Production-Comprehension Asymmetries in Child Language 9783110259179, 9783110238723

The workshop Production-Comprehension Asymmetries in Child Language held in Osnabrück in 2009 is the starting point for

211 98 2MB

English Pages 282 Year 2011

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Acknowledgements
Introduction
Testing the Aspect Hypothesis in child Tamil
An asymmetry in the acquisition of accusative clitics in child Romanian
Comprehension and imitated production of personal pronouns across languages
Comprehension and production of subject pronouns in child Dutch
Asymmetries in the processing of object relatives in child Hebrew and Italian
A comprehension delay of subject-object word order in Dutch preschoolers
Asymmetries in children’s language performance within and across modalities
Adults’ on-line comprehension of object pronouns in discourse
Production and comprehension of sentence negation in child German
Principle B delays as a processing problem: Evidence from task effects
Subject index
Recommend Papers

Production-Comprehension Asymmetries in Child Language
 9783110259179, 9783110238723

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Production-Comprehension Asymmetries in Child Language

Studies on Language Acquisition 43

Editor Peter Jordens

De Gruyter Mouton

Production-Comprehension Asymmetries in Child Language

Edited by Angela Grimm Anja Müller Cornelia Hamann Esther Ruigendijk

De Gruyter Mouton

Printed with generous support from the Center for Research on Individual Development and Adaptive Education of Children at Risk (www.idea-frankfurt.eu).

ISBN 978-3-11-023872-3 e-ISBN 978-3-11-025917-9 ISSN 1861-4248 Library of Congress Cataloging-in-Publication Data Production-comprehension asymmetries in child language / edited by Angela Grimm ... [et al.]. p. cm. ⫺ (Studies on language acquisition ; 43) Includes bibliographical references and index. ISBN 978-3-11-023872-3 (hardcover : alk. paper) 1. Children ⫺ Language. 2. Language acquisition. I. Grimm, Angela, 1971⫺ LB1139.L3P756 2011 4011.93⫺dc23 2011029557

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.d-nb.de. 쑔 2011 Walter de Gruyter GmbH & Co. KG, Berlin/Boston Typesetting: PTP-Berlin Protago TEX-Production GmbH, Berlin Printing: Hubert & Co. GmbH & Co. KG, Göttingen ⬁ Printed on acid-free paper Printed in Germany www.degruyter.com

Acknowledgements This book summarizes papers from the DGfS-workshop ProductionComprehension-Asymmetries in Child Language held at the University of Osnabrück in March 2009. The workshop considered asymmetries between children’s comprehension and production in various linguistic tasks and discussed grammatical and methodological explanations to the observed patterns. First of all, we would like to express our thanks to all participants of the workshop for the encouraging discussion during and after the meeting, to the local organizers at Osnabrück for the excellent preparation of the meeting, and to the DGfS for providing a travel grant for the invited speakers. The book would probably not have been published without the initial interest of Ursula Kleinhenz and the assistance of Julie Miess from the publisher de Gruyter and without the support of the editor of the SOLA-series, Peter Jordens. We are further very grateful to the research centre IDeA Frankfurt for kindly providing a publication grant. Our special thanks goes to all collegues who peer-reviewed the papers (in alphabetical order): Fabrizio Arosio, Sergio Baauw, Gerlof Bouma, Solveig Chilla, Tom Fritzsche, Martin Haiden, Petra Hendriks, Barbara Höhle, Angeliek van Hout, Phillip Prévost, Judith Rispens, Vesna Stojanovik, Jill de Villiers, and Ralf Vogel. Finally, we would like to thank Anna Roth for her invaluable editorial assistance. We hope that this book contributes to and encourages further discussion on methodological and theoretical explanations to unexpected, unresolved and poorly investigated patterns in children’s language. Two crucial aspects are surely the role of contextual information and children’s computation of the speaker’s intentions. With this book, we hope to have met the intentions of readers who are interested in getting a broad overview of cross-linguistic research on production-comprehension asymmetries, on the acquisition of pronouns, or, symmetrically on both. Frankfurt and Oldenburg, May 2011

Angela Grimm Anja Müller Cornelia Hamann Esther Ruigendijk

Table of contents Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

v

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Angela Grimm, Anja Müller, Esther Ruigendijk, and Cornelia Hamann

1

Testing the Aspect Hypothesis in child Tamil . . . . . . . . . . . . . . . . . . . . . . . Lavanya Sankaran

17

An asymmetry in the acquisition of accusative clitics in child Romanian Martine Coene and Larisa Avram

39

Comprehension and imitated production of personal pronouns across languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina Comprehension and production of subject pronouns in child Dutch . . . . . Charlotte Koster, Jan Hoeks, and Petra Hendriks

69

99

Asymmetries in the processing of object relatives in child Hebrew and Italian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Irena Botwinik A comprehension delay of subject-object word order in Dutch preschoolers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 Gisi Cannizzaro Asymmetries in children’s language performance within and across modalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Oda-Christina Brandt-Kobele and Barbara Höhle Adults’ on-line comprehension of object pronouns in discourse . . . . . . . . 193 Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks Production and comprehension of sentence negation in child German . . . 217 Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

viii

Table of contents

Principle B delays as a processing problem: Evidence from task effects . 247 Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin Subject index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273

Introduction Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk

Asymmetries in child language Asymmetries have long been observed in child language and it is understood that comprehension or perception usually precedes production. One example pertains to children’s lexical development: 16 months-old children produce 45 words but understand approximately 180 words (Fenson et al. 1993 for English). As to children’s grammatical development, perception studies (Höhle and Weissenborn 1998, Santelmann and Jusczyk 1998) have established that functional categories are recognized already during the ¿rst year of life but are produced consistently only after the third birthday. However, twenty-¿ve years of research on the acquisition of pronouns has shifted the focus to the possibility of production preceding comprehension. The remarkable observation in this area was that English speaking children produce pronouns in a target-like way in their third year of life (Bloom et al. 1994), but have considerable dif¿culty in pronoun interpretation up to their sixth birthday (Chien and Wexler 1990 and many others). Under the assumption that one grammatical system underlies comprehension and production (Chomsky 1982), this ¿nding fueled much theoretical speculation and lead researchers to focus on other areas where such asymmetries might be expected, such as focus particles (Müller 2010) or contrast accent (Vogel and Raimy 2002). In these linguistic phenomena, pragmatics or processing interact with syntax, and linguistic models capturing this interaction have become decisive for models of language acquisition as well as for models of the components of grammar. This book concentrates on production-comprehension asymmetries in child language in the sense that production outperforms comprehension in the same linguistic domain. It grew from the contributions to a workshop with the title “Production-comprehension asymmetries in child language” held at the 2009 Annual Meeting of the DGfS in Osnabrück. In fact, it grew from the contributions of two workshops since we also integrated a special session on the

2

Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk

acquisition of pronouns. So the papers presented here deal with the productioncomprehension asymmetry or with pronouns and often with both. The crucial problem we are addressing comes from the assumption that there is a single grammar for comprehension and production. In the context of the investigation of pronoun interpretation and on the background of what has been called its “delay”, this assumption lead to two possible explanations. One explanation explores the fact that the dif¿culties seem to arise at the interfaces, especially in areas where children have to integrate syntactic knowledge with information from discourse and the situation, i.e. with pragmatics (see Chien and Wexler 1990 and many others). The other explanation is that there are processing limitations in comprehension which do not occur in production (see Grodzinsky and Reinhardt 1993 and much subsequent work). Crucially, both these explanations open the possibility that some of the dif¿culties in comprehension are performance or task effects, either because the pragmatic context was not optimally controlled in a given experiment (see Grimshaw and Rosen 1990) or because the test conditions are too demanding for a child’s immature processing capacity. Both these possibilities raise the question to what extent these asymmetries result from methodological decisions; much recent work has addressed this question in the area of pronoun interpretation (see Elbourne 2005, Conroy et al. 2009). As to the linguistic models which can capture the empirical ¿ndings, again we are faced with two possibilities. Either the phenomenon in question is analyzed as an interface phenomenon with a clear separation of what is syntactic, semantic or pragmatic and suggestions for the interaction of these components (see Reinhardt 2006, Reuland 2001 for pronouns), or the interaction of the components is integrated into one grammar with the proviso that constraints of a pragmatic nature are “soft” constraints in the sense of Burzio (1998). So one of the questions in the focus of current research has been what a grammatical explanation of the observed asymmetries could look like. This volume sheds light on both the questions outlined above. On the theoretical side, the issue of production-comprehension asymmetries has received a great deal of attention in the last years and many studies suggest that a model of grammar as proposed by Optimality Theory is well suited to capture the asymmetry (see Hendriks and Spenader 2006, Hendriks and Koster 2010, de Villiers et al. 2006). Under the generative assumption of a narrow computational system, however, the observed asymmetries must arise from performance factors such as processing limitations, processing strategies, or task effects, an idea that has been put forward for example by Grodzinsky and Reinhart (1993), Avrutin (2006), Conroy et al. (2009), and in this volume (by Baauw et al., Botwinik,

Introduction

3

Brandt-Kobele and Höhle). One argument for this position is that researchers observed different results when the same participants were tested with different experimental paradigms in the same modality and linguistic area. As Hendriks and her collaborators point out in several papers, and also propose in this volume, methodological explanations attribute the productioncomprehension asymmetries to task effects. They put forward a grammatical explanation to production-comprehension asymmetries in the interpretation of pronouns and word order within the framework of Bidirectional Optimality Theory (Blutner 2000). The basic assumption is that children below age six (or even older) cannot compute the speaker’s alternative because they ¿rst have to acquire Theory of Mind (Wimmer and Perner 1983) or because their limited processing capacity does not allow them to compute both constraint hierarchies. Thus, at the younger ages, children must optimize unidirectionally, which leads to non-adultlike comprehension. This account predicts productioncomprehension asymmetries to occur in all linguistic areas involving pragmatic knowledge. Recent ¿ndings from the development of scalar implicatures (Papafragou and Musolino 2004, Zondervan et al. 2009, Koch, Schulz, and Katsos 2010), phrasal stress (Vogel and Raimy 2002), and focus particles (Müller 2010) are in line with the position that tasks involving pragmatic knowledge are not mastered before age ¿ve. Despite the increasing interest in the literature, the source of productioncomprehension asymmetries in child language is still an unresolved topic. In addition to the different explanations sketched above, research must also take into account cross-linguistic evidence, which turned out to be crucial in the discussion about pronouns (see McKee 1992, Jakubowicz 1984, Hamann et al. 1997, Baauw et al. this volume). As outlined in more detail in the next section, the present volume aims to consider production-comprehension asymmetries on the one hand and the acquisition of pronouns on the other from crosslinguistic and explanatory perspectives.

Aims of the volume The ¿rst goal of the volume is to bring together cross-linguistic research on production-comprehension asymmetries in syntactic, semantic, and morphological development. This results in papers addressing different linguistic areas such as the acquisition of pronominals (Hendriks, Banga, van Rij, Cannizzaro and Hoeks; Koster, Hoeks and Hendriks; Bittner, Kuehnast and Gagarina; Baauw, Zuckerman, Ruigendijk and Avrutin, Coene and Avram), aspect marking (Sankaran), negation (Wojtecka, Koch, Grimm, and Schulz),

4

Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk

relative clauses (Botwinik), word order (Cannizzaro) and inÀection marking (Brandt-Kobele and Höhle). Some of the languages studied here do already have a long research tradition in acquisition such as Dutch and German, but the volume also includes papers on lesser-studied languages such as Romanian, Bulgarian, and Tamil. The study on child Romanian is based on longitudinal production data of two children, all other studies analyzed cross-sectional data of pre-school or school-aged children or of adults. The book thus presents new quantitative information on the acquisition of selected phenomena in several linguistic domains. The second goal of the volume is to shed light on the source of the production-comprehension asymmetries. As outlined above, previous research puts forward different explanations such as limitations in pragmatic knowledge, limitations in cognitive processing, or methodological effects which may either interact with pragmatics or with processing, and different grammatical evaluation processes in perception and production which are either not equally available to the child or are beyond the child’s processing capacity (see e.g., Hendriks and Koster 2010 for an overview). The contributions in this book focus on methodological and grammatical explanations. These two approaches are brieÀy introduced below. Methodological effects Over the years, a number of experimental techniques have been established to investigate children’s linguistic behavior. For comprehension, experimental settings such as truth value judgements, syntactic and semantic priming, act out tasks and question after story tasks are in use since the 70s (McDaniel, McKee, and Cairns 1996). The fact that three out of the ten papers in this volume use eye-tracking data (Hendriks, Banga, Cannizzaro, and Hoeks; Cannizzaro, Brandt-Kobele, and Höhle) exempli¿es the increasing role of on-line measures for assessing children’s comprehension. The investigation of methodological effects tackles method-inherent problems. Both, production and comprehension experiments normally take place in non-naturalistic and highly controlled situations and test sentences are often presented in isolation or with minimal context. Therefore children ¿rst have to construct an appropriate context to master the task (Elbourne 2005, Conroy et al. 2009, Müller 2010). The non-adult performance observed in several studies might reÀect children’s inability to build a discourse model for the given situation, not their inability to interpret the test sentence proper. As Hendriks et al. (this volume) show, even adults bene¿t from contextual information to a different extent in the comprehension of pronouns, depending on the research

Introduction

5

method. This indicates that the tasks interact with the processing of discourse information. Hence, it appears that the role of the context is not straightforward in the sense of ‘the more context the better performance’ but that the kind of context and the presentation mode matter as well. Children’s performance also depends on their awareness of the goals and of the general nature of the task. Papafragou and Musolino (2003) investigated the interpretation of scalar implicatures by 5-year olds in two conditions, a) with normal context and, b) with context preceded by a training to detect infelicitous statements (see also Zondervan et al. 2009). They found a signi¿cantly better performance in the latter condition. This indicates that knowledge about the communicative expectations impacts task performance. Perception and production studies also differ with respect to the response categories. Perception studies force children to respond in a certain way. Usually, only one answer is scored as correct, and there are few error categories (e.g. as in truth value judgment tasks). Production studies, in particular spontaneous production settings, are less limited in that respect. Children can avoid dif¿cult structures and produce easier ones, hence spontaneous production never directly reÀects children’s knowledge (see Ruigendijk et al. 2010 for a close comparison of methodology in a comprehension and production study on pronouns). This explains why children often produce more errors in elicited production tasks than in spontaneous settings (e.g., Kauschke, Kurth, and Domahs in press for plural formation). The production-comprehension asymmetry thus may also be related to the fact that error analyses often use different error categories. For example, in Wojtecka et al.’s study (this volume) on sentence negation no category ‘false negatives’ is required as false negatives are hardly ever produced in normal conversation and even in elicited contexts. Thus, asymmetries between production and comprehension may also arise from researcher’s decisions on the speci¿c qualitative analysis. In sum, the differences can be diminished if the material allows children to build up an appropriate discourse model. Currently, there is no consensus if (comprehension) experiments should provide contextual information, and if so, how much context is required. Hendriks and Koster (2010: 13 – 14) argue for caution in that respect because children could use extra-grammatical information for performance on the task. In other words, it cannot be decided if a good test performance results from children’s grammatical knowledge or from their interpretation of additional situational cues. Finally, it depends on the research goals if, how much and what type of contextual information should be given (see also Hamann in press). If the focus of research is to test knowledge of a single grammatical phenomenon, conditions must be controlled as far as possible. If children’s abilities are assessed across modalities, or if the aim is to

6

Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk

know if children are in principle able to perform well on a certain structure, contextual information is required. As elicited production experiments rarely come out of the blue, this implies that the equivalent comprehension task must be enriched with context. For example, Hendriks and Koster (2010, quoting Spenader et al. 2009) point out that children signi¿cantly improved in their interpretation of pronouns if a discourse context was presented. Likewise, contextual information decreased the error rates in the interpretation of focus particles (Müller 2010), and scalar implicatures (Papafragou and Musolino 2003). These considerations – the error analysis, the role of the context, and shared knowledge of communicative goals of the task - support the position that methodological effects provide a source of the production-comprehension asymmetry in children’s performance. Grammatical explanations Recent research provides different theoretical accounts of the acquisition of production and comprehension. Constraint-based models propose that comprehension and production are generated upon an adult-like underlying representation, and that learning takes place via successive re-ranking of constraints (cf., Tesar and Smolensky 1998, Boersma and Hayes 2001). There are two types of constraints: faithfulness constraints require the output to be identical to the input; i.e., they establish a correspondence between input and output. Markedness constraints require the surface form to be unmarked, i.e., they only target the output form. Constraint-based accounts to learnability need to explain how the same grammar renders different candidates optimal in production and comprehension. As Smolensky (1996: 722) points out, the assumption of a single grammar runs into a dilemma: in comprehension (which is assumed to be adult-like or near adult-like), faithfulness constraints must dominate markedness constraints, while at the same time markedness constraints must dominate faithfulness in production (which is impoverished). Smolensky proposes that different candidate sets compete: in comprehension these are candidates sharing the output form, whereas in production competition takes place between candidates sharing the underlying form. Following the standard position that markedness constraints only apply to the surface form, they have no effect on mappings of the surface form to the underlying form, i.e. in comprehension. However, as Smolensky’s proposal presumes adult-like underlying forms, it disregards that comprehension also develops. A model where comprehension and production simultaneously develop is proposed for the acquisition of phonology in Pater (2004). Pater imposes domain-speci¿c faithfulness constraints

Introduction

7

for comprehension (C-FAITH) and production (P-FAITH); structural constraints are not speci¿c to a particular domain. If C-FAITH dominates P-FAITH, comprehension outperforms production. Essentially, Pater’s model also permits the reverse pattern, i.e., that production outperforms comprehension. At the level of syntax and semantics, comprehension-production asymmetries can better be explained by the different evaluation processes in comprehension and production. In comprehension, the form provides the input to the optimization process; in production it is the meaning. This is formalized in another constraint-based model, Bidirectional Optimality Theory (Blutner 2000). Bidirectional OT presumes that optimization takes place by associating optimal form-meaning pairs, i.e. optimization always goes in both directions (see Hendriks and Spenader 2006 for a description). For comprehension, the listener must take into account the speaker’s alternatives to optimize bidirectionally. For example, to interpret pronouns, children have to evaluate other pronouns, especially the reÀexive, in the lexicon and compare them to the associated meaning. If learners are unable to compute the alternatives, they are not able to optimize bidirectionally. Hendriks and Spenader (2006) consider children’s inability to optimize bidirectionally as a reason of why comprehension is delayed as compared to production. The insuf¿ciency to perform a bidirectional evaluation might be due to factors such as limited processing load, limited working memory capacities, or incomplete acquisition of Theory of Mind (Wimmer and Perner 1983). Children may also rely on unidirectional optimization because they have not yet mastered the mechanisms behind the bidirectional grammar (Hendriks and Koster 2010: 22). Bidirectional OT may be able to explain why comprehension sometimes lags signi¿cantly behind production. Moreover, it integrates empirical evidence on production-comprehension asymmetries and the attested pragmatic de¿cits into a single grammar without stipulating child-speci¿c constraints. However, the model needs further explication with respect to the question of how children learn the set of alternatives if grammar and lexicon have to be acquired simultaneously. The model also should further be extended to capture cross-linguistic differences in comprehension-production asymmetries. For example, it has repeatedly been shown that there is language-speci¿c variation with respect to a possible delay in the comprehension of pronouns (McKee 1992, Hamann et al. 1997, Ruigendijk 2008, Ruigendijk et al. 2010). Additionally, children acquiring Dutch, English, and Hebrew showed dif¿culties in the interpretation of pronouns but not in the interpretation of reÀexives; a pattern which could not be replicated for child German and Spanish (nor for other Romance languages). More precisely, the Dutch, English and Hebrew children, but not the

8

Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk

Spanish and Germans overextended a reÀexive meaning to the object pronoun in sentences such as ‘Bert is washing him’. To date, these cross-linguistic differences still need to be implemented in a grammatical model. Taken the tenets of constraint-based models such as bidirectional OT, cross-linguistic and inter-individual variation may be captured by different ways of re-ranking the same constraints. In sum, the evidence so far points to two main explanations to productioncomprehension asymmetries in child language. One group of researchers attributes the observed asymmetries to methodical effects, while another group refers to the framework of bidirectional OT. As will become clear by the summary of the papers in the following section, the volume represents both positions, and crucially, both directions of explanation do complement each other.

The organization of the book The volume is organized into two major sections. Section A summarizes papers on cross-linguistic evidence for production-comprehension asymmetry. Papers providing a theoretical or methodical explanation to the productioncomprehension asymmetry are collected in section B. Section A The section on cross-linguistic evidence starts with Sankaran’s study on aspect marking in child Tamil. Sankaran carried out an elicited production task and a sentence-picture-matching task with children of two different age groups and with adult controls in order to test the inÀuence of verb semantics on children’s acquisition of aspect markers. Her ¿ndings suggest a two-way asymmetry in the comprehension and production in the acquisition of the two aspect markers under investigation (kondiru, vidu). In production, the children are still acquiring kondiru while showing an adult-like performance in comprehension at the same time. In contrast, vidu is produced in an adult-like way by both age groups but performance on the vidu-sentences in the comprehension task is poor. Sankaran argues that the disadvantage of vidu is related to its dual function as perfective and inceptive marker. The study of Coene and Avram compared the development of accusative pronominal clitics to reÀexives in the spontaneous production data of two Romanian children between 21 and 36 months of age. Their data revealed a difference between 1st/2nd and reÀexives on the one hand and 3rd person clitics on the other. The former types of clitics emerge slightly later than 3rd person

Introduction

9

clitics but are produced in a target-like way from the very beginning. 3rd person clitics are found in the very early data but undergo omission still at the stage when 1st/2nd person clitics are used in an adult-like manner. Coene and Avram argue that computational dif¿culty caused by feature intervention effects best account for their ¿ndings. Their study indirectly deals with an asymmetry in comprehension and production since it has been long established (with the same methods as for English or Dutch) that Romance pronominals are interpreted correctly much earlier than English pronominals but remain problematic in production for a long time. Coene and Avram show that this asymmetry may vanish when not only 3rd person clitics are considered. Bittner, Kuehnast, and Gagarina investigated the comprehension and production of personal pronouns by 3- and 5-year-old German-, Russian- and Bulgarian-learning children. They used a question-after-story-design to evaluate comprehension and production of personal pronouns. Their study examined if children relate personal pronouns in subject position to structural prominence and/or to animacy, and if the same patterns are observed in comprehension and production. Comparing different language groups, the authors investigated cross-linguistic differences in the use of cues (e.g., animacy). Their results show that children tend to relate personal pronouns to the same cue pattern in production and comprehension using cues indicating high salience of referents (see also Elbourne 2005). Furthermore, the anaphoric capacity of personal pronouns is determined by language-speci¿c features. Section B Taking a theoretical perspective, Koster, Hoeks, and Hendriks argue that Bidirectional Optimality Theory elegantly captures the production-comprehension asymmetries. Their study investigates the processing of discourse anaphoric subject pronouns, full NPs and topic shifts by Dutch preschoolers and adults. In accordance with the Asymmetric Grammar Hypothesis, which states that asymmetries in acquisition are the result of inherent properties of the grammar, the authors argue that asymmetries between production and comprehension are the result of children’s inability to take into consideration the other person’s perspective in communication. In production, they will be overly economical and produce unrecoverable pronouns after topic shifts. When listening to a story, children will ignore NP topic shift marking and, therefore, will be at a loss as to how to interpret an ensuing subject pronoun. The results con¿rm their predictions: The children demonstrated non-adultlike production of pronouns and non-adultlike comprehension of full NPs which reÀects the asymmetric effects of the constraints of the grammar.

10

Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk

Based on data of child Hebrew and Italian, Botwinik analyzes the production-comprehension asymmetry in object relatives. Starting from the observation that comprehension lags behind production, she argues that the production-comprehension asymmetry attested by object relatives stems from the way their syntactic processing unfolds, a procedure which is crucially involved in the comprehension of object relatives, but not in their production. She claims that the processing (comprehension) of object relatives is comparable to garden path effects, involving an instance of local ambiguity. Based on the assumption that processing is done by the computational system, the nature of the syntactic clues and the point in processing at which they occur account for the level of performance on these relatives in the respective languages. Cannizzaro studied the comprehension and production of subject-object word order in Dutch 3;6-year-olds and in adults. Comprehension was assessed via a picture-selection task combined with eye-tracking, production by an elicited production task. Her data show that production outperformed comprehension in children, but not in adults. In both groups, there was a tendency for subjects to be matched to [+animate] and objects to [-animate]. Rethinking results from several previous papers, Brandt-Kobele and Höhle take a closer look on methodological issues related to the productioncomprehension asymmetry. The particular focus is on verbal inÀection. They assume that the grammatical systems for production and comprehension do not develop in an asymmetric fashion. Rather, comprehension and production require different steps of processing. Evaluation takes place structurally and heuristically; i.e., either based on structural information or on probabilistic knowledge of canonical form-meaning relations. Children might be particularly sensitive to heuristics when processing complex structures. The poor performance in comprehension results from conÀicting heuristic and structural processing strategies. In contrast, heuristics plays a minor role in production; consequently no conÀict arises between heuristics and linguistic structure. Hendriks, Banga, van Rij, Cannizzaro, and Hoeks investigated the role of the discourse context in the interpretation of pronouns. They performed a combined picture-veri¿cation and eye-tracking study with Dutch adults to test their comprehension of object pronouns and reÀexives while manipulating the discourse context. Although the adults hardly made any comprehension errors in the picture veri¿cation task, their reaction times were signi¿cantly slower when the introductory sentence did not unambiguously establish a discourse topic. This suggests that the structure of the discourse context is important for pronoun interpretation and inÀuences adults’ on-line processing of object pronouns. The authors argue against experimental artifacts as an explanation of

Introduction

11

the Delay of Principle B-Effect. Task effects can, however, explain why context effects were observed in the off-line but not in the on-line task. In a longitudinal group study, Wojtecka, Koch, Grimm, and Schulz investigated the acquisition of the sentence negator nicht ‘not’ by German preschoolers. Using elicited production data and a truth value judgment task for comprehension, the authors found that target-like production of nicht precedes its target-like comprehension. They argue that the comprehension task did not provide enough contextual information to license a certain type of sentence negation (true negatives), and that the gap between production and comprehension would be diminished if the task were embedded in a more appropriate context. Baauw, Zuckerman, Ruigendijk and Avrutin studied the role of task effects in the interpretation of pronouns. Drawing on experimental evidence from Dutch and Spanish children and from Spanish Broca’s aphasics, the authors claim that the problems of interpreting object pronouns (Pronoun Interpretation Problem; also called Delay of Principle B-Effect) is due to a processing problem, not to missing knowledge of binding or coreference principles. This explains the differences in the task performance of picture selection and truth value judgment using the same material and participants. Baauw et al. argue that the performance of these populations is strongly related to the processing load that different experimental methods impose on processing of object pronouns.

Summary The papers in this volume investigate production-comprehension asymmetries in child language and in adults by studying different languages. Comprehension is measured via truth value judgment, picture selection tasks, question after picture tasks and eye-tracking. In production, most of the studies rely on elicited productions; only one paper analyzes spontaneous production data. The volume covers a range of linguistic phenomena: pronouns, reÀexives, aspect marking, inÀection marking, non-canonical word order, negation, and relative clauses. For most of the phenomena considered here, a productioncomprehension asymmetry was found in the direction that production precedes comprehension. An exception provides the acquisition of aspect marking in child Tamil, where comprehension (of at least one of the markers) seems to be better than production. A closer examination of the results reveals that the research method plays an important role for the extent of the production-comprehension asymmetry.

12

Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk

This is particular striking in two papers of this volume. Hendriks et al. found that, in reaction times, Dutch adults showed effects of context information in comprehension but they did not do so in their eye-gaze data. Likewise, Baauw et al. observed a signi¿cantly better performance in a picture selection than in a truth value judgment task in Dutch children. Crucially, Baauw et al. used the same materials and assessed the children with the two tasks within a single test session. Thus, while there is clear evidence that tasks inÀuence children’s performances, it is still an open issue which is the optimal one for particular research questions, linguistic phenomena and age groups. The papers in this volume also reÀect different positions with respect to the source of the production-comprehension asymmetry: task effects, problems with pragmatics, processing limitations (Baauw et al., Bittner et al., Botwinik) or conÀicting processing strategies (Brandt-Kobele and Höhle) are considered as possible causes. Other papers explain the production-comprehension asymmetry by grammatical properties (Hendriks et al., Koster et al., Cannizzaro). These two perspectives are connected by the approaches using bidirectional Optimality Theory, where cognitive and/or pragmatic limitations constrain the grammatical evaluation. The more processing-oriented approaches focus on empirical data. Bringing together the two positions, this volume hopefully inspires researchers to continue bridging the gap between empirical and theoretical research on child language.

References Avrutin, Sergey 2006 Weak Syntax. In Broca’s region, Yosef Grodzinsky and Karin Amunts (eds.), 49 – 62. Oxford: Oxford University Press. Bloom, Paul, Andrew Baars, Laura Conway, and Janet Nicol 1994 Children’s knowledge of binding and coreference. Evidence from spontaneous speech. Language 70: 53 – 71. Blutner, Reinhard 2000 Some Aspects of Optimality in Natural Language Interpretation. Journal of Semantics 17: 189 – 216. Boersma, Paul, and Bruce Hayes 2001 Empirical tests of the Gradual Learning Algorithm. Linguistic Inquiry 32: 45 – 86.

Introduction

13

Burzio, Luigi 1998 Anaphora and soft constraints. In Is the Best Good Enough? Optimality and Competition in Syntax, Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky (eds.), 93 – 113. Cambridge, MA: MIT Press. Chien, Yu-Chin, and Kenneth Wexler 1990 Children’s knowledge of locality conditions on binding as evidence for the modularity of syntax and pragmatics. Language Acquisition 1, 225 – 295. Chomsky, Noam 1982 Some concepts and consequences of the theory of Government and Binding. Mass.: MIT Press. Conroy, Stacey, Eri Takahashi, Jeffrey Lidz, and Colin Phillips 2009 Equal treatment for all antecedents: How children succeed with Principle B. Linguistic Inquiry 40: 446 – 486. De Villiers, Jill, Jacqueline Cahillane, and Emily Altreuter 2006 What can production reveal about Principle B? In Proceedings of the Inaugural Conference on Generative Approaches to Language Acquisition–North America (GALANA), Kamil Ud Deen, Jun Nomura, Barbara Schulz, and Bonnie D. Schwartz (eds.), 89 – 100. (University of Connecticut Occasional Papers in Linguistics 4). Elbourne, Paul 2005 On the Acquisition of Principle B. Linguistic Inquiry 36(3): 333 – 365. Fenson, Larry, Philip S. Dale, J. Steven Reznick, Donna Thal, Elizabeth Bates, Jeffrey P. Hartung, Steve Pethick, and Judy S. Reilly 1993 The MacArthur Communicative Development Inventories: User’s guide and technical manual. San Diego: Paul H Brookes. Grimshaw, Jane, and Sara Thomas Rosen 1990 Knowledge and Obedience: The Developmental Status of the Binding Theory. Linguistic Inquiry 21: 187 – 222. Grodzinsky, Yosef, and Tanya Reinhardt 1993 The innateness of binding and coreference. Linguistic Inquiry 24: 69– 102. Hamann, Cornelia in press Binding and Coreference – views from child language. In Handbook of Generative Approaches to Language Acquisition, Jill de Villiers and Thomas Roeper (eds.). Kluwer: Springer. Hamann, Cornelia, Odile Kowalski, and William Philip 1997 The French “Delay of Principle B Effect”. In Proceedings of the Annual Boston University Conference on Language Development 21, Elizabeth Hughes, Mary Hughes, and Annabel Greenhill (eds.), 205 – 219. Somerville, Mass.: Cascadilla Press.

14

Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk

Hendriks, Petra, and Charlotte Koster (eds.) 2010 Special issue on asymmetries in language acquisition. Lingua 120 (8). Hendriks, Petra, and Jennifer Spenader 2005/6 When production preceeds comprehension. Language Acquisition 13: 319 – 348. Jakubowicz, Celia 1984 On Markedness and Binding Principles. In Proceedings of the Northeastern Linguistic Society (GSLA), Vol. 14, C. Jones and Peter Sells, (eds.), University of Massachusetts, Amherst. Kauschke, Christina, Anna Kurth, and Ulrike Domahs In press Acquisition of German noun plurals in typically developing children and children with speci¿c language impairment. Child Development Research. Koch, Corinna, Petra Schulz, and Napoleon Katsos 2010 Do children compute some or most scalar implicatures? Evidence from German. Poster presented at the COST-Action A33 “Let the children speak: Learning of Critical Skills across 25 Languages. A European-wide initiative on Language Acquisition and Language Impairment”, January 22 – 24, 2010, London. McDaniel, Dana, Cecile McKee, and Helen Smith Cairns 1996 Methods for Assessing Children’s Syntax. Cambridge: MIT Press. McKee, Cecile 1992 A comparison of pronouns and anaphors in Italian and English acquisition. Language Acquisition 2: 21 – 54. Müller, Anja 2010 Wie interpretieren Kinder nur? Experimentelle Untersuchungen zum Erwerb von Informationsstruktur. Unpublished Ph.D. diss., University of Potsdam. Papafragou, Anna, and Julien Musolino 2003 Scalar implicatures: Experiments at the Semantics-Pragmatics interface. Cognition 86: 253 – 282. Pater, Joe 2004 Bridging the gap between perception and production with minimally violable constraints. In Constraints in Phonological Acquisition, René Kager, Joe Pater, and Wim Zonneveld (eds), 219 – 244. Cambridge: Cambridge University Press. Reinhardt, Tanja 2006 Interface Strategies. Cambridge Mass: MIT Press. Reuland, Eric 2001 Primitives of Binding. Linguistic Inquiry 32 (3): 439–492.

Introduction

15

Ruigendijk, Esther 2008 Reference assignment in German preschool children. In Language Acquisition and Development: Proceedings of GALA 2007, Anna Gavarro and Maria Joao Freitas (eds.), 370 – 380. Newcastle: Cambridge Scholars Publishing. Ruigendijk, Esther, Naama Friedmann, Rama Novogrodsky, and Noga Balaban 2010 Symmetry in comprehension and production of pronouns: A comparison of German and Hebrew. Lingua 120 (8): 1991 – 2005. Santelman, Lynn and Peter W. Jusczyk 1998 Sensitivity to discontiniuous dependencies in language learners: Evidence for limitations in processing space. Cognition 69: 105 – 134. Smolensky, Paul 1996 On the Comprehension/Production Dilemma in Child Language. Linguistic Inquiry 27(4): 720 – 731. Spenader, Jennifer, Jan-Erik Smits, and Petra Hendriks 2009 Coherent discourse solves the Pronoun Interpretation Problem. Journal of Child Language 36: 23 – 52. Tesar, Bruce, and Paul Smolensky 1998 Learnability in Optimality Theory. Linguistic Inquiry 29 (2): 229 – 268. Vogel, Irene, and Eric Raimy 2002 The acquisition of compound vs phrasal stress: the role of prosodic constituents. Journal of Child Language 29: 225 – 250. Weissenborn, Jürgen, and Barbara Höhle 1998 Sensitivity to closed-class-elements in preverbal children. In Proceedings of the 22th Annual Boston Conference on Language Development, Vol.1, Annabel Greenhill, Mary Hughes, Heather Little¿eld, and Hugh Walsh (eds) 348 – 359. Somerville, Mass.: Cascadilla Press. Wimmer, Heinz, and Josef Perner (eds.) 1983 Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children’s understanding of deception. Cognition 13: 103 – 128. Zesiger, Pascal, Laurence Chillier-Zesiger, Marina Arabatzi, Lara Baranzini, Stephany Cronel-Ohayon, Julie Franck, Hans-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: 571 – 603. Zondervan, Arjen, Luisa Meroni, and Andrea Gualmini 2009 Experiments on the role of the Question under discussion for Ambiguity Resolution and Implicature computation in Adults. In Proceedings of SALT 18. http://ecommons.cornell.edu/handle/1813/5910.

Testing the Aspect Hypothesis in child Tamil Lavanya Sankaran

Introduction The acquisition of tense and aspect by ¿rst and second language learners has generated enormous interest in the past forty-¿ve years. With numerous studies being carried out in the ¿eld, certain crucial facts have come to light. Where ¿rst language acquisition is concerned it has been observed that in nearly all the child languages investigated there is a close relationship between the semantic properties of verbs and tense-aspectual markings. Children are observed to initially associate past and perfective inÀections with telic type verbs and progressive and imperfective inÀections with durative type verbs. These patterns of association have been referred to as “the Aspect Hypothesis” in the acquisition literature (Anderson and Shirai 1996). The Aspect Hypothesis in its simplest form makes the following generalizations: a) Children ¿rst use past marking (e.g., in English) or perfective marking (in Chinese, Spanish, etc.) with achievement and accomplishment verbs, eventually extending its use to activity and stative verbs. b) In languages that encode the perfective-imperfective distinction, imperfective past appears later than the perfective past, and imperfective past marking begins with stative and activity verbs, then extending to accomplishment or achievement verbs. c) In languages that have progressive aspect, progressive marking begins with activity verbs, then extends to accomplishment or achievement verbs. d) Progressive markings are not incorrectly overextended to stative verbs”. (Anderson and Shirai 1996: 533) This study tests whether these generalizations can be supported using evidence from child Tamil. A production task followed by a comprehension task was carried out with two child groups and an adult control group in order to test the inÀuence of verb semantics on children’s acquisition of aspect markers. The results also provide some insight into the long debated question of whether

18

Lavanya Sankaran

production precedes comprehension or vice-versa. This is the ¿rst time such a study has been carried out in Tamil and the ¿ndings provide valuable insight in the ¿eld. This paper is structured in the following manner: Section 2 examines previous research done in the area, section 3 explains the terminology used in the tense-aspect acquisition literature, section 4 provides insight into Tamil morphology, section 5 gives a detailed account of the present study and in section 6 the summary of results from the two experiments as well as the conclusion are discussed.

Previous research on L1 acquisition of tense and aspect Both longitudinal and experimental studies that have been carried out in various languages provide empirical evidence to support the generalizations made by the Aspect Hypothesis; French (Bronckart and Sinclair 1973), Italian (Antinucci and Miller 1976), English (Bloom, Lifter, and Ha¿tz 1980, Harner 1981, McShane and Whittaker 1988, Shirai and Anderson 1995, Johnson, and Fey 2006), Polish (Weist et al. 1984), Greek (Stephany 1981), Turkish (AksuKoç 1988), Mandarin Chinese (Li 1990) and Russian (Stoll 2005) to name a few. These studies investigating early verb morphology have helped shape the development of the theories regarding tense and aspect acquisition over the past four decades. It is crucial to remember, however, that the studies mentioned have mostly been conducted in languages where both tense and grammatical aspectual information are conÀated into a single morpheme. Languages such as English and French face this dif¿culty. For example the past marker in English conveys both perfectivity as well as past meaning and the imparfait in French conveys both the past meaning as well as imperfective aspect. Some languages, moreover, only grammaticize one of the two temporal concepts (e.g. Mandarin Chinese only marks grammatical aspect and Modern Hebrew marks only tense). These issues make it very dif¿cult for researchers to make accurate claims regarding what children use early verbal morphology for. A study in Polish by Weist et al. (1984), however, was a break-through in the ¿eld because Polish is a language where tense and grammatical aspect are marked using distinct linguistic markers. Their study serves to highlight how important it is to be able to investigate languages which grammaticize both tense and grammatical aspect as it makes it easier to discover the exact nature of the temporal concepts that are encoded in their verbal morphology. There are, however, certain drawbacks when investigating Polish and similar Slavic

Testing the aspect hypothesis in child Tamil

19

languages. In such languages the categories of tense and aspect are mutually bound and integrated within the verb stems where a choice is made between perfective and imperfective stems. Also, there is no one marker that can clearly denote tense or aspect. These issues potentially carry several analytical uncertainties. It is therefore crucial that other languages besides Polish are examined in order to be able to help clarify some fundamental questions regarding children’s use of early verb morphology. Tamil is an ideal language to study because it uses separate linguistic devices to code distinctions between tense and aspect. Tamil is also morphologically transparent, making it possible to determine which morphology is being used to mark tense and which morphology is being used to mark aspect. The fact that in theory all tense-aspectual markers can co-occur with all verb type categories in Tamil is another reason why it is advantageous to study this particular language since it allows for creative verb predicate combinations within experimental conditions. Since most of the studies conducted so far have been in Indo-European languages, this paper serves to highlight the importance of investigating the predictions of the Aspect Hypothesis in other languages, especially ones where tense and aspect are not conÀated. Different experimental methods to test the acquisition of tense and aspect The most popular methods of experimental investigation have involved production and comprehension tasks and they have in many ways proven to be the most effective. In the production tasks, children have usually been asked to describe play-situations enacted by toys or props after being given a neutralprobe question.1 Studies that have used this method include Bronckart and Sinclair (1973), Harner (1981), McShane and Whittaker (1988) and Li (1990). More recently Stoll (2005) carried out a production experiment using short ¿lms instead of toys or props. All these studies have been successful with even very young children and can help elicit a large amount of data within a short period of time. Production experiments have demonstrated that children mostly combine durative verb types such as activities with progressive-imperfective inÀections and telic verb types with perfective inÀections. Comprehension studies that examine whether children’s comprehension of inÀectional markers varies across lexical categories are also extremely successful with very young children. These studies mostly employ sentence to picture 1. Weist et al. (1984) used a non-neutral probe question when carrying out their production task, thereby creating a biased context in order to demonstrate that discourse factors can inÀuence verb-predicate patterns.

20

Lavanya Sankaran

matching tasks and those which have employed this method include Weist (1983); Weist, Wysocka, and Lyytinen (1991), and Li (1990). Stoll (1998) modi¿ed the traditional method of comprehension experimentation by using video stimuli and Wagner (2001) conducted a sentence to scene matching task where children were presented with actual-acted out events instead of pictures. The overwhelming ¿nding across all these studies is that children comprehend imperfective inÀections better with durative verb types and perfective inÀections better with telic verb types. In my study, I have incorporated both these experimental methods in order to test Tamil children’s production and comprehension of Tamil aspect markers. When carrying out such production and comprehension tasks, there is the added advantage of being able to test subjects from different age groups within a relatively short time-span in order to provide a cross-sectional developmental picture which clearly traces the changing relationship between lexical and grammatical aspect (Johnson and Fey 2006: 422–424).

Terminology Temporality conventionally involves three basic notions which are (a) the inherent temporal features/contours of a particular situation, (b) the different perspectives that can be taken and expressed with regard to the temporal course of a particular situation and (c) the temporal reference which relates the time of a situation to another time span (which is most often the time of utterance). These notions correspond to the terms lexical aspect, grammatical aspect and tense, respectively. Lexical aspect (also called aktionsart by some scholars) is encoded in the lexicon of natural language and denotes verb semantics.2 It should not be confused with the term aspect which is generally used to refer to grammatical aspect. Grammatical aspect and tense are mostly made evident by the grammatical morphological marking on the ¿nite verb. These three terms will be de¿ned in greater detail below.

2. Aktionsart is a German term that was introduced by Agrell (1908) to refer to the inherent semantic features of verbs. I shall not be using this term because it only alludes to the lexical content of verbs rather than to the lexical content of verb predicates (Klein 1994: 17).

Testing the aspect hypothesis in child Tamil

21

Lexical aspect Lexical aspect refers to the inherent temporal features of a verb that are not encoded in the morphology of a language. It is simply an intrinsic part of the semantics of the verb predicate that expresses the situation or action. Vendler (1967) proposed that the temporal features by which verb types should be categorised in a given language are telicity, durativity and dynamicity. Telicity refers to the internal structure of a situation as having an inherent endpoint or a natural completion where there is a goal, outcome or other change of state. Durativity describes a situation as lasting for a period of time, thus emphasizing that it has internal structure. Dynamicity involves change and denotes the energy required for a particular situation to exist or continue. The table below clearly illustrates how the inherent lexical aspect of verbs can be classi¿ed according to their temporal features. Table 1. Internal temporal features of verb types Situations States Activity Accomplishment Achievement Examples: States (internal states): Activities: Accomplishments: Achievements:

Telic í í + +

Durative + + + í

Dynamic í + + +

believe in fairies, know the answer laugh, swim paint a picture, eat an apple reach the mountain top, fall

Posture verbs in some languages (e.g. Mandarin Chinese and Tamil) can also be categorized under stative verbs (e.g. sit, stand). After doing some language tests and discovering that posture verb types behave like internal states, I used posture verbs instead of internal states in my study because they lend themselves to experimental manipulation better. Grammatical aspect “Aspects are different ways of viewing the internal temporal constituency of a situation” (Holt 1943: 6). It is considered a grammatical category and may be expressed by means of the inÀectional morphology of that particular language (Comrie 1976: 9). There are two main types of aspectual perspectives, the

22

Lavanya Sankaran

imperfective and the perfective. The imperfective perspective focuses on the inside of a situation without specifying its initial or ¿nal endpoints. This includes the durative/continuous and the progressive, where the latter is a subdivision of the former. Progressiveness is usually described as the combination of continuousness with non-stativity (e.g. John is running). The perfective perspective focuses on a situation from the outside, as a single unanalysable whole (e.g. John ran to the market). Interestingly, in many languages the perfective forms of certain verb types (statives in particular) can be used to indicate the beginning of a situation in order to give an inceptive reading (Comrie 1976: 19). In Mandarin Chinese, for instance, a stative predicate (e.g. tƗ hen gƗo ‘he is very tall’) in its perfective form becomes (e.g. tƗ gƗo-le ‘he became tall, has become tall’). This example illustrates that certain perfective forms refer to the inception of a situation rather than to its completion. In this paper I will use the term inceptive aspect to denote entry into a particular state, process or event. The full aspectual meaning of a sentence is thus derived from the interaction between the situation and the ‘viewpoint’ taken of that situation. Tense Tense marks temporal deixis and usually locates the time of a situation relative to the utterance time. Tense is formally marked using inÀections, but it can also be expressed through other linguistic devices such as adverbials. The three most common tenses are the past, present and future tenses. (1)

Tina cried yesterday. (past tense)

(2)

Tina is eating now. (present tense)

(3)

Tina will have some coffee after sunrise. (future tense)

Tamil morphology Tamil belongs to the southern branch of the Dravidian language family and it has been in existence since at least 500 B.C. It is spoken mainly in Tamil Nadu (India) and amongst signi¿cant minority communities in Sri Lanka, Singapore, Malaysia, Dubai, Mauritius and South Africa. Tamil is a head-¿nal language and an ideal language to study for the present purposes because (a) it grammaticizes both the temporal concepts aspect and tense and (b) it uses separate

Testing the aspect hypothesis in child Tamil

23

linguistic devices to code aspect versus tense. Interestingly, tense is obligatorily marked in Tamil while aspect is optionally marked. Since obligatory markers are usually acquired before optional ones, the likelihood therefore is that Tamil children will mark tense before they mark aspect or that they may use tense markers to denote aspectual distinctions or tense distinctions or both. This will be tested in this study along with the Aspect Hypothesis. The simplest form from which all parts of the Tamil verb are formed is known as the root or verb stem. Tamil is an agglutinating language that is morphologically transparent and inÀections are marked with suf¿xes attached to its verb stem. A simple indicative verb consists of the verb stem with the addition of one or two suf¿xes (Asher 1985: 166). An example of the indicative with aspect and tense suf¿xes is below. Verb stem + aspect + tense + person number gender (PNG) Aspect and tense markers in Tamil Aspect markers in Tamil are denoted by aspectual morphemes that are attached to the ¿nite verb stem. They follow only verbal participles and are followed by a tense marker and a PNG marker.3 The aspectual markers that I will be testing in my experiment will be the morphemes kondiru and vidu. The morpheme kondiru marks the imperfective aspect irrespective of the verb type it is associated with.4 It is considered one of the most morphologically complex aspectual markers since it is a combination of the verbal participle form of the lexical verb kol ‘to hold’ and the perfect auxiliary iru (Lehmann 1993: 207). Below is an example illustrating the use of kondiru with an activity verb type.

3. A verbal participle expresses an action that is preceding or forming part of, is simultaneous with or leading up to the action of the main verb. Verbal participles are variations of verbal stems that are morphologically unanalysable. It is important to note that although the verbal participle occurs with one of the following markers; /ttu/, /ntu/, /i/ or /PuĺPPu/, which are allomorphs of the past tense morpheme /tu/ in ¿nite verbs, these markers do not convey the past tense (Kothandaraman 2001: 45). 4. Note that Tamil does not mark the progressive perspective separately from the imperfective perspective.

24 (4)

Lavanya Sankaran Nina romba neramaa vilaiyaadi-ttiru-nthaal. Nina long time-for play-kondiru-past.3rd.SG.FEM. ‘Nina was play-ing for a long time.’

The morpheme vidu expresses the perfective aspect when associated with all non-stative verb types (Lehmann 1993: 209). Example (5) illustrates the use of vidu with an accomplishment verb type. (5)

avan viidu katti-t-taan. he house build-vidu-past.3rd.SG.MASC. ‘He built a house.’

Crucially, vidu also expresses inceptive aspect when associated with stative verb types/posture verb types. See the example below. (6)

naan ukkaarnthi -t-teen. I sit -vidu-past.1st.SG. ‘I sat.’

In example (6), when the posture verb ukkaaru ‘sit’, is inÀected with the perfective marker, vidu, it becomes ukkaarnthi-t-teen ‘I sat’ where there is focus on inceptive aspect. Both kondiru and vidu can be combined with all three tenses, the past, present and future, where the tense marker is denoted by an overt suf¿x attached to the verb.5 Tense markers in Tamil are represented by a large distribution of allomorphs that are not phonologically conditioned. Grammatical aspect need not necessarily be explicitly marked in Tamil. But, the lack of an aspectual marker does not mean that an aspectual notion is absent. Importantly, a tense marker combined with an adverbial in Tamil can imply an aspectual reading in the absence of an explicit aspectual morpheme. Take the following examples for instance. In example (7), in spite of the absence of vidu, one receives a perfective reading and in example (8), in spite of the absence of kondiru, one receives an imperfective reading.

5. Where tense is concerned, I will be mainly looking at the distribution and function of the past and present tense markers since they have relevance to the predictions of the ‘Aspect Hypothesis’.

Testing the aspect hypothesis in child Tamil (7)

Leela dabakkannu kiizre Leela adv(suddenly/with a ‘thud’) down ‘Leela fell down suddenly with a thud.’

(8)

naanku naatkalaaka mazrai vidaamal pei-kirathu. four days-for rain continuously rain-present. ‘For four days it has been raining continuously.’

25

vizru-nthaal. fell-past.3rd.SG.FEM.

Recall that aspectual marking in Tamil is an optional category but that tense marking is obligatory. When a verb is not explicitly marked for aspect, grammatical devices such as adverbials and tense markers provide the necessary cues to indicate the aspectual perspective the situation can be viewed from. In example (7) above, the past tense marker and the adverbial dabakkannu ‘suddenly/with a thud’ together help to give a perfective reading and in example (8), the present tense marker and the adverbial vidaamal ‘continuously’ together help to give an imperfective reading. This has important implications for the acquisition of aspect markers in child Tamil. The present study I carried out a production task followed by a comprehension task with one adult group and two child groups. The production task was systematically followed by the comprehension task because it was observed from the pilot experiments that the production task helped the children familiarize themselves with the verb types that were used in the comprehension task. Participants and predictions The adults were from Chennai, Tamil Nadu and were from similar socioeconomic backgrounds which means their variety of spoken Tamil was similar to one another. Their ages ranged from 20 to 27 years (M: 23;6) and they formed the control group. The two child groups comprised of an older group and a younger group and they were from a state school in Chennai, Tamil Nadu. There were 20 children in the older group and their ages ranged from 3;7 to 4;8 years (M: 4;1). The younger group was made up of 16 children and their ages ranged from 2;4 to 3;6 years (M: 3;0). There were an equal number of males and females in all three age groups and each subject was tested individually. It is predicted that the results will conform to the generalizations made by the Aspect Hypothesis, namely that the verb predicate patterns produced by the

26

Lavanya Sankaran

subjects will show strong associations between past and perfective inÀections and present and imperfective inÀections respectively. It is also predicted that the perfective past is acquired before the imperfective past, another stipulation of the Aspect Hypothesis. This latter prediction will be speci¿cally tested in the comprehension task where children will be expected to perform better with items involving vidu and the past tense compared to kondiru and the past tense. The production task The purpose of this task was to test how verb semantics inÀuences the subjects’ verb predicate patterns when they produced tense and aspect markers. Speci¿cally, I observed how children used the perfective marker vidu and the imperfective marker kondiru when asked to describe the temporal contours of different situations. Material and procedure The subjects were shown a display of toys and told that they were going to observe a series of situations enacted by them. Once a situation was enacted, the subject was required to explain to a puppet what he saw. After the initial set of instructions, the experimenter made it clear that she was not going to say anything during the experiment. This was to ensure that the verb inÀections produced by the subject would not be inÀuenced by the verb inÀections used by the experimenter. Occasionally, when a child needed encouragement, the experimenter would say athu patri sollu ‘tell me about that one’. The verb sollu is in the second person imperative form, without a tense or aspect marker, and so avoids the use of temporal inÀections. This was a neutral elicitation question creating an ambiguous context where there is no preferred tense/aspectual marker expected to feature in their response. It was used (when necessary) in order to elicit the verb predicate con¿gurations that are most natural in Tamil. There were a total of 15 situations acted out for the subjects, three of which were used as practice situations to get the subjects acquainted with the task. The 12 situations that were speci¿cally used for the experiment were categorised according to their unique temporal properties and divided equally amongst the four different situation types, namely, the posture situation, the activity situation, the accomplishment situation and the achievement situation. In other words, three different instances of each of the four situation types were enacted for the subjects. At no time was one particular situation type shown in two consecutive enactments. There was also no predictable order in which the situation types to be acted out were arranged.

Testing the aspect hypothesis in child Tamil

27

Posture situation types (e.g. sit, lie and stand) were acted out with the aid of a doll with the Àexible limbs. The situations illustrating the different postures made the inception of the situation and its resulting state evident. The activity situation types (e.g. a car going round in circles, a doll walking continuously, the actor combing her hair) were enacted to elicit process verb predicates that involved no inherent endpoint. The accomplishment situation types were enacted to make the subjects aware of the dynamicity and duration of the situations and then their ultimate result (e.g. drawing a Àower and completing the picture, the actor ¿nishes drinking a glass of water, the actor puts on a pair of socks). Finally, the achievement situation types showed the subjects a punctual resultative event (e.g. a stick breaks into two, a car crashes into a chair, a doll falls down). Each situation type was acted out a maximum of three times. The subjects’ description of the enacted situations was then coded according to three criteria; the verb type used, the type of aspect marker (if it was used at all) used and the type of tense marker used. Data analysis and results In order to test the predictions of the aspect hypothesis I ¿rst looked at the verb predicate patterning of tense markers (past and present) across the two child groups and compared it to the verb predicate patterns in the adult control group.6 Figure 1 below illustrates what was found. In ¿gure 1 it is observed that there is a strong bias in the distribution of tense markers according to verb semantics (note that the ¿gure illustrates the distribution of tense markers with the different verb types, without specifying whether aspectual auxiliaries were produced as well). All three subject groups strongly associated past inÀections with accomplishment and achievement verb types and present inÀections with posture and activity verb types. This was supported using statistical evidence from running a three-way mixed-design ANOVA. It was revealed that there was a highly signi¿cant main effect of verb type, F (3, 123) = 153.96, p < 0.001, which conform to the generalizations of the Aspect Hypothesis, namely that children are observed to initially associate past and perfective inÀections with telic type verbs and present and imperfective inÀections with durative type verbs.7 6. All the children in the two subject groups produced the target verbs just like the adults did. 7. The adult data supports the predictions of the ‘Distributional Bias Hypothesis’ since the adult group also shows a distributional bias based on verb semantics. The Distributional Bias Hypothesis posits that ‘native speakers in normal interaction with

28

Lavanya Sankaran

*POS-posture, ACT-activity, ACC-accomplishment, ACH-achievement *CHI-children, ADU-adults *misc-modals or unrecognizable markers Figure 1. The distribution of tense inÀections produced with all verb types for all subject groups

Next, an analysis of the distribution of aspect markers was carried out across all three subject groups. Figure 2 shows the distribution of vidu and kondiru across the three subject groups with all the four verb type categories.

*POS-posture, ACT-activity, ACC-accomplishment, ACH-achievement *CHI-children, ADU-adults Figure 2. The distribution of aspect markers produced with all verb types for all subject groups

other native speakers tend to use each verb morpheme with a speci¿c class of verbs, also following the Aspect Hypothesis (Anderson and Shirai 1994: 137–139).

Testing the aspect hypothesis in child Tamil

29

The data illustrated in ¿gure 2 also support the predictions of the Aspect Hypothesis since the perfective vidu is strongly associated with accomplishments and achievements and the imperfective kondiru is strongly associated with posture verb types and activities across all subject groups. Interestingly, the use of vidu is more or less equal across all three subject groups for both accomplishments (ADU: 41.7%, Older CHI: 36.7%, Younger CHI: 35.4%) and achievements (ADU: 66.7%, Older CHI: 56.7%, Younger CHI: 39.6%). While vidu is not used by any of the subject groups with activities, it is used to a small extent with posture verb types by the two child groups to express inceptive aspect. With regards to kondiru it is overwhelmingly used only by the adult group. The older children use it only 16.7% of the time with posture verb types and the younger group only 6.3% of the time (compared to the adults: 54.2%). With regards to activity verb types, the rates of use are even lower. The older children only use kondiru 3.3% of the time, while the younger children don’t use it at all. This is signi¿cantly different from the adults who use kondiru with activities 50% of the time. The results of a three-way mixed-design ANOVA revealed that the interaction between subject group and verb type was signi¿cant, F (6, 123) = 2.47, p < 0.05, meaning that the adult group differed signi¿cantly from the child groups in the way they chose to inÀect the durative verb types. Posture and activity verb types were mostly inÀected with kondiru by the adults to mark imperfectivity, while the two child groups mostly used just the present tense marker with these durative verb types to do so. In fact, the overall use of just the tense marker to express an aspectual perspective was more than the adults. Figure 3 below helps to illustrate this. As the use of aspect markers is not obligatory in Tamil, the production task also tested whether this had any inÀuence over children’s rates of use of vidu (perfective) and kondiru (imperfective). It was found that it does since the children were more likely to use the obligatory tense inÀections rather than the optional aspectual inÀections. Figure 3 clearly shows that the children from both age groups used tended to use ‘only tense’ markers compared to the adults (Younger CHI: 55.2%, Older CHI: 53.8%, ADU: 34.4%). Recall from ¿gure 1 that they used the past tense marker (found mostly with telic type verbs) to imply perfectivity and the present tense marker (found mostly with durative type verbs) to imply imperfectivity. Figure 3 also serves to illustrate that apart from the use of the perfective vidu (which the children used almost as frequently as adults), their use of kondiru was much lower than that of the adults (Older CHI: 5.8%, Younger CHI: 1.7%, ADU: 28.1%). Note that although the younger children’s overall

30

Lavanya Sankaran

*CHI- children, ADU- adults *other asp refers to other aspect markers (apart from vidu and kondiru) that were used. (These markers were mainly attitudinal) *Note: when using vidu, kondiru and other asp the subjects also inÀected for tense Figure 3. Tense vs aspect markers produced for all subject groups

use of vidu and kondiru is lower than the older children’s, there is no real difference between the two groups. Summary of results Overall, the results indicate that vidu has been acquired to an almost adult level, while kondiru still seems to be in the process of being acquired. One of the stipulations of the Aspect Hypothesis is that “in languages that encode the perfective-imperfective distinction, the imperfective past appears later than the perfective past” (Anderson and Shirai 1996: 533). This prediction seems to be borne out since both child groups only used the present tense inÀection when producing kondiru and never the past tense inÀection. The adult data, however, showed 3 instances where they used the imperfective past. The comprehension task The purpose of this task was to test how verb semantics inÀuences the subjects’ comprehension of the perfective vidu and the imperfective kondiru respectively. It also tested whether the perfective past was understood better than the imperfective past as stipulated by the Aspect Hypothesis.

Testing the aspect hypothesis in child Tamil

31

Material and procedure A sentence to picture matching task was carried out where the subjects were required to match the target sentence with the appropriate picture sequence and the responses were tabulated as being either correct or incorrect. There were altogether 24 test sentences that were presented with 24 picture sets, where each of the 4 situation types (posture, activity, accomplishment and achievement situation types) had 6 picture sets representing each situation type. In each set there were two picture sequences representing the same situation. One picture sequence corresponded with perfectivity while the other corresponded with imperfectivity. The experiment was controlled in such a way that for a particular situation type, 3 picture sets were shown to the subjects while 3 perfective test sentences (incorporating vidu and the past tense) were read out. Then, 3 other picture sets (also representing the same situation type) were shown to the subjects while 3 imperfective test sentences (incorporating kondiru and the past tense) were read out. In the task, the subject was shown one picture set (i.e. two picture sequences that formed a set) at a time.8 See ¿gure 4 below.

Figure 4. Activity verb ‘swim’

8. The subject groups were divided into halves. The ¿rst half received test sentences from Series 1 and the second half received test sentences from Series 2. The verb types that were inÀected with vidu in Series 1 were inÀected with kondiru in Series 2 and the verb types that were inÀected with kondiru in Series 1 were inÀected with vidu in Series 2. This was to control for verb-speci¿c effects.

32

Lavanya Sankaran

For each picture set (like the one above), the experimenter would describe what was going on in the two picture sequences in order that the subject would be clear as to the differences between the two. The experimenter was careful not to use the perfective and imperfective markers when describing the picture sequences. Note that the order in which the two picture sequences were placed was random. This precaution was taken to ensure that even if the children kept choosing the ¿rst picture sequence presented to them, the ¿nal results of the experiment would not be due to just chance. After describing the picture sequences, the experimenter read out the test sentence to the subject. With regards to the picture set in ¿gure 4, the test sentence was either the perfective sentence with vidu, i.e. Senthil niinji-t-taan ‘Senthil swam’ or the imperfective sentence with kondiru, i.e. Senthil niinji-ttiru-nthaan ‘Senthil was swimming’. The experimenter then said kaami ‘show me’, asking the subject to point to the picture sequence that best corresponded with the test sentence. As mentioned earlier, the verb in the test sentence was inÀected with either vidu or kondiru. It is important to note that tense was kept constant such that all the verb types were inÀected with the past tense marker, regardless of the aspect marker in the verb predicate. This was to control for any tense-effects and to ensure that the subjects would correlate the aspect marker with the appropriate picture sequence rather than the tense marker. It is crucial to note that no temporal adverbials were included in the test sentences since they might have potentially given a clue as to what perspective the aspectual markers were focusing on. Examples (9) – (12) are instances of the test sentences that were used in the experiment. The examples show how each of the verb types was inÀected with either the perfective vidu or imperfective kondiru. (9)

Kumar paduthu-t-taan. Kumar lie down-vidu-past.3rd.SG.MASC. ‘Kumar lay down.’ (posture + perfective)

(10)

Senthil niinji-ttiru-nthaan. Senthil swim-kondiru-past.3rd.SG.MASC. ‘Senthil swam.’ (activity + imperfective)

(11)

Uncle viidu katti-t-taar. Uncle house build-vidu-past.3rd.SG. ‘Uncle was build-ing a house.’ (accomplishment + perfective)

Testing the aspect hypothesis in child Tamil (12)

33

thengaa marathilirunthu vizrunthi-ttiru-nthathu. coconut tree-from fall-kondiru-past.3rd.SG.NEUT. ‘The coconut was fall-ing from the tree.’ (achievement + imperfective)

Each subject’s response was considered correct if he or she matched the test sentence with the picture sequence that best described it. The response was considered incorrect, however, in the following instances: (a) the subject matched the test sentence with the inappropriate picture sequence; (b) the subject’s response was ambiguous in that he or she chose both the picture sequences in the set; (c) the subject vacillated between one picture sequence and another; (d) the subject pointed to individual pictures in a way that demonstrated that he or she clearly did not understand the purpose of the aspectual marker in that sentence; i.e. an instance would be when a verb, i.e. ‘swim’, which was inÀected for the imperfective marker kondiru was read out and the child pointed to the picture of the boy who had come out of the swimming pool and was towelling himself dry. Refer to the activity picture set above. The picture sets from all the four situation types were scrambled up so there was no predictable order in which they were presented. Under no circumstances were two instances of the same situation type category presented consecutively. The type of aspectual marker (i.e. vidu or kondiru) that the children heard in the test sentences was also randomised. Each test sentence was read out a maximum of three times to the subjects. As in the production task the subjects were given three practice trials before the actual experiment began. Data analysis and results As expected, the adults chose the correct picture sequence every time a test sentence was read out. In other words, the adult group comprehended both vidu and kondiru with all four verb types correctly 100% of the time. Their results were compared with the results of the two child groups whose responses were not 100% correct. Note that the two child groups’ responses, however, were not signi¿cantly different. After running a three-way mixed design ANOVA with the results from the two child groups, it was observed that there was a highly signi¿cant interaction of the verb type with the aspectual marker, F (3, 102) = 16.54, p < 0.001. This implies that the child would give signi¿cantly more correct responses when presented with a certain combination of aspect marker and verb type. After collapsing the results across both child groups, it was observed that the children were more likely to give the correct response when kondiru was combined with durative verb types (posture and activity verb types) and when vidu was

34

Lavanya Sankaran

combined with telic verb types (accomplishments and achievements). This supports the predictions of the Aspect Hypothesis. Figure 5 below illustrates how certain combinations of aspect markers with different verb type categories inÀuence the type of response the children give. Note that the results of both the child groups have been collapsed and that ¿gure 5 shows the percentage of correct responses to each verb type + aspect marker combination.

*POS-posture, ACT-activity, ACC-accomplishment, ACH-achievement Figure 5. Percentage of correct responses for the verb types combined with the two aspect markers

While it is evident from the statistical analysis that verb semantics inÀuences the acquisition of aspect markers, from ¿gure 5 it seems equally evident that the children are more likely to respond correctly when presented with kondiru compared to when they are presented with vidu. The mixed design ANOVA con¿rms this by showing that there was in fact a highly signi¿cant effect of aspectual marker, F (1, 34) = 94.32, p < 0.001. This ¿nding is contradictory to what was expected. Summary of results In the production task, the children produced vidu with nearly the same frequency as the adults, while they produced kondiru with signi¿cantly less frequency. Also, when producing kondiru in the production task, it was always combined with the present tense marker, never with the past tense marker. The results of the comprehension experiment, however, are not consistent with the previous ¿ndings. It was found in fact that the children responded signi¿cantly better when presented with the imperfective past than when they were

Testing the aspect hypothesis in child Tamil

35

presented with the perfective past. The conÀicting results from the production and comprehension experiments have important implications for the debate regarding whether production precedes comprehension. These implications will be discussed in the next section.

Discussion and conclusion The production and comprehension studies carried out to investigate Tamil children’s acquisition of the aspectual markers vidu and kondiru show that there was no signi¿cant developmental effect between the two age groups tested. Children from a bigger range of age groups would need to participate in the experimental studies for there to be a better understanding of how temporal development takes place and at what stages. The overall ¿ndings reveal that the Aspect Hypothesis is supported to some extent. In the production task there was a clear pattern of co-occurrence between certain verb type categories and tense and aspect markers. Telic verb types overwhelmingly co-occurred with the perfective past markers and durative verb types overwhelmingly co-occurred with the imperfective present markers and this was observed across all the subject groups. In the comprehension task, while the adult groups responded correctly 100% of the time demonstrating that they comprehended vidu and kondiru regardless of the verb types they co-occurred with, the children responded signi¿cantly better when vidu co-occurred with a telic verb type category and when kondiru co-occurred with a durative verb type category. These ¿ndings suggest that verb semantics has an important role to play in children’s acquisition of aspect markers. Interestingly, there was one noticeable discrepancy when comparing the two experimental ¿ndings, one which did not accord with one of the Aspect Hypothesis’ predictions. It was found in the production task that the children, while producing vidu with nearly the same frequency as the adults, very rarely produced kondiru. If they did produce the imperfective marker at all, it would co-occur with the present tense inÀection. Recall that one of the stipulations of the Aspect Hypothesis is that the imperfective past is acquired after the perfective past, so the results in the production task correspond to the hypothesis. However, in the comprehension task the children performed signi¿cantly better with test sentences that incorporated kondiru with the past tense marker compared to those which incorporated vidu with the past tense marker. One explanation could be that the children probably became confused when presented with vidu because its function is not consistent across all verb type

36

Lavanya Sankaran

categories unlike the function of kondiru. The illustrative nature of the comprehension task also may have contributed to making the children especially confused as it may have highlighted the discrepancy in the function of vidu. When vidu combines with activities, accomplishments and achievements it focuses on the termination or the result of a situation, but when it combines with statives (posture verb types in both the experiments) it focuses on inceptive aspect. The imperfective marker kondiru, on the other hand, consistently presents the imperfective perspective when it combines with all four verb type categories (and this was clearly illustrated with the picture sequences). There is, therefore, a one-to-one correspondence between form and function with regard to kondiru, whereas vidu has the dual role of conveying inceptive aspect as well as termination/result. This could account for why the children responded better when presented with kondiru than when they were presented with vidu in the comprehension experiment. In summary, when examining the comprehension results in the light of what was discovered in the production task, it seems evident that while the children seemed to understand the function of the marker kondiru, they found it dif¿cult to produce its form. Recall that it is considered one of the most morphologically complex auxiliaries in Tamil (Lehmann 1993: 207). The ¿ndings from the production and comprehension experiments together reveal that perhaps comprehension precedes production where the imperfective marker kondiru is concerned. With regards to the perfective marker vidu, it seems evident that the children from both age groups were initially only able to match the form with its perfective/resultative function and not its inceptive function. In other words, the children were able to both comprehend as well as produce vidu as a perfective marker in the production task. Their performance relating to vidu suffered in the comprehension task as a result of their inability to recognise that vidu has a dual function. The ¿ndings from the two experiments thus suggest that comprehension does seem to precede production and that children are still in the process of acquiring kondiru, but that vidu, the perfective marker rather than the inceptive marker, has already been acquired to an almost adult level. In conclusion, the Aspect Hypothesis has provided the framework necessary to investigate the inÀuence of verb semantics on Tamil children’s acquisition of temporal-aspectual inÀections. Testing the Aspect Hypothesis is important because it has a bigger role to play in acquisition than just predicting the kinds of verb predicate patterns learners will produce. Researchers believe that only by understanding the precise nature of the relationship between lexical aspect, grammatical aspect and tense and by understanding what early temporal verbal morphology encodes, will they be able to: gain insight into whether the verb predicate patterns observed are innate or learned, identify what kind of

Testing the aspect hypothesis in child Tamil

37

relationship lies between conceptual development and language development, discover the processes involved during tense and aspect acquisition as well as to determine whether aspect is acquired before tense (Shirai, Slobin, and Weist 1998: 245 – 246).

References Agrell, Sigurd 1908 Aspektänderung und Aktionsartbildung beim polnischen Zeitworte. Ein Beitrag zum Studium der indogermanischen Praeverbia und ihrer Bedeutungsfunktionen. Lund: Ohlsson. Aksu-Koç, Ayhan 1988 The acquisition of aspect and modality: The case of past reference in Turkish. Cambridge, UK: Cambridge University Press. Andersen, Roger. W., and Yasuhiro Shirai 1994 Discourse motivations for some cognitive acquisition principles. Studies in Second Language Acquisition 16: 133 – 156. Anderson, Roger. W., and Yasuhiro Shirai 1996 The primacy of aspect in ¿rst and second language acquisition: The pidgin-creole connection. In Handbook of Second Language Acquisition, William C. Ritchie and Tej K. Bhatia (eds.), 527 – 570. New York: Academic Press. Antinucci, Fransesco, and Ruth Miller 1976 How children talk about what happened. Journal of Child Language 3: 627 – 643. Asher, Ron E. 1985 Tamil. London: Croom Helm. Bloom, Lois, Karin Lifter, and Jeremie Ha¿tz 1980 Semantics of verbs and the development of verb inÀection in child language. Language 56: 386 – 412. Bronckart, Jean-Paul and Hermina Sinclair 1973 Time, tense and aspect. Cognition 2: 107 – 130. Comrie, Bernard 1976 Aspect. Cambridge: Cambridge University Press. Harner, Lorraine 1981 Children talk about the time and aspect of actions. Child Development 52: 498 – 506. Holt, Jens 1943 Études d’aspect. Aarhus: Universitetsforlaget. Johnson, Bonnie W., and Marc E. Fey 2006 Interaction of lexical and grammatical aspect in toddlers’ language. Journal of Child Language: 419 – 435.

38

Lavanya Sankaran

Klein, Wolfgang 1994 Time in language. London: Routledge. Kothandaraman, Pon 2001 Tamil studies: selected papers. Ambuli Publications. Lehmann, Thomas 1993 A grammar of modern Tamil. Pondicherry Institute of Linguistic Culture. Li, Ping 1990 Aspect and aktionsart in child Mandarin. Unpublished Ph.D. diss., MaxPlanck Institute for Psycholinguistics, Nijmegen. McShane, John, and Stephen Whittaker 1988 The encoding of tense and aspect by three-to-¿ve-year-old children. Journal of Experimental Child Psychology 45: 52 – 70. Shirai, Yasuhiro, and Roger W. Andersen 1995 The acquisition of tense-aspect morphology: A prototype account. Language 71 (4): 743 – 762. Stephany, Ursula 1981 Verbal grammar in Modern Greek early child language. In Child Language: An International Perspective, Philip S. Dale and David Ingram (eds.), 45 – 57. Baltimore: University Park Press. Stoll, Sabine 1998 The role of aktionsart in the acquisition of Russian aspect. First Language 18: 351 – 377. Stoll, Sabine 2005 Beginning and end in the acquisition of the perfective aspect in Russian. First Language 32: 805 – 825. Wagner, Laura 2001 Aspectual inÀuences on early tense comprehension. Journal of Child Language 28: 661 – 681. Weist, Richard. M. 1983 Pre¿x versus suf¿x information processing in the comprehension of tense and aspect. Journal of Child Language 10: 85 – 96. Weist, Richard M., Hanna Wysocka, Katarzyna Witkowska-Stadnik, Ewa Buczowska, and Emilia Konieczna 1984 The defective tense hypothesis: On the emergence of tense and aspect in child Polish. The Journal of Child Language 11: 347 – 374. Weist, Richard. M., Hanna Wysocka, and Paula Lyytinen 1991 A cross-linguistic perspective on the development of temporal systems. Journal of Child Language 67 – 92. Vendler, Zeno 1967 Linguistics in philosophy. Ithaca, NY: Cornell University Press.

An asymmetry in the acquisition of accusative clitics in child Romanian* Martine Coene and Larisa Avram

1

Introduction

In spite of the fact that the distinction between 1st/2nd and 3rd person pronouns has been acknowledged in the literature for a long time (Benveniste 1966; Postal 1966) most studies dealing with accusative clitics chose to focus on what these pronominals had in common – de¿ciency and distribution – irrespective of their person value. This focus in the theoretical literature had an immediate resonance in the domain of acquisition. Most studies which analysed the development of accusative clitics on the basis of longitudinal data did not make any (explicit) difference between 1st/2nd vs. 3rd person, while those relying on experimental data looked exclusively at 3rd person accusative clitics.1 Among the few theoretical studies which explicitly address the differences between 1st/2nd vs. 3rd person in the domain of de¿cient pronominals is Kayne (2000). The main claim is that in French and Italian 1st/2nd person accusative clitics (m- and t-) belong to a natural class which excludes 3rd person accusative clitics (l-) but which includes the reÀexive clitic s-. According to Kayne, only 3rd person non-reÀexive accusative clitics are determiner-pronouns (D-pronouns). Similarly, Uriagereka (1995) argues that 3rd person clitics alone are of category D, whereas 1st/2nd person clitics are DPs. He also notices that reÀexive se might belong to a category different from D (p. 85). * Work on this paper was supported by CNCSIS – UEFISCU, project grant PN II IDEI 1979/2008 to Larisa Avram. 1. An anonymous reviewer points out that experimental data were related, in most cases, to binding theory, which might explain the focus on 3rd person accusative clitics and reÀexive clitics. This is indeed the case for several studies (Jakubowicz 1989; Baauw 2000; Hamann 2002, a.m.o.). It is equally true that in some previous studies relying on longitudinal data one can reconstruct from the tables the difference between the acquisition of 1st, 2nd and 3rd person accusative clitics. What we noticed, though, in previous studies is a lack of focus on the relevance for the acquisition process of the distinction between 1st/2nd person accusative clitics, on the one hand, and 3rd person accusative clitics, on the other.

40

Martine Coene and Larisa Avram

The immediate prediction which one can make with respect to acquisition is that there might be developmental differences between 1st/2nd vs. 3rd person accusative clitics, on the one hand, and developmental similarities between reÀexive clitics and 1st/2nd person non-reÀexive accusative clitics, on the other hand. The difference between reÀexive and non-reÀexive accusative clitics has been addressed in the acquisition literature (Jakubowicz 1989; Hamann, Rizzi, and Frauenfelder 1996; Jakubowick et al. 1998; Crysman and Müller 2000; Zesiger et al. 2010). The results indicate that reÀexive clitics are produced more often than non-reÀexive accusative clitics, both in longitudinal and in experimental data. But no acquisition study has explicitly compared the developmental pattern of 1st/2nd person accusative clitics to that of 3rd person accusative clitics or that of 1st/2nd person accusative clitics to reÀexives. This is precisely the aim of the present paper. On the basis of empirical data coming from child Romanian we investigate (i) whether there is a difference between the developmental pattern of 1st/2nd person accusative clitics and that of 3rd person accusative clitics, and (ii) whether one can identify a similar developmental pattern of 1st/2nd person accusative clitics and reÀexive clitics. One should mention from the very beginning that, although we invoke cross-linguistic data at various points in our analysis, the focus is on the Romanian data. Consequently, some of the theoretical conclusions may not straightforwardly extend to other Romance languages. The remaining of the paper is organized as follows: Section 2 presents the theoretical background and shows in what way the Romanian data can be integrated into previous theoretical analyses. In Section 3 we discuss the developmental facts and in Section 4 we provide an explanation for the observed phenomena. The account proposed here is based on an analysis of accusative clitics in Romanian which follows Uriagereka’s (1995) proposal for Romance and on a particular implementation of Rizzi’s Relativized Minimality (1990). The conclusions are summarized in Section 5.

2

Romance 1st/2nd person accusative clitics vs. 3rd person accusative clitics

2.1

Kayne’s analysis

Kayne (2000) argues that in French and Italian the accusative clitics m- and t- are person morphemes, whereas 3rd person accusative clitics are “determiner pronouns”, identical to the de¿nite article. The 3rd person l- is different from

An asymmetry in the acquisition of accusative clitics in child Romanian

41

m- and t- because (i) only l- clitics have a word marker reÀecting gender; (ii) only l- clitics can show number distinction; (iii) in some Italian dialects they behave differently with respect to clitic doubling; (iv) l- does not combine with the possessive morpheme which is marked for agreement with the head noun (e.g. French mon livre vs. *lon livre). Importantly, only m- and t- are marked for person, whereas l-, i.e. the traditional 3rd person, is treated, following Benveniste (1966) and Postal (1966), as non-person or as [-1st person and -2nd person]. So, 1st/2nd accusative clitics are speci¿ed exclusively for person, whereas 3rd accusative clitics are speci¿ed as [- 1st person][-2nd person] and can show number and gender agreement. The reÀexive clitic s- , on the other hand, belongs – according to Kayne – to the same class as m- and t-, with which it “patterns strongly”. Some of the arguments are that (i) the clitic forms are morphologically parallel (e.g. m, t, s, me, te, se); (ii) there is no gender or number marking on s-; and (iii) the non-clitic forms are parallel in form (e.g. moi, toi, soi). 2.2

Uriagereka (1995)

Kayne’s (2000) view is not singular. According to Uriagereka (1995), 3rd person accusative clitics, which he labels “weak determiner clitics”, are D elements, whereas 1st/2nd person accusative clitics, “strong phrasal clitics”, are DPs (see his footnote 3 and p. 112). The two types of clitic are associated with different syntactic structures: (1)

weak

DP (double)

(2)

strong

D’ D

NP

CL

pro

DP (double)

DP=D CL

On this analysis, 3rd person accusative clitics differ from 1st/2nd person accusative clitics in several important respects: (i) the former alone are base-generated as heads, as D (see (1) above), whereas the latter are base-generated as DP (see (2)); (ii) only 3rd person accusative clitics have a Speci¿er position which hosts the double. Within the “strong” DP, the double is adjoined to the DP/D

42

Martine Coene and Larisa Avram

pronominal; (iii) only 3rd person accusative clitics take a null complement, pro; (iv) importantly, 3rd person accusative clitics are not speci¿ed for Person. This de¿ciency is the one which, in Uriagereka’s analysis, motivates their movement to a projection in the C-domain, which he calls F, a projection higher than IP, which allows “attribution of reference” (p. 93), i.e. where they are referentially indexed. The associated null complement can only be licensed if the clitic is assigned referentiality in F. But, if what drives movement to F is “referentiality”, i.e. de¿ciency with respect to Person, Uriagereka’s analysis implies that only 3rd person accusative clitics move to F for referentiality reasons. Since 1st/2nd person accusative clitics are speci¿ed for Person, the motivation of movement to F, “attribution of reference”, can no longer apply. He proposes that these clitics move as phrases via adjunction scrambling (p. 114) to F or another projection (the adjunction site being subject to cross-linguistic variation). ReÀexive clitics are assigned to a class different from the class of D-clitics. In both analyses (Kayne 2000; Uriagereka 1995), 1st/2nd person accusative clitics are argued to evince morphological and syntactic properties which distinguish them from 3rd person accusative clitics. ReÀexive clitics, at the same time, are analysed as different from 3rd person accusative clitics. In what follows, we will investigate to what extent this asymmetry is found with accusative clitics in Romanian as well. 2.3

Accusative clitics in Romanian

In Romanian, a null subject language, only 3rd person accusative clitics are identical in form to the de¿nite article. The only exception is that of the 3rd person feminine singular which is identical in form to the inde¿nite article. The data are summarized in Table 1: Table 1. Article - 3rd person accusative clitics homophony in Romanian singular masc article

inde¿nite de¿nite 3rd person Acc clitic

-l-l-

plural fem o o

masc

fem

-i-i-

-le-le-

Like their French/Italian counterparts, they are marked for gender and number. In this respect, they differ from 1st/2nd person accusative clitics, m- and t-, which pattern with their French and Italian counterparts in showing no gender

An asymmetry in the acquisition of accusative clitics in child Romanian

43

or number distinction.2 Also, only m- and t- can combine with the possessive morpheme which can be marked for agreement with the head noun. The 3rd person clitic cannot: ta ‘your’ (fem sg) vs. *la. In terms of semantic features, 3rd person accusative clitics are not restricted to either [+human] or [+animate] antecedents, unlike 1st/2nd person accusative clitics. 1st/2nd person accusative clitics are always obligatory (3); 3rd person accusative clitics may be optional (4): (3)

(4)

a.

*(M) -a ajutat (pe mine). has helped (PE me) Acc clitic 1st sg ‘(S)he has helped me.’

b.

*(Te) -a ajutat (pe tine). has helped (PE you) Acc clitic 2nd sg ‘(S)he has helped you.’ (L) -am desenat Acc clitic 3rd m sg have drawn ‘(I) have drawn a child.’

pe (un) copil. PE (a ) child

In clitic doubling constructions the “double” of 3rd person accusative clitics can be a pronoun or a DP, whereas that of 1st/2nd person accusative clitics can only be a de¿nite pronoun. Notice that in (4), where the “double” is a DP, the use of the clitic is not obligatory. In any type of clitic doubling the double of a 1st/2nd person accusative clitic is a de¿nite pronoun, which can only surface marked with the preposition pe, traditionally analysed as an accusative case marker. In this case, whether the double is overt or not is heavily constrained by information structure, being associated most probably with (contrastive) focus. However, 3rd person accusative clitics are obligatory or optional in wellde¿ned contexts. At clause level, they are obligatory in combination with a left- or right-dislocated direct object (5a-b), with direct object interrogative and relative clauses introduced by care ‘which’ (6) and in combination with a strong personal or demonstrative pronoun (7). At discourse level, they 2. We adopt the line according to which number does not interfere with person in referring to speech act participants (Harley and Ritter 2002). We does not represent a plurality of Is. In some languages, in certain contexts, the same form can be used for singular and plural reference. For French, for example, Wexler (2002) argues that the 2nd person informal and formal pronouns tu/vous are distinguished from each other by Person rather than Number.

44

Martine Coene and Larisa Avram

obligatorily copy the features of a phonetically null direct object whose antecedent has referential stability and prominence (Avram and Coene 2007) and is retrievable at the interface (8): (5)

(6)

(7)

(8)

a.

Carteai , am dat *(-oi ). book-the have givenAcc clitic 3rd f sg ‘The book, I have given away.’

b.

# carteai. Am dat *(-oi ) have givenAcc clitic 3rd f sg book-the ‘I have given the book away.’

ai ales? Pe carei li PE which Acc clitic 3rd m sg have chosen ‘Which one have you chosen?’ *(Îli ) Acc clitic 3rd m sg ‘I see him/that one.’ a. b.

văd pe eli /pe acestai . see PE him /PE that one

Ce ai făcut cu măruli ?) ‘What have you done to the apple?’ am mîncat. *(Li -) Acc clitic 3rd m sg have eaten ‘I have eaten it.’

At clause level, 3rd person accusative clitics are optional when their associate is an inde¿nite pronoun (9a), an inde¿nite DP (9b), a numeral (9c), or a proper name (9d): (9)

a.

(Ii -) am văzut pe uniii . (Acc clitic 3rd m pl) have seen Acc marker some-m pl ‘I have seen some of them.’

b.

am salutat pe (Li -) (Acc clitic 3rd m sg) have greeted Acc marker ‘I have greeted a neighbour.’

c.

am ales pe al patruleai . (Li -) (Acc clitic 3rd m sg) have chosen PE the fourth one ‘I have chosen the fourth one.’

un vecini . a neighbour

An asymmetry in the acquisition of accusative clitics in child Romanian d.

45

(Oi ) avem aici pe Ruxii . (Acc clitic 3rd fem sg) have here PE Ruxi. ‘We have Ruxi here.’

The data discussed in this section indicate that in Romanian 1st/2nd person accusative clitics do not form a natural class together with 3rd person accusative clitics. In particular, only the latter are “determiner pronouns”, homophonous with the article and have a clausal antecedent; the former are inherently marked for Person, like personal pronouns; hence, their de¿ciency can only be phonologic in nature. The observed dichotomy between obligatory and optional contexts for accusative clitics gives rise to two important observations: (i) for 3rd person accusative clitics there is a substantial number of competing contexts in which the clitic can be either obligatory or optional (though, as the data indicate, one cannot speak about true optionality); no such competing contexts are found with 1st/2nd person accusative clitics. Their use is uniformly obligatory; their “double” is always a strong personal pronoun; (ii) the use of 3rd person accusative clitics interferes with information structure packaging, since they have an antecedent with referential stability and prominence. 1st/2nd person accusative clitics are deictic, which excludes the need of any (overt or covert) clausal antecedent. This has important consequences for the properties of the structures in which they occur. According to Uriagereka (1995), 3rd person accusative clitics are base-generated as the D of a post-verbal null complement, as in (1). 1st/2nd person accusative clitics, “strong phrasal clitics”, do not take a null complement pro, they are DPs (see (2) above). We therefore assume that their feature identi¿cation/matching is not related to any null complement; it is anchored into the speech-situation, possibly via null operators in the C-domain (as proposed for 1st and 2nd person pronouns by Sigurðsson (2005) or Baker (2008), or for any other deictic elements, such as Tense). That this is indeed the case can be seen in clitic structures with an epithet. In Romanian, clitic left dislocations as well as clitic constructions with a hanging topic allow an epithet in post-verbal position: (10)

[(Pe) Ion Popescu,] eu una nu -l votez pe prostănac. [(PE) Ion Popescu] I onefem not Acc clitic 3rd m sg vote PE stupid one ‘As for Ion Popescu, I for one will not vote for this moron.’

Within a 1st/2nd person context, such epithets require some adjustment. The N within the epithet DP in (11a) must take a prepositional pronominal complement (de mine ‘of me’, (11b) and (11c)) whose pronominal features will percolate to the whole phrase:

46 (11)

Martine Coene and Larisa Avram a.

[Pe mine,] nimeni nu mă ajută *pe prost. [PE me] no one not me helps PE stupid

b.

Nu mă ajută nimeni peprostul de mine. not me helps no one PE stupid.themasc of me

c.

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

1st/2nd person accusative clitics cannot have a non-pronominal associate, not even when that is an epithet. This is due to the fact that lexical DPs cannot be 1st or 2nd person (Baker 2008). Such structures show, once again, that the relationship between the speech act participant and the ‘semi-pronominal’ epithet is anchored into the speech-situation. Notice that in (11b) and (11c) there is gender marking on the epithet. Gender marking in this case can only be the copy of the features associated with the discourse participant, since 1st person pronominals do not mark gender. With 3rd person constructions, gender marking is the copy of the phi-features of the clausal/discourse topic antecedent: (12)

a.

Pe Ion Popescu, nimeni nu îl ajută pe fraieru’ PE Ion Popescu, no one not Acc clitic 3rd m sg helps PE fool. themasc ăla. thatmasc ‘Ion Popescu, no one will help that fool.’

b.

Pe Vasilica, nimeni nu o ajută pe fraiera aia. PE Vasilica, no one not Acc clitic 3rd f sg helps PE fool.thefem thatfem ‘Vasilica, no one will help that fool.’

1st/2nd person accusative clitics are speci¿ed for Person and do not have a clausal antecedent. Their interpretation is anchored into the speech-situation. One could then assume that they are interpreted via Match with an operator in the C-domain (ȁA /ȁP )3 like any other pronoun (under the assumption that the person of a pronoun is computed in syntax under ȁ-matching, Sigurðsson 2005) or, along the same line, as in Baker (2008), where ȁA and ȁP correspond to the silent operators for speaker (Agent) and hearer (Patient) in the C domain as part of the inherent speech event (Sigurðsson 2005). There is no clitic – null complement/antecedent linking at stake since they do not have a null complement. 3. In Baker’s system S (=speaker) and A (=addressee) are null arguments generated within the CP. When there is no overriding control relationship, S will designate the person who produced the CP and A the person to whom the CP was addressed.

An asymmetry in the acquisition of accusative clitics in child Romanian

47

Following Uriagereka (1995), Avram and Coene (2007, 2009) analyse Romanian accusative clitics as base-generated as the D of a post-verbal null complement, as in (13) below: (13)

[FP (Topic) F [IP [VP DP V [DP D-pro ]]]]

The properties of the two types of accusative clitics indicate, however, that the representation in (13) can be correct only for 3rd person accusative clitics. The post-verbal complement in the case of 1st/2nd person accusative clitics is a Person-marked DP: (14)

[FP (Topic) F [IP [VP DP V [DP DP ]]]]

The D-Structure of 1st/2nd person accusative clitics then is different from the one of 3rd person accusative clitics, D(eterminer)-clitics, base-generated as heads of a null DP. We adopt the analysis in Avram and Coene (2007, 2009), restricting it to 3rd person clitics. In a nutshell, we assume that the D-clitic spells-out the phi-features of a null argument, i.e. it is a copy of the null DP in complement position, whose referential stability and topic feature it inherits. The null object has an antecedent at the left periphery of the clause (the topic, which can be overt – as in dislocation structures, or null – when the antecedent was mentioned in previous discourse).4 The identi¿cation of the features of the null complement is ensured via a chain which contains the antecedent (null or overt) in the left periphery of the clause and the null complement in post-verbal position. Identi¿cation along a chain requires feature matching. In Romanian, Agreement in InÀection is pronominal, allowing pro subjects; there will always be a potential barrier between the features of the antecedent in the left periphery and those of the null DP containing the clitic. The D-clitic in post-verbal position must match the features of its antecedent over two clusters of phi-features, whose make up may be identical with the make up of the features of the antecedent: the phi-features of AgrS {person, number} and the phi-features of the VP-internally base-generated DP subject {person, gender, number}: (15)

O vede Maria. Acc clitic 3rd f sg sees Maria ‘Maria sees her.’

4. A similar analysis, according to which clitic constructions are hidden clitic left dislocation constructions with a null topic was put forth in Baauw (2000) and in Del¿tto (2002).

48 (15’)

Martine Coene and Larisa Avram [FP (Topic 3rd f sg ) F [IP Agrs3rd sg [VP Maria3rd f sg V [DP D-pro ]]]]

The structure in (15’) 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. The intervention (-like) effects of the phi-features of the overt subject DP and those of AgrS in the functional domain disrupt the feature matching relation between the null complement and the antecedent. That is why the clitic has to move to a position higher than the intervening features, possibly Uriagereka’s FP (whose Speci¿er hosts, among other things, dislocated material, non-contrastive topics, emphasis phrases). The clitic moves for identi¿cation reasons, for “referentiality”, i.e. so that the referential index of the null DP be rescued via matching with the antecedent. For Romanian clitics, there is one more factor which may be the driving force for movement. The empirical data (presented in 2.3) show that accusative clitic constructions always involve a referentially stable antecedent, which is interpreted as a topic. Movement of the D-clitic to a higher projection is forced by identi¿cation requirements as well as by the topic feature with which it is associated. It has to move to a position higher than the intervening feature(s), where the topic feature can be checked. In Romanian, accusative clitics have a [+topic] feature which requires checking. 3rd person accusative clitics move out of the DP for referentiality reasons and also in order to check their topic feature. 1st/2nd person accusative clitics are marked for Person, they are not referentially de¿cient. However, they surface in front of the verb, on a par with 3rd person accusative clitics, there are no distribution differences: all accusative clitics will surface in front of the lexical verb in ¿nite constructions, irrespective of their person feature. The question which arises is why 1st/2nd person accusative clitics, if base-generated in post-verbal position on a par with their 3rd person counterpart, move to a position in front of the lexical verb. Notice that the intervention effects which force movement in the case of 3rd person accusative clitics do not arise, since no phi-feature matching between a null DP and an antecedent is at stake. We assume that all clitic constructions have a [+topic] feature; 1st/2nd person accusative clitics will only move to a higher projection to check their [+topic] feature. Summing up, in Romanian the syntactic derivation of 1st/2nd person accusative clitics differs from the derivation of 3rd person accusative clitics. The latter are base-generated as the D of a null complement and move to FP to check their phi-features and their topic feature. The former start as full DPs, marked for Person, in post-verbal position and move to FP to check a topic feature. The two types of non-reÀexive clitics evince different morphological, syntactic, and referential properties (see Table 2).

An asymmetry in the acquisition of accusative clitics in child Romanian

2.4

49

ReÀexive vs. non-reÀexive accusative clitics in Romanian

ReÀexive clitics behave like 1st/2nd person accusative clitics with respect to lack of gender (16) and number (17) marking, accepting both singular and plural antecedents: (16)

(17)

a.

Narcis se admiră în apa lacului. Narcis REFL admires in water.the lake.Gen ‘Narcis is admiring himself in the water of the lake.’

b.

Fata se admiră în apa lacului. girl.the REFL admires in water.the lakeGen ‘The girl is admiring herself in the water of the lake.’

a.

Copilul se admiră în apa lacului. child.the REFL admires in water.the lake.theGen ‘The child is admiring himself/herself in the water of the lake.’

b.

Copiii se admiră în apa lacului. children.the REFL admire in water.the lake.theGen ‘The children are admiring themselves in the water of the lake.’

The clitic and the non-clitic forms of reÀexives and those of 1st/2nd person accusative clitics are morphologically parallel, as are their counterparts in French and Italian. Unlike 3rd person accusative clitics, reÀexive clitics are never optional. In their case, intervention effects do not arise since they require identity of phi-features with the subject DP. Moreover, Romance reÀexive clitics have been analysed as markers of reÀexivity (Dobrovie-Sorin 1998) and as such base-generated in pre-verbal position from where they can feature match with the subject DP. In this case, there are no intervention effects of potential identical feature clusters. 2.5

Predictions for acquisition

The brief analysis of Romanian clitics reveals that 1st/2nd person accusative clitics do not form a class with 3rd person accusative clitics; they seem to pattern rather with reÀexive clitics, on a par with their French and Italian counterparts. The data are summarized in Table 2 below:

50

Martine Coene and Larisa Avram

Table 2. Main properties of accusative clitics properties determiner-like/pronominal homophonous with the article marked for gender marked for number marked for person can combine with a possessive morpheme status (head/maximal projection) take a null complement movement to check referentiality movement to check a topic feature feature intervention effects optional (in well-de¿ned contexts)

1st/2nd person 3rd person nonnon-reÀexives reÀexives pronominal determiner-like í + í + í + + í + í DP í í + í í

D + + + + +

reÀexives pronominal í í í + + DP í í + í í

In terms of acquisition, the data indicate that we have every reason to assume that there might be an asymmetry between 1st/2nd person accusative clitics and 3rd person accusative clitics. As summarized in Table 2 above, they differ in terms of morphological complexity, syntactic status, referentiality, optionality and feature intervention effects. In particular, 3rd person accusative clitics are morphologically more complex than the other two types of clitics and their con¿guration involves feature intervention effects. The intervention (-like) effects of the phi-features of the subject DP and those of AgrS in the functional domain make the feature matching relation between the null complement and the antecedent computationally complex. 3rd person accusative clitics involve a higher computational load since feature matching with their antecedent has to be accomplished across two sets of identical features.5 During the early stages, children’s computational capacity is limited. We therefore predict a developmental delay of 3rd person accusative clitics due to morphological complexity and (feature) intervention effects. A higher computational load induced by intervention effects has also been argued to be the main reason of the delay in the acquisition of other structures, such as (some) direct object relative clauses (Friedmann, Belletti, and Rizzi 2009; Adani et al. 2009) or wh-questions (Guasti, Arosio, and Branchini 2008). 5. Zesiger et al. (2010) put forth an analysis similar to the one in the present paper. According to them, accusative clitics involve a crossing chain, along which the subject features can have intervention effects. We thank a reviewer for pointing this out to us.

An asymmetry in the acquisition of accusative clitics in child Romanian

51

However, one should point out that some 3rd person accusative clitics might be less computationally costly. Compare (15) above to (18): (18)

Le vede Maria. Acc clitic 3rd f pl sees Maria ‘Maria sees them.’

(18’)

[FP (Topic 3rd f pl ) F [IP Agrs3rd sg [VP Maria3rd f sg V [DP D-pro ]]]]

In (18) the intervening feature make up of the antecedent and the one of the intervening elements is the same, i.e. number, gender, person, but the value of one of the features is different (plural number for the antecedent and singular number for AgrS in InÀection and for the subject DP). The same can be seen in (19) below, where the value of more than one feature is different and where the feature make up is also different (only the make up of the antecedent has gender): (19)

(Pe Maria) noi o (pe Maria) we clitic 3rd f sg ‘Maria, we see her.’

vedem. see 1st pl

(19’)

[FP (Topic 3rd fem sg ) F [IP Agrs1st pl [VP we1st pl V [DP D-pro ]]]]

The intervention effects in con¿gurations like those in (15), (18) and (19) involve different degrees of computational complexity. We therefore expect those which involve full identity of feature value to be more problematic for children. For example, one would expect a higher number of non-target-like structures in contexts like the one illustrated in (15), where the phi-features of the subject have the same make up and the same values as the phi-features of the antecedent (3rd person feminine singular) and can, therefore, act as a (strong) intervener in the feature matching process. Such errors are also expected to last longer. Identity of feature make up with partial or no identity of feature value should be less problematic or not problematic at all.6

6. This difference in computational load may be more obvious for comprehension; however, our present study does not address the comprehension of accusative clitics.

52

Martine Coene and Larisa Avram

In the case of reÀexive clitics, there are no feature intervention effects; in their case there is always total identity of features between the clitic, the subject and the phi-features on AgrS.7 The predictions for acquisition are straightforward: (i) we expect to ¿nd an asymmetry between the acquisition of 1st/2nd person vs. 3rd person accusative clitics; (ii) since some feature intervention effects may be stronger/weaker than other, we expect this difference to be reÀected in the acquisition of clitic structures; (iii) we expect to ¿nd developmental differences between reÀexive clitics and 3rd person accusative clitics. The empirical data also revealed an asymmetry with respect to optionality: 1st/2nd person accusative clitics and reÀexive clitics are always obligatory, whereas 3rd person accusative clitics may be optional in some well-de¿ned contexts. This asymmetry can have consequences for the acquisition process. It is plausible to assume that the obligatoriness of 1st/2nd person accusative clitics and of reÀexive clitics could favour early acquisition.

3

The developmental pattern of accusative clitics in Romanian

3.1

Subjects and data

In this study, the predictions advanced in Section 2 are veri¿ed against longitudinal data coming from two corpora of monolingual child Romanian. For the present analysis, we analysed 32 transcripts of monthly 60 minute recordings of spontaneous speech between a child and a caregiver. Both child speech and child-directed speech have been transcribed in CHAT format (MacWhinney 2000). The overall number of ¿les examined for the present analysis are given in Table 3: Table 3. The data Child B. A.

Age 1;09 í 2;11 1;09 í 3;00

MLU 1.091 – 2.790 1.514 – 3.174

Nr of ¿les 16 16

7. Crysman and Müller (2000) also adopt an analysis of object clitics according to which only non-reÀexives license and identify a pro object in syntax, whereas reÀexives are created via argument absorption, a pre-syntactic process; hence, reÀexives should not interact with computational complexity.

An asymmetry in the acquisition of accusative clitics in child Romanian

53

For the B. corpus, an additional analysis was performed on the child-directed speech (i.e. the mother or father speaking with the child) covering 10,706 utterances. For coding and counting, following the method used in Avram and Coene (2007), a detailed examination of each ¿le was conducted in order to identify all the obligatory clitic contexts. The omissions which involved a de¿nite pronoun antecedent (clitics included) were the only ones counted as deviant. All the other situations (proper names included) were evaluated as adult-like. A small number of null objects in transitive environments not rescued by a clitic were considered target-like because they can be found in adult productions as well. Imitations, poetry or song fragments, as well as repetitions did not enter the analysis. Omission rates as well as rates of clitics used were calculated against the number of identi¿ed obligatory clitic contexts. Errors were calculated against the total number of clitics used. Importantly, reÀexive clitics were counted separately. 3.2

Results

The data show that 3rd person accusative clitics emerge several months before 1st/2nd person accusative clitics and reÀexive clitics.8 In the A. corpus, 3rd person accusative clitics (the feminine clitic o) are attested as early as the ¿rst recording session, at 1;09 (MLU 1.392). The ¿rst 1st/2nd person accusative clitics and reÀexive clitics are attested at 2;04 (MLU 2.136). In the B. corpus, the ¿rst 3rd person accusative clitics (the feminine clitic o) is attested at 1;10 (MLU 1.091) in post-verbal position. The ¿rst reÀexive clitic is attested at 1;11 (MLU 1.406) but no other reÀexive is attested until 2;01. One should, however, notice that all the instances of reÀexive clitics in the examined ¿les until 2;01 (4 tokens) 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 the ¿rst “genuine” reÀexives actually emerge at 2;02. 1st/2nd person accusative clitics also emerge at 1;11, but they are ¿rst attested in the formulaic te rog (‘please’-Acc clitic 2nd sg) and no other 1st/2nd person accusative clitic is attested until 2;01 (MLU 1.734). There is almost no omission of 1st/2nd person accusative clitics or reÀexive clitics after their emergence in both corpora (apart from an incidental increase in the A. corpus at 2;05). 3rd person accusative clitics, on the other hand, in spite 8. The data from child Romanian differ in this respect from what has been reported for child French on the basis of the Ivar corpus (Crysman and Müller 2000), where reÀexives are attested earlier than all non-reÀexive accusative clitics.

54

Martine Coene and Larisa Avram

of very early emergence, continue to be omitted (even though at a low rate) at a time when 1st/2nd person accusative clitics as well as reÀexive clitics are used adult-like. The omission of reÀexives decreases to 0% at an early stage: at 2;03 (MLU 1.821) in the B. corpus and at 2;08 (MLU 3.099) in the A. corpus. In both corpora, the decrease in the omission of reÀexives coincides with the decrease in the omission of 1st/2nd person accusative clitics. The statistical analysis (non-parametric paired-sample Friedman test) of the data shows that the observed differences in omission rate are highly signi¿cant in the B. corpus (p = .001). For the A. corpus, the data are less robust (the number of 1st/2nd person accusative clitic contexts is very low in some transcripts) and did not reach signi¿cance. Figure 1 below presents the omission rates of 1st/2nd person vs. 3rd person accusative clitics in the B. corpus and Figure 2 in the A. corpus:

Figure 1. Omission of accusative clitics – B. corpus

Studies of accusative clitics in longitudinal data for French also indicate a slight advantage for 1st/2nd person accusative clitics during the early stages. In the Augustin corpus (Hamann 2002, Table 7 p. 35), the ¿rst attested object clitic is a 3rd person accusative clitic (at 2;2.13), but in between 2;04.01 – 2;06.16, only reÀexive clitics and 1st person accusative clitics are attested (see also Rasetti 2003, Table 23). In the Marie corpus, after a ¿le where 1st, 2nd and 3rd person accusative clitics are attested (at 1;8.26), one notices a time span

An asymmetry in the acquisition of accusative clitics in child Romanian

55

Figure 2. Omission of accusative clitics – A. corpus

(1;09.03 – 1;11.5) when 1st/2nd person accusative clitics outnumber 3rd person accusative clitics (Rasetti 2003, Table 23). Dominguez (2003) examines the emergence of clitics in child Spanish on the basis of longitudinal data: the Maria corpus (age 1;07 – 3;11). The results indicate that the forms me, te (used both as reÀexives and as non-reÀexives) and se are acquired at approximately the same time and are used at a similar rate, i.e. in early Spanish as well the reÀexive se patterns with 1st/2nd person accusative clitics. The results for reÀexives in child Romanian are far from being singular. Several previous studies reported lower omission rates for reÀexive clitics than for accusative clitics in French (Jakubowicz 1989; Jakubowicz et al. 1996; Hamann, Rizzi, and Frauenfelder 1996; Crysman and Müller 2000; Rasetti 2003). Similar ¿ndings are reported in Zesiger et al. (2010) on the basis of experimental data; they show that reÀexive se patterns with subject clitics, not with accusative clitics. A qualitative analysis of 3rd person accusative clitics in our corpus reveals some agreement errors (illustrated in (20)): (20)

a.

Unde sînt piticii ca să *le pun aicea? where are dwarfs.them pl that Acc clitic 3rd f pl put here ‘Where are the dwarfs so that I can put them here?’ [B. 2;8]

56

Martine Coene and Larisa Avram b.

Adult: ce’ ai făcut cu ligheanu(l) ăla? what have2nd sg done with bowl.themasc sg thatmasc sg ‘What have you done to that bowl?’ Child: a spart -o. has broken clitic Acc 3rd fem sg ‘(S/he) has broken it.’

[A 1;9]

Both omission and agreement errors with 3rd person accusative clitics in child Romanian have also been reported in studies dealing with experimental data (age 2;5 – 4;0) (Avram 2001). The longitudinal corpus also contains rare 1st person accusative clitics used instead of 3rd person accusative clitics in clauses with a 1st person subject. Such errors are attested only in the B. corpus, during a very short period of time, immediately after the emergence of 1st/2nd person accusative clitics: (21)

Adult: Ce-ai făcut cu cartea? ‘What have you done with the book?’ Child: Mam pus la Kiki. Acc clitic 1st sg have1st sg put at Kiki ‘I have put myself to Kiki.’ Adult: Ce-ai făcut? ‘What did you do?’ Child: MAcc clitic 1st sg

am pus. have1st sg put

[B. 2;2]

The results from child Romanian indicate that there is a difference between the developmental pattern of 1st/2nd person accusative clitics and 3rd person accusative clitics, as predicted. The data also reveal that the developmental pattern of 1st/2nd person accusative clitics is similar to that of reÀexives, also in accordance with the predictions which we started from. 3.3

An effect of the input?

As discussed in Section 2, the use of 3rd person accusative clitics is subject to optionality in a signi¿cant number of contexts. This raises the question of whether the observed difference between the early target-like use of 1st/2nd person clitics compared to the extended omission of 3rd person accusative clitics in child language may not be a reÀex of the use of these clitics in the input. In order to answer this question, we examined all parental child-directed speech in

An asymmetry in the acquisition of accusative clitics in child Romanian

57

the B. ¿les used in the present analysis, with a view to identifying (i) whether the observed omission of 3rd accusative clitics in child speech may result from a difference in frequency of 1st/2nd person accusative clitics and 3rd person accusative clitics in the input, and (ii) whether the use of accusative clitics in child-directed speech shows any longitudinal effects, i.e. if there is an increase in the use of 3rd person accusative clitics over time that compares to the increase found in child speech. The parental data indeed show signi¿cant differences with respect to the production rate of 1st/2nd vs. 3rd person accusative clitics (Friedman paired samples, p < .001). However, in contrast to child speech, parental speech seems to favour 3rd person over 1st and 2nd person clitics. Post-hoc Wilcoxon paired-sample tests reveal a signi¿cantly higher number of 3rd accusative clitics (1st vs 3rd: p = .003, and 2nd vs. 3rd: p = .003). There are no longitudinal effects with respect to the production rate of different type of accusative clitics in child-directed speech. In addition, one should also mention that no case or agreement errors were found on any type of accusative clitic.

Figure 3. The use of accusative object clitics: child-directed speech (B. corpus)

58

Martine Coene and Larisa Avram

4

A possible account

4.1

Previous accounts and the 1st/2nd vs. 3rd person asymmetry

4.1.1 Morphological complexity The analysis of 3rd person accusative clitics indicates that they are morphologically more complex (they are marked for number and gender). This may suggest that the reason for which one ¿nds an asymmetry in acquisition could also be rooted in the different morphological properties of the various types of accusative clitics. Actually, Dominguez (2003) offers a morphological account for the Spanish data. She associates the observed difference in the acquisition of accusative clitics to morphological complexity: 3rd person accusative clitics are morphologically more complex than 1st/2nd person accusative clitics or some reÀexives; in particular, only 3rd person accusative clitics have number and gender features. Jakubowicz et al. (1998) also account for the higher frequency of se in terms of morphological complexity. At ¿rst sight, the morphological complexity account seems to be supported by the Romanian data as well. As pointed out in Section 3.2, children omit 3rd person accusative clitics (which are morphologically more complex) at a higher frequency and for a longer period of time. However, though morphological complexity may play a part, there are several reasons to believe that it cannot be the prime determinant of the developmental asymmetry under discussion. Several acquisition studies argue that similarity/difference in morphological complexity will not necessarily result in similar/different developmental patterns. Jakubowicz et al. (1996), for example, show that in both child German and child French one ¿nds clear pronominal object/ pronominal subject asymmetries, in spite of the fact that French uniformly uses clitics and German weak pronouns. Also, if one adopts an analysis according to which 3rd person accusative clitics are morphologically identical to articles, i.e. the two D elements evince the same degree of morphological complexity, one would expect their acquisition pattern to be similar. However, data coming from various languages, Romanian included, indicate that accusative clitics emerge later than de¿nite articles and children (in both monolingual and bilingual settings, both TD and SLI children) continue to omit clitics at a stage when they no longer drop articles (French: Jakubowicz et al. 1998; Hamann 2002; Greek: Marinis 2005; Italian: Bottari et al. 1993/1994, Romanian: Avram and Coene 2004). Whereas there might be some similarity in the acquisition pattern of their phi-features, articles and clitics follow different paths. Such data indicate that the acquisition of clitics

An asymmetry in the acquisition of accusative clitics in child Romanian

59

cannot be driven by the morphological properties of the respective pronominals alone. Though morphological complexity may indeed play a role, we do not think we can conclude that it provides an explanation for the asymmetry under discussion. 4.1.2 Syntactic accounts One possible syntactic account (see, among many others, Tsakali and Wexler 2003) of early clitic omission is rooted in Wexler’s (1998) Unique Checking Constraint (UCC), according to which the D-feature of a moving DP can check only against one functional category during an early stage in acquisition. For accusative clitics, Sportiche’s (1996) non-movement analysis is adopted. The clitic is base-generated in a CliticP higher than TenseP, while its associate is base-generated in direct object position, in the VP domain. In the derivation, the associate (which can be either phonetically null – a pro, or a full lexical DP) raises to SpecCliticP, where it enters a feature-checking relation with the clitic. The Unique Checking Constraint account predicts early clitic omission in French-like languages, which have past participle agreement, and where the moving associate has to check its D-feature twice: in AgroP and in CliticP. At a stage when the Unique Checking Constraint constrains the early grammar, children will either omit the CliticP or the AgroP. On the other hand, one expects no clitic omission in Spanish-like languages, which lack past participle agreement, i.e. where the associate will check its D-feature only in SpecCliticP. But, as we have already seen, the data indicate an asymmetry between 1st/2nd vs. 3rd person accusative clitics in both French (which has past participle agreement) and Romanian and Spanish (which lack past participle agreement). The Unique Checking Constraint account does not seem to be able to explain why the omission rate for 1st/2nd person accusative clitics is lower than for 3rd person accusative clitics in the early stages in both French and Romanian, or why reÀexive clitics pattern rather with the former.9 The missing CP-account (see, among many others, Müller, Crysmann, and Kaiser 1996, Crysmann and Müller 2000) relates the early omission of clitics to the optionality of the CP domain during the early stages (in accordance with Rizzi’s 1993/1994 truncation account). Clitics and weak pronouns, analysed as occupying a position in the C-domain, are omitted at a stage when the C-domain is optionally projected. There is indeed a relationship between the emergence of the C-layer of the clause and the use of early clitics. For Romanian, the acti9. As one reviewer points out, it is not clear either whether Agro and AgrPartP are one and the same projection.

60

Martine Coene and Larisa Avram

vation of the C-system seems to be a prerequisite for the emergence of object clitics (Avram and Coene 2006). But Romanian clitics continue to be omitted after the CP becomes active. Both longitudinal and experimental data show that Romanian children still omit clitics at a stage when they already produce interrogative and relative clauses. In our longitudinal data, by age 2;2 complementizers, displaced wh-phrases and relative clauses are all attested. But accusative clitics continue to be omitted. Experimental data provide similar results. In a direct object relative clause elicitation task, 5;06 – 7;00 year old subjects still omitted accusative clitics at a low rate of 3.5% (Sevcenco, Stoicescu, and Avram 2009). Even though the optionality of the CP could explain early clitic omission, it nevertheless fails to explain why 1st/2nd person accusative clitics are not omitted at a stage when their 3rd person counterparts continue to be occasionally dropped. 4.1.3 Accessibility accounts Some recent acquisition studies show that children tend to omit arguments when their referents are maximally clear from the context (discourse or situational) (Allen 2006 in Tedeschi 2007), or highly accessible. Clitic pronouns have an antecedent which is prominent in discourse, i.e. it is highly accessible. On such an analysis, if they are omitted, the missing information can be provided by discourse. But 1st and 2nd person arguments are highly accessible by de¿nition. Speaker and addressee are higher than non-participant (3rd person) on a cognitive accessibility scale (Sierwierska 2004: 46). In spite of that, the acquisition data indicate a higher percentage of omissions with 3rd person accusative clitics. Another argument against the accessibility account comes, as one reviewer suggests, from the subject/object clitics asymmetry: subject clitics, which are highly accessible, are not dropped at a time when object clitics are still omitted (see, for example, Zesiger et al. 2010). Informativeness can certainly play a role in our understanding of early clitics and of clitics in general, but we believe that it cannot account, all by itself, for the developmental asymmetry under discussion. 4.1.4 A pragmatic account Schaeffer (2000) argues that children omit object clitics because they lack the concept of shared knowledge: they make no difference between discourserelated and non-discourse related referentiality. Thus, children assume that the referent is always part of the shared knowledge of speaker and hearer and they choose the easy way out, marking referentiality through a non-linguistic

An asymmetry in the acquisition of accusative clitics in child Romanian

61

mechanism. This results in object clitic omission. While Schaeffer’s analysis can account for the early omission of 3rd person accusative clitics, it cannot straightforwardly explain why 1st/2nd person accusative clitics behave differently or why they seem to pattern with reÀexives. 4.2

Our account

We believe that the developmental asymmetry under discussion is rooted in the different properties of the three types of clitics. We propose that the asymmetry between 1st/2nd person accusative clitics and 3rd person accusative clitics can be best explained in terms of computational load, in particular in terms of feature intervention effects. We therefore propose that the developmental delay of 3rd person accusative clitics is due to (feature) intervention effects. The main difference between the three classes of accusative clitics discussed in this paper is related to intervention effects, which might increase the computational load. Since children’s early syntactic computational capacity is limited, we suggest that the developmental delay of 3rd person accusative clitics when compared to 1st/2nd person accusative clitics and reÀexive clitics boils down to intervention effects which arise only with 3rd person accusative clitics. This can explain why omissions are much more frequent and last longer with the latter. The agreement errors, however, cannot be straightforwardly accounted for as an effect of intervention effects due to two factors: (i) the obvious impossibility of manipulating gender/number in longitudinal data, so that one creates situations in which there is total/partial/no identity of feature value between the antecedent and the null DP in post-verbal position and (ii) the relatively small number of clitics. In Section 2 we also predicted that partial/no identity of feature value should be either less problematic than total identity (i.e. smaller number of omissions, smaller number of errors, problematic only during earlier stages, when the child’s computational capacity is even more reduced) or not problematic at all. But one would expect a higher number of omissions in contexts where the phi-features of the subject have the same values as the phi-features of the antecedent, i.e. where they can act as a (strong) intervener in the feature matching process: (22)

că doare burta because hurts tummy.the f sg ‘Because Bianca has a tummy ache.’

pe Bianca. pe Bianca f sg

[B. 2;2]

Unfortunately, such a prediction is also dif¿cult (if not impossible) to test fully against early longitudinal data for at least two reasons: the low number

62

Martine Coene and Larisa Avram

of attested accusative clitics and the signi¿cant number of 3rd person singular verbal forms used – in the early stages – in contexts where other inÀected forms are targeted. But a qualitative analysis of the longitudinal data, corroborated by experimental results reveals that identity of feature make up is problematic only very early and over a very short period of time. The weaker intervention effects of partial/no identity of features are reÀected in the extremely small number of 1st person accusative clitics used instead of 3rd accusative clitics in clauses with a 1st person subject. Data from experimental studies provide some interesting evidence in favour of our prediction, i.e. that when the phi-features of the DP subject are different from those of the antecedent, a higher number of clitics should be attested and a smaller number of errors. Tedeschi (2006, 2007) shows that Italian children (age 2;06 – 6;05) 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 and the 2nd person tu ‘you’ instead of 3rd person in subject position. The results show that there is almost no clitic omission for 3 – 5 year old French children in this task, while in a task using a 3rd person subject children omit the clitic 30 – 50%. In 1st/2nd person accusative clitic con¿gurations, we saw that no intervention effects arise. 1st/2nd person accusative clitics establish a relationship with a speech situation participant, not with a clausal or discourse antecedent. We also suggested that an alternative analysis of these clitics might base-generate them in FP, i.e. on this analysis there would be no movement, which suggests an even lower computational load. Whichever analysis one adopts for these clitics (movement vs. non-movement), no intervention effects arise and the computational load is less heavy. ReÀexive clitics require identity of phi-features with the subject. They are base-generated in pre-verbal position from where they can feature match with the subject DP. In this case, there are no intervention effects of potential identical feature clusters.

5

Conclusions

In the theoretical literature, a distinction has always been made between 1st and 2nd person pronouns, on the one hand, and 3rd person pronouns, on the other (Uriagereka 1995; Kayne 2000). 1st/2nd person accusative clitics belong to a class which excludes 3rd person accusative clitics but which includes reÀexive

An asymmetry in the acquisition of accusative clitics in child Romanian

63

clitics. Accusative clitics in Romanian support this distinction. We investigated longitudinal data of child Romanian with a view to testing whether this asymmetry is reÀected in a difference between the developmental path of 1st/2nd person accusative clitics and that of 3rd person accusative clitics, on the one hand, and whether the developmental path of 1st/2nd person accusative clitics is similar to that of reÀexive clitics, on the other hand. The data revealed a difference between 1st/2nd person accusative clitics and 3rd person accusative clitics: the former emerge slightly later but are practically target-like from the very beginning. 3rd person accusative clitics, in spite of very early emergence, continue to be omitted at a stage when 1st/2nd person accusative clitics are used in an adult-like manner. The developmental path of reÀexive clitics is similar to that of 1st/2nd person accusative clitics: they emerge at approximately the same time and are used target-like immediately after emergence. The starting point of our account was Uriagereka’s (1995) distinction between “strong” and “weak” clitics, which we translated into a person distinction: 1st/2nd vs. 3rd person accusative clitics. Only the latter are D elements which take a null complement, and have to move to a left periphery F projection to check their referentiality (because they lack a Person feature). Since they are base-generated in complement position as the D of a null direct object, we argued that the feature matching relation between this null DP and its antecedent “crosses” over two potential interveners, the phi-features of AgrS and those of the subject DP. This increases the computational complexity of those con¿gurations where there is identity between the phi-features of the antecedent and those of the DP subject. On the other hand, 1st/2nd person accusative clitics are DPs, i.e. they behave like pronouns. They are speci¿ed for Person and are interpreted via Match with a silent operator in the C-domain, like any other pronoun. They are not subject to intervention effects. Following Dobrovie-Sorin (1998), we analysed reÀexive clitics as base-generated in preverbal position, from where they can feature match with the subject DP. No intervention effects arise in their case either. We proposed that the developmental delay of 3rd person accusative clitics is due to feature intervention effects which plausibly cause greater computational dif¿culty.

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

64

Martine Coene and Larisa Avram

References Adani, Flavia, Heather van der Lely, Matteo Forgiarini, and Maria Teresa Guasti 2009 Grammatical feature dissimilarities make relative clauses easier: a comprehension study with Italian children. Ms. University of Milan. Avram, Larisa 2001 Remarks on the optional clitic stage. Bucharest Working Papers in Linguistics 3(1): 17 – 28. Avram, Larisa, and Martine Coene 2004 What early clitics can tell us about early subjects. In Proceedings of Gala 2003, Jacqueline van Kampen, and Sergio Baauw (eds.), 93 – 102. Utrecht: UiL LOT. 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). Avram, Larisa, and Martine Coene 2009 Null objects and accusative clitics in Romanian. Bucharest Working Papers in Linguistics XI (1): 233 – 254. Baauw, Sergio 2000 Grammatical Features and the Acquisition of Reference. A Comparative Study of Dutch and Spanish. Doctoral dissertation, Utrecht University. Baker, Mark 2008 The Syntax of Agreement and Concord. Cambridge: Cambridge University Press. Benveniste, Eugène 1966 Problèmes de linguistique générale. Paris: Gallimard. Bottari, Piero, Paola Cipriani, and Anna-Maria Chilosi 1993/94 Protosyntactic devices in the acquisition of Italian free morphology. Language Acquisition 3: 327 – 369. 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. Del¿tto, Denis 2002 On the semantics of pronominal clitics and some of its consequences. Catalan Journal of Linguistics 1: 29 – 57.

An asymmetry in the acquisition of accusative clitics in child Romanian

65

Dobrovie-Sorin, Carmen 1998 Impersonal se constructions in Romance and the passivization of unergatives. Linguistic Inquiry 29: 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, Silvina Montrul and Francisco Ordóñez (eds.), 212 – 228. Somerville, USA: Cascadilla Press. Friedmann, Naama, Adriana Belletti, and Luigi Rizzi 2009 Relativized relatives: Types of intervention in the acquisition of A-bar dependencies. Lingua 119: 67 – 88. Guasti, Maria-Teresa, Fabrizio Arosio, 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. Jakubowicz, Celia 1989 L’acquisition des anaphors et des pronoms lexicaux en français. In Grammaire générative et syntaxe compare, Jacqueline Guéron and JeanYves Pollock (eds.), 229 – 252. Paris: Editions du Centre National de la Recherche Scienti¿que. Jakubowicz, Celia, Natascha Müller, Beate Riemer, Ok-Kyung Kang, and Catherine Rigaut 1996 On the acquisition of the pronominal system in French and German. In Proceedings of the 20th Boston University Conference of Language Development Vol 1, Andy Stringfellow, Dalia Cahana-Amitay, Elizabeth Hughes and Andrea Zukowski (eds.), 331 – 342. Somerville, MA: Cascadilla Press. Jakubowicz, Celia, Lea Nash, Catherine Rigaut, and Christophe-Loic Gerard 1998 Determiners and clitic pronouns in French-speaking children with SLI. Language Acquisition 7: 113 – 160. Kayne, Richard 2000 Person morphemes and reÀexives in Italian, French, and related languages. In Parameters and Universals, Richard Kayne (ed.), 131 – 162. New York: Oxford University Press.

66

Martine Coene and Larisa Avram

MacWhinney, Brian 2000 The CHILDES Project: Tools for Analyzing Talk. Mahwah, NJ: Lawrence Erlbaum Associates. Marinis, Theodoros 2005 Subject-object asymmetry in the acquisition of the de¿nite article in Modern Greek. In Advances in Greek Generative Syntax, Arhonto Terzi and Melita Stavrou (eds.), 153 – 178. Amsterdam: John Benjamins. Müller, Natascha, Berthold Crysmann, and Georg Kaiser 1996 Interactions between the acquisition of French object drop and the development of the C-system. Language Acquisition 5: 35 – 63. 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. Postal, Paul 1966 On so-called pronouns in English. In Report of the 17th Annual Roundtable Meeting on Linguistics and Language Studies, Francis P. Dineen (ed.), 177 – 206. Washington, D.C.: Georgetown University Press. 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. Rizzi, Luigi 1990 Relativized Minimality. Cambridge, Mass.: MIT Press. Rizzi, Luigi 1993/1994 Some notes on linguistic theory and language development. Language Acquisition 3: 371 – 394. Schaeffer, Jeannette 2000 The Acquisition of Direct Object Scrambling and Clitic Placement. Amsterdam: John Benjamins. Sevcenco, Ana, Ioana Stoicescu, and Larisa Avram 2009 On the subject/object relative clause asymmetry in child Romanian. Poster presented at RASCAL, February 2009, Groningen. 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. Sportiche, Dominique 1996 Clitic constructions. In Phrase Structure and the Lexicon, Johan Rooryck and Laurie Zaring. (eds.), 213 – 276. Dordrecht: Kluwer. Tedeschi, Roberta 2006 The acquisition of object clitics: Data from an elicited production task. Annali Online di Ferrera – Lettera, Vol. 2: 31 – 42.

An asymmetry in the acquisition of accusative clitics in child Romanian

67

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). Tsakali, Vina, and Kenneth Wexler 2004 Why children omit clitics in some languages but not in others: New evidence from Greek. In Proceedings of Gala 2003, Vol. II, Jacqueline van Kampen, and Sergio Baauw (eds.), 493 – 504. Utrecht: LOT Uriagereka, Juan 1995 The syntax of clitic placement in Western Romance. Linguistic Inquiry 26(1): 79 – 123. 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. Zesiger, Pascal, Laurence Zesiger, Marina Arabatzi, Lara Baranzini, Stéphany CronelOhayon, Julie Franxk, 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: 571 – 603.

Comprehension and imitated production of personal pronouns across languages Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

1

Introduction

Across languages, little is known to what types of cues pre-school children refer when selecting and resolving pronouns across sentences and how they proceed to target use of these anaphoric means. Most studies on children’s use of pronouns investigate the knowledge of English-learning children. However, taking language-speci¿c factors seriously one has to be very careful in generalizing ¿ndings on English across languages. The English personal pronouns he and she, for instance, refer to human beings only. The feature [+human] is part of their lexical meaning. In the same way it is speci¿ed for [-human]. These speci¿cations are not given with (personal) pronouns in many other languages. The Subject Preference Hypothesis (Frederiksen 1981) claims that the preferred antecedent of an ambiguous pronoun is the grammatical subject of the preceding clause. Given that the subject of a transitive sentence prototypically presents the agent of the clause, correlation with the subject is likely to be much closer in pronouns restricted to human or even animate referents. Pronouns not restricted to animate referents can easily be shown to refer to non-subjects, cf. The forwardi is missing the ballj . Itj is slipping behind the line. There are several other aspects of language-speci¿c features which very likely affect the capacity of anaphora. One of them, for instance, is the number of referential expressions competing or overlapping in their anaphoric capacity (cf. Gundel, Hedberg, and Zacharski 1993). The correlation of (personal) pronouns with subject/ topic reference is likely to be much closer in languages not exhibiting a null pronoun. Nevertheless, the personal pronoun (PERS) is a formally less complex anaphor and is used for topic continuation in many languages. Except for null and clitic pronouns, all other pronoun types are more complex and are not as closely connected to topic continuation (Givón 1983, 1995). Therefore, salience-oriented models classify PERS as related to referents at the current centre of attention (Gundel, Hedberg, and Zacharski 1993). In linguistic

70

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

feature-oriented models, PERS is de¿ned as underspeci¿ed language unit. The referent has to be inferred by integrating information from the linguistic and non-linguistic context. Models in the framework of Centering Theory (e.g. Gordon, Grosz, and Gilliom 1993; Beaver 2004) argue for the Subject Preference Hypothesis. To some extend, this hypothesis is in line with the assumption of parallelism in antecedent-anaphor relations (e.g. Stevenson, Crawley, and Kleinmann 1994; Chambers and Smyth 1998). It meets also with the First Mention Hypothesis (Gernsbacher, Hargreaves, and Beeman 1989) which proposes that the antecedent of PERS is the ¿rst nominal phrase of the previous clause. Studies that have separated subject position from ¿rst-mentioned position present mixed results. Cowles, Walenski, and Kluender (2007), Kaiser and Trueswell (2003) found a preference for subject antecedents in English and Finish respectively, while, also for Finish, Järvikivi et al. (2005) found that both factors affect the reference relation of PERS independently. Alternatively, e.g. Givón (1983, 1995) and Ariel (1990, 2004) relate the referential function of nominal phrases to discourse structure. In this view, PERS is a prototypical means of topic continuation. Cowles (2007) presents evidence that topic status serves to enhance cognitive prominence and pronominal reference even if these factors are not converged with ¿rst position or subject role. A range of other factors as, for instance, recency of mention, grammatical aspect, type of predicates and clauses have been discussed as well. However, while the correlation of referent animacy with grammatical rules and types of referential expressions is a well-known fact (e.g. Comrie 1989; Branigan, Pickering, and Tanaka 2008) only a few studies systematically investigated the impact of referent animacy on pronoun use. Dahl and Fraurud (1996), Dahl (2000, 2008) and Juvonen (2000) observed a preference for maintaining animate subject antecedents in Swedish and Finish respectively. This was also found by Yamamoto (1999) for English and Japanese. Considering production, Fukumura and van Gompel (2010) found higher rates of pronoun use when speakers maintain an animate antecedent occurring in the presence of an inanimate antecedent in English. In sum, there is evidence that (i) referent animacy affects the resolution of PERS and (ii) PERS is used to co-refer with antecedents in structurally prominent position indicating a salient status of the respective referents in the mental representation of a situation. However, no dominant correlation with speci¿c syntactic or pragmatic feature(s) has been identi¿ed so far (cf. Kehler et al. 2007). Considering pronoun use in child language, it is worth noting that already pre-linguistic children show some sensitivity to what is in the focus of attention of their communication partners (Gergely, Bekkering, and Király 2002; Woodward 2003). Tomasello (1999, 2009) argues that the social-cognitive

Personal pronouns across languages

71

skills enabling the child to include knowledge on joint attention in interaction planning emerge until around nine months of age. Around their ¿rst birthday, infants can understand goals and intentions of others and have developed some understanding of shared knowledge (Carpenter 2009). Analyses of spontaneous speech found that children as young as 2.0 – 2.5 use less complex noun phrases when referring to objects which are in the shared focus of attention (Tomasello 2003) while using more complex ones when referring to objects newly taken into focus (Matthews et al. 2006; Bittner 2007; Rozendaal 2008). The same opposition has been found in an experimental study with German-learning 2;5-year-old children, but not with 2;0 year-olds (Wittek and Tomasello 2005). Clancy (1997) and Allen (2000) found that 2-year-olds learning Korean vs. Inuktitut omit the noun phrase more often for referents recently mentioned than for new referents. Likewise, Gomme and Johnson (1997) reported more frequent use of pronouns and ellipses for the central character in narratives of 3-year-olds. Hickmann and Hendriks (1999) provide cross-linguistic evidence for the prominence of subject referents in narratives of 4-year-olds. Grammatical subjects tend to be continued as subjects in the next clause and preferably by pronouns or subject omission. In sum, the ¿ndings on early speech production con¿rm the ¿ndings on social-cognitive skills in pre-linguistic children. The results on comprehension of intersentential pronouns provided so far are less concurrent. Song and Fisher (2005, 2007) report an impact of confounded subject role and ¿rst mention on accuracy in the production of noun and pronoun phrases and the ¿xation of referents in English-learning 2;5- and 3-year-olds. The 3-year-olds repeated more accurately and had more correct ¿xations if the presented subject continued the subject of the previous clause and was a pronoun. Even 2;5-year-olds relate PERS to the ¿rst mentioned subject. Song and Fisher (2007) conclude that children as young as 2;5 use structural features in determining referent prominence and establishing a mental representation of the situation. In contrast, Arnold, Brown-Schmidt, and Trueswell (2007) found that 3;5-year-old English-learning children show little or no signs of a bias towards ¿rst-mentioned antecedents in the comprehension of PERS. Considering the effect of animacy on pronoun use in children, to our knowledge, there are no studies on European languages. Lopes and Quadros (2005) and Averburg (2008) analysed animacy effects in the acquisition of object pronouns in Brazilian Portuguese. They report signi¿cant interactions of referent animacy and pronoun type (PERS, null pronoun). However, it is well known that already infants pay high attention to animacy and agency (e.g. Mandler 1992). Four-month-old children distinguish animate from inanimate beings referring to the in/capability of moving autonomously (Poulin-Dubois, Lepage,

72

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

and Ferland 1996). 14-month-olds have generalized the properties of drinking and sleeping throughout the animal domain, and the properties of “being keyed” and “giving a ride” throughout the vehicle domain (Mandler and McDonough 1996). Further, there is plenty of evidence that children initially base their interpretation of linguistic information on conceptual knowledge related to animacy, cf. interpretation of passive sentences (Lempert 1990), comprehension of double object constructions (Drenhaus and Féry 2008) and acquisition of subordinating verbs (Becker 2009). A caveat in the research on anaphora resolution in children is the scarcity of studies combining production and comprehension (but see Chipman and de Dardel 1974; Girouard, Ricard, and Gouin-Décarie 1997). Recently, there appeared a sample of studies conducted in the framework of Bi directional Optimality Theory (Bi-OT). Bi-OT starts out from the claim that a well-formed output results from optimizing it bi directionally from form to meaning and from meaning to form, in other words, by taking into account both the speaker’s and the hearer’s perspective (Blutner 2000). Hendriks and Spenader (2006) hypothesize that children’s processing capacities are initially restricted. Especially with linguistic structures that are low frequent or based on complex rules of form-meaning association children only perform unidirectional optimization which results in asymmetric production and comprehension outputs. Beside the well-known principle B delay, comprehension delays are reported in the acquisition of word order (Hendriks, de Hoop, and Lamers 2005) and inde¿nite subjects and objects (de Hoop and Krämer 2005). Koster, Hendriks, and Hoeks (subm.) who investigated 5- to 7-year-old Dutch-learning children found nontarget use of subject pronouns in both modalities when children were presented with stories including a topic change. Children relate the pronoun to the ¿rst topic in comprehension and produce the pronoun irrespective of whether the story exhibits a topic change or not. The authors argue that the production problem can be explained as a consequence of the children’s insuf¿cient ability to take the speaker’s perspective into account. However, the appearing heterogeneity of the reported delays raises the question whether there are other or further factors which affect the processing of grammatical structures differently in production and comprehension. One possible source of asymmetric processing of pronouns is the type of cues processed in both modalities. For instance, semantic cues might be more accessible in production than in comprehension since production starts from a complete mental representation of the situation while the appropriate linguistic structure for this representation has to be chosen from the set of possible structures. Comprehension, in contrast, requires the construction of a mental representation on the basis of the linguistic input and, thus, structural cues might be higher ranked in comprehension than in produc-

Personal pronouns across languages

73

tion. Alternatively, the hearer might rely on semantic cues, e.g. animacy or agency, and ignore the structural cues for pronoun resolution while the speaker might present the central referents more or less automatically in prominent structural position since this is the unmarked structure. In sum, there is evidence that children at the age of 3 produce PERS in structurally prominent position and assign to it referents which are in the joint focus of attention. In order to test these ¿ndings in a wider range of languages, we conducted an experiment that combines imitated production with comprehension of PERS. Speci¿cally we ask whether children relate PERS in subject position to: (a) structural prominence and/or referent animacy, (b) the same cue pattern in comprehension and production, and whether we ¿nd: (c) development over age and (d) cross-linguistic differences in cue use.

2

The Experiment

2.1

Hypotheses

The literature proposes a correlation of PERS with structurally high prominent or salient referents in adult as well as in child language. It further reports an effect of referent animacy on the maintaining of subject referents. Following this, Hypothesis A of our study is: If children make use of structural prominence cues, they will relate PERS to referents with features of high prominence. If children make use of the animacy cue, they will relate PERS to animate referents. Hypothesis B is that there are no differences in cue use between production and comprehension of PERS. Children are expected to operate on the same mental representation in both modalities and process the same cue pattern. In our experiment two aspects of cue use can change with age: i) the preferred cue type and ii) the complexity of the cue pattern. Assuming that semantic features are more accessible than structural features to the younger children, Hypothesis C1 is that younger children will more strongly rely on the semantic cue (animacy) while older children will more strongly rely on structural prominence. With respect to cue pattern complexity the results are expected to be in line with the general ¿nding of an increase in structural complexity with age.

74

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

Therefore, Hypothesis C2 is that the younger children will relate only one of the two cue types to the use of PERS while the older children will process a combination of the two. With respect to cross-linguistic comparison it is worth noting that the two Slavic languages investigated here, Russian and Bulgarian, exhibit a null pronoun, with pro-drop being obligatory in Bulgarian but facultative in Russian. In contrast, the third language, which is German, is a language without null pronoun. Additional polarization of German and the two Slavic languages arises from the existence of the so-called demonstrative 1 (DEM I) in German which partially shares its anaphoric domain with PERS. Like many other languages, the two Slavic languages exhibit only one demonstrative comparable to the demonstrative 2 of German (DEM II).1 The relevant differences in the core pronoun system of the three languages considering the types of (unstressed) pronouns and their anaphoric domains (approximately determined with respect to referent salience) are illustrated in (4). It is worth noting that in all three languages, PERS is not restricted to either animate or inanimate referents as is the case in English. (4)

German

PERS

Russian

NULL

Bulgarian

NULL

max. salient

DEM I PERS PERS

DEM II DEM DEM

min. salient referent

Hypothesis D is that the differences in the three pronoun systems cause differences in (the acquisition of) the anaphoric capacity of PERS. We expect that the correlation of PERS with high prominent referents is closest in German since here PERS is the formally least complex pronoun type. Its relation to the most salient referent should be based on cues more reliably setting apart potential antecedents than referent animacy does. Correlation of PERS with high prominent referents is expected to be weaker in Russian which exhibits a facultative null pronoun, and weakest in Bulgarian which exhibits a fully grammaticalized null pronoun. In these languages, PERS must not be related to the most salient referent and might be free for other referential relations. The 1. German DEM I is identical with the forms of the de¿nite article (der-msc, die-fem., das-ntr.) except for genitive forms and the dative plural. German DEM II (dies-) is a deictic but distance-neutral demonstrative.

Personal pronouns across languages

75

existence of DEM I in German and its non-existence in the Slavic languages is likely to strengthen the hypothesized differences. 2.2

Method and material

The experiment was designed and performed identically for all three languages. Children were presented with a short story acted out with toy puppets by the experimenter. While acting out, the experimenter was telling the story. After three initial sentences introducing two potential antecedents, the child heard the sentence entailing the investigated cues (antecedent sentence) followed by the sentence containing PERS as a subject pronoun (anaphoric sentence). The statement made by the anaphoric sentence is true for both of the two referents (e.g. being blue). Since no semantic, gender, or other cues are given, reference of PERS is ambiguous. The antecedent sentence is a simple transitive sentence consisting of subject + V¿n + object (5a). Since all referents (toy ¿gures) are visible to the child and the experimenter during the experiment, the de¿nite article has been used in the German and Bulgarian subject/object phrases. Russian does not exhibit articles; therefore, bare nouns were used. The anaphoric sentence consists of a pronoun in subject position + V¿n (either copula or lexical verb) + adverbial phrase (5b). (5)

a.

LăvătSG M DEF ‘The lion

hape3GS PRES IMPF bites

tigăraM DEF. the tiger.’

b.

TojPERS 3SG NOM M ‘He

e is

jadosanSG M. angry.’

Pronoun production was elicited by a repetition task. The child was instructed to repeat the last sentence heard to a seemingly distracted puppet introduced as absent-minded and hard of hearing. This puppet was played by the second experimenter asking What happened at the end? Pronoun comprehension was evaluated by a clari¿cation question (who is x?) asked by the distracted puppet immediately after the child repeated the anaphoric sentence. Who is appropriate to ask for animate and, metaphorically, inanimate beings in all three languages. (6)

Example of the experimental settings: Exp 1: That’s the bear and that’s the ball. The bear likes to play football. Now, the ball is in front of the bear. antecedent sentence: The bear is kicking the ball.

76

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

anaphoric sentence: Exp 2 (distracted puppet): Child: PRODUCTION Exp 2 (distracted puppet): Child: COMPREHENS.

He is white. (notice: both bear and ball are white!) Pardon? I did not get it. He is white. Who is white? The bear.

The antecedent sentence systematically varies the structural cue values (high vs. low prominence) and the semantic cue values (animate vs. inanimate referent). The 2 × 2 setting results in 4 combinations of referent properties (Table 1). We will refer to these combinations as condition A, B, C, and D. Table 1. Types of antecedent sentences (conditions) condition A B C D

antecedent features animate HP + animate LP inanimate HP + animate LP inanimate HP + inanimate LP animate HP + animate LP

example ‘the monkey is hugging the dog’ ‘the ball is touching the bear’ ‘the tractor is pushing the bus’ ‘the elephant is driving the tractor’

*HP = high prominent; LP = low prominent

The anaphoric sentence varied the three pronoun types personal pronoun (PERS), demonstrative pronoun (DEM), and null pronoun (NULL). Table 2. Types of anaphoric sentences pronoun type NULL PERS DEM

example 1 ‘_ is white’ ‘he is white’ ‘this one is white’

example 2 ‘_ is laughing loud’ ‘he is laughing loud’ ‘this one is laughing loud’

By alternating three pronoun types we aimed at detecting when children establish differences in the anaphoric capacity of these pronoun types and how the oppositions look like. We will not focus on this question here.2 However, the presence of these pronoun types in the experiment activates the children’s knowledge on (the anaphoric capacity of) each of them. Experimental studies on pronoun production suffer from the freedom the subjects have in selecting the nominal phrase out of a range of potentially appropriate types of nominal phrases. Pre-mentioning (i.e. priming) of forms or structures facilitates the 2. For treatment of the different pronoun types see Bittner and Kuehnast (in press), and the papers of Bittner, Gagarina, and Kuehnast in ZASPiL 48.

Personal pronouns across languages

77

activation of the respective forms in a following task (cf. Gershkoff-Stowe, Connell, and Smith 2006; Shimpi et al. 2007). This is especially relevant for the choice of repetition in the production part. Activation of the three pronoun types raises the possibility of substitution of the presented pronoun by one of the other types. Substitution should occur either when the child has a general preference for another pronoun type than the one presented or when the presented pronoun is not in line with the child’s mental representation of the story. Various studies have demonstrated that young children change ungrammatical or even unsuitable structures in a repetition task when having productive command of the relevant grammatical structure (e.g. Kidd et al. 2005; Costa and Friedman 2009), and, further, that they repeat words and non-words more correctly when having reached a certain level of grammatical knowledge (Dispaldro et al. 2009). The question of interest in the repetition task is in which condition children show a preference to produce PERS, either by repeating it accurately or by substituting PERS for presented NULL or DEM.3 Since we were interested in the pattern of pronoun substitutions and since the anaphoric sentences in our experiment are short and less complex, the children were not forced to repeat these sentences as accurately as they could but were asked to retell the very end of the story to the puppet. The experiment was introduced to the child by the ¿rst experimenter, ¿rst by an oral introduction of the game, the puppet and the child’s task, then by three practice items. Each child was tested with one of two randomly ordered sets of test items consisting of 12 stimuli each. About 150 children ranging from 2;6 to 5;6 in age were tested in each language. Here, we report on the results of two groups which we will call the 3- and 5-year-olds, although the older group of the German children is a little younger than 5;0 (Table 3). Table 3. Subjects

German Russian Bulgarian

n 27 25 21

3-year-olds age range mean age 2;8 – 3;4 3;0 3;0 – 3;5 3;3 2;8 – 3;4 3;1

n 26 25 31

5-year-olds age range mean age 4;6 – 5;0 4;9 5;0 – 5;5 5;5 4;10 – 5;5 5;3

3. For two reasons we were testing NULL in German although omission of the subject phrase is ungrammatical in the presented structures. First, subject or topic omission is a typical phenomenon in early child language, even in non-pro-drop languages (Hyams and Wexler 1993; Blom and van Geert 2004). Second, we aimed at holding the experimental setting parallel across languages.

78 2.3

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

Patterns of a/symmetric processing of the presented cues

The structure of the experiment determines the possible patterns and outcomes of a/symmetry in cue use across the two modalities. However, only the comprehension part provides direct access to the type of cues related with PERS since the child explicitly names the referent to whom s/he relates the anaphoric sentence. In contrast, there is no explicit mentioning of the assumed referent of PERS in the repetition task. The types of cues used in this task can only be inferred indirectly. Each value of the two cue types may be used as the only cue. Additionally, there are four possible combinations of the four values. (7)

a) b) c) d)

high prominence only low prominence only animacy only inanimacy only

e) f) g) h)

animate + high prominence inanimate + high prominence animate + low prominence inanimate + low prominence

If high prominence or low prominence features appear to be the only cues in the comprehension of PERS, symmetric processing of these cues would result in a lack of signi¿cant differences across conditions A to D in the production of PERS, since all conditions exhibit one high prominent and one low prominent referent. If children relate the comprehension of PERS to animate antecedents, symmetric processing would result in the least frequent production of PERS in condition C, since it presents only inanimate antecedents. Vice versa, symmetric processing of inanimacy would result in the least frequent production of PERS in condition A, since it presents only animate antecedents. If children relate PERS to the cue combination ‘animate + high prominent referent’ in comprehension, symmetry would result if PERS is produced more frequently in condition A and D, since only these conditions contain an animate high prominent referent. Similarly, symmetric processing of the cue pattern ‘inanimate + high prominent referent’ occurs if PERS is produced most frequently in condition B and C, since only these conditions contain an inanimate high prominent referent. Symmetric processing of the cue patterns ‘animate + low prominent’ and ‘inanimate + low prominent referent’ requires the most frequent production of PERS in conditions A and B and conditions C and D respectively, since only these conditions contain antecedents of the respective type. Table 4 summarizes the pairings of cue use that would indicate symmetric processing of the investigated cues in production and comprehension.

Personal pronouns across languages

79

Table 4. Pairings assigning symmetric application of the investigated cues

a) b) c) d) e) f) g) h)

Comprehension resolution of PERS to: high prominence low prominence animacy inanimacy animate high prominence inanimate high prominence animate low prominence inanimate low prominence

3

Results

Production (imitation) distribution of PERS: no preference for any condition no preference for any condition least frequent in C least frequent in A most frequent in A+D most frequent in B+C most frequent in A+B most frequent in C+D

The following analyses are based on the children’s production of PERS in the repetition task, not on PERS presented in the anaphoric sentence (cf. (6)). Because of limitations in working memory (e.g. Kemps, de Rammelaere, and Desmet 2000), it is rather unlikely that the sentence given by the experimenter is still available to the child after s/he has performed the repetition task. Especially the unstressed pronoun will hardly remain stored. Moreover, the sentence produced by the child in the repetition task is expected to mirror the mental representation the child established during presentation of the stimuli. 3.1

Production of PERS by the 3-year-olds

Absolute task conformity in the repetition task would give a production rate of 33.3% for each of the three pronoun types, across and within conditions. The actual rate of repetitions, however, depends on developmental preferences for single pronoun types, and, potentially, the cue constellation within the single conditions. Both factors might force the child to substitute another pronoun type or even a noun phrase for the presented pronoun. For instance, when the experimenter presents the anaphoric sentence der ist blau ‘this one is blue’, the child might substitute der by repeating Ø/er/der traktor ist blau ‘Ø/it/the tractor is blue’. Figure 1 presents the proportions of pronoun types and noun phrases produced by the 3-year-olds.

80

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

3-year-olds:repeated subject phrase (%) 60

40

PERS DEM NULL Noun

20

0 Germ

Russ

Bulg

Figure 1. Subject phrases produced in the repetition task by the 3-year-olds (%)

production of PERS (%)

In all three languages, the 3-year-olds tend to omit the subject phrase, even when presented with PERS or DEM. This can be seen from the high rate of NULL in Figure 1. Consequently, PERS production scores below the 33.3% mark. In German, it is even lower than DEM production. In order to ¿nd out whether the 3-year-olds relate PERS production to structural prominence and/or referent animacy, we analysed the distribution of PERS over the four conditions by summing up all PERS productions to 100%. Two values are distinguished in this analysis: i) repetition of PERS and ii) substitution of NULL or DEM by PERS. 3-year-olds

40 30

correct repetition

20

substitution by PERS

10 0 A

B

C GERM

D

A

B

C

D

RUSS

A

B

C

D

BULG

Figure 2. PERS production over conditions in the 3-year-olds

An analysis of variance (ANOVA) with the factor condition (A to D) on the production of PERS conducted for each language individually does not reveal signi¿cant effects. Considering German and Russian, this is due to the low number of PERS productions (Figure 1). Considering Bulgarian, this could be interpreted as evidence for an impact of structural prominence (production

Personal pronouns across languages

81

pattern (a) and (b), Table 4). However, the comparably low number of substitutions by PERS indicates high task conformity of the Bulgarian children, which makes such an interpretation highly speculative. Figure 2 shows some tendencies in PERS production which are of interest with respect to a/symmetric processing of the presented cues. The Russian data display production pattern (c) and (g), the Bulgarian data display pattern (c) evidencing a correlation of PERS production with referent animacy; PERS tends to be avoided if only inanimate referents are given (condition C). Interestingly, this pattern does not emerge in German. Here, production of and substitution by PERS do not ¿t with any of the patterns listed in Table 4.4 3.2

Comprehension of PERS by the 3-year-olds

In order to ¿nd out which cue type is more important for the resolution of PERS in the 3-year-olds we conducted a Ȥ2-test and compared the choices of (i) high prominent vs. low prominent antecedents and (ii) animate vs. inanimate antecedents after produced PERS. Table 5 gives the proportions in terms of high prominent and of animate antecedent choice. Table 5. Resolution of PERS to high prominent referents and animate referents referent/language high prominent animate

German 65% 45%

Russian 47% 65%

Bulgarian 52% 64%

While the German 3-year-olds tend to resolve PERS to high prominent referents (65%, Ȥ = 3.600 p = .058) irrespective of animacy status, the Bulgarian 3-year-olds chose the animate referent irrespective of its prominence status (64%, Ȥ = 4.738, p = .030). The Russian 3-year-olds ignore structural prominence as a resolution cue (p = .732) but show a weak tendency to choose the animate referent (65%, Ȥ = 2.941, p = .086). Figure 3 gives the percentages of choices of high prominent antecedents (subject choice) in each condition. By grouping together the conditions presenting an animacy contrast on the left side of the graph and the conditions without an animacy contrast on the right side, we aim at highlighting the speci¿c impact of each of the two cue types. An impact of animacy would be most apparent in the conditions with animate 4. Though the overall production rate is similar in all conditions, we do not interpret this as resemblance of production pattern (a) and/or (b) since the pattern of correct repetitions and substitutions are contradictory.

82

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

contrast (B and D) and an impact of structural prominence in the conditions without animacy contrast (A and C). 3-year-olds subject choice (%)

100 80 60 animS

40

inanimS

20 0 GERM

RUSS

BULG

with animacy contrast (B, D)

GERM

RUSS

BULG

no animacy contrast (A,C)

Figure 3. Resolution of PERS by the 3-year-olds

For each language, we conducted separate one-way analyses of variance (ANOVA) with the factor condition on the choice of high prominent antecedents. There are no signi¿cant differences across conditions in the data of the German 3-year-olds. The tendency to choose the high prominent referent found by the Ȥ2-test results from all of the four conditions. In contrast, there is a main effect of condition in the data of the Bulgarian 3-year-olds (F(3,57) = 3.026, p = .037). Con¿rming the overall preference for the animate referent found in the Ȥ2-test, the children chose the animate high prominent referent in condition D (73%) but the animate low prominent referent in condition B (74%) which is a highly signi¿cant difference in terms of high prominent referent choice (p = .006). Without animacy contrast, the Bulgarian children are irresolute in high prominent referent choice (condition A = 57%, condition C = 62%). The Russian children show similar rates of high prominent referent choice as the Bulgarian children. However, due to the low number of PERS productions, no signi¿cant differences across conditions turned out in the Russian 3-year-olds. 3.3

Production of PERS by the 5-year-olds

From 3 to 5 years of age, we ¿nd a considerable development in the production rates of PERS. In German, PERS develops to the most frequently produced pronoun type. In Russian, we ¿nd an increase of PERS production comparable to German, but NULL production is still more frequent. In Bulgarian, PERS and DEM production increases while NULL production decreases, resulting in similar production rates for all three pronoun types. Again this points towards high task conformity and explains the mere lack of substitutions by PERS in

Personal pronouns across languages

83

Bulgarian (see Figure 5). Figure 4 presents the proportions of pronoun types and noun phrases used as subject phrase in each language. 5-year-olds:repeated subject phrase (%) 60

40 PERS DEM NULL

20

Noun

0 Germ

Russ

Bulg

Figure 4. Subject phrases produced in the repetition task by the 5-year-olds (%)

Production of PERS (%)

Figure 5 presents the distribution of all PERS productions (100%) over the four conditions for each language, again divided into i) correct repetition of PERS and ii) substitution of NULL and DEM by PERS. 5-year-olds

40 30

correct repetition

20

substitution by PERS

10 0 A

B

C

D

GERM

A

B

C

RUSS

D

A

B

C

D

BULG

Figure 5. PERS production over conditions in the 5-year-olds

The analysis of variance (ANOVA) with the factor condition (A to D) on the production of PERS conducted for each language does not reveal signi¿cant differences. The descriptive analysis suggests that PERS production by the German 5-year-olds best ¿ts production pattern (g), indicating a preference to produce PERS in the presence of an animate low prominent referent. This preference is con¿rmed by the lowest production and substitution rates in condition C (production pattern (c)). Production pattern (c) also turns out in the

84

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

data of the Russian 5-year-olds. It ¿ts production pattern (e) pointing to a preference for PERS in the presence of animate high prominent referents. With the Bulgarian children, again, high task conformity covers possible preferences. 3.4

Comprehension of PERS by the 5-year-olds

The Ȥ2-test comparing the relation of PERS to (i) high prominent vs. low prominent antecedents and/or (ii) animate vs. inanimate antecedents reveals signi¿cant resolution to the high prominent antecedent by the German 5-yearolds (65%, Ȥ = 7.716, p = .005). The same is observed for the Russian 5-yearolds (74% subject choice, Ȥ = 21.316, p < .001). The Russian children also show a tendency to resolve PERS to the animate antecedent (60%, Ȥ = 3.8000, p = .051). For the Bulgarian 5-year-olds, we ¿nd signi¿cant resolution of PERS to the animate antecedent (59%. Ȥ = 3.973, p = .046) while resolution to the high prominent antecedent is a mere tendency (59%, Ȥ = 3.252, p = .071). Table 6. Resolution of PERS to high prominent referents and animate referents referent/language high prominent animate referent

German 65% 58%

Russian 74% 60%

Bulgarian 59% 59%

Figure 6 presents the choice of high prominent referents (subject choice) for each condition separately. Again, the two conditions with animacy contrast (B and D) are grouped together on the left side of the graph and the two conditions without animacy contrast (A and C) on the right side. 5-year-olds subject choice (%).

100 80 60 animS

40

inanimS

20 0 GERM

RUSS

BULG

with animacy contrast (B, D)

GERM

RUSS

BULG

no animacy contrast (A, C)

Figure 6. Resolution of PERS to the high prominent referent by the 5-year-olds

Personal pronouns across languages

85

The one-way analyses of variance (ANOVA) with the factor condition on the choice of high prominent antecedents conducted for each language shows that there is no signi¿cant effect in the German 5-year-olds. Thus, the tendency to resolve PERS to the high prominent referent attested by the Ȥ2-test suggests a developing correlation of PERS with high prominent referents independent of referent animacy (but see the resolution behaviour in B discussed below). For the Russian 5-year-olds, the analysis reveals a signi¿cant impact of condition on resolution to the high prominent referent (F(3,91) = 3.624, p = .016) mainly caused by different resolution behaviour in conditions A vs. B (p = .002) and D vs. B (p = .028). This is con¿rmed by the Ȥ2-test: In Russian, high prominence is the most decisive cue in PERS resolution accompanied by referent animacy. The Bulgarian 5-year-olds show the strongest impact of referent animacy on the resolution of PERS. There is a main effect of condition (F(3,107) = 9.282, p < .001). Selection of the high prominent referent is signi¿cantly more frequent in the conditions presenting an animate subject (A and D) than in each of the two other conditions (A : B/C p = .008 and p < .001; D : B/C p = .018 and p = .001). Although there is no effect of animacy on the resolution behaviour in the German 5-year-olds, comparison of Figures 3 and 6 reveals a decrease of resolution to the high prominent referent in condition B similar to that found for Russian and Bulgarian. Seemingly, across languages, the 5-year-olds disfavour inanimate high prominent referents occurring in the presence of animate low prominent referents as antecedent candidates for PERS.5 3.5

Comparison of age groups

To evaluate the development in comprehension over age we conducted a two way analyses of variance (ANOVA) with the factors condition and age group for each language separately. There is a main effect for age group (F(1,121) = 7.888, p = .006) in Russian, mainly caused by an increased selection of the high prominent referent in condition A and C in the 5-year-olds. This ¿ts the signi¿cant choice of the high prominent referent in the 5-year-olds found by the Ȥ2-tests. In Bulgarian, there is a weak interaction of condition and age group (F(3,164) = 2.595, p = .054) caused by an increase in the choice of the (animate) high prominent referent in condition A and a decrease in the choice of the (inanimate) high prominent referent in condition C. As indicated 5. The combination of an inanimate high prominence-referent (subject) with an animate low prominence-referent (object) is a marked agent-patient pattern in the accusative languages.

86

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

by the Ȥ2-test, referent animacy is the main resolution cue in both age groups in Bulgarian. For German, no signi¿cant effects and interactions of condition and age group turned out. However, the Ȥ2-test indicates that selection of the high prominent referent develops from a mere tendency in the 3-year-olds (p = .058) to a relevant resolution strategy in the 5-year-olds (p = .005). For production, the two way analyses of variance (ANOVA) with the factors condition and age group show no signi¿cant effects. However, comparison of the results over age is interesting with respect to condition C and B. Low frequency of PERS in condition C points to referent animacy as a cue for PERS production. In German, the decrease of PERS production over age suggests an increasing relevance of referent animacy. In Russian and Bulgarian, in contrast, PERS production increases in C suggesting a loss of importance of the animacy cue over age. At the same time, it decreases in the marked condition B. This also points to the emergence of high prominence as relevant cue in the Russian- and Bulgarian-speaking 5-year-olds. The cue combination ‘animate+high prominence’ emerges earlier and seems to be stronger in Russian than Bulgarian. In German, the decrease in PERS production in condition C is accompanied by an increase in conditions A and D both presenting an animate high prominent referent. While high prominence seems to be most decisive for PERS production in the German 3-year-olds, the 5-year-olds also consider referent animacy as a relevant but secondary cue which leads to the cue combination ‘high prominence+animate’.

4

Discussion

The main question of this study was whether children make use of the cue types structural prominence and/or referent animacy in the production and comprehension of PERS in subject position. The application of these cues has been tested with 3- and 5-year-old children learning German, Russian, or Bulgarian; three languages which exhibit substantial typological as well as tiny structural differences in their pronominal systems. We found that the children’s use of the two cue types is not a matter of stronger reliance on linguistic or semantic features but depends on the language-speci¿c structure of the pronoun system. While children learning a Slavic language exhibiting a null pronoun prefer to relate production and comprehension of PERS to the semantic cue referent animacy, children learning German, which does not exhibit a null pronoun, prefer to relate PERS to structural prominence cues. Further, we found development over age attested by increasing complexity of the cue pattern employed with PERS, and we found more evidence for sym-

Personal pronouns across languages

87

metric processing of the investigated cues in production and comprehension than for asymmetric processing. Before discussing the results in relation to the hypotheses presented in section 2.1 we need to say that our results on production are preliminary. It turned out that the experiment did not elicit the amount of production data for PERS needed for statistical analyses. The younger children prefer subject drop, which means that they frequently omit PERS (as well as DEM). The older children either behave highly task-compliant (Bulgarian) by repeating exactly the pronoun type presented by the experimenter or PERS production is equally spread over all conditions (Russian, German).6 Though the latter ¿nding is likely to constitute evidence of a relation of PERS to structurally prominent referents, the amount of data in each condition is low. However, the tendencies observable in production concur with the ¿ndings on comprehension in nearly all age-language groups. Hypothesis A proposed that if children relate the use of PERS to the cues presented in the experiment, they relate PERS to high prominent and/ or animate referents since these are the cue values supposed to be associated with high salience in the mental representation of referents (section 1). This hypothesis has been con¿rmed by the data. All signi¿cant results point to a correlation between PERS and these values. This holds also for the tendencies observed in production, except for the German 5-year-olds. In sum, the results provide evidence for (i) young children’s processing of feature oppositions which differentiate the salience of referents in the mental representation and (ii) the anaphoric correlation of PERS with features indicating high salience of referents from at least 3 years of age. Our ¿ndings for German are in line with the results for English-learning 2;5- and 3-year-olds reported by Song and Fisher (2005, 2007). Using various experimental methods, Song and Fisher consistently found a preference to resolve subject PERS to referents presented as subject in ¿rst mentioned position, in our terms to the high prominent referent. However, assuming that this correlation is present in the grammar of 2- to 3-year-olds across all languages cannot be upheld in the light of our results for the two Slavic languages. Even in the two Slavic languages children do relate PERS to salient referents, but use a different type of cue, which is referent animacy. The impact of the language-speci¿c structure on the anaphoric capacity of PERS is con¿rmed by the results of the 5-year-olds (see below). The results of Song and Fisher for 6. Due to a high educational standard in the Bulgarian day care system, children are used to accurately respond to educational tasks. Thus, the sentence structure in the repetition task was too simple for the Bulgarian 5-year-olds.

88

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

English included the ¿nding that there is no overall preference in 3-year-olds to rely on semantic features of referents only. This is con¿rmed by our results. The transitive sentences we used have an agent-patient pattern. If the behaviour of the 3-year-olds would be primarily determined by world knowledge, all children should relate PERS to the animate antecedent since it is presupposed to present the agent, and the agent is presupposed to be a central referent. The clear differences between the children learning a Slavic and those learning a Germanic language (which all are non-ergative languages) show that 3-yearolds use some speci¿c linguistic knowledge in the treatment of PERS. This con¿rms ¿ndings of other studies (e.g. Campbell et al. 2000). Moreover, the cross-linguistic difference in the anaphoric use of PERS suggests that 3-yearolds have established some knowledge on oppositions in the anaphoric capacity of different pronoun types. Speci¿cally, children learning a Slavic language differentiate PERS and NULL. Hypothesis B concerned the question of symmetric or asymmetric processing of the tested cues in comprehension and production. Several studies found asymmetric performance in the two modalities, speci¿cally a delay in comprehension (e.g. Chien and Wexler 1990; Baauw and Cuetos 2003; Hendriks, de Hoop, and Lamers 2005). The Bi-OT approach to language acquisition claims that this asymmetry is caused by limited processing capacities which restrict the optimization of output structures to unidirectional optimization while only bi directional optimization would generate the right output (Hendriks and Spenader 2006). Weaker performance in comprehension is assumed to result from a lack of optimization from meaning to form, i.e. the speaker’s perspective. Theoretically, asymmetries in children’s production and comprehension abilities might also arise from the use of different cues in the two modalities. However, our results do not point to such differences in cue processing but give more evidence for symmetric processing of cues in the two modalities, i.e. the null hypothesis. Table 7 and 8 summarize the ¿ndings on cue use for each language and age group separately. Table 7. Cues used with PERS in the 3-year-olds7

production comprehension a/symmetric processing:

German ? high promin. ref.(*) ?

Russian animate ref. animate ref.(*) SYMMETRIC

Bulgarian animate ref. animate ref.* SYMMETRIC

7. Stars indicate a statistically reliable correlation, stars in brackets a tendency (0.1 to 0.05), unmarked notions tendencies turning out in descriptive analyses.

Personal pronouns across languages

89

Table 8. Cues used with PERS in the 5-year-olds German production animate low promin. ref. comprehension animate high promin. ref.* a/symmetric processing: ASYMMETRIC

Russian animate high promin. ref. animate(*) high promin. ref.* SYMMETRIC

Bulgarian ? animate* high promin. ref.(*) ?

In three of the four production-comprehension pairings for which the experiment elicited analysable data, there is evidence for the use of the same cues in the two modalities. The only exception occurs with the German 5-year-olds. Bearing in mind the preliminary status of the results on production and the restricted set of cues presented in the experiment, the results suggest i) that children at the age of 3 are able to establish a consistent mental representation of agent-patient (or source-goal, Kehler et al. 2007) situations presented by simple transitive sentences, and ii) that the anaphoric capacity of PERS is bound to some consistent cues in the grammar of 3-year-olds. These cues might not necessarily be the cues tested in our experiment, though they are likely to be part of the ¿nal cue pattern. Hypothesis C proposed that there is development over age in the use of the presented cue types. Either there is a difference in the cue types used by the different age groups (hypothesis C1) or there is a difference in the complexity of the cue pattern processed by each age group (hypothesis C2). As Tables 7 and 8 show the results point to the latter. The 3-year-olds relate PERS to only one of the presented cues, i.e. referent animacy (Russian, Bulgarian) or structural prominence (German) while the 5-year-olds relate it to both referent animacy and structural prominence. Certainly, this development correlates with increasing working memory which allows processing of more information at once. In order to gain some insight into speci¿cally linguistic components of this development over age, it is worth considering the results for DEM and NULL elicited in the experiment (see footnote 2). Comparison of Figures 1 and 4 indicates a change in the composition of the pronoun system in all three languages. PERS is of marginal status in the 3-year-olds. At 5 years of age, PERS and even DEM have become part of the core pronoun system (next to NULL in the Slavic languages). This is in accord with ¿ndings of Gülzow and Gagarina (2007) and Bartsch and Bittner (2009) for pronoun use in narratives elicited from the same children analysed here. The increase in productive use of subject pronouns goes hand in hand with deeper speci¿cation of the pronoun’s anaphoric capacity. In all three languages DEM becomes related to less salient

90

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

referents, i.e. inanimate or low prominent referents, while PERS (and NULL in the Slavic languages) becomes related to the more salient referents. The ¿rst process is most prominent in German, the latter in Russian. Hypothesis D concerned the impact of language-speci¿c properties of PERS in young children. In section 1, we hypothesized that the following aspects cause differences in the anaphoric capacity of PERS across the investigated languages: (i) the existence of NULL in the two Slavic languages vs. its absence in German, (ii) the existence of DEM I in German vs. its absence in the two Slavic languages, and (iii) the different grammatical status of NULL in the two Slavic languages; while Bulgarian has obligatory subject drop in non-contrastive sentences, pro-drop is restricted to speci¿c contexts in Russian. Our results (Tables 7 and 8) con¿rm these considerations. PERS is the formally least complex anaphor in German. The formally least complex anaphor is suggested to refer to the most salient referent (Gundel, Hedberg, and Zacharski 1993; Levinson 2000; Ariel 2004). The most salient referent prototypically occurs in structurally prominent position. The existence of DEM I in German pushes the anaphoric capacity of PERS further towards maximally salient referents. This is the background for an explanation of the predominant choice of the high prominence cue by the German children (We do not have an explanation for the 5-year-olds correlation of PERS with low prominence in production.). In contrast, the existence of a formally less complex pronoun in the Slavic languages, i.e. NULL, directs the anaphoric capacity of PERS towards referents which are not necessarily the most salient ones (see also Spanish). In our experiment, the only accessible cue not related to maximal but to high salience is referent animacy. This is the background for the choice of referent animacy by the Russian and Bulgarian children. In this view, two other instances of cross-linguistic variation can be explained, i) the inner-Slavic variation in the use of referent animacy and ii) the variation in the cue patterns used by the 5-year-olds (speci¿cally in comprehension, Figure 8). The former can be traced back to the different status of pro-drop in the two languages. Since NULL in Russian is not obligatory in many contexts PERS is closer related to maximal salience in Russian than in Bulgarian. The latter goes back to a different status of the two cue types in the complex cue pattern. In German and Bulgarian, the children continue the cue used at age 3 as the main cue when 5 years of age, adding the second cue as an additional but secondary one. The Russian data indicate a change in cue preferences. While the 3-year-olds tend to relate PERS to referent animacy, the 5-year-olds use structural prominence as the main cue. A possible hypothesis on what might explain this change is the lack of reliability (Bates and MacWhinney 1987) in the use of NULL, i.e. the dif¿culty to detect the correct grammatical contexts for NULL. This affects the

Personal pronouns across languages

91

determination of the anaphoric capacity of NULL, and in consequence even that of PERS in Russian.

5.

Conclusion

An experiment provides data on only a restricted set of factors that potentially inÀuence speaker/hearer behaviour. In the experiment reported here, the children can relate the use of PERS only to structural prominence and/or referent animacy. Consequently, processing of these cues cannot be taken as mirroring the complete anaphoric capacity of PERS in the children’s grammar. Further, investigation of only one pronoun type cannot detect whether the children have established oppositions in the anaphoric capacity of different pronoun types. What our results show is: (a) Already at 3 years of age, production and comprehension of PERS is related to cues indicating high salience of referents; (b) from age 3 on, children are likely to establish a consistent mental representation of a linguistic stimulus; this gives rise to symmetric processing of the presented cues for pronoun use in production and comprehension; (c) at age 5, children are capable of processing more complex cue patterns than at age 3; (d) the anaphoric capacity of PERS is determined by language-speci¿c features, speci¿cally, it varies across languages exhibiting typologically different sets of pronoun types in the core pronoun system.

References Allen, Shanley 2000 A discourse pragmatic explanation for argument representation in child Inuktitut. Linguistics 38: 483 – 521. Ariel, Mira 1990 Accessing noun-phrase antecedents. London: Routledge. Ariel, Mira 2004 Accessibility marking: Discourse functions, discourse pro¿les, and processing cues. Discourse Processes 37 (2): 91 – 116. Arnold, Jenniffer E., Sarah Brown-Schmidt, and John C. Trueswell 2007 Children’s use of gender and order-of-mention during pronoun comprehension. Language and Cognitive Processes 22 (4): 527 – 565.

92

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

Averburg, Mayra Cristina Guimarães 2008 Aquisição em português brasileiro: O parâmetro do objeto nulo [Brazilian-Portuguese acquisition: The null object parameter]. Cadernos do Congresso Nacional de Lingüística e Filologi, 12. http://www.letras.ufrj.br/posverna/doutorado/AverbugMCG.pdf Baauw, Serge, and Fernando Cuetos 2003 The interpretation of pronouns in Spanish language acquisition and breakdown: Evidence for the “Principle B Delay” as a non-unitary phenomenon. Language Acquisition 11 (4): 219 – 275. Bartsch, Susanna, and Dagmar Bittner 2009 Asymmetries in resolution and production of pronominal anaphor in German child language. RASCAL, 24. – 25.01.2009, Groningen. http://www.let.rug.nl/rascal/index_¿les/RASCAL-Bartsch.pdf. Bates, Elisabeth, and Brian MacWhinney 1987 Competition, variation, and language learning. In Mechanisms of language acquisition, Brian MacWhinney (ed.), 157–193. Hillsdale: Erlbaum. Beaver, David 2004 The optimization of discourse anaphora. Linguistics and Philosophy 27 (1): 3 – 56. Becker, Misha 2009 The role of NP animacy and epletives in verb learning. Language Acquisition 16 (4): 283 – 296. Bittner, Dagmar 2007 Early functions of de¿nite determiners and DPs in German ¿rst language acquisition. In Nominal determination. Typology, context constraints and historical emergence. Elisabeth Stark, Elisabeth Leiss, and Werner Abraham (eds.), 215 – 240. (Studies in Language Companion Series 89) Amsterdam/Philadelphia: Benjamins. Bittner, Dagmar, and Milena Kuehnast in press Comprehension of intersentential pronouns in child German and child Bulgarian. First Language. Blom, Elma, and Paul van Geert 2004 Signs of a developing grammar: subject drop and inÀection in early child Dutch. Linguistics 42 (1): 195 – 234, Blutner, Reinhard 2000 Some aspects of optimality in natural language interpretation. Journal of Semantics 17: 189 – 216. Branigan, Holly P., Martin J. Pickering, and Mikihiro Tanaka 2008 Contributions of animacy to grammatical function assignment and word order during production. Lingua 118: 172 – 189.

Personal pronouns across languages

93

Campbell, Aimee L., Patricia Brooks, and Michael Tomasello 2000 Factors affecting young children’s use of pronouns as referring expressions. Journal of Speech, Language, and Hearing Research 43: 1337 – 1349. Carpenter, Malinda 2009 Just how joint is joint action in infancy? Topics in Cognitive Science 1: 380 – 392. Chambers, Craig G., and Ron Smyth 1998 Structural parallelism and discourse coherence. Journal of Memory and Language 39: 593 – 608. Chien, Yu-Chin, and Ken Wexler 1990 Children’s knowledge of locality Conditions in binding as evidence for the modularity of syntax and pragmatics. Language Acquisition 1 (3): 225 – 295. Chipman, Harold H., and Catherine de Dardel 1974 Developmental study of the comprehension and production of the pronoun “it”. Journal of Psycholinguistic Research 3 (2): 91 – 99. Clancy, Patricia M. 1997 Discourse motivations for referential choice in Korean acquisition. In Japanese/Korean Linguistics Vol. 6, Ho Min Sohn and John Haig (eds.), 307 – 314. CSLI Publications, Stanford. Comrie, Bernard 1989 Language universals and linguistic typology. Chicago: University Press. Costa, João, and Naama Friedman 2009 Hebrew and Arabic children going Romance: On the acquisition of word order in Semitic and Romance. In Romance Languages and Linguistic Theory I, Enoch O. Aboh, Elisabeth van der Linden, Josep Quer, and Petra Sleeman (eds.), 51 – 66. Amsterdam/Philadelphia: Benjamins. Cowles, Heidi W. 2007 The inÀuence of “aboutness” on pronominal coreference. ZASPiL 48: 23 – 38. Cowles, Heidi W., Matthew Walenski, and Robert Kluender 2007 Linguistic and cognitive prominence in anaphor resolution: Topic, contrastive focus and pronouns. Topoi 26 (1): 3 – 18. Dahl, Östen, and Kari Fraurud 1996 Animacy in grammar and discourse. In Reference and Referent Accessibility, Thorstein Fretheim and Jeannette K. Gundel (eds.), 47 – 64. Amsterdam: Benjamins. Dahl, Östen 2000 Egophoricity in discourse and syntax. Functions of Language 7: 33 – 77. Dahl, Östen 2008 Animacy and egophoricity: Grammar, ontology and phylogeny. Lingua 118 (2): 141 – 151.

94

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

de Hoop, Helen, and Irene Krämer 2005 Children’s optimal interpretations of inde¿nite subjects and objects. Language Acquisition 13 (2): 103 – 123. Dispaldro, Marco, Beatrice Benelli, Stefania Marcolini, and Giacomo Stella 2009 Real word repetition as a predictor of grammatical competence in Italian children with typical language development. International Journal of Language & Communication Disorders 44 (6): 941 – 961. Drenhaus, Heiner, and Caroline Féry 2008 Animacy and child grammar: an OT account. Lingua 118 (2): 222 – 244. Frederiksen, John R. 1981 Understanding anaphora: Rules used by readers in assigning pronominal referents. Discourse Processes 4: 323 – 347. Fukumura, Kumiko, and Roger P. G. van Gompel 2010 The effect of animacy on the choice of referring expression. Language and Cognitive Processes, ¿rst published on 13 October 2010 (iFirst). Gergely, György, Harold Bekkering, and Ildikó Király 2002 Rational imitation in preverbal infants. Nature 415 (6873): 755. Gernsbacher, Morton Ann, David J. Hargreaves, and Mark Beeman 1989 Building and accessing clausal representations: the advantage of ¿rst mention versus the advantage of clause recency. Journal of Memory and Language 28: 735 – 755. Gershkoff-Stowe, Lisa, Brenda Connell, and Linda Smith 2006 Priming overgeneralizations in two- and four-year-old children. Journal of Child Language 33 (3) 461 – 486. Girouard, Pascale C., Marcelle Ricard, and Therese Gouin-Décarie 1997 The acquisition of personal pronouns in French-speaking and Englishspeaking children. Journal of Child Language 24 (2): 311 – 326. Givón, Talmy 1983 Introduction. In Topic continuity in discourse, Talmy Givón (ed.), 1 – 41. Amsterdam/Philadelphia: Benjamins. Givón, Talmy 1995 Functionalism and grammar. Amsterdam/Philadelphia: Benjamins. Gomme, Norma J., and Carolyn E. Johnson 1997 Pronominal reference in 3-year-olds’ narratives. Proceedings of the 28th Annual child language research forum, Eve V. Clark (ed.), 123 – 132. Stanford: CSLI Publications. Gordon, Peter C., Barbara J. Grosz, and Laura A. Gilliom 1993 Pronouns, names and the centering of attention in discourse. Cognitive Science 17: 311 – 347. Gülzow, Insa, and Natalia Gagarina 2007 Noun phrases, pronouns and anaphoric reference in young children narratives. ZASPiL 48: 203 – 223.

Personal pronouns across languages

95

Gundel, Jeanette K., Nancy Hedberg, and Ron Zacharski 1993 Cognitive status and the form of referring expressions in discourse. Language 69: 274 – 307. Hendriks, Petra, Helen de Hoop, and Monique Lamers 2005 Asymmetries in language use reveal asymmetries in the grammar. In Proceedings of the 15th Amsterdam Colloquium, Paul Dekker and M. Franke (eds.), 113–118. Amsterdam. Hendriks, Petra, and Jennifer Spenader 2006 When production precedes comprehension: An optimization approach to the acquisition of pronouns. Language Acquisition 13 (4): 319 – 348. Hickmann, Maya, and Henriette Hendricks 1999 Cohesion and anaphora in children’s narratives: A comparison of English, French, German, and Mandarin Chinese. Journal of Child Language 26: 419 – 452. Hyams, Nina, and Ken Wexler 1993 The grammatical basis of Null Subjects in child language. Linguistic Inquiry 24: 421 – 459. Järvikivi, Juhani, Roger P. G. van Gompel, Jukka Hyönä, and Raymond Bertram 2005 Ambiguous pronoun resolution. Contrasting the ¿rst-mention and subject-preference accounts. Psychological Science 16: 260 – 264. Juvonen, Päivi 2000 Grammaticalizing the de¿nite article: A study of de¿nite adnonimal determiners in a genre of spoken Finnish. Doctoral dissertation (Stockholm University). Edsbruk: Akademitryck AB. Kaiser, Elsi, and John C. Trueswell 2003 The quest for a referent: Investigating the interpretation of pronouns and demonstratives in real time. Poster presented at the 16th Annual CUNY Conference on Human Sentence Processing, Boston. Kehler, Andrew, Laura Kertz, Hannah Rohde, and Jeffrey L. Elman 2007 Coherence and coreference revisited. Journal of Semantics 25 (1): 1 – 44. Kemps Eva, Stijn de Rammelaere, and Timothy Desmet 2000 The development of working memory: Exploring the complementarity of two models. Journal of Experimental Child Psychology 77: 89–109. Kidd, Evan, Elena Lieven, and Michael Tomasello 2005 The acquisition of complement clause constructions: A sentence repetition study. In Proceedings of the 32nd Stanford Child Language Forum, Eve V. Clark (ed.), 50 – 59. Stanford: CLSI Publications Koster, Charlotte, Petra Hendriks, and John Hoeks subm. Comprehension and production of subject pronouns: Evidence for the asymmetry of grammar. www.let.rug.nl/~hendriks/publ.htm Lempert, Henrietta 1990 Acquisition of passives: The role of patient animacy, salience, and lexical accessibility. Journal of Child Language 17 (3): 677 – 696.

96

Dagmar Bittner, Milena Kuehnast, and Natalia Gagarina

Levinson, Stephen C. 2000 Presumptive meanings: The theory of generalized conversational implicature. Cambridge MA and London: MIT Press. Lopes, Ruth E. Vasconcellos, and Ronice Müller de Quadros 2005 Traços semânticos na aquisição da linguagem: Há efeitos de modalidade de língua? [Semantic features in language acquisition: Are there effects of language modality?]. Revista da ABRALIN 4 (1 – 2): 75 – 108. Mandler, Jean M. 1992 How to build a baby II: Conceptual primitives. Psychological Review 99 (4): 587 – 604. Mandler, Jean M., and Larain McDonough 1996 Drinking and driving don’t mix: Inductive generalizations in one-yearolds. Cognition 59: 307 – 335. Matthews, Danielle, Elena Lieven, Anna Theakston, and Michael Tomasello 2006 The effect of perceptual availability and prior discourse on young children’s use of referring expression. Applied Psycholinguistics 27: 403 – 422. Poulin-Dubois, Dian, Anouk Lepage, and Doreen Ferland 1996 Infants’ concept of animacy. Cognitive Development 11 (1): 19 – 36. Rozendaal, Margot 2008 The acquisition of reference: A cross-linguistic study. Utrecht: LOT. Shimpi, Priya M., Perla B. Gamez, Janellen Huttenlocher, and Marina Vasilyeva 2007 Syntactic priming in 3- and 4-year-old children: Evidence for abstract representations of transitive and dative forms. Developmental Psychology 43 (6): 1334 – 1346. Song, Hyun-joo, and Cynthia Fisher 2005 Who’s “she”? Discourse structure inÀuences preschoolers’ pronoun interpretation. Journal of Memory and Language 52: 29 – 57. Song, Hyun-joo, and Cynthia Fisher 2007 Discourse prominence effects on 2.5-year-old children’s interpretation of pronouns. Lingua 117 (special issue on discourse processing), 1959 – 1987. Stevenson, Rosemary J., Rosalind A. Crawley, and David Kleinman 1994 Thematic roles, focusing, and the representation of events. Language and Cognitive Processes 9: 519 – 548. Tomasello, Michael 1999 The cultural origins of human cognition. Cambridge, Mas.: Havard University Press. Tomasello, Michael 2003 Constructing a language: A usage-based theory of language acquisition. Cambridge, Mas.: Harvard University Press. Tomasello, Michael 2009 Why we cooperate. Cambridge, Mas.: MIT Press.

Personal pronouns across languages

97

Wittek, Angela, and Michael Tomasello 2005 Young children’s sensitivity to listener knowledge and perceptual context in choosing referring expressions. Applied Psycholinguistics 26: 541 – 558. Woodward, Amanda L. 2003 Infants’ developing understanding of the link between looker and object. Developmental Science 6 (3): 297 – 311. Yamamoto, Mutsumi 1999 Animacy and reference. Amsterdam/Philadelphia: Benjamins. ZAS-Papers in Linguistics, 48 2007 Bittner, Dagmar and Natalia Gagarina (eds.). Intersentential pronominal reference in child and adult language. www.zas.gwz-berlin.de/185.html

Comprehension and production of subject pronouns in child Dutch* Charlotte Koster, John Hoeks, and Petra Hendriks

1.

Introduction

One central property of human language is that, in general, adult speakers can understand whatever they produce and adult listeners can produce whatever they understand. This observed symmetry between production and comprehension might not, however, be an inherent property of grammar. It is well-known that children sometimes understand meanings that they do not yet correctly produce. Recent studies have also provided evidence that children sometimes correctly produce forms that they do not yet understand. Such delays in comprehension have been found in areas as diverse as object pronouns, inde¿nite noun phrases, prosody and contrastive stress, word order and structural attachment (see Hendriks and Koster 2010, for discussion). Many of these delays occur relatively late in acquisition, after age 5 or even later, resulting in a gap between correct production and correct comprehension that can span several years. Such asymmetries in language acquisition present a real challenge to rulebased systems of grammar. If children know a rule of grammar, they should be able to use this rule in production and comprehension alike. So, how can asymmetries between comprehension and production in child language be explained? Taking object pronouns as an example, previous accounts of children’s acquisition have attempted to explain comprehension errors as resulting from a lack of pragmatic knowledge necessary to distinguish exceptional cases from the standard pattern (Thornton and Wexler 1999), from insuf¿cient working memory capacity for the parser to compare alternative forms and meanings (Reinhart 2006), or as an experimental artifact due to an unbalanced context * This investigation was supported by a grant from the Netherlands Organization for Scienti¿c Research, NWO, awarded to Petra Hendriks (grant no. 277-70-005) for the VICI project “Asymmetries in Grammar”. The authors thank Ellis Wubs for her assistance in carrying out the experiment and an anonymous reviewer for useful comments.

100

Charlotte Koster, John Hoeks, and Petra Hendriks

(Conroy et al. 2009). One common denominator in all these accounts is that they fail to provide a detailed explanation of children’s successful production of object pronouns. The general solution of arguing that asymmetries arise as a result of dif¿culties at the interface with other linguistic modules and do not reÀect core properties of the grammar also makes it dif¿cult to explain why certain delays occur only in particular syntactic environments or only in particular languages. The pronoun interpretation problem, for example, does not occur in Italian (McKee 1992) and, in Spanish, it occurs only in Exceptional Case Marking constructions (Baauw and Cuetos 2003). The account of object pronoun acquisition given by Hendriks and Spenader (2005/6), based on the Asymmetric Grammar Hypothesis, is capable of predicting when delays will occur and when correct production and comprehension will be achieved. Delays are argued to be a direct result of the grammar itself being inherently asymmetric and may occur in either production or comprehension, depending on local constraint properties of the grammar. Adult language tends to be symmetric because adult speakers take into account their potential listeners in their use of grammar, and vice versa, and adapt their forms and meanings accordingly. The aim of the present study is to extend the Asymmetric Grammar Hypothesis to anaphoric subjects in discourse and to test the prediction that the acquisition of subject pronouns may be delayed because of asymmetries in the grammar. The asymmetric grammar predicts a comprehension delay with object pronouns, but a production delay with subject pronouns. Speakers should avoid using a subject pronoun when this pronoun will be resolved by a listener as referring to the incorrect antecedent, for example following a topic shift. Whether or not children produce unrecoverable subject pronouns is investigated in a discourse production task using picture storybooks. In addition, the same participants are given a comprehension task to investigate their understanding of pronouns at the end of discourses that either do or do not involve a topic shift. As listeners, will they correctly interpret the topic shift marking in the discourses? This study also investigates the role of working memory in discourse production and comprehension by including a working memory task.

2.

Asymmetric Grammar Hypothesis

According to the Asymmetric Grammar Hypothesis, asymmetries in acquisition are the result of inherent properties of the grammar. Although not compatible with rule-based systems of grammar, this explanation straightforwardly

Comprehension and production of subject pronouns in child Dutch

101

follows from constraint-based systems such as Optimality Theory (e.g. Prince and Smolensky 2004). In Optimality Theory, production is viewed as a process of optimization starting with a particular input meaning and yielding the optimal output form to express that meaning. Comprehension proceeds in the opposite direction and starts with a particular input form to yield the optimal output meaning for that form. The grammar is made up of constraints which are potentially conÀicting. Resolution of conÀicts is achieved by ranking constraints in a particular order of strength and permitting a stronger constraint to have priority over a weaker one. The form or meaning that best satis¿es the set of constraints is considered the optimal form or meaning and hence is realized in the language. Because constraints are output-oriented and compare competing output candidates without necessarily considering the input, constraints can be direction-sensitive. Consequently, different constraints may be effective in production and comprehension (Smolensky 1996). For example, a constraint preferring reÀexive forms to pronoun forms irrespective of their meanings will be effective in production only. In comprehension, this constraint does not have any effect because comprehension does not involve choosing between candidate forms (the form is already given as produced by the speaker), but only between candidate meanings. Crucially, even with the constraints properly ranked, asymmetries may still arise between production and comprehension. In order to achieve the symmetry between production and comprehension observed in adult language, a person must consider both the speaker as a listener and the listener as a speaker. This is formally modeled as bidirectional optimization (Blutner 2000). Bidirectional optimization combines the two directions of unidirectional optimization: From meaning to form, and from form to meaning. Consequently, in comprehension, a listener may need to select a different meaning if the initially selected meaning would have been better expressed by a speaker with another form. In production, a speaker may need to select a different form if the initially selected form would be interpreted by a listener differently than the speaker had intended. In the framework of bidirectional Optimality Theory, language acquisition is two-fold, including both constraint ranking and bidirectional optimization. First, a very young child must discover the proper constraint ranking involved in a particular linguistic phenomenon. For example, Cannizzaro (2011) investigated subject-object word order in young Dutch children and argues that their errors in comprehension but not production may be the result of constraint mis-ranking. Even when constraints of a particular language phenomenon are properly ranked, production and comprehension can still be asymmetric in the child’s language. The second step towards adult symmetry involves achieving bidirec-

102

Charlotte Koster, John Hoeks, and Petra Hendriks

tional optimization over both form and meaning. This is not a general cognitive skill that is achieved at a speci¿c moment in development and then applicable to all aspects of language. Rather, it is dependent upon the constraints involved in a particular language phenomenon and the amount of exposure the child has to this phenomenon. For example, regarding comprehension of object pronouns, Hendriks, van Rijn and Valkenier (2007) demonstrated, using computational simulations, that this transition from unidirectional to bidirectional optimality can be achieved through suf¿cient exposure to the relevant forms. For further discussion of language acquisition in Optimality Theory, see Fikkert and De Hoop (2009). In this paper, the focus is on this second step of acquisition in which children are assumed to have acquired the correct constraint ranking but are not yet adult-like because they are not capable of bidirectional optimization. The acquisition of object pronouns illustrates this second step. Hendriks and Spenader (2005/6) argue that children’s original unidirectional and asymmetric performance on object pronouns follows from the interaction of two violable constraints. The ¿rst constraint expresses a preference for reÀexives to be bound by the local subject, and can be seen as a violable version of Principle A of Binding Theory. This constraint pertains to reÀexives only and is not sensitive to the direction of use, having the same effect on production and comprehension. A second, weaker constraint is direction-sensitive because it is formulated as a constraint on forms, and only has an effect on production. This constraint expresses a preference for referentially more economical forms over referentially less economical forms (see Burzio 1998; Gundel, Hedberg, and Zacharski 1993, for variants of this constraint hierarchy). So, reÀexives are preferred over pronouns, and pronouns are preferred over full NPs. Because Principle A is the stronger of the two constraints, reÀexives are never interpreted as disjoint to the local subject in comprehension, and a disjoint meaning is never expressed by a reÀexive in production. Furthermore, the REFERENTIAL ECONOMY constraint has the effect that a reÀexive, rather than a pronoun, is used to express a coreferential meaning. These two constraints predict children’s correct production of both reÀexives and object pronouns. With respect to the comprehension of reÀexives and pronouns, however, the same two constraints predict ambiguity for object pronouns. Because PRINCIPLE A and REFERENTIAL ECONOMY do not distinguish between possible meanings for pronouns, both the coreferential and the disjoint meaning come out as optimal. Because no Principle B speci¿cally pertaining to pronouns is assumed, based solely on these two constraints, children are predicted to correctly interpret reÀexives but to display a guessing pattern with respect to the interpretation of object pronouns.

Comprehension and production of subject pronouns in child Dutch

103

Under this account, asymmetries between comprehension and production arise as a result of the particular constraints of the grammar. These asymmetries disappear when speakers and listeners optimize bidirectionally. A listener who encounters a sentence with an object pronoun must not only select the optimal meaning for this form, but should also check whether he, as a speaker, would have produced this pronoun when intending to express this meaning. If a listener selects a coreferential meaning for a potentially ambiguous pronoun, he will then discover that this coreferential meaning is best expressed using a reÀexive. Because this form (reÀexive) is different from the heard form (pronoun), the coreferential meaning cannot be the meaning the speaker intended to express. As a result, the listener should block the coreferential meaning for the object pronoun and select the disjoint meaning instead (see Hendriks and Spenader 2005/6, for a more formal account of bidirectional optimization applied to object pronouns). Thus, the pattern that object pronouns receive a disjoint meaning (i.e. Principle B of Binding Theory) is a derived effect resulting from bidirectional optimization, and is not represented directly in the constraints.

3.

Acquisition of subject pronouns

The acquisition of referring expressions and topic structure in discourse has been the focus of research for quite some time. In production, a pioneer in the ¿eld is Karmiloff-Smith (1981) who investigated storybook narratives by French and English children between 4 and 9 years old. She found that the youngest children use pronouns indiscriminately, with no consideration for their listeners. As the children grow older, pronominalizations become governed by a broader thematic organization of their narratives. The children have to ¿rst achieve a higher level of intrasentential discourse organization before the role of discourse topic can begin to develop. The production of pronouns in discourse has been extensively investigated by other researchers as well (for example, Hickmann and Hendriks 1999; Wigglesworth 1997). In comprehension, there is also a tradition of discourse pronoun research in children of various ages (for example, Tyler 1983; Song and Fischer 2005). How children interpret pronouns in discourse context has been shown to be affected by their developing perception of discourse structure and their interpretation of the prominence of possible antecedents. The present paper investigates both production and comprehension of discourse pronouns in the same group of children. The study has a narrower scope than general discourse structure in that it is speci¿cally oriented toward

104

Charlotte Koster, John Hoeks, and Petra Hendriks

a speaker’s choice of form in relation to a listener’s interpretation of this form. The Asymmetric Grammar Hypothesis is extended here beyond object pronouns and reÀexives at sentence level to subject pronouns in discourse. This is possible because in Optimality Theory, constraints express general statements about language: Their ranking and application is not construction-dependent. If REFERENTIAL ECONOMY constrains the possible forms of a language by selecting the correct object pronoun form in potential binding environments within sentences, this constraint is expected to be effective in other areas of the language as well, for example with respect to subject pronouns at the discourse level. If a speaker wishes to select the optimal form for referring to a previously mentioned referent in subject position, based on the REFERENTIAL ECONOMY constraint alone the speaker should prefer reÀexives to pronouns and pronouns to full NPs. But because the stronger constraint, PRINCIPLE A, also applies in this case, a reÀexive is actually a suboptimal form. A reÀexive in subject position cannot be bound by the local subject. As a result, a speaker will resist using a reÀexive and will prefer using a pronoun over a full NP in subject position (cf. Hendriks et al. 2008). In discourse, an additional constraint is needed to explain the reference of subject pronouns. If the speaker intends the subject pronoun to refer to the most salient referent in the discourse, that is, the current discourse topic, it will be interpreted correctly by the listener. But if the pronoun does not refer to the most salient referent, its meaning will be unrecoverable for the listener. To explain why listeners are selective in their interpretation of pronouns in discourse, a force must be present in the language that restricts the interpretation of subject pronouns to the most salient entity in the immediate linguistic discourse, that is, to the discourse topic. This requirement has been proposed in the literature in several guises, for example as the pronoun rule in Centering Theory (Grosz, Joshi, and Weinstein 1995). In this paper, we implement this requirement as a constraint of grammar and refer to it as PROTOP: “Pronouns refer to the discourse topic” (cf. Beaver 2004; Hendriks et al. 2008; see also Hendriks et al. 2010). As listeners will preferably interpret a pronoun as referring to the discourse topic, bidirectionally optimizing speakers will take the listener’s perspective into account and use a referentially more expensive, but recoverable full NP when they need to refer to a non-topic referent. Similarly, as speakers will preferably use a pronoun, bidirectionally optimizing listeners will interpret the speaker’s choice of a suboptimal form such as a de¿nite NP as signaling reference to a non-topic. The Asymmetric Grammar Hypothesis makes predictions in relation to children and their production and comprehension of subject pronouns and topics in discourse. Since bidirectional optimization combines the speaker’s direction

Comprehension and production of subject pronouns in child Dutch

105

of optimization with the listener’s, it is more complex than unidirectional optimization. Achieving bidirectional optimization may be dependent on cognitive resources such as processing speed or working memory. In the latter case, the size of a child’s working memory may be linked to whether the child is able to apply bidirectional optimization or not. If children apply the constraints REFERENTIAL ECONOMY and PROTOP unidirectionally, the following predictions about production and comprehension can be made (see Hendriks et al. 2008; and Wubs et al. 2009, for a more formal account). In production, a unidirectionally optimizing speaker will use referentially more economical subject pronoun forms for previously mentioned referents, regardless of their current discourse topic status. Speci¿cally, children are predicted to overproduce subject pronouns. In comprehension, a unidirectionally optimizing listener will fail to interpret a full NP as signaling reference to an entity which is not at that moment the discourse topic. For this listener, a full NP can equally well refer to the discourse topic as to a non-topic, because neither of these meanings violates the two constraints. Whereas the constraints of the grammar predict that children correctly produce object pronouns, the same grammar predicts children to make errors when producing subject pronouns. So, the Asymmetric Grammar Hypothesis not only predicts the existence of asymmetries, but also the direction of these asymmetries. In the case of subject pronouns, the bidirectional adult grammar is characterized by symmetry: Adults consider the linguistic perspective of others and pair subject pronouns with discourse topics and full NPs with non-topics. In this study, whether or not children produce unrecoverable subject pronouns is investigated in a discourse production task using picture storybooks, each featuring two characters. The storybooks are structured in such a way that it encourages speakers to switch the topic twice. The ¿rst character starts out as the most likely initial topic of the speaker’s discourse, but halfway through the story the second character should become the new discourse topic. The critical question is how speakers re-introduce the ¿rst character at the end of the storybook. Will they use a full NP or will they use a pronoun? Crucially, since until that moment, the second character is the current discourse topic, a pronoun will be interpreted incorrectly by a listener as continuing to refer to this second character. In addition to the production task, participants are given a comprehension task based on recorded discourses that either do or do not involve a topic shift. The topic shift is marked by a full NP in subject position. The experimental question here is whether listeners will be able to use this linguistic marking of topic shift to interpret a potentially ambiguous pronoun at the end of the discourse.

106

Charlotte Koster, John Hoeks, and Petra Hendriks

In order to investigate the role of working memory capacity in relation to the participants’ success or failure to optimize bidirectionally, a simple wordrepetition memory task is also administered. The experimental question here is whether there is a relation between working memory and success/failure to bidirectionally optimize, that is, to consider the linguistic needs or intentions of a conversational partner. Summed up, the predictions of the Asymmetric Grammar Hypothesis tested in this study are: 1) Children prefer to produce subject pronouns over full NPs, even when referring to non-topics. 2) Children do not interpret full NPs as a topic shift signal and may therefore interpret a subsequent subject pronoun incorrectly. 3) Adult-like pronoun use results from bidirectional optimization, which in children is related to a larger working memory capacity.

4.

Method

4.1.

Participants

This study included 31 children (15 girls, 16 boys), all of whom were typically developing native speakers of Dutch, attending kindergarten classes in a public school (mean age = 5;6 years, range = 4;3 – 6;5). Two additional children were not included in the analyses because they produced almost no full utterances. The study also included an adult control group of 23 native Dutch speakers (11 women, 12 men, mean age = 24;7 years, range = 20;7 – 30;9). 4.2.

Materials production task

While looking at a storybook with one picture per page and six pictures per story, participants described what was happening in the picture story. The storybook format, with individual pictures, was chosen over ¿lms. With the presentation and description of one picture at a time, the child is encouraged to describe each scene in order. Also the overall memory load of the child’s task is lighter than if the child would ¿rst see a complete cartoon and later describe it, or if the child’s description had to match a cartoon’s presentation tempo. In this study, each participant saw the same four storybooks, all with the same internal structure designed to elicit topic shifts from the storyteller. One of the storybooks is shown in Figure 1.

Comprehension and production of subject pronouns in child Dutch

107

Figure 1. An example of a picture storybook in the production task.

The ¿rst and second pictures show the ¿rst character only. A speaker has to decide how to introduce this character and how to continue referring to it. Because the ¿rst character is the only possible referent at this point, it is the most likely initial topic of the discourse. In the third picture, the second character enters the story. In the next two pictures, this character is shown performing an action while the ¿rst character is passive. Again, a speaker must decide how to introduce the second character and how to continue referring to it. At this point in the story, the second character has become highly prominent and the speaker is likely to initiate a topic shift and establish the second character as the new topic. The ¿nal picture of the story only shows the ¿rst character. As a consequence, a speaker cannot maintain the second character as the topic and is expected to initiate a topic shift again, switching back to the original topic, the ¿rst character. Although much can be said about the introduction and maintenance of referents and topics in discourse, this study focuses on the data with respect to topic shift only. In particular, it concentrates on the second shift from the new topic (second character) back to the previous topic (¿rst character), which should occur at the end of the storytelling. A speaker who takes into account the listener’s perspective will use a full NP. On the other hand, a speaker who does not, or cannot, take into account the listener’s perspective will simply use the more economical, but unrecoverable pronominal form.

108

Charlotte Koster, John Hoeks, and Petra Hendriks

4.3.

Materials comprehension task

In the comprehension task, all participants listened to the same pre-recorded stories about two characters of the same gender. The ¿nal sentence of each story contained a potentially ambiguous subject pronoun, matching the gender of the two characters. At the end of each story, participants were asked a question about which of the two characters had done something. In total, there were eight stories composed of six sentences each. The structure of the stories differed: Four included a topic shift halfway through the story and the other four did not. The body of the stories with topic shift was designed in such a way that they were parallel in structure to the production stories (all designed to elicit a topic shift). An example of a comprehension story with topic shift is given in Figure 2. 1

Een schoonmaakster wil de eendjes gaan voeren. ‘A cleaning-lady wants to go feed the ducks.’

2

Ze haalt het oude brood uit de broodtrommel. ‘She gets the old bread out of the breadbox.’

3

Ze vraagt aan een juf om mee te gaan. ‘She asks a teacher(FEM) to go along.’

4

De juf scheurt de broodjes van de schoonmaakster in stukjes. ‘The teacher(FEM) tears the cleaning-lady’s bread into pieces.’

5

En dan geeft de juf het brood van de schoonmaakster aan de eendjes. ‘And then the teacher(FEM) gives the cleaning-lady’s bread to the ducks.’

6

Ze vindt eendjes hele lieve diertjes. ‘She thinks ducks are very sweet animals.’

Comprehension question: Wie vindt eendjes hele lieve diertjes? ‘Who thinks ducks are very sweet animals?’ Figure 2. An example of a recorded comprehension story with topic shift.

In the topic shift condition, the ¿rst character is introduced in the ¿rst sentence with an inde¿nite subject NP and is referred to with a subject pronoun in the next two sentences. The second character is introduced in the third sentence by an inde¿nite (prepositional) object NP. This character is the actor in the fourth and ¿fth sentence and is referred to with a de¿nite subject NP, while the ¿rst character is referred to here with a de¿nite non-subject NP. The ¿nal sentence of the story contains a potentially ambiguous subject pronoun, which matches the two characters in gender.

Comprehension and production of subject pronouns in child Dutch

109

The comprehension task involves answering a question about the ¿nal sentence. To answer correctly, the listener must resolve the pronoun in that sentence. If participants assume pronouns to refer to the discourse topic, their answer will clarify who they think the topic is at the end of the story. If participants think the second character became the new topic halfway through the story, they will resolve the pronoun as this second character. Four stories without topic shift are also included to compare them with the topic shift stories. An example of a comprehension story without topic shift is given in Figure 3. 1

Een clown heeft net zijn eigen gezicht geschminkt. ‘A clown has just painted his own face.’

2

Hij wil wel eens iemand anders schminken. ‘He wants to paint someone else.’

3

Hij komt in de keuken een kok tegen. ‘He comes across a cook(MASC) in the kitchen.’

4

De clown besluit de kok te schminken. ‘The clown decides to paint the cook(MASC).’

5

En dan schminkt de clown een heel stoer gezicht bij de kok. ‘And then the clown paints a real tough face on the cook(MASC).’

6

Hij vindt dat het prachtig is geworden. ‘He thinks it turned out great.’

Comprehension question: Wie vindt het prachtig geworden? ‘Who thinks it turned out great?’ Figure 3. An example of a recorded comprehension story without topic shift.

In these stories without topic shift, the ¿rst character remains the actor throughout the whole story. This character is introduced with an inde¿nite subject NP in the ¿rst sentence and is referred to by a pronoun in the second and third sentence. In order to keep these stories similar to the stories with topic shift, the ¿rst character is mentioned by a de¿nite NP in the fourth and ¿fth sentence, although as a subject rather than an object. The second character is introduced in the third sentence by an inde¿nite (prepositional) object NP and is referred to in the fourth and ¿fth sentence with a de¿nite NP. There is no topic shift: The second character is never referred to with a pronoun and also never occurs in subject position. The ¿nal sentence of the story again contains a potentially ambiguous pronoun.

110

Charlotte Koster, John Hoeks, and Petra Hendriks

To answer the comprehension question correctly, the participant should again use the structure of the preceding discourse to determine the topic referred to by the pronoun in the last sentence. If the participant considers the ¿rst character to be the topic throughout the whole story, the participant will resolve the pronoun as this ¿rst character. 4.4.

Materials memory task

The Auditory Memory Test, a subpart of the Schlichting Test for Language Production (Schlichting et al. 1995a), was also included in the test battery. This task includes word lists of increasing length that are read aloud by the tester and repeated by the participant. The words are mainly one-syllable in length, with a CVC-structure. All the words originate from the Lexilijst, a vocabulary test for 1;9 – 2;3 year old children (Schlichting et al. 1995b). 4.5.

Procedure

Each child was tested individually in a quiet room at school. The session took roughly 20 minutes, with the production task preceding the comprehension task. Between these two language tasks, the memory task was administered. Two testers were present for testing the child. One tester sat across from the child and turned the storybook pages (production) or listened to the story recordings with the child (comprehension). The second tester noted responses to the ¿nal pictures during the production task and, during the comprehension task, played the pre-recorded stories and noted the answers to the ¿nal questions. This person sat further away from the child, behind a computer screen. During the production task, it was made clear to the child that the second tester wanted to understand the story but could not see the storybook. The child’s job was to make sure that this second person understood what was happening in the stories. The ¿rst tester began the production task by showing the child an introductory page including all the storybook characters and asking the child to name them. This introductory page was used to check the child’s knowledge of the names of the story characters. Then the tester told a story based on a two-page storybook (one sentence per page) and asked the child to also tell a story, based on another two-page storybook. After the practice session, the child was once again reminded that the second tester could not see the storybook and wanted to know what was happening in the story. The child was then asked to describe four storybooks of six pictures each.

Comprehension and production of subject pronouns in child Dutch

111

The memory task was presented as a word repetition game. First, for practice, the tester read one word aloud and the child repeated it. Then the tester read aloud and the child repeated increasingly longer sets of word lists. The test ended when the child failed two lists in a row or refused to continue. The comprehension task included eight pre-recorded stories, four with topic shift and four without topic shift. These stories were presented via a computer, in two different orders. At the end of each story, the recording was stopped and the tester asked the child the question about the ambiguous pronoun in the last sentence. If necessary, the child was allowed to hear the story another time. At several moments during the testing session and upon conclusion of the whole session, the children were rewarded with colorful stickers. In general, the children gave the impression that they had no trouble understanding what was expected of them and that they enjoyed doing the tasks. Test sessions for adults were equivalent to the children’s sessions, with a few exceptions. Adults were also tested individually, but with only one tester present. They were warned that, later on, someone else would have to listen to their recordings and understand their stories. Their session lasted roughly 10-15 minutes. During the introduction to the production task, the adults did not have to name the ¿gures on an introductory page. The adults were explicitly requested to produce only one or two sentences per storybook page, since pilot testing showed that some adults tell long and detailed stories. The adults received no rewards. The total test sessions were recorded and later transcribed. All transcriptions were controlled for accuracy. As an additional check during the actual test sessions, a tester noted responses to target pictures, answers to comprehension questions and successfully repeated memory lists. 4.6.

Scoring and coding

In the production task, the focus was on how speakers refer back to a previously introduced topic after they have established a new topic. Derived from theoretical accounts of reference in discourse (Beaver 2004; Grosz, Joshi and Weinstein 1995), the following three rules were adopted for coding discourse topics: 1) the topic of an utterance is a referring expression that has been mentioned in the preceding utterance, 2) if the utterance contains one pronoun, this pronoun is the topic, and 3) if the utterance contains no pronouns or more than one pronoun, the topic is the subject of the preceding utterance. A topic shift was coded if the topic at a certain point in the discourse was different from the previous topic. Two coders (not the authors) independently scored each transcript for topic shifts and referential forms used to re-introduce

112

Charlotte Koster, John Hoeks, and Petra Hendriks

the ¿rst character. Their judgments with respect to topic shift agreed for 95.5%. Over all the topic shift productions, the judges agreed on the expression used to re-introduce the ¿rst character for 98.6%. In the cases of judgment differences or obvious errors, the authors made the ¿nal decisions as to how to code the data. The goal of the comprehension task was to investigate whether participants were sensitive to the difference between stories with and without topic shift. To investigate this sensitivity, answers to the question about the ambiguous pronoun were scored as to whether they referred to the ¿rst-mentioned character or the second-mentioned character in the story. A third category ‘other’ included any answers that did not ¿t into the ¿rst two response categories. Participants gave a variety of ‘other’ responses. For example, they sometimes mentioned a referent from a previous story or answered “both people” or “I don’t know”.

5.

Results1

5.1.

Results production task

Elicitation of a ¿rst topic shift halfway through the storybooks was quite successful. Children realized a topic shift (+TS) from the ¿rst character to the second character when talking about the third (or sometimes fourth) picture of the storybook 84% of the time (104 out of 124 stories) and adult speakers 98% of the time (90 out of 92 stories). Failure to realize a topic shift was caused by a participant either focusing too strongly on only one character, resulting in no shift, or alternating the two characters without establishing a topic. Ten children failed to realize a ¿rst topic shift one time, three children two times and one child three times. In the adult group, the two unrealized shift productions came from two different participants. Only in the productions that realize a ¿rst topic shift is it necessary for a speaker to re-introduce the ¿rst character with an NP and, therefore, further analyses include only these productions. The score of main concern was whether participants used a full NP or a pronoun when referring back to the ¿rst character in describing the ¿nal picture. For example, in the pirate story (see Figure 1), one child’s description of the last three pictures concluded with a re-introduction of the pirate using a full NP in the ¿nal utterance: de ridder heeft een visnet / hij pakt de bal d’ruit / en dan is de Piet Piraat d’r blij mee (‘the knight has a ¿sh net / he gets the ball out / and then the Pete-Pirate is 1. The results of the present analyses differ slightly from those reported in Wubs et al. 2009, due to the use of an improved scoring method.

Comprehension and production of subject pronouns in child Dutch

113

happy with it’). In contrast, another child concludes with a pronoun: dan gaat de ridder ‘m vangen / en hij heeft de bal in een net gevangen / nu heeft ie z’n bal weer terug (‘then the knight is going to get it / and he has got the ball in a net / now he has his ball back again’). To see whether children and adults differed in number of full NPs used to refer back to the previous topic, two t-tests were conducted: An independentsamples t-test on the basis of mean percentages full NPs per participant (t1-analysis), and a paired-samples t-test on the basis of mean percentages full NPs per item (t2-analysis). All percentages were normalized using an arcsine transformation. The difference between the two age groups was signi¿cant on both analyses (t1(52) = -8.5; p < .001, t2(3) = -4.7; p < .05). 95.7

Percent response types

100 80 62.4 60 40

full NP 37.6

20

pronoun 4.3

0

+TS, children

+TS, adults

Figure 4. Referring expression used in production task to refer to previous topic after topic shift (+TS).

Figure 4 gives the percentages, aggregated over participants, of full NPs and pronouns produced by children and adults when re-introducing the ¿rst character. Children used 37.6% full NPs (SE = 9%) and 62.4% pronouns (SE = 9%). Individually, eight children used pronouns exclusively, three used NPs exclusively and the other 20 children showed a mixed pattern. In contrast with the children, the adults used 95.7% full NPs (SE = 4%) and only 4.3% pronouns (SE = 4%). The adult pronoun responses were produced by different participants. 5.2.

Results comprehension task

Children and adults were also compared as to how they answered the question at the end of the two types of comprehension stories. The children’s and adults’ response percentages (aggregated over participants) for these two story types

114

Charlotte Koster, John Hoeks, and Petra Hendriks

are reported separately in Figure 5 below. For the topic shift stories, children answered 52.4%, 26.6% and 21%, respectively, with the ¿rst character, second character and ‘other’ responses. For the stories without topic, they answered 54%, 21% and 25%, respectively, with the ¿rst character, second character and ‘other’ responses. Individually, the children again showed a mixed pattern, with some children giving many correct answers, some giving mixed responses and others giving many incorrect responses. In contrast, the adult answers for the topic shift stories were 34.8%, 65.2% and 0% respectively, for the ¿rst character, second character and ‘other’ responses. For the stories without topic shift, the adults answers were 91.3%, 6.5% and 2.2% respectively, for the ¿rst character, second character and ‘other’ responses. Their patterns of responses were also mixed.

Percent response types

100

91.3

80 60

52.4

2nd character other

34.8

40

26.6 20

1st character

65.2

54

21

21

25 0

0

+TS, children -TS, children

+TS, adults

6.5

2.2

-TS, adults

Figure 5. Reference assigned to pronoun in comprehension task, in stories with (+TS) and without (-TS) topic shift.

Repeated measures ANOVAs were applied to the response percentages per participant (F1, averaged over items) and to percentages per item (F2, averaged over participants). These analyses included three factors: Response Type (1st character, 2nd character, and ‘other’ response), Story Type (with topic shift (+TS) and without topic shift (-TS)) and Age Group (children and adults). Response Type was considered a within-participants and within-items factor, Story Type was treated as within-participants and between-items, and Age Group as between-participants but within-items. As in the previous analyses, an arcsine transformation was used for all percentages. To guard against possible violations of the statistical assumption of sphericity, the Huyn-Feldt correction was used whenever the factor Type of Response was involved (Stevens 1992). We report the actual degrees of freedom that were used in the statistical test.

Comprehension and production of subject pronouns in child Dutch

115

The main effect of Response Type was signi¿cant (F1(1.8,104) = 73.1; p < .001; F2(2,12) = 45.5; p < .001), as were the two-way interactions of Response Type and Age Group (F1(1.8,104) = 12.8; p < .001; F2(1.3,12) = 6.1; p < .05) and of Story Type and Response Type (F1(2,104) = 38.4; p < .001; F2(2,12) = 17.7; p < .001). These effects, however, were quali¿ed by a signi¿cant three-way interaction of Response Type, Story Type and Age Group (F1(2,104) = 29.8; p < .001; F2(1.3,12) = 7.1; p < .05). Follow-up analyses investigating the nature of this three-way interaction showed that for children, there was no interaction between Story Type and Response Type (both F-values < 1). There was only a main effect of Response Type (F1(1.5,60) = 13.6; p < .001; F2(1.9,12) = 8.6; p < .01). Whether the children heard a story with or without topic shift, in either case they most often answered the comprehension question by naming the ¿rst character as the answer to the question (53%; SE = 9%), as opposed to choosing the other options (all p-values equal .06); the remaining responses were equally distributed over the second character (24%; SE = 8%) and the ‘other’ responses (23%; SE = 8%) (both p-values equal 1.00). In contrast, the adult data showed a highly signi¿cant interaction effect on Response Type and Story Type (F1(1.3,44) = 86.3; p < .001: F2(1.4,12) = 15.6; p < .01). Follow-up analyses showed that adults had a qualitatively different response pattern for stories with a topic shift compared to stories without a topic shift. Adults favored answering the question in topic shift stories with the second character (65.2%; SE = 9%) over the ¿rst character (34.8%; SE = 9%) (p1 < .01; p2 = 1.00) or the ‘other’ response (0%; SE = 0%) (p1 < .001; p2 < .05), and with the ¿rst character over the ‘other’ responses (p1 < .001; p2 = .10), although these differences were not always signi¿cant in the analysis by items. For stories without topic shift, the answer pattern was completely reversed: Here, adults preferred answering the question by mentioning the ¿rst character (91.3%; SE = 5%) over either the second character (6.5%; SE = 4%) (p1 < .001; p2 < .05) or the ‘other’ response (2.2%; SE = 3%) (p1 < .001; p2 < .005). The difference between using the second character versus the ‘other’ response was not signi¿cant (p-values > .30). As in the previous analyses, Bonferroni correction was applied to each set of these pairwise comparisons. 5.3.

Results memory task

Participants’ score on the memory task was the total number of word lists that they correctly repeated. For children, the mean score was 7.7 lists, with a range of 4 – 10 lists. For adults, the mean was 14 lists, with a range of 10 – 18 lists. Pearson correlation coef¿cients were calculated between the variables Memory

116

Charlotte Koster, John Hoeks, and Petra Hendriks

Score and available measures of language comprehension and production. Age was also included in the correlations to track its effects in relation to memory and language development (all correlations were subjected to a two-sided test of signi¿cance). For adults, there were no signi¿cant correlations between Age and Memory Score (r = í.17; p = .44), between Age and language scores (absolute r-values < .19; p-values > .39), or between Memory Score and language scores (absolute r-values < .25; p-values > .24). Signi¿cant correlations were found in the children’s group: The correlation between Age and Memory Score was marginally signi¿cant (r = +.33; p = .07), indicating that there was a trend for memory scores to increase with increasing age. Age as well as Memory Score was signi¿cantly correlated with a number of language measures. In production, children with higher memory scores were more prone to use a full NP instead of a pronoun when re-introducing the ¿rst character at the end of their storytelling (r = +.47; p < .01). Age predicted production performance somewhat less well (r = +.30; p = .10). In comprehension, correlations were found between children’s adult-like answers to the ¿nal question and their Memory Score. In recordings with a topic shift, children with higher memory scores were more likely to give the correct, second referent answer (r = +.37; p < .05). Memory score was not signi¿cantly correlated with the tendency to give the incorrect ¿rst referent answer (r = +.29; p = .12). However, the relation between Memory Score and the likelihood to give ‘other’ responses was negative, and highly signi¿cant (r = í.56; p < .001). This pattern of correlations was approximately the same for the relation between Age and measures of pronoun comprehension: Age and correct second referent answer were strongly correlated (r = +.53; p < .01), as were Age and ‘other’ response, which had an inverse relation (r = í.57; p < .001). Age and incorrect ¿rst referent answers were not correlated (r = +.14; p = .46). No other correlations were found between Memory Score, Age, and comprehension responses. There were also no signi¿cant correlations between production and comprehension measures in either of the two age groups (absolute r-values < .27; p-values > .14).

6.

Discussion

In this study, children and adults were tested on both production and comprehension of stories in which they demonstrated their use and interpretation of anaphoric subjects in ongoing discourse. In accordance with the Asymmetric

Comprehension and production of subject pronouns in child Dutch

117

Grammar Hypothesis, three predictions were formulated about production, comprehension and their relation to working memory capacity. The prediction in relation to production was that children would prefer to produce subject pronouns over full NPs, even when referring to non-topics. The results of the production task support this prediction. In the present study, when adult speakers re-introduce the previous topic at the end of the story, they use a full NP. This choice of form is determined by bidirectional optimization on the basis of the constraints of the grammar. Children generally do not use full NPs. Instead, they tend to use referentially more economical subject pronoun forms for previously mentioned referents. Their use of pronouns does not discriminate between referents which are the current discourse topic and referents which are not. So, as speakers, they do not seem to optimize bidirectionally: They do not take their listener’s perspective into account. The result is that a listener may be unable to recover the intended meaning from the used form. The prediction about comprehension was that children would not interpret NPs as a topic shift signal and might therefore interpret a subsequent subject pronoun incorrectly. Because only the topic shift stories signaled a topic shift, the questions in the two story types should be answered differently. For an adult listener, the correct interpretation of a full NP requires bidirectional optimization. After listening to stories without a topic shift, the adults correctly answered the question by naming the ¿rst character. Also, they generally correctly answered the question in the topic shift stories by naming the second character. For these topic shift stories, however, the adults also gave an unexpected number of answers referring to the ¿rst character. In comparison, the children’s pronoun resolution did not seem to be affected at all by the presence or absence of topic shift marking. They answered the questions equivocally for both story types, with a general preference for the ¿rst character. Their remaining answers were divided between the second character and ‘other’ responses. As in production, we conclude that children do not seem to optimize bidirectionally: They fail to notice when the speaker signals a topic shift by using a full NP rather than a pronoun. Why do the adults, who do differentiate between story types, seem to be more successful answering questions about stories without topic shift? For experimental purposes, the structure of the two comprehension stories was made as parallel as possible to each other and to the production storybooks. This structure, however, might have made the topic shift stories more opaque for listeners. One aspect of discourse structure which may be playing a role here is referent prominence, involving factors such as frequency of mention, ¿rst mention, subject status and pronominalization (Song and Fischer 2005). In the present comprehension stories, the ¿rst character might be considered more

118

Charlotte Koster, John Hoeks, and Petra Hendriks

prominent, even in the topic shift stories, which could make it more dif¿cult for listeners to note the topic shift in this story type. This could explain why the adults, who did differentiate between the two stories, were nevertheless less successful with the topic shift stories. The children’s response pattern was different from the adults’ pattern in that the children reacted almost identically to both story types. Differentiation between stories with and without topic shift depends upon two things: 1) the assumption that pronouns refer to discourse topic, and 2) the ability to interpret the full NP form and non-subject status of the ¿rst character halfway through the story as an indication of topic shift. The conclusion of the present study is that the children’s dif¿culties with answering the question about the subject pronoun are rooted in their failure to note the topic shift markings, and not in their lack of knowledge that pronouns refer to prominent antecedents. This standpoint is supported by Song and Fischer (2005), who demonstrated how discourse prominence factors of possible antecedents already affected three year old children’s comprehension of subject pronouns. For the comprehension of object pronouns in a short discourse, Spenader, Smits and Hendriks (2009) presented evidence that when there is a clearly established discourse topic, children do know that object pronouns refer to the discourse topic. They are able to use this discourse structural information to interpret a pronoun that is otherwise ambiguous for them. A complete account of children’s development of grammar must also explain why the ability of taking into consideration the other person’s perspective in communication is not fully functional in children as old as 6. In the present study, the prediction pertaining to memory stated that adult-like subject pronoun use and interpretation in discourse is the result of the completion of bidirectional optimization, which may require a larger working memory capacity. This study did not include a Theory of Mind test since children of this age are expected to already possess ¿rst order Theory of Mind abilities (for an overview, see De Villiers 2007). Focusing on working memory capacity, the ¿rst step in investigating the prediction about memory is to determine the children’s working memory capacity. As expected, results showed that the children were not able to correctly repeat as many word lists as the adults. Within the group of children, there was variation in memory scores with a tendency for memory scores to increase with age. The second step is to investigate the prediction that a larger memory capacity is related to successful bidirectional optimization in production and comprehension. A signi¿cant correlation was found between children’s higher memory scores and adult-like production of NPs to re-introduce a previous topic in discourse. Recall the production task: At the beginning of their discourse,

Comprehension and production of subject pronouns in child Dutch

119

speakers introduce a ¿rst character. Later, they shift the topic to introduce a second character. At introduction, these two characters are new to the speaker and the listener. Towards the end, however, the children must shift from the second character back to the previously introduced ¿rst character. To successfully apply bidirectional optimization, a speaker will have to keep in mind what the listener already knows and what he needs to know in order to follow this second topic shift. The correlations showed that children with higher working memory scores are thus able to consider the perspective of their listeners. Like adult speakers, when re-introducing the ¿rst character, these children more often opt for a full NP and not a referentially more economical pronoun. Two signi¿cant correlations between memory and comprehension were also found, both in the stories with topic shifts. Children with higher memory scores were able to attend better to the speaker’s story structure and recognize the topic shift, thereby giving the correct, second character as the answer to the pronoun question. Children with lower memory scores seemed to lose track of the story more often and come up with ‘other’ answers that pertained to neither story character. For the stories with no topic shift, there were no signi¿cant correlations. In the present study, working memory scores are positively correlated with children’s more adult-like performance in both production and comprehension. The presumption is that as children have more memory capacity they will be able to successfully apply bidirectional optimization. On the basis of data presented here, however, it cannot be conclusively determined whether limited working memory capacity hinders children in moving from unidirectional to bidirectional optimization or in maintaining the necessary representation of the discourse. Summarizing, the evidence presented in this study suggests that the ability to take into account the listener as a speaker and the speaker as a listener is a complex skill that children still have to develop for different linguistic phenomena during the course of language acquisition. Such bidirectional optimization is, however, crucial for obtaining a symmetric system of language in which a speaker’s preference for economy is optimally balanced with a listener’s need for clarity. In production, because children often do not bidirectionally optimize, they are overly economical and often produce unrecoverable subject pronouns. Even though they have a speci¿c referent in mind for the pronoun, they do not consider whether the listener will be able to identify that referent. In comprehension, when children are not capable of bidirectional optimization, they do not recognize the speaker’s use of a full NP as signaling a topic shift and, therefore, do not correctly interpret a subsequent pronoun. The problem

120

Charlotte Koster, John Hoeks, and Petra Hendriks

is not a local misinterpretation of the pronoun, but rather an inability to understand the topic shift marking in the prior discourse. In conclusion, the results of this study con¿rm the predictions made by the Asymmetric Grammar Hypothesis. In this account, asymmetries between comprehension and production arise as a result of the particular constraints of the grammar. Based on the constraints involved, children may show selective delays in production and comprehension, whenever they are not yet able to take into account the opposite perspective in communication. These asymmetries disappear when children, as speakers and listeners, optimize bidirectionally.

8.

References

Baauw, Sergio, and Fernando Cuetos 2003 The interpretation of pronouns in Spanish language acquisition and breakdown: Evidence for the “Principle B Delay” as a non-unitary phenomenon. Language Acquisition 11: 219–275. Beaver, David I. 2004 The optimization of discourse anaphora. Linguistics and Philosophy 27: 3–56. Blutner, Reinhard 2000 Some aspects of optimality in natural language interpretation. Journal of Semantics 17: 189–216. Burzio, Luigi 1998 Anaphora and soft constraints. In Is the Best Good Enough? Optimality and Competition in Syntax, Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky (eds.), 93–113. Cambridge, MA: MIT Press. Cannizzaro, Gisi 2011 A comprehension delay of subject-object word order in preschoolers. In Production-Comprehension Asymmetries in Child Language-SOLA43, Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk (eds.), 145 – 169. Berlin: de Gruyter. Conroy, Anastasia, Eri Takahashi, Jeffrey Lidz, and Colin Phillips 2009 Equal treatment for all antecedents: How children succeed with Principle B. Linguistic Inquiry 40: 446–486. De Villiers, Jill G. 2007 The interface of language and Theory of Mind. Lingua 117: 1858–1878. Fikkert, Paula, and Helen de Hoop 2009 Language acquisition in Optimality Theory. Linguistics 47 (2): 311–357.

Comprehension and production of subject pronouns in child Dutch

121

Grosz, Barbara J., Avarind K. Joshi, and Scott Weinstein 1995 Centering: A framework for modeling the logical coherence of discourse. Computational Linguistics 21: 203–225. Gundel, Jeanette, Nancy Hedberg, and Ron Zacharski 1993 Cognitive status and the form of referring expressions in discourse. Language 69: 274–307. Hendriks, Petra, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks 2011 Adults’ on-line comprehension of object pronouns in discourse. In Production-Comprehension Asymmetries in Child Language-SOLA43, Angela Grimm, Anja Müller, Cornelia Hamann, and Esther Ruigendijk (eds.), 193 – 216. Berlin: de Gruyter. Hendriks, Petra, Christina Englert, Ellis Wubs, and John Hoeks 2008 Age differences in adults’ use of referring expressions. Journal of Logic, Language and Information 17: 443–466. Hendriks, Petra, and Charlotte Koster 2010 Production/comprehension asymmetries in language acquisition. Lingua 120: 1887–2094. Hendriks, Petra, and Jennifer Spenader 2005/6 When production precedes comprehension: An optimization approach to the acquisition of pronouns. Language Acquisition 13: 319–348. Hendriks, Petra, Hedderik van Rijn, and Bea Valkenier 2007 Learning to reason about speakers’ alternatives in sentence comprehension: A computational account. Lingua 117: 1879–1896. Hickmann, Maya, and Henriette Hendriks 1999 Cohesion and anaphora in children’s narratives: A comparison of English, French, German and Mandarin Chinese. Journal of Child Language 26: 291–452. Karmiloff-Smith, Annette 1981 The grammatical marking of thematic structure in the development of language production. In The Child’s Construction of Language, Werner Deutsch (ed.), 121–147. London: Academic Press. McKee, Cecile 1992 A comparison of pronouns and anaphors in Italian and English acquisition. Language Acquisition 2: 21–54. Prince, Alan, and Paul Smolensky 2004 Optimality Theory: Constraint Interaction in Generative Grammar. Malden, MA: Blackwell. Reinhart, Tanya 2006 Interface Strategies. Cambridge, MA: MIT Press. Schlichting, Liesbeth, Marcia C. M. van Eldik, Henk C. Lutje-Spelberg, Sjoeke van der Meulen, and Bieuwe F. van der Meulen 1995a Schlichting Test voor Taalproductie. Nijmegen: Berkhout. 1995b Lexilijst. Nijmegen: Berkhout.

122

Charlotte Koster, John Hoeks, and Petra Hendriks

Smolensky, Paul 1996 On the comprehension/production dilemma in child language. Linguistic Inquiry 27: 720–731. Song, Hyun-Joo, and Cynthia Fischer 2005 Who’s “she”? Discourse prominence inÀuences preschoolers’ comprehension of pronouns. Journal of Memory and Language 52: 29–57. Spenader, Jennifer, Erik-Jan Smits, and Petra Hendriks 2009 Coherent discourse solves the Pronoun Interpretation Problem. Journal of Child Language 36: 23–52. Stevens, James 1992 Applied Multivariate Statistics for the Social Sciences. Hillsdale, NJ: Lawrence Erlbaum. Thornton, Rosalind, and Kenneth Wexler 1999 Principle B, VP Ellipsis, and Interpretation in Child Grammar. Cambridge, MA: MIT Press. Tyler, Lorraine K. 1983 The development of discourse mapping processes: The on-line interpretation of anaphoric expressions. Cognition 13: 309–341. Wigglesworth, Gillian 1997 Children’s individual approaches to the organization of narrative. Journal of Child Language 24: 279–309. Wubs, Ellis, Petra Hendriks, John Hoeks, and Charlotte Koster 2009 Tell me a story!: Children’s capacity for topic shift. In Proceedings Generative Approaches to Language Acquisition North America 3 (GALANA 3), Jean Crawford, Koichi Otaki, and Masahiko Takahashi (eds.), 313 – 324. Somerville, MA: Cascadilla Press.

Asymmetries in the processing of object relatives in child Hebrew and Italian Irena Botwinik

1

Introduction

It has been reported in the literature that the production of relative clauses starts quite early, around 2;2 (McKee, McDaniel, and Snedeker 1998), but their comprehension is achieved much later, around 6;0, or even later, depending on a language (Sheldon 1974; Tavakolian 1981; Roth 1984, among others). To be more precise, this asymmetry pertains especially to object relatives, whose comprehension in many languages is reported to be at chance level until a certain age (e.g. Friedmann and Novogrodsky 2004, 2007 for Hebrew; Arosio, Adani, and Guasti 2006, 2009 for Italian), in contrast to the comprehension of subject relatives, which is reported to be quite good from the start (e.g. Hamburger and Crain 1982). Focusing on the comprehension of Hebrew and Italian object relatives, the basic assumption of the study is that the production/comprehension asymmetry attested by object relatives stems from the way their syntactic processing unfolds, a procedure which is crucially involved in the comprehension of object relatives, but not in their production. Among the theories of sentence processing, two distinct approaches exist. One approach distinguishes categorically between a conscious reanalysis, leading to a perceived Garden Path effect (1), and an automatic reanalysis, not leading to such effect (2) (e.g. Pritchett 1992). Under the other approach (e.g. Frazier and Rayner 1982) a rather broad conception of the term garden path is assumed, referring to any reanalysis arising from local ambiguity. (1)

Reanalysis resulting in a perceived garden path effect: Bill [warned Todd] would die Æ Bill warned [Todd would die]

(2)

Reanalysis not resulting in a perceived garden path effect: Bill [knew Todd] would die Æ Bill knew [Todd would die]

124

Irena Botwinik

In this work I claim that the effects of potentially facilitating hints used in comprehension tasks of object relatives in Hebrew and Italian provide empirical support for the categorical approach to the garden path phenomenon. I will show that the effect of the hints on the comprehension of object relatives depends on their status as genuine hints and on the kind of reanalysis they evoke, automatic or conscious. The paper is structured as follows: In section 2 I present the facts regarding comprehension of object relatives in Hebrew and Italian, as well as the background assumptions and main hypotheses of the study. Section 3 addresses the production/comprehension asymmetry of Hebrew object relatives and the effect of the syntactic clues on their comprehension. In section 4 the effect of the disambiguation clues in Italian object relatives is discussed, and the intriguing distinction regarding the effect of the agreement clue in Hebrew vs. Italian is clari¿ed. Section 5 summarizes the main points of the study and its conclusions. 2

The facts

2.1

Comprehension of Hebrew object relatives

Based on the comprehensive study of production and comprehension of RCs in Hebrew, at the age of 4;0, object relatives (3) are produced relatively well but their comprehension is at chance level (Günzberg, Shvimer, and Friedmann 2008; Friedmann, Belletti, and Rizzi 2009).1 (3)

tar’e li et ha-para she-ha-tarnegolet menasheket show me ACC the-cow that-the-chicken kisses ‘Show me the cow that the chicken kisses.’

Importantly, the addition of a resumptive pronoun (4), which is grammatical in Hebrew, does not facilitate children’s comprehension of object relatives, whereas distinct gender agreement on the embedded verb does (5):2

1. The production of object relatives in Hebrew, although quite well, differs from the production of subject relatives. According to Günzberg et al. 2008, and Friedmann et al. 2009, object relatives are often avoided, and a subject relative or some other paraphrase is used instead. 2. The addition of a resumptive pronoun has been shown to be highly effective for children with hearing impairment (e.g. Friedmann and Szterman 2006).

Asymmetries in the processing of object relatives in child Hebrew and Italian (4)

tar’e li et ha-para she-ha-tarnegolet menasheket ota show me ACC the-cow that-the-chicken kisses her/it-ACC ‘Show me the cow that the chicken kisses.’

(5)

tar’e li et ha-yeled she-ha-yalda menasheket show me ACC the-boy that-the-girl kisses-3Fem. ‘Show me the boy that the girl kisses.’

125

An additional kind of object relatives whose comprehension might be viewed as facilitated via distinct agreement are object relatives with an arbitrary null subject, proarb (6). The ij-features of proarb in Hebrew are 3rd person plural, and hence clearly distinct from those of the head of the relative (which is singular in Günzberg et al. 2008, Friedmann et al. 2009).3 (6)

tar’e li et ha-yeled she-menashkim oto show me ACC the-boy that-kiss-3.Pl. him ‘Show me the boy that someone kisses him.’

By the age of 6, comprehension of Hebrew object relatives becomes adult-like. 2.2

Comprehension of Italian object relatives

In Italian object relatives the embedded NP can occur in the post-verbal position, rendering the sentence ambiguous between an object relative and a subject relative (7). (7)

Il ragazzo che The boy that

guarda watches

il pagliaccio … the clown

a)

‘The boy that watches the clown …’ (SR)

b)

‘The boy that the clown watches … ’ (OR)

3. In Günzberg et al. (2008) distinctive agreement on the verb (5), and object relatives including a proarb (6) are treated as separate hints. Given that the ij-features of proarb in Hebrew are 3rd person plural, thus clearly distinct from the ij-features of the head of the relative, which is singular (‘the boy’ in (6)), I treat both as instances of distinctive agreement.

126

Irena Botwinik

Disambiguation towards an object relative can be done either via placing the subject in the preverbal position (8), or by using distinct number agreement on the embedded verb (9) (Arosio, Adani, and Guasti 2006, 2009; Guasti and Arosio 2007). (8)

Il ragazzo che il pagliaccio guarda … The boy that the clown watches ‘The boy that the clown watches ’

(9)

Il ragazzo che guardano i The boy that watch-Pl. the ‘The boy that the clowns watch’

pagliacci clowns

Following examination of 4 age-groups of children, 5;3, 7;3, 9;1, and 11;3, Arosio et al. (2006, 2009) report that while the preverbal position facilitates comprehension of object relatives already at the age of 5;0, the agreement clue is ineffective until 11;0, keeping the comprehension of object relatives at chance.4 Summarizing so far, some clues facilitate comprehension of object relatives (Hebrew), or their disambiguation (Italian), while others seem to have no such effect. Probably the most intriguing contrast is between the facilitating effect of the agreement clue in Hebrew as opposed to its ineffectiveness in Italian till the age of 11. With this in mind, let me lay out the main assumptions of the study. 2.3

Background assumptions and working hypotheses

Following Pritchett (1992), Chomsky (1995), Reinhart (1999), Siloni (2004), among others, I assume that production and comprehension are done by the same component of grammar, namely the Computational System (CS). The two processes, however, are not completely identical. Informally speaking, production and comprehension differ in that in production the speaker controls the numeration, which guides the derivation, while in comprehension the speaker has no control of the whole numeration. Rather, the incoming words are assigned some structure according to the principles guiding the syntactic

4. But see Adani (2006) who shows that the performance improves, depending on the task.

Asymmetries in the processing of object relatives in child Hebrew and Italian

127

processor.5 It is possible then, depending on the nature of the processing guideline(s), that there will be structures causing some parsing dif¿culties (e.g. the Garden Path phenomenon: The doctor told the patient that he was having trouble with to leave.). Given that production of object relatives in both languages at the discussed ages is relatively good, it is reasonable to hypothesize that the chance level comprehension of these relatives should be attributed to their processing. Since in Hebrew the chance level comprehension of object relatives is attested only by young children (i.e. at the age of 4;0), I take it to stem from the mechanism involved in the formation of relative clauses at this age. In the following section I will introduce the approach to the formation of early object relatives developed in Botwinik-Rotem (2008), which is crucially implicated in the comprehension of Hebrew object relatives and the effectiveness of the syntactic clues with these structures.

3

Comprehension of the Hebrew object relatives

3.1

Formation of relative CP via externalization

One of the important observations made in Hamburger and Crain’s (1982) ground-breaking study is that due to their semantic and syntactic complexity, the acquisition of relative clauses may involve several steps. Taking this as a general guideline, Botwinik-Rotem (2008) puts forward the following claim: The adult mechanisms used in the formation of the bound variable of a relative clause, namely operator movement or operator-binding, might be preceded by a pre-operator stage where children treat a relative CP as a simple modi¿er (e.g. AP), i.e. a constituent with a slot (x) which generates modi¿cation.6 The saturation of the slot (of the CP) is achieved via identi¿cation with the external semantic argument of the modi¿ed NP (R), like in simple modi¿cation of a nominal by an AP (Higginbotham 1985), (10a).

5. I abstract away from the non-syntactic factors, which may well inÀuence the speaker’s on-line processing. 6. I assume the standard analysis of relative clauses (Sells 1984).

128 (10)

Irena Botwinik a.

Modi¿cation by a relative CP at the pre-operator stage

NP (Ri) N'

CP x = i

N (Ri)

It is further proposed that the slot of the relative CP results from externalization of one of the verb’s ș-roles (canonically, the external one, see Williams 1994). Speci¿cally, a ș-role can assume the status of a slot, if its assignment is retained. Namely, it remains on the verb, percolates up to the CP, becoming a slot (x) of the relative CP (see Reinhart and Siloni 2005 for a similar proposal in the domain of syntactic reÀexives in languages like French) (10b). b.

Formation of a relative CP via externalization (object relative)

CP θ2  x (slot) C' θ2 C

IP θ2 I' θ2 I

VP θ2 NP

V' θ2 V θ1assigned θ2 – unassigned, percolates up to the CP, becoming a slot

Non-assignment of a ș-role to an argument, being a non-conventional grammatical procedure, has to be morphologically marked, either on the verb itself or via an inÀectional element (e.g. se in French), which is closely related to the verb (Reinhart and Siloni 2005). In principle, the verb in relative clauses does not have such marking, indicating that relative clauses in adult grammar are not derived via externalization. In this relation, Botwinik-Rotem (2008) proposes that children at the discussed developmental stage are not fully familiar with the precise nature of the marking, considering the complementizer (e.g. she-, ‘that’) or the resumptive pronoun as markers of externalization. As the result, externalization would be allowed in the formation of children’s relative clauses, conceived as follows: In

Asymmetries in the processing of object relatives in child Hebrew and Italian

129

relative clauses with a gap externalization is marked by the complementizer; in relative clauses with a pronoun externalization is marked by the complementizer and resumptive pronoun. With this in mind, let us turn to the comprehension of object relatives. Recall that since the production of these relatives at the discussed age is quite good, it is reasonable to attribute their chance level comprehension as deriving from some aspect in their syntactic processing. In what follows I ¿rst present the processing model assumed here, address brieÀy the processing of Hebrew subject relatives, underlying their good comprehension, and then examine the processing of Hebrew object relatives, specifying the step in processing which underlies their chance level comprehension, and the (in)effectiveness of the relevant syntactic clues. 3.2

The processing of Hebrew object relatives

3.2.1 The model (Pritchett 1992) I assume the processing model developed in Pritchett (1992), which is based on an extensive study of the garden path phenomena. In what follows I present a simpli¿ed version of Pritchett’s model, adapted for the present purposes. The incoming words entering the processor are put in the store. The structure of the sentence is built upon the arrival of the predicate (i.e. the verb). The processing is fully automatic, with no look-ahead. Processing guideline: The processor attempts to satisfy the predicateargument relations as soon as possible; at any given moment of the processing the processor attempts to link the ș-roles (or slots) to the arguments, and to incorporate all the arguments in the store into the structure. Since the processor is part of the computational system, it is sensitive to syntactic entities such as Case, agreement, etc., i.e. these entities can inÀuence the processing, serving as hints. Reanalysis: In most cases, there is only one way to satisfy the processing guideline (i.e. the predicate-argument relations) at any processing step, leading sometimes to a wrong analysis, which requires a reanalysis. Some reanalyses are possible for the automatic processor (e.g. Bart knows the clown is crazy.), creating no (perceived) garden path effect, while others are not (e.g. Lisa warned her friends wouldn’t help.), resulting in the (perceived) garden path effect, as they involve a conscious reanalysis.7 The processing of a canonical garden path sentence is illustrated in (11). 7. See Pritchett (1992) for the de¿nitions of a ‘costly’ reanalysis, i.e. a reanalysis which cannot be done by the automatic processor.

130

Irena Botwinik

(11)

The boat Àoated down the river sank. (garden path) Processing steps: Store the boat Àoated – an ambiguous verbal form: active – ș1//passive – ș2

The processor will take the active theta-grid as it fully satis¿es the processing guideline – the ș-role of the verb is assigned, and the NP in the store, the boat, is assigned a ș-role. (Taking the passive option, namely interpreting the sentence as The boated that is Àoated down the river… will result in assigning the ș-role to the null operator in the relative clause, but the argument in the store, the boat, will still remain without a ș-role.) Structure

IP NP The boat

I' I

VP V' V floated θ1

down the river – is incorporated in the structure as a locative adjunct sank – ș1 At this point a reanalysis is required, because there is no argument for the ș-role of the main verb. This reanalysis results in a perceived garden path effect, as it cannot be done automatically by the processor (see note 7). Two equal processing options: In some sentences, at some point of processing, there actually exist two processing options which are equal in terms of predicate-argument relations. For instance, in one option all ș-roles are linked to arguments, but there is an argument in the store without a ș-role, whereas in the other processing option, all the arguments that were in the store are linked to ș-roles, but the verb has still one ș-role unlinked. Since both options satisfy the processing guideline to the same extent, some processors take one and some take the other. Let us see an example.

Asymmetries in the processing of object relatives in child Hebrew and Italian (12)

131

The boy just found disappeared again. (partial garden path) Processing steps: Store the boy just found – an ambiguous verbal form: active – ș1, ș2//passive – ș2

At this point in the processing, the processor can take either option, because each satis¿es the processing guideline to the same extent. By taking the active option, ș1 is assigned to the NP in the store, the boy, but the verb still has ș2 (Structure I). If the processor takes the passive option, namely the sentence is interpreted as The boy that was found…, the only ș-role of the verb, ș2, would be assigned to the null operator in the relative clause, but the argument in the store, the boy, will still be without a ș-role (Structure II). Upon the arrival of the verb disappeared, the ¿rst option would lead to a costly reanalysis, resulting in the perceived garden path effect, whereas the second one will result in smooth processing. Structure I

IP

NP The boy θ1

I' I

VP V' V found θ2

Structure II

NP with no θ-role

NP The boy

CP Op

C' C

IP NP

I' I

VP V' V

NP found tθ2

132

Irena Botwinik

As I will show immediately, this kind of a situation, namely the existence of two processing options, is involved in the parsing of object relatives, assuming the derivational mechanism of externalization proposed here. Good comprehension of subject relatives Let us start with the processing of Hebrew subject relatives, which leads to good comprehension of these structures. The relevant example is repeated in (13), and its processing is schematized in (14), focusing mainly on the relative CP. (13)

tar’e li et ha-para she-menasheket et ha-tarnegolet show me ACC the-cow that-kisses ACC the-chicken ‘Show me the cow that kisses the chicken.’

(14)

Processing of (13) Store Structure tar’e li et ha-para (‘show me the cow’) the main clause is built she- (‘that’) menasheket (‘kisses’) The structure of the relative CP is being built from the VP upwards. If children assume the externalization derivation, one of the verb’s ș-roles has to be externalized. Since there is no unlinked argument in the store (note that ‘the cow’ is linked to the main verb), and since externalization targets canonically the external ș-role, ș1 is externalized, becoming the slot (x) which enables the merge of ha-para (‘the cow’) and its modi¿er, the relative CP. (et) ha-tarnegolet (‘the chicken’) is merged as the complement of the verb, linked to ș2.

The processing is over, resulting in an adult-like comprehension. Note that the processing of the subject relative proceeds smoothly, as it involves no local ambiguities of any kind. As we will see right away, this is not the case with regard to the processing of Hebrew object relatives.

Asymmetries in the processing of object relatives in child Hebrew and Italian

133

3.2.3 Chance level comprehension of object relatives As mentioned, unlike subject relatives, the comprehension of Hebrew object relatives around the age of 4;0 is performed at chance level (Günzberg et al. 2008). Let us see why this is so, namely why this result is claimed here to derive from the processing of these structures, assuming the externalization mechanism. The relevant example is repeated in (15), and its processing is schematized in (16). (15)

tar’e li et ha-para she-ha-tarnegolet menasheket show me ACC the-cow that-the-chicken kisses ‘Show me the cow that the chicken kisses.’

(16)

Processing of (15) Store Structure tar’e li et ha-para (‘show me the cow’) the main clause is built she- (‘that’) ha-tarnegolet (‘the chicken’) menasheket (‘kisses’) At this point, two equal analyses in terms of predicate-argument relations arise:

Analysis I: ș1 is externalized (becoming a slot) to generate modi¿cation, while ș2 is assigned to ‘the chicken’. This analysis, if taken by the processor, would lead to incorrect parsing, because the assignment of the ș-roles is reversed: ș1, the Agent of ‘kisses’, is interpreted as ‘the cow’. Analysis II: ș1 is assigned to the chicken, while ș2 is externalized. This analysis would lead to correct parsing. Since these two parsing analyses satisfy the processing guideline to the same extent, each can be chosen by the automatic processor randomly, resulting in the attested chance level rate of object relative comprehension. Before we proceed to the relative contribution of the clues to comprehension of Hebrew object relatives, a couple of remarks are in order. First, the two analyses are equally plausible as far as (children’s) grammar is concerned, because of the following conÀicting demands. Hebrew is an SVO language, and therefore standardly, ș1 of the verb would be associated with the preverbal NP, namely the NP which is already in the store when the verb is encountered will be analyzed as the subject of the clause. At the same time, however, externalizing ș1, rather than ș2, is canonical as well. Since it is impossible to adhere to both, any choice is equally imperfect as far as canonicity is concerned. Put differently, externalization, namely the formation of the

134

Irena Botwinik

modifying CP, and structure building, namely the linking of a ș-role to the argument in the store, are equally necessary, hence the choice between them is random. Second, the existence of the two equal analyses as depicted in (16) is crucially related to the pre-operator stage in children’s derivation of RCs. The two equal analyses arise because the formation of the modi¿er (CP) via externalization involves direct and immediate manipulation of one of the verb’s ș-roles, affecting the linking of the other one. There are no two equal analyses in adult processing of object relatives, because the marking of the CP as a modi¿er is done by the relative operator, rather than via externalization. Consequently, upon the arrival of the verb, its external role (ș1) is linked to the argument already in the store (‘the chicken’, in our example), correctly analyzed as the subject, which, as mentioned above, is preverbal in Hebrew, with the subsequent linking of ș2 to the argument bound by the operator (realized either as a trace or resumptive pronoun). 3.3

Processing of Hebrew object relatives with the clues

Recall, that the two clues which I examine here are the distinct agreement on the embedded verb, either in gender or in number, and the addition of a resumptive pronoun.8 The former facilitates comprehension of object relatives, while the latter has no such effect. Based on the account of chance level comprehension of Hebrew object relatives presented in the previous section, children’s performance stems from the availability of two processing options, which exist due to externalization. Now, in principle, both clues provide the information which should have eliminated the wrong option. Yet, only the information provided by the agreement clue seems to be taken into account in children’s processing of object relatives. Let us see why this is the case. 3.3.1 The effectiveness of the agreement clue The agreement features on the verb dictate the identity of the subject, specifying the argument to be associated with ș1, namely occupying spec-IP (‘the girl’ in (17a), proarb in (17b)). As a result, Analysis I, whereby this ș-role is 8. An additional clue used in Günzberg et al. (2008), which is not addressed here, involves comprehension of free object relatives. This clue, like the agreement clue, is shown to be effective. See Botwinik-Rotem (2008) where the effectiveness of this clue is attributed to the occurrence of the overt wh-operator, rendering externalization obsolete.

Asymmetries in the processing of object relatives in child Hebrew and Italian

135

externalized, is eliminated, and only Analysis II is pursued, leading to correct comprehension. (17)

a.

tar’e li et ha-yeled she-ha-yalda menasheket show me ACC the-boy that-the-girl kisses-3Fem. ‘Show me the boy that the girl kisses.’

b.

tar’e li et ha-yeled she-menashkim oto show me ACC the-boy that-kiss-3Pl. him ‘Show me the boy that that someone kisses.’

Agreement is a robust structural hint in Hebrew object relatives because of the following two aspects of Hebrew grammar: (i) 3rd person null subjects are not licensed in Hebrew, except proarb whose ij-features are 3Pl. This cancels the possibility to analyze the preverbal position as being occupied by pro. (ii) The phenomenon of post-verbal subjects is limited (in most registers of Hebrew) to unaccusative/passive verbs (18). Object relatives necessarily include transitive verbs, disallowing post-verbal subjects (19). This restriction cancels the possibility to analyze the preverbal position as being occupied by an expletive pro (proexpl), which is assumed to realize the subject position in sentences including post-verbal subjects (Rizzi 1982). (18)

(19)

a.

ha-kos nishbera/ nishbera ha-kos the-glass broke/ broke the-glass ‘The glass broke.’

b.

shney xadarim nuku/ two rooms [were] cleaned/[were] ‘Two rooms were cleaned.’

a.

dan nika shney xadarim Dan cleaned two rooms ‘Dan cleaned two rooms.’

b.

*nika dan cleaned Dan

shney two

nuku cleaned

shney xadarim two rooms

xadarim rooms

The relevance of the aforementioned Hebrew facts is best illustrated by structures like (17b), involving a null category in preverbal position. In

136

Irena Botwinik

Hebrew, the only grammatical parsing of such structures is the one with proarb in spec-IP.9 Importantly, the restrictions Hebrew imposes on null and post-verbal subjects are absent in Italian. As we will see shortly, this difference between the languages plays an important role in the processing of object relatives and the effectiveness of the clues in Hebrew vs. Italian. 3.3.2 The non-effectiveness of the resumptive pronoun Recall that in the processing of the object relative (repeated for convenience in (20)) there are two equally possible analyses (see (16)); either ș1 of the verb kisses is externalized to generate modi¿cation, while ș2 is assigned to the chicken, or ș1 is assigned to ‘the chicken’, and ș2 is externalized. Now, the resumptive pronoun (ota ‘her’ in (20)) enters the processor after one of these options has been (randomly) taken, resulting in grammatical sentences. Therefore in order to incorporate the pronoun, some reanalysis is needed. The question is what kind of reanalysis is necessary in each case and whether this reanalysis can be done automatically or not (see note 7). (20)

tar’e li et ha-para she-ha-tarnegolet menasheket ota show me ACC the-cow that-the-chicken kisses her/it ‘Show me the cow that the chicken kisses.’

It is reasonable to assume that only an automatic reanalysis, which is done by the syntactic processor, would have a meliorating effect on the comprehension of an object relative. Under present assumptions, the pronoun marks the externalized ș-role (see section 3.1). Since the pronoun is accusative, it should be associated with externalization of the internal ș-role, namely ș2. This would be the case when ș2 is externalized, but not when ș2 is linked to the chicken (incorrect parsing). In order to incorporate the pronoun in the latter case, the whole sentence should be completely, and non-trivially (namely, consciously) reanalyzed; the previously externalized ș-role (ș1) should be linked to the chicken, while ș2 should be unlinked from ‘the chicken’ and externalized, subsequently marked by the pronoun. Given that even for adults a conscious reanalysis is not always trivial, and judging by the fact that children’s performance remains at chance, we may conclude that children do not opt for the required conscious reanalysis. Rather, 9. See Friedmann et al. (2009) for a different account of the good comprehension of Hebrew object relatives with proarb.

Asymmetries in the processing of object relatives in child Hebrew and Italian

137

they simply ignore the pronoun. After all, the sentence is fully grammatical without it. As a result, the incorrect parsing is unaffected by the addition of the pronoun. To summarize, the agreement clue is effective for two interrelated reasons. First, it is a genuine hint, namely a hint whose occurrence completely eliminates local ambiguities. Second, it occurs at the processing step where its impact does not require any kind of reanalysis, automatic or conscious, namely before the assignment of the ș-roles. In contrast, the occurrence of the resumptive pronoun is ineffective, not because the information it provides is not suf¿cient. Rather, because it occurs after ș-role assignment, and in order to use it as a hint, namely to correct the initial analysis, a conscious reanalysis is required.

4

Disambiguation clues in Italian object relatives

Before we proceed to the discussion of the clues in Italian object relatives, the following should be made clear. Unlike in Hebrew, where the contribution of the clues was tested for comprehension of object relatives by small children (around 4;0), what is examined in Italian is the contribution of two syntactic clues to the disambiguation of object relatives across 4 groups of children, 5;3–11;3. The externalization mechanism, thus, is clearly irrelevant for the following discussion, as (most of) the children in Arosio et al. (2006) are beyond the pre-operator stage. This being so, I assume that the children tested for comprehension of Italian object relatives derive RCs like adults, via operator movement. Further, keeping in line with the assumptions in Arosio et al. (2006), I adopt The Minimal Chain Principle of de Vincenzi (1991), and assume that while processing a relative clause, an operator trace is posited in the subject position.10 As in Hebrew, one of the clues used in Italian is the distinct (number) agreement on the embedded verb. Surprisingly however, unlike in Hebrew, it proved to be ineffective in Italian. In what follows I will show that this difference regarding the effectiveness of agreement stems from a more general difference 10. Under this assumption, sentence processing is head-driven, rather than predicatedriven, as assumed in Pritchett (1992). Speci¿cally, under Pritchett’s approach, the relative clause is projected only when the embedded verb is encountered. This difference, though probably important in itself, is immaterial for the present discussion.

138

Irena Botwinik

in the grammar of the two languages, affecting the status of agreement as a genuine hint for the processor. But before this, let us consider brieÀy the effective clue, namely the preverbal position of the subject. 4.1

The effectiveness of the preverbal position

Given The Minimal Chain Principle (de Vincenzi 1991), while processing an object relative with a preverbal subject (21), a trace of the relative operator is posited in the subject position of the relative clause. (21)

Fammi vedere il cane che il cavallo insegue Let-me see the dog that the horse chases ‘Show me the dog that the horse is chasing.’

(Arosio et al. 2006)

Once the preverbal subject (il cavallo ‘the horse’) enters the processor, a reanalysis is required; the preverbal NP presents a clear hint to the processor as to its subjecthood – it can be nothing but the subject. The reanalysis consists of replacing the trace by the preverbal NP. Judging by the fact that this clue facilitates disambiguation of object relatives already at the age of 5;0 (Arosio et al. 2006), this reanalysis is done automatically by the processor, probably because it does not involve reassignment of the ș-roles. It is worth mentioning that under Pritchett’s (1992) assumptions, the processing of this kind of Italian object relative does not involve any reanalysis at all, on a par with the processing of object relatives with the agreement hint in Hebrew. Rather, the occurrence of the preverbal NP signals the processor in the clearest way possible that this is the subject of the relative clause, to be projected once the embedded verb is encountered (see note 10). Let us now turn to the somewhat surprising ineffectiveness of agreement in disambiguation of Italian object relatives. 4.2

The ineffectiveness of the agreement

As hinted in passing, although agreement in Italian provides suf¿cient information to disambiguate an object relative with a post-verbal subject (22), it does not constitute a hint for the processor. This is so because of the following two interrelated features of Italian grammar: (i) Italian is a genuine null-subject language, i.e. the phenomenon is not limited like in Hebrew. As the result, the preverbal position can, in principle, be occupied by pro.

Asymmetries in the processing of object relatives in child Hebrew and Italian

(ii)

(22)

139

Post-verbal subjects are widely attested in Italian with all kinds of verbs, including the transitive ones, in contrast to Hebrew, where this phenomenon is limited to unaccusative and passive verbs.11 As the result, the preverbal position can, in principle, be occupied also by proexpl (Rizzi 1982). Fammi vedere il cane che inseguono i cavalli Let-me see the dog that chase-Pl. the horses ‘Show me the dog that the horses are chasing.’

Given the above, agreement in Italian does not constitute a valid hint for the processor, because it does not resolve the local ambiguity as to the identity of the subject. Speci¿cally, in the parsing of an object relative with a postverbal subject (22), upon the arrival of the verb, its agreement features can be interpreted as those of a null referential subject (pro), or of the post-verbal subject, coindexed with the null expletive (proexpl) in subject position. Since the identity of the subject remains unclear, the processor is not forced to ‘do the right thing’, namely replace the trace in subject position by proexpl (though it can). Irreplacement of the trace will lead to incorrect linking of the ș-roles (the head of the relative will be interpreted as the Agent of ‘chasing’), requiring reanalysis upon the arrival of the post-verbal subject (‘the horses’). Judging by children’s chance level performance, this reanalysis is not performed till the age of 11;3. Reasonably, this is so because the required reanalysis is not automatic, but rather conscious, involving reassignment of the ș-roles. Note that if on the right track, the Italian facts support the account advanced here as to the ineffectiveness of the resumptive in Hebrew object relatives; if children as old as 9;0 do not perform a conscious reanalysis, clearly such a procedure will not be performed by children as young as 4;0. To sharpen the difference between the contribution of agreement in Hebrew and the lack thereof in Italian, consider the most revealing pair of Hebrew and Italian object relatives. (23a) is the Italian object relative with post-verbal subject, while (23b) is the Hebrew object relative with an arbitrary null subject (proarb). (23)

a.

Fammi vedere il cane che inseguono i cavalli Let-me see the dog that chase-Pl. the horses ‘Show me the dog that the horses are chasing.’

( = (22))

11. As reported in Guasti and Cardinaletti (2003), post-verbal subjects in Italian relative clauses are attested by children as young as 5;0.

140

Irena Botwinik b.

tar’e li et ha-yeled she-menashkim oto show me ACC the-boy that-kiss-3Pl. him ‘Show me the boy that someone kisses.’

( = (17b))

The strings of words in the Hebrew and Italian sentences in (23) are almost identical. In both sentences the relative complementizer is immediately followed by the verb, namely the preverbal, subject position is null. This state of affairs creates local ambiguity as to the identity of the null subject. In Hebrew the ambiguity is between t and proarb, while in Italian it is between t, pro and proexpl. Despite the plural morphology on the verbs in (23), which indicates that these are object relatives, rather than subject relatives, namely that t is not an option, the processing facts in the two languages differ crucially. This is so, because by eliminating the option that the subject position is occupied by t, the agreement clue in Hebrew fully establishes the identity of the subject as proarb; recall that neither pro nor proexpl are grammatical options in the Hebrew (23b). In Italian, however, the elimination of t does not have this effect; as far as Italian grammar is concerned, both pro and proexpl are grammatical options in the analysis of (23a) (only until a certain point of the processing, of course). The agreement in Italian, thus, is not a clue for the automatic processor, because it does not force the processor to pick a single processing option, leaving it to ‘wander’ in the (small) forest of the grammatical options presented by the relevant string of words in the Italian (23a).

5

Summary and conclusion

To summarize, as shown in Table 1, in Hebrew it is the agreement clue which facilitates comprehension of object relatives at the age of 4;0, because it clearly speci¿es the identity of the subject, eliminating the wrong processing option, which exists due to externalization. This is on a par with the effectiveness of the preverbal position in Italian (Table 2), which too, presents a clear hint as to the identity of the subject, occurring early enough to evoke an automatic reanalysis. In contrast, Italian agreement does not eliminate local ambiguity, and therefore if the wrong processing option is taken by the processor it can be undone only via a conscious reanalysis, after the ș-roles have been linked. This is similar to the ineffectiveness of the resumptive in Hebrew – it occurs too late, after ș-linking. Consequently, if the initial linking was inappropriate, undoing it would require a conscious reanalysis.

Asymmetries in the processing of object relatives in child Hebrew and Italian

141

Table 1. Facilitating comprehension of Hebrew object relatives at the age of 4;0 Formation Comprehension Object Rela- Externalization chance tive Effective Externalization good agreement Ineffective Externalization chance resumptive

Processing Two equal options, one of which leads to incorrect interpretation. Clear hint as to the identity of the subject, eliminating the wrong processing option. The pronoun occurs ‘too late’, after ș-assignment. In order to undo the wrong option, a conscious reanalysis is required, involving reassignment of ș-roles.

Table 2. Disambiguating Italian object relatives at the age of 5;3-11;3

Ineffective agreement

Effective preverbal position

Formation Comprehension Op-movement chance till 11;3

Op-movement good from 5;3

Processing Not a clear hint as to the identity of the subject (pro/proexpl). An automatic reanalysis is not evoked. Upon the arrival of the post-verbal NP, a conscious reanalysis is required, involving reassignment of ș-roles. Clear hint, evoking automatic reanalysis upon the arrival of the preverbal NP.

On the conceivable assumption that before 11;0 children cannot perform conscious reanalysis, the effect of the potentially facilitating hints on the production/comprehension asymmetry attested in object relatives derives straightforwardly from the automatic as opposed to conscious nature of the required reanalysis. This provides an empirical support for the categorical distinction between the two, rendering the approach to sentence processing which attempts to de¿ne the limits of the processor, predicting this distinction, superior to the one which does not.

142

Irena Botwinik

References Adani, Flavia 2006 Re-thinking the acquisition of Relative Clauses in Italian: Towards a grammatically-based account. Journal of Child Language 34: 1–25. Arosio, Fabrizio, Flavia Adani, and MariaTeresa Guasti 2006 Children’s processing of subject and object relatives in Italian. In Language Acquisition and Development, Adriana Belletti, Elisa Bennati, Christiano Chesi, and Ida Ferrari (eds.), 15–27. Cambridge Scholars Press. Arosio, Fabrizio, Flavia Adani, and Maria Teresa Guasti 2009 Grammatical features in the comprehension of Italian Relative Clauses by children. In Merging Features: Computation, Interpretation and Acquisition, Jose M. Brucart, Anna Gavarrò, and Jaume Sola (eds.), 138– 158. Oxford University Press. Botwinik-Rotem, Irena 2008 Accounting for the comprehension of Hebrew object relatives. In Language Acquisition and Development. Proceedings of GALA2007, A. Gavarró Algueró and Maria João Freitas (eds.), 101–111. Cambridge Scholars Press. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. Frazier, Lyn, and Keith Rayner 1982 Making and correcting errors during sentence comprehension: Eye movements in the analysis of structurally ambiguous sentences. Cognitive Psychology 14: 178–210. Friedmann Naama, Adriana Belletti, and Luigi Rizzi 2009 Relativized relatives: Types of intervention in the acquisition of A-bar dependencies. Lingua 119: 67–88. Friedmann, Naama, and Rama Novogrodsky 2004 The acquisition of relative clause comprehension in Hebrew: A study of SLI and normal development. Journal of Child Language 31: 661–681. Friedmann, Naama, and Rama Novogrodsky 2007 Is the movement de¿cit in syntactic SLI related to traces or to thematic role transfer? Brain and Language 101: 50–63. Friedmann, Naama, and Ronit Szterman 2006 Syntactic movement in orally-trained children with hearing impairment. Journal of Deaf Studies and Deaf Education 11: 56–75. Guasti, Maria Teresa, and Fabrizio Arosio 2007 Children’s processing of Italian relative clauses disambiguated through agreement on the embedded verb. Interdisciplinary Approaches to Relative Clauses workshop, 13th-15th September, 2007, Cambridge.

Asymmetries in the processing of object relatives in child Hebrew and Italian

143

Guasti, Maria Teresa, and Anna Cardinaletti 2003 Relative clause formation in Romance child’s production. Probus 15: 47–89. Günzberg-Kerbel, Noa, Lilach Shvimer, and Naama Friedmann 2008 Take the chicken that the cow kissed the chicken: Comprehension and production of various relative clauses by Hebrew speaking children. Language and Brain 7. Hamburger, Henry, and Stephen Crain 1982 Relative acquisition. Language Development Vol. 1 Stan Kuczaj (ed.). Higginbotham, James 1985 On Semantics. Linguistic Inquiry 16: 547–593. McKee, Cecile, Dana McDaniel, and Jay Snedeker 1998 Relative children say. Journal of Psycholinguistic Research 27: 573–596. Pritchett, Bradley 1992 Grammatical Competence and Parsing Performance. Chicago University Press. Reinhart, Tanya 1999 The processing cost of reference-set computation: Guess patterns in acquisition. Working Papers, Utrecht Institute of Linguistics, OTS. Reinhart, Tanya, and Tal Siloni 2005 Thematic arity operations and parametric variations. Linguistic Inquiry 36 (3): 389–436. Rizzi, Luigi 1982 Issues in Italian Syntax. Dordrecht: Foris. Roth, Froma 1984 Accelerating language learning in young children. Journal of Child Language 11: 89–107. Sells, Peter 1984 Syntax and Semantics of Resumptive Pronouns. Ph.D. diss., University of Massachusetts, Amherst. Sheldon, Amy 1974 The role of parallel function in the acquisition of relative clauses in English. Journal of Verbal Learning and Verbal Behavior 13: 272–281. Siloni, Tal 2004 Garden Path: Illicit Movement. Handout of the talk given at the IATL 20 conference, Bar-Ilan University. Tavakolian, Susan 1981 The conjoined-clause analysis of relative clauses. In Language Acquisition and Linguistic Theory, Susan Tavakolian (ed.), 167–187. Cambridge, MA: MIT Press.

144

Irena Botwinik

De Vincenzi, Marica 1991 Syntactic parsing strategies in Italian: The minimal chain principle. Dordrecht: Kluwer. Williams, Edwin 1994 Thematic Structure. Cambridge, MA: MIT Press.

A comprehension delay of subject-object word order in Dutch preschoolers* Gisi Cannizzaro

Introduction The traditional view of the relationship between comprehension and production is that receptive language precedes and exceeds language production (e.g. Ingram 1974; Fraser, Bellugi, and Brown 1963). It has been shown, however, that in some cases production precedes comprehension. In a study by Chapman and Miller (1975) English-speaking children who were tested on subject-object (S-O) word order assigned subject status to animate nouns and object status to inanimate nouns in comprehension, despite conÀicting word order cues and despite their adult-like performance in production of the same types of sentences. The explanation Chapman and Miller give for this surprising comprehension delay (replicated in McClellan, Yewchuk, and Holdgrafer 1986) is that preschoolers are often incapable of reversing the relevant linguistic rules necessary for a “common basis of linguistic knowledge in both modes” (Chapman and Miller 1975: 369). This, they claim, is why children successfully produced S-O word order but employed a semantic strategy based on animacy in interpretation. An alternate explanation for this asymmetry is put forth by Hendriks, de Hoop, and Lamers (2005) in which the comprehension delay for S-O word order is the result of an incorrect ranking of constraints in the children’s grammar; speci¿cally, an animacy constraint that applies only to sentence interpretation is incorrectly ranked above a word order constraint that applies to both sentence interpretation and production. In this case, young children’s use of animacy information in comprehension is not a strategy, but rather the

* This investigation was supported by a grant from the Netherlands Organisation for Scienti¿c Research, NWO, awarded to Petra Hendriks (grant no. 277-70-005) for the VICI project “Asymmetries in Grammar.” The author would like to thank Sabine van der Ham and Hanneke Loerts for their assistance and one anonymous reviewer for helpful suggestions.

146

Gisi Cannizzaro

reÀection of a grammatical constraint at work that is also present, though lowly ranked, in adult grammars. The two explanations for this asymmetry offer different views on the nature of grammar in general. According to Chapman and Miller, children must successfully reverse linguistic rules in order to achieve adult-like language. In contrast, Hendriks et al. believe that grammar consists of a set of constraints of two different categories: The side-effect of membership in one of these categories is that some constraints promote symmetry between comprehension and production while others promote asymmetry. In their view, it is the latter of these two types of constraints that young children rank too highly, which causes differences in performance on comprehension and production. This study attempts to gain more insight into these explanations of the comprehension delay by investigating the effect that animacy properties of subject and object have on sentence production and interpretation, not only in children but also in adults. In order to take a close look at interpretation of both age groups, on-line measurements were collected in the form of eye gaze data in addition to off-line responses. In the next section of this paper, the experiment and conclusions of Chapman and Miller (1975) are described in more detail followed by Hendriks et al.’s (2005) explanation of their results. Based on these accounts, predictions are made for Dutch children and adults for both off-line and on-line responses. Two eye-tracking experiments are subsequently described and results are reported. In a general discussion, the results of the experiments are evaluated given the proposed predictions. General conclusions are then drawn about the effects of animacy in the language of both preschoolers and adults.

Theoretical background Production precedes comprehension Children’s very ¿rst multi-word utterances already conform to the word order of their target language (e.g. Brown 1973). Even before they begin to combine words, children are sensitive to S-O word order in semantically reversible sentences (Hirsh-Pasek and Golinkoff 1996). It seems from this that comprehension precedes production for S-O word order. However, upon closer inspection, it seems that there is actually evidence of a comprehension delay: Bloom (1970) identi¿ed subject-verb-object structures in the speech of children aged 3 to 5, while other studies reported a failure to comprehend S-O relations in the

A comprehension delay of subject-object word order in Dutch preschoolers

147

active voice in children of the same age (Fraser et al. 1963; de Villiers and de Villiers 1973). By manipulating the animacy of the nouns in active, semantically reversible sentences in English, Chapman and Miller (1975) attempted to determine if word order alone, or other types of cues as well, guide the assignment of subject and object by preschoolers. Using an act-out task, Chapman and Miller tested ¿fteen children between the ages of 1;8 and 2;8 on the same sentences in two tasks. In a comprehension task, the children heard a sentence spoken by the experimenter and had to act it out with two small toys. In a production task, the child watched the experimenter perform an action using the same set of toys and the child was asked to describe the action. Four conditions were tested in which the animacy properties of subject and object were manipulated: animate-animate (the boy is hitting the girl), animate-inanimate (the boy is pushing the car), inanimate-animate (the car is pushing the boy), and inanimate-inanimate (the car is pushing the truck). Table 1. Mean percentages of correct responses in Chapman and Miller (1975). Sentence Type

Comprehension

Production

[+an +an]

66.5

83.7

[+an -an]

93.8

86.3

[-an +an]

50.1

89.3

[-an -an]

65.2

82.4

Total

68.9

85.4

Table 1 shows the overall mean percentages correct for each task in each of the four sentence types. In production, the average performance did not differ with regard to sentence type, with a range of between 82% and 89% accuracy. In contrast, the results of the comprehension task reveal a striking difference in performance on [+an -an] items (93.8%) and [-an +an] items (50.1%). According to Chapman and Miller, these results indicate that comprehension and production processes of preschoolers do not yet share the same linguistic competence, probably due to an inability of children to reliably reverse intellectual operations. They conclude from this that preschoolers use semantic strategies in order to identify which noun phrase is the subject and which is the object in sentences they hear, but not when ordering noun phrases in the sentences they produce.

148

Gisi Cannizzaro

Mis-ranked constraints Utilizing the framework of Optimality Theory (Prince and Smolensky 2004), Hendriks et al. (2005) demonstrate that the comprehension delay revealed by Chapman and Miller may be explained by a grammar that is comprised of a set of ranked constraints. In Optimality Theoretic Syntax/Semantics, production and comprehension can be seen as two different directions in which the same set of grammatical constraints are applied. Speci¿cally, sentences are interpreted and expressed by applying a set of ranked, potentially conÀicting constraints to a set of possible candidates. Crucial to understanding how such a grammar can explain asymmetries is differentiating faithfulness and markedness constraints. Faithfulness constraints restrict output in relation to the input and therefore uphold symmetry between comprehension and production, while markedness constraints restrict only the output (either applying only when choosing a form to use or only when assigning meaning) and therefore uphold asymmetry between production and comprehension. Smolensky (1996) offers an account of phonological production delays in which markedness constraints on form are too highly ranked. Following from this approach, Hendriks et al. suggest that a mis-ranking of a markedness constraint on meaning above a faithfulness constraint can explain the comprehension delay found in Chapman and Miller. They propose that preschoolers have an incorrect ranking of the following constraints, with PROMINENCE (a markedness constraint) ranked above PRECEDENCE (a faithfulness constraint). (1)

PROMINENCE: The subject outranks the object in prominence, e.g. animacy

(2)

PRECEDENCE: The subject precedes the object

This ranking causes a delay in comprehension because PROMINENCE is a markedness constraint restricting which meaning should be given to a form, and therefore does not help determine what form should be used to express certain meanings. This is demonstrated in Table 2 and 3, which represent a child’s grammar in which PROMINENCE is incorrectly ranked above PRECEDENCE. (An Optimality Theory table is read as follows: The order of constraints from left to right signi¿es a decrease in rank; a star signi¿es a constraint violation, with “!” marking a fatal violation for an output candidate; and the optimal output candidate is denoted with a pointing hand.)

A comprehension delay of subject-object word order in Dutch preschoolers

149

Table 2. Child’s production (from input meaning to output form). Input: {PUSH, car(-an), cow(+an)} S-O word order ) O-S word order

PROMINENCE (MARKEDNESS)

PRECEDENCE (FAITHFULNESS) *!

Table 3. Child’s comprehension (from input form to output meaning). Input: [NP1(car(-an)) V(push¿n) NP2(cow(+an))] )

S-O interpretation O-S interpretation

PROMINENCE (MARKEDNESS)

PRECEDENCE (FAITHFULNESS)

*! *

Hendriks et al. offer the explanation that the children tested by Chapman and Miller correctly produced S-O word order across all four sentence types because the higher ranked constraint PROMINENCE does not apply in production, so PRECEDENCE is the deciding constraint, and the “S-O word order” candidate wins. In comprehension, on the other hand, PROMINENCE does play a role in determining what meaning to apply to a form. In the case of a [-an +an] sentence, a S-O interpretation violates PROMINENCE, so an O-S interpretation is the winning candidate. This is in line with the worst performance seen on [-an +an] sentences in the Chapman and Miller study.1 Animacy effects in children and adults If Chapman and Miller are correct in that once a language user has mastered general cognitive process-reversal, production and comprehension will share the same linguistic competences, then one would expect that young children may employ language interpretation (or production) strategies that would disappear with age. This means that, using the present example of S-O word order, adults should be able to effectively reverse linguistic rules and should not need to rely on any animacy information for sentence interpretation. Conversely, under the view that a markedness constraint like PROMINENCE is present in the grammar, it would not be as surprising to ¿nd animacy effects in adult language. Although in the adult grammar of English-speakers a ranking 1. In further act out studies, similar but inconsistent animacy effects on comprehension were found-effects argued to be weakened by confounding event likelihood biases (McClellan et al. 1986; Chapman and Kohn 1978).

150

Gisi Cannizzaro

of [PRECEDENCE >> PROMINENCE] is assumed, this only indicates what the forms or meanings are that “win out,” evident in the ultimate word order adults use and the meanings they give to forms. Lamers and de Hoop (2005) have shown that even if PROMINENCE is present but lowly ranked, processing effects may be found between different types of sentences in which some violate PROMINENCE and some do not. The ¿rst question the study aims to answer is whether there is indeed a comprehension delay of S-O word order in Dutch preschoolers in an improved experimental design. Dutch and English are comparable in that S-O word order in simple NV¿n N Dutch sentences functions the same as in English.2 By using a picture-pair selection task, the unlimited interpretations possible in an act out task are restricted to only two. Also, the association between comprehension and production is more direct since the target picture in the picture selection task is the same picture used in production. Finally, the additional measurement of eye gaze data informs us about the processing of the sentences in real time. Second, if there is a comprehension delay, do the animacy properties of the noun phrases play a role in the interpretations that children have? Based on the model of Hendriks et al., it is expected that sentences in which the subject is inanimate and the object is animate [-an +an] will receive O-S interpretations most often – the condition in which Chapman and Miller found only 50% correct responses. While Chapman and Miller found a difference between the [+an -an] sentences (94%) and the other two types of sentences in which the subject and object are equal in animacy (67%), the Optimality Theoretic model predicts that all three of these sentence types should result in S-O interpretations. In the [+an -an] condition, the O-S interpretation violates PROMINENCE, making a S-O interpretation the clear winner; in the [+an +an] and [-an -an] conditions, PROMINENCE is violated for both S-O and O-S interpretations, allowing PRECEDENCE to determine that a S-O interpretation is also optimal. As far as production is concerned, a difference between sentence types was not found by Chapman and Miller and is not predicted by the Optimality Theoretic model since PRECEDENCE decides in every case that S-O word order is the best form to use. Eye movements during comprehension are expected to corroborate the off-line predictions: children should show more competition from the distractor in picture selection for the [-an +an] sentences than in the other three types of sentences. 2. While OVS word order is possible in Dutch in special cases of topicalization, when answering a question, or for emphasis/contrast, experimental (e.g. Kaan 1997) and corpus (Bouma 2008) studies show that sentences in which the object precedes the subject are less preferred and occur less often (in about 14% of the cases) than sentences in which the subject precedes the object.

A comprehension delay of subject-object word order in Dutch preschoolers

151

Third, how do adult controls perform on the same tasks with the same materials as the children? Chapman and Miller would predict that adults can successfully reverse the rule about how to correctly produce S-O word order, resulting in S-O interpretations for all sentence types. The Optimality Theoretic model also predicts that the adult ranking of [PRECEDENCE >> PROMINENCE] (which holds for Dutch as well) will result in S-O word order used across all sentence types for production and S-O interpretations across all sentence types for comprehension. Speci¿cally, since PRECEDENCE is more highly ranked and it is a faithfulness constraint, it requires that the subject precedes the object in both directions of optimization. Chapman and Miller make no predictions about the sentence processing of adults. However, it follows from their view (that preschoolers are only employing a temporary strategy that uses animacy information) that adults would not show evidence of using animacy to interpret sentences. In contrast, under the view that (i) an animacy constraint is lowly ranked in adult grammar and (ii) constraint violation patterns Àuctuate differently over time for different types sentences, it is expected that adults would show evidence of using animacy information during comprehension in the form of processing differences between the sentence types (Lamers and de Hoop 2005). Speci¿cally, two predictions can be made: (1) The sentences in which an S-O interpretation never violates PROMINENCE, [+an -an], is likely the easiest sentence type to process; and (2) the sentences in which a S-O interpretation violates PROMINENCE before the second noun phrase is encountered (sentences with inanimate subjects) could show slower processing in comparison to those sentences that do not show early violations of the constraint (sentences with animate subjects).

Experiment 1a: Adults Participants In the adult control study, 19 native Dutch-speaking adults were tested (9 male, aged 20-43). Participants were students or employees of the University of Groningen. All had normal or corrected to normal vision. Materials and design There were 16 experimental items, which consisted of 8 duwen and 8 trekken sentences (‘push’ and ‘pull’, respectively). In each of the sentences for each verb, the following nouns occurred once as subject and once as object: hond,

152

Gisi Cannizzaro

koe, auto, bus, eend, vis, boot, vliegtuig (‘dog’, ‘cow’, ‘car’, ‘bus’, ‘duck’, ‘¿sh’, ‘boat’, and ‘airplane’, respectively).3 The combinations of these verbs and nouns resulted in sentences with no event likelihood biases (see note 2). Each of the following conditions occurred twice for each verb: animate subject with animate object [+an +an], animate subject with inanimate object [+an -an], inanimate subject with animate object [-an +an], and inanimate subject with inanimate object [-an -an].4 In addition there were 16 ¿ller items and 6 practice items, which included no test nouns or verbs. Thus, there were 38 sentences in total, which were pre-recorded with neutral prosody by a female voice. There were 16 illustrations that depicted the action described by each experimental test sentence and 16 distractor pictures that depicted the reversed interpretation. For example, for the sentence the car pushes the cow there was a picture of a car pushing a cow and a cow pushing a car, as seen in Figure 1. In half of the ¿ller items and in none of the practice items, the action of distractor pictures depicted the reversed interpretation of the practice target sentence. The direction of action was the same for target and distractor pictures. Direction of action and target-picture location was balanced across sentence types. The size of each picture was 320 × 240 pixels for the comprehension task and 640 × 480 pixels for the production task. The experiment was conducted in two blocks (comprehension and production). Two lists were made for the adult participants with the same items but different orders. In the experimental lists, no sentence type occurred twice in a row and no adjacent sentences shared a noun. The order of the tasks and the lists used was balanced over participant gender. The experimental items (pre-recorded sentences and pictures) were sent to a Tobii T120 remote eye tracking monitor (with resolution 1024 × 768) by a computer running the software E-Prime (version 1.2.1.844; Schneider et al. 2002). External desktop speakers presented the auditory stimuli. The eye data was collected from the Tobii monitor (at a frame rate 120 Hz) by a second computer with Tobii Studio software (version 1.2.22, later 1.5.4). Calibration was run with Tobii Studio. E-Prime collected accuracy and reaction time (RT) 3. Since these sentences are also used with preschoolers, it should be noted that according to the N-CDI user manual for Dutch (Zink and Lejaegere 2002) each of these nouns is both understood and said by between 95% and 100% of Dutch children aged 30 months, and each of the verbs is both understood and said by between 75 and 79% of Dutch children of the same age. 4. In typology of argument properties (Van Valen and Wilkins 1996) an animate-inanimate distinction is made among living things, while non-living entities are distinguished as either non-motive or independently/dependently motive. For present purposes vehicles will be referred to as “inanimate.”

A comprehension delay of subject-object word order in Dutch preschoolers

153

for the comprehension items. Tobii Studio’s user camera recorded a video of the participant during testing, which included audio either (1) recorded via a normal desktop microphone during the production task or (2) received directly from the computer running E-Prime with pre-recorded test sentences during the comprehension task.

Figure 1. Comprehension item: The car pushes the cow.

Procedure Participants’ eyes were ¿rst calibrated at ¿ve points. For the comprehension task, participants were instructed to press a button on the left or right of the keyboard to indicate whether the picture on the left or the right of the screen matched the sentence they heard, and were given three chances to practice. Each picture pair was preceded by a one-second ¿xation cross. For the production task, participants were instructed to describe the picture that appeared on the screen in a short sentence, with three chances to practice and receive feedback if necessary. They were given the example that if they saw a picture of a panda kissing a lion, they should say “The panda is kissing the lion.” They were also instructed not to say the Dutch equivalent of the construction There is/are since usually there was a clear action occurring in the pictures. The experimenter controlled presentation of the following item after the participant had produced a sentence. In order to give the participants time between the sentence productions, a ¿xation point with duration of two seconds preceded each picture. Produced utterances were scored as S-O, S-O+, O-S, O-S+, or other, with a “+” indicating that either S-O or O-S word order was preserved, but the utterance was not completely identical to the target sentence for any of the following reasons: a preposed subject was used (The car, it pushes the cow); a prepositional phrase or adverb was included (The dog pulls the cow forward); a synonym for a noun or verb was used; gaan (‘to go’) was used as an auxiliary to express ongoing action; inde¿nite articles were used or articles were omitted; or a passive construction was used. Other refers to unscorable responses such

154

Gisi Cannizzaro

as, they are swimming together or an airplane in front of a boat. S-O and S-O+ sentences were scored as correct and O-S and O-S+ sentences were scored as incorrect. Results Accuracy For comprehension, of the 304 responses for test items made by all participants, 295 were correct. Mean accuracy based on participant performance was 97.04%. To determine if there was an effect of sentence type on accuracy in comprehension, in addition to an effect of task order, a one-way repeated measures analysis was run. First, an arcsine square-root transformation for binomial responses was applied to the proportions (aggregated over participant and condition) in the data matrix. The repeated measures ANOVA was then run with the arcsine transformed proportions, with Sentence Type (four levels) as within-subject factor and Task Order (Comprehension First versus Production First) as between-subject factor. The results show that accuracy on the comprehension task was not signi¿cantly affected by the type of sentence, (F(1.58, 26.85) = 2.47, p > 0.1). Furthermore, there was no signi¿cant interaction between Task Order and Sentence Type (F < 1). For production, 286 of the 304 responses (94%) from all participants were both scorable and correct. Thus, of the scorable sentences, the participants never produced O-S word order. Of the 286 correct responses, 44 utterances (18%) were scored as S-O+, with only 4 of these sentences produced as a passive. Since all of the scorable sentences had S-O word order, there was no variation between sentence types to be tested. Reaction time Only comprehension items that were answered correctly were included in the RT analysis. Boxplots based on participant performance showed no reason to exclude any data on the basis of extreme RT values. Adults gave answers by button-press before the end of the auditory stimulus 26% of the time. On average they answered 357 ms after the end of the sentences, which had an average duration of 1596 ms (sd = 146 ms). Mean RT per sentence type were as follows (see Figure 2): 1785 ms in [+an +an]; 1726 ms in [+an -an]; 2163 ms in [-an +an]; and 2140 ms in [-an -an].

A comprehension delay of subject-object word order in Dutch preschoolers

155

Figure 2. Adult RTs per sentence type in ms.

In order to determine if there was an effect of sentence type on RT in comprehension, a one-way repeated measures ANOVA was run with the log transformed RT means (aggregated over subject and condition) with Sentence Type (four levels) as the within-subject factor. The results show that the speed at which a participant gave an answer on the comprehension task was affected by the Sentence Type (F(1.71, 30.81) = 24.21, p < .001). Pairwise comparisons with Bonferroni adjustment showed that the conditions with animate subject had signi¿cantly shorter RTs than the conditions with an inanimate subject (in all comparisons p < .001). Eye movements As with the RT analysis, only items with correct responses were analyzed. For comprehension items, Areas of Interest were de¿ned in Tobii Studio as 25% of the screen over target picture, as 25% over the distractor, and the rest was de¿ned as Not on Area of Interest. Eye data was based on the average of both eyes by Tobii Studio, which applied a default sliding radius ¿xation ¿lter to account for drift and blinking (see Tobii 1.X User Manual). Proportions of looks to Target, Distractor, or Not on Area of Interest were aggregated over participant and condition and grouped in bins of 32 ms. The data were then divided into four segments of 512 ms each, which roughly corresponded to the subject (Segment 1), the verb (Segment 2), the object (Segment 3), and beginning of sentence offset (Segment 4). The graphs in Figure 3 can be read as proportions of looking to Target or Distractor over time, going from left to right as the sentences unfolded, for each Sentence Type.

156

Gisi Cannizzaro

To determine whether there was an effect of Sentence Type on the proportions of looking to either the Target or the Distractor over time, a 2 × 4 × 4 repeated measures ANOVA was run on the arcsine transformed proportions with Area of Interest (Target versus Distractor) and Segment (1 – 4) as within-bin factors and Sentence Type (4 levels) as between-bin factors. There was a signi¿cant main effect of Area of Interest (F(1,60) = 770.3, p < .001), indicating that there was overall more looking to the Target across time and sentence types. The signi¿cant main effect of Segment (F(1.1, 65.45) = 79.0, p < .001) is not surprising, since looking behavior is expected to change over time. In order to tease apart the the interaction that was found between Area of Interest × Segment × Sentence Type (F(7.7, 155.4) = 20.5, p < .001), two oneway repeated measures ANOVAs were run for each Area of Interest, again with Segment as a within-bin factor and Sentence Type as between-bin. The analysis of looks to the Target showed that that the main effect of Segment was maintained (F(1.3, 76.1) = 133.78, p < .001) as well as the interaction of Segment and Sentence Type (F(3.8, 76.1) = 20.5, p < .001). Contrasts on this interaction term showed that looking to the Target always increased from Segment 1 to Segment 2 regardless of Sentence Type (F < 1), but after that there was an effect of Sentence Type (Segment 2 to 3 (F (3, 60) = 39.9,

Figure 3. Adults’ mean proportions of looking to Target/Distractor over 4 Segments, per Sentence Type. Error bars: +/í 1.00 SE.

A comprehension delay of subject-object word order in Dutch preschoolers

157

p < .001), Segment 3 to 4 (F(3, 60) = 19.1, p < .001). In order to investigate the interactions in Segments 2 to 4, a one-way independent ANOVA was run for each Sentence Type comparing the 4 Segments. The multiple comparisons with Bonferroni corrections based on these ANOVAs showed a decrease in looking to the Target from Segments 2 to 3 and from 3 to 4 for the sentences with an animate subject (in all comparisons p < .001 except Segment 2 to 3 in [+an +an], p = .028). In contrast, comparisons showed an increase in looking to the Target from Segments 2 to 3 for the sentences with an inanimate subject ([-an +an]: p = .002, [-an -an]: p = .001). Looking continued to decrease from Segment 3 to 4 in the [-an -an] condition (p < .001), but did not change signi¿cantly in the [-an +an] condition (p > .10). Thus, while looking to the target generally increased and then decreased, this pattern was protracted in the conditions with an inanimate subject. The analysis of looks to the Distractor also maintained the main effect of Segment (F(1.2, 76.5) = 77.5, p < .001), but there was no longer an interaction of Segment and Sentence Type, F(3.8, 76.5) = 1.059, p > .10. Post hoc pairwise comparisons with Bonferroni correction showed the effect of Segment was the same for all Sentence Types, namely, that there was no difference in looking in the ¿rst two Segments (p < .10) and then there was a pattern of decreased looking from Segments 2 to 3 (p = .007) and from 3 to 4 (p < .001). There was also a signi¿cant main effect of the between-bin factor of Sentence Type. A one-way independent ANOVA was run comparing the 4 Sentence Types over collapsed Segments; multiple comparisons showed the amount of looking to the Distractor in the [+an -an] sentences was lower than all of the other conditions (for all comparisons p < .001). Therefore, the general pattern of looking to the distractor was a decrease over time, but in the [+an -an] sentences the amount of looking to the Distractor was overall lower than for the other three sentence types. Discussion Experiment 1a Adults had no problem with answering accurately on either task. They correctly interpreted sentences as subject-initial 97% of the time and correctly used S-O word order every time they produced a target-like construction. The overall tendency to ¿xate on the target rather than the distractor in the picture selection task across sentence types (the main effect of Area of Interest) was in line with the interpretation adults indicated with their off-line answers. While there was no effect of sentence type on the answers they gave, there was such an effect on how fast they gave these answers. If the subject was animate, adults were able to respond more quickly than if the subject was

158

Gisi Cannizzaro

inanimate, which indicates that the animacy properties of the subject play a role in time needed to process simple sentences. The delay in giving an answer can explain the late drops in looking to the target in the sentences with inanimate subject: In the other two conditions, [+an +an] and [+an -an], the participants had already made up their minds since proportions of looks to anything were already dropping off. The gaze data of the adult participants shows a further distinction that can be made between the sentence types. It was in the [+an -an] condition that throughout the sentence, looks to the distractor remained the lowest in comparison to the other three conditions. In sum, the results from the adult study reÀect the predictions of the model in which accuracy would be at ceiling, sentences with inanimate subjects would be delayed, and the [+an -an] items would be facilitated. This supports the idea that there is animacy constraint in the adult grammar that has no inÀuence on ultimate sentence interpretations, but that does have an inÀuence on the processing of the sentences.

Experiment 1b: Children Participants There were 26 children aged 2;10 to 4;0 (18 male, mean age = 3;6.5 sd = 3.25 months) who participated in the study.5 Participants were recruited through daycares in Groningen. The children were monolingual Dutch-speakers with normal hearing, vision, and history of health. Materials and design The 16 experimental items (pictures and sentences) were identical to those in Experiment 1a, also with regard to correct side and direction of action, but easier practice items were used and no ¿llers were included. The sentences were presented orally, and there was only one list order. In order to avoid confusion between the two verbs (which are semantic opposites), one verb was used in the ¿rst block and the other verb in the second, with some children receiving duwen ¿rst and some trekken. The comprehension task was always given before production in each block. The computer set up for Experiment 1b was the same for Experiment 1a except that Tobii Studio’s user 5. Young two-year-olds were not tested since those pilot tested did not pass the pretests, or they did not point or speak during the experiment itself.

A comprehension delay of subject-object word order in Dutch preschoolers

159

camera and the desktop microphone were used to record what was said by the assistant and the child during the entire experiment. Procedure Each child was pre-tested to see if he or she was able (1) to name and identify each of the objects that appear in the experimental and practice items, (2) to successfully point to one of two pictures, and (3) to demonstrate the action of pushing and pulling after having seen it modeled with a small toy pig and truck. In the verb demonstration, the assistant performed the action to one object and asked the child to perform the same action to the other object. The action of pulling was modeled with use of a string, which represented the rope seen in the experimental pictures. Crucially, the assistant used phrases like “This is pulling. Now you try” in order not to model word order. The child and parent then moved to a chair in front of the eye tracker, where the child sat on the parent’s lap. The parent wore dark glasses that prevented the eye tracker from registering the parent’s eyes and prevented any unconscious inÀuencing of the child’s answers. Once the child was positioned about 70 cm from the screen, the researcher ran the calibration. In the comprehension task, the child was presented with two pictures that appeared side-by-side on the screen and was simultaneously delivered the experimental sentence orally by the assistant in a neutral manner. Only during the practice items did the assistant give instructive feedback. During the test phase, if the child did not point, the test sentence was repeated a maximum of two times. Responses were scored as one of the following: clear to left, clear to right, a change from left to right, a change from right to left, both, or no response. If the child changed an answer, the second response was used. Points clearly intended to signify both pictures (often reinforced with a verbal both) were scored as incorrect. The responses were recorded by both the assistant via E-Prime and by the researcher on a score sheet during testing. Directly after the 8 comprehension items, the elicited production task began. The puppet closes his eyes before a single picture appeared and it was the task of the child to tell the puppet what was on the screen. The assistant asked “What’s happening here?” instead of “What do you see?” to prevent simple object naming. Only during the practice items did the assistant give instructional feedback. During the test phase, if the child gave no response, the elicitation question was repeated a maximum of two times. If the child gave an incomplete (but not incorrect) answer, a second elicitation was repeated a maximum of two times, such as “Can you tell me more?” The researcher wrote down the responses of the child and consulted the video recordings after testing

160

Gisi Cannizzaro

to con¿rm the responses. Utterances were scored in the same way as adults, but included the category no response. To check for scoring reliability, a second categorization of productions was conducted by an independent scorer; agreement was very high (Cohen’s ț = .99). The ¿rst block was followed by a break. Results Of the 26 children tested, 19 remained eligible for inclusion in the analysis. While all children passed the pre-tests, seven did not get at least two out of ¿rst three practice items correct on one or both tasks. The mean age of the remaining set of children was 3;6.5 (sd = 3.3 months). Accuracy For comprehension, of the 304 expected responses for test items made by all children, there was only one item for which no answer was given; of the remaining responses, there was 198 scored as correct, and 105 as incorrect. For production, there were no refusals to answer and 4 responses were deemed unscorable (for example the boat and the ¿sh are bumping against each other). This left 265 productions to be scored as having S-O word order, and 35 as O-S. Of the S-O responses, 39% (104 utterances) were scored as S-O+, mostly due to use of gaan as an auxiliary. Proportions of correct answers based on participant analysis per sentence type (with counts in parentheses) are shown for both tasks in Table 4. Table 4. Mean percentages of correct responses for children in Dutch study Sentence Type

Comprehension

Production

[+an +an] [+an -an] [-an +an] [-an -an]

72.37 (55) 65.79 (50) 59.21(45) 63.16 (48)

89.47 (67) 90.79 (69) 88.16 (65) 85.11 (64)

Total

65.13 (198)

88.38 (265)

To see if there was a difference between the two tasks and type of sentence, a 2 × 4 repeated measures ANOVA was run on the arcsine transformed means (over participants), with Task (Production vs. Comprehension) and Sentence Type (four levels) as within-participants factors. The analysis revealed a signi¿cant main effect of Task (F(1,18) = 32, p < .001) but no signi¿cant

A comprehension delay of subject-object word order in Dutch preschoolers

161

main effect of Sentence Type, (F(3,54) = 1.488, p > .05). There was also no signi¿cant interaction of Task and Sentence Type (F < 1). This means that children performed better on production than comprehension regardless of type of sentence. Eye movements A threshold was set which required that more than 50% of a participant’s items had to have at least 60% tracking of at least one eye. Four children did not meet this requirement and were removed from the eye data analysis. Only data from items with correct responses were analyzed. The same Areas of Interest were used as in Experiment 1a and the same data treatment was applied to allow an analysis of gaze data over bins of time. Four segments were created, but because children took longer to answer than adults, the segments were made to be 1024 ms, twice the size of the segments in the adult analysis. The auditory stimuli were presented orally by the assistant, so the gaze data for the children was not tightly time-locked to the auditory stimulus. Figure 4 shows bar plots of proportions of looking to the Target or the Distractor per Sentence Type for the 4 Segments, each of which correspond to about one second. To determine whether there was an effect of Sentence Type on the proportions of looking to either the Target or the Distractor over time, the same 2 × 4 × 4 repeated measures ANOVA was run as with the adults on the arcsine transformed proportions with Area of Interest (Target versus Distractor) and Segment (1 í 4) as within-bin factor and Sentence Type (4 levels) as betweenbin factors. There was a signi¿cant main effect of Area of Interest (F(1,124) = 483.5, p < .001), a reÀection of overall more looking to the Target across time and sentence types. There was also a signi¿cant main effect of Segment (F(3,372) = 151.0, p < .001), indicating that there were changes in looking over time. The three way interaction of Area of Interest × Segment × Sentence Type was signi¿cant as well (F(9, 372) = 25.1, p < .001), so two additional repeated measures ANOVAs were run for each Area of Interest, again with Segment as a within-bin factor and Sentence Type as between-bin. The analysis of looks to the Target showed that the main effect of Segment was preserved, (F(1.7, 372) = 168.3, p < .001), as well as the interaction of Segment and Sentence Type (F(4.9, 372) = 14.88, p < .001). Contrasts on the main effect of Segment showed that across Sentence Types there was a signi¿cant effect of Segment (Segment 1 to 2 : F(1, 124) = 244.7, p < .001; Segment 2 to 3: F(1,124) = 53.6, p < .001; Segment 3 to 4: F(1, 124) = 73.9, p < .001). Speci¿cally, pairwise comparisons with Bonferroni corrections showed the following pattern over time: looking to the Target always increased from

162

Gisi Cannizzaro

Segment 1 to Segment 2 and from 2 to 3 before decreasing in the last Segment (for all comparisons p < .001). Contrasts on the interaction term indicated no interaction of Segment and Sentence Type when comparing the ¿rst two segments, but a signi¿cant interaction was found found when comparing Segment 2 to 3 (F(3, 124) = 62.4, p < .001) and Segments 3 to 4 (F(3, 124) = 24.4, p < .001). In order to investigate the interactions in Segments 2 to 4, a one-way independent ANOVA was run for each Sentence Type comparing the 4 Segments. The multiple comparisons with Bonferroni corrections based on these ANOVAs showed that looking increased from Segment 2 to 3 only for the [+an +an] sentences (p < .001), and with no signi¿cant changes for the other three sentence types ([+an -an] and [-an -an]: p > .10; [-an +an]: p < .10). Decreases in looking were found between Segment 3 and the ¿nal segment for only the sentences with an animate subject (both p’s < .001) and no signi¿cant change in looking was found for the sentences with inanimate subject (both p’s > .10). In other words, the same pattern was found here as with adults: While looking to the target over time generally increased and then decreased over time, it was in the sentences with inanimate subjects that these looks did not eventually drop signi¿cantly. The analysis of looks to the Distractor also showed both a main effect of Segment (F(1.4, 175.17) = 50.9, p < .001) and an interaction of Segment and Sentence Type (F(4.2, 175.17) = 10.79, p < .001). Contrasts on the main effect

Figure 4. Children’s mean proportions of looking to Target/Distractor over 4 Segments, per Sentence Type. Error bars: +/í 1 SE.

A comprehension delay of subject-object word order in Dutch preschoolers

163

showed that the main effect was signi¿cant when moving from the Segment 1 to 2 (F(1, 124) = 84.2, p < .001) and 2 to 3 (F(1, 124) = 192.5), p < .001), but there was no signi¿cant effect of Segment when moving from 3 to 4 (F(1,124) = 2.4, p > .10). Multiple comparisons of the Segments with Bonferroni correction showed that the general pattern was for looks to the Distractor to increase from Segment 1 to 2 and then decrease from 2 to 3, with no pattern when moving to the remaining segment. Therefore, looking to the distractor occurred early and then decreased. In order to ¿nd the source of the interaction that was found, one-way independent ANOVAs comparing Sentence Types were run per Segment. Multiple comparisons with Bonferroni correction showed a higher amount of looking to the Distractor in the [+an +an] condition relative to the other types of sentences in the Segment 2 (for all comparisons p < .001) This means that of all the looks to the Distractor in Segment 2, the [+an +an] sentences received the most looks. Discussion Experiment 1b The overall performance of children on production (88.38%) was better than their performance on comprehension (65.13%). Type of sentence did not seem to make a difference in the word order the children produced, with performance on each type above 85%, which is in line with what Chapman and Miller found with English-speaking children. However, there was also no signi¿cant effect of type of sentence on the comprehension task, which Chapman and Miller did ¿nd (with the best performance in [+an -an] and the worst performance on [-an +an]). The gaze data of the children who answered correctly was similar to that of adults in that there was an overall tendency to look towards the Target in the picture selection task. Although the segments used in analysis were of different sizes for the children and adults, the looking pattern of children did reÀect that of adults in the later two segments, in which looks to the Target were more sustained in the conditions with inanimate subject. In the discussion of the results from adults it was suggested that this is due to the fact that the adults have already made up their minds about the sentences with an animate subject, so proportions of looks to anything dropped. This drop in looks to either Target or Distractor can also be seen for children in the last segment in the conditions with animate subject.

164

Gisi Cannizzaro

General discussion The ¿rst question the study aimed to answer was whether there is a comprehension delay of S-O word order in Dutch preschoolers. Results showed that children performed better on the elicited production task than on the picture selection task, indicating that there is a comprehension delay for S-O word order in these children. (See Cannizzaro 2010 for a discussion of how adequately this delay can be explained by traditional approaches to language acquisition versus by the re-ranking model.) The second question was whether the animacy properties of the noun phrases play a role in the interpretations that children have in their ofÀine responses. For production, children performed well on all sentence types, which is in line with what Chapman and Miller (1975) found and with what the Optimality Theoretic model of Hendriks et al. (2005) predicts. For comprehension, however, while Chapman and Miller found extremely good performance in the [+an -an] sentences and bad performance on [-an +an] sentences, the children in this study made no signi¿cant distinction between any of the sentence types. The Optimality Theoretic model predicts more mistakes on [-an +an] sentences than was found here and predicts less mistakes on the three sentence types than was shown here. One explanation that can be given for why the children did not give more O-S interpretations to sentences with an inanimate subject and animate object is because the constraints in the grammar of three year olds are likely in the state of gradually being re-ranked (Boersma and Hayes 2001) since constraints are hypothesized to be correctly ranked around the age of 4 (Hendriks and Spenader 2006). This does not explain, however, why the children would not give S-O interpretations more often to the other three sentence types, since regardless of the ranking of PRECEDENCE and PROMINENCE, this interpretation is the winner. On-line results for children were not in accordance with the predictions made. Children did not suffer from more competition from the distractor picture via looks during the picture selection in the [-an +an] sentences than in the other three types of sentences. Instead, there was an early preference for the distractor in the [+an +an] sentences. The fact that in the distractor picture for these items an animate entity was also performing an action could be the explanation, but then this does not explain why there was not also a larger look to the distractor in the [-an +an] sentences. With regard to the looking to the Target, there was a difference in their looking behavior between sentences with animate subject and sentences with inanimate subject, a pattern comparable to that of adults.

A comprehension delay of subject-object word order in Dutch preschoolers

165

The third question was how do adults controls perform on these tasks. Adults, not surprisingly, performed well on comprehension and production. However, their speed of response and gaze patterns during comprehension revealed that they make a distinction between sentences in which the animacy of subjects and objects are manipulated. They took longer to respond to sentences in which the subject was inanimate, and they were least distracted by the picture depicting an O-S interpretation in the [+an -an] condition. If young children’s use of animacy is a strategy that disappears as they mature, then why would adults show evidence of using animacy information to help them interpret sentences faster? In sum, while the children did show poorer performance on comprehension, there was not as strong of an effect of animacy as the Optimality Theoretic model predicted. One possible reason signi¿cant animacy effects were found by Chapman and Miller and not here is that the distance on the animacy hierarchy between the animate and inanimate nouns they used was greater. Chapman and Miller used not only animals but also human beings as animate entities. The result is a greater difference in the likelihood to agentivize between the humans and the vehicles in their study versus between the animals and vehicles in this study. Furthermore, vehicles – though non-living – are capable of dependent motion (see note 5). While non-motive entities (like building) or even other dependently motive entities (like rock or cup) are less likely to agentivize than vehicles, including these in the materials would have resulted in sentences with push and pull that are not plausibly reversible. Therefore the most controlled sentences with push and pull involve animals and vehicles – entities that may be too near on the animacy hierarchy to result in reliable animacy effects in children. Even though there is a relative closeness in animacy of the entities used in the materials, it could be the case that the children actually were using the animacy properties of subject and object in their interpretation. As was already pointed out, the children tested here were older than the children tested in Chapman and Miller and were perhaps too advanced in their gradual constraint re-ranking to still show the predicted effects of animacy. In order to test this possibility, an experiment with a picture selection task in which two-year-olds reliably point is necessary. A second possibility that would allow for inclusion of two-year-olds who have dif¿culty with pointing would be a comprehension task in which no decision is required of the young children and only looking behavior is measured. Only when two-year-olds have been tested can we be sure that the differences between the sentence types do not have an inÀuence on their answers. Furthermore, by testing children without the task of pointing, it can be determined whether the comprehension delay is simply a task effect

166

Gisi Cannizzaro

of the picture selection; in other words, if children perform just as well on comprehension as on production in a looking-only task, then perhaps there is no actual underlying linguistic delay. Word order in Dutch is more Àexible than in English. Nevertheless, the comparability of Dutch and English when it comes to word order in the types of sentences investigated here is upheld by the fact that in this study Dutch adults always used SVO word order in production and always interpreted sentences as subject-initial in comprehension. The difference between Dutch and English word order could, however, explain why the Dutch children did not exhibit an effect of animacy. In Dutch, OVS word order is possible in simple sentences only in the rare cases when discourse and information structure call for it (see note 3). Perhaps whatever constraint that allows the topic to come ¿rst in violation of normal S-O word order plays a role here, in which case the Optimality Theoretic model and resulting predictions need to be adjusted accordingly. While the children’s gaze patterns did not meet the speci¿c predictions of the Optimality Theoretic model concerning the different sentence types, patterns did show that the animacy of the subject made a difference in how soon their looking to either target or distractor would drop off, an effect also found in adults. This effect of subject animacy was predicted for adults based on violations of an animacy constraint over time (de Hoop and Lamers 2005), so if children also show an effect of the animacy of the subject, it could be that children process sentences in a similar fashion to adults. This is a topic for future research that could be more carefully inspected if the stimuli for children were pre-recorded and therefore time locked to the gaze data.

Conclusion In this study, children and adults were tested on the same sentences in comprehension and in production. The animacy properties of subject and object in the sentences were manipulated to test the effect this had on word order in sentence production and comprehension. A closer look was taken at sentence interpretation by additionally measuring on-line behavior. Results showed that while adults exhibited no effect of S-O animacy on S-O word order production or comprehension, their on-line responses during comprehension revealed a sensitivity to the animacy manipulations. This was argued to be support for a low-ranked animacy constraint in their grammar, namely, PROMINENCE. Children performed better on production than comprehension, but it remains unclear what is causing this asymmetry. Even though there was no signi¿cant animacy effect in their off-line responses on the comprehension task, we cannot

A comprehension delay of subject-object word order in Dutch preschoolers

167

be certain whether this asymmetry is due to a task effect of picture selection or whether the asymmetry is after all a result of children using animacy information in sentence interpretation. The latter possibility must be investigated further with younger children, as well as with a model that accounts for possible cross-linguistic differences between Dutch and English S-O word order. The similarity in adults’ and children’s processing of the different sentence types supports the hypothesis that an animacy constraint is present in both grammars.

References Bloom, Lois 1970 Language development: Form and function in emerging grammars. Cambridge, MA: The MIT Press. Boersma, Paul, and Bruce Hayes 2001 Empirical tests of the gradual learning algorithm. Linguistic Inquiry 32: 45–86. Bouma, Gerlof 2008 Starting a sentence in Dutch: a corpus study of subject- and object-fronting. Ph.D. diss, University of Groningen. Brown, Roger 1973 A ¿rst language: the early stages. Cambridge MA: Harvard University Press. Cannizzaro, Gisi 2010 Animacy and early word order. In Language Acquisition and Development: Proceedings of GALA 2009, João Costa, Ana Castro, Maria Lobo, and Fernanda Pratas (eds). Newcastle upon Tyne, UK. Cambridge Scholars Publishing. Chapman, Robin S., and Lawrence L. Kohn 1978 Comprehension strategies in two and three year olds: Animate agents or probable events? Journal of Speech and Hearing Research 21: 746–761. Chapman, Robin S., and Jon F. Miller 1975 Word order in early two and three word utterances: Does production precede comprehension? Journal of Speech and Hearing Research 18: 355–371. Fraser, Colin, Ursula Bellugi, and Roger Brown 1963 Control of grammar in imitation, comprehension and production. Journal of Verbal Learning and Verbal Behavior 2: 121–135.

168

Gisi Cannizzaro

Hendriks, Petra, Helen de Hoop, and Monique Lamers 2005 Asymmetries in language use reveal asymmetries in the grammar. In Proceedings of the 15th Amsterdam Colloquium, Paul Dekker and Michael Franke (eds.), 113–118. Amsterdam: Institute for Logic, Language, and Computation. Hendriks, Petra, and Jennifer Spenader 2006 When production precedes comprehension: An optimization approach to the acquisition of pronouns. Language Acquisition 13 (4): 319–348. Hirsh-Pasek, Kathy, and Roberta M. Golinkoff 1996 The origins of grammar: Evidence from early language comprehension. Cambridge, MA: The MIT Press. Ingram, David 1974 The relationship between comprehension and production. In Language perspectives: Acquisition, retardation, and intervention, Richard Schifelbusch and Lyle Lloyd (eds.), 313-334. Baltimore: University Park Press. Kaan, Edith 1997 Processing subject-object ambiguities in Dutch. Ph.D. diss, University of Groningen. Lamers, Monique, and Helen de Hoop 2005 Animacy information in human sentence processing: An incremental optimization of interpretation approach. In Lecture Notes in Arti¿cial Intelligence 3438, Henning Christiansen, Peter R. Skadhauge and Jørgen Villadsen (eds.), 158–171. Berlin/Heidelberg: Springer-Verlag. McClellan, Janet, Carolyn Yewchuk, and Gary Holdgrafer 1986 Comprehension and production of word order by 2-year-old children. Journal of Psycholinguistic Research 15: 97–116. Prince, Alan, and Paul Smolensky 2004 Optimality Theory: Constraint interaction in generative grammar. Blackwell. As Technical Report CU-CS-696-93, Department of Computer Science, University of Colorado at Boulder, and Technical Report TR-2, Rutgers Center for Cognitive Science, Rutgers University, New Brunswick, NJ, April 1993. Schneider, Walter, Amy Eschman, and Anthony Zuccolotto 2002 E-Prime User’s Guide. University of Pittsburgh: Psychology Software Tools, Inc. Smolensky, Paul 1996 On the comprehension/production dilemma in child language. Linguistic Inquiry 27: 720–731. Tobii Technology 2008 Tobii Studio 1.X User Manual. Danderyd, Sweden: Tobii Technology AB.

A comprehension delay of subject-object word order in Dutch preschoolers

169

van Valin, Robert D. Jr., and David P. Wilkins 1996 The case for ‘effector’: Case roles, agents, and agency revisited. In Grammatical constructions: Their form and meaning, David P. Wilkins (ed.), 289–322. Oxford University Press. de Villiers, Jill, and Paul de Villiers 1973 Development of the use of word order in comprehension. Journal of Psychological Research 2: 331–341. Zink, Inge, and Maryline Lejaegere 2002 N-CDIs: Lijsten voor Communicatieve Ontwikkeling. Aanpassing en hernormering van de MacArthur CDIs van Fenson et al. 1993 [A CDI user’s manual with normative and validity data for Dutch]. Acco, Leuven/Leusden.

Asymmetries in children’s language performance within and across modalities Oda-Christina Brandt-Kobele and Barbara Höhle

Introduction In order to communicate successfully via spoken language we have to be able to understand utterances directed to us as well as to encode our thoughts into sounds when talking to others. Young children acquiring a language have to learn both: to decode what a given linguistic expression means in language comprehension, and to encode a meaning linguistically in language production. Comprehension tasks are often considered as measuring what children know about language while production tasks reveal how children put this knowledge into use (Bates, Dale, and Thal 1995). In support of this, it is generally observed that children ¿rst understand a linguistic expression, like a word or a sentence, before they use it productively (Clark and Hecht 1983). Some recent experimental ¿ndings have nevertheless pointed to a reverse picture, where children’s production abilities concerning a speci¿c linguistic structure are found to be present earlier than their respective comprehension abilities. In this chapter, we will ¿rst provide a survey of studies that indicate such an asymmetric developmental path, followed by some proposals as to how to account for this pattern. Second, we explore one of these proposals, which deals with methodological issues, in more detail and report on a test case in a domain in which the production-preceding-comprehension pattern has also been found, namely that of verb inÀection. Third, we discuss how methoddependent within-modality asymmetries, which we among others have found, might help in understanding cross-modality asymmetries.

Production-comprehension asymmetries in child language The often assumed pattern that the comprehension of a given linguistic structure precedes its production is reÀected in various ¿ndings in child language research: on the lexical level, children’s receptive vocabularies are found to be more elaborate than their productive ones at any given age

172

Oda-Christina Brandt-Kobele and Barbara Höhle

(e.g. Goldin-Meadow, Seligman, and Gelman 1979; Fernald and Marchman 2006). In the domain of syntax, researchers found children to be sensitive to the grammaticality of sentence structures they do not yet produce themselves (e.g. Santelmann and Jusczyk 1998; Höhle et al. 2006), and to understand the meanings of transitive and intransitive verb frames at an age when they hardly produce two-word utterances (Naigles 1996). These kinds of asymmetries in child language can be accounted for by the assumption of a discrepancy between children’s linguistic competence as revealed by comprehension data and their linguistic performance as revealed by production data. An alternative route of explanation is to assume two different grammars, a more elaborate one for comprehension and an inferior one for production (e.g. Smolensky 1996). A puzzling case on the other hand exists when children’s production abilities for a linguistic structure seem to be better or in place earlier than their comprehension abilities. This cross-modal asymmetry has been observed in the acquisition of some linguistic expressions, such as pronouns, focus particles, and restrictive modi¿ers. English pronouns are perhaps the best known instance of the productionpreceding-comprehension pattern. While children produce pronouns (him, her) from the age of 2 or 3 years onwards, adult-like comprehension is usually not found before the age of 6 years (Chien and Wexler 1990; Hendriks and Spenader 2006; Sekerina, Stromswold, and Hestvik 2004). This holds for pronouns in main clauses, in which the pronoun is not co-indexed and thus not coreferential with the preceding NP (Tigger is washing him), as well as for pronouns within prepositional phrases (Tigger put the box behind him). In the former case, children often interpret the personal pronoun like a reÀexive one. In the latter case, the sentence is ambiguous in adult English, as the pronoun can refer to the sentence-internal referent or to an external referent mentioned in the discourse. Children younger than 6 years of age do not grasp the two meanings that these sentences have in adult English (Sekerina et al. 2004). The pattern of better performance in the production of pronouns than in their comprehension is even found in children older than 6 years (De Villiers, Cahillace and Altreuter 2004). A comparable cross-modal asymmetry has been found for the acquisition of focus particles. The interpretation of sentences containing focus particles is highly dependent on the prosodic structure of the sentence. If the word carrot is stressed in a sentence like Tigger only gave a CARROT to Piglet, the listener has to infer that Piglet received only a carrot from Tigger and nothing else. If on the other hand the word Piglet receives primary stress (Tigger only gave a carrot to PIGLET), the listener has to infer that only Piglet received a carrot and no one else did (e.g. SzendrĘi 2004). In production experiments asking for

Asymmetries in children’s language performance within and across modalities

173

descriptions of picture pairs that only differ in one single attribute, children below age 4 produce contrastive stress in an adequate way (Hornby and Haas 1970), but children at that age do not seem to be able to make use of contrastive stress when interpreting sentences with a focus particle like only (e.g. SzendrĘi 2004; Gualmini, Maciukaite, and Crain 2003). Interestingly, children perform better when discourse information in the preceding context adds to the prosodic information (Gualmini et al. 2002). Thus, prosodic information alone does not determine children’s comprehension of sentences containing the focus particle only at age 5 or 6, even though they use this information productively at age 4. Similar problems arise for German children when processing sentences containing the focus particle auch (‘also’). Again, interpretation of such sentences is highly dependent on prosodic information. If the focus particle is accented as in the sentence Tobi hat AUCH eine Puppe (‘Tobi has ALSO a doll’), the interpretation calls for a subject-alternative set, i.e. there must be someone else besides Tobi who also has a doll. If the focus particle is not accented (Tobi hat auch eine PUPPE ‘Tobi has also a DOLL’), the interpretation calls for an object-alternative set to make the sentence felicitous, i.e. there has to be something else that Tobi owns. Even though children were found to produce accented and unaccented auch as early as 2 to 3 years (Nederstigt 2003), they still do not perform adult-like by the age of 5 to 6 years in a picture-selection task (Hüttner et al. 2004). The use of restrictive modi¿ers is another area in which this productioncomprehension asymmetry has been observed. Hurewitz et al. (2000) aimed to further explore children’s inability to comprehend a prepositional phrase like on the napkin (in sentences like Put the frog on the napkin in the box) as a restrictive modi¿er. As found by Trueswell et al. (1999), children aged 4 to 5 years always interpret such an ambiguous PP as a destination of the verb put, even if the referential scene is calling for a modi¿er interpretation. Hurewitz and colleagues presented children with arrays of objects that always contained two animals of the same kind (e.g. two frogs) as well as short stories about these animals. The 4- to 5-year old children produced restrictive modi¿er phrases in 87% of all cases, when asked which particular animal performed a speci¿c action mentioned in the short story. But in the act-out comprehension task that immediately followed the production task, the same children interpreted a PP as a restrictive modi¿er in only 22% of all cases. Thus, while English-speaking children aged 4 to 5 years use restrictive modi¿ers productively to disambiguate the members of a set, they fail to realize that a PP is used to determine a speci¿c referent during language comprehension (Hurewitz et al. 2000). All of these cases in which production seems to be ahead of comprehension have led to intense discussions about what such cross-modal asymmetries

174

Oda-Christina Brandt-Kobele and Barbara Höhle

might reveal about the child’s underlying linguistic system. One possible way of dealing with them is to assume the existence of dissociations in the grammatical systems which underly the performance in language production and comprehension. This does not necessarily imply different grammars for comprehension and production, but rather assumes that comprehension lags behind in those cases where the speaker’s perspective has to be taken into account (Hendriks and Spenader 2006). A second approach renders children’s computational restrictions and processing limitations responsible for comprehension failures in areas where production seems to be ahead (Grodzinsky and Reinhardt 1993). A third approach considers the asymmetry to be an artifact of methodological problems in traditional comprehension tasks (Bloom et al. 1994). This last approach will be further explored in the remainder of this chapter and tested in a linguistic domain in which asymmetries are also found, namely in the area of verb inÀection. We focus on children’s ability to use the number information provided by verb inÀection to infer the number of an ambiguous sentence subject and we show that the evaluation of children’s competence to successfully use verb inÀection in sentence comprehension is strongly dependent on the experimental method employed. We will end with some considerations what this method-dependency of experimental results might imply for the production-comprehension asymmetry. Production and comprehension of verb inÀections Subject-verb agreement is an overt indication of one of the basic dependencies syntactic constituents enter into, and is required in most languages (Nicols 1986). Typically, subject-verb agreement involves the cross-referencing of morphosyntactic features of the subject by means of inÀectional endings on the verb. Thus far, child language researchers have primarily been interested in the emergence of inÀectional morphemes because the presence (or absence) thereof in early multi-word utterances is considered to provide evidence for (or against) the existence of functional categories in child grammar. This relates to the debate about whether children’s early grammars differ from adult grammars (e.g. Radford 1990; Lebeaux 1988) or not (e.g. Verrips and Weissenborn 1992; Guasti 1993). Another debate focusses on the impact of verb inÀection on the acquisition of verb movement, which is, for example, obligatory in German and Italian (Clahsen 1986; Verrips and Weissenborn 1992; Guasti 1993). The richness and complexity of the inÀectional paradigm varies from language to language. Cross-linguistic research has shown that a low number of different inÀectional forms, as in English, does not necessarily prove to be

Asymmetries in children’s language performance within and across modalities

175

advantageous for children learning the language, since earlier mastery of such items has been found for languages with rich and diverse inÀectional systems, as for example Spanish or Italian (e.g. Guasti 2002, Phillips 1995). This might be due to the fact that in languages with a richer inÀectional system the mapping between form and meaning is often more transparent than in languages with a poorer inÀectional system. Such cross-linguistic ¿ndings are based mainly on production data, gathered either as samples of spontaneous speech (e.g. Brown 1973; Poeppel and Wexler 1993; Clahsen 1986) or through elicited production tasks (e.g. Song, Sundara, and Demuth 2009; Leonard et al. 1992). Fewer studies have been concerned with the comprehension of verb inÀections (e.g. Johnson, de Villiers, and Seymour 2005; Leonard, Miller, and Owen 2000; Perez-Leroux 2006), and only a handful of studies have compared production and comprehension abilities within a single group of children (Fraser, Bellugi, and Brown 1963; Keeny and Wolfe 1972). When testing the comprehension of verb inÀections, usually the number contrast is considered (but see for example Beyer and Hudson Kam (2009) for testing the comprehension of the tense contrast). Interestingly, when comparing the time-course found in production studies and the one found in comprehension experiments, the picture of a production-comprehension asymmetry emerges. Concerning the productive modality, Brown (1973) studied the appearance of various morphological markers in spontaneous speech samples of three children acquiring English. He de¿ned ‘mastery’ to be present when a child is producing a grammatical morpheme correctly in at least 90% of its obligatory contexts. He found that mastery of the 3rd person singular agreement morpheme -s emerged between the ages of 2;2 (years;months) and 3;10. Rice and Wexler (2002) examined the production of the same morpheme in different age groups using an elicited production task. They found mastery in Brown’s sense not before the age of 4;0. One reason for the later mastery when employing elicited production tasks might lie in the selection of verbs and their frequency in a child’s mental lexicon. In a very recent study, Song et al. (2009) examined the inÀuence of phonological factors on children’s production of 3rd person singular -s. Here, longitudinal spontaneous speech data as well as elicited imitations from children between the ages 1;3 and 3;6 delivered results in line with the ¿ndings of Brown (1973) and Rice and Wexler (2002). In addition, they found children to be better at producing the verb inÀection in simple codas than in complex codas, as well as in utterance-¿nal as compared to utterance-medial positions, which suggests that children’s use of grammatical morphemes is conditioned not only by grammatical but also by prosodic factors.

176

Oda-Christina Brandt-Kobele and Barbara Höhle

For German-speaking children, only spontaneous speech data have been reported so far. Clahsen (1986) examined the productive use of verb inÀections by two German children in the age range of 1;6 to 3;6 by means of frequency analysis. Correct use (in Brown’s sense) did not emerge before age 2;11 for the 3rd person singular agreement marker -t. For 3rd person plural -n, mastery was found a little later, namely at age 3;1. In a similar study, Poeppel and Wexler (1993) analyzed the spontaneous speech of one German child at age 2;1. They found that in all utterances containing a singular subject and a verb, the verb was correctly inÀected. The very small number of agreement errors all consisted of a plural subject combined with a verb inÀected with the 3rd person singular -t. Based on these observations, the authors concluded that the agreement system is basically in place at age 2. Very similar ¿ndings were gathered by Rice, Noll, and Grimm (1997), who analyzed the spontaneous speech of German-speaking SLI-children and normally developing controls aged 2;6. The control children produced correct 3rd person singular subject-verb agreement up to 99% of the time. In line with Poeppel and Wexler (1993) as well as Verrips and Weissenborn (1992), they concluded that the agreement system is in place early (around age 2) in German-learning children.1 The conclusion the agreement morphology is acquired early is supported by other researchers investigating different languages (e.g. Phillips 1995; Hoeckstra and Hyams 1998). Concerning the receptive modality, only a few studies that focus on verb inÀection have been conducted. Using the head-turn preference paradigm, Soderstrom, Jusczyk, and Wexler (2002) examined infants’ very early sensitivity to the presence of the 3rd person singular English -s. They presented 19-month olds with sentences that either contained correct subject-verb agreement (At the bakery, the team bakes bread) or incorrect subject-verb agreement (*At the bakery, the team bake bread), where the subject was singular, but the verb was lacking the -s morpheme. They found signi¿cantly longer looking times to the grammatical than to the ungrammatical passages, which was interpreted as indicating early sensitivity to the presence of verbal -s in sentences containing a singular subject. Crucially, children at this age are only starting to use verb inÀection productively. These ¿ndings do not support the hypothesis 1. The discrepant ¿ndings of Clahsen (1986) and Poeppel and Wexler (1993) as well as Rice et al. (1997) can be explained by different ways of proceeding in data analysis. While Clahsen included non¿nite verbs in the analysis and thus counted occurrences of subject plus non¿nite verb as agreement violations, Poeppel and Wexler as well as Rice and colleagues restricted their analyses to utterances containing ¿nite verbs and thus overt inÀected verb forms.

Asymmetries in children’s language performance within and across modalities

177

that the children are actually aware of the manifestation of subject-verb agreement, over the hypothesis that they are responding to low-level distributional properties of the input (Soderstrom 2008). A different line of research has focused on the question of when children are able to associate verb inÀections with the corresponding morphosyntactic categories to infer meaning. In order to accomplish this, researchers have tested children’s use of the verbal inÀection cues in inferring the grammatical number of the sentence subject. Johnson et al. (2005) conducted a two-choice picture selection task, in which the only cue to subject number was the presence or absence of the 3rd person singular -s. To neutralize the number information on the subject, they only used verbs starting with an s-initial consonant cluster, which was coarticulated with the preceding subject noun as usual in rapid speech (e.g. The duck swims in the pond. vs. The ducks swim in the pond.). Each sentence was presented with two pictures, one depicting one actor, and the other one depicting two or more actors performing the action described by the verb (e.g. one vs. two ducks swimming in a pond). The results of the 3- to 6-year old children revealed an interesting picture: only the 5- and 6-year old children performed signi¿cantly above chance-level, but still well below 100% correct. The 3- and 4-year old children did not show any evidence of inferring the subject number by solely relying on the verb inÀection information. The authors concluded that the presence of the 3rd person singular -s is not a transparent marker for subject number for English speaking children. Although the children in this study were not explicitly tested on their production abilities, it should be evident that they are well within the age range in which these agreement morphemes are produced correctly. One explanation put forward by the authors relates the children’s failure to the poverty of the English agreement system in general, in which the 3rd person singular is the only form distinguished on main verbs. This assumption predicts that children acquiring a language with a richer morphological system should show better or earlier comprehension of verb inÀections. Exactly this was tested by Perez-Leroux (2006) in a study with Spanishspeaking children. The same visual and verbal material (translated into Spanish) as in the Johnson et al. (2005) study was used. Sentences without overt subjects were presented, which is possible in Spanish as it is a pro-drop language. This left the verb inÀection as the only cue to subject number, again presented in a singular (Nada en el charco ‘[The duck] swims on the pond’) or in a plural condition (Nadan en el charco ‘[The ducks] swim on the pond’). Spanish-speaking children aged 3- to 6-years were tested in a two-choice picture selection task. The results from the Spanish-speaking children were remarkably similar to the ones obtained from the English-speaking children. While the 3- and 4-year old

178

Oda-Christina Brandt-Kobele and Barbara Höhle

children displayed chance-level performance in both number conditions, the 5- and 6-year old children were better than chance-level, but only in the plural condition. Thus, a richer morphological system does not seem to enhance children’s comprehension of verb inÀection as a marker for subject number (Perez-Leroux 2006). Testing the production-comprehension asymmetries within the same children Three studies have directly compared production and comprehension of verb inÀection, thereby using the same material and testing the same children. Fraser et al. (1963) studied the number contrast as encoded by verb inÀections and auxiliaries. To avoid double number marking, they presented nouns that are invariant for number (e.g. deer, ¿sh). The 3;1- to 3;7-year old children were tested with an elicited production task, an imitation task, and a two-choice picture selection task. The picture pairs used in all three tests differed only in the number of actors performing the action denoted by the verb. Fraser and colleagues found comprehension to be ahead of production, since the children obtained higher performance scores in the comprehension task. But their conclusion was strongly challenged by Johnson et al. (2005) and Keeny and Wolfe (1972), whose close inspection of the data revealed that the comprehension scores did not differ from chance level. Keeny and Wolfe (1972) employed a production, a comprehension, and an imitation task to test English-speaking children between the ages of 3;0 and 4;11. In the imitation task, children were asked to repeat grammatical and ungrammatical sentences, the latter with incorrect subject-verb agreement. Children’s production data was assessed via spontaneous speech samples, and their comprehension of agreement morphemes was assessed in various conditions, among them a picture selection task. The production data revealed more than 90% correct use of verb inÀections. In the imitation task, children repeated signi¿cantly more grammatical than ungrammatical sentences verbatim, while most of the ungrammatical ones were corrected, thereby indicating grammatical sensitivity in the productive domain. The comprehension abilities were examined thoroughly in three conditions, altogether yielding a much lower degree of performance than found in the two productive tasks. In the pictorial condition, children were presented only with the inÀected verb, and had to point to a picture showing either one or two animals. Their performance did not exceed chance-level here. In the sentence condition children were presented with two cues to subject number (the number marked sentence subject plus the agreeing verb) and had to react via picture pointing again. Performance was

Asymmetries in children’s language performance within and across modalities

179

slightly better than chance-level, showing that adding the subject information slightly enhanced performance. In the verbal condition, children were again presented with only the inÀected verb and had to indicate verbally whether one or two animals were performing the mentioned action. Interestingly, children’s performance in this verbal task was better than chance-level as well. This ¿nding provides a ¿rst indication that response type or task demands can inÀuence children’s comprehension performance. In a more recent study, Leonard and colleagues (2000) examined the comprehension of verb agreement morphology in English-speaking SLI- and normally developing children. In one experiment, the children heard either sentences in which the number cues on the verb were the only basis of interpretation (one-cue condition), or sentences in which the subject noun and the verb were overtly inÀected for number (multiple-cue condition). Two groups of normally developing children were tested, one ranging from 4;0 to 7;2 and the other ranging from 4;0 to 5;2. Interestingly, while the older group performed better than chance in both conditions (with better performance in the multiplecue than in the one-cue condition), the younger group exceeded chance performance only in the multiple-cue condition. Altogether, the comprehension abilities of all groups of children were inferior to their production abilities, which were measured using spontaneous speech samples and yielded over 90% correct agreement productions (Leonard et al. 2000). Explanations that have been put forward to account for such late comprehension despite much earlier production deal with the kind of knowledge necessary for each modality. Leonard and colleagues (2000) suggest that a ‘sense’ about the matching of various inÀection forms is suf¿cient for production, but that a more complete understanding of inÀectional forms is necessary for comprehension. This is similar to the knowledge of distributional properties that Soderstrom (2008) supposes to underlie infants’ preference for grammatical sentences in head-turn experiments. Keeny and Wolfe (1972) suggest that grammatically correct performance in spontaneous speech, imitation, and the verbal condition of the comprehension task could be based entirely on syntactic knowledge of agreement, while in the pictorial condition of the comprehension task, semantic knowledge had to be taken into account as well, to relate a linguistic rule to an external referent. So far, the ¿nding that children’s comprehension abilities with respect to verb inÀections lag behind their productive abilities seems to be a robust one, demonstrated within and across different populations of children.

180

Oda-Christina Brandt-Kobele and Barbara Höhle

Task-dependent effects on children’s language comprehension As reported above, Keeny and Wolfe (1972) found different result patterns depending on the type of response the children had to give, with verbal responses being superior to picture pointing reactions. The general ¿nding that the paradigm used to measure language comprehension in children has a high impact on the results obtained will be laid out in the following section. We will focus on discrepancies between comprehension scores gained with traditional methods, like picture pointing, and results obtained with techniques measuring children’s looking behavior, which have been used more intensely in child language research during the last decade (e.g. Fernald et al. 2008; Hirsh-Pasek and Golinkoff 1996; Trueswell 2008; Trueswell and Gleitman 2007; Snedeker 2009). Recent eye-tracking studies on children’s language processing have repeatedly revealed performance levels higher than those found in more traditional ofÀine tasks (e.g. Sekerina et al. 2004). To examine the inÀuence of task dependent factors on children’s ability to exploit verb inÀection as a cue to the number of the sentence subject, we conducted two experiments in our lab that combined an eye-tracking setup employing the preferential looking paradigm with a picture selection task (for a more detailed description of the experiments see Brandt-Kobele and Höhle 2010). In the ¿rst experiment, the pure eye-tracking experiment, German-speaking children aged 3 to 4 years had to watch pictures presented on an eye-tracking screen and listen to test sentences without having to make an overt response. In the second experiment, the children were asked to point to the picture matching the auditorily presented sentence while their eye gaze was tracked. In both experiments, the same pairs of sentences were presented as auditory stimuli. The sentences of a pair differed only in the number of the verb, which was inÀected either for 3rd person singular (Sie fütter-t einen Hund ‘She is feeding a dog’) or for 3rd person plural (Sie fütter-n einen Hund ‘They are feeding a dog’). Number information on the subject noun was avoided by using 3rd person singular female (sie) and 3rd person plural (sie) pronouns, which are homophonous in German. The visual stimuli used in both experiments consisted of pairs of colored pictures, depicting the proposition expressed by the sentences, with the two pictures of a pair differing only in the number of actors performing a given action (one girl vs. two girls). Thus, for the singular inÀected verbs, the 1-actor picture served as target and the 2-actor picture as distractor. For the plural inÀected verbs this was reversed. For the analysis of the eye-tracking data, looking behavior was measured in a baseline phase (before sentence presentation) and in a testing phase (after sentence presentation), both lasting 3 seconds. Between these phases the screen was black for 2 seconds while the

Asymmetries in children’s language performance within and across modalities

181

sentences were presented. In both experiments, 4 practice and 8 testing trials (4 per number condition) were presented on an eye-tracking monitor. In the ¿rst experiment, 28 children aged 3;0 to 4;1 were instructed to watch the screen and listen to the auditorily presented sentences. The children’s looking patterns based on ¿xation durations revealed correct interpretation of the subject number based solely on verb inÀection information.2 In the baseline phase of both number conditions, a strong preference for the 2-actor picture was found. After the presentation of a sentence with a singular inÀected verb, this initial preference vanished and a preference for the one-actor picture emerged. After the presentation of a sentence containing a plural inÀected verb on the other hand the initial preference for the 2-actor even got enhanced. Even when only considering the testing phase, a clear preference for the matching picture was found in both number conditions. This result pattern indicating correct interpretation of the verb inÀection contrasts with the ¿ndings from English- and Spanish-speaking children reported above, who did not show comprehension of verb inÀections before age 5 in a picture selection task (Johnson et al. 2005; Perez-Leroux 2006). To examine the impact of different methodologies, 28 additional children aged 3;2 to 4;4 (mean 3;8) were tested in a combination of a picture selection task with eye-tracking. These children had been instructed to point to the matching picture while their eye gaze was tracked. Their pointing reactions to the target picture in both number conditions did not differ from chancelevel (53% for the singular sentence condition, 48% for the plural sentence condition). The looking pattern in the testing phase itself did not reveal a clear preference for the matching picture in either number condition, but changes in the looking behavior from the baseline to the testing phase depending on the type of sentence were observed. In the baseline phase, a clear preference for the 2-actor picture was visible irrespective of number condition, just as in the aforementioned experiment. This preference decreased sharply in the testing phase when a singular sentence had been presented, but not after the presentation of a plural sentence. Thus, children did not show any evidence for correct comprehension of verb inÀection, when tested with a picture selection task, and even their looking patterns in such a task were less conclusive than the looking patterns in the pure eye-tracking condition from the ¿rst experiment. Altogether, the eye-tracking data are interpreted as providing evidence that German-speaking children aged 3 to 4 years are able to infer the number of an ambiguous sentence subject from the number information given by the verb 2. Only ¿xations on either presented picture that have a minimal duration of 100 ms and a maximal radius of 30 pixels were used to calculate ¿xation durations.

182

Oda-Christina Brandt-Kobele and Barbara Höhle

inÀection. In contrast, no such evidence could be provided in the picture selection task suggesting that this task imposes additional demands on the child, that mask or otherwise interfere with their ability to interpret the sentences. Our results are in line with other studies that ¿nd discrepant comprehension abilities depending on the experimental paradigm used. Sekerina et al. (2004) examined the processing of pronouns in adults and children using pointing reaction, reaction time and eye gaze as dependent variables. Adults pointing reactions indicated that English pronouns are referentially ambiguous. In sentences such as The boy put the box behind him, the pronoun was interpreted as referring to the sentence-internal referent (the boy) in 80% of all cases, while it was interpreted as referring to an external referent (a previously mentioned man) in about 20% of all cases. Adults’ eye-tracking data replicated these ¿ndings by showing that participants detect the referential ambiguity of pronouns online. The 5- to 6-year old children, on the other hand, almost always choose the sentence-internal referent when tested via the picture selection task. Thus, based on these pointing reactions, one would have to assume that Englishspeaking children aged 5 to 6 years do not construct a sentence-external interpretation and thus are not sensitive to the referential ambiguity of English pronouns. But children’s looking behavior revealed a pattern quite similar to the one found in adults. Their eye gaze patterns showed that they were indeed taking an external referent into consideration – a pattern not detectable by just recording ofÀine pointing reactions. Comparable ¿ndings for sentences with focus particles come from Höhle et al. (2009), who examined the comprehension of accented and unaccented auch (‘also’) employing the eye-tracking paradigm. As mentioned earlier, adult-like comprehension of sentences with this focus particle has not been found before the age of 5 or even 7 years (e.g. Hüttner et al. 2004). When Höhle and colleagues tested German children’s comprehension of the focus particle auch within the eye-tracking paradigm, they found ¿xation patterns that reÀect correct processing of these particles in 3-year old children. Another study conducted by Beyer and Hudson Kam (2009) examined the comprehension of the English 3rd person -s inÀection as a tense marker, again employing a picture selection task as well as eye-tracking measures. Test sentences contained the present tense marker -s or the past tense marker -ed and additional control sentences contained time marking adverbs (today, yesterday). In the picture selection task, 7-year old children showed high performance in all conditions, while 6-year olds did not exceed chance-level in the present tense condition. In contrast, the same 6-year olds performed far better than chance for the past-tense condition and the control sentences. Because slower processing with the present tense marker -s was assumed to be at the

Asymmetries in children’s language performance within and across modalities

183

heart of the 6-year olds comprehension problems, a second experiment was conducted, combining a picture selection task with an eye-tracking measure. Concerning the picture selection task, the results from the ¿rst experiment were replicated. The 7-year olds again showed good performance in all testing conditions. The 6-year olds performed at chance-level in the present tense condition, with all other conditions better than chance. But the eye-tracking data revealed the same looking pattern for the 6- and 7-year old children in the present tense condition. Children of both groups looked signi¿cantly more to the target picture, the 7-year olds 600 ms and the 6-year olds 1200 ms after the disambiguating verb inÀection had been presented. The authors interpreted their ¿ndings as being more compatible with the assumption of a processing de¿cit causing the 6-year-olds low performance in the picture selection task rather than a lack of linguistic knowledge. Altogether, studies comparing children’s comprehension abilities using different experimental methods show adult-like patterns with some techniques and performance levels inferior to those of adults with other techniques. Such task-dependent effects have to date been found for three of the four mentioned linguistic domains in which production seems to precede comprehension: pronouns, focus particles and verb inÀections (but not prepositional modi¿ers). Various explanations for method-dependent effects will be presented in the following discussion. Performance asymmetries within and across modalities The factors that are responsible for the discrepant ¿ndings obtained with different experimental methods are still far from being completely understood, but several explanations have been put forward. Before presenting some of these, we want to stress one aspect concerning the complexity of the linguistic material mentioned in this survey. It seems that late comprehension is usually found for linguistic structures that involve some kind of ambiguity. English pronouns are clearly ambiguous, since they can refer to sentence-external or -internal referents, depending on syntactic environment and sentence prosody (Bloom et al. 1994). Restrictive modi¿ers are at least temporarily ambiguous, because their syntactic role only becomes clear later on in the sentence (Trueswell et al. 1999; Hurewitz et al. 2000). Sentences containing focus particles like only and auch are ambiguous as well, because the focus particles can be associated with various constituents in the sentence (e.g. Höhle et al. 2009). Finally, sentences containing only verb inÀections as a cue to subject number are temporarily ambiguous, because the subjects used in the experimental studies are ambigu-

184

Oda-Christina Brandt-Kobele and Barbara Höhle

ous with respect to number. It is not directly clear why ambiguity should result in late comprehension abilities when tested in pointing tasks and better performance when tested indirectly as with eye gaze measures, but pursuing this question more intensely seems worthwhile. In the following, we will review some accounts that have been proposed to explain task-dependent effects on children’s comprehension, culminating in our own explanation. A straightforward account lies in the assumption that task demands per se are responsible for the lack of comprehension found using certain methodologies (e.g. Bloom et al. 1994; Grimshaw and Rosen 1990; Höhle et al. 2009). Task demands are often suggested to be at issue when children show weak performance in such tasks that involve metalinguistic abilities, e.g. in grammaticality judgement tasks. In the case of the picture selection task, which has been widely used in the areas of research considered in this article, this could mean that the pointing task is too demanding for children. It is very unlikely that the cognitive demands imposed by the gesture itself can account for all of the dif¿culties, since infants point at people and objects, usually even before they produce referential speech (e.g. Tomasello, Carpenter, and Liszkowski 2007). But successful performance in a picture selection task demands further abilities, namely storing linguistic and visual information in parallel, comparing different kinds of representations, and making a decision between the pictures presented. All these factors might be less essential in pure eye-tracking tasks. As Gerken and Shady (1996) point out, we do not yet know whether preferential looking and picture pointing tap into the same underlying processes, with preferential looking being merely an easier version of the same task. Considering that picture pointing is widely and successfully used to assess older children’s linguistic knowledge at both lexical and syntactic levels (e.g. TROG-2, Bishop 2003), it is rather implausible that nothing but task demands prevents older children from showing comprehension of linguistic structures they already produce. This explanation includes the hypothesis that children possess the linguistic competence under investigation, but that experiments only tap performance, which is inevitably affected by task demands (Grimshaw and Rosen 1990). A further explanation takes the distinction between competence and performance for granted as well, but views children’s processing capacities and inferential resources as not ef¿cient enough for adult-like comprehension. On this account, the underlying architecture of the children’s parsing mechanism is considered to be adult-like, and the difference is supposed to lie in the ef¿ciency of a child’s parser to integrate and use pragmatic and discourse information in sentence comprehension (Goodluck 1990; Sekerina et al. 2004). The fact that children avoid revising an initial commitment to a sentence parse

Asymmetries in children’s language performance within and across modalities

185

or interpretation is called on to explain divergent ¿ndings across response measures (e.g. Lorsbach et al. 1998). Children might take an alternative interpretation of a sentence into consideration during sentence comprehension, as seen in eye-tracking data, but because of limited processing capacities they only entertain one interpretation which ¿nally guides their ultimate reaction (Sekerina et al. 2004; Hurewitz et al. 2000). The consideration that children have dif¿culties integrating discourse information into the process of sentence comprehension is challenged by ¿ndings that show enhanced performance when suf¿cient discourse information is provided (Gualmini et al. 2003; Crain and Fodor 1993). The idea that different interpretations are tapped via on-line and off-line measures will be further explored in the windup of this section. Another explanation for task-dependent effects on children’s comprehension relates to the concepts of cue validity and cue cost. The Competition Model of sentence processing (Bates and MacWhinney 1987; Bates, Devescovi, and Wulfeck 2001) views language comprehension as a process whereby multiple cues are integrated during sentence interpretation. Cue validity refers to the information value of a linguistic form, while cue cost refers to the amount and type of processing associated with the activation of that form. Importantly, cue validity and cue cost can change over time. Thus it may well be that adults can make use of a speci¿c cue in deriving an intended meaning, while that same cue is not utilized by children. This would not necessarily mean that children do not know or have not acquired this speci¿c linguistic structure, but that they cannot use it as a reliable cue for sentence comprehension. For example, Beyer and Hudson Kam (2009) observe that children fail to infer the tense of the clause from the presence English present tense marker -s, despite their sensitivity to the verb inÀection as revealed by their eye-gaze data. They propose that the -s is only a weakly reliable cue to tense in English. Similarly, Johnson et al. (2005) suggest that verb inÀection as an agreement marker does not survive to the syntactic level of LF (Chomsky 1995) and is thus not available for sentence comprehension in children. Adults and older children who are able to use verb inÀection as a cue for interpreting the subject number are thought to make recourse to metalinguistic abilities in the ful¿llment of the task. The general idea that unreliable cues are not employed in sentence interpretation in young children is also maintained by Trueswell et al. (1999) and Hurewitz et al. (2000). Closely linked to this explanation is one that calls on heuristics which are applied in sentence comprehension processes. Townsend and Bever (2001) as well as Ferreira (2003) claim that adults process complex sentences twice, once syntactically, based on the structural properties of a given sentence, and once heuristically, based on probabilistic knowledge of canonical form-meaning

186

Oda-Christina Brandt-Kobele and Barbara Höhle

relations in a given language. With the latter kind of processing they only come up with gross interpretations, which are only by chance correct. According to Ferreira (2003) heuristics come into play especially in cases where sentences are complex, where no further contextual support for interpretation is given, and where a decision has to be made with limited time and resources. Adults’ misinterpretation of sentences which are observed in ambiguous or noncanonical structures like passives (e.g. Ferreira 2003) are considered as being a product of the parallel work of the two processing mechanism. If the heuristically based mechanism overrules the syntactic one – which may be the case in syntactically more complex sentences – then misinterpretations can result. We suggest that sentence processing models which view language comprehension as a process of integrating structural and statistical knowledge about language (as proposed by Townsend and Bever (2001) and Ferreira (2003)) can also account for the discrepancies across different tasks that we have found in children’s sentence comprehension performance. The employment of heuristics in children’s sentence processing has already been proposed by Clark and Hecht (1983), and is also suggested by Hurewitz et al. (2000). Assuming that children’s sentence interpretation is guided by structural as well as by heuristic interpretation mechanisms, we think that discrepancies across different tasks can be explained so that speci¿c task demands reveal the inÀuence of these two mechanisms in different ways. Children’s eye-gaze data with longer ¿xations on the target picture suggest that – at least temporarily – a correct interpretation of the sentence based on the structural information provided by the input is constructed. But when children have to point to the corresponding picture, their response does not necessarily seem to be guided by that interpretation. One way to account for this would be that in accomplishing the picture pointing task, the outcomes of the heuristic parsing mechanism come into play. These may interfere with the outcomes of the syntactic mechanism if both arrive at conÀicting interpretations. The temporal relation between the two measurements seems to be relevant here. While the eye-gaze data as an on-line measure are in close temporal proximity to the processing of the input, the pointing task as an off-line task lags behind the processing mechanism. Possibly this temporal delay – which is also caused by the additional metalinguistic aspects of the pointing task – opens the space for the interference of heuristic interpretations. Considered in this way, our data suggest that children’s capacities for an analysis of the linguistic structure of a sentence is comparable to those of adults, but that heuristically based interpretations step more heavily into the process of sentence interpretation when they are given suf¿cient time.

Asymmetries in children’s language performance within and across modalities

187

These considerations also provide an account for the asymmetries observed across children’s production and comprehension skills. If these heuristics are considered to be speci¿c for the processing mechanisms that are involved in the decoding of speech, but do not enter into the mechanisms that are involved in language encoding during production, no conÀicts between heuristics and linguistic structure should occur in the productive modality. This would suggest that there may be conditions in which children’s productions can reveal their underlying structural syntactic competence in a more direct way than can their comprehension skills. This does not mean that children’s production capacities cannot be affected by heuristics and by task demands at all. Actually, in the area of the production of verb inÀection the studies reported in this paper suggest that performance in elicited speech production tasks can underestimate children’s spontaneous language performance (Brown 1973; Rice and Wexler 2002). Additional evidence for the effects of task demands in production is for example provided by Crain and Fodor (1993) who showed that children’s ability to produce by-passives increases when provided with appropriate discourse conditions. Thus, performance asymmetries within modalities have to be taken into account when asymmetries between comprehension and production are discussed. This is necessary because the asymmetries found within modalities challenge the determination of the age of acquisition of various linguistic structures, which in turn forms the basis of the postulated cross-modal asymmetries. Considering this, cross-modality asymmetries might actually just be an artifact of the different methods employed. But furthermore, the knowledge about within-modality asymmetries helps to determine whether asymmetries between comprehension and production can truly tell us something about children’s underlying grammatical system, and whether this system is different for comprehension and production.

Conclusion To summarize, we suggest that cross-modal asymmetries in children’s language processing do not provide evidence that the underlying grammatical systems for production and comprehension develop in an asymmetric fashion. The asymmetries observed so far have to be carefully interpreted, thereby taking into account the kind of data and the procedures used to elicit them. A full understanding of the development of linguistic competence in children has to integrate considerations about the development of processing mechanisms, which may have different characteristics in production and comprehension.

188

Oda-Christina Brandt-Kobele and Barbara Höhle

This has to include a closer inspection and evaluation of the speci¿c demands that experimental procedures typically used in language acquisition research impose on the children’s processing mechanisms, an understanding of the characteristics of children’s processing mechanisms, which may still deviate from those employed by adults, and of their effects on children’s performance in a given task. Incorporating a wider range of techniques into language acquisition research, like eye-tracking or electrophysiological methods, which are suitable to tap into the mechanisms involved in speech processing, will provide us with a more complete picture of the child’s development into a competent language producer and comprehender.

References Bates, Elizabeth, and Brian MacWhinney 1987 Competition, variation, and language learning. In Mechanisms of language acquisition, Brian MacWhinney (ed.), 157 – 194. Hilsdale, NJ: Lawrence Earlbaum. Bates, Elizabeth, Philip S. Dale, and Donna J. Thal 1995 Individual differences and their implications. In The handbook of child language, Paul Fletcher and Brian McWhinney (eds.), 96 – 151. Blackwell, Oxford. Bates, Elizabeth, Antonella Devescovi, and Beverly B. Wulfeck 2001 Psycholinguistics: A cross-language perspective. Annual Review of Psychology 52: 369 – 398. Beyer, Tim, and Carla L. Hudson Kam 2009 Some cues are stronger than others: The (non)interpretation of 3rd person present -s as a tense marker by 6- and 7-year-olds. First Language 29 (2): 208 – 227. Brandt-Kobele, Oda-Christina, and Barbara Höhle 2010 What asymmetries within comprehension reveal about asymmetries between comprehension and production: The case of verbal inÀection in language acquisition. Lingua 120: 1910 – 1925. Brown, Roger 1973 A ¿rst language: The early stages. Cambidge, MA: Harvard University Press. Chien, Yuchin, and Kenneth N. Wexler 1990 Children’s knowledge of locality conditions on Binding as evidence for modularity of syntax and pragmatics. Language Acquisition 13: 225 – 295. Chomsky, Noam 1995 The minimalist program. Cambridge, MA: MIT Press.

Asymmetries in children’s language performance within and across modalities

189

Clahsen, Harald 1986 Verb inÀections in German child language: Acquisition of agreement markings and the functions they encode. Linguistics 24: 79 – 121. Clark, Eve Vivienne, and Barbara F. Hecht 1983 Comprehension, production, and language acquisition. Annual Review of Psychology 34: 325 – 349. Crain, Stephen, and Janet D. Fodor 1993 Competence and performance in child language. In Language and Cognition: A Developmental Perspective, Esther Dromi (ed.), 141 – 171. Norwood, NJ: Ablex. De Villiers, Jill, Jacqueline Cahillane, and Emily Altreuter 2004 What can production reveal about Principle B?. In Proceedings of GALANA 2004, Generative Approaches to Language Acquisition North America, Kamil Ud Deen, Jun Nomura, Barbara Schulz, and Bonnie D. Schwartz (eds.), 89 – 100. University of Connecticut, Occasional Papers in Linguistics, 4. Fernald, Anne, and Virginia Marchman 2006 Language learning in infancy. In Handbook of Psycholinguistics, 2nd Ed, Matthew J. Traxler, and Morton A. Gernsbacher (eds.), 1027 – 1071. San Diego, CA: Academic Press; 2006. Fernald, Anne, Renate Zangl, Adriana Medina-López-Portillo, and Virginia A. Marchman 2008 Looking while listening: Using eye movements to monitor spoken language comprehension by infants and young children. In Developmental Psycholinguistics: On-line methods in children’s language processing, Irina A. Sekerina, Eva M. Fernandez, and Harald Clahsen (eds.), 97 – 135. Amsterdam: John Benjamins. Fraser, Colin, Ursula Bellugi, and Roger Brown 1963 Control of grammar in imitation, comprehension and production. Journal of Verbal Learning and Verbal Behaviour 2: 121 – 135. Gerken, LouAnn, and Michelle E Shady 1996 The picture selection task. In Methods for assessing children’s syntax, Dena McDaniel, Cecile McKee, and Helen Cairns (eds.), 211 – 235. Cambridge, MA: MIT Press. Goldin-Meadow, Susan, Martin E. Seligman, and Rochel Gelman 1976 Language in the two-year-old. Cognition 4: 189 – 202. Goodluck, Helen 1990 Knowledge integration in processing and acquisition: Comments on Grimshaw and Rosen. In Language Processing and Language Acquisition, Lyn Frazier and Jill de Villiers (eds.), 369 – 382. Dordrecht: Kluwer Academic Publsiher.

190

Oda-Christina Brandt-Kobele and Barbara Höhle

Gualmini, Andrea, Simona Maciukaite, and Stephen Crain 2003 Children’s insensitivity to contrastive stress in sentences with ONLY. In Proceedings of PLC 25 (PWPL 8.1), Sudha Arunachalam, Elsi Kaiser, and Alexander Williams (eds.), 71 – 110. Philadelphia: University of Pennsylvania. Guasti, Maria Teresa 1993 Verb syntax in Italian child grammar: Finite and non¿nite verbs. Language Acquisition 3 (1): 1 – 40. Guasti, Maria Teresa 2002 The growth of grammar. Cambridge, MA: MIT Press. Hendriks, Petra, and Jennifer K. Spenader 2006 When production precedes comprehension: An optimization approach to the acquisition of pronouns. Language Acquisition 13 (4): 319 – 349. Hirsh-Pasek, Kathy, and Roberta M. Golinkoff 1996a The origins of grammar: Evidence from early language comprehension. Cambridge, MA: MIT Press. Hoekstra, Teum, and Nina Hyams 1998 Aspects of root in¿nitives. Lingua 106: 81–112. Höhle, Barbara, Michaela Schmitz, Lynn M. Santelmann, and Jürgen Weissenborn 2006 The recognition of discontinuous verbal dependencies by German 19month-olds: Evidence for lexical and structural inÀuences of children’s early processing capacities. Language Learning and Development 2: 277 – 300. Höhle, Barbara, Frauke Berger, Anja Müller, Michaela Schmitz, and Jürgen Weissenborn 2009 Focus particles in children’s language. Production and comprehension of auch ‘also’ in German Learner from 1 year to 4 years of age. Language Acquisition 16: 36 – 66. Hornby, Peter A., and Wilbur A. Haas 1970 The use of contrastive stress in children. Journal of Speech and Hearing Research 13, 395 – 399. Hüttner, Tanja, Heiner Drenhaus, Ruben van de Vijver, and Jürgen Weissenborn 2004 The acquisition of the German focus particle auch ‚too’: Comprehension does not always precede production. Proceedings of the 28th Annual Boston University Conference on Language Development. On-line supplement http://www.bu.edu/linguistics/APPLIED/BUCLD/supp.html. Johnson, Valerie E., Jill de Villiers, and Harry N. Seymour 2005 Agreement without understanding? The case of third person singular /s/. First Language 25(3): 317 – 330. Kenney, Terrence J., and Jean Wolfe 1972 The acquisition of agreement in English. Journal of Verbal Learning and Verbal Behaviour 11: 698 – 705.

Asymmetries in children’s language performance within and across modalities

191

Lebeaux, David 1988 Language acquisition and the form of grammar. Ph.D. dissertation, University of Massachusetts at Amherst. Leonard, Laurence B., Umberta Bortolini, M. Christina 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 B., Carol A. Miller, and Amanda J. Owen 2000 The comprehension of verb agreement morphology by English-speaking children with speci¿c language impairment. Clinical Linguistics and Phonetics 14 (6): 465 – 481. Naigles, Letitia 1996 The use of multiple frames in verb learning via syntactic bootstrapping. Cognition 58: 221 – 251. Nederstigt, Ulrike 2003 Auch and noch in child and adult German. Berlin/New York: Mouton de Gruyter. Perez-Leroux, Ana T. 2005 Number problems in children. In Proceedings of the 2005 Canadian Linguistics Association Annual Conference. Claire Gurski (ed.) http://ling.uwo.ca/publications/CLA-ACL/CLA-ACL2005.htm. Phillips, Colin 1995 Syntax at age two: Crosslinguistic differences. MIT Working Papers. Poeppel, David, and Kenneth N. Wexler 1993 The full competence hypothesis of clause structure in early German. Language 69: 1 – 33. Radford, Andrew 1990 Syntactic theory and the acquisition of English syntax. Cambridge, MA: Blackwell. Rice, Mabel L., Karen Ruff Noll, and Hannelore Grimm 1997 An extended optional in¿nitive stage in German-speaking children with speci¿c language impairment. Language Acquisition 6 (4): 255 – 295. Rice, Mabel L., and Kenneth N. Wexler 2002 Test of early grammatical impairment. San Antonio, TX: Psychological Corporation. Santelmann, Lynn M., and Peter W. Jusczyk 1998 Sensitivity to discontinuous dependencies in language learner: Evidence for limitations in processing space. Cognition 69: 105 – 134. Sekerina, Irina A., Karin Stromswold, and Arild Hestvik 2004 How do adults and children process referentially ambiguous pronouns. Journal of Child Language 31 (1): 123 – 152.

192

Oda-Christina Brandt-Kobele and Barbara Höhle

Smolensky, Paul 1996 On the comprehension/production dilemma in child language. Linguistic Inquiry 27: 720 – 732. Soderstrom, Melanie, Kenneth N. Wexler, and Peter W. Jusczyk 2002 English-learning toddler’s sensitivity to agreement morphology in receptive grammar. In Proceedings of the 26th Annual Boston University Conference on Language Development, Barbora Skarabella, Sarah Fish, and Anna H.-J. Do (eds.), 643 – 652. Somerville, MA: Cascadilla Press. Soderstrom, Melanie 2008 Early perception–late comprehension of grammar? The case of verbal –s: A response to de Villiers and Johnson (2007). Journal of Child Language 35: 671 – 676. Song, Jae Yung, Megha Sundara, and Katherine Demuth 2009 Effects of phonology on children’s production of 3rd person singular -s. Journal of Speech, Language and Hearing Research 52 (3): 623 – 642. SzendrĘi, Kriszta 2004 Focus and the interaction between syntax and pragmatics. Lingua 114 (3): 229 – 254. Tomasello, Michael, Marlinda Carpenter, and Ulf Liszkowski 2007 A new look at infant pointing. Child Development 78 (3): 705 – 722. Trueswell, John C. 2008 Using eye movements as a developmental measure within psycholinguistics. In Developmental Psycholinguistics: On-line methods in children’s language processing, Irena A. Sekerina, Fernandez, Eva M., and Clahsen, Harald (eds.), 73 – 96. Amsterdam: John Benjamins. Trueswell, John C. and Lila Gleitman 2007 Learning to parse and its implications for language acquisition. In Oxford Handbook of Psycholinguistics, Gareth Gaskell (ed.), 635 – 656. Oxford: Oxford University Press. Trueswell, John C., Irina A Sekerina, Nicole M. Hill, and Marian L. Logrip 1999 The kindergarden-path effect: Studying on-line sentence processing in young children. Cognition 73: 89 – 134. Verrips, Maaike, and Jürgen Weissenborn 1992 Routes to verb placement in early German and French: The independence of ¿niteness and agreement. In The acquisition of verb placement, Jürgen Meisel (ed.), 283 – 331. Dordrecht: Kluwer Academic Publishing.

Adults’ on-line comprehension of object pronouns in discourse* Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

Introduction A well-known ¿nding in the literature on language acquisition is that Englishspeaking children as old as 6 frequently misinterpret object pronouns as co-referring with the local referential subject. The percentage of errors with respect to this so-called Delay of Principle B-Effect, however, varies substantially across studies, even when studies are considered that investigate the same language. Conroy et al. (2009) showed that in English the Delay of Principle B-Effect disappears when an elaborate context is presented in which the correct referent and the correct sentence interpretation are made accessible. They conclude from this that English-speaking children possess knowledge of Principle B but are hindered by a discourse context in which the potential referents and interpretations are not appropriately balanced. A similar disappearance of the Delay of Principle B-Effect was shown for Dutch by Spenader, Smits, and Hendriks (2009). However, rather than presenting children with an elaborate context, they used a short introductory sentence that unambiguously established the correct referent as the discourse topic. They * Contact information Petra Hendriks, Jacolien van Rij, Gisi Cannizzaro and John Hoeks: CLCG, University of Groningen, P.O. Box 716, 9700 AS Groningen, The Netherlands, [email protected], [email protected], [email protected], [email protected]. Contact information Arina Banga: CLS, Radboud University Nijmegen, P.O. Box 9103, 6500 HD Nijmegen, The Netherlands, arina.banga@ mpi.nl. This investigation was supported by a grant from the Netherlands Organisation for Scienti¿c Research, NWO, awarded to Petra Hendriks (grant no. 277-70005) for the VICI project “Asymmetries in Grammar”. The authors thank Robbert Prins and Petra van Berkum for drawing the pictures used in the experiment, and Colin Phillips, Jennifer Spenader, the editors Cornelia Hamann and Esther Ruigendijk, and three anonymous reviewers for their useful comments and discussion.

194

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

interpret their results as indicating that children’s grammar underdetermines the interpretation of pronouns and conclude that Principle B is not a rule of the grammar, but rather is a derived effect resulting from a mature hearer’s ability to consider the perspective of the speaker. Because children’s interpretations only appear to conform to Principle B if the discourse structure provides a clear topic, Spenader et al. further conclude that children’s comprehension of pronouns is sensitive to discourse structure and that children are helped by a coherent discourse. As their analysis is formulated within the framework of Optimality Theory (OT), the relevant discourse conditions are integrated in the grammar. The different and partly contradictory conclusions of these two studies raise questions regarding English and Dutch children’s knowledge of Principle B and the exact contribution of discourse context to pronoun interpretation. The aim of the present study is to shed more light on the second issue by performing an eyetracking study with Dutch adults. In this study, we test adults’ comprehension of object pronouns and reÀexives while manipulating the discourse context. This paper is organized as follows. First, we discuss the off-line studies of Conroy et al. and Spenader et al. with children in more detail. On the basis of these studies, we formulate predictions with respect to the off-line and on-line behavior of adults. We then discuss our eyetracking study with Dutch-speaking adults and present the results of our study. These results are discussed in the light of the formulated hypothesis and predictions. Finally, the implications of our results for the study of anaphora in child language are discussed.

Theoretical background Delay of Principle B-Effect Many experiments in various languages have established that children who correctly interpret reÀexives from the age of four or ¿ve have trouble interpreting pronouns correctly until the age of 6;6 or even later (e.g. Chien and Wexler 1990). Consider the following examples: (1)

This is Mama Bear. This is Goldilocks. Is Mama Bear washing herself?

(2)

This is Mama Bear. This is Goldilocks. Is Mama Bear washing her?

Adult’s on-line comprehension of object pronouns in discourse

195

Children consistently interpret herself in (1) as referring to Mama Bear, thereby showing knowledge of Principle A of Binding Theory, which governs the use and interpretation of reÀexives. At the same time, when presented with (2), the same children frequently choose Mama Bear as the referent for her. This suggests that they do not yet have knowledge of Principle B of Binding Theory, which governs the use and interpretation of pronouns. This pattern in children’s responses is often referred to as the Delay of Principle B-Effect. Discourse matters Many studies have tried to provide an explanation for the observed asymmetry between children’s behavior with Principle A and their behavior with Principle B. One strategy, accepting the observed asymmetry, is to argue that the cause for children’s errors with pronouns lies outside the grammar, for example in their lack of pragmatic knowledge (Chien and Wexler 1990; Grodzinsky and Reinhart 1993; Thornton and Wexler 1999) or insuf¿cient processing resources (Reinhart 2006). Another strategy is to accept this asymmetry but argue that the cause for the asymmetry lies in the properties of the grammar, rather than in extra-grammatical aspects of comprehension (Hendriks and Spenader 2005/6). A third strategy, adopted by Conroy et al. (2009), is to argue that the observed asymmetry between pronouns and reÀexives largely is a reÀection of shortcomings of the experimental tests used. Conroy et al. (2009) carried out three experiments to test the validity of the Delay of Principle B-Effect as well as the widely assumed asymmetry between quanti¿ed antecedents and referential antecedents. The children in their experiments, in which they employed a truth value judgment task (TVJT), watched an experimenter act out a story with props, and then had to judge whether a statement about the story produced by a puppet, such as I think that Grumpy painted him, was true or not. The stories were constructed in such a way as to satisfy a number of conditions. First, a potential antecedent for the coreferential interpretation as well as the disjoint interpretation should be available in the discourse (the Availability Assumption). Second, the story should make the correct disjoint interpretation of the pronoun a genuine potential outcome at some point (the Disputability Assumption). Under these conditions, children made very few errors in the ¿rst experiment, accepting the coreferential interpretation in only 11% of trials. Conroy et al. argue that this provides evidence that children know Principle B. In a second experiment, where the pronoun him was replaced by the possessive noun phrase his costume, children were found to accept reference to the subject in 80% of the trials. According to Conroy et al., this discon¿rms the idea that

196

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

children did so well on the ¿rst experiment because they have a general dispreference for bound variable interpretations of pronouns. Reintroducing some of the shortcomings of previous experiments such as Thornton and Wexler’s (1999) in the third experiment, children’s percentage of incorrect coreferential interpretations increased to 56%. Conroy et al. argue that the results of these three experiments show that children have knowledge of Principle B but are hindered by a context that is not suitably balanced. Although children seem to know and generally respect Principle B, they do make more errors in experiments (in roughly 15-30% of trials) than would be expected if Principle B acted as a strong constraint on children’s interpretations. This ‘residual’ Delay of Principle B-Effect is a real effect, Conroy et al. claim, and may be related to the recent ¿nding in on-line studies of pronoun resolution in adults that adults temporarily consider ungrammatical coreferential antecedents in Principle B contexts (Badecker and Straub 2002; Kennison 2003; Runner, Sussman, and Tanenhaus 2003). If these results, obtained by eyetracking and self-paced reading measures, are correct, then Principle B acts as a late ¿lter on the processing of pronouns, after the grammar has generated multiple interpretations (but see Nicol 1988, who did not ¿nd evidence for a coreferential interpretation of the pronoun in her cross-modal priming study; see also Nicol and Swinney 1989). Because children ¿nd it more dif¿cult than adults to inhibit an initial but incorrect interpretation, Conroy et al. argue, this may make them prone to error in their interpretation of pronouns. Like Conroy et al., Spenader et al. (2009) recognize the importance of the discourse context for a correct assessment of children’s knowledge of Principle B. However, rather than implementing the Disputability Assumption in the test materials in the form of an elaborate context story, Spenader et al. implemented this condition as part of their experimental design. In their experiment with Dutch children, the child was told by a puppet that the computer had been built by the experimenter, but the puppet believed that the computer was built wrong. The child was then asked to help repair the computer. So it was made plausible that the pictures and sentences in their picture veri¿cation task could – but might not – match. This design allowed Spenader et al. to focus on the other condition that Conroy et al. argue to be crucial in investigating the Delay of Principle B-Effect, namely the Availability Assumption. In Conroy et al.’s study, the coreferential referent and the disjoint referent were introduced and subsequently referred to in a very elaborate story context, in which these two referents interacted with several other characters. This makes it very dif¿cult to determine the relative salience of the two referents. For this reason, Spenader et al. chose to compare a classic but rather unnatural introduction of the two potential referents, as in Chien and Wexler’s (1990) study, with an introduc-

Adult’s on-line comprehension of object pronouns in discourse

197

tion that is more coherent in terms of Centering Theory (Grosz, Joshi, and Weinstein 1995): Classic Condition: (3) Here you see an elephant and an alligator. The elephant is hitting him/himself. Single Topic Condition: (4) Here you see an alligator. The elephant is hitting him/himself.

In the Classic Condition, the coreferential and the disjoint referent are introduced in a conjunction and can therefore be taken to be equally salient. As a result, the structure of the linguistic discourse does not provide the listener with any clues as to which of these two referents is to be preferred as the topic of the second sentence, i.e. the test sentence. In the Single Topic Condition, only the disjoint referent is introduced in the ¿rst sentence. As a result, only this referent is a potential topic of the test sentence, according to the de¿nitions of Centering Theory. Furthermore, one of the rules of Centering Theory posits that if there is only one pronoun present in the utterance, this pronoun refers to the topic of the utterance (or backward-looking center, in the terminology of Grosz et al. 1995: 214). Thus the discourse structure is neutral regarding the interpretation of the pronoun in (3) but promotes the correct interpretation of the pronoun in (4), independently of Principle B. Overall, in Spenader et al.’s study children’s comprehension of reÀexives was signi¿cantly better than their comprehension of pronouns, consistent with the existence of a Delay of Principle B-Effect. But whereas children’s comprehension of reÀexives was similar across conditions, the Delay of Principle B-Effect was only observed with pronouns in the Classic Condition (31% errors, compared to 14% errors with reÀexives in this condition, which was signi¿cantly different). In the Single Topic Condition, the Delay of Principle B-Effect had disappeared completely. Children’s comprehension of pronouns in this condition (17% errors) was not signi¿cantly different from their comprehension of reÀexives (18% errors). Spenader et al. conclude from these results that, for children, pronouns can receive a coreferential as well as a disjoint meaning if the inherent bias of a natural – coherent – discourse context is neutralized. From this, it follows that children are not yet able to apply Principle B. This explanation is compatible with the optimality theoretic (OT) account of the Delay of Principle B-Effect proposed in Hendriks and Spenader (2005/6). According to this OT account, Principle B is not a constraint of the grammar but rather is a derived or emergent effect. Principle B emerges when hearers

198

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

optimize bidirectionally and also consider the speaker’s perspective. The OT account formalizes the reasoning that if the speaker would have wanted to express a coreferential meaning, the best option for the speaker according to the constraints of the grammar would have been to use a reÀexive. If the speaker did not use a reÀexive but rather used a pronoun, the hearer may conclude that the coreferential meaning is not the meaning intended by the speaker, and hence block this interpretation for the pronoun. Assuming that children are not yet able to take into account the speaker’s perspective in their linguistic optimization (cf. de Hoop and Krämer 2005/6; Hendriks and Spenader 2005/6), this explains why children do not discard the coreferential interpretation for a pronoun. At the same time, the OT grammar predicts that children’s production of pronouns in the Classic Condition is adult-like. This latter prediction was con¿rmed by the results of Spenader et al.’s elicited production task.1 So, why do children perform so much better with pronouns in the Single Topic Condition than in the Classic Condition? In the Single Topic Condition, as in all other contexts where the disjoint referent is much more salient than the coreferential referent, a general preference for pronouns to refer to the discourse topic may facilitate selection of the disjoint referent over the coreferential referent. Spenader et al. suggest that this preference is incorporated in the grammar as a violable constraint, which we refer to here as PROTOP. Assuming that the more highly ranked constraints of the grammar do not decide between a coreferential and a disjoint meaning for a pronoun, this weaker constraint PROTOP will determine the selection of the antecedent. In optimality theoretic 1. The reviewers suggest that a potential problem for the analyses of both Conroy et al. and Spenader et al. may be the observed cross-linguistic differences in the Delay of Principle B-Effect. Although the Delay of Principle B-Effect has been observed in simple transitive sentences in English and Dutch, this effect does not seem to arise in similar sentences in the closely related language German (see Ruigendijk, 2008). These cross-linguistic differences are problematic for Conroy et al.’s account, which appeals to processing factors for the ‘residual’ Delay of Principle B-Effect effect. It is highly unlikely that speakers of German are better at inhibiting incorrect interpretations than speakers of Dutch or English. However, crosslinguistic differences do not seem to be a priori incompatible with Spenader et al.’s OT account of the Delay of Principle B-Effect. In OT, languages are assumed to differ as a result of a different ranking of the constraints of the grammar, which may give rise to different lexical inventories and different locality conditions on reÀexive binding. Because acquisition delays are predicted to arise in those cases where unidirectional optimization yields a different result than bidirectional optimization when using the same grammar, different grammars may yield different predictions regarding the Delay of Principle B-Effect.

Adult’s on-line comprehension of object pronouns in discourse

199

terms, this is known as ‘the emergence of the unmarked’: usually only the effects of stronger constraints are visible within a language, but in particular circumstances a weaker constraint becomes crucial. Because PROTOP promotes selection of the discourse topic as the antecedent of a pronoun, its application results in selection of the correct antecedent in the Single Topic Condition in Spenader et al.’s study. This is because, in their study, the discourse topic introduced in the ¿rst sentence always is the correct antecedent. In the Classic Condition, on the other hand, the constraint PROTOP will lead to guessing behavior. Because the coreferential and the disjoint referent are equally salient in the ¿rst sentence, PROTOP will not be able to decide between these two referents. Hence, children will simply select one of these referents at chance. In contrast to children, adults are not dependent on the presence of contextual cues to arrive at the correct interpretation of a pronoun. Rather, they are able to discard the coreferential meaning by reasoning about the speaker’s alternative linguistic options. Therefore, they select the disjoint referent also in the Classic Condition. Adults’ processing of pronouns According to Conroy et al. (2009), there are two different sources for children’s Delay of Principle B-Effect errors: (1) an unbalanced discourse context, and (2) a failure to inhibit the initially activated but incorrect coreferential interpretation. Only children are hindered by an unbalanced context. Adults have no problems applying Principle B in these situations. Conroy et al. do not provide an explanation for why adults cannot be contextually coerced into an ungrammatical interpretation. However, an obvious explanation, which is compatible with their account of the Delay of Principle B-Effect, is that adults’ application of Principle B is independent of discourse context. Spenader et al. (2009), on the other hand, argue that Principle B is a derived effect which requires that hearers take into account the speaker’s perspective. Although children are unable to derive Principle B, a preference for the pronoun to refer to the discourse topic may nevertheless direct children toward the correct interpretation. This preference is argued to have the form of a weak constraint that is part of the grammar. But if this constraint is part of the grammar, its effects may show up in adults’ on-line processing as well. On the basis of these two different positions regarding the Delay of Principle B-Effect, we can formulate a number of predictions regarding adults’ and children’s on-line processing of pronouns. Both for adults and for children, we predict that pronouns are initially ambiguous and activate the coreferential as well as the disjoint referent. This is in line with results of earlier process-

200

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

ing studies (Badecker and Straub 2002; Kennison 2003; Runner, Sussman, and Tanenhaus 2003) as well as with the theoretical assumptions of the two accounts under consideration. If a preference for pronouns to refer to the topic is a constraint of the grammar, as Spenader et al. contend, we expect the effects of this preference not to be con¿ned to children’s off-line responses. Rather, this preference may also show up in children’s and adult’s on-line processing. In particular, we predict that children as well as adults will experience more processing dif¿culty interpreting pronouns in the Classic Condition than in the Single Topic Condition. Because this preference only pertains to the interpretation of pronouns, we expect no effects of context when they process reÀexives. In contrast, a prediction that seems compatible with Conroy et al.’s account of the Delay of Principle B-Effect is that effects of context are observable neither in adults’ on-line processing of reÀexives nor in their on-line processing of pronouns. In this study, we focus on adults’ on-line processing and investigate the predictions regarding adults’ on-line comprehension of pronouns and reÀexives. In the next section, we describe the details of our eyetracking experiment with Dutch adults, in which we use similar materials to Spenader et al. (2009).

Experiment Participants Twenty-¿ve adult native speakers of Dutch participated in this study. Most of them were university students, who participated voluntarily. One participant was excluded from data analysis because his other native language was Frisian, and pronouns in Dutch and Frisian have different properties. Therefore, our analyses are based on the data of the resulting 24 participants: 13 men and 11 women (mean age 22 years, age range 18 – 27). Materials and design Each item consisted of a pre-recorded sequence of two sentences in combination with a picture. The sequences of sentences consisted of an introductory sentence followed by the test sentence. The pictures displayed two animals of approximately equal size engaged in a self-oriented or other-oriented action (see Figure 1). The recorded sentences were spoken by a female student, who had a neutral Dutch accent. The ¿rst sentence of each sentence pair served as an introduction of the character(s) and the setting, whereas the second sentence contained an

Adult’s on-line comprehension of object pronouns in discourse

201

anaphor which referred back to one of the two referents introduced in the previous sentence, or to the single referent introduced in the previous sentence. The following verbs were used in the second sentence, which was the test sentence: aankleden ‘to dress’, bijten ‘to bite’, kietelen ‘to tickle’, schilderen ‘to paint’/ tekenen ‘to draw’, slaan ‘to hit’, vastbinden ‘to tie’, wijzen naar ‘to point at’ and schminken ‘to make up’.

Figure 1. A picture displaying a self-oriented action (left) and a picture displaying another-oriented action (right), adapted to black and white print.

Two factors were manipulated in the experimental items. The ¿rst factor was Type of Introductory Sentence: The introductory sentence introduced both animals (Classic Condition, or C) or only one of the animals (Single Topic Condition, or S). The second factor was Type of Anaphor: The sentences contained a pronoun (P) or reÀexive (R). An example of each of the four resulting experimental conditions is given in (5) – (8). Classic Condition+Pronoun (CP): (5) Een aap en een schildpad zijn op het strand. De aap kietelt hem. ‘A monkey and a turtle are on the beach. The monkey is tickling him.’ Classic Condition+ReÀexive (CR): (6) Een aap en een schildpad zijn op het strand. De aap kietelt zichzelf. ‘A monkey and a turtle are on the beach. The monkey is tickling himself.’ Single Topic Condition+Pronoun (SP): (7) Een schildpad is op het strand. De aap kietelt hem. ‘A turtle is on the beach. The monkey is tickling him.’ Single Topic Condition+ReÀexive (SR): (8) Een schildpad is op het strand. De aap kietelt zichzelf. ‘A turtle is on the beach. The monkey is tickling himself.’

Four versions of the experiment were constructed by a Latin square design, so that each list contained four items of each condition and one version of each

202

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

item. This was done by using each of the 8 verbs twice, but with different pairs of animals. As a result, each version of the experiment consisted of 16 experimental items, distributed equally over four blocks. In addition, to distract the attention of the participants from the goal of the experiment, 32 ¿ller items were included which were also preceded by an introductory sentence but did not contain an anaphor. Of these ¿llers, 18 contained a transitive verb and a de¿nite description as the direct object, and 14 contained an intransitive verb. See Banga (2008) for all four lists as well as a detailed discription of the distribution of experimental items and ¿llers over the lists. For half of the experimental items, the test sentence and the picture matched, whereas for the other half of the items the test sentence and the picture did not match with respect to the orientation of the action (self-oriented vs. other-oriented). Also for the ¿llers, half of the sentences and pictures matched, whereas the other half did not match with respect to the meaning of the predicate. The matching items are expected to lead to a ‘yes’ response by participants, whereas the mismatching items are expected to lead to a ‘no’ response. The similarities and differences between items in the four conditions are listed in Table 1. The pictures for pronouns in a match situation and reÀexives in a mismatch situation display other-oriented actions. The pictures for pronouns in a mismatch situation and reÀexives in a match situation display self-oriented actions. The sentences with pronouns and reÀexives in the same context condition are identical until the onset of the anaphor. Table 1. Speci¿cation of the experimental conditions. Condition CP-Match CP-Mismatch SP-Match SP-Mismatch CR-Match CR-Mismatch SR-Match SR-Mismatch

Action in picture other-oriented self-oriented other-oriented self-oriented self-oriented other-oriented self-oriented other-oriented

Context sentence Classic Classic Single Topic Single Topic Classic Classic Single Topic Single Topic

Anaphor pronoun pronoun pronoun pronoun reÀexive reÀexive reÀexive reÀexive

To obtain eye-movement data for the correct and the incorrect referent, two Areas of Interest (AoI) were de¿ned in each picture (see Figure 2).

Adult’s on-line comprehension of object pronouns in discourse

203

Figure 2. De¿ning the areas of interest (AoIs). AoIs were drawn by hand at approximately 1 cm distance around the animals on the pictures.

The ¿rst AoI is the referent that is the subject of the test sentence (NP1), which also is the patient of a self-oriented action. The second AoI is the other referent (NP2), which is the patient of another-oriented action or the bystander in case of a self-oriented action. The borders of the AoIs were drawn by hand at approximately 1 cm distance around the two animals on the pictures, except when the animals (almost) touched each other. In that case the borders of the AoIs were drawn closer to the edges of the animals, so that there would be no overlap between the AoIs. Procedure Using a remote Tobii T120 eyetracker in combination with two computers, we measured adults’ accuracy, reaction times and eye movements during a picture veri¿cation Task. One computer with E-Prime (Schneider, Eschman, and Zuccolotto 2002) was used for stimuli presentation and collection of the accuracy and reaction time data. The other computer with Tobii Studio software was used for collecting the gaze data (in this study, at a frame rate of 60 Hz). The remote Tobii eyetracker is integrated in a 17 inch TFT monitor, has no visible or moving tracking devices, and allows a freedom of head movement of 44 × 22 × 30 cm. The eye-movement data reported are an average of both eyes. Participants were tested individually in a room without any windows, in which the eyetracker and associated equipment were located. They were seated in front of the Tobii eyetracker monitor, with a keyboard attached to the E-Prime computer placed before them. The experimental procedure involved four parts. The ¿rst part was the calibration, the second part was a training session of three items through which the participants were familiarized with the experimental procedure, the third part was the actual experiment, and the fourth part was a questionnaire requesting some basic personal information that the

204

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

participant was asked to complete. Men and women were equally distributed over the four different versions of the experiment. The participants were told that they were going to listen to a series of short stories while looking at pictures on the computer screen. The picture always appeared on the screen ¿rst, followed by the start of the sound ¿le after approximately 1000 ms. The participants were asked to register as fast and as accurately as possible whether the second sentence of the story they heard corresponded to the picture they saw. If it did, the participants had to press the ‘Q’ key on the keyboard; if it did not, they had to press the ‘P’ key. The entire test session lasted about 25 minutes.

Results Responses Although it is to be expected that adults hardly make any errors in the picture veri¿cation task, the accuracy of responses was analyzed to con¿rm this expectation.2 In total 335 responses were given, of which 322 (96%) were correct and only 13 (4%) were incorrect. For the four conditions, the proportions of correct responses out of the total amount of responses were calculated for each participant and for each item. The mean proportions of correct responses and their standard deviations, all based on participant analysis, are listed in Table 2. Table 2. Mean proportion of correct responses (standard deviations), in the picture veri¿cation task. Pronoun Classic 0.93 (0.14)

Single Topic 0.99 (0.07)

Classic 1.00 (0.00)

ReÀexive Single Topic 0.91 (0.18)

2. Because accidentally, for one item in all conditions, the wrong audio ¿le and picture were combined (resulting in a mismatch item rather than a match item, but with an incorrect agent rather than an incorrect patient), this item (Item 2) was removed from our analysis. Another item (Item 15) was removed from our analysis of the SP condition only because the verb/action for ‘to dress’ was used rather than for ‘to make up’.

Adult’s on-line comprehension of object pronouns in discourse

205

The proportions were ¿rst arcsine-transformed. Repeated Measures ANOVAs were then run on these transformed proportions with Type of Introductory Sentence (classic versus single topic) and Type of Anaphor (pronoun versus reÀexive) as within-participants and within-items factors. No signi¿cant main effects were found for Type of Introductory Sentence, or Type of Anaphor (all F-values < 1). There was also no signi¿cant interaction of Type of Introductory Sentence × Type of Anaphor (p-values > .10). So the few errors that were made were equally distributed across conditions.3 Reaction times In addition to their responses in the picture veri¿cation task, we also measured participants’ reaction times on the task. Differences in reaction times between conditions are generally considered to be an indication of differences in processing dif¿culty. If a preference for pronouns to refer to the topic is part of the grammar (cf. Spenader et al. 2009), we expect pronouns to take more time to be interpreted in the Classic Condition than in the Single Topic Condition. If, on the other hand, context does not play a role in adults’ on-line processing of pronouns (a possibility compatible with Conroy et al. 2009), we do not expect any signi¿cant differences between the two conditions. For reÀexives, both accounts predict that there will not be any differences between the two conditions. Only correct responses were included in the analysis of reaction times. Although no accuracy differences were found between the four conditions, this was done to rule out the possibility that incorrect responses may have inÀuenced reaction times. For the four conditions, the mean reaction times and the standard deviations were calculated for each participant and for each item. Reaction times were measured from the onset of the anaphor (i.e. the disambiguating word hem ‘him’ or zichzelf ‘himself’) until the response given. The mean reaction times and standard deviations are listed in Table 3 and are graphically presented in Figure 3. Numbers are based on participant analysis.

3. Although the use of a de¿nite article without previous mention of the referent is generally believed to be infelicitous, the fact that participants readily accepted sentences such as (7) and (8) indicates that visual information can render a referent suf¿ciently familiar to license the use of a de¿nite article.

206

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

Table 3. Mean reaction times in milliseconds (standard deviations), measured from onset of anaphor. Pronoun Classic 1642 (473)

Single Topic 1301 (364)

Classic 1319 (355)

ReÀexive Single Topic 1290 (327)

Repeated Measures ANOVAs were run on the reaction times with Type of Introductory Sentence (classic versus single topic) and Type of Anaphor (pronoun versus reÀexive) as within-participants and within-items factors. The main effect of Type of Introductory Sentence was signi¿cant by participants, but did not reach signi¿cance in the analysis by items, F1(1,23) = 8.6, MSE = 78391, p = .008; F2(1,13) = 1.7, MSE = 92938, p = .21. Similarly, the effect of Type of Anaphor was signi¿cant by participants but not by items, F1(1,23) = 8.1, MSE = 100830, p = .009; F2(1,13) = 2.9, MSE = 88211, p = .11. These effects, however, were quali¿ed by a signi¿cant interaction of Type of Introductory Sentence x Type of Anaphor, F1(1,23) = 6.9, MSE = 84236, p = .015; F2(1,13) = 9.9, MSE = 120030, p = .008. Paired t-tests with Bonferroni correction showed that responses to pronouns in the Classic Condition took longer than responses in any of the other three conditions (all p-values < .005). Reaction times in these latter three conditions were not signi¿cantly different.

Reaction times (ms)

2500 2000 1500

Classic Single Topic

1000 500 0 Pronoun

Reflexive

Figure 3. Mean reaction times measured from onset of anaphor.

Adult’s on-line comprehension of object pronouns in discourse

207

Eye movements In addition to collecting reaction times, we recorded the eye movements of the participants, as the timing and pattern of looks to potential referents has been argued to provide information about which referents are considered during the comprehension of pronouns and reÀexives (Runner, Sussman, and Tanenhaus 2003; Sekerina, Stromswold, and Hestvik 2004). Figures 4 – 7 below present the graphical results for the four conditions, distinguishing between match situations and mismatch situations. The ¿gures show the proportion of ¿xations averaged over participants from the presentation of the picture until after a response has been given following the second sentence. The average duration per condition of the introductory sentence is represented by the left horizontal bar above the graph, and the average duration per condition of the second sentence (the test sentence) by the right horizontal bar. The onset of the anaphor is indicated by ‘A’ and the mean reaction time by ‘RT’. The target is always the correct referent, and the distractor the other referent. The category ‘other’ includes all looks outside these two Areas of Interest. As can be seen from Figures 4 – 7, the pictures have a considerable effect on the eye movements. If the picture presents another-oriented action, as with pronouns in the match conditions and reÀexives in the mismatch conditions, the pattern of looks is very similar. In these cases, participants looked at the two referents approximately equally often, somewhat more than 40% of times. This pattern is clearly distinct from the pattern that can be observed with pictures presenting a self-oriented action. For pronouns in the mismatch conditions as well as reÀexives in the match conditions, more looks (almost 60%) are to the agent of the self-oriented action, which is at the same time the patient of the action. Fewer looks (roughly 30%) are to the other referent. This effect of the pictures results in a dominance of looks to the correct referent (the target) for the reÀexive but to the incorrect referent (the distractor) for the pronoun. In addition to these general looking patterns, we considered two speci¿c measures in the participant’s eye-movement data: (1) mean proportions of observation length to the correct referent, and (2) mean time to ¿rst ¿xation on the correct referent. Observation length is an overall measure of sentence interpretation that aggregates all looking times to a given referent (or to be more precise: within the area of interest de¿ned for that referent) from the onset of the anaphor until the participant has given a response. The onset of the anaphor is the disambiguating point in the sentence and is in most items the

208

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

last word of the sentence.4 Because of individual differences in reaction times (i.e. time between onset of anaphor and actual response), observation lengths are normalized by dividing the observation length for the correct referent by the sum of the observation length for the correct referent and the observation length for the incorrect referent. If participants look less at the correct referent in a particular condition as compared to another condition, we may assume that they ¿nd the correct referent a less probable antecedent for the anaphor in this condition than in the other condition. The second measure, mean time to ¿rst ¿xation on the correct referent, is the time in milliseconds from the onset of the anaphor until the start of the ¿rst ¿xation on the correct referent (or to be more precise: within the area of interest de¿ned for the correct referent). It is an early measure of interpretation that yields an indication of how much time it took the participant to zoom in on the correct referent for the ¿rst time after the onset of the anaphor. The faster this process, the easier accessing the correct interpretation seems to be. So if participants take less time to ¿xate on the correct referent in a particular condition as compared to another condition, we may assume that they ¿nd it easier to access the correct referent in this condition.

Figure 4. Pronouns in Classic Condition (CP).

4. Two of the verbs that were used require a particle (aankleden ‘to dress’ and vastbinden ‘to tie’). These particles (aan and vast) are usually placed in sentence-¿nal position in Dutch main clauses and hence follow the anaphor in the test sentences. However, as these verbs were distributed equally across conditions, their inclusion did not inÀuence the results.

Adult’s on-line comprehension of object pronouns in discourse

Figure 5. ReÀexives in Classic Condition (CR).

Figure 6. Pronouns in Single Topic Condition (SP).

Figure 7. ReÀexives in Single Topic Condition (SR).

209

210

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

Mean proportions of observation length are listed in Table 4 and are graphically presented in Figure 8. Table 4. Mean proportions of observation length (standard deviations), measured from onset of anaphor until correct response. Pronoun Classic 0.39 (0.16)

Single Topic 0.43 (0.34)

Classic 0.73 (0.18)

ReÀexive Single Topic 0.69 (0.12)

Observation length (proportions)

The proportions were ¿rst arcsine-transformed. Repeated Measures ANOVAs were then run on these transformed proportions with Type of Introductory Sentence (classic versus single-topic) and Type of Anaphor (pronoun versus reÀexive) as within-participants and within-items factors. There was no main effect of Type of Introductory Sentence (both F-values < 1), but a main effect of Type of Anaphor was found, F1(1,21) = 41.5, MSE = 0.27, p < .001; F2(1,15) = 17.0, MSE = 0.55, p = .001. In the reÀexive conditions (M = 0.71), higher proportions of observation length were to the intended referent compared to the pronoun conditions (M = 0.41). No interaction between Type of Introductory Sentence × Type of Anaphor was found (p-values > .10). 1 0.8 0.6

Classic Single Topic

0.4 0.2 0 Pronoun

Reflexive Type of anaphor

Figure 8. Mean proportions of observation length, measured from onset of anaphor until correct response.

Adult’s on-line comprehension of object pronouns in discourse

211

The mean times to ¿rst ¿xation on the correct referent were also analyzed. These data are listed in Table 5 and are graphically presented in Figure 9. Table 5. Mean time to ¿rst ¿xation in milliseconds (standard deviations), measured from onset of anaphor. Pronoun Classic 358 (278)

Single Topic 323 (278)

Classic 73 (87)

ReÀexive Single Topic 85 (103)

Repeated Measures ANOVAs were run on the mean times to ¿rst ¿xation on the correct referent with Type of Introductory Sentence (classic versus singletopic) and Type of Anaphor (pronoun versus reÀexive) as within-participants and within-items factors. No main effect of Type of Introductory Sentence emerged (both F-values < 1), but a main effect of Type of Anaphor was found, F1(1,22) = 39.9, MSE = 39387, p < .001; F2(1,12) = 15.8, MSE = 65887, p = .002. For the reÀexive conditions (M = 79 ms), the time to the ¿rst ¿xation on the correct referent was shorter than for the pronoun conditions (M = 340 ms). No interaction between Type of Introductory Sentence × Type of Anaphor was found (F-values < 1).

Mean time to first fixation (ms)

700 600 500 400

Classic

300

Single Topic

200 100 0 Pronoun

Reflexive Type of anaphor

Figure 9. Mean time to ¿rst ¿xation, measured from onset of anaphor.

212

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

Discussion Adults’ on-line comprehension of anaphora The central question of our study is whether and how discourse context inÀuences adults’ on-line comprehension of pronouns. On the basis of the literature, two competing hypotheses were formulated. The ¿rst hypothesis, compatible with Conroy et al.’s (2009) partly reductionist account of the Delay of Principle B-Effect, is that discourse context has no signi¿cant effects on adults’ on-line comprehension of pronouns. The second hypothesis, derived from Spenader et al.’s (2009) optimality theoretic explanation of the Delay of Principle B-Effect, is that constraints on local discourse coherence not only have effects on children’s off-line responses but also on adults’ on-line processing of pronouns. We looked at adults’ accuracy, reaction times and eye movements during a picture veri¿cation task. As both accounts would predict, the adult participants in our study hardly made any comprehension errors. However, their reaction times with pronouns in the Classic Condition were signi¿cantly longer than their responses in the other three conditions. As we found a signi¿cant interaction, and not just main effects of either linguistic information or visual information, it is rather unlikely that these reaction times are mere reÀections of processes having to do with decision making. Instead, we assume that these differences in reaction times to a large extent reÀect differences in the complexity of processes underlying sentence comprehension. This would indicate that the adults experienced more processing dif¿culties with pronouns in the Classic Condition than in the Single Topic Condition. Thus, the results of our on-line study with adults appear to be consistent with the results of Spenader et al’s off-line study with children. They found that children make more errors on pronoun interpretation in the Classic Condition than in the Single Topic Condition. Because children’s errors and adults’ processing dif¿culties occur in the same type of sentence, it is not implausible that they stem from the same source. That is, adults’ longer reaction times for pronouns in the Classic Condition may provide support for Spenader et al.’s suggestion that the observed preference for pronouns to refer to the topic is part of the grammar. Participants’ eye movements however showed a different picture. The two speci¿c measures we looked at only showed a main effect of type of anaphor: Participants looked for a shorter time at the correct referent for the pronoun than for the reÀexive, and it took them longer to ¿xate on the correct referent for the pronoun than for the reÀexive. Surprisingly, the two measures showed no effects of context. Participants did not look for a shorter time at the correct referent for the pronoun in the Classic Condition than in the Single Topic

Adult’s on-line comprehension of object pronouns in discourse

213

Condition, nor did it take them longer to ¿xate on the correct referent for the ¿rst time. These results then seem to support the account of Conroy et al., since adults do not appear to be inÀuenced by discourse context in their looking behavior when interpreting pronouns or reÀexives. So the reaction times appear to support an analysis according to which discourse context has a signi¿cant effect on adults’ processing of object pronouns, whereas the eye-movement data appear to support an analysis according to which discourse context has no effects at all. How are we to reconcile these different results? Generally speaking, eye movements do not seem to be as strongly linked to processing dif¿culty as reaction times. It is possible that the eye movements in our study have been inÀuenced by the task. This possibility is supported by the participants’ patterns of looking to the correct and incorrect referent over time. As can be seen from Figures 4 – 7, the pictures had a considerable effect on the eye movements. The effects of the pictures are large enough perhaps to have masked potential context effects in the eye-movement data. On all numerical measures, pronouns in the Classic Condition seem most dif¿cult for participants. Pronouns in the Classic Condition gave rise to the longest reaction times, the shortest observation lengths to the correct referent, and the longest times to ¿rst ¿xation on the correct referent. Although only the result from reaction times is signi¿cant, the tendencies displayed in the eye-movement data are certainly not incompatible with the result from reaction times. If particants’ processing dif¿culties in our study are more accurately reÀected by their reaction times than by their eye movements, this has implications for the use of eyetracking in the study of language. Researchers have used eyetracking in combination with a wide array of tasks, such as picture selection tasks (Sekerina et al. 2004), action based tasks (Runner et al. 2003), and truth value judgment-tasks and picture veri¿cation tasks (Arnold et al. 2000). The general consensus seems to be that eyetracking is less sensitive to task effects than other on-line or off-line methods, because of the underlying assumption that “the mind is going where the eye is going” (Trueswell and Gleitman 2007). However, our study suggests that picture veri¿cation tasks, but perhaps also other psycholinguistic tasks, may inÀuence participants’ eye movements to a considerable extent. Implications for child language research The present study investigated adults’ on-line processing of pronouns and reÀexives. It revealed certain similarities between adults’ on-line processing and children’s off-line interpretation. These similarities indicate that for children as well as adults pronouns in the Classic Condition are more dif¿cult

214

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

to interpret than pronouns in the Single Topic Condition or reÀexives in either condition. These results follow from the predictions of Spenader et al. (2009), who assume that children are not yet able to apply Principle B of Binding Theory but use cues from discourse context whenever they can to interpret pronouns. Adults’ reaction times for pronouns in the Single Topic Condition were comparable to their reaction times for reÀexives in either condition. This suggests that adults are helped by a coherent discourse too. Our study also displayed differences between adults’ on-line processing and children’s off-line interpretation. Whereas children’s off-line interpretation of object pronouns is heavily inÀuenced by discourse context, as was shown by Conroy et al. (2009) and Spenader et al. (2009), we did not ¿nd any effects of discourse context in the eye-movement data of the adults. The results of the adults may follow from the account of Conroy et al. On the basis of the off-line results of their experiments, Conroy et al. conclude that children are hindered by an unbalanced context. The adult controls in their experiments, on the other hand, did not seem to be hindered by an unbalanced context in their off-line responses. If the effects of context are assumed to be caused by factors related to the truth value judgment task that disappear with age, it is predicted that adults are not inÀuenced by the discourse context in their on-line processing either. However, we suggested that the absence of context effects in adults’ eye-movement data may have been caused by particular task effects associated with the picture veri¿cation task. To make more sense of the different conclusions that can be drawn from adults’ reaction times and their eye movements, and to shed more light on children’s knowledge of Principle B, it would be useful to study children’s on-line processing of pronouns and reÀexives in relation to the structure of the discourse. If children show the same pattern of eye movements as the adults in our study, this could suggest that eye movements are not directly tied to a participant’s interpretation of an anaphor and may reÀect a different set of (task- or materials-induced) processes. On the other hand, differences between children’s eye movements and adults’ eye movements may point at a different contribution of discourse context in children’s and adults’ interpretation of pronouns.

Conclusions In this study we tested Dutch adults on a picture veri¿cation task while recording their reaction times and monitoring their eye movements. As we were interested in the effects of discourse context on the interpretation of object pronouns

Adult’s on-line comprehension of object pronouns in discourse

215

and reÀexives, we manipulated the structure of the introductory sentence. Although the adults in our study hardly made any comprehension errors, their reaction times were signi¿cantly slower when the introductory sentence did not unambiguously establish a discourse topic. This suggests that the structure of the discourse context is important for pronoun interpretation and inÀuences both children’s off-line interpretation of object pronouns (resulting in Delay of Principle B-Effect errors in particular contexts) and adults’ on-line processing of object pronouns (resulting in slower reaction times in the same contexts). These results seem to be at odds with approaches that try to reduce the Delay of Principle B-Effect to an experimental artifact. On the other hand, adults’ eye movements did not provide signi¿cant evidence for effects of discourse context on their processing of object pronouns. We believe that this may have been caused by task effects.

References Arnold, Jennifer E., Janet G. Eisenband, Sarah Brown-Schmidt, and John C. Trueswell 2000 The rapid use of gender information: Evidence of the time course of pronoun resolution from eyetracking. Cognition 76: B13 – B26. Badecker, William, and Kathleen Straub 2002 The processing role of structural constraints on the interpretation of pronouns and anaphors. Journal of experimental psychology: Learning, memory and cognition 28: 748 – 769. Banga, Arina 2008 On-line measures in adults’ comprehension of anaphors in coherent and incoherent discourse. MA thesis, University of Groningen. Chien, Yu-Chin, and Kenneth Wexler 1990 Children’s knowledge of locality conditions on binding as evidence for the modularity of syntax and pragmatics. Language Acquisition 1: 225 – 295. Conroy, Anastasia, Eri Takahashi, Jeffrey Lidz, and Colin Phillips 2009 Equal treatment for all antecedents: How children succeed with Principle B. Linguistic Inquiry 40: 446 – 486. de Hoop, Helen, and Irene Krämer 2005/6 Children’s optimal interpretations of inde¿nite subjects and objects. Language Acquisition 13: 103 – 123. Grodzinsky, Yosef, and Tanya Reinhart 1993 The innateness of binding and the development of coreference. Linguistic Inquiry 24: 69 – 101.

216

Petra Hendriks, Arina Banga, Jacolien van Rij, Gisi Cannizzaro, and John Hoeks

Grosz, Barbara J., Aravind K. Joshi, and Scott Weinstein 1995 Centering: A framework for modelling the local coherence of discourse. Computational Linguistics 21: 203 – 225. Hendriks, Petra, and Jennifer Spenader 2005/6 When production precedes comprehension: An optimization approach to the acquisition of pronouns. Language Acquisition 13: 319 – 348. Kennison, Shelia M. 2003 Comprehending the pronouns “her”, “him” and “his”: Implications for theories of referential processing. Journal of Memory and Language 49: 335 – 352. Nicol, Janet 1988 Coreference processing during sentence comprehension. Ph.D. diss., MIT. Nicol, Janet, and David Swinney 1989 The role of structure in coreference assignment during sentence comprehension. Journal of Psycholinguistic Research 18: 5 – 19. Reinhart, Tanya 2006 Interface Strategies: Optimal and Costly Computations. Cambridge, MA: MIT Press. Ruigendijk, Esther 2008 Pronoun interpretation in German kindergarten children. In Proceedings of GALA 2007, Anna Gavarró and M. João Freitas (eds.), 370 – 380. Newcastle upon Tyne: Cambridge Scholars Publishing. Runner, Jeffrey T., Rachel S. Sussman, and Michael K. Tanenhaus 2003 The inÀuence of Binding Theory on the on-line reference resolution of pronouns. Poster presented at the NELS 34. Stony Bruck, NY. Schneider, Walter, Amy Eschman, and Anthony Zuccolotto 2002 E-Prime User’s Guide. University of Pittsburgh Psychology Software Tools, Inc. Sekerina, Irina A., Karin Stromswold, and Arild Hestvik 2004 How do adults and children process referentially ambiguous pronouns? Journal of Child Language 31: 123 – 152. Spenader, Jennifer, Erik-Jan Smits, and Petra Hendriks 2009 Coherent discourse solves the Pronoun Interpretation Problem. Journal of Child Language 36: 23 – 52. Thornton, Rosalind, and Kenneth Wexler 1999 Principle B, VP Ellipsis, and Interpretation in Child Grammar. Cambridge, MA: MIT Press. Trueswell, John C., and Lila R. Gleitman 2007 Learning to parse and its implications for language acquisition. In Oxford Handbook of Psycholinguistics, M. Gareth Gaskell (ed.), 635 – 656. Oxford: Oxford University Press.

Production and comprehension of sentence negation in child German* Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

1

Introduction

The acquisition of sentence negation has been well-studied since the late seventies (Clahsen 1988; de Villiers and Tager Flusberg 1975; Déprez and Pierce 1993; Dimroth 2009; Drozd 1995; Hamann 2000; Meisel 1997; Penner, Tracy, and Weissenborn 2000; Stromswold and Zimmermann 2000; Wode 1977). Independent of the language to be acquired, negation particles, for example no and not in English, belong to the ¿rst elements children produce in the early multiword stage (e.g. Wode 1977). It has been argued that already three-year-olds use negation elements target-like with respect to semantic and syntactic properties (cf. Clahsen 1988 for German). In contrast to the extensive research on the production of negation, few studies have investigated children’s comprehension of negated sentences. Moreover, to our knowledge studies have not looked at production and comprehension of sentence negation in the same children, thus leaving open the question of whether there is an asymmetry in the acquisition path between both modalities. Taking as a starting point the traditional assumption that comprehension precedes production (cf. Clark 1993), we would predict that at age three children have no dif¿culty understanding negated sentences. On the other hand, it has been claimed that production may precede comprehension. This position has received support for example from ¿ndings on the acquisition of binding * Acknowledgements: This research was carried out in the Center for individual development and adaptive education of children at risk (IDeA, chair M. Hasselhorn), funded by the federal state government of Hessen (LOEWE initiative).We thank the children, their parents, and the research assistants for gathering the data, and the Frankfurt language acquisition group for helpful support. In addition, we thank the two anonymous reviewers for valuable comments and questions that helped clarify the ideas presented in this paper.

218

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

principles (Delay of Principle B-effect, Baauw et al. this volume), of in¿nitival complements (Grimm, Schöler, and Wintermantel 1975), and of focus particles (Müller, Schulz, and Höhle 2010). The evidence to date is still limited, as few studies have investigated comprehension and production in the same children, and few studies have used a number of different methods within the same modality to reach converging evidence. Therefore, a production-comprehension asymmetry may not be present within the same group of subjects or may be speci¿c to a particular linguistic task. Regarding negation, production may be expected to precede comprehension for the following reasons. At the early multiword stage negation particles are functionally closely linked to focus particles in that they serve as bootstrapping elements to project functional categories. Accordingly, negation particles form necessary prerequisites for acquiring complex syntactic structures, and are already produced around age two (Penner, Tracy, and Weissenborn 2000). In contrast, errors seem to persist late in development in formal comprehension tasks requiring children to handle logical negation (Feldman 1972). Previous studies on the acquisition of negation could only partially address the question of a production-comprehension asymmetry. First, no study tested production and comprehension of sentence negation with the same group of children. Second, no study provided longitudinal data for production and comprehension examining larger groups of monolingual children. Therefore, we investigated the acquisition of negation in monolingual learners of German, using a combined cross-sectional and longitudinal design. To compare production and comprehension skills, the same children were tested with an elicited production paradigm and a truth value judgment task. Both tasks are part of the standardized language test Linguistische Sprachstandserhebung – Deutsch als Zweitsprache (LiSe-DaZ, Schulz and Tracy 2011). All participants were tested twice in a six-month interval at age 3;7 and age 4;2. The data are part of the larger research project MILA1, which compares the acquisition of morphosyntactic, semantic, and phonological abilities of monolingual children and early second language learners of German. The paper is structured as follows: Section 2 provides theoretical background on sentence negation. Section 3 summarizes previous studies dealing with the acquisition of sentence negation. Our research hypotheses are formulated in section 4. The design and results of the present study are presented in 1. The children take part in the project MILA (The relation between migration background and language impairment in children’s language achievement, Principal Investigator: Petra Schulz), which is part of the Research Center IDeA.

Production and Comprehension of Sentence Negation in Child German

219

section 5. Finally, section 6 provides a general discussion of the ¿ndings and suggests directions for future research.

2

Syntactic, semantic, and pragmatic aspects of sentence negation

Unlike contrastive negation, which serves to deny a speci¿c constituent, sentence negation denies the whole proposition of a sentence. Sentence negation is assumed to occur in all languages, but languages use different lexical and syntactic means to express negation.2 In the following, core syntactic, semantic, and pragmatic aspects of negation are summarized that were relevant for our acquisition study of sentence negation in German. Sentence negation belongs to the class of non-anaphoric negation (Bloom 1970). Non-anaphoric negation is de¿ned as denoting a “negative relationship [that] holds between neg and some part or the whole of the sentence or phrase with which neg occurs in construction” (Wode 1977: 90). In contrast, anaphoric negation corrects parts of an aforementioned utterance. In German, both negation types have in common that the negation is expressed by a free morpheme, nein ‘no’ in anaphoric negation (as a sentence equivalent); nicht ‘not’, kein ‘no’ (as a determiner), and niemals ‘never’ in the case of non-anaphoric negation. Sentence negation is exclusively expressed by the negation particle nicht ‘not’ in German. The position of nicht is ¿xed, but depending on the sentence type and the required movements of verb and objects, it may appear sentence ¿nally or sentence internal. This is illustrated in the following examples ((1a – c) taken from LiSe-DaZ, Schulz and Tracy 2011). (1)

a.

Die Ente schwimmt nicht. the duck swims not ‘The duck is not swimming.’

b.

Das Mädchen gibt dem Jungen den Ball nicht. the girl gives the-DAT boy the-ACC ball not ‘The girl does not give the ball to the boy.’

c.

Der Junge fährt nicht mit dem Skateboard. the boy drives not with the-DAT skateboard ‘The boy is not riding a skateboard.’

2. Sentence negation can be realized either via morphological marking or as a negation particle (German, French) or via negative verbs as in some African languages (Dahl 1993).

220

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz d.

Ich sehe, dass die Ente nicht schwimmt. I see that the duck not swims ‘I see that the duck is not swimming.’

(1a) to (1d) show that the position of the negation particle nicht is closely related to the placement of the ¿nite verb. Whereas sentence negation occupies a ¿xed position within the sentence structure, the position of the ¿nite verb varies with respect to sentence type. Since German is a V2-language, the ¿nite verb in a main clause moves to the second position regardless of which constituent is placed in sentence-initial position. In this case, the ¿nite verb precedes nicht (1a – c). In a subordinate clause, the ¿nite verb stays sentence¿nal, and the negation particle precedes the ¿nite verb (1d). Moreover, in German sentence negation may be realized either directly adjacent ((1a), (1c), (1d)) or non-adjacent (1b) to the verb. This is in contrast to the cross-linguistic tendency to realize sentence negation only adjacent to the ¿nite verb (Meisel 1997). De¿nite arguments and semantically [+ speci¿c] inde¿nite arguments of the verb must be scrambled in front of the negator (Steube 2006). Accordingly, to achieve an interpretation of nicht as sentence negation in (1b) the scrambled constituents dem Jungen and den Ball have to be reconstructed within the scope of the negator (Sudhoff 2008). Prepositional adjuncts like mit dem Skateboard in (1c), on the other hand, do not scramble over the negation particle nicht. The same restriction holds for example for the extended verbal complex consisting of constituents as predicates (2a), as well as for modal adverbials (2b) that cannot scramble over the sentence negation particle nicht (Steube and Sudhoff 2007). (2)

a.

Die Studentin war nicht in der Universität. The student was not in the university. ‘The student was not at the university’

b.

Die Studentin tanzt nicht gut. The student dances not good. ‘The student doesn’t dance well’

The precise syntactic analysis of the sentential negator nicht is still a matter of debate. Researchers agree that nicht projects a maximal phrase of its own (i.e., a NegP (Pollock 1989) or an adjunct (Haider 2004)), and that it occupies a ¿xed position within the sentence structure. In the following, we assume that nicht is located in the speci¿er of NegP that directly dominates VP (Meisel 1997 or the extensive discussion in Hamann 2000), and the head of NegP is empty.

Production and Comprehension of Sentence Negation in Child German

221

Figure 1. Syntactic structure of negation in German (cf. Meisel 1997)

Concerning its semantic properties, sentence negation differs from contrastive negation, which in German is expressed by the phonologically identical particle nicht. As mentioned above, sentence negation denies the whole proposition of a sentence, while contrastive negation has scope only over the constituent it is adjoined to (Steube and Sudhoff 2007), as illustrated in (3a) (repeated from (1b)) vs. (3b).3 Focused constituents are marked by capital letters. (3)

a.

Das Mädchen gibt dem Jungen den Ball nicht. the girl gives the-DAT boy the-ACC ball not ‘The girl does not give the ball to the boy.’

b.

Das Mädchen gibt dem Jungen nicht den BALL, the girl gives the-DAT boy not the-ACC ball, sondern … but… ‘The girl does not give the BALL to the boy, but …’

3. In contrast to Steube and Sudhoff (2007), Jacobs analyzes contrastive negation as adsentential; it can only be adjoined to verbal projections as VP, IP, and CP (Jacobs 1982: 276f).

222

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

In (3a) the proposition ‘the girl gives the ball to the boy’ is denied, i.e. the sentence meaning is ‘it is not the case that the girl gives the ball to the boy’. (3b) means ‘the thing that the girl gives to the boy is not the ball’ OR ‘the girl gives something to the boy and it is not a ball’. While sentence negation occupies a ¿xed position in the sentence structure, contrastive negation is analyzed as an adjunct that is adjoined to the negated constituent. This constituent is focused, and is marked with contrasting stress. Typically, contrastive negation offers an alternative ending, which might be added by sondern, but as sondern den Schläger ‘but the racket’. Turning to pragmatic properties of sentence negation, negated sentences can be used in two different types of contexts. They can be a true description of a situation (henceforth called ‘true negatives’) as in (4) or a false description of a situation (henceforth called ‘false negatives’) as in (5) (examples taken from Wason 1972: 20). (4)

57 is not an even number.

True negative

(5)

92 is not an even number.

False negative

Note that in order to evaluate (4) and (5) the actual discourse context in which the negated mathematical statements are being uttered does not play a role. However, often negated sentences are used in situations where the actual context decides about truth or falsity of a negated statement. Thus, the interpretation of a negated sentence requires matching the negated proposition with the actual discourse context. This is demonstrated in examples (6) and (7) (picture taken form LiSe-DaZ, Schulz and Tracy 2011). (6)

True negative: Ibo schläft nicht. Ibo sleeps not ‘Ibo is not sleeping.’

(7)

False negative: Ibo steht nicht. Ibo stands not ‘Ibo is not standing.’

Given that the picture above depicts a boy named Ibo, (6) is a true negative because it describes the situation on the picture correctly. (7) is a false negative because Ibo is in fact standing. Clark (1976) assumes that more processing steps are necessary to evaluate a true negative as compared to a false negative.

Production and Comprehension of Sentence Negation in Child German

223

Thus, processing true negatives in language comprehension is assumed to be harder than processing false negatives. As can be inferred from (6) and (7) above, in everyday discourse false negatives seem more natural and thus are much more frequently used than true negatives (cf. Wason 1972). As Wason states, “a negated sentence typically denies what is supposed to be true” (Wason 1972: 35). The sentence in (6) “Ibo is not sleeping” might be a true negative describing the depicted situation, but that is not the reason for saying it. Rather, it denies the assumption of the listener that Ibo is sleeping and thus could be said to function as a false negative regarding the listener’s belief. A true negative regarding the context would be the statement Ibo is not sleeping in a situation where nobody expects Ibo to sleep. Hence, using a negated sentence to con¿rm a proposition is contrary to its natural function and is therefore unexpected in language production. Additionally, Wason (1972: 28) states that “negation functions within an af¿rmative context”. Accordingly, verifying sentence (8) without a given af¿rmative context is harder than verifying it with an af¿rmative context such as “4 is even and …” (Wason 1972: 28). (8)

5 is not even

Summarizing this section, sentence negation possesses complex syntactic, semantic, and pragmatic properties that all have to be mastered by the language learner.

3

Previous acquisition studies

In section 3.1 previous studies on the production and the comprehension of sentence negation are summarized; section 3.2 describes previous research on the comprehension of negated sentences. 3.1

Production of sentence negation

Several studies have investigated the production of negation in monolingual language acquisition. For German, two studies by Clahsen (1988) and Wode (1977) are the most comprehensive ones. Based on spontaneous production data of two monolingual children, Wode proposes four developmental stages. At Stage I, children use phonetic variants of nein ‘no’ in isolation. Around age 1;7, children enter Stage II realizing nein in a clause-external position. Following Wode, this pre-sentential nein can either function as anaphoric nega-

224

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

tion or as a non-anaphoric negation that is established at the semantic level, but is not fully acquired at the syntactic level.4 At Stage III, nicht ‘not’ is used sentence internally, gradually replacing the non-anaphoric nein-structures. Although main aspects of the syntactic realization of sentence negation are in place now, children still fail to obligatorily scramble de¿nite noun phrases at Stage III. Thus, they realize ¿nite verb and negator in adjacent positions resulting in a contrastive reading as in (9), while actually intending the interpretation to be a sentence negation as in (10). (9)

*Das Mädchen isst nicht den The girl eats not the ‘The girl is not eating the candy.’

Bonbon. candy.

(10)

Das Mädchen isst den Bonbon nicht. The girl eats the candy not. ‘The girl is not eating the candy.’

Clahsen (1988) analyzed the spontaneous speech data of six children. Unlike our assumptions, Clahsen analyses neg as a head instead of a speci¿er of NegP. According to his account, INFL and neg build a syntactic island, which cannot be affected by movement operations at this stage. Scrambling takes place once the children master subject-verb agreement,5 and thus become aware that the INFL-neg complex forms a derived structure. As a consequence, the unanalyzed unit is split into its lexical and syntactic constituents (Clahsen 1988: 15). When obligatory scrambling takes place, at the latest by age three, children enter Stage IV in Wode’s classi¿cation, and sentence negation is produced in an adult-like way. Recent ¿ndings by Dimroth (2009) and Winkler (2006) going back to Penner (2000) provide further support for a close relationship between negation and ¿niteness in child language. More precisely, Dimroth (2009) argues that sentence negators such as nicht provide stepping stones for the acquisition of ¿niteness. In accordance with this view, ¿niteness is marked more frequently in negated sentences than in af¿rmative ones around age two (Dimroth 2009, but see Hamann 2000 for a discussion of counterevidence). These results 4. In contrast, Drozd analyzes pre-sentential negation in English as target-like nonanaphoric metalinguistic exclamatory sentence negation (Drozd 1995). 5. Drenhaus’ sentence repetition data (2000) shows that children up to age 6;1 do not obligatorily scramble objects in front of negation. This contradicts the ¿ndings of Clahsen (1988) and Wode (1977), using both the method of spontaneous speech analysis.

Production and Comprehension of Sentence Negation in Child German

225

further strengthen the assumption that the negation particle nicht is recognized early in acquisition. 3.2

Comprehension of sentence negation

Most comprehension studies up to date focused on older children and investigated the interplay of sentence negation and other structural phenomena, including de¿nite and inde¿nite noun phrases (Rumain 1988), scrambling of inde¿nite noun phrases (Krämer 1998; Unsworth 2005), and scope ambiguities caused by universal quanti¿cation (e.g. Gualmini 2004; Musolino and Lidz 2006). Few studies address whether and how sentence negation is processed in younger children. Using a preferential looking paradigm, Gilkerson, Hyams, and Curtiss (2004) found that 14- to 25-months-old monolingual children acquiring American English have a basic understanding of sentence negation. When presented with a negated sentence, children looked for a shorter period of time at the picture that was negated than at the same picture accompanied with an af¿rmative sentence. To react this way, children had to identify the negator and possess at least some understanding of its negating properties. Accordingly, the authors conclude that children as young as 14 months are able to distinguish between sentences with and without the negation element not. The studies by Wason (1965) and also by de Villiers and Tager Flusberg (1975) demonstrated that context and pragmatics of negated sentences crucially affect their interpretation for children. Their starting point was evidence from adult processing demonstrating that negated sentences are more dif¿cult to parse than af¿rmative sentences, and that true negatives are more dif¿cult to parse than false negatives (Wason 1972). Replicating Wason’s study (1965) with younger subjects, de Villiers and Tager Flusberg used a sentence completion design with children aged two to ¿ve years. Six or seven objects or drawings were placed in front of a child, with one differing from the others (e.g., seven cars and one baby bottle). Then the experimenter pointed to one of the objects and asked “This is a __?” or “This is NOT a __?”, thus eliciting true af¿rmatives and true negatives. Following the exceptionality hypothesis (Wason 1965), de Villiers and Tager Flusberg hypothesized that answering a negative probe for the exceptional item is more plausible, and therefore easier than answering a negative probe for one of the similar items. This hypothesis was con¿rmed for all age groups. Only implausible negatives (i.e. denying the common characteristic of the similar items) were associated with long response times and higher error rates. If embedded in a plausible context, four-year

226

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

old children showed no differences in the interpretation of true negatives and af¿rmative sentences (de Villiers and Tager Flusberg 1975). For German, data on the comprehension of negated sentences is only available from the norming data of the standardized language test SETK 3 – 5 (Grimm 2001).6 Children aged 3;0 to 3;11 are given a picture choice task with two test items and children aged 4;0 to 5;11 are given an act out task, again with two items. The three-year-olds responded correctly in about 76% of the cases, compared to 85 % correct responses in the four- and ¿ve-year-olds. This data does not allow detailed conclusions to be drawn about the acquisition of negation in German, as only a few items were tested, and as additional linguistic phenomena such as relative clause formation were included in the test material.

4

Research questions and hypotheses

Focusing on children’s comprehension of sentence negation, the present study addressed four questions. The ¿rst was whether there is an asymmetry between production and comprehension of negated sentences in child German. Extending the argumentation that production may precede comprehension (Baauw et al. this volume; Müller, Schulz, and Höhle 2010) to sentence negation, we formulate as Hypothesis 1: (H1)

Monolingual German speaking children who produce nicht still have a non target-like comprehension of sentence negation.

The second question concerned the syntactic aspects of negated sentences. Recall that in German sentence negation may appear adjacent or non-adjacent to the ¿nite verb. Therefore, we were interested in exploring how these syntactic factors affect comprehension of negated sentences. Assuming that verb nonadjacent negation is more complex than verb adjacent negation because the former but not the latter requires object scrambling, we state as Hypothesis 2:

6. Ofner (2007) investigated the acquisition of contrastive negation with a sentence completion task in monolingual German speaking children. Four- to six-year-old children had dif¿culty interpreting structures containing contrastive negation. The children performed correctly to contrastive negation of the subject in 62% of the cases, in contrast to only 33% correct responses to contrastive negation on the verb and 35% correct responses to contrastive negation on the object.

Production and Comprehension of Sentence Negation in Child German

(H2)

227

Comprehension of verb adjacent negation is easier than of verb nonadjacent negation.

To date, detailed research on the acquisitional stages children go through towards mastering true and false negatives is still lacking. Wason (1972) found that in adults false negatives are relatively easy to interpret, while true negatives are more prone to errors. For English-speaking four-year-olds, dif¿culties with the interpretation of true negatives were observed as well (de Villiers and Tager Flusberg 1975). We thus formulate as Hypothesis 3: (H3)

Comprehension of false negatives is easier than of true negatives.

Our last question addressed the interaction of syntax (i.e. the structural position) and pragmatics (i.e. context matching) of sentence negation, using a regression analysis. Since many pragmatic abilities that require context integration are not yet in place in pre-schoolers (Hickmann 2000; Schulz 2007), as Hypothesis 4 we state: (H4)

The pragmatic factor context matching predicts children’s performance on negated sentences better than the syntactic factor structural position.

5

The study

5.1

Participants

We tested 34 typically developing monolingual German-speaking children (11 girls, 23 boys) in two test rounds. All participants take part in a larger study on the acquisition of German as the ¿rst or the second language. The age ranged from 3;5 to 4;1 (M = 3;7, SD = 2 months) in the ¿rst test round, and from 4;0 to 4;5 (M = 4;2, SD = 1.8 months) in the second test round. All children had an age-appropriate non-verbal IQ, with a mean of 89 (SD = 13) assessed by the non-verbal part of the K-ABC (Melchers and Preuss 2003). All children attended regular kindergardens, and according to a parent questionnaire did not show any signs of language impairment or language delay, of hearing problems or psycho-social deprivation.

228

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

5.2

Pretest

A pretest administered at the beginning of the ¿rst test round (age 3;7) served to select the children for the present study. Only children who produced the negation particle nicht consistently participated in the main test. Production of the negation particle nicht was assessed via the subtest Elicited Production of the standardized language test LiSe-DaZ (Schulz and Tracy 2011), which comprises 14 test items in total. Different sentence types (main, subordinate, questions) and word classes (modal and auxiliary verbs, prepositions, conjunctions) are elicited. Three test items aim at eliciting the negation particle nicht. Example (11) illustrates a typical test item for sentence negation. After a lead-in sentence, children are prompted to complete the utterance so that it conveys the intended meaning. Due to the occurrence of nicht in a preceding sentence this may be seen as a repetition task rather than as a production task. However, note that this sentence and the prompt for the child are separated by an intervening sentence. Therefore, simple repetition is impossible. Moreover, many studies have shown that children are successful at imitation only if this structure is already present in the child’s grammar (MacDaniel et al. 1996). (11)

Experimenter: Lise und Ibo gehen jetzt weiter. Den Hund wollen sie aber nur mitnehmen, wenn er nicht mehr wegrennt. ‘Lise and Ibo continue their walk. They want to take the dog with them if he doesn’t run away anymore.’ Lead-in:

Ibo sagt zu dem Hund: Du darfst nur mitkommen, … Ibo says to the-DAT dog: You can only come with … ‘Ibo says to the dog: You can only come with us…’

Child:

wenn du nicht mehr wegrennst. if you not anymore run away ‘if you don’t run away anymore.’

As expected, all 34 children consistently used nicht appropriately in the elicited contexts at the ¿rst test round at age 3;7 and were thus included in the main study. Comparing children’s performance to the age norms, they all reach T-values of 42 and higher and thus fall within the normal range. Material and procedure of the main test Children’s comprehension of sentence negation was tested with the subtest Comprehension of Negation of the standardized language test LiSe-DaZ (Schulz and Tracy 2011), using a variant of the truth value judgment task. Importantly, the truth value judgement task does not show children’s prefer-

Production and Comprehension of Sentence Negation in Child German

229

ence, but measures which interpretation is actually possible for the child (cf. Gordon 1996). The comprehension of negation task works as follows: The experimenter shows the child a picture introduced by a short lead-in sentence, and then a hand puppet makes a statement about the picture, while the child is looking at the picture. The task of the child is to decide whether the puppet’s utterance is correct or incorrect with respect to the picture. Two example test items are given in (12) and (13). (12)

Experimenter: Guck mal, hier ist ein Mädchen, ein Junge und ein Ball. ‘Look, there are a girl, a boy, and a ball.’ Puppet: Das Mädchen gibt dem Jungen den Ball nicht. The girl gives the-DAT boy the-ACC ball not. ‘The girl isn’t giving the ball to the boy.’ Experimenter: Stimmt das? ‘Is that right?’ Child: Ja, das stimmt. ‘Yes, that’s right.’ (13)

Experimenter: Guck mal, hier ist Lise, ein Rucksack und Ibo. ‘Look, there are Lise, a backpack, and Ibo.’

230

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz Puppet:

Lise gibt Ibo den Rucksack nicht. Lise gives Ibo the-ACC backpack not. ‘Lise isn’t giving the backpack to Ibo.’ Experimenter: Stimmt das? ‘Is that right?’ Child: Nein, das stimmt nicht. ‘No, that’s not right.’

Each of the participants was tested individually by trained student assistants in a quiet room in their kindergarden. There was an interval of six months between the two test rounds (M = 5.5, SD = 1.5). The sessions were video-recorded for later data check against the onsite-coding and for further individual analyses. No response-contingent feedback was given by the experimenter. When the child failed to supply an answer, test items were repeated once. A total of 12 negated sentences were presented to each child in a 2 × 2 design.7 Two test conditions varied the syntax and two the pragmatics of the negated sentence, each comprising six test items. An overview of the design is given in Table 1. The order of presentation was random. Table 1. Number of items in the subtest “Comprehension of Negation” by condition Syntax

Pragmatics

Verb adjacent

Verb non adjacent

Total items

True negative

3

3

6

False negative

3

3

6

Total items

6

6

12

Regarding the factor Syntax, the position of the negation particle with respect to the ¿nite verb was varied. (14) illustrates the verb adjacent position, and (15) the verb non-adjacent position of nicht.

7. Note that af¿rmative sentences were not included in the standardized test.

Production and Comprehension of Sentence Negation in Child German

231

(14)

Experimenter: Guck mal, hier ist ein Junge und hier sind Bausteine. ‘Look, there is a boy and there are bricks.’ Puppet: Der Junge spielt nicht mit den Bausteinen. The boy plays not with the-DAT bricks. ‘The boy is not playing with the bricks.’ Experimenter: Stimmt das? ‘Is that right?’ Child: Nein, das stimmt nicht. ‘No, that’s not right.’ (15)

Experimenter: Guck mal, hier sind ein kleiner Junge und eine ‘Look, there are a small boy and a Puppet: Der kleine Junge füttert im Park die Ente The small boy feeds in park the-ACC duck ‘The small boy is not feeding the duck in the park.’ Experimenter: Stimmt das? ‘Is that right?’ Child: Ja, das stimmt. ‘Yes, that’s right.’

Ente. duck.’ nicht. not.

Regarding the factor Pragmatics, the matching between the visual context and the puppet’s statement was varied. True negatives are illustrated in (12) and (15): Here the puppet’s statement con¿rms the situation depicted in the picture

232

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

and hence requires an af¿rmative response. False negatives are illustrated in (13) and (14). In this case the puppet’s statement had to be denied.8 5.4

Coding of responses

Children’s responses were coded as correct and incorrect. Depending on the pragmatic conditions the target response was ‘yes’ to a true negative, and ‘no’ in the case of a false negative. Table 2 summarizes which responses were coded as target-like and which as incorrect for these two conditions. Table 2. Types of correct and incorrect responses to the test items Condition TRUE NEGATIVE ‘The girl doesn’t give the ball to the boy.’

Correct responses Yes. That’s right. No, she is not giving it to him. She is kicking the ball away. No. FALSE NEGATIVE ‘Lise doesn’t give the That’s not right. backpack to Ibo.’ She is giving it to him.

5.5

Incorrect responses No. That’s not right. She is giving it to him. Yes. That’s right. No, she is not giving it to him.

Results

5.5.1 General results This section summarizes the quantitative results. A more detailed analysis with respect to the four hypotheses follows in the subsequent subsections. Figure 2 illustrates the high proportion of correct responses (age 3;7: mean = 9.2, SD = 2.12; age 4;2 mean = 9.6, SD = 2.07). Comparing children’s performance with the age norms in the standardized test, across the two test rounds all but three children fall within the age-appropriate norm, with T-values of 41 or above. A comparison of children’s performance across the two test rounds yielded no signi¿cant difference. As a group, children performed above chance at both test rounds (Age 3;7: t(33) = 8.718, p < .001; Age 4;2: t(33) = 10.078, p < .001; One-Sample T-test).

8. Note that the role of the context supplied by the picture differs in the two pragmatic conditions. In false negatives, the picture depicts the agents, objects, and the activity, which are then referred to in the puppet’s statement. In true negatives, however, the verb used in the puppet’s true negative statements is not given in the picture.

Production and Comprehension of Sentence Negation in Child German

233

Figure 2. Proportion of correct responses in the subtest Comprehension of Negation

Overall, these results show that the children could identify the negation particle nicht within the negated sentence. If the children had simply ignored the sentence negation, a reverse response pattern would have emerged (i.e. answering ‘yes’ instead of ‘no’ and vice versa to all test items). This pattern was not observed in any of the children. 5.5.2 Production and comprehension (H1) H1 hypothesized that children who produce nicht still have a non target-like comprehension of sentence negation. As mentioned in Section 5.2, all children who participated in this study produced nicht appropriately in the contexts used in an elicited production task. Table 3 illustrates children’s individual performance in the comprehension task for both test rounds. All children had ¿ve or more correct responses. For the present purposes, mastery was then calculated for each child, de¿ning mastery as performance above chance. Based on binomial distribution, mastery was reached if a child responded correctly to nine or more out of the 12 test items. According to this de¿nition, 71% of the children (24 of 34) mastered negation at the ¿rst test round, and 74 % (25 of 34) showed mastery six months later. Put differently, at age 4;2 still 26 % of the children had not reached the criterion for mastery. This indicates that although all children used sentence negation appropriately at the age of three, comprehension of negation is not necessarily mastered at the age of four.

234

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

Table 3. Number of subjects by responses correct for both test rounds in the comprehension task Number of correct responses 0/12 1/12 2/12 3/12 4/12 5/12 6/12 7/12 8/12 9/12 10/12 11/12 12/12

Age 3;7 í í í í í 2 5 1 2 5 8 8 3

Age 4;2 í í í í í 1 3 2 3 6 7 3 9

5.5.3 Comparing verb adjacent and verb non adjacent negation (H2) The factor Syntax varied the position of negation particle and ¿nite verb. The proportion of correct responses to verb adjacent and verb non-adjacent sentence negation are depicted in Figure 3. Children performed signi¿cantly better on the verb adjacent condition at age 3;7 (Z = í2.769, p = .006, Wilcoxon-test), and at age 4;2 (Z = í3.099, p = .002, Wilcoxon-test).

Figure 3. Proportion of correct responses to verb adjacent and verb non-adjacent sentence negation

Production and Comprehension of Sentence Negation in Child German

235

Table 4. Mastery (3) and non-mastery (í) of verb adjacent and verb non-adjacent negation by the children

3 verb adjacent 3 verb non-adjacent í verb adjacent 3 verb non-adjacent 3 verb adjacent í verb non-adjacent í verb adjacent í verb non-adjacent sum

N First test round (age 3;7)

N Second test round (age 4;2)

14

14

0

1

11

13

9

6

34

34

To investigate the developmental stages in this syntactic aspect of negation, mastery of verb adjacent negation and verb non-adjacent negation was calculated for each child. In both conditions mastery was again de¿ned as performance above chance. Mastery was reached if a child responded correctly in at least ¿ve of the six test items. Table 4 summarizes the distribution of mastery across the two conditions. 14 out of the 34 children (41%) mastered verb adjacent and verb nonadjacent negation at age 3;7 and 4;2 respectively.9 11 children (32%) mastered verb adjacent, but not verb non-adjacent negation in the ¿rst test round. In the second test round, three of them also mastered verb non-adjacent negation, six did not improve their performance, and two had not reached the criterion for mastery in both conditions. Interestingly, there still were nine children (26%) in the ¿rst test round and six (18%) in the second test round, who did not reach mastery in either condition. As expected, mastery of verb non-adjacent negation without mastery of verb adjacent negation occurred very infrequently (only one child at age 4;2). These results suggest that syntactic aspects of sentence negation are not fully acquired at the age of four. First, children mastered negation in the sentences in which the negator nicht is adjacent to the ¿nite verb. Then mastery of verb 9. Looking at the 14 children mastering both conditions in the ¿rst test round, 10 of them also performed well in the second test round. Three out of 14 mastered verb adjacent negation only, and one child unexpectedly reached mastery of verb nonadjacent negation but not of verb adjacent. Such a response pattern was not found in any child in the ¿rst test round.

236

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

non-adjacent negation requiring object scrambling is reached. This acquisition order is indicated by the fact that, in both test rounds, more children mastered verb adjacent than verb non-adjacent negation. However, note in this analysis the factor Syntax was considered without taking into account the contribution of the factor Pragmatics. 5.5.4 Comparing true and false negatives (H3) In this analysis, the proportion of correct responses to the condition true negative (i.e. a negated sentence that con¿rms the given situation) and the condition false negative (i.e. a negated sentence that is false in a given situation) are compared. The results are presented in Figure 4. Children’s performance on false negatives was signi¿cantly better than on true negatives at both test rounds (Age 3;7: Z = í2.445, p = .014, Age 4;2: Z = í3.341, p = .001, Wilcoxon-test).

Figure 4. Proportion of correct responses to true and false negatives

To look into the individual response patterns, we calculated children’s mastery for true negatives and for false negatives. Again, the criterion for mastery was performance above chance, here de¿ned as giving at least ¿ve out of 6 correct responses per condition. Children’s mastery patterns are depicted in Table 5. 14 out of 34 children (41%) mastered false and true negatives in the ¿rst test round, and 16 (47%) in the second test round.10 In the ¿rst test round, there was only one child (3%) failing in both conditions; this child reached mastery of false negatives at the second test round. 15 children (29%) mastered false nega10. Out of 14 children mastering both conditions in the ¿rst test round, 10 showed the same performance in the second test round, four reached mastery of only false negatives, and one child mastered true negatives but not false.

Production and Comprehension of Sentence Negation in Child German

237

tives, but not true negatives in the ¿rst test round. In the second test round, ¿ve of them also mastered true negatives, nine did not improve their performance, and one child reached mastery of only true negatives. Table 5. Number of children with mastery (3) and non-mastery (í) of false negatives and true negatives (N=34)

N First test round (age 3;7)

N Second test round (age 4;2)

3 false negatives 3 true negatives

14

16

í false negatives 3 true negatives

4

2

3 false negatives í true negatives

15

16

í false negatives í true negatives

1

0

sum

34

34

As expected, mastery of true negatives was rarely observed. Only four of the 34 children mastered true negatives but not false negatives at age 3;7, and only two of these 34 at age 4;2.11 In sum, these results provide evidence that false negatives are acquired before true negatives. This is indicated by the fact that, in both test rounds, more children mastered false but not true negatives than vice versa. Moreover, the number of children mastering false but not true negatives increases over the interval of six months. This suggests that children not mastering either condition in the ¿rst test round mastered false negatives more successfully than true ones at the second.

11. These children consistently responded with “yes” to all test items. Such a response pattern may reÀect a yes-bias, which has been argued to occur if a child is not able to perform a task (Siegal 1997).

238

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

5.5.5 Syntactic and pragmatic factors as predictors for comprehension of negated sentences (H4) Our data show that children performed better on verb adjacent than on verb non-adjacent negation, and that false negatives were easier to interpret than true negatives. Next, we investigated how these syntactic and pragmatic factors interact. Two hierarchical multiple regression analyses were performed to uncover which factor (Syntax or Pragmatics) predicted children’s comprehension of negation better in the two test rounds. The results of the regression analysis for the ¿rst test round are depicted in Table 6. For our model, only the pragmatic factor is a signi¿cant predictor of children’s performance in the test at age 3;7. Table 6. Results of the regression analysis for syntactic and pragmatic factors at the ¿rst test round b

Standard error b

ȕ

t

Signi¿cance

Constant Syntax

2.735 í.294

.262 .166

í.151

í 1.774

.078

Constant Syntax Pragmatics

3.574 í.294 í.559

.347 .159 .159

í.151 í.288

í 1.847 í 3.509

.067 .001

Step 1

Step 2

The regression results for the second test round in Table 7 indicate that both factors Syntax and Pragmatics signi¿cantly predict children’s performance. Based on the value of t, we can assume that the contribution of the predictor Pragmatics was greater than the contribution of the predictor Syntax. Table 7. Results of the regression analysis for syntactic and pragmatic factors at second test round b

Standard error b

ȕ

t

Signi¿cance

Constant Syntax

2.985 í.382

.267 .169

í.192

í2.264

.025

Constant Syntax Pragmatics

4.265 í.382 í.853

.332 .153 .153

í.192 í.428

í2.506 í5.591

.013 .001

Step 1

Step 2

Production and Comprehension of Sentence Negation in Child German

239

In sum, the results of the regression analyses for both test rounds indicate that children’s performance on comprehension of negation is better predicted by the factor Pragmatics. 5.5.6 Summary of the results The children who participated in the study all produced nicht appropriately at the age of three using an elicited production task. However, many children still had dif¿culty interpreting negated sentences at the age of four. These ¿ndings suggest an asymmetrical development with production preceding comprehension. With respect to the syntactic factor, signi¿cant differences between verb adjacent and verb non-adjacent negation were found in both test rounds (age 3;7 and 4;2). Regarding the pragmatic aspects, an examination of the individual comprehension data revealed an advantage for false negatives over true negatives. Moreover, the regression analyses indicated that the pragmatic factor Context Matching is a better predictor for children’s performance on comprehension of negation than the syntactic factor Position.

6

Discussion

With a focus on comprehension, this study investigated the acquisition of sentence negation in German-speaking children. All children who participated in this study produced the negation particle nicht appropriately in the contexts supplied in the subtest Elicited Production of LiSe-DaZ (Schulz and Tracy 2011). Children’s comprehension of negation was assessed using the true value judgement task Comprehension of Negation of LiSe-DaZ. With respect to the relationship between production and comprehension, the results suggest an asymmetrical acquisition. While all children produced nicht target-like in the ¿rst test round at age 3;7, they still had dif¿culty interpreting negated sentences at age 4;2. Thus, production of nicht does not coincide with target-like comprehension of sentence negation, con¿rming hypothesis (H1). A further ¿nding of the study is that the order of the negation particle and the ¿nite verb affect children’s comprehension of negated sentences. Verb adjacent negation was easier to interpret than verb non-adjacent negation in both test rounds, con¿rming (H2). This result may be due to the fact that non-adjacent negation is more complex than verb adjacent in that only the latter requires object scrambling. To correctly interpret the negated sentence, scrambled objects have to be reconstructed within the domain of negation, which still seems to be problematic at age four.

240

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

Regarding the pragmatic aspects of negation, children performed signi¿cantly better on false negatives than on true negatives, con¿rming hypothesis (H3). These results corroborate previous ¿ndings in adults (Wason 1972). However, in view of the fact that four-year-old children in de Villiers and Tager Flusberg’s study (1975) had little dif¿culty interpreting true negatives, this difference between the true and false negatives is surprising. It may be that the truth value judgement task we used is more taxing than the sentence completion task used by de Villiers and Tager Flusberg. In truth value judgement tasks, the child has to match a speci¿c reading of a sentence to a picture. Therefore, the child needs to understand the sentence and to evaluate whether the sentence correctly describes the picture. In sentence completion tasks no such decision about a sentence picture matching is required.12 Concerning the interaction between syntactic and pragmatic aspects of sentence negation, we found that children’s performance on comprehension of negation can better be predicted by the pragmatic factor. Thus, hypothesis (H4) can be con¿rmed. According to Wason (1972) and de Villiers and Tager Flusberg (1975), context plays a major role in the interpretation of negated sentences. If there is no context or if the presented context does not licence the negation, false negatives outperform true negatives, since only false negatives are used in everyday speech. Our ¿ndings con¿rm this. In both sessions, children who did not master true negatives performed well on false negatives. A similar tendency to reject the test sentences is reported by Gualmini (2004), studying the interaction of negation and quanti¿cation. In line with Wason (1972), Gualmini argues “that children’s non-adult behaviour documented in previous studies results from the failure to satisfy the felicity conditions associated with negative statements” (Gualmini 2004: 957f). Our results thus indicate that the lead-in sentences used in the truth value judgement task do not license both types of negatives equally well. Especially, the context given was not speci¿c enough to licence the true negative condition. Summarizing our ¿ndings, we suggest the following acquisition path for the comprehension of sentence negation, taking as the point of departure Wode’s (1977) Stage IV:

12. In addition, one reviewer raises the question of whether false negatives may be more dif¿cult to reject than true negatives in general. In false negatives, it is crucial that the action depicted in the picture is recognized as matching the verb in the test sentence. However, our results do not support this hypothesis.

Production and Comprehension of Sentence Negation in Child German

A. B. C. D.

241

Target-like production of negated sentences Correct interpretation of false negatives with verb adjacent negation Correct interpretation of false negatives with verb non-adjacent negation Correct interpretation of true negatives

To reach converging evidence across tasks, future studies should employ a variety of methods with the same children and within and across the modalities comprehension and production. In addition, further studies are needed with younger children to explore when children ¿rst start interpreting negated sentences correctly and how acquisition of production and comprehension interact in children before age three.

References Baauw, Sergio, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin (this volume) Principle B Delays as a processing problem: Evidence from task effects. Becker, Angelika 2005 The semantic knowledge base for the acquisition of negation and the acquisition of ¿niteness. In The Structure of Learner Varieties, Henriëtte Hendriks (ed.), 263 – 314. Berlin: Mouton de Gruyter. Bloom, Lois Masket 1970 Language development: form and function in emerging grammars. Cambridge, Mass. MIT Press. Clahsen, Harald 1988 Kritische Phasen der Grammatikentwicklung. Eine Untersuchung zum Negationserwerb bei Kindern und Erwachsenen. Zeitschrift für Sprachwissenschaft 7 (1): 3 – 31. Clark, Eve V. 1993 The Lexicon in Acquisition. Cambridge: Cambridge University Press. Clark, Herbert H. 1976 Semantics and Comprehension. The Hague, Paris: Mouton. Crain, Stephen, Weijia Ni, and Laura Conway 1994 Learning, parsing and modularity. In Perspectives on sentence processing, Charles Clifton, Jr., Lyn Frazier, and Keith Rayner (eds.), 443 – 467. Hillsdale, NJ: Lawrence Erlbaum. Déprez, Viviane, and Amy Pierce 1993 Negation and Functional Projections in Early Grammar. Linguistic Inquiry 24 (1): 25 – 67.

242

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

de Villiers, Jill, and Helen B. Tager Flusberg 1975 Some facts one simply cannot deny. Journal of Child Language 2: 279 – 286. Dimroth, Christine 2009 Stepping stones and stumbling blocks. Why negation accelerates and additive particles delay the acquisition of ¿niteness in German. In Functional Categories in Learner Language, Christine Dimroth, and Peter Jordens (eds.), 135 – 168. Berlin, New York: Mouton de Gruyter. Drenhaus, Heiner 2002 On the acquisition of German nicht ‘not’ as sentential negation: Evidence from Negative Polarity items. In Proceedings of the 26. Boston University Conference on Language Development, Barbora Skarabela, Sarah Fish, and Anna H.-J. Do (eds.), 166 – 174. Somerville, MA: Cascadilla Press. Feldman, Sarah Shirley 1972 Children’s understanding of negation as a logical operation. Genetic Psychology Monographs 85: 3 – 49. Gilkerson, Jill, Nina Hyams, and Susan Curtiss 2004 On the scope of negation: More evidence for early parameter setting. In Proceedings of GALA 2003, Jacqueline van Kampen and Sergio Baauw (eds.), 175 – 185. Utrecht: LOT. Grimm, Hannelore, Hermann Schöler, and Magret Wintermantel 1975 Zur Entwicklung sprachlicher Strukturformen bei Kindern. Weinheim: Beltz. Grimm, Hannelore 2001 SETK 3–5. Sprachentwicklungstest für drei- bis fünfjährige Kinder. Diagnose von Sprachverarbeitungsfähigkeiten und auditiven Gedächtnisleistungen. Göttingen: Hogrefe. Gordon, Peter 1996 The truthvalue judgement task. In Methods for assessing children’s syntax, Dana McDaniel, Cecile McKee, and Helen Smith Cairns (eds.), 211 – 231. Cambridge, Mass: MIT Press. Gualmini, Andrea 2004 Some knowledge children don’t lack. Linguistics 42 (5): 957 – 982. Haider, Hubert 2004 Wie viel Syntax braucht die Semantik, und wie viel Semantik enthält die Syntax? Tidsskrift for Sprogforskning 2 (2): 71 – 90. Hamann, Cornelia 2000 Negation, In¿nitives, and Heads. In The Acquisition of Scrambling and Cliticization, Susan M. Powers and Cornelia Hamann (eds.), 423 – 477. Dodrecht: Kluwer.

Production and Comprehension of Sentence Negation in Child German

243

Hickmann, Maya 2000 Pragmatische Entwicklung. In Sprachentwicklung. Enzyklopädie der Psychologie, CIII, Band 3, Hannelore Grimm (ed.), 193 – 227. Göttingen: Hogrefe. Jacobs, Joachim 1982 Syntax und Semantik der Negation im Deutschen. Munich: Wilhelm Fink. Krämer, Irene 1998 Children’s Interpretation of Inde¿nite Object Noun Phrases: Evidence from the Scope of Negation. In Linguistics in the Netherlands 1998. Renée van Bezooijen and René Kager (eds.), 163 – 174. Amsterdam/ Philadelphia: John Benjamins. McDaniel, Dana, Cecile McKee, and Helen Smith Cairns 1996 Methods for Assessing Children’s Syntax. MIT Press. Musolino, Julien, and Jeffrey Lidz 2006 Why children aren’t universally successful with quanti¿cation. Linguistics 44 (4): 817 – 852. Meisel, Jürgen M. 1997 The acquisition of the syntax of negation in French and German: Contrasting ¿rst and second language development. Second Language Research 13(3): 227 – 263. Müller, Anja, Petra Schulz, and Barbara Höhle 2010 How the understanding of focus particles develops: Evidence from child German. 32. Jahrestagung der Deutschen Gesellschaft für Sprachwissenschaft (DGfS), Berlin. Nederstigt, Ulrike 2003 Auch and noch in child and adult German. Berlin, New York: Mouton de Gruyter. Ofner, Daniela 2007 Children’s Understanding of Negation and Focus – Development of an Experiment. M.A. Thesis, University of Mannheim. Penner, Zvi, Rosemarie Tracy, and Jürgen Weissenborn 2000 Where scrambling begins: Triggering object scrambling at the early stage in German and Bernese Swiss German. In The Acquisition of Scrambling and Cliticization, Susan M. Powers and Cornelia Hamann (eds.), 127 – 164. Dodrecht: Kluwer. Pollock, Jean-Yves 1989 Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry 20: 365 – 424. Rumain, Barbara 1988 Syntactics of Interpretation of Negation: A Developmental Study. Journal of Experimental Child Psychology 45: 119 – 140.

244

Magdalena Wojtecka, Corinna Koch, Angela Grimm, and Petra Schulz

Schulz, Petra 2007 Erstspracherwerb Deutsch: Sprachliche Fähigkeiten von Eins bis Zehn. In Diagnostik am Schulanfang (=Entwicklungslinien der Grundschulpädagogik. Band 3), Ulrike Graf and Elisabeth Moser Optiz (eds.), 67 – 86. Baltmannweiler: Schneider Hohengehren. Schulz, Petra, and Rosemarie Tracy 2011 Linguistische Sprachstandserhebung – Deutsch als Zweitsprache (LiSeDaZ). Göttingen: Hogrefe. Schulz, Petra, Rosemarie Tracy, and Ramona Wenzel 2008 Entwicklung eines Instruments zur Sprachstandsdiagnose von Kindern mit Deutsch als Zweitsprache: Theoretische Grundlagen und erste Ergebnisse. In Kinder und Jugendliche mit Migrationshintergrund – Empirische Befunde und Forschungsdesiderate, Bernt Ahrenholz (ed.), 17 – 41. Freiburg im Breisgau: Fillibach. Siegal, Michael 1997 Knowing children. Hove: Psychology Press. Steube, Anita 2006 The inÀuence of operators on the interpretation of DPs and PPs in German information structure. In The Architecture of Focus, Valéria Molnár, and Susanne Winkler (eds.), 489 – 516. Berlin/New York: Mouton de Gruyter. Steube, Anita, and Stefan Sudhoff 2007 Negation und Fokuspartikeln in der Informationsstruktur der deutschen Standardsprache der Gegenwart. In Von der Pragmatik zur Grammatik, Sandra Döring and Jochen Geilfuß-Wolfgang (eds.), 87 – 108. Leipzig: Leipziger Universitätsverlag. Stromswold, Karin, and Kai Zimmermann 1999 Acquisition of Nein and Nicht and the VP-Internal Subject Stage in German. Language Acquisition 8 (2): 101 – 127. Sudhoff, Stefan 2008 Zum relativen Skopus von Negation und Fokuspartikeln im deutschen Mittelfeld. In Beiträge zu Sprache und Sprachen 6: Vorträge der 16. Jahrestagung der Gesellschaft für Sprache und Sprachen, Karin Pittner (ed.), 317 – 328. Munich: lincomeuropa. Thornton, Rosalind 1996 Elicited production. In Methods for assessing children’s syntax, Dana McDaniel, Cecile McKee, and Helen Smith Cairns (eds.), 77 – 102. Cambridge, Mass: MIT Press. Tracy, Rosemarie, and Dieter Thoma 2006 Deutsch als frühe Zweitsprache: Zweite Erstsprache? In Kinder mit Migrationshintergrund. Spracherwerb und Fördermöglichkeiten, Bernt Ahrenholz (ed.), 58 – 79. Freiburg im Breisgau: Fillibach.

Production and Comprehension of Sentence Negation in Child German

245

Unsworth, Sharon 2005 Child L2, Adult L2, Child L1: Differences and Similarities. A study on the acquisition of direct object scrambling in Dutch. Doctoral dissertation. Utrecht: LOT. Wason, Paul C. 1965 The Contexts of Plausible Denial. Journal of Verbal Learning and Verbal Behavior 4: 7 – 11. Wason, Paul C. 1972 In Real Life Negatives are False. Logique et Analyse 15:17 – 38. Winkler, Stefanie 2006 Finiteness in ¿rst and second language acquisition. Annual Report. MaxPlanck Institut für Psycholinguistik, Nijmegen. Wode, Henning 1977 Four early stages in the development of L1 negation. Journal of Child Language 4: 87 – 102.

Principle B delays as a processing problem: Evidence from task effects* Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

1.

Introduction

Several studies show that 3-to-6-year-old children often allow object pronouns to be identi¿ed with local c-commanding antecedents (1a). The same children perform highly adult-like on the interpretation of reÀexives (Chien and Wexler 1990). This phenomenon has, erroneously, often been called the Delay of Principle B Effect. (1)

a.

John touched him.

[50% adult-like performance]

b.

John touched himself.

[100% adult-like performance]

Recently, this phenomenon, which we will call the Pronoun Interpretation Problem (Coopmans 2000), has been challenged as being an experimental artifact (Conroy et al. 2009). In this paper we will claim that the Pronoun Interpretation Problem is real, but also that it is due to a processing problem, not to missing knowledge of binding or coreference principles. To support this claim we will present experimental evidence from Dutch and Spanish children and Spanish agrammatic patients suffering from Broca’s aphasia. We will show

* This article was supported by the project Comparative Psycholinguistics, which is funded by the Netherlands organization for Scienti¿c Research (NWO). We are indebted to the children, their parents, and teachers of the Anne Frank School, in Bunnik, Netherlands, and of the Colegio San Ignacio, Oviedo, Spain. We also thank the staff and patients of University of Málaga/CIMES in Málaga and Urbamin hospital in Elcano, Spain. Special thanks go out to Alicia Pérez, Marcelo Berthier, Fernando Cuetos, Christina Green and Julián Hinojosa for helping us organize our study in Oviedo, Pamplona and Málaga. We thank Joke de Lange and Nada Vasic for their usefull comments.

248

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

that the performance of these populations is highly sensitive to the processing load that different experimental methods impose on them.

2.

Pronoun Interpretation Problem as a syntax-discourse interface problem

It is nowadays widely accepted that children’s dif¿culties with pronoun interpretation in sentences such as (1a) are not due to any problem with Principle B of the Binding Theory, but with a syntax-discourse interface principle that constrains local coreference.1 This principle, which was called Principle P by Chien and Wexler (1990) and Rule I by Grodzinsky and Reinhart (1993), limits cases of local corereference (also called pragmatic coreference) to contexts such as (2). (2)

Everybody hates Oscar. John hates him, Mary hates him, even Oscar hates him.

Apparently, children often do not apply this principle, to the effect that local coreference readings are accepted by them across the board. The question that language acquisition researchers have tried to answer is why children often seem not to apply Rule I/Principle P. We can distinguish roughly three kinds of approaches: 1. Incomplete acquisition approach 2. Experimental artifact approach 3. Limited processing resources approach The ¿rst approach considers the Pronoun Interpretation Problem the result of an incomplete acquisition process. This view was defended by Chien and Wexler (1990), who claimed that children had not yet acquired Principle P. If Principle P is absent in children’s grammar, nothing prevents them from accepting correference readings in constructions such as (1a). The second approach views the Pronoun Interpretation Problem as mainly the result of de¿ciencies in the experimental design. Conroy et al. (2009) 1. Children’s performance in sentences with a quanti¿ed subject, at least in English, seems to improve as well (Chien and Wexler 1990), which led these authors to argue (correctly in our view) that children’s errors are not due to the lack of grammatical knowledge of Principle B.

Principle B delays as a processing problem: Evidence from task effects

249

showed that children perform much more adultlike on the interpretation of pronouns when certain de¿ciencies in the experimental design are corrected. This means that according to this approach children’s linguistic competence is not different from adults, although they may be more sensitive to de¿ciencies in the experimental design than adults. The third approach assumes that although 5-year-olds have adult-like linguistic competence, they often have dif¿culties with the application of this implicit knowledge, due to their more limited processing capacity. This approach is defended by Grodzinsky and Reinhart (1993), who claim that children know the binding principles and Rule I, but due to their limited processing resources they often fail in their atempt to apply Rule I. A similar approach is followed by Avrutin to account for both the coreference results and other aspects of child language, such as root in¿nitives and determiner omission (Avrutin 1999, 2004). In the next sections we will argue that both the incomplete acquisition approach and the experimental artefact approach fail to account for some of the (experimental) results on children’s interpretation of pronominal anaphora. We will also show that recent experimental results provide additional evidence for our claim that the Pronoun Interpretation Problem is due to children’s dif¿culties with the use of syntactic and interface principles that regulate referential dependencies.

3.

Incomplete acquisition

Chien and Wexler’s (1990) claim that the Pronoun Interpretation Problem is the result of the absence in children’s grammar of Principle P, a pragmatic principle that constrains local coreference, cannot easily explain why this delay is absent in spontaneous production. As Bloom et al. (1994) show, children seem to respect Principle B in their spontaneous production. If a principle that constrains local coreference were absent in little children, one expects to ¿nd instances of locally coreferring pronouns at least occasionally. A second argument against incomplete acquisition as an account for children’s dif¿culties with the interpretation of pronouns comes from aphasia. Grodzinsky et al. (1993) have shown that the Pronoun Interpretation Problem is not limited to children, but also shows up in English speaking patients with agrammatic Broca’s aphasia. Similar results have also been found for Dutch and, to some extent, in Spanish agrammatic Broca’s aphasics (Ruigendijk et al. 2006; Vasic 2006; Baauw and Cuetos 2003). Unlike children, agrammatics are not in the process of language acquisition, which means that their failure to

250

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

reject local coreference in sentences such as (1a) cannot be due to incomplete knowledge. Although some researchers have tried to explain patients’ errors in terms of “loss of linguistic knowledge”, others point at the variability in performance between and within patients as an argument in favor of processing accounts (Avrutin, Lubarksky, and Green 1999; Kolk 1987). It is of course possible that the Pronoun Interpretation Problem in children and in agrammatics are independent phenomena. However, pronominal anaphora are not the only area where children and agrammatics show similarities. Other areas are the production of articles and tense, the interpretation of passives and the interpretation of wh-elements. This justi¿es a single approach that accounts for the linguistic performance of both populations (Avrutin 2006).

4.

Experimental artifact

Conroy et al. (2009) argue that the Pronoun Interpretation Problem that was found in children by Thornton and Wexler (1999) and other studies that used the truth value judgement task is due to de¿ciencies in the experimental setup. In the particular variety of the truth value judgement task used by Thornton and Wexler, children were presented stories involving two or more individuals of the same gender. These stories had a ¿nal scene in which one of the characters performs an action on either himself (e.g. a boy washing himself), or on another individual in the scene (e.g. a boy washing another male individual). In trials that test children’s interpretation of pronouns this ¿nal scene is paired with the test sentence that contains an object pronoun (the boy is washing him). The child is asked to indicate whether the test sentence correctly describes the ¿nal scene. If a child accepts local pronominal coreference, she will accept The boy is washing him as a true statement about a scene in which a boy is washing himself. Children who, like adults, reject local coreference, will reject the sentence as a true statement of what happened in the ¿nal scene. Conroy et al. (2009) argue that the reason why many studies that used a truth value judgement task found a strong Delay of Principle B-Effect was due to the fact that they failed to satisfy a pragmatic felicity condition that Crain and Thornton (1998) call the condition of plausible dissent. This condition states that in order to reject a sentence as a true description of a situation, the opposite situation, which would lead to acceptance of the sentence, should be under consideration at some point. In the case of pronominal anaphora this means that in order to reject The boy is washing him as a true statement about a scene in which a boy is washing himself, the situation in which the boy is washing somebody else should have been presented as a potential outcome of the

Principle B delays as a processing problem: Evidence from task effects

251

story that is used as a context of the test sentences. Not satisfying the condition of plausible dissent will make the sentence internal antecedent pragmatically a more likely antecedent of the pronoun than the sentence external antecedent. Conroy et al. (2009) claim that in many studies on pronominal anaphora this is exactly what happens. For example, in Thornton and Wexler’s (1999) study children were presented a story in which Bert asked a reindeer to brush him. In the end the reindeer refused, and Bert decides to brush himself. Then the child heared the test sentence Bert brushed him. The problem with this story is that it was all about Bert asking a reindeer to brush him; the possibility of Bert brushing the reindeer was not a realistic possible outcome of the story. This made the reindeer a relatively unaccessible antecedent of the pronoun. As a result, the child is ‘forced’ to choose the sentence internal antecedent, leading to 58% yes-responses. Conroy et al. (2009) present experimental evidence showing that when the condition of plausible dissent is satis¿ed, which makes both antecedents equally salient, the percentage yes-responses drops dramatically to 11%. However, if the Pronoun Interpretation Problem is really an artifact, it is dif¿cult to explain why it is subject to cross-linguistic variation. For example, it does not show up in the Spanish counterparts of (1a), neither in children, nor in agrammatics (Baauw, Escobar, and Philip 1997; Baauw and Cuetos 2003). (3)

La niña la señala the girl her points-at ‘The girl is pointing at her.’

[90% adultlike performance]

If the Pronoun Interpretation Problem were merely an artifact, i.e. the result of a de¿cient experimental design, one would expect Spanish children and agrammatics to show the same rate of yes-responses as their Dutch or English peers.2 Also, the Pronoun Interpretation Problem appears to be structure sensitive; it is much stronger in Exceptional Case Marking sentences than in normal transitive sentences, such as (1a) (Philip and Coopmans 1996). (4)

Jan ziet hem dansen. ‘John sees him dance.’

[20% adultlike responses]

2. Note that Conroy et al. (2009) brieÀy discuss the cross-linguistic differences with respect to the Pronoun Interpretation Problem and recognize that these indicate that there may be something real about the Pronoun Interpretation Problem in acquisition, but that its effect is much milder, even in English, if the experiment fully respects the condition of plausible dissent.

252

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

In fact, Exceptional Case Marking sentences are the only type of sentences that give rise to a Pronoun Interpretation Problem in Romance children and agrammatics (Baauw and Cuetos 2003). (5)

La niña la ve bailar. ‘The girl sees her dance.’

[50% adultlike performance]

This seems to suggest that the Pronoun Interpretation Problem is not just an artefact, but a real developmental phenomenon sensitive to structural properties of languages.

5.

Pronoun Interpretation Problem as a processing problem

Grodzinsky and Reinhart (1993) already claimed that the Pronoun Interpretation Problem that shows up in (1a) is not the result of the lack of knowledge about Principle B. Instead, they claim that it is due to a limited ability to apply a syntax-discourse interface principle called Rule I. This rule states that local coreference is only possible if it can be semantically distinguished from a binding construal. Since this is not the case in (1a), adults will reject a local coreference reading for this sentence. Importantly, Rule I requires the comparison of two different construals of the same sentence; a binding construal, and a coreference construal. Grodzinsky and Reinhart (1993) argue that, due to their limited processing capacity, children often fail in their attempt to make this comparison. The same can be claimed for agrammatic Broca’s aphasics (Grodzinsky et al. 1993; Avrutin 1999). As a result, Rule I breaks down, and children revert to a guessing strategy, sometimes giving a yes-response, sometimes a no-response. Rule I is only invoked when local coreference is an option. As a result it does not play a role in the interpretation of Spanish sentences such as (3). The reason is that Spanish weak pronouns are syntactic clitics, and as Baauw (2000); Baauw and Cuetos (2003) argue, syntactic clitics do not allow local coreference relations. Consequently, the coreferential reading of sentences such as (2) is much harder to get in a language with syntactic clitics, such as Spanish. If local coreference is no option for syntactic clitics, Rule I will not be invoked in any sentence containing syntactic clitics. If Rule I is not invoked, it cannot break down either, which explains the highly targetlike performance of Spanish children and agrammatics on (3). Children and agrammatics’performance on Exceptional Case Marking constructions can also be explained as the result of processing limitations. As we

Principle B delays as a processing problem: Evidence from task effects

253

showed in section 4, Exceptional Case Marking constructions lead to additional errors, both in children and agrammatics. For Spanish children and agrammatics this is the only context that gives rise to a Pronoun Interpretation Problem. This contrast between Exceptional Case Marking and simple transitive sentences cannot be easily explained by standard Binding Theory (Chomsky 1981), but does follow naturally from Reinhart and Reuland’s ReÀexivity model (Reinhart and Reuland 1993). In this alternative model, Principle B is a condition on reÀexive predicates. This means that this principle has nothing to say about the binding relation between John and him in (6): John and him belong to different verbal predicates, so that no reÀexive predicate is formed. (6)

Johni saw [himi dance]

Yet, the reÀexive interpretation in (6) is incorrect. Reuland (2001) argues that the ungrammaticality of a reÀexive reading in (6) is not due to any binding principle but to an economy condition on referential dependencies. Reuland (2001) argues that referential dependencies can be established at different levels of the language system. They can be established in narrow syntax (computational system), semantics or discourse (pragmatics). Local coreference, as in (2), is an example of a referential dependency established in discourse. The binding relation between the Dutch anaphor zich and Jan in (7) is an example of a referential dependency established in narrow syntax. (7)

Jan zag [zich dansen] |_______| A-Chain

‘John saw himself dance.’

According to Reuland (2001), zich needs to establish a feature checking relation with a local determiner phrase, in this case Jan. As a byproduct of this syntactic operation, a referential dependency is established in the form of an A-Chain. In principle, the subject of the embedded Exceptional Case Marking clause can also be a pronoun. In that case feature checking and A-Chain formation are not an option, but pronouns, such as hem ‘him’ can establish a ‘normal’ bound variable relation with the main clause subject determiner phrase as in (8). (8)

Jani zag [hemi dansen] ‘John saw him dance.’

254

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

This would represent a referential dependency establish at the level of semantics. Recall, that Principle B does not bar this construal, since hem and Jan are not co-arguments of the same predicate. Nevertheless, unlike (7), (8) is ungrammatical, just like (6). Reuland (2001) argues that the ungrammaticality of (8) is due to an economy principle which states that dependencies established in narrow syntax (A-Chains) are cheaper than dependencies established outside narrow syntax, for example, in semantics (bound-variable relations). Stated differently; (8) is ungrammatical because (7), which is semantically equivalent, is cheaper. But if syntactic operations are cheaper than extra-syntactic operations, why do children and agrammatics often accept (8)? Avrutin (1999, 2004) argues that syntax need not be the cheapest option in populations that are known to have a limited capacity to process syntactic operations, such as children and agrammatic Broca’s aphasics. If syntax is no longer the cheapest option to establish referential dependencies, extra-syntactic operations become an equally economical option, which will lead children and agrammatics often to accept (8) with a reÀexive interpretation. Note ¿nally that Dutch children are predicted to accept a reÀexive interpretation of pronominal Exceptional Case Marking sentences more often than Spanish children, as shown in (4) and (5). In Dutch pronominal Exceptional Case Marking sentences not only the bound varial construal (8) is available, but also a local coreference construal. As a result, children who manage to block a bound variable reading (8), by activating the A-Chain construal (7), may still establish a local coreference relation between the embedded subject pronoun and the main clause subject determiner phrase, and subsequently fail in their atempt to apply the relevant principle that constrains local coreference (Rule I). Spanish children, on the other hand do not have this latter option; they either succeed in blocking the bound variable construal, or they fail, but there is no local coreference construal possible. In fact, the differences between Spanish and Dutch give rise to the following probability calculation (Baauw 2000). Suppose Dutch children manage to block the bound variable construal 50% of the time, and that 50% of the time these children also manage to block a coreference construal. This predicts that a reÀexive interpretation of pronominal Exceptional Case Marking sentences will be blocked 25% of the time, which comes close to the roughly 20% rejections of a reÀexive interpretation that Philip and Coopmans (1996) found. If Spanish children block the bound variable construal 50% of the time, 50% rejection of a reÀexive interpretation of pronominal Exceptional Case Marking sentences is predicted, since these children will not have access to a local coreference construal. This comes close

Principle B delays as a processing problem: Evidence from task effects

255

to the 40% to 60% rejections found in several studies (Baauw 1997; Baauw 2000; Baauw and Alija 2005).

6.

Pronoun Interpretation Problem and task effects

If the Pronoun Interpretation Problem is due to limited processing resources of children and agrammatics, it is expected that children and agrammatics will perform more targetlike on experimental methods that impose low processing load, such as picture selection tasks, than experimental methods imposing high processing load, such as truth value judgement tasks. We assume that when the child hears a sentence containing a pronoun, he will try to ¿nd an antecedent for this pronoun that satis¿es both Principle B and constraints on local coreference. This means children will initially consider extra-sentential antecedents to resolve the pronoun, but may consider the sentence internal antecedent, depending on contextual and task related factors. In a picture selection task a child sees two or three pictures and has to choose the one that best depicts the test sentence (e.g. John is washing him). One of the pictures represents a “reÀexive” action, another one a “transitive” action (a third one represents an unrelated action). The child is invited to point at the picture that matches the sentence. In the truth value judgement variety that we used (sometimes called a picture veri¿cation task) a child sees a single picture and has to judge whether an accompanied sentence correctly describes this picture. Some pictures elicit a yes-response in healthy adults (non-reÀexive action), some elicit a no-response (reÀexive action). A major difference between the two methods is that in the picture selection task, the child has the freedom to choose the picture that best matches the sentence. Since we assume that children initially try to ¿nd an extra-sentential antecedent for the pronoun, they will tend to avoid to select pictures, i.e. readings, that are complicated processingwise, such as local coreference readings (due to Rule I etc.). In the truth value judgement task, on the other hand, the child does not have this freedom; she is ‘forced’ to consider a particular reading, and judge whether it is acceptable. This means that the child has no opportunity to avoid readings that require more processing resources, such as a local coreference reading. In order to test our prediction that children and agrammatics perform more targetlike on Picture-Selection than on truth value judgement, we will present experimental results of recent picture selection experiments with Spanish children and agrammatics, with the aim to compare them with previous truth value judgement results. We will also present the results of a combined truth value judgement –picture selection task with Dutch speaking children.

256

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

7.

Picture selection experiments Spanish

7.1.

Picture selection Spanish children

In this experiment 32 Spanish speaking children, ranging from 5;3 to 6;2 years old, with a mean age of 5;9 from a primary school in Oviedo (Asturias, Spain) participated. The test materials consisted of 3 sentence types (reÀexives in transitive sentences, pronouns in transitive sentences and Exception Case Marking constructions). Every sentence type was tested with 8 trials, each containing different verbs. In addition 38 ¿ller items were included. The total number of test items was 62, administered to the child in one session of 15 to 20 minutes. Before this session some practice items were administered. In addition 8 healthy adults were tested. The child task consists of pointing at the picture that matches the test sentence. For a complete list of experimental items, see Appendix A. In Figure 1 we provide an example of a picture selection trial. Test sentence: Primero la niña y la mujer bailaron, y luego la niña la pinta. ‘First the girl and the mother danced and then the girl painted her.’ Target response: rightmost picture

Figure 1. Trial condition B

In Table 1 we provide the percentage correct responses on the 3 main conditions: As can be seen, Spanish children performed highly targetlike on sentence type B, i.e. sentences such as (3), and A, i.e. similar sentences with the reÀexive pronoun se. Their performance on pronominal Exceptional Case Marking sen-

Principle B delays as a processing problem: Evidence from task effects

257

tences was much lower. An ANOVA shows that the difference between the three conditions is signif¿cant (F = 19,117, p < 0,0001). A Tukey-Kramer posthoc test shows that A does not differ signi¿cantly from B, but that Exceptional Case Marking differs from B and A (A vs. ECM, p < 0,001; B vs. ECM, p < 0,001). The adult controls performed at ceiling level on all 3 conditions. Table 1. B = pronominal transitive; A = reÀexive transitive; ECM = pronominal ECM A 93 (SD = 0,10)

B 94 (SD = 0,11)

ECM 75 (SD = 0,25)

In Figure 1 we compare the picture selection results on conditions B and Exceptional Case Marking with the results of earlier studies carried out with truth value judgement tasks on children of around 5 years of age. truth value judgement studies: TVJ-1 (Baauw, Escobar, and Philip 1997; Baauw and Cuetos 2003), TVJ-2 (Baauw 2000) and TVJ-3 (Baauw and Alija 2005). The results of the truth value judgement studies are percentage correct no-responses on “no” trials. Standard deviations are provided below the ¿gure. 100 90 80 70 60 B

50

ECM 40 30 20 10 0 PS

TVJ-1

TVJ-2

TVJ-3

SD condition B: PS 0,11; TVJ-1 0,23; TVJ-2 0,29; TVJ-3 0,24 SD condition ECM: PS 0,25; TVJ-1 0,36; TVJ-2; 0,38; TVJ-3 0,34 Figure 2. B = pronominal transitive; ECM = pronominal ECM

As Figure 2 shows, children scored at ceiling on normal transitive sentences, independently of the test method. Their performance on Exceptional Case Marking was sensitive to task effects; around 50% in truth value judgement tasks and 75% in picture selection tasks.

258

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

7.2.

Picture selection Spanish agrammatics

In this experiment seven male patients with agrammatic Broca’s aphasia participated, with ages that ranged from 42 to 68 years, and with a mean age of 53;9 years. All patients suffered a Cerebro Vascular Accident in the left hemisphere of the brain, and were diagnosed as agrammatic Broca’s aphasics by the Barcelona Test (Peña Casanova 1991) in one case, and by the Western Aphasia Battery (Kertesz 1982) in the other six cases. This diagnosis was con¿rmed by their speech therapists on the basis of their clinical observation. Six patients were recruited from the medical research centre CIMES, of the University of Málaga, in the south of Spain. One patient was recruited from the Ubarmin clinic in Elcano, near Pamplona, Navarra. In addition, six non-brain-damaged controls were tested. The age of the controls ranged from 36 to 75 years, and with a mean age of 53;8. The test materials consisted of 3 sentence types (reÀexives in transitive sentences, pronouns in transitive sentences and pronominal Exceptional Case Marking constructions). The materials and the way of presenting them were similar to those of the child study, except that every sentence type was tested with 15 trials, each containing different verbs. In adition 55 ¿ller items were included. The total number of test items was 100, administered to the patient in one session of 20 to 25 minutes, with a brief pause after the ¿rst 50 items. Before the session some practice items were administered. In Appendix B a complete list of experimental items is provided. In Appendix C further (clinical) details about the patients are provided. In Table 2 we provide the percentage correct responses on the 3 main conditions: Table 2. B = pronominal transitive; A = reÀexive transitive; ECM = pronominal ECM A 76 (SD = 0,09)

B 89 (SD = 0,08)

ECM 61 (SD = 0,36)

An ANOVA shows that there is an effect of condition (F = 4,563, p < 0,0336). A Tukey-Kramer posthoc test shows that B differs from ECM (p < 0.05), but that A does not differ signi¿cantly from B, nor does A differ from ECM. In Figure 3 we compare this result with Baauw and Cuetos (2003), a truth value judgement study. The results on B and ECM of the truth value judgement study represent the percentage no-responses on “no” trials. Standard deviations are provided below the ¿gure.

Principle B delays as a processing problem: Evidence from task effects

259

100 90 80 70 60

B

50

ECM

40 30 20 10 0 PS

TVJ

SD condition B: PS 0,08; TVJ 0,21 SD condition ECM: PS 0,36; TVJ 0,21 Figure 3. B = pronominal transitive; ECM = pronominal ECM

As can be seen, whereas the patients in both studies performed highly targetlike on condition B, there is a strong contrast in performance on the Exceptional Case Marking condition; patients tested with a truth value judgement task perform much less targetlike on this condition.3 Although both the child and patient studies indicate that picture selection tends to lead to higher targetlike performance than truth value judgement studies, there are factors that make direct comparison of these studies dif¿cult; the number of participants is different, the age range of the participants and the materials used are often different, and there are some differences with respect to the exact way the task was administered (in some the test sentences were “pronounced” by a puppet, in other the investigator pronounced them, sometimes the test sentence was phrased as a question, sometimes as declarative sentence). In order to make the results of the two test methods fully comparable the same children should be tested with the same materials during the 3. One reviewer notes that children’s performance on the conditions ECM and A in the picture selection task is more targetlike than the patients’ performance on these conditions. We believe that the difference in ECM may be accidental, since it is similar to the variation that one ¿nds between different experiments with children (see Figure 2). This variation is probably related to differences in materials, exact age of children and the way the experiment was presented. The differences in condition A seem more real, and have also been found in Baauw and Cuetos (2003), in which patients were tested with a truth value judgement task. Since the SE anaphor se in Spanish Principle A sentences can be considered a reÀexive-marking morpheme (Baauw et al. 2006 ), some patients problems with this condition might point at a problem with reÀexivization itself or with the use of morphological morphemes that mark reÀexivity.

260

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

same test session. This is what we did in the experiment presented in the next section. 7.3.

Combined picture selection – truth value judgement experiment

In this experiment 20 Dutch speaking children were administered a test in which picture selection and truth value judgement items alternate. The children ranged from 4;11 to 6;0, with a mean age of 4;10. The test consisted of 36 items, 3 sentence types (pronouns in transitive sentences, pronouns in Exceptional Case Marking sentences, and ¿llers). In case of the picture selection items, each sentence type was tested with 6 trials, each containing a different verb. In case of the truth value judgement items, the “yes” items were tested with three trials, the “no” items with six items, each containing a different verb. The children were tested in one session of 20 minutes. In the B condition trials of the truth value judgement task children heard sentences such as Een groene en een paarse tele-tubby…Ik denk dat de paarse hem knuffelt ‘A green and a purple tele-tubby…I think that the purple one is hugging him’. In the “yes” trials they were were shown a picture of purple tubby hugging a green one, in the “no” trials the purple tubby hugged himself. For the picture selection tasks each trial consisted of two pictures, one of which was correct. In the case of the Exceptional Case Marking condition children heard a question such as In welk plaatje ziet de paarse tubby hem touwtje springen? ‘In which picture does the purple tuby see him jump-rope?’ At the same time they saw two pictures, one representing a green tele-tubby jump-roping with his back to the mirror, while a purple tubby was watching him (correct picture), in the other one the purple tubby was jump-roping watching himself in the mirror, while the green tubby sat down watching him (wrong picture). In Appendix D we provide a complete list of test items. In Figure 4 we present the results of the experiment, in terms of percentage targetlike performance. Note that for the truth value judgement task only correct no-responses on no-items are presented. Standard deviations are provided below the ¿gure. For both Exceptional Case Marking and B there is more target-like performance on picture selection (80% targetlike performance on B, 53% targetlike performance on ECM) than truth value judgement (60% targetlike no-responses on B, 33% targetlike non-responses on ECM). An ANOVA shows that there is a signi¿cant effect of methodology (F = 19.504, p = .000) as well as an effect of structure (F = 30.009, p = .000).

Principle B delays as a processing problem: Evidence from task effects

261

90 80 70 60 50

B

40

ECM

30 20 10 0 PS

TVJ

SD condition B: PS 0,23; TVJ 0,36 SD condition ECM: PS 0,21; TVJ 0,28 Figure 4. B = pronominal transitive; ECM = pronominal ECM

8.

Discussion

The results support a processing account for the Pronoun Interpretation Problem. If children’s (and agrammatics’) problem were one of missing knowledge, methodology should not have mattered; knowledge that is not available cannot be revealed by any methodology. The experimental results con¿rm our prediction that children (and agrammatics) perform less targetlike on truth value judgement tasks than on Picture Selection task. This is in line with our prediction; Truth-Value-Judgement tasks force children (and patients) to consider a speci¿c reading of the test sentence, represented by a picture. When this reading involves the establishment of a referential dependency between a pronoun and a local antecedent, this gives rise to frequent non-targetlike acceptance of a local coreference interpretation (in the case of Dutch and English), or a bound-variable construal (in Exceptional Case Marking sentences), in populations that do not have suf¿cient processing resources to always constrain these readings, such as children and agrammatics (Grodzinsky and Reinhart 1993; Reuland 2001). In picture selection tasks, on the other hand, subjects can always choose a picture corresponding to the non-reÀexive reading for a pronominal test sentence, which does not imply any processing dif¿culty. Note that this does not mean that children will always perform poorly on truth value judgement tasks. As Conroy et al. (2006) show, when a truth value judgement task is pragmatically optimalized, by making the

262

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

sentence external reading more prominent, this may help children to ignore the local coreference reading and provide correct “no” responses. Finally, the present results also allow us to speculate about production. In production the child is not forced to consider the possibility of local coreference/binding reading (as in truth value judgement); it is not even “invited” to do so (as in picture selection tasks). A child who wants to express a “reÀexive” meaning will not likely select a pronoun. If a child wants to express a nonreÀexive meaning, pronouns are the only option.4 As a result, the Pronoun Interpretation Problem does not show up in production.

4. In an elicited production task Baauw et al. (2006) show that Dutch 5-year old children do not often use SE anaphors, such as zich, but do often use SELF anaphors, like zichzelf. The limited use of zich is due to the fact that it involves establishing a referential dependency in narrow syntax, whereas SELF anaphors involve extrasyntactic referential dependencies. In any case, pronouns were hardly used to express a reÀexive interpretation.

Principle B delays as a processing problem: Evidence from task effects

263

Appendix A Part 1 # 1

Type Item A Primero la mujer y la niña cantaron, y luego la niña se pintó. ‘First the woman and the girl sang, and then the girl painted herself.’ Primero la mujer y la niña rieron, y luego la mujer la tocó.

2

B

3

‘First the woman and the girl laughed, and then the woman touched her.’ ECM Primero el hombre y el niño lloraron, y luego el niño lo vio beber algo. ‘First the man and the boy sang, and then the boy saw him drink something.’ Primero el hombre y el niño saludaron, y luego el niño lo rascó.

7

B

8

‘First the man and the boy greeted, and then the boy scrached him.’ ECM Primero el hombre y el niño patinaron, y luego el hombre lo vio comer algo.

B

‘First the man and the boy skated, and then the man saw him eat something.’ Primero la mujer y la niña caminaron, y luego la mujer la lavó.

18

A

‘First the woman and the girl wlked, and then the woman washed her.’ Primero la mujer y la niña rieron, y luego la mujer se tocó.

19

‘First the woman and the girl laughed, and then the woman touched herself.’ ECM Primero la mujer y la niña rieron, y luego la mujer la vio bailar.

12

22 23 28

29

B

‘First the woman and the girl laughed, and then the woman saw her dance.’ Primero la mujer y la niña leyeron un libro, y luego la niña la pellizcó.

A

‘First the woman and the girl read a book, and then the girl pinched her.’ Primero el hombre y el niño saludaron, y luego el niño se rascó.

A

‘First the man and the boy greeted, and then the boy scratched himself.’ Primero el hombre y el niño anduvieron en bici, y luego el niño se untó con crema.

‘First the man and the boy rode a bike, and then the boy smeared cream on himself.’ ECM Primero la mujer y la niña aplaudieron y luego la niña la vio fregar el suelo. ‘First the woman and the girl applauded, and then the girl saw her scour the Àoor.’

264

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

Part 2 # 32

Type Item B Primero la mujer y la niña cantaron, y luego la niña la pintó.

33

A

34

‘First the man and the boy cried, and then the man embraced himself.’ ECM Primero el hombre y el niño bebieron algo, y luego el niño lo vio andar en bicicleta.

‘First the woman and the girl sang, and then the girl painted her.’ Primero el hombre y el niño lloraron, y luego el hombre se abrazó.

‘First the man and the boy drank something, and then the boy saw him ride a bike.’ Primero el hombre y el niño anduvieron en bici, y luego el niño lo untó con crema.

37

B

38

‘First the man and the boy rode a bike, and then the boy smeared cream on him.’ ECM Primero el hombre y el niño comieron algo, y luego el niño lo vio llorar.

41

A

45

‘First the man and the boy crawled, and then the man dressed himself.’ ECM Primero la mujer y la niña bailaron, y luego la mujer la vio reir.

‘First the man and the boy ate something, and then the boy saw him cry.’ Primero el hombre y el niño anduvieron a gatas, y luego el hombre se vistió.

‘First the woman and the girl danced, and then the woman saw her laugh.’ Primero el hombre y el niño anduvieron a gatas, y luego el hombre lo vistió.

47

B

50

A

55

‘First the mother and the girl read a book, and then the girl pintched her.’ ECM Primero la mujer y la niña cantaron, y luego la niña la vio leer un libro.

58

B

‘First the mother and the girl sang, and then the girl saw her read a book.’ Primero el hombre y el niño lloraron, y luego el hombre lo abrazó.

62

A

‘First the man and the boy cried, and then the man embraced him.’ Primero la mujer y la niña caminaron, y luego la mujer se lavó.

‘First the man and the boy crawled, and then the man dressed him.’ Primero la mujer y la niña leyeron un libro, y luego la niña se pellizcó.

‘First the woman and the girl walked, and then the woman washed herself.’

Principle B delays as a processing problem: Evidence from task effects

265

Appendix B Part 1 # 1

type A

Item Primero la mujer y la niña aplaudieron, y luego la mujer se fotogra¿ó.

2

B

‘First the mother and the girl applauded, and then the mother photographed herself.’ Primero la mujer y la niña rieron, y luego la mujer la tocó.

5

‘First the mother and the girl laughed, and then the mother touched her.’ ECM Primero el hombre y el niño lloraron, y luego el niño lo vio beber algo. ‘First the man and the boy cried, and then the boy saw him drink something.’ Primero el hombre y el niño saludaron, y luego el niño lo rascó.

10

B

11

‘First the man and the boy greeted, and then the boy scratched him.’ ECM Primero la mujer y la niña fregaron el suelo, y luego la niña la vio cantar.

A

‘First the woman and the girl scoured the Àoor, and then the girl saw her sing.’ Primero la mujer y la niña cantaron, y luego la niña se pintó.

15

B

‘First the woman and the girl sang, and then the girl painted herself.’ Primero la mujer y la niña fregaron el suelo y luego la niña la mordió.

17

‘First the woman and the girl scoured the Àoor, and then the girl bit her.’ ECM Primero el hombre y el niño patinaron, y luego el hombre lo vio comer algo.

14

‘First the man and the boy skated, and then the man saw him eat something.’ Primero el hombre y el niño patinaron, y luego el hombre lo ¿lmó.

19

B

20

‘First the man and the boy skated, and then the man ¿lmed him.’ ECM Primero la mujer y la niña rieron, y luego la mujer la vio bailar.

24

B

‘First the woman and the girl laughed, and then the woman saw her dance.’ Primero la mujer y la niña caminaron, y luego la mujer la lavó. ‘First the woman and the girl walked, and then the woman washed her.’

266 29

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin ECM Primero el hombre y el niño saludaron, y luego el hombre lo vio jugar al futbol.

31

B

‘First the man and the boy greeted, and then the man saw him play football.’ Primero la mujer y la niña leyeron un libro, y luego la niña la pellizcó.

32

A

‘First the woman and the girl read a book, and then the girl pintched her.’ Primero el hombre y el niño bebieron algo, y luego el niño se acarició.

34

‘First the man and the boy drank something, and then the boy petted himself.’ ECM Primero la mujer y la niña aplaudieron y luego la niña la vio fregar el suelo.

A

‘First the woman and the girl applauded, and then the girl saw her scour the Àoor.’ Primero la mujer y la niña rieron, y luego la mujer se tocó.

B

‘First the woman and the girl laughed, and then the woman touched herself.’ Primero la mujer y la niña aplaudieron, y luego la mujer la fotogra¿ó.

45

A

‘First the woman and the girl applauded, and then the woman photographed her.’ Primero el hombre y el niño saludaron, y luego el niño se rascó.

46

‘First the man and the boy greeted, and then the boy scratched himself.’ ECM Primero el hombre y el niño bebieron algo, y luego el niño lo vio andar en bicicleta.

38

40

48

A

‘First the man and the boy drank something, and then the boy saw him ride a bike.’ Primero la mujer y la niña fregaron el suelo, y luego la niña se mordió. ‘First the woman and the girl scoured the Àoor, and then the girl bit herself.’

Principle B delays as a processing problem: Evidence from task effects

267

Part 2 51

53

ECM Primero el hombre y el niño jugaron al futbol, y luego el hombre lo vio andar a gatas.

A

‘First the man and the girl played football, and then the man saw him crawl.’ Primero el hombre y el niño anduvieron en bici, y luego el niño se untó con crema.

B

‘First the man and the boy rode a bike, and then the boy smeared himself with cream.’ Primero el hombre y el niño bebieron algo, y luego el niño lo acarició.

58

A

‘First the man and the boy drank something, and then the boy petted him.’ Primero el hombre y el niño lloraron, y luego el hombre se abrazó.

64

‘First the man and the boy drank something, and then the man embraced himself.’ ECM Primero el hombre y el niño comieron algo, y luego el niño lo vio llorar.

69

B

‘First the man and the boy ate something, and then the boy saw him cry.’ Primero la mujer y la niña cantaron, y luego la niña la pintó.

70

A

72

‘First the man and the boy crawled, and then the man dressed himself.’ ECM Primero la mujer y la niña bailaron, y luego la mujer la vio reir.

75

‘First the woman and the girl danced, and then the woman saw her laugh.’ ECM Primero la mujer y la niña caminaron, y luego la mujer la vio aplaudir.

77

‘First the woman and the girl walked, and then the woman saw her applaud.’ Primero el hombre y el niño jugaron al futbol, y luego el hombre se grabó.

55

78

81

A

B

B

‘First the woman and the girl sang, and then the girl painted her.’ Primero el hombre y el niño anduvieron a gatas, y luego el hombre se vistió.

‘First the man and the boy played football, and then the man recorded himself.’ Primero el hombre y el niño anduvieron en bici, y luego el niño lo untó con crema. ‘First the man and the boy rode a bike, and then the boy smeared cream on him.’ Primero el hombre y el niño anduvieron a gatas, y luego el hombre lo vistió. ‘First the man and the boy crawled, and then the man dressed him.’

268 83 86

88

90

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin ECM Primero la mujer y la niña cantaron, y luego la niña la vio leer un libro. B

‘First the woman and the girl sang, and then the girl saw her read a book.’ Primero el hombre y el niño jugaron al futbol, y luego el hombre lo grabó.

A

‘First the man and the boy played football, and then the man recorded him.’ Primero la mujer y la niña leyeron un libro, y luego la niña se pellizcó.

‘First the woman and the girl read a book, and then the girl pintched herself.’ ECM Primero el hombre y el niño anduvieron a gatas, y luego el hombre lo vio patinar. ‘First the man and the boy crawled, and then the man saw him skate.’ Primero la mujer y la niña caminaron, y luego la mujer se lavó.

92

A

96

‘First the woman and the girl walked, and then the woman washed herself.’ B-15 Primero el hombre y el niño lloraron, y luego el hombre lo abrazó.

99

A

‘First the man and the boy cried, and then the man embraced him.’ Primero el hombre y el niño patinaron, y luego el hombre se ¿lmó. ‘First the man and the boy skated, and then the man ¿lmed him.’

Appendix C Patient Sex

Age

MA

M

66

JO

M

46

JA

M

42

PA

M

68

JV

M

54

CA

M

41

JU

M

50

Months Lesion/localization Classi¿cation WAB/ post onset Barcelona test 54 CVA. Left frontotemporopariWAB: Broca etal (persylvian) hemorrhage 34 CVA. Left frontotemporopariWAB: Broca etal (persylvian) hemorrhage 15 CVA. Left frontotemporopaWAB: Broca rietal (persylvian) hemorrhage. 22 Left frontotemporoparietal WAB: Broca (persylvian) isquemic CVA. 13 Left frontal persylvian WAB: Broca isquemic CVA. 41 Arterotrombotic isquemic WAB: Broca CVA of the left middle cerebral artery. 15 Arterotrombotic persylvian Barcelona: Broca ischemic CVA.

Principle B delays as a processing problem: Evidence from task effects

Appendix D # 3

Condition* ECM-Y

Method TVJ

Test sentence Ziet TW hem eten?

4

B

PS

‘Does TW see him eat?’ In welk plaatje kietelt Lala haar?

5

ECM-N

TVJ

‘In which picture does Lala tickle her?’ Ziet TW hem touwtjespringen?

7

B-N

TVJ

‘Does TW see him jump-rope?’ Fotografeert Lala haar?

8

ECM

PS

‘Does Lala take a picture of her?’ In welk plaatje ziet Lala haar dansen?

9

B-N

TVJ

‘In which picture does Lala see her dance?’ Knuffelt TW hem?

10

ECM

PS

‘Does TW hug him?’ In welk plaatje ziet TW hem touwtjespringen?

12

B

PS

‘In which picture does TW see him jump-rope?’ In welk plaatje aait Po haar?

13

ECM-N

TVJ

‘In which picture is Po petting her?’ Ziet TW hem eten?

14

B

PS

‘Does TW see him eat?’ In welk plaatje raakt TW hem aan?

15

ECM-N

TVJ

‘In which picture does TW touch him?’ Ziet Lala haar dansen?

17

B-Y

TVJ

‘Does Lala see her dance?’ Knijpt Dipsy hem?

18

ECM

PS

‘Does Dipsy pinch him?’ In welk plaatje ziet Lala haar touwtjespringen?

19

B-Y

TVJ

‘In which picture does Lala see her jump-rope?’ Knuffelt TW hem?

20

ECM

PS

‘Does TW hug him?’ In welk plaatje ziet TW hem eten?

21

B-N

TVJ

‘In which picture does TW see him eat?’ Kietelt Lala haar?

22

B

PS

‘Does Lala tickle her?’ In welk plaatje fotografeert Lala haar? ‘In which picture does Lala take a picture of her?’

269

270

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

23

ECM-Y

TVJ

Ziet Lala haar touwtjespringen?

24

B-N

TVJ

‘Does Lala see her jump-rope?’ Aait Po haar?

25

ECM

PS

‘Does Po pet her?’ In welk plaatje ziet Dipsy hem dansen?

26

B-N

TVJ

‘In which picture does Dipsy see him dance?’ Raakt TW hem aan?

27

ECM-N

TVJ

‘Does TW touch him?’ Ziet Dipsy hem dansen?

28

B

PS

‘Does Dipsy see him dance?’ In welk plaatje knuffelt TW hem?

30

ECM-N

TVJ

‘In which picture does TW hug him?’ Ziet Po haar eten?

31

B

PS

‘Does Po see her eat?’ In welk plaatje knijpt Dipsy hem?

32

ECM-Y

TVJ

‘In which picture does Dipsy pinch her?’ Ziet Dipsy hem dansen?

33

B-Y

TVJ

‘Does Dipsy see him dance?’ Fotografeert Lala haar?

34

ECM

PS

‘Does Lala take a picture of her?’ In welk plaatje ziet Po haar eten?

35

ECM-N

TVJ

‘In which picture does Po see her eat?’ Ziet Lala haar touwtjespringen?

36

B-N

TVJ

‘Does Lala see her jump-rope?’ Knijpt Dipsy hem? ‘Does Dipsy pinch him?’

* TVJ conditions marked with Y are “Yes” conditions, those marked with N are “No” Conditions

References Avrutin, Sergey 1999 Development of the Syntax-Discourse Interface, Dordrecht, The Netherlands: Kluwer Academic. Avrutin, Sergey 2004 Optionality in Child and Aphasic Speech. Lingue e Linguaggio 1 (1): 67 – 93.

Principle B delays as a processing problem: Evidence from task effects

271

Avrutin, Sergey 2006 Weak Syntax. In Broca’s region, Yosef Grodzinsky and Karin Amunts (eds.), 49 – 62. Oxford: Oxford University Press. Avrutin, Sergey, Stuart Lubarsky, and Jennifer Greene 1999 Comprehension of Contrastive Stress by Agrammatic Broca’s Aphasics. Brain and Language 70: 163 – 186. Baauw, Sergio 2000 Grammatical Features and the Acquisition of Reference. A Comparative Study of Dutch and Spanish. Ph.D. diss., Utrecht: LOT 39. Baauw, Sergio, and Maira Alija. 2005 La Interpretación de los Pronombres Masculinos y Femeninos por Niños Neerlandeses y Españoles [The Interpretation of Masculine and Feminine Pronouns by Dutch and Spanish Children]. Cognitiva 3 – 28. Baauw, Sergio, and Fernando Cuetos 2003 The Interpretation of Pronouns in Spanish Language Acquisition and Breakdown: Evidence for the “Principle B Delay” as a Non-Unitary Phenomenon. Language Acquisition 11 (4): 219 – 275. Baauw, Sergio, Linda Escobar, and William Philip 1997 A Delay of Principle B Effect in Spanish Speaking Children: The Role of Lexical Feature Acquisition. In Proceedings of the GALA ’97 Conference on Language Acquisition, Antonella Sorace, Caroline Heycock, and Richard Shillcock (eds.), Edinburgh, Scotland: Human Communication Research Centre. Baauw, Sergio, Marieke Kuipers, Esther Ruigendijk, and Fernando Cuetos 2006 The Production of SE and SELF Anaphors in Spanish and Dutch Children. In The Acquisition of Syntax in Romance Languages, Vicenç Torrens and Linda Escobar (eds.), 3 – 21. Language Acquisition and Language Disorders 41, Amsterdam/Philadelphia: John Benjamins. Chien, Yu-Chin, and Kenneth 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, Noam 1981 Lectures on government and binding. Foris Publications: Dordrecht. Conroy, Anastasia, Eri Takahashi, Jeffrey Lidz, and Colin Phillips 2009 Equal Treatment for all Antecedents: How Children Succeed with Principle B. Linguistic Inquiry 40: 446 – 486. Coopmans, Peter 2000 Het Projectieprobleem in de Praktijk [The Projection Problem in Practice]. Utrecht University, Utrecht, The Netherlands. Crain, Stephen, and Rosalind Thornton 1998 Investigations in Universal Grammar: A Guide to Experiments on the Acquisition of Syntax and Semantics. Cambridge, Mass.: MIT Press.

272

Sergio Baauw, Shalom Zuckerman, Esther Ruigendijk, and Sergey Avrutin

Grodzinsky, Yosef, and Tanya Reinhart 1993 The Innateness of Binding and Coreference. Linguistic Inquiry 24: 69 – 102. Grodzinsky, Yosef, Kenneth Wexler, Yu-Chin Chien, Susan Marakovitz, and Julie Solomon 1993 The breakdown of binding relations. Brain and Language 45: 396 – 422. Kertesz, Andrew 1982 Western Aphasia Battery. New York: Harcourt Brace Jovanovich. Kolk, Herman 1987 A theory of grammatical impairment in aphasia. In Natural Language Generation, Gerard Kempen (ed.), Dordrecht: Kluwer. Peña Casanova, J. 1991 Normalidad, Semiología y Patología Neuropsicológicas. Programa Integrado de Exploración Neuropsicológica. Test Barcelona, Barcelona: Masson. Reinhart, Tanya, and Eric Reuland 1993 ReÀexivity. Linguistic Inquiry 24: 657 – 720 Reuland, Eric 2001 Primitives of Binding. Linguistic Inquiry 32 (3): 439 – 492. Ruigendijk, Esther, Nada Vasiü, and Sergey Avrutin 2006 Reference Assignment: Using Language Breakdown to Choose Between Theoretical Approaches. Brain and Language 96 (3): 302 – 317. Thornton, Rosalind, and Kenneth Wexler 1999 Principle B, VP Ellipsis and Interpretation in Child Grammar. Cambridge, Mass.: MIT Press. Vasiü, Nada 2006 Pronoun Comprehension in Agrammatic Aphasia. Ph.D. diss., Utrecht University. LOT 140. Utrecht: LOT.

Subject index

agreement, 134, 137–139 animacy, 71–72, 145–151, 158, 164–167 area of interest (AoI), 155, 207–208 Aspect Hypothesis, 17 aspect grammatical, 21–22 lexical, 21 Asymmetric Grammar Hypothesis, 9, 100, 104–106 bidirectional optimization, 101–106, 117–120 Centering Theory, 70, 197 clitic accusative, 39–45, 50, 61–63 omission, 53–56, 59–61 reÀexive, 39, 52, 61–63 Romanian, 42–48, 50, 61–63 syntactic, 252 comprehension delay, 72, 99–100 Delay of Principle B-Effect, 193–200, 212, 247 discourse coherence, 193–194, 197, 212, 214 structure, 103, 117 topic, 103–113, 117–118, 193, 197–200, 215 eye movements, 207, 212–215 eye-tracking, 181, 184–185, 188 externalization, 127–128, 134, 140 ¿rst mention hypothesis, 70 garden path effect, 123, 130 imperfective, 22 intervention, 50–52, 61–63 listener’s perspective, 104–105, 107, 117 negation verb adjacent, 220, 230, 234, 238, 239

verb non-adjacent, 220, 230, 234, 238, 239 false, 222, 231, 236, 238, 240 true, 222–223, 239–241 observation length, 207–208 object relatives Hebrew, 124–126, 134–136 Italian, 125–126, 137–139 Optimality Theory, 3, 7, 72, 101–102, 104, 148–151, 164–166, 194, 197–198, 212 Parallelism, 70 perfective, 22 post-verbal subject, 135, 139 PRECEDENCE, 148–151 Principle A, 102, 104, 195 Principle B, 102–103, 195–197, 199, 214, 252–153 processing options, 130, 134, 140–141 production delay, 100, 120, 148 PROMINENCE, 148–151, 166 pronoun ambiguous, 69, 103, 105, 118, 199 object, 193–194, 213–215, 247, 250 person, 39–40, 69–71 resumptive, 128–129, 136–137 subject, 72, 100, 103–106, 117–119 unrecoverable, 104–105, 119 REFERENTIAL ECONOMY, 102, 104–105 reÀexive, 194–195, 197–198, 212–215 sentence negation, 217–223, 239–241 Subject Preference Hypothesis, 69–70 structural prominence, 73–74, 87, 90–91 subject pro-drop, 74, 87, 89–91 tense marker, 18–19, 22–25, 35–37, 182, 185 topic shift, 100, 117–120

274

Subject index

truth value judgment task, 195, 213, 214, 228–229, 240, 250, 255, 260–161 unidirectional optimization, 101–102, 105, 119 word order, 145–147, 149–151, 164–167

working memory, 105–106, 110, 118–119 production-comprehension asymmetries, 1–4, 171–174, picture selection task, 182, 184, 255, 261