Pronouns and Clitics in Early Language 9783110238815, 9783110238808

Citation preview

Pronouns and Clitics in Early Language

Studies in Generative Grammar 108

Editors

Henk van Riemsdijk Harry van der Hulst Jan Koster

De Gruyter Mouton

Pronouns and Clitics in Early Language Edited by

Marı´a Pilar Larran˜aga Pedro Guijarro-Fuentes

De Gruyter Mouton

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

ISBN 978-3-11-023880-8 e-ISBN 978-3-11-023881-5 ISSN 0167-4331 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 in the Internet at http://dnb.dnb.de. ” 2012 Walter de Gruyter GmbH & Co. KG, Berlin/Boston Typesetting: PTP-Berlin Protago TEX-Production GmbH, Berlin Printing: Hubert & Co. GmbH & Co. KG, Göttingen 앝 Printed on acid-free paper 앪 Printed in Germany www.degruyter.com

Contents Clitics: to be or not to be . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pedro Guijarro-Fuentes and Pilar Larrañaga

1

Collecting data and measuring competence . . . . . . . . . . . . . . . . . . . . . . . . María Pilar Larrañaga and Pedro Guijarro-Fuentes

11

On the prosodic structure of articles in L1 acquisition of Spanish . . . . . . . Conxita Lleó

45

Third person clitic production and omission in Romance SLI . . . . . . . . . . Anna Gavarró

79

Which category replaces an omitted clitic? The case of European Portuguese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 João Costa, Maria Lobo and Carolina Silva (A)symmetries in the production of object clitics by Romanian children . . 131 Martine Coene and Larisa Avram Clitics in the acquisition of inalienable possession in French . . . . . . . . . . 175 Katrin Schmitz Children’s Acquisition of Subject Markers in isiXhosa . . . . . . . . . . . . . . . 209 Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers The Acquisition of Subject and Object Clitics in Swahili . . . . . . . . . . . . . 237 Kamil Deen Endocliticization and the Lexical Integrity Hypothesis: Insights from Degema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Ethelbert Emmanuel Kari

vi

Contents

Afterword: On clitic omission and the acquisition of subject clitic pronouns. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 Anna Cardinaletti Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

Clitics: to be or not to be Pedro Guijarro-Fuentes and Pilar Larrañaga

1.

General Overview

The term ‘clitic’ is commonly applied as a generalization of the traditional categories proclitic and enclitic (for clitics emerging at the beginning and end, respectively, of an associated host word or phrase). They are bound elements which, in their phonological behavior, bear a resemblance to af¿xes, but in their grammatical function look like independent words. That is, clitics can be considered as morpho-syntactic elements that are ‘neither clearly independent words, nor clearly af¿xes’ (cf. Zwicky 1977: 1); they are phonologically weak and often show special syntactic behaviour as well as a tendency towards af¿xation and disappearance and are hence often subject to linguistic variation. This can be observed for instance in modern French (cf. Jean (n’)aime pas les pommes), where the realisation of the clitic negation particle ne varies according to intra- and extra-linguistic factors, (such as the morpho-syntactic and phonological environment, the socio-demographic and geographic origin of the speakers, and the communication situation). More speci¿cally, the seminal book by Kayne (1975) stated that clitics and noun phrases differ in a number of ways in French as the former are restricted in their position (1), do not appear in positions where nominal phrases appear (2), the complex verb + clitic cannot be interrupted (3), clitic clusters cannot be split up (4), are not stressed and cannot be stressed (5), cannot appear in isolation, for example, as answer to a question (6) cannot be modi¿ed (7) and, ¿nally, cannot be co-ordinated (8). Subsequent research in various language families such as Romance, Germanic, Slavic, (see the volume by van Riemsdijk 1999 and references therein) and some exotic languages (such as Degema and Swahili investigated by the authors of the papers in this volume) has shown that Kayne’s observations hold true for the languages which happen to have clitics. The following examples in French and Spanish (1–8) illustrate the above mentioned features.

2

Pedro Guijarro-Fuentes and Pilar Larrañaga

(1)

[Sp] Pongo [el pantalón]DO [a la niña]IO/ [Se]IO [lo]DO pongo [Fr] Je mets [le pantalon]DO [à la ¿lle]IO / Je [le]DO [lui]IO mets “I put the trousers to the girl on’/I put them her on”

(2)

[Sp] Voy a ofrecer uno a Arantxa/*Voy a dar lo le [Fr] Je vais offrir un à Arantxa/*Je vais offrir le lui “I will offer one to Arantxa”

(3)

[Sp] Voy a lavarme después/*Voy a lavar después me [Fr] Je vais me laver après/* Je vais laver après me “I will wash myself later”

(4)

[Sp] ¡Dámelo ahora mismo!/*Dame ahora mismo lo [Fr] Donne-le moi immédiament/*Donne-le inmédiatement moi “Give it to me immediately!”

(5)

[Sp] Voy a ponerlo/*Voy a ponerlo [Fr] Je vais le mettre/*Je vais le mettre “I will put it”

(6)

[Sp] ¿Qué has comprado?/ *Lo [Fr] Qu´as-tu acheté?/*Le “What did you buy?/*It”

(7)

[Sp] Cojo el abrigo de papá /*[Lo] [de papá] cojo [Fr] Je prends le manteau de papa /*Je [le] [de papa] prends “I take daddy’s coat”

(8)

[Sp] *Lo y la he visto [Fr] *Je le et l´ai vus “I have seen him and her”

The publication of Richard Kayne’s 1975 book on French syntax not only marked the beginning of much research on the French language but also on more general theoretical issues with particular reference to clitics. Over the years, however, clitics have not only come to play a central role in formal linguistics, but also in language acquisition too which is the main focus of the present edited volume. In the ¿eld of diachronic syntax, for instance, clitics have been considered to represent an intermediary stage between a full lexical item and an inÀexional af¿x (Bouzouita, 2008). There is, nevertheless, support

Clitics: to be or not to be

3

that Kayne’s tests of clitic-hood should be contested. For instance, in African French, clitics may be stressed, and in the history of Italian, the internal position of clitics has changed (Poletto, 1999). It is also necessary to modify the basically Romance conception of clitics (Romance languages have verb-adjacent clitics): Clitics in the Germanic languages are not primarily pronouns, and their host is not necessarily a verb (Mavrogiorgos, 2010) (more on this below). So thus, descriptively speaking, clitics are often phonologically reduced forms with a special paradigmatic and/or syntactic distribution compared to other parts of speech (e.g. free pronouns). At the same time, there has been an intensive discussion as to whether clitics in general must be dealt with in morphology or syntax. In that regard, there have been three main linguistic approaches to the derivation of clitics at the spell-out: the movement approach according to which clitics are originated in the canonical position and then undergo movement to its surface position associated with the verb (Belletti, 1990; Cardinaletti, 1994). The other approach is the ‘base-generation’ which proposes that clitics initiate in a clitic position connected with the verb (Jaeggli, 1986; Borer, 1984). The third is the most recent one which is based on the hypothesis that clitics are functional categories (Uriagereka, 1995) and the movement of the clitics is triggered by the strength of the morphological features encompassed in the clitics. More speci¿cally, clitics are standardly assumed to be ambiguous categories sharing XP and X0 properties which move from argument XP positions within VP and adjoin to T0 (Chomsky 1995: 249). Given Diesing’s (1992) Mapping Hypothesis, they must move out of VP, as they are inherently referential (cf. Uriagereka 1995). Traditional grammars have stated that clitics are subject or object pronouns whose distributional features make them different from personal pronouns. What grammars have adequately described, has reached the explanatory adequacy with Cardinaletti and Starke‘s (1999) proposal. These scholars contend that clitics and strong pronouns have to be divided into three classes. Clitics differ from personal pronouns in that they lack two layers of complexity. This is why clitics cannot be stressed and cannot refer on their own. Thus, Cardinaletti and Starke (1999) explain the differences between clitics and pronouns by means of external properties. Depending on the side of the host the clitics attach to, they have been labelled as proclitic and enclitic, both of which have been studied in this volume for a number of languages. That is, clitics have been believed to be attaching to the periphery of the host. However, Kari shows in this volume that clitics can be inserted into the host in Degema, a phenomenon called endocliticization. He discusses at length the status of this special type of clitic to show that endocliticization exists, although it is not a widespread phenomenon in the languages of the world. In a very preliminary study, he shows that endoclitics pose a challenge for the Lexical Integrity Hypothesis.

4

Pedro Guijarro-Fuentes and Pilar Larrañaga

Moreover, clitics are studied from quite different linguistic perspectives. For instance, while phonologists are interested in prosodic questions about clitics at the interface between phonology and syntax, morphologists and syntacticians usually investigate the paradigmatic and/or linear distribution of clitics with respect to morphosyntactic features and to syntactic rules of well-formedness (see papers in the present volume and references therein). One characteristic, however, that concerns all linguistic perspectives mentioned here, is the explicit or at least implicit interest in the interaction between different components of the grammar, since clitics are a locus classicus for acquisition studies. Therefore, the present collection tries to bring together researchers working on clitics from diverse research ¿elds focusing on a range of languages of the Romance family (i.e., Spanish, Portuguese, French and Romanian) and a few more exotic languages such as Degema, Swahili and Xhosa. In particular, all papers included herein address clitics and related expressions from various perspectives such as phonology and prosody, morphology, syntax and semantics with particular reference to ¿rst language acquisition. The article by Larrañaga and Guijarro-Fuentes focuses on methodological issues concerning data collection and processing. The point of departure of this piece of research is the fact that clitics (subject and object clitics as well as articles) have a high frequency in spoken and written Spanish as the Davies Corpus (www.corpusdelespanol.org) shows. The authors raise the issue of whether clitics are distributed in early language acquisition in the same way as they appear in the Davies corpus and whether the amount of clitics impacts on MLU in any signi¿cant way. Moreover, they investigate whether interlocutors’ MLU mirrors the children’s MLUs. The authors show that the distribution of clitics in both children and interlocutors very much resembles the distribution in the Davies corpus. Furthermore, they show that there is a high correlation between the interlocutor’s MLU and the children’s MLU. Since a large amount of the MLU by the children is due to the interaction of interlocutor and child, they concluded that it is not possible to tease apart how much of the MLU is due to genuine language development and how much is due to the interlocutor’s MLU. In addition to this, they showed that clitics do not play a major role in the MLU score. Hence, any increase of MLU in the children’s performance must be due to other factors that need to be investigated in the future. The paper by Lleó is the one paper which deals with prosodic issues concerning clitic acquisition. Lleó analyses the performance of three monolingual Spanish children growing up in Madrid from the onset of the recordings until the age of 2;3 approximately. Articles as well as verbal clitics were taken into account. Lleó sets out to determine the prosodic structure of lexical words and clitics, whereby the main question is whether clitics belong to the PW (prosodic

Clitics: to be or not to be

5

word) in an initial stage in order to develop to a part of the PPh (phonological phrase) at some later stage. As expected, lexical items constitute PWs, whilst articles and verbal clitics go through different stages before they attain targetlike prosodic structure. Lleó shows that the structures all subjects use can be accounted for by assuming three different stages, namely a ¿rst stage in which children use proto-articles and a second and third stage where children use articles. In terms of prosodic structure, the proto-articles belong to the PW in the very ¿rst stage by two of the children. In the second stage, the articles belong to the PW to conclude with a third stage in which the articles belong to the PPh. Lleó con¿rms the results on previous studies on Romance languages, namely that children acquiring Romance languages use articles at an early age. However, children need around one year in order to attain the 100% level of use. In other words, article use does not increase abruptly but steadily over a year or so. More importantly, homophonous article and verbal clitics are not acquired with the same ease since verbal clitics are acquired more rapidly. Gavarró studies the acquisition of clitics in Catalan in a population suffering from SLI. She tests the hypotheses of whether clitics (articles and verbal clitics) are acquired as a class in Catalan SLI due to the fact that both articles and verbal clitics share a number of phonological properties and whether de¿cits in the production of third person clitics found in normal populations are subject to cross-linguistic variation as is the case in normally developing populations. Gavarró shows, as it has been shown for other languages, that clitics and articles are affected differently in the language of the population studied. Moreover, Gavarró compares her Catalan data with data in other languages and, thus, concludes that third person clitics are rather problematic in some languages such as Italian, French including Catalan, but pose no problems in Spanish both in normal developing children and in children with SLI. The paper by Costa, Lobo and Silva focuses on early acquisition of clitics in Portuguese. The authors review a number of papers which have studied the acquisition of clitics in Portuguese and come to the conclusion that children drop clitics in early Portuguese and, more interestingly, that clitic drop is not a uniform phenomenon. They report on a pilot study by Costa and Lobo (2010) in which they tested 20 ¿ve year old children and a group of 12 adults on the interpretation of null subjects and objects and VP ellipsis on a comprehension test. The adults perform target-like in that they accepted sloppy readings with object drop and VP ellipsis and rejected such a reading in subject drop contexts. Interestingly, the children perform equally badly in all three conditions. Costa, Lobo and Silva come to the conclusion that the poor performance is due to the fact that they are not able to assign the correct interpretation to the null category, since other longitudinal studies on Portuguese have shown that

6

Pedro Guijarro-Fuentes and Pilar Larrañaga

children are aware of the existence of null elements. The difference between pro and variables is at this stage irrelevant according to Costa, Lobo and Silva since children reject strict readings for objects and assign pro readings to variables. Therefore, children have to acquire the target-like interpretation of null elements in order for illicit clitic drop to decrease. Coene and Avram investigate the acquisition of reÀexive and non-reÀexive clitics in early Romanian. The authors make the observation that reÀexive and non-reÀexive clitics are not acquired in the same way since an acquisition asymmetry is attested between 1st and 2nd on the one side and the 3rd person non-reÀexive clitics. The latter appear later and are dropped even at a later stage as opposed to 1st and 2nd person clitics. Coene and Avram attribute this ¿nding to the syntactic status of the 3rd person clitics which are determiner-like and base generated in the VP (one of the linguistic positions to the derivation of clitics reviewed earlier on). They argue that the relationship between Agree and this position is computationally more costly because of the crossing of the phi-features. First and 2nd person clitics are assumed to be pronominal in nature and licensed by an operator in C as any other deictic category. ……‘”†‹‰ –‘Coene and Avramǡ–Š‹•ƒƒŽ›•‹•‡š’Žƒ‹•the asymmetry. Schmitz investigates the acquisition of dative and reÀexive clitics in monolingual French and bilingual German-French children in order to test whether cross-linguistic inÀuence can account for her results. Schmitz investigates both types of clitics by looking at di-transitive and inalienable body part constructions. The results of two elicitation tasks with monolingual French and bilingual German-French children are discussed in this study. The results show that clitic pronouns are not the preferred nominals in early child language, neither with direct nor indirect objects. Moreover, indirect objects are dropped more often and for a longer period of time than direct objects by all subjects. Interestingly, reÀexive pronouns are always used if the possessors of body parts are present. Schmitz’s tentative explanation is that the anaphoric interpretation of “se” is accounted for by a Voice Head necessary to introduce an external argument. The following three articles deal with subject and object clitics in African languages. Smouse, Gxilishe, de Villiers and de Villiers review the existing literature in early acquisition of isiXhosa, a language of the Bantu family spoken in Africa. As for many other languages in the Bantu family such as Zulu, Chichewa and Swahili the status of clitics is not settled yet so that, on the one hand, some scholars contend that clitics have pronominal status and, on the other hand, other researchers attribute them af¿xal status according to the authors. Smouse, Gxilishe, de Villiers and de Villiers suggest that early acquisition data could contribute to elucidation of this controversy. The authors report

Clitics: to be or not to be

7

some results by Gxilishe et al. (2007a; b) and conclude that subject markers (SM) have af¿xal status in isiXhosa. The authors also report on research carried out by de Villiers and Gxilishe (2008) and Gxilishe et al (2009) who investigated whether children were able to retrieve number information by the sole means of having access to subject and object clitics. The outcome of the tests was that the 38 tested children between the ages of four and six show no sensitivity to SM or OM since they perform at chance. The reported authors conclude that the results favour the hypothesis that the clitics have af¿xal status. Smouse, Gxilishe, de Villiers and de Villiers cast doubts about the methodology followed and report on a study conducted by Smouse (submitted) with an act-out methodology. The results with this method show a higher sensitivity in SM (0.65) than reported in previous studies depending on the cue used. Deen investigates the acquisition of subject and object clitics Swahili by four children (two male and two female) in Nairobi between the ages of 1;08– 2;11. His data are very interesting since the object clitics are acquired prior to the subject clitics. He argues that subject clitics are not supplied in the input in all contexts especially when the subject clitic is discourse salient and this may be causing the delay in acquisition. As opposed to the subject clitics, object clitics are realized in all required contexts. Deen argues that object clitics are acquired earlier because they are consistently realized by the adults although they are less frequent and less salient in discourse. Kari presents the very ¿rst study on the acquisition of clitics in Degema by four monolingual and bilingual children aged between two and four using an elicitation task with pictures and informal observation of spontaneous speech. The case of Degema is of crucial interest to gain insights in the intricate world of clitics, since clitics have been analyzed as external to their hosts, either in proclitic or enclitic position. The clitics studied by Kari are endoclitics, clitics that are inserted inside the host which challenge the Lexical Integrity Hypothesis, which states that words are internally coherent and cannot be rearranged according to syntactic rules. The results show that the children studied drop the subject clitics even at advanced ages. Kari’s study is the very ¿rst study on acquisition of Degema and longitudinal as well as cross-linguistic studies should con¿rm his data.

2.

Concluding remarks

One original aspect of the present volume is that it addresses theoretical, methodological and empirical issues as regards the class of clitic pronouns and DPs. It has not only addressed syntactic issues, but also semantic, pragmatic and

8

Pedro Guijarro-Fuentes and Pilar Larrañaga

phonological questions that have been under-represented in the acquisition literature up to today. An interesting aspect of this book is that its focus is not only on Romance languages, but it also reports on acquisition data of more exotic languages which have not attracted the attention of researchers until very recently. In that respect, we believe that it no doubt contributes to our understanding of clitic acquisition due to the fact that exotic languages have been taken into account. Nevertheless, based on some of the ¿ndings reported by the studies included herein a series of questions remains to be investigated in order to determine why children opt to choose clitics or pronouns or DPs or drop them. Furthermore, and despite the nice contribution that these papers represent, very little is known about the contexts in which adults omit clitics if their variety allows for it and, consequently, how children acquire these varieties. For interested researchers willing to explore that line of thought, the work of López (2009) is crucial in that respect as López claims that clitics are topics. However, there is no consensus amongst researchers about the concept of topic and how to use it in order to investigate language acquisition in particular. To sum up, the papers of this volume show that clitic is not equal clitic, that is the behaviour of clitics very much depends on the syntactic function they ful¿l. Whether a weak pronominal is a subject, object or reÀexive clitic is crucial, since the syntactic function clitics ful¿l plays a crucial role in the course of language acquisition. It has been observed for the languages studied in this volume that very clear acquisition asymmetries exist depending on the function clitics have. The papers in this volume propose some answers to issues such as asymmetries in clitic acquisition such as the acquisition of reÀexives, but their results raise a number of unresolved issues which will have to be dealt with in future research. One interesting line of research could be whether the locus of the clitic, with respect to its host, boosts its acquisition or, on the contrary, causes delay.

References Belletti, A. 1990 Borer, H. 1984

Generalized Verb Movement, Rosenberg and Sellier: Torino. Parametric syntax: case studies in Semitic and Romance languages. Dordrecht: Foris.

Bouzouita, M. 2008 At the Syntax-Pragmatics Interface: Clitics in the History of Spanish. In: Language in Flux: Dialogue Coordination, Language Variation, Change and Evolution, Cooper, Robin and Ruth, Kempson (eds), London: College Publications, 221–263.

Clitics: to be or not to be

9

Cardinaletti, A. 1994 On Cliticization in Germanic Languages. In H. Van Riemsdijk and L. Rizzi (eds), Clitics and their hosts, ESFEurotyp, Tilburg University. Cardinaletti, A. and M. Starke 1999 The Typology of Structural De¿ciency: A Case Study of the Three Classes of Pronouns. In H. van Riemsdijk (ed.) Clitics in the Languages of Europe, vol. 8 of Language Typology, Mouton de Gruyter, Berlin, pp. 145–233. Chomsky, N. 1995 The Minimalist Program, MA: MIT Press. Costa, A. and M. Lobo 2010 Clitic omission is null object: evidence from comprehension. In J. Costa, A. Castro, M. Lobo and F. Pratas, orgs. Proceedings of GALA2009. Newcastle: Cambridge Scholars Publishing. 96–106. Diesing, M. 1992 Bare Plural Subjects and the Derivation of Logical Representations. Linguistic Inquiry 23, pp. 358–380. Gxilishe, Sandile, de Villiers, Peter A. and de Villiers, Jill G. 2007a The acquisition of subject agreement in Xhosa. In A. Belikova, L. Meroni, and M. Umeda (Eds.), Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition North America (GALANA). Somerville, MA: Cascadilla Press. pp. 114–123. Gxilishe, Sandile, Jill G. de Villiers and Peter A. de Villiers 2007b Acquisition of tense in Xhosa: the long and short of it. In H. CauntNulton, S. Kulatilake, and I. Woo (Eds.), BUCLD 31: Proceedings of the 31st annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press. pp. 274–285. Jaeggli, O. 1986 Three issues in the theory of clitics: Case, doubled NPs and extraction. Syntax and Semantics, 19, pp. 15–42. Kayne R.S. 1975 French Syntax. The Tranformational Cycle. Cambridge: The MIT Press. López, L. 2009 A Derivational Syntax for Information Structure. Oxford: Oxford University Press. Mavrogiorgos, M. 2010 Clitics in Greek: a minimalist account of proclisis and enclisis. John Benjamins. Poletto, Cecilia 1999 The internal structure of Agrs and subject clitics in the Northern Italian dialects. In Henk van Riemsdijk (ed.). Clitics in the Languages of Europe, pp. 581–620. Berlin: Mouton de Gruyter.

10

Pedro Guijarro-Fuentes and Pilar Larrañaga

Uriagereka, J. 1995 Syntax of clitic placement in Western Romance. Linguistic Inquiry, 26, 79–123. Van Riemsdijk, H. (ed.) 1999 Clitics in the Languages of Europe, vol. 8 of Language Typology, Mouton de Gruyter, Berlin. Zwicky, A. M. 1977 On Clitics. Bloomington, IN: Indiana University Linguistics Club.

Collecting data and measuring competence* María Pilar Larrañaga and Pedro Guijarro-Fuentes

1.

Introduction

Any researcher in the ¿eld of ¿rst (FLA), second language (SLA) and bilingual (BFLA) acquisition seeks to achieve various goals. These include the provision of an accurate analysis of the phenomenon at stake and achievement of explanatory adequacy. In order to ful¿l these goals, a variety of data collection methods exists, i.e., naturalistic vs. diverse experimental methods. It is beyond any doubt that scholars seek to ¿nd insights into the competence of the speakers they are investigating, but this can only be achieved by looking at their performance. For that purpose, most researchers resort to either naturalistic or experimental approaches. Both approaches have advantages and drawbacks, but we mainly focus on the naturalistic data collection in relation to clitics use in the present paper. For instance, the ¿rst problem researchers in second language acquisition are confronted with is how to discriminate amongst learner groups in terms of their overall pro¿ciency level in the target language. Since this problem is not a trivial one, a whole plethora of literature has been devoted to this issue over the last two decades or so (e.g., Daller, Milton and Treffers-Daller, 2007, to name one of the most recent ones). The range of subjects discussed ranges from the adequacy of tests used, their validity and reliability in order to approach the topic under investigation to the most appropriate external measures to discriminate amongst learner groups, such as Type-Token-Ration (TTR) and C-test to mention but only two independent measures. The issue of comparability between subjects in early ¿rst language acquisition (and for that matter, bilingual ¿rst language acquisition) is a more intricate one1. It is well known that age in ¿rst language acquisition is not a good predictor of language development due to the fact that there are large extra individual * We wish to express our gratitude to Michael Daller who has helped us with the whole statistical analysis. Of course, all remaining errors remain our own. 1. We won’t be discussing other acquisition scenarios such as language attrition, heritage language acquisition and so on; however, that does not mean that the metho-

12

María Pilar Larrañaga and Pedro Guijarro-Fuentes

divergences. Some children start with their ¿rst one word utterances as early as 1;03, others start at least a year later. However, age and Mean Length of Utterance (henceforth, MLU) correlate very strongly as numerous studies have shown (Miller and Chapman 1981, Klee and Fitzgerald, 1985; Scarborough, Wyckoff, and Davidson 1986). In order to be able to compare children, it is more reasonable to have some external metrics to ascertain that we are comparing linguistic stages and not ages. Some linguistic measures used in SLA and other language scenarios such as the C-test cannot be used in FLA because this kind of test is normally a written test. The TTR is a very controversial measure because it is highly text length dependent (Klee, 1992; Malvern and Richards, 1997) and is only used as a “secondary measure” in FLA, if at all. The most widely accepted measure is MLU, a measure that will be discussed at length below. As regards clitics, Section 3 discusses issues related to their syntactic status since it is not quite clear whether they belong to the ¿rst (i.e., words) or the second class (i.e., morphemes) in Spanish. Although widely accepted by the research community (Casielles, Andruski, Kim, Nathan and Work 2006), MLU has some general drawbacks worth discussing (Section 4) in relation with the issue of clitics. The present paper analyzes the linguistic performance of two normally developing monolingual Spanish children with special focus on the use of clitics and their impact on the total MLU and the impact of the interlocutor’s MLU on the child’s MLU (Section 5 and 6). Given these preliminary insights into some of the shortcomings of the linguistic measures used to date and available to the research community, in the rest of the present paper, we will focus on the issue of data collection before addressing some well known and probably universal properties of clitics, focusing on Spanish in Section 3, and then looking at some general MLU problems in Section 4. Section 5 is devoted to our own empirical study and Section 6 is the overall concluding discussion.

2.

Sampling and design

In this section, we will discuss some general issues which are linked to sampling, research design and some speci¿c problems related to clitics. To start with, two different approaches exist, a naturalistic and an experimental approach to language acquisition. Starting with the former, one issue the redologies employed to collect data from those language scenarios do not deserve a proper scrutiny and investigation.

Collecting data and measuring competence

13

searcher has to tackle when dealing with language is sample size with respect to frequency of the investigated phenomenon. It is well known that some linguistic phenomena have a high frequency, whilst others are not very frequent at all. Hence, the probability of ¿nding a low frequency phenomenon in a 20 minute sample is much lower than of ¿nding a high frequency phenomenon. One way to solve this shortcoming would be to do an intensive study of one child over a short period (let us say, recording 5–10 hours per week over a period of 3–4 weeks) (Tomasello and Stahl, 2004) in order to collect as much data as possible. However, this is not always possible, and the period studied indeed remains very short. Many researchers therefore opt to collect data every fortnight for about 30–45 minutes for a period of two years or more. Of course, the more data the researcher can obtain from a given child, the more ‘evidence’ it provides the researcher about language development. One important fact to bear in mind is that it is crucial to possess enough data in order to be able to determine the age of acquisition of the phenomenon at hand quantitatively. It seems plausible that highly frequent2 phenomena will have a greater impact than those with a low frequency in the ultimate MLU, the topic at hand in this paper. We know that clitics are highly frequent in child language (Larrañaga, 2000), but wish to know whether this is true for adult language. In order to determine their frequency, we counted some word classes and clitics in the oral Davies Corpus accessible in www.corpusdelespanol.org.The author of the corpus has made available a wide variety of epochs and genres. For our purposes, we have restricted ourselves to the oral corpus in an attempt to keep as far as possible to the type of sample presented in this paper and the range of studies in the present volume, although our methodological concerns will not cover entirely the number of problems encountered in all the studies reported here. In addition to the clitics, we collected some data with regard to verb frequency as well as other word classes in order to compare their frequency to the frequency of clitics to assess how often clitics were used on the whole. Thus, we decided to calculate the frequency of auxiliaries and some full verbs. The choice of full verbs was determined by the frequency of use of the verbs in early language acquisition studies in Spanish. Hence, the verbs hacer “to make”, poner “to put”, quitar “to remove”, coger “to take”, abrir “to open” and cerrar “to close” were considered. First of all, let us consider the total number of verbs, nouns and pronouns (see Table 1).

2. “Unsurprisingly, learners are more likely to learn the frequent words than the infrequent words” Daller et al. (2007: 10).

14

María Pilar Larrañaga and Pedro Guijarro-Fuentes

Table 1. Total amount of words (20.000.000 words) in The Davies Corpus Verb

Noun

Pronoun Article Adjective Adverb Preposition Conjunction

775,945 718,266 218,726 448,327 214,497 338,2313

214,479

495,070

Table 2. Frequency of selected verbs, articles and clitics in Spanish Verb/lemma ser haber estar ir hacer poder querer deber poner coger abrir quitar

Frequency (n) 123,529 62,264 37,487 28,148 25,246 21,213 8,025 5,090 4,998 927 794 661

Word el4 la los las me te se lo le les nos os

Frequency (n) 144,737 125,426 51,071 31,936 26,481 9,355 53,593 41,407 14,605 3,992 8,705 538

Table 1 does not necessarily mirror the word class distribution of children acquiring Spanish, since it is well known that adjectives, adverbs, prepositions and conjunctions are not very frequent in their oral productions. We are, however, not aware of any study which has made an accurate count of the mentioned word classes in early language acquisition in Spanish. Despite its limitations, this table 1 shows that verbs and nouns make up the majority of the words counted in the corpus. Pronouns and articles (see also Table 2) together build a class that is nearly as frequent as the nouns themselves. Moreover, from some recent acquisition studies on Spanish (Larrañaga, 2000) we are aware of pronouns and articles as categories that emerge rather early in child language (as in any languageȌ and have a high frequency of use. Hence, the impact of pronouns and articles on MLU in early language cannot be negligible and requires a serious scrutiny when calculating its scores. Nevertheless, as Table 2 shows, not all clitics are equally frequent. The article el is the most frequent one, singular clitics are far more frequent than the plural ones. For example, os 3. The corpus includes the negation amongst the adverbs. 4. The results with el include el in isolation 97,715, al 17,111 and del 29,911 tokens.

Collecting data and measuring competence

15

is almost 30 times less frequent than el. For the purposes of this study, we will conÀate all clitics. In sum, one can conclude that clitics are quite frequent in oral Spanish. So then, how large should a sample be? In this respect, Tomasello and Stahl (2004:102) point out that “[…] for high-frequency phenomena, for instance, children’s use of copulas or pronouns in English, the typical samples used in the study of child language are no doubt adequate – at least for some kinds of analyses.” Hence, data available in CHILDES are adequate for the present study. With regard to research designs in language acquisition research, the different studies may be catalogued depending on whether researchers are more interested in the descriptive or qualitative data analysis, correlational, and/or experimental data collection. Hence, the method of data collection also differs from the very use of some very restrictive linguistic tasks on one end of the spectrum, and naturalistic inspection on the other. Different methodologies are designed to meet different expectations and purposes. On the one hand, researchers have made ample use of studying a language in its natural environments or habitats without trying to manipulate or control anything. That is, naturalistic corpus data collection consists of a large sample of the spontaneous speech of one or more young children acquiring whichever language. In these type of studies, the researcher limits himself/herself to observe the language (acquisition and/or development) under study in its natural setting without attempting to inÀuence or control/manipulate it (e.g., Guijarro-Fuentes and Ortiz-Lopez, 2008; Larrañaga and Guijarro-Fuentes, in press; Ortiz-Lopez and Guijarro-Fuentes, 2008 amongst others on the acquisition of clitics in Spanish and other related languages). The observations are conducted in a naturalistic setting, observing either adults or children at different ages and stages of language development in the same or different setting so that language Àows without (consciously) paying attention to any particular linguistic feature. One of the main weaknesses of the studies in which naturalistic data is employed is

16

María Pilar Larrañaga and Pedro Guijarro-Fuentes

that the appearance of a particular feature is very limited and that the researchers inadvertely prompt the linguistic feature under investigation5,6. There are several other problems which one has to scrutinise when collecting a naturalistic corpus. One is the stage of acquisition to study. Generally speaking, from a developmental and theoretical point of view, the most interesting stages are the earliest stages or the earliest that the researcher can have access to (prior to the age of two and a half or three years, by which time most children are capable of producing adult-like structures with clitics); child syntax (or in our case, the presence or absence of clitics) can differ from adult syntax. So, an ideal time to study is the period between one and a half and three years of age. Nevertheless, if one can study the whole period by conducting a longitudinal study, this is much better. A problem related to early stages of language acquisition when dealing with clitics is to what extent one should count the so called proto-articles. As to the status of early vowels preceding nouns, Kuchenbrandt (2008) studied longitudinal the data of 3 bilingual German-Spanish, 3 monolingual German and 3 monolingual Spanish children. She analyzed the very early vowel-like prenominal items which have often been referred to as ¿llers or proto-articles in previous studies and raised the issue of whether they had any grammatical function. Her results show that children combine certain vowels with certain genders consistently. Kuchenbrand concluded that these not yet target-like vowels had grammatical content from early on. If a sound string has a grammatical function, it would be reasonable to count it for the MLU calculations. However, anyone who has worked with early learners knows that children at early stages make gender errors as do the children studied in this paper. So, the question 5. We leave to one side the advantages or disadvantages of using longitudinal or crosssectional studies. Designing a cross-sectional study would involve making a number of recordings of various different children in a given age-range (e.g. 1;6 to 2;6 years) acquiring a particular language; whereas a longitudinal study of one or more children would involve recording weekly samples of the child’s spontaneous speech in interaction with his or her parents/caretaker/s. The types of studies present their own bene¿ts depending on the linguistic phenomena under investigation. One clear protracted Àaw of the longitudinal studies, however, is that the child in question may reach the two-word stage much later than one would predict; in addition, the child’s speech language may initially be relatively unintelligible in addition to other logistics problems such as the investigated child may go away on holiday, become ill for a certain period of time making it impossible to continue the data collection. 6. Other clear disadvantages of using naturalistic data are that it can take an inordinate amount of time to collect, transcribe, and tag the data (if one decides to do so for future generations or other researchers to make use or pro¿t from it).

Collecting data and measuring competence

17

arises as to when a certain mismatch between prenominal vowel and noun is attributed to not yet acquired gender and must be counted for MLU because it already ¿lls a syntactic position. In other words, how do we differentiate between a ¿ller with no syntactic function and a prenominal clitic whose function is settled but the realization is still not target-like? Hence, if one is studying the acquisition of clitics together with their position within a particular structure, and the child one is studying, does not produce many utterances with clitics and/or different word orders, it may be a good idea to supplement the naturalistic corpus data with some more controlled elicited data employing other elicitation techniques, including, but not limited to picture naming, word/sentence completion, sentence repetition and so on. Depending on the age of the participants, other types of tests worthy of employment are those tests designed to measure subject reaction time to a certain stimuli. In sum, it is evident that language competence is best assessed by a multiple combination of pinpointing methodologies which can be designed to pin-down different language components: subjects – children and adults alikemay develop a particular skill faster than others (for instance, reading vs. writing) or even not allǡ•ince subjects may never develop one of the skills in the target language (Guijarro-Fuentes and Ortiz-Lopez, 2008). In addition to some of the already outlined drawbacks of the naturalistic data collection, some children may not produce a very wide range of structures (e.g., clitic constructions) and in some cases may produce mainly one-word utterances. An additional constraint of the naturalistic data collections is that some ‘perfect’ children may be so advanced (even at a very young age) that their grammar is virtually error-free. If that is the case, it is nearly impossible to evaluate the child’s grammar in terms of any frequent or, for that matter, infrequent grammatical errors (either of errors of commission or of errors of omission in terms of the use of clitics). By nature, studies which employ experimental data collection on the use of clitics are generally speaking more precise and, hence, have conclusive power (Einsenchlas, 2003). This kind of study is particularly useful in attesting and investigating linguistic features (e.g., clitics) that otherwise would escape the eye of the researcher. In a true experimental study, all (dependent) variables are held constant except for the independent variable, which is manipulated adding a great power to the potential ¿ndings and conclusions. Thus, any unpredictable ¿ndings would be solely attributed to the pure interaction between the dependent and independent variable. This type of study does not escape criticism, however, from the research community ranging from the design of the experiment itself to the sample of subjects in an experiment. Certain measures must be taken to make sure that there are no experimental shortcomings.

18

María Pilar Larrañaga and Pedro Guijarro-Fuentes

For the purposes of the present paper, we decided to use naturalistic data because the phenomenon under investigation has a high frequency and according to Tomasello and Stahl (2004), a longitudinal sampling including 30 recordings of around 30 minutes for each child is appropriate for these purposes – see Section 5.1.

3.

Clitics as a challenging “word class”: the case of Spanish

It is well known that Romance languages possess clitics of various types. In this paper and more speci¿cally in this section, we will concentrate on clitics in Spanish, but many of the points made in what follows hold undoubtedly for other languages, which possess clitics with a similar distribution (see papers included herein the present volume). If we examine traditional grammars and more recent generative approaches, the Spanish words “me, te, se” are referred to as pronouns, object pronouns, weak pronouns or clitics. The different terms used to refer to this class of words show that the ones that used these terms for the ¿rst time to refer to them focused on different aspects of the clitics. Let us look at these properties in more detail. Table 3. Pronominal clitic system of Spanish Accusative Me 1st person sing. Te 2nd person sing. Lo 3rd person masc. sing. La 3rd person fem. sing. Nos 1st person pl. Os 2nd person pl. Los 3rd person masc. pl. Las 3rd person fem. pl.

Dative Me 1st person sing. Te 2nd person sing. Le 3rd person sing. Nos Os Les Se

1st person pl. 2nd person pl. 3rd person pl. 3rd person sing./pl.

Clitics are monosyllabic in Spanish. They have no word stress and cannot receive contrastive stress, hence, they attach to a host, the verb. Table 3 gives an overview of the system of object pronoun clitics7 in Spanish, both accusative

7. Spanish pronominal clitics are weak, unstressed pronouns which, unlike the strong (or free) forms, cannot be used in isolation, cannot receive contrastive stress, and cannot be topicalized, conjoined or modi¿ed. Clitics in Spanish always appear joi-

Collecting data and measuring competence

19

and dative. Clitics in Spanish are scored for person, number and gender – and politeness – features (third person in the singular as well as the plural). The position of the clitics within the Spanish sentence is multiple. In descriptive terms, clitics in Spanish always materialize with conjugated verb forms substituting the missing direct object as shown in (1a and b); (1b) would be ungrammatical without the presence of a clitic8. (1)

a.

Izaskun compró esa casai. ‘Izaskun bought that house.’

b.

Izaskun lai compró. ‘Izaskun CL-DO bought.

As mentioned above, the syntactic and morphological properties of clitics have been the object of intense theoretical debate in Romance languages. Considering some of the linguistics description with regard to the constraints ordering of both full-phrase complements (as in (1a)) and cliticized complements (as in (1b)) in Spanish, we depart from the idea that Spanish is a head initial language, a verb theta-marks a DP complements to its right (Zagona, 2002). The derivation of (1a) implies the movement of V-to-INFL and its complement DP shifts to the right of the verb as shown in (2). (2)

V’ V

DP

Compró

esa casa

In contrast, in the derivation of (1b) whose surface structure has no overt DP in its complement position, the verb is taken as transitive and the clitic has person and number features which correspond to the omitted DP complement, which is to say, third person feminine singular. For our purposes, we take the following stand in order to account for the position to which the clitic moves, that is,

ning phonetically to the verb either as proclitic to tensed verbs or as enclitic to in¿nitive, gerund and imperative forms (Zagona, 2002). 8. We leave aside the reasons why (1b) would be ungrammatical as they do not have any bearing on the present paper.

20

María Pilar Larrañaga and Pedro Guijarro-Fuentes

that the clitic is adhered to the verb when it undergoes movement9 as illustrated in (3). (3)

V’ V CL1 la

DP V

ti

compró

The attachment of the clitic, la, to the verb, compró, is due to the fact that the verb is the clitic ‘host’, that is, the constituent on which the clitic is phonologically dependent on (Zagona, 2002:186). Once the clitic adheres to V, the complex verb would then move to INFL. Furthermore, for the present investigation we adopt the most recent generative linguistic analysis that assumes that clitics in Spanish have syntactic status by virtue of their being the head of their own functional categories. This proposal is very much in line with the movement analysis aforementioned according to which clitics originate in argument positions. Clitics, therefore, are considered to be the head of DP. Moreover, it has been proposed that there is an object-related AGR10 head that checks Case and other features of the objects, and that object clitics are generated in that head (AgrOP and AgrIOP) (Franco, 1993; Uriagereka, 1995). They act as af¿xes of morphological agreement. Following Uriageraka (1995), clitics move from their position as object (AgrO) to a functional category FP, that is above AgrS some place in CP, where they check their referential features. Clitics are assumed to instantiate the AgrO whose features include gender and number. The movement

9. There have been, nevertheless, other linguistic approaches to the derivation of clitics at the spell-out: the movement approach according to which clitics are originated in the canonical position and then undergoes movement to its surface position associated with the verb (Belletti, 1990; Cardinaletti, 1994). The other approach is the ‘base-generation’ which proposes that clitics initiate in a clitic position connected with the verb (Jaeggli, 1982; Borer, 1984). The ¿nal approach is the most recent one which is based on the hypothesis that clitics are functional categories (Uriagereka, 1992, 1995) and the movement of the clitics is triggered by the strength of the morphological features encompassed in the clitics. 10. As Agr-S or Agreement-subject involves subject agreement features, the AgrO would involve object agreement features in the form of a clitic. This clitic would be the spell-out of the features for number, gender and case copied from a complement onto its governing head.

Collecting data and measuring competence

21

to F happens because of the morpho-phonological and prosodic properties that permit the association with a null or pro element, as illustrated in (4). (4)

FP

F

AgrSP

Agr

TP

T

AgrOP

AgrO

VP

V

DP clitic

As demonstrated in (4), clitics move to an abstract AgrO constituent in order to satisfy the D-feature checking of that head and movement is accounted for on the basis of D-feature checking. With regard to the collocation (order), clitics attach to the left periphery of the ¿nite verb (proclisis) and to the right periphery of in¿nite forms (in¿nitive and gerund) and build a prosodic word. If clitics cluster either in the proclisis or in the enclisis, the cluster cannot be broken down. More speci¿cally, in Spanish, clitics can appear in preverbal position when the verb on which they depend is conjugated (5a)11. Nevertheless, with in¿nitives and gerunds (non-conjugated verbal forms), the clitics must appear in a post-verbal position, joined to the in¿nitive (5b): (5)

a.

Juan losi vio (a los niñosi). John los-CL-DO see-past (the children). ‘John saw them.’

b.

Juan comprende el alemáni, pero hablarloi no le es fácil. John understands German, but speakit-CL-DO not him-clitic is easy. ‘John understands German, but it is not easy for him to speak it.’

11. Examples borrowed from Guijarro-Fuentes and Ortiz-Lopez (2008).

22

María Pilar Larrañaga and Pedro Guijarro-Fuentes

Clitics pronoun may be raised on constructions containing two verbs (6a–c): (6)

a.

Carmen quiere comprarlo vs. Carmen lo quiere comprar. Carmen wants buy it-CL-DO vs. Carmen it-CL-DO wants buy. ‘Carmen wants to buy it.’

b.

Carmen está viéndola vs. Carmen la está viendo. Carmen is watching it-CL-DO vs. Carmen it-CL-DO is watching. ‘Carmen is watching it.’

c.

Carmen puede visitarlo vs. Carmen lo puede visitar. Carmen can visit him-CL-DO vs. Carmen him-CL-DO can visit. ‘Carmen can visit him.’

The examples given under (6) illustrate that clitics in Spanish can stay behind in the lowest position (attached to the in¿nitive or gerund) (that is, enclitic to the non-¿nite verb) or be raised over a series of verbs until they occupy a preverbal position (that is, proclitic to the higher ¿nite verb). In addition, in Spanish, syntactic constructions with a clitic are possible, when the direct or indirect object is a strong pronoun or with a NP (7a–c): (7)

a.

Loi vi a éli. Him-CL-DO saw to him. ‘I saw him.’

b.

Lei dieron un premio a éli. Him-CL-IO gave a prize to him. ‘They gave him a prize.’

c.

Koldo lei habló a Izaskun. Lewis her-CL-DO talked to Izaskun. ‘Koldo talked to Izaskun.’

Whereas the clitic pronouns are compulsory in (7a) and (7b), it is optional in (7c). Nevertheless, an object clitic is compulsory in constructions in which the object is in the position of the topic (8): (8)

Las cartasi lasi dejé en mi despacho. The letters them-CL-DO left on my of¿ce. ‘I left the letters on my of¿ce.’

Collecting data and measuring competence

23

Finally, clitics in Spanish cannot show in nominal or prepositional constructions where the obligatory presence of a strong pronoun is required. In sum, the facts presented so far would support the syntactic status of clitics in Spanish. Let us now move on to some distributional facts that will contribute to the understanding of the main point in the present paper, i.e., the issue of the unit to be counted. Spanish has strong subject (9) and object (10) personal pronouns as well as weak subject (11) and object pronouns (12), referred to as clitics in the current literature. (9)

Tú bailarás un tango You will dance tango.

(10) A mí no me gusta el potaje I do not like soup (11) ¡Te vas a caer! You will fall down (12) (Tú) se lo quieres comprar (You) want to buy it to for him Strong subject pronouns can be omitted because Spanish is a so-called prodrop language, and as a matter of fact only around 30% of the subjects are realized in today’s spoken language (Grinstead, 2000). Nevertheless, the omission rate varies from around 20% to 70% from recording to recording (Bel, 2001). Subject and object clitics, however, are compulsory with the only exception being direct object clitics, which must be converted in a few contexts, i.e., with a co-indexed wh-word in the same utterance and with implicit objects. That is, one can assume that the frequency of these words is relatively high in current Spanish. As also mentioned above, the distributional property of most relevance for our purposes is that clitics in Spanish can occupy two positions: the proclitic and the enclitic position (see examples (6) above). (13) Se lo quieres comprar him/her it want buy (You) want to buy it to him/her

24

María Pilar Larrañaga and Pedro Guijarro-Fuentes

(14) Quieres comprárselo want buy-him/her it (You) want to buy it to him/her According to the current descriptive linguistic analyses (see above), proclitics are written separately from their host, whilst enclitics are written together with their host. That is and if we adhere to the most commonly accepted de¿nition of a word, a group of letters with meaning between two spaces, we would count ¿ve words in the ¿rst case (13) and only two in the second one (14). This procedure implies that clitics are counted as words. If, on the contrary, we count them as morphemes, the count in both cases will be identical. Since many researchers working on Spanish calculate MLU counting words (Casielles et al., 2006), one should think about how proclitic vs. enclitic clitics are distributed in spoken corpora. From previous studies, in early language acquisition (Larrañaga, 2000), it is known that proclisis is by far more frequent than enclisis. Articles can also clitizice phonologically to their host, in most cases the noun. The complete list of Spanish articles is: el-la, los-las, un-una, unos-unas. For the present study, we took only monosyllabic articles into account because bisyllabic words do not clitizize. Furthermore, the plural articles unos-unas are not frequent and would not impact in the total amount of articles.

4.

Counting MLU

MLU (mean length of utterance)12 goes back to Nice (1925) who proposed calculating the utterance length in words. As languages differ considerably as to their morphology and in order to be able to compare highly inÀected languages with languages with poor morphology, Brown (1973) suggested using the morpheme as a counting unit MLUm (morphemes) in 100 utterances of a transcript having ignored the ¿rst page of it. The quotient between the total number of morphemes in 100 utterances divided by 100 is MLUm. There is no consensus amongst acquisitionists as to whether to use words or morphemes as counting units since some languages are more inÀectional than others, e.g., Basque vs. Chinese and the ultimate decision depends upon the language under investigation. Brown (1973:53f.) proposed it because:

12. As will be discussed in the main text, MLU may be measured in morphemes, words or syllables adding an additional methodological conundrum.

Collecting data and measuring competence

25

‘The mean length of utterance (MLU) is an excellent simple index of grammatical development because almost every new kind of knowledge increases length: the number of semantic roles expressed in a sentence, the addition of obligatory morphemes, coding modulations of meaning, the addition of negative forms and auxiliaries used in interrogative and negative modalities, and, of course, embedding and coordinating.’

Indeed, the main advantage of MLU is that it is fairly easy to calculate. It only involves counting words. In the computer era this task has become even easier with the computer program clan13 which can calculate MLU in various languages per mouse click, which saves time and is more accurate than a calculation by hand. Some languages are highly inÀecting and MLUw would not make any sense, so that researchers count morphemes instead. In order to calculate MLU on the basis of morphemes MLU(m), on the contrary, the text must be lemmatised, which is extremely time consuming. This could also be done for Spanish; however, some studies have shown that MLU in morphemes (MLUm) and MLU in words (MLUw) are highly correlated: Aguado (1995) for Spanish 0.992, p = 0.0000. In the same vein, López Ornat (2003) demonstrated that MLU(m) and the syllable based MLU(s) correlated. Hence, some researchers opt to calculate MLU(w) if they are investigating syntactic issues, unless Spanish is to be compared with an even more inÀecting language such as Basque where a morpheme based MLU is more adequate (Larrañaga, 2000). MLU is nowadays a widely accepted score amongst acquisitionists, although scholars such as Scarborough (1990), Hickey (1991) have listed some drawbacks worth noting here. The main problem is that MLU works very well in a language like English which has segmental morphemes and for which it was developed, but it is dif¿cult to use with synthetic languages, where morphemes are not incremental. This problem is acknowledged to a lesser extent in languages like English where only a limited number of words belong to the class of portmanteau allomorphs, hence, having only a minor impact on the ¿nal score. Early studies on MLU conducted by Brown (1973) suggested that qualitative changes in language pro¿ciency could be mirrored by a quantitative measure such as MLU. He identi¿ed 5 stages in early language development presented in Table 4 below.

13. Clan is available at http://childes.psy.cmu.edu/

26

María Pilar Larrañaga and Pedro Guijarro-Fuentes

Table 4. Brown’s stages Stage I Stage II Stage III Stage IV Stage V

Age in months 15–30 28–36 36–42 40–46 42–52+

MLU 1.75 2.25 2.75 3.50 4.00

MLU range 1.5–2.0 2.0–2.5 2.5–3.0 3.0–3.7 3.7–4.5

Brown proposed not to count imitations and self-repetitions. At ¿rst sight, his suggestion looks self-explanatory; however, it is not easy to decide what counts as an imitation or self-repetition. Any researcher would agree that if a child utters an identical utterance immediately after the adults’ utterance this counts as an imitation. The same goes for self-repetitions. Nevertheless, there is a considerable grey zone here, because children do imitate or repeat previous parts of speech and it is not at all clear how far in the discourse we should look back to classify a given utterance as imitation or self-repetition. To our knowledge nobody has raised this issue until now, and it is by no means trivial. Another issue which has not received much attention is whether repetitions or imitations of parts of the previous speech count as such. It goes without saying that if a child retakes a word from the discourse of the interlocutor, this would not count as repetition/imitation. However, what if the child repeats more than one word? Where is the line between what counts as a repetition/imitation and productive production located? It is far from transparent. And, for the topic under scrutiny, can te lo quiso dar be viewed as repetition of quiso dártelo ‘(I) wanted to give it to you’?. Most researchers would agree that universal counting conventions for MLU are a desideratum not yet achieved in order to be able to compare studies in FLA and BFA14. Ideally, one should be able to compare all languages, but languages differ structurally in such a way that comparisons can be tricky. Structural differences susceptible to impact MLU values are also attested in genetically close language families or subfamilies, so that comparisons in terms of MLU can only be viewed as approximations. We are then left with conventions for one 14. A clear methodological advantage for those morpheme based MLU studies is that these studies allow comparisons of different distant typological languages that differ with respect to what they realize as clitics (Larrañaga and Guijarro-Fuentes, in press). Nevertheless, clear methodological drawbacks for other studies with morpheme-based MLU as some languages, (for instance, English) have little or no morphology compared to other languages which possess a lot (e.g., any Romance language including Spanish).

Collecting data and measuring competence

27

and the same language. If all researchers working on one language could agree on the conventions to use, we would move a step forward as regards comparability in a given language. Recently, ‘Richards and Malvern (2007:86)’ have suggested that “where possible, individualised coding is more appropriate than a ‘one analysis for all subjects’ approach”. In other words, counting conventions should be clearly stated for the research question being studied. If one adheres to this suggestion, theoretically, it is possible to obtain different MLU scores for one and the same recording because of the diverging counting conventions. The question is, in such a case, what does the score obtained reÀect if we arrange conventions according to the topic studied? In what follows, we will discuss some general issues that have been reported in the literature. Brown proposed a set of principles according to which the morphemes that ought to be counted were in most cases tuned to English grammar. Some of the rules he proposed have subsequently been abandoned or adapted, even for English. In fact, one word negations and af¿rmations are no longer counted in every single occurrence. As for Spanish, Brown’s rules have been subsequently revised by Linares (1981). He gives a much more elaborate account of how to calculate MLU, one of the differences being that diminutive morphemes are counted in Spanish, as opposed to Brown who advised not to do this. In what follows, we will concentrate on two key issues that Linares proposed: the computation of all NP related issues and the verb, as these two domains have a major impact on MLUm values. We will ignore all issues surrounding compounds, augmentatives, superlatives or adverbials in the present paper since that will all be scored as one point because we will adhere to the MLUw calculation. The endings for masculine and feminine –o and –a respectively are counted as bound morphemes on nouns, adjectives, pronouns and articles, only if the root can host both. A word like puerto ‘harbour’ would not be credited one point for –o because the counterpart with –a is missing, i.e., it has a completely different meaning. The adjective verde ‘green’ would not be credited one point for gender, because it is invariant. The plural ending –s or –es is also counted in nouns, adjectives, pronouns and articles. As to the verbs, the roots are credited one point, the in¿nitives, gerunds and participles are credited two points and all inÀected forms are credited 5 points for person, number, tense, mood and the root. These few remarks show that, as other languages have been investigated, adaptations have had to be incorporated. The most important issue with MLU is, however, that there is, as of today no consensus amongst researchers of one and the same language as to which counting conventions to use and researchers do not even state clearly how they count. López Ornat (2003) is an exception to

28

María Pilar Larrañaga and Pedro Guijarro-Fuentes

the latter since she showed the calculating rules15. She calculated MLU(m) and MLU s(yllables). The above mentioned problems make it very dif¿cult to compare scores in different studies in the same language. The comparison across different languages is, therefore, impossible. Next, we will outline a wide range of general problems and limitations that have been discussed in the literature mainly by researchers working in the clinical domain16. It has been noted by various scholars that there are different practices amongst researchers as to how morphemes are supposed to be counted which is detrimental to the comparability of studies even in the same language. The ¿rst problem refers to the number of utterances that should be taken into account. Some scholars adhere to Brown’s rules and count only the ¿rst 100 utterances (e.g., Linares, 1981; Linares-Orama and Sanders, 1997; Hickey, 1991), discounting the ¿rst 15 utterances of the ¿rst page of the transcript, others take into account the whole transcript (Plunkett, 1993) to mention but a few. The main issue is that we do not know what inÀuences the different practices as to the number of utterances will have in the ¿nal score. Scarborough (1990:8f.) has shown that text length affects the type-based score index of productive syntax (IPS) she proposed. In an attempt to assess the reliability of the measure it has been shown by Klee and Fitzgerald (1985) that the section of the transcript that is drawn to calculate MLU positively affects the ultimate score. In their crosssectional study, Klee and Fitzgerald (1985:259) made the observation that ’the child’s stage af¿liation may vary as much as three stages in 13 of the 18 children’. Moreover, the reliability of MLU decreases as age increases. Klee and Fitzgerald (1985), amongst others, stated that MLU’s reliably accounts for structural development until around the age of 3.6. Notably, Brown himself insisted upon the fact that MLU did not predict anything beyond the stage V, where children can use subordination and co-ordination. One of the most relevant problems of MLU is not linked to the computation itself. Dewaele (2000:2) points out that the problem is ‘celui de la segmentaion du discours’. 15. It is not clear to us, how morpheme count in her Table 6 has to be interpreted. Whilst the article (l)o(s) ‘the’ is credited one point, the article una ‘a’ is credited 2 in the morpheme count. To us, it seems clear that the missing –s is a phonological issue that should not interfere in the morpheme count since the use of (l)o(s) is evidence for the fact that the child is targeting a plural because the singular and plural morpheme have different roots. 16. For other readers interested in that line of research, they are referred to Eisenberg et al. (2001) and references therein.

Collecting data and measuring competence

29

Dewale discusses this issue at length concluding that discourse segmentation is mainly done applying subjective criteria that may largely differ from the ones used by children. Utterance segmentation can only be done using a range of criteria such as prosodic, semantic and syntactic and rely very much upon the accuracy of the transcriber. Moreover, researchers are reluctant to state which segmentation criteria they used (see Thordardottir and Weismer 1998 for an exception). For example, Rosenthal Rollins, Snow and Willet (1996:249) are an exception in that segmentation criteria are made explicit ‘Utterance boundaries were based primarily on intonation contour and secondarily on pause duration’. Yet another transcribing related issue is the fact that many ad hoc decisions are made when transcribing a recording and when it comes to computing MLU. In this respect, Thordardottir and Weismer (1998:3) suggest that the computing rules must be ‘such that they can be applied consistently, thus minimizing the need for ad hoc decisions’. In the same vein, Crystal (1974) points out that criteria must be applied consistently, if one wishes to compare across different studies and across languages. Even if ad hoc decisions are reduced to a minimum, researchers, in general, do not keep track of them. Thus, we are far from knowing the extent to which those decisions may affect the ultimate MLU score. In sum, there has been a considerable amount of discussion about which utterances to include in the MLU computation, with still no consensus as of today. Some researchers include sub-lexical items in their computations, others exclude them totally. Furthermore, it is not at all clear whether utterances that contain one or two sub-lexical items should be considered for the computation or not. If they should, it is not clear which criteria should be applied. The validity of MLU has been criticised ever since the measure was proposed. Brown conceived the score as a measure of general linguistic development, although it is not quite clear what this general development involves. Scarborough (1990:2) points out that MLU ‘content validity has been seriously challenged on the grounds that qualitative aspects of syntactic development are not directly evaluated’. Plunkett (1993) who studied two Danish speaking children showed that the MLU score was affected differently across transcriptions depending on whether sub-lexical items were counted or not. However, the inclusion of sub-lexical form in the calculation raised the MLU score throughout the study. Plunkett (1993:49) de¿ned sub-lexical items as ‘sequences of sound which cannot be identi¿ed with, nor contain target lexemes’. An issue that has been raised by several researchers is that one word-morpheme utterances have an effect upon the ultimate score. Since one-morpheme utterances do not involve the ordering of morphemes, they do not involve syntax. Therefore, Klee and Fitzgerald (1985) proposed ignoring one-morpheme

30

María Pilar Larrañaga and Pedro Guijarro-Fuentes

utterances in the MLU calculations in order to avoid all pragmatic ‘noise’. In their cross-sectional study, the number of single-morpheme utterances (SMU) ranged between 16% and 45%. The new score was renamed as MSL (mean syntactic length) because it excluded all SMU. The effect was that the MSL score rose considerably and could predict syntactic growth better. Another problem that acquisitionists face, especially those involved in bilingual language acquisition, is that it is dif¿cult to decide which morphemes to count17. Researchers investigating BFLA are confronted with the problem of how to count the morphemes. All scholars are aware of the fact that some languages are more inÀectional than others. Theoretically, the highly inÀected languages should reach higher MLUs. Moreover, some languages have a very developed morphology in one grammatical domain and lack morphology in another area. If we are to compare these two languages, it may turn out that the highly inÀected domain compensates for the less developed one. The problem is that this factor cannot be controlled by only counting the morphemes. This issue is widely ignored by many acquisitionists who take the validity of MLU for granted and compare very different languages without further thought. However, some researchers take a radical position stating that ‘such comparisons may be useless at best and deceptive at worst’ (Hickey 1991:569). It seems advisable to regard MLU as a purely intra-linguistic device, allowing comparisons of the same child’s language over time, and between children acquiring the same language. If one reviews the few studies that have been explicit about how the MLU score has been calculated, it can be concluded that some researchers count gender and number in the NP, others omit it from the computation making comparisons dif¿cult. However, the worst problem associated with MLU is that researchers do not comment on how they calculate MLU and as Hickey (1991:556) states ‘its values are frequently reported without comment’. Finally, a plethora of research has been devoted to the issue of motherese or child directed speech (CDS) ever since it was proposed by Snow (1972). In qualitative terms, it is known that parents, siblings and caregivers tune their language to the child’s language in such a way that their utterances are simpler. 17. A different method to measure language development in bilingual ¿rst language acquisition studies is by looking at the size and growth of the lexicon in the two (or even more) languages and/or number of utterances per session (Müller and Kupisch, 2003). In addition, in dealing with BFLA other methodologies issues dif¿cult to control for are language dominance, language preference, language balance/usage, and so on, which make the task of the researcher even harder. Some researchers have based language dominance on the basis of MLU, but even those studies have not escaped criticisms (Arencibia-Guerra, 2008).

Collecting data and measuring competence

31

Most of the studies on CDS have been conducted on English and most notably with impaired language. Our sample differs signi¿cantly from the ones found in the literature in that we study two normal developing Spanish speaking children. Since we are not aware of any studies which have tested the impact of the input on the child’s output in Spanish, this paper will ¿ll this gap.

5.

Present study: Research questions

From all the above, a number of conclusions can be drawn: MLU is a widely accepted measure (Hickey, 1991) but suffers from a number of drawbacks. It is beyond any doubt that different word classes impact differently on its ¿nal score as we have seen in a cursory frequency analysis of some word classes above. Moreover, the impact of the input has not been analyzed in a systematic manner for Spanish. Clitics and articles are grammatical words that potentially could increase the MLU score if we believe Brown’s words that every new grammatical or semantic role increases length. As of today, it is not known what the impact of clitics on MLU is in Spanish language development. Therefore, this study aims at ¿lling these two gaps. Hence, we will address two research questions: (a) What is the impact of the clitics on the ¿nal MLUw score in early Spanish? (b) What is the impact of the input on clitic use score in early Spanish? 5.1.

Methodology

In order to address the above research questions, two children have been analysed, Irene and Magín from the Irene and Aguirre corpora available in Childes, http://childes.psy.cmu.edu. Magín is a male monolingual Spanish child who was recorded from the age of 1;07 until the age of 2;10,2418. Irene is a monolingual female Spanish child who was recorded from the age of 0;11,01 to 3;2,19. In the case of Magín, 29 recordings of different duration are available, Irene has 60 recordings. We calculated all values without excluding repetitions and self-repetitions, and counting all one-word utterances. For the purposes of this paper, we counted as clitics, object and subject clitics as well as monosyllabic articles. The list of clitics and articles is as follows: me, te, se, lo, la, los, las, le, les, nos, os, el, un. We excluded some articles such 18. This is a ¿rst exploratory study. Whether or not our results can be replicated remains open for future research.

32

María Pilar Larrañaga and Pedro Guijarro-Fuentes

as una, unos, unas because they are disyllabic and do not clitizise. The plurals are not frequent in oral discourse and, hence, their impact on MLU is negligible. The former is more frequent but its impact will be discarded. In order to calculate MLU, we used the tools available in CHILDES, the programme clan. We calculated MLU in words for three reasons: (i)

As we have argued above, clitics in Spanish have a syntactic status and not a morphological status (see section 3). (ii) The available ¿les in CHILDES are not lemmatised and lemmatisation is time consuming. (iii) And more importantly, according to (Aguado, 1995) there is a high correlation between MLUm and MLUw in Spanish. We counted the clitics using the tool freq19 available in clan. 5.2

Results

Figure 1 represents the MLU(w) for both children and both mothers where age has been matched. The MLU values of all transcripts in one and the same month were conÀated in order to compare both children in an age matched Figure. Several observations can be drawn from the ¿gure:

Figure 1. Age matched MLU by Magín, Irene and their respective mothers.

19. Freq counts the total frequency of a given word or morpheme.

Collecting data and measuring competence

33

a) As expected, MLU increases with age in both children. b) Irene has a higher MLU than Magín at all stages. Irene’s mother has a higher MLU than Magín’s mother. c) The MLU of the mothers differs considerably, by approximately 2 points. d) Interestingly, the MLU score of the mothers increases with age as well. The increase with age mirrors the MLU increase by their respective children. e) From age 2;6 onwards the MLU scores do not increase linearly, but oscillate both in the child Irene and her mother. In the case of Magín, the increase is linear. Note, however, that his recordings end at the age of 2;7 (There is one more recording at 2;10). We conducted a correlation analysis using Pearson and Spearman in order to evaluate the magnitude of the impact. Magin’s and his mother’s as well as Irene’s and her mother’s MLU show a high correlation which is signi¿cant at the 0,01 level with both methods of analysis. Rho = 0,857 in Magín’s case (n = 29) and 0,636 in Irene’s case (n = 27). In other words, 73,4% of the variance in Magín and 40,45% in Irene’s case can be accounted for by the interaction mother-child. Hence, it can be stated that the length of the utterances of children and mothers is linked in a way in which it is impossible to tease apart how much of the MLU is induced by the input the children are exposed to and how much of the variance is due to the measure points taken into account. The impact of the input on clitic use score was the second research question. In order to assess the inÀuence of the clitics on the ultimate MLU of children and mothers we counted all clitics and divided the total number of clitics by the number of utterances and by the total amount of words used. In a ¿rst step we will compare the impact of clitics of the children against their respective mothers. In a second step we will compare the impact of the clitics against the MLU. With respect to the impact of the clitics per utterance, one can make a number of interesting observations according to Figures 2 and 3: a) The ratio clitic per utterance is lower for children than for their mothers. b) The ratio clitic per utterance increases over time in all four subjects. c) The impact of clitics ranges from 0 clitics per utterance at the beginning of the study to around 0,6 for Magín and around 0,8 for Irene at the end of the studied period. d) The number of clitics per utterance increases over time for the parents but the increase is much lower than in the case of the children. In fact, the increase is of around 0.4 whilst the increase in the case of the children is around 0.8.

34

María Pilar Larrañaga and Pedro Guijarro-Fuentes

Figure 2. Clitic per utterance by Irene and Irene’s mother.

Figure 3. Clitic per utterance by Magín and Magín’s mother.

Bivariate correlation analyses were conducted in order to check whether the quotient clitics per utterance of the children and their respective mothers were statistically linked. In Magín’s as well as in Irene’s case a high correlation signi¿cant at the 0.01 level can be reported. A high variance amounting to 49,84 % can be attributed to the interaction between mother and child in the case of Irene (rho of Spearman 0,706) and 41,47% in the case of Magín (rho of Spearman 0,644). We were also interested in the question of whether the total number of words used was linked to the total number of clitics. The quotient between

Collecting data and measuring competence

35

the total number of clitics and the total number of words was correlated in a bivariate analysis. No signi¿cant correlation can be reported for Magín and his mother according to the Pearson correlation, but the analysis conducted using spearman throws signi¿cant results at the 0,05 level with rho equalling 0,409. In other words, 16,72% of the variance can be attributed to the interaction between Magín and his mother. By contrast, the quotient clitics per total number of words is positively correlated in the case of Irene and her mother using both Pearson (p = 0,05) and Spearman (p = 0,01). Rho of Spearman amounts to 0,497 whereby the variance found accounts to 24,70%. Summarizing our main ¿ndings so far, a high correlation between the MLU, quotient clitics per words and clitics per utterance of children and their mothers can be reported. Let us now look at the ratio clitic per utterance plotted against the MLU in order to examine whether MLU is affected in any signi¿cant way by the presence of clitics over time. Figures 4 and 5 show the MLU and clitics per utterance by Irene and Magín and their respective mothers, according to age. These ¿gures have not been age matched but the effects shown in Figure 1 prevail. Figures 4 and 5 demonstrate the following:

Figure 4. MLU and clitic per utterance by Magín and his mother.

36

María Pilar Larrañaga and Pedro Guijarro-Fuentes

Figure 5. MLU and clitic per utterance by Irene and her mother.

a) The ratio clitic per utterance increases in all subjects over time. b) The increase of this ratio is higher in the case of children than in their mothers. c) This increase remains in all instances below the one point mark. d) The increase of this ratio is lower than the increase in MLU at all times. e) And, more importantly, the impact clitic per utterance does not affect MLU in any signi¿cant way at any time. Its impact does not increase; on the contrary, it decreases over time. In order to test whether the quotient clitics per utterance correlates with the MLU, we conducted a bivariate correlation analysis using Pearson and Spearman. For both Magín and Irene a high correlation between MLU and the quotient clitics per utterance signi¿cant at the 0,01 level can be reported. Moreover, a large amount of the variance can be attributed to the interaction MLU and clitics per utterance in Irene (rho of spearman 0,909, variance 82,62%) and Magín (rho of spearman 0,643, variance 41,34%). Finally, we tested whether the ratio total amount of clitics per total amount of words were statistically linked. For both Irene and Magín statistically signi¿cant results at the 0.01 level can be reported. Moreover, a large amount of the variance can be accounted for by the interaction of MLU and the ratio clitics per words, i.e. 37,33% (rho of Spearman 0,611) for Irene and 41,34% (rho of Spearman 0,643) for Magín. In sum, it can be showed that MLU and clitics, clitics and number of words and number of utterances are linked for the children. But, what about the mothers? A bivariate correlation analysis shows that MLU and the ratio clitics per utterance are signi¿cantly correlated at the 0,01 level (Irene’s mother rho of

Collecting data and measuring competence

37

Spearman 0,417, variance 17,39%, Magín’s mother 0,619, variance 38,32%), whereas MLU and the ratio clitics per words are not.

6.

Concluding discussion

We set out this study with the idea of investigating two questions: a) What is the impact of the clitics on the ¿nal MLUw score in early Spanish? b) What is the impact of the input on clitic use score in early Spanish? In what follows, we will present the two main ¿ndings and discuss their relevance for the computation of MLU and clitic use. To begin with, the results obtained in the previous section clearly demonstrate that the impact of clitics (for our purposes subject and object clitics and articles) on the total MLU increases over time, but is minor at all times. More importantly, they show that the impact always remains below the one point mark, at around 0.6 for both mothers and children. Despite the increase over time, the increase of MLU is much higher overall so that the impact of clitic per utterance remains negligible. More importantly, we were able to show that both MLU and the ratio clitic per utterance increase in both children, but these measures also increase with about the same gradient in the mothers. Let us now discuss these results. It is somewhat surprising that clitics have such a minor impact on the whole MLU. Articles are compulsory words in a Spanish NP if we exclude generic readings that are not very frequent in Spanish. Moreover, we counted subject and object clitics. These two types of words are very frequent grammatical words that allegedly increase the MLU in a signi¿cant way. We have shown that this was a mistaken belief. Let us now consider the Davies corpus in more detail repeated here as Table 5. In this corpus, pronouns and articles amount to 19,47% of the total words. Table 5. Total amount of words (20.000.000 words) in The Davis Corpus Verbs

Nouns

Pronouns Articles

775,945 718,266 218,726 6,38%

448,327 13,09%

Adjectives 214,497

Adverbs Prepositions 338,23120 214,479

20. The corpus includes the negation amongst the adverbs.

Conjunctions 495,070

38

María Pilar Larrañaga and Pedro Guijarro-Fuentes

If we compare the Davies corpus with our own data (Table 6) an observation comes to light. Around 6% to 7% fewer pronouns and articles are used by all subjects. This discrepancy, however, has a simple explanation. We did not count the articles una, unas-unos, the stressed subject pronouns and demonstrative pronouns. We do not know the exact impact of these words on our corpus, but it is plausible that they represent around 5% of the total amount in which case the distribution of pronouns and articles would be almost identical in the Davis corpus and our own. Table 6. Pronouns and articles Irene Irene’s mother Magín Magín’s mother

Total words 38,994 77,258 25,573 37,591

% clitics (n) 12,8 (n = 5,002) 12,3 (n = 9,491) 14,34 (n = 3,667) 14,07 (n = 5,289)

Daller et al. (2007:10) make the correct observation that “verbs such as make or do, prepositions such as in and on and pronouns such as I or you are used a lot by every speaker”. That is to say, these types of words are far more frequent than others in adult speech and as a consequence it can be assumed that the same is true in child language as well. In our case, articles and pronouns amount to around 13% of the entire corpus, whichever data we use. However, this overall high frequency of clitics has only a minor inÀuence in the ultimate MLU. This amounts to saying that other word classes impact more on utterance length than clitics and articles. If the distribution of pronouns and articles is similar in both corpuses, it is very plausible that nouns and verbs represent the gross of the MLU at all stages of development. Finally, and since articles and clitics are purely grammatical words, one is tempted to argue that lexical items impact more on the total MLU than grammatical words. We did not investigate this point in this paper and leave it open for future work. From the results concerning the input, the ¿rst issue that comes to mind is whether the increase in MLU by the mothers is a kind of motherese effect which includes features as high intonation contour, lexical choice, diminutives etc21. It is well known that parents attune their language to the competence and 21. We are cautious with some of our claims made herein as one of the reviewers reminds us. To verify or falsify them, the reviewer in question made the following observation, which we agree with, that in a future study researchers interested in the topic would need to look at MLU in more depth since we have assumed for Spanish

Collecting data and measuring competence

39

needs of their children in early stages (Snow, 1972). However, we are not aware of the fact that the attuning would involve the adaptation in terms of length of utterance. The other possibility is, however, that children adapt their language to the one of the interlocutor in some way. Our data show clearly, that Irene’s MLU is higher than Magín’s at all stages. Interestingly Irene’s mother has a higher MLU than Magín’s mother as well. This clearly demonstrates that the MLU level of the mothers impacts on the MLU of the children. From the data we have collected, we are not able to quantify how much of the MLU increase by the children is due to their language development and how much of it is due to the input they are exposed to. If children’s MLU increase is not only due to their language development but also mirrors the MLU of the mothers to some extent, researchers working with normal developing children should seriously think about whether to use this measure or not, since it is not clear at all how much of the increase can be attributed to genuine language development. MLU is widely used by investigators researching syntax in First Language Acquisition (FLA), mainly following Brown’s suggestion that an increase in pro¿ciency results in an increase in the MLU score. As a matter of fact, a large number of studies have been able to show that MLU increases with age and increasing language pro¿ciency, something that is also con¿rmed in our own study. One of the main ¿ndings is that around age 2, when MLU surpasses the level of 2, children start using inÀected utterances across languages. Hence, there is a major developmental turning point in FLA at this age; the child overcomes the one word stage and is able to combine words. According to Brown, stage V in language acquisition is attained when subordination and co-ordination are acquired. If different studies are reviewed (Ezeizabarrena, 1996; Barreña, 1995, to name but only two), one encounters one consistent ¿nding: the category I(nÀ) is acquired around MLU 2–2.5. For category C(omplementizer) MLU 3–3.5 is found. Taking into account our own ¿ndings about the role input plays, one should question whether MLU reliably depicts language development. We need to be language development that mother MLU drives child MLU; however, we are also aware of the fact that the literature on child-directed speech for mainstream English speakers has shown that mothers attune to their child’s language level. Thus, there are reasons for thinking that there might be a transactional pattern here with a signi¿cant impact of child MLU on parent MLU rather than a uni-directional impact of mother to child across languages. However, we leave this issue for a future study. Nevertheless, what our data is clearly instructing us is on relative frequency of clitics in mother and child language - data that shows the same attunement pattern that has previously been reported for utterance length, that mother’s clitic use seems to stay just above child use.

40

María Pilar Larrañaga and Pedro Guijarro-Fuentes

cautious with our conclusions based on two children, but it would be interesting if future studies could investigate whether the MLU of the parents correlates with the MLU of normally developing children in order to discard the hypothesis that we are facing two idiosyncratic cases. In conclusion, the present study was able to show that clitics’ impact on the entire MLU can be regarded as minor. Moreover, a topic which has not attracted the attention of researchers working with a normally developing population seems to play a major role which cannot yet be quanti¿ed. Further studies with a larger population will have to investigate whether the characteristics of input indeed affect the MLU of children. Since the range of variance which can be explained by the interaction child-interlocutor varies considerably from child to child, it would be advisable to include the interlocutor’s MLU in future studies so as to be able to explain eventual differences between children’s performances in language development. Moreover, naturalistic settings facilitate collecting a given amount of data, but it would be advisable to supplement these data with data coming from experimental settings in order to capture low frequency phenomena.

References Aguado, G. 1995 El desarrollo del lenguaje de 0 a 3 años. Madrid: CEPE. Arencibia-Guerra, L. 2008 Sprachdominaz bei bilingualen Kindern mit Deutsch und Französisch, Italienisch oder Spanisch als Erstprachen. PhD dissertation, Bergische Wuppertal University. Barreña, A. 1995 Gramatikaren jabekuntza-garapena eta haur euskaldunak [El proceso de adquisición de la gramática y los niños vascos]. Bilbao: Servicio Editorial de la Universidad del País Vasco. Bel, A. 2001 Sujetos nulos y sujetos explícitos en las gramáticas iniciales del castellano y el catalán. Revista Española de Lingüística, 31, 2,. 537–562. Belletti, A. 1990 Generalized Verb Movement: Aspects of Verb Syntax. Turin: Rosenberg and Sellier. Borer, H. 1984 Parametric Syntax. Dordrecht: Foris Publications.

Collecting data and measuring competence Brown, R. 1973

41

A ¿rst language: The early stages. London: George Allen and Unwin Ltd. Cardinaletti, A. 1994 On the internal structure of pronominal DPs. Linguistic Review 11, 195– 219. Casielles, E., J. Andruski, S. Kim, G. Nathan and R. Work 2006 Syntactic and Discourse Features of Subjects in Child Spanish: Evidence from Spanish/English Bilingual Acquisition. In D. Bamman, T. Magnitskaia, and C. Zaller. BUCLD 30: Proceedings of the 30th annual Boston University Conference on Language Development 72–83. Crystal, D. 1974 Review of Brown, A First Language. Journal of Child Language 1. 289– 306. Daller, H., J. Milton and J. Treffers-Daller (eds.) 2007 Modelling and Assessing Vocabulary Knowledge. Cambridge: Cambridge University Press. Dewaele, J.-M. 2000 Saisir l’insaisissable? Les mesures de longueur d’enoncés en linguistique appliquée. IRAL 38. 17–38. Einsenchlas, S. 2003 Clitics in Spanish. First Language 23, 193–211. Eisenberg, S., T. McGovern and C. Lundgreen 2001 Use of MLU for identifying language impairment in preschool children: a review. American Journal of Speech-Language Pathology, 10, 323– 342. Ezeizabarrena, M.J. 1996 Adquisición de la morfología verbal en euskera y castellano por niños bilingües. Bilbao: Universidad del País Vasco. Franco, J. 1993 On Object agreement in Spanish. Doctoral Dissertation. University of Southern California, Los Angeles. Grinstead, J. 2000 Case, inÀection and subject licensing in child Catalan and Spanish. Journal of Child Language 27, 119–155. Guijarro-Fuentes, P. and L. Ortiz-Lopez 2008 Creole/Spanish contact and the acquisition of clitics on the DominicanHaitian border. International Journal of Bilingualism 12 (4), 231–262. Hickey, T. 1991 Mean length of utterance and the acquisition of Irish. Journal of Child Language 18. 553–569. Jaeggli, O. 1982 Topics in Romance Syntax. Dordrecht: Foris Publications.

42

María Pilar Larrañaga and Pedro Guijarro-Fuentes

Klee, T. 1992

Developmental and diagnostic charactristics of quantitative measures of children’s language production. Topics in Language Disorders 12. 28– 41. Klee, T. and M. D. Fitzgerald 1985 The relation between grammatical development and mean length of utterance in morphemes. Journal of Child Language 12. 251–269. Kuchenbrandt, I. 2008 Cross-linguistic inÀuences in the acquisition of grammatical gender? Arbeiten zur Mehrsprachigkeit 86. Folge B. Larrañaga, M.P. 2000 Ergative Sprachen, akkusative Sprachen: der Erwerb des Kasus bei bilingualen Kindern. Frankfurt:Vervuert. Larrañaga, P. and P. Guijarro-Fuentes in press Clitics in L1 bilingual acquisition. First Language. Linares, N. 1981 Rules for calculating mean length of utterance in morphemes in Spanish, in: Erickson, J.G. and D.R. Osmark, Communication Assessment of the Bilingual Bicultural Child. Baltimore: University Park Press. 291–295. Linares-Orama, N. and L.J. Sanders 1997 Evaluation of Syntax in three-year-old Spanish-speaking Puerto Rican children. Journal of Speech and Hearing Research 20. 350–357. López Ornat, S. 2003 Learning earliest grammar: evidence of grammar variations in speech before 22 months. In S.Montrul and F.Ordóñez (eds.) Linguistic Theory and Language Development in Hispanic languages. Sommerville, MA:Cascadilla Press, 254–274. Malvern, D. and B. Richards 1997 A new measure of lexical diversity. In A. Ryan and A. Wray Evolving models of language . Clevedon, UK: Multilingual Matters. 58–71. Miller, J.F. and R.S. Chapman 1981 The Relation between Age and Mean Length of Utterance in Morphemes. Journal of Speech and Hearing Research 24, 154–161. Müller, N. and T. Kupisch 2003 Zum simultanen Erwerb des Deutschen und des Französischen bei (un) ausgeglichen bilingualen Kindern. Vox Romanica 62. 145–169. Nice, M.M. 1925 Length of sentences as a criterion of a child’s progress in speech. Journal of Educational Psychology, 16, 370–379. Ortiz-Lopez, L. and P. Guijarro-Fuentes 2008 Objetos directos (nulos) en la frontera dominico-haitiana: ¿contacto de lenguas o motivaciones internas? Spanish in Context 5 (1), 110–141.

Collecting data and measuring competence Plunkett, K. 1993

43

Lexical segmentation and vocabulary growth in early language acquisition. Journal of Child Language, 20, 43–60.000. Richards, B., and R. Malvern 2007 Validity and threats to the validity of vocabulary measurement. Modelling and Assessing Vocabulary Knowledge. Cambridge: Cambridge University Press. 79–92. Rosenthal Rollins, P., C. Snow and J.B. Willet 1986 Predictors of MLU: semantic and morphological developments. First Language 16. 243–259. Scarborough, H.S. 1990 Index of productive syntax. Applied Psycholinguistics. 1–22. Scarborough, H.S., J. Wyckoff and R. Davidson 1986 A reconsideration of the relation between age and mean utterance length. Journal of Speech and Hearing Research, 29, 394–399. Snow, C.E. 1972 Mothers’ speech to children learning language. Child Development, 43, 549–565. Thordardottir, E.T. and S.E. Weismer 1998 Mean length of utterance and other language sample measures in early Icelandic, in: First Language 18, 1–32. Tomasello, M. and D. Stahl 2004 Sampling children’s spontaneous speech: How much is enough? Journal of Child Language, 31, 101–121. Uriagereka, J. 1992 Extraction Parameters: A case study on underspeci¿cation. Unpublished MS. University of Maryland. Uriagereka, J. 1995 Aspects of the syntax of clitic placement in Western Romance. Linguistic Inquiry 26: 79–123. Zagona, K. 2002 The Syntax of Spanish. Cambridge: Cambridge University Press.

On the prosodic structure of articles in L1 acquisition of Spanish Conxita Lleó

1.

Introduction

It is common knowledge in the ¿eld of L1 acquisition that children begin by producing lexical items and omitting function words. There have been several attempts to explain this phenomenon, which have generated a debate about how to interpret the missing category. In the last years, a prosodic point of view has been introduced into the debate, according to which function words are not produced, because they are not licensed prosodically (Demuth 2001, Lleó 2001a, 2001b, 2001c, 2003a, Lleó and Demuth 1999). That is, children begin producing simple prosodic structures, namely words that can be represented by means of one single foot, and once these structures become more complex, there will be room for functional categories. However, what is the nature of the evolution of function words in the process of acquisition? What is their prosodic status? How does their prosodic status evolve? We know the initial state, when function words are absent, and the ¿nal one, when they are produced target-like, but what lies in between the beginning and the end? These questions pose a real challenge to phonologists trying to understand the acquisition process. In this article, we will discuss some aspects of the structure of function words, especially articles, within the L1 acquisition process of Spanish. In the nineteen eighties and nineties, the missing functional elements were described within the model of Principles and Parameters as syntactically nonexistent (Radford 1990). According to this view, children begin by acquiring lexical items, and functional elements come in later, because they are dependent on the maturation of syntactic principles licensing their occurrence. Once syntactic principles are in place, functional elements are predicted to appear rapidly. However, reality delivers a slow picture: it takes many months before a child produces 100% of functional elements. Lately, there have been other proposals, based on prosody. Whereas lexical words are organized around metrical feet, belonging to the PW (Prosodic Word), and are the bearers of stress, functional words are often unstressed and may ¿ll extrametrical positions, like unfooted syllables, i.e. syllables not dominated by the foot, but

46

Conxita Lleó

directly dominated by the PW or the PPh (Phonological Phrase). Only when children have developed the necessary prosodic positions in the PW and in the PPh will they be able to produce function words. As Demuth, Patrolia, Song and Masapollo (2012) put it, “children tend to produce their ¿rst grammatical morphemes in phonologically simple, ‘unmarked’ contexts.” This has been called the Prosodic Licensing Hypothesis (Demuth et al. 2012). However, the pace with which marked prosodic positions are acquired depends on factors like certain prosodic characteristics of the target language, as well as on the individual tendencies of a child. The ¿rst function words that children acquire are articles, as, in general, children begin to produce nouns before producing verbs. From the point in time when children begin to produce their ¿rst articles, until they provide all necessary articles in an utterance, several months (even one whole year) may elapse. From a prosodic point of view, when children begin to develop extrametrical or marked prosodic positions, these are still unsteady, and children may go back to previous production patterns, before a certain prosodic position is available. Moreover, functional morphemes, i.e. articles, tend ¿rst to be produced as default forms, which resemble simpli¿ed versions of the target form, generally keeping the vowel. These simpli¿ed forms, i.e. proto-articles, gradually become real articles. They may also involve allomorphy, as, for example, the nasal consonant of the inde¿nite article undergoes assimilation of PA (Place of Articulation) in Spanish. Previous studies on the acquisition of articles in several languages, like English, French or Spanish, have offered different prosodic descriptions of the articles in the child’s development and different explanations for their evolution (see e.g. Demuth and McCullough 2009 for English; Demuth and Tremblay 2008 for French; Demuth et al. 2012, and Lleó 2001a, 2001c, and 2003a for Spanish). In all three languages, articles begin to appear at about 1;7, according to Demuth et al. (2012), but they develop at different pace depending on the prosodic structure of the language. In the acquisition of English and French articles, the number of syllables of the following noun is generally a conditioning factor for the production or omission of the article, and both French- and English-speaking children need a relatively long time to produce articles with disyllabic nouns, as in this context articles are unfooted syllables (Demuth and Tremblay 2008). In Spanish, Demuth et al. (2012) studied the development of articles in two Spanish children of the CHILDES database, Irene and Emilio. At a certain age, Irene at 1;7 and Emilio at 2;1, they produced more (monosyllabic) articles with disyllabic than with trisyllabic nouns. The two children had different outcomes concerning trisyllabic nouns. Whereas Emilio did not exhibit truncation and produced all three syllables of the

On the prosodic structure of articles in L1 acquisition of Spanish

47

noun before producing articles, Irene seemed to alternate between producing a disyllabic (truncated) version of the target noun preceded by the article, or the non-truncated trisyllabic noun without article. The authors propose that at this early stage, the article is a constituent of the PW, and only later will belong to the higher PPh. On the one hand, Lleó (2001a) looked at the development of articles in one Spanish child, María, who produced ¿ller syllables and proto-articles for quite a long period of time. This protracted usage of the unspeci¿ed phonological forms of ¿ller syllables points to lack of autonomy of these functional elements, and thus to their treatment as being constituents of the PW. On the other hand, Lleó (2003b) presents some data from another monolingual Spanish child, José, who starting at about 2;0 seems to have already developed the PPh node, and to place some of his articles there. Lleó (2006) shows that another Spanish child, Miguel, masters the PPh node at 1;5, and assumes that some articles are already adjoined to this node at a very early age. Goad and Buckley (2006) analyze the early production of articles in French by the child Clara, and propose that whereas the initial unstressed syllable of lexical items belongs to the PW, the article is already placed into the PPh. Goad and Buckley (2006: 114) contend that because of the different treatment given by the child to these syllables – the one belonging to the functional word and the other one to the lexical item – the functional element cannot belong to the PW. But as shown in Lleó (2001a) and below (Section 3.1), some children, like José and María, during a certain time do treat functional and lexical syllables alike, justifying the PW stage for certain function words, at least for some children. The present article describes the target articles (Section 1.1) and their target prosodic structure (Section 1.3). After de¿ning some criteria that a child form has to ful¿ll, in order to be considered an article or a proto-article (Section 1.2), the study is presented, beginning with the research questions (Section 2.1) and a brief description of the corpus analyzed (Section 2.2). The child data that offer evidence in favor of certain prosodic representations of articles at various stages of development are presented and analyzed in Sections 3.1 through 3.7. Finally, Section 4 presents the discussion and some concluding remarks. 1.1.

Some functional categories in the target language Spanish

A brief description of target articles will allow us to get a picture of what the end state of their acquisition should be like. Spanish has two articles, de¿nite and inde¿nite, varying for gender and number. Articles always precede the noun and, in the case of the de¿nite article, because it does not have a stress of its own, it is cliticized to the noun, and is thus proclitic. The inde¿nite article is

48

Conxita Lleó

generally described as being stressed, which is especially clear in those cases, in which it is disyllabic (see Navarro Tomás 1963: 193).1 When necessary, verbal clitics will also be drawn upon for comparison. There are ¿rst-, second- and third-person verbal clitics in Spanish. They can appear as enclitics or proclitics. The latter precede ¿nite verb forms. We will only consider proclitics here, as they constitute a comparable counterpart to the articles. Note that within articles, we describe the de¿nite ones, which are always considered to be clitics, as well as the inde¿nite ones, which in the target language are supposed to bear stress, and whose clitic status is thus questionable. Table 1. List of Spanish functional words, analyzed in the present study ARTICLES De¿nite Inde¿nite Masc. Fem. Masc. Fem. Sg. el la un una Pl. los las unos unas

1.2.

VERBAL CLITICS 2nd Pers. 3rd Pers. 1st Pers. Masc./Fem. Masc./Fem. Masc. Fem. me te lo/le/se la/le/se nos os los/les/se las/les/se

Criteria for the status of articles and proto-articles

A careful consideration of articles, from their emergence as ¿ller syllables to their full-Àedged production, delivers a continuous development to more speci¿city regarding segmental properties and to more autonomy regarding prosodic properties (see Lleó 2001a). This apparent continuous development obscures the point at which a form can be properly considered to be an article. We will distinguish those default or segmentally underspeci¿ed forms, which are generally considered ¿llers,2 from real articles, even if not yet completely Àedged out, in which case they are proto-articles. For that purpose, we will de¿ne some criteria that will allow us to identify articles and proto-articles. In Lleó (2001c) criteria were formulated, which with some modi¿cations can be applied here, too. 1) Segmental accuracy. A target-like form can clearly be identi¿ed as an article. The problem, again, is posed in the opposite case, if segments are not target-like. This can be due to a limited phonological inventory, so that consonants (and/or vowels) may be substituted by default ones, like laryngeals. Because articles often appear in syllables that are unstressed and prosodically defective, as in the case of the unstressed syllable of the Spanish de¿1. The plural forms can be both stressed and unstressed, according to Navarro Tomás (1963: 193–194). 2. See Lleó (2001b) for arguments in favor of treating ¿ller syllables as proto-articles.

On the prosodic structure of articles in L1 acquisition of Spanish

49

nite article, segments ¿lling those positions are also relatively unspeci¿ed. 2) In some cases, especially in trisyllabic nouns, the initial unstressed syllable has a default onset. There are two situations: the initial syllable onset is linked to the PA of the following syllable onset. Alternatively, the onset is laryngeal, or there is no onset. In the former case, the initial syllable gets its PA by spreading from right to left, and is interpreted as belonging to the lexical item. In the latter case (laryngeal or no onset) it will be considered a proto-article, especially if the following two syllables have already been produced by the child as a truncated form of the target trisyllabic noun. 3) Are the codas of articles produced at the same pace as the codas of lexical items? A faster pace of coda production in articles would speak for a greater autonomy of articles. Conversely, a slower pace of coda production in articles would point to lack of autonomy. 4) Articles should not be followed by a rhythmic pause, i.e., they should build some type of prosodic unit with the following noun. 1.3.

Prosodic structures for functional items

Regarding the prosodic theoretic background of the present paper, we will apply a hierarchical model of prosody, based on the Prosodic Hierarchy, according to which, prosodic constituents are organized as in (1), where only those nodes that are relevant to the present article are represented. Another important theoretical issue concerns the prosodic status of function words, which was discussed by Selkirk (1996) in a seminal work that has inspired much research on acquisition. The alternative structures are presented in (2). We will concentrate on those structures that represent possible candidates of prosodic constituency for functional items. From all possible function words, we will focus on those that are acquired ¿rst, namely, articles. (1) Some constituents of the Prosodic Hierarchy (Selkirk 1984; Nespor and Vogel 1986) Intonational Phrase

IP

Phonological Phrase

PPh

Prosodic Word

PW

Foot

Ft

Syllable

m

50

Conxita Lleó

(2) Alternative prosodic structures for functional items a. b. c. d.

free clitic af¿xal clitic internal clitic prosodic word

(fnc (lex)PW )PPh ((fnc (lex)PW )PW)PPh ((fnc lex)PW )PPh ((fnc)PW (lex)PW )PPh

The decision about the prosodic level at which an item should be adjoined and its concomitant prosodic representation is generally based on phonological processes and phonotactic constraints. We will thus try to use the available evidence, to de¿ne the more adequate prosodic structure of the items under analysis. In Spanish, phonological processes are applied to all levels of the Prosodic Hierarchy, i.e. to large constituents, up to the IP (Intonational Phrase). Accordingly, Spanish is a grouping language in Lleó and Vogel (2004)’s terms, which makes it dif¿cult to ¿nd compelling evidence for prosodic constituency based on such processes. In other words, in Spanish there hardly are any processes which only apply to a small constituent like the PW. However, phonological processes are applied to different degrees, depending on the prosodic level of the constituent. For example, assimilation of the PA of nasals obligatorily applies within the PW and also to the PPh comprised of inde¿nite article plus noun,3 but it applies only optionally between PPhs. The inde¿nite article obligatorily undergoes PA assimilation, which entails that it should belong to a PW or to a PPh, but no evidence is provided to decide which one of the structures in (2) is the most adequate one. However, there is a phonotactic fact that can contribute an important criterion to the decision, namely PW-initially, only the trill /r/ can appear, the tap /5/ being excluded. In order to exclude the tap /5/ initially, a noun like rana ‘frog’ must constitute a PW, also when preceded by the de¿nite article, la rana ‘the frog’. That is, independently of whether the article builds a PW together with the noun, or not, it is crucial that a noun like rana has the status of a PW, and has its own PW node, to exclude the tap /5/ from the left edge of the noun. According to the phonotactis of /5/-/r/, we thus must reject structure (2c), because in that structure the lexical item rana does not constitute a PW by itself. We can also appeal to stress as a criterion for prosodic constituency. Here, we have to distinguish between the de¿nite article and the inde¿nite article, as 3. This constituent is considered a CG (Clitic Group) in Nespor and Vogel (1986), and a PPh in most other work on the topic. We will represent it as a PPh, in order to avoid the proliferation of levels. Moreover, most of the studies that we are referring to here dispense with the CG.

On the prosodic structure of articles in L1 acquisition of Spanish

51

the former is unstressed but the latter is generally attributed stress, especially in the feminine form una. The unstressed character of the de¿nite article is especially clear when it combines with a trisyllabic noun comprised of an unstressed syllable followed by a trochee, as por example la pelota ‘the ball’. In such cases, the (de¿nite) article (la) together with the ¿rst unfooted syllable of the noun (pe) does not receive secondary stress, which means that a) the two syllables do not constitute a foot, and b) the clitic is not a PW, nor part of a PW. This entails that in the case of the de¿nite article, (2d) should be rejected, because if the de¿nite article were a PW it would be able to receive stress. Moreover, the structure (2b), where fnc corresponds to one syllable, should be rejected for the disyllabic inde¿nite article, because the two syllables would be split between the two adjoined PWs, and would thus be unable to build a foot, and to receive stress. Having excluded (2b), (2c) and (2d) for various reasons, we are only left with (2a), where the lexical unit is adjoined to PW, and the clitic is not. In order to account for both articles (de¿nite and inde¿nite) with the same structure, we select the free clitic status. Thus, the most adequate structure for the Spanish articles is (2a). However, in order to account both for the monosyllabic (de¿nite) and the disyllabic (inde¿nite) article, structure (2a) should be split into two structures, which we present as (3a) and (3b).

(3a). free clitic: de¿nite article

(3b). free clitic: inde¿nite article

PPh

fnc

PPh PW

Ft

PW

lex

fnc

lex

Verbal clitics should arguably be analyzed by means of structure (3a) as free clitics, as well, because they are invisible to stress assignment. Actually, in enclitic position, verbal clitics offer a strong argument against being part of the PW, for the following reason. The verb receives stress in one of the last three syllables, as Spanish abides by the three-syllable window. Stress assignment is independent of the number of enclitics following, as e.g. entrégamelo “give it to me”, cárgaselo “charge it to him/her”, devuélvemelo “give it back to me”. In proclitic position they do not count for stress, either. Neither de¿nite articles nor verbal clitics receive secondary stress, independently of the number of pretonic syllables in the utterance. We can thus assume that, when they ¿ll the proclitic position, verbal clitics do also belong to the class of free clitics (3a).

52

Conxita Lleó

Summarizing, (3a) and (3b) are the target structures of articles, and (3a) is that of verbal clitics, too. Accordingly, these are the structures that children must acquire in Spanish to produce articles and verbal clitics. However, it is possible that in the course of acquisition children go through stages, in which the functional words under study are represented differently, building other constituent structures.

2.

The present study

2.1.

Research questions

Most of the research done on the prosody of function words in L1 acquisition has focused on articles, which are the ¿rst functional elements to be produced by young children. In spite of much work done in this domain, many issues are still unresolved. On the basis of data produced by three very young monolingual Spanish children, involving ¿ller syllables, proto-articles and articles, we want to tackle the two research questions posed in the Introduction, which are recapitulated here for convenience: 1) What is the nature of the evolution of articles in the process of acquisition? 2) What is their prosodic status, and how does this evolve? These research questions can be expanded as follows: 1. What evidence is there to propose a stage in which articles (and proto-articles) belong to the PW or to the PPh? 2. What is the difference between lexical and functional elements in prosodic terms? 3. Is the course of article acquisition in Spanish the same for all children? i.e. do articles follow the same evolutionary path in the course of acquisition of L1 Spanish, independently of individual differences? In order to answer these research questions, we have conducted several analyses on the data produced by three Spanish-speaking children, beginning with the production of the ¿rst proto-articles until about 2;3, when articles and proto-articles were abundantly produced. We have included some verbal clitics in the present study. However, the data available on their acquisition is not as rich as that on articles, and their analysis will mainly be complementary to that of articles. 2.2.

Subjects and data

Utterances produced by three monolingual Spanish children, José, María and Miguel, have been auditorily analyzed from the beginning of word production until about age 2;3. The focus of analysis was on the research questions pre-

On the prosodic structure of articles in L1 acquisition of Spanish

53

sented in Section 2.1. The children were recorded in Madrid, as part of a joint project of the Universidad Autónoma de Madrid and the Universität Hamburg.4 Recordings were carried out about once a month by a researcher, who visited the child in his/her home, and interacted and played with him/her. Some words were elicited with the help of books and toys. Data were transcribed by phoneticians and students with training in phonetics. Transcriptions were orthographic for the adults’ utterances, and phonetic (based on the IPA characters and some diacritica) for the child’s utterances. They were ¿rst done with pencil and paper, and then introduced into Fourth Dimension for the Mac, in the form of a lexicon. For the purpose of this paper, because many ¿ller syllables had not been transferred to the computarized data bank, the pencil-and-paper transcriptions were all inspected and utterances containing bare nouns and nouns preceded by an article, a proto-article or a ¿ller syllable were extracted. All contexts requiring an article were counted for each recording session, which allowed us to calculate the percentages of article production. Following Lleó (2006), we consider that a certain prosodic constituent has been acquired, once the corresponding utterance is produced at least twice in a single recording session. Several further calculations as to the relation of articles, proto-articles and ¿ller syllables to the nouns, number of syllables, coda production, etc., were carried out. Only those items satisfying criterion (4) of Section 1.2 were classi¿ed as (proto)articles.

3.

Segmental and prosodic analyses of articles produced by the Spanish children

We will ¿rst, in Section 3.1, quantify some phenomena related to the co-occurrence of articles, depending on a) whether they are monosyllabic or disyllabic, and b) whether they occur with nouns of various numbers of syllables: monosyllables, disyllables, and multisyllables (trisyllables and quadrisyllables). We will also analyze the pace of coda production for articles in comparison to coda 4. The project was supported by the DAAD with grants to Antonio Maldonado and to Conxita Lleó. Each project side had also national support: from the Ministry of Education and Science, on the Spanish side, and from the DFG (German Research Foundation) on the German side. I am very grateful both to the DFG and the DAAD for their support. The data collected within those projects constitute the PAIDUS corpus, comprised of the utterances of 5 monolingual Spanish-speaking children and 5 monolingual German-speaking children. All children were recorded from the beginning of word production until 2;0 or 2;6 in the case of Spanish and until ca. 3;0 in the case of German.

54

Conxita Lleó

production in lexical items, in Section 3.2. These analyses will give us a segmental and prosodic overview of the degree of autonomy of (proto)articles. Finally, in Sections 3.3 to 3.6, we will give a qualitative (rather than quantitative) analysis, showing concrete occurrences of the developing prosodic structures of articles. 3.1.

Occurrence of monosyllabic and disyllabic articles with monosyllabic, disyllabic and multisyllabic nouns.

Tables 2 to 4 show percentages (and raw numbers in parentheses) of monosyllabic and bisyllabic articles preceding nouns according to numbers of syllables – monosyllables, disyllables and multisyllables (tri- and quadrisyllables) – as they were uttered by the three monolingual Spanish children, José, María and Miguel, respectively. In the “monosyllabic” article category, we have considered the masculine de¿nite el and masculine inde¿nite un, produced with or without the coda consonant. “Fillers” refer to ¿ller syllables (see e.g. Lleó 2001b); they have not been counted in the category “articles”. The feminine de¿nite article la has not been included either, because it is often not distinguishable from a ¿ller syllable (see Lleó 2001a). The “disyllabic” article category refers to una and uno. Here, it should be noticed that una is the target-like production for the inde¿nite article in the feminine form, whereas uno is generally a creation on the part of the child, substituting for the target form of the inde¿nite article in the masculine, un.5 The percentages appearing with the articles indicate the percentages of occurrence of the monosyllabic and bisyllabic articles, respectively. The percentages referring to nouns according to number of syllables show the occurrence of each type of article depending on whether the nouns are monosyllables, disyllables or multisyllables. The last column called “Fillers” indicates raw numbers of ¿ller syllables, that is, those segmentally default syllables that ¿ll the position of a target article. Note that these are only relevant at the earliest stages of development, as they are continuously being replaced by proto-articles and/or articles. We thus consider three stages of articles, from a segmental point of view: the initial or ¿ller stage, the proto-article stage (e.g. el and un, produced without the ¿nal coda) and the ¿nal or full article stage, in which articles are produced target-like. These three segmental stages serve to classify children’s utterances segmentally, depending on how close the forms uttered by the children are to the target articles, but they do not constitute self-contained stages, as the three segmental stages show much overlap, and partly even occur simultaneously. 5. In a few cases, uno may correspond to the target plural form unos, produced without the ¿nal /s/.

On the prosodic structure of articles in L1 acquisition of Spanish

55

Table 2. Percentages (and raw numbers) of monosyllabic and disyllabic articles cooccurring with nouns classi¿ed according to number of syllables for the child José José Nr. syllables 1;7–1;10 Monosyllables Nouns Disyllables Multisyllables José 1;11 Monosyllables Nouns Disyllables Multisyllables José 2;0 Monosyllables Nouns Disyllables Multisyllables José 2;2 Monosyllables Nouns Disyllables Multisyllables

Monosyllabic Article 100% (9) 0% 89% (8) 11% (1) Monosyllabic Article 88% (45) 13% (6) 73% (33) 13% (6) Monosyllabic Article 77% (20) 0% 65% (13) 35% (7) Monosyllabic Article 54.5% (72) 3% (2) 54% (39) 43% (31)

Disyllabic Article 0%

Fillers 20

Disyllabic Article 12% (6) 0% 33% (2) 67% (4) Disyllabic Article 23% (6) 0% 50% (3) 50% (3) Disyllabic Article 45.5% (60) 5% (3) 53% (32) 42% (25)

Fillers 26

Fillers 0

Fillers 0

The three children begin by producing many ¿ller syllables instead of articles and proto-articles, but whereas Miguel already produces full-Àedegd articles very soon, the other two children, José and María, begin by producing few target-like articles and many ¿ller syllables. Most nouns produced with monosyllabic articles are disyllables, as these are by far the most frequent words in Spanish children’s lexica. This is shown in Figure 1, which contains the mean percentages of word types according to number of syllables, in the early cumulative lexica of the Spanish children. Clearly, disyllables are by far the most frequent words in the Spanish children’s lexica, constituting more than 70% at age 1;6, and almost 60% at age 2;0. Disyllabic articles become very frequent at different time points depending on each individual child, but they always emerge after monosyllabic articles: in the case of José at 2;2 (with 60 tokens and 45.5% of all articles and protoarticles), María not earlier than 2;2–2;3 (with only 7 tokens and 50%), and Miguel at 1;10 (with 22 tokens and 35%).

56

Conxita Lleó

Table 3. Percentages (and raw numbers) of monosyllabic and disyllabic articles cooccurring with nouns classi¿ed according to number of syllables for the child María María Nr. syllables 1;7–1;10 Monosyllables Disyllables Nouns Multisyllables María 2;0–2;1 Monosyllables Disyllables Nouns Multisyllables María 2;2–2;3 Monosyllables Disyllables Nouns Multisyllables

Monosyllabic Article 100% (8) 0% 87.5% (7) 12.5% (1) Monosyllabic Article 100% (19) 0% 74% (14) 26% (5) Monosyllabic Article 50% (7) 0% 71% (5) 29% (2)

Disyllabic Article 0%

Fillers 65

Disyllabic Article 0%

Fillers 35

Disyllabic Article 50% (7) 0% 12% (1) 86% (6)

Fillers 30

Table 4. Percentages (and raw numbers) of monosyllabic and disyllabic articles cooccurring with nouns classi¿ed according to number of syllables for the child Miguel Miguel 1;7–1;8 Nouns

Nr. syllables Monosyllables Disyllables Multisyllables

Miguel 1;10 Nouns

Monosyllables Disyllables Multisyllables

Miguel 2;0 Nouns

Monosyllables Disyllables Multisyllables

Miguel 2;2 Nouns

Monosyllables Disyllables Multisyllables

Monosyllabic Article 95% (61) 31% (19) 56% (34) 13% (8) Monosyllabic Article 65% (40) 12.5% (5) 62.5% (25) 25% (10) Monosyllabic Article 67% (14) 0% 71% (10) 29% (4) Monosyllabic Article 63% (41) 15% (6) 70% (29) 15% (6)

Disyllabic Article 5% (3) 0% 100% (3) 0% Disyllabic Article 35% (22) 9% (2) 64% (14) 27% (6) Disyllabic Article 33% (7) 0% 86% (6) 14% (1) Disyllabic Article 37% (24) 4% (1) 54% (13) 42% (10)

Fillers 28

Fillers 17

Fillers 0

Fillers 0

On the prosodic structure of articles in L1 acquisition of Spanish

57

Except in the case of Miguel, the Spanish children at ¿rst do not use monosyllables preceded by articles. Although most of the words that are preceded by proto-articles at these early stages are disyllables, it is to be noted that the mean percentages of multisyllables (tri- and quadrisyllables) that co-occur with articles, in the case of José and María, are higher than expected according to Figure 1. That is, although multisyllables reach 22% of all word types at age 2;0, José shows 43% of multisyllables preceded by a monosyllabic article and 42% of multisyllables preceded by a disyllabic article. It is also interesting that María, who is the child with fewer (proto)articles, reaches 86% of multisyllables preceded by bisyllabic articles, at 2;2–2;3. Miguel exhibits percentages that agree with those of Figure 1 at age 1;10, as he produces a relatively low number of multisyllables (27%) preceded by a bisyllabic article. However, at 2;2, he produces a much higher percentage (42%) of multisyllables preceded by a bisyllabic article. A plausible explanation for these unexpected high values of disyllabic articles will be presented in Section 3.3. 100 90 80 70 60 monos y llables

50

dis y llables tris y llables

40

quadris y llables

30 20 10 0 1;6

1;8

1;10

2;0

Figure 1. Mean percentages of word types according to number of syllables in the early cumulative lexica of the Spanish children.

3.2.

Treatment of codas

Spanish has a clear preference for open syllables, which constitute about 72% of all syllables (Delattre 1965: 42). Consonants produced in coda position are preferably sonorants, and fricative /s/, plus a few other obstruents, which appear within words rather than at the right word edge. Articles have some obligatory codas, namely /l/ in the masc. sing. of the de¿nite, /n/ in the masc. sing. of the inde¿nite, and /s/ in all plural forms. Table 5 shows numbers and percentages

58

Conxita Lleó

of codas produced by the three children out of target codas, separately for codas belonging to lexical items (indicated as “Lex codas”) and for codas belonging to the article (indicated as “Art codas” and underscored on a gray background). In general, article codas increase over time, whereas lexical codas show much variability along the time period studied. María does not follow this developmental pattern, because she produces very few codas, hardly showing any differences between the two types of codas. José begins producing some more article codas than lexical ones at 1;10, but the difference is not large. At 2;0, his article codas reach more than 50%, clearly going beyond the lexical ones. Miguel, who produces many more lexical codas than the other two children, at 2;2 begins to produce more article than lexical codas; this trend goes further at 2;3, with 64% of lexical codas and 91% of article codas. As predicted by criterion (3) in Section 1.2, an intensi¿ed production of article codas points to the prosodic autonomy of articles. The results of Table 5 thus provide some evidence for the prosodic autonomy of articles, in the case of José and Miguel. This result will be con¿rmed on the basis of the following analyses. Table 5. Production of codas belonging to the lexical item or to the article José Lex codas Art codas 1;4 1;5 1;6 2/6 33% 1;7 4/31 13% 1;8 1;9 8/69 12% 1;10 12/78 15% 1/3 1;11 4/108 4% 8/39 2;0 17/128 13% 31/60 2;1 2;2 41/115 36% 24/43 2;3 21/74 28% 19/34 2;4

3.3.

María Lex codas Art codas 1/8 12% 0/2 0% 4/25 16% 2/8 0/44 0% 2/5

33% 0/53 21% 52% 1/28 12/43 56% 4/32 56% 16/70 9/66

Miguel Lex codas Art codas

4/16 25% 4/8 40% 17/19 16/23

25% 50% 0/3 0% 89% 2/8 25% 70% 10/17 59%

0%

2/4

50% 49/72 68% 5/21 24%

4% 28% 12% 23% 14%

1/3 4/10 0/4 3/8 2/15

33% 22/28 79% 9/16 56% 40% 0% 56/83 67% 32/45 71% 38% 25/39 64% 10/11 91% 13%

Prosodic stage I: PW stage

At 1;6 José already produces target trisyllables, like gorila ‘gorilla’ and zapato ‘shoe’, and the quadrisyllable, cocodrilo ‘crocodile’. The target number of syllables of trisyllables is maintained, the onset of the initial syllable of zapato being assimilated to the following labial. The quadrisyllable is reduced to a

On the prosodic structure of articles in L1 acquisition of Spanish

59

trisyllable ¿rst, being later (in the same recording session) produced with its four syllables, with assimilation of the coronal consonant to the dorsal (4c). The ¿rst proto-articles make their appearance at 1;7, with two tokens, as shown in (4a). José also produces many forms uttered as putative articles, like those in (4b), produced as non-high vowels preceded by a laryngeal consonant; they constitute ¿ller syllables, without segmental or prosodic autonomy. According to criterion (2) above, whereas the initial syllables of (4c) are considered part of the lexical item, on the basis of the harmony process applied to them, the initial syllables in (4b) are putatively considered proto-articles, in contrast to those of (4c), and because these trisyllabic items have also been produced by José with two syllables, as in (4d). That is, multisyllabic target words can be produced with three syllables, as in (4c) or truncated to bare disyllables, as in (4d), the latter being then eligible for the addition of a proto-article, as in (4b). (4)

José (1;7) un dedo los dedos

[٧єn.‫ۉ‬de.˕o] [lƥ.‫˕ۉ‬e.˕o]

‘a ¿nger’ ‘the ¿ngers’

b.

camello cocodrilo guitarra pajarito trompeta tijeras

[hƥ.‫˕ۉ‬ȳ.˕͝ƥ] [‫˕ۉ‬œ.‫ۉ‬ʪi.‫ˇ׀‬ƥ] [a.‫ۉ‬dĎ.‫ۉ‬dĎ] [hđ.‫ۉ‬ki.‫׀‬ƥ] [hƥ.‫˕ۉ‬e.˕œ] [˕Ď.‫́˕ۉ‬.˕Ď]

‘camel’ ‘crocodile’ ‘guitar’ ‘birdy’ ‘trumpet’ ‘scissors’

c.

tijeras zapato cocodrilo

[ba.‫ۉ‬ƒȪ.Ď] [bđ.‫ۉ‬ƒa.bˇĎ], [bˇa.‫ۉ‬bˇa.‫ۉ‬bˇa] [ʪօ.ʪֽ.‫ۉ‬ʪo], [ʪo.‫ۉ‬ʪi.ʪo] [ʺu.ʺu.‫ۉ‬gy.٧ʺƥ]

‘scissors’ ‘shoe’ ‘crocodile’

d.

amarillo ardilla chupete gacela

[‫ۉ‬di.jƥ] [‫ۉ‬di.hœ], [‫ۉ‬zi.jĎ] [dˇȳ.‫ۉ‬dˇȳ] [‫ۉ‬pȳ.hđ:]

‘yellow’ ‘squirrel’ ‘paci¿er’ ‘gazelle’

a.

At age 1;9 the child seems to be busy mastering trisyllabic lexical items, whose initial unstressed syllable takes the form of a harmonized onset, with the same PA as the following consonant.6 Some examples for this type of templatic formation (see e.g. Velleman and Vihman 2002, or Vihman 1996, on such tem6. This is similar to the cases described in Lleó (1990) for the child Laura.

60

Conxita Lleó

plates) are shown in (5). We thus consider these harmonized initial syllables (5a) as belonging to the lexical item, whereas those in (5b) are considered as proto-articles. However, the two types of initial syllables are hardly distinguishable from one another, except for the initial segment. That is, both syllables are treated similarly, in the sense that their onset is ¿lled by default: by consonantal harmony in the case of the syllable of the lexical item, and by a laryngeal consonant in the case of the proto-article. Moreover, some disyllabic target nouns are expanded to trisyllables, adding an initial syllable with a labial onset (5c). These forms might be interpreted as the result of a trisyllabic requirement on the part of the child, i.e. a preference for trisyllables over disyllables. They may also involve a possible confusion on the part of the child between the unfooted syllable of such trisyllabic nouns and the unstressed syllable of the de¿nite article, as only nouns beginning with /l/ receive the addition of an initial syllable. (5)

José (1;9) abuela Alberto conejo caramelos elefante jirafa mariposa sombrero trompeta zapato

[ba.‫ۉ‬pe.ja] [bĎ.‫ۉ‬bȳ.tֽ] [no.‫ۉ‬ni.no] [ma.‫ۉ‬mȳ.no] [bĎ.‫ۉ‬bȳ.tˇi], [bi.‫ۉ‬pƥ.de] [pĎ.‫ۉ‬bĎ.pa] [pa.‫ۉ‬bƥ.ta] [ba.‫ۉ‬ƒȳ.jo], [ba.‫ۉ‬bȳ.lo] [pa.‫ۉ‬bȳ.ta], [bu.‫ۉ‬bȳ.ta] [pĎ.‫ۉ‬pĎ.tˇo]

‘grandma’ ‘Albert (name)’ ‘rabbit’ ‘bonbons’ ‘elephant’ ‘giraffe’ ‘butterÀy’ ‘hat’ ‘trumpet’ ‘shoe’

b.

mariposa sombrero zapato

[Ď.‫ۉ‬bƥ.ta] [٧ƥ.‫ۉ‬p.Ď.di] [Ď.‫ۉ‬pĎ.tu]

‘butterÀy’ ‘hat’ ‘shoe’

c.

(la) lengua (el) león (el) loro (el) plato

[pa.‫ۉ‬pȳi ۶.da] [bĎ.‫ۉ‬jƥn] [pƥ.‫ۉ‬po.jo] [ba.‫ۉ‬bȳ.ta], [bu.‫ۉ‬bȳ.to]

‘tongue’ ‘lion’ ‘parrot’ ‘plate’

a.

This very limited production of articles, totally bound to the PW of the lexical item is also manifested in a few productions of the verbal proclitic se, whose initial /s/ is incorporated into the process of assimilation of the segments belonging to the PW, or treated as a default onset, either substituted by the coronal stop /t/, or by a laryngeal. See examples in (6).

On the prosodic structure of articles in L1 acquisition of Spanish

(6)

José (1;9): verbal clitics se ha roto [dȪ.dƥ.‫ۉ‬tֽ] se cae [sȪ.‫ۉ‬kai ۶], [٧ȳ.ʪai ۶], [ha.kai ۶] se ha caído [‫׳‬a.‫׳‬a.‫ۉ‬i.ju] se ha ido [‫ۉ‬ta.ijo], [tֺ.hiju] no se toca [‫ۉ‬tete‫ۉ‬tœta]

61

‘it broke’ ‘it is falling down’ ‘it fell down’ ‘it’s gone’ ‘don’t touch’

Clitic se in (6) is thus segmentally very similar to the proto-articles. However, prosodically, se seems to be treated differently than nominal clitics, as articles preferably build a trisyllabic unit with the noun, whereas se, besides occurring in disyllabic and trisyllabic constructions, also appears in quadrisyllabic constructions twice, in se ha caído and no se toca. These two utterances, comprised of two feet each, correspond to the beginnings of the production of Phonological Phrases. At age 1;10 more structures with two feet make their appearance, generally disyllabic nouns preceded by a disyllabic determiner (7b). Notice that such determiners are demonstratives, but not yet articles. At this stage, articles tend to be monosyllabic, or reduced to a monosyllable, as in the case of una nena ‘a girl’ in (7a). Groupings as those in (7b) constitute the ¿rst PPhs in the nominal domain; they are comprised of two PWs, as both words bear stress, but do not yet contain articles. (7) a.

b.

José (1;10) la rana la tele la vaca un papá (u)n mono una nena

[hє.‫ۉ‬na.nah] [a.‫ۉ‬tˇe.Ȫ] [٧æ.‫ۉ‬ƒa.pˇa] [٧օ.ƒa.‫ۉ‬pˇa] [٧m‫ ݙ‬.‫ۉ‬mє.nƥh] [na.‫ۉ‬nȳ.na]

‘the frog’ ‘the TV’ ‘the cow’ ‘a daddy’ ‘a monkey’ ‘a girl’

otro mono otro cuento éste (es) mío

[‫ۉ‬٧ֽj.tˇֽ.‫ۉ‬mƥ.nƥ] [‫ۉ‬o.to.‫ۉ‬pe.o] [‫ۉ‬٧Ȫ.tˇȪ.‫ۉ‬ḿ.ƥh]

‘another monkey’ ‘another’ ‘this is mine’

Some verbal clitics appear at this age (1;10), too, as exempli¿ed in (8). They follow the same type of pattern as the unfooted syllable of nominal multisyllables, with a laryngeal onset or an onset that may result from an assimilation of Manner to the following coronal consonant (as in se ha roto ‘it broke’), thus segmentally showing lack of autonomy. The last two examples in (8) contain two feet each, being thus similar to the nominal PPhs of (7b).

62

(8)

Conxita Lleó

José (1;10): verbal clitics me asusta [٧æ.‫ۉ‬dօ.tˇĎ] [mĎ.‫ۉ‬du.tˇĎ] te asusta [tĎ.‫ۉ‬tօ.ta] se ha roto [tƥ.‫ۉ‬tƥ.tˇƥ], [hȪ.‫ۉ‬to:.tˇƥ] lo ha roto [‫ۉ‬lƥ.ƥ.‫ۉ‬tƥ.tƥ] no me gusta [ƒđ.ma.‫ۉ‬Ҹu.tˇah]

‘it frightens me’ ‘it frightens you’ ‘it broke’ ‘(s)he has broken it’ ‘I don’t like it’

María’s data at age 2;0 show strong similarities to José’s data of the previous months. She produces many target trisyllabic lexical items (comprised of an unstressed syllable followed by a trochee) prosodically target-like. On the basis of criterion (2) above, the initial syllables of cases as those in (9a), with a labial or a coronal consonant in the onset, are considered as belonging to the lexical item. By the same criterion, the initial syllable of (9b) corresponds to a protoarticle, followed by the noun reduced to a disyllable. This is especially clear in those cases in which the truncated noun is also produced as a bare noun, as in the case of biberón ‘(milk) bottle’ or chupete ‘paci¿er’. The trisyllabic prosodic structures of (9c) are trochaic-shaped target nouns preceded by a proto-article. (9)

María (2;0) caballo patata zapatilla

[bƥh.‫ۉ‬kai ۶.do‫]ݡ‬ [ba.‫ۉ‬ta.tĎ] [ba.‫ۉ‬ti.jĎ]

‘horse’ ‘potato’ ‘slipper’

b.

biberón chupete babero zapato

[hȳ.ƒe.ƒi], [‫ۉ‬de.ji] [hæ.‫ۉ‬ti.di] [đ.‫ۉ‬val.ðo‫]ݡ‬ [˕a.‫ۉ‬bĎ.to‫]ݡ‬

‘(milk) bottle’ ‘paci¿er’ ‘bib’ ‘shoe’

c.

la piña el guau guau el puré

[٧a.‫ۉ‬pi:.jĎ] [hȪ.‫ۉ‬ʪĎ:.u۶Ď] [đ.bo.‫ۉ‬je]

‘the pineapple’ ‘the dog’ ‘the puree’

a.

Most of María’s articles at this stage are proto-articles, without the coda consonant, or just ¿ller syllables. In the same manner as in José’s data, the prosodic nature of such proto-articles and ¿ller syllables is similar to that of the initial syllable of trisyllabic lexical items, in the sense that their onset is a default consonant or a null one. This thus offers evidence in favor of these early protoarticles being adjoined to the PW, in the same way as the initial syllables of trisyllabic nouns are. The only disyllabic determiner at this age appears in otro Àan [‫ۉ‬o.do: Ȫ.‫ۉ‬fĎ] ‘another pudding’. Notice that between the demonstrative and the monosyllabic noun, a proto-article is produced; this quadrisyllabic structure

On the prosodic structure of articles in L1 acquisition of Spanish

63

may be anouncing a new stage, in which PPhs are already produced; but being only represented by one single utterance, the PPh cannot yet be considered as mastered. At age 2;2, María keeps the same prosodic structures that we have just seen, but also produces some innovations in the domain of the PW, namely nouns with 4 or 5 syllables in the target language are produced with four syllables (10a). Moreover, target trisyllabic nouns may retain their three syllables and at the same time be preceded by a (proto)article or a ¿ller syllable (10b). Since María already produces PWs comprised of four syllables, the four syllables of the forms in (10b) should plausibly be analyzed as PWs with a special initial syllable, a type of default that corresponds to the proto-article. From 1;10 on, there is some independent evidence in favor of the existence of the PPh node in the verbal domain, as in the utterance in (10c), with a multisyllabic verbal phrase comprised of two trochaic-shaped words, each with a different unitary PA (namely labial and dorsal), preceded by a segmentally autonomous clitic (with palatal PA). It appears that what may be true for the verbal phrase does not have to apply to the nominal phrase. In other words, PPhs are better entrenched in the verbal than in the nominal phrase, as the former can contain many syllables, but the latter do not go beyond four syllables. (10) a. b.

c. 3.4.

María (2;2) caramelito aguacate

[kƥ.me‫ݡ‬.‫ۉ‬li.tƥ] [Ď.փo.‫ۉ‬kđ.tȳ]

‘little candy’ ‘avocado’

la piscina un patito el vestido la comida en el culito

[a.ƒi.‫ۉ‬si.nĎ] [Ď.pĎ.‫ۉ‬ti.tˇo‫]ݡ‬ [hĎ.be‫ݡ‬.‫ۉ‬ti.ðo], [hæ.ve.‫ۉ‬di.do] [lє.ko.‫ۉ‬mi.dĎh] [lo‫ݡ‬.bo.‫ۉ‬li.to‫]ݡ‬

‘the swimming-pool’ ‘a little duck’ ‘the dress’ ‘the food’ ‘in the little tush’

María (1;10) se llama queca [je.‫ۉ‬ma.ma.‫ۉ‬ke‫ݡ‬.kĎ]

‘her name is doll’

Prosodic stage II: rhythmic stage

At age 2;0 José continues to produce disyllabic demonstratives followed by a target disyllabic noun, and he also begins to produce disyllabic articles followed by a target disyllabic noun, as illustrated in (11a), or by a target trisyllabic noun truncated to a disyllable, as shown in (11b). Disyllabic determiners also co-occur with prosodically more complex nouns; in that case, quadrisyllabic nouns are preferred over trisyllabic ones, and, actually, some target trisyllables

64

Conxita Lleó

are converted into quadrisyllables, as in casita ‘little house’ (11c). This shows a clear tendency to build phrases comprised of trochees. The last case in (11c) shows that the missing syllable (to build the ¿rst trochee of the phrase) is not added to the noun, but to the article. (11)

José (2;0) una foca una mano una llave

[u.na.‫ۉ‬pƥ.pa] [‫ۉ‬٧u.na.‫ۉ‬ma.nƥ] [օ.na.‫ۉ‬bƥ.dȳ]

‘a seal’ ‘a hand’ ‘a key’

b.

una espada

[٧օ:.na.‫ۉ‬pa.lĎ], [٧u.ЈĎ.pa.ðĎ]

‘a sword’

c.

una casita este caramelo un caracol

[‫ۈ‬օ.dĎ.‫ۈ‬a.da.‫ۉ‬ki.ʪa] [٧́ç.te.ma.ha.‫ۉ‬me.lo] [٧u:.la.‫ۉ‬qa.qa.‫ۉ‬qo]

‘a little house’ ‘this candy ‘a snake’

a.

This tendency to building pairs of syllables, i.e. sequences of trochees, is clearly manifested in the IP, too, as shown by the examples in (12), with the addition of one syllable in most cases. These multisyllabic constructions must constitute PPhs, as there are no words comprised of more than four syllables in this child’s inventory. (12) José (2;0): IPs Este (está) roto Este está roto (La) guitarra (está) rota Toca la guitarra El caballo trota éste

[ȳ.te.do.tƥ], [٧ȳ.tȳ.ӥo.to] ‘This one (is) broken’ [٧́.t́.ta.ta.‫ۉ‬lo.to] ‘This one is broken’ [‫ۈ‬ʪ́.di.‫ۉ‬tđ.lae۶.‫ۉ‬to.ta] ‘(The) guitar (is) broken’ [‫ۉ‬to.tĎ.‫ۉ‬٧a.ʺi.‫ۉ‬ta.lĎ] ‘(He) is playing guitar’ [٧ȳ.pa.‫ۉ‬ba.͡o.‫ۉ‬to.tĎ.‫ۉ‬ȳ.tȳ] ‘The horse trots, this one’

At 2;2 José further develops the tendency that had begun at 2;0, namely he abides by the number of syllables: if a noun is disyllabic or quadrisyllabic, there is a preference to combine it with a disyllabic article (13a), whereas if it is trisyllabic, the preference is for a monosyllabic article (13b). Many of the examples in (13) are target-like, but some of them have been modi¿ed to abide by the number of syllables, as in the case of globo ‘balloon’, oso ‘bear’, bebé ‘baby’, pequeñito ‘small’, with the addition of one syllable to the determiner, and casita ‘little house’ and ballena ‘whale’ with deletion of one syllable from the determiner. Thus, the child is producing binary groupings of disyllabic feet, many of them obtained by ¿xing the number of syllables of the noun or the article.

On the prosodic structure of articles in L1 acquisition of Spanish

(13) a.

b.

65

José (2;2) un globo un oso un bebé una bicicleta unos calcetines una aspiradora un pequeñito

[٧օ.nƥ.‫ۉ‬vƥ.vƥ] [u.nƥ.‫ۉ‬ӫƥ.bƥ], [u.no.‫ۉ‬ƥ.dƥ:] [‫ۉ‬٧u.nƥ.ƒȳ.‫ۉ‬ƒȳ] [u.na.‫ۈ‬bi.јi.‫ۉ‬klȳ.ta] [‫ۉ‬٧u.nƥ.pa.pȳ.‫ۉ‬tsi.nȳ] [‫ۉ‬٧u.na.‫ۉ‬pi.ӫa.‫ۉ‬do‫ݡ‬.ӥa] [u.nƥ.pƥ.tȳ.‫ۉ‬ni.tƥ]

‘a balloon’ ‘a bear’ ‘a baby’ ‘a bicycle’ ‘some socks’ ‘a vacuum cleaner’ ‘a small one’

un juguete un caballo el caballo los caballos un elefante los elefantes una casita una ballena el vestido

[օn.po.‫ۉ‬vȳ.tȳ] [um.pa.‫ۉ‬ja.to], [‫ۉ‬nօ.pa.ƒal.jo] [‫ۉ‬٧ȳ.pa.‫ۉ‬ƒa.jƥ] [lo.pa.‫ۉ‬ƒa.jo] [u.ven.‫ۉ‬pˇa.јȳ] [nƥ.pˇe.‫ۉ‬pđ.pe] [nĎ.‫׀‬Ȫ.‫ۉ‬di.ta] [na.ma.‫ؒۉ‬ȳ.la] [ȳ.bȳl.‫ۉ‬ti.do]

‘a toy’ ‘a horse’ ‘the horse’ ‘the horses’ ‘an elephant’ ‘the elephants’ ‘a little house’ ‘a whale’ ‘the dress’

At 2;4, María’s lexical items show a clear tendency to align a labial PA to the left of the word, as exempli¿ed in (14a). Articles are not affected by this alignment constraint. As in the previous stage, there are alternations between articles and proto-articles (a syllable without onset or with a laryngeal onset), as shown in (14b) and (14c), respectively. (14)

María (2;4) tomate toma dejamos toma, venga

[po.‫ۉ‬ma.tˇe] [‫ۉ‬bˇo.ma] [me.‫ۉ‬xa:.mƥ] [‫ۉ‬po.ma. ‫ۉ‬meІ.‫׀‬a]

‘tomato’ ‘take it’ ‘we leave (it)’ ‘take it, come on’

b.

la(s) ¿cha(s) la manita la pipa un gato

[la.‫ۉ‬fˇ́.shđ] [la.mã.‫ۉ‬ṇ.tˇĎ] [la.‫ۉ‬pˇ́.pˇa] [un.‫ۉ‬kĎ.tƥ]

‘the tab(s)’ ‘the little hand’ ‘the pipe’ ‘a cat’

c.

el mar [ȳ‫ݡ‬.‫ۉ‬ƒa], [ȳ.‫ۉ‬ma] ‘the sea’ el bolso [æ.‫ۉ‬ƒƥ:.so] ‘the bag’ el chupete (de) María [æ.sƥ.‫ۉ‬pˇe.tˇȳ:.mđ.‫ۉ‬i:Ď] ‘Maria’s paci¿er’

a.

Besides these proto-forms of function words, (15) shows utterances that have become very frequent at this stage: production of the feminine inde¿nite article

66

Conxita Lleó

una (15a) and conversion of the masculine inde¿nite article to uno instead of the target form un (15b). Notice that some of such uno forms might correspond to the plural form unos, produced without the coda consonant; examples of expected plurals are shown in (15c); however, those forms in (15b) plausibly correspond to singular forms. (15)

María (2;4) una pala una papilla una carterita a comprar una camiseta

[օ.nđ.‫ۉ‬pa.l:a] [٧u.na.pab.‫ۉ‬ṕ.ja] [‫ۉ‬٧օ‫ܗ‬.na.ka.tˇȳ.‫ۉ‬ӥ́.tˇa:] [a.kˇõ.‫ۉ‬pa.‫ۉ‬օ.na.ʺa.ḿ.‫ۉ‬ԙȳ‫ڴ‬.tˇĎ]

‘a shovel’ ‘a pap’ ‘a little purse’ ‘to buy a t-shirt’

b.

un sol un tren un barco un reloj un queso un muñeco un pájaro un plato de papilla

[u.nƥ.‫ۉ‬sol] [օ.nօ.‫ۉ‬tȳn], [‫ۉ‬օ.nƥ.‫ۉ‬tˇȳ‫ܗ‬n] [‫ۉ‬օ.no.‫ۉ‬ƒa.ko] [‫ۉ‬٧.օ nƥ.lȳ.‫ۉ‬ӫƥ:l] [օ.nօ.‫ۉ‬kȳ.so] [օ.n‫ڴ‬o.‫ۉ‬ke‫ݡ‬.ko] [‫ۉ‬օ.nօ.pa.‫ۉ‬xa.ӫօ] [u.nƥ.‫ۉ‬pĎ.tœ.pa.‫ۉ‬pi.i ۶a]

‘a sun’ ‘a train’ ‘a boat’ ‘a watch’ ‘a cheese’ ‘a doll (masc.)’ ‘a bird’ ‘a plate of pap’

c.

unos ojos unos patines

[u.nƥ.‫ۉ‬lƥ.xƥ] [‫ۉ‬օ.nƥ.pa.‫ۉ‬tˇ́:.nȪ]

‘some eyes’ ‘some skates’

a.

Notice that the masculine form uno created by the child combines with nouns of any number of syllables: monosyllabic, disyllabic, trisyllabic, as well as a whole NP, as in plato (de) papilla ‘plate of pap’. Such forms, una as well as uno, are segmentally independent from the PW, since they do not enter into the assimilation of PA, into the alignment of labial, or into laryngeal substitution. These forms are segmentally target-like. Thus, from a segmental point of view, they are autonomous, and this autonomy is attributable to the prosody, as well. The form uno co-occurs with PWs like patines ‘skates’ and pájaro ‘bird’ (the latter stressed by the child in the middle syllable), comprised of a trochee and the initial unfooted syllable, as well as with a whole NP like plato (de) papilla. It is thus plausible to consider these articles as being adjoined to the PPh. Articles such as those in (15) are segmentally autonomous. Prosodically, the child seems to assume that function words should constitute a whole PW, in the same manner as lexical items do. The binarity constraint (in terms of syllables) is playing a crucial role, as PWs should comprise a minimum of a foot, and feet must be disyllabic in this child’s system.

On the prosodic structure of articles in L1 acquisition of Spanish

3.5.

67

Prosodic Stage III: PPh stage

José’s examples in (16) show that monosyllabic articles (with or without coda) co-occur with disyllabic, but also trisyllabic and quadrisyllabic nouns, constituting PWs comprised of proto-articles plus trochees, i.e. sequences of two disyllabic feet, or even a PPh with ¿ve syllables, as in the last example of un conejito. (16) José (2;0) un/el cuento un/el gato un/el reloj un coche el perro el caballo un lapicero un conejito

[օ.‫ۉ‬pe.dƥ], [ȳ.‫ۉ‬pȳ.tƥ] [u.ta.to], [ȳ.‫ۉ‬la.to] [օ.le.‫ۉ‬lƥ], [٧ȳ.le.‫ۉ‬lƥ] [u.tƥ.tȳ:] [ȳ.‫ۉ‬pȳ.lƥ] [٧ȳ.pĎ.‫ۉ‬ƒa.́͡ ۶o] [٧օn. pi.‫ۉ‬te.ӥo] [٧օ.ʺօ۶ȳ.te.‫ۉ‬ti.tƥ]

‘a/the story’ ‘a/the cat’ ‘a/the watch’ ‘a car’ ‘the dog’ ‘the horse’ ‘a pencil’ ‘a little rabbit’

Thus, the evidence gathered here makes clear that at 2;0, José already has feet organized into PWs and these organized into PPhs. Monosyllabic articles appear abundantly, most of them uttered with the vowel /u/ or with the vowel /e/, belonging to the inde¿nite and de¿nite article, respectively. These forms are clearly distinguished from those produced in previous stages, limited to default vowels, generally low ones. This means that initial vowels began having the function of articles in a non-autonomous manner, hardly distinguishable from the initial syllable of lexical items, and thus belonging to the PW. Once these vowels are more speci¿ed towards /u/ or /e/, their status is shown to be different from the initial one, not belonging to the PW anymore, but being part of the PPh, as there is independent evidence that the child has the ability to produce PPhs, as shown by the examples. Furthermore, the inde¿nite article (both disyllabic and monosyllabic) has clearly gained prosodic autonomy, and it is often used preceding longer NPs, independently from the number of syllables, as shown in the examples in (17a). Such NPs can also be preceded by the monosyllabic de¿nite article, independently from the number of syllables (17b).

68

Conxita Lleó

(17) a.

b.

3.6

José (2;0) un elefante grande una tarta de manzana un payaso pequeñito un payaso grande una pelota amarilla

[٧u.nƥ.pȳ.‫ۉ‬pˇa.te.‫ۉ‬ma.nȳ] [‫ۉ‬٧u.n:a.‫ۉ‬ta.ta.de.m:a.‫ۉ‬ta.na] [u. pa.‫ۉ‬ja.to. be.ke.‫ۉ‬ni.do] [٧օm.‫ۉ‬pa.ja.to.‫ۉ‬la.nȳ] [٧u.na.pȳ.‫ۉ‬lo.tĎ.a.ma.‫ۉ‬i.ja]

‘a big elephant’ ‘an apple pie’ ‘a little clown’ ‘a big clown’ ‘a yellow ball’

el gato malo la mano de papá

[e.la.tƥ.‫ۉ‬ma.lo‫]ݡ‬ [la.‫ۉ‬ma.no. de.‫ۉ‬pa.pa‫]ڴ‬

‘the mean cat’ ‘papa’s hand’

Overlapping stages in Miguel’s data

Miguel’s articles develop faster than those of the other children: At 1;5 he already produces a few target-like articles, as shown in (18). However, long target words are reduced to disyllables, as in the example of (18c), and he already combines two trochaic feet in one prosodic unit (18b), namely a PPh. (18)

Miguel (1;5) un perro (guau-guau) un reloj un/el lápiz un gallo (kikirikí)

[օ:n.va.vĎ] [u.lȳ.lƥ] [uІ.kĎ.ṕ:], [lĎ.lĎ.ṕ:] [օ:n.‫ۉ‬ki.k:Ȫ]

‘a dog’ ‘a watch’ ‘the/a pencil’ ‘a rooster’

b.

otro nene una vaca

[ƥ.dօ.‫ۉ‬nȳ.nȳ] [81e.kƥkĎ]

‘another boy’ ‘a cow’

c.

mandarina

[‫ۉ‬hi.n:a]

‘mandarin’

a.

At 1;6, Miguel keeps on shortening trisyllabic and quadrisyllabic words to disyllables. In one case, though, he adds one syllable to a trisyllabic word, thus producing the sequence of two syllabic trochees in one single PW: caballo ‘horse’ [ʪahȪ.‫ۉ‬bai ۶.jo]; however, being one single quadrisyllabic token, we cannot consider it yet as being acquired. At 1;8 he already produces many articles and proto-articles: along with default syllables (only comprised of a low vowel), as in (19a), he produces the vowels /e/ and /u/ often with the corresponding coda (19b). He also produces many cases of disyllabic determiners belonging to demonstratives (19c), and a few disyllabic articles (19d). It should be noted that all cases of disyllabic determiners combine with disyllabic words, either disyllabic in the target language, or converted to a disyllable by the child, as in the case of mariposa (19d). Thus, he also seems to abide by the bisyllabic tendency shown by the other two children in Section 3.4, although its manifestation in Miguel’s data is not as compelling as in the case of José.

On the prosodic structure of articles in L1 acquisition of Spanish

(19)

Miguel (1;8) la mariposa el coche el sol la pelota

[hđ.‫ۉ‬bo.za] [Ȫ.‫ۉ‬Ϧœ.u۶ȳ] [Ȫ.‫ۉ‬ðƥl] [˕Ď.‫ۉ‬bƥ.kĎ]

‘the butterÀy’ ‘the car’ ‘the sun’ ‘the ball’

b.

el pez el perro el helado la mariposa una bici

[el.bes] [Ȫl.bȳ.Ӱo] [el.le.‫ۉ‬lĎ.ðo] [lđ.‫ۉ‬bo.zđ] [hȪm.‫ۉ‬ƒi.јe]

‘the ¿sh’ ‘the dog’ ‘the ice-cream’ ‘the butterÀy’ ‘a bike’

c.

este oso otra luna otro pato otro tic-tac

[Ȫð.de.‫ۉ‬bƥ.zo] [٧ƥ.do.‫ۉ‬lu.nĎ] [٧o.dօ.‫ۉ‬ba.do] [‫ۉ‬ԁo.do.diʪ.‫ۉ‬daʪ]

‘this bear’ ‘another moon’ ‘another duck’ ‘another watch’

d.

una luna [˕Ȫ.nĎ.‫ۉ‬lu.nđ] una pala [٧́.nȪ.‫ۉ‬ba.lĎ] una mariposa [˕n‫ݙ‬.na.‫ۉ‬bƥ.ƣa]

a.

69

‘a moon’ ‘a shovel’ ‘a butterÀy’

At age 2;0 Miguel has already mastered articles, especially the inde¿nite one, producing a minimal number of proto-articles. The disyllabic una combines not only with disyllables, as it was the case in the previous recording (19), but also with trisyllabic nouns, as shown by the examples in (20a). Examples in (20b) contain monosyllabic forms of both articles, inde¿nite and de¿nite. (20) a.

b.

Miguel (2;0) una casa una pera una galleta una pelota

[u.nĎ.‫ۉ‬kđ.sĎ] [‫؝ۈ‬.na.‫ۉ‬be.Ď] [u.nĎ.‫׳‬đ.‫͡ۉ‬ȳ.dđ] [u.‫ۈ‬nĎ.be.‫ۉ‬ӫa.dĎ]

‘a house’ ‘a pear’ ‘a cookie’ ‘a ball’

un globo un bizcocho la nariz la tortuga los peces

[un.‫ۉ‬Ϧƥ.lƥ] [uϤ.bi.‫ۉ‬ʪƥ.ԙo] [lđ.nai ۶] [lđ.dֺ.‫ۉ‬du.hđ] [lƥԓ.‫ۉ‬bȳ.ðe]

‘a balloon’ ‘a cake’ ‘the nose’ ‘the turtle’ ‘¿sh (pl.)’

The form una already combines with full PWs, as trisyllabic nouns are already comprised of a trochaic foot plus the initial unfooted syllable, and thus together with the article constitute a PPh. Monosyllabic articles belong to the PPh, too, for the following reason. Miguel does not yet produce quadrisyllabic nouns at

70

Conxita Lleó

this age. His ¿rst target quadrisyllable produced target-like, i.e. with its four syllables, will be produced at 2;2 (see Lleó 2006: 213). This leads to the conclusion that the production of four syllables in (20b), with words like bizcocho ‘cake’ and tortuga ‘turtle’ preceded by the article, is organized into PPhs. Thus, the fact that the child still abides by a constraint against quadrisyllabic PWs in the nominal domain provides evidence for the monosyllabic article belonging to the PPh and not to the PW. At age 2;2 Miguel produces many disyllabic target-like inde¿nite articles in the feminine form; they appear with nouns of various numbers of syllables (21a). The masculine form of the inde¿nite article is often target-like, too, as it keeps its coda consonant in many cases (21b). Only in a few cases is there a reduction of the feminine article or of the plural form (21c); since this only occurs when the article combines with long words, the number of syllables seems to be the (prosodic) cause of their reduction. The de¿nite article is also used target-like (21d), but it appears much less often. (21)

Miguel (2;2) una cama una pelota una cuchara

[˕u.nĎ.‫ۉ‬ʪĎ.mĎ] [‫ۉ‬hu.nĎ.pȳ.‫ۉ‬lƥ.dĎ‫]ݡ‬ [u.‫ۈ‬nĎ.‫׀‬u.‫ۉ‬sai ۶.jȪ]

‘a bed’ ‘a ball’ ‘a spoon’

b.

un barco un gato un payaso un gatito un regalo un teléfono un cumpleaños un globo roto un tractor amarillo un lápiz para pintar

[õ.‫ۉ‬ƒđ.ko] [uІ.‫ۉ‬ʪĎ.do] [um.ba.‫ۉ‬ja.so] [˕զІ. ‫׀‬đ.‫ۉ‬ji.do‫]ݡ‬ [un. de.‫ۉ‬hĎ.lo] [un.de.‫ۉ‬lȳ.fo.no] [uІ. kuϤ.ble.‫ۉ‬Ď.Єo] [un.ʪlo:.‫ۉ‬o.do] [hun.dĎ.‫ۉ‬dœ.˕Ď.mĎ.‫ۉ‬i.jo] [un.‫ۉ‬lđ.bih. pđ.pin.‫ۉ‬tđh]

‘a boat’ ‘a cat’ ‘a clown’ ‘a kitty’ ‘a present’ ‘a phone’ ‘a birthday’ ‘a broken balloon’ ‘a yellow tractor’ ‘a pencil to draw’

c.

una pelota unos zapatos unos calcetines una(s) moneda(s)

[na.me‫ݡ‬.‫ۉ‬lo‫ݡ‬.dĎ] [u.no. sĎ.‫ۉ‬bđ.do] [nօ.ԙa.ԙe.‫ۉ‬di.nȪ] [‫؝‬n. mo.‫ۉ‬nej.dæ‫]ڴ ݡ‬

‘a ball’ ‘some shoes’ ‘some socks’ ‘a/some coin(s)’

d.

el gato I mira la cacerola La muñeca en el coche del pirata

[ȳl.‫׀ۉ‬Ď‫ݡ‬.to] [i.‫ۉ‬mi.Ď. lđ. kĎ‫ݡ‬.se.‫ۉ‬o‫ݡ‬.lĎ] [lđԁ.bu.‫ۉ‬Єȳ.ʪĎ. lȗ.ʪƥ.Ղe] [dȳ.dȳl. bi.‫ۉ‬jđ.dĎ]

‘the cat’ ‘and look at the pot’ ‘the doll in the car’ ‘the pirate’s’

a.

On the prosodic structure of articles in L1 acquisition of Spanish

3.7

71

Summary of the data: Prosodic treatment of function words

Two of the three children, José and María have a very similar start, although at different time points: José at 1;7, María at 2;0. They begin to produce what can be considered proto-articles: the latter are closely related to the initial syllable of trisyllabic nouns stressed in the second syllable (amphibrachic-shaped). These initial unstressed syllables tend to be reduced to a default consonant (a laryngeal) followed by a low vowel, and their occurrence is limited at ¿rst to precede target disyllabic nouns, and target trisyllables reduced to disyllables by the child. Such proto-articles are best interpreted as internal clitics, corresponding to the structure (2c), as they do not exhibit prosodic autonomy, behaving as default syllables. This structure had been dubbed inappropriate for the target language (in Section 1.3) because of the lexical item not being exhaustively dominated by a PW-node. The third child, Miguel, begins to produce articles and some proto-articles earlier than the other two children, at 1;5. His nouns are ¿rst produced with a maximum of two syllables, showing relatively high numbers of unfooted syllable truncation (see Lleó 2002: 296). Based on this child’s tendency to produce disyllabic nouns, we proposed that the additional syllable of the article is not adjoined to the PW. This leads us to reject (2c) for this child, as he does not have such complex PWs in his system. Whereas José and María include proto-articles within the PW, Miguel prefers to position them within the PPh. That is, a sequence of three syllables, comprised of a proto-article followed by a disyllabic noun, is treated as a constituent of the PW by the former children and as a constituent of the PPh by Miguel. Because these early articles are not stressed, (2d) must be rejected, too. This leaves only the free clitic interpretation (3a) for Miguel, which coincides with the prosodic representation that we have attributed to the de¿nite article forms of the target language. After the initial stage (1;7–1;9), in which articles are dominated by the PWnode, which they share with the lexical item, José goes on to produce articles and proto-articles with trisyllables and quadrisyllables, at 2;0. At this stage, José produces a few words with an upper bound of four syllables. This means that the additional syllable of the article can no longer be considered as part of the PW, but is attached to the PPh. However, the most striking aspect at this stage (until 2;2) is José’s preference to produce sequences of trochees: the numbers of syllables of nouns and/or articles are modi¿ed to satisfy this binarity constraint. José often substitutes uno for target un. Clearly, the inde¿nite article, under the fem. form una and especially the masc. form uno, is an independent PW. María and Miguel also go through a stage with a preference for sequences of trochees, but they do not act upon the number of syllables the way that José did. They do have one result in common, though: María (at age 2;4)

72

Conxita Lleó

and Miguel (at age 2;2) also tend to produce the form uno instead of the target masc. un. In the case of Miguel, there does not seem to be any reason to assume a stage of internal clitic (2c), given that at the ¿rst stages, he prefers to produce truncated multisyllabic nouns that are then preceded by an almost targetlike article. The best representation for this (at 1;6) is (3a), because PWs have an upper bound of two syllables, in this child’s utterances, whereas PPhs can already be quadrisyllabic. At 2;0 articles clearly belong to the PPh, because some of them are stressed, and the PWs have not yet reached four syllables. Thus, Miguel is the ¿rst child to reach the PPh for the representation of articles. Clearly, articles are not integrated into the PW, they are very soon Àeshed-out segmentally, and violate the NOCODA constraint more frequently than lexical words. The other two children go through a PW stage for articles. All children go through a transitional stage, in which syllables seem to primarily be organized in trochees, but this is more compelling for José, who ¿xes the number of syllables both in the noun or the article, in order to reach an even number. For all children, the inde¿nite article shows a clear tendency to be produced with two syllables, which for the masc. sg. involves the addition of one syllable. The inde¿nite article thus receives the status of a PW. For these children, more proto-articles than full articles are produced. Thus, in the cases here analyzed, prosodic autonomy goes hand in hand with segmental autonomy, as proto-articles in the PW have laryngeal onsets or no onsets, and they lack codas, as well. We can summarize the child data on articles in the following Table 6, which indicates the prosodic structures of articles for the three children at the various stages, representing the prosodic level to which articles are adjoined. Table 6. Stages in the development of articles José Art Indef Stage I

Art Def

1;7 (2c) Proto-Art] PW

Stage II 2;0 (2d) Art=PW

2;0 (2c) Art] PW

Stage III 2;2 (3a/b) Art] PPh

María Art Indef

Art Def

Miguel Art Indef

Art Def

2;0 (2c) Proto-Art] PW

1;5 (3a/b) Art] PPh

1;8 (3a/b) Art] PPh

2;4 (2d) Art=PW

2;0 (2d) Art=PW

2;0 (3a/b) Art] PPh

2;4 (2c) Art] PW

2;6 (3a/b) Art] PPh

2;2 (3a/b) Art] PPh

On the prosodic structure of articles in L1 acquisition of Spanish

4.

73

Concluding discussion

We have presented articles in Spanish, and have tried to establish their prosodic representation, based on the literature and on some relevant criteria. From the various (four) representations presented in (2), we have arrived at the conclusion that the most adequate structure to account for them was (2a), which we have divided into (3a) and (3b). These prosodic structures are able to account for the unstressed as well as for the stressed articles. With these structures in mind and some criteria to analyze the clitica produced by the children, utterances of three monolingual Spanish children were analyzed and their prosodic structure was tentatively established. This has led us to three different stages, with two children sharing them and another child showing a faster development. Stage I, in which articles are purported to be a constituent of the PW, is shared by two children. The other two stages are shared by the three children: Stage II contends that children abide by a binary number of syllables, with a tendency to produce trochees and to substitute uno for un; it is strongly documented especially in the data of one of the children. Stage III is close to the target system, in which articles are constituents of the PPh; it is reached by all three children, although at different time points. With regard to the research questions formulated in Section 2.1 on the evolution of the prosodic status of articles, we have shown that they develop from segmentally unspeci¿ed and prosodically dependent categories to more targetlike categories, acquiring segmental speci¿cation and prosodic autonomy. In two cases, we have seen that, at ¿rst, monosyllabic articles were treated as belonging to the PW, in a comparable way to the ¿rst syllable of trisyllabic nouns. That is, both syllables were produced as default syllables, in the case of the lexical items by means of spreading, and in the case of the articles by laryngeal onsets. Disyllabic articles give way to a stage in which children prefer to abide by binary feet, but this is stronger in the case of one of the three children. Finally, all children go to the target-like stage, in which articles are immediately dominated by the PPh node. One of the children skips stage I and simultaneously begins at stages II + III, treating the inde¿nite article as a PW, and adjoining all articles to the PPh. As for the more speci¿c questions from Section 2.1, we have seen evidence that for two children articles were constituents of the PW, and for the third child, articles directly became constituents of the PPh. We have further seen that lexical items constitute PWs, whereas articles are only either part of the PW at the earliest stages, or of the PPh later on. Interestingly, though, there is a stage at which all children attribute PW status to the inde¿nite article, and do not only often use the feminine form una, but also convert the masc. form un to

74

Conxita Lleó

uno, to abide by binary prosodic constituents. The comparison between articles and other functional categories has been limited to a few verbal clitics. There, in the case of both children with a PW stage for articles, verbal clitics seem to belong to the PPh earlier than articles. That is, for some children, articles are closely intertwined with nouns prosodically in the NP, whereas the VP may develop the PPh status sooner. Finally, we have seen that whereas two children go through three different stages, the third child can be considered to jump into the second and third stage, simultaneously. This study has con¿rmed what previous studies have also shown, namely, that children acquiring a Romance language produce certain functional items, especially articles, at an early age (Lleó 2001d, Lleó and Demuth 1999), as one of the subjects already produces some articles at age 1;5. However, articles do not develop very fast: their production does not go from 0% to ca. 100%, but rather during about one year the percentage of article production develops slowly but steadily. Articles grow quantitatively and qualitatively: not only do percentages grow, but also segments, as, for example, codas make the articles progressively target-like. As has been pointed out before (e.g. Demuth and McCullough 2009), linguistic development based on syntax and on syntactic parameters predicts a fast increase of functional word production, once a certain parameter is set. Thus, other explanations have been looked for, and the prosodic licencing hypothesis seems to be the most adequate one (Demuth et al. 2012): along the prosodic development, more constituents are progressively available, and the child can get hold of new levels and new constituents, which allow him to anchor new material. This gives place to the development that we have seen in Sections 3.3 to 3.5. Children acquiring Spanish produce articles very soon, in spite of the fact that the prosodic structure of such items in the target language is not straightforward. Most of the phonological processes of the Spanish language take place at higher levels of the Prosodic Hierarchy, and thus do not provide much evidence to argue for or against certain constituency at lower levels of the Prosodic Hierarchy. Perhaps this very dif¿culty is precisely what allows children acquiring Spanish to produce function words so soon, because their prosodic structure does not matter as much as it matters in languages, like the Germanic ones, in order to produce the right forms. There is a growing number of studies dealing with prosodic structures and how they determine certain pronunciations (see e.g. Peperkamp 1996, Itô and Mester 2009). The Spanish child does not encounter many dif¿culties with the target-like production of his/her function words target-like, as the only relevant phenomena are stress and the phoneme /r/, which is the last phoneme to be acquired. Thus, if the child picks let us say

On the prosodic structure of articles in L1 acquisition of Spanish

75

“the wrong” prosodic structure, this does not matter much, because it will not have destructive consequences for his/her pronunciation of the target language. There are certainly other phenomena, which favor production of function words, namely, the presence of long words in the child’s lexicon from the very beginning. In the early Spanish lexicon (see Figure 1) there are hardly any monosyllabic words; most words are disyllabic with a relatively large presence of trisyllables and some quadrisyllables. Having to organize such words, i.e. represent them in the mental lexicon, stimulates producing unstressed syllables, and extrametrical syllables, like the unfooted syllables both of lexical and functional items. In this study, we have assumed a bottom-up model of prosodic structure acquisition, in the sense that syllables are acquired before feet, these are acquired before PWs, which are acquired before PPhs, and so on. However, as already noted in Lleó (2006) this does not mean that the PW must be totally acquired before the PPh in its turn is acquired. For instance, a child who cannot produce cocodrilo or hipopótamo, and reduces these words to trisyllables, may be able to say el coche (de) papá ‘daddy’s car’ without dif¿culty. That is, the child is able to produce ¿ve syllables, when they are parsed as a PPh, but not when they belong to a PW. The reason for this mismatch is probably working memory, as PPhs do not need to be stored in the lexicon the way PWs do. For reasons not entirely “prosodic” a higher constituent may get to be more complex sooner than a lower constituent, which does not pose any dif¿culty to a bottom-up model. Moreover, it is important to note that the model assumed here is hierarchical, as only a hierarchical model of prosodic structure is able to account for the structural difference between na paposa and una posa for target una mariposa ‘a butterÀy’, produced by the same child at two different times. That is, both na paposa and una posa are comprised of four syllables, the former organized as a m plus a trisyllabic PW, and the latter as a Ft plus a disyllabic PW. References Delattre, Pierre 1965 Comparing the Phonetic Features of English, German, Spanish and French. Heidelberg: Julius Groos. Demuth, Katherine 2001 Prosodic constraints on morphological development. In Approaches to Bootstrapping: Phonological, Syntactic and Neurophysiological Aspects of Early Language Acquisition, Jürgen Weissenborn, and Barbara Höhle

76

Conxita Lleó

(eds.), 3–21. (Language Acquisition and Language Disorders Series 24). Amsterdam: John Benjamins. Demuth, Katherine and Ann Tremblay 2008 Prosodically-conditioned variability in children’s production of French determiners. Journal of Child Language 35: 99–127. Demuth, Katherine and Elizabeth McCullough 2009 The prosodic (re)organization of children’s early English articles. Journal of Child Language 36: 173–200. Demuth, Katherine, Meghan Patrolia, Jae Yung Song and Matthew Masapollo 2012 The development of articles in children’s early Spanish: Prosodic interactions between lexical and grammatical form. Linguistic interfaces and language acquisition in childhood, Jason Rothman and Pedro GuijarroFuentes (eds.), First Language 32 (1–2): 17–37. Goad, Heather and Meaghen Buckley 2006 Prosodic structure in child French: evidence for the foot. Catalan Journal of Linguistics 5: 109–142. Itô, Junko and Armin Mester 2009 The extended prosodic word. In Phonological Domains: Universals and Deviations, Janet Grijzenhout, and Baris Kabak (eds.), 135–194. Berlin, New York: Mouton de Gruyter. Lleó, Conxita 1990 Homonymy and reduplication revisited: On the extended availability of two strategies in phonological acquisition. Journal of Child Language 17: 267–278. Lleó, Conxita 2001a Determining the Acquisition of Determiners: On the Innateness of Functional Categories. In Features and Interfaces in Romance. Essays in Honor of Heles Contreras, Julia Herschensohn, Enrique Mallén, and Karen Zagona (eds.), 189–202. Amsterdam/ Philadelphia: John Benjamins. Lleó, Conxita 2001b Early ¿llers: undoubtedly more than phonological stuf¿ng. Journal of Child Language 28: 262–265. Lleó, Conxita 2001c The Transition From Prenominal Fillers to Articles in Spanish and German. In Research on Child Language Acquisition: Proceedings of the 8th Conference of the International Association for the Study of Child Language (Donostia, Spain, Juli 1999), Margareta Almgren, Andoni Barreña, María-José Ezeizabarrena, Itziar Idiazabal, and Brian MacWhinney (eds.), 713–737. Somerville, Mass.: Cascadilla Press. Lleó, Conxita 2001d The interface of phonology and syntax: The emergence of the article in the early acquisition of Spanish and German. In Approaches to Bootstrapping: Phonological, lexical, syntactic and neurophysiological as-

On the prosodic structure of articles in L1 acquisition of Spanish

77

pects of early language acquisition, Vol. 2, Jürgen Weissenborn, and Barbara Höhle (eds.), 23–44. Amsterdam/Philadelphia: John Benjamins. Lleó, Conxita 2002 The role of Markedness in the Acquisition of Complex Prosodic Structures by German-Spanish Bilinguals. International Journal of Bilingualism 6 (3): 291–313. Lleó, Conxita 2003a Child Prosody and Filler Syllables: Looking into Spanish through the Optimal Window of Acquisition. In Linguistic Theory and Language Development in Hispanic Languages. Papers from the 5th Hispanic Linguistics Symposium and the 4th Conference on the Acquisition of Spanish and Portuguese, Silvina Montrul, and Francisco Ordóñez (eds.), 229– 253. Somerville, Mass.: Cascadilla Press. Lleó, Conxita 2003b Prosodic licensing of coda in the acquisition of Spanish. Probus 15 (2): 257–281. Lleó, Conxita 2006 The Acquisition of Prosodic Word Structures in Spanish by Monolingual and Spanish-German Bilingual Children. Language and Speech 49 (2): 207–231. Lleó, Conxita and Katherine Demuth 1999 Prosodic Constraints on the Emergence of Grammatical Morphemes: Crosslinguistic Evidence from Germanic and Romance Languages. In Proceedings of the 23rd Annual Boston University Conference on Language Development, Vol. 2, Annabel Greenhill, Heather Little¿eld, and Cheryl Tano (eds.), 407–418. Somerville, Mass.: Cascadilla Press. Lleó, Conxita and Irene Vogel 2004 Learning new segments and reducing domains in German L2 Phonology: The role of the Prosodic Hierarchy. International Journal of Bilingualism 8: 79–104. Navarro Tomás, Tomás 1963 Manual de pronunciación española. 11th ed. Madrid: Centro Superior de Investigaciones Cientí¿cas. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris Publications. Peperkamp, Sharon 1996 On the Prosodic representation of Clitics. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 102–127. Berlin: Akademie Verlag. Radford, Andrew 1990 Syntactic theory and the acquisition of English syntax. Oxford: Basil Blackwell.

78

Conxita Lleó

Selkirk, Elizabeth 1984 Phonology an Syntax: The relation between sound and structure. Cambridge, Mass.: MIT Press. Selkirk, Elizabeth 1996 The prosodic structure of function words. In Signal to Syntax: Bootstrapping from Speech to Grammar in Early Acquisition, James L. Morgan, and Katherine Demuth (eds.), 187–213. Mahwah, N.J.: Lawrence Erlbaum Associates. Velleman, Shelley L. and Marilyn M. Vihman 2002 Whole-word phonology and templates. Language, Speech and Hearing Services in Schools 33: 9–23. Vihman, Marilyn M. 1996 Phonological Development. The Origins of Language in the Child. Cambrige, Mass. and Oxford, U.K.: Blackwell.

Third person clitic production and omission in Romance SLI Anna Gavarró

1.

Introduction

Speci¿c Language Impairment is a pathology that affects language acquisition in the absence of any other impairment and in recent years has received much attention as a valuable source of information on the (innate) language faculty. In the words of Leonard (2004: 402) “Speci¿c Language Impairment (SLI) is a term that is applied to children who show a signi¿cant de¿cit in their spoken language ability with no obvious accompanying problems such as hearing impairment, mental retardation or neurological damage. This type of language disorder is regarded as developmental in nature because affected children exhibit language learning problems from the outset.” Leonard estimates that SLI affects around 7% of 5-year-olds according to epidemiological data and attributes it to a genetic source, given that children with SLI are more likely than typically developing children to have parents or siblings with SLI, and that concordance rates for SLI are also higher for monozygotic twins than they are for dizygotic twins. In this paper I corroborate the ¿nding in the literature that clitics do not constitute a natural class in acquisition, contrary to the claims of Leonard’s (1998) Surface Hypothesis, according to which phonologically de¿cient elements such as clitics are all vulnerable in SLI. I undertake a comparison of SLI in the Romance languages, and argue that for some of them, but only some, the omission or deviant production of clitic pronouns may be used as an SLI marker. Furthermore, I provide an account of the phenomenon along the lines of the Extended-UCC hypothesis (Wexler 2003), according to which early child grammar presents some computational limitations (in feature checking) that extend in time in SLI children.1 1. Earlier versions of this paper were presented at the 37th Linguistic Symposium on Romance Languages in Pittsburgh, the Colloquium of Generative Grammar in Girona and the COST IS0804 2010 Meeting in Athens. The author wishes to acknowledge two anonymous reviewers and the editors for their comments on a previous

80

Anna Gavarró

In the ¿rst section I provide the theoretical background to this study. In section 2 new results on clitic production and omission in Catalan SLI, both longitudinal and experimental, are presented; I compare the results to those on Spanish found in the literature, and propose an analysis of the ¿ndings along the lines of Wexler (2003, to appear). In section 3 the analysis proposed is shown to extend to languages other than Catalan and Spanish, and section 4 presents the conclusions.

2.

Background

Rice (2003) and Rice and Wexler (1996) argued that the optional use of tense markers (third person –s, regular past tense –ed) in SLI English-speaking children is the most outstanding feature of this linguistic disruption in English. The existence of a prolonged optional in¿nitive stage has been substantiated for a number of other languages, namely Dutch (de Jong 1999), German (Clahsen 1989, Clahsen, Bartke, and Göllner 1997) and Swedish (Hansson and Nettelbladt 1995). The optional in¿nitive stage is found to have all the properties of the optional in¿nitive stage in typically developing (TD hereafter) children except for its timing: children with SLI tend to stay in the optional in¿nitive stage for an extended period (see for example the results of Wexler, Schaeffer, and Bol 2004 on child Dutch). The claim in Wexler (1998) is that the optional in¿nitive stage results from the interaction of two principles, the Unique Checking Constraint (1) – a principle subject to maturation – and Minimise Violations (2). The Unique Checking Constraint states that children go through a period of grammatical development in which the derivations they license involve only one instance of featurechecking by a given DP; young children, therefore, have immature grammars which constrain their computational capacities. Minimise Violations, on the other hand, holds for all grammars, adult and child, and guarantees that of two alternative derivations the one that violates fewer grammatical principles (optimally none) will be the one selected by the speaker. (1)

Unique Checking Constraint (on children in Optional In¿nitive stage) The D-feature of DP can only check against one functional category. version of the paper, the ¿nancial support provided by project HUM2006–13295– C02–01, and the help of Anna Tàpias, who acted as experimenter in the experiment reported in section 2.1. I am also greatly indebted to the subjects who took part in the experiments reported.

Third person clitic production and omission in Romance SLI

(2)

81

Minimise Violations Given an LF, choose a numeration the derivation of which violates as few grammatical properties as possible. If two numerations are both minimal violators, either one may be chosen.

Following Wexler (to appear) SLI children only differ from TD children in a slower maturation out of the stage at which the UCC is operative. It is well known that the optional in¿nitive stage is of negligible incidence in null subject languages such as Italian and Spanish (Rizzi 1993/94, Torrens 1995 a.o.), a fact that can be accounted for given the properties of the verbal inÀection in those languages (in standard analyses of Italian, Spanish and all null subject languages, agreement is an interpretable feature and does not need to be checked out, therefore no double checking is required).2 Given this general perspective, what are our expectations for SLI in Romance? We do not expect tense marking to be substantially impaired in the Romance null subject languages even in SLI, given the absence of widespread root in¿nitives in TD children. This is in fact what has been found for one of the Romance languages investigated, Italian (see Leonard, Sabbadini, Leonard, and Volterra 1987, Leonard, Bortolini, Caselli, McGregor, and Sabbadini 1992, Bottari, Cipriani, Chilosi, and Pfanner 1998, Leonard and Bortolini 1998). Bottari, Cipriani, and Chilosi (1996) report no root in¿nitives at all in 20 out of 27 Italian SLI children, and a very low percentage in the remaining seven. For a Romance language which is non null-subject, such as French, the prediction of the UCC is that problems with the expression of ¿niteness should be found in SLI children – a prediction that is borne out, as shown in Hamann, CronelOhayon, Dubé, Frauenfelder, Rizzi, Starke, and Zesiger (2003). In the search for other markers of SLI in Romance, Jakubowicz, Nash, Rigaut, and Gérard (1998) found a high incidence of object clitic omission in French SLI. This parallels object clitic omission in TD children, but lasts for a much longer period. In contexts requiring a clitic (because the referent of a DP is old, known information, with a close linguistic antecedent), children pro2. Some authors argue that a form analogous to the root in¿nitive is indeed to be found in the Romance null subject languages: a form identi¿ed variously as the in¿nitive (Buesa 2007), the imperative (Salustri and Hyams 2003) or a bare form homophonous with the third person singular (Pratt and Grinstead 2007 and authors cited therein).The lack of unanimity in identifying such root analogue is in itself indicative that the phenomenon is at best much weaker than what is found in the non-null subject languages. Most of the work arguing for an optional in¿nite stage in the null subject languages is based on the analysis of spontaneous production; those having studied it experimentally have found its incidence to be very low (5% in Davidiak and Grinstead 2004’s study of child Spanish).

82

Anna Gavarró

duce either a full DP (3) or omit the clitic (4), although they may also produce a target-like sentence with a clitic (5). (Examples (3) and (4) are taken from Jakubowicz and Nash, to appear.) (3)

Il met son tee-shirt. he puts-on his tee-shirt (target: Il le met ‘He puts it on’)

(Jér, SLI3b, 7;3)

(4)

Elle grate. she scratches (target: Elle la gratte ‘She scratches it’)

(5)

Il le cache. he it hides ‘He hides it.’

(Ben, SLI3b, 5;6)

The percentage of each of the options in (3–5) for the populations of TD and SLI children in French is presented in Table 1, reworked from Jakubowicz and Nash (to appear). The group of children in Table 1 correspond to different age groups (TD three- and four-year-olds; SLI children in group SLI1 are the oldest, mean age: 9;8, and least impaired; group SLI3 is the youngest, mean age: 6;10, and most impaired; SLI2 is in between the two for age (mean age: 8;2) and severity of impairment). Table 1. Percentage of response types produced in accusative clitic contexts, French Subject group

Clitic

Clitic omission

Full DP

TD 3-y-o TD 4-y-o SLI3a SLI3b SLI2 SLI1

42.6% 79.3% 8% 13.4% 22% 43.4%

15.7% 4.3% 66.4% 20.6% 14.8% 4.7%

19.7% 9.7% 8% 29.1% 37.6% 27.3%

Other DP ReÀexive clitic 17% 5% 16.8% 15.4% 11.6% 8.7%

5% 1.7% 0.8% 21.7% 14% 2.7%

As we can see, apart from the response types exempli¿ed above, children produced other transitive structures or an inappropriate reÀexive clitic. Note that French has a property shared by some, but not all, Romance languages: when a third person clitic occurs, if the verb is in a perfect tense the participle agrees with the object:

Third person clitic production and omission in Romance SLI

(6)

a.

Jean a ouvert la fenêtre. Jean has opened the-fem window(fem) ‘Jean has opened the window.’

b.

Jean l’a ouverte. Jean CL has opened-fem ‘Jean has opened it.’

83

The analysis proposed by Wexler (to appear), Wexler, Gavarró, and Torrens (2004) and Gavarró, Torrens, and Wexler (2010) capitalises on this property of French cliticisation, and treats object clitic omission in child French as the result of the operation of the UCC in conjunction with Minimise Violations. Here I brieÀy sketch the analysis proposed. Let us assume along the lines of Sportiche (1996) that a sentence with an object clitic presents a null DP object in canonical position that must be raised to Spec, ClP, to check an uninterpretable feature of Cl, so that the derivation may converge at the interface. In addition, in a participle-agreement language another uninterpretable feature must be deleted in vP – the feature which gives rise to participle agreement. The con¿guration is represented in (7): (7)

CIP Spee XP' Cl

Cl' TP T' T

vP v' v

VP V' DP pro'

84

Anna Gavarró

Therefore, while two uninterpretable features must be deleted in a participleagreement language (e.g. French), only one feature must be deleted in a nonparticiple-agreement language. The UCC only visibly affects the ¿rst set of languages, not the second; this prediction of the hypothesis is borne out for Catalan and Spanish, as shown in Wexler, Gavarró, and Torrens 2004; compare the rate of object clitic omission for TD children in these two languages3: Table 2. Production of third person clitics, Catalan and Spanish TD children Catalan, present tense 1–2 year olds 3 year olds 4–5 year olds Spanish, present tense 2 year-olds 3 year-olds 4 year-olds

Clitic

Clitic omission

Full DP

22.6% 68.2% 95.7%

74.2% 25% 4.2%

3.2% 6.8% 0

100% 97.5% 100%

0 2.5% 0

0 0 0

In this approach, object clitic omission is predicted for the Romance languages with participle agreement (such as French and Catalan, already illustrated, but also Italian), but not for Romance languages without participle agreement (e.g. Spanish and Romanian). There are nevertheless some differences between Catalan, French and Italian with respect to participle agreement: in many Catalan varieties (those tested by Wexler, Gavarró and Torrens 2004, Gavarró et al. 2010) participle agreement is optional, as shown in (8). (8)

El pare ha obert les ¿nestres. Les ha obert/obertes. the father has opened the windows CL-fem-pl has opened/opened-fem-pl

In some varieties of French, participle agreement is apparently also becoming optional; in all varieties agreement is only audible with some verbs (e.g. ouvrir ‘open’, where ouvert ‘open’ and ouverte ‘open-fem’ are pronounced different3. Some authors have challenged this view, and contended that child Spanish presents clitic omission (see, for example, Fujino and Sano 2002, Castilla, Pérez-Leroux, and Eriks-Brophy 2008 and Castilla and Pérez-Leroux 2010) . We have argued that this conclusion stems from a combination of factors: the experimental method used to elicit clitics, the recount procedure and, most importantly, the fact that the varieties of Spanish investigated allow null objects (for a detailed review, see Gavarró et al. 2010).

Third person clitic production and omission in Romance SLI

85

ly) but not with many others (e.g. voir ‘see’, where vu ‘seen’ and vue ‘seen-fem’ are homophonous). In Italian, participle agreement is both obligatory with third person clitics and always phonologically overt, so that it is more robust than in French and most varieties of Catalan. However, these differences seem to be immaterial in acquisition, and we assume that the underlying mechanisms of participle agreement and cliticisation remain the same for the three languages under scrutiny in spite of super¿cial differences. In Gavarró et al. (2010) we show in detail how the predictions of the UCC are borne out, even beyond Romance, for the acquisition of pronominal clitics and participle agreement in typically developing children. Turning now to the broad issue of clitics, in Catalan, Spanish, and many other Romance languages, third person pronominal clitics are homophonous with de¿nite articles. For TD children, Guasti, de Lange, Gavarró, and Caprin (2004) and Guasti, Gavarró, de Lange, and Caprin (2008) have shown that articles develop very early in the Romance languages (as opposed to Germanic languages). Compare the rates of determiner omission for Catalan, Italian and Dutch at three developmental stages: Table 3. Mean percent omission of articles in different periods of linguistic development in utterances with verbs (from Guasti et al. 2004), TD children

Catalan Italian Dutch

Stage 1: 1–100 words 45% 52% 88%

Stage 2: 101–200 words 6% 17% 54%

Stage 3: >200 words 1% 23%

Determiner omission is very high in the ¿rst stage for Catalan and Italian, but signi¿cantly higher in Dutch; the contrast between Catalan and Italian on the one hand and Dutch on the other remains in stages 2 and 3. The different patterns of development of articles are attributed by Chierchia, Guasti, and Gualmini (1999) and Guasti et al. (2008) to the setting of the parameter which determines the realisation of arguments as NPs or DPs in a language (Chierchia’s so called Nominal Mapping Parameter). To the extent that Catalan and Spanish belong to the same language class, that of the Romance setting of the Nominal Mapping Parameter, we would not expect any contrast between Catalan and Spanish regarding determiner production, in either SLI or TD children. Moreover, were SLI only to affect the maturation out of the UCC-stage, we would not expect SLI children to develop Ds any later than TD children. In contrast, Leonard’s (1998) Surface Hypothesis states that SLI children have dif¿culty with elements of scarce phonetic saliency, such as clitics; if this hypothesis was correct,

86

Anna Gavarró

the prediction would be that homophonous determiners and pronominal clitics would show the same behaviour in children with SLI. Given this background, in the next section I address the following questions: (i) Are clitics, due to their phonological properties, singled out in SLI as a class? And (ii) Is the de¿cit attested in third person pronominal clitics in SLI universal, or is it subject to cross-linguistic variation, as it is in TD children?

3.

Catalan and Spanish, a point of comparison

As in Wexler, Gavarró, and Torrens (2004) and Gavarró et al. (2010) I take Catalan and Spanish as a term of comparison, given a crucial difference between the grammars of these two languages, namely the presence/absence of participle agreement. 3.1.

Some new results on Catalan

Here I present some new data from Catalan that correspond to the spontaneous productions of two Catalan-speaking children with SLI available in the CHILDES database (MacWhinney and Snow 1985). These children were selected after an initial screening of 2000 three-year-olds in state schools (through the school support units CREDA); the children initially selected for language speci¿c problems were administered tests for audition, intelligence (WPPSI), and a protocol for the detection of SLI. Details of the recordings of the two children appear in Table 4.4 Table 4. Data source for Catalan spontaneous production J, transcript 1 J, transcript 2 A, transcript 1 A, transcript 2

Age in months 45 57 43 58

MLU 1.5 4.7 1.3 3.1

4. I am grateful to Mònica Sanz, of the Departament de Psicologia Bàsica, Universitat de Barcelona, for providing information on the selection procedure of the children in the CHILDES database, recorded by the team of Miquel Serra at the Universitat de Barcelona. The database does not include any Catalan-speaking child with SLI other than the ones in this paper; the remaining SLI children are all native speakers of Spanish.

Third person clitic production and omission in Romance SLI

87

The transcripts of the Catalan-speaking children were analysed for the presence/absence of (i) root in¿nitives, (ii) determiners and (iii) object clitics. As is standard, immediate repetitions of adults were discarded, and repetitions of the same sentences by the child himself were computed only once. In the analysis of the verbal productions of our two SLI Catalan-speaking children, only seven (1.25%) cases of root in¿nitive were found, out of a total of 556 verbal productions. Given that Catalan is a null subject language, this falls out from the predictions of the UCC. The results for determiner and object clitic production appear in Table 5: Table 5. Results for determiner and 3rd person object clitic production, Catalan, SLI

J, tr 1 J, tr 2 A, tr 1 A, tr 2 Total

Target-like Article omis- 3 clitic prod. article prod. sion 12 (25%) 35 (75%) 0 265 (96.4%) 10 (3.6%) 4 (28.6%) 18 (47.4%) 20 (52.6%) 1 (100%) 98 (90.7%) 10 (9.3%) 5 (35.7%) 393 (84%) 75 (16%) 10 (33.3%)

3 clitic omission 1 (100%) 10 (71.4%) 0 9 (64.3%) 20 (66.6%)

Clitic contexts 1 14 1 14 30

In relation to direct object clitics, we ¿nd on average an omission rate of 66.6%, even though there are relatively few clitic contexts by which to evaluate clitic production/omission. Pronominal clitics do not reach levels above 28.6–35.7% in the second transcript even though the children were respectively 4;9 and 4;10 years old at the time of the recordings; this sharply contrasts with the behaviour of TD children, who produce 95.7% of pronominal clitics for the group of 4 to 5 year-olds, as shown above. The full inventory of de¿nite articles in Catalan is (e)l, la, els, les, for masculine singular, feminine singular, masculine plural and feminine plural respectively (recall that they are homophonous with object clitics). As shown in Table 5 article production appears to be delayed in the ¿rst transcript, but by the second transcript J and A are omitting very few de¿nite articles (3.6% by J, 9.3% by A). Although no statistical analysis of these data is possible due to the few data points available, there is a clear contrast between the production of pronominal clitics and the production of de¿nite articles. Given the scarcity of spontaneous production data for Catalan SLI, compounded with the problems associated with the analysis of spontaneous production, we ran an elicitation experiment with ¿ve SLI children: P and C, both aged 5, N, aged 9, S, aged 12 and Y, aged 15. They were attending a speech therapist’s consultancy and had been diagnosed with SLI by standard practice.

88

Anna Gavarró

The elicitation test used was designed in the context of COST Action A33 (Varlokosta et al. in preparation); the children were asked to complete a sentence by the method exempli¿ed in (9): (9)

La nena renta la girafa i la girafa està neta. Com és que la girafa està neta? La girafa està neta perquè la nena… the girl washes the giraffe and the giraffe is clean now. Why is the giraffe clean? The giraffe is clean because the girl… Expected response: La renta. cl-fem washes ‘washes it.’

Each child was tested on 12 experimental items (see the appendix for the full list) plus ¿llers. Following the procedure of COST A33, twenty TD children were tested for Catalan (mean age: 65 months; age range: 61–70). Here we report the results for Catalan TD children (Table 6) as a point of reference for the new SLI results (Table 7): Table 6. Results of elicitation for Catalan, TD children 5-year-olds

Target clitic 206/240 85.8%

Omission 0

Dative cl 31/240 12.9%

Full DP 1/240 0.4%

Other 2/240 0.8%

Full DP

Other 1/12 1/12

Table 7. Results of elicitation for Catalan, SLI children Target clitic P (age 5) 3/12 C (age 5) 3/12 N (age 9) 5/12 S (age 12) 11/12 Y (age 15) 1/12 total 23/60 (38.3%)

Omission

0

Dative cl 8/12 8/12 7/12

1/12 10/12 33/60 (55%) 2/60 (3.3%)

2/60 (3.3%)

The SLI children tested were all older than the ones in the CHILDES database, with two only slightly older and the remaining three older by ¿ve years or more. With the exception of S, all of them behaved differently from the TD children whose results have been presented so far: although the participants did not omit third person object clitics, they did not exhibit a target-like behaviour given that they produced many li dative clitics. This answer type is exempli¿ed in (10):

Third person clitic production and omission in Romance SLI

89

(10) … perquè la mare li pentina. because the mother cl-DAT combs ‘because the mother combs his/her hair.’ 3.2.

Previous results on Spanish

If we turn to Spanish, de la Mora (2004) carried out various elicitation tasks with ten Mexican Spanish-speaking children, together with ten age-matched and ten MLU-matched children, all from Mexico, DF. Age and MLU of the SLI children appear in Table 8: Table 8. Data source for Spanish Child MEDH MEDF MEET FELB FELG MERC MEJC MEJH FEPS FEAC

Age in months 66 63 74 73 69 60 51 73 47 59

MLU 3.6 3.5 2.3 2.5 2.9 2.8 1.8 3.2 2.7 3.2

De la Mora (2004) provides results on object clitics for two contexts, spontaneous production (in an experimental setting) and ‘prompted response’: Table 9. Percentage of object clitic production, Spanish, SLI children

Spontaneous production Prompted response

Target-like clitic prod.

Unmatched clitic prod.

Clitic omission

Full DP

Other

157 (37%)

68 (9%)

39 (9%)

150 (36%)

6 (2%)

191 (45%)

72 (17%)

34 (8%)

120 (29%) 3 (0.7%)

Object clitic omission ranges between 8% and 9%; this contrasts with the result found for the younger Catalan children. The overall accuracy in the production of object clitics is, according to de la Mora (2004), only 45% in the prompted response condition, but her error category includes all errors (and reÀects most-

90

Anna Gavarró

ly errors involving gender and number, on which we have nothing to say5); if we focus on omission, Spanish SLI cannot be characterised by the omission of object clitics. Nor is there evidence of inappropriate dative or reÀexive clitics. It is worth pointing out that de la Mora’s tasks give rise to a high number of full DPs in contexts in which they would appear not to be felicitous; we do not have an explanation for that, but in her results large numbers of full DP were also produced by the age- and MLU-matched children, so it is not a feature that characterises SLI; in the results for Spanish TD children in Wexler, Gavarró, and Torrens (2004) no full DPs were produced by the Spanish-speaking children, and only a marginal number were produced by the Catalan-speaking ones – so the reason for the considerable number of full DP in child Mexican Spanish may relate to the experimental design and procedure. De la Mora (2004) also quanti¿es the production of the de¿nite articles la, los and las, as shown in Table 10. Table 10. Percentage of target-like production of de¿nite articles, Spanish, SLI children and controls Group SLI MLU AGE

la 91.49% (43/47) 100% (54/54) 98.61% (71/72)

los 72.73% (8/11) 100% (42/42) 93.75% (30/32)

las 87.5% (7/8) 100% (10/10) 91.67% (22/24)

Total 85% (56/66) 100% (106/106) 96% (123/128)

Overall, accuracy in the production of de¿nite articles is 85% for the children with SLI, a rate which is slightly different for age- and MLU matched children (since they produce target-like forms in 96% and 99% of cases respectively). In this respect, there seems to be no contrast between Catalan (with a production rate of 84%) and Spanish SLI. In fact, the Catalan-speaking SLI children produce articles very consistently after MLU 1.5, and the results by de la Mora also indicate relatively little difference between SLI and TD children. Previous work on object clitics in SLI in Spanish provides results similar to those yielded by de la Mora (2004). Former work by Serra Raventós and Bosch Galceran (1993) based on spontaneous production does not provide details of the errors produced by ¿ve Spanish-speaking children with SLI (ages 6;8 to 8;1), but indicates that “errors with pronouns were less frequent than expected” and “consisted of omission, inappropriate use, some errors in agreement and doubling” – note that pronouns in that study included clitics other than third 5. But see Grinstead, Cantú-Sánchez, and Flores-Ávalos (2008) for an analysis of the errors in nominal number morphology in Spanish SLI.

Third person clitic production and omission in Romance SLI

91

person object clitics (while in this paper we restrict ourselves, as in most of the literature, to third person); interestingly, there was no statistically signi¿cant difference between the children with SLI and their controls in clitic production. Bedore and Leonard (2001) report an experimental study of ¿fteen Mexican Spanish-speaking children aged between 3;11 and 5;6, with a mean MLU of 2.88 (MLU range: 2.46–3.62); these children were tested together with ageand MLU-matched children. The results for object clitic production are shown in Table 11. (The fact that a relatively high number of non-responses were produced is due to the elicitation procedure, in which the children were asked a rather open question.) Table 11. Object clitic responses (target in bold), Spanish SLI children and controls Response Masc. sg. Fem. sg. Masc. pl. Fem.pl. Omission No resp. Total MLU matched Omission Age matched Omission

Masc. sg. 41 21 4 0 10 14 90

Fem. sg. 5 55 0 5 7 18 90

Masc. pl. 31 13 17 3 15 11 90

Fem. pl Target 7 43 3 14 14 9 90

14

8

12

11

4

2

5

1

As can be seen, omission only amounts to 46/360 responses, i.e. there is a 12.7% omission rate, which resembles de la Mora’s (2004) omission rate of 8–9%. Bedore and Leonard (2001) also provide results for MLU-matched and age-matched groups of children. The percentage of omission of object clitics for the MLU-matched group is close to that of the SLI children: 45/360 responses, 12.5%. The age-matched children produced a very low rate of clitic omission, 12/360 responses, 3%, as we would expect given the results for Spanish in Wexler, Gavarró, and Torrens (2004). Despite the general claim by Bedore and Leonard (2001) that object clitic production is an impaired domain in a language like Spanish, presumably due to errors in the production of gender/number morphemes, the error pattern in Spanish is very different from that in Catalan.

92

3.3.

Anna Gavarró

A comparison

I undertake a comparison of the new results for Catalan with those in the literature for Spanish. This is a ¿rst approximation to the topic before, in future work, the same experiment is run in Catalan and Spanish with a similar and higher number of matched subjects. For the moment, let us compare the spontaneous production of the Catalanspeaking children A and J to the results for Spanish of de la Mora (2004). I leave aside the ¿rst recording of A and J, given the small size of the sample. J’s MLU in the second transcript, 4.7, is higher than those of all the Spanishspeaking children, whose MLU ranges from 1.8 to 3.6; yet, he omits 71.4% of third person object clitics, while the Spanish-speaking children omit between 8% and 9% (depending on the elicitation technique). A’s MLU in the second transcript is 3.1, higher than that of six of the Spanish-speaking children, and lower than that of four; his omission rate is 64.3%, again much higher than the 8–9% average omission for the Spanish-speaking children. Our experimental results indicate that with Catalan-speaking children older than 4, there is no object clitic omission, but target-like performance represents only 38.3% of production. There is one child, S, aged 12, who has adult-like performance, and such cases can also be found in the records for other languages such as French and Italian. The remaining four children revert to the dative clitic li instead of the third person object clitic. No such phenomenon is mentioned in the literature in relation to child Spanish. How well do these results match the predictions of the UCC? If we compare what TD and SLI children do in Spanish, we can see that the results reported are as one would expect from the UCC: there is virtually no clitic omission in child Spanish, and very little in Spanish SLI. In contrast, there is considerable omission in TD children under the age of 4 in Catalan and high rates of omission in young Catalan SLI children. Object clitic omission may thus be a marker of SLI only in Catalan, not in Spanish. There is something, at ¿rst sight, unexpected in the original experimental results for older SLI Catalan-speaking children: the lack of third person clitic omission. There is, on the other hand, a phenomenon not previously attested in the high degree of occurrence seen here: the production of dative clitics when third person objects ought to appear. In the results by Jakubowicz and Nash (to appear) reported above, children produced an unexpectedly high number of reÀexive clitics instead of third person object clitics (at a rate of 21.7% for the SLI1b group). By contrast, the presence of dative clitics in Catalan was marginal in the results for TD children in Wexler, Gavarró, and Torrens (2004), and is also found, to a lesser extent, in the TD children in Table 4 above.

Third person clitic production and omission in Romance SLI

93

Why should Catalan SLI children revert in particular to dative clitics in place of third person object clitics? Gavarró and Mosella (2009) studied the acquisition of dative clitics in Catalan with an elicitation task similar to that used in Wexler, Gavarró, and Torrens (2004), Gavarró et al. (2010). Forty Catalanspeaking children between 2 and 5 were tested, and they produced target-like clitics very consistently after the age of 2. Table 12 presents the results. Table 12. Production and omission of dative clitics, Catalan, TD children Age group 2-year-olds 3-year-olds 4-year-olds 5-year-olds Adults

Omission 35% 8% 3% 0% 0%

Production 65% 91% 97% 100% 100%

These results were analysed in the light of the UCC and we argued that omission was very low (after a threshold MLU) because a dative clitic did not involve the checking of more than one uninterpretable feature. Therefore the UCC did not have any effect on the derivation of a sentence with a dative clitic, and children produced dative clitics without dif¿culty. Based on the analysis of Gavarró and Mosella (2009), SLI children may revert to a dative clitic because a dative clitic is not problematic for a grammar subject to the UCC. That is, just as children omit clitics at an early age (whether TD or SLI) to avoid a UCC violation, they may produce a different third person clitic also in order to avoid a UCC violation. The resulting sentence converges because the child has not failed to eliminate any uninterpretable feature, but the sentence is still deviant because a dative clitic has been produced instead of an accusative clitic. We can conclude that, in Spanish, the three existing studies on clitic production in SLI produce very consistent results, all indicating low rates of object clitic omission; this contrasts with what we have found to be the case for the spontaneous productions of two Catalan-speaking children with SLI, who omit third person object clitics in over half of the clitic contexts, at an age when TD children have ceased to omit pronominal clitics. The older children tested omitted no clitics, but instead produced dative clitics in place of object clitics. Given the limitations of the new data, to give further support to our hypothesis we now turn to the results available on pronominal clitic production in other languages.

94

4.

Anna Gavarró

Beyond Catalan and Spanish

By hypothesis, we expect object clitic omission and/or replacement to occur for an extended period in SLI in participle agreement languages. We will start with two such languages, French and Italian, and will proceed with languages without participle agreement, for which the prediction is early target-like behaviour. In French, third person pronominal clitics are homophonous with de¿nite articles. When compared to those in Catalan, there is one difference of no consequence: there is no gender marking in the plural forms of French de¿nite articles and pronominal clitics. Jakubowicz et al. (1998) discuss clitic acquisition in general and argue that, if the rates of disruption of article and object clitic production are not coincident, we can discard the possibility that the phonological features of these elements are at the source of the disruption (as Leonard’s (1998) Surface Hypothesis claims). Jakubowicz (2003) and Jakubowicz et al. (1998) indeed observed that de¿nite article production and pronominal clitic pronoun production follow different developmental paths. The results already presented in Table 1 from Jakubowicz and Nash (to appear) indicate an overall 66% omission rate of object clitics in young French children with SLI; the groups of older children with SLI display decreasing rates of omission: 20.6%, 14.8%, 4.7% (still higher than the 4.3% omission rate of TD four-year-olds). Those same children produced target-like determiners at rates between 81.8% and 100%. Paradis and Crago (2001) and Paradis, Crago, and Genesee (2003, 2005/2006) have also studied third person object clitic omission and production in the acquisition of Quebec French, and maintain that clitic omission is a marker of SLI in French. Italian is the other participle agreement language that has been investigated. Leonard et al. (1992) examined the semi-spontaneous productions of ¿fteen SLI children aged 3;8 to 5;7, with an MLU ranging from 2.7 to 4.2. These children produced only 26% of the clitics expected, and otherwise omitted them; the MLU-matched controls in the study produced 66.3% of the expected clitics, and the age-matched controls 91.7%. The analysis of SLI in Italian of Bottari et al. (1998) is based on the analysis of the productions of eleven children whose mean MLU was 2.6, similar to the mean MLU of the Catalan-speaking children and Spanish-speaking children studied here. Regarding object clitic production, their results are given in Table 13. We can see in the results of Bottari et al. (1998) that some children hardly produced any clitic contexts at all (similarly to the children in our study, who only produced one clitic context in the ¿rst transcript). For those who did pro-

Third person clitic production and omission in Romance SLI

95

duce clitic contexts, omission was generally high, with only two children with relatively low levels of omission: DG (8% omission) and AR (19% omission). The remaining six children omitted pronominal clitics at rates between 30.7% and 83%. The levels of determiner omission are higher than those of pronominal clitics for all the Italian subjects in this study, in contrast with what is found in the majority of studies reported here – for reasons that remain unclear. The dissociation of determiner and object clitic production is also pointed out by Bottari et al. (1998) as it was for French by Jakubowicz et al. (1998). Table 13. Production of clitics, Italian, SLI children SS MFun MFan EG JM MFab JT SG DG AR PF Mean

MLU 2.7 1.5 2.2 2.8 2.3 2 4 3.7 3.5 2.4 2.2 2.6

Clitic 3 0 25 1 0 1 8 4 11 17 9

Omission 0 3 54 0 0 5 7 2 1 4 4

% Omission – – 68% – – 83% 46% 33% 8% 19% 30.7% 41.1%

Cipriani, Bottari, Chilosi, and Pfanner (1998) undertake the longitudinal study of a child with SLI from age 6;2 to 13;2 (MLU at 6;2: 2.3; at 13;2: 4.3). In the spontaneous productions of this child, clitic omission was consistently high (from 50 to 75% of clitic contexts) until age 9;4. At age 11;2 it dropped to 7% and after age 12 descended further to 4%. Bortolini, Caselli, Deevy, and Leonard (2002) and Bortolini, Arfé, Caselli, Degasperi, Deevy, and Leonard (2006) investigated possible clinical markers for SLI in Italian. The eleven pre-school SLI children tested produced on average less than 20% target object clitics. The authors concluded that children with SLI were signi¿cantly less likely to produce target clitics in obligatory clitic contexts than TD children; the error type in elicited production was omission. The pattern of clitic omission/disruption in Catalan resembles that in French and Italian. If we assume that SLI is a grammatically based disruption, the result of a protracted UCC-stage, the facts of clitic omission/disruption in the Romance language with participle agreement are predicted (as pointed out for French and Italian in Wexler 2003). For other Romance languages without par-

96

Anna Gavarró

ticiple agreement, such as Romanian, we would expect very low rates of object clitic omission: this is in fact the case for TD Romanian-speaking children (see Babyonyshev and Marin 2004; I refer the reader to Babyonyshev and Marin 2006 for a full discussion of the work on TD Romanian speaking children, including the work of Avram 1999). To my knowledge no study of Romanian SLI has as yet been undertaken. It is important to stress that the analysis put forward here generalises over languages outside the Romance family, as nothing in the hypothesis ties it to a particular language type: it is a more abstract characteristic of a grammar, namely the presence of an uninterpretable feature in vP, that determines whether clitic omission/replacement will take place or not. To show how the analysis holds for a language outside Romance, we brieÀy consider recent results from Greek. Greek is a language with clitics and without participle agreement; TD children acquiring Greek do not omit object clitics (see Tsakali and Wexler 2004). However, work on object clitics in Greek SLI has appeared to be inconclusive. For example, while Tsimpli (2001) reported levels of clitic omission over 90%, other studies have reported low omission rates (5% omission rate in Terzi 2007). Recently, Manika, Varlokosta, and Wexler (2011) re-examined previous studies and concluded that variation in the results from one study to the next had its source in varying methodologies and subject selection. They then conducted an elicitation experiment with seventeen SLI children (aged 4;10 to 8;1) and thirty-two control TD children, and found no statistical difference in the performance of the two groups, who produced the expected clitics over 95% of the time. Greek SLI children pattern with Spanish SLI children, not with Catalan, French or Italian children. Finally, Stavrakaki and Chrysomallis (2011) tested a bilingual FrenchGreek child aged 9. According to them, the child performed at ceiling with Greek object clitics, but showed dif¿culty in French. If we assume that a true bilingual child (i.e. not affected by cross-linguistic inÀuence) performs like a monolingual child for each of his/her languages, then we predict that a bilingual child may for example omit clitics in Catalan, Italian or French, but not in Greek or Spanish. This prediction appears to be borne out in the French-Greek child tested by Stavrakaki and Chrysomallis (2011).

Third person clitic production and omission in Romance SLI

5.

97

Conclusion

Let us go back to our research questions: (13) a.

Are clitics, due to their phonological properties, singled out in SLI as a class?

b.

Is the de¿cit encountered in third person pronominal clitics universal, or is it subject to cross-linguistic variation, as it is in TD children?

With respect to the ¿rst question, several authors stated before me that the same phonological unit varies in the degree of affectation in SLI depending on whether it is a determiner or a clitic pronoun, a claim corroborated by our own results. Hence, the phonological properties of clitics cannot be the source of impairment, and clitics do not constitute a class when we consider acquisition, whether impaired or not. In answer to the second question, our own results and those in the literature allow us to conclude that object clitics appear to be problematic in only a subset of languages, Catalan, French and Italian, but not in others, such as Spanish. So problems with object clitics are subject to cross-linguistic variation in SLI, just as they are in TD children (Gavarró et al. 2010), in spite of the fact that the literature often refers to object clitics as a problematic domain for children with SLI irrespective of the language. If we assume the Extended-UCC analysis of SLI, what we expect for Romance is that SLI will have different visible consequences depending on how the participle agreement parameter is set in a particular language. Third person object clitic production is problematic if the participle agreement parameter is set positively, but not when the parameter is set negatively (meaning that vP does not have an uninterpretable feature). The empirical evidence presented here indicates that, when the participle agreement parameter is set positively, this may result not only in omission but also in replacement of a third person object clitic by a dative clitic, i.e. the UCC may have varying effects on production.

98

Anna Gavarró

Appendix Test items (presented together with 5 ¿llers), elicitation task 1. La princesa ha tapat el soldat. Ara, el soldat no té fred. Com és que el soldat no té fred? El soldat no té fred perquè la princesa… The princess has covered the soldier. Now the soldier isn’t cold. How come the soldier isn’t cold? He isn’t cold because the princess … 2. La mare ha pentinat la nena i la nena ha quedat molt maca. Com és que la nena ha quedat molt maca? La nena ha quedat molt maca perquè la mare… The mother combed the girl’s hair and now the girl looks very pretty. How come the girl looks pretty? She looks pretty because the mother… 3. El nen eixuga l’hipopòtam. L’hipopòtam aviat estarà sec. Com és que l’hipopòtam estarà sec aviat? L’hipopòtam estarà sec aviat perquè el nen… The child dries the hippopotamus. The hippopotamus will soon be dry. Why will the hypopotamus soon be dry? It will be dry because the child … 4. La nena desperta el nen i el nen es posa a plorar. Com és que el nen es posa a plorar? El nen es posa a plorar perquè la nena… The girl wakes the boy up and the boy bursts out crying. Why is boy crying? The boy is crying because the girl … 5. El pintor pinta la nena i la nena està contenta. Com és que la nena està contenta? La nena està contenta perquè el pintor… The painter paints the girl and the girl is happy. Why is the girl happy? She is happy because the painter… 6. El nen mulla el gat i el gat està xop. Com és que el gat està xop? El gat està xop perquè el nen… The boy sprays the cat and the cat is wet. Why is cat wet? The cat is wet because the boy… 7. L’home ha pintat la casa i la casa ara és blava. Com és que la casa ara és blava? La casa ara és blava perquè l’home… The man has painted the house blue. Why is the house blue now? The house is blue because the man… 8. El gos llepa el gat i el gat està content. Com és que el gat està tant content? El gat està content perquè el gos… The dog licks the cat and the cat is happy. Why is the cat happy? The cat is happy because the dog …

Third person clitic production and omission in Romance SLI

99

9. L’abella ha lligat el grill i el grill ara no pot saltar. Com és que el grill no pot saltar? El grill no pot saltar perquè l’abella… The bee has tied the cricket and the cricket can’t jump. Why can’t the cricket jump? It can’t jump because the bee… 10. La nena renta la girafa i la girafa està neta. Com és que la girafa està neta? La girafa està neta perquè la nena… The girl washes the giraffe and the giraffe is clean now. Why is the giraffe clean? The giraffe is clean because the girl… 11. El nen ha menjat un tros de pastís i el tros de pastís ja no hi és. Com és que el tros de pastís ja no hi és? El tros de pastís ja no hi és perquè el nen… The boy has eaten a piece of cake and the piece of cake is gone. Why is the piece of cake gone? The piece of cake is gone because … 12. La nena ha caçat una papallona i la papallona no pot volar. Com és que la papallona no pot volar? La papallona no pot volar perquè la nena… The girl has caught a butterÀy and now the butterÀy can’t Ày. Why can’t the butterÀy Ày? It can’t Ày because the girl… References Avram, Larisa 1999 Clitic omission in child language and multiple spell-out. Poster presented at GALA 1999, Potsdam. Babyonyshev, Maria and Stefania Marin 2004 Object clitics in Romanian. In Proceedings of the 29th Boston University Conference on Language Development, A. Brugos, M. R. Clark-Cotton, and S. Ha (eds.), 49–60. Somerville: Cascadilla Press. Babyonyshev, Maria and Stefania Marin 2006 Acquisition of pronominal clitics in Romanian. Catalan Journal of Linguistics 5: 17–44. Bedore, Lisa and Laurence Leonard 2001 Grammatical morphology de¿cits in Spanish-speaking children with speci¿c language impairment. Journal of Speech, Language, and Hearing Research 44: 905–924. Bortolini, Umberta, Barbara Arfé, M. Cristina Caselli, Luisa Degasperi, Patricia Deevy and Laurence Leonard 2006 Clinical markers for speci¿c language impairment in Italian: the contribution of clitics and non-word repetition. International Journal of Language and Comunication Disorders 6: 695–712.

100

Anna Gavarró

Bortolini, Umberta, M. Cristina Caselli, Patricia Deevy and Laurence Leonard 2002 Speci¿c language impairment in Italian: ¿rst steps in the search of a clinical marker. International Journal of Language and Communication Disorders 37: 77–93. Bottari, Piero, Paola Cipriani and Anna M. Chilosi 1996 Root in¿nitives in Italian SLI children. In Proceedings of the 20th Boston University Conference on Language Development, A. Stringfellow, D. Cahana-Amitay, E. Hughes, and A. Zukowski (eds.), 75–86. Somerville: Cascadilla Press. Bottari, Piero, Paola Cipriani, Anna M. Chilosi and Lucia Pfanner 1998 The determiner system in a group of Italian children with SLI. Language Acquisition 7: 285–315. Buesa, Carlos 2007 Early root in¿nitives in a null-subject language: a longitudinal case study of a Spanish child. In Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition (GALANA), A. Belikova, L. Meroni, and M. Umeda (eds.), 39–50. Somerville: Cascadilla Press. Castilla, Anny and Ana T. Pérez-Leroux. 2010 Omissions and substitutions in Spanish object clitics: developmental optionality as a property of the computational system. Language Acquisition 17(2): 2–25. Castilla, Anny, Ana T. Pérez-Leroux and Alice Eriks-Brophy 2008 Omission in early Spanish clitics’. In Language Acquisition and Language Development. Proceedings of GALA 2007, A. Gavarró, and M. J. Freitas (eds.), 112–122. Newcastle, Cambridge Scholars Publishing. Chierchia, Gennaro, Maria Teresa Guasti and Andrea Gualmini 1999 Nouns and articles in child grammar and the syntax/semantics map’. Paper presented at GALA, Potsdam. Cipriani, Paola, Piero Bottari, Anna M. Chilosi and Lucia Pfanner 1998 A longitudinal perspective on the study of speci¿c language impairment: the long term follow-up of an Italian child. International Journal of Language and Communication Disorders 33: 245–280. Clahsen, Harald 1989 The grammatical characterization of developmental dysphasia. Linguistics 27: 897–920. Clahsen, Harald, Susanne Bartke and Sandra Göllner 1997 Formal features in impaired grammars: a comparison of English and German SLI children. Journal of Neurolinguistics 10: 151–171. Davidiak, Elena and John Grinstead 2004 Root non-¿nite forms in child Spanish. Poster presented at Generative Approaches to Language Acquisition, GALANA 2004, Hawai’i.

Third person clitic production and omission in Romance SLI

101

De la Mora, Juliana 2004 Direct object clitics in Spanish-speaking children with and without Speci¿c Language Impairment. Master’s thesis, University of Alberta. De Jong, Jan 1999 Speci¿c Language Impairment in Dutch: InÀectional Morphology and Argument Structure. Enschede: Print Partners Ipskamp. Fujino, Hanako and Tetsuya Sano 2002 Aspects of the null object phenomenon in child Spanish’. In The Acquisition of Spanish Morphosyntax, A. T. Pérez Leroux, and J. Muñoz Liceras (eds.), 67–88. Dordrecht, Kluwer. Gavarró, Anna, Vicenç Torrens and Ken Wexler 2010 Object clitic omission: two language types. Language Acquisition 17(4): 192–219. Gavarró, Anna and Marta Mosella 2009 Testing syntactic and pragmatic accounts of clitic omission. In Proceedings of the Third Conference on Generative Approaches to Language Acquisition, J. Crawford, K. Otaki, and M. Takahashi (eds.), 25–35. Somerville, Cascadilla Press. Grinstead, John, Myriam Cantú-Sánchez ,and Blanca Flores-Ávalos 2008 Canonical and epenthetic plural marking in Spanish-speaking children with Speci¿c Language Impairment. Language Acquisition 15: 329–349. Guasti, Maria Teresa, Anna Gavarró, Joke de Lange and Claudia Caprin 2008 Article omission across child languages. Language Acquisition 15: 89– 119. Guasti, Maria Teresa, Joke de Lange, Anna Gavarró and Claudia Caprin 2004 Article omission: across child languages and across special registers. In Proceedings of GALA 2003, J. van Kampen, and S. Baauw (eds.), 199– 210. Utrecht: LOT Occasional Series, vol. 1. Hamann, Cornelia, Stéphanie Cronel-Ohayon, Sébastien Dubé, Ulrich Frauenfelder, Luigi Rizzi, Michal Starke and Pascal Zesiger 2003 Aspects of grammatical development in young children French children with SLI. Developmental Science 6: 151–158. Hansson, Kristina and Ulrika Nettelbladt 1995 Grammatical characteristics of Swedish children with SLI. Journal of Speech and Hearing Research 38: 589–598. Jakubowicz, Celia 2003 Hypothèses psycholinguistiques sur la nature du dé¿cit dysphasique. In Les dysphasies, Ch. L. Gérard, and V. Brun (eds.), 23–70. Paris: Masson. Jakubowicz, Celia and Lea Nash To appear Why accusative clitics are avoided in normal and impaired language development. In Essays in Syntax, Morphology and Phonology in SLI, C. Jakubowicz, L. Nash, and K. Wexler (eds.). Cambridge, MA: The MIT Press.

102

Anna Gavarró

Jakubowicz, Celia, Lea Nash, Catherine Rigaut and Christophe -Loic Gérard 1998 Determiners and clitic pronouns in French-speaking children with SLI. Language Acquisition 7: 113–160. Leonard, Laurence 1998 Children with Speci¿c Language Impairment. Cambridge, MA: The MIT Press. Leonard, Laurence 2004 The MIT Encyclopedia of Communication Disorders. Cambridge, MA: The MIT Press. Leonard, Laurence and Umberta Bortolini 1998 Grammatical morphology and the role of weak syllables in the speech of Italian-speaking children with speci¿c language impairment. Journal of Speech, Language, and Hearing Research 41: 1363–1374. Leonard, Laurence, Umberta Bortolini, M. Cristina Caselli, Karla K. McGregor and Letizia Sabbadini 1992 Morphological de¿cits in children with speci¿c language impairment: the status of features in the underlying grammar. Language Acquisition 2: 151–179. Leonard, Laurence, Letizia Sabbadini, Jeannette Leonard and Virginia Volterra 1987 Speci¿c language impairment in children: a cross-linguistic study. Brain and Language 32: 233–252. MacWhinney, Brian and Catherine Snow 1985 The child language data exchange system. Journal of Child Language 12: 271–295. Manika, So¿a, Spyridoula Varlokosta and Ken Wexler 2011 The lack of omission of clitics in Greek children with SLI: an experimental study. In Proceedings of the 35th Boston University Conference on Language Development, N. Danis, K. Mesh, and H. Sung (eds.), 427– 439. Somerville: Cascadilla Press. Paradis, Johanne and Martha Crago 2001 The morphosyntax of SLI in French: Evidence for an extended optional default account. Language Acquisition 9: 269–300. Paradis, Johanne, Martha Crago and Fred Genesee 2003 Object clitics as a clinical marker of SLI in French: evidence from FrenchEnglish bilingual children. In Proceedings of the 27th Boston University Conference on Language Development, B. Beachley, A. Brown, and F. Conlin (eds.), 638–649. Somerville: Cascadilla Press. Paradis, Johanne, Martha Crago and Fred Genesee 2005/2006 Domain-general versus domain-speci¿c accounts of Speci¿c Language Impairment: evidence from bilingual children’s acquisition of object pronouns. Language Acquisition 13: 33–62.

Third person clitic production and omission in Romance SLI

103

Pratt, Amy and John Grinstead 2007 Optional in¿nitives in child Spanish. In Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition (GALANA), A. Belikova, L. Meroni, and M. Umeda (eds.), 351–362. Somerville: Cascadilla Press. Rice, Mabel 2003 A uni¿ed model of Speci¿c Language Delay: grammatical tense as a clinical marker of unexpected variation. In Language Competence Across Populations, Y. Levy, and J. Schaeffer (eds.), 63–95. Mahwak, NJ: Lawrence Erlbaum. Rice, Mabel and Ken Wexler 1996 Toward tense as a clinical marker of speci¿c language impairment in English speaking children. Journal of Speech and Hearing Research 39: 850–863. Rizzi, Luigi 1993/94 Some notes on linguistic theory and language development: The case of root in¿nitives. Language Acquisition 3(4): 371–393. Salustri, Manola and Nina Hyams 2003 Is there an analogue to the RI stage in the null subject languages?. In Proceedings of the 27th Boston University Conference on Language Development, B. Beachley, A. Brown, and F. Conlin (eds.), 692–703. Somerville: Cascadilla Press. Serra Raventós, Miquel and Laura Bosch Galceran 1993 Análisis de los errores de producción en los niños con trastorno especí¿co del lenguaje. Revista de Logopedia, Foniatría y Audiología XIII(1): 2–13. Sportiche, Domique 1996 Clitic constructions. In Phrase Structure and the Lexicon, J. Rooryck, and L. Zaring (eds.), 213–276. Dordrecht: Kluwer. Stavrakaki, Stavroula and Marie-Annick Chrysomallis 2011 Subject-verb agreement, object clitics and wh-questions in bilingual French-Greek SLI: the case study of a French-Greek-speaking child with SLI. Clinical Linguistics and Phonetics 25(5): 339–67. Terzi, Arhonto 2007 Achimedes I Corpus (Prepositions in Normal, Early and Impaired Language). Patras: TEI. Torrens, Vicenç 1995 The acquisition of inÀection in Spanish and Catalan. Papers in Language Processing and Acquisition, MIT Working Papers in Linguistics 26: 452– 472.

104

Anna Gavarró

Tsakali, Vina and Ken Wexler 2004 Why children omit clitics in some languages but not in others: new evidence from Greek. In Proceedings of GALA 2003, J. van Kampen, and S. Baauw (eds.), 493–504. Utrecht: LOT. Tsimpli, Ianthi M. 2001 LF-Interpretability and Language Development: A Study of Verbal and Nominal Features in Greek Normally Developing and SLI Children. Brain and Language 77(3), 432–432. Varlokosta, Spyridoula, Adriana Belletti, João Costa, Naama Friedmann, Anna Gavarró, Kleanthes Grohmann, Maria Teresa Guasti, Laurie Tuller, Maria Lobo, Darinka Andjelkovic, Núria Argemí, Larisa Avram, Sanne Berends, Valentina Brunetto, Hélène Delage, María-José Ezeizabarrena Segurola, Iris Fattal, Ewa Haman, Kristine Jensen de Lopez, Napoleon Katsos, Lana Kologranic, Nadezda Krstiü, Jelena Kuvac Kraljevic, Aneta MiĊkisz, Mixaela Nerantzini, Clara Queraltó, Zeljana Radic, Sílvia Ruiz, Uli Sauerland, Anca Sevcenco, Magdalena SmoczyĔska, Eleni Theodorou, Heather van der Lely, Angeliek van Hout, Alma Veenstra, John Weston, Maya Yachini and Kazuko Yatsushiro In preparation A cross-linguistic study of the acquisition of clitic and pronoun production. Ms. Wexler, Kenneth 1998 Very early parameter setting and the unique checking constraint: A new explanation of the optional in¿nitive stage. Lingua 106: 23–79. Wexler, Kenneth 2003 Lenneberg’s dream: learning, normal language development, and speci¿c language impairment. In Language Competence Across Populations, Y. Levy, and J. Schaeffer (eds.), 11–61. Mahwak, NJ: Lawrence Erlbaum. Wexler, Kenneth To appear The Unique Checking Constraint as the explanation of clitic omission in SLI and normal development. In Essays in Syntax, Morphology and Phonology in SLI, C. Jakubowicz, L. Nash, and K. Wexler (eds.). Cambridge, MA: The MIT Press. Wexler, Ken, Anna Gavarró and Vicenç Torrens 2004 Feature checking and object clitic omission in child Spanish and Catalan. In Romance Language and Linguistic Theories, R. B. Bennema, B. Hollebrandse, B. Kampers-Manhe, and P. Sleeman (eds.), 253–70. Amsterdam/Philadelphia: John Benjamins. Wexler, Kenneth, Jeannette Schaeffer and Gerard Bol 2004 Verbal syntax and morphology in typically developing Dutch children and children with SLI: how developmental data can play an important role in morphological theory. Syntax 7(2): 148–198.

Which category replaces an omitted clitic? The case of European Portuguese João Costa, Maria Lobo and Carolina Silva

This paper reports on a set of experiments run in order to assess clitic production in the acquisition of European Portuguese. The fact that children omit clitics across the board in this language led to an investigation of the acquisition of null objects. The ¿ndings suggest that children overuse null objects, and that this overuse is due to poor knowledge of the type of category involved in the null object construction.

1.

Introduction

A well known fact about child language is that pronominal clitics are omitted in children’s early productions in many languages (cf. Jakubowicz et al. 1998, Wexler, Gavarró and Torrens, 2003; Tsakali and Wexler, 2003; Babyonyshev and Marin, 2005, among others). There is less consensus regarding the universality of clitic omission. For instance, in Spanish, there is an ongoing debate regarding the existence of omission and its duration (cf. e.g. Castilla et al. 2008). In recent work, we contend that there is evidence to claim that clitic omission is not a uniform phenomenon (Costa, Lobo and Silva 2009, Costa and Lobo 2010b). Another area in which there is little consensus – and not much discussion – is on the nature of the (empty) category replacing the omitted clitic in languages with omission. The goal of this paper is twofold: ¿rst, we review previous work on clitic production by children acquiring European Portuguese, showing that there is omission in this language. Second, we contribute to the debate regarding the nature of the empty category, by arguing that children acquiring European Portuguese produce a null object, as in the target adult grammar, when they fail to produce a clitic. Based on this discussion, we try to ¿nd evidence for or against continuity in the representation of the null object in child language, by inspecting whether it is a pro, a variable or whether children’s omissions of complements might be target deviant instances of VP-ellipsis.

106

João Costa, Maria Lobo and Carolina Silva

This paper is organized as follows: in Section 2, some background information on clitics and null objects in European Portuguese is provided. Section 3 presents the results of previous studies on the production of different types of clitics by children acquiring European Portuguese. Section 4 discusses a set of comprehension experiments con¿rming the idea that European Portuguese speaking children know that the language has null objects. Finally, in Section 5, we present some preliminary results of a study aimed at determining whether European Portuguese children know the nature and properties of the empty category involved in the null object construction.

2.

Clitics and null objects in European Portuguese

Like other Romance languages, European Portuguese has pronominal clitics. Clitics can be accusative, as in (1a), but also dative (1b), reÀexive (1c), or nonargumental (1d): (1)

a.

O Pedro viu-o. the Pedro saw it/himacc “Pedro saw it/him.”

b.

O Pedro deu-lhe um livro. the Pedro gave him/herdat a book “Pedro gave him/her a book.”

c.

O Pedro lavou-se. the Pedro washed-SEreÀ “Pedro washed himself.”

d.

O Pedro porta-se mal. the Pedro behaves-SEinherent bad “Pedro behaves badly.”

Some clitics vary in gender, person and number. This is shown for the accusative clitic in (2) and (3): (2)

Number and gender variation a. O Pedro viu-o. the Pedro saw himacc/masc “Pedro saw him.”

Masculine singular

Which category replaces an omitted clitic? The case of European Portuguese

(3)

107

b.

O Pedro viu-a. the Pedro saw heracc/fem

c.

O Pedro viu-os. the Pedro saw themacc/masc “Pedro saw them.”

Masculine plural

d.

O Pedro viu-as. the Pedro saw themacc/fem

Feminine plural

Person variation a. O Pedro viu-me. the Pedro saw meacc

Feminine singular

1st person singular

b.

O Pedro viu-te. the Pedro saw youacc

2nd person singular

c.

O Pedro viu-o. the Pedro saw himacc

3rd person singular

d.

O Pedro viu-nos. the Pedro saw usacc

e.

O Pedro viu-vos. the Pedro saw youacc

2nd person plural

f.

O Pedro viu-os the Pedro saw themacc

3rd person plural

1st person plural

Unlike what occurs in the majority of Romance languages, clitics are not always proclitic. Depending on the syntactic environment, clitics can be proclitic, enclitic or mesoclitic. As stated in Duarte and Matos (2000), there is a limited set of proclisis triggers, and enclisis emerges elswhere. Mesoclisis is restricted to future and conditional inÀection, and occurs in enclisis contexts. In (4), we list the set of proclisis triggers (cf. Duarte and Matos 2000). In (5), an example of proclisis (5a), enclisis (5b) and mesoclisis (5c) is given. (4)

Proclisis triggering environments: Negation Wh-questions Embedded sentences with a lexical C Certain adverbs in preverbal position Preverbal quanti¿ed subjects Affective operators

108

João Costa, Maria Lobo and Carolina Silva

(5)

a.

Proclisis Eu não te vi. I not you saw “I didn’t see you”.

b.

Enclisis Eu vi-te. I saw you “I saw you.”

c.

Mesoclisis Eu ver-te-ei. I see-you-fut “I will see you.”

Besides clitics, European Portuguese also has null objects (Raposo 1986). A null object construction can be used in the exact same pragmatic environment in which a clitic is legitimate: whenever the referent is given information: (6)

A:

E o Pedro? what about Pedro?

B:

a.

Não o vi. not him saw “I haven’t seen him.”

b.

Não ’ vi. not saw “I haven’t seen him.”

As described in Raposo (1986) and Costa and Duarte (2003), null objects do not always freely alternate with clitics. In reÀexive contexts, null objects are ruled out (7): (7)

A:

E o Pedro? what about Pedro?

B:

a.

Não se lavou. not SE washed “He hasn’t washed himself”

b.

*Não ’ lavou. not washed

Which category replaces an omitted clitic? The case of European Portuguese

109

The same holds for 1st and 2nd person clitics. (8) illustrates this for 1st person: (8)

A Maria viu*(-me). the Maria saw me “Maria saw me.”

The availability of the null object construction is also constrained by certain syntactic environments. As argued in Raposo (1986), null objects are ruled out in strong island environments: (9)

A:

E o Pedro? what about Pedro

B:

a.

*Fiquei aborrecido quando abracei. got annoyed when hugged

b.

Fiquei aborrecido quando o abracei. got annoyed when him hugged “I got annoyed when I hugged him.”

The observation that null objects are ruled out in strong island contexts motivates Raposo’s (1986) proposal that null objects are variables. As such, they cannot be bound by a topic operator in the matrix sentence across a strong island boundary. Pronominal forms, instead, can establish coreference relations across islands. We will return to the different properties of pronouns and variables below. For the time being, it is enough for us that the reader keeps in mind the idea that the null object is a variable, which explains its limited distribution and its ungrammaticality in contexts like islands – a context in which pronouns cannot be bound. The main aspects to be retained from this sketchy description of the behavior of clitics and null objects in European Portuguese are: a) European Portuguese has pronominal clitics, like other Romance languages; b) European Portuguese clitics vary in syntactic function, gender, number and person; c) European Portuguese is a null object language; d) In certain environments, null objects freely alternate with clitics; e) In speci¿c environments, null objects are ruled out; f) The analysis of null objects as variables explains their sensitivity to strong island environments.

110

João Costa, Maria Lobo and Carolina Silva

Given this description, it should be clear now that a child exposed to Portuguese faces several challenges, since acquiring clitics in Portuguese involves, at least, detecting its form and functions, ¿nding out that the clitic can alternate with a null object, and discovering the correct distribution of null objects, partly by determining the nature of the empty category involved in null object constructions.

3.

Clitic production in the acquisition of European Portuguese

In a series of studies by Ken Wexler and associates (Wexler, Gavarró and Torrens, 2003; Tsakali and Wexler, 2003; Babyonyshev and Marin, 2005), it is argued that children omit clitics in early stages of acquisition, but omission is not universal. These authors argue that it correlates with the availability of past participle agreement in the language. For instance, French is a language with past participle agreement (10), and as such it is expected to have clitic omission: (10) a.

Les fenêtres… Je les ai repeintes. The windows… I themACC-FEM-PL have repainted-FEM-PL “The windows…. I repainted them.”

Spanish, on the other hand, does not have past participle agreement (11), and is not expected to have clitic omission: (11) a.

Las ventanas… las he repintado The windows… I themACC-FEM-PL have repainted-MASC-SG “The windows…. I repainted them.”

These authors’ proposal predicts that there should be no clitic omission in European Portuguese, since this language patterns like Spanish in what concerns past participle agreement. In a series of studies (Costa and Lobo 2006, Silva 2008, Carmona and Silva 2007, Costa, Lobo, Silva and Carmona 2008), we tested this prediction and found no evidence to support it, because children acquiring European Portuguese do omit clitics. However, two problems arise: ¿rst, the rates of omission were not comparable to those found in other languages. It was found that children acquiring European Portuguese omit more clitics than children acquiring other languages with omission, and that omission lasts longer than in other languages, since 5 year olds still omit clitics, whereas in other languages, omission rates drop

Which category replaces an omitted clitic? The case of European Portuguese

111

much earlier. Varlokosta et al. (in preparation), a comparative work on 16 languages, shows that omission rates are higher in early Portuguese than in other languages. Varlokosta et al. make the observation that children acquiring European Portuguese, at the age of 5, produce clitics in 20% of their utterances, whereas Italian and Catalan children, at the same age, and in the same experimental setting, produce 92,9% and 98% of clitics, respectively. The second problem is that the data do not receive a self-evident interpretation. Recall that European Portuguese has null objects. As such, when a child utters a sentence without a clitic, one does not know whether the child is omitting a clitic as in other languages, or producing a target null object construction. In order to ¿nd the source of the problem, we tested a wider range of clitics. In particular, we decided to test the production of clitics in contexts in which null objects are ruled out in order to determine whether the children omit in accordance with a target null object grammar. If so, their rates of omission should have been nearly null in those contexts in which the clitic is compulsory. In the remainder of this section, we summarize the results obtained for clitic production of different types. The results presented come from Silva (2008, 2010). Following the elicitation procedure ¿rst used in Schaeffer (1997), and ¿rst adapted for European Portuguese in Costa and Lobo (2006, 2007a), Silva (2008, 2010) tested 73 children aged between 3 and 6 years and 6 months old. The tests of clitic production were also applied to a control group composed of 15 adults. The following table gives details about the children and adults tested: Table 1. Participants information Children [3,0 – 4,0] [4,0 – 5,0] [5,0 – 6,0] [6,0 – 6,5]

Mean age 3 years and 7 months 4 years and 5 months 5 years and 6 months 6 years and 3 months

Total —— Adults Mean age [24,0 – 27,0] 25 years and 2 months

Female 11 16 9 5

Male 4 11 12 5

Total 15 27 21 10

41 Women 9

32 Men 6

73 Total 15

As just mentioned, the aim of Silva’s (2008) study was to test different types of clitics, in order to assess correlations with the null object construction. As such, the following types of clitics were tested:

112

João Costa, Maria Lobo and Carolina Silva

Table 2. Conditions tested in the experiment. Tests per type of clitic Accusative Dative ReÀexive Non-argumental Enclisis Proclisis Islands Enclisis Proclisis Islands Enclisis Proclisis Enclisis Proclisis 1st S 1st S 1st S 1st S 1st S 1st S 1st S 1st S 1st S 1st S st st st st st st st st st 1 P 1 P 1 P 1 P 1 P 1 P 1 P 1 P 1 P 1st P nd nd nd nd nd nd nd nd nd 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2nd S nd nd nd nd nd nd 2 P 2 P 2 P 2 P 2 P 2 P —— —— —— —— 3rd S, f 3rd S, f 3rd S, f 3rd S 3rd S 3rd S 3rd S 3rd S 3rd S 3rd S rd rd rd 3 S, m 3 S, m 3 S, m 3rd P, f 3rd P, f 3rd P, f 3rd P 3rd P 3rd P 3rd P 3rd P 3rd P 3rd P, m 3rd P, m 3rd P, m 3rd P

As can be seen in the table, the conditions include those contexts in which clitics and null objects alternate, and those in which only clitics are a valid option. Two elicitation procedures were used. For 1st and 2nd person clitics, a puppet interacted with the child and with some toys. For 3rd person clitics, the clitics were induced through the presentation of pictures. Both procedures had in common the relevant fact that the referent for the pronoun was made highly accessible. Below, we present an example for each of the procedures taken from Silva (2010):

Figure 1. Example of test item for eliciting 1st person clitics.

Which category replaces an omitted clitic? The case of European Portuguese

113

Figure 2. Example of test item for eliciting 3rd person clitics.

For a detailed presentation of the results for each and every condition, we refer the reader to Silva (2008, 2010). For the purposes of this paper, it is suf¿cient that we focus on the results obtained in the 3rd person accusative condition – the clearest case in which clitics and null objects are in free variation, comparing them with a summary of results in which clitics should be obligatorily produced. Let us start by considering the results for the 3rd person (singular and plural) accusative condition in non-compulsory contexts (enclisis and proclisis) summarized in table 3: Table 3. Results for 3rd person accusative clitics in non-compulsory contexts.

Clitic Null form DP Strong pronoun

3rd person ACCUSATIVE CLITICS [3,0 – 4,0] [4,0 – 5,0] [5,0 – 6,0] [6,0 – 6,5] Control group 23/240 99/432 146/336 74/160 229/240 9,58% 22,92% 43,45% 46,25% 95,42% 210/240 298/432 149/336 84/160 3/240 87,50% 68,98% 44,35% 52,50% 1,25% 6/ 240 33/432 41/336 2/160 8/240 2,50% 7,64% 12,20% 1,25% 3,33% 1/240 2/432 0/336 0/160 0/240 0,42% 0,46% 0% 0% 0%

114

João Costa, Maria Lobo and Carolina Silva

Two major conclusions emerge from these results. First, children’s production of clitics is different from what adults do (chi-square tests: p < 0.001 for all comparisons between each children’s group and the control group). There is an evident developmental effect from 3 to 5 years old (chi-square tests: p < 0.001). From 5 to 6, the development is much slower (chi-square test: p = 0,897). However, at age 6 the rate of production is still far from what is observed for adults1 (as mentioned above, chi-square test: p < 0.001). Second, as pointed out above, it is not the case that clitic omission ceases at around age 3, as argued for other languages. Recall, however, that these data do not allow us a clear interpretation of the facts as clitic omission, since the “null form” row may be read as a case for assuming that children produce target null objects. More revealing data come in the following table, containing a synthesis of the results for those contexts in which clitics are obligatory and null objects are ruled out. Table 4. Results for 3rd person accusative clitics in strong islands (compulsory contexts).

Clitic Null form DP Strong pronoun

3rd PERSON ACCUSATIVE CLITICS IN STRONG ISLANDS [3,0 – 4,0] [4,0 – 5,0] [5,0 – 6,0] [6,0 – 6,5] Control group 4/120 33/216 49/168 32/80 78/120 3,33% 15,28% 29,17% 40% 65% 77/120 76/216 27/168 20/80 0/120 64,17% 35,18% 16,07% 25% 0% 38/120 107/216 92/168 28/80 42/120 31,67% 49,54% 54,76% 35% 35% 1/120 0/216 0/168 0/80 0/120 0,83% 0% 0% 0% 0%

The results of these tables are interesting. First, they reveal that clitics are also omitted in contexts in which null objects are ruled out. However, they show that the rate of omission is lower in compulsory contexts than in non compulsory contexts: [3,0 – 4,0] – 87,50% vs. 64,17%; [4,0 – 5,0] – 68,98% vs. 35,18%; [5,0 – 6,0] – 44,35% vs. 16,07%; [6,0; 6,5] – 52,50% vs. 25 %. In Costa and Lobo (2006), the fact that children omit clitics, even in contexts in which null objects are ruled out, was taken as an indication that children’s clitic omission differs from adult null objects. However, this account fails to provide a plausible explanation for the different omission rates, and does not 1. The adults were, possibly, sensitive to the experiment since a higher production of null objects was expected in contexts where they freely vary with clitics.

Which category replaces an omitted clitic? The case of European Portuguese

115

make any prediction as to the nature and category of the empty element in the contexts in which clitics were not produced. In fact, in the absence of a clitic, several outcomes are possible: either the child produces a kind of truncated structure and no empty category is involved, or the child is ¿lling in the object position with some kind of empty category. The ¿rst hypothesis predicts that children treat structures without a complement as intransitive, whereas the second one predicts that children know that the relevant structures are transitive. In other words, under the second hypothesis, children know how to interpret null object constructions. If the second alternative is on the right track, one still has to know what type of null object is involved: a pronominal or a variable. Summarizing, the results obtained in the studies on children’s production of clitics in European Portuguese open up space for a certain degree of analytical indeterminacy. The options are: a) Children omit clitics in compulsory contexts, because they have clitic omission of the type found in other languages. b) Children omit clitics in compulsory contexts, because they know that European Portuguese is a null object language, but they overuse null objects. Some comprehension tests were run, in order to understand whether children acquiring European Portuguese display knowledge of the existence of the null object construction. If they do not, the second hypothesis can be immediately discarded. In the next section, we brieÀy present the results of the experiments carried out in Costa and Lobo (2008, 2010a), which try to provide answers to this issue.

4.

Null objects in comprehension

French children are known to omit clitics (cf. Grüter 2006 for a literature review). Grüter (2006) raised a question about French quite similar to the one we raised for European Portuguese: can it be that French children are producing a kind of null object construction? If so, as conjectured by Grüter (2006), this means that their grammar includes the possibility of generating transitive structures without an explicit element. As such, they are supposed to interpret a verb without a complement transitively. In order to test this prediction, Grüter (2006) developed a test in which children were asked to judge the truth value of verbs without a complement in transitive contexts. Children acquiring French rejected transitive interpretations for verbs that did not appear with any type of

116

João Costa, Maria Lobo and Carolina Silva

complement. This result can be interpreted as a consequence of the absence of a null object construction in their grammar. Costa and Lobo (2008, 2009) adapted Grüter’s test for European Portuguese, and assessed whether children could assign transitive interpretations to a verb without a complement in the following environments: a) Simple clauses; b) Strong islands; c) ReÀexive contexts. The purpose of testing environment a) is twofold: ¿rst, it allows for a clear comparison with the results obtained for French; second, it allows us to determine whether children acquiring European Portuguese have null objects. Environments b) and c) are those in which null objects are illegitimate in the adult grammar (out of these, only environment b) was tested for French, and no differences were found with respect to context a)). Testing the comprehension of verbs without a complement is interesting, since one can assess consistency between production and comprehension, and add a further argument to the idea that omission in European Portuguese is a case of overuse of the null object construction. In order to test the comprehension of null objects, Grüter’s test was adapted to European Portuguese. In this test, a selection of verbs that can be either intransitive or transitive are used. A transitive situation is presented with a sentence describing it (uttered by a puppet) containing only subject and verb. If the child’s grammar has no null objects, the child must judge such a description as false. If, on the other hand, the child’s grammar includes null objects, she will accept a sentence with no explicit internal argument, considering it to be true. Let us consider two examples of test items: a potentially ambiguous sentence is presented to the child in a context making one of the interpretations true or false. After the presentation of an image, the child is asked to judge whether a sentence uttered by the puppet is true or false. In the test items, the following transitive/intransitive alternation verbs were used: mergulhar ‘dive’ (‘x mergulha’ or ‘x mergulha y’), adormecer ‘fall asleep’/‘put to sleep’ (‘x adormece’ or ‘x adormece y’), acordar ‘wake up’ (‘x acorda’ or ‘x acorda y’), baloiçar ‘swing’ (‘x baloiça’ ou ‘x baloiça y’). As shown in the following examples, both uses are possible for all these verbs: (12) a.

O Rui mergulhou. the Rui dove ‘Rui dove.’

Which category replaces an omitted clitic? The case of European Portuguese

b.

(13) a.

b.

(14) a.

b.

(15) a.

b.

O Rui mergulhou o irmão the Rui dove the brother ‘Rui put his brother in the pool’

117

na piscina. in the pool

O Rui adormeceu. the Rui fell asleep ‘Rui fell asleep.’ O Rui adormeceu o bebé. the Rui fell asleep the baby ‘Rui put the baby to sleep.’ O Rui acordou. the Rui woke up ‘Rui woke up’ O Rui acordou o bebé. the Rui woke up the baby “Rui woke up the baby.” O Rui baloiçou na cadeira. the Rui swang in the chair ‘Rui swang in the chair.’ O Rui baloiçou o bebé na cadeira. the Rui rocked the baby in the chair ‘Rui rocked the baby in the chair.’

The test was preceded by a familiarization period with the task, with the puppet, with the images and with the verbs used in both variants (transitive and intransitive). In the period of adaptation to the experimental setup, the images were presented to the child and the four verbs were used in the transitive and in the intransitive constructions in order to ensure that the verbs, in their different constructions, were known to the child, and in order to make the interpretation of the drawings easier. In the following set of images, one can see the type of material used in the elicitation of judgements on null objects in reÀexive environments:

118

João Costa, Maria Lobo and Carolina Silva

Figure 3. Example of pictures used in the comprehension of null objects

For these images, the experimenter ¿rst introduces the characters, and then he describes the situation in the ¿rst drawing making the characters highly available in the discourse scale. Then, the puppet would utter a sentence like “He is washing” (without the reÀexive clitic), which is not a legitimate description for the image on the right. The results of this experiment, run on 19 children aged between 3;6 and 5;9, are summarized in the following ¿gure, in which the rates of adult-like responses are given2: 3;6-5;9 (n=19) 100% 80% 60% 40% 20% 0% Pron (V)

Pron (V)

Pron=Refl (F)

Pron=Refl (F)

Refl (V)

Refl (V)

Refl=Pron (F)

ON (V)

Refl=Pron (F)

ON (V)

ReflN (F)

ReflN (F)

DP (V)

DP (V)

TransRefl (F) TransRefl (F)

Figure 1. Rates of correct comprehension of pronouns, reÀexives and null arguments 2. Pron (V): Condition for pronouns with pronominal interpretation (True); Pron=ReÀ (F): Condition for pronouns with reÀexive interpretation (False);ReÀ (V): Condition for reÀexives with reÀexive interpretation (True); ReÀ=Pron (F): Condition for reÀexives with pronominal interpretation (False); ON (V): Condition with accusative null objects (True); ReÀN (F): Condition with null reÀexive (False); DP (V): Control condition with full DP (True); TransReÀ (F): Control condition with full DP and reÀexive interpretation (False).

Which category replaces an omitted clitic? The case of European Portuguese

119

The most important conditions to focus on in the graph are ON(V), in which we tested the comprehension of null objects in simple sentences, and REFLN(F), the condition in which null objects should be ruled out. As can be seen from the graph, children are successful in the mastery of null objects in simple transitive clauses. This is obvious, given the high success rate in the interpretation of transitive verbs without a complement as transitive. This contrasts with the behavior of children acquiring French described in Grüter (2006). In the present study, the comparison between ON(V) and REFLN(F) shows that children fail to understand that reÀexives are not good candidates for null objects. Note that no clear individual differences were found, and, as such, the difference between these two conditions is quite strong. These results are consistent with the results obtained in production tasks. Recall that children acquiring Portuguese fail to reject the ungrammatical structures that they produce. From these observations, several interesting conclusions can be drawn: a) First, it becomes obvious that omission is not a uniform phenomenon, since different speakers of different languages display different behaviors towards omission. In particular, in French, no acceptance was found in these contexts, very much different from European Portuguese. As such, it is legitimate to conjecture that the high rate of omission in this language is due to the overuse of null objects, whereas in French the inability to judge the same kind of stimulus transitively reÀects that no null object grammar is at play. b) Second, the hypothesis that the high rates of omission found in European Portuguese are due to an overuse of the null object construction is con¿rmed, since the contexts of illegitimate production and comprehension overlap. c) Third, the rate of illegitimate acceptance of null objects in reÀexive contexts is very similar to the rate of illegitimate acceptance of null objects in strong islands, found in Costa and Lobo (2008). This consistency in results again parallels what has been found in production. If the analysis we are proposing is on the right track, the combination of the results from production and comprehension converge in showing that children know that European Portuguese is a null object language, and overuse the null object option. If that is so, the differential behavior in clitic production, when this language is compared to others, can be easily explained: the rate of clitic omission and the ages at which children converge with the control group are different from what has been found for other languages, because clitic omission

120

João Costa, Maria Lobo and Carolina Silva

in this language is of a different nature. Unlike in other languages, clitic omission in European Portuguese is a kind of null object construction. Note that this is an interesting result, since it con¿rms the idea that much of the syntactic knowledge is acquired very early (Wexler 1998), and gives support to the claim that there is great continuity between child and adult grammar (Guasti 2002). However, note that the evidence referred to so far does not yet provide a crystal-clear answer regarding the nature of the empty category involved. One possibility raised in Silva (2010) is that the null category assumed by children is a pro, and not a variable. This would explain why there are more contexts in which the null object is allowed for the child than for the adult. What is at stake, then, is to determine the degree of continuity involved in the mastery of the null object construction. Do children posit the same empty category as adults? Is the null object a pro or a variable in child grammar? If the null object is pro, the degree of continuity is smaller than one might think. Still another option available in the target grammar is that children use a VPellipsis construction (as in example 16a). VP-ellipsis differs from null object in that it is allowed in islands (16b), with reÀexives (16c), and it requires verbal parallelism (16d) (Matos 1992)3: (16) a.

O Pedro tinha escrito o artigo e eu também tinha. the Pedro had written the article and I also had “Pedro had written the article, and I had too.”

b.

O Pedro escreveu o artigo, porque eu também escrevi. the Pedro wrote the article because I also wrote “Pedro wrote the article, because I also did.”

c.

O Pedro lavou-se e eu também lavei. Pedro washed SE and I also washed “Pedro washed himself, and I did too.”

d.

*O Pedro comprou o livro, porque eu roubei. the Pedro bought the book, because I stole

3. Santos (2006), unlike Matos (1992), rejects the idea that VP-ellipsis requires verbal identity. The fact that sentences like (16d) are ungrammatical is a standard argument to distinguish VP-ellipsis and null objects, since there is no verbal identity, and the ellipsis is illegitimate, because of the island context. If, on the contrary, a relaxation of the identity requirement were possible, this sentence should be ¿ne under a VP-ellipsis reading, since VP-ellipsis is allowed in island contexts. Since, in the experiment, we use cases similar to these ones, we will follow Matos in accepting the idea that VP-ellipsis requires identity.

Which category replaces an omitted clitic? The case of European Portuguese

121

The pilot study presented in the next section tries to shed some light on the nature of the null category involved in children’s performances.

5.

Null object: pro, variable or ellipsis?

As discussed in the previous section, there is good evidence for saying that clitic omission in child Portuguese is an overuse of the null object option, a construction independently available in the target grammar. If this analysis is on the right track, there is continuity between children’s knowledge and the properties of adult grammar. However, in order to determine the degree to which child language and adult grammar are alike, it is important to ¿nd out whether the representation assumed by the child for null objects matches the one of adults. In order to do so, Costa and Lobo (2010a) developed a test to assess whether children ascribe to null objects the interpretation expected if it is a variable. In this section, we present the test and explore some pilot results. This test is highly inspired in the ¿ndings by Miyagawa (2010), who described that pro and variables behave differently in anaphoric retrieval contexts. In particular, a null subject in an embedded environment cannot have a sloppy reading. This can be seen in the following example: (11) O Pedro disse que os pais iam a cavalo e o Rui disse que ’ iam a pé. the Pedro said that the parents went on horse and the Rui said that went on foot “Pedro said that his parents went on horse, and Rui said that his (parents) went by foot”. Possible reading: Rui said that Pedro’s parents went walking. Impossible reading: Rui said that his parents went walking. The only reading available for this sentence is the strict one, in which the null category retrieves the reference of the matrix antecedent. This contrasts with null objects: (12)

O Pedro abraça os pais e o Afonso beija ’. the Pedro hugs his parents and Afonso kisses

Sentence (12) is ambiguous, since the object can be co-referential with an antecedent from the ¿rst coordinate (in the case Afonso kisses Pedro’s parents), or with Afonso’s parents (in the reading in which Pedro hugs his parents,

122

João Costa, Maria Lobo and Carolina Silva

and Afonso kisses his parents). The ¿rst reading is generally known as strict, whereas the second one is known as sloppy. Recall from the previous section that VP-ellipsis is another option to be considered in the interpretation of children’s performances. As such, it is relevant to know how this construction behaves with respect to the strict/sloppy readings. As shown in (13), VP-ellipsis allows both strict and sloppy readings: (13) O Pedro abraça os pais e o Afonso também abraça. the Pedro hugs his parents and Afonso also hugs This sentence is ambiguous, since it can mean that Afonso hugs Pedro’s parents (strict reading) or that he hugs his own parents (sloppy reading). Given this, it becomes obvious that the sole difference between null objects in simple sentences (as in 12) and VP-ellipsis in the same context is the requirement of verbal parallelism. Since we know that pro only allows for strict readings (Miyagawa 2010) (recall the description of the sentence in (11)), if a child consistently rejects sloppy readings for null object, there will be good reasons to assume that he/ she does not know that the null object is a variable. If, on the other hand, the child knows that the null subject can only have strict readings, rejecting sloppy readings for it, and accepting them for null objects, we will have good evidence to postulate that children know the category of the null arguments they use. If the child gives non-adult answers for null subjects, accepting sloppy readings, however, we will have to assume that the interpretation of empty categories in child grammar is generally subject to maturation. In order to test this, Costa and Lobo (2010a) developed a truth value judgment task testing the following conditions: a) b) c) d) e) f)

Strict readings for null subjects (True); Sloppy readings for null subjects (False); Strict readings for null objects (True); Sloppy readings for null objects (True); Strict readings for VP ellipsis (True); Sloppy readings for VP ellipsis (True).

The test was designed in the following way: pictures of situations involving two children and their parents were presented to the participants. A puppet uttered a statement about the picture, which had to be judged true or false. The sentences were of the type illustrated in (11), (12) and (13). Consider, for instance, the following test item for condition a). The picture below was shown:

Which category replaces an omitted clitic? The case of European Portuguese

123

Figure 4. Example of picture used in the strict/sloppy readings comprehension task

The test sentence for this picture was: O Moreno disse que os pais estavam sentados e o Loiro disse que tinham chapéu. the dark haired kid said that the parents were sitting and the blond kid said that had hat. “The dark-haired kid said that his parents were sitting, and the blond kid said that they had a hat.” This sentence is true, because only the strict reading is available for subject pro in the adult grammar. Let us now see an example of a test item for the null object condition under the sloppy reading. First, a picture like the following is shown:

Figure 5. Example of picture used in the strict/sloppy readings comprehension task

124

João Costa, Maria Lobo and Carolina Silva

The test sentence for this picture was: O Louro abraça os pais e o Moreno beija. the blond kid hugs the parents and the dark haired kid kisses “The blond kind hugs his parents and the dark-haired kid kisses them.” This sentence is true, because the sloppy reading is available for null object in adult grammar. In what follows, we present the results of a pilot of this test (Costa and Lobo 2010a). In the pilot, twenty 5 year old typically developing monolingual children acquiring European Portuguese, aged between 5;0 and 5;10 participated (8 boys and 12 girls). The mean age was 5;5. Children were tested individually in a quiet room. No response-contingent stimulus was given, and children were only rewarded after completion of the whole task. Data were coded and transcribed by two experimenters. Additionally, the test was run on a control group consisting of 15 adults aged between 23 and 46 years old. The test contained 5 items for each of the conditions listed above (30 items for the sum of the whole conditions) and 17 ¿llers. As can be seen in the following table, the adult control group performed as expected, associating sloppy readings with VP-ellipsis and null object only, and rejecting such sloppy reading in null subject contexts: Table 5. Results of the control group in the strict/sloppy readings comprehension test Sloppy reading for null subject (False) Strict reading for null subject (True) Strict reading for null object (True) Sloppy reading for null object (True) Strict reading for VP ellipsis (True) Sloppy reading for VP ellipsis (True)

%target responses 96% (72/75) 100% (75/75) 100% (75/75) 100% (75/75) 98,6% (74/75) 98,6% (74/75)

In the following table, we present the general results per condition for children: Table 6. Results for children in the strict/sloppy readings comprehension test Sloppy reading for null subject (False) Strict reading for null subject (True) Strict reading for null object (True) Sloppy reading for null object (True) Strict reading for VP ellipsis (True) Sloppy reading for VP ellipsis (True)

%target responses 51% (51/100) 64% (64/100) 71% (71/100) 71% (71/100) 69% (69/100) 86% (86/100)

Which category replaces an omitted clitic? The case of European Portuguese

125

These preliminary results from the pilot study are very interesting. As is clear from the results in the table, there is no clear difference between conditions. In fact, children have a poor performance in all conditions, and do not reach ceiling performances in any of them (unlike what was found for adults, cf. Costa and Lobo 2010a). Crossing these results with those of previous experiments and the results of Santos (2006) for VP-ellipsis can help us understand this performance. First, one cannot argue that the bad results are a consequence of a lack of knowledge on null object, null subject or VP-ellipsis. Independent research reveals that children acquiring European Portuguese know the grammar of null objects (cf. the studies mentioned in the previous sections). The same holds for null subjects, since they produce sentences with null subjects from early on (Gonçalves 2005). Finally, Santos (2006) convincingly shows that children know VP-ellipsis, since they produce it in verbal answers to yes-no questions, and are able to retrieve the antecedent of an elided VP in simple contexts. In short, children’s dif¿culties cannot be attributed to the inexistence of each of these constructions in their grammar. We contend that these data show that children have problems interpreting the null category involved in these constructions. In other words, they have null subjects, null objects, and VP-ellipsis, but fail to assign them an adult interpretation, and are not able to distinguish between strict and sloppy readings. If we go back to our original question – what is the nature of the category involved in children’s object omissions? – we are now closer to an answer. Three possibilities emerge: i) children use pro in object position, but do not know its interpretation; ii) children use a variable in object position, but do not know its interpretation; iii) children use a VP-ellipsis, and disrespect the requirement of verbal parallelism. Option iii) can be easily discarded on the basis of Santos’ (2006) results, since she showed that children’s production of VP-ellipsis in verbal answers to yesno questions provides positive evidence for compliance with the parallelism requirement. However, as shown by these results, their interpretation of VPellipsis is not target, since there are obvious differences between children’s performance and the ceiling results of adults. Options i) and ii) can, actually, be merged, since, on the basis of the evidence we have, they cancel each other out. pro and variables can be distinguished on the basis of their distribution and interpretation. The data from previous experiments show that children annul the differences with respect to distribution, since they license an alleged variable in islands. The preliminary data from

126

João Costa, Maria Lobo and Carolina Silva

Costa and Lobo (2010a) shows that they also annul the difference with respect to interpretation, since they assign pro readings typical of variables (the sloppy one), and they sometimes reject strict reading for null objects. Given this, and on the basis of the data available so far, it is legitimate to state that the difference between pro and variables is irrelevant at this stage of language development. Considering all this, one can hypothesize that for children to reach an adult knowledge of the types of null categories involved in subject and object position, they have to acquire the difference in interpretation that distinguish them, which arguably involves acquiring further properties distinguishing operatorvariable relations from pronominal binding, a matter we leave for further research.

6.

Conclusions

In this paper, we aimed at presenting a summary of previous results obtained in the study of clitic production by European Portuguese children. It was shown that the results of the production tasks indicated that clitic omission in European Portuguese is different from what occurs for other languages. As mentioned in previous studies, the results obtained for European Portuguese show that clitic omission is not a uniform phenomenon cross-linguistically, and, because of this, no clear conclusions may be drawn from the European Portuguese data in order to support or counter the views of Wexler, Gavarró and Torrens (2003) on omission. We also reported that the results from comprehension studies provide good evidence to say that null objects are available early, and known to the children. This implicit knowledge of null objects provides further evidence for a different treatment of clitic omission, and is a good argument in favor of continuity in language acquisition. In order to determine the degree of continuity, we tested whether children master the difference between pro, variable and VP-ellipsis. Preliminary results from a pilot experiment indicate that children do not yet know the interpretive option associated with each one of these null constructions. As a consequence, we did not yet obtain crystal-clear evidence for determining the nature of the category involved in clitic omission contexts, but we are now able to state that the difference between pro and variable is not a relevant distinctive factor for children. This is promising as an explanation for the overuse of null object. By hypothesis, and comparing these results with those obtained for VP-ellipsis (for which there is independent evidence for the early mastery of its syntax), we

Which category replaces an omitted clitic? The case of European Portuguese

127

can suppose that children know the syntax of null subject and null object, and the only problem they have is in the assignment of the correct interpretation. Children’s task for null objects will be to narrow down the range of possible interpretations, excluding, for instance, reÀexive and 1st and 2nd person options.

References Babyonyshev, Maria and Stefania Marin 2005 The Acquisition of Object Clitic Constructions in Romanian. In Gess, Randall S. and Edward J. Rubin (eds.), Theoretical and Experimental Approaches to Romance Linguistics; 21–40. Carmona, Jaqueline and Carolina Silva 2007 A aquisição de clíticos dativos em PE: teste piloto. In A. Coutinho and M. Lobo (eds.) Textos Seleccionados. XXII Encontro Nacional da APL. APL/Colibri, Lisboa; 199–210. Castilla, A., A. T. Pérez-Leroux and A. Eriks-Brophy 2008 Omission in Early Spanish Clitics. In A. Gavarrò e M. J. Freitas, (eds.) Language Acquisition and Development. Proceedings of GALA2007. Newcastle: Cambridge Scholars Publishing; 112–122. Costa, João and Maria Lobo 2006 A aquisição de clíticos em PE: omissão de clíticos ou objecto nulo?. In XXI Encontro Nacional da Associação Portuguesa de Linguística. Textos seleccionados. Lisboa: APL; 285–293. Costa, João and Maria Lobo 2007a Clitic omission, null objects or both in the acquisition of European Portuguese?. In S. Baauw, F. Drijkonongen and M. Pinto, eds. Romance Languages and Linguistic Theory 2005. Amsterdam: John Benjamins; 59–72. Costa, João and Maria Lobo 2007b Complexidade e omissão de clíticos: o caso dos reÀexos. In A. Coutinho and M. Lobo (eds.) XXII Encontro Nacional da APL. Lisboa: APL/Colibri; 303–313. Costa, João and Maria Lobo 2008 Omissão de clíticos na aquisição do português europeu: dados da compreensão. In Textos Seleccionados, XXIII Encontro Nacional da Associação Portuguesa de Linguística, APL/Colibri, Lisboa; 143–156. Costa, João and Maria Lobo 2009 Clitic Omission in the Acquisition of European Portuguese: Data from comprehension. In A. Pires and J. Rothman eds. Minimalist Inquiries into Child and Adult Language Acquisition: Case Studies across Portuguese, Berlin/New York: Mouton de Gruyter; 63–84.

128

João Costa, Maria Lobo and Carolina Silva

Costa, João and Maria Lobo 2010a Objecto nulo na aquisição do PE: pro ou variável? Paper presented at 26º Encontro Nacional da Associação Portuguesa de Linguística. Costa, João and Maria Lobo 2010b Clitic omission is null object: evidence from comprehension. In J. Costa, A. Castro, M. Lobo and F. Pratas, orgs. Proceedings of GALA2009. Newcastle: Cambridge Scholars Publishing. Costa, João, Maria Lobo, Jaqueline Carmona and Carolina Silva 2008 Clitic omission in European Portuguese: correlation with null objects? In A. Gavarró and M. J. Freitas (orgs). Proceedings of Generative Approaches to Language Acquisition 2007, Newcastle: Cambridge Scholars Publishing; 133–143. Costa, João, Maria Lobo and Carolina Silva 2009 Null objects and early pragmatics in the acquisition of European Portuguese. Probus 21; 143–162. Duarte, Inês and Gabriela Matos 2000 Romance Clitics and the Minimalist Program. In Costa, J. (org.) Portuguese Syntax. New Comparative Studies. Oxford: Oxford University Press; 116–142. Gonçalves, Fernanda 2005 Riqueza morfológica e aquisição da sintaxe em português europeu e brasileiro, PhD dissertation, Univ. Évora. Grüter, T. 2006 Object clitics and null objects in the acquisition of French. PhD Dissertation, McGill University, Montreal. Guasti, Maria-Teresa 2002 Language Acquisition. The growth of grammar. Cambridge, Mass.: MIT Press. Jakubowicz, Celia, L. Nash, C. Rigaud and C. Gérard. 1998 Determiners and Clitic pronouns in French speaking children with SLI. Language Acquisition 7.2; 113–160. Matos, Gabriela 1992 Construções de elipse do predicado em português – SV nulo e despojamento. PhD dissertation, Universidade de Lisboa Miyagawa, Shigeru 2010 Primacy of Person Agreement: Revisiting Jaeggli and Sa¿r’s Morphological Uniformity for Pro-drop. Talk given at MIT, ms. Raposo, E. P. 1986 On the Null Object Construction in European Portuguese. In Studies in Romance Linguistics. orgs. Jaeggli e Silva-Corvalán. Dordrecht: Foris; 373–390.

Which category replaces an omitted clitic? The case of European Portuguese

129

Santos, Ana Lúcia 2006 Minimal Answers. Ellipsis, syntax and discourse in the acquisition of European Portuguese. PhD dissertation. Univ. de Lisboa. Schaeffer, J. 1997 Direct object scrambling in Dutch and Italian child language, UCLA Dissertations in Linguistics, 17. Silva, Carolina 2008 Assimetrias na aquisição de clíticos diferenciados em português europeu. MA dissertation. Faculdade de Ciências Sociais e Humanas, Universidade Nova de Lisboa. Silva, Carolina 2010 Asymmetries in the acquisition of different types of clitics in European Portuguese. In V. Torrens, L. Escobar, A. Gavarró, J. Gutiérrez (eds.) Movement and Clitics: Adult and Child Grammar. Newcastle, Cambridge Scholars Publishing; 361–388. Tsakali, Vina and Kenneth Wexler 2003 Why children omit clitics in some languages but not in others: new evidence from Greek. Paper presented at Generative Approaches to Language Acquisition 2003, Utrecht University, Utrecht, The Netherlands, ms. Varlokosta, Spyridoula, Adriana Belletti, Maria Teresa Guasti, João Costa, Kleantes Grohmann and Laurie Tuller. in preparation A Cross-Linguistic Study of the Acquisition of Clitic and Pronoun Production, ms. Wexler, Kenneth 1998 Very early parameter setting and the unique checking constraint: A new explanation of the optional in¿nitive stage. In Lingua 106; 23–79. Wexler, Kenneth, Anna Gavarró and Vicent Torrens 2003 Feature checking and object clitic omission in child Catalan and Spanish. In R. Bok-Bennema, B. Hollebrandse, B. Kampers-Manhe e P. Sleeman (eds.) Romance Languages and Linguistic Theory. Selected Papers from Going Romance 2002. Amsterdam, John Benjamins.

(A)symmetries in the production of object clitics by Romanian children Martine Coene and Larisa Avram*

1.

Introduction

In the literature dealing with object clitics one can identify two main research options. Some studies focus on what pronominal clitics have in common – de¿ciency and distribution – and try to provide a unifying analysis (see, for example, Cardinaletti and Starke 1999). Others focus on the different properties evinced by the various types of object clitics. Besides the uncontroversial distinction between reÀexive and non-reÀexive clitics, Accusative and Dative object clitics have also been argued to have different status (see, for example, Sportiche 1996). Within the class of Accusative clitics, Uriagereka (1995) and Kayne (2000) make a distinction between 3rd person and 1st/2nd person Accusative clitics (ACs). According to Kayne, the former are Determiner-clitics. Uriagereka (1995) argues that 3rd person clitics occur in D, on a par with articles, and take a null complement, whereas 1st/2nd person clitics are strong phrasal clitics which project as DPs. Recently, the issue of the developmental path of 1st/2nd person vs. 3rd person ACs has also been addressed (Avram and Coene 2007, Coene and Avram 2011, Silva 2010). Longitudinal data from child Romanian reveal a higher omission rate for 3rd person than for 1st/2nd person ACs. The goal of the present paper is to investigate the early production of reÀexive and non-reÀexive object clitics in Romanian with a view to getting a better understanding of the role of the person feature in the acquisition of clitics. The central question targets the 1st/2nd vs. 3 person asymmetry in the early acquisition of different types of clitics: do we ¿nd the same person asymmetry in the acquisition of all types of object clitics? If we do, it means that this is a genuine person asymmetry. In terms of learnability, the hypothesis would be that the category of person plays a central part in the acquisition of pronominal clitics. But if this developmental person asymmetry is attested only with some types of clitics, one would have to identify and explain its cause in relation to features other * Work on this paper was supported by CNCSIS – UEFISCU, project grant PN II IDEI 1979/2008 to Larisa Avram.

132

Martine Coene and Larisa Avram

than person. In terms of learnability, this will raise the question of how the child copes with the particular identi¿ed cause of the asymmetry. Ideally, whether the same person asymmetry can/cannot be found across the various types of clitics might offer a hint with respect to the way in which the person feature is encoded in the syntax of clitics as well as with respect to the way in which it may interfere with other syntactic properties of this type of pronominal objects. The paper is organized as follows: Section 2 brieÀy lists and illustrates the obligatory clitic contexts in Romanian. The data and the method are presented in Section 3. Section 4 contains the results: (i) the early production of nonreÀexive Accusative clitics; (ii) the production of early non-reÀexive Dative clitics; (iii) the production of reÀexive Accusative clitics. Section 5 addresses the 1st/2nd vs. 3rd person distinction across the various types of clitics in more depth and offers an account of the attested asymmetries. The conclusions are summarized in Section 6.

2.

Obligatory clitic contexts in Romanian

2.1.

Preliminary remarks

One important step in the study of the acquisition of clitics is the identi¿cation of those contexts in which the use of the clitic is obligatory in the target grammar. This is even more important in a language like Romanian, where the use of Accusative and Dative object clitics is subject to optionality in many contexts. Moreover, there are important differences between previous studies which described the use of clitics and the actual use of clitics in the spoken language. 2.2.

Obligatory contexts for Accusative object clitics

In Romanian, third person ACs are obligatory in: (i) left dislocation structures with D-linked direct objects (1); (ii) direct object relative clauses (both restrictive and non-restrictive) introduced by the relative pronoun care ‘who, which’ (2); (iii) wh-questions with care ‘which’ (3); (iii) sentences in which the postverbal complement position is phonetically empty (the so-called single clitic constructions) and the clitic has a (salient) discourse antecedent (4). ACs are also obligatory in clitic doubling constructions, when the double is a de¿nite pronoun (5), but optional in any other clitic doubling construction (6). (1)

a.

Cartea # am dat *(-o). book.the # have given clitic 3rd fem sg ACC ‘The book, I have given away.’

(A)symmetries in the production of object clitics by Romanian children

(2)

(3)

b.

o carte # am dat *(-o). a book # have given clitic 3rd ‘A book, I have given (away). ’

c.

pe Ion # *(l-) pe Ion # clitic 3rd masc sg ACC ‘Ion, I have seen.’

a.

Mărul pe care *(l-) am mîncat. apple.the pe which clitic 3rd sg masc ACC have eaten ‘The apple which I have eaten.’

b.

Fata, pe care *(o) útia ... girl. the pe whom clitic 3rd sg masc ACC knew ‘The girl, whom he knew, ....’

Pe care

*(l-)

133

[under speci¿c reading] fem sg ACC am văzut. have seen

ai

ales?

pe which/whom clitic 3 sg masc ACC have chosen ‘Which one have you chosen?’ rd

(4)

A:

Ce-ai făcut cu mărul? ‘What have you done to the apple?’

B:

*(L-) clitic 3rd masc sg ACC ‘I have eaten it.’

am mîncat. have eaten

(5)

*(O) avem aici pe ea/aceasta. clitic 3rd fem sg ACC have here pe her/this fem sg ‘We’ve got her/this one here.’

(6)

a.

(O) avem aici pe Ruxi. (clitic 3rd fem sg ACC) have here pe Ruxi. ‘We’ve got Ruxi here.’

b.

(L-) a văzut pe copil. (clitic 3rd fem sg ACC) has seen pe child ‘(S)he has seen the child.’

First and second person ACs are always obligatory, since their antecedent is always a pronoun.

134

Martine Coene and Larisa Avram

2.3.

Obligatory contexts for Dative object clitics

Third person Dative object clitics (DCs) are obligatory: (i) when the indirect object has been fronted to sentence initial position (7); (ii) in indirect object relative clauses (both restrictive and non-restrictive) introduced by the relative pronoun care ‘who, which’ (8); (iii) in sentences in which the post-verbal complement position is phonetically empty and the null indirect object has a (salient) antecedent in the preceding discourse (9). In the so-called clitic doubling constructions, the clitic is obligatory when the double is a de¿nite pronoun (10). The clitic is optional in any other clitic-doubling construction in which the double is in post-verbal position (11) (see also Rădulescu Sala 2008) as well as in wh-questions with cui ‘cineDAT’ (12): (7)

Lui Ion *(i-) am dat cartea. rd to Ion clitic 3 sg DAT have given book-the ‘I have given the book to Ion.’

(8)

Omul căruia *(i-) am dat man-the who DAT clitic 3rd sg DAT have given ‘The man to whom I gave the book has left.’

(9)

A:

L-ai văzut pe Ion? ‘Have you seen Ion?’

B:

Da; úi *(i-) am dat cartea. yes and clitic 3rd sg DAT have given book-the ‘Yes, I have; and I gave him the book.’

cartea book-the

(10) *(I-) am dat cartea lui/ acestuia. clitic 3rd sg DAT have given book-the him DAT/ this DAT ‘I gave the book to him/to this one.’ (11) (I-) am dat clitic 3rd sg DAT have given ‘I gave the book to Ion.’

cartea lui Ion. book-the to Ion

(12) Cui (i-) ai dat cartea? rd who DAT clitic 3 sg DAT have given book-the ‘To whom have you given the book?’

First and second person DCs are never subject to optionality.

a plecat. has left

(A)symmetries in the production of object clitics by Romanian children

3.

Data and method

3.1.

Data

135

The present study relies on data provided by three longitudinal corpora of child Romanian, consisting of weekly 60 minute audio recordings of natural unstructured conversations of monolingual Romanian children: a girl (B.1;3–3;2) and two boys (A. 1;9–3;6 and I. 2;0–3;5), transcribed in Childes format (MacWhinney 2000). For the present analysis, we examined one recording per month for the period 2;0–2;11 for child B., 1;9–2;11 for child A., and 2;0–2;11 for child I. In each case, the ¿rst ¿le examined corresponds to the one in which the ¿rst clitic was attested. The data are summarized in Table 1 below: Table 1. Longitudinal data Child A. B. I.1

3.2.

Age 1;9–2;11 2;0–2;11 2;0–2;11

MLU 1.514–3.174 1.392–2.790 1.407–3.720

Nr of 60’ recordings 15 12 8

Method

The method used for coding and counting is similar to the one used for the analysis of ACs in Avram and Coene (2009). This involved a detailed examination of each transcript in order to identify all the obligatory clitic contexts. We investigated the production of Accusative and Dative non-reÀexive clitics by looking at occurrences and omissions only in obligatory contexts. We distinguished between 3rd person and 1st/2nd person object clitics. In the analysis of the acquisition data, we considered clitics to be omitted only when they were not used in an obligatory clitic context2. In optional clitic contexts, omission of 1. The I. corpus was recorded and transcribed by Ioana Stoicescu, University of Bucharest, whom we thank for generously sharing her data with us. 2. One reviewer recommends that the data should also be analyzed according to the type of construction in which the clitic is attested: with an overt clause-mate antecedent, as in clitic left dislocation and clitic doubling structures, or with a null clause-mate antecedent, as in single clitic constructions. Though this might indeed open the path to important ¿ndings in relation to the potential dif¿culties according to type of clitic construction, the goal of the present study is to investigate a potential person asymmetry in the acquisition of clitics, irrespective of the type of structure in which the clitic is used. Besides, a uniform analysis of ACs might be desirable both from a descriptive (see also Sportiche 1996, Del¿tto 2002) and from

136

Martine Coene and Larisa Avram

the clitic was considered grammatical, even when adults used the clitic in an identical context in the same conversational turn3. For reÀexive clitics4 (RCs), which are always obligatory, we counted only the Accusative forms. The attested number of reÀexive Dative clitics was much too small to be taken into account. No distinction was made between varian explanatory point of view. In terms of description, a unifying analysis would have the advantage of identifying what clitics have in common, irrespective of the type of structure in which they occur. From the point of view of acquisition, identifying the “obligatory clitic context” might lead to the identi¿cation of what the child has to “see” in the input in order to be able to acquire ACs. But we took the reviewer’s suggestion into account and counted the clitics in structures with an overt clause-mate associate separately. The analysis is, unfortunately, less relevant for the person asymmetry under discussion, due to the low number of ACs attested in such structures. The attested ACs in the three corpora appear mainly in single clitic constructions. For example, in the B. corpus, in the ¿les analysed for the present study, only 11.9% of the attested ACs are in clitic left dislocation and clitic doubling structures; as expected, they are mostly 3rd person clitics. This seems to weaken the strength of the analysis from the point of view of the person asymmetry. These data have been used in the qualitative analysis (see section 4.3). 3. Because of this methodology, the data reported in this paper for the A. and the B. corpus may differ from what was presented in previous studies (Avram and Coene 2006, 2007). 4. One reviewer asks why accusative reÀexives and non-reÀexives have been treated as different classes. Besides the standard differences between these two types of pronominals, such as binding, for example, one important difference is related to optionality. In Romanian, only 3rd person non-reÀexive ACs are optional; reÀexive clitics are always obligatory. Optionality has been argued to be a potential important factor in acquisition. Also, Romanian 3rd person accusative reÀexives are, unlike their non-reÀexive counterpart, neutral with respect to number (behaving in this respect as well as reÀexives in other Romance languages). If one takes morphological make-up to have consequences for syntactic behavior, this difference might turn out to be relevant for the developmental path of clitics. The same reviewer also suggests reÀexives in unaccusative structures should be excluded from the analysis because they carry aspectual features, unlike the other reÀexives. However, treating reÀexives as aspectual markers or identifying different classes of reÀexive clitics on the basis of aspectual features is not uncontroversial. Moreover, for the purpose of the present study, the relevant property, which is shared by all the reÀexives which we took into account in the analysis, is the nature of the relation between the clitic and the subject of the sentence. Zesiger et al. (2010), for example, propose that reÀexives differ from non-reÀexives in that they do not give rise to crossing chains, as in non-reÀexive constructions, where the object chain crosses the subject chain. ReÀexive clitic constructions contain nested chains of subjects and objects.

(A)symmetries in the production of object clitics by Romanian children

137

ous types of reÀexive clitics (reÀexive/reciprocal, passive, unaccusative/inchoative) because the number of some types of reÀexives was either absent in some corpora or too small to allow a comparative analysis. We distinguished between 1st/2nd person vs. 3rd person Accusative reÀexives.

4.

Results

4.1.

Non-reÀexive clitics

4.1.1. Accusative clitics 4.1.1.1. The data The data reveal that the ¿rst ACs are attested early in the three corpora. In the B. corpus, the ¿rst ACs emerge as soon as the child has entered the two word stage. The ¿rst clitic, placed in post-verbal position, is attested at 1;10: (13) apa … pus-o water-the put clitic 3rd fem sg ACC [B. 1;10] ‘I have put the water.’ An examination of the next recording sessions reveals, however, that no other AC is attested until 2;0. This is why we only analyzed the data beginning with this age, which we take to be the age of the ¿rst clear instance of repeated use (as de¿ned in Snyder 2007:77), i.e. the age at which a clitic has been used at least twice, in order to exclude incidental use (see also Stromswold 1996). ACs are correctly placed from the earliest stage5. However, random omissions are attested in the examined ¿les until 2;11 (14). The data are summarized in Figure 1. (14) a.

b.

Ăsta trebuie să *(-l) apăr. this must să SUBJ (clitic 3rd sg masc ACC) defend ‘I must defend this one.’ [B. 2;7] Nu *(mă) vezi no clitic 1st sg ACC see 2nd sg

pe mine aici? pe me here

‘Don’t you see me here?’ [B; 2;10]

5. During the period 2;0,03–2;0,21 the only clitic used is the feminine clitic o ‘her’, placed in post-verbal position, irrespective of the gender feature of the antecedent. It is at 2;1 that ACs other than o ‘her’ are attested.

138

Martine Coene and Larisa Avram

Figure 1. Production of ACs in the B. corpus

In the A. corpus ACs are found in the ¿rst recording session, i.e. as early as 1;9. Just as in the B. corpus, the ¿rst attested AC is the feminine o ‘her’ in postverbal position: (15) Adult: ce-ai făcut cu ligheanul? ‘What have you done to the bowl?’ Child: *spart-o. has broken clitic 3rd sg fem ACC [ = bowl MASC] ‘I have broken it.’ [A. 1;9] As already pointed out for corpus B., early emergence is followed by a stage when ACs are used in a non-adult like way. They are randomly omitted until 2;8: (16) Adult: úi ce-a făcut moúu(l) cu cocoúu(l)? ‘And what did the old man do to the rooster?’ Child: *(l-) a bătut. clitic 3rd masc sg ACC has beaten ‘He has beaten (it).’ [A. 2;0] (17) aia , am călcat *(-o). that one have stamped clitic 3rd fem sg ACC ‘I have stamped on that one.’ [A. 2;8]

(A)symmetries in the production of object clitics by Romanian children

139

The data are summarized in Figure 2:

Figure 2. Production of ACs in the A. corpus

As can be seen in Figure 3 below, ACs are attested at 2;0 (MLU 1.407) in corpus I. However, the only clitic attested (a 1st person sg clitic) occurs three times with the same verb in what can be a formulaic construction (lasă-mă ‘let me’). This is why we take 2;1 as the age when the ¿rst freely used ACs are attested. But random omissions are found until 2;11, the last ¿le examined: (18) Adult: Dar ieri ai văzut-o? ‘But did you see her yesterday?’ Child: Da. Da’ din maúină pînă la Colentina n-am văzut *(-o). yes but from car to Colentina not have seen clitic

‘Yes, I did. But I could not see (her) from inside the car until we reached Colentina.’ [I. 2;11]

Figure 3. Production of ACs in the I. corpus

140

Martine Coene and Larisa Avram

As can be seen in Figure 4 below, the omission rate of ACs in the three corpora is similar6:

Figure 4. Omission of ACs in the A, the B and the I. corpus

No clitic placement errors have been found in any of the corpora. At ¿rst sight, the analysis of the data indicates that 3rd person ACs emerge slightly earlier than 1st/2nd person ACs. In the B. corpus, for example, the ¿rst 3rd person AC (the feminine clitic o) is attested at 1;10 (MLU 1.091), in post-verbal position, and the ¿rst repeated use (FRU, Snyder 2007) is attested at 2;0. First/ second person ACs are ¿rst attested in the formulaic te rog (‘please-Acc clitic 2nd sg) at 1;11; no other 1st/2nd person AC is attested until 2;1 (MLU 1.734). The data in the A. corpus also indicate that 3rd person ACs emerge earlier than 1st/2nd person ACs. Whereas 3rd person ACs (the feminine clitic o) are attested as early as the ¿rst recording session, at 1;9 (MLU 1.392), the ¿rst 1st/2nd person ACs are attested at 2;4 (MLU 2.136), actually one month after the ¿rst pre-verbal 3rd person AC. Because of the small number of tokens it is impossible to safely assert whether this is an instance of ordered acquisition or whether the two types of ACs emerge at the same age. However, one can notice that both A. and B. go through a very short stage (lasting from age 1;10 to age 2;1 in corpus B. and from age 1;9 to age 2;2 in corpus A.) when the only AC used is the feminine singular o ‘her’, even when the antecedent has masculine (or neuter, i.e. masculine in the singular) gender, and that it is placed exclusively in post-verbal position (see 15 above).7 Interestingly, in both corpora the emergence of 1st/2nd 6. At the time when we analyzed the data, we had no access to the recordings of child I. for the period 2;3–2;8. 7. There is no similar stage attested in the I. corpus.

(A)symmetries in the production of object clitics by Romanian children

141

person ACs coincides with the emergence of 3rd person ACs in pre-V position. This may indicate that the early o clitics are used as default forms and that actually all ACs, irrespective of their Person feature, emerge at the same time. The conclusion is reinforced by the data coming from the I. corpus, where the ¿rst repeated use of 1st/2nd and 3rd person ACs is attested in the same ¿le, at 2;1. In terms of emergence, then, there does not seem to be any Person asymmetry. But the investigation of the omission rate of ACs in obligatory contexts indicates an asymmetry between 1st/2nd person and 3rd person ACs (see also Avram and Coene 2010, Coene and Avram 2011). In the B. corpus, the omission rate of 3rd person ACs is signi¿cantly higher than the omission rate of 1st/2nd person ACs (Wilcoxon Signed Rank test, p = .047). After 2;8, ACs are used almost adultlike, but the few omissions which are attested after that age are exclusively 3rd person ACs. The data are summarized in Figure 5 below:

Figure 5. Production of 1st/2nd vs. 3rd person ACs in the B. corpus

Similarly, in the A. corpus, there is almost no omission of 1st/2nd person ACs, apart from an incidental increase in omission rate at 2;5. Third person ACs, on the other hand, in spite of very early emergence, continue to be omitted (even though at a low rate) when the omission rate is 0% for 1st/2nd person ACs. The data are summarized in Figure 6. In the I. corpus the analysis of the used/omitted ACs reveals the same person asymmetry. The omission rate of 1st/2nd person ACs is lower than the omission rate of 3rd person ACs (Wilcoxon Signed Ranks test, p = .018). The data are summarized in Figure 7:

142

Martine Coene and Larisa Avram

Figure 6. Production of 1st/2nd vs. 3rd person ACs in the A. corpus

Figure 7. Production of 1st/2nd vs. 3rd person ACs in the I. corpus

There is a signi¿cant 1st/2nd vs. 3rd person asymmetry in the three corpora. Gender and number agreement errors are attested, but they are extremely rare: (19) a.

b.

cînd termin eu Ġi*o when ¿nish I clitic 2nd sg DAT clitic 3rd fem sg ACC dau ăstea. give these 3rd fem pl ‘When I have ¿nished I’ll give you these.’ [B. 2;6] *îi

bag úi pe fetiĠe.

clitic 3rd masc pl ACC put and pe girls DIM fem pl ‘I am putting the little girls inside as well.’ [I. 2;8]

(A)symmetries in the production of object clitics by Romanian children

143

There are no agreement errors with 1st or 2nd person ACs. Actually, plural forms are attested only for the 3rd person8. It is worth pointing out that the data reveal an asymmetry between the omission rate of 3rd person singular and 3rd person plural ACs (Wilcoxon Signed Ranks, B. p = .018, A. p = .036, I. p = .068). However, the production rate of 3rd person plural is higher in corpora B. and I., whereas A. uses more 3rd person singular clitics. The data are summarized in Table 2 and Figure 8: Table 2. Production of 3rd person sg vs. 3rd person pl ACs in the three corpora Child B. I. A.

Total contexts 3rd sg

Total 3rd sg clitics production

Total contexts 3rd pl

Total 3rd pl clitics production

151 93 269

78.14% (n = 118) 83.87% (n = 78) 60.96% (n = 164)

32 19 28

90.62% (n = 29) 94.73% (n = 18) 39.28% (n = 11)

Figure 8. Production of 3rd person sg vs. 3rd person pl ACs in the three corpora

8.

There is one single 1st person plural AC found in the examined ¿les in the B. corpus. This is why the data only compare the use of 1st/2nd AC vs. 3rd person AC in the singular. However, this does not mean that the plural form with 1st/2nd person is absent from the child’s grammar; plural forms other than 3rd person are attested in RC contexts, and they are homophonous with the Accusative non-reÀexive form.

144

Martine Coene and Larisa Avram

4.1.1.2. Preliminary summary The data available in the three corpora reveal that in most respects the acquisition of ACs in Romanian is similar to what has been found for other languages. They emerge very early (before age 2;0 and at an MLU < 2) and are never misplaced. This is similar to what has been reported in longitudinal studies of several other languages, among which Spanish (Domínguez 2003) and Greek (Marinis 2000). In two corpora, during a very short stage, only the feminine clitic o ‘her’, placed exclusively in post-verbal position, is attested. The fact that only the post-verbal clitic o ‘her’ is used during this early stage may be taken as evidence for one of the following two claims: (i) both children prefer Merge over Move, leaving the clitics in situ, or (ii) the feminine clitic o is used as a default form during this early stage. This clitic is exceptional among the other Romanian ACs in that it is the only one which can be placed either in preor in post-verbal position in ¿nite constructions: (20) a.

o vede clitic 3rd fem sg ACC sees ‘He/she sees her.’

b.

a văzut -o has seen clitic 3rd fem sg ACC ‘He/she has seen her.’

c.

ar vedea -o aux see clitic 3rd fem sg ACC ‘He/she would see her.’

In terms of article-clitic homonymy, o is the only AC which is homophonous with the inde¿nite article, not with the de¿nite, as is the case with the other 3rd person ACs. It also represents the default clitic; it occurs in frozen constructions where it does not always have a referential value: a o úterge ‘to run away’, a o lua pe coajă ‘to get spanked’, etc. It is also the clitic which can have a clausal antecedent. The exclusive use of o in post-verbal position during this early stage might indicate that this has to be taken as a language speci¿c phenomenon, rooted in the properties of the feminine clitic o. The early clitic o might be a default clitic, used in the process of paradigm building. A similar phenomenon is discussed in Domínguez (2003). She observes that the Spanish child Maria goes through a stage when one clitic is used as a “substitute”. The clitic which Maria uses during the early stages is the masculine lo. Zesiger, Zesiger, Arabatzi, Baranzini, Cronel-Ohayon,Franck, Frauenfelder, Hamann, and Rizzi (2010) also report that, in experimental data, French children use the

(A)symmetries in the production of object clitics by Romanian children

145

masculine singular clitic as a default. For Romanian, however, the number of tokens is too small; the existence of a short “default clitic” stage is not backed up by signi¿cant statistics. Moreover, as mentioned in the previous section, such a stage is attested only in two of the examined corpora. Clitic omission, however, is attested in all the corpora. ACs are omitted during a short period of time. As children’s MLU increases, clitic production increases as well (see, for similar ¿ndings for Italian, Caprin and Guasti 2009) and by age 3;0 the omission rate drops below 5%. But, interestingly, the rate of omission is not the same for all the ACs. There is an obvious asymmetry with respect to the omission of 1st/2nd vs. 3rd person ACs. All ACs emerge concurrently but the omission rate is statistically signi¿cant only with 3rd person ACs. 1st/2nd person ACs are hardly ever omitted. The comparison of the data in the three corpora reveals a highly signi¿cant difference (Wilcoxon p = .001). The data are summarized in Figure 9:

Figure 9. Overall use of 1st/2nd person vs. 3rd person ACs in the three corpora

The Romanian data are not singular in this respect either. Studies of ACs in longitudinal data for French (Hamann 2002, Rasetti 2003) also indicate a slight advantage for 1st/2nd person ACs during the early stages. In the Augustin corpus, the ¿rst attested object clitic is a 3rd person AC (at 2;2.13), but in between 2;4,01–2;6,16, only RCs and 1st person ACs are attested (Hamann 2002:35,Table 7, Rasetti 2003, Table 23). In the Marie corpus, after a ¿le where 1st, 2nd and 3rd person ACs are attested (at 1;8,26), one notices a time span (1;9,03–1;11,5) when 1st/2nd person ACs outnumber 3rd person ACs (Rasetti 2003, Table 23). Along the same line, Silva (2010) provides experimental data from European Portuguese which indicate that children produced more 1st/2nd person ACs than 3rd person ACs. One more asymmetry which was found in the data was the one between 3rd person singular and 3rd person plural ACs. The omission rate is much lower with plurals in two corpora, B. and I., but higher in the third one. The asymmetry found in the B. and the I. corpora is in line with previous ¿ndings for

146

Martine Coene and Larisa Avram

other languages, such as French (Rasetti 2003), Spanish (Domínguez 2003) and European Portuguese (Silva 2010). 4.1.2. Dative clitics 4.1.2.1. The data DCs are ¿rst attested slightly later than ACs. In the B. corpus, the ¿rst DC in the examined ¿les is found at 2;0, but no other DCs are found until 2;3. This is why we take 2;3 as the ¿rst instance of repeated use in this corpus (21). The ¿rst DC in the A. corpus is found at 1;11 but no other DC has been found in the examined ¿les until 2;2 (22). This is why we take 2;2 as the ¿rst instance of repeated use in the A. corpus. The ¿rst DC is attested in the ¿rst ¿le available (2;0) in the I. corpus, but in a structure which could well be formulaic. We take 2;2 as the ¿rst instance of free use, since it is in this ¿le that a DC is used with more than one verb (23): (21) Dă -mi mie. give clitic 1st sg DAT me 1st sg DAT ‘Give (it) to me’. [B. 2;3] (22) Dă -i apă! give clitic 3rd sg DAT water ‘Give her/him some water.’ [A. 2;2] (23) îmi pui ăs(t)a? clitic 1st sg DAT put this ‘Will you put this to me?’ [I. 2;3]

As can be seen in Figure 10 below, in the B. corpus the production of DCs reaches a 100% rate as early as 2;7, earlier than with ACs. But there is no signi¿cant difference between ACs and DCs in the B. corpus (Figure 11). In the I. corpus, a 100% production rate with DCs is reached at an even earlier age (Figure 12); however, there is no signi¿cant difference between ACs and DCs either (Figure 13). The data in the A. corpus differ from what was found in the other two corpora; with A., one ¿nds a signi¿cant difference between ACs and DCs: A. uses signi¿cantly more ACs than DCs (Wilcoxon Signed Ranks z = í2.275; p = .023). The results might, however, be misleading; the vast majority of omitted DCs in the A. corpus appears in the context of one and the same verb which is

(A)symmetries in the production of object clitics by Romanian children

147

used very often. This is why it is plausible to assume that the data cannot be taken as evidence in favour for or against a DC/AC developmental difference.

Figure 10. The production rate of ACs and DCs in the B. corpus

Figure 11. The overall production rate of ACs and DCs in the B. corpus

With respect to person, at ¿rst sight, in both the B. corpus and the I. corpus, there seems to be a slight difference between 1st/2nd person and 3rd person DCs, with 3rd person DCs being omitted over a longer period of time and at a higher rate:

148

Martine Coene and Larisa Avram

Figure 12. The production rate of ACs and DCs in the I. corpus

Figure 13. The overall production rate of ACs and DCs in the I. corpus

Figure 14. The production rate of ACs and DCs in the A. corpus

(A)symmetries in the production of object clitics by Romanian children

149

Table 3. Target like use of 1st/2nd person vs. 3rd person DCs Child B. I.

DC 1st/2nd 84.29% (n = 102) 88.23% (n = 60)

DC 3rd 25% (n = 1) 83.33% (n = 15)

The comparison is summarized in Figure 15 for B. and in Figure 16 for I.:

Figure 15. The production rate of 1st/2nd vs. 3rd person DCs in the B. corpus

Figure 16. The production rate of 1st/2nd vs. 3rd person DCs in the I. corpus.

But the difference does not reach signi¿cance in any of the two corpora, most probably because of the small number of tokens. In the A. corpus the difference is marginally signi¿cant (z = í1,680, p = .093), with 1st/2nd DCs being omitted less than 3rd person DCs overall.

150

Martine Coene and Larisa Avram

Figure 17. The production rate of 1st/2nd vs. 3rd person DCs in the A. corpus

Case errors are attested only in two corpora: the Accusative form is used in Dative contexts. In the B. corpus, case errors are attested at a stage when the omission rate shows target-like use. Across the ¿les, out of 103 clitics used in Dative contexts, 63.72% (n = 65) are Accusatives. The case errors are attested exclusively with 1st and 2nd person DCs: (24) a.

trebuie să *mă must să SUBJ clitic 1st sg ACC ‘you must give me’.

dai. give 2nd sg

b.

a zis că *mă fură iepurii. has said that clitic 1st sg ACC steals rabbits ‘She said she will steal my rabbits.’

c.

Vreau să *te povestesc. want să SUBJ clitic 2nd sg ACC tell 1st sg ‘I want to tell you a story.’ [B. 2;10]

In the I. corpus, the percentage of case errors is much lower. Out of the 75 clitics used in Dative contexts, only 5.33% (n = 4) are realized as ACs. No case errors are found in the A. corpus. 4.1.2.2. Preliminary summary Summing up, DCs are attested very early but slightly later than ACs in all the three corpora. The production rate of DCs reaches 100% shortly after ¿rst use, but in terms of overall production the difference between ACs and DCs does not reach signi¿cance. The Romanian data are similar to the ones reported for

(A)symmetries in the production of object clitics by Romanian children

151

European Portuguese, where the omission rates for ACs and DCs are comparable (Silva 2010). As regards Romanian early clitics the comparison has to take into account two facts: case error substitutions and omission/production rates. With respect to case errors, the performance of ACs is better. With respect to omission/production rates, it is still the performance of ACs which is slightly better, though the difference does not reach signi¿cance. The Romanian children seem to behave like the Portuguese children in this respect and to differ from Italian children. Caprin and Guasti (2009) report that Italian children omit ACs more than DCs. There is a slight 1st/2nd vs. 3rd person asymmetry with DCs across subjects, but it only reaches (marginal) signi¿cance with one child. In this respect, the Romanian data differ from the ones reported for European Portuguese. Silva (2010) reports an asymmetry between 1st/2nd vs. 3rd person both in the case of Accusative and in the case of Dative clitics. Case errors are attested only with 1st/2nd person DCs and the rate is subject to individual variation. 4.1.3. ReÀexive clitics 4.1.3.1. The data RCs emerge early. In the A. corpus, the ¿rst RCs are attested at 2;4 (MLU 2.136). In the B. corpus the ¿rst RC is attested at 2;1. One should, however, notice that all the instances of RCs at 2;1 are all 2nd person sg reÀexives which occur with the same verb used in a formulaic-like imperative (du-te ‘go-reÀ 2nd sg’). Therefore, we believe that B. produces the ¿rst “genuine” reÀexives at 2;2. In the I. corpus, RCs are attested at 2;1. What is striking is that there is almost no omission of RCs after their emergence. In the B. corpus, the production of reÀexives increases to 100% as early as 2;3 (MLU 1.821), only one month after the ¿rst genuine reÀexive was used. The data are summarized in Figure 18. The same picture emerges from the I. corpus, as can be seen in Figure 19. In the A. corpus as well the production rate of RCs increases to 100% early, at 2;8 (MLU 3.099). The data are summarized in Figure 20 below. The comparison of RCs with ACs in the B. corpus reveals a signi¿cant difference between reÀexives and non-reÀexive accusatives with respect to production rate; the former have a higher percentage of target-like use than the latter. This indicates that reÀexives are acquired earlier than ACs (Wilcoxon Signed Ranks z = í2,135, p = .033). In the A. and the I. corpora, the results seem to go along the same line, however without reaching signi¿cance. But all

152

Martine Coene and Larisa Avram

the three children have a higher percentage correct use at an earlier age, and RCs are used in a target-like manner earlier than both ACs and DCs.

Figure 18. The use of RCs in the B. corpus

Figure 19. The use of RCs in the I. corpus

There is a slight person asymmetry within RCs, but in this case it is reversed: it is the 3rd person clitic whose omission rate is lower, as can be seen in Figure 21 for B., Figure 22 for A. and Figure 23 for I. But the difference does not reach signi¿cance (B. p = .180; A. p = .593; I. p = .248).

(A)symmetries in the production of object clitics by Romanian children

153

Figure 20. The production rate of RCs in the A. corpus

Figure 21. The use 1st/2nd person vs. 3rd person RCs in the B. corpus

Figure 22. The use 1st/2nd person vs. 3rd person RCs in the A. corpus

A comparison of the developmental path of RCs and ACs reveals that the observed differences in omission rate are highly signi¿cant in the B. corpus, where RCs are acquired earlier than both ACs (p = 0.033) and DCs (p = 0.043). Figure 24 shows that RCs reach 100% production rate shortly after emergence:

154

Martine Coene and Larisa Avram

Figure 23. The use 1st/2nd person vs. 3rd person RCs in the I. corpus

Figure 24. The production of ACs, DCs and RCs in the B. corpus

In the other two corpora, the longitudinal data go along the same line (without reaching signi¿cance though), as can be seen in Figures 25 and 26. 4.1.3.2. Preliminary summary RCs (taking together 1st, 2nd and 3rd person contexts) reach a 100% target-like use in required contexts earlier than ACs or DCs and do so shortly after emergence. This ¿nding comes as no surprise, as several previous studies report lower omission rates for RCs. For example, several previous studies report lower omission rates for RCs than for ACs in French (Jakubowicz 1991, Jakubowicz, Müller, Riemer, Kang, and Rigaut 1996, Hamann, Rizzi and Frauenfelder

(A)symmetries in the production of object clitics by Romanian children

155

1996, Crysman and Müller 2000, Rasetti 2003) and lower omission rates for RCs than for both ACs and DCs in Italian (Caprin and Guasti 2009). The Romanian data are similar to the Italian data in this respect. No 1st/2nd person vs. 3rd person asymmetry has been found with RCs. As in the case of the other types of clitics, no clitic misplacement has been attested.

Figure25. The production of ACs, DCs and RCs in the A. corpus

Figure 26. The production of ACs, DCs and RCs in the I. corpus

156

Martine Coene and Larisa Avram

5.

Discussion

5.1.

Starting point

The longitudinal data which we have examined reveal that the omission/production rate for various object clitics in child Romanian is not uniform. The following (a)symmetries have been attested: (i) the omission rate of 1st/2nd person ACs is lower than that of 3rd person ACs but it is similar to that of RCs overall; (ii) there is a signi¿cant person asymmetry with ACs but not with DCs or RCs; (iii) the omission rate of RCs is lower than the one of DCs and of ACs overall; (iv) case errors are attested only with 1st/2nd person DCs. In what follows we will try to provide an account of these developmental (a)symmetries. 5.2.

Context-dependent vs. discourse-dependent referentiality

Concerning the 1st/2nd person vs. 3rd person AC asymmetry, one common difference that is often made between 1st/2nd vs. 3rd person pronominals in general is that only the former are inherently deictic; their interpretation is always discourse-dependent in that their reference is determined by the changing of discourse roles (Harley and Ritter 2002). 3rd person pronouns are variables whose reference is context-dependent and not indifferent to the ‘referential intentions of the speaker’ (Schlenker 2003:51). Also, speaker and addressee are higher than non-participant (3rd person) on a cognitive accessibility scale (Siewierska 2004:46). Their referent is highly accessible. This distinction might be taken as a starting point in an account of the person asymmetry within ACs which we found in our acquisition data. Children may deal differently with discourserelated (deictic) and context-related referentiality. There are, however, several facts which indicate that this difference alone cannot help us disentangle the acquisition facts. One is that the same “person effect” would be predicted across the various types of object clitics, a prediction which is not borne out by our data. With DCs, we found no signi¿cant difference between 1st/2nd person and 3rd person clitics. With RCs, if there is a difference (though not statistically signi¿cant), it goes in the opposite direction, with 3rd person RCs being less omitted than 1st/2nd person RCs. This suggests that the observed asymmetry might either not be a person asymmetry at all or that it might be rooted in a coalition of factors, person included.

(A)symmetries in the production of object clitics by Romanian children

5.3.

157

First and second person vs. third person ACs

5.3.1. Previous accounts As mentioned in section 1, there are – to our knowledge – two previous studies which explicitly address the difference between 1st/2nd person clitics and 3rd person clitics: Uriagereka (1995) and Kayne (2000). The two studies converge on one important conclusion: 3rd person clitics do not belong to the same class as 1st/2nd person clitics. They have different etymology, different morphological make up, different categorial status, and different syntactic properties. But the details of their analyses, as well as the conclusions with respect to the categorial status of the two classes of clitics are different. According to Kayne, 1st/2nd person clitics are NPs, with an empty D9, whereas 3rd person clitics are DPs. One important ingredient in his argumentation is gender. Since 1st/2nd person clitics do not specify gender, they cannot project D. Consequently, they are NPs. 3rd person clitics, on the other hand, are speci¿ed for gender and, hence, they project as DPs. Uriagereka’s analysis builds on the synchronic and diachronic relationship between 3rd person clitics and determiners. 3rd person ACs in Romance, which derive from the Latin demonstratives, are Ds which take a null complement (25). They are, using his terminology, weak determiner clitics, which are referentially de¿cient. (25)

DP (double)

D’

D

NP

CL

null

1st/2nd ACs are not referentially de¿cient, they specify person and behave like strong phrasal clitics. They are DPs (26):

9. The same analysis is assigned to French clitics, uniformly, by Déchaine and Wiltschko (2002).

158

(26)

Martine Coene and Larisa Avram

DP DP = D CL

The important element in Uriagereka’s analysis seems to be referential de¿ciency. In what follows, we will show in what way the two main ingredients in these two studies, gender and referential de¿ciency, can be integrated in one analysis. 5.3.2. The features in D and types of clitics Our account of the asymmetry which was found in the acquisition data builds on the important insight that the notion of ‘pronominal clitic’ does not refer to a homogeneous class of pronominal elements. In particular, ACs do not represent a homogeneous class. It can be argued, on the basis of morphological make up and syntactic properties, that it is necessary to differentiate between these two classes. A ¿ner-grained classi¿cation has the advantage of explaining the observed developmental asymmetry. The two studies brieÀy mentioned in the previous subsection built the proposed differentiation on the morphological make up of the DP. But D encodes two types of features: person features, responsible for the referential index of the DP, and phi-features (number and gender). Along the line of Longobardi (2008), we assume a distinct representation of the person head; this projection hosts the feature associated with referentiality, as suggested in Chomsky (1999), where it is associated with one variant of D: ‘Similarly D, – or at least one variant of D – might be associated with referentiality in some sense, not just treated as an automatic marker of “nominal category” (Chomsky 1999:35 fn10). Following this insight, we take the “traditional” DP to contain two projections: Person Phrase and øPhrase (with the possibility of further splitting the øP10 into a Number and a Gender Phrase). The former is the projection of the referentiality feature. The latter is a projection of the number and gender features:

10. This projection could be argued to encode individuation, i.e. features which are independent of the discourse (Bianchi 2006).

(A)symmetries in the production of object clitics by Romanian children

(27)

159

PersP Pers°

øP

1st/2nd person clitics are deictic. They have a Person feature (or, alternatively, a Participant feature, as one reviewer suggests), which is licensed by an operator in the C-domain, like any other deictic element, as proposed by Sigurðsson (2005). It is commonly assumed that 1st/2nd person clitics are not speci¿ed for gender (or, arguably, for number). This is what made Kayne (2000) analyse them as NPs. Though it is true that they do not encode gender, it is not entirely correct to say that they are indifferent to the gender feature. We would like to propose that they contain an underspeci¿ed øP which will be assigned a value via the same computation process through which the person or participant feature is licensed. The implicit assumption is that if a nominal is speci¿ed for person, it will also be speci¿ed for number and gender (see also Shlonsky 1989 for a similar assumption with respect to the possible projection of a split IP). This is reÀected in gender marking on epithets (see also Coene and Avram 2011). In (28) below, if the double contains an epithet, it has to be adjusted, since the double of 1st/2nd person ACs can be only a pronoun11. (28a) is ungrammatical precisely because the double is a lexical DP. In (28b) and (28c) the semi-pronominal epithet is marked for gender and number: (28) a.

(Pe mine,) nimeni nu mă ajută *pe prost. pe me no one not me helps pe stupid

b.

Nu mă ajută nimeni pe not me helps no one pe

prostul de mine. stupid-the masc of me

c.

Nu mă ajută nimeni pe proasta de mine. not me helps no one pe stupid-the fem of me ‘No one will help me, the fool that I am.’

We therefore assume that 1st/2nd person ACs project as a Person Phrase which contains an underspeci¿ed øP,whose feature(s) is/are context-dependent. This is in line with Uriagereka’s (1995) analysis, according to which 1st/2nd person clitics are DPs. But at the same time it also captures the important property that they are not speci¿ed for gender (or number); these features are valued via discourse anchoring, in the computation of the person/participant feature.

11. Lexical DPs cannot be 1st or 2nd person (Baker 2008).

160

Martine Coene and Larisa Avram

In terms of binding, they are subject to Condition B. In (29) below, the 1st person AC does not corefer with the DP which c-commands it in its local domain: (29) Vasile m - a desenat. st Vasile clitic 1 sg ACC has drawn ‘Vasile has drawn me.’ Local binding obtains only in the case of reÀexives. We take the property of 1st/2nd person non-reÀexive ACs illustrated in (29) to reÀect their pronoun status. They differ from strong pronouns, though, in that they cannot be the complement of a preposition (see Cardinaletti and Starke 1999). Their Person Phrase (DP) status also accounts for the fact that they are used as arguments. In Romanian, 3rd person ACs are identical in form to the de¿nite article12. Unlike 1st/2nd person ACs, they lack a referential index. They enter the derivation as articles, as D. Semantically, they do not have ¿xed reference or individuation. They simply spell-out the phi-features of the NP complement. They are not restricted to either [+ human] or [+ animate] antecedents. Their antecedent, whose features they inherit, can be either a nominal or a pronoun. They are not inherently de¿nite either. Their antecedent can be both de¿nite and inde¿nite. In Romanian, this property may be said to be reÀected not only in the compatibility of 3rd person ACs with both de¿nite and inde¿nite antecedents (Dobrovie-Sorin 1994) but also in the contemporary paradigm, which contains clitics identical both to the de¿nite and to the inde¿nite article: Table 4. Article - 3rd person ACs homophony Singular masc article

Inde¿nite De¿nite 3rd person Acc clitic

-l -l-

plural fem o o

masc

fem

-i-i-

-le -le-

Unlike 1st/2nd person ACs, D-clitics lack a Person/participant feature. We therefore assume that 3rd person ACs do not project a Person Phrase. They inherit the phi-features of the so-called antecedent in the derivation. This is in line with the de¿ciency of clitics in Uriagereka (1995) and Cardinaletti and Starke 12. The only exception is that of the 3rd person feminine singular AC which is identical in form to the inde¿nite article.

(A)symmetries in the production of object clitics by Romanian children

161

(1999). Since they can have a lexical DP as an antecedent and lexical DPs can only be 3rd person, D-clitics, which spell out the features of the antecedent, are interpreted as 3rd person by default. But they enter the derivation as ø, whose feature values are, via inheritance, those of the NP complement. Following Uriagereka (1995), we assume that 3rd person ACs enter the derivation as the D (without a Person domain) of a null complement (see 25 above). When D takes an overt NP as its complement, it is used as an article. If its NP complement is covert, as in (25) above, it is used pronominally. We adopt a movement analysis for the Romanian ACs; ACs are base-generated in post-verbal position. The antecedent, overt or null, is placed in the Speci¿er of a TopicP position, at the left periphery (for details, see Avram and Coene 2009)13. On such an analysis, all clitic constructions (single clitic, clitic doubling, clitic left dislocation structures) are treated in a uniform way (see also Del¿tto 2002). This captures the uniform interpretive property of clitics across all the structures in which they are obligatory: they all have a D-linked antecedent. 5.3.3. Types of clitics and intervention effects As we have said, 3rd person/Determiner-clitics, base-generated as heads of a DP, enter an Agree relation with a null NP. The D-clitic is a “copy” of the null NP, whose phi-features must be retrieved from a (clausal or discourse) antecedent. The antecedent may be an overt clause-mate (as in clitic left dislocation or topicalization structures) or null, in which case it was mentioned in previous discourse. It is precisely this Agree (at a distance) operation which is argued to give rise to dif¿culties with Determiner-clitics. The feature Agree operation involves a chain which includes the antecedent and the null complement across the subject DP: (30) [FP antecedent F [IP DPi I [VP ti V [DP D-null NP ]]]]

As can be seen in (30), the null complement in post-verbal position must “recover” the features of its antecedent over possible “inheritable” phi-features, i.e. the phi-features of the DP subject: {person, gender, number}14. The dif13. Our analysis departs here from the one in Uriagereka (1995) or from the “big DP analysis” where the double is placed in the Speci¿er of the DP. 14. The intervention effects are not affected by the position in which the subject surfaces (pre- or post-verbally) since the Agree relation is between the antecedent in

162

Martine Coene and Larisa Avram

¿culty arises when the phi-features in the intervening cluster are identical to those of the antecedent, as in (31) below: (31) [FP (antecedent FEM SG ) F [IP DP F EMSG V-T/Agrs3rd sg [VP ti V [DP D-null NP ]]] The structure in (31) has the Àavour of Relativized Minimality con¿gurations (Rizzi 1990), where a local relation between X and Y is disturbed when Z, a potential candidate for the local relation, intervenes. In this case, feature matching between X and Y is disturbed when Z, a feature cluster which is a potential “matcher”, intervenes. In (31) the intervention (-like) effects of the phi-features of the subject DP may affect the Agree relation between the null complement and its antecedent, adding to the computational load. (see also Coene and Avram 2011). This is what makes 3rd person ACs dif¿cult. And since their dif¿culty is associated with computational load, we expect clitic omission as well as agreement errors to be found with children, but also, occasionally, with adults. Our prediction is borne out by the data. As we saw, Romanian children omit 3rd person ACs and rare agreement errors are also attested. Occasional clitic omission or agreement errors are found with adults, especially when clitics are part of more complex structures. As one reviewer points out, this proposal also makes the prediction that the clitic should not be omitted when the features of the intervening subject have values different from those of the antecedent. Indeed, according to our proposal, the feature make up of the con¿guration can give rise to stronger or weaker intervention effects. For example, one would expect a lower rate of clitic omissions in con¿gurations like the ones in (32) below, where the antecedent of the clitic and the intervening subject have a different feature make up (as in 32a) or different feature values (as in 32b): (32) a.

b.

(Pe Maria) o desenez eu. pe Maria fem sg clitic 3rd fem sg ACC draw I ‘Maria, I draw.’ (Pe Maria) o desenează elevii. pe Maria fem sg clitic 3rd fem sg ACC draw students-the masc pl ‘Maria, the students draw.’

SpecTopP and the null NP in post-verbal position, before the D-clitic moves to the left-periphery, i.e. it is always across the intervening subject. If the subject is null, i.e. pro, it still has phi-features and it will also count as an intervener.

(A)symmetries in the production of object clitics by Romanian children

163

On the other hand, one would expect a higher clitic omission rate in con¿gurations like the one in (33) (which illustrates clitic omission in the B. corpus), where there is identity of feature values between the antecedent and the intervening subject: (33) că

doare burta

(pe Bianca)

because hurts tummy-the fem sg pe Bianca fem sg ‘Because Bianca has a tummy ache.’ [B. 2;2,13]

This prediction, for obvious reasons, is dif¿cult (if not impossible) to test against early longitudinal data. In the B. corpus, however, over a very short period of time (2;1-2;2), 1st person ACs are used instead of 3rd person ACs in clauses with a 1st person subject: (34) Adult: Ce-ai făcut cu cartea? ‘What have you done to the book?’ Child: *Mam pus la Kiki. clitic 1st sg ACC have put at Kiki ‘I have put it on Kiki.’ [B. 2;2,06] This indicates that intervention effects may arise, during the early stages, even when the feature make up of the antecedent and of the intervening subject is different. Data coming from experimental studies also show that this prediction is borne out. Avram and Coene (2010) discuss Accusative clitic production in Romanian in a task which elicited the use of 3rd person and 2nd person ACs. The results reveal a discrepancy between 2nd and 3rd person contexts: no omission was attested with 2nd person clitics. With 3rd person clitics the omissions were attested only in constructions with a 3rd person subject, i.e. when there was feature value identity between the antecedent and the intervening subject: (35) Pentru că Vasile (l-) a lovit. because Vasile clitic 3rd masc sg ACC has hit ‘Because Vasile has hit him.’ Gender errors were attested when the features of the antecedent and of the intervening subject had the same make up (number, gender) but not the same values. In (36) below, the clitic does not spell out the gender feature of the antecedent, which is masculine, but the gender feature of the intervening subject, which is feminine:

164

Martine Coene and Larisa Avram

(36) Vasile rîde pentru că Anca *o gîdilă Vasile masc is laughing because Anca fem her tickles ‘Vasile is laughing because Anca is tickling her.’ These results are also backed up by data coming from Italian and French. Tedeschi (2006, 2007) reports that Italian children (age 2;6–6;5) have problems linking the clitic to its antecedent in an elicitation task using a 3rd person subject: object clitics occasionally agree with the subject (instead of the object), which indicates that children treat the Accusative clitic as a sort of reÀexive. Pîrvulescu and Belzil (2008) used an acting task with tangible objects in which they elicited ACs in structures with the 2nd person tu ‘you’ instead of 3rd person in subject position. The 3–5 year old French children who participated in this task used the clitic almost 100%, while in a task using a 3rd person subject children omitted the clitic 30-50%. Summing up, we argue that what makes 3rd person ACs dif¿cult is the fact that they are part of a feature Agree chain which can be affected by feature intervention (see also Friedmann, Belletti and Rizzi 2009, Adani, van der Lely, Forgiarini and Guasti 2009, where intervention effects are argued to be the main reason of the delay in the acquisition of (some) direct object relative clauses or Guasti, Arosio and Branchini 2008, for the role of intervention in the acquisition of wh-questions ). 1st/2nd person ACs, on the other hand, enter the derivation speci¿ed for Person and they are licensed by an operator in the C-domain (Sigurðsson 2005), or via licensing by functional projections in one or several Person ¿elds, as suggested by Bianchi (2006) or Ciucivara (2009). There are no feature interveners between a 1st/2nd person clitic and their C-domain operator. 5.3.4. Types of clitics and movement One further important difference between the two types of ACs is related to movement. As already stated, we are adopting a movement analysis of clitics for Romanian, according to which ACs are base-generated in post-verbal position and surface in the left-periphery of the clause. According to Uriagereka (1995:93), the person de¿ciency of clitics motivates their movement to a projection in the C-domain, which he calls F, a projection higher than IP, which allows attribution of reference, i.e. a projection where they are referentially indexed. The associated null complement can only be licensed if the clitic is assigned referentiality in FP. But, if what drives movement to FP is referentiality, i.e. de¿ciency with respect to Person, Uriagereka’s analysis implies that only 3rd person ACs move there for referentiality reasons. Since 1st/2nd ACs are

(A)symmetries in the production of object clitics by Romanian children

165

speci¿ed for Person, an interpretable feature on DPs on minimalist assumptions, the motivation of movement to FP, attribution of reference, can no longer apply. Uriagereka proposes that these clitics move as phrases via adjunction scrambling (p. 114) to FP or another projection (the adjunction site being subject to cross-linguistic variation). Romanian ACs have been analysed as markers of topicality and d-linking (Avram and Coene 2009, for example, argue that they are markers of d-linked topicality); their movement to the left-periphery is associated with information structure effects. If one follows Uriagereka’s proposal, one could say that 3rd person ACs move for referentiality reasons and, possibly, also in order to check a topic feature15. 1st/2nd person ACs, on the other hand, are not referentially de¿cient. They move exclusively to check a topic feature. What both structures have in common is the interpretive effect of movement. One reviewer asks whether it is not precisely this added interpretive feature, more obvious in clitic left dislocation and in clitic doubling constructions, which makes clitics dif¿cult to acquire, and not the intervention effect of the features of the subject. However, there is cross-linguistic evidence that children are sensitive to information structure properties at a very early stage (de Cat 2002). In Romanian as well displaced topics are attested early (Avram and Coene 2006). Chomsky (2000: 120–121) mentions two types of interpretive principles: (i) what he calls deep interpretive principles, which include action-event structure or predicateargument relations and (ii) interpretive principles which are discourse-oriented, such as topic-comment, new-old information, focus. The former can also be found in language-like systems, but the latter are claimed to be speci¿c to human language. It comes as no surprise that children show sensitivity to these principles at a very early age. In answer to the question we started from, we would say that rather than being a source of dif¿culty, the interpretive property 15. One reviewer mentions that clitic doubling constructions in Romance languages may involve a focus interpretation (as argued, for example, by Isac 2003 for Romanian), while clitic left dislocation is associated with a topic interpretation. However, as shown by Dobrovie-Sorin (1994), Cornilescu (2001, 2002), Alboiu (2000), Isac (2003), in Romanian clitic left dislocation structures can be associated both with a topic and a focus interpretation. What requires the use of the clitic is not a topic or a focus feature, but the d-linked property of the direct object whose phi-features the clitic spells out (Avram and Coene 2009). Since Romanian is a focus in situ language (Cornilescu 2001), (contrastive) focus can be expressed in any syntactic position, provided it receives appropriate stress; the double of a clitic, in a clitic doubling construction, can but need not be associated with focus. In Romanian, the DPs in the Topic Phrase can have interpretations ranging from topic, to links and informational focus (Cornilescu 2001, 2002).

166

Martine Coene and Larisa Avram

of clitic constructions may actually have a bootstrapping effect, leading to the early acquisition of ACs. The intervention effect account also receives support from the analysis of clitics in structures with an overt antecedent, where retrieval of the phi-features of the antecedent is easier. Though the number of such clitic structures in the examined longitudinal data is very low, one can notice, nevertheless, that the production rate is higher than in single clitic constructions with a covert antecedent: 76% in the B. corpus, 83.7% in the I. corpus, and 78% in the A. corpus. Summing up, we propose that the developmental asymmetry between 1st/2nd person ACs and 3rd person ACs can be accounted for in terms of the different categorial status of the two types of ACs. 3rd person ACs are D-pronouns, with the feature make-up of determiners, whereas the other ACs are Person Phrases. This suggests that the reason for which one ¿nds an asymmetry in their acquisition is rooted in their different feature make up reÀected in their syntactic properties. One should mention at this point that the analysis of ACs which we are proposing may be both language speci¿c and type of clitic speci¿c. For example, no such asymmetry need exist with DCs, which are always referential, usually human and de¿nite (Farkas 1978). The fact that the asymmetry can be accounted for in terms of morphological make up allows for cross-linguistic variation. 5.4.

No 1st /2nd vs. 3rd person clitic asymmetry with DCs

The investigation of our longitudinal data reveals that there is signi¿cant 1st/2nd person vs. 3rd person clitic asymmetry with ACs but not with DCs. We propose that this difference is due to the fact that in Romanian DCs have the same feature make-up as 1st/2nd person ACs. In particular, they are not speci¿ed for gender, not even with 3rd person, i.e. they do not have the phi-features of nonpronominal DPs. 3rd person DCs are prototypically associated with a [+animate] [+affected] feature16, unlike their accusative counterpart. This indicates 16. A reviewer draws our attention that in some constructions 3rd person DCs can have a [-animate] associate, as in part-whole constructions like: (i) Iam rupt piciorul (mesei). have broken leg.the (table.the dat) clitic dat 3rd sg. ‘I have broken the leg of the table.’ This is indeed the case. As mentioned in the text, we take the prototypical DC to be associated with both animacy and affectedness. In constructions like the one in (i), it might be the affectedness feature, in conjunction with the part-whole (possession) interpretation, which allows the use of a DC. It seems that a DC can be more freely used in such constructions if the indirect object is interpreted as affected. When the

(A)symmetries in the production of object clitics by Romanian children

167

that DCs are uniformly pronominals; there is no Determiner-pronoun (with de¿cient referentiality) vs. pronominal distinction within this clitic class. We found one difference, though, between the acquisition of ACs and DCs in two corpora: DCs were the only ones which were substituted. Both B. and I. make case substitution errors; they use ACs in DC contexts. But they only do it with 1st and 2nd person clitics. We propose that the case errors which were found can be accounted for in terms of movement. Ciucivara (2009) also adopts a movement analysis of clitics; she argues that Romanian dative and accusative clitics are base-generated in post-verbal position, as in (37) below; the indirect object (IO) is generated above the direct object (DO) in the Speci¿er of a light applicative verb. Both the IO and the DO clitic undergo movement to a Kase ¿eld and a Person ¿eld in the functional domain. What is relevant for our discussion is the fact that clitics move to a Kase ¿eld in which Dative case is assigned in a projection higher than the one where Accusative case is assigned: (37)

KaseP K-dat K-acc VP ApplP IO

Appl’ Appl

DO

According to Ciucivara (2009), a Dative clitic (an IO) moves to K-dat after the DO has moved to K-acc. We propose that if the K-acc position, closer to the applicative VP, has not been taken by an AC, the child will opt for shortest Move: the IO will move only as high as K-acc. This will result in a case substitution indirect object cannot be interpreted as affected, some adjustment of the context may be needed. Compare (i) above to: (ii) ? Iam privit piciorul (mesei). have looked leg.the (table.the dat) clitic dat 3rd sg ‘I looked at the leg of the table.’

168

Martine Coene and Larisa Avram

error: an AC instead of a DC. The fact that case substitution can be accounted for in terms of shortest Move ¿nds support in the lack of case errors in clitic clusters, where the K-acc position has already been ‘taken’ by the AC, forcing the DC to move to the higher available case assigning position. Importantly, we found case substitution errors only with 1st/2nd person DCs, most probable because they have the same status (i.e. they are pronominal clitics) and the same feature make-up (in particular, they are speci¿ed for Person) as 1st/2nd person ACs. Third person ACs are actually Determiner-pronouns. 5.5.

RCs vs. DCs vs. ACs

In child Romanian, RCc are used adult-like shortly after emergence. The omission rate is lower than the one of DCs and ACs. It has been argued in the literature that RC constructions cannot be analysed on a par with non-reÀexives; the derivation of reÀexive structures differs from the standard derivation of transitive structures in more than one respect. Dobrovie-Sorin (1998), for example, argues that reÀexives are base-generated in pre-verbal position, from where they can feature match with the subject DP. On such an analysis, RCs do not involve movement and there are no intervention effects of potential identical feature clusters. ReÀexivity is feature identity. If both movement and feature intervention effects are computationally costly, this might explain why RCs are acquired faster as well as why there is no 1st/2nd vs. 3rd person developmental asymmetry in their case.

6.

Conclusions

We investigated the acquisition of reÀexive and non-reÀexive object clitics in Romanian, on the basis of longitudinal data. The analysis revealed a difference between 1st/2nd person ACs and 3rd person ACs only with respect to non-reÀexive ACs: 3rd person ACs emerge slightly later but are not target-like from the very beginning. In spite of very early emergence, they continue to be omitted at a stage when 1st/2nd ACs are used in an adult-like manner. The production rate of 1st/2nd person ACs is signi¿cantly higher than the one of 3rd person ACs. No such person asymmetry has been found in the case of DCs or RCs. We argued that the observed facts can be accounted for in terms of a distinction among the various types of clitics depending on their speci¿cation of person, number and gender, reÀected in their categorial status, in conjunction with an approach to feature intervention effects rooted in Rizzi’s Relativized Minimality (1990). Third person ACs were argued to be the only object clitics

(A)symmetries in the production of object clitics by Romanian children

169

which are Determiner pronouns; they are not inherently speci¿ed for person, and they inherit the number and gender features of the antecedent. Since they are base-generated in post-verbal position as the D of a null direct object, the Agree relation between this null object and its antecedent “crosses” over the intervening phi-features of the subject DP. This increases the computational complexity of those con¿gurations where there is identity between the phifeatures of the antecedent and those of the intervening DP subject. No feature intervention effects arise with clitics other than 3rd person ACs. st nd 1 /2 ACs and all DCs were argued to be pronominal and they were analyzed as Person Phrases/DPs, licensed by an operator in the C-domain, like any deictic category. For RCs we adopted an analysis according to which they are generated in pre-verbal position from where they feature match with the subject DP.

Acknowledgements We would like to thank two anonymous reviewers for their extremely helpful comments on an earlier version of this paper. The remaining errors are, as always, ours.

References Alboiu, Gabriela 2000 The Features of Movement in Romanian, Bucharest: Editura UniversităĠii din Bucureúti. Adani, Flavia, Heather van der Lely, Matteo Forgiarini and Maria Teresa Guasti 2009 Grammatical feature dissimilarities make relative clauses easier: comprehension study with Italian children. Ms. University of Milan. Avram, Larisa and Martine Coene 2006 The Complementizer Phrase in child Romanian: An early discourse-anchor. In Language Acquisition and Development. Proceedings of GALA 2005, Adriana Belletti, Elisa Bennati, Cristiano Chesi, Elisa DiDomenico, and Ida Ferrari (eds.), 28–34. Newcastle: Cambridge Scholars Publ. Avram, Larisa and Martine Coene 2007 Object clitics as Last Resort: Implications for language acquisition. In The Acquisition of Romance Languages. Selected Papers from the Romance Turn II, Sergio Baauw, Jacqueline van Kampen, and Manuela Pinto (eds.), 7–26. Utrecht: LOT (LOT Occasional Series 8).

170

Martine Coene and Larisa Avram

Avram, Larisa and Martine Coene 2009 Null objects and Accusative clitics in Romanian. Bucharest Working Papers in Linguistics XI (1): 233–254. Avram, Larisa and Martine Coene 2010 On the role of phonological de¿ciency and discourse accessibility in the acquisition of 2nd and 3rd person pronouns. Paper presented at The Romance Turn 4. Workshop on the Acquisition of Romance Languages, University of Tours, 24–27 August, 2010. Baker, Mark 2008 The Syntax of Agreement and Concord. Cambridge: Cambridge University Press. Bianchi, Valentina 2006 On the syntax of personal arguments. Lingua 116: 2023–2067. Caprin, Claudia and Maria Teresa Guasti 2009 The acquisition of morphosyntax in Italian: A cross-sectional study. Applied Psycholinguistics 30: 23–52. Cardinaletti, Anna and Michal Starke 1999 The typology of structural de¿ciency: A case study of the three classes of pronouns. In Clitics in the Languages of Europe, van Riesmsdijk, Henk (ed.), 145–233. Berlin: Mouton de Gruyter. Chomsky, Noam Derivation by phase. MIT Occasional Papers in Linguistics 18. Cambridge, Mass.: MITWPL Minimalist inquiries: The framework. In Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michaelis, and Juan Uriagereka (eds.), 89–156. Cambridge, Mass.: The MIT Press. Ciucivara, Oana 2009 A syntactic analysis of pronominal clitic clusters. The view from Romanian. Ph.D. diss., New York University. Coene, Martine and Larisa Avram 2011 An asymmetry in the acquisition of Accusative clitics in Romanian. In Comprehension-Production Asymmetries in Child Language. Proceedings der AG 3 der 30. Jahrestagung der DGFS in Berlin, Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk (eds.), 39–67. Berlin/New York: Mouton de Gruyter, SOLA 43. Cornilescu, Alexandra 2001 Direct objects at the Left Periphery. Bucharest Working Papers in Linguistics 3(1):1–18. Cornilescu, Alexandra 2002 Clitic doubling and parasitic gaps in Romanian. Paper presented at Going Romance, November 2002, Groningen.

(A)symmetries in the production of object clitics by Romanian children

171

Crysman, Berthold and Natascha Müller 2000 On the non-parallelism in the acquisition of reÀexive and non-reÀexive object clitics. In The Acquisition of Scrambling and Cliticization, Susan Powers and Cornelia Hamann (eds.), 207–236. Dordrecht: Kluwer. De Cat, Cécile 2002 French Dislocation. Ph.D. diss., University of York. Déchaine, Rose-Marie and MartinaWiltschko 2002 Decomposing pronouns. Linguistic Inquiry 33 (3): 409–442. Del¿tto, Denis 2002 On the semantics of pronominal clitics and some of its consequences. Catalan Journal of Linguistics 1: 29–57. Dobrovie-Sorin, Carmen 1994 The Syntax of Romanian. Comparative Studies in Romance. Berlin: Mouton de Gruyter. Dobrovie-Sorin, Carmen 1998 Impersonal se constructions in Romance and the passivization of unergatives. Linguistic Inquiry 29 (3): 399–437. Domínguez, Laura 2003 Interpreting reference in the early acquisition of Spanish clitics. In Linguistic Theory and Language Development in Hispanic Languages: Papers from the 5th Hispanic Linguistics Symposium and the 4th Conference on the Acquisition of Spanish and Portuguese, Silvia Montrul and Francisco Ordóñez (eds.), 212–228. Somerville, Mass.: Cascadilla Press. Farkas, Donka 1978 Direct and indirect object reduplication in Romanian. In Papers from the Seventeenth Regional Meeting of the Chicago Linguistics Society (CLS) 14, 88–97. University of Chicago. Friedmann, Naama, Adriana Belletti and Luigi Rizzi 2009 Relativized relatives: Types of intervention in the acquisition of A-bar dependencies’. Lingua 119 (1): 67–88. Guasti, Maria-Teresa, Arosio Fabrizio and Chiara Branchini 2008 Interference effects and the production of subject and object wh-questions in early Italian. Paper presented at The Romance Turn III. Workshop on the Acquisition of Romance Languages, University of Southampton, September 18–20, 2008. Hamann, Cornelia 2002 From Syntax to Discourse. Pronominal Clitics, Null Subjects and In¿nitives in Child Language. Dordrecht/Boston/London: Kluwer. Hamann, Cornelia, Luigi Rizzi and Ulrich Frauenfelder 1996 The Acquisition of subject and object clitics in French. In Generative Perspectives on Language Acquisition, Harald Clahsen (ed.), 309–334. Amsterdam/Philadelphia: John Benjamins.

172

Martine Coene and Larisa Avram

Harley, Heidi and Elisabeth Ritter 2002 Person and number in pronouns: A feature-geometric analysis. Language 78 (3): 482–526. Isac, Daniela 2003 Identi¿cation focus vs. contrastive focus: A syntactic distinction. In Romance Languages and Linguistic Theory 2001. Selected Papers from ‘Going Romance’ Amsterdam, 6–8 December 2001, Josef Quer, Jan Schroten, Mauro Scoretti, Petra Sleeman, and Els Verheugd (eds.), 113– 130. Amsterdam: John Benjamins. Jakubowicz, Célia 1991 L’acquisition des anaphors et des pronoms lexicaux en français’. In Grammaire générative et syntaxe comparée, Jacqueline Guéron and Jacques-Yves Pollock (eds.), 229–252. Paris: Editions du Centre National de la Recherche Scienti¿que. Jakubowicz, Célia, Natascha Müller, Beate Riemer, Ok-Kyuing Kang and Catherine Rigaut 1996 On the acquisition of the pronominal system in French and German. In Proceedings of the 20th BUCLD, Andy Stringfellow, Dalia Cahana-Amitay, Elizabeth Hughes, and Andrea Zukowski (eds.), vol. I, 331–342. Somerville, Mass.: Cascadilla Press. Kayne, Richard S. 2000 Person morphemes and reÀexives in Italian, French, and related languages. In Richard Kayne Parameters and Universals, 131-162. New York: Oxford University Press. Longobardi, Giuseppe 2008 Reference to individuals, person, and the variety of mapping parameters. In Essays on Nominal Determination, Alex Klinge and Henrik Müller (eds.), 189–211. Philadelphia/Amsterdam: John Benjamins. MacWhinney, Brian 2000 The CHILDES Project: Tools for Analyzing Talk. Mahwah, NJ: Lawrence Erlbaum Associates. Marinis, Theodoros 2000 The acquisition of clitic objects in modern Greek: Single clitics, clitic doubling, clitic left dislocation. ZAS Papers in Linguistics 15: 260–283. Pîrvulescu, Mihaela and Isabelle Belzil 2008 Object clitic omission in French-speaking children: Effects of the elicitation task. Paper presented at Generative Approaches to Language Acquisition, North America, September 2008, University of Connecticut, US. Rasetti, Lucienne 2003 Optional categories in early French syntax: A developmental study of root in¿nitives and null arguments. Ph.D diss., University of Geneva.

(A)symmetries in the production of object clitics by Romanian children

173

Rădulescu Sala, Marina 2008 Complementul indirect. In Gramatica Academiei, Valeria GuĠu Romalo (ed.), vol. II, 418–437. Bucharest: Editura Academiei Române. Rizzi, Luigi 1990 Relativized Minimality. Cambridge, Mass.: MIT Press. Schlenker, Philippe 2003 A plea for monsters. Linguistics and Philosophy 26 (1):29–120. Shlonsky, Ur 1989 The hierarchical representation of agreement. Unpublished Manuscript. University of Geneva. Siewierska, Anna 2004 Person. Cambridge: Cambridge University Press. Sigurðsson, Halldor Ármann 2005 The syntax of Person, Tense, and speech features. Rivista di Linguistica 16(1): 219–251. Silva, Carolina 2010 Asymmetries in the acquisition of different types of clitics in European Portuguese. In Movement and Clitics: Adult and Child Grammar, Vincenç Torrens, Linda Escobar, Anna Gavarró, and Juncal Guttiérez (eds.), 361–388. Newcastle: Cambridge Scholars Publishing. Snyder, William 2007 Child Language. The Parametric Approach. Oxford/New York: Oxford University Press. Sportiche, Dominique 1996 Clitic constructions. In Phrase Structure and the Lexicon, Johan Rooryck and Laurie Zaring (eds.), 213–276. Dordrecht: Kluwer. Stromswold, Karin 1996 Analyzing children’s spontaneous speech. In Methods for Assessing Children’s Syntax, Dana McDaniel, Cecile McKee, and Helen Smith Cairn (eds.), 23-53. Cambridge, Mass.: The MIT Press. Tedeschi, Roberta 2006 The acquisition of object clitics: Data from an elicited production task. Annali Online di Ferrera – Lettera, vol. 2: 31–42. Tedeschi, Roberta 2007 Referring expressions in early Italian. A study on the use of lexical objects, pronouns and null objects in Italian pre-school children. In The Acquisition of Romance Languages. Selected Papers from the Romance Turn II, Sergio Baauw, Jacqueline van Kampen, and Manuela Pinto (eds.), 201–216. Utrecht: LOT (LOT Occasional Series 8). Uriagereka, Juan 1995 The syntax of clitic placement in Western Romance. Linguistic Inquiry 26(1): 79–123.

174

Martine Coene and Larisa Avram

Zesiger, Pascal, Laurence Zesiger, Marina Arabatzi, Lara Baranzini, Stéphany CronelOhayon, Julie Franck, Ulrich Frauenfelder, Cornelia Hamann and Luigi Rizzi 2010 The acquisition of pronouns by French children: A parallel study of production and comprehension. Applied Psycholinguistics 31(4): 571–603.

Clitics in the acquisition of inalienable possession in French Katrin Schmitz

1.

Introduction

1.1.

Aims of the paper

All researchers working in the ¿eld of ¿rst language acquisition in Romance languages claim that children omit the reÀexive clitic se (see Müller et al. 2006; Schmitz and Müller 2008 for an overview). The present paper pursues the investigation of the acquisition of reÀexive and non-reÀexive object clitics in French*. The astonishing result of the investigation presented is that the French speaking children under study produce the reÀexive clitic se with possessors of inalienable body part constructions while they prefer lexical direct non-reÀexive objects. The children also omit clitics – in that case, clitics are ambiguous with respect to their nature. From the viewpoint of information, the aforementioned observation is also interesting: reÀexive clitics referring to possessors in inalienable body part constructions as e.g. elle se lave les mains ‘she washes her hands’ are highly redundant since se refers back to the subject and is thus easily retrievable, unlike omitted non-reÀexive object clitics. Much work has been done in the domain of Romance clitics, both from a theoretical and an empirical perspective: Gabriel and Müller (2005) developed a typology of clitics which differentiates between subject and reÀexive and nonreÀexive object clitics. Both object clitic types share the important property of * The data have been taken from the following research projects under the direction of Prof. Dr. N. Müller, ¿nanced by the DFG (Deutsche Forschungsgemeinschaft): 1) Frühkindliche Zweisprachigkeit: Italienisch-Deutsch und Französisch-Deutsch im Vergleich (Hamburg, 1999–2005); 2) Die Architektur der frühkindlichen bilingualen Sprachfähigkeit: Italienisch-Deutsch und Französisch-Deutsch in Italien, Deutschland und Frankreich im Vergleich (Wuppertal, 2005–2008). For further information see Müller et al. (2011). I thank Natascha Müller, Nadine Eichler, Veronika Jansen, Laia Arnaus Gil, Marisa Patuto and Christian Timm for discussing an earlier draft of this paper with me.

176

Katrin Schmitz

lacking a (N)ominal layer in contrast to subject clitics but differ with respect to their phrasal (non-reÀexive) or head (reÀexive) character in this model. Schmitz and Müller (2008) provided evidence for the validity of the typology which is able to explain the subject-object asymmetry observed in both monolingual and bilingual children acquiring French (and Italian) clitics. However, dative clitics are not mentioned in the model, i.e. it does not split up the non-reÀexive object clitics any further. The investigation of this issue has thus to move further ahead and the present paper integrates two complex constructions which contain both clitizisable direct objects and indirect objects or reÀexives, namely constructions of inalienable possession and double object constructions in both French monolingual and German-French bilingual children using data from two elicitation tasks testing these constructions (among others). Both constructions can be considered to be strongly related from a semantic viewpoint: double object constructions expressing mostly transfers of things, i.e. alienable possession, while inalienable possession deals with body parts and other inalienable entities. The present paper will, however, not study the acquisition of these two complex constructions in the ¿rst place but concentrates on the use of clitics (in contrast to other means of realizing the objects or their omissions) in monolingual and bilingual ¿rst language acquisition of French. An important question is whether the syntax of clitics is really completely independent from the semantic properties of the involved verbs and/or properties of the constructions they appear in. The present paper tries to answer this question by comparing data from monolingual French and bilingual FrenchGerman children. If the syntax-semantics interface is involved, cross-linguistic inÀuence is more likely to occur than in syntactic constructions which do not involve semantics or pragmatics (see criteria for the occurrence of cross-linguistic inÀuence formulated by Hulk and Müller 2000). So far, the syntax of clitics is considered to be a purely syntactic phenomenon (Schmitz and Müller 2008) which is less likely to trigger cross-linguistic inÀuence in bilingual children than phenomena at the interfaces (eg. subject and object realizations vs. omissions which are subject to both syntactic and pragmatic constraints, Müller et al. 2011). The second criterion for the probability of cross-linguistic inÀuence, namely an overlap of the two languages in a given grammatical domain, has been reformulated recently by Müller and Patuto (2009): on the basis of new observations in the domain of subject realizations, they reformulated it in a way that cross-linguistic inÀuence occurs only if the surface strings of two languages, A and B, are analyzable in terms of the syntactic derivation of one language (which is less complex). For the present paper it is important that the different position of German(ic) non-clitic object pronouns and French object clitics prevents the bilingual children from using the same syntactic analysis.

Clitics in the acquisition of inalienable possession in French

177

Therefore, we do not expect cross-linguistic inÀuence to occur in the domain of object clitics, be they direct or indirect, reÀexive or non-reÀexive. The present study shows that there is no cross-linguistic inÀuence in the constructions under study but very individual and restricted use of non-standard French inalienable body part constructions in mainly one child. Summarizing so far, the paper has two aims, namely a) to integrate Dative and ReÀexive clitics in the typology proposed by Gabriel and Müller (2005) and to explain the differences in the use of the different types of object clitics, observed by studying the special properties of reÀexive clitic se both in theoretical and empirical perspective in inalienable body part constructions, and b) to investigate whether cross-linguistic inÀuence may occur in the use of reÀexive clitics in these constructions as an indicator of an interface character. The paper is organized as follows: The remainder of section 1 brieÀy presents inalienable body part constructions in French and German focusing on those chosen for the elicitation with young children. Section 2 presents in more detail the typology of (the syntax of) clitics and important morphological and referential properties of French object clitics as well as previous results from studies on the acquisition of object clitics in French monolingual and bilingual children. Section 3 presents the empirical data from the two cross-linguistic studies mentioned above. Section 4 presents a syntactic account of the ¿ndings. Section 5 concludes the article. 1.2

Inalienable body part constructions

Inalienable possession is taken to be a semantically based dependency relation between two elements since an inalienable object is inherently de¿ned in terms of another object of which it is a part (see e. g., Vergnaud and Zubizarreta 1992; Alexiadou 2002). Body parts are treated as typical inalienable possessed nouns since body parts are de¿ned with respect to an individual. Seiler (1983) highlights the fact that alienable possession calls for a possessor who is acquiring the possessed entity while inalienable possession represents an inherent, intimate possession that does not need to be acquired. There is much variation among speakers concerning nouns other than the mentioned body parts and kinship terms which may also function as inalienables by extension (Vergnaud and Zubizarreta 1992). French is among the Romance languages most intensely studied regarding inalienable possession constructions in various theoretical frameworks (see overviews in Lamiroy 2003; Guéron 2006; Nakamoto 2009). It displays a variety of constructions expressing inalienable possession for body parts, partly also for kinship terms, clothes and other objects. Nakamoto (2009) presents

178

Katrin Schmitz

the following types of constructions which are directly contrasted with their glossed German counterparts in order to see parallelisms which might create overlaps in the perspective of the bilingual child (possessor marked bold, body part marked by italics): (1)

Nominative possessor, body part accusative (Nakamoto 2009: 2a): Il lève les mains. He raises the hands Er hebt die Hände.1 ‘He raises his hands’

(2)

Dative possessor, body part being accusative (2a) or oblique (2b) (Nakamoto 2009: 4a/5a): a. Jean s’ est lavé les mains. John self is washed the hands *Hans sich hat gewaschen die Hände Hans hat sich die Hände gewaschen. ‘John washed his hands’ b.

Pierre lui a collé un pansement sur le dos. Peter her has applied a plaster on the back Peter hat ihr ein PÀaster auf den Rücken geklebt. ‘Peter applied a plaster on her back’

(3)

Accusative possessor, body part oblique (Nakamoto 2009: 6a) Elle l’ a embrassé sur le front. She him has kissed on the forehead *Sie ihn hat geküsst auf die Stirn Sie hat ihn auf die Stirn geküsst. ‘She kissed him on the forehead’

(4)

Nominative possessor, body part contained in a small clause2. (Nakamoto 2009: 7c): Elle a les cheveux longs. She has the hairs long Sie hat das Haar lang ‘She has long hair.’

1. The parallel structure to the English one is also possible in German: er hebt seine Hände ‘He raises his hands’. 2. A word of caution is in order here: English “she has long hair” is ambiguous between two readings whereas French and German distinguish them: the ¿rst

Clitics in the acquisition of inalienable possession in French

179

The comparison of French and German inalienable body part constructions in the examples above shows that both languages differ mainly with regard to the order of the constituents in (2) and (3) since French uses preverbal case-marked clitics whereas German uses postverbal, case-marked (Dative) or underspeci¿ed (sich) object pronouns. This difference however, will prevent bilingual children from using the same syntactic analysis for both languages.3 In particular, the constructions in (2) and (3) contain actions that small children are familiar with but there is not yet a body of research on the acquisition of these structures in French. The present paper will focus on structures such as those in (2) and (3), exploiting data sets in which the dative possessors tend to be pronouns and the described activities those of cleaning own body parts or those of family members (see presentation in section 3.2). It should be added that there is variation regarding inalienable body part construction in Brussels French, i.e. in a language contact situation (French and a Germanic language): Nadasdi (1989) investigates the role of language contact in some grammatical phenomena in the French variety of Brussels arguing – contra Baetens Beardsmore (1971) – that the deviations4 from Standard French should in the ¿rst place be explained by the variation within French before the inÀuence of another language, in this case Flemish, is made responsible. Among the grammatical phenomena that are being investigated, the use of posor main reading corresponds to French elle a des cheveux longs (and in German Sie hat lange Haare) whereas the secondary meaning of the English construction corresponds to French elle a les cheveux longs (lit. “she has the hair long”) and German sie hat/trägt die Haare lang (lit. she has/wears the hair long). This secondary predication is probably not yet available to small children and these constructions are not part of the test item sets used in the present study. 3. One reviewer has reservations about young bilingual children being able to see this difference and to refrain from treating se and sich the same way since these pronouns share properties. There is, however, evidence in favour of this kind of knowledge reported in Müller and Patuto (2009) and Schmitz, Patuto, and Müller (2011) from the domain of subject realization: Unlike German-Italian bilingual children, French-Italian bilinguals do not exhibit cross-linguistic inÀuence from the non-null-subject language (German/French) on the null-subject language (Italian) in terms of a higher rate of realized subjects compared to monolingual Italian peers. The investigated French-Italian children realize very early that there is no overlap: Italian lacks subject clitics and French does not allow argument omissions at all (while German does by allowing topic drop). 4. Nadasdi calls those deviations simpli¿cations which seems inadequate since he deals with different constructions not with reductions of one construction leading to the other.

180

Katrin Schmitz

sessive adjectives in body part constructions is discussed. Nadasdi (1989: 3) reports the following examples of a pleonastic use of possessive adjectives: (5)

Pleonastic use of possessive adjectives in the French dialect of Brussels a. Elle me frotte mon dos She me rubs my back St. French: elle me frotte le dos ‘She rubs my back’ b.

Elle a mal à son She has hurting at her St. French: elle a mal au fois ‘She has her liver hurting’

fois liver

Nadasdi (1989: 3) further shows that deviations from the Standard French construction do not only occur in spoken Brussels French but also in Canadian French varieties such as the variety of Gaspésie in Quebec, concerning the replacement of the “reÀexive pronoun + de¿nite article construction” by the possessive adjective (Nadasdi 1989: 3–4): (6)

Possessive adjective instead of reÀexive pronoun + def. article: a. Je brosse mes dents tous les jours (Québec) I brush my teeth all the days St. French: Je me brosse les dents tous les jours ‘I brush my teeth every day’ b.

Il lave ses mains He washes his hands St. French: Il se lave les mains ‘He washes his hands’

(Brussels dialect)

For the present paper, this variation is interesting because of the possibility that one might also ¿nd the kind of structures described by Nadasdi, namely an avoidance of the standard French construction with the reÀexive clitic pronoun and the use of a possessive adjective, in the data from the bilingual German/ French children. There is variation in Germanic as well concerning inalienable body part constructions: Broekhuis and Cornips (1997) document different kinds of these constructions in Dutch varieties which correspond to the options found in the French varieties. The eastern varieties of Dutch abundantly use a variant which is very rare in Standard Dutch (Broekhuis and Cornips 1997: 186, 1)

Clitics in the acquisition of inalienable possession in French

(7)

Ik was hemdat de I wash him the

181

handen hands

In Standard Dutch, the the possession relation must be expressed by using a genitival possessor: … zijn hands (… his hands). In sum, there is a split both within Germanic languages (Dutch and German allow both constructions, depending partly on regional variation) and within French which is not necessarily the result of the inÀuence of another language. We may expect both options to show up in the data from the children without necessarily assuming an inÀuence from language contact. Although the paper deals with French, a short glance will be given at the German inalienable body part constructions in the investigated bilingual children in order to see if these constructions are likely to inÀuence the French ones in these children. For the further discussion of the constructions, the Standard French construction elle se lave les mains, the term external possession construction (extposs) by Vergnaud and Zubizarreta (1992) will be used, whereas the more frequent Germanic construction elle lave ses mains will be termed internal possession construction (int-poss) according to these authors. The following section deals with the syntactic, morphological and referentical properties of the different types of French object clitics.

2.

French object clitics and their acquisition

2.1.

The syntax of clitics

This section brieÀy presents a recent approach to the syntax of clitics by Gabriel and Müller (2005) which combines the assumptions on the external and internal syntax of clitics proposed by Cardinaletti and Starke (1999, 2000) and Déchaîne and Wiltschko (2002). The external and internal de¿ciency of the clitics is integrated and related to strong pronouns in the following typology:

182

Katrin Schmitz

(8)

A typology of clitics in relation to strong pronouns (Gabriel and Müller 2005) LUI DP D

ΦP Φ

NP N IL ΦP

Φ

LE/LO ΦP

SE/SI

Φ

Φ

NP N

LUI represents the strong pronoun and possesses a N-layer just as the subject clitic IL which is considered to be a pro-ijP with the most articulated structure. LE/LO is a pro-ijP as well but lacks the N-layer, while the reÀexive pronoun SE/SI is a ij-head. Clitics are a class of ij-expressions and differ from strong pronouns in that the former are of the categorical type ij while the latter are DPs. In this respect, clitics represent a homogeneous class. The important point of this approach, however, is that a further subdivision of the class of clitics is proposed which can explain many acquisition data. Subject clitics are arguments, reÀected in the presence of a N(ominal) layer while non-reÀexive and reÀexive object clitics are purely functional categories lacking the N-layer. A further difference between the latter is that reÀexives are heads whereas nonreÀexive object clitics are maximal projections. This model makes two predictions for acquisition (Schmitz and Müller 2008: 37) (A) If the acquisition process is guided by the external syntactic architecture of pronouns, strong (subject and object) pronouns – D-elements – should be acquired either before or after clitic pronouns, i.e. all elements of the category ĭ. (B) If children rely on the internal syntactic architecture of pronouns, one can predict that strong pronouns and subject clitics – elements which possess a N-layer – should be acquired together, either before or after object clitics which lack the N-layer.

Clitics in the acquisition of inalienable possession in French

183

What Schmitz and Müller (2008) found on the basis of their investigation of subject and (direct) object and reÀexive pronouns in both longitudinal crosscorresponds exactly to the prediction B, namely the following acquisition orders in their investigation of longitudinal and cross-sectional data from both monolingual French and Italian and bilingual French-German and Italian-German children. The relevant ¿ndings for the present paper are: – Subject clitics appear before object clitics. – ReÀexive clitics appear much later than subject clitics. – The appearance of reÀexive clitics in relation to non-reÀexive clitics is random, i.e., they can appear simultaneously or one before the other. Schmitz and Müller (2008) were, however, aware of the need to re¿ne this approach by implementing the fact that the different forms of clitics do not encode all ĭ-features (gender, number, person), the reÀexive se/si being the extreme case. The morphological shape of clitics, as well as other important properties, will be dealt with in the following subsection. 2.2.

The morphological and referential properties of French object clitics

According to Kayne (2000: 131 ff.), the apparently monomorphemic non-clitic pronouns can be further decomposed. He then extends this analysis to object clitics. Only the analysis concerning the latter will be brieÀy presented here: Kayne discusses the personal pronouns me (‘me’) and te (‘you’) which he decomposes in the person morphemes m- and t-, followed by an epenthetic vowel –e which is not pronounced when these pronouns occur pre-verbally and the verb begins with a consonant (Jean m’invite/Jean t’invite ‘John invites me/ you’). Due to the epenthetic character of the vowel, these pronouns cannot appear in postverbal position: *Invite-me/invite-te! ‘Invite me/Invite you(rself)’). In this case, a strong variant, again decomposed in the personal pronoun and the –oi ending m-oi/t-oi is used (invite-moi/invite-toi). Interestingly, it is possible to use the object clitics le, la in the position where me/te cannot be used: Invite-le/ Invite-la (‘Invite him/Invite her’). In this case, according to Kayne (2000: 133), the –e and -a are not epenthetic and le/la are truly bimorphemic. The next important difference concerning m-/t- versus l- involves a discussion of the term “third person”: Kayne (2000: 139) argues against this term and in favor of the term “determiner pronoun”, especially for the French object clitics le, la, les in e.g. Jean le/la/les voit (‘John sees him or it/her or it/them’) since they are identical in form to the de¿nite article. Besides these direct object clitics, also the French dative clitics lui (3.sg.), leur (3.pl.) and the nonclitic “third person” pronouns lui (3. mask. sg.), elle (3. fem. sg.), eux (3. mask. pl.)

184

Katrin Schmitz

and elles (3. fem. pl.) contain a l-. Kayne (2000: 140 f.) gives several reasons to assume that this l- is not to be grouped with m- and t-: (a) There is no corresponding gender distinction with ¿rst or second person: Jean me/*ma voit (‘John sees me’). French Dative clitics do not show a gender distinction either. (b) The determiner pronouns in l- often express plural by adding the usual plural morpheme (accusative clitic: les, subject clitics ils (mask. pl.) and elles (fem.pl.) and non-clitic elles (fem. pl.). According to Kayne (2000: 140), m- and t- combine with plural morphemes to express ¿rst- or second-person plural: *Jean mes/tes voit. *Jean parle de mous/tous. (c) The ¿nal argument concerning French is that l- never combines with the possessive morpheme that can show gender agreement with the head noun: *lon livre (opposed to mon/ton livre ‘my/your book’). Integrating now reÀexive s-, Kayne (2000: 142 ff.) states that it strongly patterns together with m- and t- (and with n- and v-5, but not with l-) concerning all properties investigated for m- and t-. So far, the proposal by Kayne (2000) could explain various dif¿culties that the children prove to have with the different object clitic types delaying their use from a morphological stance until after having acquired the determiners as has been shown by Müller and Kupisch (2007): They observed that (direct) object clitics appear at the sentence level when the target-deviant determiner omissions have almost disappeared from the child’s language. It can also be taken to support the proposal by Schmitz and Müller (2008) that the (potentially reÀexive) m-/t-/s-clitics are represented as the most de¿cient form (ij-heads) and the bimorphemic non-reÀexives le/la/les as ijP’s. Since both groups of clitics display their own dif¿culties in terms of agreement (l-clitics yield gender and number marking, m-/t-/s-clitics appear also as (non-clitic) part of possessive pronouns), it could also be argued that no acquisition order in terms of complexity can be formulated. Rowlett (2007: 122 f.) establishes basically the same two groups of object clitics as Kayne, but in terms of referentiality: While the object clitics me (m’), te (t’), nous, vous differ from the 3rd person object proforms le (l’), la (l’), les, lui, leur in being potentially coreferential with the subject, that is, reÀexive/ reciprocal (Je me lave ‘I wash myself’ tu te salis ‘you make yourself dirty’ nous nous adorons ‘we admire ourselves’ vous vous parlez ‘you speak to your5. The (clitic) object pronouns nous and vous have not been presented here since they are almost never produced by little children, in particular not in the context of an elicitation task.

Clitics in the acquisition of inalienable possession in French

185

selves’), the 3rd person object clitics cannot be co-referential. Where the 3rd person object is co-referential with the subject, a distinct 3rd person reÀexive/ reciprocal clitic se (s’) is used which is highly underspeci¿ed with regard to gender, number, and function (reÀexive, reciprocal, direct or indirect object): elle s’habille ‘she dresses herself’ vs. ils se parlent ‘they speak to themselves/ each other’. This reÀexive property allows se to express a possessor in relevant constructions (which, of course, extends to all clitics of the m-/t-/s-/n-/v-group, henceforth represented by se). Summarizing this subsection, there are differences both on morphological and referential grounds which motivate a distinction of m-/t-/s-clitics on the one hand and l-clitics on the other. The crucial question is whether the investigated children make their decisions on clitic use based on the above discussed properties. The following subsection deals with previous ¿ndings on the acquisition of object clitics in ¿rst language acquisition. 2.3.

The acquisition of French object clitics: omissions and realizations

Research on early child French (and Italian) showed that object clitics are omitted until the age of approximately 2;4 in both monolingual and bilingual ¿rst language acquisition (see the overview in Müller et al. 2006). The relevant ¿ndings from the studies which used the same tests as presented in this paper (see section 3 for relevant details of the methodology) are presented in what follows. Müller et al. (2006) compare the development of (direct) object clitics in monolingual French and Italian children with regard to possible language-speci¿c differences. The authors developed the ¿shing test presented in section 3.2 and administered it to 21 French and 19 Italian monolingual children at the age of two to three years. The children were grouped according to a qualitative and a quantitative criterion: the qualitative criterion concerned the ability to produce determiners in obligatory contexts; since some of the children did not omit determiners but performed weakly in the test, the number of answers constituted the quantitative criterion. They show that very young children acquiring French indeed behave differently from their Italian monolingual peers: they tend to repeat the object DP in obligatory contexts for clitics while Italian children omit the object altogether. The difference decreases in the course of the development: in all groups omissions decrease in favor of clitics. Müller et al. explain the different behavior with a language-speci¿c nature of the object omissions, using a typology of different empty pronominals (pro) proposed by Zushi (2003) (see Müller et al. 2006: 80 ff. for a detailed presentation). The French child starts out with

186

Katrin Schmitz

an NP-pro, for which no feature checking takes place and then moves on to a grammar for which agreement morphemes play no role in feature checking, with exceptions that are lexically de¿ned. In contrast, the Italian child starts out with a ĭP-pro and ¿rst assumes that agreement morphemes play a role in feature checking of objects (as is indeed the case for null subjects in the null subject language Italian according to Zushi’s approach). In other words, young children acquiring French have a syntactic representation of object omissions which corresponds to adult Japanese and they use a pragmatic strategy in order to license empty objects. Italian children, on the contrary, use a syntactic structure which is similar to the adult grammar of Brazilian Portuguese, i.e. they license objects via AGR. With regard to object clitics, the Italian children converge quickly with the adult system while the French children continue to use the pragmatic strategy. Eichler (2008) conducted the same test with 8 bilingual French-German children at the age of four to six years and investigated both their subject and object omissions and realizations. As in the monolingual study, the participants were split in two groups, based again on a quantitative and a qualitative criterion which, however, differ slightly from the criteria used in Müller et al. (2006). A child showing a low MLU (” 1,8) and a low standard deviation (” 1,4), both measured in the spontaneous part of the recording session, belongs to group 1 child, whereas a higher MLU and standard deviation quali¿es for group 2. Furthermore, this group classi¿cation corresponds to the existence of evidence for the C(omplementizer) system in the group 2 children, a qualitative criterion used also by Schmitz (2006). With regard to objects, Eichler observes that the bilingual children differentiate between direct, indirect and locative objects. In both languages, they omit the indirect and locative objects for a longer period, in particular in French where more of these objects are omitted in a target-deviant manner than in German. If objects are realized, they are mostly determiner phrases (DP’s). The study by Schmitz (2006) concentrates on the acquisition of the complex argument structure in double object constructions based both on longitudinal and cross-sectional studies. For the latter, the author developed a picture story with 16 activities (intransitive, transitive, ditransitive verbs and reÀexive verbs allowing an inalienable body part construction) and presented it to 14 GermanFrench bilingual children, among others, at the age of three to four years. The overall results of her study are the following, starting with the monolingual and bilingual longitudinal studies of the French speaking children: The monolingual French child omits the locative argument for a longer time and to a higher degree than the other object types whereas the monolingual German child omits the indirect object for a longer time and to a higher degree than the

Clitics in the acquisition of inalienable possession in French

187

other objects. All children used the ditransitive verbs correctly regarding their semantics and started to produce at least one of the objects very early without preference for a particular object type. The asymmetry concerning German is reÀected in the German production of the French-German children whereas the asymmetry of French is rarely present in the French of the bilingual children. An analysis of the case errors regarding Dative case shows that the problem with indirect objects cannot be explained by the slow acquisition of Dative, since indirect objects are realized target-like at a developmental stage in which many errors regarding Dative case marking still occur. The test data con¿rm the strong asymmetry of omissions between object types in German whereas the Romance languages (French and Italian) do not show the asymmetry found in the monolingual longitudinal studies. The omissions of direct objects decrease ¿rst. Regarding the realizations of objects and their respective ordering, the overall result is that the ordering of objects, both with respect to their respective order and to their placement in relation to the verb, does not represent a problem. The French monolingual child avoids clitic clusters by realizing locative and indirect objects mostly as (clitic) pronouns but direct objects rather lexically. Also the French-German bilingual children avoid clitic clusters. Finally, Schmitz (2006) could not explain her ¿ndings for the French-German bilingual children by cross-linguistic inÀuence. From these empirical results, one might conclude that there have to be fundamental differences between direct and indirect objects from the perspective of acquisition in the bilingual children. Both bilingual and monolingual French learning children tend to avoid the respective non-reÀexive object clitics in the earlier acquisition phases, favoring (direct and indirect) DP objects or omitting them all together. 2.4.

Hypotheses for the present paper

From the properties and observations reported in this section we should expect for the acquisition of clitic use in inalienable body part constructions that 1) Dative and ReÀexive clitics (i.e. se as bene¿ciary from the verb’s action) are often omitted. Se as possessor is more often realized than l-clitics which are mostly realized as DP’s or omitted, i. e., the children are able to distinguish the different referential properties of the object clitics (if not also their morphological shape). 2) Cross-linguistic inÀuence from German in the bilingual children should not occur.

188

Katrin Schmitz

3.

The study

3.1.

Participants and procedure

This study uses two elicitation tasks that were administered to different groups of children: Müller et al. (2006) tested monolingual French children living in France whereas Schmitz (2006) and Eichler (2008) tested German-French bilingual children living in Germany at the time of testing. Eichler (2008) uses the same test as Müller et al. whereas Schmitz uses a different one. Both will be presented in detail in section 3.2. In order to distinguish the tests already described in section 2.3, the test conducted by Müller et al. and Eichler will be called “¿shing test” (due to its presentation as a ¿shing game) and the test from Schmitz will be called “double object test” (since these constructions were the main objectives). The presentation here maintains the group division as proposed in the studies presented above since the two test items sets are not identical6: DP use and MLU in the young monolingual children, evidence for the C-system in the bilingual children (in the older children in the ¿shing game combined with MLU, see also section 2.3). The following tables 1 and 2 show numbers, age, MLU (as far as available7) and the number of answers of the children who produced the relevant items. The average indication (Ø) shows that, in principle, the production of the advanced bilingual group from the ¿shing game corresponds exactly to the advanced bilingual group from the double object test. Furthermore, the difference in production of the relevant items is less marked between the less and the more advanced group in the double object test than in the ¿shing test (bilingual children).

6. I do not follow one of my anonymous reviewers who proposed to group the children together since both tests contain slightly different items as ¿llers. Furthermore, the advantage of getting larger numbers for the analysis concerns only the bilingual children and does still not allow the calculation of percentages for all children. The distinction between most of the bilingual children would be lost as well, since 16 out of the 18 bilingual participants would belong to the same group based on the two criteria (MLU, answered test items). 7. If MLU values are not available this is due to a missing spontaneous speech part in the recording. Missing MLU values are marked “N.A.” (not available).

Clitics in the acquisition of inalienable possession in French

189

Table 1. French speaking children in the ¿shing test (data extracted from Müller et al. 2006, Eichler 2008) Name

KEV MIC THE N. AGA Ø TOM R. CHA ALE QUE B. SER MED Ø AN EM Ø KE JO KA AM Ø

Age

MLU

No. of answered test items of the list above Monolingual French children Group 1 2;6,27 1.9 4 2;11,16 1.5 1 2;10,20 2.5 2 2;9,24 4.5 1 2.6 2 Monolingual French children Group 2 2;11,23 2.3 2 3;1,15 N.A. 4 2;8,4 4.7 4 2;11,19 3.7 4 3;1,14 2.2 3 2;6,29 2.3 2 3.0 3 Bilingual French children Group 1 6;1,2 1.2 2 4;4,14 2.5 4 1.85 3 Bilingual French children Group 2 5;4,18 4.4 4 6;2,14 3.2 7 6;5,0 3.5 8 2;11,0 3.9 5 3.75 6

In all cases, the children were recorded in a separate room in the respective kindergarten. Note that elicitation tasks do not allow the measurement of the development of the two languages in the bilingual child and the (temporary) dominance of one language with regard to the other to be safely stated. In most cases, wherever possible, a spontaneous speech part was recorded as well which served to establish the MLU listed above.

190

Katrin Schmitz

Table 2. French-German bilingual children in the double object test (data extracted from Schmitz 2006)8 Name

Age

ZO AL VI MC Ø

3;5,6 3;2,27 3;4,19 2;9,30

VA SO CH TO PE CE AN EM Ø

3;8,7 4;0,25 3;2,2 3;6,15 4;1,1 3;3,12 3;6,7 3;11,8

MLU

No. of answered test items of the list above

4.7 2.6 2.7 3.8 3.45

7 3 6 4 5

4.7 4.1 3.9 3.6 3.0 3.4 3.0 4.5 3.77

6 7 3 4 4 3 7 7 5

Group 1

Group 2

3.2.

The test items

The present study uses a subset of the test items of two elicitation tasks, both designed to test the children’s production of (pronominal) objects with various types of dyadic and triadic verbs. The ¿rst test, documented in more detail in Müller et al. (2006), was presented as a ¿shing game during which the children had to ¿sh for pictures with various actions and to answer questions of the type: “What is X doing with Y?” (transitive verbs) or “What is happening?” (intransitive verbs) inviting the children to use object pronouns. The 15 test items represent various activities with different kinds of verbs (intransitive, transitive, ditransitive and reÀexive verbs). Müller et al. conducted this test with French and Italian monolingual children at the age of 2–3 years. Adhering to the same experimental procedure, Eichler (2008) tested a small number of bilingual German-French children at the age of 4–6 years.

8. The children AN and EM in the ¿shing game are not identical to the children AN and EM in the double object test (both tests were conducted in different cities and kindergardens).

Clitics in the acquisition of inalienable possession in French

191

The relevant items of this test for the present study are those containing inalienable body part constructions (henceforth abbreviated as IABP in titles and tables) with one possessor different from the subject (item [1]) and two possessors identical to the subject of the sentence (items [2,3]), double object constructions [4,5,8])9 as well as reÀexive and two-place mettre (to put) constructions ([6,7]) which are easily extended to three-place mettre, in particular by the bilingual French-German children since mettre les pantalons ‘put on the trousers’ corresponds to German sich die Hose anziehen ‘to put oneself the trousers’.10 The relevant items for the present study are listed below: [1]

La mère lave le visage à l’enfant ‘The mother washes the children’s face’

[2]

L’ours se lave/brosse les dents ‘The bear brushes his teeth’

[3]

Le chat se lave le visage ‘The cat washes his face’

[4]

La mère souris donne des gâteaux à l’enfant/la petite souris ‘The mouse mother gives cakes to the little mouse’

[5]

La ¿lle met des graines sur une assiette ‘The girl puts seeds on a plate’

[6]

La ¿lle met le maillot de bain ‘The girl puts on the swimsuit’ (alternatively accepted: reÀ. la ¿lle s’habille ‘the girl dresses herself’11)

9. The term “double object constructions” covers all constructions with 3-place verbs, i.e. verbs with a direct and an indirect object (e.g. donner ‘give’) or with a direct and a locative object (e.g. mettre ‘put’). Both types can be expressed in French only via a construction with a DP for the direct object and a PP for the indirect/locative object. 10. The ¿shing test was also applied to Italian speaking children where the corresponding structure resembles the German one: mettersi i pantaloni, i.e. with reÀexive si. 11. See the discussion of the results of the bilingual children in section 3.3. for the relevance of including these constructions.

192

Katrin Schmitz

[7]

La souris met/enlève la chaussette/La souris s’habille (see [6]) ‘The girl put on/takes off the sock’

[8]

La ¿lle met le crocodile dans la valise ‘The girl puts the crocodile in the suitcase’

The second experiment is a continuous picture story designed for the elicitation of triadic verbs familiar to young children about a day in the life of an invented girl named Lucy and her friends Marc and Rosa, used in Schmitz (2006). Among the ¿llers, transitive and inalienable body part constructions are used which are of interest for the present study. The test contained 16 items and was conducted with bilingual children only. The relevant items for this study are listed below: items [9,10,11] constitute inalienable body part constructions, [12] a transitive construction with a benefactive adjunct, [13–17] double object constructions. [9]

Lucy se lave le visage ‘Lucy washes her face’

= item [3] from the ¿shing test

[10] Lucy se lave/brosse les dents ‘Lucy brushes her teeth’

= item [2] from the ¿shing test

[11] Lucy se frotte/gratte les yeux ‘Lucy rubs her eyes‘ [12] Marc ouvre la porte à Lucy ‘Marc opens Lucy the door’ [13] Lucy met le gâteau dans une boîte ‘Lucy puts the cake in a box’ [14] Lucy donne le gâteau à Marc ‘Lucy gives Marc the cake’ [15] Lucy donne l’argent à la caissière ‘Lucy gives the money to the cashier’

Clitics in the acquisition of inalienable possession in French

193

[16] Rosa met la tasse sur la table ‘Rosa puts the cup on the table’ (alternatively accepted: Rosa met/dresse la table ‘R. sets/lays the table’) [17] Lucy offre des biscuits à Rosa ‘Lucy offers biscuits to Rosa’ Summarizing the item types and comparing both test sets with respect to the maximal amount of possible realizations), table 3 indicates the correspondences of the items: Table 3. Comparison of test items Test Fishing test Doc-test

3.3

IABP 3 [1,2,3] 3 [9,10,11]

DOC 3 [4,5,8] 5 [13–17]

ReÀ. 2 [6,7] –

Other 2 [6,7]12 1 [12]

The results

As indicated above, only a subset of the test items was used. The analysis of the answers excludes those children who did not provide answers to the items relevant here. Answers without a ¿nite verb form were excluded since they involve target-like omissions of subjects and objects. Answers which departed largely from the requested answer13 were excluded as well. This happened in two cases. The presentation of the results proceeds in several steps: First, a short quantitative analysis shows the production of inalienable body part and double object constructions with the respective realization or omission of objects by groups of items (3.3.1.). In the second subsection, the quantitative analysis of the use of clitics vs. DP as direct, indirect and reÀexives is given by group (3.3.2.). Finally, a qualitative analysis of the two constructions in the data of the monolingual and bilingual children with particular regard to the question of cross-linguistic inÀuence is provided both for French (3.3.3.) and for the constructions in German from the bilingual children (3.3.4.).

12. Table 3 registers twice the items [6, 7], since the reÀexive alternates of the two-place mettre constructions have been accepted as well. 13. E.g. when the child described a detail of the given picture which was not asked for and the description did not contain the relevant constructions.

194

Katrin Schmitz

3.3.1. The realization of objects in IABP and DOC This subsection brieÀy presents the realizations of the test items presented in section 3.1. The tables 4 (¿shing test) and 5 (double object test) indicate whether the items have been realized as expected or differently (e.g. an inalienable body part constructions as reÀexive, simple transitive/without object, the latter being comprised of “other”, see section 3.3.2 for examples of each kind). Table 4. Realizations of the test items in the ¿shing test Group Mon-G1

Mon-G2

Bil-G1

Bil-G2

Item No. 1,2,3 4,5,8 6,7 1,2,3 4,5,8 6,7 1,2,3 4,5,8 6,7 1,2,3 4,5,8 6,7

IABP 0 0 0 5 0 0 2 0 0 7 0 0

ReÀ. 1 0 1 1 0 4 0 0 1 2 0 4

DOC 0 0 0 0 2 0 0 5 0 0 5 0

other 2 0 2 1 1 6 1 0 1 3 0 4

Table 4 shows that target-like inalienable body part constructions in the ¿shing test do occur, apart from group Mon-G1 (0/3 answers), in the other groups: Mon-G2 in 5/7 answers (71%), Bil-G1 in 2/3 answers (67%), Bil-G2 in 7/12 answers (58%). Target-like double object constructions occur quite rarely in the ¿shing test: maximally two in advanced monolingual children and 5 in the two bilingual groups. Table 5. Realizations of the test items in the double object test (only bilingual children)

Group G1

G2

Item No. 9,10,11 12 13–17 9,10,11 12 13–17

IABP 1 0 0 9 0 0

ReÀ. 1 0 0 2 0 0

DOC 0 0 3 0 0 4

other 3 4 6 5 2 15

Clitics in the acquisition of inalienable possession in French

195

Target-like inalienable body part constructions do also occur in both groups of the doc test. G1 in 1/5 answers (20%), G2 in 9/16 answers (56%). Compared with Bil-G2 from the ¿shing test, we can see that both advanced bilingual groups pattern very much alike. This also holds for the production of target-like double object constructions in the double object test: both groups use this construction rarely (less than offered items, only 3 resp. 4 answers). Overall, we observe a restricted use and frequent object omissions. The following subsection investigates in more detail the use of clitic pronouns in the data of the children. The examples presented above hint to the fact that the use of the reÀexive clitic does not represent a problem per se. although omissions do occur. 3.3.2. The realization of clitic vs. other objects This section presents the investigation of the use of Dative and ReÀexive clitics compared to both direct object clitics and the respective lexical (DP and PP) objects in order to see if there is a general avoidance of these clitic types. The ¿gures 1 to 3 show, separated by type of object, that all children, whether monolingual or bilingual, tend to realize direct and indirect objects as lexical DP resp. PP whereas the possessor-related argument in inalienable body part constructions is exclusively realized by clitics. The abbreviation “str.pr.” refers to strong pronouns as e.g. à moi ‘(to) me’ (opposed to the clitic me ‘me’).

Figure 1. Clitics vs. other DO realizations by the different tested groups

196

Katrin Schmitz

Figure 2. Clitics vs. other IO realizations by the different tested groups

Figure 3. Clitics vs. other possessor realizations by the different tested groups

The following examples illustrate the Dative pronouns in the data of both tests which are largely avoided as ¿gure 2 shows:

Clitics in the acquisition of inalienable possession in French

(9)

197

Dative pronouns a.

elle lui donne à manger/

b.

elle lui prend ça/

c.

elle lui donne le gâteau/

d.

elle le donne à lui/

(QUE B. Mon-G2, 2;11,19) (CHA, Mon-G2, 3;1,15) (AN G2, 3;6,7) (EM, G2, 3;11,8)

These examples concentrate on item [4] of the ¿shing test (9a,b) and item [14] of the doc test (9c,d). 3.3.3. Qualitative description of IABP While the previous sections described only quantitative aspects of the results, here the qualitative perspective is taken, in particular regarding certain “German”-looking structures in the data of the bilingual children from both elicitation tasks. In the ¿rst step, however, the variety of realizations of inalienable body part constructions by the monolingual and bilingual children in the two tests will be illustrated by the following examples for the individual items of both elicitation tasks. (10) examples for item [1] a.

ben elle le lave/

(Tom R., Mon-G2, 2;11,23)

b.

elle lave la ¿lle/

(KA, Bil-G2, 6;5,0)

(11) examples for item [2] a.

i’ va se laver/

b.

il se brosse les dents/

c.

il brosse les dents/

d.

i’ se brosse les dents/

(AGA, Mon-G1, 2;9,24) (QUE B., Mon-G2, 2;11,19) (AN, Bil-G1, 6;1,2) (AM, Bil-G2, 2;11,0)

(12) examples for item [3] a.

il est en train d’ s’ laver les mains/

(ALE, Mon-G2, 2;8,4)

b.

i’ s’ nettoie/

c.

un chat qui s’ – qui s’ – qui s’ lave la bouche/ (EM, Bil-G1, 4,4,14)

d.

il se lave/

(QUE B. Mon-G2, 2;11,19) (KA, Bil-G2, 6;5,0)

198

Katrin Schmitz

(13) examples for item [9] a.

se lave/

(ZO, G1, 3;5,6)

b.

nettoie son visage/

c.

elle se nettoie le visage/

(VA, G2, 3;8,7)

d.

elle va se laver/

(AN, G2, 3;6,7)

(AL, G1, 3;2,27)

(14) examples for item [10] a.

elle brosse les dents/

(ZO, G1, 3;5,6)

b.

elle brosse les dents/

(PE, G2, 4;1,1)

c.

et après elle s’est lavée les dents/

(SO, G2, 4;0,25)

(15) examples for item [11] a.

elle se frotte les yeux (veut pas) sophie/

b.

elle se frotte les yeux/

(MC, G1, 2;9,30) (AN, G2, 3;6,7)

The examples (10) to (15) clearly show that, if omissions occur, inalienable body part constructions are reduced to transitive or reÀexive constructions by omitting an object, mostly the (clitic) possessor or the possessed body part. Only one example of structures like those presented by Nadasdi (1989) for Canadian French could be found, namely (13b), containing a possessive adjective in the body part-DP (int-poss-construction). The second step of the analysis concerns structures which could be assumed to be caused by cross-linguistic inÀuence, more precisely by a transfer of structures from German. Concerning the ¿shing game, such structures occur with double object constructions including donner as well as with 2place-mettre (in the target system for the expression of ‘to put on clothing pieces’) and the reÀexive verb s’habiller ‘to dress’. With regard to the double object constructions, the lexical indirect objects are realized as a DP instead of a PP in some of the realizations of item [4], but only by two of the children (16a,b): (16) a.

elle donne la petite souris les biscuits/

(JO, 6;2,14)

b.

donne la petite souris les noisettes/

(KA, 6;5,0)

c.

elle donne des gateaux à s a p’tite ¿lle souris/

(KE 5,4;18)

d.

elle donne ça à la petite souris/

(AN 6;1,2)

Clitics in the acquisition of inalienable possession in French

199

There are also a few constructions with an additional reÀexive pronoun with 2place mettre or with an additional direct object DP combined with the reÀexive s’habiller which rather complicate the structure. Both errors occur only in the eldest bilingual children as the following examples show: (17) [picture 6] elle s’ habille un maillot de bain/ she self puts a swimsuit German: Sie zieht sich einen Badeanzug an/ ‘She puts herself a swimming suit’ target: elle met un maillot de bain.(2place mettre) or alternatively elle s’habille (reÀ.) German: sich x anziehen ‘to put oneselfdat xacc’

(JO 6;2,14)

(18) [picture 7] elle se – elle s’ habille ses chaussettes/ (AN 6;1,2, G1) she self puts her socks German: Sie zieht sich ihre Socken an ‚She puts herself her socks‘ Target: elle met ses chaussettes/elle s’habille (19) [picture 7] elle se met les chaussettes/ she self puts the socks German: sie zieht sich ihre Socken an Target: elle met les/ses chaussettes/

(KA 6;5,0, G2)

To summarize, the errors described above which appear to be caused by crosslinguistic inÀuence (transfer) from the German structures are restricted to a few items and they are also produced by only two of the bilingual children. Concerning the double object test, there were no parallel examples to (16)–(19). Cross-linguistic inÀuence should appear across the tested populations since it is predictable on the basis of a systematic overlap of the languages in a grammatical domain which does not hold in the present case (see section 1.2). The low number and concentration in two children of these constructions is therefore interpreted as evidence against cross-linguistic inÀuence.

200

Katrin Schmitz

3.3.4. IABP in German This section brieÀy looks at qualitative aspects of the realizations of inalienable body part constructions in German with respect to the use of external and internal possessor constructions. Only constructions with a ¿nite verb are included since in¿nitives may involve target-like argument omissions. The relevant items of the ¿shing test are: [1]

Die Mutter wäscht dem Kind das Gesicht ab. ‘The mother washes the child’s face’

[2]

Der Bär putzt sich die Zähne. ‘The bear brushes his teeth’

[3]

Die Katze wäscht sich das Gesicht. ‘The cat washes his face’

It becomes evident that item [1] is mostly realized with the child as (animate) direct object, only (20e) is an int-poss-construction (‘she washes her face’), however without naming the possessor referred to, namely the child. For the other two items, both constructions are used, preferably however the ext-possconstructions with sich. Interestingly, it is always the same child, JO, who prefers the “Germanic” int-poss construction (20e, 21e, 22d): (20) examples for item [1]: a.

die, putzt das kind/

(AN, 6;5,2, G2)

b.

sie wäscht das kind/

(KA, 6;5,14, G2)

c.

die wäscht sie blitzsauba/

(KE 5;4,21, G2)

d.

sie putzt, das kleine mädchen/

(FR, 3;9,28, G2)

e.

die mutter wäscht ihr gesicht/

(JO 6;2,14, G2)

(21) examples for item [2]: a.

putzt sich die zähne/

(AN 6;5,2, G2)

b.

der putzt sich die zähne/

(KA 6;5,14, G2)

c.

ehm, der putzt sich die zähne/

(KE 5;4,21, G2)

d.

der putzt sich die zähne/

(FR 3;9,28, G2)

e.

der putz seine zähne/

(JO 6;2,14, G2)

Clitics in the acquisition of inalienable possession in French

201

(22) examples for item [3]: a.

putzt sich den gesicht/

(AN 6;5,2, G2)

b.

die wäscht sich das gesicht/

(KA 6;5,14, G2)

c.

die wäscht sich den gesicht/

(KE 5,4,21, G2)

d.

sie wäscht ihr gesicht/

(JO 6;2,14)

Relating the results from the child JO for German to those for French in the previous section supports the interpretation of a very individual choice of one child and against a systematic cross-linguistic inÀuence which should affect more than one bilingual child. The relevant items from the double object test are the following ones: [9]

Lucy/sie wäscht sich das Gesicht ‘Lucy washes her face’

[10] Lucy putzt sich die Zähne ‘Lucy brushes her teeth’

= item [3] from the ¿shing test

= item [2] from the ¿shing test

[11] Lucy reibt sich die Augen ‘Lucy rubs her eyes‘ Again, only utterances with ¿nite verbs have been considered (see above). Concerning item [9], there is only one realization of the expected construction (23d), two reÀexive (23 a,c) and one transitive (23e) and an omission of all objects (23b). The most frequently realized item is [10] with two expected extposs-constructions (24a,d), three transitive constructions (24b,c,e) and one intposs-construction (24f). Item [11] is realized in (25a) as described by Nadasdi for the Brussels French construction, namely with both the reÀexive pronouns sich and a possessive DP (ihre augen ‘her eyes’) while (25b) is reduced again to a transitive construction: (23) examples for item [9]: a.

die wascht sich/

(VA, 3;8,21, G2)

b.

die – die wascht/

(SO 4;0,11, G2)

c.

die macht sich sauber/

(TO, 3;6,29, G2)

d.

sie putzt sich die zähne/

(CE 3;3,23, G2)

e.

sie reibt – sie putzt ihre backe/

(PE 4;2,16, G2)

202

Katrin Schmitz

(24) examples for item [10]: a.

sie putzt sich die zähne/

(AN, 3;10,18, G1)

b.

die zähne putzt sie/

(MC, 3;10,20, G1)

c.

und da, putzt noch die zähn- die zähne/

(VA, 3;8,21, G2)

d.

die wa[s]t sich die zähne/

(TO 3;6,29, G2)

e.

sie putzt die zähne/

(CE 3;3,23, G2)

f.

emm, die putzt – die putzt ihre zähne/

(PE 4;2,16, G2)

(25) examples for item [11]: a.

sie reibt sich ihre augen/

b.

reibt die augen/

(AN, 3;10,18, G1) (VI, 3;4,26, G1)

To summarize this short overview of the corresponding German test items, the results can be considered to show a clear language separation. Both types of inalienable body part constructions do occur as assumed, but in German, the picture is less homogeneous and can be interpreted in two ways: 1) the children seem to lag behind their French production of these constructions since they produce fewer of the expected constructions in German or 2) the properties of German reÀexive (and non-clitic) pronouns combined with the Germanic preference of int-poss-constructions lead to a far less homogeneous picture than in French. The second interpretation appears to be the more plausible one, since the children know about the different options in German and are quite advanced with respect to age and development: except for the child PE who produces one target-deviant determiner omission in the spontaneous speech part, none of the bilingual children omits determiners in a target-deviant way in German any more. The next section tries to clarify the syntactic status of se in these constructions and whether it represents the possessor. Furthermore a tentative syntactic approach is proposed. 4.

A tentative syntactic account of inalienable body part constructions in child French

The ¿rst important question for a syntactic account of se in the inalienable body part constructions is whether it is itself the possessor argument or a functional category linked to it. According to Vergnaud and Zubizarreta (1992: 618) who label this kind of possessive construction “ext-poss construction” (see section 1

Clitics in the acquisition of inalienable possession in French

203

and their example (56) on leur a lavé les mains ‘one washed them their hands’). Vergnaud and Zubizarreta do not discuss in detail the question of the status of the dative complement aux enfants in Le médecin a examiné la gorge aux enfants (‘the doctor examined the children’s throat’) but refer to Kayne (1975) who argues that this complement is to be analyzed as a malefactive/benefactive complement of the verb (i.e. a complement affected by the action or state referred to) and not as a possessor moved out of the direct object and into the VP. Mentioning an argument from Rizzi (p.c.), Vergnaud and Zubizarreta argue in favor of a small clause representation which would, according to Rizzi, possibly explain the ¿xed order between the direct object and the dative complement (excluding *Le médecin a examiné aux enfants la gorge, see Vergnaud and Zubizarreta 1992: footnote 34). Such a proposal is also supported by Alexiadou (2002: 179) who proposes that the semantically and syntactically dependent possessors on the possessed noun form a complex DP together with it. Both proposals, however, explicitly work only with lexical entities and do not assume the possessor-related clitic to be a head which has been argued for above. A more interesting proposal which takes the clitic se and its particular characteristics (strong underspeci¿cation, various possible interpretations) compared to other object clitics into account is that of Labelle (2008) who argues that the French anaphoric and reciprocal se appear with verbs or predicates that are semantically transitive, that se does not absorb an argument and that no case absorption takes place. Rather, se is a Voice head which introduces the external argument of the verb to syntax and the referent of the object is determined on the basis of that of the subject. Labelle follows Kratzer (1996) in assuming that each structure with a verb has to contain a Voice head (a slightly different conception of the vP). In active sentences, the voice head introduces the type of external argument required by the predicate. Non-active sentences contain another type of Voice which does not introduce an external argument. This approach would correctly represent the aforementioned anaphoric reading of se (reference to the subject) and also avoid the problems with case that Schmitz (2006) reports from her data, in particular with respect to Dative. For the speci¿c use of se in inalienable body part constructions where se is considered to be the possessor of the de¿nite DP object, Labelle (2008) states that se is not an argument of the verb in the lexicon and therefore not lexically related to the dative. For the syntactic implementation, she refers to Pylkkänen (2008) who proposed two types of applicatives, introducing either a recipient (‘to the possession of x’) or a source (‘from the possession of x’). Since inalienable possessors denote entities in possession of the object, they fall into this class which Pylkkänen represents syntactically below the VP level but higher than the object of the verb (“low applicative”). According to Labelle, it is the

204

Katrin Schmitz

applicative head in this approach which adds a dative argument without which the VP would be saturated and not able to combine with se. The structure for Labelle’s example Luc se coupe les cheveux ‘Luc SE cuts the hair = Luc cuts his hair’) is the following (phrase level labels higher than VP added by the author): (26)

VoiceP

Luc

Voice’

Voice se

VP

V coupe

ApplP

Appl poss

les cheveux

The other clitic types identi¿ed by Schmitz and Müller (2008) can easily be integrated in this approach: subject clitics, being DP’s with a nominal layer (and thus early acquired) are inserted in Spec VoiceP, unlike non-reÀexive Dative and Accusative clitics which are ijP’s according to Gabriel and Müller (2005) and purely functional categories. As such, they would not be original arguments in this lexical ¿eld (everything below VoiceP) but inserted later in the functional ¿eld, possibly in TP, adjoined to the moved ¿nite verb. The crucial point of this analysis which covers only the lexical domain of argument/ applicative insertion is that reÀexive clitics (and other object clitics with potential reÀexive readings as e.g. me, te) are thus inserted much earlier and in a less complex domain (according to Jakubowicz 2002) which could explain the early and homogenous realization in the French-speaking children whereas they show more dif¿culties with non-reÀexive object clitics. There is positive evidence in favor of a tendency to apply an anaphoric interpretation of pronouns from the domain of subjects: Schmitz, Patuto and Müller (2011) analyzed subject omissions and realizations of bilingual FrenchGerman, Italian-German and Italian-French children (and their monolingual peers) and found also some “odd omissions and realizations” consisting of ¿rst or second person pronouns (clearly recognizable from the context in the case

Clitics in the acquisition of inalienable possession in French

205

of subject pronoun omission) combined with a third person verb form, i.e. syntactically third person. Schmitz, Patuto, and Müller (2011) show that all children misinterpret the deictic ¿rst and second person pronouns as anaphoric (as third person pronouns are). In German and French, these empty subjects are topics and result from a discourse licensing strategy that the children abandon later. In Italian, the null subject language, subject realizations of ¿rst or second person co-occur with a third person verb form like in the target-deviant utterance *Io ora scende (‘I now goes down’). Possibly, also the domain of objects is concerned and discourse licensing via anaphoric reading of pronouns is a temporary strategy. In inalienable body part constructions which favor the use of anaphorically interpretable object clitics this would then lead to early realization of se. Furthermore, the general approach by Kratzer (1996) implies that children have to know about event and argument structure. The fact that French and German speaking children initially omit subjects and objects does not prevent them from knowing about events. These omissions can be explained by their above mentioned strategy of discourse licensing of the arguments of the event expressed by the verb instead of a syntactic licensing. The semantic and syntactic adaptation to all types of se made by Labelle (2008) explains the anaphoric reading of se in inalienable body part constructions (ext-poss-type) which is already produced very early by French speaking children.

5.

Conclusion

The present paper has investigated the acquisition of Dative and ReÀexive clitics in monolingual French and bilingual German-French children by looking in particular at constructions in experimental settings inviting to the use of these clitics, namely inalienable body part constructions and double object constructions. Using the data from two elicitation tasks, the sample of the present paper is quite small but still allows some interesting results: – Both monolingual and bilingual children prefer lexical direct and indirect objects to clitic pronouns – as in the studies presented – and omit indirect objects more often and for a longer time than direct objects. – At the same time, reÀexive clitics are used throughout if possessors of body parts are produced. – There is no cross-linguistic inÀuence across the board as predicted.

206

Katrin Schmitz

These results are explained by an early preference for anaphoric pronouns (or an anaphoric interpretation of them) and a syntactic derivation which explicitly acknowledges this character of reÀexive se by considering it a Voice head, necessary for the introduction of an external argument. The possessor is then introduced by a low applicative projection according to Pylkkänen (2008). Since the present study is rather exploratory and worked with few data, more longitudinal data are needed for a developmental perspective, but inalienable body part constructions should also be tested with more and different possession structures.

References Alexiadou, A. 2002 Some notes on the structure of alienable and inalienable possessors. In: Coene, M. and Y. D’Hulst (eds.), From NP to DP, Volume 2: The expression of possession in noun phrases. Amsterdam/Philadelphia: John Benjamins, 167–188. Baetens Beardsmore, H. 1971 Le Français regional de Bruxelles. Brussels: Presses Universitaires de Bruxelles. Broekhuis, H. and L. Cornips 1997 Inalienable possession in locational constructions. Lingua 101, 185–209. Cardinaletti, A. and M. Starke 1999 The typology of structural de¿ciency: A case study of the three classes of pronouns. In: Van Riemsdijk, H. (ed.), Clitics in the languages of Europe. Berlin: Mouton de Gruyter, 145–233. Cardinaletti, A. and M. Starke 2000 Overview: The grammar (and acquisition) of clitics. In: Powers, S. and C. Hamann (eds.), The acquisition of scrambling and cliticization. Dordrecht: Kluwer, 165–186. Déchaîne, R.-M. and M.Wiltschko 2002 Decomposing pronouns. Linguistic Inquiry 33 (3), 409–442. Eichler, N. 2008 Frühkindliche Zweisprachigkeit: Argumentauslassungen bei bilingual deutsch-französisch aufwachsenden Kindern. [Early bilingualism: Argument omissions in bilingual German/French children]. Unveröffentlichte Magisterarbeit, Bergische Universität Wuppertal. Gabriel, C. and N. Müller 2005 Zu den romanischen Pronominalklitika: Kategorialer Status und syntaktische Derivation. In Kaiser, G. (ed.), Deutsche Romanistik – generativ, pp. 161–180. Tübingen: Narr.

Clitics in the acquisition of inalienable possession in French Guéron, J. 2006

207

Inalienable Possession. In: Everaert, M. and H. v. Riemsdijk (eds.), The Blackwell Companion to Syntax, vol. 2, ch. 35. Malden: Blackwell, 589– 638. Hulk, A. and N. Müller 2000 Crosslinguistic inÀuence at the interface between syntax and pragmatics. Bilingualism: Language and Cognition 3 (3), 227–244. Jacubowicz, C. 2002 Functional categories in (ab)normal language acquisition. In: I. Lasser (ed.), The Process of Language Acquisition. Frankfurt a.M.: Lang, 165– 202. Kayne, R. S. 1975 French Syntax: The transformational cycle. Cambridge, Mass.: MIT Press. Kayne, R. S. 2000 Parameters and Universals. Oxford: Oxford University Press. Kratzer, A. 1996 Severing the external argument from its verb. In: Rooryck, J. and L. Zaring (eds.), Phrase structure and the lexicon. Dordrecht: Kluwer Academic Publisher, 109–137. Labelle, M. 2008 The French reÀexive and reciprocal se. Natural Language and Linguistic Theory, doi 10.1007/s11049-008-9053-1 [downloaded August 25, 2010]. Lamiroy, B. 2003 Grammaticalization and external possessor structures in Romance and Germanic languages. In: Coene, M. and Y. D’Hulst (eds.), From NP to DP, vol. 2, Amsterdam: Benjamins, 257–280. Müller, N. and T. Kupisch 2007 Acquisition des déterminants et des clitiques objets chez des enfants bilingues (français-allemand). Acquisition et Interaction en Langue Etrangère (AILE), 25, 45–68. Müller, N., K. Schmitz, T. Kupisch and K. Cantone 2006 Null arguments in monolingual children: A comparison of Italian and French. In: Torrens, V. and L. Escobar (eds.), The Acquisition of Syntax in Romance Languages. Amsterdam: Benjamins, 69–93. Müller, N., K. Schmitz, T. Kupisch and K. Cantone 2011 Einführung in die Mehrsprachigkeitsforschung. Tübingen: Narr. Third, revised edition. Müller, N. and M. Patuto. 2009 Really competence-driven cross-linguistic inÀuence in bilingual ¿rst language acquisition? The role of the language combination. In: P. Bernardini, V. Egerland and J. Granfeld (Hgg.) Mélanges plurilingues offerts à Suzanne Schlyter à l’occasion de son 65ème anniversaire. Études

208

Katrin Schmitz Romanes de Lund 85. Språk- och litteraturcentrum Romanska, Lunds Universitet, 299–319.

Nadasdi, T. 1989

Déviation et simpli¿cation linguistique dans le français bruxellois. Journal of the Atlantic Provinces Linguistic Association 11, 1–18.

Nakamoto, T. 2009 Inalienable possession constructions in French. Lingua 120 (2010), 74– 102. Pylkkänen, L. 2008 Introducing arguments. Cambridge, Mass.: MIT Press. Rowlett, P. 2007 The Syntax of French. Cambridge: Cambridge University Press. Schmitz, K. 2006 Zweisprachigkeit im Fokus. Der Erwerb der Verben mit zwei Objekten durch bilingual deutsch-französisch und deutsch-italienisch aufwachsende Kinder. [Focussing bilingualism: The acquisition of verbs with two objects in bilingual German/French and German/Italian children]. Tübingen: Narr. Schmitz, K. and N. Müller 2008 Strong and clitic pronouns in monolingual and bilingual language acquisition: Comparing French and Italian. Bilingualism: Language and Cognition 11 (1), 19–41. Schmitz, K., M. Patuto and N. Müller 2011 The null-subject parameter at the interface between syntax and pragmatics: Evidence from bilingual German-Italian, German-French and Italian-French children. To appear in: Guijarro-Fuentes, Pedro and Jason Rothman (eds.), Interfaces in Child Language Acquisition. Special issue of First Language. Prepublication in OnlineFirst (10.05.2011), doi 10.1177/0142723711403880. Seiler, H. 1983 Possessitivity, subject and object. Studies in Language 7 (1), 89–117. Vergnaud, J.-R. and M. L. Zubizarreta 1992 The De¿nite Determiner and the Inalienable Constructions in French and English. Linguistic Inquiry 23, 595–652. Zushi, M. 2003 Null Arguments: the Case of Japanese and Romance. Lingua 113, 559– 604.

Children’s Acquisition of Subject Markers in isiXhosa Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

1.

Introduction

The study of African languages is a rapidly growing area of investigation in linguistics. It has become clear that there is a rich fund of information in the large number of African languages falling into several families that can inform the science of the universal and particular features of human language. Although the Bantu language family has no historical relation to the family of Romance languages, fruitful discussion has begun among researchers seeking to illuminate some of the special features that are shared (De Cat and Demuth 2008), especially in the area of concern here, namely clitics. But the de¿nition of clitic within Bantu is not without controversy. In order to understand how studying acquisition can contribute to the debates on the nature of Bantu morphosyntax, we open this chapter with a brief introduction of the various proposals regarding the status of subject markers (SM) and object markers (OM) in Bantu. 1.1 The status of the subject, object, SM and OM in isiXhosa IsiXhosa, being a typical Bantu language has 15 noun classes, normally referred to as genders in other languages. The numbering system of these noun classes is a result of systematic studies of nouns in Bantu starting in the 19th century and later developed by Doke (1954) and Meinhof (1948) amongst others. The nouns in isiXhosa are grouped in such a way that the singular and plural pairs fall into two adjacent noun classes. For example, the plural form of nouns in class 1 is labeled as class 2 and the plural form of nouns in class 3 is labeled as class 4. The gaps in the cataloging are a consequence of the uniform labeling of nouns across Bantu languages. This uniform labeling system

210

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

makes comparisons between the different Bantu languages more systematic. See Table 1 for a complete list of noun classes in isiXhosa. Table 1. IsiXhosa noun classes and their grammatical markers. Noun Class 1 1a 2 2a 3 4 5 6 7 8 9 10 11 14 15

Noun umntu umama abantu oomama umvundla imivundla iblomu amablomu isihlangu izihlangu incwadi iincwadi uthando ubusi ukutya

Noun Pre¿x umuabaooumimiiliamaisiiziiniiuluubuuku-

Subject Marker uubabauiliasiziizilubuku-

Object Marker -m-m-ba-ba-wu-yi-li-wa-si-zi-yi-zi-lu-bu-ku-

Absolute Pronoun yena yena bona bona wona yona lona wona sona zona yona zona lona bona kona

Gloss person mother people mothers hare hare (Pl.) Àower Àowers shoe shoes book books love honey food

The basic word order for isiXhosa is SVO. The verb hosts a number of pre¿xes and suf¿xes, amongst which fall the causative, applicative, reciprocal and passive. The ¿nal ending of the verb, often called the ¿nal vowel, also marks tense, aspect, modality and negation. Af¿xes associated with agreement, tense, aspect and mood are also found on the left edge of the verb. The subject marker and the object marker make it possible to move the lexical arguments of the verb to various positions in relation to the verb and the other verbal arguments. The examples below illustrate these facts. (1)

Umama ufund- ela abantwana incwadi 1.mother 1.SM- read- APPL- FV 2.children 9.incwadi1 ‘Mother is reading the children a book’

1. SM = Subject Marker; OM = Object Marker; APPL = Applicative; FV = Final Vowel; TNS = Tense; A number (as in 1.SM) indicates noun class membership

Children’s Acquisition of Subject Markers in isiXhosa

211

(2)

Umama uyabafund1.mother 1.SM- TNS- 2.OM- read‘Mother is reading to them’

ela APPL- FV

(3)

Umama akaba- fund- eli 1.mother 1.NEG.SM- OM- read- APPL- NEG ‘Mother is not reading to them’

(4)

U- ya - ba- fund –el- a incwadi abantwana umama SM- TNS- OM- read- APPL -FV 9.book 2.children 1.mother Lit ‘She-them-read-to book children mother’ ‘Mother is reading the children a book’2

In sentences (1–4), the subject umama enters into an agreement relation with the verb. A similar relationship between the direct object abantwana and the verb is expressed in sentences (2–4). The relation is expressed through the subject marker (SM) and the object marker (OM) encoding features of person, number and class (gender). The order and position of the SM and OM is ¿xed. Whereas the SM is obligatory in isiXhosa, the lexical subject and OM are optional. The structural representation of the verb and its arguments that we adopt in this study follows from Koopman and Sportiche’s (1991) proposal. In this framework the arguments of the verb are merged within the VP shell. The movement of the verb and its arguments is driven by the need to check features (Rizzi 1997; Chomsky, 1995). The subject is assumed to start off in spec v (See Figure 1). Since the subject DP is the closest, it is the one that is attracted to Spec, AgrS to check the phifeatures which are spelled out as SM on AgrS. The movement of the subject is triggered by EPP (subject requirement) on AgrS. As will be described, there are a number of positions associated with the subject in Bantu. The differences rest on whether the subject is regarded as a Topic, in which case it would be in Spec Top (above AgrSP in ¿gure 1). If it is a standard subject it can be on spec AgrS or Spec T, depending on whether one adopts the split inÀection hypothesis or whether the various inÀection markers are merged on T. For the purposes of this paper, we adopt the split inÀection framework as it better captures the order of these af¿xes in Bantu.

2. Italics added for emphasis

212

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers AgrSP

DP/pro

AgrS’ AgrS

TP

SM

T’ T

vP

v’ v

VP V’ < V>

DP

Figure 1. IsiXhosa subject and verb movement.

The verb also has features such as tense, aspect and mood that need to be valued. The verb then moves through various heads driven by these features and sensitive to locality constraints on movement. The verb which bears inÀectional features associated with tense, aspect and mood, moves through the heads, including v which has been proposed to host argument-changing af¿xes such as causative, applicative, reciprocal and passive attested in Bantu languages (Baker 1988). This feature-checking relationship and the realization of the SM and OM features is what this study seeks to explore. The ¿rst question that arises with respect to Bantu relates to whether the SM and OM are reduced pronouns (pronominal clitics) or agreement markers. The second question has to do with the status of the subject in Bantu. Based on the literature, the question about the status of SM/OM can be answered in several ways. Some scholars treat the SM and OM as pure agreement markers (see Buell 2005 (for Zulu), Deen 2006a (for some varieties of Swahili) amongst others). Other scholars subscribe to the notion of SM being ambiguous between pronominal clitic and agreement marker whereas the OM is unambiguously a pronominal clitic (Bresnan and Mchombo 1986 and 1987; Keach 1995; but see Woolford 2000 for an opposing view). In the third approach the SM and OM are both treated as pronominal clitics (Zwart 1997; Zeller 2008). Let us ¿rst look at how the three approaches differ structurally ((2) a; b from Deen 2006b: 226). The tree structure in (2)a represents the possibility of SM as an agreement marker.The SM is the head of AgrS, and the subject occupies the spec position. In this instance the SM functions purely as an agreement marker which enters

Children’s Acquisition of Subject Markers in isiXhosa

213

into an agreement relation with the subject in spec AgrS. If the subject is null, pro retains the appropriate phi features and licenses SM. The tree structure in 2(b) represents the possibility of SM as a pronominal af¿x. SM occupies the subject position in spec AgrS. AgrS is a null head in this instance. The SM has a pronominal function in this case. In this approach a lexical subject is represented as occupying the speci¿er of the Topic node which is above AgrSP. The SM is in spec AgrS, which performs the function of the subject. In this position the SM bears the theta role associated with the topic. A further possibility exists, combining 2(a) and 2(b). In this con¿guration the SM is in AgrS, but the subject, which can either be a lexical subject or a null pro, occupies the speci¿er of Topic. That is, the Topic could license the SM either in AgrS or spec of AgrS. The possibilities have fueled studies on the nature of the subject in Bantu. The treatment of subjects as a topic in Bantu gathered momentum after Bresnan and Mchombo’s (1986) proposal that the SM in Chichewa is ambiguous between the subject agreement and anaphoric agreement. The OM on the other hand, is treated as a pure incorporated pronoun in Chichewa. The central claim of this analysis is that there is no grammatical relation between left-dislocated topics and the incorporated clitic pronoun. The association between the subject and the SM obtains through an anaphoric binding relation. Let us now look more closely at the three main approaches to SM and OM in Bantu. The view of SM and OM as agreement markers has received more attention in the literature. Buell’s (2005) study of Zulu morpho-syntax discusses the interaction between SM, tense, mood and aspect. Buell concludes that since subject markers interact3 with the various inÀectional heads, they require an agreement analysis because pronouns do not interact in this way with inÀec3. For example, isiZulu marks compound tenses by means of an auxiliary verb and a lexical verb. The auxiliary and lexical verb both bear a subject marker: Wa(1.SM)be e-(1.SM)hambile (he/she had left). The SM in this sentence is different from SM

214

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

tion. In addition, using evidence from non-agreeing subjects, Buell concludes that these subjects are not in Topic position. Buell further provides evidence for the treatment of OM as an agreement marker, based on VP-ellipsis in Zulu (Buell 2005). Buell notes that under VP ellipsis, the OM in Zulu becomes obligatory. Using linear order and constituency tests, Buell further suggests that the position of the subject and object in both agreeing and non-agreeing subject sentences is the same. He claims that the various alterations observed between the SM and OM are driven by constituency rather than a focus feature (Buell 2005)4. Zwart (2000) on the other hand proposes that SM and OM are pronouns rather than agreement markers in Swahili, a view which is opposed by Deen (2006b). Zwart argues that: The widespread use of the subject agreement marker outside of verbal morphology suggests that the subject agreement marker is not an agreement marker in the strict sense, i.e. an af¿x with no other function than to mark the congruence of the subject and the verb. Rather, the distribution of the agreement marker suggests that its status is closer to that of a pronoun. Zwart (2000:4)

Zwart compares the SM to personal pronouns in Swahili. He proposes that if SMs are pronouns, then they are not personal pronouns but rather resumptive pronouns which are comparable to clitics or bound demonstratives found in other languages (Zwart 2000). In addition, Zwart points out that the nature of the various tense and aspect markers can be used as evidence against an analysis of SM as an agreement marker. He claims that because the tense morphemes have a lexical origin, they still act as verbs which host the proclitic subject markers as well as the enclitic relative marker. The strongest argument against an agreement analysis is put forward by Zeller (2008) in a study of isiZulu that examines the lack of agreement between the verb and the subject when the expletive is used. Zeller is of the view that the SM is only licensed if the subject has moved out of vP. The difference between Zeller and other proponents of the pronominal analysis of SM is that by treating SM as an antifocus marker, he is able to do away with stipulations associated with lack of agreement in locative inversion as well as pro-drop in Bantu. These stipulations were used as support for the topic analysis of subjects in the Indicative positive, suggesting an interaction with the remote past morpheme(See Buell 2005:51). 4. See Buell (2005) chapter 5 for a full discussion of agreeing and non-agreeing subjects and object.

Children’s Acquisition of Subject Markers in isiXhosa

215

in Bantu. Zeller proposes that the SM in Bantu is not a morphological reÀex but rather a case of clitic doubling. The SM doubles the subject whenever the subject has moved into spec T. More importantly the SM is treated as a functional nominal head n* which is merged with the subject before the subject moves to spec T. In this analysis, the SM is regarded as a determiner-like element which is the highest functional head of the DP, which may or may not have a DP complement. Since SM is a functional element, it can move independently of the subject. This movement is determined by the features5. When the SM is bare (does not have a DP complement), the SM bears the theta-role associated with the subject, which he proposes is overtly realized as the SM, giving rise to a pronominal interpretation (Zeller 2008). Zeller concludes that “what has traditionally been called subject “agreement” in Bantu is rather a case of clitic doubling whose effects on the information structure of the sentence are comparable to those that have been observed in other clitic-doubling languages such as Albanian and Greek” (Zeller 2008: 247). As a close relative of isiZulu, isiXhosa might be subject to a similar treatment. These possibilities are represented in Table 2. The issues of the status of OM are also unresolved (Bresnan and Mchombo 1986; Buell 2005), but there is perhaps more agreement that it should be regarded as a pronominal clitic. Buell (2005) provides a summary of the complex conditions of this use where he also discusses its interaction with the so-called long and short forms of tense in isiZulu. Table 2. The proposed Positions for SM in Bantu Language Proposal Chichewa Bresnan and Mchombo Chichewa Bresnan and Mchombo isiXhosa, Visser, isiZulu, Buell, Swahili Deen Swahili Zwart isiZulu

Zeller

Position of SM Head of AGR Head of AGR Head of AGR

Position of lexical subject Spec of Topic n/a

Position of empty subject Spec of AGR n/a

Spec of AGR

Spec of AGR

Spec of AGR Head of T(InÀ)

Spec of Topic Spec of T(InÀ)

n/a pro IS SM

Pronoun or agreement? Agreement Anaphoric pronoun Agreement

Anaphoric pronoun Pronominal clitic/ (clitic doubling)

5. See Zeller (2008) for an elaborate discussion of the anti-focus features associated with SM in Bantu.

216

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

1.2

The absolute pronoun in isiXhosa

So far, we have brieÀy described the relation between the SM, OM and lexical subject and object. IsiXhosa also has a set of pronouns, often referred to as absolute pronouns in Bantu. The absolute pronoun may occur with or without the lexical subject in isiXhosa. A list of absolute pronouns is given in Table 1 above. The absolute pronoun in isiXhosa has been classi¿ed as one of the nominal modi¿ers which behave in the same manner as quanti¿ers (Du Plessis and Visser 1992). Visser (2008:18) classi¿es the absolute pronoun as a nominal modi¿er “with an inherent lexical semantic de¿niteness property.” As such, the absolute pronoun has relative freedom in terms of its position in relation to the subject; that is, it can Àoat away from its head. Let us look at some examples illustrating the distribution of the absolute pronoun in isiXhosa. (5)

Abazali bona bayasebenz- a (…abantwana bayafunda) 2.parents 2.them 2.SM- TNS- work- FV ‘Parents, as for them, they work’ (..the children study)

(6)

Bona abazali bayasebenz- a 2.them 2.parents 2.SM- TNS- work- FV ‘As for them, the parents, they work’

(7)

Abazali bayasebenz- a bona 2.parents 2.SM- TNS- work- FV 2.them ‘The parents, as for them, they work’

(8)

Bona bayasebenz- a 2.them 2.SM- TNS- work- FV ‘As for them, they work’

(9)

*Bona ya-sebenz-a 2.them -TNS-work-FV Intended: “they work”

In sentence (5) the absolute pronoun occurs in a post-nominal position. In (6) it occurs before the head noun to further emphasize the noun. In (7) the absolute pronoun occurs post-verbally. Sentence (8) illustrates the use of the absolute pronoun where the lexical subject is omitted. Finally, sentence (9) shows that the absolute pronoun cannot be used independent of the SM. In terms of features, the pronoun agrees with the noun in number (singular/ plural) and by noun class (gender) in the same way the SM and OM do. How-

Children’s Acquisition of Subject Markers in isiXhosa

217

ever, the absolute pronoun differs from the SM and OM in a number of ways. Structurally, the position of the absolute pronoun in relation to the noun it modi¿es is the same, regardless of whether the subject is treated as a ‘subject’ or a ‘topic’. As noted above, this may not be the case with SM. Secondly, whereas the SM interacts with inÀection (such as tense, aspect and mood), the absolute pronoun does not. Finally, unlike the SM, the absolute pronoun is optional. What is relevant for our purposes here is that in the absence of a lexical subject, as in sentence (8), the absolute pronoun provides additional information about the subject. In a sense, it adds the same information (features) as the SM. 1.3

How work on acquisition of isiXhosa might contribute to the debate about SM and OM.

The preceding brief introduction to the competing treatments of SM and OM in the literature shows that the status of SM and OM as clitics is not a straightforward one. Outside the con¿nes of Bantu studies, various researchers have demonstrated several structural connections between Bantu and Romance languages (Cardinaletti 2008; Hartford 2008; Labelle 2008; and Marten, Kempson and Bouzouita 2008, amongst others). Of particular interest is the parallelism drawn between the subject and object clitics, in terms of structural position, features, and verb movement. Although these studies assume different theoretical frameworks, the conclusion is that the freedom regarding word order possibilities in Bantu and Romance languages can be accounted for by using the same formal tools. We hope that the study of comprehension of SM and OM in Bantu can contribute further to the debates on the structural connections between languages that share certain features. The remainder of this chapter is as follows: First, we will provide some details of how these subjects and object markers are acquired in isiXhosa by L1 learners. We will call them “markers” because the status of these forms as clitics versus agreement af¿xes is unresolved. However, we believe that the acquisition data might contribute insights to the theoretical discussions. Since the work has only begun, the account will be preliminary, but we plan to map out the path that research on this question will need to take, especially to integrate work on naturalistic speech with experimentation. Work on many languages has incorporated experimental studies with young native speakers to illuminate the nature of their grammars. Experimental work on the acquisition of African languages is very much in its infancy, but we hope to provide a road map of what can be achieved by taking a broader approach. Secondly, we discuss the path of acquisition of noun classes (NC), subject markers, (SM) and object markers (OM) in children learning isiXhosa as a ¿rst

218

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

language, though OM seems to be used too rarely in our samples of children’s speech to speak with con¿dence about its course. Then we report on experimental studies that investigate whether SM in isiXhosa behaves like agreement af¿xes in other languages or whether they behave more like pronouns. We also report on an experimental study which tested whether children can retrieve number information about the subject and object from verbal morphology alone. We conclude the chapter by proposing future work that might contribute to debates about the nature of these markers cross-linguistically.

2.

The acquisition of NC, SM and OM in isiXhosa-speaking children

In our investigation of children’s acquisition of SM and OM in isiXhosa we analyzed samples of their spontaneous speech between ages 12 and 39 months (Gxilishe, de Villiers, and de Villiers 2007a, 2007b). The spontaneous speech data came from longitudinal samples of conversation from monolingual isiXhosa-speaking toddlers in a township outside of Cape Town collected every one to two months – for ¿ve children from 12 to 28 months, and for another 6 children from 24 to 39 months. The children were recorded in naturalistic settings interacting with a familiar adult research assistant whose ¿rst language was isiXhosa. Transcripts of the speech of all of the participants in the conversations were made and checked by two native-speakers of the language. The transcripts were combined into 6-month age bands to generate suf¿cient utterances for reliable analyses of stages of acquisition of different syntactic features. For the one-year-old cohort, there were 1155 child utterances in the total sample; for the two-year-olds a total of 1485. For each utterance containing a lexical verb we coded the noun class of the target subject and/or object, whether the nouns were explicitly expressed or not. If a subject or object noun was present we noted whether the children marked its class with the correct NC pre¿x. This analysis did not include cases where the subject nouns required copulative pre¿xes, which replace the NC pre¿x in isiXhosa. Whether the subject noun was explicitly expressed or not, we coded whether the child correctly provided the obligatory subject marker for that noun class. Totaled across all the children there were 295 obligatory contexts for subject agreement spread across the different age bands, varying from a low of 36 obligatory contexts for age 18–24 months to a high of 87 for age 24–30 months. For each child at each age band we calculated the percentage of correct subject markers provided in obligatory contexts, the percentage of markers omitted, and the percentage of contexts in which the wrong NC marker was supplied.

Children’s Acquisition of Subject Markers in isiXhosa

219

Figure 2. Acquisition of Noun Class and Subject Agreement Markers (from Gxilishe et al. 2007a)

Gxilishe et al. (2007a; b) report several important ¿ndings. First, across the age range of 12 to 39 months there was steady and parallel development in supplying the correct NC pre¿x on the nouns and the target SM on the verbs (see Figure 2), with each reaching a mean close to 80% correct use in obligatory contexts at age 36–39 months. Second, there was no difference at any age in the likelihood of the children supplying the correct SM on the verb whether the subject was explicitly expressed or not (see Figure 3). Across all the children at all ages, the correct SM was supplied 31.8% of the time when the subject noun was not expressed versus 27.8% of the time when the subject noun was present. Similarly, the SM on the verb was omitted 22.7% of the time when the subject noun was absent versus 18.2% of the time when the noun was expressed. This argues against any account in which the SM is “copied” from the subject noun onto the verb, at least on the surface. In addition, the SMs did not appear to be acquired in any piecemeal lexical item-by-item process, neither verb by verb nor noun class by noun class. Correct marking of subject noun class on the verb increased in a probabilistic fashion across many verb roots and noun classes at the same time. For example, between 24 and 30 months, the six individual children correctly produced SMs for 4 to 7 noun classes on between 5 and 16 verb roots. And for two-thirds of the verb roots to which they correctly attached SMs, the children also omitted them on those same verbs. Similar gradual and probabilistic acquisition of subject marking was reported by Deen (2005) for Swahili.

220

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

Figure 3. Acquisition of Subject Agreement Marking on the Verb with Explicit or Empty Subjects (from Gxilishe et al. 2007a)

Finally, errors in which the wrong SM was substituted (i.e. the marker for a different NC) on the verb were remarkably rare, as reported previously by Demuth (2003) for Sesotho and Siswati: 139 out of 143 errors observed across all the children in obligatory contexts for subject agreement were errors of omission in which no marker was provided (97.2%). In contrast, errors of this type predominate in adult learners of isiXhosa as a second language. The children’s data are certainly compatible with the position of Du Plessis and Visser (1998) that the morpheme in isiXhosa is a subject agreement marker with either an explicit or null subject, in that there is no difference in the children’s likelihood of supplying the SM on the verb as a function of whether the subject is overt or null. Deen (2005, 2006) draws a similar conclusion for the status of the agreement pre¿x in child data from the Nairobi dialect of Swahili. But the argument is not decisive: the form could still be a pronominal clitic. There is some minimal data on OM in a paper on children’s knowledge of the long and the short tense in isiXhosa, in Gxilishe et al. (2007b). They tested whether the same young speakers of isiXhosa knew the conditions under which the so-called long and short forms of the tense were obligatory in isiXhosa, some of which rely on having an OM. There were several appropriate uses of OM in the children’s speech, that is, they used OM with the long but not the short form. However, the sample of such utterances was too small to constitute a study of OM in itself.

Children’s Acquisition of Subject Markers in isiXhosa

2.1

221

Previous experimental work

Experimental studies of isiXhosa-speaking children have just begun, and their focus has been to investigate whether SM in isiXhosa behaves like agreement suf¿xes in other languages (English, Spanish), or more like pronouns. It seems to be true that languages move from having overt pronouns, to having pronominal clitics, to having agreement af¿xes (Givón, 1976). English has been argued to have undergone this process, and spoken French may be in transition to agreement af¿xes from pronominal clitics (the 3rd person singular forms il, elle (he,she)) (Legendre, Culberston, Barriere, Nazzi and Goyet, 2010). In question is the status of the SM marker in isiXhosa: does it behave more like agreement, or more like a pronoun, in children’s comprehension? First, there is important experimental work in developmental psycholinguistics to set the stage for the debate. A study by Johnson, de Villiers and Seymour (2005) on English has suggested that young English children do not seem to use the agreement marker on the verb to establish subject number. They designed a set of stimuli to disguise the number marking on the noun by having the verb begin with an /s/, so that the third person /s/ or its absence was the only cue to choosing the plural versus singular subject picture, e.g. (10) The cats sleep in the bed. (11) The cat sleeps in the bed. Children aged three or four, and many ¿ve-year-olds, were at chance in the picture choice task on sensitivity to the 3rd person /s. By about age 5–6 years, children began to show sensitivity, but only to the singular verb form /s/ (11), not the plural /Ø/ (10). But English has a weak system of agreement, and the condition of the experimental setup is very atypical, as the marking is usually redundant with the noun marking. Pérez-Leroux (2006) repeated the experiment in a pro-drop language, Dominican Spanish, but her 3 and 4-year-old Spanish-speaking children also could not use the verb marking to choose the right picture even when the subject noun was null (pro-drop), and the verb marking was an essential cue. Only at around age 5–6 did the Spanish children begin to show sensitivity, but then only to the plural marking. In both English and Spanish, the paradigm is asymmetrical: English has /Ø/ for the plural, and Spanish has /Ø/ for the singular. This was interpreted by Pérez-Leroux as evidence that children were in fact sensitive to overt marking (–s on third singular verbs in English, –n on third plural verbs in Spanish).

222

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

Brandt-Kobele and Höhle (2010) tested German-speaking children on verb agreement inÀections using a picture choice task. Although German is not a pro-drop language, the personal pronouns for 3rd person singular female (sie) and 3rd person plural (sie) are homophones, making the sentence ambiguous unless the inÀection marker of the verb is parsed. The verbs were either inÀected for 3rd person singular (-t) or 3rd person plural (-n), (12) Sie fütter-t einen Hund. Pronoun-3SG feed-3SG a dog “She is feeding a dog.” (13) Sie fütter-n einen Hund. Pronoun-3PL feed-3PL a dog “They are feeding a dog.” In their Experiment 2, German children at 36–48 months showed no evidence of distinguishing plural and singular pictures on the basis of the verb inÀections on the verb in the sentence presented. But, the main point of their paper was to argue that the picture choice methodology may entail task demands that interfere with children’s comprehension. Their experiments used an eye-tracker to track the children’s eye-movements as the sentence was presented, on the assumption that the gaze response is absolutely minimal in its demands. The results of their Experiment 1 reveal that the children at 3–4 years did respond differentially to the verb agreement, in that they moved from a strong bias to looking at the plural picture (argued to be informationally richer) towards the singular picture in the case when a singular agreement was presented. In Experiment 2 on pointing, they also tracked eye-movements and showed a similar trend but not nearly as strong, and interestingly, a dissociation between eyegaze and pointing. Their paper raises important questions about the methodology of picture choice, which might indeed introduce response selection demands that go beyond comprehension. As in other areas of infant development, there is considerable disagreement about what eye gaze reveals, and how “deep” the understanding goes. For example, some scholars have argued for a dual system of processing (e.g. Apperley, 2010), in which the ¿rst level (Level 1), captured in eye gaze, is a kind of automatic registration of signi¿cant stimuli, but perhaps insuf¿cient for the child to make a decision or to engage or drive the motor response systems such as pointing. Level 2 is the level of integration of that information with the other response systems.

Children’s Acquisition of Subject Markers in isiXhosa

223

Legendre, Barriere, Goyet and Nazzi (2010) studied comprehension in children acquiring French, which unlike the other three languages discussed, can be argued to have pre-verbal agreement marking, under one analysis of weak subject pronouns (Legendre et al. 2010a). They argue that in the languages studied (English, Spanish, German) overt verbal agreement is marked by a single consonant in (word-¿nal) coda position. They point out that numerous studies have found evidence that perception of consonants is affected by their position in a syllable and in a word, and cite Swingley (2009). Swingley characterizes some of the challenge as follows: “Word-¿nal consonants are, in general, less clearly articulated; they are heard only after perception of the initial parts of the word has led children to consider an interpretation; and they enjoy less of the (hypothesized) bene¿t of membership in dense phonological neighborhoods.” (266). For this

reason, studies of languages with preverbal marking may contribute to the understanding of what morphosyntactic features children can detect. In modern spoken French, inÀected verbs in a frequent verb class do not distinguish person and number in their spoken form, except for ¿rst and second person plural. According to Legendre et al, the preverbal clitics are distinct for all persons and number, and have, therefore, taken up this function. This fact was exploited to create stimuli in which the number was carried only by the liaison between the ¿nal /z/ on the pronoun “ils” (they) versus “il” (he). They chose verbs that the children knew already, but as a condition they had to be vowel-initial, with phonologically identical third person singular and third person plural forms, number agreement being only signaled by liaison between the pronoun and the verb. Furthermore, in a clever design, they motivated the looking by using nonsense words and objects (“le voube”, “le taque”) so when the stimulus was presented, the participants were motivated to identify the novel object using the cue from subject number e.g. (14) Il embrasse le voube He kisses the voob or (15) Ils embrassent le taque. They kiss the tak Pro¿cient users of French make use of their knowledge of il (singular) versus ils (plural) to assign a singular or plural interpretation. The rather astonishing ¿nding of this study is that French 30-month-olds (but not 24-month-olds)

224

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

could look appropriately in these circumstances based on the cue from liaison, an average of 6.8% difference from baseline looking. This is more robust than the ¿nding in Brandt-Kobele and Höhle (2010) and at a younger age. Some difference might be attributed to the dynamic nature of the events presented in the French task, whereas in the other picture choice studies the pictures were static and sometimes movement had to be inferred (Legendre et al., 2010b). But the primary attribution made by the authors is that the success is because the critical stimulus is at the front of the verb, not the end. Like Brandt-Kobele and Höhle (2010), Legendre et al. also look at pointing behavior in the same age group, using an experimenter to encourage pointing at the scenes but minimizing nonverbal cues. The authors asked whether the 6.8% increase in looking time observed between baseline and test in IPLP revealed a suf¿cient understanding of number agreement to guide a pointing response, assuming that the pointing task also requires both decisional and motor planning. The 30-month-olds in French, unlike any other study to date, showed an ability to point at the right matching picture. The level of performance was not high, but above chance (average 61%)6. Setting aside the thorny question of what eye-gaze reveals about comprehension, we have a pattern of failure in 3- to 5-year-olds in explicit picture choice tasks in three languages that have ¿nal verb agreement, and one moderate success, in French. The complication that arises is the dispute over the nature of the initial element in French, the weak pronoun, traditionally considered a cliticized pronoun, but which Legendre et al. (2010a) argue has evolved into an agreement marker, an af¿x, in spoken French. If it were instead a pronoun, could that explain the superior performance? Some preliminary data testing pronouns in English speaking children aged 3–5 years revealed much better success in comprehending the number carried by overt pronouns than on 3rd /s/ (Gxilishe, Smouse, Xhalisa and de Villiers 2009). Here we discuss results of similar experimental work on isiXhosa, in the hope of illuminating the issues with yet another language. If isiXhosa is a pro6. A secondary analysis of the published data computed the sensitivity to the singular marker used in Johnson et al. (2005) and Gxilishe et al. (2009), to compare with the French results in Legendre et al. (2010b). Using the same index, namely choosing the singular picture when it was appropriate/total choice of the single picture, the data are not so far apart. However, since the ages vary, it is hard to compare the outcomes. For eyegaze in French at 30 months, sensitivity is .54, for pointing, it is .58. We do not know if by these indices the results are different from chance =.50. In Johnson et al for Mainstream English at age 3, the sensitivity using the pointing task was .47 and at 4, .56 ; for isiXhosa (Gxilishe et al. 2009) it was .56 for 4 year olds using pointing. None of these were signi¿cantly different from chance.

Children’s Acquisition of Subject Markers in isiXhosa

225

drop language, then SM should be a salient and important cue to the subject, as agreement is in Spanish. Unlike English and Spanish, the cues to plural and singular in isiXhosa are both overt, that is, zero morphology for one function is not found. However, morphology is complex because of the large set of noun classes. It can be argued that isiXhosa shares some of the properties of French but within a much richer agreement paradigm. Unlike the rare cue from liaison in French, the SM is almost invariably present in isiXhosa, but also at the beginning of the verb. 2.2

The comprehension of SM and Pronouns in isiXhosa

De Villiers and Gxilishe (2009) and Gxilishe et al. (2009) asked whether children speaking isiXhosa could use the SM marking on the verb to determine subject number. It is a much more dif¿cult problem in the case of isiXhosa, given the variety of forms and their dependence on noun class. The argument was that if SM is more pronominal in nature, then perhaps isiXhosa children would be able to predict the subject number from the SM alone, unlike the cases of the other languages. Of course, it is also different in that SM in isiXhosa is preverbal (like spoken French, according to Legendre et al. 2010a), and the agreement in English and Spanish consists of postverbal suf¿xes. Gxilishe et al. (2009) tested whether SM for children speaking isiXhosa patterned like subject agreement or like pronouns. They predicted that if it behaves like subject agreement, then children might show poor sensitivity to subject number carried only by SM. If it behaves like pronouns, then children might show good sensitivity to subject number carried only by SM. They also tested OM using a similar design, to see if number agreement could be carried by the agreement morphemes attached to the verb. Though Buell (2005) argues that they have properties that qualify them as object agreement morphemes, most theorists of Bantu consider these object morphemes to be pronominal clitics. In particular, when the morpheme co-occurs with a lexical object, the lexical object is displaced out of the clause, suggesting that a Principle A violation is being avoided. The question of how children retrieve number from object markers has not been tested previously. The study was designed as an experiment to test whether children can retrieve number information about the subject and the object from the verb morphology alone. This was done as in Johnson et al. (2005) and Pérez-Leroux (2006) using 8 sets of pictures for the subject marking test and 8 sets for the object marking test. The pictures were selected based on frequency of use of verbs associated with them amongst children aged between 4 and 6 as well as on cultural appropriateness. The pictures represent the most common variety of

226

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

singular-plural agreement pairs in isiXhosa. The participants involved in this study were 38 isiXhosa speaking children aged 4 to 6 years attending daycare or preschool in the townships near Cape Town. The researchers were native speakers of isiXhosa who tested the children individually in their schools. After some warm up pictures to ensure that the children could attend and point to pictures, the participants were asked to listen to a sentence and then choose the picture that corresponded to the sentence that they heard. The children usually spontaneously repeated the sentences without any problems, and produced the morphemes in question in their repetition. They received general positive feedback regardless of their choices. In the SM test, for example, one picture showed one rabbit snif¿ng at Àowers, and a second picture showed two rabbits snif¿ng at Àowers. After indicating the rabbits and the Àowers to bring them into the discourse, the child was told to “point to the picture where”: (16) Unukisa amablomu 3.SM- sniff Àowers “It sniffs at the Àowers”. The task was very similar for OM. For instance one picture showed a woman watering a single Àower, and one showed a woman watering three Àowers. After saying about the pair of pictures: (17) Jonga… Oo-mama, … ama-blomu See 2a-women 6.Àowers “See… Women, Àowers”. The child is asked to ‘show the picture where’: (18) Umama

u-

1a.mother

ya- wa –

1a.SM-TNS- 6.OM-

nkcenkceshel-

a

waters –

FV

“Mother waters them”. The alternative question would be: (19) Umama

u-

ya-

li-

nkcenkceshel- a

1a.mother 1a.SM- TNS- 5.OM- waters –

“Mother waters it”.

FV

Children’s Acquisition of Subject Markers in isiXhosa

227

The responses were analyzed in two ways, as in earlier work (Johnson et al. 2005) First, the number of correct choices was noted. Then the children’s sensitivity was coded, namely, whether the child picked the singular picture only when the singular sentence was presented. Sensitivity is the number of singular pictures chosen when singulars were presented divided by the total number of times a singular is presented, and likewise for the plural. Chance performance is therefore a sensitivity of .5. This latter index takes account of bias say, towards the plural picture. Given these three properties: isiXhosa has rich agreement, it is a pro-drop language in which the SM carries essential information, and the marker is preverbal (Legendre et al. 2010b) one might expect performance to be better than the languages studied to date, or at least on a par with French. In fact, average sensitivity hovered around chance. Although there was a signi¿cant change in sensitivity to the singular forms between ages four and ¿ve, the level of performance at each age was no greater than .5 by a single sample t-test. This was the case for both singular and plural forms. For the morphemes carrying object information, both singular and plural sensitivity was at chance for each age group, and the children did not seem to get better with age over this age range. Neither was it true that some children could do the task and other children could not. The data were distributed as one would expect by chance. The isiXhosa-speaking children’s data do resemble the data from English-speaking, Spanish-speaking and German-speaking children in that the children showed no sensitivity to the morphemes on the verb that carry number agreement with the subject, at least in an overt decision task. Gxilishe et al. (2009) speculate that these data could contribute to debates about the nature of the SM marker in isiXhosa, in that they resemble data from agreement markers (suf¿xes) in other languages, rather than pronouns. However, several nagging questions remained, especially in the light of the superior results in other languages from eyegaze tasks, and these are addressed in new work (Smouse, submitted). First, it needed to be established how isiXhosa speaking children would behave with absolute pronouns, rather than SM alone. If isiXhosa speaking children proved insensitive to number agreement on strong pronouns, the result would cast doubt on the meaning of the result with the SM forms. Clearly, children should be tested on both the SM and absolute pronouns. When an absolute pronoun is used in isiXhosa, the information in the SM is essentially duplicated.

228

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

The second issue is methodological: is the picture choice task at fault here? Consider the simple case, where you, as participant, are asked to point to a picture that shows (20) The duck swims on the pond. The plural picture also has a duck swimming on a pond, but you, as an adult, recognize that the singular picture is a better choice. Similarly, if you are asked which picture shows: (21) The ducks swim on the pond. You could take the singular picture as a duck representative of a plurality, though again, there would be a calculation that the experimenter probably wants you to choose the best example. So picture choice tasks inevitably entail something more than grammar, something along the line of Gricean implicatures. We are reluctant to ascribe all of the effect to this problem, just because English-speaking children succeed with pronouns, for example, and when asked to describe the pictures, always provide the right subject number. But it is important to try other, perhaps more sensitive, methodologies. Eye-tracking, though producing subtle results (Brandt-Kobele and Höhle 2010; and Legendre et al. 2010b), was not practically feasible, nor would it contribute to the larger goal of helping to provide language assessments for children who speak languages like isiXhosa. 2.3

Act-out comprehension as an alternative method

Smouse (submitted) adopted an act-out methodology instead of picture choice, with the same age group of 4–6 year old isiXhosa speakers (N = 37). In this new method, children were given a choice of plastic cut-out ¿gures and props to manipulate. Children’s comprehension of sentences with and without absolute pronouns was tested using a series of Act-Out activities. As before, the tests were administered by two native speakers of isiXhosa. The participants were told that they would listen to an instruction/sentence, they would have to pick up the characters and act-out what they heard. The SM test consisted of 8 laminated pictures representing the 4 noun classes. Each noun class had four sentences. The researcher put two pictures representing the singular and plural nouns of one noun class on the table and read out the action. For example, a picture of a cat might be presented along with two dogs, but critically, the dogs were glued together so could only act as a plurality. Table 3 presents a typical example with SM alone providing the clue

Children’s Acquisition of Subject Markers in isiXhosa

229

to subject number. Notice that noun class was neutralized since the two nouns were from the same noun class. Table 3. SM alone sentences ‘Masidlale. Uyabona kukho umakazi, kukho oomalume, ibhedi, indlu, isofa, ita¿le. Ngamanye amaxesha siza kukhetha umakazi, ngamanye amaxesha siza kukhetha oomalume. Masibabeke apha. Ndiza kukuxelela ukuba wenze ntoni. Ndifuna undibonise.’ ‘Let’s play. We have aunt, uncles, a bed, a house, a sofa and a table. Sometimes we will choose aunt, and sometimes we will choose uncles. Let’s put them here. I’ll tell you what happens. I want you to show me’ 1) Jonga, ibhedi. Uhlala ebhedini. Look, a bed. He/she sits on the bed. 2) Jonga, indlu. Baya emva kwendlu. Look, a house. They go behind the house.

In addition, a pronoun test was presented, also consisting of 8 laminated pictures representing 4 classes. These were the same pictures as the ones used for the SM test. The procedures for the pronoun test were exactly the same as those of the SM test. The difference was that the pronoun sentences contained an absolute pronoun. Table 4 provides an example. Table 4. Pronoun + SM sentences ‘Masidlale. Uyabona kukho umakazi, kukho oomalume, ibhedi, indlu, isofa, ita¿le. Ngamanye amaxesha siza kukhetha umakazi, ngamanye amaxesha siza kukhetha oomalume. Masibabeke apha. Ndiza kukuxelela ukuba wenze ntoni. Ndifuna undibonise.’ ‘Let’s play. We have aunt, uncles, a bed, a house, a sofa and a table. Sometimes we will choose aunt, and sometimes we will choose uncles. Let’s put them here . I’ll tell you what happens. I want you to show me’. 3) Jonga, ibhedi. Yena uhlala ebhedini. Look, a bed. As for her, him, she/he sits on the bed. 4) Jonga, indlu. Bona baya emva kwendlu. Look, a house. As for them, they go behind the house.

First, there is a very strong effect of the kind of cue used, with absolute pronouns being understood much better than SM alone, regardless of whether the items were singular or plural. Although not perfect on pronouns, the isiXhosa

230

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

children showed the same disparity between pronouns and agreement markers as the English children reported in Gxilishe et al. (2009). Yet, the isiXhosaspeaking children also showed better sensitivity to SM using this methodology than those in the previous work using picture-choice. On average, their sensitivity was .65, signi¿cantly above the .5 expected by chance. Even the four year olds were above chance in sensitivity to number in SM, although their accuracy was not very high. As with the other languages, it is perplexing to see the massive gap between performance in spontaneous production, where it is near 100% at age four, and comprehension, which is considerably lower. Johnson et al. (2005) speculated that children could not retrieve the information in uninterpretable features, such as those carried in English agreement suf¿xes, in comprehension. In production, the features are checked and then the information is deleted before Logical Form. In comprehension, the children cannot access the information that led to the agreement from the target of agreement alone. But if pronominal clitics reÀect notional number like strong pronouns, the information might be accessible. However, other researchers dispute the production-comprehension disparity, arguing that the eye-tracker results reveal successful earlier comprehension, and/or that the estimates of production from spontaneous speech are exaggerated compared to the level of performance one might get from say, elicited production, especially with nonce verbs (Brandt-Kobele and Höhle 2010; Legendre at al. 2010b). Clearly more work is needed here on elicited production.

3.

Future work

We are left then with an unresolved puzzle: clearly SM is not as strong a carrier of notional number as pronouns are, but children as young as four do show some sensitivity to the number information carried by SM. The improvement in performance of the participants compared to earlier work lies in the adoption of the act-out methodology, suggesting that the picture choice task might present challenges beyond the grammar to young children. It is noticeable that the performance is about the same for French and isiXhosa, that is, considerably weaker than production but at least better than chance. Both languages have preverbal clitic forms that might be in transition between weak pronouns and agreement af¿xes, and the children seem to be better at a younger age than their counterparts in Spanish and English who have suf¿xal agreement. Perhaps some more regularities will emerge if future studies can employ closely matched methodologies, and even more language varieties are included.

Children’s Acquisition of Subject Markers in isiXhosa

231

First, it is generally agreed that children learning languages with clitic pronouns do not show a delay in obedience to Principle B, now often called the “pronoun interpretation problem”, or PIP. That is, in languages with strong pronouns, children go through a prolonged period (up to age 6 or 7) when they misinterpret a sentence with a pronoun in the following kind of scenario: (22) Here are Bert and Elmo. Elmo hit him. That is, they allow the pronoun him to co-refer with Elmo, apparently violating Principle B of the binding rules (Chien and Wexler 1990; see Hamann 2010, for a review). Such errors do not occur when clitic pronouns are used in languages, such as Italian (McKee 1992), French (Hamann, Kowalski, and Philip 1997; Jakubowicz 1989), and Spanish (Baauw, Escobar, and Philip 1997). In the course of working out why there is this difference, several other properties of clitic object pronouns have been pointed to, such as the fact that they are high in the functional structure of the clause in Romance languages and also referentially de¿cient, in the sense that they cannot be used deictically, never take focal stress, and depend on discourse. Contrast these with English pronouns, which can in certain unusual cases be coreferential “accidentally”: (23) Everyone admires Bob. Jane admires him, Fred admires him, and even Bob admires him! In Romance, clitics and strong pronouns occupy different positions and have different properties. Evidence shows that children never misplace clitics in production, so if they never mistake clitics for strong pronouns, they know that clitics cannot take accidental coreference as in (23), and therefore do not show the PIP effect in (22) (Hamann 2010). Doing parallel work in isiXhosa on binding in children might shed more light on the status of object clitics. Second, some exciting work is underway on what other features SM might carry in isiXhosa. Number information is obscured to some extent by the variation in af¿xes across noun classes. But perhaps SM carries noun class information more readily. Smouse and her colleagues are currently testing whether young children can retrieve the noun class information from an SM marker alone, i.e. with no overt subject. If noun class information is more robustly represented than number, we will gain further insights into the nature of these markers and their status. For example, number information might be subsumed under noun class information in isiXhosa. In addition, we intend to undertake a comparison of the child L1 learners with those acquiring isiXhosa as an L2. A recent claim has been made that

232

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

uninterpretable features of an L2 that were not selected during a critical period in L1 learning might present a special challenge to second language learners (Hawkins and Hattori 2006; Tsimpli 2003). An interesting case to study might be English-speakers learning isiXhosa as an L2, looking at their relative sensitivity to number features (which might be acquired already, though minimally, for their L1) and gender or noun class, which should be much harder if not impossible to recover from the target of agreement, namely, the SM on the verb. How sensitive would L2 learners of isiXhosa be on these tasks?

4.

Conclusion

Work on acquisition of Bantu languages such as isiXhosa might contribute to linguistic debates about several phenomena in both acquisition and linguistic theories. So far, the evidence from spontaneous speech suggests that children command the subject marker at an early age in production, with very rare substitutions. The form seems to be productive, and not tied to particular lexical items. In addition, supplying the SM is just as good when the subject is dropped (pro-drop) as when it is overt, making surface copying an unlikely mechanism. In comprehension, two studies have shown weak performance in retrieving number information from the subject and object markers, in contrast to successful spontaneous production. Furthermore, SM is clearly different from absolute pronouns in this regard. The data on SM resemble data from French, in which a preverbal pronominal clitic also serves an agreement function. The four year old isiXhosa-speaking children are beginning to show some sensitivity to the number carried by SM on more re¿ned tests. Further research is planned to explore these issues, to compare Romance languages and isiXhosa using similar methods, and to compare the success of L1 and L2 learners.

References Apperley, I. 2010

Mind Readers: The Cognitive Basis of “Theory of Mind”. Hove: Psychology Press. Baauw, S., L. Escobar and W. Philip 1997 A delay of Principle B-effect in Spanish speaking children: The role of lexical feature acquisition. In A. Sorace, C. Heycock and R. Shillock

Children’s Acquisition of Subject Markers in isiXhosa

233

(Eds.), Proceedings of GALA 1997: Language Acquisition: Knowledge Representation and Processing. University of Edinburgh, 16–21. Baker, M. C. 1988 Incorporation: A Theory of Grammatical Function Changing. Chicago: The University of Chicago Press. Brandt-Kobele, O.C. and B. Höhle 2010 What asymmetries within comprehension reveal about asymmetries between comprehension and production: The case of verb inÀection in language acquisition. Lingua, 120 (8), 1910–1925. Bresnan, J and S. Mchombo 1986 Grammatical and anaphoric agreement. In Anne Farley, Peter. T. Farley and Karl Mc Cullough (Eds.), Papers from the Parasession on Pragmatics and Grammatical Theory. CLS 22. Vol. 2. Chicago Linguistic Society. University of Chicago. 278–297. Bresnan, J and S. Mchombo 1987 On topic, pronoun and agreement in Chichewa. Language, 63, 741–782. Buell, L. 2005 Issues in Zulu verbal morphosyntax. Doctoral dissertation, University of California, Los Angeles. Cardinaletti, A. 2008 On different types of clitic structures. In C. De Cat and K. Demuth (Eds.), The Bantu-Romance Connection. Amsterdam: Benjamins. 41–82. Carstens, Vicki. 2008 DP in Bantu and Romance. In C. De Cat and K. Demuth (Eds.), The Bantu-Romance Connection. Amsterdam: Benjamins. 131–162. Chien, Y. C. and K. Wexler 1990 Children’s knowledge of locality conditions in binding as evidence for the modularity of syntax and pragmatics. Language Acquisition, 1, 225– 295. Chomsky, N. 1995 The Minimalist Program. Cambridge, MA: MIT Press. De Cat, C. and K. Demuth (Eds.). 2008 The Bantu-Romance Connection. Amsterdam: Benjamins. Deen, K. U. 2005 The Acquisition of Swahili. Amsterdam: Benjamins. Deen, K. U. 2006a The clitic pronoun in Swahili. Poster presented at the workshop: The Bantu-Romance Connection, Leeds University. Deen, K. U. 2006b Subject agreement in Nairobi Swahili. In J. Mugane, J.P. Hutchison, and D.A. Worman (Eds.), Selected Proceedings of the 35th Annual Conference on African Linguistics: African Languages and Linguistics in Broad Perspectives. Somerville, MA: Cascadilla Press. 225–233.

234

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

Demuth, K. 2003

The acquisition of Bantu languages. In D. Nurse and G. Phillipson (Eds.), The Bantu Languages. Surrey: Curzon Press. 209–222. de Villiers, J. G. and S. Gxilishe 2009 The acquisition of number agreement in English and Xhosa. In J.M. Brucart, A. Gavarró and J. Sola (Eds.), Merging Features: Computation, Interpretation and Acquisition. London: Oxford University Press. 91–116. Doke, C. 1954 The Southern Bantu Languages. London: Oxford University Press. Du Plessis, J. A. and M. Visser 1992 Xhosa Syntax. Pretoria: Via Afrika. Du Plessis, J. A. and M. Visser 1998 Xhosa syntax. Stellenbosch Communications in African Languages. #7. Givón, T. 1976 Topic, pronoun and grammatical agreement. In C.N. Li (Ed.), Subject and Topic. New York: Academic Press. 149–188. Gxilishe, S., P. A. de Villiers and J.G. de Villiers 2007a The acquisition of subject agreement in Xhosa. In A. Belikova, L. Meroni, and M. Umeda (Eds.), Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition North America (GALANA). Somerville, MA: Cascadilla Press. 114–123. Gxilishe, S., J. G. de Villiers and P.A. de Villiers 2007b Acquisition of tense in Xhosa: the long and short of it. In H. CauntNulton, S. Kulatilake, and I. Woo (Eds.), BUCLD 31: Proceedings of the 31st annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press. 274–285. Gxilishe, S., M. Smouse, T. Xhalisa and J.G. de Villiers 2009 Children’s insensitivity to information from the target of agreement: the case of Xhosa. Proceedings of the 3rd GALANA Conference. Somerville, MA: Cascadilla Press. 46–53. Hamann, C. in press The acquisition of binding. In T. Roeper and J.G. de Villiers (Eds.), Handbook of Generative Approaches to Acquisition. Dordrecht: Springer. Hamann, C., O. Kowalski and W. Philip 1997 The French ‘delay of Principle B’ effect. In E. Hughes, M. Hughes, and A. Greenhill (Eds), BUCLD 21: Proceedings of the 21st Boston University Conference on Language Development. Somerville, MA: Cascadilla Press. 205–219. Hartford, C. 2008 The Bantu-Romance in verb movement and verbal inÀection morphology. In C. De Cat and K. Demuth (Eds.). The Bantu-Romance Connection. Amsterdam: Benjamins. 111–128.

Children’s Acquisition of Subject Markers in isiXhosa

235

Hawkins, R. and H. Hattori 2006 Interpretation of English multiple wh-questions by Japanese speakers: a missing uninterpretable features account. Second Language Research, 22 (3), 269–301. Jakubowicz, C. 1989 Linguistic theory and language acquisition facts: Reformulation, maturation or invariance of binding principles. Paper presented at conference on Knowledge and Language, Groningen, May. Johnson, V. E., J. G. de Villiers and H. N. Seymour 2005 Agreement without understanding? The case of third person singular /s/. First Language, 25, 317–30. Keach, C. 1995 Subject and Object markers as agreement and pronoun incorporation in Swahili. In A. Akinlabi (Ed.), Theoretical Approaches to African Linguistics. Trenton, NJ: African World Press. 109–116. Koopman, H. and D. Sportiche 1991 The position of subjects. Lingua, 85, 211–258. Labelle, M. 2008 Pronominal object markers in Bantu and Romance. In C. De Cat and K. Demuth (Eds.). The Bantu-Romance Connection. Amsterdam: Benjamins. 83–109. Legendre, G., J. Culbertson, I. Barriere, T. Nazzi and L. Goyet 2010a Experimental and empirical evidence for the status and acquisition of subject clitics and agreement marking in adult and child Spoken French. In V. Torrens, L. Escobar, A. Gavarró, and J.G. Mangado (Eds.), Movement and Clitics. Newcastle: Cambridge Scholars Publishing. 333–360. Legendre, G., I. Barriere, L. Goyet and T. Nazzi 2010b Comprehension of infrequent subject-verb agreement forms: Evidence from French-learning children. Child Development, 81 (6), 1859–1875. Marten, L., R. Kempson and M. Bouzouita 2008 Concepts of structural underspeci¿cation in Bantu and Romance, in C. De Cat and K. Demuth (Eds.). The Bantu-Romance Connection. Amsterdam, Netherlands: Benjamins. 3–39. McKee, C. 1992 A comparison of pronouns and anaphors in Italian and English acquisition. Language Acquisition, 2, 21–54. Meinhof, C. 1948 Grundzüge einer vergleichenden Grammatik der Bantusprachen. Hamburg: Dietrich Reimer. Second edition. First published 1906. Pérez-Leroux, A. T. 2006 Recovering plurality. Paper presented at workshop: The acquisition of the syntax and semantics of number marking, GLOW, Barcelona, April.

236

Mantoa Smouse, Sandile Gxilishe, Jill G. de Villiers and Peter A. de Villiers

Rizzi, L. 1997

Smouse, M. in press

The ¿ne structure of the left periphery. In L. Haegeman (Ed.), Elements of Grammar: Handbook of Generative Syntax. Dordrecht: Kluwer. 281– 337. Uninterpretable features in comprehension: Subject-verb agreement in isiXhosa. South African Journal of African Languages, 2012, 32 (2)

Swingley, D. 2009 Onsets and codas in one-year-olds’ word recognition. Journal of Memory and Language, 60, 252–269. Tsimpli, I. 2003 Clitics and determiners in L2 Greek. In J.M. Liceras, H. Zobl and H. Goodluck (Eds.), Proceeding of the 6th Generative Approaches to Second Language Acquisition Conference. Somerville, MA: Cascadilla Press. 331–339. Visser, M. (2008). De¿niteness and speci¿city in the isiXhosa determiner phrase. South African Journal of African Languages, 28 (1), 1–25. Woolford, E. 2000 Agreement in disguise. In V. Carstens and F. Parkinson (Eds.), Trends in African Linguistics. Trenton, NJ: Africa World Press. 103–117. Zeller, J. 2008 The Subject marker in Bantu as an antifocus marker*. Stellenbosch Papers in Linguistics, Vol. 38. 221–254. Zwart, J. 1997 Rethinking Subject Agreement in Swahili. NWO/University of Groningen. http://www.let.rug.nl/zwart/docs/nels97.pdf (accessed 21 October, 2011).

The Acquisition of Subject and Object Clitics in Swahili Kamil Deen

1.

Introduction

Swahili is a Bantu language spoken in Eastern Africa (Kenya, Tanzania, Uganda, as well as parts of Central African Republic and northern Mozambique). Clitics in Swahili (and other Bantu languages) occur as pre¿xes on the verb, marking both subject reference and direct object reference. As with clitic forms in other languages, a debate exists as to whether these clitics are incorporated pronouns or grammatical agreement forms. In this chapter, we investigate this issue with respect to Swahili, drawing on a series of diagnostic tests established by Bresnan and Mchombo (1987). To anticipate, it is concluded that both subject clitics (SC) and object clitics (OC) are grammatical agreement morphemes, unlike other well-known Bantu languages such as Chichewa (Bresnan and Mchombo, 1987). Besides the question of the status of clitics in Swahili, questions arise over what the developmental trajectories of these two clitics are. We ¿rst consider predictions made on the basis of well-known and well-regarded facts about child language more broadly. The four predictions all state that SC will be acquired before OC. The results presented in this chapter go counter to these predictions: The key ¿nding is that Swahili speaking children acquire OC earlier than SC. The explanation put forward is that OC occurs reliably and consistently in obligatory contexts in the input to children, while SC is occasionally omitted. This variability in the input leads to a gradual, and somewhat slower development of SC relative to OC. The suggestion, therefore, is that rare structures in the input (such as OC, relative to SC) might nevertheless be acquired early and with little dif¿culty if the context in which they occur is predictable.

2.

The Verbal Complex in Swahili

Swahili is an agglutinative language, with considerable verbal pre¿xing and suf¿xing. The unmarked word order is S-V-O, as shown in (1) below, in which

238

Kamil Deen

the subject (Juma) canonically occurs preverbally and the object (Mariam) occurs postverbally. The verb occurs in a verbal complex which consists of a subject clitic (a-) on the left periphery, followed by tense (-na-), an object clitic (syllabic -m-) and then the verb root itself (-pend-). The verb is followed by (in this case) one suf¿x which indicates mood (in this case indicative –a). Other suf¿xes are possible (always occurring between the verb root and the mood ¿nal vowel), including a passive suf¿x, causative suf¿x, applicative suf¿x, stative suf¿x, reciprocal suf¿x, etc., none of which are directly relevant to this current chapter (see Ashton, 1947; Myachina, 1981; Vitali, 1981, amongst others). Subject Verbal Complex (1) Juma a - na - m – pend - a Juma SC3s-Pres- OC3s- like - IND ‘Juma likes Mariam’

Object Mariam Mariam

(2)

Ø (null object)

Ø A - na - m – pend – a (null subject) SC3s - Pres- OC3s- like- IND ‘(He) likes (her).’

Note that the argument associated with the clitic may occur overtly, as in example (1) - so called clitic-doubling - but may also be omitted. Thus null subject and/or null object sentences are possible (2). 2.1

Subject Clitic

The SC always occurs as the ¿rst morpheme in the verbal complex, except in certain negative paradigms, in which case it is the second morpheme. It agrees with the grammatical subject in person and number when the subject is [+human], as in (1–2), and in noun class otherwise (3). Noun class agreement consists of both lexical noun class, as well as number. Limiting ourselves to the case of human subjects, there are six different forms of SC, given in Table 1. 1. Like grammatical gender in many European languages, Swahili classi¿es nouns into a series of 15 noun classes. These classes are marked with identifying pre¿xes which encode their class membership as well as number (singular/plural). Any nominal modi¿ers, like adjectives or demonstratives, typically carry agreement with this noun class, and more importantly for our purposes, the SC and OC also agree with this noun class. Here, we follow the standard practice of numbering these noun classes from 1–15. So in example (3a), kiti is of noun class 7, indicated by the pre¿x ki-.

The Acquisition of Subject and Object Clitics in Swahili

(3)

a.

Ki – ti ki – li – 7–chair SC7 – past – ‘The chair fell.’

a.

Vi – ti vi – li – 7pl–chair SC7pl – past – ‘The chairs fell.’

239

anguk– a fall – IND anguk– a fall – IND

SC in standard Swahili (so-called Kiswahili Sanifu) is obligatory in every indicative sentence of Swahili, except those which carry certain unusual tense markers (e.g., the habitual hu- marker). Furthermore, in the dialect of Swahili under investigation here (Nairobi Swahili), native speaker adults omit SC in approximately 5% of verbal utterances (see Deen, 2002; 2005 for details). However, such omission only occurs when the referent of the subject is clear from discourse. This will be discussed more in section 5. All SC morphemes are syllabic and consist of either a vowel or a consonant and a vowel. Table 1 shows the [+human] morphemes in the paradigm – by far the most frequent SC clitics in child directed speech. Table 1. [+human] SC clitics in Nairobi Swahili. SC Morpheme -ni-u-a-tu-mu-wa-

Referent 1st person, singular 2nd person, singular 3rd person, singular 1st person, plural 2nd person, plural 3rd person, plural

SC morphemes (and indeed OC morphemes, see below) occur in a ¿xed position with respect to the verb: all SC occur preverbally (without exception), preceding tense and OC. Thus unlike clitics in some European languages, clitics may not move in Swahili. However, like clitics in European languages, SC cannot be stressed nor topicalized, nor occur as the answer to a question. These are all typical properties of clitics, and so SC is often described as such. 2.2

Object Clitics

OCs, unlike SCs, do not occur in every indicative sentence in Swahili. Rather, OCs occur only when the direct object is speci¿c. That is, when the direct object is speci¿c, OC is obligatory, and when the direct object is non-speci¿c,

240

Kamil Deen

OC is obligatorily absent. This is formalized in Deen (2006, p.231) as The Speci¿city Condition. A consequence of this is that when OC is present (4a), the interpretation of the object is obligatorily speci¿c, while when OC is absent (4b), the interpretation of the object is obligatorily non-speci¿c. The Speci¿city Condition:

(4)

If the object is speci¿c, OC is obligatory, and if the object is non-speci¿c, OC is obligatorily absent.

a.

Juma a– li– mw– on– a m – tu [+OC] Obligatorily Juma SC3s– past– OC3s– see– IND 1–person Speci¿c ‘Juma saw the person / *a person.’

b.

Juma a– li– on– a m – tu Juma SC3s– past– see– IND 1–person ‘Juma saw a person / *the person.’

[-OC] Obligatorily Non-speci¿c

Thus OC and SC differ in that the former is subject to the Speci¿city Condition, and therefore does not arise in every utterance containing a verb, while the latter is not subject to the Speci¿city Condition, and occurs in all utterances containing a verb, unless omitted through discourse. The morphological shape of OC is similar to SC in that all morphemes are syllabic, although the 3rd person singular OC has no vowel nucleus but is a syllabic nasal, as shown in table 2. Table 2. [+human] OC clitics in Nairobi Swahili. OC Morpheme -ni-ku-m-tu-mu-wa-

3.

Referent 1st person, singular 2nd person, singular 3rd person singular 1st person, plural 2nd person, plural 3rd person, plural

Status of the Clitics

A typical problem that arises in the analysis of clitics (in many languages) is determining their status with respect to being pronominal versus agreement. That is, clitics are by their very nature similar to pronouns in familiar languages like

The Acquisition of Subject and Object Clitics in Swahili

241

English, as well as verbal agreement. In Swahili, as in many other languages, there is a debate as to what the status of the clitics are in the language (see, for example, Keach, 1995; Seidl and Dimitriadis, 1997; Woolford, 1999; Deen, 2006). One way of analyzing these clitics is to view them as pronominal in nature, and therefore arguments of the verb. On this view, the overt nominals that accompany the verb (what are referred to above as ‘subject’ and ‘object’) must then be viewed as topics or adjuncts of some kind, and not arguments of the verb. The alternative line of thinking is that the overt nominals that occur with the verb are in fact the true arguments of the verb (e.g., the preverbal DP is the subject of the sentence, and the postverbal DP is the object of the sentence). On this view, the clitics are then not pronominal, but must be viewed as agreement markers on the verb. These two approaches will henceforth be referred to as the Pronominal Analysis and the Agreement analysis, respectively. Pronominal Analysis: A clitic is the overt manifestation of the grammatical argument, akin to he, she, they, etc. in English. Thus in a sentence like anampenda (1 above), ‘a-’ is the third person singular pronoun, in the grammatical subject position (and similarly, ‘-m-‘ is the third person direct object incorporated pronoun). If this is the case, then the preverbal DP (in 1, Juma) is a topic, and not an argument of the verb. This last point is crucial to many of the tests put forward by Bresnan and Mchombo (1987) and discussed below. Agreement Analysis: A clitic is a grammatical reÀex on the verb, indicating verbal agreement with the true argument. So, in a sentence like (1) above, ‘a-’ is a verbal agreement pre¿x which agrees with the grammatical subject ‘Juma’ in person and number (and similarly, ‘-m-‘ is grammatical agreement with the direct object, Mariam). Thus the preverbal DP (‘Juma’) is the true grammatical subject, and not a topic. In the next two sections, we investigate these two analyses for each of the clitics. 3.1

Status of SC

In this section, we focus on whether SC is a subject pronoun or whether it is a verbal af¿x indicating agreement with the grammatical subject. Four arguments are provided that suggest that SC in Swahili is an agreement marker and not a pronoun. The ¿rst two of these arguments come from Keach’s (1995)

242

Kamil Deen

insightful application of Bresnan and Mchombo’s (1987) tests for the status of clitics, while the latter two arguments are independently motivated. Bresnan and Mchombo (1987), in their seminal article on Chichewa, provide a series of tests that can be used to assess the status of clitics. They point out that one good way to determine the status of a clitic in a language like Chichewa (a Bantu language spoken in Malawi) and other related languages is to determine whether the preverbal DP is a topic or not. The logic of this is that if the preverbal DP can be established as a topic, this will in turn show that SC must be the grammatical subject. If, on the other hand, the preverbal DP is the grammatical subject, the SC must be an agreement marker. Bresnan and Mchombo (1987) put forward a series of properties of topics and focused elements that can be used to assess whether the preverbal DP is a topic or a subject, and thereby determine the status of the SC. These properties are given in (5): (5)

(i) Relative pronouns bear TOPIC function; (ii) Questioned constituents bear FOCUS function; (iii) An argument cannot bear both TOP and FOC function in the same clause.

Keach (1995) uses these properties and provides two tests as to whether SC is pronominal or agreement in nature. First, if the preverbal DP in Swahili is a topic then, by 5(ii) and 5(iii), subject wh-questions should not be possible. This is because subject wh-questions are by (5ii) focused elements, and if the preverbal DP is a topic, it would be both a focused element and a topic at the same time, thereby violating 5(iii). In Chichewa, as Bresnan and Mchombo point out, subject wh-questions are not possible, suggesting that in Chichewa (and other related Bantu languages like Sesotho, see Demuth, 1992, amongst others) the preverbal DP is indeed a topic, and that SC is a pronoun. But in Swahili, as Keach points out, subject wh- questions are perfectly acceptable, showing that Swahili differs from Chichewa, and that SC in Swahili is not in fact pronominal, but more akin to agreement: (6)

nanii ai – me –end–a ? who SC3s-pr.prf-go-IND who has gone?

Keach’s second test involves idiomatic expressions. As Bresnan and Mchombo point out, it is well known that the subjects of idioms resist topicalization. For example, the idiom the cat is out of the bag loses its idiomatic meaning when

The Acquisition of Subject and Object Clitics in Swahili

243

the subject is topicalized (As for the cat, it is out of the bag). Likewise, as Keach points out, topicalizing the preverbal DP from within an idiom causes the loss of the idiomatic meaning in Swahili: (7)

a.

ni – li – ¿kir – i kuwa Mtindi u me – va - a Asha SC1s – past – think– IND that brew SC3– pr.prf.– wear- IND Asha ‘I thought that Asha was drunk.’ (lit: I thought that the brew has covered Asha)

b.

*Mtindi ni – li – ¿kir – i kuwa [t] u - me – va - a Asha Brew SC1s- past- think- IND that SC3- pr.prf.-wear-IND Asha (lit: (As for) the brew, I thought that it has covered Asha)

This suggests that the preverbal DP (in this case, Mtindi, ‘brew’) is not a topic, but a grammatical subject, because were it a topic to begin with, further topicalization should not have an effect on the idiomatic meaning of the expression. In turn, this suggests that the SC is an agreement marker and not a pronoun. In addition to these two arguments put forward by Keach, a third argument comes from what kinds of elements may occur in the preverbal position. It is well known that quanti¿ers may not occur in topic positions, as quanti¿ers are inherently focus-type elements (see Lasnik and Stowell, 1991; Rizzi, 1992). In Swahili, this restriction also holds. In (8a), the object (kila kitabu) is in object position, and is ungrammatical when topicalized, as in (8b). This shows that quanti¿ed elements cannot occur in topic position in Swahili – a fact that will now be used as a diagnostic for the status of SC in Swahili. (8)

a.

a– li – nunu – a kila kitabu SC3s – past – buy – IND every book She bought every book.

b.

* kila kitabu, a – li – (ki) – nunu– a [t] every thing SC3s – past – (OA7) – buy – IND [t] Every book, she bought.

Given the facts in (8) in which topics cannot be quanti¿ed, if the preverbal DP is a topic, it should be ungrammatical as a quanti¿er. This is in fact not true, as shown in (9). Once again, this suggests that the preverbal DP is not a topic, but a true subject. This in turn suggests that SC is an agreement morpheme, not a pronoun.

244

Kamil Deen

(9)

a.

kila mtoto a – li– nunu – a ki – tabu every child SC3s– past – buy – IND 7–book Every child bought a book..

b.

kila ki–tabu ki – li – nunuli – every 7–book SC7– past– buy – Every book was bought by a child..

w– a na mtoto passive–IND by child

A fourth argument also comes from the tests put forward by Bresnan and Mchombo (1987), who point out that topics cannot be the answer to a whquestion: (10) a.

Question:

Who arrived late?

b.

Answer:

??As for John, he arrived late.

c.

Answer:

John arrived late.

As (10b) shows, a topic as an answer to a wh-question is ungrammatical or dispreferred, while a sentence containing a regular grammatical subject (10c) is acceptable. In Swahili, if the preverbal DP is a topic, it should not be possible for it to occur as an answer to a wh-question, but rather the only possibility should be one in which the SC occurs, but the preverbal DP does not (i.e., clitic doubling should be disallowed). However, this is not the case, as (11) shows. (11) a.

b.

Question: nani a – li – ¿k – a who SC3s– past– arrive– IND ‘Who arrived early?’ Answer:

mapema early

Juma a – li – ¿k – a mapema Juma SC3s– past– arrive– IND early ‘Juma arrived early.’

(11b) shows that a sentence containing both the preverbal DP and SC is an acceptable answer to a wh-question, suggesting that the preverbal DP is not a topic, but a true grammatical subject.

2. As pointed out by an editor, the postverbal position in Swahili is a focus position, in which the preverbal nominal may occur postverbally with a focused interpretation, as in ali¿ka Juma mapema. This shows that this nominal (Juma) cannot be a topic since it would bear both topic function and focus function at the same time.

The Acquisition of Subject and Object Clitics in Swahili

245

In sum then, four arguments have been presented that all suggest that the preverbal DP in Swahili is not a topic, but a true grammatical subject. If this is the case, then the SC cannot also be a subject, but must be a verbal agreement marker – a position that will be adopted for the remainder of this chapter. 3.2

Status of OC

We turn now to the status of OC in Swahili. Two arguments are provided which suggest that OC is in fact agreement, akin to SC, and not pronominal in nature. First, OC may occur with an overt direct object (i.e., clitic doubling is possible). If it were the case that OC is a pronoun, then it is unclear what the status of the overt ‘direct object’ would be. If it were some kind of adjunct, perhaps postposed, then we would expect a marked prosodic pattern to be associated with the presence of an overt object, something akin to ‘comma’ intonation. However, this is not the case: when the direct object is present, there is no marked prosodic contour across the sentence overall, nor over the direct object in particular. Second, as we saw with idiomatic expressions in the analysis of SC, with the insertion of OC, idiomatic objects do not lose their idiomatic meaning. The logic of this argument is as follows. If the postverbal DP is a genuine direct object when OC is absent, but a topic/postposed adjunct when OC is present, then the insertion of OC into an idiomatic expression should disrupt the idiomatic meaning of that idiom, since the object has been moved out of the idiom. As (12) shows, this is not the case: the insertion of OC does not disrupt the idiomatic meaning of an idiom. (12) a.

b.

ni -

lipig– a pasi SC1s- past- hitIND iron ‘I ironed’ (lit.: ‘I hit iron.’) ni -

li - i pig- a past- OC3- hit- IND ‘I ironed it’ (lit.: ‘I hit it (with) iron.’) SC1s-

pasi iron

(12a) is an idiomatic expression which occurs without OC, and (12b) occurs with OC. Note that the absence of OC means that a non-speci¿c reading is intended (as formulated in the Speci¿city Condition, see section 2.2), and this is

246

Kamil Deen

reÀected in the gloss. (12b), on the other hand, takes a speci¿c reading, but in both cases, the idiomatic meaning is not disrupted. If OC were an incorporated pronoun, then the object in (12b) would have to be a kind of postposed topic of some kind, which would break up the idiomatic meaning. The fact that the idiomatic meaning is not disrupted suggests that the postverbal nominal in (12) is not in fact postposed (neither in 12a nor in 12b), but is in the same (direct object) position in both sentences. This, in turn, suggests that OC is not an incorporated pronoun, but a true agreement marker. An anonymous reviewer suggests that the manner in which the Speci¿city Condition (exempli¿ed in 4 above) is presented may be misleading, and an alternative view of the Speci¿city Condition may lead to a pronominal analysis of OC. The reviewer’s argument goes as follows. Assume that OC is a pronominal clitic which has a requirement that the DP with which it is cross-referenced must always be speci¿c (the Speci¿city Condition). That is, rather than viewing the Speci¿city Condition as a condition on when OC may occur, the Speci¿city Condition states that when OC does occur, the postverbal nominal must be speci¿c. This would then mean that a sentence with OC and a direct object (cliticdoubling) would be a case of right dislocation, along the lines of the English sentence Juma bought it, the book. While this view of the Speci¿city Condition is feasible, the associated pronominal analysis of OC is not. First, if the postverbal DP were in fact right-dislocated, the fact that no prosodic evidence of this dislocation is present would be puzzling and cross-linguistically odd. As mentioned above, the postverbal DP does not in fact carry so-called ‘comma’ intonation. Additionally, and importantly, other constituents (e.g., the subject) may be right-dislocated, in which case they always carry the typical comma intonation. This suggests that the direct object is not dislocated, but occurs in its base position. Second, the evidence from idiomatic expressions discussed above also suggests that the postverbal DP is not dislocated. Taken together then, this evidence shows that both clitics in Swahili are agreement markers, and not pronominal arguments of the verb. The overt DPs that accompany the verb are true arguments of the verb, with which the verb agrees.

4.

The Acquisition of SC and OC

In this section we investigate the acquisition of SC and OC, and then compare their relative development. First we discuss the method and the data, followed by predictions for the acquisition of SC and OC.

The Acquisition of Subject and Object Clitics in Swahili

4.1

247

Method and Data

The method employed in this study is the analysis of naturalistic data which consist of recordings from four children in Nairobi, Kenya aged 1;8 to 2;11. Swahili was the primary language in the home of all four children. Recordings were made once every two weeks or so, with the experimenter, the child and at least one parent present in most recordings. Each recording session lasted between 1 hour and 1.5 hours. Below is the relevant information for each child, including the age ranges of each child, the number of recordings, and the start and end mean length of utterances (MLU). The start MLU is the MLU calculated at the beginning of the child’s data recording period, and the end MLU is the MLU calculated at the end of the data recording period. Thus for Haw, her MLU at age 2;2.01 was 1.54, and her MLU at 2;6.05 was 2.46. Table 3. Children’s names, age ranges and number of recordings Child Haw (girl) Mus (boy) Fau (girl) Has (boy)

Age Range 2;2.01 – 2;6.05 2;0.16 – 2;10.10 1;8.19 – 2;2.07 2;10.13 – 2;11.25

Number of recordings 07 23 10 04

Table 4. Mean Length of Utterances for the four children Child Haw Mus Fau Has

Starting MLU 1.54 1.52 2.97 3.15

Ending MLU 2.46 3.57 3.35 4.23

Three measures of grammatical maturity were used to group the data into four developmental stages: (i) Mean Length of Utterance (MLU), (ii) V ratio (the proportion of verbal utterances to the overall number of utterances, a measure used by, for example, Valian, 1991), and (iii) the proportion of Mono-syllabic Placeholders, or ¿ller syllables (Peters, 2001; Bottari, Cipriani and Chilosi, 1993).The corpora could be broken down into four stages, with three children falling into three different stages, and the fourth child spanning three stages. 3. The other languages that occurred in the homes were typically English from television, and occasionally another ethnic languages such as Kikuyu or Luo. However, all parents reported that the children did not understand any language other than Swahili, and knew very few words in other languages.

248

Kamil Deen

This is shown below in table 5, and the criteria for the establishment of these stages are given in table 6. See Deen (2002, 73; 2005) for details on the establishment of stages. Henceforth, data will be described in terms of these stages. Table 5. Division of corpora into four stages Stage 1 Stage 2 Stage 3 Stage 4

Hawa, Mustafa01 Mustafa02 Mustafa03, Fauzia Hassan

Table 6. Criteria for the establishment of stages Stage 1 Stage 2 Stage 3 Stage 4

4.2

MLU 3.5

V-Ratio .25

%MPH >25% 15–24% 5–14%