed in section 2.2, on the procedural meaning of the subjunctive: as described in Relevance Theory, procedural meaning is abstract and difficult to verbalise and, unlike conceptual meaning, it is accessible instantaneously (Blakemore 1983; Wilson and Sperber 1993). Turning again to the research questions formulated in section 3.1, we conclude that, indeed, L2 interpretable features are acquirable, regardless of whether they are present or not in the L1 (Interpretability Hypothesis, Tsimpli and Dimitrakopoulou 2007; Feature Reassembly Hypothesis, Lardiere 2008, 2009), since the Spanish L2 learners from our study successfully chose the expected interpretations for the majority of the proposed utterances. Nevertheless, ungrammatical options are not entirely ruled out in the L2 groups, suggesting different processing tendencies between native and non-native speakers, along the lines of Clashen and Felser (2006) and VanPatten (2007).
166
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
For the non-natives, interpreting mood, whether or not it is a feature present in the L1, clearly presents the kind of difficulty predicted by the Interface Hypothesis. Additionally, as the IH claims, integrating information from different cognitive modules (linguistic and pragmatic) also constitutes a source of variability, or instability, for natives, even though the underlying grammatical knowledge accessed in such instances is substantially different. The results of the metalinguistic task presented above have provided some insight into the processes by which the mental representation of contextual information during interpretation leads to this variability. However, the limitations of our study must be acknowledged here. The amount of data and the population sample are obviously too small to claim any generalizability of our conclusions. Additionally, due to the partially qualitative nature of the metalinguistic task, our findings should be validated by an independent, preferably quantitative measure. Clearly, more research is needed on the differences and similarities of native and non-native knowledge of grammar and interface integration. In our account, we have appealed to the framework of Relevance Theory for the description of the semantics of mood, and of the mechanisms by which utterances are interpreted through inferential processes that operate on linguistic and contextual information. This framework provides theoretical notions that are advantageous for our purposes: by analysing mood as an instance of procedural encoding, it becomes possible to substantiate an explanation of why contextual information plays a fundamental role in determining the effects of mood alternation on the overall interpretation of an utterance. In addition, it offers a means for specifying the discourse-contextual conditions that lead to the specific effects of mood on the interpretation of a given utterance. Thus, using the RT framework for considering the diversity among L1 Spanish speakers themselves with regards to their responses to mood alternation interpretation tasks, and looking at this diversity in relation to the behaviour of the Spanish L2 learners, has offered new evidence in favour of the procedural nature of the semantics of mood. It is our hope that in future studies, both L1 development and L2 acquisition studies may benefit from this approach to the systematisation of the interaction between pragmatics and linguistic representation.
6 References Ahern, Aoife. 2004. El subjuntivo: significado e inferencia. Un estudio basado en la Teoría de la Relevancia. PhD Dissertation. Madrid: UNED. Ahern, Aoife. 2006. Spanish mood, metarrepresentation and propositional attitudes. In Ken Turner and Klaus von Heusinger (eds.), Where Semantics Meets Pragmatics, 445–471. Amsterdam: Elsevier.
Mood interpretation in Spanish
167
Ahern, Aoife. 2010. Propositional attitudes in relevance theory: an overview. In Wałaszewska, Ewa, Marta Kisielewska-Krysiuk y Agnezska Piskorska (eds.), In the Mind and across Minds: a Relevance-Theoretic Perspective on Communication and Translation, 147–167. Cambridge: Cambridge Scholars. Ahern, Aoife, José Amenós-Pons and Pedro Guijarro-Fuentes. 2014. Interfaces in the interpretation of mood alternation in L2 Spanish. Morpho-phonology, semantics and pragmatics. Eurosla Yearbook 14. 173–200. Ahern, Aoife and Manuel Leonetti. 2004. The Spanish subjunctive: procedural semantics and pragmatic inference. In Rosina Márquez-Reiter and María Elena Placencia (eds.) Current Trends in the Pragmatics of Spanish, 35–57. Amsterdam / Philadelphia: John Benjamins. Allen, Mary J., and Wendy M. Yen. 2002. Introduction to Measurement Theory. Long Grove, IL: Waveland Press. Amenós-Pons, José. 2010. Los tiempos de pasado del español y el francés: semántica, pragmática y aprendizaje de E/LE. Perspectivas desde la Teoría de la Relevancia. PhD Dissertation, Madrid: UNED. Amenós-Pons, José. 2015. Spanish ‘Imperfecto’ vs. French ‘Imparfait’ in Hypothetical Clauses: A Procedural Account. Cahiers Chronos 27. 235–263. Ballestero, Carmen. 2010. La construction avec aunque : définition, selection modale, traductions françaises. PhD Dissertation, Université de la Sorbonne Nouvelle-Paris III. Bedel, Jean-Marc. 1997. Grammaire de l’espagnol moderne. Paris: Presses Universitaires de France. Blakemore, Diane. 1986. Semantic constraints on relevance. Oxford: Blackwell. Borgonovo, Claudia and Philippe Prévost. 2003. Knowledge of Polarity Subjunctive in L2 Spanish. In Beachley, Barbara, Amanda Brown, and Frances Conlin (eds.), Proceedings of the 27 Boston University Conference on Language Development, 150–161, Boston: Cascadilla Press. Borgonovo, Claudia, Bruhn de Garavito, Joyce, Guijarro-Fuentes, Pedro, Prévost, Philippe and Elena Valenzuela. 2006. Specificity in Spanish: The syntax/semantics interface in SLA. In Eurosla Yearbook 6. 57–78. Amsterdam: John Benjamins. Borgonovo, Claudia, Joyce Bruhn de Gravito and Philippe Prévost. 2008. Methodological issues in the L2 acquisition of syntax/semantics phenomenon: How to assess L2 knowledge of mood in Spanish relative clauses. In Joyce Bruhn de Garavito and Elena Valenzuela (eds), Proceedings of the Hispanic Linguistics Symposium, 13–24. Somerville, MA: Cascadilla Press. Bruhn de Garavito, Joyce and Elena Valenzuela. 2008. Eventive and stative passives in Spanish L2 acquisition: a matter of aspect. Bilingualism: Language and Cognition. 11. 323–336. Burrow, J. A. and Torlac Turville-Petre (eds.). 2005. A Book of Middle English. Oxford: Blackwell. Clahsen, Harald and Claudia Felser. 2006. How native-like is non-native language processing? Trends in Cognitive Sciences 10 (12). 564–570. Chomsky, Noam. 2007. Of minds and language. Biolinguistics 1, 9–27. Damourette, Jacques and Edouard Pichon. 1911/1936. Des mots a la pensée. Essai de grammaire de la langue française. Paris: d’Artey, vol. 5. De Saussure, Louis. 2003. Temps et pertinence. Brussels: Duculot. Escandell-Vidal, Victoria and Manuel Leonetti. 2011. On the Rigidity of Procedural Meaning. In Victoria Escandell-Vidal, Manuel Leonetti and Aoife Ahern (eds), Procedural Meaning: Problems and Perspectives, 81–103. Bingley: Emerald.
168
Aoife Ahern, José Amenós-Pons and Pedro Guijarro-Fuentes
Escandell-Vidal, Victoria, Manuel Leonetti and Aoife Ahern (eds). 2011. Procedural Meaning: Problems and Perspectives. Bingley: Emerald. Fábregas, Antonio. 2014. A guide to subjunctive and modals in Spanish. Questions and analyses. Borealis 2.3. 1–94. Hawkins, Roger and Hajime Hattori. 2006. Interpretation of English multiple wh-questions by Japanese speakers, a missing uninterpretable feature account. Second Language Research 22. 269−301. Hopp, Holger. 2004. Syntactic and interface knowledge in advanced and near-native interlanguage grammars. In Foster-Cohen, Susan et al (eds.), Eurosla Yearbook 4. 67–94. Amsterdam: John Benjamins. Gerboin, Pierre, and Christine Leroy. 1991. Grammaire d’usage de l’espagnol contemporain. Paris: Hachette. Grevisse. Maurice. 2011 [1936]. Le bon usage de la langue française, 15th Edition. Bruxelles. De Boeck-Duculot. Jary, Mark. 2009. Relevance, assertion and possible worlds: A cognitive approach to the Spanish subjunctive. In De Brabantier, Philippe and Mijail Kissine (eds), Utterance interpretation and cognitive models, 235–177. Bingley: Emerald. Kiparsky, Paul and Carol Kiparsky. 1971. Fact. In Steinberg, Danny and Leon Jakobovits (eds), Semantics, an Interdisciplinary Reader in Philosophy, Linguistics and Psychology, 345– 370. Cambridge: Cambridge University Press. Lardiere, Donna. 2008. Feature Assembly in Second Language in Second Language Acquisition. In Juana M. Liceras, Helmut Zobl, and Helen Goodluck, (eds.), 2008: The role of formal features in second language acquisition. New York: Lawrence Erlbaum Associates. 106– 140. New Jersey: Lawrence Erlbaum. Lardiere, Donna. 2009. Some thoughts on a contrastive analysis of feature in second language acquisition. Second Language Research 25.2. 173–227. Lozano, Cristóbal. 2006a. Focus and split intransitivity: The acquisition of word order alternations in non-native Spanish. Second Language Research 22. 1–43. Lozano, Cristóbal. 2006b. The development of syntax-discourse interface: Greek learners of Spanish. In Vicenç Torrens and Linda Escobar (eds.), The Acquisition of Syntax in Romance Languages, 371–399. Amsterdam: John Benjamins. Moeschler, Jacques. 1998. Pragmatique de la référence temporelle. In Jacques Moeschler (ed.), Le temps des événements, 157–180. Paris: Kimé. Montrul, Silvina. 2004. Subject and object expression in Spanish heritage speakers: A case of morpho-syntactic convergence. Bilingualism: Language and Cognition 7.2. 125–142. Montrul, Silvina. 2008. Incomplete Acquisition in Bilingualism. Re-examining the Age Factor. Amsterdam: John Benjamins. Montrul, Silvina. 2011. Interfaces and incomplete acquisition. Lingua 212.4. 591–604. Montrul, Silvina and Maria Polinsky. 2011. Why not heritage speakers? Linguistic Approaches to Bilingualism, 1.1. 58–62. Montrul, Silvina and Celeste Rodríguez-Louro. 2006. Beyond the syntax of the Null Subject Parameter: A look at the discourse-pragmatic distribution of null and overt subjects by L2 learners of Spanish. In Torrens, Vicenç and Linda Escobar (eds.), The Acquisition of Syntax in Romance Languages, 401–418. Amsterdam: John Benjamins. Montrul, Silvina and Roumyana Slabakova. 2002. On aspectual shifts in L2 Spanish. In Skarabela, Barbora, Sarah Fish, and Anna H.-J. Do (eds.), BUCLD 26 Proceedings, 631– 42. Somerville, MA: Cascadilla Press.
Mood interpretation in Spanish
169
Montrul, Silvina and Roumyana Slabakova. 2003. Competence similarities between natives and near-native speakers: An investigation of the Preterit/Imperfect contrast in Spanish. Studies in Second Language Acquisition 25. 351–398. Nicolle, Steve. 1997. A Relevance-theoretic account of be going to. Journal of Linguistics 33. 355–377. Palmer, Frank R. 2001 [1979]. Mood and Modality, 2nd Edition. Cambridge: Cambridge University Press. Quer, Josep. 2001. Interpreting Mood. Probus 13. 81–111. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. 1972. A Grammar of Contemporary English. Essex: Longman. Riegel, Martin. 1994. Grammaire méthodique du français. Paris: Presses Universitaires de France. Rhis, Alain. 2013. Subjonctif, gérondif et participe présent en français. Une pragmatique de la dépendence verbale. Bern: Peter Lang. Rouchota, Villy. 1994. The subjunctive in Modern Greek: Dividing the labour between semantics and pragmatics. Journal of Modern Greek Studies 12. 185–201. Sorace, Antonella. 2004. Native language attrition and developmental instability at the syntaxdiscourse interface: Data, interpretations and methods. Bilingualism: Language and Cognition 7. 143–145. Serratrice, Ludovica, Antonella Sorace and Sandra Paoli. 2004. Transfer at the syntax-pragmatics interface: subjects and objects in Italian-English bilingual and monolingual acquisition. Bilingualism: Language and Cognition 7. 183–205. Sorace, Antonella. 2011. Pinning down the concept of ‘interface’ in bilingualism. Linguistic Approaches to Bilingualism 1. 1–33. Sorace, Antonella and Francesca Filiaci. 2006. Anaphora resolution in near-native speakers of Italian. Second Language Research 22. 339–368. Sorace, Antonella and Ludovica Serratrice. 2009. Internal and external interfaces in bilingual language development: Beyond structural overlap. International Journal of Bilingualism 13. 195–210. Sperber, Dan and Deirdre Wilson. 1986/1995. Relevance. Communication and Cognition. Oxford: Blackwell. Tsimpli, Ianthi Maria and Maria Dimitrakopoulou. 2007. The Interpretability Hypothesis: Evidence from Wh-Interrogatives in Second Language Acquisition. Second Language Research 23. 215–242. Tsimpli, Ianthi Maria and Antonella Sorace. 2006. Differentiating Interfaces: L2 performance in syntax-semantics and syntax-discourse phenomena. In Bamman, David, Tatiana Magnitskaia, and Colleen Zaller (eds), Proceedings of the 30th Boston University Conference on Language Development, 653–664. Somerville, MA: Cascadilla Press. Tsimpli, Ianthi Maria, Antonella Sorace, Caroline Heycock and Francesca Filiaci. 2004. First language attrition and syntactic subjects: A study of Greek and Italian near-native speakers of English. International Journal of Bilingualism 8. 257–277. VanPatten, Bill. 2007. Input processing in adult SLA. In VanPatten, Bill and Jessica Williams (eds.), Theories in Second Language Acquisition, 115–135. Mahwah, NJ: Erlbaum. Wilson, Deirdre and Dan Sperber. 1993. Linguistic form and relevance. Lingua 90. l 1–25.
Tim Diaubalick and Pedro Guijarro-Fuentes
7 How do German speakers acquire the Tense-Aspect-System in Spanish as a Second Language? An empirical study Abstract: Recent hypotheses in SLA, such as the Interpretability Hypothesis (Hawkins and Hattori 2006) or the Feature Reassembly Hypothesis (Hwang and Lardiere 2013; Lardiere 2009), attribute major effects to the learner’s L1 regarding the success in their L2. Focusing on these approaches, an empirical study with German L2 learners of Spanish was conducted. The comparison between the involved tense systems reveals great differences: while the Spanish past tenses involve the feature of (im)perfectivity that must be marked morphologically (Leonetti 2004, Zagona 2007); in German there is no evidence that there are any morphological aspectual differences at all (Schwenk 2012, Heinold 2015), not even a grammaticalized progressive. The goal of our study was, therefore, to analyze in how far it is possible to acquire an uninterpretable feature not instantiated in the learner’s L1 in adult SLA in comparing different proficiency levels. It turned out that L2 learners had indeed problems to acquire all nuances persisting until advanced levels. Our L2 learners seemed to rely heavily on temporal adverbials in a way that exceeded any similar effects observed in earlier studies. The German language often attributes higher values to lexical elements, such as particles and adverbs, than to verbal morphology. This represents a possible cause for the observed behaviour.
1 Introduction The acquisition of the Spanish past tense system in Second Language Acquisition (SLA) is still a current research ground, where several generative and cognitive theories were developed (e.g. Comajoan 2013; Domínguez, Tracy-Ventura, Arche, Mitchell and Myles 2013; Salaberry 2008, 2011; Salaberry and Comajoan 2013 among others) and can be discussed from various perspectives. General generative SLA-theories such as the Interpretability Hypothesis (IH) (see Hawkins and Hattori 2006; Tsimpli and Dimitrakopoulou 2007) and Tim Diaubalick, Wuppertal University and University of the Balearic Islands Pedro Guijarro-Fuentes, University of the Balearic Islands
172
Tim Diaubalick and Pedro Guijarro-Fuentes
the Feature Reassembly Hypothesis (FRH) (Hwang and Lardiere 2013; Lardiere 2009) highlight the role of the speaker’s native language (L1). While the IH predicts that uninterpretable features which are not present in a learner’s L1 are impossible to acquire even at late learning stages, the FRH states that target-deviate constructions occur due to form-to-meaning mapping problems. Specific studies regarding the acquisition of the Spanish tense system, in contrast, usually concentrate on postulating learning and development stages (Comajoan 2013; Salaberry 2011). They are often based on native English speakers1, but the role of the L1 is not always highlighted. In the present chapter, German speaking participants of L2 Spanish were chosen in order to test the aforementioned hypotheses and their tenets2. German is a language with a verbal system which results quite different from both the Spanish and the English system, and is therefore an adequate candidate to test the predictions on the acquisition of (un)interpretable features. The results from the empirical study reported herein seem to suggest that some of the findings in aspect studies regarding development are de facto based on the characteristics of English, and do not necessarily predict the acquisition of language learners of other types, including German. That is, the native language of a second language learner indeed plays a crucial role. This chapter is organized as follows: After giving a description of the two linguistic systems involved in the study (Spanish and German) in Section 2, Section 3 brings together previous research findings in general SLA and in specific tense-aspect studies, and ends with particular research questions for the present study. After presenting the methodology in Section 4, Section 5 presents the results obtained in the conducted study. Finally, the chapter concludes with a discussion and some suggestions for future research.
2 Linguistic background 2.1 Spanish At a very descriptive level, in Spanish, as in other Romance languages like Italian or French, grammatical aspect is a phenomenon which is consequently marked 1 However, interested readers are referred to Comajoan (2013), Díaz et al. (2008), Salaberry and Comajoan (2013), Shirai (2013), among others, as noteworthy exceptions and discussions on the role of L1 and research in L2 tense and aspect based on L2 learners, other than English speaking ones. 2 As an anonymous reviewer pointed out, there are indeed studies that contrast different L1–L2 settings and include German (see von Stutterheim and Carroll 2006). However, the combination L1 German-L2 Spanish has not been investigated in the way we are doing here.
How do German speakers acquire the Spanish TA-system?
173
in the past tenses (Zagona 2007). The most important contrast lies between the Imperfect (‘cantaba’ ‘I was singing’/‘I would sing’/‘I used to sing’), which primarily marks habitual or continuous past actions, and the Preterit (‘canté’ ‘I sang’), which is used for punctual and completed actions. Moreover, the distinction involves many further semantic interpretative features which are to be acquired in order to use the forms adequately. The grammatical aspect has to be distinguished from the lexical one. The latter, also referred to as aktionsart3, is a semantic property of every verbal action which is inherent and universal, although not always overtly marked morphologically. Vendler (1957) divided verbal phrases into four groups; states, activities, accomplishments and achievements. One lexical aspect feature is telicity which concerns the existence of an inherent endpoint (cf. Comrie 1976). This feature is crucial for separating states and activities from accomplishments and achievements. An example for a telic event is a verbal phrase like to run a mile that has a clearly marked endpoint upon which the action is completed. Actions like to read, however, do not have a natural endpoint and are therefore atelic. In contrast to this, regarding the grammatical aspect, Comrie (1976) distinguishes between imperfective, perfective and perfect aspect. While the Imperfect makes “explicit reference to the internal temporal structure of a situation, viewing a situation from within” (Comrie 1976: 24), a perfective tense “represents the actions pure and simple, without any additional overtones” (Comrie 1976: 21). The perfect aspect then, often seen as special case of perfectivity, is used for events with a major relation to the present. These three categories seem to be more or less congruent with the Spanish tenses of Imperfect, Preterit and Present Perfect. Nevertheless, there is an ongoing debate about the nature of the Spanish Imperfect, which questions the importance of perfectivity in the Romance verbal system at all, reducing the relevant features to only anteriority, simultaneity and posteriority. Studying Italian, Giorgi and Pianesi (2004) postulate the Romance Imperfect as a present in the past (Giorgi and Pianesi 2004: 260), explaining the characteristics by means of the temporal relationships of anaphoricity and anchoring. That is, the Imperfect, being an anaphoric tense, lacks any interpretation if no time topic, which anchors the reference point, is established. Various examples are adduced to show that, while the Preterit has a rather robust interpretation, the Imperfect is highly affected by various contexts, both linguistically and pragmatically:
3 Note, however, that the terms are not interchangeable for every researcher (cf. Salaberry and Ayoun 2005)
174
Tim Diaubalick and Pedro Guijarro-Fuentes
(1)
Alle tre Mario prendeva il tè. A las tres Mario tomaba el té. At the three Mario drank.IMPF the tea. ‘At three Mario had tea.’
(2)
Mario prendeva il té alle tre. Mario tomaba el té a las tres. Mario drank.IMPF the tea at the three. ‘Mario had tea at three.’ (Italian examples and English translation taken from Giorgi and Pianesi 2004: 275, italics in orig.)
While in (1) all possible readings of an imperfective tense, i.e. continuity, habituality and futurity, are available, the position of the adverbial in (2) blocks the continuous interpretation. Nothing similar happens with the corresponding perfective tense. Another proof can be seen in the following contrast: (3)
Gianni ha sognato che c’è stato un terremoto. Gianni soñó que hubo un terremoto. Gianni dreamed.PRET that had.PRET an earthquake. ‘Gianni dreamed that there has been an earthquake.’
(4)
Gianni ha sognato che c’era un terremoto. Gianni soñó que había un terremoto. Gianni dreamed.PRET that had.IMPF an earthquake ‘Gianni dreamed that there was an earthquake.’ (Italian examples and English translation taken from Giorgi and Pianesi 2004: 281, italics in orig.)
While the anaphoric Imperfect takes the dream as an anchor placing the earthquake inside the dream, the perfective tense refers to an event in the real world, which was additionally dreamed of by Gianni. Conversely, García Fernández (1999) and Leonetti (2004) give some examples which contradict the prominence of anaphoricity. While the indirect speech indeed seems appropriate to demonstrate that the Imperfect logically completes the usages of the Pluperfect as an Antepretérito (or ‘Pre-Preterit’) and the Conditional as a Pospretérito (or ‘Post-Preterit’), and thus adopts the role of a CoPreterit, some particular sentences show that this is not always the case (García Fernández 1999). See examples (5) and (6):
How do German speakers acquire the Spanish TA-system?
(5)
Juan nos contó que María estaba triste el día de su boda. (García Fernández 1999: 177) Juan us told.PRET that Maria was.IMPF sad the day of her wedding. ‘John told us that Mary was being sad on her wedding day.’
(6)
Napoleón dijo que Ana Bolena era de Edimburgo. (García Fernández 1999: 178) Napoleon told.PRET that Ana Bolena was.IMPF from Edimburgo. ‘Napoleon said that Anne Boleyn was from Edinburgh.’
175
It is further claimed that the lexical aspect exhibits an influence, e.g. preventing states from being interpreted as simultaneous, which is shown via (7): (7) *Ana Bolena era de Edimburgo cuando se casó con Enrique VIII. (García Fernández 1999: 179) Ana Bolena was.IMPF from Edimburgo when herself married.PRET with Enrique VIII. ‘Anne Boleyn was from Edinburgh when she got married to Henry VIII.’ Leonetti (2004), likewise, concludes that in reality anaphoricity is an effect, whereas the underlying contrast between Imperfect and Preterit is of an aspectual nature. According to Leonetti (2004), imperfectivity expresses non-bound actions without fixed limits. Therefore, to be adequately interpreted, a contextual frame establishing borders between which the event is localized is needed. A perfective tense, in comparison, always presents the action as being completed. In any case, there is one effect in Spanish that undoubtedly must be seen as aspectual: coercion. For example, when stative verbs like saber (‘to know’, ‘to have information’) or conocer (‘to know’, ‘to be acquainted with’) are combined with a perfective tense, their interpretation is shifted to an eventive meaning (‘to get to know’). Similar phenomena can occur with extralinguistic elements that, likewise, coerce the reading of some verbs into different categories, as the following examples taken from Leonetti (2004: 483) show: (8)
Estaban a buen precio. . . Were.IMPF to good price. . . ‘They were at a good price. . .’
(9)
Encendió un cigarrillo. La fiebre le daba al tabaco un gusto de miel. lit-up.PRET a cigarette. The fever him gave.IMPF to-the tobacco a taste of honey. ‘He lit up a cigarette. The fever gave the tobacco a taste of honey’
176
Tim Diaubalick and Pedro Guijarro-Fuentes
While in (8), a possible reference could be a strange look given by a friend seeing a new pair of shoes, so that the point of buying is established as implicit reference point, (9) shows that the verb daba (‘gave’) is to be understood as parallel to the nonverbally expressed act of smoking. In summary, the Preterit involves the semantic features of anteriority in relation to the moment of speech and, according to the argumentation presented at last, perfectivity. The Imperfect, on the other hand, is characterized as being either parallel to a past action, or as imperfective, and involves, thus, anteriority, simultaneity and imperfectivity. The grammatical aspect, following the generativist approach (Chomsky 2001), presents a functional category AspP, which, assuming Pollock’s Split-InflHypothesis (c.f. Pollock 1989), should be separated from the temporal phrase TP and the agreement phrase AgrP (see Figure 1). The interpretable feature [±perfective] is checked here via verb raising through an uninterpretable reflection of the same feature on the morphological verb form. According to Chomsky (2001), grammatical and lexical aspects are spelled out at different phases (since the latter one is part of the vP); thus, the mentioned coercion effect must clearly take place on the logical form (LF).
Figure 1: Tense and Aspect representation
How do German speakers acquire the Spanish TA-system?
177
2.2 German The German language belongs to the group of the so-called non-aspectual languages (“Nichtaspektsprachen”, Schwenk 2012: 32), and most of the contrasts must be expressed lexically (cf. Schwenk 2012: 37). The two major past tenses are the Present Perfect (‘ich habe gemacht’ ‘I have done’) and the Preterit (‘ich machte’ ‘I did’). Whereas there is no consensus about the nature of interchangeability, it seems indeed to be agreed upon that aspect does not play a major – if any – role in the distinction (cf. Schwenk 2012; Heinold 2015). It is possible, on the contrary, to characterize the tenses by establishing relationships between the pure temporal variables: i.e., Time of Utterance, Topic Time and Time of Situation (Schwenk 2012). The only features characterizing the tense system are anteriority, simultaneity/inclusion and posteriority. Possible aspectual contrasts are only an effect of different combinations of those temporal variables. In the colloquial language, however, the tenses are treated as completely interchangeable in almost all cases, the Present Perfect being the dominant one (see DUDEN 2006, Heinold 2015). Generally speaking, one can affirm that for a German speaker there is no obvious difference between the competing past tense forms; the contrast is restricted to only the most formal and written registers. Instead, German marks most of the contrasts lexically; using periphrasis, particles or certain adverbial complements (cf. Sánchez Prieto 2002). To summarize the difference between the two systems, it is clear that the Spanish language presents more complex contrasts than German, since there is indisputably a greater amount of morphological verb forms and the role of aspect is far more prominent. Both systems share the semantic interpretable features of anteriority, simultaneity and posteriority, but only the Spanish tenses may additionally be characterized by the perfectivity, completeness and boundness suggesting the presence of AspP. Keeping in mind, however, the presented discussion about the role of the aspectual features in the Spanish system, the systems might resemble each other if aspect is seen as only effect of different temporal combinations. This has consequences for the possible outcome for a non-native German learner of L2 Spanish, because, if aspectual contrasts are only effects, they do not have to be acquired as such. In that case, only the aspectual coercion, as presented above, would present a problem that indeed is connected to aspectual features.
178
Tim Diaubalick and Pedro Guijarro-Fuentes
3 State of the art – second language acquisition of the past tense system 3.1 Previous research and competing hypotheses An ongoing debate in generative SLA research is whether the Universal Grammar (UG) is still accessible. In the last decade, many different approaches and hypotheses came about that made various predictions regarding the possible outcome, the ultimate attainment, and the development stages in acquiring an L2 (see White 2003 for a review). Recent SLA theories distinguish between grammatical features that depending on the learner’s L1 are acquirable or not. Within this explorative line of enquiry, there is the Interpretability Hypothesis (IH, see Hawkins and Hattori 2006, Tsimpli and Dimitrakopoulou 2007), according to which uninterpretable features (i.e., semantically valueless features relevant only for syntactic processing; see Chomsky 2001) are impossible to acquire, if absent in their L1, after a critical period that depends on the learner’s age. Universal principles and processing mechanisms of linguistic input, in contrast, are claimed to be available life-long, i.e., new interpretable features are acquirable, regardless of whether they are absent or present in the L1. Thus, if an uninterpretable feature already exists in the learner’s native grammar, no acquisition problems are to be expected. If, however, the same feature does not form part of the L1, then its acquisition is unattainable and the learner has to fall back upon target-deviant strategies which may turn out to be different when testing a very concrete structure (see also Hawkins and Chan 1997). Following the same line of enquiry (that is, availability of individual uninterpretable and interpretable features), Lardiere (2009) proposes the Feature Reassembly Hypothesis. Lardiere claims that acquiring an L2 grammar is not a question of whether features are still available for selection from a universal inventory. The relevant question is how features are assembled and mapped to lexical items taking into consideration particular language-specific conditions under which they are phonologically realized. Thus, two languages can select the same formal features such that a native speaker of language A acquiring language B would not need to ‘reset’ parameters in the target grammar. However, how a particular feature is assembled and the conditions of its expression in each of the two languages may be quite different. The learning task would then consist of appropriately reconfiguring or reassembling formal and semantic bundles in the L2 lexicon, and determining the specific conditions under which their properties may or may not be morphophonologically expressed. That is, in
How do German speakers acquire the Spanish TA-system?
179
addition to acquiring new features, the adult learner must redeploy the morphological expression of individual features from the way they are employed in the native language. More recently, Hwang and Lardiere (2013) provide new evidence for the FRH which can be summarized in the following steps: First, a complete transfer from the learner’s native grammar is realized, i.e., the learner assumes a 1:1 correspondence between his native system and the foreign one. The specific features of a given grammatical form are thus underspecified and can appear in contexts where the target system disallows them. Next, the learner integrates continuously the target features and, by this, carries out a reconfiguration or reassembly. In the very last stage, a complete mastery of the target system is declared, thus negating any resisting L1-effects in the ultimate attainment. Particularly, tense aspect features are an intriguing phenomena to explore in terms of lexical feature assembly and reassembly, as stated by the Feature Reassembly Hypothesis. In light of these hypotheses, it is important to investigate distinct linguistic features in different L1–L2 configurations. To that end, we investigated the acquisition of the Spanish tense aspect system by German adult learners of Spanish at distinct proficiency levels adopting the above generative hypotheses4. Related to that, many of these specific hypotheses are based on English speaking participants in their original formulation, yet they do not always discuss L1-effects as a main reason for the observed specific behaviour5. While it is not explicitly claimed that the findings relate universally to all languages, it remains a partially open question whether the postulated patterns and learning stages really apply generally to all language learners, or only reflect a special setting. The generativist hypotheses, nevertheless, stress the importance of considering both grammar systems (L1 and L2) involved in the acquisition process. According to Hawkins and Chan (1997) and Hawkins and Hattori (2006), learners can use deviated strategies to overcome the problems which result from not being able to acquire new uninterpretable features. These strategies can in many cases result in an adequate output which seems to equal the target grammar. Only some peripheral contexts show that the strategies are actually defective. The findings presented above can, thus, be regarded as the results of such strategies. In that case, however, L1 effects should be crucial since the IH is
4 For a revision of those hypotheses conducted outside the generativist framework (such as the Lexical Aspect Hypothesis or Default Past Tense Hypothesis) see Comajoan (2013), Salaberry (2011), Salaberry and Comajoan (2013), amongst others. 5 In the case of his Default Past Tense Hypothesis, however, Salaberry (2008) admits that it is based on the English L1-system the learners have. Nevertheless, it is unclear if this means that speakers of other languages do or do not follow different patterns.
180
Tim Diaubalick and Pedro Guijarro-Fuentes
strongly dependent on a L1–L2-comparison. McManus (2011, 2015) shows that, in the case of French, the L1 does indeed have an influence on the acquisition of the verbal system. It is, therefore, necessary, also in the case of Spanish, to examine speakers of several other backgrounds to strengthen the observations. This is intended to be done in the present study, concentrating on native speakers of German. For the purpose of this chapter, one of the studies about feature reassembly which should especially be mentioned, is the one conducted by Slabakova and Montrul (2003), who analyzed form-to-meaning mapping problems concerning the features of specificity and genericity. Their findings are based on English speaking participants, and thus the results may depend on special EnglishSpanish situations. The two grammar systems possess one ambiguous and one unambiguous tense to express the features [+habitual], [+punctual] and [+continuous]. The difference, however, consists in the feature bundling. The unambiguous forms – the Past Progressive for continuity in English and the Preterit for punctuality in Spanish – do not match and a L1 transfer would lead to a targetdeviate system (cf. Slabakova and Montrul 2003: 188). A feature reassembly is necessary. In order to investigate the nature of this reassembly, Slabakova and Montrul use both constructions with obvious habituality-punctuality contrasts, as well as a poverty-of-stimulus phenomenon, which presents a valid test ground for the UG access (cf. Schwartz and Sprouse 2000). The Spanish grammar allows constructions with impersonal subjects that can refer either to a generic action or to a specific event. Since the Preterit in Spanish implies specificity, it automatically hinders any generic interpretation. That is, in the imperfective construction ‘se comía’ two readings are still available, which correspond to the English translations ‘one ate’ (generic) or ‘we ate/we were eating’ (specific) (cf. Slabakova and Montrul 2003: 167). The corresponding construction with the unambiguous Preterit, however, being ‘se comió’ is necessarily specific (‘we ate’). Thus, there are two possible forms for the specific contexts, while genericity can only be expressed by one of them. Since this contrast is almost absent in foreign language classes (see Slabakova and Montrul 2003: 174), it is highly improbable to have been acquired by learning explicit rules. In their study, an intermediate and an advanced group of non-native Spanish learners, as well as a native control group, had to perform a truth judgment task on the basis of a given context which was followed either by a habitual action, by a single event, or by a sentence containing an impersonal subject. Although the authors point out that the test design does not reflect a natural language situation and might therefore have led to rather odd results in some contexts (Slabakova and Montrul 2003: 191), the findings mainly corroborated the theory of domain-specific learning mechanisms which follow from a still open UGaccess.
How do German speakers acquire the Spanish TA-system?
181
In a more recent study, Slabakova and Montrul (2008) tested more conditions and found out that those coercion effects that include an implication of pragmatics and discourse factors seem to be more problematical for language learners. However, although not all of them, some of the study participants did show to have acquired all conditions completely. It is therefore still justified to speak in favor of an open UG-access.
3.2 Research Questions and predictions for the present study In view of the different SLA-theories presented above, the general research questions which underlie the present study are as follows: (i) What features are crucial when German learners acquire the Spanish tense system? That is, are the aspectual features perfectivity, completeness and boundness necessary to be acquired? Does the interpretability of those features play any role? (ii) Is access to the Universal Grammar still possible? The interpretability of the features at play, as already mentioned, is heavily dependent on the underlying characteristics that define the contrasts between the past tenses. Recall that it is not clear, whether the Imperfect is indeed marked as imperfective, or if it in fact is only an anaphoric tense. This means that if aspect is only a mere effect of underlying temporal relationships, it does not have to be acquired as such. In this case, the German speakers would only encounter features already known from their mother tongue, i.e., anteriority, simultaneity and posteriority, but in a different constellation. If, however, the aspectual features such as perfectivity, boundness and completeness are the cause for the observed contrasts between the Spanish past tenses, then the German and the Spanish tense systems differ strongly. Involving uninterpretable formal features, thus, the only possibility – according to the IH – for the German learners would be a domain-general strategy in order to overcome the nonaccessibility of the aspectual features, which would then be visible in a development. In the case of the FRH, an observable development is to be expected in both possibilities. We can therefore specify the two research questions and state the following: (i’) Are there group and individual differences between native and non-native speakers in the use and production of past tenses in Spanish, i.e., do learners show to use target-deviate feature configurations that can be related to a transfer from the native language?
182
Tim Diaubalick and Pedro Guijarro-Fuentes
(ii’) Are there group and individual differences among the L2 learners with different proficiency levels in the distribution of the past tense in Spanish, i.e., is a transfer effect in the beginning and a reassembly task in the later stages visible? (iii’) Is there any evidence of L2 learners having acquired the uninterpretable features which involves associating a certain set of interpretable features? That is, is there any evidence of L2 learners having reassembled successfully the new target features, moving from their L1 tense/aspect system to a target/L2? If so, are there differences between proficiency levels? Does knowledge of a particular feature in our participants’ first language (German) have an influence on the reconfiguring and then the reassembly of the features in the target language (Spanish)? According to Hawkins and Liszka (2003), who conducted a study about acquiring tense as a new feature by Chinese learners, the underlying properties which are responsible for verb raising (c.f. Chomsky 2001) must contain uninterpretable features. Similar processes must be at play to cause verbal raising to grammatical AspP that can only be caused by unvalued and uninterpretable features (see Chomsky 2001). Some of the aspectual features must, thus, be uninterpretable. In that case, the IH would predict an impossibility of acquiring the complex aspectual features, since German does not present any of them at all. If assuming a total UG-access in SLA, neither impersonal subjects nor coercion contexts should cause any difficulties. In this case, the answers to research question (i’) should be “no”, at least in what concerns the most advanced speakers. According to the FRH, we expect learners to start with a feature configuration known from their native system (German, in our case) before they start to approach the target grammar continuously. Thus, question (ii’) can be answered with “yes”. Since the FRH does not distinguish between accessibility of interpretable and uninterpretable features, question (iii’) should expectatively be answered by a “yes”, again at least in the most advanced learners. Assuming the IH, on the contrary, one has to distinguish two cases: 1. If the Imperfect is indeed marked as imperfective, it will produce major problems in being acquired. In this case, we expect a difference from native-speakers, and a target-deviate learning-strategy, resulting in an acquisitional development that does not arrive at a native-like competence. Since this strategy might be based on the native German system which prefers the lexical expression over the morphology, we would expect a lexical based strategy.
How do German speakers acquire the Spanish TA-system?
2.
183
If the Imperfect, however, is rather anaphoric and contains only temporal features, no problems of feature accessibility should be expected and both the IH and the FRH would predict similar findings.
4 Methodology 4.1 Participants To collect data, offline questionnaires were distributed among groups of native and non-native speakers of Spanish. The participants consisted of 71 learners at three different proficiency levels (i.e., low-intermediate, high-intermediate and advanced) plus a Spanish monolingual speaking control group (n = 16). All participants of the experimental group were enrolled in full-time philology studies in a German university. The age varied between 20 and 28 years. 38 participants were attending a B1-level class according to the CEFR (Common European Framework Reference), and can therefore be classified as low-intermediate learners with less than two years of previous exposure to Spanish. A further 18 participants were enrolled in a course of the superior Bachelor’s module, thus having been exposed to Spanish for one or two more years and being on a B2level, which makes them high-intermediate speakers of Spanish. The third group consisted of 15 Master’s students who partially have spent a semester abroad, thus having a greater exposure to Spanish and being on a C1/C2-level. The latter group will be referred to as advanced learners. In addition to the linguistic task, participants had to complete a short ethnolinguistic questionnaire. According to its findings, participants indicated German as native and English as first foreign language. Spanish was mentioned as third language in most cases, a small number of the students additionally indicated knowledge in French or Classic Latin. Knowledge of these additional languages does not have any bearing on our findings as English, the majority language by our L2 learners, does have a tense aspect system very similar to their L1 German. All participants started learning Spanish after puberty. Besides the experimental group, a group of 16 Spanish Madrilenian philology students was used as a native monolingual control group. Their age varied between 19 and 28 years and the gender proportion was balanced. The data of these participants were recorded in Spain with none of the subjects having spent a considerable time abroad, and hence with very little or no possible interfering of any additional languages. All data from the ethnolinguistic and linguistic tasks was collected in written form without a time limit. Questionnaire materials were afterwards digitalized
184
Tim Diaubalick and Pedro Guijarro-Fuentes
and made anonymous. According to the nature of the linguistic tasks, which is described in the next section, all data were translated into nominal or numerical values in order to be statistically analyzed.
4.2 Linguistic tasks To test the participants’ knowledge of the usage of tense aspect in Spanish, a Sentence Completion Task and a Grammatical Judgement Task were used6. All items included in all tasks were accurate representations of all the features which motivate the tense aspect in Spanish presented in Section 2 above. The vocabulary used in the test items were selected from the textbooks used in class so that all L2 participants were familiar with them. With regard to the completion linguistic task, it included a sentence-level production task consisting of two parts and made of 18 gap-sentences with infinitives presented in brackets that tested different combinations of (a)telicity and (im)perfectivity in a given context with illustrating matching pictures as additional information in order to counterpart any bias results for using decontextualized sentences; participants were asked to fill the gaps in with the corresponding conjugated verb form. Examples (10)–(12) are sample items. As part of the production task, there was also a free production task in which participants were asked to tell a little anecdote using certain past-indicating adverbials without further requirements. (10)
Ayer una mujer _____ (cruzar) la calle. Yesterday a women (cross-INF) the street ‘Yesterday a woman crossed the street.’ (telic, perfective)
(11)
Su amiga _____ (cruzar) la misma calle cuando fue atropellada por un camión. Her friend (cross-INF) the same street when was.PRET hit by a truck ‘Her friend was crossing the same street when she was hit by a truck.’ (telic, imperfective)
6 Due to space limitations, no appendix containing all items for both linguistic tasks can be provided. Although we acknowledge that the use of descontextualized items in some of the components of both linguistic elicitation tasks can give rise to some reservations as one of the reviewers rightly pointed out, we are confident that the combination of multiple elicitation tasks nevertheless strengthen our findings and rest any potential bias results. Furthermore, the validity of our results is strengthened by having distinct tasks whose format is totally different, which also allowed us to control task effects. Nevertheless, future research should take into account some of the risks in designing linguistic tasks with the specific purpose of eliciting tense aspect – see chapters in Salaberry and Comajoan (2013) for a thorough discussion on these methodological issues.
How do German speakers acquire the Spanish TA-system?
(12)
185
María _____ (dibujar) pájaros, mientras Pablo _____ (dibujar) elefantes. María (draw-INF) birds, while Pablo (draw-INF) elephants. ‘María drew birds, while Pablo drew elephants.’ (atelic, imperfective)
Written instructions were presented in Spanish for all groups, but the experimental group was allowed to state orally answered questions regarding methodological matters in German. Instructions clearly indicated how to complete the different components of the production task by giving examples. Participants were instructed to indicate their “first intuition” and not to go back and change their answers. The Grammaticality Judgment task, on the other hand, contained 30 testing items, half of which were classified as ungrammatical; the task also included ten distracters without past tenses and fourteen additional fillers also testing other aspects of the Spanish grammar (54 items in total). Participants were asked to judge test sentences such as those in (13) to (18) below. Three conditions in particular were to be tested in this task. First, the aspectual coercion effect was present in five grammatical and five ungrammatical items, pre-eminently employing the verbs saber and conocer. In order to fully master these effects, an acquisition of all aspectual features, including the uninterpretable ones, is necessary. Ten further items included impersonal subject constructions presenting, as argued, a poverty-of-stimulus phenomenon. Both these conditions require an interaction of the two phases vP and CP: in the case of the coercion sentences, the interaction is caused by the direct relation between grammatical and lexical aspect. The items containing impersonal subjects, on the other hand, require an interaction between the proper subject and the verb form. A third set of ten items consisted of standard usages, usually taught in language classes, which contrasted habitual with punctual events, testing if there are some general problems with the past tenses. Items were to be rated via a numeric 3-point Likert scale. Due to the familiarity of the test subjects with and following the evaluating system used in German schools and universities, in which a smaller number means higher evaluation, the translations of the numerical assessments were: 1 = “grammatical”, 2 = “strange” and 3 = “ungrammatical”. Furthermore, a fourth scale point was included with the option “I don’t know”. Additionally, the participants were asked to correct the ungrammatical sentences in order to guarantee that it is really the verb form which is responsible for the ungrammaticality. As in the Completion Task, written instructions were given in Spanish and oral additions in German were facilitated when requested. Participants were asked to judge the acceptability of the sentences based on their first intuition, and not to go back and change their answers later. There was no time limit for the task.
186
Tim Diaubalick and Pedro Guijarro-Fuentes
Grammatical items: (13)
Quedaba con mis amigos todos los días. Stayed.IMPF with my friends all the days. ‘I met my friends every day.’ (habitual, Imperfect)
(14)
Ayer probamos un nuevo bar. Se comió muy bien allí. Yesterday tried.PRET a new bar. Self ate.PRET very well there. ‘Yesterday we tried a new bar. One ate (we ate) very well there.’ (specific subject, Preterit)
(15)
Nos conocimos en un viaje a Portugal. Ourselves knew.PRET in a trip to Portugal. ‘We got to know each other during a journey to Portugal.’ (eventive context, Preterit)
Ungrammatical items: (16) *Al final encontraba mi monedero. At-the end found.IMPF my purse. ‘In the end, I was finding my purse.’ (punctual, Imperfect) (17) *Todos me decían que se comió muy bien allí. All me said.IMPF that self ate.PRET very well there. ‘Everybody told me that one ate very well there.’ (generic subject, Preterit) (18) *Ayer todavía lo supe, pero se me ha olvidado. Yesterday still it knew.PRET, but self me has forgotten.PERF ‘Yesterday I still knew it, but I forgot it then.’ (stative context, Preterit) As pointed out by one of the reviewers, given that the construction of testing items, which – especially in the Grammatical Judgment Task – cannot always be presented with a whole context to avoid participants’ fatigue, the research design must therefore be validated by the monolingual control group. To appropriately be able to talk about target-deviate production or interpretation of tense-aspect features, in the following we will not only present the categories “grammatical” vs. “ungrammatical”, but will also always refer to the comparison with the control group.
How do German speakers acquire the Spanish TA-system?
187
5 Results 5.1 Production tasks The following tables show the results of the production tasks which tested the effect of telicity. All verb forms given by the participants were coded according to the tense form used. Since orthography and misspelling was not in the centre of attention, false written forms were considered as belonging to one of the tense categories if an assignation was unequivocal. All non-classifiable forms were not further considered in the analysis. Both the absolute number of tokens that resulted from counting all relevant gaps in the exercises, and the percentage are given. That way, individual and inter-group differences are summarized together. Table 1: Telic verb, perfective context
Preterit Imperfect Present Perfect Other
low intermediate (n = 38)
high intermediate (n = 18)
advanced (n = 15)
control group (n = 16)
166 (77%) 28 (13%) 14 (6%) 9 (4%)
92 7 2 2
82 (92%) 6 (7%) 1 (1%) 0 (0%)
89 (94%) 6 (6%) 0 (0%) 0 (0%)
(89%) (7%) (2%) (2%)
Table 2: Telic verb, imperfective context
Preterit Imperfect Present Perfect Other
low intermediate
high intermediate
advanced
control group
52 (47%) 50 (45%) 5 (5%) 3 (3%)
13 (25%) 37 (71%) 1 (2%) 1 (2%)
20 (44%) 25 (56%) 0 (0%) 0 (0%)
5 (10%) 43 (90%) 0 (0%) 0 (0%)
low intermediate
high intermediate
advanced
control group
70 (64%) 31 (28%) 3 (3%) 6 (5%)
16 (31%) 31 (61%) 1 (2%) 3 (6%)
7 (16%) 36 (84%) 0 (0%) 0 (0%)
3 (8) 36 (92) 0 (0) 0 (0)
Table 3: Atelic verb, imperfective context
Preterit Imperfect Present Perfect Other
First of all, considering all three tables, it should be noticed that none of the conditions demonstrated a direct transfer from the German grammar. Since the Present Perfect is the dominant tense in the spoken German language, a transfer should have resulted in a marked overuse of the corresponding Spanish verb forms. However, no strong overgeneralization of this tense (below 6% in all
188
Tim Diaubalick and Pedro Guijarro-Fuentes
cases) is found in the data, although there are evidently some individuals who indeed prefer the present perfect. To produce statistically more valid results, a Chi-Square-Test could be performed over a reduced table which summarized perfective tenses (Preterit, Perfect, Pluperfect) against imperfective tenses (Imperfect, Present). The clear dominance of Preterit and Imperfect, opposed to the very marginal role of other forms, justifies this simplification for statistical purposes. The values obtained are illustrated in Table 4. Table 4: Chi-Square-Test on the production data test condition
chi-square-value
telic verb in perfective context
χ2(3) = 5.785; no significance
telic verb in imperfective context
p < 0.001 χ2(3) = 30.408; highly significant difference
atelic verb in imperfective context
p < 0.001 χ2(3) = 68.957; highly significant difference
p = 0.123
It is safe to say that while the combination of telicity and perfectivity did not produce significance, the combinations that involved imperfectivity turned out to be highly significant. Considering the significant results, it is safe to say that the group differences are stronger than the individual ones mentioned above, an observation that should be kept in mind for the discussion section.
5.2 Grammatical Judgment Task The results of the Grammatical Judgment Task resulted in the following average ratings as shown in Figures 2, 3 and 4.
Figure 2: Ratings of habitual and punctual contexts
How do German speakers acquire the Spanish TA-system?
189
Figure 3: Ratings of coercion contexts
Figure 4: Impersonal subject constructions
Furthermore, standard deviations were calculated in order to account for possible individual differences in the ratings. These are listed in table 5. Table 5: Mean Rate (and Standard Deviation) per condition and proficiency level
habitual/punctual (grammatical) habitual/punctual (ungrammatical) coercion (grammatical) coercion (ungrammatical) impers. subject (grammatical) impers. subject (ungrammatical)
low intermediate
high intermediate
advanced
control
1.33 (0.37) 2.31 (0.54) 1.60 (0.43) 1.96 (0.54) 1.76 (0.36) 1.95 (0.47)
1.16 (0.42) 2.55 (0.47) 1.18 (0.30) 2.27 (0.48) 1.72 (0.23) 2.14 (0.48)
1.20 (0.26) 2.61 (0.34) 1.31 (0.25) 2.29 (0.45) 1.82 (0.40) 2.22 (0.57)
1.08 (0.14) 2.57 (0.49) 1.08 (0.12) 2.61 (0.41) 1.78 (0.32) 2.15 (0.37)
190
Tim Diaubalick and Pedro Guijarro-Fuentes
Several One-Way-ANOVAs and Kruskal-Wallis-Tests were conducted to examine the data for possible significances in order to decide if inter-group differences or individual differences were decisive. Additionally, Tukey’s HSD tests were conducted to form subgroups and, thus, to give an answer to research question (i’) and (ii’), i.e., if there are differences to the native speakers and between groups. While the Kruskal-Wallis-Test makes fewer assumptions to the data (an ordinal scale is enough) and might therefore be the more appropriate method for the analysis, it is more convenient for an understanding of the results to present mean values and assuming, thus, an interval scale that allows the ANOVA. Nevertheless, it shall be mentioned that both tests resulted in the same conclusions, i.e., the assumption of an interval scale is not compulsory for the interpretation of the data. The different conditions were compared to answer research question (iii’). In general, coercion and impersonal subject contexts indeed caused more problems than the standard contexts, suggesting a more complex processing of interaction of phases. The only significant category within the standard contexts was the ungrammatical combination of a punctual action with an imperfective verb form (F(3, 82) = 2.781; p = 0.046). In this case, learners rated the item higher than the control group. Regarding the impersonal subject contexts, results are somewhat different, since the control group did not heavily distinguish between grammatical and ungrammatical contexts either, i.e., this condition is a proof for the strong necessity of a control group in aspect studies. Only the grammatical category of specific subjects with a Preterit verb form led to a significance (F(3, 81) = 3,71; p = 0,015), being lower rated by the learners than by the control group. The significance in precisely that condition is a result that should be kept in mind. For purposes of completeness, Table 6 contains the results of the ANOVAs that did not provide significances in standard and impersonal subject contexts. Table 6: ANOVA tests in non-significant categories Condition
F value (ANOVA)
significance measure
standard contexts (grammatical) standard contexts (ungrammatical) impersonal subjects (grammatical)
F(3, 81) = 1.575 F(3, 82) = 1.548 F(3, 81) = 0.675
p = 0.202 p = 0.208 p = 0.570
The important coercion contexts showed significance in almost every condition, as summarized in Table 7. The post-hoc test results (Tukey’s HSD) are also presented in this table.
How do German speakers acquire the Spanish TA-system?
191
Table 7: ANOVA tests for the coercion contexts in the grammatical judgment task Condition
F value (ANOVA)
significance measure
sub-grouping according to Tukey’s HSD
stative, imperfective (grammatical)
F(3, 79) = 8.687
p < 0.001
low interm.
high interm advanced control
eventive, perfective (grammatical)
F(3, 82) = 5.009
p = 0.003
low interm. advanced
high interm. advanced control
grammatical sentences as a whole
F(3, 82) = 9.597
p < 0.001
low interm. advanced
high interm. advanced. control
eventive, imperfective (ungrammatical)
F(3, 83) = 2.473
p = 0.067
(non applicable)
stative, perfective (ungrammatical)
F(3, 81) = 8.711
p < 0.001
low interm. high interm. advanced
control
ungrammatical sentences as a whole
F(3, 83) = 5.902
p = 0.001
low interm. high interm. advanced
high interm. advanced control
As could be seen in the mean values of the coercion conditions (Figure 3), although the low intermediate learners differ at the greatest, the other ones did not reach the native level either. Note that there is only one condition that did narrowly fail significance (eventive, imperfective).
6 Discussion and Conclusion Before turning to the general implications of our findings, we should answer the specific research questions (Section 3.1). (i’) Are there group and individual differences in the use and production of past tenses? As the results reveal, there are definitely differences between native and non-native speakers. The ANOVA-values showed that these differences concern both inter-group and intra-group variations, as the means and the deviations are significantly different in many of the tested conditions. Therefore, it is probable that some kind of transfer or interference from the L1 system is at play. While standard deviations proved that individual differences were attested, too, in all of these groups, these were not high enough to exceed the significant differences between groups. (ii’) Are there group and individual differences among the L2 learners? Yes, there is clearly a visible development with increasing proficiency. This can suggest that learners either employ a target-deviate strategy, as predicted by
192
Tim Diaubalick and Pedro Guijarro-Fuentes
Hawkins and Hattori (2006), to get closer to the Spanish grammar, or that they successfully begin a reassembly task (Hwang and Lardiere 2013; Lardiere 2009). Additionally, as Tukey’s HSD tests revealed, the low intermediate learners and the control group behaved significantly differently in all conditions, while the other two experimental groups showed signs for convergence in some aspects. Since the native level globally is, however, not reached by even the most advanced speakers, we cannot corroborate a full attainment of the target-like feature arrangement. (iii’) Is there any evidence of L2 learners having acquired the uninterpretable features? Especially the coercion condition in the grammatical judgment task leads one to suspect that the uninterpretable features connected to perfectivity that are responsible for a verb rising to AspP are not acquired since even advanced speakers do not reach native level. Although standard contexts seem unproblematic, it is highly probable that this observation is due to domaingeneral strategies not derived from UG that are based on rule-guided learning since no understanding of feature interactions is necessary in these cases and target-deviate strategies can resolve the problems. Furthermore, it was shown that the impersonal subject constructions also cause problems, which shed doubts onto UG-accessibility, so indeed other strategies than UG principles may be at play (cf. Schwartz and Sprouse 2000). In particular, the clearest condition (specific subject, Preterit form), i.e., the undoubtedly grammatical one, led to the greatest significance and a clear distinction between all non-native groups and the native speakers was proven. Our data, therefore, speak in favour of the Interpretability Hypothesis since a verification of a successful reassembly task is not possible. However, the IH predicts not only the inaccessibility of uninterpretable features in the target system, but also a target-deviate strategy employed by the learners to overcome these. In order to support this strategy, some additional analyses must be conducted. Recall that the German grammar makes a major use of lexically-based items to express meanings that in Spanish are coded morphologically. This is why it seems plausible that the strategy employed by the German speakers is also based on the lexicon. That is, the meaning of lexical elements characterizing aspect and time seem more easily acquirable than grammatical features. Table 8 based on both production tasks seems to confirm this assumption. Table 8: verbal form chosen in a sentence that contained a known signal word for the Preterit
Preterit Imperfect
low intermediate
high intermediate
advanced
control group
69 (65%) 37 (35%)
28 (57%) 21 (43%)
28 (64%) 16 (36%)
15 (32%) 32 (68%)
How do German speakers acquire the Spanish TA-system?
193
Table 9: verbal form chosen in those sentences without any signal words
Preterit Imperfect
low intermediate
high intermediate
advanced
control group
64 (63%) 38 (37%)
27 (56%) 21 (44%)
24 (54%) 20 (46%)
20 (42%) 28 (58%)
Learners on all proficiency levels show precisely the opposite distribution of forms as does the control group in those contexts that contained Preterit signals (as for example ayer ‘yesterday’, la semana pasada ‘last week’ or en el año X ‘in the year X’). These signal words are therefore partially misguiding and an orientation towards them clearly presents, thus, a non-native and target-deviate strategy. Conversely, in those sentences which did not contain any adverbial at all, no significance was found. Thus, there is clearly a trigger effect when a signal word appears. This effect, somewhat neglected in the literature (see, e.g., Quesada 2013 who argues in favour of a combination of this effect with orientations towards other features as lexical aspect), resulted highly significant: it thus proves a major influence of the adverbial complements, often taught as guidelines in Spanish classes. According to the presented findings, the German speakers indeed show a different pattern than the English speaking Spanish learners known from previous studies. The specific past tense hypotheses7 are therefore no more than target-deviate strategies predicted by the IH, since the orientation towards the lexical aspect or other linguistic elements produces target-like constructions in some contexts, but does not equal the native strategy. In this case, this implies the interpretable aspectual features as perfectivity, boundness and completeness and their uninterpretable formal reflections, present neither in the English nor in the German grammar. The strategies, however, evidently differ among various kinds of language learners. The IH can only manage to explain the data if the formal uninterpretable features are indeed crucial for an aspectual contrast. Most of the features connected to aspectual marking are of a semantic nature and therefore interpretable, but the uninterpretable features, as explained, are necessary for verb raising (cf. Figure 1 outlined in Section 2.1), therefore the IH can explain the majority of the observations found here. Nevertheless, it depends on the actual existence of a grammatical AspP, on which there is, as presented in previous sections, no agreement on the part of the grammarians. Only the approach that characterizes the Imperfect as imperfective (García Fernández 1999, Leonetti 7 As for example the ones reported in Comajoan (2013): Lexical Aspect Hypothesis, Default Past Tense Hypothesis, amongst others.
194
Tim Diaubalick and Pedro Guijarro-Fuentes
2004) can explain the observed data. The approach that declares the Imperfect as a mere anaphoric tense (Giorgi and Pianesi 2004), on the other hand, fails to explain the totality of the data since it declares anteriority, simultaneity and posteriority as the relevant features, which are indeed present in German, English and Spanish, and should not produce major problems. The FRH (Lardiere 2009, Hwang and Lardiere 2013), in contrast, is capable of explaining why for the German participants, who would start with a full transfer of the feature arrangement known from their mother tongue, it is difficult to map different features onto two forms which in the L1 are expressed by only one tense. However, since a direct transfer is neither observed in the previous studies that examined English speakers nor in the present chapter, the FRH fails to explain why the differences between the German and English learners are as reported. Furthermore, although an acquisitional development in line with the FRH is attested, the native level is not reached in the end, and the problems with the coercion effects, in particular, persist until late stages. We have, therefore, no possibility to corroborate the FRH in all its implications and cannot confirm an eventual mastery of all features. All in all, our study shows that further research might have to be led in another direction. Perhaps the acquisition of the Spanish past tenses presents a phenomenon that is rather connected to the lexicon than to the interpretability of features. Furthermore, future research has to clarify the difference attested between the two subgroups in the Germanic language group. A tentative thought, for example, may be that while English and Spanish, as opposed to German, share the macro-parameter of having morphological marking of aspect (recall the progressive and the “used to”-construction), the difference lies in the microparameter of marking perfectivity (cf. Uriagereka 2007). Thus, another type of parameter-reset might be necessary, comparing German and English learners. This thought, obviously, has to be further developed. Finally, it should also be mentioned that the effect of instruction rather than the L1 effects might have been tested. In that case, the target-deviate strategies would not differ in relation to the nature of the L1 grammar but according to explicitly learned rules8. Since the investigated German university learners presented a rather homogeneous group that partly shared the same language classes in which signal words may have been highlighted, the data do not allow confirming the presented conclusions in a strong way. Rothman (2008) warns that the concepts of rule learning and language acquisition must be kept separated. Especially in the case of temporal markers and signal words, it is 8 See Rothman 2008 for a further discussion on this topic.
How do German speakers acquire the Spanish TA-system?
195
highly probable that in fact domain-general learning of explicit rules is at play, instead of an influence that is exercised by the German grammar. Therefore, it is necessary to expand the study to other groups (like noninstructed learners, and learners with another linguistic background) and other phenomena affected by aspect effects (e.g. Differential Object Marking, future tenses) in order to examine a possible robustness of the observations. On the other hand, it is necessary to compare learners of other L1s in order to verify that the observed patterns are indeed due to German. Additionally, following Comajoan (2005), it is helpful to integrate non-tutored learners, i.e., immigrants that acquire Spanish in a Spanish-speaking environment without the help of a teacher. This group in particular will help to exclude the observation of mere rule-based learning. The present study may have great pedagogical implications; one must rethink the method of distributing lists of signal words which might guide the learner into a wrong direction. Instead, the concept of lexical aspect should perhaps be taught.
References Chomsky, Noam. 2001. Derivation by phase. In: Kenstowicz, Michael (ed.): Ken Hale: A life in language. Cambridge, MA: MIT Press, 1–52. Comajoan, Llorenç. 2005. The Early L2 Acquisition of Past Morphology: Perfective Morphology as an Aspectual Marker or Default Tense Marker? In: Eddignton, David (ed.), Selected Proceedings of the 6th Conference on the Acquisition of Spanish and Portuguese as First and Second Languages. Somerville, MA: Cascadilla Proceedings Project, 31–43. Comajoan, Llorenç. 2013. Tense and aspect in second language Spanish. In Geeslin, Kimberly L. (ed.), Handbook of Spanish Second Language Acquisition. Boston: Wiley-Blackwell, 235– 252. Comrie, Bernard. 1976. Aspect. An Introduction to the study of verbal aspect and related problems. Cambridge: Cambridge University Press. Diaz, Lourdes, Bel, Aurora & Konstantina Beklou. 2008. Interpretable and Uninterpretable Features in the Acquisition of Spanish Past Tenses. In: Liceras, Juana M., Zobl, Helmut & Helen Goodluck (eds.), The role of formal features in Second Language Acquisition. New York/London: Lawrence Erlbaum Associates, 484–512. Domínguez, Laura, Tracy-Ventura, Nicole, Arche, María J., Mitchell, Rosamond, & Florence Myles. 2013. The role of dynamic contrasts in the L2 acquisition of Spanish past tense morphology. Bilingualism: Language and Cognition, 1(1), 1–20. DUDEN. 2006. Duden Band 4. Die Grammatik. Mannheim/Leipzig/Wien/Zürich: Dudenverlag, Bibliographisches Institut & F. A. Brockhaus AG, 7th edition. García Fernández, Luis. 1999. Sobre la naturaleza de la oposicion entre pretérito imperfecto y pretérito perfecto simple. Lingüística Española Actual, LEA 21 (2), 169–188. Giorgi, Alessandra & Fabio Pianesi. 2004. On the speaker’s and the subject’s temporal representation: the case of the Italian imperfect. In: Guéron, Jacqueline & Jacqueline Lecarme (eds.), The Syntax of Time. Cambridge, MA / London: MIT Press, 259–298.
196
Tim Diaubalick and Pedro Guijarro-Fuentes
Hawkins, Roger & Cecilia Y. H. Chan. 1997. The partial availability of Universal Grammar in second language acquisition: the ‘failed functional features hypothesis’. Second Language Research, 13(3), 187–226. Hawkins, Roger & Hajime Hattori. 2006. Interpretation of English multiple wh-questions by Japanese speakers: a missing uninterpretable feature account. Second Language Research, 22(3), 269–301. Hawkins, R., & Sarah Liszka. 2003. Locating the source of defective past tense marking in advanced L2 English speakers. Language Acquisition and Language Disorders, 30, 21–44. Heinold, Simone. 2015. Tempus, Modus und Aspekt im Deutschen. Tübingen:narr. Hwang, Sun Hee & Donna Lardiere. 2013. Plural-marking in L2 Korean: a feature-based approach. Second Language Research 29, 57–86. Lardiere, Donna. 2009. Some thoughts on the contrastive analysis of features in second language acquisition. Second Language Research, 25(2), 173–227. Leonetti, Manuel. 2004. Por qué el imperfecto es anafórico. In: García Fernández, L. & B. Camus Bergarache. (ed.). El pretérito imperfecto. Madrid: Gredos, 481–508. McManus, Kevin. 2011. The Development of Aspect in a Second Language: What Role for the First Language? Saarbrücken: Lambert Academic Publishing. McManus, Kevin. 2015. L1–L2 differences in the acquisition of form-meaning pairings: A comparison of English and German learners of French. Canadian Modern Language Review/ La Revue canadienne des langues vivantes 71 (2), 51–77. Pollock, Jean-Yves. 1989. Verb Movement, Universal Grammar, and the Structure of IP. Linguistic Inquiry 20, 365–424. Quesada, Margaret. 2013. The Primacy of Morphology in the Acquisition of Tense and Aspect in L2 Spanish Narrative Structure. In: Howe, Chad, Blackwell, Sarah E. & Margaret Lubbers Quesada (eds.), Selected Proceedings of the 15th Hispanic Linguistics Symposium, Somerville, MA: Cascadilla Proceedings Project, 62–77. Rothman, Jason. 2008. Aspect selection in adult L2 Spanish and the Competing Systems Hypothesis. When pedagogical and linguistic rules conflict. Languages in Contrast, 8(1), 74–106. Salaberry, M. Rafael & Dalila Ayoun. 2005. The development of L2 tense-aspect in the Romance languages. In: Salaberry, M. Rafael & Dalila Ayoun (eds.), Tense and Aspect in Romance Languages: Theoretical and applied perspectives. Amsterdam/Philadelphia: John Benjamins, 1–33. Salaberry, M. Rafael. 2008. Marking Past Tense in Second Language Acquisition: A Theoretical Model. New York: Continuum. Salaberry, M. Rafael. 2011. Assessing the effect of lexical aspect and grounding on the acquisition of L2 Spanish past tense morphology among L1 English speakers. Bilingualism, Language and Cognition, 14(2), 184–202. Salaberry, M. Rafael & Llorenç Comajoan. 2013. Research Design and Methodology in Studies on L2 Tense and Aspect, Berlin: Mouton de Gruyter, 119–156. Sánchez Prieto, Raúl. 2002. Transferencias e interferencias lingüísticas en el uso de los tiempos verbales de indicativo en español y alemán. In: González Martín, Vicente (ed.), Hacia la unidad en la diversidad. Difusión de las lenguas europeas. Salamanca: Diputación de Salamanca, 467–480. Schwartz, Bonnie D. & Rex A. Sprouse. 2000. When syntactic theories evolve: consequences for L2 acquisition research. In: John Archibald (ed.), Second language acquisition and linguistic theory. Oxford: Blackwell, 156–186.
How do German speakers acquire the Spanish TA-system?
197
Schwenk, Hans-Jörg. 2012. Die Vergangenheitstempora im Deutschen und ihr semantisches Potential. Lublin Studies in Modern Languages and Literature 36, 35–49. Shirai, Yasuhiro. 2013. Defining and coding data: Lexical aspect in L2 studies. In: Salaberry, M. Rafael & Llorenç Comajoan (eds.), Research Design and Methodology in Studies on L2 Tense and Aspect, Berlin: Mouton de Gruyter, 271–308. Slabakova, Roumyana & Silvina A. Montrul. 2003. Genericity and Aspect in L2 Acquisition. Language Acquisition 11 (3), 165–196. Slabakova, Roumyana & Silvina A. Montrul. 2008. Aspectual Shifts: Grammatical and Pragmatic knowledge in L2 Acquisition. In: Liceras, Juana M., Zobl, Helmut & Helen Goodluck (eds.), The role of formal features in second language acquisition, New York/London: Lawrence Erlbaum Associates, 106–140. Stutterheim, Christiane von & Mary Carroll. 2006. The impact of grammaticalised temporal categories on ultimate attainment in advanced L2–acquisition. In: Heidi Byrnes (ed.), Educating for advanced foreign language capacities. Constructs, curriculum, instruction, assessment, Washington DC: Georgetown University Press, 40–53. Tsimpli, Ianthi M., & Maria Dimitrakopoulou. 2007. The interpretability hypothesis: Evidence from wh-interrogatives in second language acquisition. Second Language Research, 23(2), 215–242. Uriagereka, Juan. 2007. Clarifying the Notion of ‘Parameter’. Biolinguistics, 1, 99–113. Vendler, Zeno. 1957. Verbs and Times. The Philosophical Review 66 (2), 143–160. White, Lydia. 2003. Second language acquisition and Universal Grammar. Cambridge: Cambridge University Press. Zagona, Karen. 2007. Some effects of aspect on tense construal. Lingua, 117(2), 464–502.
Pedro Guijarro Fuentes, M. Carmen Parafita Couto, Rocio Pérez-Tattam, and Myrthe Wildeboer*
8 The non-native acquisition of Spanish (un)interpretable features by Dutch L1 learners Abstract: This article investigates Dutch learners (N = 27) of non-native Spanish and assesses their knowledge of (un)interpretable features (i.e., number, gender, collocation of adjectives) through their performance under experimental conditions. Current debates about second language acquisition (SLA) have led to different hypotheses: (i) the Interpretability Hypothesis (Hawkins and Hattori 2006) which claims that uninterpretable features will not be completely acquired, (ii) the Interface Hypothesis (Sorace 2011) which claims the exact opposite and supports a full acquisition of uninterpretable features, and (iii) the Feature Reassembly Hypothesis (Hwang and Lardiere 2013; Lardiere 2009), suggesting that acquiring (un)interpretable features is a very demanding task for L2 learners. To test these hypotheses, this study explores the L2 acquisition of both types of features by looking at certain Spanish features such as number, gender, and focus/contrast that are related to the determiner phrase. The Feature Reassembly Hypothesis seems to best explain the results of our two experimental tasks, although the results do not provide clear-cut support for any of the hypotheses under study. The variability in our data is discussed in relation to Rebuschat and Williams’ (2012) proposal that explicit and implicit learning may affect language acquisition in different ways.
1 Introduction Not all features in a language have the same status and therefore a distinction can be made between interpretable and uninterpretable features (Chomsky 2007). * The names of the authors are in alphabetical order. All authors contributed equally to this article. Pedro Guijarro-Fuentes, University of the Balearic Islands M. Carmen Parafita Couto and Myrthe Wildeboer, Leiden University Rocio Perez-Tattam, Swansea University
200
Pedro Guijarro Fuentes et al.
Uninterpretable features are features that only carry pure syntactic properties, like gender and number concord between the determiner and the noun, while interpretable features, on the other hand, also carry meaning and thus are more relevant to the semantic and pragmatic components in a language. This study follows previous studies by Guijarro-Fuentes, specifically replicating Guijarro-Fuentes’ (2014a) study on the acquisition of [+/–] interpretable features in second language acquisition (henceforth SLA) of Spanish to order to investigate the development of both uninterpretable features (i.e., gender/number determiner-noun concord)1 and interpretable features (i.e., adjectives can have a Focus/Contrast feature2, perhaps due to interface conditions, triggering movement to a FocusP in pre-nominal position) by L1 Dutch speakers. These phenomena differ or do not exist in Dutch (more on this in Section 2). Guijarro-Fuentes (2014a, 2014b) highlights different hypotheses in recent SLA and bilingualism research about the degree of acquisition of either interpretable (semantic) or uninterpretable (syntactic) features. One of these hypotheses is the Interpretability Hypothesis (Hawkins and Hattori 2006; Tsimpli and Dimitrakopoulou 2007), according to which uninterpretable features (narrow syntax) that are absent in the learner’s L1 are no longer available for adults in SLA, because they are susceptible to critical period constraints (e.g., Johnson and Newport 1989; Smith and Tsimpli 1995, among others). This will lead to incomplete acquisition, as well as computation and representation deficits of second language uninterpretable features. Interpretable features (semantic/ pragmatic) remain accessible for second language learners, regardless of whether they are present or absent in the learner’s L1. Tsimpli and Dimitrakopoulou (2007) explain this difference in accessibility by claiming that interpretable features are represented in both the language system and in the LF-interface. Therefore, the interpretable features in LF will be accessible top-down and bottomup and consequently not be subject to critical period constraints. The second hypothesis, the Interface Hypothesis (Sorace 2011), claims the exact opposite – namely, that interpretable features give rise to more problems than uninterpretable features. Variability in a second language is not due to deficits in uninterpretable features, but rather to deficits in interpretable features. 1 Whereas the number feature is usually regarded as an interpretable feature of N’s as it adds to semantic interpretation, it is uninterpretable in D and A. Although it has no bearing to the purpose of the present study, recall that in Chomsky’s (2015) recent work these features are used for feature checking or for AGREE: an interpretable feature checks an uninterpretable feature. However, the interpretable feature of pre-nominal or postnominal adjectives seems to be of a different type (more on this Section 2). 2 That is, this is an LF interpretable feature that is not realized in all adjectives, and that in turn makes certain adjectives move, while nouns remain in situ (Demonte 1999, 2008).
The non-native acquisition of Spanish (un)interpretable features
201
This hypothesis has been studied in adult L2 learners (Belletti, Bennati, and Sorace 2007), simultaneous bilinguals (Sorace et al. 2009; Montrul 2004), and attrited L1 speakers (Tsimpli, Sorace, Heycock, and Filiaci 2004) (also see recent studies on the topic in Rothman and Slabakova 2011). For instance, Belletti, Bennati and Sorace (2007) tested the Interface Hypothesis by studying the nonnative competence of near-native L2 Italian participants at a very high level of attainment. English-Italian bilinguals performed elicitation and spontaneous production tasks designed to elicit null and overt pronominal subjects, structures which are sensitive to syntactic and pragmatic constraints. Results revealed different sources of divergence between the native and non-native grammars. Participants made use of remaining non-native interpretable features, which are located in the syntax-discourse interface, of L2 Italian, which shows a possible influence of English, their first language. Belletti et al. (2007) have provided evidence for the need to formalize the syntax-discourse interface within a model of linguistic theory, as some of the remaining difficulties experienced by speakers at this level are located within this interface. A third hypothesis, the Feature Reassembly Hypothesis (FRH, henceforth) (Hwang and Lardiere 2013; Lardiere 2009), suggests that second language readjustment is a slow and arduous acquisition task for both interpretable and uninterpretable features. A second language learner who wants to assemble particular lexical items of the second language is required to reconfigure features from the way they are represented in the first language into new configurations on possibly quite different types of lexical items in the second language (Lardiere 2009). Spinner (2013) tests the FRH in English learners of Swahili and their ability to acquire grammatical gender and number in this language. In Swahili, the root of every noun displays gender and number markers. 38 English students of Swahili took part in an elicited production task. Results of her study demonstrated that English learners of Swahili were unable to deconstruct the prefix of the noun as a gender and number marker. The FRH proposes two possible explanations for this result. The first explanation is about the location of the number marker in the L1. The number marker in English – as all other inflectional markers – appears as a suffix. Perhaps the English learners searched for a suffix marker and when no overt marker was detected, they analysed it as a null form. The second explanation implies that it is difficult to identify either a gender marker or a number marker when they are both conflated on a single morpheme (Spinner 2013). Among recent studies testing this hypothesis, Spinner (2013) suggests that the FRH best explained her results However, Ahern, Amenos-Pons, and Guijarro-Fuentes (2014) also have pointed out that feature reassembly in L2 Spanish is not completely trouble-free, even in cases of shared features. A group of adult French (N = 48) and English (N = 40)
202
Pedro Guijarro Fuentes et al.
L2 learners of Spanish were tested on their ability to detect and interpret variation in sentence meaning linked to indicative/subjunctive mood alternation in certain types of Spanish if-conditional sentences. French shares these features of morphological mood whereas English does not3. Participants were asked to conduct an interpretation task of twenty if-conditional constructions and ten although-concessive multiple-choice items. The authors found evidence of feature re-assembly of formal and semantic properties, but with upper-intermediate and advanced L2 acquisition the interface integration becomes an issue, regardless of L1. Integration of semantics, morpho-syntax and pragmatics is a source of variability, regardless of overall proficiency. Guijarro-Fuentes (2014a, 2014b) and other researchers have compared these different learnability hypotheses in order to see which one would account best for their data. Guijarro-Fuentes looked at the L2 acquisition of [+/–] interpretable features (gender, number, focus/contrast in adjective placement) by 34 French (advanced and intermediate) learners of Spanish (Guijarro-Fuentes 2014a) and by 32 Chinese (advanced and intermediate) learners of Spanish (Guijarro-Fuentes 2014b). Whereas French and Spanish are both Romance languages and share certain features related to the DP, Chinese and Spanish do not share DP features. Chinese only allows for pre-nominal adjectives and has no nominal agreement (Guijarro-Fuentes 2014b). The French and Chinese L2 learners of Spanish were tested using a Grammaticality Judgment Task (GJT) and Semantic Context-Based Collocation Task (SCBCT), amongst others, and their performance was compared to a control group of 12 Spanish native speakers. The results of the experiments from both studies show that both the French and Chinese learners of Spanish were unable to (re)select the proper uninterpretable and interpretable features, notwithstanding the resemblances between French and Spanish. According to Guijarro-Fuentes, this suggests that both uninterpretable and interpretable features are not totally accessible by adult L2 learners, whether they are present in the L1 (Guijarro-Fuentes 2014a) or not (GuijarroFuentes 2014b). This is contrary to the claims of the Interpretability Hypothesis regarding accessibility of uninterpretable features (which depends on whether they are present or not in the L1), and appears to be more in line with the predictions of the Feature Reassembly Hypothesis with regards to a gradual process of L2 reanalysis for both [+/–] interpretable features. Prentza and Tsimpli (2013) compare the Interpretability Hypothesis and the Interface Hypothesis by studying the optionality of second language pronominal interpretation, focusing on 3 Due to space limitations, we cannot go into many details, but interested readers are referred to this study in question where it is shown in depth in which aspects Spanish and French differ and why the acquisition of Spanish should be difficult even for the L1 French learners.
The non-native acquisition of Spanish (un)interpretable features
203
adjunct CP clauses and VP-coordination structures. 72 Greek learners of English and a control group of 25 English native speakers participated in the experiment. Two production tasks were designed: a sentence completion task and a cloze task. The data showed that both the intermediate and the advanced learners produced significantly more ungrammatical null subject pronouns than the English native speakers. Recall that, on the one hand, the Interface Hypothesis claims that the syntactic aspect of the phenomenon can be acquired, but the processing load of coordinating different types of information may inhibit L2 acquisition and cause target-deviant performance. On the other hand, the Interpretability Hypothesis suggests that the narrow syntactic features will remain unavailable for the L2 learners, due to maturational effects. The Interpretability Hypothesis explains the results better because of the fact that the interface is not solely responsible for the L2 optionality in pronominal production and interpretation. Thus, pronouns in L2 should first be considered as a challenge for the grammar and secondly for the interface functions (Prentza and Tsimpli 2013). In order to evaluate all the above-mentioned hypotheses, this study will examine the development of certain interpretable and uninterpretable features that are related to the determiner phrase (DP) by L1 Dutch adults learning Spanish as an L2, replicating Guijarro-Fuentes’ (2014a) study of the acquisition of [+/–] interpretable features in L2 Spanish. Similarly to the Guijarro-Fuentes (2014a) study, this study will focus on the uninterpretable features gender and number, and the interpretable Focus/Contrast feature vs. syntactic knowledge with respect to adjective placement. The contributions of our study are firstly to add a different group of L2 learners with the view of testing Guijarro-Fuentes’ (2014a, 2014b) findings in other language pairs; and secondly to test the abovementioned hypotheses in the Dutch-Spanish language pair, where gender and number are not as fully comparable as in the French-Spanish language pair (more in Section 2). Thus, our predictions will be based on Guijarro-Fuentes (2014a), but we anticipate some differences in our results given the asymmetry between Dutch and French compared to French and Spanish. Finally, we test the contrast between interpretable and non-interpretable features within one phenomenon: syntactic vs. semantic knowledge of adjective placement, as in Guijarro-Fuentes (2014a, 2014b). This article is organized as follows: Section 2 will discuss the theoretical considerations of the number and gender concord, and the collocation of adjectives; in Section 3 we will look at the acquisition of interpretable and uninterpretable features, and present our research questions and predictions; Section 4 will be an overview of the methodology and Section 5 will be an overview of the results. At the end of the article there will be a conclusion and discussion.
204
Pedro Guijarro Fuentes et al.
2 Theoretical considerations 2.1 Word order in the Determiner Phrase Dutch has a subject-object-verb (SOV) order with verb-second (V2) (Sybesma 2002), whereas Spanish has a dominant subject-verb-object (SVO) order, with some exceptions (Zagona 2002). Despite surface differences in word order, it is a widely accepted hypothesis that in both Romance and Germanic, pre-nominal adjectives are generated in pre-nominal position and post-nominal adjectives are derived through noun movement. This means that the surface order is derived when the noun moves leftwards to a higher functional projection, resulting in a noun-adjective word order (Kayne 1994; Cinque 1999; Alexiadou, Haegeman and Stavrou 2007). In Spanish and other Romance language such as French, prenominal and post-nominal adjective word orders within the DP are both possible. We adopt Guijarro-Fuentes’ (2014) proposed analysis of the internal structure of Spanish DPs with pre-nominal adjectives vs. DPs with post-nominal adjectives, as illustrated in Figure 1.
Figure 1: A syntactic tree of pre-nominal adjectives (left tree) and post-nominal adjectives (right tree) (Guijarro-Fuentes 2014a: 104)
According to Guijarro-Fuentes (2014a), Spanish DPs instantiate the functional features Word Marker (WM) (Bernstein 1993) and Num. Uninterpretable gender and number feature checking in these categories trigger noun raising, as shown
The non-native acquisition of Spanish (un)interpretable features
205
in Figure 1. If the adjective merges in an adjunct position to NP, the noun raises over the adjective, giving rise to post-nominal adjective word order. In cases of pre-nominal word order, the noun rises to the same position (Num) but the AdjP adjoins to NumP. In Spanish and other Romance language such as French, many adjectives can vary between a pre-nominal and a post-nominal position. According to Demonte (1999, 2008), Spanish adjectives in post-nominal position have an intersective/ restrictive interpretation; they refer to a subset of properties on the noun. Thus example (1a) refers to only the brave women and not the cowardly ones. Spanish adjectives in pre-nominal position have a non-intersective/non-restrictive interpretation; they refer to the whole of objects that are denoted by the noun. Thus example (1b) applies to all women, since the assumption is that all women are brave. (1)
a.
Las mujeres valientes (only intersective/restrictive interpretation: brave person)
b.
Las valientes mujeres (non-intersective/non-restrictive interpretation) ‘The brave women’
Demonte (1999, 2008) attributes these added interpretive semantic properties in Spanish to the presence of a [Focus/Contrast] interpretable feature in prenominal adjectives, an analysis adopted in Guijarro-Fuentes (2014b) to contrast adjective placement in Spanish and Chinese. This feature triggers movement of the adjective to a pre-nominal position in Spanish, as shown in Figure 2. The placement of adjectives in Spanish depends on the existence of extra functional categories besides the NP and the DP: [nP] and [FocP], both positioned above NP, as shown in Figure 2. As mentioned previously, the noun raises to Num (or the head of nP) in Spanish, so post-nominal adjectives must have a restrictive interpretation (Demonte 1999, 2008). Thus Spanish expresses semantic differences such as restrictive vs. non-restrictive interpretation of adjectives syntactically by means of pre-nominal and post-nominal adjective word orders in the DP. At the same time, some quality adjectives in Spanish (i.e., temporal modifiers, nationality, shape, colour, origin or relational) only appear in one position, usually post-nominal but sometimes pre-nominal. This is due to the lack of a focus/ contrast interpretable feature (examples 2a–d), explaining why they are only allowed in one fixed syntactic position:
206
Pedro Guijarro Fuentes et al.
Figure 2: A syntactic tree representing the extra functional categories associated with Focus/ Contrast (Guijarro-Fuentes 2014b: 150)
(2)
a.
La comida italiana ‘The Italian cuisine’
b. *La italiana comida ‘The Italian cuisine’ c.
El cálido sol ‘The warm sun’
d. *El sol cálido ‘The warm sun’ In Dutch, pre-nominal adjective word order is the only possible order within the DP. This has been explained in terms of lack of overt noun movement (Cinque 1994), but given that Dutch also has uninterpretable gender and number features that can trigger noun raising, it could also be due to the noun moving to a lower position than the adjective (Guijarro-Fuentes 2014, but also Cinque 1994, 2010, among others). In contrast to Spanish, it cannot express semantic differences such as restrictive vs. non-restrictive interpretation of adjectives syntactically by means of pre-nominal and post-nominal adjective word orders in the DP.
2.2 Agreement in gender and number At a purely descriptive level, both Dutch and Spanish distinguish two types of gender. Dutch differentiates between common gender, nouns that are always
The non-native acquisition of Spanish (un)interpretable features
207
preceded by the Dutch article de, and neuter gender, nouns that are always preceded by the Dutch article het. Besides definite determiners, other elements inside and outside the DP are marked with grammatical gender, for example demonstratives, relative pronouns, first person plural possessives, wh-phrases and attributive adjectives. The Dutch indefinite article een can precede both types of nouns. The Dutch definite plural nouns are always preceded by the article de, regardless of gender. Thus, the gender distinction disappears with plural nouns. Attributive adjectives are inflected with a schwa in all cases except with singular, indefinite neuter nouns (see for example Figure 3).
Figure 3: Examples of Dutch common and neuter nouns in combination with definite/indefinite articles and attributive adjectives (Orgassa 2009: 64; also Kester 1996)
In order for a learner to determine the gender of Dutch nouns and then the corresponding correct form (with or without a schwa), it is necessary to know whether the DP contains a definite or indefinite determiner or whether it is a plural. In other words, the syntactic structure of the DP is a cue to the gender of the noun in Dutch. According to Unsworth (2008) the uninterpretable gender features of determiners and adjectives are checked by the interpretable gender feature of the Dutch nouns. The noun moves covertly (i.e. at Logical Form, LF) to the head of the Number Phrase, to check the features of the adjective in a spec–head relation (in Romance languages this movement is overt), and to the head of the DP, D, to check the features of the determiner in a head–head relation (Unsworth 2008), see Figure 4. Spanish has a more transparent gender system. The vast majority of nouns that end with an –a are feminine and nouns that end with an –o are masculine (Teschner and Russell 1984). The determiner and the adjective concord with the gender of the noun in the same way. The feminine determiner and the majority of feminine adjectives take an –a, whereas the masculine determiner and the majority of masculine adjectives take an –o (see examples 3 and 4). Both Dutch and Spanish distinguish between singular and plural nouns. Pluralization in
208
Pedro Guijarro Fuentes et al.
Figure 4: A syntactic tree of the Dutch DP (Unsworth 2008: 369)
Figure 5: A syntactic tree of the Spanish DP (Rothman et al. 2010: 3)
Spanish is indicated by –s on the noun, the determiner and the adjective (see example 4) (Teschner and Russell 1984). (3)
Esta mesa negra ‘This black table’
(4)
Estos niños divertidos ‘These funny children’
As mentioned earlier, functional categories, such as Number Phrase and Word Marker are situated between the DP and the NP, heading their own phrases and checking and valuing number and gender features, respectively (Bernstein 1993). Figure 5 illustrates the internal structure for Spanish DPs.
The non-native acquisition of Spanish (un)interpretable features
209
3 Interpretable and uninterpretable features and language acquisition Considering the acquisition of interpretable and uninterpretable features, monolingual Spanish-speaking children master gender agreement in noun phrases with almost 100% accuracy by the age of 3 (Pérez Pereira 1991). However, gender marking and agreement is problematic for non-native (L2) adult learners, especially in spontaneous and elicited oral production, even at advanced levels of proficiency. This has been documented not only in L2 Spanish (Fernandez 1999; Hawkins and Franceschina 2004; Franceschina 2005; McCarthy 2007), but also in L2 French (Carroll 1989; Dewaele and Veronique 2001; Gess and Herschensohn 2001; Guillelmon and Grosjean 2001) and in L2 Dutch (Sabourin, Stowe and de Haan 2006). It is generally accepted that second language learners whose first language lacks grammatical gender find gender agreement more difficult to master than second language learners whose first language possesses gender (Franceschina 2005; Sabourin et al. 2006). However, there is counterevidence from other L2 and bilingual studies. For instance, bilingual children learning French and German (both languages have gender marking) seem to acquire grammatical gender at the same rate as monolingual children (Müller 1994). Bruhn de Garavito and White (2003) studied the L2 acquisition of gender in Spanish by French-speaking learners, and compared their results to those of Hawkins (1998) – he investigated English-speaking learners of L2 French, in which only the L2 has gender marking. The results of the French-speaking learners of Spanish and of the English-speaking learners of French were in fact very similar, with both groups showing a similar gender error rate of around 30% (Bruhn de Garavito and White 2003). Research about adjective placement is less extensive. Previous studies on the topic include Gess and Herschensohn (2001) who studied the acquisition of English L2 learners of French, and Parodi, Schwartz and Clahsen (2004), who looked at L2 acquisition of German by Korean, Turkish, Italian, and Spanish learners. According to both studies L2 learners can attain a target like order. Other more recent studies, which specifically focus on the semantic constraints related to adjective placement, reveal that L2 learners of English can attain target knowledge of semantic consequences of adjective placement in French and in Spanish (Anderson 2001, 2007a, 2007b, 2008; Rothman et al. 2010). Anderson’s studies focus on the development of native-like knowledge of adult L2 French learners. His studies provide evidence that L2 French learners have acquired the properties that are related to the syntactic-semantic interface, while these properties were undetermined in the input received by the L2 learners.
210
Pedro Guijarro Fuentes et al.
Rothman et al. (2010), who observed English learners of Spanish, examined the acquisition of new L2 DP features with 21 English intermediate learners of Spanish and 24 English advanced learners of Spanish, and compared their data to a Spanish control group of 15 monolingual Spanish speakers. Rothman et al. (2010) investigated whether the English learners of Spanish performed in line with the native control group across experiments and across features. According to their findings, English advanced and intermediate learners of non-native Spanish show evidence of having attained both the uninterpretable features related to the Spanish DP and the interpretable features such as the collocation of adjectives. They concluded that features within syntactic-semantic interfaces are obtainable. And finally, as described in previous sections, the work of Guijarro-Fuentes (2014a, 2014b) looks at the L2 acquisition of [+/–] interpretable properties of the Spanish DP (gender, number, adjective placement in connection to semantic interpretation) by French and Chinese learners.
3.1 Research questions and predictions In this study, we test the Interface Hypothesis, the Interpretability Hypothesis and the Feature Reassembly Hypothesis with regard to the L2 acquisition of [+/–] interpretable properties of the Spanish DP (gender, number, syntactic knowledge of adjective placement, focus/contrast, i.e. adjective placement in connection to semantic interpretation) by Dutch learners. A successful L2 acquisition should include the following aspects: (a) All nouns have grammatical gender and number features, and there is number and gender agreement between nouns, determiners and adjectives (which requires the acquisition of new uninterpretable features, i.e., [uGender] and [uNumber]). Dutch possesses grammatical gender and number features and the functional category where they are checked, but the gender distinction disappears with plural nouns, as explained previously. Dutch learners of Spanish have to generalize gender and number marking to all NPs. (b) Adjectives in Spanish can be either pre- or post-nominal. Adjectives in Dutch are strictly pre-nominal due to the noun moving to a lower position than the adjective, as explained previously. Dutch learners need to learn that the noun rises over the adjective, giving rise to post-nominal adjective word order in Spanish. They also need to acquire the Focus/Contrast feature that is associated with non-restrictive interpretation in Spanish, giving rise to pre-nominal adjective word order in Spanish. The present study thus addresses the following research questions and makes the following predictions:
The non-native acquisition of Spanish (un)interpretable features
211
(1) Are there differences in performance between native and non-native speakers of Spanish across uninterpretable and interpretable features (i.e. gender, number, adjective collocations)? The Interpretability Hypothesis would predict difficulties with uninterpretable Gender and Number features, and with syntactic knowledge of adjective placement, given the differences between the two languages, but not with interpretable Focus/Contrast features, which remain available regardless of whether they are present or absent in the L1. The Interface Hypothesis would predict difficulties with interpretable Focus/Contrast features (semantic/pragmatics interface) rather than uninterpretable Gender and Number features, or syntactic knowledge of adjective placement (narrow syntax). The Feature Reassembly Hypothesis would predict problems with both uninterpretable and interpretable features, due to the reconfiguration of features from the way in which they are represented in the first language into new configurations, on different types of lexical items in Spanish. (2) Are there differences in performance between native and non-native speakers of Spanish across uninterpretable and interpretable features in the same construction (i.e. syntactic vs. semantic knowledge of adjective placement)? Were L2 learners able to select the appropriate meaning or syntactic position for each and every adjective, based on the syntactic position in which the adjective was provided in the test item (experiment 1)? Or was their selection based on the meaning given in the context (experiment 2)? Testing these three learnability hypotheses for L2 acquisition is the central aspect of our study. At the same time, we are dealing with performance data from participants who have learned Spanish in very different environments (e.g. natural settings – L1 vs. institutional settings – L2). Therefore, we also take into account explicit vs. implicit learning of uninterpretable and interpretable features. Based on our experience and knowledge on the content of L2 Spanish curriculum and textbooks, the features of interest of this study which are explicitly taught in a classroom context are gender and number, which are uninterpretable on determiners and adjectives but interpretable on nouns. As to the position of Spanish adjectives, certain adjectives are explicitly taught as being lexically specified to appear in only one position. Some of these adjectives are very frequent in the textbooks and in the classroom context (e.g., origin, colour), others are less so (i.e., relational adjectives). With regard to adjectives which have a focus/contrast interpretable semantic feature, only adjectives that allow a distinct target interpretation for each position, pre-nominal or postnominal, were included in the present study, avoiding other adjectives (such as pobre ‘poor’) which are very often taught to L2 learners as being lexically
212
Pedro Guijarro Fuentes et al.
specified to appear in either position depending on their semantic interpretation (i.e., un hombre pobre ‘a poor man’ vs. un pobre hombre ‘miserable man’)4. Thus, since this semantic contrast (see examples in (1)) is almost absent in the foreign language class, it is highly improbable for it to have been acquired by learning explicit rules. In the process of acquiring a language, a distinction can be made between implicit learning (acquiring knowledge without intending to and without becoming aware of the knowledge that is acquired) and explicit learning (instructed to look at rules or patterns) (Rebuschat and Williams 2012). This distinction can help us understand which features of the L2 are amenable to explicit instruction and to what extent this amenability interacts with individual characteristics such as one’s first language (L1) or language-related cognitive capacities, and to understand how different learning contexts affect development (Andriga and Rebuschat 2015).
4 Method In this study we adopted the same (or a very similar) methodology as GuijarroFuentes (2014a, b) in terms of the experimental design (Grammaticality Judgment Task and Context Based Collocation Task) and the statistical analyses of the data.
4.1 Participants 27 Dutch learners of Spanish participated in the experiment5. There were 22 female and 5 male participants and they were between 18 and 34 years old (mean age = 21.07). Participants were students of the University of Leiden and the University of Utrecht. The highest level of education reported ranged from high school to a Master’s degree. The length of exposure is quite broad: most of the participants learned Spanish as an adult (N = 14), some of them learned Spanish in high school (N = 12) or in primary school (N = 1). The control group consisted of 20 monolingual Spanish speakers. 17 of them were born in Spain, 2
4 Judy, Guijarro-Fuentes and Rothman (2008) provide a brief explanation along those lines too. Besides, albeit for L2 French, for those readers interested in the way certain adjectives are taught and the deterministic way they are acquired, we refer them to Anderson (2007b). 5 In total we had 31 learners, 3 of which learned Spanish before the age of two (therefore no sequential learners) and one scored at an advanced level (unable to achieve statistical results with one participant). Those were excluded from the analysis, resulting in 27 learners.
The non-native acquisition of Spanish (un)interpretable features
213
in Colombia and 1 in Nicaragua. There were 14 female and 6 male monolingual Spanish speakers and they were between 25 and 51 years old (mean age = 31.2). The highest level of education of the control group was either a Bachelor’s or a Master’s degree. All participants- monolingual and L2 speakers- completed a standardized Spanish proficiency test (an adapted version of the DELE) that consisted of a multiple choice test and a cloze test. The multiple choice test consisted of 30 sentences, with a gap either at the end or in the middle of the sentence. There were four possible answers per sentence. The cloze test consisted of a text of 340 words with 20 gaps in it. For each gap there were 3 answer possibilities. In total, the proficiency test consisted of 50 items. Based on the composite score results, the participants were divided into three levels of proficiency: intermediate, upper-intermediate and advanced (see Table 1). Table 1: The proficiency of the Dutch learners of non-native Spanish L2 learners (N = 27) Composite score Advanced
Range N Mean
47–50 0 0
Upper-intermediate
Range N Mean
37–46 13 40
Intermediate
Range N Mean
23–36 14 29,3
4.2 Materials The experiment consisted of three parts: an ethno-linguistic background questionnaire, a proficiency test, and an experimental test. The ethno-linguistic background questionnaire contained questions about the participants’ age, gender, occupation, education, language use, and self-reported language proficiency. The experimental test consisted of two linguistic tasks, namely a Grammaticality Judgment Task (GJT) designed to test only uninterpretable features (narrow syntax) and a Context Based Collocation Task (CBCT) intended to test interpretable features (syntax-semantic interface). The grammaticality judgment task contained 12 different conditions (see Table 2). For each condition there
214
Pedro Guijarro Fuentes et al.
Table 2: The 12 experimental conditions in the Grammaticality Judgment Task Conditions
Examples
1. Completely ungrammatical (distractors)
Yo escriben dos cartas. ‘I write (pl) two cards.’
2. Completely grammatical (distractors)
Ayer nosotros jugamos en el parque. ‘Yesterday we played in the park.’
3. Good noun-adjective agreement
Ellas prefieren comida fresca. ‘They prefer fresh food.’
4. Bad gender noun-adjective agreement
¿Quién preparó este plato deliciosa? ‘Who prepared this delicious (f) dish (m)?’
5. Bad number noun-adjective agreement
Su vida académicas es muy divertida. ‘His academic (pl) life (sg) is very funny.’
6. Good determiner-noun agreement
La banda es muy popular. ‘The band is very popular.’
7. Bad gender determiner-noun agreement
Iván necesita lavar el ropa. ‘Iván needs to wash the (m) clothes (f).’
8. Bad number determiner-noun agreement
El platos son de cerámica. ‘The (sg) plates (pl) are ceramic.’
9. Good postnominal adjective
Tengo un lápiz azul. ‘I have a blue pencil.’
10. Bad pre-nominal adjective
No me gusta esa fea casa. ‘I hate that ugly house.’
11. Bad postnominal adjective
Ese hombre es un soldado mero ‘That man is a mere soldier.’
12. Good pre-nominal adjective
El supuesto asesino fue arrestado ayer. ‘The alleged murdered was arrested yesterday.’
were five sentences. So, there were 60 items6 in total. The gender and number agreement conditions only contained transparent or canonical nouns (i.e. nouns ending in –o, which designates masculine gender, or nouns ending in –a, which usually designates feminine gender) – because of their overt morphophonemic cues to gender, canonical nouns present less difficulties in this respect than opaque or non-canonical nouns, which have consonant or other vowel endings. 6 Space limitations prevent us from including an appendix with all the test conditions for both tasks. The examples provided in the main text are a fair representation of the items for each condition. For clarification to one of the reviewers, in the GJ task we included conditions in which we tested obligatory adjective placement (syntax/uninterpretable) in order to draw comparisons with other conditions in the CBCT which tested variable adjective placement (semantic/interpretable).
The non-native acquisition of Spanish (un)interpretable features
215
Participants were instructed to judge each sentence on its grammaticality. Participants could judge the sentence as “completely grammatical”, as “ungrammatical, but unsure how to correct it” or as “completely ungrammatical”. For sentences not classified as “grammatical” the participants had to write down what they thought the correct grammatical sentence should be – their answers are not analysed in this article due to limitations of length and scope. For the purposes of the analysis, we only included answers where the participants demonstrated a correct judgment of what was wrong in the sentence. Answers where the participants thought something else was wrong were excluded, as they were considered an incorrect judgment of the sentence. The context based collocation task consisted of 10 sentences with an adjective between brackets at the end. In these sentences there was a gap before and after the noun, leaving both possibilities up to the participants. In designing this task, there were five contexts supporting a pre-nominal adjective and five contexts supporting a post-nominal context. Recall that all quality adjectives included in this task could potentially occur in both contexts, only the semantic context will determine in which position the adjective can appear. See below for examples or each context. All sentences with translations are at the end of the chapter as an appendix. Contexts supporting pre-nominal adjective En general, la dinastía Ming era una de las más fuertes y exitosas en la historia del mundo. Los ____ Ming____ conquistaron muchas tierras. (aventurero) “adventurous” Contexts supporting post-nominal adjective No todo chico en mi escuela es atlético, pero los____ chicos ____ corren 7 millas cada día. (fuerte) “strong” All the items for both tasks were put in an online survey by means of the Qualtrics platform7. The order of the proficiency tests and the experimental tasks was randomized. The sentences within the multiple choice test, the grammaticality judgment task, and the context based collocation task were also randomized, to exclude any learning effect. The order of the sentences for the cloze test was always the same, since we needed to follow the structure of the text. The ethnolinguistic background questionnaire was given after all other tasks had been completed, in order to avoid making the participants aware of their language use in the experimental tasks.
7 httpsː//uleidenss.eu.qualtrics.com/ControlPanel/
216
Pedro Guijarro Fuentes et al.
4.3 Procedure and data analysis The link of the online survey was sent to the participants by e-mail. The text needed for the cloze test was attached as a pdf-file in that e-mail. We asked participants to fill in the survey as soon as possible and asked them to reply when they had finished the survey. Participants received a remuneration of 5 euros for their participation. In the data analysis, we compared the number of correct responses provided by the intermediate L2 learners, upper-intermediate L2 learners and native speakers of Spanish in the GJT and the CBCT. To assess the variance of answers provided by each group in the GJT, we used the standard deviation, which is represented by the error bars on the graphs. To determine whether there were any statistically significant differences, we compared the performance of the intermediate L2 learners, upper-intermediate L2 learners and the native speakers of Spanish using a one-way ANOVA. All analyses used a significance level of α = 0.05.
5 Results 5.1 Group results 5.1.1 Grammaticality judgment task The first part of the grammaticality judgment task tested for the acquisition of L2 gender and number features, which are uninterpretable on determiners and adjectives, but interpretable on nouns. Figure 6 shows the average numbers of tokens (N = 5) correctly judged by each of the three groups for the following types of sentences: (a) good noun-adjective agreement, (b) bad gender nounadjective agreement, (c) good determiner-noun agreement, and (d) bad gender determiner-noun agreement. The conditions “good noun-adjective agreement” and “good determiner-noun agreement” refer to a concordance in number as well as a concordance in gender and act as a baseline for both the gender and number conditions. Figure 6 shows an apparent improvement by proficiency except in the good determiner-noun agreement condition, in which the intermediate L2 learners perform slightly better than the upper-intermediate L2 learners. A one-way ANOVA showed that there is a significant difference among the three groups in the bad gender noun-adjective agreement condition but not the bad gender determinernoun condition. (Guijarro-Fuentes (2014a) obtains the opposite result with his
217
The non-native acquisition of Spanish (un)interpretable features
Figure 6: Grammaticality judgment results (gender), with error bars of one standard deviation
French learners of Spanish, who show a significant difference for bad gender determiner-noun condition but not bad gender noun-adjective agreement condition). A two-sample t-test was conducted to determine where the differences were, and these results are shown in Table 3, with the significant differences highlighted. Table 3: Grammaticality judgment test (intergroup comparison) Good NA t ANOVA
p
Bad gender NA df
1,408 0,256 2
t
p
5,443 0,008
Good DetN
df
t
p
Bad gender DetN df
t
df
0,809
2
NSSp vs Up- InterSPL2Dutch
*
*
*
20,412