Crosslinguistic Perspectives on Argument Structure: Implications for Learnability

Citation preview

CROSSLINGUISTIC PERSPECTIVES ON ARGUMENT STRUCTURE

CROSSLINGUISTIC PERSPECTIVES ON ARGUMENT STRUCTURE

MELISSA BOWERMAN PENELOPE BROWN

Cover design by Tomai Maridou

Lawrence Erlbaum Associates Taylor & Francis Group 270 Madison Avenue New York, NY 10016

Lawrence Erlbaum Associates Taylor & Francis Group 2 Park Square Milton Park, Abingdon Oxon OX14 4RN

© 2008 by Taylor & Francis Group, LLC Lawrence Erlbaum Associates is an imprint of Taylor & Francis Group, an Informa business Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number-13: 978-0-8058-4194-7 (Hardcover) Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging-in-Publication Data Bowerman, Melissa. Crosslinguistic perspectives on argument structure : implications for learnability / edited by Melissa Bowerman and Penelope Brown. p. cm. Includes bibliographical references and index. ISBN 978-8058-4194-7—0-8058-4194-6 (cloth : alk. paper) ISBN 978-1-4106-1645-6—1-4106-1645-2 (e book) 1. Language acquisition. 2. Grammar, Comparative and general—Syntax. 3. Grammar, Comparative and general—Verb. 4. Semantics. I. Bowerman, Melissa and Brown, Penelope. II. Title. P118.B6535 2007 415—dc22 Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the LEA and Routledge Web site at http://www.routledge.com

2006025579

Contents

Preface

1

I.

Introduction Melissa Bowerman and Penelope Brown

vii 1

VERB MEANING AND VERB SYNTAX: CROSSLINGUISTIC PUZZLES FOR LANGUAGE LEARNERS

2

A Person, a Place, or a Thing? Whorfian Consequences of Syntactic Bootstrapping in Mopan Maya Eve Danziger

29

3

The Pitfalls of Getting from Here to There: Bootstrapping the Syntax and Semantics of Motion Event Coding in Yukatek Maya Jürgen Bohnemeyer

49

4

Making Sense of Complex Verbs: On the Semantics and Argument Structure of Closed-Class Verbs and Coverbs in Jaminjung Eva Schultze-Berndt

69

5

Figure–Ground Indeterminacy in Descriptions of Spatial Relations: A Construction Grammar Account Sotaro Kita

89

6

Learning Verbs without Boots and Straps? The Problem of ‘Give’ in Saliba Anna Margetts

111

v

vi

II.

CONTENTS

PARTICIPANTS PRESENT AND ABSENT: ARGUMENT ELLIPSIS AND VERB LEARNING

7

Same Argument Structure, Different Meanings: Learning ‘Put’ and ‘Look’ in Arrernte David Wilkins

141

8

Verb Specificity and Argument Realization in Tzeltal Child Language Penelope Brown

167

9

Interacting Pragmatic Influences on Children’s Argument Realization Shanley E. M. Allen

191

III. TRANSITIVITY, INTRANSITIVITY, AND THEIR ASSOCIATED MEANINGS: A COMPLEX WORK-SPACE FOR LEARNABILITY

10

Intransitive Verbs in Ewe and the Unaccusativity Hypothesis James Essegbey

213

11

He died old dying to be dead right: Transitivity and Semantic Shifts of ‘Die’ in Ewe in Crosslinguistic Perspective Felix K. Ameka

231

12

Acquiring Telicity Crosslinguistically: On the Acquisition of Telicity Entailments Associated with Transitivity Angeliek van Hout

255

13

The Acquisition of the English Causative Alternation Melissa Bowerman and William Croft

279

14

What Adverbs Have to Do with Learning the Meaning of Verbs Angelika Wittek

309

15

Event Realization in Tamil Eric Pederson

331

Author Index

357

Language Index

363

Subject Index

365

Preface

This volume came about as the result of an unusual opportunity: the presence at the Max Planck Institute for Psycholinguistics of two kinds of researchers who normally do not tend to cross each others’ paths, namely, specialists in first language development and field linguists working on a wide variety of lesser-known languages. These researchers formed an interdisciplinary collaborative project on argument structure and its acquisition, with project members bringing together their expertise to jointly explore what (if any) aspects of argument structure are universal, and to rethink the possible role of argument structure in children’s lexical and syntactic development. From looking at adult and/or child data in a variety of languages, some of them poorly known to Western linguists, it has become clear that there is much more diversity in this domain than specialists in language acquisition have realized. The findings create challenges for influential “bootstrapping” theories of language acquisition, and they highlight new acquisition puzzles for which existing theories provide no solution. Comparative linguists will also find some surprises here, argument structure patterns that provide new grist for the typologists’ mill. The argument structure project culminated in a workshop in 1998, at which papers were presented and five invited discussants provided commentary and links to broader theoretical issues. Each paper presented an argument-structure-related problem in a particular language, drawing on original fieldwork and on either adult data or child data, or both, from a total of 13 languages from nine different language families (three Mayan, two Australian, one West African, three Germanic, one Oceanic, one Dravidian, and Inuktitut and Japanese). The five discussants—Eve Clark, Cynthia Fisher, Adele Goldberg, Ken Hale, and Beth Levin—provided detailed commentary on the papers presented. We are extremely grateful to these five discussants, who provided a broader crosslinguistic perspective and saved us from innumerable errors and infelicities. Unfortunately, space has precluded the inclusion of additional chapters by our discussants in this volume, but all the papers vii

viii

PREFACE

presented at the conference are included, after having been rewritten in light of the commentaries and discussions so that the perspectives of our commentators are (hopefully) reflected. Themes shared across chapters are highlighted in the introduction as well as through cross-referencing in individual chapters. The resulting volume is an integrated interdisciplinary attack on the relationship between language structure and language learning.

CHAPTER 1

Introduction Melissa Bowerman and Penelope Brown Max Planck Institute for Psycholinguistics

Verbs are the glue that holds clauses together. As elements that encode events, verbs are associated with a core set of semantic participants that take part in the event. Some of a verb’s semantic participants, although not necessarily all, are mapped to roles that are syntactically relevant in the clause, such as subject or direct object; these are the arguments of the verb. For example, in John kicked the ball, ‘John’and ‘the ball’are semantic participants of the verb kick, and they are also its core syntactic arguments—the subject and the direct object, respectively. Another semantic participant, ‘foot’, is also understood, but it is not an argument; rather, it is incorporated directly into the meaning of the verb. The array of participants associated with verbs and other predicates, and how these participants are mapped to syntax, are the focus of the study of ARGUMENT STRUCTURE. At one time, argument structure meant little more than the number of arguments appearing with a verb, for example, one for an intransitive verb, two for a transitive verb. But argument structure has by now taken on a central theoretical position in the study of both language structure and language development. In linguistics, argument structure is seen as a critical interface between the lexical semantic properties of verbs and the morphosyntactic properties of the clauses in which they appear (e.g., Grimshaw, 1990; Goldberg, 1995; Hale & Keyser, 1993; Levin & Rappaport Hovav, 1995; Jackendoff, 1990). Increasingly, this interface is understood as being mediated by a rich representation of event structure based on causal dynamics, the internal temporal structure of events (aspect), or both (e.g., Croft, 1991, 1998, in press; Dowty, 1979; Erteschik-Shir & Rapoport, 2005; Pustejovky, 1991; Tenny, 1994; Tenny & Pustejovsky, 2000; Van Valin & LaPolla, 1997). Although theorists differ in how they represent argument-structure-related properties of language, they often agree on the view that, across languages, there are strong consistencies in the number of arguments associated with verbs with certain kinds of meanings, and in the typical mapping of these arguments to syntactic roles (e.g., Keenan, 1976; Perlmutter & Rosen, 1984; see Pinker, 1989: 94–95 for 1

2

BOWERMAN AND BROWN

discussion). Strong similarities across languages—universals or near-universals—demand explanation, and it has been a familiar step, since Chomsky’s work of the 1960s and 1970s, to locate this explanation in the child’s innate capacity for language acquisition. Universals, according to this way of thinking, reflect children’s a priori expectations about the structure of language. Children come to the acquisition task with inborn knowledge of those abstract aspects of grammar that are universal, and this knowledge enables them to home in quickly on how these universals are instantiated in their particular language. Individual languages, for their part, are constrained to conform to the universal architecture, because if they did not, children would be unable to learn them. Given this hypothesized link between linguistic universals and the capacity for language acquisition, proposals about universals of argument structure have caught the attention of language acquisition researchers. If children have innate expectations about argument structure—in particular, about the typical correspondences between the semantic and syntactic roles associated with verbs—they could draw on this information to solve a number of learning problems. This thought lies at the heart of several influential hypotheses, to be reviewed shortly, about how children set up their initial phrase structures, acquire the meanings of novel verbs, and figure out which verbs can occur in which syntactic frames. Despite the emphasis on universals, most of the research on the proposed role of innate argument structure knowledge in language acquisition has revolved around English and closely related languages. It is not yet clear whether children could, in fact, use the hypothesized universals of argument structure to acquire the structures that confront them in a broad range of typologically distinct languages. The present volume attempts to address this concern. The research it presents came about as the result of a unique situation: a cooperation, within the framework of the Argument Structure Project at the Max Planck Institute for Psycholinguistics, between crosslinguistically minded language acquisition specialists and field linguists working on a diverse set of lesser-known languages. Working together over a period of several years, our group of linguists, psychologists, and anthropologists explored claims about argument structure universals, and used findings to jointly analyze and rethink the possible role of argument structure in children’s language development. The work often challenges current proposals, especially in showing that there is more crosslinguistic diversity in the domain of argument structure than most specialists in language acquisition have realized. But it also suggests new directions for finding solutions, and it calls attention to argument structure acquisition puzzles that have so far been neglected. In this chapter, we introduce some of the proposals, controversies, and problems that have inspired and motivated the authors of this volume, and we summarize the organization and contents of the book. Section 1 here overviews the so-called “bootstrapping” hypotheses, according to which innate knowledge of argument structure plays a critical role in language acquisition. Section 2 further sets the stage by reviewing some current major controversies surrounding these hypotheses. Finally, section 3 lays out the plan of the book, and highlights the key findings of the specific chapters.

1.

INTRODUCTION

3

1. THE PROPOSED ROLE OF ARGUMENT STRUCTURE UNIVERSALS IN LANGUAGE ACQUISITION What are the language acquisition challenges that children might solve by drawing on inborn knowledge of argument structure? There have been two broad lines of theorizing. Both presuppose that children come to the language acquisition task with expectations about the linking between the semantic and syntactic roles associated with verbs, but they differ in their assumptions about the typical direction of learners’inferences. One scenario goes from meaning to syntax. In this approach, children are hypothesized to use their nonlinguistic cognitive understanding of the world to determine the structure of everyday events like ‘running’, ‘hitting’, and ‘giving’, including how many participants there are in the event and what their semantic function, or thematic role, is (e.g., agent, patient, recipient). Children then draw on their innate knowledge to predict how noun phrases representing these participants should be mapped to syntactic roles. The second scenario reverses the direction of inferencing, going from syntax to meaning rather than meaning to syntax. In particular, this proposal is about how children learn the meaning of verbs. On encountering a new verb, children are hypothesized to use the syntactic frame(s) in which it appears to predict some basic aspects of its meaning. Proposals of the first type are lumped loosely under the rubric “semantic bootstrapping,” whereas those of the second type are termed “syntactic bootstrapping.” 1.1. Semantic Bootstrapping: Using Meaning to Predict Syntax Cracking into Grammar. Following Chomsky’s claim that universals of language are innate, many child language researchers have assumed that children have inborn knowledge of putatively universal syntactic categories and relationships such as “noun”, “verb”, “subject”, and “direct object”, and—importantly—of the abstract syntactic properties associated with these constructs. But this would buy children nothing unless they had some way to identify concrete instances of these constructs in the speech around them. To explain how this identification takes place, Pinker (1984, building on Grimshaw, 1981) proposed that children’s inborn linguistic toolkit includes not only information about syntactic categories and relationships, but also some cognitively simple “semantic flags” by which these elements can be recognized. He termed this use of semantics to make good guesses about syntax “semantic bootstrapping.” According to this hypothesis, for example, children will initially assume that if a word names a concrete object it is a noun, and if it names an action it is a verb. Of course, not all nouns pick out objects and not all verbs name actions. But this procedure could help children identify a starter set of nouns and verbs correctly; from here they can go on to notice the morphosyntactic elements associated with these forms, and so to identify instances of the categories that lack the default semantics.

4

BOWERMAN AND BROWN

For instance, learners of English will notice that words that name objects—hence, nouns—typically occur in contexts like a/the/another/my___, or two___ and ___-s, and words that name actions—hence, verbs—occur in contexts like ___-ed and is / are____-ing. This morphosyntactic knowledge will enable children to identify bath as a noun and stay as a verb even though these words do not name an object or a dynamic action, respectively. Children have to figure out not only the part-of-speech membership of the words in the sentences they hear, but also their syntactic function. According to the semantic bootstrapping hypothesis, “semantic flags” help with this problem, too (Pinker, 1984). For example, suppose that the syntactic constructs “subject” and “direct object”—by hypothesis inborn—are initially cued by the relational roles “agent” and “patient”. When a learner of English sees a dog biting a cat, her nonlinguistic understanding of the event tells who does what to whom. If she now hears the sentence The dog is biting the cat, she can infer that the dog (naming the agent) is the subject and the cat (naming the patient) is the direct object. Once she has identified a few subjects and direct objects like this, she will discover how these constituents are typically ordered with respect to the verb, whether and how they are case-marked, and so on. And this knowledge will now allow her to identify other constituents as subjects and direct objects even when they lack the default agent–patient semantics, as in Mary heard a noise or The costume frightened the dog. With semantic bootstrapping, then, the child gets a toe in the door of grammar by an initial simple mapping between words and basic semantic types. Constraining Argument Structure Overgeneralizations. The basic logic of semantic bootstrapping has been applied not only to the initial stages of grammar construction, but also—in a more elaborate form—to a thorny problem arising later in the course of language development: how children avoid overgeneralizations of argument structure alternations. Many verbs of English and other languages appear in more than one syntactic frame, and whole groups of semantically related verbs often show similar patterns of frame alternations. Children become sensitive to these alternation patterns in the course of language acquisition, and sometimes apply them too liberally. Two kinds of alternations, and a few errors based on them, are shown in (1)–(2) (from Bowerman, 1982a, 1982b; see also Pinker, 1989). 1. Causative Alternation (cf. The stick broke/Harry broke the stick) a. You staggered me. (After mother pulls on child’s arm when child stumbles. 3;10 [age in years; months]) b. I saw a witch and she disappeared them. (Pretending a witch has made some blankets disappear. 4;8) 2. Locative Alternation (cf. Mary sprayed paint on the wall/Mary sprayed the wall with paint)

1.

INTRODUCTION

5

a. Can I fill some salt into the bear? (= fill the bear [a bear-shaped salt shaker] with salt. 5;0) b. Pour, pour, pour. Mommy, I poured you (waving empty container near M. M: You poured me?) Yeah, with water. (= I poured water on you. 2;11) It is widely acknowledged that children are rarely corrected for such errors. Why then do they stop making them—how do they end up with adult-like intuitions about which verbs do and don’t undergo a certain alternation? This question has given rise to much debate (e.g., Bowerman, 1988; Braine & Brooks, 1995; Pinker, 1984, 1989; Randall, 1990). According to one line of reasoning, the adult state is reached by gradual learning. Children discover an alternation pattern by a process of abstraction and schema formation after exposure to a sufficient number of exemplars of it, and they overgeneralize it because they do not yet know the relevant semantic, morphophonological, or idiosyncratic constraints. With increasing linguistic experience, they fine-tune the pattern to reflect these constraints, and errors fade out (Bowerman, 1982a, 1988; Brooks & Tomasello, 1999; Goldberg, 1995; Goldberg, Casenhiser, & Sethuraman, 2004). According to a very different approach, children get a tremendous boost toward the adult system through aspects of their innate knowledge (see Pinker, 1989, for the theory, and Gropen, Pinker, Hollander, & Goldberg, 1991a, 1991b, for some applications). In this approach Pinker updates his (1984) view of how verb meaning is related to argument structure. Rather than relying on simple semantics–syntax correspondences like “if agent, then subject” (as described under Cracking into Grammar, earlier), which requires reference to a fixed list of atomic thematic roles (e.g., agent, theme, location, source, goal…), with each noun phrase in a clause assigned to just one role, he adopts the decompositional approach to verb meaning found in, for example, Jackendoff (1987, 1990) and Levin and Rappaport Hovav (1995). Here, thematic roles are positions in a semantic representation of verb meaning that is structured around a set of primitive meaning elements such as “cause,” “go,” and “be,” with each thematic role triggering its own linking rule, for example, “first argument of ‘cause’ is subject,” “second argument of ‘cause’ [the ‘affected object’] is direct object.” A particular argument can bear more than one thematic role because it can participate in more than one semantic substructure in the verb’s semantic representation; for example, it can be both the second argument of “cause” and the first argument of “go.” As an example, consider the English verbs fill and pour. Both verbs specify the causation of a change of some sort, but they differ in which argument is specified to undergo the change (i.e., is the “affected object”): pour means roughly “cause X (a liquid or particulate mass) to go downward in a stream,” whereas fill means something like “cause Y (a container) to go to a state of being full by means of causing X to go into Y.” By virtue of this meaning difference, the verbs take different direct objects: X for pour, but Y for fill.

6

BOWERMAN AND BROWN

In this account, the syntactic treatment of a verb’s arguments is a direct projection of the verb’s meaning. Children are hypothesized to have foreknowledge of both predicate primitives like “cause” and “go” and of the specific linking rules associated with them, and to use these in building semantic representations for verbs. This means that if learners represent a verb’s meaning correctly, they will automatically be able to link its arguments correctly. But sometimes they misunderstand the meaning of a verb. For instance, if they think that fill specifies the caused movement of something (i.e., means something like ‘pour’), then they will link the moving entity to object position, making errors like (2a); and if they think that pour is about the caused state-change of the poured-upon entity (e.g., it goes to a state of being filled or covered), they will make errors like (2b). Once children have worked out the verb’s meaning through repeated observation of the verb’s use in context, errors 1 will cease (Gropen et al., 1991a, 1991b; Pinker, 1989). Pinker did not christen this process with a convenient name, but Gleitman (1990) subsumed it (despite Pinker’s 1994 objections) under an (extended) notion of “semantic bootstrapping,” because in this procedure, as in the original semantic bootstrapping procedure (see Cracking into Grammar, earlier) the child uses meaning to make predictions about form. 1.2. Syntactic Bootstrapping: Using Syntax to Predict Verb Meaning In the approaches just considered, learners acquire the meanings of verbs simply by observing the kinds of events they are paired with. Critics have argued that this procedure is grossly inadequate for verb learning (Gleitman, 1990; Gillette, Gleitman, Gleitman, & Lederer, 1999; Landau & Gleitman, 1985). For example, many verbs encode situations that are not directly observable at all (think, know). And many real-world scenes can be described by different verbs depending only on the perspective the speaker is adopting (The cat CHASED the mouse/The mouse FLED from the cat; The peddler SOLD John a pot/John BOUGHT a pot from the peddler. Even if the child has all the relevant concepts ahead of time, like ‘sell’ and ‘buy’, how does she know which is the one intended? To solve this problem, Gleitman and her colleagues in a sense turn Pinker’s semantic bootstrapping argument on its head. They agree with Pinker that children have foreknowledge of universal syntactic-semantic correspondences, but they propose that, rather than using meaning to predict syntax, children use syntax to narrow in on the general kind of meaning a new verb is likely to have. In particular, syntactic bootstrapping theorists suggest that children can use the number of arguments a verb appears with, and the syntactic arrangement of these arguments, to make a good ballpark guess about the verb’s meaning. For example, a novel verb used in a frame like “X ___ Y prep. Z” (e.g., John GORPED the ball onto the table) is likely to pick out an action of causing something to go somewhere in a certain manner (e.g., ‘throw’). In contrast, a novel verb with a

1.

INTRODUCTION

7

complementizer (e.g., John GORPED that the ball was on the table) is likely to specify an event of cognition or perception (e.g., ‘think’ or ‘see’) (Gleitman, 1990). Once children have been able to narrow down the range of possible meanings of the new verb, they can observe its contexts of use to get more detailed information about what kind of motion, perception, etc., is at issue. Gleitman termed this process “syntactic bootstrapping,” to highlight that the starting point for the child is not meaning, as in semantic bootstrapping, but rather syntax. Observation of a verb in a single frame is often compatible with more than one kind of meaning; for example, the transitive frame NP1___NP2 is found with verbs from a variety of semantic classes, such as caused state change (The child broke the vase), surface contact (Mary wiped the table), perception (I hear a strange noise), and cognition (I know that man). But proponents of syntactic bootstrapping have argued that the ambiguity associated with a single frame can be drastically reduced by taking into account a fuller range of the frames a verb appears in (e.g., Fisher, Hall, Rakowitz, & Gleitman, 1994; Gleitman, 1990; Naigles, 1996; Naigles & Hoff-Ginsberg, 1995, 1998). For example, some verbs that might be encountered in a transitive frame will also turn up in an intransitive frame (The child broke the vase/The vase broke), whereas others will not (Mary wiped the table/*The table wiped), or, if they do, will arrange their arguments differently (John read the newspaper/*The newspaper read/John read). And these differences will correspond systematically to differences in meaning (see Levin, 1993; Levin & Rappaport Hovav, 1995, on how semantic classes of English verbs are distinguished by the argument structure alternations they can appear in). Children can, then, by hypothesis, use information from multiple frames to home in on what kind of meaning a novel verb might have. 2. SOME CONTROVERSIES AND QUESTIONS ABOUT ARGUMENT STRUCTURE AND LEARNABILITY The two kinds of bootstrapping proposals we have just reviewed deal insightfully with fundamental problems to be solved in accounting for language acquisition, and they have been highly influential in the language acquisition literature. But they have also been subject to important empirical and theoretical challenges. In this section we sketch some of the major controversies surrounding the bootstrapping hypotheses, and raise additional questions bearing on the role of argument structure in children’s language acquisition. 2.1. Do Children Behave in Accordance with the Proposed Innate Knowledge? One basic question for the bootstrapping theories is whether children’s use and comprehension of language accord with the predictions that flow from existing hypotheses about what is innate. This is not a primary topic of the present volume,

8

BOWERMAN AND BROWN

which focuses more on whether bootstrapping assumptions are tenable in the face of crosslinguistic variation in argument structure and argument realization. But it is worthwhile briefly reviewing the controversy. On the one hand, a number of experimental studies of novel verb learning, conducted mostly with learners of English, have suggested that young children do make systematically different predictions about a new word’s meaning depending on what syntactic frame or frames it has been presented in (see Fisher & Gleitman, 2002, and Naigles, 1998, for reviews of a number of studies, and Gertner, Fisher, & Eisengart, 2006, and Fernandes, Marcus, Di Nubila, & Vouloumanos, 2006, for more recent experimental evidence). Also supporting the idea that children have predispositions for argument mapping is the finding that when deaf children develop self-made “home-sign” systems, they are systematic in their handling of argument structure (Goldin-Meadow & Mylander, 1998). On the negative side of the ledger, however, several studies of spontaneous speech and of elicited production or comprehension have found no evidence for sensitivity to hypothesized constraints. In one test, Bowerman (1990) examined early word combinations in the spontaneous speech of children learning English. She reasoned that if learners are helped by innate linking rules, as proposed by Pinker (1984, 1989), they should start to combine arguments earlier and more accurately (in terms of word order) with verbs that link canonically (i.e., consistently within and across languages) than with verbs that link noncanonically (variably or counter to prevailing patterns). (Examples of verbs of the former type are prototypical agent–patient verbs like break and open; an example of the latter type is get, as in John got a present [from Mary], where the recipient rather than the agent links to subject position.) But there was no advantage for canonical verbs: as soon as the children began to combine verbs with subject or object arguments at all, they did so equally accurately for verbs of all semantic types. In a second explicit test of proposals for innate linking rules, Brinkmann (1993, 1997) investigated whether learners of German follow a rule invoked by Gropen et al. (1991a, 1991b) that links an “affected object” (second argument of the primitive predicate “cause”) to direct object position (see 1.1 earlier). In elicited production experiments with locative verbs like schmieren ‘smear’, rieseln ‘drizzle’, and werfen ‘throw’, she found no evidence for an inborn “ affected object” linking rule: the youngest children were not influenced at all by the relative “affectedness” of the goal entity in their choice between this argument and the theme (i.e., the moving entity) as the direct object, and even the older children were influenced only to a limited extent. The most sustained attack on the proposal that children are helped by innate knowledge of linking has come from Tomasello and his colleagues (Akhtar, 1999; Akhtar & Tomasello, 1997; Jaakkola & Akhtar, 2000; Tomasello, 1992, 2000a, 2000b; Lieven, Pine, & Baldwin, 1997). On the basis of both early spontaneous speech and experiments with novel verbs modeled in particular syntactic frames, they argue that children’s early grammars are organized entirely around individual verbs

1.

INTRODUCTION

9

and other predicates. Thus, 2-year-olds may know how to combine hit with nominals specifying the “hitter” and/or the “hittee,” but show no evidence for relational concepts like “agent” and “patient” that apply across a number of verbs, much less for 2 completely verb-general constructs like “subject” and “direct object.” In summary, evidence from various spontaneous speech studies and experiments does not yet converge on a consistent picture of whether children bring knowledge of argument linking with them to the language acquisition task, or instead learn linking patterns gradually over time on the basis of exposure to the target language. This state of affairs has led to spirited interchanges, as in Fisher’s (2002) rebuttal to Tomasello (2000a) and Tomasello and Abbot-Smith’s (2002) response. Resolving the controversy can be expected to keep language acquisition researchers busy for some time to come. 2.2. How Universal Are the Alignments between Semantics and Syntax? From a theoretical point of view, the viability of bootstrapping proposals rests ultimately on the accuracy of the claim that semantic-syntactic correspondences are 3 universal, at least in broad outline: only universal aspects of grammar could plausibly be inborn, and so could serve as unlearned jumping-off points from which children could “bootstrap” further knowledge. This is well recognized by the “bootstrapping” authors. For example, according to Gleitman (1990: 35), The first proviso to the semantic usefulness of syntactic analysis for learning purposes is that the semantic/syntactic relations have to be materially the same across languages. Otherwise, depending on the exposure language, different children would have to perform completely different syntactic analyses to derive aspects of the meaning. And that, surely, begs the question at issue.

Although syntactic bootstrapping theorists are aware of the potential problems presented by language-specific mappings (e.g., Fisher et al., 1994), they argue that there is enough crosslinguistic consistency to at least give children a start into the grammar of their language. But the evidence on universality is in fact still somewhat sketchy, and “problem cases” for bootstrapping proposals—that is, linking patterns that depart from those familiar to speakers of English and other European languages—have played relatively little role in the acquisition literature.4 In a move that accommodates some crosslinguistic variation in linking, Fisher has proposed that in the very early stages of acquisition, before children know yet how to identify a subject or direct object in their language, they engage in a precursor to full-fledged syntactic bootstrapping that she terms “structure mapping” (Fisher, 1996, 2000; Fisher et al., 1994). To carry out structure mapping, children do not need to know anything about syntactic roles or to recognize conceptual similarities between the agents of different actions or the patients undergoing various motions or state changes. It is enough for them to assume that there will be an analogical match between the number of noun phrases accompanying a newly

10

BOWERMAN AND BROWN

encountered verb and the number of participants in the event described by the verb: “Children can roughly interpret sentences by bringing the noun phrases of a sentence into one-to-one alignment with the arguments of a conceptual predicate derived from observation of events” (Fisher, 1996: 74). This proposal has the advantage that it imposes fewer requirements on how languages map between semantic and syntactic roles, and so can tolerate a certain amount of crosslinguistic variation. But it does still presuppose that languages will agree on the number of semantic participants there are in events of various types (e.g., one for ‘laughing’, two for ‘pushing’, three for ‘giving’), and that this number will be reflected in the number of noun phrases in sentences describing these events. One important goal of the contributors to this volume has been to expand the crosslinguistic database available to researchers who want to test and explore the explanatory potential of semantic or syntactic bootstrapping procedures. Throughout the volume, the chapters bring information about lesser-known languages to bear on bootstrapping issues, showing kinds of crosslinguistic variability that challenge current proposals and so may help researchers to build more comprehensive models in the future. 2.3. Does the Input Provide Enough Information for the Syntactic Bootstrapping Procedure to Work? In its classic form, syntactic bootstrapping assumes that children get crucial information about the meaning of a verb from the number of arguments it occurs with and the syntactic roles to which these are assigned. This hypothesis may seem plausible for a language like English, in which noun phrases representing a verb’s arguments are reliably present.5 But how could syntactic bootstrapping work for children learning one of the many languages of the world that allow extensive argument ellipsis—the omission of arguments whose referents can easily be recovered through previous discourse or nonlinguistic contextual information? This question was raised influentially by Rispoli (1995) in his work on Japanese. In analyzing child-directed speech from nine Japanese caregivers, Rispoli found that the vast majority of transitive sentences (90%) had no or only one overt argument, usually not case-marked; only 1% had two overt case-marked arguments. (Narasimhan, Budwig, and Murty [2005] find a similar dearth of arguments in child-directed Hindi.) If children often encounter verbs accompanied by none or only a subset of their arguments, how can they use syntactic frames to make good guesses about the verbs’ meanings? One approach to this problem is to note that even though in any one utterance a verb may appear with none or only one of its arguments, across discourse turns all of its arguments may eventually be displayed. For example, Clancy (1996) found that in a Korean adult–child interaction involving a board and plastic shapes that adhered to it, the verb pwuthita ‘stick, affix’ was, over time, combined by the adult

1.

INTRODUCTION

11

with all three of its arguments: for example, ‘shall auntie stick?’ (agent); ‘stick this’ (theme); ‘stick there’ (location). A second proposal invokes sources of information other than argument structure that a child might use in distinguishing between verbs with different kinds of meanings. For instance, Rispoli (1987, 1995) found that in Japanese input to children, the distinction between transitive and intransitive verbs is associated with a cluster of properties such as the animacy of the theme and patient referents and the speech act of the utterance in which the verb occurs. (See also Wittek, chapter 14 of this volume, for the suggestion that another source of evidence is the adverbials that co-occur with a verb.) More recent work has pinpointed two other possible sources of help for language learners faced with massive argument ellipsis. First, crosslinguistic studies show that even when children are very young, they are remarkably sensitive to the discourse factors that influence when arguments can be ellipsed (Allen, 2000 and this volume [Inuktitut child language]; Choi, 1998; Clancy, 2003 [Korean]; Narasimhan et al., 2005 [Hindi]; see also papers in Bavin, 2000). As the authors of these studies observe, this sensitivity to contexts of ellipsis might make it possible for children to infer the covert presence of ellipsed arguments in the adult input. Second, it has been pointed out that even though arguments may often be omitted in the input to children, verbs of different semantic classes are still probabilistically associated with different argument profiles; for example, transitive verbs are more likely than intransitive verbs to appear with two noun phrases (Lee & Naigles, 2005, working on Mandarin; Gleitman, Cassidy, Nappa, Papafragou, & Trueswell, 2005). Whether this kind of probabilistic information is sufficient is unclear, however: one of the authors of this introduction (Brown) notes that learners of her field language, Tzeltal Maya, might have to wait for years for any statistical evidence that ditransitive verbs have more arguments than transitive verbs. (See also Demuth, Machobane, and Moloi, 2000, on how learners of Sesotho could acquire restrictions on applicative arguments for which there is very little overt evidence in the input; also see Wilkins’s discussion of a paradox for probabilistic evidence in chapter 7.) 2.4. Multiple Frames and Multiple Verb Senses An intriguing but understudied set of problems in the acquisition of verb meaning and argument structure revolves around the issue of multiple subcategorization frames. Syntactic bootstrapping theorists have noted that hearing a verb in a single frame, for example, transitive, is often compatible with membership in a variety of semantic classes. But—as noted earlier—they argue that this ambiguity can be resolved if children base their guesses about a verb’s meaning not on its appearance in a single frame, but on the whole range of frames it occurs in. Although intuitively appealing, this proposal is fraught with problems. In particular, as Grimshaw (1994) and Pinker (1994) point out, different subcategorization frames are often associated with different senses of a verb. (Grimshaw gives the

12

BOWERMAN AND BROWN

example of two frames for English shoot: She shot the burglar [an NP complement] and The burglar shot out of the room [a PP complement]). Because the number and nature of senses that a verb takes on in a particular language is determined to a large extent by historical accidents, there is no predictable relationship between the meaning of the verb and its full set of subcategorization frames, only a relationship between a particular sense of the verb and its frames. At present, it is unclear how children decide how many senses a verb has and which sets of frames are associated with which senses (see Gropen, Epstein, & Schumacher, 1997, for one suggestion). The child’s problem of determining how many senses a verb has raises complex questions about the learning of argument structure alternations such as the locative or the dative. In these alternations, a verb such as spray or give is saliently associated with two frames, for example, John sprayed paint on the wall and John sprayed the wall with paint; Mary gave a book to Wilma and Mary gave Wilma a book. If children use multiple frames to triangulate in on a verb’s meaning, they should be looking for a meaning that is compatible with both frames. But this outcome is at odds with the hypothesis (Levin & Rappaport Hovav, 1995; Pinker, 1989) that the reason these verbs have two frames is that they in fact have two meanings, each one of which projects only one syntactic frame.6 To learn how the alternation works, and which verbs can undergo it, the child must distinguish the two meanings (Pinker, 1989; see Constraining Argument Structure Overgeneralizations in 1.1, earlier). 2.5. Argument Structure and Event Structure Bootstrapping claims have focused on aspects of verb meaning that are often reflected in relatively coarse aspects of syntax and verb argument structure: for example, the tendency for actions of agents on patients to be expressed with two-place verbs, and for actions of caused transfer to be encoded with three-place verbs. But argument structure has complex associations with other important aspects of verb meaning and clause interpretation, in particular, those implicated in studies of event structure, such as lexical aspect and causativity. Linguists are still heavily debating the proper handling of these notions in linguistic theory (e.g., see Croft, in press, and chapters in Erteschik-Shir & Rapoport, 2005, and Tenny & Pustejovsky, 2000). Although there have been many studies of the acquisition of lexical aspect (see Li & Shirai, 2000, and Shirai, Slobin, & Weist, 1998, for overviews), only a few have looked at the developmental relationship between aspect and argument structure, and these have tended to revolve around the aspectual concomitants of the distinction between two kinds of intransitive verbs, unaccusatives and unergatives (e.g., Randall, van Hout, Weissenborn, & Baayen, 2004; van Hout, 1996). Causativity has been more extensively investigated in connection with the acquisition of argument structure, but within the relatively narrow perspective of whether children associate transitivity with causativity (e.g., Lidz, Gleitman, & Gleitman, 2003; Naigles, Gleitman, & Gleitman, 1993); or causativity with transitivity (e.g., Bowerman, 1982a, 1982b, 1988; Brooks & Tomasello, 1999; Pinker, 1989).

1.

INTRODUCTION

13

Several chapters of this volume, especially 10, 11, 12, 14, and 15, break new ground in their efforts to understand the relationship, in both language and language acquisition, between argument structure and event structure. 3. PLAN OF THE BOOK This volume is divided into three sections. Section I focuses on verbs, examining, in particular, crosslinguistic variation in the relationship between verb meaning and verb syntax, and weighing the significance of this variation for the bootstrapping proposals discussed earlier. Section II investigates the role of arguments, especially the learning problems associated with massive argument ellipsis. In section III, verbs and arguments come together in an examination of a cluster of issues revolving around the construct of (in)transitivity and associated meanings such as causativity, control, and telicity; here the problem of multiple senses and multiple syntactic frames is also considered. Section I. Verb Meaning and Verb Syntax: Crosslinguistic Puzzles for Language Learners. The chapters in this section, all based on adult language data, show that there is more variation in the relationship between verb meaning and syntax than is often supposed. This creates problems for bootstrapping proposals as they now stand, and the authors consider how children may deal with these challenges. In chapter 2, Danziger takes up the problem of what kinds of meanings are expressed in nouns as opposed to verbs. She shows that actions like ‘run’, ‘laugh’, and ‘jump’are treated as structurally analogous to possessed nouns in Mopan Maya; for example, to express ‘I am jumping’, one says, very roughly, ‘My jumping continues’ (with the semantic information in ‘jump’ turning up in a nominal argument of an inflected predicate). This systematic violation of the supposedly canonical encoding of action meanings as verbs creates intriguing problems for both semantic and syntactic bootstrapping proposals. For example, if children use semantics to predict form class (semantic bootstrapping) they would misidentify actional nominals as verbs. And if they rely on a verb’s syntactic frames to get at verb meaning (syntactic bootstrapping), how will they discover the actional meanings in nominals? Ultimately, syntactic bootstrapping is argued to give the best results, but possibly at the cost of giving up noun and verb as universal categories. Chapter 3 (Bohnemeyer) shows problems for bootstrapping proposals in the domain of motion event coding. Yukatek Maya lacks the formal clues differentiating motion and non-motion events that the syntactic bootstrapping hypothesis depends on, such as a motion verb’s ability to appear in trajectory-expressing frames like ‘X VERBED from Y to Z’. At the same time, the meanings of Yukatek motion verbs differ systematically from those of their Indo-European counterparts in a way that cannot—counter to the claims of the other bootstrapping hypothesis, semantic

14

BOWERMAN AND BROWN

bootstrapping—be learned solely by observing the use of the verbs paired with real-world situations. An adequate explanation for how these forms are acquired can be achieved only by postulating a strong ongoing interaction in the child’s language processing between formal evidence, semantic evidence, and evidence from the behavior of the forms in discourse. Chapter 4 (Schultze-Berndt) considers the challenge for both semantic and syntactic bootstrapping posed by the phenomenon of complex predicates. In the Australian Aboriginal language Jaminjung, along with many other languages in the area, most verbal predicates are formally and semantically complex, consisting of one member of a small closed class of semantically generic inflecting verbs, plus one or more members of a second lexical category, “coverbs,” which mainly occur together with an inflecting verb. The meanings of the interacting forms are mutually defining: the semantic contribution of each element to the whole can only be determined by taking into account the meanings of the range of elements of the other class with which it co-occurs. A unidirectional model of acquisition by bootstrapping from either syntax to semantics or semantics to syntax is implausible in a case like this; acquisition must proceed by a dialectic between the two. Still another intriguing puzzle for bootstrapping theories is raised in chapter 5 (Kita). Approaches to argument linking on which bootstrapping theories are based assume that for a verb in any particular syntactic frame, each semantic argument role is linked to a particular grammatical role. But in a number of languages, including Japanese, Tzeltal Maya, and Likpe (Kwa, West Africa), the linking for some verbs is left more free, that is, it is “underspecified.” For example, consider the spatial configuration of meat on a skewer. If we encode this situation with the English verb pierce in a sentence like The skewer pierced the meat, the nominals naming the “pointed entity” (piercer) and the “penetrated entity” (piercee) must be subject and object; the reverse linking—*The meat pierced the skewer—is not possible. Speakers of Japanese, in contrast, can say the equivalent of either “The skewer pierced [sasuru] at the meat” or “The meat pierced at the skewer.” Their choice of which argument to link to which position is influenced by pragmatic considerations, but, crucially, it is not prespecified by either the meaning of the verb or more general linking rules. Underspecificity creates an important indeterminacy at the semantic/syntactic interface, with implications, as Kita shows, for both linguistic theory and language acquisition theory. Chapter 6 (Margetts) throws up a problem for the assumption, shared by both semantic and syntactic bootstrapping, that there is a universal set of basic verbs whose meanings are obvious concepts that would be formed by children anywhere, and whose syntax follows transparently from their meaning. (So that meaning, once established, can predict syntax, or, alternatively, syntax provides a good clue to meaning.) One often-cited example is the concept of ‘give’, which supposedly straightforwardly predicts three arguments: a giver, a thing given, and a recipient (Gleitman, 1990; Pinker, 1989). But in the Oceanic language Saliba, the ‘give’concept is divided between two verbs, which differ in argument structure. The verb

1.

INTRODUCTION

15

glossed as ‘give’ for third-person recipients is, as expected, ditransitive. But ‘give’ for first- and second-person recipients has a single object, the “gift” NP, and the recipient is indicated with a directional suffix comparable to ‘hither’ or ‘thither’ (depending on whether the recipient is the speaker or the addressee). Margetts discusses the problems this arrangement might present for learners of Saliba, and suggests some factors that may facilitate the learning task. The five chapters in section I all illustrate areas where widespread universalist expectations about the relationship between syntax and verb semantics are not met, and so where a unidirectional inference from either meaning to form or form to meaning cannot straightforwardly guide children to the correct form–function mappings. Additional examples of unexpected mappings between syntax and verb meanings, and suggestions for how children might deal with them, are presented later in the book; see in particular chapters 7, 10, and 11. Section II: Participants Present and Absent: Argument Ellipsis and Verb Learning. The first section of the volume shows that the relationship between a verb’s meaning and its argument structure is less direct than is often assumed. An additional problem often noted for syntactic bootstrapping, as discussed earlier, is argument ellipsis. If arguments are often missing in a language, how can a child identify a verb’s argument structure to begin with, and so get the syntactic bootstrapping process off the ground? The three chapters in this section are all based on developmental data in lesser-studied languages, as well as analyses of the relevant structures in adult speech. The section is kicked off by Wilkins in chapter 7, a “swing” chapter in the sense that it continues the section I focus on proposed correspondences between verb semantics and syntax, but shows how this topic is connected to issues of argument realization. Wilkins considers two verbs in the Australian language Arrernte, which are roughly glossable as ‘put’ and ‘look’. These belong to distinctly different semantic classes (transfer vs. perception), and so should— according to syntactic bootstrapping assumptions articulated by Gleitman (1990)—have different argument structures. But their argument structure is in fact identical: both are three-argument verbs, taking an ergative NP (the putter/looker), an absolutive NP (the thing put/seen), and a dative NP (the place of the thing put/seen). The two verbs do, however, differ strikingly in the frequency with which these arguments are actually realized in adult speech and where they are typically positioned with respect to the verb, and these differences are related to the meanings the verbs express. Wilkins suggests that children could use this information to infer that the verbs belong to different semantic classes. Intriguingly, he finds that even quite old children make errors in marking the crosslinguistically unusual “place seen” argument of Arrernte ‘look’, which suggests that some mappings of syntax and semantics may indeed be more “natural” than others, even if not universal.

16

BOWERMAN AND BROWN

Chapter 8 (Brown) shows that patterns of ellipsis may offer cues not only to the general semantic class of a verb but also to its semantic “richness.” For example, in the Mayan language Tzeltal, object arguments are more often omitted for verbs like k’ux ‘eat crunchy stuff’ than for verbs like tun ‘eat (anything)’, probably because the meaning of the verb itself narrows the listener’s search for the intended referents of unmentioned arguments. This pattern of argument realization is observable in the speech of both adults and children of age three-and-a-half to four years, suggesting that already at this age children might use rate of argument ellipsis as a cue to the specificity of verb meaning. Taken together, the chapters by Wilkins and Brown show that what argument ellipsis takes away with one hand (by making it harder to identify a verb’s arguments) it may to some extent give back with the other: patterns of argument ellipsis—for example, differences across verbs in which arguments are typically expressed or omitted—could provide potentially useful clues to the verb’s meaning. Chapter 9 (Allen) asks not what children can learn about verb meaning from patterns of ellipsis in adult speech, but rather what children know about the multiple discourse-pragmatic factors that influence decisions by fluent speakers about when to provide or omit arguments. Examining child data from Inuktitut, and reviewing related data from Korean (Clancy, 1996, 2003), Allen shows that even very young children are highly sensitive to a wide range of factors that condition argument ellipsis, and so might be able to use this information to infer the existence of “missing” arguments in adult speech. Section III. Transitivity, Intransitivity, and Their Associated Meanings: A Complex Work-Space for Learnability. No property of verbs has played a greater role in the linguistic and language-development literature than transitivity. The ability—or lack of it—to take a direct object is one of the quintessential concepts in the study of argument structure, verb meaning, and event structure. What can children infer about the meaning of a verb from the fact that it appears with only one argument? With two arguments? How do they decide whether a transitive verb presents an event as telic (culminating at an inherent end point) or atelic (an unbounded activity, process, or state)? How do they figure out how transitivity is related to causativity and state change? In this section, the themes of the first and second sections of this volume—verbs and verb syntax, and the realization of event participants—are intertwined, and it becomes increasingly clear how profoundly language acquisition can be influenced by the typological properties of the language being learned. Chapter 10 (Essegbey) considers the universality of claims that intransitive verbs fall into two formally distinct classes (the Unaccusativity Hypothesis and related proposals). These claims have been influential not only in linguistics but also in acquisition research, where children have been hypothesized to come to the acquisition task anticipating the split (e.g., Randall et al., 2004; van Hout, 1996). But

1.

INTRODUCTION

17

as Essegbey shows, the distinction is irrelevant in Ewe (Kwa, West Africa), where all intransitives fall into a single class which has a constructional meaning to do with “lack of control” on the part of the event participant picked out by the subject noun phrase. Meanings involving a single participant who is in control of his action—which in English and other European languages are typically expressed with intransitives of the unergative class, as in John jumped—are in Ewe expressed with transitive constructions (e.g., ‘Kofi “jump” hill’ = Kofi jumped [on purpose]). This arrangement raises challenges for bootstrapping proposals, which assume that there will be a match between the number of event participants a verb is associated with and the number of noun phrases it appears with (Fisher, 1996, 2000; Fisher et al., 1994; Gleitman, 1990). Essegbey suggests that learners may be able to use syntactic bootstrapping to discover the meanings of new verbs in Ewe—but only after they have already learned the language-specific association between transitivity and control in their language. Chapter 11 (Ameka) also considers problems posed by Ewe for the universality of the Unaccusativity Hypothesis, and explores, through an in-depth examination of the supposedly unaccusative verb ‘die’ in Ewe and other languages, how cultural/historical processes can bring about a complex variety of meaning shifts that are not transparently related to syntax (cf. also section 2.4, earlier, on multiple frames). Ameka shows that Ewe ‘die’ can appear as both a one-place (intransitive) and a two-place (transitive) predicate. It has at least three senses, but these do not line up neatly with the number of arguments the verb appears with, for example, the same sense is involved in both a one-place construction like ‘she died’ and a two-place cognate-object construction like ‘she died a wicked death’, whereas different senses may be expressed by formally identical two-place constructions, for example, ‘this garment die dirt’ [=is dead dirty: intensity] vs. ‘he die ear to the matter’ [=he turned a deaf ear/does not want to hear: negative desiderative]. Although the senses of Ewe ‘die’ cannot be predicted from the verb’s syntactic frame, Ameka shows that they are related, albeit indirectly, to the properties of the event participant(s), such as whether or not they are animate. To learn the various meanings of ‘die’, he suggests, a child must pay close attention—as also proposed by Gropen et al. (1997)—to properties of the verb’s participants. Chapter 12 (van Hout) takes up the problem of how children discover a transitive verb’s telicity entailments. Telicity is an aspectual notion having to do with whether an event is presented as having a natural endpoint (e.g., break a pot, write a letter: telic) or as unbounded (love Mary, write letters: atelic). It is a property of verbs, and often of entire verb phrases. Van Hout shows that languages signal telicity in a variety of ways, some more closely related to a verb’s syntactic frames than others. For example, in English, Dutch, and Finnish, the form of a verb’s direct object NP, and in particular this NP’s count/mass syntax, provides critical cues to telicity (compare John wrote LETTERS [atelic] with John wrote THE LETTER [telic]). In Polish and Russian, in contrast, it is the verb form itself that signals telicity (for example, PISAL pis’mo ‘wrote, was writing letter’ [imperfective verb form: atelic] vs. NA-PISAL

18

BOWERMAN AND BROWN

pis’mo ‘wrote letter’[perfective verb form: telic]). On the basis of experiments, van Hout concludes that children have more trouble grasping telicity entailments when they are signaled by the direct object nominal (“compositional telicity,” in her terminology) than when they are marked directly on the verb (“predicate telicity”). The interaction of telicity with direct object marking gives rise to drawn-out learning problems. The final three chapters of the volume revolve around causativity, and in particular causative-transitive verbs. Chapter 13 (Bowerman and Croft) uses spontaneous speech corpora collected longitudinally from two learners of English over an extended period—about age 2 to 12 years—to examine the nature and time course of causative overgeneralizations of the causative-inchoative alternation (e.g., Don’t giggle me; see also examples (1a) and (1b) earlier). Overgeneralizations of argumentstructure alternations have given rise to intense debates on the nature of language learning, as discussed earlier in section 1.1—in particular, whether alternations are mastered through general learning principles such as schema abstraction and fine-tuning (Goldberg, 1995; Goldberg et al., 2004; Tomasello, 2003), or with assistance from inborn knowledge of the relationship between verb meaning and argument linking (Gropen et al., 1991a, 1991b; Pinker, 1989). Bowerman and Croft find that their subjects’ overall pattern of error-making, including the order in which errors with verbs of different semantic classes die out, accords poorly with Pinker’s model. The data point instead to a gradual learning process, and especially to the importance of the language-usage-based mechanism of entrenchment (here, simply hearing a verb repeatedly over time only in an intransitive frame). Chapters 14 (Wittek) and 15 (Pederson) introduce an intriguing new acquisition puzzle for transitive causative verbs such as kill, open, and break: how do children decide whether the state change associated with such a verb is actually entailed by the verb’s meaning? It is a commonplace observation that such verbs entail their state change—for instance, you can’t say *John broke a plate, but the plate didn’t break. But Wittek’s study of German-speaking children shows that learners do not necessarily recognize this entailment: unlike adults, they often accept sentences with causative verbs as descriptions of events in which an action was performed that could be expected to bring about the relevant state change, but the state change did not occur. Wittek suggests that children may interpret causal state-change verbs similarly to verbs like wash: the verb implies a certain state change—‘clean’, in this case—but it does not actually entail it (one can say I washed the shirt, but it is still dirty).7 In crosslinguistic perspective, German children’s failure to recognize the state-change entailment in verbs like ‘break’ and ‘kill’ is not anomalous: adult speakers of Tamil, tested with the same materials, also tend to accept such sentences even when the state change is not realized (chapter 15, Pederson). In the linguistic literature, it has been argued that in some languages—for example, Tamil and Mandarin—transitive causative verbs do not, in fact, entail a state change (Ikegami, 1985; Talmy, 1991). But Pederson argues that this conclusion is wrong

1.

INTRODUCTION

19

for Tamil. He suggests that the basic meaning of ‘break’ and other transitive causative verbs is similar in Tamil and English, but that Tamil is more tolerant than English in allowing speakers to stretch these verbs to situations of near-realization of an event. This tolerance is related, argues Pederson, to larger differences between Tamil and English in the availability of explicit linguistic devices to confirm or deny that an event has been realized; for example, English is richer than Tamil in “denial” markers, as in John ALMOST killed Harry. Together, these two studies show that the status of the state change associated with a transitive verb may be unclear. Bootstrapping theories of how verb meanings are learned offer little help with this problem, but an effective disambiguator comes from a surprising source (Wittek, chapter 14): when children are exposed to a novel verb coupled with a single instance of the adverb wieder ‘again’, in a context where the adverb must refer not to the repetition of the action but to the re-establishment of a prior state (as in The door blew open but John closed it again), they can immediately infer that the state change associated with the new verb is entailed, and not merely implicated. Cross-Cutting Themes. Cutting across the three major sections of this volume there are several additional themes worth pulling out for explicit mention: • Apparent mismatches between a verb’s number of semantic participants and its number of arguments (chapters 4, 6, 7, 10, and 11). A notable subgroup of chapters within this category discusses mismatches involving verbs that express events of “externally caused transfer or change of possessor,” such as ‘put’ and ‘give’ (chapters 4, 6, 7): e.g., ‘put’ or ‘give’ as a two- argument verb, and ‘look’ as a three-argument verb identical in argument structure to ‘put’. • Crosslinguistic variation in the semantics and syntax of expressions of motion (chapters 3, 4, 5). For example, are “motion verbs” always associated with distinctive path-trajectory syntactic frames? Do they always inherently specify “motion,” or do they sometimes treat motion as involving change of location or configuration? • Emphasis on the need to assume a continual interaction between meaning and form in children’s acquisition of verb meaning and syntax, rather than a one-way street from semantics to syntax (semantic bootstrapping) or from syntax to semantics (syntactic bootstrapping) (chapters 3, 4, 10). • The existence of multiple sources of information that could provide useful cues to verb meaning (in addition to, or other than, the traditional focus of bootstrapping proposals, i.e., number and syntactic arrangement of the verb’s arguments, and the distinction between arguments that are noun phrases or propositions; cf. Gleitman et al., 2005). For example, the set of elements a complex verb collocates with (chapters 3, 10); adverbs, which may modify a specific aspect of a verb’s meaning (chapter 14); statistical patterns

20

BOWERMAN AND BROWN

of argument omission and realization (chapters 7, 8, 9); semantic properties of event participants (e.g., animate or not) (chapter 11); the syntax of noun phrases encoding event participants (important for telicity entailments) (chapter 12); and possible inferences about meaning based not only on what is said, but also what is not said (chapter 15). • The relevance for learners of the larger system in which a particular aspect of language is embedded. This includes cultural practices which may make the “unexpected” presence or absence of arguments understandable (chapters 6, 7, 11); for example, Wilkins (chapter 7) shows how the surprising existence of a “place seen” argument for the Arrernte ‘look, see’ verb makes sense in the traditional cultural setting of the language. But it also includes the position of a particular to-be-learned element of language in the system of contrasts drawn by a language’s larger linguistic ecology, which can influence patterns of pragmatic inferencing (chapter 15). • The appeal of the construction grammar approach to argument structure. A number of the contributors to this volume suggest that construction grammar (Croft, 2001; Goldberg, 1995) can deal insightfully with within-language systematicities in argument structure, without presupposing that the regularities embodied in argument-structure constructions are necessarily universal (chapters 3, 4, 5, 10, 11). • Attention to the fine-scale mechanisms underlying language learning, such as what drives generalization and retreat from overgeneralization, and whether there are elements that can rapidly trigger certain kinds of inferences about meaning (chapters 5, 13, 14). 4. CONCLUSION This volume provides the first sustained crosslinguistic examination of a set of densely interwoven issues revolving around the role of argument structure in language acquisition. Eight of its chapters are typological in nature, examining the basic structures of a variety of languages with language-learnability issues in mind. Six chapters are devoted more directly to child language itself. Taken together, these chapters make clear that detailed work on crosslinguistic variation is critical if questions about language acquisition are to be handled insightfully. But they also show that despite the challenges that crosslinguistic variation raises for theories of language acquisition, especially for the bootstrapping hypotheses as they now stand, these challenges have plausible solutions. In particular, the contributors to the volume find that both semantic and syntactic bootstrapping can provide the child with valuable information, but that these two procedures must often work together in a dialectic to arrive at a satisfactory outcome, rather than running off separately (see also Grimshaw, 1994). These two procedures are also likely to interact continually with a wider set of clues to the meaning and syntactic structure of verbs, including, for example, statistical

1.

INTRODUCTION

21

patterns of argument ellipsis, adverbs, co-verbs, and other elements that co-occur with verbs, information about event participants, cultural practices, and conventional pragmatic inferences. This mix is rich, and it will not be easy to disentangle the role of the various ingredients. But taken together, the chapters of this volume help point the way to a more sophisticated understanding of the learnability issues associated with argument structure. And that is an outcome worth striving for. ACKNOWLEDGMENTS We are grateful to Cynthia Fisher and David Wilkins for their insightful comments on an earlier draft of this chapter. NOTES 1

A fuller explanation of Pinker’s proposal about how innate knowledge constrains children’s argument structure is presented in chapter 13. 2 According to Jaakkola and Akhtar (2000: 38), “Findings like these raise some questions about the proposal that even young children have knowledge of general “linking rules” between grammatical and semantic categories of the type posited in theories of semantic and syntactic bootstrapping…These theories propose: that children have a general understanding of abstract semantic and syntactic categories (like agent and subject); that they are able to correctly assign these categories based on the structures of the event and the sentence; and that they know the correct mapping between the two (e.g., agent–subject). In contrast to this proposal, the current studies showed that … children are not able to use the abstract transitive frame to figure out who is who. In other words, they either do not have the abstract categories of agent or subject, do not know that the agent is the subject, or do not know where the subject is located in the sentence.” 3 Or at least aspects that are easily parameterized—i.e., that are hypothesized to take one of two or a small number of values, such as “subject is/is not obligatory”. 4 The single exception is morphological and syntactic ergativity, a family of patterns in which it is arguably patients rather than agents that get linked to subject position. See Marantz (1984), Pinker (1984), Pye (1990), and Van Valin (1992) on how children could learn ergative languages. 5 Even for these languages there is the thorny question of how children distinguish between arguments and obliques (Fisher et al., 1994: 368; Grimshaw, 1994: 417; see Randall, 1990, for one suggestion). For example, in He read the book ON HIS BED, the PP on his bed is an oblique, and so has no significance for the meaning of read, whereas in He put the book ON HIS BED, the same PP is an argument of put that is intimately related to its meaning. 6 E.g., the first spray is hypothesized to mean roughly “cause Y [here, paint] to move in a certain manner”, whereas the second spray means something like “cause Z [the wall] to go to the state of being covered (by causing X to go onto it in a certain manner)”. In both versions of the verb, it is the affected object (second argument of

22

BOWERMAN AND BROWN

“cause”) that is the direct object, but which argument this is differs in the two versions of the verb, because, by hypothesis, the versions have different meanings. 7 Chen (2005) has recently used Wittek’s test materials with learners of Mandarin. In this language, caused state changes are typically expressed by resultative verb compounds (RVCs), e.g., zhai-xia ‘do.picking.action-descend’ (= pick, pick off/down). In an RVC it is the second verb that confirms that the state change came about; the first verb, e.g. ‘pick’, ‘break’, ‘kill’, does not, in itself, entail a state change. Ironically, Mandarin learners often made the opposite error to that of German learners: rather than incorrectly failing to recognize that a verb like ‘pick’, ‘break’, or ‘kill’ entails a state change, they incorrectly failed to recognize that it does not entail a state change. Chen concludes that children’s early strategies for interpreting verb meaning are not language-independent, but are influenced at a young age by the specific lexicalization patterns of the target language. REFERENCES Akhtar, N. (1999). Acquiring basic word order: Evidence for data-driven learning of syntactic structure. Journal of Child Language, 26, 339–356. Akhtar, N., & Tomasello, M. (1997). Young children’s productivity with word order and verb morphology. Developmental Psychology, 33, 952–956. Allen, S. (2000). A discourse-pragmatic explanation for argument representation in child Inuktitut. Linguistics, 38, 483–521. Bavin, E. (Ed.). (2000). A functional approach to ellipsis in child language. Linguistics, 38, 569–589. Bowerman, M. (1982a). Evaluating competing linguistic models with language acquisition data: Implications of developmental errors with causative verbs. Quaderni di Semantica, 3, 5–66. Bowerman, M. (1982b) Reorganizational processes in lexical and syntactic development. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art (pp. 319–346). New York: Academic Press. Bowerman, M. (1988). The ‘no negative evidence’ problem: How do children avoid constructing an overly general grammar? In J. A. Hawkins (Ed.), Explaining language universals (pp. 73–104). Oxford: Blackwell. Bowerman, M. (1990). Mapping thematic roles onto syntactic functions: Are children helped by innate linking rules? Linguistics, 28, 1291–1330. Braine, M. D. S., & Brooks, P. J. (1995). Verb argument structure and the problem of an overly general grammar. In M. Tomasello & W. E. Merriman (Eds.), Beyond names for things: Children’s acquisition of verbs (pp. 353–376). Hillsdale, NJ: Lawrence Erlbaum Associates. Brinkmann, U. (1993). Nonindividuation versus affectedness: What licenses the promotion of the prepositional object? In E. V. Clark (Ed.), Proceedings of the 25th Annual Child Language Research Forum (pp. 158–170). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Brinkmann, U. (1997). The locative alternation in German: Its structure and acquisition. Amsterdam: John Benjamins. Brooks, P. J., & Tomasello, M. (1999). How children constrain their argument structure constructions. Language, 75, 720–738. Chen, J. (2005). Interpreting state-change: Learning the meaning of verbs and verb compounds in Mandarin. In A. Brugos, M. R. Clark-Cotton, & S. Ha (Eds.), Supplement to the Proceedings of the 29th Boston University Conference on Language Development, Boston (http://128.197.86.186/posters/29/ChenBUCLD2004.pdf).

1.

INTRODUCTION

23

Choi, S. (1998). Verbs in early lexical and syntactic development in Korean. Linguistics, 36, 755–80. Clancy, P. M. (1996). Referential strategies and the co-construction of argument structure in Korean acquisition. In B. Fox (Ed.), Studies in anaphora (pp. 33–68). Philadelphia: John Benjamins. Clancy, P. M. (2003). The lexicon in interaction: Developmental origins of Preferred Argument Structure in Korean. In J. W. Du Bois, L. E. Kumpf, & W. J. Ashby (Eds.), Preferred Argument Structure: Grammar as architecture for function (pp. 81–108). Amsterdam: John Benjamins. Croft, W. (1991). Syntactic categories and grammatical relations: The cognitive organization of information. Chicago: University of Chicago Press. Croft, W. (1998). The structure of events and the structure of language. In M. Tomasello (Ed.), The new psychology of language: Cognitive and functional approaches to language (pp. 67–92). Mahwah, NJ: Lawrence Erlbaum Associates. Croft, W. (2001). Radical construction grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press. Croft, W. (in press). Aspectual and causal structure in event representation. In V. M. Gathercole (Ed.), Routes to language: Studies in honor of Melissa Bowerman. Mahwah, NJ: Lawrence Erlbaum Associates. Demuth, K., Machobane, M., & Moloi, F. (2000). Learning word-order constraints under conditions of object ellipsis. Linguistics, 38, 545–568. Dowty, D. R. (1979). Word meaning and Montague Grammar: The semantics of verbs and times in Generative Semantics and in Montague’s PTQ. Dordrecht, The Netherlands: D. Reidel. Erteschik-Shir, N., & Rapoport, T. (2005). The syntax of aspect: Deriving thematic and aspectual interpretation. Oxford: Oxford University Press. Fernandes, K. J., Marcus, G. F., Di Nubila, J. A., & Vouloumanos, A. (2006). From semantics to syntax and back again: Argument structure in the third year of life. Cognition, 100, B10–B20. Fisher, C. (1996). Structural limits on verb mapping: The role of analogy in children’s interpretations of sentences. Cognitive Psychology, 31, 41–81. Fisher, C. (2000). From form to meaning: A role for structural alignment in the acquisition of language. In H. W. Reese (Ed.), Advances in child development and behavior (Vol. 27, pp. 1–53). San Diego, CA: Academic Press. Fisher, C. (2002). The role of abstract knowledge in language acquisition: A reply to Tomasello (2000). Cognition, 82, 259–278. Fisher, C., & Gleitman, L. (2002). Language acquisition. In H. Pashler & R. Gallistel (Eds.), Steven’s handbook of experimental psychology: Vol 3. Learning, motivation, and emotion (3rd ed., pp. 445–496). New York: Wiley. Fisher, C., Hall, D. G., Rakowitz, S., & Gleitman, L. (1994). When it is better to receive than to give: Syntactic and conceptual constraints on vocabulary growth. In L. Gleitman & B. Landau (Eds.), The acquisition of the lexicon (pp. 333–375). Cambridge, MA: MIT Press. Gertner, Y., Fisher, C., & Eisengart, J. (2006). Learning words and rules: Abstract knowledge of word order in early sentence comprehension. Psychological Science, 17, 684–691. Gillette, J., Gleitman, H., Gleitman, L., & Lederer, A. (1999). Human simulations of vocabulary learning. Cognition, 73, 135–176. Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55. Gleitman, L. R., Cassidy, K., Nappa, R., Papafragou, A., & Trueswell, J. C. (2005). Hard words. Language Learning and Development, 1, 23–64. Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press.

24

BOWERMAN AND BROWN

Goldberg, A. E., Casenhiser, D. M., & Sethuraman, N. (2004). Learning argument structure generalizations. Cognitive Linguistics, 15, 289–316. Goldin-Meadow, S., & Mylander, C. (1998). Spontaneous sign systems developed by deaf children in two cultures. Nature, 391(6664), 279–281. Grimshaw, J. (1981). Form, function, and the language acquisition device. In C. L. Baker & J. J. McCarthy (Eds.), The logical problem of language (pp. 165–183). Cambridge, MA: MIT Press. Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press. Grimshaw, J. (1994). Lexical reconciliation. In L. Gleitman & B. Landau (Eds.), The acquisition of the lexicon (pp. 411–430). Cambridge, MA: MIT Press. Gropen, J., Epstein, T., & Schumacher, L. (1997). Context-sensitive verb learning: Children’s ability to associate contextual information with the argument of a verb. Cognitive Linguistics, 8, 137–182. Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991a). Affectedness and direct objects: The role of lexical semantics in the acquisition of verb argument structure. Cognition, 41, 153–195. Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991b). Syntax and semantics in the acquisition of locative verbs. Journal of Child Language, 18, 115–151. Hale, K., & Keyser, S. J. (1993). On argument structure and the lexical expression of syntactic relations. In K. Hale & S. J. Keyser (Eds.), The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger (pp. 53–109). Cambridge, MA: MIT Press. Ikegami, Y. (1985). ‘Activity’—‘accomplishment’—‘achievement’—A language that can’t say ‘I burned it but it did not burn’ and one that can. In A. Makkai & A. K. Melby (Eds.), Linguistics and philosophy: Essays in honor of Rulon S. Wells (pp. 265–304). Amsterdam: John Benjamins. Jaakkola, R., & Akhtar, N. (2000). Assessing children’s knowledge of word order with familiar and novel verbs. In E. V. Clark (Ed.), Proceedings of the 30th Annual Child Language Research Forum (pp. 33–40). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Jackendoff, R. (1987). The status of thematic relations in linguistic theory. Linguistic Inquiry, 18, 369–411. Jackendoff, R. (1990). Semantic structures. Cambridge, MA: MIT Press. Keenan, E. (1976). Toward a universal definition of “subject.” In C. Li (Ed.), Subject and topic (pp. 303–333). New York: Academic Press. Landau, B., & Gleitman, L. R. (1985). Language and experience: Evidence from the blind child. Cambridge, MA: Harvard University Press. Lee, J. N., & Naigles, L. R. (2005). The input to verb learning in Mandarin Chinese: A role for syntactic bootstrapping. Developmental Psychology, 41, 529–540. Levin, B. (1993). English verb classes and alternations. Chicago: University of Chicago Press. Levin, B., & Rappaport Hovav, M. (1995).Unaccusativity: At the syntax–lexical semantics interface. Cambridge, MA: MIT Press. Li, P., & Shirai, Y. (2000). The acquisition of lexical and grammatical aspect. New York: Mouton de Gruyter. Lidz, J., Gleitman, H., & Gleitman, L. (2003). Understanding how input matters: Verb learning and the footprint of universal grammar. Cognition, 87, 151–178. Lieven, E., Pine, J., & Baldwin, D. (1997). Lexically-based learning and early grammatical development. Journal of Child Language, 24, 187–219. Marantz, A. (1984). On the nature of grammatical relations. Cambridge, MA: MIT Press. Naigles, L. R. (1996). The use of multiple frames in verb learning via syntactic bootstrapping. Cognition, 58, 221–251.

1.

INTRODUCTION

25

Naigles, L. R. (1998). Developmental changes in the use of structure in verb learning. In C. Rovee-Collier, L. Lipsitt, & H. Haynes (Eds.), Advances in infancy research (Vol. 12, pp. 298–318). London: Ablex. Naigles, L. R., Gleitman, H., & Gleitman, L. R. (1993). Children acquire word meaning components from syntactic evidence. In E. Dromi (Ed.), Language and cognition: A developmental perspective (pp. 104–140). Norwood, NJ: Ablex. Naigles, L. R., & Hoff-Ginsberg, E. (1995). Input to verb learning: Evidence for the plausibility of syntactic bootstrapping. Developmental Psychology, 31, 827–837. Naigles, L. R., & Hoff-Ginsberg, E. (1998). Why are some verbs learned before other verbs? Effects of input frequency and structure on children’s early verb use. Journal of Child Language, 25, 95–120. Narasimhan, B., Budwig, N., & Murty, L. (2005). Argument realization in Hindi caregiver– child discourse. Journal of Pragmatics, 37, 461–495. Perlmutter, D. M., & Rosen, C. (1984). Studies in relational grammar. Chicago: University of Chicago Press. Pinker, S. (1984). Language learnability and language development. Cambridge, MA: Harvard University Press. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Pinker, S. (1994). How could a child use verb syntax to learn verb semantics? In L. Gleitman & B. Landau (Eds.), The acquisition of the lexicon (pp. 377–410). Cambridge, MA: MIT Press. Pustejovsky, J. (1991). The syntax of event structure. In B. Levin & S. Pinker (Eds.), Lexical and conceptual semantics (Cognition Special Issues, pp. 47–82). Cambridge, MA: Blackwell. Pye, C. (1990). The acquisition of ergative languages. Linguistics, 28, 1291–1330. Randall, J. (1990). Catapults and pendulums: The mechanics of language acquisition. Linguistics, 28, 1381–1406. Randall, J., van Hout, A., Weissenborn, J., & Baayen, H. (2004). Acquiring unaccusativity: A cross-linguistic look. In A. Alexiadou, E. Anagnostopoulou, & M. Everaert (Eds.), The unaccusativity puzzle: Explorations of the syntax-lexicon interface (pp. 332–353). Oxford: Oxford University Press. Rispoli, M. (1987). The acquisition of transitive and intransitive action verb categories in Japanese. First Language, 7, 183–200. Rispoli, M. (1995). Missing arguments and the acquisition of predicate meanings. In M. Tomasello & W. E. Merriman (Eds.), Beyond names for things: Young children’s acquisition of verbs (pp. 331–352). Hillsdale, NJ: Lawrence Erlbaum Associates. Shirai, Y., Slobin, D. I., & Weist, R. M. (Eds). (1998). The acquisition of tense/aspect morphology. First Language, 18 (Special Issue). Talmy, L. (1991). Path to realization: A typology of event conflation. In L. A. Sutton, C. Johnson, & R. Shields (Eds.), Proceedings of the 17th Annual Meeting of the Berkeley Linguistics Society (pp. 480–519). Berkeley, CA: Berkeley Linguistics Society, University of California. Tenny, C. L. (1994). Aspectual roles and the syntax–semantics interface. Dordrecht, The Netherlands: Kluwer Academic. Tenny, C. L., & Pustejovksy, J. (Eds.). (2000). Events as grammatical objects: The converging perspectives of lexical semantics and syntax. Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Tomasello, M. (1992). First verbs. Cambridge, MA: Cambridge University Press. Tomasello,M.(2000a).Doyoungchildrenhave adult syntactic competence?Cognition,74,209–253. Tomasello, M. (2000b). The item-based nature of children’s early syntactic development. Trends in Cognitive Science, 2, 156–163.

26

BOWERMAN AND BROWN

Tomasello, M. (2003). Constructing a language. Cambridge, MA: Harvard University Press. Tomasello, M., & Abbot-Smith, K. (2002). A tale of two theories: Response to Fisher. Cognition, 83, 207–214. van Hout, A. (1996). Event semantics of verb frame alternations: A case study of Dutch and its acquisition. Ph.D. dissertation, University of Brabant, Tilburg, The Netherlands. Van Valin, R. D., Jr. (1992). An overview of ergative phenomena and their implications for language acquisition. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 3, pp. 15–37). Hillsdale, NJ: Lawrence Erlbaum Associates. Van Valin, R. D., Jr., & LaPolla, R. J. (1997). Syntax: Structure, meaning and function. Cambridge: Cambridge University Press.

PART I

Verb Meaning and Verb Syntax: Crosslinguistic Puzzles for Language Learners

CHAPTER 2

A Person, a Place, or a Thing? Whorfian Consequences of Syntactic Bootstrapping in Mopan Maya Eve Danziger University of Virginia

1. INTRODUCTION When Semitic, Chinese, Tibetan, or African languages are contrasted with our own, the divergence in analysis of the world becomes more apparent; and, when we bring in the native languages of the Americas, where speech communities for many millenniums have gone their ways independently of each other and of the Old World, the fact that languages dissect nature in many different ways becomes patent. The relativity of all conceptual systems, ours included, and their dependence on language stand revealed. (Whorf, 1940/1956: 214)

Over the last half-century, a compelling argument from child language has played a crucial role in discrediting claims for linguistic relativity such as this one. It is claimed (Chomsky, 1975) that the detailed structures of human languages are logically unlearnable under the conditions in which normal human children in fact routinely do acquire fluency in their native language (Bowerman, 1988). If the most important grammatical structures cannot in fact be learned, but are nevertheless undoubtedly acquired, it follows that the central facts of human language must be available to the child innately. And if this is so, it is easy to see that languages all over the world must be, at some deep level, and appearances sometimes to the contrary, fundamentally similar to one another. Whorf and his companion proponents of linguistic relativity were, it seems, just plain wrong. Even with a universal grammar to rely on, however, some problems in accounting for child language acquisition remain. For example, if we ask how children know the meanings of words in their language, we may well suppose that part of this knowledge is a matter of innate access to universal syntactic categories and functions, but we must 29

30

DANZIGER

still ask, for example, how English children pair the particular phonology /rən/ with the 1 English concept RUN (Gleitman, 1990). Since phonological forms clearly differ across languages, innateness can be only part of the answer here.“Bootstrapping” solutions to this problem propose that a universal and innate linkage exists between the meanings of lexical items and the syntactic form classes to which they belong. This linkage is crucial in allowing children to match phonological strings with the appropriate meanings and functions in their native languages. Bootstrappers suggest (Pinker, 1987, 1988, 1989, 1994) that, in any language, if a linguistic form refers to a Concrete Object, then it will function syntactically as a daughter node of an NP, falling into the familiar and universal form class “Noun.” Names for Actions or Changes of State will in their turn be universally linked to the form class “Verb,” which itself is defined by its syntactic privileges of occurrence as a VP daughter. Finally (at least in early versions; cf. Pinker, 1987: 406), “Names for Attributes” will be linked to the universal syntactic class “Adjective.” Once again, Whorf could not have been more wrong. Immediately following the stirring manifesto already quoted, Whorf continues: Let us consider a few examples. In English we divide most of our words into two classes, which have different grammatical and logical properties. Class 1 we call nouns, e.g., ‘house, man’; class 2, verbs, e.g., ‘hit, run’.… Our language thus gives us a bipolar division of nature. But nature herself is not thus polarized. If it be said that ‘strike, turn, run’ are verbs because they denote temporary or short-lasting events, i.e., actions, why then is ‘fist’ a noun? It also is a temporary event. Why are ‘lightning, spark, wave, eddy, pulsation, flame, storm, phase, cycle, spasm, noise, emotion’nouns? They are temporary events. If ‘man’and ‘house’are nouns because they are long-lasting and stable events, i.e., things, what then are ‘keep, adhere, extend, project, continue, persist, grow, dwell,’ and so on doing among the verbs? If it be objected that ‘possess, adhere’ are verbs because they are stable relationships rather than stable percepts, why then should ‘equilibrium, pressure, current, peace, group, nation, society, tribe, sister’, or any kinship term be among the nouns?

Whorf argues here that speakers do not derive their sense that a particular experience has the properties of an Action, say, rather than those of an Object, from its inherent nature. Rather, speakers treat an experience as having an inherent Action component just because it is treated syntactically as a Verb in their language. But examples like these of Whorf’s are readily accounted for under current theories of a universal syntax–semantics link in the lexicon. Naturally, says the current theory, in every language and in all syntactic classes we can expect that there will be found many semantically anomalous items. There may well be forms referring to non-Concrete non-Objects in the Noun class, for example (cf. the English abstract Nouns). These will be learned by children by semantic extension from the core members of each category. What is crucial for the universal syntax–semantics linkage is simply that linguistic items that do refer to genuine Concrete Objects should not fail to function as syntactic Nouns, and that those referring to clear-cut Actions should not fail to function as syntactic Verbs. In addition, it is important to recognize that form class derivation is an active process across languages; the underived

2.

A PERSON, A PLACE, OR A THING?

31

lexical root must conform to the universal linkage (e.g., English run encodes an Action Concept and should therefore be a Verb, but runner, visibly derived from run by the addition of extra morphology, is under no such constraint). 2. HOW BOOTSTRAPPING WORKS Bootstrappers differ among themselves in their views as to exactly how the child first comes to understand that a certain phonological string indeed refers to an Action, say, and not to a Concrete Object. Under semantic bootstrapping, Pinker (1987, 1988, 1989, 1994) proposes that experiential phenomena (such as, for example, what it means to HIT or to RUN, or to find oneself in the presence of a HORSE, HOUSE, or MAN) appear to all of us in universally similar ways, complete with attendant subsidiary semantics of the type ‘Concrete Object’ or ‘Action’. Children learn to associate word-length phonological strings with these straightforward nonlinguistic experiences, and from there it is a short step, via the universal syntax– semantics linkage, into the acquisition of native language syntax. But certainly Quine (1969) has been famously at pains to point out the logical difficulties of imagining that the referents of words are unambiguously available from context. Gleitman (1990) offers similar arguments why starting with naive appreciation of real-world semantics is implausible from a learnability point of view. In essence, any linguistic reference to even the most mundane real-world scene offers multiple possibilities of interpretation. Is the speaker interpreting the scene with reference to the Concrete Objects that are present, or to the Actions? How, in short, does the child determine “that it is the phonological object /run/, not /horse/ or /marathoner/—that is to be mapped onto the action concept” (Gleitman, 1990: 4)? Brown (1957) had early shown that young children will use syntactic information in order to approach the meaning of a novel word.2 The hypothesis of syntactic bootstrapping (Gleitman, 1990) follows up on these findings to take the radical and interesting step of proposing that children use their innate knowledge of the universal mapping between syntax and semantics in order to deduce semantic meaning from syntactic occurrence, as well as the other way around. According to syntactic bootstrapping, a young child may know more about the syntax of a given phonological string (Aha! There goes a Verb!) than about its semantics. But thanks to the universal syntax–semantics linkage, children can begin to acquire increasingly precise refinements of lexical meaning by reasoning from the syntactic and morphological position in which each string routinely occurs. In a convincing illustration, Gleitman (1990: 5) shows how a great deal of the semantic distinction between the English verbs to see and to look lies simply in their “syntactic privileges of occurrence” in an actual corpus. The verb to see always takes two arguments (Subject and Object; it is a Transitive Verb), whereas the verb to look always takes only one (the Subject; it is an Intransitive Verb). Says Gleitman (1990: 27), “The range of subcategorization frames has considerable potential for partitioning the verb set

32

DANZIGER

semantically, and…language learners have the capacity and inclination to recruit this information to redress the insufficiencies of observation.” A final refinement (Fisher, Hall, Rakowitz, & Gleitman, 1994; Maratsos, 1990) proposes that although the meanings of Verbs are acquired through syntactic bootstrapping, perhaps the meanings of Nouns are indeed acquired through semantic bootstrapping from self-evident experience with Concrete Objects. This of course merely re-poses much of the original problem. For how is the child to know which phonological strings are to count as Nouns and which are not? For many theorists, the answer once again places syntax before semantics (cf. Gleitman, 1990: 41–42). Quine notwithstanding, other theorists (Maratsos, 1990; see also Gentner, 1982; Gentner & Boroditsky, 2001) meanwhile continue to argue that Concrete Object semantics are especially apparent to young children and that Nouns are distilled from the input data on this basis, perhaps leaving the rest to be acquired according to syntactic occurrence. Regardless of the preferred variant, however, it is well understood that all bootstrapping theories of lexical acquisition depend absolutely on the postulate that there exists a natural link between semantic meaning and syntactic function in the lexicon of all languages. This linkage must be readily available (and is therefore probably “innate”; Fisher et al., 1992; Gleitman, 1990: 29) to all children embarking on the process of language acquisition. The linkage functions always and everywhere to match semantic concepts of the type Concrete Object, Action, and Attribute with syntactic objects of the type Noun, Verb, and Adjective, respect3 ively. Semantically anomalous members of any syntactic category (e.g., abstract nouns in English) are learned later in the acquisition process, by extension from nonanomalous cases. In general, word roots in the mental lexicon come labeled with a part-of-speech affiliation, much as they conventionally do in the dictionary. 3. PARTS OF SPEECH AND LINGUISTIC DIVERSITY In a clear challenge to this position, a small but persistent lobby within the field of descriptive linguistics continues to argue that the distinction between Noun and Verb as lexical categories cannot be empirically defended across all languages (Foley, 2005; Sapir, 1917, 1921; Sasse, 1993; Swadesh, 1938). Evidence, usually from non-European languages, is offered to query the position that those classes of lexical forms that encode Concrete Object semantics also correspond universally to a syntactic category recognizable as a Noun, or that those classes that encode Action concepts always correspond to a universal syntactic category of Verb. The most classic examples show that in some languages the same lexical roots can have either nominal or verbal function, as these are defined by their privileges of morphosyntactic occurrence. Such issues of Noun–Verb ambiguity are certainly well known to and often remarked on by students of the modern Mayan languages, and by specialists in the study of the classic Mayan inscriptions (Bohnemeyer, 1998; Bricker 1981a; Brown, 1998; Bruce, 1968; Hofling & Tesucún, 1997; Lois & Vapnarsky, 2003; Smailus, 1989).

2.

A PERSON, A PLACE, OR A THING?

33

Nowadays, however, to use even puzzling non-European data to pose an outright challenge to the universality of Noun and Verb constitutes very much the minority position in linguistics. Most linguists who have considered the problem since Whorf have concluded, with a discernible component of relief, that Noun and Verb can indeed be defended as crosslinguistically universal categories (Croft, 1993; Grimshaw, 1990; Hopper & Thompson, 1984; Langacker, 1990). The relief is traceable to the assumption that by defending Noun and Verb we can avoid what are taken to be the obvious—and obviously undesirable—Whorfian implications of any postulated Noun–Verb relativity (see Sasse, cited in Croft, 1993).4 An important component of the crosslinguistic defense of Noun and Verb has been the recognition (Hopper & Thompson, 1984) that we should not expect every occurrence in discourse even of a canonical Noun candidate form (e.g., English man, horse, or house) actually to behave syntactically as a Noun. Even English is full of possibilities for Nouns to adapt to Verbal function (as in to man the ship). In this respect, single examples from unfamiliar languages may be misleading, because they do not show all the possibilities of occurrence of a candidate form. To fully establish the existence of Noun, for example, as a semantic–syntactic linked category in the lexicon of any language, it is enough to show that in some reasonably routine way—but by no means necessarily always—the expected Concrete Object semantics is coupled with canonical Noun-like occurrence in syntax. To dispute the existence of such a category, on the other hand, a few examples of some noncanonical syntax–semantics match will not do. What is necessary is a demonstration that the canonical semantics–syntax co-occurrence never occurs in the language. This rallying of linguists to defend the universal categories of Noun and Verb certainly appears reassuring from the bootstrappers’ point of view. But in what follows I use linguistic data from Mopan Maya, an indigenous Central American language spoken by several thousand people in Guatemala and Belize, and currently being acquired as a first language by their children, to show that although it indeed remains possible to defend the existence of universal lexical categories of Noun and Verb for Mopan Maya, to do so leads us into unexpected philosophical terrain.5 In particular, in its current formulation, syntactic bootstrapping as applied to Mopan Maya forces us to consider very seriously Whorf’s proposal that the intuitive philosophical division between Action and Object is not a property of nature, but a construal drawn from the structure of Indo-European languages. 4. BOOTSTRAPPING INTO MOPAN MAYA: A HYPOTHETICAL EXCURSION Let us examine the syntactic form classes of Mopan from the point of view of a hypothetical Mopan child endeavoring to bootstrap his or her way into adult command of the language. We will assume the existence of universal syntactic

34

DANZIGER

categories “Noun,” “Verb,” and an attenuated “Adjective,” defined syntactically in the first instance by their possibilities of occurrence in the syntactic functions of Argument and Predicate (which to my knowledge have not been disputed in the linguistic literature), respectively. We will further assume a universal semantics–syntax linkage such that “names for concrete objects and for people are universally nouns” (Pinker, 1987: 406). Verbs will be semantically identifiable because they include “Names for Actions or Changes of State,” and Adjectives because they show 6 us “Names for Attributes.” We will find that in certain critical respects, the everyday linkage of semantics to syntax in Mopan grammar is quite unlike that of English. Most importantly, a large group of Mopan phonological strings that function as the lexical translation equivalents of canonical English Action concept words (like to run, to jump, to move) consistently function in Mopan as syntactic Nouns and not as syntactic Verbs. As root forms, these strings have no Verbal privileges of occurrence. Therefore, any hypothesized strategy of pure semantic bootstrapping (“if it names an Action, it must be a Verb”) simply will not lead the child to acquisition of adult Mopan. But if acquisition proceeds through syntactic bootstrapping (“if it’s a Noun, it refers to a Concrete Object”), we can only conclude that these strings (RUN, JUMP, MOVE, and so on) are not among those that have Action concept semantics in the minds of adult Mopan speakers. It is possible to avoid such a Whorfian conclusion, while maintaining the hypothesis of universal semantics–syntax linkage in the lexicon, only by sacrificing the traditional categories Noun and Verb completely. 4.1. Finding Universal Semantic–Syntactic Form Classes in Mopan 4.1.1. Nouns If we assume, with Maratsos (1990), Gentner (1982), and others, that Nouns are early recognized by children across languages by their relative absence of inflectional morphology and by their repeated occurrence in the presence of Concrete Objects, we will have done well by the Mopan child. In Mopan, a formally identifiable class of lexical items often referring to Concrete Objects (ready translation equivalents of English words like horse, man, house, etc.) exists, and indeed is relatively meager in morphological possibilities. Such forms, henceforth the Nouns of Mopan, can be inflected with an optional suffix for plurality, but with little else. So for Mopan tz’imin ‘HORSE’, we find tz’imin-oo’ ‘HORSES’.7 An optional preposed determiner indicates definiteness (a tz’imin ‘ THE HORSE ’). One of a series of person-indicating elements may also appear in the same position to indicate possession (in tz’imin ‘ MY HORSE’, u tz’imin ‘ HIS/HER HORSE’, etc.). Some Nouns (e.g., otoch ‘ HOUSE’, ni’ ‘ NOSE’, na’ ‘ MOTHER’) do not normally occur without a possessor expressed in this way;

2.

A PERSON, A PLACE, OR A THING?

35

others are not so constrained (cf. Danziger, 1996a). Except in cases of nominal predication (‘ IT’S A HORSE!’), Mopan Nouns do not function as main predi8 cates in clauses. They appear instead as syntactic Arguments to other forms, and refer to participants in the discourse (see examples 2, 4, 5, 7, and 9 below). Forms in this common class include ja’ ‘ WATER’; k’ubeete ‘ BUCKET’; maax ‘MONKEY’; miis ‘ BROOM’; winik ‘MAN’, and many others (consult Ulrich & Ulrich, 1976, for more examples). 4.1.2. Verbs Equally satisfactory to the bootstrapper is the fact that in Mopan it is not difficult to identify two semantic–syntactic classes that express “Action or Change of State” and that function exactly as we expect universal Verbs to do. Transitive and Intransitive Verbs are morphologically quite distinct in Mopan, but the members of both classes routinely play the role of main predicates in clauses, and take obligatory inflections both for person and for aspect/mood. Transitive Verbs. Mopan Transitive Verbs take obligatory morphological inflections to cross-reference their two dependent arguments, and each occurrence of such a Verb is also obligatorily inflected for one of four distinct aspect/mood statuses. Forms in this large Mopan class include ad ‘SAY SOMETHING’; b’et ‘MAKE OR DO SOMETHING’; il ‘SEE SOMETHING’; jätz’ ‘STRIKE SOMETHING’; jeb’ ‘OPEN’; muk ‘BURY SOMETHING’; pa’‘BREAK’; si’‘GIVE SOMETHING AWAY’; tz’a ‘PUT OR PLACE SOMETHING’ (consult Ulrich & Ulrich, 1976, for more examples). For Transitive Verbs, Completive Aspect inflection gives the clearest expression to all of the obligatory morphological slots: 1. Mopan Transitive Verb, Completive Aspect9 Uy-il-aj-ech 3ACT-see-CMPL-2PTNT ‘S/he saw you.’ 2. Mopan Transitive Verb, Completive Aspect, with Lexical Noun Phrase Argument 10 Uy-il-aj-Ø

in

3ACT-see-CMPL-3PTNT 1POSS

tz’imin horse

‘S/he saw my horse.’

Intransitive Verbs. Like the Transitives, the Intransitive Verbs also occur as main predicates and are marked for aspect/mood status. They take inflectional marking for a single dependent argument. Forms in this class include b’el ‘GO’; ch’i’

36

DANZIGER

‘GROW’; ch’ud ‘GET WET’; em ‘DESCEND’; jan ‘EAT’; ka’al ‘BEGIN’; ka’n ‘GET TIRED’; k’ax ‘FALL’; k’och ‘ARRIVE ELSEWHERE’; tal ‘COME’; uch ‘HAPPEN’ (consult Ulrich & Ulrich, 1976; Danziger, 1996b: 409–410, for more examples). For Intransitive Verbs, Subjunctive Mood inflection gives the clearest expression to all of the obligatory morphological slots:11 3. Intransitive Verb, Subjunctive Mood …ka’

tal-ak-ech

COMP

come-SUBJ-2PTNT

‘…that you should come.’ 4. Intransitive Verb, Subjunctive Mood, with Lexical Noun Phrase Argument … ka’

tal-ak-Ø

in

tz’imin

COMP

come-SUBJ-3PTNT

1POSS

horse

‘… that my horse should come.’

It will be of particular relevance for the later argument to note the existence of the regular Mopan Intransitive Verb uch ‘HAPPEN’: 5. Intransitive Verb uch, Subjunctive Mood with Lexical Noun Phrase Argument (This example is from Ulrich & Ulrich, 1982: 27, line 71) wa

ka’

uch-uk-Ø

a

yuklaja…

Q

COMP

happen-SUBJ-3PTNT

ART

earthquake

‘… if an earthquake should happen.’

4.1.3. Adjectives Showing no inflection for aspect/mood status, but still appearing as main predicates in clauses and exhibiting marking with a person suffix for a single argument, is a class of forms encoding Names for Attributes and other State Concepts. Forms in this class include, for example, Mopan b’ak ‘TO BE SKINNY’; b’es ‘TO BE MUTE’; b’ik’ ‘TO BE COARSE’; b’ox ‘TO BE BLACK’; ke’en ‘TO BE LOCATED’; kook ‘TO BE DEAF’; koom ‘TO BE TIGHT’; ko’oj ‘TO BE EXPENSIVE’; nooch ‘ TO BE LARGE’; saak ‘TO FEAR, BE AFRAID’; wi’ij ‘TO BE HUNGRY’; yan ‘TO EXIST’ (consult Ulrich & Ulrich, 1976, and Danziger, 1996b: 407–409, for more examples). As in many languages, these forms pattern somewhat like Verbs and somewhat like English Adjectives. Unlike Mopan Nouns, however, they do not appear as referring arguments in clauses. Let us call these simply the Stative Predicates of Mopan.12

2.

A PERSON, A PLACE, OR A THING?

37

6. Stative Predicate: Name for Attribute Nooch-ech be.large-2PTNT ‘You are great.’ 7. Stative Predicate with Lexical Noun Phrase Argument Nooch-Ø

in

tz’imin

be.large-3PTNT

1POSS

horse

‘My horse is large.’

One frequently occurring predicate in this class is the regular Stative Predicate tan ‘BE CONTINUOUS’, a form that will recur in the discussion that follows. 8. Stative Predicate tan K’ua tan-ech

aleeb’e?

What be.continuous-2PTNT

today

‘What are you doing today?’ 9. Stative Predicate tan with Lexical Noun Phrase Argument Tan-Ø

a

ja’

be.continuous-3PTNT

ART

water

‘It’s raining.’ (lit.: ‘The water is continuous.’)

4. 2. Summary We have seen that the syntactic category Noun, complete with Concrete Object semantics as required by the universal semantics–syntax linkage, can readily be defended in Mopan Maya. Equally, the syntactic category Verb, with the requisite semantics of “Action or Change of State,” has not been difficult to discover in this language. We might also readily point out that those Verb forms with Action semantics tend to take Transitive marking, whereas those that encode Changes of State take Intransitive marking. Even the problematic category Adjective has found a match in Mopan that is consistent with the current crosslinguistic understanding that “Names for Attributes” often have highly predicative properties. The survey of Mopan form classes has been brief, but incorporates the necessary attempt at exhaustivity.13 The problem for the Mopan child who attempts to learn this language on the basis of a universal syntax–semantics linkage based on Noun and Verb arises when he or she grapples with those lexical items perhaps most commonly cited in acquisition

38

DANZIGER

accounts as good examples of “Action Concept” encoding and as canonical Verbs. These are Mopan words such as alka’ ‘RUN’; awat ‘YELL’; ok’ot ‘DANCE’; siit’ ‘JUMP’; t’an ‘SPEAK’; che’ ‘LAUGH’; peek ‘MOVE’; and many others (consult Ulrich & Ulrich, 1976; Danziger, 1996b: 410–411, for more examples). These forms can be characterized as those having something like Action semantics, but with reference to only a single participant.14 In Mopan, such forms never have the syntactic privileges of occurrence associated with Verbs in the adult language. Instead, they function morphosyntactically like the Mopan Nouns we have already described. For RUN, JUMP, FLY and many others, therefore, semantic bootstrapping gives the wrong outcome for adult Mopan (forms that name Actions should by hypothesis be syntactic Verbs, but in this language they are not). Syntactic bootstrapping meanwhile will yield correct adult Mopan forms, but logically fails to generate any hypothesis of Action semantics for these apparently most canonical Action Concept forms! An account of the acquisition of these forms by Mopan children under a syntax–semantics linkage based on the categories Noun and Verb therefore cannot be achieved under semantic bootstrapping alone. An account based on syntactic bootstrapping, on the other hand, if based on a universal syntax–semantics link via the categories Noun and Verb, forces us to the Whorfian conclusion that the words RUN, JUMP, FLY, and so on do not have Action semantics in the minds of adult Mopan speakers. 5. MORPHOSYNTAX OF MOPAN MAYA ‘RUN’ AND ITS FELLOWS In Mopan Maya, ‘RUN’ is expressed phonologically as alka’, an unexceptional form that belongs to a coherent and readily identifiable form class. The morphosyntactic possibilities for the members of this class are such that they contrast reliably with the members of the classes identified earlier as Transitive Verbs, as Intransitive Verbs, and as Stative Predicates (“Adjectives”). In particular, in clear contrast to both Transitive and Intransitive Verbs, the single-argument Action concept forms (RUN, etc.) never inflect for aspect or for mood.15 And these forms never appear with the pronoun suffix that always cross-references the Argument of a Stative predicate. These descriptive statements do not represent preferences or statistical likelihoods. In adult Mopan it is a grammatical error of the first order to inflect RUN in the same way that HIT (Transitive Verb), GO (Intransitive Verb), and 16 EXIST (Adjective/Stative) are inflected. All this could just mean that we have in hand yet another class of Verbs in Mopan. But, fatally for bootstrapping via Noun and Verb, the morphosyntactic patterning of forms in the Mopan RUN class is in fact not such as to allow them to contrast readily with the class identified as Nouns earlier (MAN, HORSE, and so 17 on). In order to form a complete clause, Mopan RUN must, like any Noun, occur as a lexical Argument to another form, which is itself inflected for person and sometimes for aspect/mood. This other form functions as the main predicate of

2.

A PERSON, A PLACE, OR A THING?

39

the clause. Most common in that position are Intransitive Verbs (like uch ‘HAPand Statives (like tan ‘BE CONTINUOUS’) whose own internal semantics 18 convey some form of temporal contour. Like certain of the Nouns (e.g., HOUSE, NOSE, MOTHER), the non-transitive ‘Action-Concept’ forms (RUN, etc.) always 19 occur with a pronoun prefix that indicates personal possession.

PEN’)

10. Intransitive Verb uch, Subjunctive Mood, with Lexical Argument alka’ ‘RUN’ Compare with examples 4 and 5 shown earlier. … Ka’ uch-uk-Ø

in

alka’

COMP happen-SUBJ-3PTNT

1POSS

run

‘… that I should run.’ (lit.:‘… that my running should happen.’) 11. Stative Predicate tan with Lexical Argument alka’ ‘RUN’ Compare with example 9 shown earlier. Tan-Ø

in

be.continuous-3PTNT 1POSS

alka’ run

‘I run; I’m running.’ (lit.: ‘My running is continuous.’)

To reiterate, these examples illustrate a regular and very frequent pattern of occurrence for a large class of lexical items in the language. In addition, in Mopan there is no alternative form of words or inflections that would yield a more robust Verbal analysis for underived alka’ ‘RUN’ and the many other forms like it. 5.1. Bootstrapping Mopan ‘Run’ from Nouns and Verbs Because many canonical Action Concepts are not expressed by Verbs in this language, it seems clear that semantic bootstrapping alone cannot fully account for the acquisition of adult Mopan. Any Mopan child who embarks on a semantic bootstrapping strategy to acquire it will, by definition, have no difficulty in identifying Object, Action, Change of State, and Attribute semantics in the world. This child can readily sort through the available input to correctly assign phonological strings that refer to Concrete Objects to the syntactic Noun class, strings that refer to Changes of State to one of the Verb subclasses (Intransitive Verb), and strings that refer to Attributes to the Stative class. In addition, this child will be correct in assigning many phonological strings that indicate Action concepts to the Transitive Verb class. But we can predict that he or she will make many errors when assigning a syntactic affiliation to phonological strings that refer to the frequent situations in which Action is accomplished by only one participant. Our semantic bootstrapper persistently attempts to inflect these sorts of phonological strings as Verbs—perhaps using the existing Transitive or Intransitive aspect/mood suffixes to do so. This type of predicted error, diagnostic of the semantic bootstrapping approach, ought to be readily identifiable in any corpus of spontaneous Mopan child language.20

40

DANZIGER

The problem for acquisition arises of course when we ask how semantic bootstrappers could ever stop making errors of this type. At some stage the child must allow the syntactic facts of the language to override his or her innate endowment of expectations for the syntax–semantics link. This is not merely a matter of extending the contents or functions of a universal category already correctly learned via bootstrapping (as has been argued for semantic outliers like English justice, or non-canonical syntactic possibilities like English to man). No, semantic bootstrappers who find themselves learning Mopan must simply come (and, we suppose as usual, without negative evidence) actually to produce and understand lexical forms that flatly contradict the universal linkage which dictates that clear-cut Action semantics entails Verbal syntax. Some mechanism or strategy other than pure semantic bootstrapping must come into play to ensure that this happens. This mechanism could be syntactic bootstrapping. The child who uses syntactic bootstrapping to learn the Mopan lexicon (either for all phonological strings or only for those that do not refer to Concrete Objects) will in fact fare better than the semantic bootstrapper in terms of actually acquiring the adult language. This child will correctly use the Verbal syntax of Transitive and Intransitive Verb occurrences to construct Action and Change of State meanings for the phonological strings involved. And he or she will not make the mistake of attempting to impose Verbal syntax on alka’ ‘RUN’and forms like it, because morphosyntactic facts are the starting point, and not the end point, of the acquisition process under this strategy. On the contrary, by the rules of syntactic bootstrapping, the child can readily appreciate that these forms have Noun syntax, and will perhaps even attempt to apply them to Concrete Object referents on this basis. This child would therefore, we predict, make frequent acquisition errors that confuse reference to Action and to Object (e.g., alka’ ‘RUN’ used when HORSE is the intended referent).21 As acquisition progresses, we can assume that the child (like children everywhere) becomes able to understand that Concrete Object semantics does not constitute an ideal conceptual match for all of the phonological strings that populate the syntactic Noun class. The other members will eventually come to be understood as semantically anomalous Nouns, in a way perhaps analogous to the English child’s eventual understanding that some Nouns represent abstract entities and not Concrete Objects. Intriguingly from a Whorfian perspective, however, nowhere in this syntactic bootstrapping progression toward correct adult usage has there occurred a moment in which the Mopan child has considered the RUN forms for “Action Concept” semantics. It is the forms which are inflectable for aspect/mood, and which appear as clause predicates (like GIVE, MAKE, or COME, for example), that are being industriously scanned for this type of meaning. But not RUN. And not WALK, not KICK, not JUMP, not FLY. In none of these cases do the syntactic privileges of occurrence trigger any hypothesis of “Action Concept” semantics, by the rules of the syntactic bootstrapping game.

2.

A PERSON, A PLACE, OR A THING?

41

5.2. Theoretical Countermoves and Their Practical Failure There is in short no apparent reason why a Mopan child following a syntactic bootstrapping strategy based on Noun and Verb should ever arrive at the notion that RUN, YELL, LAUGH, JUMP, MOVE, and others constitute Actions or Events in the sense that we seem intuitively to understand those terms. On the other hand, semantic bootstrapping cannot alone account for acquisition of adult Mopan. It is possible, and perhaps even historically defensible, to hypothesize that all of the Mopan nontransitive Action-concept occurrences (RUN, etc.) actually represent deeper, more truly Verbal structures. There are various ways to declare this true for Mopan; the two most plausible involve invoking a surface-invisible process of nominalization, or analyzing the offending clauses as cases of serial predication.22 These analyses or others may perhaps save the theoretical situation for descriptive 23 linguistics. But none of them will help a child actually to acquire this language using bootstrapping based on Noun and Verb, because no theoretical account will change the facts of lexical occurrence that a child actually encounters and must eventually acquire. Because, under these Deep Verb solutions, no surface criteria of form can distinguish Noun from Verb in a great many Mopan cases, “syntactic” form class assignment, for a given lexeme (e.g., MOTHER, RUN), must be permitted to depend entirely on semantics. At best, the distinction between Noun and Verb in actual Mopan utterances could under such analyses be brought down to a matter of lexical collocation. Forms that follow predicates with temporal semantics ( TO HAPPEN, TO CONTINUE) can be declared Verbs. Forms that follow predicates with more existential semantics ( TO EXIST, TO DIE) can be declared Nouns. But such an account cannot be translated into a satisfactory bootstrapping strategy, because it leaves nothing on the syntax side of the necessary syntax–semantics link. To illustrate, let us propose, following the logic of the Deep Verb solution, that, faced with Action concepts in apparent Noun position, the Mopan child reverts to semantic bootstrapping. The form class that includes RUN will be analyzed as a Verb class. But how is the semantic bootstrapping child now to understand the semantics of forms in this same syntactic class that express HORSE, BUCKET, NOSE, MOTHER, and so on? Perhaps as “extended” or anomalous semantic Actions? If this is the solution, note that the Whorfian casualty this time is Concrete Object semantics itself. The syntactic bootstrapper, meanwhile, also can make little practical use of the Deep Verb view. It is difficult to imagine how the child might identify a distinct syntactic class (Verbs for Action concepts but not for Concrete Objects) when such a class does not exist in adult production. How is the child to learn word meanings from syntax when the phonological strings alka’ ‘RUN’ and tz’imin ‘HORSE’ have such similar privileges of morphosyntactic occurrence?

42

DANZIGER

At this point the defender of bootstrapping might be inclined to propose a dialectic course of acquisition in which the child simply switches back and forth from syntactic to semantic bootstrapping at need. As we have seen, it is possible that “switches” from syntactic to semantic bootstrapping and back could be visible in the type of acquisition errors made and not made by Mopan children. But “bootstrapping” is surely no longer really the right word for the procedure. If either end of the syntax–semantics link can be invoked as the starting point whenever necessary, then presumably neither end ever actually required assistance from the other: nothing substantial ever really needed to be acquired! In particular, in this case, knowing when to switch from a semantic to a syntactic starting point and back again would appear to require prior knowledge (“innate”?) of the peculiar syntax– semantics link that obtains in Mopan. 6. CONCLUSION The question of child language acquisition has assumed the status of a crucial theoretical concern in linguistic theory since the Chomskyan revolution of the mid-20th century. A central plank of the argument in favor of a single universal grammar underlying all human languages has been the claim that universal mechanisms are needed in order to account for child language acquisition. The most fruitful approach to characterizing such universal mechanisms so far appears to invoke bootstrapping mechanisms that depend on a universal syntax–semantics link. But at the heart of the Mopan acquisition problem sketched here lies the fact that the universal syntax–semantics link that is necessary to all bootstrapping proposals does not hold in this language. What we find in Mopan is that the syntactic form class encoding many of the canonical “Action concepts” most frequently cited in the language acquisition literature is syntactically identical with the one whose members encode reference to clear and equally canonical Concrete Objects. Because this is so, semantic bootstrapping based on the proposed universal linkage is not a viable building block for the acquisition of adult Mopan. And syntactic bootstrapping yields the possibly disturbing prediction that forms like RUN, JUMP, and FLY do not have Action Concept semantics for Mopan speakers. To avoid such predictions, language acquisition researchers will have to abandon Noun and Verb as categories of reflection, discussion, or investigation. So perhaps there is no universal linkage between syntax and semantics in the lexicon, and language acquisition via bootstrapping is not the answer. Or, more hopefully, perhaps there is a universal linkage, but our current understandings of syntax, semantics, or both have not so far been adequate to characterize it. To do so, we would have to search for a set of semantic contrasts that are crosslinguistically better attested to correlate with syntax than the contrast between Action and Object. This requires rethinking the linguistic categories and processes taken for granted in the more European-based linguistic theories that are the ones most commonly used by acquisition theorists. A clear starting-point from the Mopan perspective is that typology of linguistic form classes in which the semantics end of the

2.

A PERSON, A PLACE, OR A THING?

43

semantics–syntax linkage is anchored in Aktionsart or lexical aspect (DeLancey, 1985; Dowty, 1979; Foley & Van Valin, 1984; Levin & Rappaport, 1992; VanValin, 1990; Vendler, 1967; see also Danziger, 1996b), a semantics concerned with de24 grees of duration, control, telicity, and relationality. Whatever the solution adopted, it is at any rate clear from the Mopan data that such familiar semantic notions as Concrete Object, Action, and Attribute are in fact folk characterizations, and are not the semantic units in terms of which syntax is actually articulated across all languages. This last point naturally then leads us to consider to what extent our own linguistic intuitions are themselves language-influenced. The strong initial conviction that Action and Object were the obvious and only candidates for basic universal semantic classes perhaps indeed had something of the Whorfian about it too. As Whorf put it (1940/1956: 215): It will be found that an “event” to us means “what our language classes as a verb” or something analogized therefrom. And it will be found that it is not possible to define ‘event, thing, object, relationship,’ and so on, from nature, but that to define them always involves a circuitous return to the grammatical categories of the definer’s language.

To conclude, the hypothesis that Mopan children acquire their language by bootstrapping via universal categories of Noun and Verb has led us straight toward, and not away from, the radical consequences indicated by Whorf. But language acquisition via bootstrapping can perhaps be saved, and arguably without Whorfian consequences. Ironically, however, we can probably only do so if we are willing to abandon the traditional syntactico-semantic categories Noun and Verb as either universal in the structure of languages, or innate in the minds of language-learning children. NOTES 1

In my text, words and expressions from actual languages, including English, are mentioned in the text using italics (e.g., run, alka’). Concepts hypothesized to correspond to word meanings are presented in small caps, with glosses in quotation marks (e.g., ‘RUN’). The higher-order mental categories into which these glosses are hypothesized to fit (e.g., Action, Object) are capitalized. In direct quotes from other authors, however, the orthographic conventions of the original are respected. 2 Children in Brown’s study pointed to different likely referents for novel words (e.g., blick) according to the syntactic category in which the word was presented to them (e.g., some blick, a blick, blicking). As the reference to Brown’s classic study makes clear, the term syntax as used in the bootstrapping context must be broadly interpreted to include inflectional morphology as well as strict positioning of a given element in the phrase or sentence. This is quite correct from a crosslinguistic point of view, because the syntax-morphology boundary is far from uniform in the world’s languages. 3 Although it now appears quite clear that “Adjective” is a problematic category from the universal semantics–syntax linkage point of view (Dixon, 1982; Maratsos, 1990; etc.).

44 4

DANZIGER

Quite contrary to his own stated intentions (cf. Whorf, 1940/1956: 218), Whorf has certainly been read more often as a preacher of determinism and constraint than as one of liberation through self-knowledge. 5 Mopan data and conclusions are based on my own field research with this language, conducted since 1986. All Mopan examples in this chapter are from my own corpus unless otherwise indicated. With the exception of some proper names, orthographic conventions for Mopan are as described in England and Elliott (1990). In this orthography, most of the consonants have roughly their Spanish values. Apostrophes represent glottal stops, or glottalization of the preceding consonant. The symbol ä denotes the mid-central vowel (schwa). 6 See Danziger (1996) and Ulrich, Ulrich, and Peck (1986) for more on Mopan grammar. 7 See Lucy (1992) for discussion of optional pluralization and other syntactic issues (including the relativity of the Mass/Count distinction) as they apply to Yukatek Maya. The issues are very similar in Mopan. They do not affect the current argument. 8 Mopan has no copula verb. As we have seen, the fact that forms referring to Concrete Objects and functioning as syntactic Arguments in many contexts may also function as Predicates in no way of course detracts from their identification as Nouns for bootstrapping purposes. 9 Abbreviations used in interlinear glossing: ACT, Transitive Subject; ART, Article; CMPL, Completive; COMP, Complementizer; POSS, Possessor; PTNT, Patient (Transitive Object/some Intransitive Subjects); Q, Interrogative/Conditional; SUBJ, Subjunctive; 1, First Person; 2, Second Person; 3, Third Person. For more on Mopan phonology and grammar see Ulrich et al. (1986) and Danziger (1996b). 10 The third person nonplural form of the Mopan patient-marking suffix in this environment is morphological zero (see example 2). 11 In Mopan Intransitive Verbs, the single participant (“Subject”) is often represented by the same pronoun suffix that also represents the Transitive Direct Object (cf. example 3). This does not affect the present argument. See Danziger (1996b) for full discussion. 12 Many of these can undergo derivation to form Intransitive (e.g., GET BIG from BIG) or Transitive (MAKE SOMETHING BIG from BIG) Verbs. 13 The form class of Positional Verbs, which appear to have a semantics ambiguous between voluntary “Action” and involuntary “Change of State,” has not been discussed, nor need it be for the purposes of the present argument. Forms in this class inflect for aspect/mood and include tin ‘SIT’, wa’ ‘STAND’, ch’uy ‘HANG’, sätz ‘STRETCH’, b’uy ‘COIL’, and many others. See Danziger (1996b) for a more complete list and for further discussion. 14 In other treatments, these might be called the Unergatives, Activity, or Manner predicates of Mopan (cf. Danziger, 1996b: 387; Perlmutter, 1978). 15 This is one of the major differences between Mopan and the other languages of the Yucatecan subfamily (Danziger, 1996b). 16 Derivational morphology can be added to alka’ ‘RUN’ and its fellows to yield Transitive Verbs (in the case of RUN, the semantic result is CHASE). We have

2.

A PERSON, A PLACE, OR A THING?

45

already seen that derived forms are not the relevant forms for confirming the universal syntax–semantics linkage (recall English run, runner). 17 In fact, familiar nouns from European languages are routinely borrowed into the RUN class (Verbeeck, 1997). This, for example, is where we find Mopan pieesta ‘PARTY’, a clear relative of the Spanish Noun fiesta. 18 Also very frequent in this position is the subjunctive of the Positional Verb wa’ ‘STAND, STAY’, yielding a habitual reading. 19 The same prefix also normally indicates the Subject of a Transitive Verb and can sometimes indicate the single participant of an Intransitive Verb (Danziger, 1996b; Ulrich, Ulrich & Peck, 1986). A supposition of two homonymous pronoun series, one for Nouns and one for Verbs, is the best solution for preserving our traditional form-class dichotomy. If we adopt this solution, then syntactic evidence from elsewhere in the language (non-existence of aspectmood inflection, appearance as clause Argument to inflected temporal predicate) dictates that the Activities (RUN, JUMP, FLY, etc.) must appear with the Nominal series of pronoun affixes. The homonymy solution is not of course completely satisfactory, especially in light of general typological and Mayan historical data (we know that the current state of affairs arose directly from a nominalization process; see Robertson, 1992). To abandon homonymy here, however, is certainly to confound the categories Noun and Verb in Mopan even further (cf. Bricker, 1981b)—to the point of proposing a syntactic continuum all the way from forms expressing semantic Accomplishments (here, the Transitive Verbs) to those expressing Concrete Object semantics. 20 Unfortunately no acquisition data for Mopan are available. Brown (1998) notes that children do not make inflectional errors when acquiring Noun–Verb ambiguous forms in the related language Tzeltal Maya. 21 For Quinean reasons, such errors as these would be difficult to identify. Without empirical access to the referential intentions of children, they would be virtually invisible in any spontaneous language corpus. It is possible that experimental contexts could be designed to test for them. 22 Even less efficacious in accounting for language acquisition is the observation that Noun–Verb ambiguous forms in extant languages can usually be shown to have historically originated from and/or to be progressing toward less ambiguous configurations (cf. Croft, 1993). This is almost certainly true of Mopan too, but is of little help to the several thousand children who are today acquiring the language in its current form. 23 Although it is not good news for such analyses that in Mopan’s sister languages, where it remains possible to inflect forms like RUN as Verbs, more, not less, morphology is necessary in order to do so (Bohnemeyer, 1998; Lehman, 1993; Lucy, 1994; Straight, 1976). 24 At present we have no idea what this sort of analysis would yield if employed to interpret acquisition across languages—for example, how it would play itself out in the literature of the crosslinguistic competition between so-called “Noun” and “Verb” in the young child’s vocabulary (Bassano, 2000; Brown, 1998; Choi & Gopnik, 1995; Gentner, 1978, 1982; Nelson, Hampson, & Kessler Shaw, 1993; Tardif, 1995).

46

DANZIGER

REFERENCES Bassano, D. (2000). Early development of nouns and verbs in French: Exploring the interface between lexicon and grammar. Journal of Child Language, 27, 521–559. Bohnemeyer, J. (1998). Time relations in discourse: Evidence from a comparative approach to Yukatek Maya. PhD dissertation, University of Tilburg, Tilburg, The Netherlands. Bowerman, M. (1988). The “no negative evidence” problem: How do children avoid constructing an overly general grammar? In J. A. Hawkins (Ed.), Explaining language universals (pp. 73–101). Oxford: Blackwell. Bricker, V. R. (1981a). Grammatical introduction. In E. Po’ot Yah (Ed.), Yucatec Maya verbs (Hocoba dialect)/Los verbos del Maya Yucateco (Dialecto de Hocoba) (pp. vi–xlvii). Latin American Studies Curriculum Aids, Center for Latin American Studies. New Orleans, LA: Tulane University. Bricker, V. R. (1981b). The source of the ergative split in Yucatec Maya. Journal of Mayan Linguistics, 2, 83–127. Brown, P. (1998). Children’s first verbs in Tzeltal: Evidence for an early verb category. Linguistics, 36, 713–753. Brown, R. (1957). Linguistic determinism and the part of speech. Journal of Abnormal and Social Psychology, 55, 1–5. Bruce, R. D. (1968). Gramática del Lacandón. Departamento de Investigaciones Antropológicas Publicaciones 21. Mexico City: Instituto Nacional de Antropología e Historia. Choi, S., & Gopnik, A. (1995). Early acquisition of verbs in Korean: A crosslinguistic study. Journal of Child Language, 22, 497–529. Chomsky, N. (1975). On cognitive capacity. In N. Chomsky, Reflections on language (pp. 3–35). New York: Random House. Croft, W. (1993). A noun is a noun is a noun—or is it? In J. S. Guenter, B. A. Kaiser, & C. C. Zoll (Eds.), Proceedings of the 19th Annual Meeting of the Berkeley Linguistics Society: Semantic Typology and Universals (pp. 369–381). Berkeley, CA: Berkeley Linguistics Society, University of California. Danziger, E. (1996a). Parts and their counter-parts: Social and spatial relationships in Mopan Maya. Journal of the Royal Anthropological Institute (N.S.), Incorporating MAN, 2, 67–82. Danziger, E. (1996b). Split intransitivity and active-inactive patterning in Mopan Maya. International Journal of American Linguistics, 62, 379–414. DeLancey, S. (1985). On active typology and the nature of agentivity. In F. Plank (Ed.), Relational typology. Trends in Linguistics, Studies and Monographs (Vol. 28, pp. 47–60). Berlin: Mouton de Gruyter. Dixon, R. M. W. (1982). Where have all the adjectives gone? In R. M. W. Dixon (Ed.), Where have all the adjectives gone? (pp. 1–62). Berlin: Mouton de Gruyter. Dowty, D. R. (1979). Word meaning and Montague Grammar. Dordrecht, The Netherlands: D. Reidel. England, N. C., & Elliott, S. R. (1990). Lecturas sobre la lingüística Maya. Antigua, Guatemala: Centro de Investigaciones Regionales de Mesoamérica. Fisher, C., Hall, D. G., Rakowitz, S., & Gleitman, L. R. (1994). When it is better to receive than to give: Syntactic and conceptual constraints on vocabulary growth. Lingua, 92, 333–375. Foley, W. A. (2005). Do humans have innate mental structures? Some arguments from linguistics. In S. McKinnon & S. Silverman (Eds.), Complexities: Beyond nature and nurture. Chicago: University of Chicago Press. Foley, W. A., & Van Valin, R. D., Jr. (1984). Functional syntax and universal grammar. New York: Cambridge University Press.

2.

A PERSON, A PLACE, OR A THING?

47

Gentner, D. (1978). On relational meaning: The acquisition of verb meaning. Child Development, 49, 988–98. Gentner, D. (1982). Why nouns are learned before verbs: Linguistic relativity versus natural partitioning. In S. A. Kuczaj (Ed.), Language development, Vol. 2: Language, thought, and culture (pp. 301–334). Hillsdale, NJ: Lawrence Erlbaum Associates. Gentner, D., & Boroditsky, L. (2001). Individuation, relativity, and early word learning. In M. Bowerman & S. C. Levinson (Eds.), Language acquisition and conceptual development (pp. 215–256). Cambridge: Cambridge University Press. Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55. Grimshaw, J. (1990). Argument structure. Cambridge MA: MIT Press. Hofling, A., & Tesucún, F. F. (1997). Itzaj Maya-Spanish-English dictionary/Diccionario Maya Itzaj-Espanol-Ingles. Salt Lake City: University of Utah Press. Hopper, P. J., & Thompson, S. A. (1984). The discourse basis for lexical categories in universal grammar. Language, 60, 703–752. Langacker, R. W. (1990). Nouns and verbs. Language, 63, 53–94. Lehmann, C. (1993). Predicate classes in Yucatec Maya. Función, 13–14, 195–272. Levin, B., & Rappaport Hovav, M. (1992). A preliminary analysis of causative verbs in English. Lingua, 92, 35–77. Lois, X., & Vapnarsky, V. (2003). Les racines verbo-nominales du maya yucateque. Faits de Lanque, 21, 41–70. Lucy, J. A. (1992). Grammatical categories and cognition: A case study of the linguistic relativity hypothesis. Cambridge: Cambridge University Press. Lucy, J. A. (1994). The role of semantic value in lexical comparison: Motion and position roots in Yucatec Maya. Linguistics, 32, 623–656. Maratsos, M. (1990). Are actions to verbs as objects are to nouns? On the differential semantic bases of form class categories. Linguistics, 28, 1351–1379. Nelson, K., Hampson, J., & Kessler Shaw, L. (1993). Nouns in early lexicons: Evidence, explanations and implications. Journal of Child Language, 20, 61–84. Perlmutter, D. M. (1978). Impersonal passives and the unaccusative hypothesis. In J. J. Jaeger, A. C. Woodbury, F. Ackerman, C. Chiarello, O. D. Gensler, J. Kingston, E. E. Sweetser, H. Thompson, & W. Kenneth (Eds.), Proceedings of the 4th Annual Meeting of the Berkeley Linguistics Society (pp. 157–189). Berkeley, CA: Berkeley Linguistics Society, University of California. Pinker, S. (1987). The bootstrapping problem in language acquisition. In B. MacWhinney (Ed.), Mechanisms of language acquisition (pp. 399–441). Hillsdale, NJ: Lawrence Erlbaum Associates. Pinker, S. (1988). Learnability theory and the acquisition of a first language. In F. S. Kessel (Ed.), The development of language and language researchers (pp. 97–119). Hillsdale, NJ: Lawrence Erlbaum Associates. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Pinker, S. (1994). How could a child use verb syntax to learn verb semantics? Lingua, 92, 377–410. Quine, W. V. (1969). Speaking of objects. In Ontological relativity and other essays (pp. 1–25). New York: Columbia University Press. Robertson, J. S. (1992). The history of tense/aspect/mood/voice in the Mayan verbal complex. Austin: University of Texas Press. Sapir, E. (1917). Review of Uhlenbeck. International Journal of American Linguistics, 1, 82–86. Sapir, E. (1921). Language: An introduction to the study of speech. New York: Harcourt Brace.

48

DANZIGER

Sasse, H. J. (1993). Syntactic categories and subcategories. In J. Jacobs, A. von Stechow, W. Sternefeld, & T. Venneman (Eds.), Syntax: An international handbook of contemporary research (Vol. 1, pp. 646–686). Berlin: Walter de Gruyter. Seiler, H. (1983). Possessivity, subject and object. Studies in Language, 7, 89–117. Smailus, O. (1989). Gramática del Maya Yucateco Colonial. Hamburg: Wayasbah-Verlag. Straight, S. H. (1976). Decompositional structure in Yucatec verbs. Mayan Linguistics, 1, 189–201. Swadesh, M. (1938). Nootka internal syntax. International Journal of American Linguistics, 9, 77–102 Tardif, T. (1995). Nouns are not always learned before verbs, but why? Evidence from Mandarin Chinese. In E. Clark (Ed.), Proceedings of the 26th Annual Child Language Research Forum (pp. 224–230). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Ulrich, M., & Ulrich, R. (1976). Diccionario Bilingüe: Maya Mopán y Español; Español y Maya Mopán. Guatemala City: Summer Institute of Linguistics. Ulrich, M., & Ulrich, R. (1982). Stories of the sacred, serious, sensational and silly from Mopan Maya texts. Guatemala City: Summer Institute of Linguistics. Ulrich, M., Ulrich, R., & Peck, C. (1986). Mopan Mayan verbs. Guatemala City: Summer Institute of Linguistics. Van Valin, R. D., Jr. (1990). Semantic parameters of split intransitivity. Language, 66, 221–260. Vendler, Z. (Ed.). (1967). Linguistics in philosophy. Ithaca, NY: Cornell University Press. Verbeeck, L. (1997). Linguistic acculturation in Mopan Maya. Lincom Studies in Native American Linguistics 5. Munich, Germany: Lincom Europa. Whorf, B. L. (1956). Science and linguistics. In J. B. Carroll (Ed.), Language, thought, and reality: Selected Writings of Benjamin Lee Whorf (pp. 207–219). Cambridge, MA: MIT Press. (Original work published 1940.)

CHAPTER 3

The Pitfalls of Getting from Here to There: Bootstrapping the Syntax and Semantics of Motion Event Coding in Yukatek Maya Jürgen Bohnemeyer The State University of New York at Buffalo

1. INTRODUCTION According to Landau and Gleitman’s (1985) syntactic bootstrapping hypothesis, children are guided in the acquisition of motion and state-change expressions by certain morphosyntactic clues which distinguish their meanings. In particular, source- and goal-denoting expressions such as into and out of identify (literal and metaphoric) motion event expressions. From the presence of these clues, children are able to predict that the expression encodes motion rather than state change. It is shown in this article that children acquiring Yukatek Maya cannot rely on such morphosyntactic clues to differentiate between motion and state-change meanings. Yukatek is a Native American language spoken by approximately 800,000 people living on the Yucatan peninsula in Mexico and Belize. In this language, the referential ground in a motion event, that is, the object or place with respect to which motion is described, is expressed by obliques which distinguish neither dynamicity (‘move to/from’ vs. ‘be at’) nor directionality (source vs. goal), and the verbs deployed in such constructions to assert change of location are morphologically members of a class of dedicated change-of-state verbs. So there is no morphosyntactic difference in Yukatek between the translations of ‘enter the house’and ‘die in the house’. The semantic bootstrapping hypothesis of Pinker (1984, 1989), in contrast, predicts that children start from universal cognitive representations and learn to package these into language-particular semantic representations. These are then encoded according to universal linking rules. However, the semantics of motion event constructions in Yukatek does not seem to fall inside what Pinker assumes to be 49

50

BOHNEMEYER

crosslinguistically invariant. The crosslinguistic variation in semantic construal that Pinker’s account allows is circumscribed by the scope of Talmy’s (1985, 2000) conflation typology. With Talmy, Pinker assumes that semantic representations of motion events invariably involve a “figure” moving along a “path” relationally defined with respect to a series of grounds (such as the “source” and “goal” of the motion event). But in Yukatek, motion is construed as location change of the figure with respect to single grounds. Motion from source to goal is not encoded as a single event in Yukatek (e.g., ‘She went from A to B’), but is represented as a sequence of a departure event and an arrival event, where the path traversed in between is left to implicature (e.g., ‘She left A, and then she arrived at B’). So Yukatek children have to learn to construe motion events for encoding in a way that is more different from how English children learn to construe motion events for encoding than Pinker assumes possible. Given that English and Yukatek children learn these different ways of construing events from listening to adult speakers talking about real world events that will be in many cases broadly similar across the two environments, it seems inevitable to conclude that children must pay more attention to language-particular structures than the semantic bootstrapping hypothesis assumes necessary. While neither semantic bootstrapping nor syntactic bootstrapping is considered a one-way road by its proponents, the two hypotheses do hold that children need, as a first approximation, only semantic input plus innate knowledge of linking rules and syntax to come up with reasonably good predictions of argument structures (semantic bootstrapping), and that they only need argument structure input and innate knowledge of syntax and linking rules to come up with reasonably good predictions of the “ball park meanings” (e.g., motion vs. physical state change) of verbs (syntactic bootstrapping). In contrast, the picture that emerges from the discussion of motion event encoding in Yukatek suggests that semantic learning and syntactic learning are more closely intertwined than the proponents of both bootstrapping hypotheses assume. Taking the evidence from Yukatek and English in a comparative perspective, it seems likely that children acquire verb meanings and argument structures in tandem. 2. THE SYNTACTIC BOOTSTRAPPING HYPOTHESIS In Landau and Gleitman’s (1985) proposal, the distinction between motion and nonmotion meanings plays a central role. Landau and Gleitman (1985: 130–136) argue extensively and forcefully that all languages differentiate these meanings with morphosyntactic clues that may guide learners in the acquisition of motion and nonmotion verbs. Landau and Gleitman are aware of important differences in the encoding of motion events across languages. However, they hold that, no matter how, all languages do distinguish motion from state change in one way or other: In English, both a verb ‘satellite’ and a preposition are generally required to express the path: John ran out (satellite) of (preposition) the house. But in Atsugewi there is a set of satellites (appearing as verb suffixes), used without a preposition,

3.

THE PITFALLS OF GETTING FROM HERE TO THERE

51

which play these roles: for example, suffixes expressing ‘into a liquid’, ‘down into the ground’, or ‘horizontally onto an object above the ground.’ Summarizing, languages vary in which meaning components are characteristically conflated within the verb, and in the surface syntactic or morphological resources for expressing these various meaning components.…Though languages differ from one another, each language apparently is restricted in the choices taken. If so, the learner can depend on the notional conflations a language characteristically exhibits to guide inductions about the meanings of new verbs; and he can depend on the surface reflexes (satellites, prepositions, etc.) of the verbs to determine just how these notions will likely be mapped into individual lexical entries. (Landau & Gleitman, 1985: 148–149)

From Landau and Gleitman’s proposal, the following hypothesis can be derived (to be falsified by the Yukatek facts): motion event constructions have formal properties that distinguish them from other constructions. These differences guide learners to map the motion meaning onto the motion construction, and, more specifically, onto the motion verb that contains the central lexical information in the construction. Specifically: motion event expressions are formally sensitive to the “path” component of the motion event, i.e., the distinction between motion to, from, into, out of, and past a ground, etc. Languages vary in how they signal and distinguish path relations, but learners can always rely on the fact that they do signal and distinguish path relations. 3. THE MORPHOSYNTAX OF YUKATEK MOTION EVENT EXPRESSIONS This section investigates the formal properties of motion verbs and ground-denoting obliques in Yukatek and discusses the implications of the findings for the syntactic bootstrapping hypothesis. The upshot is that there is no formal reflex of path distinctions in Yukatek, contrary to the prediction derived in the previous section from Landau and Gleitman (1985). 3.1. Morphological and Syntactic Properties of Motion Verbs in Yukatek In order to understand how Yukatek grammar treats motion verbs, the basic facts of verb form classes in the language need to be considered. Yukatek verbs are divided into a number of distinct form classes. These distinctions have to do with the realization of aspect-mood marking on the verb. For present purposes, the mechanisms involved may be likened to distinctions among conjugation classes in Latin and Romance languages, or to processes of auxiliary selection in languages like Dutch (Zaenen, 1993), German (Shannon, 1992), and Italian (Van Valin, 1990). The details are of no particular concern here (but see Bohnemeyer, 2002, 2004, for extensive discussion).1 The system of morphosyntactic predicate classes distinguished by these processes is summarized in Figure 3.1.

52

BOHNEMEYER

FIGURE 3.1. Yukatek formal predicate classes.

There is one class of transitive verb stems and there are four classes of intransitives. The labels assigned to the intransitive classes stand for the semantic traits that motivate the classes. “Inactive,” “positional,” and “inchoative” verbs encode state changes, while active verbs express “activities” in the sense of Vendler (1967) and Dowty (1979) (see Bohnemeyer, 2004). The active verb class features typical activity verbs like ‘dance’ and ‘play’, manner-of-motion verbs like ‘roll’ and ‘run’, and ‘emission’ verbs (Levin & Rappaport Hovav, 1995) such as ‘shine’ (light emission), ‘buzz’ (sound emission), and ‘urinate’ (bodily emission). Active verbs constitute a large class in Yukatek. Moreover, the class is open in the sense that it freely accommodates Spanish loan verbs. Only the active, transitive, and (to a lesser extent) inchoative classes have this property. The inactive class includes verbs of physical state change comparable to English be born, “phase verbs” equivalent to English begin and end, and verbs of “inherently directed motion” (Levin & Rappaport Hovav, 1995) resembling English come, go, enter, and exit. There is a closed class of no more than perhaps 100 roots that produce inactive stems without derivation. Positional verbs express nonpermanent spatial properties of objects, animals, and people, including shape (e.g., ‘bulge’), disposition (e.g., ‘be coiled around something’), distribution (e.g., ‘be scattered’), configuration (e.g., ‘be between two things’), posture (e.g., ‘sit’, ‘stand’, ‘lie’), and orientation (e.g., ‘lie face-down’). This class includes 100–150 roots in Yukatek (see Bohnemeyer & Brown, 2007). The members of the last set, inchoative verbs, are all derived from stative predicates (corresponding to English adjectives) and nouns, and express the process of entering the state denoted by the base (e.g., ‘be big’ > ‘grow’). This class is open in the sense that most nouns and stative predicates—both themselves open classes—produce inchoatives, and also in the sense that the inchoative derivation also operates to some extent on stative predicates borrowed from Spanish.

3.

THE PITFALLS OF GETTING FROM HERE TO THERE

53

The evidence presented in Bohnemeyer (2004) in support of the analysis that the inactive, inchoative, and positional verbs express state changes comes from their aspectual behavior. Semantic tests show that these verbs entail a transition between two states, a source state and a result state (e.g., ‘be alive’ and ‘be dead’ in the case of kim ‘die’), such that the event encoded by the verb is completed once the theme or patient enters the result state.2 Active intransitives differ from the three classes of state-change verbs in their argument structure properties. To produce transitive stems, active roots take an “applicative” suffix -t. The semantic effect of this alternation is the addition of an “applied object.” In contrast, state-change roots causativize to produce transitive stems. This alternation is marked by different suffixes in the three subclasses. The semantic effect of this alternation is the addition of a causer argument. Of the five morphological verb classes introduced in Figure 3.1, only the inchoative class does not host verbs that regularly occur in the expression of motion events. The dynamic verb forms of positional roots refer to the process of entering the spatial configuration expressed by the base (e.g., the process of assuming a posture) and only in this sense denote “motion”; in the remainder of this chapter, they are ignored. Transitive verb stems express caused motion, that is, motion events portrayed as caused by a participant different from the moving entity (e.g., putting, inserting, throwing, tossing). What from the point of view of English appears to be the most basic case of a motion scenario, motion of an object or animate being without an external cause, is expressed in Yukatek using intransitive verbs of the active and inactive classes. Table 3.1 lists some active and inactive verbs that frequently figure in the expression of motion events. TABLE 3.1 The Distribution of Motion Verbs in the Active and Inactive Classes Active

Inactive

péek

‘move’

bin

‘go’

sùut

‘turn’

tàal

‘come’

xíimbal

‘walk’

máan

‘pass’

áalkab

‘run’

u’l

‘return’

síit’

‘jump’

lúuk’

‘leave’

balak’

‘roll’

k’uch

‘arrive’

xíiknal

‘flutter, fly’

na’k

‘ascend’

bàab

‘swim’

em

‘descend’

òokot

‘dance’

òok

‘enter’

...

…

hóok’

‘exit’

lúub

‘fall’

líik’

‘rise’

54

BOHNEMEYER

The English glosses in Table 3.1 invite an informed guess to the effect that active verbs occurring in motion event descriptions express “manner of motion” in the sense of Talmy (1985, 2000). In contrast, the inactive verbs in the right column express an aspect of the “path” in Talmy’s parlance: a feature of location change with respect to a ground. Thus, the inactive motion verbs denote “inherently directed motion” (Levin & Rappaport Hovav, 1995). This presumed semantic difference between active and inactive motion verbs is confirmed by a striking semantic asymmetry in the behavior of verbs of the two classes vis-à-vis ground-denoting obliques. Consider the examples in (1):3 1. a. Le=ch’íich’=o’ túun DET=bird=D2

xíiknal

PROG:A.3 fly

y=óok’ol

le=che’=o’

A.3=TOP

DET=tree=D2

‘The bird is flying [i.e., circling!] above the tree.’ b. Le=ch’íich’=o’ h-em DET=bird=D2

PRV-descend(B.3.SG)

u=xíiknal

te=che’=o’.

A.3=fly

LOC:DET=tree=D2

‘The bird flew down from the tree [lit.: it descended from the tree flying].’ c. Le=ch’íich’=o’ h-na’k DET=bird=D2

PRV-ascend(B.3.SG)

u=xíiknal

te=che’=o’

A.3=fly

LOC:DET=tree=D2

‘The bird flew up to the tree [lit.:it ascended to the tree flying].’

When active motion verbs are combined with ground-denoting obliques (1a), the resulting interpretation is not change of location with respect to the ground, but only location of the motion event as a whole. Only inactive motion verbs can express change of location with respect to the ground, assigning to the latter a semantic role such as “source” (as with em ‘descend’ in (1b)), “goal” (as with na’k ‘ascend’ in (1c)), or “via” (cf. Jackendoff, 1983), in the case of máan ‘pass’. There are various ways to combine reference to change of location with reference to manner of motion. In the simplest case, the two verbs appear in independent sentences which are simply coordinated or juxtaposed (‘The bird flew, and it ascended/descended to/from the tree’). The two verbs can also be combined into one sentence. In this case, the active motion verb may appear as a “gerundial” subordinate to the main verb (as in (1b–c), translated ‘it ascended/ descended flying’).4 Because only the inactive motion verbs assign source, goal, or via roles to the ground-denoting oblique, it seems fair to conclude that only they express “inherently directed motion”. These inactive motion verbs are the focus of the rest of this chapter. They are termed change-of-location verbs. The semantics of these verbs is examined in detail in section 5.

3.

THE PITFALLS OF GETTING FROM HERE TO THERE

55

3.2. Implications of the Distribution of Yukatek Motion Verbs across Form Classes In English, verbs may be assumed to form complex lexical entries together with what Talmy (1985, 2000) calls “satellites,” that is, particles such as up, down, in, out. (Note that Yukatek does not have any such satellites, unlike many other Mayan languages.)5 If so, then learners of English have direct morphological evidence to the effect that a given verb has a change-of-location meaning. If, on the other hand, satellites are assumed not to form lexicalized collocations with verb stems (as argued, e.g., in Ruhl, 1989: 163–172), then English does not show a morphological distinction of motion verb classes. This means that, in English, children are not led by any morphological facts to assumptions about the meanings of these verbs. By contrast, Yukatek learners should be biased by the morphological pattern of the change-of-location verbs to assume that the semantics of these verbs is in some respect similar to the semantics of verbs that lexicalize uncaused state changes in the physical domain, such as ‘be born’, ‘die’, or ‘explode’, and that the semantics of the change-of-location verbs is in the same respect dissimilar to the semantics of activity verbs (e.g., equivalents of sing and dance) and transitive verbs denoting caused state changes (e.g., equivalents of make, break, drink, etc.). By the same token, unlike their English-learning peers, Yukatek learners should be biased by the morphological facts of their language to assume a semantic difference between the change-of-location verbs translating ‘come’, ‘go’, ‘enter’, ‘exit’, and so on and the manner-of-motion verbs translating ‘run’, ‘swim’, ‘fly’, and so forth.6 3.3. The Expression of the Referential Ground in Yukatek Motion Event Coding The referential ground is always referred to by an oblique in a Yukatek motion event description. This holds with three exceptions: tàal ‘come’and u’l ‘return’both assign a goal role to the deictic center; this may be expressed by a deictic adverb like here, but usually remains unexpressed. In addition, bin ‘go’ assigns a source role either to the deictic center (in which case it again remains unexpressed) or to a location that cannot be specified in the same clause, but has to be retrieved anaphorically from context.7 Ground-denoting obliques are usually headed by a preposition or relational noun. The most important of the prepositions and relational nouns that appear in this context are listed in Table 3.2. Like other Mayan languages, Yukatek lacks an elaborate set of genuine prepositions (cf. Kaufman, 1990: 78). Aside from ti’ ‘LOC’ and ich(il) ‘in’, all relators listed in Table 3.2 are relational nouns (see Lehmann, 1996, for details). The generic preposition ti’, somewhat elusively glossed ‘LOC’ in the examples, is a semantically almost empty adverbializer which does not distinguish between a

56

BOHNEMEYER TABLE 3.2 Spatial Relators in Yukatek Ground-Denoting Obliques

Form Class

Relators

Meaning in Ground-Denoting Obliques

Prepositions

ti’ ich(il)

‘LOC’ (generic preposition) ‘in(side)’

Relational nouns that may directly head ground-denoting obliques

óok’ol àanal iknal

‘upper side’, ‘on (top of)’, ‘above’ ‘(at the) bottom (of)’, ‘under’ ‘at’, ‘vicinity’

Relational nouns that require the generic preposition ti’ or the relational suffix -il to form ground-denoting obliques

chúumuk nak’ ts’u’ (ba’)pàac h (ak)táan tséel xno’h xts’i’k háal xùul yáam tòoh

‘(at the) center (of)’ ‘belly’, ‘(at) mid-height’ ‘(at the) core (of)’ ‘(at the) back (of)’, ‘behind’, ‘outside’, ‘around’ ‘(in) front (of)’, ‘before’, ‘opposite’ ‘(at the) side (of)’ ‘(to the) right (of)’ ‘(to the) left (of)’ ‘(at/on the) edge (of)’ ‘(at the) end (of)’ ‘interstice’, ‘between’ ‘(in the) direction (of)’

spatial point of reference, a recipient, beneficiary, or experiencer, a purpose, and a number of other readings. The function of ti’simply consists in relating any kind of peripheral participant (with the exception of a comitative or instrumental participant) to the event core expressed by the verbal complex. Ti’may generally be translated as ‘with respect to’. It is easily demonstrated that the operators listed in Table 3.2 do not express path relations. Consider the examples in (2). Both òok ‘enter’ and hóok’ ‘exit’ are equally possible with both ich ‘in’ and ti’‘LOC’. The same holds for the existential predicate yàan employed in (2c) to express stative location. Hence, the preposition is sensitive neither to the source–goal distinction nor even to the dynamicity of the event core (cf. also Goldap, 1992, and Lehmann, 1992). 2. a. Le=kàaro=o’ DET=cart=D2

h-òok

ich/ti’

le=kàaha=o’.

PRV-enter(B.3.SG)

in/LOC

DET=box=D2

‘The cart, it entered [lit.: in] the box.’ (or rather: it entered with respect to the box’s inside)

3.

THE PITFALLS OF GETTING FROM HERE TO THERE b. Le=kàaro=o’ DET=cart=D2

57

h-hóok

ich/ti’

le=kàaha=o’.

PRV-exit(B.3.SG)

in/LOC

DET=box=D2

‘The cart, it exited [lit.: in] the box.’ (or rather: it exited with respect to the box’s inside) c. Le=kàaro=o’ DET=cart=D2

ti=yàan

ich/ti’

le=kàaha=o’.

LOC=EXIST(B.3.SG)

in/LOC

DET=box=D2

‘The cart, it is in the box.’ (or rather: it is located with respect to the box’s inside)

Prepositions or relational nouns heading ground-denoting obliques merely serve to specify a spatial region of the ground, such as the inside of the cardboard box in the examples in (2) if ich(il) is chosen. If for whatever reason no particular region is selected, ti’ takes over, leaving the spatial properties of the ground to inference. If the ground is not referred to by a phrase headed by a preposition or relational noun, but for example by a deictic adverb equivalent to here or there, there is likewise no formal reflex of either the distinction between motion and location or the distinction between different path roles such as source and goal. 3.4. Implications of the Expression of the Referential Ground for the Syntactic Bootstrapping Hypothesis Because the oblique specifying the referential ground in a motion event does not distinguish between stationary location and change of location, and the verb used to express change of location has the same formal properties as an inactive verb expressing change of state in the physical domain,8 there is no morphological difference between (3) and (4) as shown next, and no syntactic difference that could be read off of constituent order. This means that, contrary to what is predicted by the syntactic bootstrapping hypothesis, Yukatek children have no formal clue that would allow them to determine that (3), but not (4), expresses motion. 3. Ts’o’k TERM

uy=òok-ol

ich

le=nah=o’.

A.3=enter-INC

in

DET=house=D2

‘He has entered the house.’ 4. Ts’o’k TERM

u=kim-il

ich

le=nah=o’.

A.3=die-INC

in

DET=house=D2

‘He has died in the house.’

58

BOHNEMEYER

Notice that (3) cannot be understood as locating the entire entering event inside the house, as (4) locates the dying event inside the house. Under this analysis, (3) would not distinguish between entering into the house (where the source state of the entering event is outside the house) and entering a room or compartment inside the house (where the theme is located inside the house at both the source and the target state of the entering event). But native speakers systematically reject the latter type of interpretation. The interpretation of the prepositional phrase ich le naho’ in (3) is by necessity different from the interpretation of the same prepositional phrase in (4). When combined with a verb that lexicalizes change of location, the ground denoted by the oblique is assigned a path role of source, goal, or via. With a verb that does not lexicalize change of location, no such interpretation arises. The interpretation of the ground-denoting oblique strictly depends on the semantics of the verb. Only once Yukatek-learning children have established the change-of-location verbs as a lexical category, based on semantic evidence, can they assign the correct interpretations to utterances of the structure of (3) and (4) and use these in an adult-like manner. 4. THE SEMANTIC BOOTSTRAPPING HYPOTHESIS The point advanced above with respect to Landau and Gleitman’s syntactic bootstrapping approach is that it is not capable of accounting for the acquisition of motion expressions in Yukatek, because the formal clues distinguishing motion from state change that the syntactic bootstrapping proposal relies on do not exist in Yukatek. The evidence to be presented now against the semantic bootstrapping hypothesis is of a different nature. There is no evidence suggesting that children could not learn the morphosyntactic properties of motion event expressions in Yukatek in the way that Pinker (1989) suggests children learning any language would (essentially, by application of universal linking rules to semantic event representations). However, Pinker’s proposal entails that learners do not require evidence of the morphosyntactic treatment of event expressions in the adult language to construct semantic representations of the events they are learning to encode. Pinker (1989) does not claim that semantic representations are language-independent, and he actually stresses the differences between semantic and cognitive representations (in contrast to Pinker, 1984). However, he assumes that semantic differences across languages reduce to variation in idiosyncratic properties among otherwise corresponding lexemes, and to differences in lexicalization patterns as studied by Talmy (1985, 2000). With the aid of “child-friendly” parental input, children should still be able to map their prelinguistic cognitive event representations onto verbs by application of a process of “event-category labeling,” without having to take in extensive evidence from the morphosyntactic properties of the verbs in the adult language: First, there is the innocuous assumption that children’s perceptual and cognitive mechanisms are enough like adults’ (at least in situations in which they interact with their parents) that they construe the world in pretty much the same way that the adults speaking to them do. Second, there is a somewhat stronger assumption:

3.

THE PITFALLS OF GETTING FROM HERE TO THERE

59

that in parent-to-child speech, the parent uses words whose semantic representations correspond closely to the child’s conceptual representation for that situation, so that event-category labeling and analogous processes for other grammatical entities will generally be accurate. (Pinker, 1989: 362)

In the motion domain, this means that, given beneficial input, it should be obvious to both English- and Yukatek-learning children how to form the appropriate semantic representation of a motion scene, so that they can then proceed to structurally encode this representation following linking rules. The cognitive representation of motion that Pinker assumes to feed into the “event-category labeling” process is “a certain schematization of motion whereby a moving object is idealized as a point traversing some trajectory” (1989: 177), which Pinker repre-

FIGURE 3.2. Conceptual representation of motion events according to Pinker (1989:177). Reprinted by permission.

FIGURE 3.3. Semantic representation of manner of motion in English according to Pinker (1989: 182). Reprinted by permission.

FIGURE 3.4. Semantic representation of manner of motion in Spanish according to Pinker (1989: 182). Reprinted by permission.

60

BOHNEMEYER

sents as in Figure 3.2. The predicate GO here stands for the event type of “a thing moving along a path” (1989: 176) and THING stands for the moving entity. “Event-category labeling” then maps this conceptual representation onto language-particular semantic representations. The main crosslinguistic difference among such representations has to do with the integration of manner of motion along the lines of Talmy’s (1985, 2000) lexicalization typology. Spanish-learning children acquire semantic representations like the one for ‘roll’ depicted in Figure 3.4 as the only way to frame manner of motion, whereas English-learning children in addition acquire representations like the one in Figure 3.3, which Pinker considers “created” from the one in Figure 3.4 according to a “lexical rule”. Here, the open brackets represent argument positions to be filled according to the linking rules. The representation in Figure 3.4 is intended to license The ball rolled, which is fine in both English and Spanish, whereas the one in Figure 3.3 is intended to license The ball rolled down the hill, which is not permitted in Spanish. The difference between the framing in Figure 3.3 and the one in Figure 3.4 covers the amount of crosslinguistic variation in motion semantics that Pinker acknowledges, and he contends that children can cope with this variation and still acquire semantic representations of motion events without inspecting the morphosyntactic properties of the verbs and argument structures involved in coding these representations.9 It is this assumption that is to be argued against in the following section. The difference in the semantic construal of motion scenes across English and Yukatek cannot be accounted for by a mere lexical-semantic rule that derives the Yukatek-type representation from the English-type one or vice versa, in the way Pinker assumes Figure 3.3 to be derived from Figure 3.4. In Yukatek, motion is not framed at all as “a moving object (…) traversing some trajectory”, the cognitive representation of motion that Pinker assumes is universally mapped onto semantic representations by “event-category labeling.” Instead, motion is represented as location change with respect to single grounds. It is argued later in this chapter that Yukatek children could not possibly tune in to this Yukatek way of framing motion without examining the morphosyntactic properties of motion event expressions (in particular, the properties of ground-denoting obliques), contrary to the semantic bootstrapping hypothesis. 5. THE SEMANTICS OF YUKATEK MOTION EVENT EXPRESSIONS It has been shown in section 3 that ground-denoting obliques in Yukatek motion clauses do not formally distinguish among distinct path functions such as “source”, “goal”, and “via”. Instead, any ground-denoting obliques can be assigned any of these roles by the change-of-location verbs (whereas other verbs cannot assign path roles at all). This has the consequence that no verb can combine with more than one ground-denoting oblique in a single clause.10 This follows from the fact that every change-of-location verb assigns exactly one path role. Moreover, even if the change-of-location verbs could assign multiple path roles, no mechanism would be

3.

THE PITFALLS OF GETTING FROM HERE TO THERE

61

FIGURE 3.5. First frame of the motion event video stimulus described in (5).

in place to determine which role is assigned to which ground-denoting oblique, because the form of the oblique does not reflect the role assigned to it. And because there are no serial verb constructions in Yukatek that license combinations of multiple change-of-location verbs in single clauses, Yukatek motion clauses only encode location changes with respect to single grounds. Consequently, scenarios of a figure traveling from source to goal have to be distributed across at least two clauses, one encoding departure from the source, the other arrival at the goal. Consider, for example, (5), a description of the motion scenario depicted in Figure 3.5: 5. Ba’l=e’, thing=TOP

be’òora=a’ t-inw=il-ah=e’,

hun-p’éel

now=D1

one-CL.IN

PRV-A.1=see-CMP(B.3.SG)=TOP

chan

áasul

ba’l

k-u=p’áat-al

t-u=xùul

DIM

blue

thing

IMPF-A.3=await\ACAUS-INC

LOC-A.3=end

le=tu’x

h-luk’

le=chan

ba’l

chak=o’,

DET=where

PRV-leave(B.3.SG)

DET=DIM

thing

red(B.3.SG)=D2

k-u=bin

u=balak’=e’,

k-u=ts’o’k-ol=e’,

IMPF-A.3=go

A.3=roll=TOP

IMPF-A.3=end-INC=TOP

k-u=máan

y-iknal

hun-p’éel

chan

ba’l

chak

xan=e’,

IMPF-A.3=pass

A.3–at

one-CL.IN

DIM

thing

red(B.3.SG)

also=D3

k-u=ts’o’k-ol=e’,

k-u=k’uch-ul

y-iknal

le=triàangulo…

IMPF-A.3=end-INC=TOP

IMPF-A.3=arrive-INC

A.3-at

DET=triangle

‘But, this time, I saw a blue thing, it remains at the end where the red thing left, [the red thing] goes rolling, then it passes by a thing which is also red, then it arrives at the triangle…’

Indo-European languages provide the option of a Yukatek-like framing as well, as the English translation of (5) illustrates. However, this hardly seems the most natural way in English to describe the motion event represented in Figure 3.5. More pertinently, in Indo-European languages this construal is merely a borderline case of a kind of construal whereby the figure undergoes incremental location change along the path as the event progresses through time. The path is encoded with its be-

62

BOHNEMEYER

ginning and end points assigned to source- and goal-denoting obliques that may be copresent in the clause (‘The red ball rolled from the square (via the house) to the triangle’). Motion event descriptions of this type entail a “homomorphic” (i.e., many-to-one) mapping from subevents into the “subpaths” traversed by the figure during the subevents (cf. Jackendoff, 1996; Krifka, 1998). This construal is not encoded at all in Yukatek. In Yukatek, motion is represented, not as a homomorphic mapping from the course of the event into the path, but as location change with respect to single grounds. Traversal of the path that connects events of departure, passing, and arrival, and even the occurrence of motion during the corresponding time intervals, is left to implicature. This has some striking consequences for the conditions under which Yukatek “motion” descriptions can be used. Elicitation with a variety of different stimuli (some of which are presented in Bohnemeyer, 1997) has shown that the change-of-location verbs òok ‘enter’, hóok’ ‘exit’, na’k ‘ascend’, em ‘descend’, and máan ‘pass’ do not entail but only implicate motion of the theme argument. In scenarios in which the ground moves instead of the figure, these verbs are still applicable to the event, provided the implicature that the figure moves is explicitly canceled. For example, if a cardboard box is placed upside down over a toy car so that the car ends up inside, it is perfectly acceptable in Yukatek to say ‘The box was moved, and the car entered it’. Here, reference to the motion of the box serves to block the implicature that the car moved. Even when there is no motion involved at all, for example in animations in which a figure “beams” into or out of a spatial configuration, any of a variety of different resultative or perfect forms of the change-of-location verbs can still be used in reference to the configuration. For instance, although an event of the toy car materializing inside the box cannot be referred to as the car ‘entering’ the box, it is perfectly acceptable to say ‘The car has entered the box’ once the beaming event is completed. It has been claimed in section 4 that the difference in semantic representations of motion scenes between Yukatek and English cannot be accounted for by a mere lexical rule that derives one type of representation from the other, in the way that Pinker assumes the English-type representation in Figure 3.3 to be derived from the Spanish-type representation in Figure 3.4. The justification for this claim is that the basic event type of movement along a path captured by the GO predicate in the representations depicted in Figures 3.2–3.4 is not instantiated in semantic representations of motion in Yukatek.11 Let us assume now, with Pinker, that Yukatek and English children bring to the task of language acquisition the same prelinguistic cognitive representations of motion “whereby a moving object is idealized as a point traversing some trajectory” (Pinker, 1989: 177). Can Yukatek children derive semantic representations from these conceptual representations that license the relevant argument structure properties of change-of-location verbs (in particular, the fact that they take no more than one ground-denoting oblique, expressing location change with respect to that single ground), merely by “event-category labeling” relying on beneficial input? Suppose a child sees the scenario in Figure 3.5 and then hears the description in (5).

3.

THE PITFALLS OF GETTING FROM HERE TO THERE

63

Would that input be sufficient to prevent Yukatek children from deriving English-style lexical-semantic representations for Yukatek verbs? Certainly not. Nothing in (5) prevents Yukatek children from assuming that change-of-location verbs could occur with multiple ground-denoting obliques the way that English motion verbs do, even if the change-of-location verbs in (5) happen to occur only with single ground-denoting obliques. And if Pinker is correct in assuming that Yukatek children derive their semantic representations from conceptual representations of “a moving object (…) traversing some trajectory”, then Yukatek children should expect that the path of a motion event can be mapped onto a series of ground-denoting obliques within the clause denoting the event. Of course, children’s predictions become much more accurate once their database includes information about the frequency with which motion verbs occur with multiple ground-denoting phrases (high in English, zero in Yukatek)—but this information is assumed unnecessary for learning semantic representations that license the argument structure properties of verbs according to the semantic bootstrapping hypothesis. But in order to predict adult-like semantic representations that would not clash with the uses discussed earlier, in which the figure does not actually move, Yukatek children clearly have to perform an even more detailed analysis of the ground-denoting obliques with which change-of-location verbs occur.12 In particular, they have to take on board the fact that Yukatek ground-denoting obliques show no formal reflex of path or locative roles. Given this information, they can conclude that path relations are exclusively expressed in verbs in Yukatek, and on this basis they can infer the correct semantic analysis of change-of-location verbs. 6. THE DIALECTIC OF SEMANTIC AND SYNTACTIC LEARNING Contrary to what is predicted by the syntactic bootstrapping hypothesis, Yukatek children cannot learn the basic semantic difference between motion expressions and descriptions of physical state changes relying on formal clues, because such formal clues—morphosyntactic reflexes of motion path roles outside the verb— are lacking in Yukatek. On the contrary, to determine whether a verb assigns a path role to a ground-denoting oblique, as a change-of-location verb does, or a stationary locative role, or whether it is merely compatible with a locative oblique, as any other verb is (including, of course, manner-of-motion verbs, which are also used in reference to motion events!), Yukatek learners have to have access to the semantics of the verb first. However, in contradiction to what is assumed by the semantic bootstrapping hypothesis, Yukatek children cannot derive semantic representations of change-of-location verbs that license the correct argument structure properties of these verbs solely by event-category labeling of preverbal cognitive representations with the aid of “child-friendly” parental input. Yukatek-learning children could not derive

64

BOHNEMEYER

the appropriate semantic representations, namely location change with respect to single grounds, from this input—especially if, as Pinker assumes, they start from the same cognitive representations of continuous locomotion along a path as their English peers do—unless they take in evidence from the fact that change-of-location verbs only ever occur with single ground-denoting obliques, and that these obliques do not formally reflect the path roles assigned to them. It would appear, then, that Yukatek children need information about the argument structure properties of change-of-location verbs to determine their semantics, and information about the semantics of these verbs to determine their argument structures. This suggests a learning mechanism that offers a synthesis of syntactic and semantic bootstrapping, by exploiting both the constraining power of semantic representations on syntactic ones and that of syntactic representations on semantic ones. However, in order to avoid circularity, syntactic learning and semantic learning have to have distinct roles in the mechanism. One such mechanism is “lexical reconciliation,” proposed by Grimshaw (1994). The reconciliation process is summed up in a nutshell in (6): 6. Lexical reconciliation (based on Grimshaw, 1994: 423) a.

The learner hears a sentence, observes an event she assumes the sentence to refer to, isolates the verb in the sentence, and forms candidate semantic representations of the verb that fit both the observed event and whatever she knows about the sentence’s lexical content. b. On the basis of the candidate semantic representations and the universal and/or language-specific linking rules she assumes, the learner predicts the verb’s argument structure. The predicted argument structures are then checked against the observed syntactic properties of the sentence, and candidate semantic representations that predict argument structures not fitting the observed facts are rejected.

Reconciliation preserves the full predictive power of semantic bootstrapping, but it does not rely on mere “concept labeling” for the formation of the semantic input. The input representations for the phase of reconciliation that corresponds to semantic bootstrapping are derived from potentially fairly intricate mappings between perceived utterances and perceived events. The predictive power of syntactic bootstrapping is exploited by a checking mechanism whose function it is to filter out candidate input representations that are inconsistent with the perceived syntactic properties of the target verb, given the learner’s assumptions about linking. Learning semantic representations that are not discriminated by syntactic properties that linking is sensitive to—such as the distinction between motion and state change in Yukatek—is no longer a problem, because the semantic input to verb learning comes from an independent source. Grimshaw’s lexical reconciliation mechanism avoids the pitfalls syntactic bootstrapping runs into when applied to the Yukatek data. But to the pitfalls of semantic bootstrapping it offers only partial remedy. Under reconciliation, Yukatek

3.

THE PITFALLS OF GETTING FROM HERE TO THERE

65

children are no longer burdened with initial interpretations of change-of-location verbs exclusively as labels for conceptual representations of translational motion, such as those in Figures 3.3 and 3.4, which children learning English and Spanish presumably start out with. But there is no reason to assume that such representations are not among the candidate input representations the learner attempts to map to the verb’s argument structure, and it is not clear that the observed syntactic facts of the verb and the learner’s knowledge of linking rules are sufficient to reject them. The problem is that even though the child observes change-of-location verbs only with single ground-denoting obliques (say, expressions of the source), no single instance of this observation rules out the interpretation that another ground (e.g., the goal) is merely left unexpressed even though present in the semantic representation. To realize that this is not the case, the learner might need to determine that the form of the ground-denoting oblique does not discriminate a path function like source or goal, unlike in English or Spanish. But this realization can apparently only come from a generalization across uses of the same ground-denoting oblique in different path roles. From this the learner might conclude that Yukatek clauses cannot include more than one ground-denoting oblique as a matter of principle. This might eventually lead her to discard translational-motion representations for the meanings of the change-of-location verbs. So it appears that in order to arrive at the correct language-particular semantic representations, semantic and syntactic learning have to proceed in tandem to an extent that exceeds even what Grimshaw assumes to be part of lexical reconciliation. 7. CONCLUSIONS It could be argued on the basis of the facts of motion event expressions in Yukatek that both bootstrapping hypotheses are wrong. But a more appropriate conclusion seems to be that both proposals are in fact right. If there is a sense in which the two hypotheses are falsified by the Yukatek data, it is in the assumption that they are mutually exclusive. This is a conclusion that has been arrived at before, on the basis of other sets of data, for example, by Grimshaw (1994). However, the amount of crosslinguistic variation in both semantic framing of events and predicate argument structures uncovered in the present study suggests that semantic learning and morphosyntactic learning proceed in tandem to an extent that goes even beyond what Grimshaw envisioned. NOTES 1

The Yukatek verb classes have also been studied intensively by Lehmann (1993), Lucy (1994), and Krämer and Wunderlich (1999). As far as the issues dealt with here are concerned, these authors have reached the same conclusions as Bohnemeyer (2004).

66 2

BOHNEMEYER

An exception are “degree achievement” verbs (Dowty, 1979) such as the inchoative nohochtal ‘grow’. Such verbs do not entail a definite end state, unless the extent to which the theme or patient undergoes the change is specified (e.g., ‘grow five inches’). 3 The orthographic representation in this chapter is morphemic rather than morphophonemic. The orthography applied is based on Lehmann (1996). In the interlinear morpheme glosses, ‘-’ is used for for affixes and ‘=’ for clitics. Abbreviations in the glosses include the following: 2, second person; 3, third person; A, set-A (‘ergative’/possessor) clitics; ACAUS, anticausative derivation; B, set-B (‘absolutive’) suffixes; CL, (numeral/possessive) classifier; CMP, completive aspect; D1, proximal deixis; D2, distal/anaphoric deixis; DET, definite determiner; DIM, diminutive (particle); EXIST, existential/locative/possessive predicate; IMPF, imperfective aspect; IN, inanimate (classifier); INC, incompletive aspect; LOC, generic preposition; PROG, progressive aspect; PRV, perfective aspect; REL, relational derivation (nouns); SG, singular; TERM, terminative aspect-mood marker; TOP, topic marker. 4 Alternatively, the manner verb may be fronted in a special manner-focus construction; cf. Bohnemeyer & Stolz (2006) for details. 5 Many Mayan languages have so-called “directional” morphemes grammaticalized out of motion verbs; cf. Kaufman (1990: 82–83), and Zavala (1993). 6 These points have been stressed by Lucy (1994). 7 With most change-of-location verbs, the ground is frequently not specified at all in the clause that contains the verb, but either retrieved from context by inference or simply left unspecified. In five “Frog Story” narratives collected by Christel Stolz (cf. Bohnemeyer & Stolz, 2006), I counted a total of 158 inactive change-of-location verbs. Of these, only one-third (52) were accompanied by ground-denoting obliques. In 25 cases (16%), the verb appeared in a “motioncum-purpose” construction, in which instead of a ground, a “goal event” is specified (as in to go shopping). And in 51% of all instances, neither a ground nor a goal event was specified. The only member of the set of inactive change-of-location verbs that rarely occurs without a ground-denoting oblique is na’k ‘ascend’. 8 There is in fact one difference: the three most frequent inactive motion verbs, bin ‘go’, tàal ‘come’, and máan ‘pass’, are all irregularly zero-marked in one aspect-mood category which on all other state-change verbs is overtly marked. But I do not see how this could help a child determine that these verbs express change of location. 9 As far as I can see, Pinker does not explain how English-learning children acquire the representation in Figure 3.3 while Spanish-learning children do not. But for the sake of the argument, I will assume that Pinker is correct in his supposition that these lexical-semantic representations can be acquired without evidence from argument structure properties. 10 There is one exception: direction obliques headed by tu tòohil ‘in the direction of’ (see Table 3.2), that is, ‘toward’ or ‘away from’, can be combined with obliques encoding source, goal, or via roles. But because direction specifications do not entail change of location (cf. Jackendoff, 1983: 165), their presence in a clause does not affect the location change information entailed by the clause.

3.

THE PITFALLS OF GETTING FROM HERE TO THERE

67

11

Lucy (1994: 641) points to the framing of motion as state change in Yukatek, but holds that continuous-locomotion readings can still be obtained with the progressive aspect. But this misses the point that the path from source to goal cannot be encoded. Progressives of motion clauses refer to pre-states of departure, arrival, or passing events, but even progressives cannot portray a moving entity as being en route from source to goal in Yukatek. 12 It may be argued that Yukatek children can predict the semantics of change-of-location verbs on the basis of observations of nonmotion uses. However, such uses are highly infrequent. A Yukatek child may never observe a single instance of such usage until age 4 or even much later, and initial evidence suggests that children’s use of change-of-location verbs is already adult-like at age 4. REFERENCES Bohnemeyer, J. (1997). Yucatec Mayan lexicalization patterns in time and space. In M. Biemans & J. van der Weijer (Eds.), Proceedings of the CLS opening academic year 1997/1998 (pp. 75–105). Tilburg, The Netherlands: Center for Language Studies. Bohnemeyer, J. (2002). The grammar of time reference in Yukatek Maya. Munich, Germany: Lincom Europa. Bohnemeyer, J. (2004). Split intransitivity, linking, and lexical representation: The case of Yukatek Maya. Linguistics, 42, 67–107. Bohnemeyer, J., & Brown, P. (2007). Standing divided: Dispositional verbs and locative predications in two Mayan languages of Mexico. Linguistics, 45(5/6), 1105–1151. Bohnemeyer, J., & Stolz, C. (2006). Spatial reference in Yukatek: A survey. In S. C. Levinson & D. P. Wilkins (Eds.), Grammars of space (pp. 273–310). Cambridge: Cambridge University Press. Dowty, D. R. (1979). Word meaning and Montague grammar. Dordrecht, The Netherlands: D. Reidel. Goldap, C. (1992). Morphology and semantics of Yucatec space relators. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung, 45, 612–625. Grimshaw, J. (1994). Lexical reconciliation. Lingua, 92, 411–430. Jackendoff, R. (1983). Semantics and cognition. Cambridge, MA: MIT Press. Jackendoff, R. (1996). The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language and Linguistic Theory, 14, 305–354. Kaufman, T. (1990). Algunos rasgos estructurales de los idiomas Mayances. In N. C. England & S. R. Elliot (Eds.), Lecturas sobre la lingüística maya (pp. 59–114). Antigua, Guatemala: Centro de Investigaciones Regionales de Mesoamérica. Krämer, M., & Wunderlich, D. (1999). Transitivity alternations in Yucatec, and the correlation between aspect and argument roles. Linguistics, 37, 431–480. Krifka, M. (1998). The origins of telicity. In S. Rothstein (Ed.), Events and grammar (pp. 197–235). Dordrecht, The Netherlands: Kluwer. Landau, B., & Gleitman, L. R. (1985). Language and experience. Cambridge, MA: Harvard University Press. Lehmann, C. (1992). Yukatekische lokale Relatoren in typologischer Perspektive. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung, 45, 626–641. Lehmann, C. (1993). Predicate classes in Yucatec Maya. Función, 13–14, 195–272. Lehmann, C. (1996). Possession in Yucatec Maya. Munich, Germany: Lincom Europa. Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax–lexical semantics interface. Cambridge, MA: MIT Press.

68

BOHNEMEYER

Lucy, J. A. (1994). The role of semantic value in lexical comparison: Motion and position roots in Yucatec Maya. Linguistics, 32, 623–656. Pinker, S. (1984). Language learnability and language development. Cambridge, MA: Harvard University Press. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Ruhl, C. (1989). On monosemy: A study in linguistic semantics. Albany: State University of New York Press. Shannon, T. (1992). Split intransitivity in German and Dutch: Semantic and pragmatic parameters. In R. Lippi-Green (Ed.), Recent developments in Germanic linguistics (pp. 97–113). Amsterdam, The Netherlands: John Benjamins. Talmy, L. (1985). Lexicalization patterns. In T. Shopen (Ed.), Language typology and syntactic description, Vol. 3: Grammatical categories and the lexicon (pp. 57–149) Cambridge: Cambridge University Press. Talmy, L. (2000). Toward a cognitive semantics, Vol. 1: Concept structuring systems. Cambridge, MA: MIT Press. Van Valin, R. D., Jr. (1990). Semantic parameters of split intransitivity. Language, 66, 221–260. Vendler, Z. (1967). Verbs and times. In Z. Vendler (Ed.), Linguistics in philosophy (pp. 97–121). Ithaca, NY: Cornell University Press. Zaenen, A. (1993). Unaccusativity in Dutch: Integrating syntax and lexical semantics. In J. Pustejovsky (Ed.), Semantics and the lexicon (pp. 129–161). Dordrecht, The Netherlands: Kluwer. Zavala, R. (1993). Clause integration with verbs of motion in Mayan languages. Unpublished MA thesis, University of Oregon, Eugene.

CHAPTER 4

Making Sense of Complex Verbs: On the Semantics and Argument Structure of Closed-Class Verbs and Coverbs in Jaminjung Eva Schultze-Berndt University of Manchester

1. INTRODUCTION It is widely assumed that verb learning should be more difficult than noun learning, because the mapping between observable situations in the real world and verbs is—supposedly—less straightforward than the mapping between objects in the real world and nouns (Gentner, 1978). This assumption has lead to the proposal, known as the “syntactic bootstrapping hypothesis,” that children use the syntactic context that a verb occurs in—specifically, its argument structure—to determine aspects of its meaning (e.g., Fisher, Hall, Rakowitz, & Gleitman, 1994; Gleitman, 1990; Landau & Gleitman, 1985; Naigles & Hoff-Ginsberg, 1998). However, this strategy of verb learning presupposes that verbal predicates are, by and large, single lexical items. When this is the case, it is possible to draw direct conclusions about the meaning of a predicate on the basis of the syntactic frame or frames it occurs in. In this chapter, the assumptions underlying the syntactic bootstrapping hypothesis are tested against a language with a closed class of verbs. Only 35 verbs can serve as simple predicates, and consequently, the large majority of verbal predicates are complex. They are formed with one of the closed-class verbs and a member of a second lexical category, termed coverbs here, which is distinct from both verbs and nominals. Complex verbal predicates of this type constitute an areal feature found in several dozen, mostly unrelated, Northern Australian Aboriginal languages.1 The language under investigation here is Jaminjung, which belongs to one of the Non-Pama69

70

SCHULTZE-BERNDT

Nyungan language families, Jaminjungan or “Yirram.” Jaminjung is a severely endangered language, now spoken by around 100 people who are spread over a large area in the Victoria River District in Northwest Australia. All data are taken from my own fieldwork, extending over 26 months in total. Because Jaminjung is no longer acquired by children, this chapter does not include actual observations on acquisition. Rather, it suggests a plausible strategy for verb learning that would accommodate the particular lexical and grammatical structure of Jaminjung. The chapter is structured as follows. Section 2 provides the basic facts about the lexical classes involved in complex predicate formation in Jaminjung, and a brief description of the main morphosyntactic features of this language. Section 3 states the methods and assumptions on which the further argumentation is based. The framework adopted for the description of argument structure in Jaminjung is construction grammar. This allows for a distinction between semantic participants and morphosyntactic arguments that will be of particular relevance for the description of argument sharing in complex predicates. Language-specific definitions of core arguments and central participants are also proposed in this section. Sections 4 and 5 illustrate the approach proposed here for a small subset of simple and complex predicates. In section 4, the meaning and argument structure of one of the closedclass verbs is discussed in some detail. It is demonstrated that a verb that at first sight corresponds to a verb of transfer like English put does not entail motion, but only change of configuration, and that this fact cannot be deduced from the argument structure of this verb. Section 5 outlines a strategy for identifying the meaning and argument structure of coverbs and of establishing classes of coverbs. Section 6 presents some conclusions, discussing the implications of the findings for models of language acquisition. 2. SIMPLE AND COMPLEX VERBS IN JAMINJUNG Jaminjung is a language with fewer than 35 verbs.2 Thus, lexical items that can take verbal inflections and that may constitute a simple predicate in a finite clause form a closed class in this language. Verbs can be combined with members of an open class of words that do not inflect, and that are formally distinct from both verbs and nominals. For lack of a better term, these are referred to as ‘coverbs’ (other terms used in the literature on Northern Australian languages with a similar structure include ‘preverb’, ‘verbal particle’, ‘verb base’, and ‘uninflected verb’). Coverbs cannot be used as simple predicates in finite clauses; they mainly occur as part of complex verbs consisting of a verb and one or sometimes two coverbs (see Schultze-Berndt, 2000, for details). The structure of simple and complex predicates is illustrated in examples (1) to (5). The verb in each of these examples is based on the root -(ng)angu ~ -angga, which has been given the tentative gloss ‘GET/HANDLE’.3 It can easily be identified as the verb because it takes obligatory pronominal prefixes and inflects for tense, aspect, and mood, either by affixation or by stem suppletion.

4.

MAKING SENSE OF COMPLEX VERBS

71

In (1) and (2), -(ng)angu functions as a simple predicate. The two examples illustrate the semantically general nature of the closed-class verbs in Jaminjung. In (1), the verb can be translated as ‘get’, ‘catch’, ‘obtain’; this is the default reading for -(ng)angu as a simple verb. However, both the context and the nature of the Undergoer argument make it clear that (2) should not be interpreted as ‘get my body’, but rather as ‘handle my body in the appropriate way considering that I am sick’, which is ‘rubbing’.4 1. lawarra rifle.fish(ABS )

jungulug

gan-angu

Nalyarri-ni

one(ABS)

3sgA:3sgU-GET/HANDLE.PST

[name]-ERG

‘Nalyarri caught one rifle fish’ 2. ban-ngangu IMP:2sg:1sg-GET/HANDLE

ngarrgina

mayi

1sg:POSS

body(ABS)

‘(maybe I am sick), rub my body’

One may therefore conclude that -(ng)angu ‘GET/HANDLE’ does not entail ‘caused motion towards the Actor’ or ‘bringing into possession’, but only ‘manipulation’ (see Schultze-Berndt, 2000: 293–308, for a more detailed argument). This general meaning is consistent with the use of -(ng)angu as part of a large number of complex predicates, including those shown in (3) to (5). Here the verb is combined with different coverbs, durd ‘hold, grasp (of a single entity)’, ning ‘break off’, and dirrg ‘be tied up’. Although most coverbs, like those illustrated here, semantically correspond to verbs in many other languages, they do not take verbal inflections, and have to combine with a semantically compatible inflecting verb. 3. ngarrgina-ni 1sg:POSS-ERG

jalig

ganurr-angga-m

durd

child

3sgA:3pl U-GET/HANDLE-PRS

hold.one

‘my son picks them up (one by one)’ (referring to people in need of assistance) 4. mununggu string(ABS )

ngunggina

ning=biya

nganj-angu

2sg: POSS

break.off=now

2sgA:3sgU-GET/HANDLE.PST

‘you broke your fishing line’ (by pulling it) 5. dirrg be.tied.up

gan-angga-m 3sgA:3sgU-GET/HANDLE-PRS

‘she ties it up’ (e.g., a bundle of clothes, or someone’s hair)

The examples also illustrate some further morphosyntactic features of Jaminjung. In a complex predicate, the coverb usually immediately precedes the verb, as in (4) and (5), but this order may be reversed, as in (3). Clitics, as in (4), and

72

SCHULTZE-BERNDT

very occasionally other constituents, may intervene between the two components. Complex predicates in Jaminjung therefore do not constitute a single word, but are of a phrasal nature. Arguments are encoded as obligatory pronominal prefixes on the verb. Two different prefix paradigms distinguish intransitive verbs (which take a prefix representing a single argument) from transitive verbs like -(ng)angu ‘GET/HANDLE’ (which take prefixes representing the Actor and the Undergoer). In addition, arguments may be encoded as case-marked noun phrases; as (3) shows, case has to be marked only once in the noun phrase. Case marking follows an ergativeabsolutive pattern; the absolutive is unmarked, and the ergative is marked by a suffix. However, Jaminjung is not a syntactically ergative language; there is no evidence for either an absolutive syntactic pivot or a syntactic pivot of the subject type. As these examples also show, the word order of predicates and arguments is free, and lexical arguments can be freely omitted. For example, in (2) and (4) only the absolutive argument is lexically expressed, in (3) only the ergative-marked argument is lexically expressed, and in (5) no lexical arguments are present. The main questions to be addressed in the remainder of this chapter are the following. What would be plausible strategies for a child to adopt in order to learn the meaning of the components of complex predicates in Jaminjung? In particular, how can the contribution of the coverbs be identified, because they almost exclusively occur as part of complex predicates? For example, it is not immediately obvious that the coverb dirrg in (5) is really a monovalent predicate describing the state of being tied up (as suggested by the gloss), rather than a bivalent predicate describing the activity of tying something up. Both meanings would be compatible with the semantics of the verb -(ng)angu proposed earlier, as well as with the interpretation of the whole utterance. To make this point clearer, some further coverb–verb combinations are listed in (6) and (7); here, only a gloss for the complex verb as a whole is provided. Again, it is not immediately obvious which component of the complex verb makes which semantic contribution to the interpretation of the whole. In particular, it is not clear what the meaning of the verbs -arra in (6) and -yu(nggu) in (7) could be. 6. a. b. c. d.

dirrg gan-arra-ny ‘tie up (in a location)’ (e.g., a dog) jubard gan-arra-ny ‘shut in (tr.), enclose in a location’ wirriny gan-arra-ny ‘turn over (tr.)’ dalag gan-arra-ny ‘send away, set free’

7. a. b. c. d.

bawa gani-yu ‘shout’ nunaj gani-yu ‘nod’ butharl gani-yu ‘be sad’ burrngburrng gani-yu ‘boil, bubble’

At this point, one might be tempted to say that complex predicates in Jaminjung can only be given an interpretation as a whole, and that the closed-class verbs, when

4.

MAKING SENSE OF COMPLEX VERBS

73

combined with a coverb, are more or less meaningless carriers of verbal inflections (as has indeed been proposed in many descriptions of languages of the area). One therefore would have to simply learn the inflecting verb appropriate for use with each coverb. However, although there is no doubt that complex predicates are lexicalized to a large degree (see section 3), there is quite a lot of evidence that the majority of complex predicates are not just formally complex, but also semantically compositional, in the way already proposed for the predicates in (3) to (5). The main evidence for compositionality is that most coverbs may combine with several different verbs, with systematic contrasts in meaning. For example, with the coverb dirrg ‘be tied up’, the use of the verb -(ng)angu, as in (5) shown earlier, leads to the interpretation that something is tied up ‘with itself’, like a bundle of clothes or one’s hair; this is consistent with the verb’s meaning of ‘manipulate, handle’. With the same coverb, the verb -arra, as in (6a), yields the interpretation that something is tied to a location, like a boat or a dog (on the semantics of -arra see section 4). Another set of verbs contrasting with a single coverb is provided in (8). The different transitive verbs combined with the coverb ning ‘break off’ in (8a) to (8c) express differences in the manner of causation (compare also (4)), while the intransitive verb -(i)jga in (8d) signals absence of an external cause (see also (18) in section 5). 8. a. b. c. d.

ning gani-ma ‘break something off by hitting it; kill (an animal)’ ning gana ‘chop off, kill by hitting with an edge’ ning gani-wa ‘bite off’ ning ga-jga-ny ‘break off (intr.), stop’

Further evidence for the compositionality of complex predicates comes from the observation that speakers use the closed-class verbs creatively in combination with verbs from Kriol (the English-based creole language spoken in the area), which are borrowed into Jaminjung as coverbs. Examples of loans combining with the verb -(ng)angu ‘GET/HANDLE’ are wajim ‘wash something’, bayim ‘buy something’, braitenim ‘frighten someone’, and tijim ‘tease someone’. The choice of a given verb with a loan is clearly based on semantic compatibility. It therefore makes sense to assume that a speaker of Jaminjung masters the meaning of closed-class verbs and coverbs as individual lexical items, not just that of complex verbs as a unit. For the language learner, this results in a seemingly paradoxical situation: because there are only around 30 simple verbs, in a large percentage of clauses the event is expressed by a complex verb. Here a learner has to figure out which component of the complex verb corresponds semantically to which component of the event. However, in order to figure out the contribution of either one of the elements, the learner already has to know the meaning of the other. 3. BASIC ASSUMPTIONS Before investigating more closely a possible strategy of verb (and coverb) learning in Jaminjung, I briefly outline the basic assumptions underlying the argumentation.

74

SCHULTZE-BERNDT

First, the view of the lexicon adopted here deviates from the mainstream view according to which the grammar describes all expressions that can be derived by general rules, while the lexicon is the repository of everything that is idiomatic in a language, that is, unanalyzable morphemes and noncompositional complex expressions. Complex predicates (including light verb constructions and the particle verbs of English and other European languages) have proved a notorious problem for theoretical frameworks that posit a strict division between a lexicon as a “list” and syntax as “rules”; see, e.g., the discussions in Mohanan (1994: 234–236), Goldberg (1996), Ackerman and LeSourd (1997), and Hampe (1997). In the alternative view adopted here, following, for example, Pawley and Syder (1983), Langacker (1987: 41ff.), and Goldberg (1996), the lexicon comprises all expressions that are conventionalized in a language. This is true even when these expressions are semantically transparent and fully sanctioned by the grammatical knowledge of any speaker of this language. In this view, there is no contradiction in saying that an expression is remembered as a fixed, conventionalized expression (“stored in the lexicon”), while at the same time the patterns (or rules, if one prefers) on which the expression is built are available for productive, creative use (“part of the grammar”). Second, monosemy, that is, isomorphism between form and meaning (see, e.g., Haiman, 1980; Kirsner, 1985; Ruhl, 1989; Taylor, 1990), is adopted as a heuristic guideline for semantic analysis (although not as an absolute principle). The method resulting from adopting a monosemic bias was stated by Ruhl (1989: 234) as follows: “Assume that any meaning that is not present in all contexts of a word is not part of the word’s inherent meaning.” In other words, the meaning—comprising only the lexical semantic invariants—and the interpretation of a linguistic expression have to be carefully distinguished. The interpretation is arrived at by taking into consideration the meaning of all expressions in the context and the meaning of the construction itself, as well as by applying pragmatic principles. This leads to the third assumption made here: syntactic constructions are meaningful signs in their own right. This approach is based on structuralist works (e.g., Bloomfield, 1933/1970; Hockett, 1958; Frei, 1962) and more recent works in the frameworks of cognitive grammar and construction grammar, including Langacker (1987, 1990), Zwicky (1987), Fillmore (1988), Fillmore, Kay, and O’Connor (1988), Taylor (1989), Goldberg (1995), Michaelis and Lambrecht (1996), and Kay and Fillmore (1999). The description of argument structure adopted here follows the approach of Goldberg (1995). Participants of predicates on the semantic level (i.e., those roles inherent in the lexical semantics of a relational predicate) are distinguished from argument slots of a construction on the morphosyntactic level. Participants of predicates (here, of verbs and coverbs) can be linked to the argument slots of a construction if they are semantically compatible with the constructional argument role (cf. Goldberg, 1995: 50). This approach also lends itself to the representation of argument sharing in cases where two or more relational lexemes constitute a

4.

MAKING SENSE OF COMPLEX VERBS

75

complex predicate: two or more semantic participants can be mapped directly onto a single constructional argument role. Participants that share the same argument slot have to be semantically compatible with one another, as well as with the argument role (see section 5). For Jaminjung, both bound pronominals and case-marked noun phrases can be regarded as constituting argument structure constructions in their own right, with somewhat different but overlapping functions. For the purposes of the description of argument sharing in Jaminjung, only the central participants of verbs and coverbs (those that are “profiled,” in Goldberg’s terminology) are considered. Central participants can be defined as those participants that are expressed as core arguments across all constructions that a given verb or coverb occurs in. This is not to deny that verbs and coverbs may have other participants that are inherent in their meaning, but are not expressed as core arguments. Core arguments are defined on a language-specific, purely formal basis. For Jaminjung, all bound pronominals, and in addition all absolutive noun phrases (whether or not cross-referenced on the verb), can be considered as core arguments; for a justification, see Schultze-Berndt (2000: 147–163). 4. MAKING SENSE OF CLOSED-CLASS VERBS As already described in section 2, the majority of verbal predicate types in Jaminjung are complex. However, the 35 verbs are frequently used as simple predicates; in fact, these account for approximately 40% of all tokens of verbal predicates in actual texts. In order to disentangle the contribution of each of the components of Jaminjung complex verbs, it therefore seems like a reasonable strategy to start from the closed-class verbs. According to the syntactic bootstrapping hypothesis, the morphosyntactic frame or frames that a verb occurs in should help us to determine its meaning. For example, a verb of transfer should usually occur with three arguments, representing the participants traditionally labeled ‘agent’, ‘theme’, and ‘location’ (cf., e.g., Landau & Gleitman, 1985: 126ff.). Jaminjung only has a single verb that, at least at first sight, meets the description of a verb of transfer. This is the verb -arra, glossed as ‘PUT’. This verb is formally transitive, in that it takes a transitive pronominal prefix marking Actor and Undergoer. According to the definitions of central participants and core arguments provided in section 3, the ‘agent’ and the ‘theme’ participants of -arra are therefore always expressed as core arguments, and can be considered central participants, inherent in the meaning of this verb. As for the ‘location’ participant that we might expect the verb to have, it is not expressed as a core argument, but only by a noun phrase marked with a peripheral case (locative or allative). Moreover, a location argument is usually not expressed at all. A text count in a corpus of around 4,000 clauses, including narratives and conversational data, revealed that in more than 75% of the clauses containing this verb in its basic sense, no location was expressed, that is, the clauses were of the

76

SCHULTZE-BERNDT

type illustrated in (9a). In only 25% of the uses of this verb was the location expressed, for example as a locative-marked noun phrase, as in (9b). 9. a. gugu water(ABS )

nga-w-arra=biyang,

ba-wurr-ijga

1sgA:3sgU-POT-PUT=now

IMP -2pl- GO

‘I will turn on the water now, go away’ (lit.: ‘I will put water (here) now’) (the speaker is about to turn on sprinklers in a park, asking people to move) b. gugu water(ABS )

nga-w-arra

guyug-gi

1sgA:3sgU-POT-PUT

fire-LOC

‘I will put water on the fire (to boil)’

The argument structure of -arra ‘PUT’ therefore differs from that of a trivalent verb like -ngarna ‘GIVE’, illustrated in (10). This verb also takes a transitive pronominal prefix, which (usually) represents the ‘giver’ and the ‘recipient’ participants. Because lexical arguments are not obligatory, the third participant, the ‘theme’, is not obligatorily expressed. However, if lexically represented, it is linked to a noun phrase in absolutive case (thanthu marlayi in (10)), rather than a noun phrase marked with a spatial case. According to the definition of core arguments provided in section 3, the ‘theme’ is thus also represented as a further, third core argument (a ‘secondary object’ in the sense of Dryer, 1986). 10. ngayug 1sg(ABS )

bun-ngarna-ny

thanthu

marlayi

3plA:1sgU-GIVE-PST

DEM ( ABS )

woman(ABS )

‘me, they gave me that woman’ (as a wife)

Although for -ngarna ‘GIVE’, the basic meaning and participant roles can therefore be inferred from its argument structure, this is doubtful for -arra ‘PUT’. A learner of Jaminjung (a child, or a linguist) therefore has to take other evidence into consideration in order to arrive at a lexical semantic representation of this verb. As it turns out, -arra ‘PUT’ does indeed have a ‘location’ participant as part of its meaning. However, it also turns out that it is not, strictly speaking, a verb of transfer, because it does not entail caused change of location of a theme. Rather, -arra only entails that a theme is caused to change its locative relation with respect to a location. This somewhat cumbersome paraphrase is necessary to capture the observation that the theme can undergo a change of locative relation without moving itself. For example, -arra ‘PUT’ can be used to describe the state of affairs where someone causes an entity to be located at her back by turning her back to it, as in (11a). Of course, much more frequently a change of locative relation is achieved by causing the theme to move, as in (11b).

4.

MAKING SENSE OF COMPLEX VERBS 11. a. langiny wood(ABS )

77

ngagaj-gi

gan-arra-m

back-LOC

3sgA:3sgU-PUT-PRS

‘he has his back turned to the tree’ (lit.: ‘he puts the tree at his back’) (referring to the spatial configuration of a toy man and a toy tree) b. langiny wood(ABS )

ba-rra

nawij-gi

IMP (2sg A :3sg U )- PUT

neck-LOC

‘put a stick at your neck’ (in a dance)

The examples in (11) also show that this difference in interpretation is not reflected in the argument structure: in both cases, the ‘location’ participant is represented as a locative-marked noun phrase. (However, the allative case rather than the locative case could have been used on the Location argument to unambiguously indicate that the theme actually moves.) On the basis of the use of -arra ‘PUT’ as a simple verb, we can thus propose the semantic characterization in (12a), from which the participant roles in (12b) can be extracted. 12. Semantic characterization for -arra ‘PUT’ a. -arra ‘an agent causes a theme to change its locative relation with respect to a location’ b. -arra

If this step is to help us at all in determining the meaning and argument structure of coverbs, we have to make the assumption that a verb is likely to be monosemous, rather than taking on a completely different meaning (or turning into a meaningless element) as soon as it combines with a coverb, unless there is strong evidence to the contrary. This was referred to in section 3 as the “monosemic bias.” When examining complex verbs that are formed with the verb -arra, such as the ones listed in (6) earlier, one can indeed arrive at a compositional interpretation that is consistent with the semantic characterization just proposed for -arra. For example, the complex verb in (6a), dirrg ganarrany ‘tie something up (in a location)’, does not entail that the theme moves to a different location, but only that it is caused to assume the locative relation encoded by the coverb dirrg ‘be tied up’. Thus, (6a) could refer to the event of tying up a dog even when the dog does not change its location. To give a further example, the coverb jubard in (6b) specifies the locative relation as one where the theme is enclosed in the location. As shown in (13), the theme may be caused to be in a position of being ‘enclosed’ either by moving the container, that is, the location (13a), or by moving the theme into the container (13b).

78

SCHULTZE-BERNDT 13. a. jubard shut.in

nganth-arra-ny

kap-gi

2sgA:3sgU-PUT-PST

cup-LOC

‘you shut it in in the jar’(addressee was pretending to catch a fly with a jar) b. jubard shut.in

gan-arra-m

envelope-gi

3sgA:3sgU-PUT-PST

envelope -LOC

‘she is putting it in an envelope’ (comment on a video-taped scene of someone writing a letter)

The verb -arra ‘PUT’, either as part of a complex verb as in (13), or as a simple verb as in (11), thus does not have a ‘path’ component, but only a component of change of configuration. In this respect it differs from a true verb of transfer such as English put. Similar cases of verbs that pragmatically correspond to English verbs of motion and transfer but that differ dramatically in their semantics and also their argument structure properties are discussed by Bohnemeyer (this volume) and Kita (this volume). These examples show that the meaning of verbs that can be used to describe the same real-world scenes cannot be assumed to be the same crosslinguistically, even if they have a similar argument structure. This could be further illustrated by a discussion of the meaning of the other closed-class verbs in Jaminjung, which, however, is beyond the scope of the present chapter. 5. MAKING SENSE OF COVERBS Even after having isolated, in the manner outlined in section 4, the semantic contribution of the closed-class verbs to the complex predicates in Jaminjung, we cannot be sure what exactly the contribution of the coverb is. For example, the combination of the coverb wirriny and the verb -arra ‘PUT’ in (6c) can be translated as ‘turn something over’. However, we still do not know whether wirriny is a monovalent dynamic predicate (‘turn [intr.]’), a bivalent dynamic predicate (‘turn something’), or a monovalent stative predicate (‘be upside down’). All three meanings would be consistent with the semantic characterization proposed in section 4 for the verb -arra and the interpretation of the complex predicate under a compositional analysis. Answering the question of what the coverbs “really” mean is more difficult than it is in the case of verbs. As already indicated in section 2, coverbs do not occur on their own in finite clauses, but have to be accompanied by a closed-class verb. Although coverbs may constitute the only predicate in nonfinite clauses (e.g., purposive clauses), they never occur with a full range of argument expressions in this case. Still, there is a strategy that allows one to determine more precisely the meaning of coverbs. This is to examine the range of verbs that a given coverb may combine with. As an illustration of this approach, consider the examples in (14). The two coverbs jarr and jurrb are in semantic opposition: jarr can only be predicated of singular entities, and jurrb only of nonsingular entities. Both combine with the same verb, -arra ‘PUT’, to form complex verbs that translate as ‘put something

4.

MAKING SENSE OF COMPLEX VERBS

79

down’. Tentative glosses for the two coverbs are therefore ‘put down one entity’ for jarr and ‘put down many entities’ for jurrb. 14. a. jungulug one(ABS )

jarr

gan-arra-m

?put.down.one

3sgA:3sgU-PUT-PRS

‘he/she puts down one’ (piece of firewood) b. jirrama, two(ABS )

jurrb

gan-arra-m

jirrama

?put.down.many

3sgA:3sgU-PUT-PRS

two(ABS )

‘two, he/she puts down two’ (pieces of firewood)

Both coverbs may also combine with other transitive verbs, such as -unga ‘LEAVE’. In addition, jurrb is also attested with the stative intransitive verb -yu ‘BE’. Example (15) is from a narrative where a boy and a dog, who have been looking for their pet frog, finally find it sitting amidst a frog family. From the context it is quite clear that there was no agent that ‘transferred’ the group of frogs. This suggests that jurrb means ‘be together, be (of many entities in the same place)’ rather than ‘put down many entities’. 15. malara=ma frog(ABS )= SUBORD

jurrb

ga-yu

be.together

3sg-BE.PRS

‘… where the frogs are (together)’

The coverb jurrb can thus be characterized as a coverb of spatial configuration (Jaminjung has a number of other coverbs encoding a complex configuration, e.g., murruny ‘be heaped up, be in a heap’). Its single central participant can be labeled as ‘figure’; possibly, this coverb also has a (peripheral) ‘location’ participant. This conclusion, of course, is only valid under a monosemic bias. Theoretically, one could postulate two senses for a coverb like jurrb, for example, ‘be together’ and ‘put down (of multiple entities)’. More generally, the reasoning behind our approach to determining the meaning and argument structure of the coverbs can be characterized as follows. According to the criteria proposed in section 3, central participants of coverbs can be identified as those that are expressed as core arguments across all constructions that a coverb can enter into. For those coverbs that combine with both transitive and intransitive verbs, only one participant is expressed as a core argument across all constructions. This is the participant represented, in the combination with an intransitive verb, by the pronominal prefix on the verb, and (optionally) by an absolutive noun phrase. Coverbs that allow a combination with both transitive and intransitive verbs thus have to be regarded as monovalent. When a monovalent coverb combines with a transitive verb, the verb contributes an additional participant, which is represented as a second core argument. The first core argument represents a participant of both the verb and the coverb, according to the assumptions about argument sharing outlined in section 3.

80

SCHULTZE-BERNDT

A schematic representation of argument sharing between the bivalent verb -arra ‘PUT’ and the monovalent coverb jurrb ‘be together’ is provided in Figure 4.1. Much of the notation in this and the other figures is adopted from Mohanan (1994, 1997) and from Goldberg (1995). The single ‘figure’ participant of the coverb is compatible with the second, ‘theme’ participant of the verb. Both participants are simultaneously represented by the Undergoer pronominal prefix, as well as (optionally) by an absolutive noun phrase. The uppermost “box” in Figure 4.1 represents a case-marking construction, consisting of the predicate and a case-marked argument (here in the absolutive). The combination of coverb and verb is thought of as instantiating a complex verb construction. (Recall that under the model of lexicon and grammar outlined in section 3, there is no contradiction in saying that a complex predicate instantiates a grammatical construction and is lexicalized at the same time.) The complex verb construction should be thought of as occupying the V slot in the case-marking construction. The bound pronominal construction, consisting of a verb root and its A and U prefix, is represented at the bottom of Figure 4.1, framed by a box with bolded lines. It should be thought of as embedded in the Verb slot of the complex verb construction (something that is not adequately captured by the notation). The representation of the two components of the complex predicate with their respective participant roles, which constitute the fillers of both argument structure constructions, is placed in between the boxes representing the other constructions. The ordering of the boxes that represent the constructions in this and the other figures should not be taken to represent any hierarchical ordering, because all constructions are simultaneously present. Free word order is such a pervasive feature of Jaminjung that it is not represented here. The arguments precede

FIGURE 4.1. Argument sharing of a monovalent coverb with a bivalent verb (example 14b).

4.

MAKING SENSE OF COMPLEX VERBS

81

the predicate only to allow a clearer mapping to the bound pronominal construction, where the order (pronominal prefix followed by the verb root) is, of course, fixed. By analogy, we expect the coverb jarr in (14a) to have the same valency as jurrb, the only difference being the singular number of the ‘theme’. However, jarr ‘put down one entity’is never combined with the intransitive verb -yu ‘BE’, or with other intransitive verbs, and all my attempts to form such combinations were rejected by speakers. From this we can conclude that jarr is a bivalent coverb that itself encodes a caused change of location. It has two central participants, an ‘agent’ and a ‘theme’, which have to be expressed as core arguments, and possibly a third, peripheral, ‘location’ participant. The integration of the bivalent coverb jarr ‘put down one entity’into a two-argument construction with the verb -arra ‘PUT’is illustrated in Figure 4.2. Here we find a total overlap, both semantically and in morphosyntactic expression, between the two central participants of the verb and those of the coverb. Both share the A and the U prefix of the verb, and in addition they share any argument that is represented by a noun phrase cross-referenced by these prefixes (in this example, only an absolutive noun phrase). Thus, by taking into consideration the set of verbs that a given coverb may combine with, it is possible to obtain indirect evidence for the coverb’s semantic valency. Direct evidence for the valency of coverbs, on the other hand, is found only in the case of a small number of coverbs that contribute an extra argument to the argument structure of the resulting complex verbs. One of these is yurrg ‘show, teach’. This coverb exclusively combines with the verb -arra ‘PUT’, familiar from previous

FIGURE 4.2. Argument sharing of a bivalent coverb with a bivalent verb (example 14a).

82

SCHULTZE-BERNDT

examples; here, arguably, the verb takes on a secondary sense of metaphorical transfer. The resulting complex verb, just like -ngarna ‘GIVE’ as a simple verb (see example (10)), is trivalent; that is, it allows for the expression of three core arguments. As shown in (16), the ‘agent’ is encoded as the Actor, and the ‘recipient’ as the Undergoer (and optionally, but not in this example, by an absolutive noun phrase), whereas the ‘theme’ (here: gagawuli) is (optionally) represented by an absolutive noun phrase, but is not cross-referenced on the verb. 16. mulurru-ni

gagawuli

yurrg

old.woman-ERG yam(ABS ) show

gan-garra-ny

Gilwi-ni

3sgA:1sgU-PUT-PST

-LOC

‘the old woman showed me yam in Gilwi’

Here we can see very clearly that the coverb yurrg has an influence on the overall argument structure of the complex verb. Because -arra ‘PUT’, by itself, does not allow for the expression of three core arguments, yurrg has to be a trivalent coverb. That is, it has three central participants, a ‘shower’, a ‘recipient’, and an ‘entity shown’, which all have to be expressed as core arguments. This is schematically represented in Figure 4.3. So far, I hope to have demonstrated that allowing for a compositional integration of the participant roles of verbs and coverbs is a way to make sense of complex verbs in Jaminjung. Specifically, although coverbs do not occur as simple predicates in finite clauses, the range of combinations that they form with members of the closed class of verbs, as well as the argument structure of the resulting complex verbs, allow us to draw conclusions about their meaning and valency. By this method (which

FIGURE 4.3. Argument sharing of a trivalent coverb with a bivalent verb (example 16).

4.

MAKING SENSE OF COMPLEX VERBS

83

could only be demonstrated here for one verb and a few coverbs), we arrive at classes of coverbs that pattern alike both in their combination with verbs and in their argument structure properties. A few of these classes—those that are exemplified in one way or another in this chapter—are listed in (17); for an exhaustive treatment of Jaminjung predicate classes see Schultze-Berndt (2000: chapter 6). 17. Some coverb classes a. Spatial configuration:

waga ‘sit’, dirrg ‘be tied up’ (5, 6a), jubard ‘be shut in, be enclosed’ (6b, 13), jurrb ‘be together’ (14b, 15)

b. Transfer:

dalag ‘send away’(6d), jarr ‘put down one entity’(14a)

c. Transfer of a message: yurrg ‘show, teach’ (16), yanggi ‘ask’ d. Change of state:

ning ‘break off, stop’ (4, 18), lag ‘split’, digirrij ‘die’

e. Internal cause:

bawa ‘shout’, nunaj ‘nod’, butharl ‘be sad’, burrng-burrng ‘bubble, boil’ (7)

Coverbs of spatial configuration are monovalent; they have a ‘theme’ or ‘figure’ as their only central participant, but have an additional, peripheral, location participant. This class includes the coverb jurrb ‘be together, be (in a place, of many entities)’ in (14b), while the coverb jarr ‘put down one entity’ in (14a) was identified as a bivalent coverb of transfer, belonging to a class that also contains dalag ‘send away’. Coverbs of transfer occur mainly or exclusively with the verb -arra ‘PUT’ whose semantics was characterized in section 4 as ‘cause something to change its locative relation with respect to a location’. This also holds for the small class of trivalent coverbs of ‘transfer of a message’, which includes yurrg ‘show, teach’, illustrated in (16), and yanggi ‘ask’. With these coverbs, the verb -arra ‘PUT’ takes on a metaphorical sense. Coverbs from the change-of-state class pattern like ning ‘break off’ in (8), repeated in (18). 18. a. b. c. d.

ning gani-ma ‘break something off by hitting it; kill (an animal)’ ning gana ‘chop off, kill by hitting with an edge’ ning gani-wa ‘bite off’ ning ga-jga-ny ‘break off (intr.); stop’

Coverbs of this type enter into a causative alternation: they combine with a number of transitive verbs of contact and force (18a–c) as well as with an intransitive verb (18d) in inchoative function (the general motion verb -(i)jga is used here in a change-of-state reading). By the criteria proposed so far, coverbs of change of state have to be regarded as monovalent rather than bivalent, because they allow the combination with intransitive verbs—in other words, a coverb like ning cannot mean ‘break off’ (tr.). They also have to be regarded as dynamic rather than stative, because they do not allow for a combination with a stative verb like -yu ‘BE’; that is, ning has to mean ‘break off’ (intr.) rather than ‘be broken off’.

84

SCHULTZE-BERNDT

Finally, the rather heterogeneous set of coverbs in (7), shown earlier, which all combine—mostly exclusively—with the verb -yu(nggu), is again listed in illustration of the class of coverbs of “Internal Cause” in (17e). “Internal cause,” to be understood in the sense of Levin and Rappaport Hovav (1995), is also the central semantic component of the verb -yu(nggu), which variously translates as ‘say’, ‘do’, ‘become’, or ‘manifest’(see Schultze-Berndt, 2000: 349–369, for a characterization of the semantics of this polyfunctional verb). This is just a subset of the coverb classes that can be established by applying the strategy proposed in this section. In the light of these classes of coverbs, Jaminjung does not seem all that exotic any more, because predicate classes of position, change of state, transfer, caused change of possession, and internal cause—with the predicates falling into these classes having very similar semantics—have all been identified for many languages, including better known languages like English (Levin, 1993). One strategy for identifying predicate classes in English has been to consider alternations in argument structure. This strategy is not viable for Jaminjung coverbs, which do not enter an argument frame on their own. Rather, in Jaminjung, predicate classes can be identified by the sets of closed-class verbs that the coverbs co-occur with, that is, by a kind of lexical alternation. 6. IMPLICATIONS In the literature on language acquisition there seems to be a convergence of views, and also a lot of empirical evidence, in favor of a procedure for verb learning that is based on already-learned associations of semantically-based predicate classes with certain patterns of argument structure. It seems reasonable that if a child associates certain argument structure constructions with a certain semantic class, the meaning of a new verb can be narrowed down on the basis of the argument structure(s) that it is found with. However, as the debate about “syntactic” versus “semantic” bootstrapping has shown (see, e.g., Pinker, 1994), the question still remains of how these associations are established in the first place. For those who maintain strict versions of either “syntactic bootstrapping” (Gleitman, 1990) or “semantic bootstrapping” (Pinker, 1994), the association, that is, the linking of verbal participant roles with argument expressions, has to be innate. However, this claim seems very problematic in light of the crosslinguistic diversity of both verb semantics and argument structure (cf. also the contributions by Bohnemeyer, Danziger, Essegbey, Kita, Margetts, and Wilkins in this volume). In this chapter, this question was addressed from the point of view of a—hypothetical—learner of Jaminjung, a Northern Australian language that has a closed class of semantically generic inflecting verbs, and an open class of uninflected coverbs that almost exclusively occur in complex verbs together with an inflecting verb. According to models of language where lexicon and grammar are strictly separated, a learner of Jaminjung faces the paradox of having to learn the meaning of the closed-class verbs in order to learn the meaning and argument structure of the coverbs, and vice versa.

4.

MAKING SENSE OF COMPLEX VERBS

85

In this chapter I have presented an outline of a plausible strategy of verb learning in Jaminjung. First, it was shown for the verb -arra ‘PUT’ that the meaning of a closed-class verb cannot necessarily be derived from its argument structure in a straightforward way. This verb has a location participant, which, however, is not overtly expressed as an argument in the majority of its uses. Furthermore, not syntax, but only the observation of the whole range of applications of this verb to real-world situations makes it possible to conclude that -arra has no semantic component of motion or path; that is, it is not strictly speaking a verb of transfer, but only entails caused change of locative relation with respect to a location. Second, it was demonstrated for a small subset of coverbs that predicate classes can be established for Jaminjung on the basis of alternations of coverbs with sets of closed-class verbs, and that the meaning of coverbs can be narrowed down in this way. The resulting coverb classes coincide to a large extent with predicate classes found in many other languages. These findings, therefore, do not in any way invalidate the claim that children learning a language would use the linguistic context that a “verbal word” occurs in, in order to narrow down its meaning. However, it is doubtful whether this context can be reduced to “syntax.” For Jaminjung, it seems plausible that the meaning of the uninflected coverbs can partly be determined, not in terms of argument structure alone, but by taking into consideration the range of inflected, closed-class verbs that the coverbs occur with. But of course the meaning of these verbs will have to be determined as well. Any unidirectional model of acquisition by “bootstrapping”—either from syntax to semantics, or vice versa—becomes implausible in this case. The conclusion that acquisition of grammar and lexical meaning would have to go hand in hand is consistent with a model of language according to which grammatical constructions are meaningful signs in their own right, with language-specific meanings. According to this model, there is no principled difference between learning the meaning of an argument structure construction in order to then use it to narrow down the meaning of verbs in English, and learning the meaning of a closed- class verb in order to be able to determine the meaning of coverbs in Jaminjung. ACKNOWLEDGMENTS For comments on previous versions of this chapter and/or more general discussions on the topic, I thank Felix Ameka, Jürgen Bohnemeyer, Melissa Bowerman, Penelope Brown, James Essegbey, Nikolaus Himmelmann, Frances Kofod, Silvia Kutscher, Christian Lehmann, Bill McGregor, Rachel Nordlinger, Jane Simpson, David Wilkins, Stephen Wilson, and Roberto Zavala. I also express my gratitude to the Aboriginal Communities in Bulla Camp, Timber Creek, and Mirima (Kununurra), and to the Jaminjung and Ngaliwurru speakers who taught me some of their language, in particular Violet Balidi, Dolly Bardbarriya, Daisy Bitting, Judy Marchant, the late Dinah and Duncan McDonald, Doris Pannikin, Iza Pretlove, Doris and Laurie Roberts, Eileen Roberts, and the late Margaret

86

SCHULTZE-BERNDT

Wilinygari. Financial support for my field work, which is gratefully acknowledged, came from the DAAD (Deutscher Akademischer Austauschdienst) for 6 months in 1993, and from the Max Planck Institute for Psycholinguistics in Nijmegen between 1994 and 1997. NOTES 1

Published descriptions of other languages of this area which are very similar to Jaminjung in the structure of their “verbal” lexicon and with respect to complex predicate formation include Hoddinott and Kofod (1988), McGregor (1993), Merlan (1994), Tsunoda (1981), and Wilson (1999). One of the best known Aboriginal languages, Warlpiri, is also similar in this respect, but has a somewhat larger class of verbs with just over 100 members (cf. Hale, 1982; Hale, Laughren, & Simpson, 1995; Nash, 1982, 1986; Simpson, 1991). 2 More precisely, there are 26 reasonably frequent verbs and 9 additional verbs that are very infrequent or obsolete, of which two only occur in the Ngaliwurru dialects. 3 Glosses of verbs are written in capitals throughout, because they should not be taken as proper representations of the verbs’ meanings. 4 The following abbreviations are used in the interlinear glosses: 1, 2, 3, first/ second/third person; A, Actor bound pronoun; ABS, Absolutive case; DEM, Neutral demonstrative, usually ‘given’; du, Dual; ERG, Ergative case; excl, Exclusive; POT, Potential mood; IMP, Imperative; incl, Inclusive; LOC, Locative case; POSS, Possessor case; pl, Plural; PRS, Present tense; PST, Past (perfective); sg, Singular; SUBORD, Subordinating clitic; U, Undergoer bound pronoun. A hyphen is used to indicate morpheme boundaries and an equals sign (=) to indicate clitic boundaries. 5 Because Jaminjung verbs do not have an infinitive form, the past perfective form of verbs with the third person singular Actor/third person singular Undergoer prefix is used as a citation form throughout. 6 In this chapter, no particular theoretical relevance is assigned to the labels used to represent participant roles. These should be regarded as abbreviations for roles specific to a predicate, or a predicate class, and do not have a status as primitives of the analysis. 7 A small number of complex verbs formed with -arra ‘PUT’ cannot be analyzed in this way and have to be considered truly idiomatic; see Schultze-Berndt (2000: 241–248) for details. 8 We would expect that it is not possible for coverbs and verbs to combine if none of their participants overlap; that is, the coverbs cannot introduce a set of arguments independent of the arguments of the verb. This expectation is indeed borne out by the Jaminjung data. 9 Presumably, the recipient participant of the coverb can be identified with the (peripheral) location participant of the verb, interpreted metaphorically.

4.

MAKING SENSE OF COMPLEX VERBS

87

REFERENCES Ackerman, F., & LeSourd, P. (1997). Towards a lexical representation of phrasal predicates. In A. Alsina, J. Bresnan, & P. Sells (Eds.), Complex predicates (pp. 67–106). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Bloomfield, L. (1970). Language. London: George Allen & Unwin Ltd. (Original work published 1933.) Dryer, M. S. (1986). Primary objects, secondary objects, and antidative. Language, 62, 808–845. Fillmore, C. J. (1988). The mechanisms of “construction grammar.” In S. Axmaker, A. Jaisser, & H. Singmaster (Eds.), Proceedings of the 14th Annual Berkeley Linguistics Society (pp. 35–55). Berkeley. CA: Berkeley Linguistics Society, University of California. Fillmore, C. J., Kay, P., & O’Connor, M. C. (1988). Regularity and idiomaticity in grammatical constructions: The case of let alone. Language, 64, 501–538. Fisher, C., Hall, D. G., Rakowitz, S., & Gleitman, L. (1994). When it is better to receive than to give: Syntactic and conceptual constraints on vocabulary growth. Lingua, 92, 333–375. Frei, H. (1962). L’unité linguistique complexe. Lingua, 11, 128–140. Gentner, D. (1978). On relational meaning: The acquisition of verb meaning. Child Development, 49, 988–998. Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55. Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Goldberg, A. E. (1996). Words by default: Optimizing constraints and the Persian complex predicate. In J. Johnson, M. L. Juge, & J. L. Moxley (Eds.), Proceedings of the 22nd Annual Berkeley Linguistics Society (pp. 132–146). Berkeley, CA: Berkeley Linguistics Society, University of California. Haiman, J. (1980). The iconicity of grammar: Isomorphism and motivation. Language, 56, 515–540. Hale, K. (1982). Some essential features of Warlpiri main clauses. In S. Swartz (Ed.), Papers in Warlpiri grammar, in memory of Lothar Jagst (pp. 217–315). Berrimah, Northern Territory, Australia: Work Papers of SIL-AAB, Series A, Vol. 6. Hale, K., Laughren, M., & Simpson, J. (1995). Warlpiri. In J. Jacobs, A. von Stechow, W. Sternefeld, & T. Vennemann (Eds.), Syntax: An international handbook of contemporary research (Vol. 2, pp. 1430–1451). Berlin: Walter de Gruyter. Hampe, B. (1997). Towards a solution of the phrasal verb puzzle: Considerations on some scattered pieces. Lexicology, 3, 203–243. Hockett, C. F. (1958). A course in modern linguistics. New York: MacMillan. Hoddinott, W. G., & Kofod, F. M. (1988). The Ngankikurungkurr language (Daly River area, Northern Territory). Canberra: Pacific Linguistics D-77. Kay, P., & Fillmore, C. (1999). Grammatical constructions and linguistic generalizations: The What’s X doing Y? construction. Language, 75, 1–33. Kirsner, R. S. (1985). Iconicity and grammatical meaning. In J. Haiman (Ed.), Iconicity in syntax (pp. 249–270). Amsterdam: John Benjamins. Landau, B., & Gleitman, L. R. (1985). Language and experience. Cambridge, MA: Harvard University Press. Langacker, R. W. (1987). Foundations of cognitive grammar, Vol. 1, Theoretical prerequisites. Stanford, CA: Stanford University Press. Langacker, R. W. (1990). Concept, image, and symbol: The cognitive basis of grammar. Berlin: Mouton de Gruyter.

88

SCHULTZE-BERNDT

Levin, B. (1993). English verb classes and alternations: A preliminary investigation. Chicago: University of Chicago Press. Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax-lexical semantics interface. Cambridge, MA: MIT press. McGregor, W. B. (1993). Gunin/Kwini. Munich, Germany: Lincom Europa. Merlan, F. (1994). A grammar of Wardaman: A language of the Northern Territory of Australia. Berlin: Mouton de Gruyter. Michaelis, L., & Lambrecht, K. (1996). Toward a construction-based theory of language function: The case of nominal extraposition. Language, 72, 215–247. Mohanan, T. (1994). Argument structure in Hindi. Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Mohanan, T. (1997). Multidimensionality of representation: Noun-verb complex predicates in Hindi. In A. Alsina, J. Bresnan, & P. Sells (Eds.), Complex predicates (pp. 431–472). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Naigles, L. R., & Hoff-Ginsberg, E. (1998). Why are some verbs learned before other verbs? Effects of input frequency and structure on children’s early verb use. Journal of Child Language, 25, 95–120. Nash, D. (1982). Warlpiri preverbs and verb roots. In S. Swartz (Ed.), Papers in Warlpiri grammar, in memory of Lothar Jagst (pp. 165–216). Berrimah, Northern Territory, Australia: Work Papers of SIL-AAB, Series A, Vol. 6. Nash, D. (1986). Topics in Warlpiri grammar. New York: Garland. Pawley, A., & Syder, F. H. (1983). Two puzzles for linguistic theory: Nativelike selection and nativelike fluency. In J. C. Richards & R. W. Schmidt (Eds.), Language and communication (pp. 191–225). London: Longman. Pinker, S. (1994). How could a child use verb syntax to learn verb semantics? In B. Landau & L. R. Gleitman (Eds.), The acquisition of the lexicon (pp. 377–410). Cambridge, MA: MIT Press. Ruhl, C. (1989). On monosemy: A study in linguistic semantics. Albany, NY: State University of New York Press. Schultze-Berndt, E. (2000). Simple and complex verbs in Jaminjung: A study of event categorisation in an Australian language. PhD dissertation, University of Nijmegen, Nijmegen, The Netherlands. Simpson, J. (1991). Warlpiri morphosyntax: A lexicalist approach. Dordrecht, The Netherlands: Kluwer Academic. Taylor, J. R. (1989). Linguistic categorization: Prototypes in linguistic theory. Oxford: Clarendon Press. Taylor, J. R. (1990). Schemas, prototypes, and models: In search of the unity of the sign. In S. L. Tsohatzidis (Ed.), Meanings and prototypes: Studies in linguistic categorization (pp. 521–534). London: Routledge. Tsunoda, T. (1981). The Djaru language of Kimberley, W.A. Canberra: Pacific Linguistics B-78. Wilson, S. (1999). Coverbs and complex predicates in Wagiman. Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Zwicky, A. (1987). Constructions in monostratal syntax. Chicago Linguistic Society, 23, 389–401.

CHAPTER 5

Figure–Ground Indeterminacy in Descriptions of Spatial Relations: A Construction Grammar Account Sotaro Kita University of Birmingham

What has to be specified in the semantic representation of a verb in order to be able to account for its distributional characteristics? This chapter explores this question by examining how spatial relationships and motion are described in Japanese and other languages. The focus is on two aspects of the distributional characteristics of a verb: (a) a set of constructions the verb can appear in and (b) semantic types of NPs that can co-occur with the verb in a specific position in a given construction. More specifically, the aim of this chapter is to provide an account in terms of Construction Grammar (Goldberg, 1995) for what I call the Figure–Ground indeterminacy phenomenon (this phenomenon was first discussed by Brown, 1994). I will also discuss the implications of this phenomenon for language acquisition. 1. WHAT IS FIGURE–GROUND INDETERMINACY? 1.1. Definition of Figure–Ground Indeterminacy “Figure” and “Ground” are terms used in Talmy’s work on semantics of spatial expressions to distinguish an entity to be described spatially and an entity serving as a reference point, respectively. More specifically, Figure is the entity that “has unknown spatial… properties to be determined” and Ground is the entity that “acts as a reference entity, having known properties that can characterize the Figure’s unknowns” (Talmy, 2000: 315). Examples (1) and (2) exemplify how Japanese typically encodes Figure and Ground in a motion event.1 89

90

KITA 1. densha-ga train-Nom

eki-ni

tsui-ta

station-Loc

arrive-Pst

‘The train (Figure) arrived at the station (Ground).’ 2. Hanako-ga Hanako-Nom

terebi-o

daidokoro-ni

oi-ta

TV-Acc

kitchen-Loc

put-Pst

‘Hanako put a TV (Figure) in the kitchen (Ground).’

In Japanese, a typical way to mark Ground is with the locative postposition, ni (see Kita, 2006, for more details). The Figure is typically marked with the nominative postposition in a spontaneous motion event description, as in (1), or with the accusative postposition in caused motion events, as in (2). The Figure in (1) and (2) is the entity that changes its location. In sentences with tsuku (the citation form for tsui- in (1)) ‘to arrive’ and oku (the citation form for oi- in (2)) ‘to put’, the entity that changes its location must be the Figure, and the end location of the entity must be the Ground. Thus, if the Figure NP and Ground NP in such sentences are interchanged, the sentences may become infelicitous, as in (3) and (4), and they cannot refer to the same events as the original sentences ((1) and (2)). 3. # eki-ga station-Nom

densha-ni

tsui-ta

train-Loc

arrive-Pst

‘The station arrived at the train.’ 4. # Hanako-ga Hanako-Nom

daidokoro-o

terebi-ni

oi-ta

kitchen-Acc

TV-Loc

put-Pst

‘Hanako put a kitchen in the TV.’

However, there is a class of Japanese verbs that allow just such an exchange of the Figure NP and the Ground NP. Sasaru (sasat- in (5)) ‘to pierce (intransitive)’ and sasu (sashi- in (6)) ‘to pierce (transitive)’ are such verbs. These are what I call verbs with Figure–Ground indeterminacy. 5. a. niku-ga meat-Nom

kushi-ni

sasat-ta

skewer-Loc

pierce(intr)-Pst

‘The meat (Figure) pierced the skewer (Ground).’ b. kushi-ga skewer-Nom

niku-ni

sasat-ta

meat-Loc

pierce(intr)-Pst

‘The skewer (Figure) pierced the meat (Ground).’ 6. a. Taroo-ga Taroo-Nom

kushi-o

niku-ni

skewer-Acc meat-Loc

sashi-ta pierce(tr)-Pst

‘Taroo puts the skewer (Figure) into the meat (Ground)’ (so as to achieve a piercing configuration).

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS b. Taroo-ga Taroo-Nom

niku-o

kushi-ni

sashi-ta

meat-Acc

skewer-Loc

pierce(tr)-Pst

91

‘Taroo puts the meat (Figure) on the skewer (Ground)’ (so as to achieve a piercing configuration).

There are two important things to note about the pairs of examples in (5) and (6). First, both the verb and the construction in (a) and (b) are identical. Second, (a) and (b) are both semantically valid descriptions of the same set of events. The two verbs in (5) and (6) specify that two distinct types of event participants (i.e., piercer and piercee) are involved. However, the verbs do not determine which one must be the Figure and which one must be the Ground. In other words, the verbs exhibit Figure–Ground indeterminacy. Despite the indeterminacy, one participant may be pragmatically preferred as a Figure relative to the other participant, for a given context. I discuss this issue in more detail later. The pairs in (5) and (6) are different in a couple of important respects from the celebrated “argument structure alternations” in English, as in (7) and (8), which are sometimes characterized as involving a “reversal” of Figure and Ground (e.g., Pinker, 1989; Talmy, 2000). 7. a. The bees swarmed in the tree. b. The tree swarmed with bees. 8. a. John loaded hay onto the truck. b. John loaded the truck with hay.

In (7) and (8), the (a) and (b) sentences instantiate different constructions. In general, different constructions have different semantics (Goldberg, 1995), and it has indeed been claimed that (a) and (b) have different semantics. The (b) variant has been claimed to have a “holistic” effect, as first observed by Anderson (1971). Thus, in (7b) the whole tree is covered with bees, and in (8b) the truck is fully loaded with hay. The (a) variant does not exhibit this effect. This leads to the analysis, for example, of (8b) as an instance of caused change of state with a particular means, and (8a) as an instance of caused change of location (e.g., Goldberg, 1995; Rappaport & Levin, 1988). In contrast to these locative alternation examples in English, the Japanese examples (a) and (b) in (5) or (6) share an identical construction and identical semantics (in the sense that they can refer to the same set of events). Furthermore, the pairs in (5) and (6) differ from pairs of examples involving a potentially symmetrical predicate, namely, a predicate whose semantics is general enough to cover a construal of an event as involving a symmetrical relation between two participants. Karamaru ‘to entangle (intransitive)’is such a predicate. Example (9a) with karamaru (karamat- is a conjugational variant) entails that the referent of the nominative NP, the red thread, moved and entangled itself on the white thread. The sentence is also applicable to a scene involving a symmetrical relation (henceforth, a symmetrical scene), in which two equivalent participants (i.e., two threads)

92

KITA

both move and get into mutual entanglement. Such a scene can be described by (9a) as well as (9b), and crucially by a construction with a conjoined subject, (9c). Unlike (9a) and (9b), (9c) is applicable only to symmetrical scenes. 9. a. aka-i

ito-ga

shiro-i

red-Prs thread-Nom white-Prs

ito-ni

karamat-ta

thread-Loc entangle(intr)-Pst

‘The red thread entangled with the white thread.’ b. shiro-i white-Prs

ito-ga

aka-i

ito-ni

karamat-ta

thread-Nom

red-Prs

thread-Loc

entangle(intr)-Pst

‘The white thread entangled with the red thread.’ c. aka-i red-Prs

ito-to thread-and

shiro-i white-Prs

ito-ga thread-Nom

karamat-ta entangle(intr)-Pst

‘The red thread and the white thread entangled.’

The pair (9a) and (9b) may look superficially similar to the pair (5a) and (5b); however, these two pairs are fundamentally different. The pair (9a) and (9b) can describe the same scene only if the scene is symmetrical. This becomes apparent in (10a,b). Example (10) has the same verb and construction as (9), but one of the participants, “the stick,” is not a flexible object, and the construed relation between the two participants cannot be symmetrical. Thus, if you exchange the Figure NP and the Ground NP in (10a), then the resulting sentence, (10b), is infelicitous. The nonsymmetrical nature of the relation is also underscored by the infelicity of (10c). 10. a. aka-i red-Prs

ito-ga

boo-ni

karamat-ta

thread-Nom

stick-Loc

entangle(intr)-Pst

‘The red thread entangled onto the stick.’ #b. boo-ga stick-Nom

aka-i

ito-ni

karamat-ta

red-Prs

thread-Loc

entangle(intr)-Pst

‘The stick entangled onto the red thread.’ #c. aka-i red-Prs

ito-to

boo-ga

karamat-ta

thread-and

stick-Nom

entangle(intr)-Pst

‘The red thread and the stick entangled.’

In contrast to (9a) and (9b), (5a) and (5b) do not refer to a symmetrical scene, as evidenced by the fact that a construction with a conjoined subject as in (11) cannot refer to the same scene as (5a) and (5b). Unlike two entangled threads, a piece of meat and a skewer are not equivalent entities in the piercing configuration. More generally, sasaru, unlike karamaru, can never refer to a symmetrical scene.

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS 11.

# niku-to meat-and

kushi-ga

sasat-ta

skewer-Nom

pierce(intr)-Pst

93

‘The meat and the skewer pierced each other.’ (The sentence is infelicitous for the intended meaning.)

Similarly to the discussion of the difference between sasaru and karamaru, the pair (6a) and (6b) should be distinguished from the pair (12a) and (12b), which involve a lexical causative of karamaru, namely, karameru (karame- in (12)) ‘to entangle (transitive)’. 12. a. Taroo-ga Taroo-Nom

aka-i

ito-o

red-Prs

thread-Acc white-Prs thread-Loc entangle(tr)-Pst

shiro-i

ito-ni

karame-ta

‘Taroo entangled the red thread onto the white thread.’ b. Taroo-ga Taroo-Nom

shiro-i

ito-o

aka-i

white-Prs

thread-Acc red-Prs

ito-ni

karame-ta

thread-Loc entangle(tr)-Pst

‘Taroo entangled the white thread onto the red thread.’ c. Taroo-ga Taroo-Nom

shiro-i

ito-to

aka-i

ito-o

white-Prs

thread-and

red-Prs

thread-Acc entangle(tr)-Pst

karame-ta

‘Taroo entangled the white thread and the red thread.’

Crucially, unlike (12a) and (12b), (6a) and (6b) do not refer to a scene with a symmetric relation between the accusative NP referent and the locative NP referent. This is evidenced by the fact that (12c) is felicitous, but (13) is not. 13. # Taroo-ga Taroo-Nom

niku-to

kushi-o

sashi-ta

meat-and

skewer-Acc

pierce(tr)-Pst

‘Taroo pierced the meat and the skewer each other.’ (The sentence is infelicitous for the intended meaning.)

Thus, the defining features for the central case of Figure–Ground indeterminacy, as exemplified by (5) and (6), are as follows: (a) Figure and Ground are construed to be in a nonsymmetrical relationship with respect each other, and (b) when two versions of a sentence are created by exchanging the two NPs referring to Figure and Ground (keeping the verb and the construction constant), the two versions are both semantically valid descriptions of the same set of events. 1.2. Pragmatic Preference for the Choice of Figure and Ground Even though the (a) variant and the (b) variant of (5) and (6) can refer to the same set of events, one variant may be pragmatically preferred to the other in some contexts. This is because certain characteristics of participants of an event or a state make one

94

KITA

participant more appropriate for Figure and the other more appropriate for Ground. Talmy (2000) suggests the list of such characteristics shown in Table 5.1. Sometimes these characteristics are aligned and make one choice of Figure and Ground much more felicitous than the other, but sometimes they compete with each other and cancel each other’s effect. In addition, a preference for one choice of Figure and Ground is determined by the relative suitability for Figure (or for Ground) of one event participant over the other event participant. Take example (14), involving the expression “to be near”. 14. (Talmy, 2000: 314) a. The bike (Figure) is near the house (Ground). ?b. The house (Figure) is near the bike (Ground).

The expression refers to a symmetrical relation between two participants; thus, it is possible to describe the same scene in two versions, (a) and (b). But (14a) is more felicitous than (14b) because a bike is typically more movable and smaller than a house. In other words, characteristics (a) and (b) in Table 5.1 jointly reinforce the preference for the bike as Figure and the house as Ground. However, if the house were in fact a movable miniature toy house and the bike were a regular life-size bike, then the bike would be a better Ground and the house would be a better Figure. If the bike were in fact a sculpture-bike fixed on the ground, the bike would become an even better Ground. In the Figure–Ground indeterminacy pairs in (5) and (6), a similar phenomenon can be observed. Suppose that the skewer and the meat have a typical size for Japanese cuisine: say, a 15–cm-long skewer and pieces of meat in 2.5–cm cubes. The preferred interpretation of (5) and (6) is that it is the Figure that moved and the Ground that stayed in place, although this does not have to be the case. However, TABLE 5.1 Characteristics of Participants in an Event or a State That Make Them More Appropriate for Figure or for Ground (Talmy, 2000, pp. 315–316)

Figure

Ground

a) more movable

more permanently located

b) smaller

larger

c) geometrically simpler, (often) pointlike in its treatment

geometrically more complex in its treatment

d) more recently on the scene/ in awareness

more familiar/expected

e) of greater concern/relevance

of less concern/relevance

f) less immediately perceivable

more immediately perceivable

g) more salient, once perceived

more backgrounded, once Figure is perceived

h) more dependent

more independent

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS

95

the preference for this interpretation can be shifted by changing the size of one of the participants and consequently its salience. Suppose that the same skewer moved and pierced a stationary orange (say, 8 cm in diameter). The rough proportions of this scene can be seen in Figure 5.1, which is taken from the animation stimuli for linguistic elicitation in Levinson (2002). When sentences equivalent to (5a) and (5b), or (6a) and (6b) are used to describe the scene, the preference for the skewer as the Figure will be stronger than in the case of piercing meat pieces. Another example of overriding the preference for a moving participant as Figure is when a human body part is involved in a piercing event. Imagine a scene in which somebody steps on a nail sticking out of the floor and the nail penetrates her foot. Both (15a) and (15b) are compatible with this scene. However, (15a) is more felicitous, even though the foot is more movable and the nail is more fixed. 15. a. kugi-ga nail-Nom

ashi-ni

sasat-ta

leg/foot-Loc

pierce(intr)-Pst

‘The (carpentry) nail pierced the leg/foot.’ ?b. ashi-ga leg/foot-Nom

kugi-ni

sasat-ta

nail-Loc

pierce(intr)-Pst

‘The leg/foot pierced the (carpentry) nail.’

This is because other characteristics listed in Table 5.1 support the choice of Figure and Ground shown in (15a). Due to the characteristics (d), (f), and (g), there may be a general bias toward a human body as Ground. In this example, characteristic (b) provides additional support for the human body as Ground. When these characteristics supporting (15a) weaken, the preference for (15a) weakens. For example,

FIGURE 5.1. A piercing scene from Levinson’s (2002) Moverb animations.

96

KITA

(15b) becomes a better choice if the amputated foot of a person (perhaps better, of a doll) moves and gets pierced by the nail. To summarize, in a pair of sentences involving a verb with Figure–Ground indeterminacy, one variant may be pragmatically preferred over the other because one participant in the event or the state may be seen as more suitable as Figure (or as Ground) than the other participant. 2. THE TREATMENT OF FIGURE–GROUND INDETERMINACY IN CONSTRUCTION GRAMMAR In this section, I discuss the syntax–semantics nexus in Japanese sentences with various classes of intransitive and transitive verbs mentioned above. The key facts to be captured are the following. Let us take the case of the intransitive verbs. The “arriving” verbs, as in (1), and the intransitive “piercing” verb, as in (5), (as well as the intransitive “entangling” verbs, as in (9) and (10)), appear in the identical construction. The “arriving” verb requires the “arriver” to be the Figure and the “arrival location” to be the Ground. In contrast, for the “piercing” verb, there is no fixed mapping between event participants (i.e., “piercer” and “piercee”) and Figure–Ground slots in the construction. In other words, for the “piercing” verb, the unification of verb semantics and the construction semantics is underspecified. I discuss how this Figure–Ground indeterminacy can be treated within the framework of Construction Grammar (Goldberg, 1995). A construction is the basic theoretical building block in Construction Grammar. It is a form–meaning pair whose structure and/or meaning are not predictable from its components (Fillmore, Kay, & O’Connor, 1988). Constructions are basic units of language, and they can consist of a single morpheme or a combination of words or higher-order syntactic categories. Goldberg (1995) extended this approach to argument structure, suggesting that basic sentence types should be treated as constructions. For example, the English Caused Motion Construction is a syntax–semantics pairing, as illustrated in Figure 5.2, in which the subject of a clause is paired with the cause of motion in semantics, the direct object with the theme of motion, the oblique argument with the goal, and the verb with the predicate, CAUSE-MOVE. When this construction merges with a compatible verb, the empty slot in the middle layer of Figure 5.2 is filled by semantic specification of the

FIGURE 5.2. English Caused Motion Construction (a simplified version of Figure 2.6 in Goldberg, 1995: 52).

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS

97

verb, namely, the predicate (PRED) and participant roles. Note that Figure 5.2 is a 2 simplified version of Goldberg’s original figure. Verbs have their own semantic specifications. Verb semantics includes “participant roles”; for example, the English verb put has three participants with distinct roles, . In addition, verb semantics includes information that allows one to establish a specific set of semantic relations between a verb and an abstract semantic predicate such as CAUSE-MOVE; for example, put is an “instance” of CAUSE-MOVE. This allows the construction and the verb to be unified, as in Figure 5.3. In order for the participant roles of a verb and the argument roles of a construction to be fused, the roles must be semantically compatible with each other (“Semantic Coherence Principle,” Goldberg, 1995). In the example in Figure 5.3, the putter role is compatible with the agent role, the puttee role with the theme role, and the put.place with the goal role. Let us now turn to the Japanese transitive verbs in question. The putting verb, as in (2), the transitive piercing verb, as in (6), and the transitive entangling verb, as in (12), all share an argument structure for which we can stipulate a construction called “Caused Locative Change”, as in Figure 5.4, which is similar to the English Caused Motion Construction.3 In this construction, Figure is associated with accusative, and Ground is associated with oblique. The meaning of this construction is that something causes the Figure and Ground to be in a particular locative relationship. A specific instance of this construction is the Caused Locative Change with Goal Construction, in which Goal is associated with locative. The putting verb in (2), the transitive piercing verb in (6), and the transitive entangling verb in (12) all instantiate this construction. The difference among the sentences with the three classes of transitive verbs is captured in the semantic representation of the verbs. This requires us to enrich verb semantics with information that Goldberg (1995) did not stipulate. In Goldberg (1995), the semantic specification of a verb consists of two things: (a) a list of participant roles and (b) an indication of which roles are “profiled.” Profiled participants roles are “focal points within the scene, achieving a special degree of prominence” (p. 44). In English, they are obligatorily expressed in finite

FIGURE 5.3. Composite fused structure: Caused Motion + put (a simplified version of Figure 2.7 in Goldberg, 1995: 52).

98

KITA

FIGURE 5.4. Japanese Caused Locative Change constructions.

clauses unless some arguments are suppressed, as in the passive construction. In Japanese, a pro-drop language, a profiled participant has to be specified in the interpretation even if its surface realization is null. Profiling status is a feature of participant role. In order to represent the difference among the three classes of Japanese transitive verbs in question, we need additional features of participant roles. The putting verb, oku, has three participant roles, as in (16). In addition, the theme has the feature [+Figure] and the end location has [+Goal]. When the verb is fused with the Caused Locative Change with Goal Construction, the feature specifications just described force the theme to be unified with Figure, and the end location with Goal, which is an instance of Ground. 16. oku [+Figure]

[+Goal]

The transitive piercing verb, sasu, also has three participant roles. However, the “piercing object” role and the “pierced object” role are not specified for [+Figure] or [+Ground], as shown in (17). 17. sasu

This allows these participant roles to be unified with either Figure or Goal in the Caused Locative Change with Goal Construction. That is, it is possible to associate the piercing object with accusative and the pierced object with locative, or the piercing object with locative and the pierced object with accusative. Which association is preferred is determined by the pragmatic factors listed in Table 5.1.

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS

99

The transitive entangling verb, karamaru, has three participant roles. The entangler role is marked for [+Figure] and the entanglee role is marked for [+Ground], as in (18). The feature specifications account for the contrast between (10a) and (10b); namely, the entangler can be the Figure, but the entanglee cannot. 18. karameru [+Figure]

[+Goal]

Set(entanglers) Õ Set(entanglees)

Now, how do we capture the potentially symmetrical nature of the predicate, as illustrated in (12)? The crucial thing to capture is the relationship between potential entanglers and entanglees. An entangler has to be a long flexible object (e.g., a thread). An entanglee can also be a long flexible object, but it does not have to be. For example, a stick can be an entanglee. That is to say, the set of potential entanglers is a subset of the set of potential entanglees. This subset relationship is specified in (18) as a part of the lexical entry for karamaru.4 When an entanglee is an entity that could also qualify as an entangler, such as a thread in (12), two alternative descriptions can be given to the same situation, as shown in (12a) and (12b). Furthermore, a construction with a conjoined accusative noun phrase, as in (12c), becomes the third alternative description. Thus, two alternatives for the transitive entangling verbs, as in (12a) and (12b), and two for the transitive piercing verbs, as in (6a) and (6b), are licensed for different reasons. For the transitive entangling verb, the subset relationship between potential entanglers and entanglees makes it possible to have the two alternatives. For the transitive piercing verb, the lack of Figure–Ground specification for the participant roles makes it possible to have the two alternatives. The intransitive verbs of arriving, piercing, and entangling can be given essentially the same analyses as the transitive verbs, as shown in (19)–(21). 19. tsuku [+Figure]

[+Goal]

20. sasaru 21. karamaru [+Figure]

[+Goal]

Set(entanglers) Õ Set(entanglees)

These verbs unify with the Locative Change constructions shown in Figure 5.5, which constitute a “subpart” (Goldberg, 1995) of the Caused Locative Change Construction in Figure 5.4. In order to capture the distributional facts regarding arguments of the three classes of transitive verbs and the three classes of intransitive verbs, I have used Construction

100

KITA

FIGURE 5.5. Japanese Locative Change Constructions.

Grammar à la Goldberg (1995). The Construction Grammar approach is suitable for this purpose since it allows one to specify verb semantics and the semantics of argument structures separately, unlike other approaches where argument structure semantics is represented in the lexical entry of a verb (e.g., Pinker, 1989; Rappaport & Levin, 1988). In the current analysis, the differences among classes of Japanese verbs sharing an argument structure are attributed to differences in verb semantics. The meaning contributed by the shared argument structure is specified separately. The specification of verb semantics is enriched, as compared to that in Goldberg (1995). In particular, it is suggested that verb semantics specifies not only which participant roles are profiled but also which participant roles must be Figure and Ground, and a possible subset relationship between potential candidates for participant roles. Figure–Ground indeterminacy can arise because some verbs do not specify which participant roles should be Figure and Ground. 3. OTHER JAPANESE VERBS IN THE SIX CLASSES OF VERBS There are other Japanese verbs that have essentially the same properties as the six classes of verbs discussed earlier. Many verbs have essentially the same properties as tsuku ‘to arrive’in (19). Hairu (hait- in (22)) ‘enter’is an example. There are also many verbs that have essentially the same properties as oku ‘to put’ in (16). Noseru (nose- in (23)) ‘to put.on.top.of’ is one such verb. 22. Hanako-ga Hanako-Nom

heya-ni

hait-ta

room-Loc

become.inside-Pst

‘Hanako entered the room.’

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS 23. Hanako-ga Hanako-Nom

sara-o

tana-ni

nose-ta

dish-Acc

shelf-Loc

put.on.top-Pst

101

‘Hanako put a dish on the shelf.’

Verbs that have properties equivalent to sasaru ‘to pierce (intransitive)’, as in (20), or sasu ‘to pierce (transitive)’, as in (17), are rare. Nukeru ‘to un-pierce (intransitive)’ and nuku ‘to un-pierce (transitive)’ are such verbs. Nukeru (nuke- in (24a,b)) and nuku (nui- in (25a,b)) share important characteristics with sasaru and sasu, respectively, including Figure–Ground indeterminacy. However, unlike for sasaru and sasu, the Ground for nukeru and nuku is marked by the ablative postposition. 24. a. niku-ga meat-Nom

kushi-kara

nuke-ta

skewer-Abl

unpierce(intr)-Pst

‘The meat came off the skewer.’ b. kushi-ga skewer-Nom

niku-kara

nuke-ta

meat-Abl

unpierce(intr)-Pst

‘The skewer came out of the meat.’ 25. a. Hanako-ga Hanako-Nom

niku-o

kushi-kara

nui-ta

meat-Acc

skewer-Abl

unpierce(tr)-Pst

‘Hanako took the meat off the skewer.’ b. Hanako-ga Hanako-Nom

kushi-o

niku-kara

nui-ta

skewer-Acc

meat-Abl

unpierce(tr)-Pst

‘Hanako took the skewer out of the meat.’

There are some verbs that belong to the same class as karamaru ‘to entangle (intransitive)’, as in (21), and karameru ‘to entangle (transitive)’, as in (18). Butsukaru (butsukat- in (26)) ‘to collide’ and butsukeru (butsuke- in (27)) ‘to make collide’, respectively, are such examples. Sentences (26a,b) entail that the referent of the nominative NP moved and hit the referent of the locative NP. Similarly, sentences (27a,b) entail that the referent of the accusative NP moved and hit the referent of the locative NP. These two pairs of sentences, (26a,b) and (27a,b), can also refer to a scene where both the red and white balls moved and collided with each other. This symmetrical scene can also be described by (26c) and (27c) with a conjoined NP. 26. a. aka-i red-Prs

booru-ga

shiro-i

booru-ni

butsukat-ta

ball-Acc

white-Prs

ball-Loc

collide-Pst

‘The red ball collided with the white ball.’

102

KITA b. shiro-i white-Prs

booru-ga

aka-i

booru-ni

butsukat-ta

ball-Nom

red-Prs

ball-Loc

collide-Pst

‘The white ball collided with the red ball.’ c. aka-i red-Prs

booru-to

shiro-i

booru-ga

butsukat-ta

ball-and

white-Prs

ball-Nom

collide-Pst

‘The red ball and the white ball collided with each other.’ 27. a. Hanako-ga

aka-i

Hanako-Nom red-Prs

booru-o

shiro-i

ball-Acc

white-Prs ball-Loc

booru-ni butsuke-ta make.collide-Pst

‘Hanako made the red ball collide with the white ball.’ b. Hanako-ga Hanako-Nom

shiro-i

booru-o

white-Prs ball-Acc

aka-i

booru-ni butsuke-ta

red-Prs

ball-Loc

make.collide-Pst

‘Hanako made the white ball collide with the red ball.’ c. Hanako-ga Hanako-Nom

aka-i

booru-to

shiro-i

red-Prs

ball-and

white-Prs ball-Acc make.collide-Pst

booru-o

butsuke-ta

‘Hanako made the red ball and the white ball collide with each other.’

4. FIGURE–GROUND INDETERMINACY CROSSLINGUISTICALLY Figure–Ground indeterminacy is not a quirky feature of Japanese. The same phenomenon has been observed in at least a few other languages. It was first reported by Brown (1994) in a Mayan language, Tzeltal. (Brown called it “Figure–Ground ambiguity.”) According to Brown (1994), the Tzeltal locational construction takes the form of “V-Stative ta NP NP.” The first element in the construction is a “dispositional verb” with the stative suffix. The second element is a prepositional phrase denoting Ground, with a generic preposition ta. The final element is a noun phrase denoting Figure, which is often omitted because the referent is clear from the context. Example (28) is a Tzeltal sentence exemplifying the locative construction. 28. (Brown, 1994: 776)5 pach-al

ta mexa

bojch

be.positioned-Stative

at table

gourd.bowl

‘The gourd bowl is on the table.’

A closed class of dispositional verbs (more than 200 in number) can be used in the locative construction, and they typically encode the characteristic shape, orientation, or configuration of the Figure and/or Ground. For example, in (28) with the verb pach, the Figure is ‘bowl-shaped and upright’. Interestingly, Brown found that

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS

103

a small subset of dispositional verbs allow not only the Figure but also the Ground to fulfill the shape characteristics specified. Pach is, in fact, one such exceptional verb, as shown in (29). 29. (Brown, 1994: 776) pach-al

ta bojch

te mantzanae

be.positioned-Stative

at gourd.bowl

the apple

‘The apple is in the gourd bowl.’

In terms of the analysis proposed in this chapter for such a phenomenon, the semantics of the verb, pach, has two participant roles. One role is specified for specific shape characteristics, and the second role is not. Neither of the roles is specified for the Figure or Ground status, as in (30). 30. pach

The two participant roles can be fused with either of the two NPs in the locative construction: the NP in the ta-prepositional phrase (Ground) or the sentence-final NP (Figure). In the Tzeltal dispositional verbs that do not exhibit Figure–Ground indeterminacy, the participant role with a specific shape characterization is specified for [+Figure], and the other role is specified for [+Ground]. This ensures that the NP referring to an entity with specific shape characteristics appears in the Figure position in the locative construction. Bohnemeyer (1997) reports data that suggest that the ‘entering’ verb in Yukatek (another Mayan language spoken in Mexico) has Figure–Ground indeterminacy. He elicited from his consultant descriptions of scenes that he demonstrated with a toy car and a box. When he moved the car into the box, the consultant described the scene as in (31). 31. (Bohnemeyer, 1997:84) h

òok

le

chan

káaro

te’l

ich

le

kàaha-o’

PRV

enter(B.3.SG)

Def

small

cart

there

in

Def

box-D2

‘The little cart went into the box (lit.: entered there in the box).’

In (31), the subject NP of the ‘enter’ verb, òok, refers to the containee, which changed its location in his demonstration. This is similar to the Japanese ‘enter’ verb, as we saw in (22), and to English expressions such as go into and enter. However, when Bohnemeyer pushed the box so that the stationary car ended up inside, the consultant described the scene again with the ‘enter’ verb, but in a causativized form. Furthermore, the consultant put the NP referring to the box in such a position, as in (32), that the literal translation would be ‘You made the box enter with respect to the little cart’.

104

KITA

32. (Bohnemeyer, 1997: 84) t-aw

òok-s-ah

PRV-A.2 enter-Caus-CMP(B.3.SG)

le

kàaha te

chan

kàaro-o’

Def

box

small

cart-D2

Loc:Def

‘You made the box enter with respect to the little cart.’

Crucially, the speaker accepted sentence (33) as a description of the same situation. 33. (Bohnemeyer, personal communication) le

kàaha

Def box

he’la

h

òok

te’l

ich

le

kàaro-o’

here

PRV

enter(B.3.SG)

there

in

Def

cart-D2

‘The box here, it entered into the little cart (lit.: entered there in the little cart).’

In (33), the ‘enter’verb is used without a causative morpheme, and the NP referring to the box is in the subject position. (Note that, unlike in (31), the subject NP in (33) is topicalized at the sentence-initial position, deviating from the Yukatek canonical word order of VSO.) When asked what happened to the car (in the same scene), the consultant replied as follows: 34. (Bohnemeyer, 1997: 84) le

chan

kàaro-o’ h

òok

xan

te’l

ich le

Def small cart-D2 PRV enter(B.3.SG) also there in

Def

kàaha-o’ box-D2

‘The little cart, it entered the box (lit.: entered there in the box), too.’

In (34), the sentence structure is identical to (33), but the subject NP refers to the car. (The subject NP is again topicalized in the sentence-initial position.) The contrast between (33) and (34) suggests that the semantics of the Yukatek ‘enter’ verb does not specify which participant role must be the Figure and which the Ground.6 This contrasts with the Japanese ‘enter’ verb, hairu (see (22)), in which Figure and Ground are linked to specific participant roles. Thus, the containee always has to be marked by the nominative and the container has to be marked by the accusative. (Yukatek òok and Japanese hairu are, however, similar in that they can both apply to a scene in which a container moves and a stationary containee ends up inside. See Kita, 1999, for more on this phenomenon.) Note further that the choice of Figure in the preceding Yukatek examples suggests that Talmy’s features for Figure–Ground preferences in Table 5.1 are also relevant in Yukatek. When there is a choice, the moving entity is likely to be chosen as the Figure. The first descriptions given by the consultant follow this principle, as in (31) and (32). And when an entity is the discourse topic, features (d) and (e) in Table 5.1 support the choice of topicalized entity as the Figure. For example, when Bohnemeyer asked what happened to the car, setting up the car as the discourse topic, the consultant chose the car as Figure in his reply (34).

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS

105

A Figure–Ground indeterminacy in a different language is reported by Ameka (1999): in Likpe (a Kwa language spoken in Ghana), the ‘be.at’ verb, kpé, exhibits Figure–Ground indeterminacy when it is used with a possessive 7 function. That is, either the possessor or the possessed can be either the subject or the object, as in (35a,b). And it is impossible to have a subject with conjoined NPs, as in (36). 35. (Ameka, 1999: 23) a. n-kpé

lé-siabí

1SG-be.at

CM-knife

‘I have a knife.’ b. lé-siabí CM-knife

kpé

mə

be.at

1SG

‘I have a knife.’ (To be faithful to the Likpe order: ‘A knife belongs to me’.) 36. (Ameka, personal communication) * mɔ 1SG

kú

lé-siabí

kpé

Com

CM-knife

be.at

‘I and knife belong to each other.’

Ameka also reports Figure–Ground indeterminacy in another Likpe verb, t k , ‘to make surface-contact’, as in (37a,b). And again, a subject with conjoined NPs is not possible, as in (38). 37. (Ameka, 1999: 29) a. kɔ´pu

ə-m ´ ə´

cup

Agr-Det

t ək ´ ə´

ku´-tsε

make.surface.contact

CM-crack

‘The cup has a crack.’ b. kú-tsε CM-crack

t əkə ´ ´

kɔ´pu

ə-m ´ ə´

make.surface.contact

cup

Agr-Det

‘There is a crack on the cup.’ 38. (Ameka, personal communication) *kɔ´pu cup

ə-m ´ ə´

kú

kú-tsε

t ək ´ ə´

Agr-Det

Com

CM-crack

make.surface.contact

‘The cup and the crack make surface contact with each other.’

Thus, various languages exhibit Figure–Ground indeterminacy in verbs referring to different kinds of events. More importantly, two verbs in two different

106

KITA

languages that can describe roughly the same set of events may differ with respect to Figure–Ground indeterminacy. A case in point is the contrast between the Yukatek ‘enter’ verb, which has Figure–Ground indeterminacy, and the Japanese ‘enter’ verb, which does not. This indicates that whether a certain participant role (e.g., the theme role for an ‘enter’ verb) must appear as Figure or Ground does not fall out automatically from the content of the underlying predicate for a verb (e.g., ENTER). The specifications of linking of participant roles to Figure and Ground have to be explicitly stated for each verb. However, it is still an open question whether crosslinguistic generalizations can be made as to what type of predicates tend to exhibit Figure–Ground indeterminacy. If a freely moving object is universally likely to be construed as Figure (see Table 5.1), it is, for example, possible that verbs for piercing-type situations are crosslinguistically more likely to exhibit Figure–Ground indeterminacy than verbs for entering-type situations. This is because in piercing-type situations, two objects constrain each other’s movement whereas in entering-type situations, the container constrains the containee’s movement but not vice versa. Another question that arises is whether languages with certain typological features are more likely to have verbs with Figure–Ground indeterminacy. Unfortunately, documentation of Figure–Ground indeterminacy is currently too limited to further pursue these questions. 5. IMPLICATIONS FOR LANGUAGE ACQUISITION AND CONCLUDING REMARKS What implications does Figure–Ground indeterminacy have for language acquisition? It raises the classic question of how children constrain the use of a verb when very little negative evidence (i.e., information about what not to do) is provided (what Pinker, 1989: 7–8, calls “Baker’s paradox”). For example, how do children learn (without negative evidence) that (39b) is not acceptable, given that they hear (40a), (40b), and (39a)? 39. a. John put the paint on the wall *b. John put the wall with the paint 40. a. John sprayed the paint on the wall b. John sprayed the wall with the paint

An analogous question with the same logical structure can be raised for the Japanese piercing verbs. How do Japanese children learn (without negative evidence) that (3) is infelicitous, given that they hear (1) and (5a,b)(repeated here)?

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS 3. #eki-ga station-Nom

densha-ni

tsui-ta

train-Loc

arrive-Pst

107

‘The station arrived at the train.’ 1. densha-ga train-Nom

eki-ni

tsui-ta

station-Loc

arrive-Pst

‘The train (Figure) arrived at the station (Ground).’ 5. a. niku-ga meat-Nom

kushi-ni

sasat-ta

skewer-Loc

pierce(intr)-Pst

‘The meat (Figure), the skewer (Ground) pierced.’ b. kushi-ga skewer-Nom

niku-ni

sasat-ta

meat-Loc

pierce(intr)-Pst

‘The skewer (Figure) pierced the meat (Ground).’

Although the Japanese problem has the same logical structure as the English problem, there are some important differences between the two. First, the semantics of verbs alone cannot fully predict Figure–Ground indeterminacy (cf. the ‘entering’ verbs in Japanese vs. Yukatek), so it is not possible to posit innate universal knowledge about what kind of verbs exhibit Figure–Ground indeterminacy. In addition, Figure–Ground indeterminacy is confined to a small set of verbs. It is an exception, rather than a general pattern. Thus, children learning Japanese are unlikely to make the assumption that all the verbs compatible with the Japanese Locative Change Construction exhibit Figure– Ground indeterminacy. This suggests the hypothesis that children may be extremely conservative in generalizing characteristics of a verb to other verbs under certain circumstances. More specifically, children may generalize a characteristic for a set of verbs to a new verb only when they see a sufficiently large number of verbs participating in the pattern of form–meaning pairing. Japanese verbs with Figure–Ground indeterminacy may not reach this critical mass to trigger generalization by children. If this is the case, the pattern we see in (5a) and (5b) would be learned for that specific verb based on positive evidence, and would not spread to other verbs. This line of thinking leaves some open questions. What counts as a sufficiently large base to trigger generalization? Does the critical mass depend on the nature of the form–meaning pairing (e.g., the degree of transparency of the pairing)? Figure–Ground indeterminacy highlights a general question to be investigated in the future: what mechanisms prevent children from getting caught up in numerous highly localized patterns in a language, but ensure that they do not miss patterns that are important for making generalizations?

108

KITA

ACKNOWLEDGMENTS I thank Felix Ameka and Jürgen Bohnemeyer for their insights into the phenomenon discussed in this chapter and for providing me with published and unpublished examples in the languages they study. I also thank Melissa Bowerman and Penny Brown for their comments on an earlier version of the chapter. I also benefited from valuable comments from the members of the MPI Argument Structure Group and the participants of the Argument Structure Workshop at the MPI in 1998. NOTES 1

The following abbreviations are used: A, Cross-reference set A (for Yukatek; see Bohnemeyer, 2002 and this volume, for further information on Yukatek in general, and the glossing system for the Yukatek examples cited in this chapter); Abl, Ablative; Acc, Accusative; Agr, Agreement; B, Cross-reference set B (for Yukatek); Caus, Causative (for Yukatek); Com, Comitative (for Likpe); CM, Noun class marker (for Likpe); CMP, Completive (for Yukatek); Dobj, Direct object; D2, Distal deixis (for Yukatek); Def, Definite (for Yukatek); Det, Determiner (for Likpe); (intr), intransitive; Loc, Locative; Nom, Nominative; Obl, Oblique; Pst, Past; Prs, Present; PRV, Perfective; SG, Singular; (tr), transitive. 2 For the sake of simplicity, Figure 5.2 and the other figures in this chapter do not make the three distinctions that Goldberg made in her original figure. First, Goldberg’s figure distinguished profiled and nonprofiled argument roles of a construction (the profiled roles are linked to either the subject, the direct object, or the indirect object). Second, Goldberg’s figure distinguished argument roles that are obligatorily fused with a participant role of a verb and those that are not. Finally, Goldberg’s figure specified possible semantic relationships, such as “an instance of,” between the semantic predicate of the construction and that of the verb to be fused with the construction. 3 The construction is not called “Caused Motion” because this Japanese construction does not entail that the Figure moves. It only entails that the locative relationship between Figure and Ground changes. See Kita (1999, 2006) for more details. 4 We do not include features such as [+long] and [+flexible] as part of the lexical representation for current purposes because such physical features do not have direct impact on what kind of constructions the verb can fuse with. 5 This gloss is based on personal communication from Penny Brown. In Brown (1994: 760), pach is glossed as ‘be located’ of wide-mouthed container canonically ‘sitting’. 6 Note, however, that Figure–Ground indeterminacy for òok is accepted only by roughly the half of the consultants Bohnemeyer worked with (Bohnemeyer, personal communication). 7 This verb has several uses: locative (where it has the ‘be.contained’ reading), existential, and possessive functions.

5.

FIGURE–GROUND INDETERMINACY IN DESCRIPTIONS

109

REFERENCES Ameka, F. K. (1999). Spatial information packaging in Ewe and Likpe: A comparative perspective. In S. Neumann (Ed.), Comparing African spaces (FAB 11) (pp. 7–34). Cologne, Germany: Rudiger Köppe. Anderson, S. R. (1971). On the role of deep structure in semantic interpretation. Foundations of Language, 6, 197–219. Bohnemeyer, J. (1997). Yucatec Mayan lexicalization patterns in time and space. In M. Biemans & J. v.d. Weijer (Eds.), Proceedings of the CLS opening academic year 1997–1998 (pp. 73–106). Tilburg, The Netherlands: Center for Language Studies. Bohnemeyer, J. (2002). The grammar of time reference in Yukatek Maya. Munich, Germany: Lincom Europe. Brown, P. (1994). The INs and ONs of Tzeltal locative expressions: The semantics of static descriptions of location. Linguistics, 32, 743–790. Fillmore, C. J., Kay, P., & O’Connor, M. C. (1988). Regularity and idiomaticity in grammatical constructions: The case of let alone. Language, 64, 501–538. Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Kita, S. (1999). Japanese Enter/Exit verbs without motion semantics. Studies in Language, 23, 317–340. Kita, S. (2006). A grammar of space in Japanese. In D. Wilkins & S. C. Levinson (Eds.), Grammars of space (pp. 437–474). Cambridge: Cambridge University Press. Levinson, S. C. (2002). Motion verb stimulus, version 2. In S. C. Levinson & N. Enfield (Eds.), Manual for the field season 2001 (pp. 9–12). Nijmegen, The Netherlands: Max Planck Institute for Psycholinguistics. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Rappaport, M., & Levin, B. (1988). What to do with theta-roles. In W. Wilkins (Ed.), Syntax and semantics, Volume 21: Thematic relations (pp. 7–36). San Diego: Academic Press. Talmy, L. (2000). Toward a cognitive semantics, Vol. 1: Concept structuring systems. Cambridge, MA: MIT Press.

CHAPTER 6

Learning Verbs without Boots and Straps? The Problem of ‘Give’ in Saliba Anna Margetts Monash University

1. INTRODUCTION Expressions of ‘giving’have repeatedly been the focus of linguistic research and discussion. Besides Dixon’s (1982) classic study of nuclear and non-nuclear verbs, ‘giving’ expressions have been the subject of detailed semantic analysis and crosslinguistic comparison in the recent literature (Newman, 1996, 1997). They are part of basic vocabulary, and through their frequent use they show morphological irregularities in many cases. Saliba contributes an unusual case of stem suppletion to the ongoing discussion, and provides a counterexample to the crosslinguistically common yet not universal mapping between ‘transfer’events and ditransitive verbs. Saliba is a Western Oceanic language of the Suauic family that belongs to the Papuan Tip Cluster (Ross, 1988: 190–212). It is spoken by about 1,000 people on Saliba Island at the Eastern tip of Papua New Guinea and it is acquired as children’s first language. Saliba is a nominative-accusative and head-marking language with SOV word order for lexical elements. The verb obligatorily carries a subject prefix and, if it is transitive, an object suffix.1 In Saliba, two incomplete paradigms of ‘give’ verbs complement each other to form a single suppletive paradigm; that is, similarly to English go–went, they form a conjugation paradigm whose members differ in sound shape and which are historically derived from independent verbs. The choice of verb depends on the person of the recipient. ‘Giving’ events with a first- or second-person recipient are expressed by one verb, whereas ‘giving’ to a third person is expressed by another. The two verbs differ in argument structure. One is transitive; the other may be transitive or ditransitive. In addition, the two ‘give’ verbs encode different participants with their object suffix: one verb always encodes the theme, and the other can encode the theme or the recipient. 111

112

MARGETTS

The Saliba data raise some interesting questions about the definition of the terms “paradigm” and “suppletion”. In addition, the investigation of the Saliba ‘give’ verbs may have implications for current theories about language acquisition, in particular for the two “bootstrapping” hypotheses entertained most prominently by Gleitman (1990) and Pinker (1987, 1989). The Saliba paradigm of ‘give’could provide a test case for the accuracy of these theories, in that the two “bootstrapping” proposals predict different error patterns in the acquisition of this paradigm. 2. THE PARADIGM OF ‘GIVE’ Searching for a translation equivalent of English ‘give’in Saliba, we need to consider at least three verbs. Each of them is incomplete in its expression of ‘giving’and forms only a partial conjugation paradigm. Two of the three verbs can express the concept of ‘giving’only to a first- or second-person recipient. The third verb can express ‘giving’ only to a third person. As I discuss shortly, one of the three verbs is semantically distinct from the others. The other two verbs complement each other to build a suppletive paradigm of ‘give’, which is split according to the person of the recipient. The two verbs which express ‘giving’to a first or second person are le ‘give’and hai ‘take/get’; ‘giving’ to a third person is expressed by mose-i ‘give’. Speakers generally state that le ‘give’ and hai ‘take/get’ have the same meaning, and they are indeed interchangeable in many contexts. But there is a slight semantic difference between them and, as I discuss shortly, only le ‘give’ is a true suppletion partner of mose-i ‘give’. For comparison, a discussion of hai ‘take/get’ is included in sections 3.2 and 4.3. Evidence for the suppletive split in the paradigm of ‘give’ comes both from text examples and elicitation. The Saliba translation equivalents of English ‘give’, as in ‘he gave X one basket’, are presented in (1) to (4). ‘Giving’ events involving a first or second person recipient are expressed with le ‘give’plus a directional suffix, as in 2 (1) and (2). ‘Giving’ events with a third person recipient are expressed with mose-i ‘give’, as in (3) and (4), which obligatorily takes the applicative suffix. 1. Bosa basket

kesega ye

le-ya-ma3

one

give-3SG.O-hither

3SG

First-Person Recipient (SG/PL)

‘He gave me/us one basket.’ 2. Bosa basket

kesega ye

le-ya-wa.

one

give-3SG.O-to.addr

3SG

Second-Person Recipient (SG/PL)

‘He gave you (SG/PL) one basket.’ 3. Bosa basket

4

Third-Person Singular Recipient

kesega ye

mose-i-ø.

one

give-APP -3SG.O

3SG

‘He gave him/her one basket.’ 4. Bosa basket

kesega ye

mose-i-di.

one

give-APP -3PL.O

3SG

‘He gave them one basket.’

Third-Person Plural Recipient

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

113

The clauses in (1) and (2) with le ‘give’are transitive and encode the theme of the transfer as the object. The recipient is implied by the directional suffixes but not encoded as a syntactic argument. In contrast, the verbs in (3) and (4) are ditransitive: the recipient is encoded as an object argument by the pronominal suffix on the verb, the theme occurs as a second object, a bare NP without a postposition. 2.1. Problems: What Constitutes a Paradigm? Stem suppletion is crosslinguistically most commonly attested with high-frequency verbs, as for example, go–went or am–is in English. So in a way, it is not surprising to find a suppletive paradigm for ‘give’. But the Saliba case of suppletion is unusual in several respects. In fact, it is unusual enough that the suitability of the notions of “paradigm” and “suppletion” for the Saliba ‘give’verbs deserves discussion. A definition of “suppletion” needs to make reference both to the semantics and the shape of stems. Consider for example the following definitions: Stem suppletion can be identified as the phenomenon whereby two or more stems, phonologically disparate but synonymous, are distributed complementarily within a paradigm of related word forms. (Carstairs-McCarthy, 1994:4410) Suppletion is a relation between two segmental linguistic signs X and Y of language L such that the semantic difference between X and Y is maximally regular, while the formal difference between them is maximally irregular. (Mel’
uk, 1992:97)

Besides referring to both form and meaning, definitions of suppletion commonly refer to inflection. Carstairs-McCarthy (1994:4410) states that “[s]uppletion is usually considered to occur only in inflectional morphology,” and Bybee (1985: 91) describes suppletive paradigms as “inflectional paradigms that have forms built on two or more stems that are etymologically from different sources.” Thus, suppletive paradigms show an “inflectional split” (Bybee, 1985), that is, a split between the suppletive stems according to an inflectional category of the verb, for example, tense in the case of English go–went. One criterion that the Saliba ‘give’ verbs satisfy easily is the maximal irregularity of their formal difference. But there are a number of problems with calling the Saliba case a suppletive paradigm. The first problem is that the two verbs that are putative suppletion partners differ in argument structure. There is not merely a change of verb stems which differ in sound shape (and historically in meaning), as with go–went. In the Saliba case, the stem suppletion affects the overall structure of the clause and determines how many participants are expressed as syntactic arguments. This means that the suppletion is not restricted to the lexical level, but has an effect on the syntax of the clause. This is an unusual feature for suppletion in an inflectional paradigm.

114

MARGETTS

A second rare feature is that the Saliba paradigm of ‘give’ shows suppletion according to the person of the recipient. This is noteworthy, given Bybee’s (1985) work, which shows that suppletion along “the person agreement line” is crossling5 uistically the rarest case, and only sparsely attested in the languages of the world. A final problem is that the Saliba paradigm is split according to person agreement with the recipient, but le ‘give’ does not actually agree with the recipient participant. As I discuss shortly, the directional suffixes imply a person distinction of the recipient but they are not expressions of syntactic arguments. This means that the “inflectional split” depends on a category for which one of the verbs does not actually inflect. Despite these problems, there is evidence that the two verbs, le ‘give’and mose-i ‘give’, stand in a paradigmatic relationship and are true suppletion partners. The verbs are in complementary distribution and both exhibit some crucial morphological anomalies. Rather than being defective randomly, they exactly fill each other’s distributional gaps (cf. section 4). This speaks for an analysis of the verbs as members of a suppletive paradigm despite their difference in argument structure. Evidence in support of their paradigmatic relation also comes from the verbs’ semantics and use. 3. SEMANTICS AND USE When we are faced with two words that are morphologically different but seemingly synonymous, an obvious possibility is that there is, after all, a difference in meaning or possibly in dialect. Could the verbs le vs. mose-i refer to different types of ‘giving’ events? Crosslinguistically, semantic differences are attested between ‘give’verbs, for example, along the lines of Dixon’s (1982) distinction between nuclear and nonnuclear verbs, or according to register of speech and politeness (e.g., Martin, 1975: 352–354, on Japanese). However, as I show next, the Saliba ‘give’ verbs differ in neither of these parameters. 3.1. Le ‘give’ versus Mose-i ‘give’ The choice between le and mose-i is clearly not dialectal because the same speakers use both forms. Nor does Saliba have a refined system of politeness registers, and synchronically, it is not the case that one ‘give’ verb is more polite than the other. There is simply no stylistic choice, because the forms are in complementary distribution on grammatical grounds. Historically, politeness may well have played a role in promoting a preference for indexing first and second person participants indirectly, by means of directional suffixes, a strategy commonly found in Oceanic languages. But this still does not explain the suppletion of the ‘giving’ expressions, because other ‘transfer’ verbs (e.g., ‘send’, ‘carry’, and ‘lead’) also take directional suffixes to indicate a recipient but without showing an inflectional split.

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

115

Besides possible differences in dialect or politeness register, further semantic parameters must be considered. Among the parameters I have tested are animacy of the theme, temporary versus permanent transfer, Dixon’s (1982) distinction between nuclear and non-nuclear verbs, and event properties such as manner or completion of transfer. Both verbs, le and mose-i, typically describe a direct transfer of the theme from the previous controller/owner to the recipient. Sending objects by mail or via an intermediate person is not considered a good instance of le or mose-i ‘give’. In such cases the verb hetamali ‘send’ is preferred. In the default usage, both ‘give’ verbs describe transfer of an object into the recipient’s hands. In addition, the verbs do not differ in animacy requirements of the theme: both le ‘give’ and mose-i ‘give’ typically take inanimate themes, as in (1) to (4) shown earlier, but also allow animate themes, as in (5) and (6). 6 5. Natu-m child-2SG.P

ku

le-ya-ma

ya

2SG

give-3SG.O-hither 1SG

kitahetete-ø. look.after-3SG.O

‘Give me your child, I’ll look after it.’ 6. Natu-m

ku

child-2SG.P 2SG

mose-i-ø

ye

kitahetete-ø.

give-APP -3SG.O

3SG

look.after-3SG.O

‘Give her your child, she’ll look after it.’

Both verbs can express temporary or permanent transfer of the theme as in (7) and (8): 7. a. Bolo ne ball

DET

ku

le-ya-ma

memelahi kabo ku

hai-uyo-i-ø.

2SG give-3SG.O-hither afternoon TAM 2SG get-back-APP-3SG.O

‘Give me the ball, you’ll get it back in the afternoon.’ b. Bolo ne ball

DET

ku

le-ya-ma

yo-gu.

2SG give-3SG.O-hither CL1–1SG.P

‘Give me the ball (for me to keep).’ 8. a. Bolo ne ball

DET

ku

mose-i-ø

2SG give-APP-3SG.O

memelahi kabo ku

hai-uyo-i-ø.

afternoon TAM 2SG get-back-APP-3SG.O

‘Give him the ball, you’ll get it back in the afternoon.’ b. Bolo ne ball

DET

ku

mose-i-ø

2SG give-APP-3SG.O

yo-na. CL1-3SG

‘Give him the ball (for him to keep).’

Permanency of transfer can explicitly be expressed with the verbs kainauya-i ‘give as gift’, duwa-i ‘give in exchange for help’ (e.g., to a midwife), and kune ‘exchange important gifts’.7 These terms can be considered non-nuclear expressions of ‘giving’, whereas both le and mose-i are nuclear expressions (Dixon, 1982). It is

116

MARGETTS

a feature of non-nuclear expressions that they are semantically more specific and can be defined in terms of the semantically more general nuclear expressions. This is the case for the Saliba non-nuclear ‘give’ verbs just shown. For example, kune ‘exchange important gifts’ was explained in terms of both le and mose-i in the text example in (9). 9.

…

se

lao-ma

puwaka

se

le-ya-ma

3PL

go-hither

pig

3PL

give-3SG.O-hither

‘… they come and give us a pig, na

hinage

bena

puwaka

ta

mose-i-di.

CONJ

also

OBLIGAT

pig

1INC

give-APP-3PL.O

and then we are obliged to also give them a pig (in due course).’

Finally, apparently neither le nor mose-i ‘give’ entails successful transfer of the theme, although both clearly imply this by default. Examples (10) and (11) were considered correct, unproblematic sentences by Saliba speakers. 10. Bisikete biscuit

ye

le-ya-ma

na

nige

ya

hai-ø.

3SG

give-3SG.O-hither

CONJ

NEG

1SG

take/get-3SG.O

‘She gave me a biscuit but I didn’t take it.’ 11. Bisikete biscuit

ye

mose-i-ø

na

nige

ye

hai-ø.

3SG

give-APP -3SG.O

CONJ

NEG

3SG

take/get-3SG.O

‘She gave him a biscuit but he didn’t take it.’8

Except for the person distinction of the recipient, the two verbs do not show any semantic differences, and do indeed refer to the same type of ‘giving’ events. For instance, they can describe one and the same real-world event, as in example (12), where the difference between the expressions in (a) and (b) is basically one of perspective. A mother asks her daughter to give her an object and the grandmother, who is present, also addresses the child to express her agreement with the mother’s request. 12. Mother to child: a. Kanuwa sweet.potato

ku

le-ya-ma!

2SG

give-3SG.O- hither

‘Give me the sweet potato!’

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

117

Grandmother to child: b. Kanuwa sweet.potato

ku

mose-i-ø!

2SG

give-APP -3SG.O

‘Give her the sweet potato!’

The examples show that the same request is expressed with a different verb depending on whether the recipient is a speech act participant or not. Further evidence for le and mose-i ‘give’ being suppletion partners is that they allow the same semantic extensions and occur in the same metaphoric and idiomatic expressions. The text examples in (13) show that both verbs occur in an expression that corresponds to English ‘talk to someone in language X’; it translates literally as ‘give someone language X’. This constitutes a semantic extension of the literal ‘transfer’ meaning of ‘give’. It is worth mentioning that both (13a) and (13b) come from the same speaker and the same text, and they refer to the same realworld situation. In (13a) the speaker talks about the recipient in the third person, whereas in (13b) he takes the recipient’s perspective and switches to direct speech (as indicated by i-wane ‘she said’). 13. a. Tabu PROHIB

dimdim

kwa mose-mose-i-ø,

white.person 2PL

nige ye

RED-give-APP-3SG.O NEG

henuwa-ø …

3SG like-3SG.O

‘Don’t speak English with her, she doesn’t like it …’ (lit.: don’t give her white people[’s language] …) b. …i wane … nige ya 3SG say

NEG

henuwa-ø

dimdim

kwa le-ya-ma …

1SG like-3SG.O white.people 2PL give-3SG.O-hither

‘She said “I don’t want you to talk English to me….”’ (lit.: don’t give me white people[’s language] …)

To summarize, not only are le ‘give’ and mose-i ‘give’ in complementary distribution, but their semantics and use also support their analysis as suppletion partners in one paradigm. Both are nuclear expressions in terms of which non-nuclear ‘give’ verbs are defined; they show no difference in the animacy requirements of the theme; both can express temporary or permanent transfer; neither entails that the transfer is completed; and both verbs occur in the same semantic extensions. The verbs differ essentially in that they describe events from different perspectives. 3.2. Hai ‘take/get’ As mentioned, there is a verb hai ‘take/get’ which speakers consider synonymous with le ‘give’in certain contexts. Like le, this verb can only express ‘giving’to a first

118

MARGETTS

or second person. But there is a slight semantic difference between hai, on the one hand, and le and mose-i, on the other. The verb hai ‘take/get’ is preferred in situations where the theme is out of the agent’s reach. There is a connotation of ‘fetching’ the object before the actual ‘giving’ event takes place. In contrast, le ‘give’ is preferred when the agent already holds the theme or has it within reach. Consider example (14): 14. a. Nigwa

ku

b. Nigwa

hai-ya-ma!

2SG take/get-3SG.O-hither

knife

knife

‘(Go and) get me the knife!’

ku

le-ya-ma!

2SG give-3SG.O-hither

‘Give me the knife!’

But this seems to be a connotation rather than an entailment because hai ‘take/get’ can also be used when the theme is within the agent’s reach. It is interchangeable with le ‘give’ even in (15), where the theme nima-m ‘your hand’ is the agent’s body part. 15. a. Nima-m hand-2SG.P

ku

le-ya-ma

ya

mulamula-i-ø.

2SG

give-3SG.O-hither

1SG

medicine-APP-3SG.O

ya

mulamula-i-ø.

‘Give me your hand, I’ll put medicine on it.’ b. Nima-m hand-2SG.P

ku

hai-ya-ma

2SG

take/get-3SG.O-hither 1SG

medicine-APP-3SG.O

‘Give me your hand, I’ll put medicine on it.’

Like le ‘give’, the verb mose-i ‘give’ implies that the theme is within the agent’s reach. To express a situation with a third-person recipient in which the theme is outside the agent’s reach, two clauses are needed, as in (16). 16. Nigwa knife

ku

hai-ø

ku

mose-i-ø.

2SG

take/get-3SG.O

2SG

give-APP -3SG.O

‘Get the knife (and) give it to him.’

The text example in (17) demonstrates the same strategy: First the ‘getting’ of the object is described by hai ‘take/get’ and then the ‘giving’ itself is expressed by mose-i ‘give’. 17. kaikaiwa ye stick

3SG

hai-ø

ye

lao-ma

ede

take/get-3SG.O

3SG

go-hither

PRSP

yo-na

golowa

wa

ye

mose-i-ø.

CL1 – 3 SG.O

younger.brother

GIVEN

3SG

give -APP- 3SG.O

‘He got a stick, came and gave it to his small brother.’

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

119

To summarize, the verbs hai ‘take/get’and le ‘give’are not truly in free variation. They differ semantically and are not part of the same paradigm. 4. ARGUMENT STRUCTURE OF ‘GIVE’ VERBS The argument structure of the Saliba ‘give’ verbs and their morphological anomalies deserve discussion. All three verbs of ‘giving’ can indicate three participants, an agent, a theme, and a recipient, but they differ in how these participants are encoded. First, there is variation in which participant—the recipient or the theme—is encoded by the verb’s object suffix. Second, although the agent and the theme are always encoded as syntactic arguments, the recipient may be expressed as an argument or oblique, or merely be pragmatically implicated by a directional suffix. As the distinction between arguments and obliques plays a role in the following discussion, I briefly outline the relevant criteria. In Saliba, arguments can be distinguished from obliques by the lack of postpositional markers and, in the case of inner-core arguments (Margetts, 1999), by pronominal cross-referencing on the verb. Topicalization and relativization do not establish argumenthood, and there is no morphological passive. Saliba object marking in ditransitive clauses is sensitive to the distinction between primary and secondary objects (Dryer, 1986). The primary object is expressed by the object suffix on the verb, whereas the secondary object is not cross-referenced, but may appear in the clause as a bare NP, without a postposition. Both primary and secondary objects are syntactic arguments of the verb. 4.1. ‘Giving’ to a Third Person: Mose-i ‘give’ ‘Giving’ events that involve a third person recipient are expressed by mose-i ‘give’. This verb can occur in alternative syntactic frames, reminiscent of the “dative alternation” in English. The verb is flexible as to which participant, the recipient or the theme, is encoded by its object suffix. 4.1.1. Recipient as Object When mose-i ‘give’ encodes the recipient with its object suffix, the verb is ditransitive and both the recipient and the theme are expressed as core arguments. The agent and the recipient are inner-core arguments, marked by the pronominal affixes (and optionally by lexical NPs). The theme is the secondary object and is encoded as an outer-core argument and not indicated on the verb (Margetts, 1999). Clauses in which mose-i encodes the recipient as its primary object are represented in Schema 6.1.

120

MARGETTS

SCHEMA 6.1. Ditransitive clauses with mose-i ‘give’, object suffix = recipient.

An example is given in (18), where the theme object is expressed by an overt NP. 18. Bosa 3 basket

kesega

ye1

mose-i-di2 .

one

3SG

give-APP -3PL.O

‘He gave them one basket.’

Most Saliba ditransitive verbs are derived from transitive verbs by the causative prefix and show the same syntactic frame as (18). Consider the clause in (19): 19. Tautau picture

ne

ye

he-kita-gau.

DET

3SG

CAUS -see-1 SG.O

‘He showed me the picture.’

Ditransitive verbs derived by the causative prefix can only occur in the frame represented in Schema 6.1, they do not allow syntactic variation. By analogy with other ditransitive verbs, Schema 6.1 can therefore be considered the main syntactic frame of mose-i ‘give’.9 4.1.2. Theme as Object When mose-i ‘give’encodes the theme with its object suffix, there are two possibilities for encoding the recipient. These are represented in Schemas 6.2(a) and 6.2(b). In Schema 6.2(a) the recipient is expressed as an oblique with the general postposition unai, which also marks locations, instruments, and so on. In Schema 6.2(b) the recipient appears as an outer-core argument, a bare NP without a postposition that is not cross-referenced on the verb. Following Schema 6.2(a), the clause in (20a) shows the recipient as an oblique marked by the postposition unai. The verb is transitive and has two arguments. The clause in (20b) expresses the recipient as a syntactic argument—following Schema

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

121

SCHEMA 6.2(a). Transitive clauses with mose-i ‘give’, object suffix = theme.

SCHEMA 6.2(b). Ditransitive clauses with mose-i ‘give’, object suffix = theme.

6.2(b)—and the verb is ditransitive. There is no clear difference in meaning between the two clauses. 20.

Bosa

labui

wa

haedi?

Question:

basket

two

GIVEN

where

‘Where are the two baskets?’ a. Ya 1SG

mose-i-di-ko

ka-gu

kaha

wa

unai.10

give-APP-3PL.O-PERF

CL 2-1 SG.P

sibling

GIVEN

PP.SG

mose-i-di-ko

ka-gu

kaha

wa.

give-APP-3PL.O-PERF

CL 2-1 SG.P

sibling

GIVEN

‘I gave them to my sister.’ b. Ya 1SG

‘I gave them to my sister.’

122

MARGETTS

In (20b), the secondary object occurs following the verb in a position typical for obliques. The recipient in Schema 6.2(b) thus shares its lack of morphological marking with canonical objects, but its position in the clause with obliques. Schema 6.2(b) is exceptional in that mose-i ‘give’ is the only verb occurring in this frame. One other verb, kainauya-i ‘give as a gift’, allows the alternation between Schema 6.1 and Schema 6.2(a), but it cannot occur in the frame of Schema 6.2(b). The recipient of kainauya-i ‘give as a gift’ must either be marked as a primary object or as an oblique with a postposition. It may not occur as a secondary object. 21. Ya 1SG

kainauya-i-di

ka-gu

kaha

wa

*(unai).

make.gift-APP- 3PL.O

CL2-1 SG.P

sibling

GIVEN

PP.SG

‘I gave them (as a gift) to my sister.’

This suggests that Schema 6.2(b) must be considered a separate syntactic frame rather than merely an instance of dropping the postposition unai. Otherwise one would expect kainauya-i ‘give as a gift’ to allow the same variation (with and without the postposition). Note also that other semantic roles marked by unai, such as instrument or source, may not occur without the postposition. In examples from natural discourse it is typically impossible to determine in which syntactic frame mose-i ‘give’ appears. The frames can only be distinguished through the lexical expression of the objects, and often neither the theme nor the recipient is expressed lexically in the clause. Even when they are, it is not necessarily transparent which one is cross-referenced on the verb because theme and recipient of mose-i ‘give’ are both third person and cross-referenced by the same morpheme. Even if the two participants differ in number, the cross-referencing on the verb may be ambiguous, because Saliba, like other Oceanic languages, allows certain plural objects to be cross-referenced by singular suffixes.11 However, when mose-i ‘give’ occurs as part of a complex verb with two stems, the verb is transitive and must follow Schema 6.2(a). Only theme participants may occur as the objects of such verbs. Examples with the complex stems mose-gabae ‘give away’ and mose-uyo-i ‘give back’ are given in (22) and (23). 22. Moni money

ye

mose-gabae-ø.

3SG

give-away-3SG.O

‘He gave the money away.’

23. Moni

ye

mose-uyo-i-ø.

money 3SG give-back-APP-3SG.O ‘He returned the money.’

To summarize, mose-i ‘give’ is ditransitive when it encodes the recipient as the primary and the theme as the secondary object. In contrast, when the theme is the primary object, the verb can be either transitive or ditransitive.

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

123

4.1.3. Morphological Anomalies of Mose-i ‘give’ As mentioned, mose-i ‘give’ is morphologically defective in several respects. The verb can be analyzed into the simplex stem mose plus the applicative suffix -i, but mose never occurs as a simplex stem without the applicative and example (24) is ungrammatical: 24. *Ye 3SG

mose. give

‘He gave.’

Mose is one of a few stems that take the applicative suffix obligatorily. That the final -i vowel is indeed the applicative and not part of the verb root can be seen in complex verbs as in (22) and (23), shown earlier, where mose is followed by a second stem. The applicative suffix is sanctioned only on the last stem of a complex verb (Margetts, 1999). A further anomaly is that the applicative suffix generally derives transitive verbs but not ditransitive ones. Input to suffixation with the applicative are intransitive verbs as in (25), or nouns as in (26), but not transitive verbs. 25. a. Ye 3SG

lupo.

b. Ye

trick/lie

3SG

‘He lied.’ 26. a. Gado-gu

lupo-i-go. trick/lie-APP-2SG.O

‘He lied to you.’ ye

magu.

throat-1SG.P 3SG low.tide

b. Bagi necklace

ye

gado-i-ø.

3SG throat/neck-APP-3SG.O

‘I’m thirsty.’

‘She put on a bagi (necklace).’

(lit.: ‘My throat has low tide.’)

(lit.: ‘She throated a bagi.’)

Thus, morphologically, mose ‘give’ behaves like an intransitive verb or a noun stem in that it can take the applicative, but distributionally, mose-i ‘give’ behaves like a ditransitive verb occurring in clauses with three syntactic arguments. Because its morphological and distributional features do not align, the underived stem mose ‘give’ is basically unclassifiable in terms of valency. Besides these anomalies concerning the applicative suffix, mose-i ‘give’ is morphologically defective in not taking first or second person object suffixes or the directional suffixes -ma ‘hither’ and -wa ‘toward addressee’. Most Saliba verbs that imply any kind of motion or transfer can take these suffixes. Speakers rejected sentences like (27) and (28) and instead suggested clauses with le ‘give’.

124 27.

MARGETTS *Ye 1SG

mose-i-gau.

28.

*Ya

give-APP -1SG.O

1SG

‘He gave (it to) me.’

mose-i-ya-wa. give-APP -3SG.O-to.addr

‘I gave it to you.’

To summarize, mose-i ‘give’ deviates from the general morphological pattern of Saliba verbs in several ways. First, the simplex stem mose obligatorily takes the applicative suffix, and the underived stem cannot be classified as intransitive or transitive. Second, mose-i ‘give’ does not allow first- or second-person object suffixes. Third, the verb cannot take the directional suffixes, whereas transfer verbs generally can. 4.2. ‘Giving’ to First or Second Person: Le ‘give’ ‘Giving’ events with first- or second-person recipients are expressed by le ‘give’ plus a directional suffix. As opposed to mose-i ‘give’, le ‘give’ can occur in only one syntactic frame: it invariably encodes the theme with its object suffix and the verb is transitive but never ditransitive. The recipient is not a syntactic argument of the verb and is only implied by the directional suffix (section 5). The recipient may explicitly be encoded, but only as an oblique marked by a postposition. ‘Giving’ expressions with le are represented in Schema 6.3. Examples are given in (29) and (30).

SCHEMA 6.3. Transitive clauses with le ‘give’, object suffix = theme.

29. Aee INTRJ

taba

bihiya-ne 2

kesega

ku1

le-ya 2 -ma …

if

banana-DET

one

2SG

give-3SG.O-hither

‘Oh, if you could give me one banana …’ 30. Kaiteya 1

ye 1

le-ya 2 -wa

kali-m3 -wai?

who

3SG

give-3SG.O-to.addr

toward-2SG.P -to

‘Who gave it to you?’

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

125

4.2.1. Morphological Anomalies of Le ‘give’ The form le ‘give’ is morphologically defective in two respects: first, it allows only third-person object suffixes; second, it cannot occur without a directional marker. The examples in (31a) and (32a) show that first- and second-person object suffixes are ungrammatical. This is true even if they are followed by a directional marker, as in the (b) versions. Such a gap in object inflection is otherwise not attested for transfer verbs. 31. a.

* Ye 3SG

le-gau.

b.

give-1SG.O

*Ye 3SG

le-gau-ma. give-1SG.O-hither

‘He gave me.’ 32. a. * Ye 3SG

le-go.

b.

give-2SG.O

*Ye 3SG

le-go-wa. give-2SG.O- to.addr

‘He gave you.’

Examples (33) and (34) show that le ‘give’ cannot occur without a directional suffix. 33. a.

*Ye 3SG

le-ø.

b.

give-3SG.O

‘He gave it.’ 34. a.

*Ye 3SG

le-di. give-3PL.O

‘He gave them.’

Ye

le-ya-ma.12

3SG

give-3SG.O-hither

‘He gave it to me.’ b.

Ye

le-di-wa.

3SG

give-3PL.O-to.addr

‘He gave them to you.’

Saliba verb stems are typically disyllabic. Because le ‘give’ has only one syllable, it could be that it requires a suffix in order to maintain a certain syllable structure or intonation pattern. But the unacceptability of (35) with the perfect suffix -ko shows that it is not merely a phonological constraint that rules out examples (33a) and (34a). The stem does not call for just any suffix, but specifically requires the directional markers. 35.

*Ye

le-di-ko.

3SG

give-3PL.O-PERF

‘He gave them already.’

This constraint that le ‘give’needs to take a directional suffix can also be observed in complex verbs where two or more stems combine to form a morphologically

126

MARGETTS

complex verb with a single set of inflectional affixes. The form uyo ‘back/again’(and its transitivized version uyo-i) may occur as the last stem in a complex verb, as in (36), where it follows hetamali ‘send’. The compound stem is followed by the object suffix and the directional -ma ‘hither’. 36. Ye

hetamali-uyo-i-ya-ma.

3SG

send-back-APP- 3SG.O-hither

‘He sent it back to me.’

In contrast, le ‘give’ allows uyo-i ‘back/again’ only as a particle following the verb, as in (37a). The ungrammatical example in (37b) shows that le ‘give’ cannot build a complex verb parallel to (36) in which uyo-i ‘back/again’ intervenes between le and the directional suffix.13 37. a. Ye 3SG

le-ya-ma

uyo-i.

give-3SG.O-hither back-APP

‘He gave it back to me.’

b.* Ye

le-uyo-i-ya-ma.

3SG give-back-APP-3SG.O-hither ‘He gave it back to me.’

The bond between le ‘give’ and the directional suffix seems to be closer than for other verbs. Further evidence for this tight bond comes from speech errors. On two different occasions, a small child (age ca. 2;6) and an adult speaker (perhaps after rather too many hours of elicitation) produced the ungrammatical verb in (38a). Although the child was corrected by another speaker, the adult speaker corrected herself shortly after producing the verb. In both cases, the fully inflected verb, including the singular object suffix, was suffixed with an additional plural object suffix -di. The verb resulting from the speech error has two object suffixes and is ungrammatical. The correct form is given in (38b). 38. a. *Ku 2SG

le-ya-ma-di! give-3SG.O- hither-3PL.O

‘Give them to me!’

b. Ku 2SG

le-di-ma! give-3PL.O- hither

‘Give them to me!’

It appears that in these errors speakers reanalyze the sequence le-ya-ma ‘give-it-hither’as leyama-ø ‘give-it’, analogously to verbs like (39a). The sequence of verb stem, object suffix, and directional is reinterpreted as a simplex transitive stem carrying the zero allomorph of the third-person singular object suffix. The zero suffix is then replaced by -di in the plural, analogously to (39b). 39. a. Ku 2SG

kita-ø. see-3SG.O

‘You saw it.’

b. Ku 2SG

kita-di. see-3PL.O

‘You saw them.’

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

127

Because there are only two instances of such errors in my data, the evidence is rather anecdotal; nevertheless, the errors reinforce the evidence that le ‘give’ has a closer bond to the directional suffix than other verbs do. It appears that the combination of le ‘give’ plus the directional suffix tends to be treated as a lexical unit. 4.3. The Verb Hai ‘take/get’ Like le ‘give’, the verb hai ‘take/get’expresses transfer of an object and encodes the theme participant with its object suffix. To express the concept of ‘giving’it needs a directional suffix, as in (40) and (41). The examples have the same structure as the ‘giving’ expressions with le ‘give’ represented in Schema 6.3, shown earlier. 40. Oh INTRJ

kagutoki

ka-gu

labiya

ko

hai-ya-ma, …

my.thanks

CL2-1 SG.P

sago

2SG

take/get-3SG.O-hither

‘Oh thank you, you gave me my sago, …’ 41. Ka-m CL2-2SG.P

keyaka

ya hai-ya-wa,

unai kabo ku

numa…

coconut.bowl 1S take/get-3SG.O-to.addr PP.SG TAM 2SG drink

‘I’ll give you a coconut bowl and you’ll drink from it, …’

Like le ‘give’, hai ‘take/get’ can express ‘giving’ only to a first- or second-person, not a third-person, recipient. This is so even though hai ‘take/get’ is not morphologically defective in any way. In contrast to le ‘give’, hai allows object suffixes of all person distinctions and it can occur without a directional suffix, as shown in (42). 42. Tukeli cough

ye

hai-go.

3SG

take/get-2SG.O

‘You got a cough.’ (lit.: ‘cough got you.’)

The verb hai ‘take/get’ behaves morphologically exactly in parallel to other verbs of transfer. So why can it not express ‘giving’to a third person? There is a crucial semantic difference between hai ‘take/get’ and other transfer verbs: hai ‘take/get’ entails transfer ‘to subject’. If not otherwise specified, the subject will be interpreted as the final recipient of the transfer, as in (43) and (44). 14 43. Wau now

maimaina

wa

ka

hai-ø

vine

GIVEN

1EX

take/get-3SG.O 1EX pound-3SG.O

‘We take the vine and pound it …’

ka

tutu-ø …

128

MARGETTS

44. … ka 1EX

lao

ka

keli

kabo

lam-na

go

1EX dig

TAM

root-3SG.P GIVEN 1EX take/get-3PL.O

wa

ka

hai-di …

‘… we go and dig and get the roots …’

The concept of ‘giving’—that is, the transfer of the theme to a participant other than the subject—can only be expressed if the verb carries a directional marker as in (40) and (41). Other transfer verbs, such as hetamali ‘send’ and bahei ‘carry’, do not have an entailment of transfer to subject. For these verbs, the absence of a directional suffix tends to be interpreted as directionality ‘to third person’, as in (45) (cf. section 5). 45. Yo-na CL1-3 SG.P

letter

ya

hetamali-ya-ko.

letter

1SG

send-3SG.O-PERF

‘I sent her a letter already.’

For hai ‘take/get’ the absence of a directional suffix is interpreted as directionality ‘to subject’ rather than ‘to third person’. Compare (46) with (45): 46. Yo-na CL1-3 SG.P

letter

ya

hai-ya-ko.

letter

1SG

take/get-3SG.O-PERF

‘I got her letter already.’ *‘I gave her the letter already.’

The fact that other transfer verbs can express transfer to a third person whereas hai ‘take/get’ cannot is thus due not to any morphological anomalies but to a different interpretation of the absence of a directional marker. 5. DIRECTIONAL SUFFIXES The directional suffixes obviously play an important role in the Saliba expression of ‘give’. So, how are they used and what do they do? Primarily, they express the directionality of an action or event. But, as I discuss shortly, in certain contexts they can be interpreted as expressing a person distinction, similar to pronominal suffixes (see also Margetts, 2003). The semantic entailment of the Saliba directionals can be described as ‘hither, toward speaker’15 for -ma and ‘toward addressee’ for -wa. Consider (47) to (49): 47. a. Ku 2SG

lao-ma! go-hither

‘Come (here)!’

b. Aa, INTRJ

ya

lao-wa.

1SG

go-to.addr

‘Yes, I’m coming (to you).’

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA 48. a. Ye 3SG

b. Ye

kita-lao-ma. see-go-hither

3PL

hedede-lao-wa.

3SG

‘He looked over to me.’ 49. a. Se

129

b. Se

bahe-i-ya-ma. carry-APP -3SG.O-hither

‘They brought it.’

talk-go-to.addr

‘He told you.’ bahe-i-ya-wa.

3PL

carry-APP -3SG.O-to.addr

‘They brought it to you.’

The suffixes entail directionality TOWARD but not necessarily TO the speaker or addressee. With second person subjects the addressee-based form -wa ‘toward addressee’ can also express motion away from the speaker, as in (50).16 50. Kwa 2PL

wasabu

kwa

lao-wa.

go.away

2PL

go-to.addr

‘Clear off, go away!’

When speaker and addressee are positioned in the same location, the speaker-based form has preference over the addressee-based one. A motion event toward both speaker and addressee will be described by -ma ‘hither’rather than -wa ‘toward addressee’.17 Apart from the binary contrast between ‘toward speaker’ and ‘toward addressee’, the suffixes also engage in a three-way contrast. In this case the absence of a directional marker is interpreted as the third member of the opposition. It can express directionality ‘away from speaker’ as in (51) (especially with motion verbs), or ‘toward third person’ as in (52) (with verbs involving an addressee or recipient). Note that the absence of a directional suffix is meaningful only through the contrast with -ma and -wa. 51. Se 3PL

lao-ko. go-PERF

‘They left already.’

52. Ya 1SG

hedede-lao. tell-go

‘I told her.’

Often the only reference to an event participant in a clause is by means of a directional suffix. The Saliba directionals can thus be used to evoke the notion of an event participant which is not lexically expressed, either as a syntactic argument or as an oblique. Verb semantics determine the semantic role of the implicated participant. With intransitive verbs, the participant implicated by a directional suffix is typically an addressee or stimulus, as in (48), shown earlier. With transitive verbs, a directional suffix hints at the existence and identity of a recipient. Examples of this are the

130

MARGETTS

‘giving’expressions with le ‘give’and hai ‘take/get’discussed in sections 4.2 and 4.3 and the clauses with ‘bring’ in (49). The examples in (53) demonstrate the three-way opposition between -ma ‘toward first person’, -wa ‘toward second person’, and the absence of a directional, which is interpreted as ‘toward third person’. 53. a. Leta wa ye letter GIVEN 3SG ‘He sent the letter to me/us.’ b. Leta wa ye letter GIVEN 3SG ‘He sent the letter to you.’ c. Leta wa ye letter GIVEN 3SG ‘He sent the letter to her.’

hetamali-ya-ma. send-3SG.O-hither hetamali-ya-wa. send-3SG.O-to.addr hetamali-ø send-3SG.O

The interpretation of directionality toward a third person in (53c) can be strengthened by adding the general locative postposition unai ‘in/at/to third person singular’, which is, however, not obligatory. If the recipient is overtly expressed, it occurs as an oblique marked by unai, or, in the case of first and second person recipients, by the circumposition kali- -wai following the verb. Other transfer verbs follow the same pattern. They are transitive, take the theme as their object, and encode the recipient indirectly by means of the directional suffixes (or their absence). The distinction between a recipient and a goal depends on verb semantics and the context of the clause. 54. Ku bahe-i-ya-ma! 2SG carry-APP -3SG.O-hither ‘Bring it here/to me!’ 56. Ye tabe-i-ya-ma. 3SG pull-APP -3SG.O-hither ‘He pulled it hither.’

55. Ye woya-i-ya-wa. 3SG lead-APP -3SG.O-to.addr ‘He led it to you.’ 57. Ye tu-ya-wa. 3SG throw-3SG.O-to.addr ‘He threw it to you.’

The directional suffixes can in fact derive transfer verbs from verbs that lack the semantic component of transfer. Consider (58) and (59): 58. Ye 3SG

duwui-ya-ma. dive-3SG.O-hither

‘He dived and got it hither.’

59. Ye 3SG

kuli-ya-wa. write-3SG.O-to.addr

‘He wrote it to you.’

Although speakers interpret the clauses in (53) to (59) as differing in the person of the recipient, the directional suffixes do not indicate person agreement as

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

131

pronominal affixes do. There are two distinctions which are consistently made in all Saliba pronominal paradigms which the directionals cannot express: that between singular versus plural, and, for the first person plural, that between inclusive versus exclusive. Directionals clearly differ from the cross-referencing pronouns on the verb. They are not expressions of syntactic arguments but merely implicate the notion of an event participant. Clauses with directionals are interpreted according to the context of the utterance. In some instances it is unclear whether the directionals are used to imply a recipient or literally denote a direction or goal. After all, the semantic entailment is only directionality toward speaker or addressee. As a consequence, the clause in (60) is underspecified as to whether the speaker is actually the recipient of the theme or merely witnesses the transfer to a location nearby. 60. Leta letter

wa

ye

bahe-i-ya-ma.

GIVEN

3SG

carry-APP -3SG-hither

‘She brought me the letter.’ OR ‘She brought the letter here.’

There are strategies to specify an intended recipient reading, for example, by adding a possessive expression as in (61), which, in combination with the directional suffix, reinforces the interpretation of the speaker as the recipient of the theme (cf. Margetts, 2002). 61. Yo-gu CL1-1 SG.P

leta

wa

ye

bahe-i-ya-ma.

letter

GIVEN

3SG

carry-APP -3SG-hither

‘She brought (me) my letter.’

The directional suffixes can be used to linguistically signal event participants that are not expressed as arguments in the clause. They entail directionality but implicate the presence of a participant. 6. CONCLUSIONS AND IMPLICATIONS FOR LANGUAGE ACQUISITION I have shown that, in Saliba, the concept of ‘giving’ is expressed by two verbs. Despite the difference in argument structure, their semantics and use as well as their complementary distribution prove that the two ‘give’ verbs, le and mose-i, are suppletion partners in a single paradigm. What makes this case of suppletion such a complex matter is that it is not restricted to the domain of the lexicon, but affects clausal syntax. I also showed that transfer events are typically not expressed by ditransitive verbs in Saliba but by transitive ones.

132

MARGETTS

These findings raise questions about the crosslinguistic universality of the alignment between event types and syntactic structures—in particular, between transfer events and clauses with three syntactic arguments. Such an alignment constitutes the basis for crucial claims in the recent literature on language acquisition about whether children derive verb meaning from syntax or vice versa. Two current proposals have been termed syntactic versus semantic bootstrapping, and are represented most prominently by Gleitman (1990) (syntactic) and Pinker (1987, 1989) (semantic). Both proposals assume a crosslinguistically universal mapping between syntax and semantics. What makes the Saliba data especially interesting is that the paradigm of ‘give’ could provide a test for the two bootstrapping hypotheses. The two proposals predict different error patterns in the acquisition of ‘give’ verbs for Saliba-speaking children. According to Gleitman’s (1990) syntactic bootstrapping proposal, the child will insert the component of ‘transfer’ into the semantic entry for a verb if the verb is observed with three arguments. This will happen for mose-i but not for le: the differences in argument structure will cause the child to falsely assume different underlying meanings for these verbs. According to the semantic bootstrapping proposal (Pinker, 1989), children would successfully associate both verbs with the same meaning by connecting them to the same kinds of real world events. But this meaning would falsely project the same argument structure for both verbs. We could expect, then, that children would produce ungrammatical ditransitive clauses with le ‘give’ and hai ‘take/get’ to express ‘giving’ to first and second person. Similar errors would be expected with other transfer verbs such as ‘send’ and ‘bring’, which also express a recipient with the directional suffixes. The syntactic bootstrapping approach predicts that this type of error should not occur.18 Do Saliba children have problems in acquiring the syntax and semantics of the two ‘give’verbs? Do they make errors? The acquisition of Saliba has not been studied, so these remain open questions for the time being. But we can identify some factors that may help children learn that le and mose-i are not semantically distinct. The discourse occurrence of the ‘give’ verbs is actually very similar. Despite the difference in argument structure, both verbs typically occur in clauses with only one lexically expressed participant (if any). This is in most cases the theme object. Lexical expressions of the recipient are more likely in clauses with mose-i than with le, but such cases are rather rare overall. This shows that the difference in argument structure is not prominently manifested in the discourse patterns of the two verbs and is potentially hard to observe. Table 6.1 summarizes the occurrence and distribution of lexically expressed event participants independent of their argument status in the text corpus. In addition to the similarity in discourse patterns, there are other factors that may help Saliba children in the acquisition of ‘give’. Giving and bringing things to people are very common tasks assigned to the smallest family members in

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

133

TABLE 6.1 Participant Realization with ‘Give’ Verbs in Text Data mose-i ‘give’

le (+ DIR) ‘give’

Instances of ‘give’ verbs in text corpus:

37

12

Participants lexically expressed:

None

16

3

Theme only

16

9

Recipient only

3

—

Theme and Recipient

2

—

All three participants

—

—

Saliba-speaking communities. Especially the ingredients for betelnut chewing are endlessly passed back and forth, most commonly by sending small children. It is a way to integrate children who are too young to perform household duties into the daily activities. The requests with le or mose-i ‘give’ directed to a small child are typically accompanied by pointing or beckoning gestures, which may help the child to understand the task even before the verbs are fully understood. Commonly the child is turned to face the intended recipient before being sent off to deliver the object. Children who are too small to walk or too shy to perform the task are sometimes given the object to hold and then carried to the recipient to “deliver” the object. So the caregiving adults have nonlinguistic strategies that help children acquire the semantics of the ‘give’ verbs. Adults’ comments on events may also help children acquire the distinction between le and mose-i. Interactions such as the one discussed earlier in (12) describe the same event from different perspectives. In this example a mother asked a child to give her an object using one ‘give’ verb (ku le-ya-ma ‘give it to me’) and the grandmother commented on the request, addressing the child using the other verb (ku mose-i ‘give it to her’). During a short sample period I have taken notes on the language production and comprehension of one Saliba-learning child at the age of 2;0 to 2;1.19 These very limited data suggest that comprehension of both ‘give’ verbs is well established at this age. The little girl reacted accurately to requests with both verbs. She produced just a few tokens of mose-i ‘give’ (without NP) and none of le ‘give’. Most of her utterances accompanying ‘giving’ events or requests for things simply contained possessive expressions without verbs. Requests for an object were generally expressed by ka-gu ‘mine (edible thing)’, alone or followed by the name of the object, as in (62) and (63).20

134

MARGETTS

62. Ka-gu!

63. Ka-gu

baela!

CL2-1 SG.P

CL 2-1 SG.P

banana

‘Mine!’

‘My banana!’

Telling the addressee to give an object to a third person was also expressed in this way: 64. Bubui granny

ka-na! CL2-3 SG.P

‘Granny’s!’ (Give it to Granny!)

In contrast to Tomasello’s (1997) data on an English-speaking child’s first talk about possession and exchange, possessive expressions in Saliba are not predominantly used in situations where control over an object is disputed, such as when someone is trying to take an object away from the child. The possessive forms describe both exchanges and requests for an object. In the child’s utterance in (65), the possessive expression is used similarly to English ‘Here you are!’ accompanying the transfer of an object from speaker to addressee. 65. Mother to child: Bubui

ka-na

lugu

ku

mose-i-ø!

granny

CL2-3 SG.P

mustard

2SG

give-APP -3SG.O

‘Give granny the mustard!’ Child giving the mustard to grandmother: Ka-m! CL2-2 SG.P

‘Yours!’ (Here you are! Take this!)

The use of possessive words to describe such exchanges is very common in adult language, and the child’s utterances in (62) to (65) are grammatically and pragmatically normal. Adult speakers are, however, more likely to combine the possessive words with verbal ‘giving’ expressions, as in the mother’s utterance in (65). In summary, the discourse patterns of le and mose-i ‘give’ are not saliently different and there are linguistic and nonlinguistic patterns in the interaction between speakers which may help children to learn about the semantics of the two Saliba ‘give’ verbs. In Saliba, ‘giving’ expressions and transfer events in general neither require nor prefer clauses with three syntactic arguments. Transfer events are mostly expressed by syntactically transitive verbs. The assumption of a universal alignment between event types and syntactic structures clearly needs rethinking in the light of these findings.

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

135

ACKNOWLEDGMENTS I am indebted to Melissa Bowerman, who inspired much of this work with her interest and long discussions. The members of the Argument Structure Group at the MPI and other participants of the workshop held at the MPI Nijmegen in 1998 where this paper was originally presented gave valuable comments and criticisms. Thanks especially to Ken Hale for his commentary. I am also grateful for comments by Penny Brown, Nikolaus Himmelmann, Bill McGregor, and John Newman. In Papua New Guinea, I’d like to thank the people of the Sawasawaga ward of Saliba Island. Without their support my research would not have been possible. My thanks are also due to the National Research Institute and the University of Papua New Guinea and the authorities in Milne Bay Province. The research on which this chapter is based was made possible by a PhD fellowship from the Max Planck Institute for Psycholinguistics in Nijmegen. NOTES 1

In this chapter I draw on data collected during six trips to Papua New Guinea (PNG) between 1995 and 2000. Besides elicited material the data corpus consists of about 6 hours of transcribed spoken language and includes a variety of text types from male and female speakers from a wide age range. The following abbreviations have been used:

Abbreviations APP

applicative

NEG

negative

CL1, CL2

possessive classifiers

O

object

COND

conditional marker

OBLIGAT

obligation marker

CONJ

conjunction

P

possessive

DEM

demonstrative

PERF

perfect

DET

determiner

PL

plural

DUR

duration

PP

postposition

EMP

emphatic

PROHIB

prohibition

EX

exclusive

PRSP

presupposition marker

GIVEN

givenness marker

RED

reduplication

INC

inclusive

SG

singular

INTRJ

interjection

TAM

tense/aspect/mood

IRREAL

irrealis marker

2 3

Alternatively, speakers replace le ‘give’ in (1) and (2) with hai ‘take/get’. I follow the SIL trial orthography for Saliba (Oetzel & Oetzel, 1997), in which the subject prefix is written separately while the object suffix is written as bound to the verb.

136 4

MARGETTS

The word-final allomorph of the third singular object suffix is zero, the non-final allomorph is -ya. 5 Comrie (2003) points out that suppletion according to person agreement seems to be crosslinguistically less rare with ‘giving’ expressions than with other verbs. He reports cases of this from Tsez (northeast Caucasian) and Malayalam (Asher & Kumari, 1997: 348). As in Saliba, the split in these languages is between first and second versus third person. Barai (Papuan) shows a case of suppletion in the paradigm of ‘give’according to the number of the theme object (Newman, 1996: 27). Note, however, that, in contrast to Saliba, none of these languages shows a difference in argument structure between the suppletion partners. 6 Unless otherwise indicated, the examples in this section are elicited. 7 The term kune is the Saliba equivalent of kula in other Milne Bay languages. Saliba Island was never part of the famous Kula Ring, but the concept of kune shows close similarity to the known kula exchanges. Kune refers to ‘giving’pigs, yams, bagi necklaces, or mwali arm shells. Kune gifts require a counter-gift and put the recipient in debt toward the giver. The exchanges are considered to establish kawa keha mamohoina ‘good relationships’ between the involved parties. For references on the Kula Ring see Leach and Leach (1983). 8 A less literal translation would be ‘She tried to give’ or ‘offered’ a biscuit. 9 There is no way to determine the frequency of the different syntactic frames in which mose-i ‘give’ may occur; cf. the discussion later in this section. 10 Most Saliba postpositions inflect for number (and some also for person); cf. Margetts (1999). 11 For the examples in this section I chose objects for which the marking on the verb must agree in number, that is, nouns that are overtly number-marked or that have human referents. 12 As a reminder: the third singular object suffix is -ø word-finally, but -ya in nonfinal position. 13 Hetamali ‘send’ is probably acceptable in the construction in (37a), but (36) is clearly preferred. 14 The notion of ‘TRANSFER TO SUBJECT’ is entailed in the semantics of hai ‘take/get’ rather than being a matter of pragmatics. The directional suffixes do not overrule the notion of transfer ‘to subject’; they merely indicate that the subject is not the endpoint of the transfer. 15 The shorter gloss ‘toward speaker’ is used here for ‘toward speaker OR DEICTIC CENTER’. 16 The semantic analysis of the directional suffixes is based on the text corpus as well as on elicitations. I used the elicitation tool designed by Wilkins and Hill (1993) for the use and semantics of COME and GO expressions. The application of this tool and results for two languages, Mparntwe Arrernte (Australia, PamaNyungan) and Longgu (Oceanic), are described in Wilkins and Hill (1995). Note that the analysis of the Saliba directionals is based on additional field work and differs substantially from the analysis presented in Margetts (1999). 17 The same preference holds for the Saliba demonstratives, where the speakerbased proximal form has precedence over the addressee-based form if an object is equidistant from speaker and addressee (Margetts, 2004).

6.

THE PROBLEM OF ‘GIVE’ IN SALIBA

137

18

The absence of such errors would not necessarily support the syntactic bootstrapping hypothesis. 19 Notes were taken over a period of two months during three visits of several days each. 20 During the sample period, the use of ka-, the classifier for food items, was overextended to nonfood items, which in adult speech require the general classifier yo-. REFERENCES Asher, R. E., & Kumari T. C. (1997). Malayalam. London: Routledge. Bybee, J. (1985). Morphology: A study of the relation between meaning and form. Amsterdam: John Benjamins. Carstairs-McCarthy, A. (1994). Suppletion. In R. E. Asher & J. M. Y. Simpson (Eds.), The encyclopedia of language and linguistics (Vol. 8, pp. 4410–4411). Oxford: Pergamon Press. Comrie, B. (2003). Recipient person suppletion in the verb ‘give’. In M. R. Wise, T. N. Headland, & R. M. Brend (Eds.), Language and life: Essays in memory of Kenneth L. Pike (pp. 265–281). Dallas: SIL International and the University of Texas at Arlington. Dixon, R. M. W. (1982). The semantics of giving. In R. M. W. Dixon (Ed.), Where have all the adjectives gone? And other essays in semantics and syntax (pp. 117–139). Berlin: Mouton de Gruyter. Dryer, M. S. (1986). Primary objects, secondary objects and antidative. Language, 62, 808–845. Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55. Leach, J., & Leach, E. (Eds.). (1983). The Kula: New perspectives on Massim exchange. Cambridge [Cambridgeshire]; New York: Cambridge University Press. Margetts, A. (1999). Valence and transitivity in Saliba, an Oceanic language of Papua New Guinea. PhD dissertation, University of Nijmegen, The Netherlands. Margetts, A. (2002). The linguistic encoding of three-participant events in Saliba. Studies in Language, 26, 613–636. Margetts, A. (2004). Spatial deictics in Saliba. In G. Senft (Ed.), Deixis and demonstratives in Oceanic languages (pp. 37–57). Canberra: Pacific Linguistics. Martin, S. (1975). A reference grammar of Japanese. New Haven, CT: Yale University Press. Mel’
uk, I. (1992). Suppletion. In W. Bright (Ed.), International encyclopedia of linguistics (Vol. 4, pp. 97–98). Oxford: Oxford University Press. Newman, J. (1996). Give: A cognitive linguistic study. Berlin: Mouton de Gruyter. Newman, J. (Ed.). (1997). The linguistics of giving. Amsterdam: John Benjamins. Oetzel, S., & Oetzel, R. (1997). Orthography and phonology description of Saliba. Unpublished manuscript. Ukarumpa, Papua New Guinea: Summer Institute of Linguistics. Pinker, S. (1987). The bootstrapping problem in language acquisition. In B. MacWhinney (Ed.), Mechanisms of language acquisition (pp. 399–441). Hillsdale, NJ: Lawrence Erlbaum Associates. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Ross, M. (1988). Proto Oceanic and the Austronesian languages of Western Melanesia. Pacific Linguistics, Series C-98. Canberra: Pacific Linguistics. Tomasello, M. (1997). One child’s early talk about possession. In J. Newman (Ed.), The linguistics of giving (pp. 348–373). Amsterdam: John Benjamins. Wilkins, D. P., & Hill, D. (1993). Preliminary ‘COME’ and ‘GO’ questionnaire. In E. Danziger (Ed.), Cognition and space kit (version 1.0) (pp. 29–46). Nijmegen, The Netherlands: Max Planck Institute for Psycholinguistics. Wilkins, D. P., & Hill, D. (1995). When “go” means “come”: Questioning the basicness of basic motion verbs. Cognitive Linguistics, 6, 209–259.

PART II

Participants Present and Absent: Argument Ellipsis and Verb Learning

CHAPTER 7

Same Argument Structure, Different Meanings: Learning ‘Put’ and ‘Look’ in Arrernte David P. Wilkins Center for Aphasia and Related Disorders V. A. Northern California Health Care System, Martinez, CA

1. INTRODUCTION This chapter takes as a starting point Gleitman’s (1990) suggestion that, universally, a verb meaning ‘put’ occurs naturally in structures with three noun phrases, but a verb meaning ‘look’does not, so these structures provide children with a clue to the verbs’ meanings. One means of assessing her proposals is to examine them crosslinguistically, and I do so by drawing on data from Mparntwe Arrernte, a Central Australian Aboriginal language. It is not my intention to assess the validity of Gleitman’s syntactic bootstrapping hypothesis per se, but to explore the more general notion—held by both syntactic bootstrapping theorists and semantic bootstrapping theorists—that there are more versus less natural alignments of syntax and semantics. According to Gleitman’s syntactic bootstrapping hypothesis, children are able to infer aspects of verb meaning from the syntactic frames in which the verb occurs. Gleitman recognizes that this hypothesis requires taking a very strong position concerning the universal alignment of semantic and syntactic relations: The first proviso to the semantic usefulness of syntactic analysis for learning purposes is that the semantic/syntactic relations have to be materially the same across languages. Otherwise, depending on the exposure language, different children would have to perform completely different syntactic analyses to derive aspects of the meaning. And that, surely, begs the question at issue. (Gleitman, 1990: 35; my emphasis)

Gleitman highlights verbs meaning ‘put’ and ‘look’ to illustrate her claim that there are universal alignments of syntax and semantics: Verbs that describe externally caused transfer or change of possessor of an object from place to place (or from person to person) fit naturally into sentences with 141

142

WILKINS

three noun phrases, for example, John put the ball on the table. This is just the kind of transparent syntax/semantics relation that every known language seems to embody.…That is, “putting” logically implies one who puts, a thing put, and a place into which it is put; a noun phrase is assigned to each of the participants in such an event. In contrast, because one can’t move objects from place to place by the perceptual act of looking at them, the occasion for using look in such a structure hardly, if ever, arises.…Restating this more positively, the component ‘transfer’ is inserted into a verb’s semantic entry in case it is observed to occur in three noun-phrase structures. This happens for /put/ but not for /look/. (Gleitman, 1990: 30; my emphasis)

As I show here, however, verbs meaning ‘put’ and ‘look’ in Arrernte defy this commonsense reasoning. The discussion has three main parts. First, I present some information about Mparntwe Arrernte and outline the available tests for identifying arguments in this language. This reveals that the verbs arrerne- ‘put’ and are- ‘look, see’ both take three NP arguments with the case array {ERGative, ACCusative, DATive} (where the Dative argument provides the location that the Accusative argument is associated with). Second, I examine a corpus of texts from Arrernte adults to determine whether the frequency in occurrence and positioning of the arguments in actual use would suggest that the verbs have different kinds of meanings. Finally, I analyze and compare production data from five adults and fifteen Arrernte children aged 6–10 to determine whether the children are treating the verbs in a syntactically mature way. The data force us to reevaluate the claim that “syntactic/semantic relations are materially the same across languages”, although they are consistent with the position that there are indeed more and less natural alignments of syntax and semantics. 2. BACKGROUND TO THE GRAMMAR OF ‘PUT’ AND ‘LOOK’ IN ARRERNTE In this chapter I examine data collected from Eastern and Central (i.e., Mparntwe) Arrernte speakers living in Alice Springs (Mparntwe) and Santa Teresa (Ltyentye Apurte). Eastern and Central Arrernte are, from a linguist’s standpoint, dialects of one language, which is simply labeled Eastern Arrernte. Eastern Arrernte belongs to the Arandic group within the Pama-Nyungan family of Australian languages. It is estimated that there are approximately 2000 speakers of Eastern Arrernte, making it one of the healthiest Australian languages. The language is in daily use and children are still learning it as a first language, although many are now bilingual in English. In fact, it is used as the primary medium in the early years of instruction at the Yipirinya School in Alice Springs and the Ltyentye Apurte (Santa Teresa) school, and some of the data presented here were collected at these two schools with the kind help of Arrernte teachers. Substantial linguistic work has been done on Eastern Arrernte (i.e., the Eastern and Central Arrernte varieties). There is an extremely good dictionary (Henderson

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

143

& Dobson, 1994), a learner’s guide (Green, 1994), and a reference grammar (Wilkins, 1989). Text collections can be found in Turner (1994), Wilkins (1989), and Henderson (1986). Further research on the language includes Harkins and Wilkins (1994), Henderson (1998, 2002), and Wilkins (1986, 1988, 1991, 1993b, 1997, 2000, 2006; Wilkins & Hill, 1995). 2.1. Grammatical Sketch of Arrernte The grammar of Arrernte has many typical Pama-Nyungan features. It is an agglutinating language that employs only suffixes, no prefixes. It has an extensive case system, and the ordering of phrases within a sentence is pragmatically determined and does not convey basic grammatical information. Nominals display a split case-marking pattern that basically conforms to the observations of Silverstein (1976). Common nouns show an ergative/accusative pattern and pronouns show a nominative/accusative pattern. Word order within noun phrases is fixed and case is marked on the final element of the phrase no matter what it is (i.e., case marking is by peripheral attachment). Clause-level syntax shows a clear tendency towards a nominative-accusative grammar. Arrernte is distinct from many Australian languages in that it has not retained the ancestral verb conjugations, and it does not have bound pronominals on the verb to reference the arguments of a clause. The Arrernte verb has a relatively complex structure, with seven distinct positions in the stem: the verb root, a slot for derivational suffixes, four slots for other quasi-inflectional suffixes, and a final slot for obligatory verb inflections. The only obligatory elements are the verb root and one of the final suffixes. For main verbs, the obligatory final inflection will be a tense or mood suffix, whereas for dependent verbs it may be a switch-reference suffix (indicating whether the subject of the dependent clause is the same or different from that of the main clause) or one of a set of other suffixes that mark the clause as a complement or as a temporal, causal, or conditional adjunct to the main clause. 2.2. ‘Put’ and ‘Look’ in Arrernte The Arrernte translation equivalents of ‘put’ and ‘look’ are arrerne- and are-. Both are highly frequent verb roots that appear regularly in interchanges with young children. Table 7.1 presents the top 20 most frequent verb roots found in adult Arrernte texts, including conversations with children and narratives for them; are‘look, see, visit’ is the second most frequent, while arrerne- ‘put’ is the 12th. Arrerne- ‘put’is a transfer verb and are- ‘look, see’is a perception verb. Of course, these words are not exact semantic equivalents to the English verbs put and look, but they are semantically similar enough to them that, according to Gleitman’s proposal, one would expect arrerne- and are- to differ in argument structure, and have a corresponding difference in syntactic patterning. To give an idea of the range of meaning of each verb, I follow Henderson and Dobson’s (1994) dictionary presentation of their different senses. Then text examples are given in (1) and (2).

144

WILKINS TABLE 7.1 Top 20 Arrernte High-Frequency Verbs from a Corpus of about 50,000 Words (approximately 325 texts)

1.

ane- intr. ‘sit, stay, be’ [601]

2.

are- tr. ‘see; look; visit’ [369]

3.

alhe- intr. ‘go’ [move] [341]

4.

arlkwe- tr. ‘eat’ [276]

5.

angke- intr. ‘say; speak; make noise characteristic of a thing or animal’ [195]

6.

ile- tr. ‘tell a story; tell s.o. to do something’ [166]

7.

arrate- intr. ‘appear; come out; arrive at’ [142]

8.

mpware- tr. ‘make; do’ [141]

9.

ine- tr. ‘get; pick up; get hold of’ [139]

10.

aknge- tr. ‘take; carry’ [122]

11.

unthe- intr. ‘look for; wander around; go in search of’ [107]

12.

arrerne- tr. ‘put’ [101]

13. a. apetye- intr. ‘come’ [94] 13. b. atwe- tr. ‘hit, chop, kill’ [94] 14.

irrpe- intr. ‘enter; go into’ [76]

15.

alpe- intr. ‘go back; return’ [75]

16.

iwe- tr. ‘throw away’ [68]

17.

unte- intr. ‘hurry off’ [move along at speed in the manner characteristic of subject] [66]

18.

ite- tr. ‘cook; start a fire’ [59]

19.

anthe- tr. ‘give’ [58]

20. a

awe- tr. ‘hear; understand’ [54]

20 b.

itnye-/atnye- intr. ‘fall’ [54]

Note. The number in square brackets is the number of occurrences in the corpus.

are- 1a. look at something, see, watch, notice, have a look at something; 1b. visit someone, see them; 1c. meet with someone, meet up with them; 1d. find something or someone; come across it; 2. look for something; 3. look to be a certain way; 4. shine on something, light it up arrerne- put something (down), put it on something, put it in something, place it somewhere

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

1. Examples of are- ‘look, see’: a. Inteye cave

145

1

kwene-ke

re

arrentye

re-nhe

are-ke

inside-DAT

3sgS[=ERG]

demon

3sg-ACC

look/see-pst

ankwe-inte-rlenge. asleep-lie-DS ‘He saw the cannibal inside the cave sleeping.’ [DAT ERG ACC V-main ClausComp] b. Arelhe-le woman-ERG

tyarte-ke

are-rlane-rlenge

intelhentye-Ø ...

shirt-DAT

look/see-CONT-DS

design/pattern-ACC

‘A woman was looking at the design on one of the shirts when ...’ [ERG DAT V-dependent ACC] c. Iwerre-ke path-DAT

anwerne

aherre

arunthe-Ø

are-ke.

2plS[=ERG]

kangaroo

several-ACC

look/see-pst

‘On the way we saw some kangaroos.’ [DAT ERG ACC V-main] d. …ante …and

apwerte-ke

are-rle

kere

arrwe-ke.

hill/rock-DAT

look/see-GenEvt

game

rock.wallaby-DAT

‘…and one looks in the hills for rock wallabies.’ [DAT V-main]. (Both the ERG and ACC arguments are ellipsed in this sentence; the second DAT NP is an adjunct referring to e.g. ‘signs/tracks/evidence of rock wallabies’.) e. Kenhe

utyerrke

nhenge-Ø

athetheke-arle are-me-le

but

wild.fig

REMEMB-ACC

red-FOC

arne-ke,

kele

mpenge

anteme.

tree-DAT,

O.K.

ripe

now

look/see-pres-SS

‘But when the wild fig looks red on the tree, then it’s ripe.’ [ACC 2ndPRED V-dependent DAT] f. Tyape

utnerrengatye-Ø

akenhe utnerrenge-ke-arlke are-me.

edible.grub emu.bush.grub-ACC BUT

emu.bush-DAT-too look/see-pres

‘But the utnterrengatye grub is also found on the emu bush.’ [ACC DAT V-main]

146

WILKINS

g. Re 3sgS[=ERG]

are-ke

mameye

ikwerenhe-Ø

kwene-ke

look/see-pst

mum

3sgGEN-ACC

inside-DAT

mantere anteye

irlwe-lhe-rlenge.

clothes

take.off-REFL-DS

also

‘She saw/ watched/ looked at her mum inside also changing her clothes.’ [ERG V-main ACC DAT ClausComp]

2. Examples of arrerne- ‘put’: a. Arrweketye mape-Ø woman group-NOM mpware-me

anthepe-irre-tyeke

lane-me,

awelye

dance-INCH-PURP

sit-pres,

ceremony

arrerne-me urlatye-ke, ulyepere-ke,…

arlkenye-Ø

make/do-pres stripe/painting-ACC put-pres

breast-DAT thigh-DAT

‘When women are going to dance, or when they’re doing their own ceremonies, a striped design is put on their breasts and thighs.’ [ACC V-main DAT] b. Ante and

yanhe-ke

waytpele

sign

yanhe-Ø

there-DAT

white.person

sign

that(mid)-ACC put-pst…

arrerne-ke…

‘And (they) put/placed that “whitefellah” sign there (mid-distant).’ [DAT ACC V-main] c. Table table

yanhe-ke there-DAT

arrerne-warr-Ø-aye!

The

fridge-ke

put-plS-IMP-EMPH

1sgERG

fridge-DAT

arrerne-tyenhenge. put-SBSQNT ‘Put (the food) on the table there! I’ll put (it) in the fridge (later)’. [DAT V-main; ERG DAT V-dependent]

2.3. Identifying Arguments in Arrernte Arguments are not obligatorily expressed in Arrernte and massive argument ellipsis is in fact the norm. In the text examples already shown in (1) for are- ‘look, see’, more arguments are realized (i.e., explicit) than is usual; this is to demonstrate that sentences with ‘look, see’, just like those with ‘put’, take a third NP, marked as Dative. This Dative noun phrase is semantically required by the verb, so it is understood even if not explicitly expressed; that is, it is an argument. But how do we know that a Dative noun phrase with are- ‘look, see’ is an argument? It is notoriously problematic to establish argument structure, even for a

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

147

language like English, which, unlike Arrernte, allows relatively little ellipsis of NPs. As Comrie (1993) points out: A crucial distinction in studying predicate-argument structure is that between argument and adjunct .…The basic intuition behind this distinction is relatively clear, though difficulties arise as soon as one tries to make it more explicit, and there is as yet no generally accepted solution to these difficulties. (Comrie 1993: 906)

One has to find appropriate tests for each language. I accept for current purposes the Gleitman (1990)/Pinker (1984) tests whereby English see (a perception verb) has two arguments and put (a transfer verb) has three. But what distinguishes arguments and adjuncts in Arrernte? The properties of Arrernte grammar relevant to establishing argument structure include the following: • There are 14 distinct case forms, which attach to the end of a noun phrase. If a noun phrase is realized it must be marked for case, to indicate what role its referent plays in the state of affairs described by the utterance. Cases of interest here include Nominative, Accusative, Ergative, Dative, Allative, and Ablative. • Phrase order in an utterance is free. That is, word order does not give grammatical information; it is pragmatically determined. This means that order cannot be used to determine argument structure. • The only form of agreement on the verb is number agreement for subject (i.e., Nominative or Ergative arguments). Thus, the subject argument is easily identified. • Noun phrases may be freely omitted, usually but not always subject to retrievability from discourse context. There is no NP that must surface with a verb. Thus, the obligatory occurrence of a given phrase with a predicate is not a possible diagnostic of argument status. • Predicates assign case to their arguments, but not all case-marked NPs are arguments of the predicates they occur with. Some case forms mark arguments in some instances, adjuncts in others. (Note that prepositional phrases in English can show a similar ambiguity: e.g., The boy put the bag ON THE TABLE [argument: integral to the meaning of put] and The boy saw the bag ON THE TABLE [adjunct: incidental to the meaning of see].) How do we know when a case-marked NP in Mparntwe Arrernte functions as an argument (i.e., is determined by the semantic structure of the predicate), rather than as an adjunct (adding extra information in the utterance beyond that given by the predicate)? There are a number of diagnostics in this language, but the clearest involves relative clauses. There are two ways of forming a relative clause, both with

148

WILKINS

the schematic structure shown in (3a): one is used with arguments, as illustrated in (3b), and the other is used with adjuncts, as is discussed shortly. 3. a. [[N]Hd

[X-arle

b. Pmere

re-arle

alhe-ke-arle

3sgS-REL

go-pst-REL

place

(Y)

Vb FINITE (-arle)]SREL]NP …

‘The place where she went…’

In the relative clause used with arguments, there is simply a gap where we might expect the role played by the referent of the head NP to be marked. This is the case in (3b), as can be seen in the analysis of this sentence in example (4b), where the gap is indicated as [Ø]. (The main clause that corresponds to the relative clause in (4b) is shown first in (4a).) 4. Allative argument (gapped) a. Re

pmere-werne

alhe-ke

place-ALL

go-pst

3sgS

‘She went to(ward) the place.’ b. [[Pmerei]Hd place

[re-arle

Øi

alhe-ke-arle]SREL]NP…

3sgS-REL

[gapped-ALL]

go-pst-REL…

‘The place [where] she went ...’ [gapped Allative in relative clause indicates that the Allative NP is an argument of alhe- ‘go’.]

In the relative clause used with adjuncts, the role of the head NP in the clause is indicated by a case-marked pronominal form (i.e., a pronominal copy of the head NP); this is illustrated in (5b) (again preceded by its main-clause counterpart in (5a)): 5. Allative adjunct (pronominal copy) a. Re 3sgS

pmere-werne

are-ke.

place-ALL

look/see-pst

‘She looked toward the place.’ b. [[Pmerei]Hd place

[re-arle

ikwere-wernei

are-ke-arle]SREL]NP…

3sgS-REL

3sg-ALL(pro-copy)

look/see-pst-REL…

‘The place that she looked toward…’ [Allative pronominal copy in relative clause indicates that the Allative NP is an adjunct of are- ‘see’. A Dative NP with this verb, in contrast, is an argument, as we see in section 2.4.]

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

149

These two types of relative clause are not interchangeable—for example, the gap in (4b) cannot be replaced by a case-marked pronominal, nor the pronominal in (5b) by a gap. The logic of the two forms can be understood informally as follows. If the NP that is relativized is an argument of the verb in the relative clause, it is easy to recover its role in the clause because this role is already implicit in the verb’s meaning (this is what sanctions the gap). In contrast, if the NP is an adjunct, its role cannot be recovered, so it must be explicitly marked (hence, the case-marked pronominal). Often one does not have to actually carry out the gapping test, or any other diagnostic, to determine whether an NP is an argument; argument status (hence behavior on the diagnostics) can be accurately predicted on the basis of the case marking alone. Thus, NPs marked with Ergative, Accusative, or Nominative case are always arguments (and always gap in relative clauses); NPs marked with Genitive, Proprietive, Associative, Afterative, Aversive, or Comitative case are never arguments (and never gap). But certain case-marked NPs—Datives, Locatives, Ablatives, Allatives, and Instrumentals—can be either arguments or adjuncts, depending on the verb. We have seen this for Allatives in examples (4b) (gap—an argument) and (5b) (pronominal copy—an adjunct). Parallel examples for Datives are given in (6b) (argument) and (7b) (adjunct) and for Ablatives in (8a) (argument) and (8b) (adjunct). 6. Dative argument (gapped) a. Ayenge

merne-ke

unthe-me.

1sgS

food-DAT

look.for-pres

‘I’m looking for bush tucker.’ (i.e., fruit and vegetables) b. [[Mernei]Hd

[ayenge-arle

Øi

unthe-me-arle…]SREL]NP

food

1sgS-REL

[gapped-DAT] look. for-pres-REL …

‘The bush tucker that I’m looking for…’ 7. Dative adjunct (pronominal copy) a. Ayenge

merne-ke

alhe-me.

1sgS

food-DAT

go-pres

‘I’m going for bush tucker.’ b. [[Mernei] Hd [ayenge-arle food

1sgS-REL

ikwere-i

alhe-me-arle…] SREL]NP

3sg-DAT[pro-copy] go-pres-REL …

‘The bush tucker that I’m going out for….’

150

WILKINS

8. a. Ablative argument (gapped) Re

lhe-ke

3sgS

go-pst

[[artwei]Hd [unte-arle pwerte man

2sgS-REL money

ine-ke-

Øi

[gapped-ABL] get-pst-

arle]SREL -kerte]NP REL

PROP

‘She left with the man who you got/took money from.’ b. Ablative adjunct (pronominal copy) Anwerne

lhe-ke

1plS

go-pst

[re-rle

ikwere-ngei

place

3sgS-REL

3sg-ABL[pro-copy]

[[pmerei]Hd

artwe

mperlkere

are-ke-arle]SREL

-werne] NP

man

white

see-pst-REL

ALL

‘We went to the place where he saw the white man from.’

To summarize, sometimes we can tell from the case alone whether the case-marked NP is an argument or an adjunct. But there are several case forms—including Dative, Locative, Ablative, Allative, and Instrumental—that are compatible with either argument or adjunct status; in these cases, the relative clause test is definitive. The application of the relative clause test allows us to formulate a hierarchy of access to relativization in Mparntwe Arrernte, as shown in Figure 7.1.

FIGURE 7.1. Hierarchy of access to relativization in Mparntwe Arrernte.

The relative clause test is the clearest (formal) window on verb argument structure I could find, but there are myriad other facts about the uses of each verb that are consistent with it. For example, the results of the relative clause test are fully in accord with the range of potential interpretations of a highly frequent form of nominalization.2 There is no special relation between the nominalization and the relative clause test beyond the fact that they are both especially sensitive to the semantics of verbs in a way that reveals the relevant set of participant roles.

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

151

2.4. Argument Structure of Arrerne- ‘put’ and Are- ‘look, see’ As Gleitman’s hypothesis leads us to expect, the Arrernte equivalent of ‘put’, arrerne-, has the following argument structure: {ERG, ACC, DAT}. That is, it has three NP arguments, a ‘putter’ (ERG), a ‘thing put’ (ACC), and a ‘place put’ (DAT). This is true of other similar transfer verbs such as ‘give’. But are- ‘look, see’ (and indeed certain other perception verbs) also take three NP arguments, which goes counter to Gleitman’s predictions. The “unexpected” third argument is a Dative noun phrase indicating the endpoint of the sight path (which coincides with the location of the object). Example (9) shows that the Dative NP is gapped in a relative clause with are- ‘look, see’, establishing that it is an argument. This example contrasts with the earlier example (5b), in which an Allative NP with are- ‘look, see’—indicating the trajectory of the sight path but not its endpoint—must be marked with a pronominal copy in its relative clause, showing it to be an adjunct. 9. Dative argument with are- ‘look, see’ (gapped) Unte

kwenhe

unthe-tyeke

2sgS

ASSERT

look.for-PURP

[[arnei]Hd [unte-arle

thipe

2sgS[=ERG]-REL bird

tree

nyengke

Øi

are-me]SREL

-ke]NP,…

zebra.finch

[gapped-DAT]

see-pres

-DAT,…

‘You should look for the tree in which you see zebra finches,…

(The sentence underlying the SREL in this example is something like Unte thipe nyengke arne-ke are-me [2sgS[=ERG] bird zebra.finch tree-DAT see-pres] ‘You see a zebra finch in a tree’.) So ‘look, see’ in Arrernte, like ‘put’, needs three arguments: a perceiver, a perceived thing, and a place where the perceived object is. In contrast, in the English sentence ‘He saw the bird in the tree’, the prepositional phrase ‘in the tree’ is an adjunct. The argument properties of these two verbs can be roughly represented as in Table 7.2.

TABLE 7.2 Argument Properties of Arrerne- ‘put’ and Are- ‘look, see’, and Related Arrernte Verbs Ergative

Accusative

Dative

put-ter

thing put

place where put

perceiver

thing perceived

place of thing perceived

Transfer verbs e.g., arrerne- ‘put’ Perception verbs e.g., are- ‘look, see’

152

WILKINS

Example (10) shows parallel sentences with arrerne- ‘put’and are- ‘look, see’. 10.

ERG a. Artwe-le man-ERG

VERB

DAT

irrtyarte

ACC re-nhe

arrerne-ke

ilthe-ke

spear

3sg-ACC

put-pst

house-DAT

3

‘The man put the spear in the house.’ b. Artwe-le man-ERG

irrtyarte

re-nhe

are-ke

ilthe-ke

spear

3sg-ACC

look-pst

house-DAT

‘The man saw the spear in the house.’ [i.e., describes state of affairs where, from outside the house, the man saw the spear that was located inside the house.]

To summarize, the Mparntwe Arrernte verbs meaning ‘put’ and ‘look’ share the 4 same 3-NP argument structure frame. This is contrary to the expectations of Gleitman (1990: 30), who proposed that (a) a verb of “externally caused transfer”, such as ‘put’, logically implies three participants (a putter, a thing put, and a place into which it is put), whereas a verb of perception, such as ‘see’, does not, and (b) a child will therefore insert the component ‘transfer’ into a verb’s semantic entry if it is observed in structures with three noun phrases, but not if it occurs only with two. A child will be seriously misled if she assumes that because are- ‘look, see’ has three arguments, like arrerne- ‘put’, it encodes a meaning of transfer. Like Gleitman (1990) and Pinker (1989), I assume that a verb’s argument structure is a projection of its meaning, and that the Arrernte verb are- ‘look, see’—just like arrerne- ‘put’—has a meaning that implies three participants. In this sense its meaning does differ from that of English look or see, just as its number of participants suggests. But this meaning difference is quite subtle, and falls well below the radar level picked up by the difference between two and three arguments. In fact, in Arrernte, ‘transfer’is not the meaning that should immediately be associated with the three-argument frame {ERG, ACC, DAT}, even though all sentences instantiating this frame are concerned with expressing the “endpoint” location of the Accusative argument. The “endpoint” (place of thing seen) of an event of visual perception is important enough to warrant core participant status in the meaning of are- ‘look, see’ in Arrernte, but this is not true for look or see in English. I come back to this point in the Discussion. 3. USE OF ‘PUT’ AND ‘LOOK’ IN ADULT ARRERNTE: STATISTICAL PATTERNS OF ARGUMENT REALIZATION Under the classic syntactic bootstrapping story (Gleitman, 1990), children learning Arrernte have a problem, because, as described in section 2.4, the argument structure of ‘put’and ‘look’cannot help them to differentiate the meanings of these verbs and might well mislead them as to the meaning of ‘look’. Perhaps children are not misled because they cannot actually detect the argument structures of these verbs, either because the morphosyntactic behaviors that distinguish arguments from ad-

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

153

juncts are too subtle for them or because arguments are so often ellipsed in this lan5 guage. In this case, however, they have a different problem: if they can’t determine argument structure, they are not getting the help with verb meanings that syntactic bootstrapping is supposed to offer. But perhaps there is information in a verb’s syntactic frames that goes beyond what can be gleaned from simply knowing the verb’s argument structure (or, alternatively, knowing how many noun phrases typically appear with it, regardless of whether they are arguments or adjuncts—Fisher, 1996, 2002). In this section I explore the hypothesis that the realization of the Accusative and Dative arguments of arrerne- ‘put’ and are- ‘look, see’ pattern statistically differently in adult speech, in a way that is related to the verbs’ meanings. To determine the statistical behavior of the two verbs, I examined a corpus of Arrernte adult texts. This corpus comprises approximately 325 texts, each one from 20 to 2,000 words long, for a total of about 50,000 words. Produced by about 30 adults between 20 and 70 years old, it covers a range of text genres, including conversations, narratives, procedural texts, expository texts, and hortatory texts. In only a few of the texts is the speech actually directed to young children, so the corpus cannot be said to be representative of Arrernte input to language learners. Still, we should be able to get a sense of whether there are differences in the clausal organization of NP arguments with arrerne- ‘put’ and are- ‘look’ that could, in principle, suggest different kinds of meanings to language learners. The clauses I examined were restricted in the following ways: • Only main-clause uses of the two verbs were examined. Dependent-clause uses were excluded for a number of reasons: (a) crosslinguistically, dependent clauses often have different organizational properties (in terms of order and occurrence of arguments) than main clauses; (b) in Arrernte, dependent clauses tend to show fewer overt NPs than main clauses (i.e., there is a greater degree of argument ellipsis in dependent clauses); and (c) I assume that, because of the various salient prosodic, structural, semantic, and discourse properties of main clauses, children attend to their structure earlier than that of dependent clauses. • Only clauses containing an Accusative or a Dative argument, or both, were included. Clauses with only the Ergative argument (the one who looks or puts), or with no surface arguments at all, were excluded, because it is the pattern of the Accusative and Dative arguments rather than of the Ergative argument that is most likely to distinguish these two verbs. The full set of clause types and their frequency in the corpus is given in the Appendix; the examples there show nicely how varied word order can be, and how freely arguments can be omitted. The frequency with which Accusative and/or Dative arguments appear at the surface with ‘look’ and ‘put’ is presented in Table 7.3. In clauses with overtly realized Accusative and/or Dative arguments, Accusative occurs 96% of the time with ‘look, see’ and only 60% with ‘put’. Dative shows the opposite pattern: it occurs only 30% of the time with ‘look, see’, but 81% with ‘put’.

154

WILKINS TABLE 7.3 Frequency of Accusative (ACC) and Dative (DAT) Arguments with Are- ‘look, see’ and Arrerne- ‘put’ in a Corpus of Adult Speech are- ‘look, see’

arrerne- ‘put’

[n=174]

[n=53]

Only ACC, no DAT

69.5%

18.9%

Only DAT, no ACC

4.0%

39.6%

Both ACC and DAT

26.4%

41.5%

In general, arguments placed immediately before or after the verb are semantically more tightly linked to the verb than arguments that are more distant. For utterances that contain both Accusative and Dative arguments, Table 7.4 shows the distribution of argument placement for our two verbs. Placement does differ: with arrerne- ‘put’ it is the Dative argument that is far more likely to occur immediately adjacent to the verb, whereas with are- ‘look, see’ it is the Accusative argument. So the “place” argument is semantically more closely linked to ‘put’ than to ‘look’, consistent with what we would expect on the basis of the argument structure of put and look in English. There are, then, statistical differences in the realization and positioning of the Dative and Accusative arguments with arrerne- ‘put’ and are- ‘look, see’. Learners are sensitive to differences in frequency and other statistical properties of sentences (cf. Bowerman, 1990; Rispoli, 1987, 1991; Tomasello, 2003, and much other research within usage-based approaches to language acquisition). For the child trying to infer verb meaning, the raw number of a verb’s arguments may be less important than the relative salience of these arguments, as cued by both the frequency with which they are explicitly mentioned and their position relative to the verb. 4. CHILDREN’S USE OF ARRERNE- ‘PUT’ AND ARE- ‘LOOK, SEE’ In section 2, I showed that, contrary to Gleitman’s (1990) proposal, the Arrernte equivalents of ‘put’ and ‘look’ do not differ in argument structure: both take three NP arguments, case-marked for Ergative, Accusative, and Dative. So children could not simply use occurrence with three noun phrases as a guide to differences in meaning between the two verbs. But in section 3, using counts from a corpus of adult narrative texts, I demonstrated that the ‘put’ and ‘look’ verbs show statistical differences in the surface occurrence and ordering of arguments. A child could,

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

155

theoretically, use these differences as a guide to meaning differences. In this section I explore this proposal with two forms of production data. First, I examined the spontaneous occurrence of arrerne- ‘put’ and are- ‘look, see’in 4 hours of videotaped classroom interaction, where the activities highlighted ‘putting actions’ and ‘looking actions’. Classroom interaction at this age is primar6 ily in Arrernte, with some code switching. The subjects were five adults and fifteen children between 6 and 10 years of age. The use of arguments with the two verbs is summarized in Table 7.5. Both children and adults regularly produced both Accusative and Dative arguments with arrerne- ‘put’, but only adults used both arguments with are- ‘look, see’ (all five adults contributed to the eight instances of are- with both arguments). Even for adults, though, most uses of are- ‘look, see’ included only an Accusative argument. Children produced are- ‘look, see’ only with an Accusative argument, as though it were a simple two-argument transitive verb. TABLE 7.4 Placement of Accusative and Dative Arguments with Respect to the Verbs Are- ‘look, see’ and Arrerne- ‘put’ are- ‘look, see’

arrerne- ‘put’

[n = 46]

[n = 22]

Only DAT adjacent to verb

19.6%

54.5%

Only ACC adjacent to verb

50.0%

18.2%

Both DAT and ACC adjacent

28.3%

22.7%

2.2%

4.5%

Neither adjacent

TABLE 7.5 Realization of Accusative and Dative Arguments in Naturalistic Classroom Interaction

arrerne- ‘put’

are- ‘look, see’

ACC and DAT

ACC only

DAT only

[18 tokens]

12

0

6

Children [21 tokens]

11

1

9

[31 tokens]

8

23

0

Children [26 tokens]

0

26

0

Adults Adults

Note. Subjects: 5 adults and 15 children: age 10 (3), age 9 (1), age 8 (3), age 7 (5), age 6 (3). Four hours of taping.

156

WILKINS

Of course, from these results we cannot tell whether or not children know or use the ‘look’ verb with three arguments (i.e., with a Dative argument, as well as an Ergative and an Accusative argument). To explore this question, I presented the same adults and children with a task modeled on Berman and Slobin’s (1994) “Frog Story” protocol. In this task the child (or adult) first sits together with a native speaker (a teacher), and looks page by page through a wordless picture book drawn by a Warlpiri man (the “Bird Story”). Then he or she goes through the book again page by page and tells the teacher the story depicted in the pictures. The first picture shows three boys, and the second shows the boys chasing a bird. The third picture is the critical one: it introduces a baby bird in a nest for the first time (see Figure 7.2). This picture tends to provoke three-argument uses of the ‘look’ verb. The results show a clear difference between the adults and children. Four of the five adults used are- ‘look, see’ in their description of the picture, and all of them provided all three arguments: Ergative (the seer), Accusative (the seen), and Dative (the location of the thing seen). For example: 11. Kele imarte itne O.K. then

antywe kwene-ke

3plS[=ERG] nest

inside-DAT

thipe

akweke-Ø

bird

small- ACC now

anteme

are-mele. see-narr.pres ‘So, after that, they see a baby bird inside a nest.’ [ERG DAT ACC X V]

FIGURE 7.2. Picture of bird in nest.

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

157

Thirteen of the 15 children used are- ‘look, see’ in their descriptions. But only five of them provided the Dative argument that gives location (one 8-year-old, three 7-year-olds, one 6-year-old). For example: 12. Mape-le group-ERG

thipe

akweke

nyente-Ø

nest-ke

are-me.

bird

small

one-ACC

nest-DAT

see-pres

‘The group of them are looking at one baby bird in a nest.’ [Age 7]

Four of the children omitted the Dative argument entirely (two 10-year-olds, one 7-year-old, one 6-year-old). For example: 13. Re 3sgS[=ERG]

anteme

are-ke

rtake-rtake

anyente-Ø.

now

see-past

duck

one-ACC

‘Then he saw one duck.’ [Age 10]

Strikingly, the final group of four children who used are- ‘look, see’ in describing the picture attempted to provide a location argument, but did so ungrammatically (one 10-year-old, two 8-year-olds, one 6-year-old). Three of the errors involved selecting the wrong case marker for the intended meaning: instead of the appropriate Dative case, two children used the Ablative case form -nge and one used the Locative case form -le. Some of these sentences are possible in adult speech, but the Ablative- and Locative-marked NPs are adjuncts, not arguments, and they carry a different meaning, as illustrated in (14): 14. Child errors: a. Itnareye 3plS[=NOM]

ntwe-ke

nest

akwene-nge

are-ke.

run-past

nest

inside-ABL

see-pst

‘They ran and looked inside the nest.’ [Age 8] [This would mean that the whole event took place in the nest.] b. Itnareye 3plS[=ERG]

mape

anteme

rtake-rtake-Ø

are-ke

nest-nge.

group

now

duck-ACC

see-pst

nest-ABL

‘The group of them now looked at the duck in the nest.’ [Age 8] [This would mean that the boys were in the nest looking at the bird outside it.] c. Thipe-Ø bird-ACC

are-me

nest-le.

see-pres

nest-LOC

‘(They’re) looking at the bird in the nest.’ [Age 6] [This would mean that the boys and bird are both in the nest.]

158

WILKINS

The fourth child’s error involves lack of case marking altogether: d. *Itne 3plS[=ERG]

are-ke

thipe

nhenge-Ø

nest-

see-pst

bird

REMEMB-ACC nest-[ ]

ane-me. sit-pres

‘They saw the bird in the nest sitting.’ [Age 10]

Of course, all noun phrases are marked for case, and all NPs functioning as place locations require an overt case ending. (Only the subjects of transitive and intransitive verbs and the objects of transitive verbs may, depending on noun or phrase type, occur with a zero case marker.) This child is clearly adding a location phrase, but appears to be unsure how to mark it, so leaves it unmarked. Elsewhere this same child appropriately marks the location adjuncts of verbs other than are- ‘look, see’, so the source of the problem seems to be confusion over the adjunct/argument status of this NP. To summarize, naturalistic observations in a classroom situation show that all the children and adults used arrerne- ‘put’ in its correct frame, with Dative case marking the NP that refers to the place where something is put. No examples of are‘look, see’with ‘Dative of location’turned up in the children’s spontaneous speech, but relevant data were elicited with a wordless picture book. This data revealed that, even as late as 6–10 years of age, children are still having difficulty with the treatment of the syntactic argument of are- ‘look, see’ that refers to the location of the thing seen. Children’s syntactic errors with are-‘look, see’are intriguing because they are at odds with the argument structure of are- in adult Arrernte. Even quite old children apparently do not know that a noun phrase naming the location of the thing seen is an argument of this verb. This suggests that the meaning they attribute to this verb is not—counter to Gleitman (1990)—an inference from the verb’s argument structure. Instead, the errors are in line with the meaning and the two-argument structure that is “expected” for a perception verb like ‘look’. This meaning is also compatible with the statistical characteristics of adult speech, in the sense that the argument children are having trouble with—the Dative, or “place of the thing seen”—is the one most commonly omitted in adult production. In contrast, children have no trouble with the Dative argument for ‘put’, which is very frequent in adult speech. 5. DISCUSSION According to both syntactic bootstrapping (Gleitman, 1990) and semantic bootstrapping (Pinker, 1984), children come to the language-learning task with foreknowledge of “natural” syntactic/semantic alignments. Unusual alignments should be harder to learn, and—to the extent that children rely on argument structure to infer verb meaning—they might lead children to make incorrect predictions about verb meaning. In this study, children seem to have no trouble recognizing that the Arrernte verb are- ‘look, see’ has to do with visual perception; its unusual three-NP

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

159 7

argument structure does not stand in the way of acquiring this basic meaning. But they do have trouble with its unusual syntax: despite 8 or 10 years of evidence to the contrary, they often treat are- as if it were a two-argument verb. This suggests that—just as Gleitman and Pinker assume—certain argument-structure alignments between semantics and syntax may indeed be more “natural” for learners than oth8 ers. Let us pursue this hypothesis. If we assume that some alignments are more natural than others, we can speculate that adult languages could depart from these alignments at points for a variety of reasons that influence the relative salience of particular event participants in different ways. Thus, some event participants could be (“naturally”) salient for individuals (including children), but over time, cultural salience could come to override this natural salience, pulling for verb meanings, and their associated argument structures, that are unusual in crosslinguistic perspective. What is culturally salient will become apparent to the learner through extended exposure to the community of usage, and it will come to seem increasingly “natural”. The 6- to 10-year-old Arrernte children I investigated have not gotten the meaning and associated argument structure of are- ‘look, see’ quite right yet. For this verb, the “place seen” participant is more central to the verb’s meaning than is true for many languages—it is, after all, an argument, not an adjunct. This centrality is perhaps related to the culture’s traditional hunter-gatherer activities, where speakers frequently need to localize the object seen in its place and talk about its location so that others may find it, stay upwind of it, and so on. This cultural motivation for the “place seen” argument is apparent in examples (1d)-(1f) shown earlier, which have to do with hunting and gathering practices, and we can infer that it underlies (1c) as well (‘On the way-DAT we-ERG saw some kangaroos-ACC’): here, the Dative argument lets the listeners know where to find the kangaroos if they should wish to hunt for them. If this analysis is correct, we might expect children’s active participation in the relevant cultural activities to bring them solidly into the associated community of practice (in this society children start joining in hunter-gatherer activities intensively at about the age of 8–10). An additional important influence may be the learner’s increasing appreciation of the place of are- ‘look, see’ within a wider web of Arrernte perception verbs that all treat the “place of the thing perceived” as an argument. Although the Arrernte verb are- ‘look, see’ has a crosslinguistically unusual argument structure, its statistical pattern of argument realization is more in line with the semantic/syntactic alignment that is “expected” for a perception verb: the argument encoding “place of the thing seen” is only mentioned about a third of the time. By comparison, the argument encoding “place of the thing put” is mentioned more than eighty percent of the time for the ‘put’verb (see Table 7.3). Arrernte ‘look’and ‘put’ can thus be differentiated on statistical grounds, even if not in their basic argument structure: the pattern of argument realization for ‘look’ is similar to what we

160

WILKINS

anticipate for a perception verb, and this may contribute both to children’s initial ballpark identification of its meaning and to the fact that even quite old children do not yet recognize that the “place seen” participant is actually an argument of are‘look, see’, not merely an adjunct. Statistical patterns of argument realization are often invoked in contemporary theorizing about syntactic bootstrapping, but in a distinctly different way than I have just 9 done. According to the syntactic bootstrapping model, argument structure is a reliable guide to a verb’s general semantic class; for example, three-argument verbs have consistently different kinds of meanings from two-argument verbs. Languages with massive argument ellipsis have been seen to pose a challenge to this model, because if arguments are not reliably present, how can children determine how many arguments a verb has, and use this information to hypothesize appropriate meanings (e.g., Rispoli, 1995)? Statistics have been invoked to solve this problem (e.g., Lee & Naigles, 2005): as long as verbs of different classes can be probabilistically distinguished on the basis of their number of arguments—for example, ditransitive verbs are at least more likely than transitive verbs to appear with three arguments—then children will have sufficient information to infer appropriate meanings. The problem for this reasoning raised by Arrernte are- ‘look, see’ is that the verb’s three-argument structure might initially mislead learners, rather than being a reliable guide to the verb’s meaning. In particular, if children behave according to Gleitman’s (1990) proposal, they would hypothesize that are- ‘look, see’—like arrerne- ‘put’, a verb with an identical three-argument array—encodes “externally caused transfer”. A way out of this dilemma is suggested by the infrequency in adult speech of the unexpected “place seen” argument (relative to the parallel “place put” argument of ‘put’): if children are initially led by statistical patterns of argument realization to assume that ‘look, see’ has only two arguments, then they might more easily hit on the correct perception verb meaning. This line of reasoning thus invokes statistical patterns of argument realization to account not for how children identify a verb’s argument structure even when its arguments are often ellipsed, but for how they can overlook an argument (“the place seen”) that is in fact explicitly or implicitly present, in order to hypothesize a two-argument-style perception verb meaning for are‘look, see’. We have, then, two different proposals for how children might deal with probabilistic evidence for a candidate argument. According to the proposal for overcoming ellipsis, learners would necessarily take even minimal probabilistic evidence seriously as confirmation of the argument’s existence. According to my proposal, in contrast, learners could ignore such evidence, at least for a while; this would allow them to view are- ‘look, see’initially as a two-argument verb, and so to hypothesize a suitable meaning. But when then will children take statistical evidence seriously and when will they ignore it? Here, let us return to the notion that some argument structure alignments are cognitively more “natural” for learners than others. For instance, by

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

161

hypothesis it is natural for a verb like ‘put’ to take an endpoint (“place put”) argument, so even minimal probabilistic evidence for this argument should be convincing to the child. But it is less natural for a verb like ‘look’to take an endpoint (“place seen”) argument, so children may require more evidence, presented over a longer period of time, before they are swayed from their initial assumption that this verb has only two arguments. Of course, once a more unusual pattern of argument structure is acquired, it will seem completely natural to speakers who are fluent in the larger linguistic and cultural context of their language. But such patterns may be harder to acquire because they initially “go against the grain,” and learners may err in the direction of more common patterns of argument alignment—for example, by 10 treating “place seen” as an adjunct, as discussed in section 4. To conclude, let us consider the relevance of the data discussed in this chapter for the larger issue of where we might find the most fruitful clues to what is cognitively natural for language learners. Notice that if we simply compare the grammars of English and Arrernte directly, the difference in the argument structure of ‘put’ and ‘look’ is falsely exaggerated. Although Arrernte’s verb ‘look, see’ seems exotic in having three arguments, it looks from a statistical, usage-based point of view much more like its English two-argument counterpart see: in particular, the unexpected “place seen” argument is mentioned explicitly rather infrequently (especially in comparison to the parallel “place put” argument of the Arrernte verb ‘put’). This finding suggests that statistical patterns of argument occurrence, and the syntactic/semantic alignments these reflect, may be materially more similar across languages than a comparison of the languages’reference grammars would indicate. Although how a language is used is obviously related to its grammatical system, usage patterns and grammar are not identical—and it is usage that may give us the more accurate clues to what is “natural” for language learners. Of course, eventually children figure out the grammar of their language, even those elements that are difficult or unexpected from the point of view of their developing cognitive system. In this learning process, statistical patterns of language use may often align well with the grammatical structure to be acquired. But in cases where they do not, like that of Arrernte ‘look, see’, statistics could provide a bridge between what children expect and what is needed (and felt to be “natural” in the socially shared communication system controlled by adults). Thus, children might initially rely heavily on statistical patterns of argument realization in adult speech in making guesses about the argument structure and meaning of verbs, but as they get further into the structure of their language as an integrated sociocultural tool for communication, and discover finer grained evidence about argument structure, they may sometimes need to fine-tune their hypotheses. ACKNOWLEDGMENTS I thank Melissa Bowerman, Penny Brown, and Cynthia Fisher for their helpful comments on earlier drafts of this chapter. Melissa’s constant questions about the

162

WILKINS

logic of the argumentation made for a far better chapter than the reader otherwise would have encountered. With respect to the research for the chapter, I am especially indebted to teachers at the Yipirinya and Ltyentye Apurte schools, and members of the Intelyape-lyape Akaltye early Arrernte curriculum development project. Two members of the Arrernte community, Rosalie Riley and Veronica Dobson, deserve special thanks for their help in taping and interpreting the data discussed in section 4. Field trips relating to this work were financed by the Max Planck Gesellschaft, and the research was nurtured in the crucible of the Language and Cognition group and Argument Structure project of the Max Planck Institute for Psycholinguistics. NOTES 1

The following abbreviations are used in the interlinear glosses: 1, ‘first person’, 2, ‘second person’, 3, ‘third person’, ACC, ‘accusative’, ABL, ‘ablative’, AFT, ‘afterative’, ALL, ‘allative’, ASS, ‘associative’, ASSERT, ‘assertion’, AVS, ‘aversive’, BUT, ‘focus change: but, although, whereas, as well as’, CONT, ‘continuous’, ClausComp, ‘clause complement’, DAT, ‘dative’, dl, ‘dual’, DS, ‘different subject’, EMPH, ‘emphatic’, ERG, ‘ergative’, FOC, ‘focus’, GEN, ‘genitive’, GenEvt, ‘generic event’, Hd, ‘head’, INCH, ‘inchoative’, IMP, ‘imperative’, LOC, ‘locative’, narr.pres, ‘narrative present’, NOM, ‘nominative’, pl, ‘plural’, pres, ‘present’, pro-copy, ‘pronominal copy’, PROP, ‘proprietive’, pst, ‘past’, PURP, ‘purposive’, REFL, ‘reflexive’, REL, ‘relative clause’, sg, ‘singular’, S, ‘subject’(of both transitive and intransitive verbs; relevant for pronouns, which—except for 1sg—show nominative/accusative marking; in glosses, the case that would be seen if the NP were a simple common noun is also shown, e.g., 3sgS[=ERG]), SREL, ‘subject relative clause’, REMEMB, ‘you remember the one’, SBSQNT, ‘subsequent’, SS, ‘same subject’, V, ‘verb’, 2ndPRED, ‘secondary predicate’. 2 This form of nominalization reduplicates the last part of the verb stem and adds -nhe to both original and copy (e.g., arrerne- ‘put’ becomes arrerne-nherne-nhe). The resultant form always refers to someone or something that is habitually involved in the performance of the verb action described by the stem. The involvement in the action may be in any of the standard argument roles semantically associated with the verb. Thus, Arrernte ‘hit’ when nominalized in this way can mean either ‘a hitter’ (e.g., a boxer) or ‘a thing hit’ (e.g., a punching bag). These are the only two interpretations it can have, and, consistent with the Relative Clause test, Arrernte ‘hit’ has only two arguments. Similarly, Arrernte ‘faint’, which by the test has only one argument, has only one possibly interpretation in the reduplication (‘someone/something who faints’). What happens with arrerne- ‘put’? Not surprisingly, arrerne-nhe-rne-nhe has three possible interpretations: ‘someone/something that habitually puts’(e.g., a sorter); ‘something that is habitually put/placed’ (e.g., markers [place keepers] in a game); or ‘a place where things are habitually put’ (e.g., a storage area). Most notably for our discussion here, the nominalization of are- ‘look, see’, are-nhe-are-nhe,

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

163

also has three possible interpretations: ‘someone/something that always sees/ watches’ (e.g., a lookout person); ‘something that is habitually seen/looked at’ (e.g., a film); or ‘a place where a thing is habitually seen’ (e.g., a cinema). In other words, the number and type of interpretations of this form of nominalization is fully consistent with the number of arguments the Relative Clause test reveals. 3 Note that the Dative NP of are- ‘look, see’ does not need to be adjacent to the Accusative NP; its word order, just like that of the Dative argument of arrerne- ‘put’, is free. 4 Are- ‘look, see’, unlike arrerne- ‘put’, may take a complement clause indicating the action perceived (see examples 1a and 1g), and one could argue that this clause is an argument of the verb. Thus the argument structures of the two verbs would be partially distinct. In truth, I am unclear about the syntactic status of such “perception complements” (the Relative Clause test cannot be used with complement clauses). Still, in terms of the three-NP structure, the complement clause does not replace any of the arguments but can co-occur with them, and far more often than not there is no complement clause. 5 I thank Cynthia Fisher for drawing my attention to this possibility. 6 Although this is a bilingual community with a bilingual education program, children in the earliest grades (from age 6 on) are taught primarily in Arrernte. Their parents talk to them mainly in Arrernte, and sometimes in English. There appears to be no effect of English on their syntax in Arrernte, only lexical borrowing (Wilkins, 1993a). 7 I have no idea when children acquire a meaning for this verb; on the grounds of frequency, it is likely by at least age 3. 8 Another possible explanation for the children’s errors is interference from English (the children are to varying degrees bilingual). But see note 6. 9 The following argument was developed in collaboration with Melissa Bowerman, who noted the need to reconcile the “standard” account of how children could use statistical patterns of argument realization with my own “cognitive naturalness” argument. 10 These data are in line with Bowerman’s (1985) more general hypothesis that errors in acquiring unusual systems may take the form of patterns that are crosslinguistically more common.

164

WILKINS

APPENDIX: CORPUS INVESTIGATION OF ARRERNE- ‘PUT’ AND ARE- ‘LOOK’ IN ADULT SPEECH—PERCENTAGE OF DIFFERENT ARGUMENT REALIZATIONS arrerne- ‘put’ [n = 53]

are- ‘look’ [n = 174]

All three arguments ERG ACC DAT V

5.7% [3]

0.6% [1]

ERG ACC V DAT

5.7% [3]

4.6% [8]

ERG DAT V ACC

1.9% [1]

0

ERG DAT ACC V

3.8% [2]

1.1% [2]

ERG V ACC DAT

0

6.9% [12]

ACC ERG V DAT

5.7% [3]

0

ACC ERG DAT V

1.9% [1]

0.6% [1]

ACC V ERG DAT

0

0.6% [1]

ACC DAT ERG V

0

0.6% [1]

DAT ERG V X ACC

1.9% [1]

0

DAT ERG V ACC

0

1.7% [3]

DAT ERG ACC V

1.9% [1]

1.1% [2]

ERG DAT V ERG ACC V ERG ACC X V

7.5% [4] 1.9% [1] 0

1.1% [2] 18.4% [32] 1.7% [3]

ERG V DAT ERG V ACC ERG V X ACC ACC ERG V ACC V ERG ACC DAT V ACC V DAT ACC X V DAT DAT ERG V DAT ACC V

9.4% [5] 0 0 3.8% [2] 0 7.5% [4] 1.9% [1] 1.9% [1] 1.9% [1] 1.9% [1]

0.6% [1] 10.9% [19] 0.6% [1] 8.0% [14] 0.6% [1] 3.4% [6] 1.1% [2] 0.6% [1] 1.1% [2] 1.1% [2]

DAT V ACC V ACC DAT V ERG ACC

0 0 0

Two arguments

1.7% [3] 0.6% [1] 0.6% [1]

Just one argument ACC V

11.3% [6]

17.8% [31]

ACC X V DAT V DAT X V

1.9% [1] 11.3% [6] 1.9% [1]

1.1% [2] 0.6% [1] 0

7.5% [4] 0

0.6% [1] 9.8% [17]

V DAT V ACC

7.

LEARNING ‘PUT’ AND ‘LOOK’ IN ARRERNTE

165

REFERENCES Allen, S. E. M., & Schröder, H. (2003). Preferred Argument Structure in early Inuktitut spontaneous speech data. In J. W. Du Bois, L. E. Kumpf, & W. J. Ashby (Eds.), Preferred argument structure: Grammar as architecture for function (pp. 301–338). Amsterdam: John Benjamins. Berman, R. A., & Slobin, D. I. (Eds.). (1994). Relating events in narrative: A crosslinguistic developmental study. Hillsdale, NJ: Lawrence Erlbaum Associates. Bowerman, M. (1985). What shapes children’s grammars? In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition, Vol. 2: Theoretical issues (pp. 1257–1319). Hillsdale, NJ: Lawrence Erlbaum Associates. Bowerman, M. (1990). Mapping thematic roles onto syntactic functions: Are children helped by innate linking rules? Linguistics, 28, 1253–1289. Comrie, B. (1993). Argument structure. In J. Jacobs, A. von Stechow, W. Sternefeld, & T. Vennemann (Eds.), Syntax: Ein internationales Handbuch zeitgenössischer Forschung/ An international handbook of contemporary research, 1. Halbband/Vol. 1 (pp. 905–914). New York: Walter de Gruyter. Fisher, C. (1996). Structural limits on verb mapping: The role of analogy in children’s interpretation of sentences. Cognitive Psychology, 31, 41–81. Fisher, C. (2002). Structural limits on verb mapping: The role of abstract structure in 2.5-year-olds’ interpretations of novel verbs. Developmental Science, 5, 55–64. Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55. Green, J. (1994). Learner’s guide to Arrernte. Alice Springs: Institute for Aboriginal Development. Harkins, J., & Wilkins, D. P. (1994). Mparntwe Arrernte and the search for lexical universals. In C. Goddard & A. Wierzbicka (Eds.), Semantic and lexical universals (pp. 285–310). Amsterdam: John Benjamins. Henderson, J. (Ed.). (1986). Arrernte ayeye (Arrernte stories). Alice Springs: Yipirinya School Council and Institute for Aboriginal Development. Henderson, J. (1998). Topics in Eastern and Central Arrernte grammar. PhD dissertation, University of Western Australia. Henderson, J. (2002). The word in Eastern/Central Arrernte. In R. M. W. Dixon & A. Y. Aikhenvald (Eds.), Word: A cross-linguistic typology (pp. 100–124). Cambridge: Cambridge University Press. Henderson, J., & Dobson, V. (1994). Eastern and Central Arrernte to English dictionary. Alice Springs: Institute for Aboriginal Development. Lee, J. N., & Naigles, L. (2005). The input to verb learning in Mandarin Chinese: A role for syntactic bootstrapping. Developmental Psychology, 41, 529–540. Pinker, S. (1984). Language learnability and language development: The acquisition of argument structure. Cambridge, MA: Harvard University Press. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Rispoli, M. (1987). The acquisition of transitive and intransitive action verb categories in Japanese. First Language, 7, 183–200. Rispoli, M. (1991). The mosaic acquisition of grammatical relations. Journal of Child Language, 18, 517–551. Rispoli, M. (1995). Missing arguments and the acquisition of predicate meanings. In M. Tomasello & W. E. Merriman (Eds.), Beyond names for things: Young children’s acquisition of verbs (pp. 331–352). Hillsdale, NJ: Lawrence Erlbaum Associates. Silverstein, M. (1976). Hierarchy of features and ergativity. In R. M. W. Dixon (Ed.), Grammatical categories in Australian languages (pp. 112–171). Canberra: Australian Institute of Aboriginal Studies.

166

WILKINS

Tomasello, M. (2003). Constructing a language. Cambridge, MA: Harvard University Press. Turner, M.-M. (1994). Arrernte foods: Foods from Central Australia—Nhenhe-areye anwerne-arle arlkweme. Alice Springs: IAD Press. Wilkins, D. P. (1986). Particle/clitics for criticism and complaint in Mparntwe Arrernte (Aranda). Journal of Pragmatics, 10, 575–596. Wilkins, D. P. (1988). Switch-reference in Mparntwe Arrernte (Aranda): Form, function, and problems of identity. In P. Austin (Ed.), Complex sentence constructions in Australian languages (pp. 141–176). Amsterdam: John Benjamins. Wilkins, D. P. (1989). Mparntwe Arrernte: Studies in the structure and semantics of grammar. PhD dissertation, Australian National University, Canberra. Wilkins, D. P. (1991). The semantics, pragmatics and diachronic development of “associated motion” in Mparntwe Arrernte. Buffalo Papers in Linguistics, 207–257. Wilkins, D. P. (1993a). Assessment of Central Arrernte children’s vernacular language use at Yipirinya School: What impact is English having on the acquisition of Arrernte? Report commissioned by the Yipirnya School Council. Yipirinya School: Alice Springs. Wilkins, D. P. (1993b). Linguistic evidence in support of a holistic approach to traditional ecological knowledge. In N. Williams & G. Baines (Eds.), Traditional ecological knowledge: Wisdom for sustainable development. Canberra: CRES 71–93. Wilkins, D. P. (1997). The verbalization of motion events in Arrernte (Central Australia). In E. V. Clark (Ed.), Proceedings of the 28th Annual Child Language Research Forum (pp. 295–308). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Wilkins, D. P. (2000). Ants, ancestors and medicine: A semantic and pragmatic account of classifier constructions in Arrernte (Central Australia). In G. Senft (Ed.), Systems of nominal classification (pp. 147–216). Cambridge: Cambridge University Press. Wilkins, D. P. (2006). Towards an Arrernte grammar of space. In S. C. Levinson & D. P. Wilkins (Eds.), Grammars of space (pp. 24–62). Cambridge: Cambridge University Press. Wilkins, D. P., & Hill, D. (1995). When “go” means ”come”: Questioning the basicness of basic motion verbs. Cognitive Linguistics, 6, 209–259.

CHAPTER 8

Verb Specificity and Argument Realization in Tzeltal Child Language Penelope Brown Max Planck Institute for Psycholinguistics

1. INTRODUCTION How do children learn a language whose arguments are freely ellipsed? The Mayan language Tzeltal, spoken by an indigenous population of Mayan Indians in southern Mexico, is such a language. The acquisition pattern for Tzeltal is distinctive in at least two ways: (a) verbs predominate even in children’s early production vocabulary, and (b) these verbs are often very specific in meaning. This runs counter to the patterns found in most Indo-European languages, where nouns tend to predominate in early vocabulary and children’s first verbs tend to be “light” or semantically general. In this chapter I explore the idea that noun ellipsis and semantically “heavy” verbs are related: the “heavy” verbs restrict the range of possibilities that their nominal arguments can refer to and so allow recovery of “missing” nouns. The particular proposal I explore here is this: the predominance of semantically specific verbs in Tzeltal children’s first transitive verb combinations is related to patterns of argument ellipsis in the adult language, and it suggests a different verb-learning strategy for Tzeltal children than for children learning English. The data on which this analysis is based are samples drawn from a longitudinal corpus of the spontaneous production of four Tzeltal children, interacting with caregivers, recorded between the ages of roughly 1 and 4 years. In what follows, section 1 reviews the earlier findings for Tzeltal child language in light of claims in the literature about how children learn verbs. This establishes the problem to be explained. In section 2, I sketch the linguistic background for Tzeltal and survey earlier work on argument realization in child and adult speech, to establish the plausibility of my proposed solution to the problem. Section 3 takes up my candidate explanation: it spells out what I mean by semantic specificity in 167

168

BROWN

Tzeltal verb meanings and presents the results of testing my hypothesis that Object (henceforth O) arguments will be realized differently with transitive verbs of different semantic “weight.” Section 4 discusses the findings in the light of Mayan language typology, and section 5 assesses the theoretical implications of these results for theories of word learning and of the pragmatics of argument realization. 1.1. Explaining Early Tzeltal Child Language Two characteristics of early child language production in Tzeltal are strikingly different from that reported for the majority of children learning Indo-European languages. First, Tzeltal is very much a “verb-friendly” language for learners, with verbs predominant in the children’s early vocabulary even at the one-word stage. A study of two Tzeltal children’s initial productive vocabulary development1 showed that they do not have a big burst in noun vocabulary (as measured by their spontaneous productions) before they have a burst in verb vocabulary (Brown, 1998a, 1998b): New verbs and nouns appear together from the earliest recordings, at first in roughly equal numbers in each session. For both children, … before their MLU exceeds 1.5, new verbs outnumber new common nouns (not including proper names) in their cumulative production vocabulary.…And by the time morpheme combinations are frequent,…still well before the MLU 2.0 point, new verbs outnumber all new nouns, including proper names. (Brown 1998b: 720)

This picture is generally at odds with acquisition patterns in the Indo-European languages, where most child language studies have concentrated. The Tzeltal child data are actually more in line with the picture emerging for Korean and Chinese (Choi, 1997, 1998; Tardif, 1996, 2006), as well as for other Mayan languages (de León 1999a, 1999b, 2001; Pfeiler, 2003a, 2003b), so perhaps Tzeltal is not so exotic after all from a broader perspective than that of Indo-European. Second, among Tzeltal children’s very early verbs are many semantically specific ones; in fact, for transitive and positional verb roots, semantically general verb types are outnumbered by semantically specific ones in the children’s speech. This contrasts with the general tendency observed in children learning Indo-European languages to depend in their early productions on a rapidly increasing noun vocabulary with heavy reliance on just a few general verbs like ‘make’, ‘do’, ‘give’, and ‘get’ (Clark, 1993). By semantic specificity I have something quite precise in mind. A typological characteristic of Mayan languages (and in fact, many other Amerindian languages) is that many basic-level transitive and positional verb roots incorporate into their semantics features of the arguments with which such verbs can co-occur. There are, for example, many different verbs for eating, or carrying, or breaking, or setting something down, or tipping something over, depending on the shape, substance, position, or orientation of the objects corresponding to the internal or (sometimes)

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

169

to the external arguments. These specific verbs are basic-level in the sense of being monomorphemic, in frequent everyday use, and not specialized to a particular register. Sometimes there is a superordinate term for the semantic domain (e.g., there are nine specific verbs in the eating domain but there is also a verb meaning ‘eat anything’). Often there is no superordinate term (e.g., there is none for verbs for different kinds of ‘cut’, ‘break’, ‘carry’, or ‘be-positioned’). However, it is these specific verbs that appear prominently among the earliest words in Tzeltal children’s vocabularies (e.g., lo’ ‘eat soft things’, we’ ‘eat tortilla-like things’, k’ux ‘eat crunchy things’), whereas the superordinate terms (where they exist, e.g., tun ‘eat anything’) do not. This kind of semantic specificity is restricted to transitive and positional roots; intransitives are all semantically general in the sense that they do not place restrictions (other than, in some cases, animacy) on what their single argument can be. 1.2. The “Light Verb” Hypothesis The appearance of many specific verbs among the earliest words in Tzeltal children’s vocabularies conflicts with the presumption that it is easier to learn general (or “light”) verbs (like ‘give’, ‘get’, ‘do’, ‘make’, ‘want’), based on (putatively) universal meanings, than it is to learn verbs with language-specific meanings (Clark, 1993; Goldberg, 1995). Of course, even though Tzeltal children produce many semantically specific verbs among their first verbs, it is possible that initially they have overgeneralized “light” meanings for these verbs; this, however, does not appear to be the case. I have examined the contexts in which specific verbs are used in order to see whether the children attend to the specificity of such verbs from when they first start using them (at the one-word stage), or whether the semantics of their early word use shows evidence of biasing from universal (general) concepts. The evidence from the children’s production (although comprehension might be a different story) suggests that these verbs are semantically specific, restricted to an appropriately limited set of contexts, from the beginning of production. For example, Lus (2;0), complaining that the dog has carried off her lollipop, says lut bel tz’i’ ‘the dog carried it off’, using the verb lut, which means ‘carry long thing in mouth’; she does not overuse this verb for carrying off (or stealing) non-long things or carrying them not in the mouth.2 The Tzeltal data also go against the proposal by Anat Ninio (1996, 1999), based on English and Hebrew data, that children initially rely on semantically general—or “light” —“pathbreaking” verbs to break into argument structure when they begin combining words into sentences. Ninio claims that semantically light verbs (like ‘do’, ‘make’, ‘give’, ‘go’) are the first to start children off on syntactic learning, because their meanings involve little more than highly general relations between arguments. These are thus “pathbreaking verbs” that surface whenever there is a significant advance in verb syntax, leading the way “precociously” with a significant time lag until other verbs are used in the pattern. New syntactic learning,

170

BROWN

Ninio argues, is initially item-based and lexical with later categorical knowledge based on a generalization from a few pathbreaking verbs (Ninio, 1996, 1999). So, Ninio’s claim is that, although the young child may know some specific verbs, when combining words into new constructions she at first relies on general verbs. Other researchers have also argued for the primacy of semantically general verbs in initiating syntactic learning (e.g., Hollebrandse & van Hout, 1994, 1998; Goldberg, 1995, 2005; Goldberg, Casenhiser, & Sethuraman, 2004). However, the Tzeltal children’s few general-purpose (or “light”) verbs do not appear to play this role. For two children whose early productions were examined in detail (Brown, 1998b), at least for transitive and positional verbs, semantically general-purpose verbs do not lead the way either in early morpheme combinations as a whole or in specific constructions. In fact, for these two children it is actually semantically specific verbs that play a dominant role in early combinations with transitive argument structure.3 In contrast, the children’s early intransitive verbs are indeed semantically general—for example, ‘exist’, ‘go’, ‘come’; these, however, are among the five most frequent verbs in adult speech and would be expected to appear early on frequency grounds alone. A somewhat different approach to children’s early light verbs is found in the work of Adele Goldberg. In early data from English-learning children (Goldberg et al., 2004), Goldberg has found that the most frequent and earliest acquired verbs correspond to the semantic prototype of the construction—that is, she finds overwhelming predominance of light verbs in their particular constructions (give in the ditransitive construction S V REC O; put in the caused motion construction S V O Obl; go in the intransitive motion construction S V Obl). Other verbs do occur in these constructions but much less frequently. The children’s production matches the input, where light verbs predominate numerically in these constructions too. Goldberg’s explanation is this: the fact that light verbs are semantically general means that they are widely applicable, across many contexts, and they code scenes basic to adults and children everywhere. Because they predominate in the construction, children come to associate the meaning of that verb with the constructional meaning. Thus in Goldberg’s picture of acquisition, the child moves from initially categorizing verbs on the basis of input utterances into what are at first verb-centered constructional categories (verb islands); these then get generalized to the construction according to their frequency in the language. So where light verbs are the most frequent, one would expect those meanings to predominate in the meanings of the construction—as they do, even in Tzeltal, for intransitives. The problem with generalizing this picture to Tzeltal is that for transitive and positional verbs, these early verb island constructions are not predominately made with light verbs.4 1.3. The Role of Frequency Why do we find the predominance of semantically specific verbs in early Tzeltal child production? One obvious candidate for explaining their presence among

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

171

children’s first verbs is frequency. Perhaps, just as general verbs are the most frequent ones in English, specific verbs are the most frequent ones in Tzeltal. But a check of verb type frequencies in a Tzeltal sample of child-directed speech 5 (CDS) indicates that this cannot be the whole story, a story that differs for transitive and intransitive verbs. Frequency may be the explanation for which intransitive verbs are first learned by Tzeltal children; these include the most frequent verbs in the language: ay ‘exist’, ba ‘go’, and tal ‘come’, in a way reminiscent of English. But, although the most frequent transitive verbs in adult Tzeltal speech are indeed also semantically general (e.g., ‘want’, ‘see’, ‘give’, ‘do/make’), just as in English, these are not necessarily the transitive verbs that first enter into children’s combinations. Rather, these are specific verbs (e.g., lo’ ‘eat soft things’), which are not among the most frequent 20 verbs in the language (see Brown, 1998b). So in what I discuss here, we leave aside intransitives and focus on semantically specific transitives that restrict the kinds of things they can take as patient (O) arguments. The Tzeltal children’s emphasis on specific verbs casts some crosslinguistic doubt on the core role of highly frequent general-purpose verbs in the early stages of language acquisition—at least as a universal strategy. Such a strategy may work well for particular kinds of languages; it may be a good one for learners of English and Hebrew. But it does not seem to be for Tzeltal. 1.4. Argument Ellipsis: The Current Proposal These two properties of Tzeltal child production data—many very early verbs, and among them many semantically specific ones—have prompted me to consider whether they might be related to another feature of Tzeltal as actually spoken: Tzeltal is a language with massive argument ellipsis.6 One obvious potential consequence of semantic specificity in transitive verbs is that, for these verbs, the search space for arguments is radically reduced, and therefore overt noun phrase arguments may be even less often required in natural speech than they are for general verbs. If, for example, you say ya jk’an jlo’ ‘I want to eat (it)’ using the verb -lo’, which is specific to soft foods like bananas, it may be much less often necessary in context to specify exactly what you want to eat. You do not have to look very far to see what the referent of the O argument could be. If, however, you say ya kich’ ‘I get (it)’, using the light verb -ich’ with which a very wide range of arguments is possible, it may be more often necessary to spell out what it is you intend to get. This might then be a semantic factor interacting with the pragmatic factors influencing argument realization in adult speech. If so, and if young children are sensitive to such properties of the input, perhaps the degree of argument realization could be a clue to young children that the meaning of the verb somehow incorporates (at least some features of) the meaning of its O argument. In other words, if a child at this age is able to make use of pragmatic information—what her interlocutor is assuming about what she (the child) needs to know—she may be able to reckon that if the

172

BROWN

O argument does not have to be mentioned explicitly, she must be able to recover it either from the context (because, e.g., it is physically present and the focus of attention) or from verb semantics (because the verb actually means ‘eat-squishy-things’, for example, and in her small child world there are not many different candidates for squishy things to be eaten). In this chapter, I argue that the semantic specificity of Tzeltal verbs is indeed a possibly crucial ingredient in Tzeltal children’s early transitive verb learning. This is for two reasons: • Concreteness: Tzeltal verb semantics is relatively concrete, specific, “nouny.” So whatever makes concrete nouns relatively easy to learn in other languages may also make these Tzeltal verbs easy to learn: their referents are more concrete, the range of contexts they apply in is more easily delimited. • Redundancy: Information about the O argument’s referent is carried both in the verb and in the Object NP. When the latter is ellipsed, this information is still carried in the verb. In this language, one therefore might well expect more NP ellipsis with semantically specific verbs, both in the input and in children’s speech. The specific hypothesis to be tested is the following: For Tzeltal adults and children, the O argument of a transitive verb is realized lexically less often when the verb is specific (like lo’ ‘eat soft things’), and more often when the verb is general (like ich’ ‘get’). Or to put it the other way around: we will find more ellipsis of O arguments with semantically specific verbs.

Before considering this hypothesis, we need some background information about what the Tzeltal child needs to learn about the structure of transitive sentences in Tzeltal, and about constraints on argument realization. 2. BACKGROUND 2.1. The Basic Tzeltal Transitive Sentence Tzeltal is a VOS language with obligatory aspect marking that is different for transitive and intransitive verbs, and with ergative/absolutive person cross-referencing on the verb.7 Argument structure (at least transitive vs. ditransitive vs. intransitive) is therefore always morphologically coded in the input. The minimally required morphology for transitive verbs is given in (1) and exemplified in (2); overt subject and object NPs are optional:8

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

173

1. ASPECT + ERG + VERB STEM + ABS (+ OBJ NP) (+ SUBJ NP) 2. ya ICP

s-nutz-on

(tz’i’).

3ERG-chase-1ABS

(a dog)

‘It is chasing me.’ (‘A dog is chasing me.’)

Core arguments are obligatorily cross-referenced on the verb; thus (2) is perfectly grammatical with no overt NPs expressed. Arguments may be overtly realized in two additional ways: by overt pronouns (which in adult speech are emphatic, and relatively rare), as in (3), and by a full noun or noun phrase, as in (4). 3. jo’on I/me

ya

s-nutz-on

(te tz’i’-e)

ICP

3ERG-chase-1ABS

(ART dog-CL)

‘(It’s) me it (the dog) is chasing.’ 4. ya ICP

s-nutz-0

y-ajwal

te tz’i’-e

3ERG-chase-3ABS

3ERG-master

ART dog-CL

‘The dog is chasing its master.’

2.2. Argument Realization in Tzeltal Adult and Early Child Language Tzeltal nominal arguments can be freely ellipsed if their referents are clear in the context. Alternatively, for first and second person they can be expressed by an independent pronoun; in adult speech this is used only for special emphasis. Or they can be realized lexically, with a full NP. How do Tzeltal children realize arguments in their early productions? Some examples from a child of 26–29 months, with MLU less than 2.0, are given in Table 8.1 to give a sense of the range of argument realization patterns in young children’s productions. No examples were found in these early child data where both A and O are lexical. Otherwise, children from about the age of 2;0 produce utterances with all of these argument realization patterns. What then constrains their argument ellipsis? One important candidate is preferred argument structure (PAS), the tendency to ellipse arguments according to a universal pattern—to provide only one new (informative) argument per clause, usually in S or O position (Du Bois, 1987, 2000; Du Bois, Kumpf, & Ashby, 2003). But do children know about PAS? Could they use it as a clue to verb meaning? An affirmative answer is suggested by the work of Shanley Allen for children learning Inuktitut (Allen, 2000, and this volume; Allen & Schröder, 2003), and Pat Clancy for Korean (Clancy, 1993, 1996, 2003). They

174

BROWN TABLE 8.1 Examples of Child Utterances with Overtly Expressed Lexical NPs (MIK, 2;2—2;5)*

Child Speech

Adult Target

Ellipsed [=Null] O and A: ya _tzak

ya j-tzak

ASP grasp

ICP 1E-grasp

‘(I) grasp (it).’

‘I grasp (it).’

Only O is lexical: _pet i _tz’i’-tik

ya j-pet i j-tz’i’-tik

hold.in.arms this dog-1pl.incl

ICP 1E-hold.in.arms this 1E-dog-1pl.incl

‘(I) hold (our) dog.’

‘I hold our dog.’

Only A is lexical: _s-we’ _papa-tik

ya s-we’ j-papa-tik

3E-eat father-1pl.incl

ICP 3E-eat 1E-father-1pl.incl

‘(Our) father eats (it).’

‘Our father eats (it).’

A is pronominal, O is lexical: _lo’ lo’bal jo’on

ya j-lo’ lo’bal

eat banana I

ICP 1E-eat.soft.things banana

‘I eat a banana.’ OR:

‘I eat a banana.’ OR:

‘I eat my banana.’

ya j-lo’ j-lo’bal ICP 1E-eat.soft.things my-banana ‘I eat my banana.’

O is pronominal, A is null: _lutz jo’on

ya a’-lutz-on

cuddle I

ICP 2E-cuddle-1A

‘(You) cuddle me.’

‘You cuddle me.’

* Abbreviations: O = Object of Transitive; A = Agent of Transitive; _ Indicates missing cross-referencing and aspect morphemes which are obligatory for adults.

have demonstrated in detail the importance of pragmatic factors influencing argument ellipsis in child language. Similar findings come from Tzeltal, my analysis of the early child samples for two children showed some evidence that they are already sensitive by age 2;4–2;5 to the PAS constraints on argument realization: they are two to three times more likely to represent the O argument lexically than the A argument (Brown, 1998a). Narasimhan, Budwig, and Murty (2005) have found the same to be true of Hindi children.

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

175

This work shows that even very young children attend to PAS constraints. Thus the child from about 18 months appears to be a little Gricean, able to assess how explicit to be in relation to what can be presumed in the context. Because already at the two-word stage Tzeltal kids attend (in some sense) to PAS constraints, we may infer that they are sensitive to to some degree to the informational status of utterances, to given versus new information, and to what can be presumed in context as understood vs. what needs to be spelled out lexically. Therefore it seems not unreasonable to suggest that children could attend to degrees of argument realization in utterances in relation to semantic specificity. The question of whether, in addition to pragmatic factors, semantic factors like verb specificity also play a role in argument realization is of course a complicated issue to assess, partly because children vary individually in their word-learning strategies, but also because pragmatic reasons for argument ellipsis might be expected to obscure any effect (if in fact there is any) of semantic specificity. Such pragmatic factors include recency of mention, presence in context, query, contrast, person, and animacy—things that affect the pragmatic saliency of information at any point in the discourse (Allen, 2000, this volume). Despite these reservations, I wanted to see if Tzeltal children’s realization of arguments is related to the semantic specificity of the verb, because if indeed it turns out that verb specificity enhances the degree of O argument ellipsis in Tzeltal, this might be an important factor in the predominance of verbs in children’s early utterances. There are two reasons for why this might be so: • Because with a lot of ellipsis, verbs predominate numerically and “stand out” in the input. • More controversially, because with relatively nouny semantics (object property features encoded in the verb), it is somehow easier to learn the semantics of transitive verbs in Tzeltal than in Indo-European languages. So let’s look at how O arguments are realized in Tzeltal. 3. REALIZATION OF O ARGUMENTS WITH TRANSITIVE VERBS OF DIFFERENT SEMANTIC WEIGHT 3.1. Hypothesis Recall the hypothesis to be tested: for Tzeltal adults, and children, the O argument is realized lexically less often when the verb is specific (like lo’ ‘eat soft things’), and more often when the verb is general (like ich’ ‘get’). 3.2. Data The data examined consist of an adult sample of parental input utterances to Tzeltal children aged 3;7 to 3;8, and data from four children (Lus, Xan, Mik, and Xaw) at

176

BROWN

an age (3;4–3;9) when they are talking in fully grammatical sentences. (See Table 8.2 for details.) All samples are drawn from naturally-occurring conversation in the 9 children’s own homes. 3.3. Method First I selected samples of naturally occurring speech of about 800 utterances from the input speech of three adults, and 800 utterances for each child, and coded them for (a) grammatical transitivity, (b) syntactic role of arguments—Agent of transitive verb (A), Subject of intransitive verb (S), and Object of transitive verb (O), (c) whether the O arguments of transitive verbs were realized lexically, pronominally, or null (ellipsed), and (d) semantics of the verb. The verbs were classified into one of three semantic categories: general, specific, and “other,” those that do not fit into either category. Setting aside the “other” category, where semantic specificity makes no predictions about how the O argument will be realized, I then checked whether semantically specific transitive verbs are more likely to get null O argument realization than semantically general verbs.10 Semantic Weight. Verbs were categorized as semantically “heavy” or “light” on language-internal grounds: “general” (or “light”) verbs are those able to apply to a wide range of arguments. For these there are no selection restrictions on O (or virtually none); these are the classic light verbs in the literature: for example, ‘give’, ‘put’, ‘get’, ‘do’, ‘make’. Many other Tzeltal verbs are equally indifferent as to the nature of their O arguments: you can, for example, ‘look at’, ‘see’, ‘search for’, ‘fear’, ‘want’, almost anything, regardless of its specific properties. “Specific” (or “heavy”) verbs are language-specifically restricted to particular kinds of O arguments; they can only apply to certain classes of arguments in Tzeltal. For example, as mentioned earlier, the eating verbs are “heavy” in Tzeltal: each verb subcategorizes for a particular class of things eaten. This trait appears similarly for verbs of carrying, holding, breaking, inserting, opening, TABLE 8.2 Tzeltal Child Data Summary

Child

Age

Number of Verbal Utterances in Sample Coded

Number of Coded Utterances with Transitive Verbs

XAN

3;4

557

268

LUS

3;9

601

272

MIK

3;7

856

340

XAW

3.9

723

343

ADULTS

[to children 3;7–3;8]

504

282

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

177

positioning, and many others. Note that some “heavy” verbs may be specific in what O arguments they take by virtue of being specific about other things—for example, specific manners or places where the action occurs; lut ‘hold/carry long thing in mouth’, or lik ‘hold/carry by handle from above’, might appear to involve manner as much as O object restrictions. Yet restricting what you are talking about to what you can hold in your mouth or by a handle from above places obvious restrictions on what the O argument can refer to (a small object, a bucket or bag with a handle on top, respectively). In contrast, “other” is a hodge-podge set of verbs that select in ways that do not restrict what the O argument can refer to except perhaps on very general grounds like animacy or instrument, and they are not members of the crosslinguistically classic set of general verbs. They provide no basis for predicting how their O arguments would tend to be treated; for this reason they are set aside and the analysis compares only general versus specific verbs as defined earlier. Examples of verbs in each category found in the data examined are given in Table 8.3. Note that, because of the language-specific restrictions on multiple features of these semantically specific verbs, their glosses are necessarily shortcuts; many of these would require a line or two to specify exactly what the restrictions on the O object are.

TABLE 8.3 Semantically General/Specific and “Other” Verbs in Tzeltal General (“Light”)

Other

Specific (“Heavy”)

ak’ ‘give, put’

al ‘tell’

boj ‘cut [with machete or knife]’

a’y ‘feel, hear, experience’

bislun ‘fix, make (it) work’ [inanimate O]

busk’in ‘tip out, spill [small objects, from container]’

ich’ ‘get, bring, take’

butz’ ‘kiss, suck on’

chotan ‘stand (it) up [four-legged object]’

il ‘see’

chol ‘line up in a row’

chejpan ‘stand (it) up [sack]’

k’abu ‘look at’

chon ‘sell’

jatz ‘rip [cloth, paper]’

k’an ‘want’

chuk ‘tie up, put in jail’

jojk’an ‘hang up [from handle/ strap]’

le ‘search for’

ik’ ‘take with’ [animate O]

ok’esan ‘play [noise-making object or wind instrument]’

lok’ta ‘photograph’

ixlan ‘play with’ [inan. O]

kuch ‘carry [on headstrap or back]’

mulan ‘like (it)’

jel ‘exchange, trade’ [inan. O]

k’ok ‘break off, pluck [e.g., fruit, from stem]’

pas ‘do, make’

jip ‘throw something’

k’ux ‘eat [crunchy things]’ (continued)

178

BROWN Table 8.3 (continued)

General (“Light”)

xi’ ‘fear (it)’

Other

Specific (“Heavy”)

jojk’o ‘ask’

lik ‘lift/carry [by handle, from top]’

joyinta ‘encircle’

lim ‘spread out [cloth object]’

kananta ‘look after, protect’ lo’ ‘eat [soft things like bananas]’ k’ej ‘put away’ [inan. O]

luch ‘hit [with head], charge, gore’

lajin ‘finish’

metz’an ‘lay it down [on its side]’

lo’lo ‘tease’ [animate O]

nap’ ‘stick onto sticky surface’

lok’es ‘take out’ [inan. O]

nuj ‘invert, place upside-down [bowl-shaped object]’

mak ‘cover’ [e.g., with cloth] pay ‘boil [in water]’ maj ‘hit’

pet ‘carry [in arms]’

man ‘buy’ [inan. O]

p’ij ‘break [stems of plants]’

mayli ‘wait for’

set’ ‘cut [rope/wire with e.g. scissors]’

mil ‘kill, ruin’

tam ‘pick up small thing fallen to ground’

na’ ‘know’

tek’an ‘step on [two-footed]’

nak ‘hide (it)’

ti’ ‘eat [meat], bite’

net’ ‘push against’

tik’ ‘insert [through narrow opening]’

nit ‘pull [on rope/string]’

toch ‘rip, break off [e.g. bark, plaster]’

nutz ‘chase’ [animate O]

top’ ‘break [e.g. pottery or glass]’

otzes ‘insert, make (it) enter’ tuch ‘stand upright [long thin object]’ paj ‘compare’

tuy ‘cut [meat]’

poj ‘steal, take away from’

tzak ‘take in hand, grasp’

s-tak’ ‘be able to’

tz’ot ‘twist-insert [long/thin object]’

tzaj ‘choose, sort’

tz’ap ‘insert [long sharp object]’

tzob ‘gather together’

tz’it ‘wipe clean [bowl, with finger]’

tz’ibu ‘write’ [inan. O]

tz’us ‘close [hinged thing like door]’

ut ‘tell, do’

t’uxan ‘make fall [from standing-up stance]’

wo’ ‘toast [in front of fire]’

uch’ ‘drink’

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

179

3.4. Results Rates of argument realization in each of the three possible forms (null, i.e., cross-referencing alone, vs. pronominally, vs. lexically) in samples of adult input speech during the course of everyday interaction, and in the children’s samples, are shown in Figure 8.1. As this figure shows, semantically specific verbs do indeed receive overt O arguments less often, that is, they have a higher rate of argument ellipsis than general verbs in Tzeltal. This is true for the adults somewhat more (p < .01) than for the children (p < .05). There is considerable variation across the children: for two of them the results are like those of the adults, significant at the p < .01 level; for the other two the effect is weaker. We must therefore conclude that this pattern is variable across children at this age, and sensitive to pragmatic factors (topic, presence of referent in the discourse context, etc.) as well as to semantic ones. Yet these results are striking in light of the many different factors conditioning the overt realization of arguments. In addition to the obvious fact that pragmatic saliency makes it often unnecessary to mention the argument of general verbs as well as of specific ones, several other complications might lead one to expect a messier picture. One is that there is, under certain conditions, a tolerance for redundancy. In the adult sample it is clear that speakers sometimes keep using the lexical noun representing the O argument even on repeating the verb four or five times in the same context. In some cases these are idioms where the NP is never ellipsed even on repetition; these idioms were omitted from the counts (e.g., pas choke ‘make a crash [cars]’; nup’betik sk’alel ‘blow on his fire’ [i.e., get him angry]; laj yo’tan, ‘his heart finished’ [i.e., ‘he is done’]). But in addition, when stating a general rule one often repeats the noun: e.g., ya stuy jk’abtik ‘it slices our

FIGURE 8.1. O argument realization in Tzeltal, for adults and children.

180

BROWN

hands’, net’ bel jk’abtik ‘it pulls away at our hands’, xat’ jk’abtik ‘it splits our hands’, lajin jk’abtik ‘it finishes off our hands’ are all comparable ways of stating the rule that lamina (corrugated iron) can cut you. There are thus various reasons for using the explicit O argument noun even when it’s obvious in the context, which could obliterate any effect of specificity. There are also other reasons for not needing to specify the O argument—even semantically general verbs may have, in practice, in the limited contexts applicable to these children, only a handful of possible arguments, obvious in context. Nonetheless, the results clearly support the hypothesis: Tzeltal adults are indeed less likely to represent the O argument of transitive verbs lexically if it is a specific verb (like ‘eat soft things’) than if it is a general verb (like ‘get’). So are Tzeltal children by around age 3;6, although to a lesser degree than the adults (p < .05). It might be objected that this effect—less overt argument realization with specific verbs than with general verbs—could be due not to the nature of the verb semantics but to the well-known pragmatic constraints on argument realization: that arguments are less likely to be overtly realized when they refer to old (given) information, or to things physically present in the context, or to pragmatically prominent things. But there seems to be no reason to expect these contextual conditions to cluster with semantically specific verbs—no reason to think that a specific verb like lut ‘hold/carry long-thin object in mouth’ is more likely to be uttered in contexts where the object is given information, or is physically present, say, than a general verb like ich’ ‘get’. Rather, the richness of information carried in the verb seems likely to affect the probability of a felt need, in a given context, to explicitly spell out what the O argument refers to, even when pragmatic prominence would push in the other direction. The pattern of argument ellipsis in adult speech offers the Tzeltal child the possibility of inferring from rate of argument ellipsis to verb semantics: object ellipsis suggests recoverability, not only from the context, but also from the verb meaning (specific vs. general). We may conclude that argument realization (the rate of Object NP ellipsis) could provide a clue to verb semantics (specific vs. general) for the Tzeltal child, and hence function as an aid to verb learning. The finding reported here does not in itself establish that verbs are easier in Tzeltal because their semantic specificity is reflected in rates of argument ellipsis. But it does suggest a plausible link that could be explored in further research; namely, that the Mayan trait of coding certain object properties in verbs, instead of in nouns, may provide some of the basis for the earliness of Tzeltal children’s verbs. This is because the reference of such semantically specific verbs is restricted to a relatively coherent set of extensions, delimited by the properties of the referents of the nominal arguments the verbs can take. For verbs like these, a child has to construct categories of verb meanings that are less diverse—are applicable to fewer contexts—than for semantically general verbs. In essence, in order to master the verb’s meaning the child has to generalize from first hearing a word in a particular context to fewer different kinds of contexts; she has to divide the world up into

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

181

smaller categories of verb-relevant contexts. A corollary of this view is that because these categories are language specific, there is a large role for the input language in the learning process. 4. DISCUSSION: FACTORS INFLUENCING CHILDREN’S EARLY VERBS 4.1. Light versus Heavy Verbs in Early Child Verb Learning There are two opposing proposals (see Bowerman, 2005) about the level of abstraction that children find easier in learning verb meanings: (a) general/ light verbs are easier because their meanings do not go beyond pure argument structure (see section 1.2), versus (b) semantically specific verbs are easier because there is less to extend, so children do not have to work so hard to figure out what the boundaries of the verb’s semantic category are. The Tzeltal data examined to date make me lean toward the latter, at least for this language. But it may not always be the case. For example, we may speculate that in a language with obligatory NPs (like English), a light verb strategy is sensible; in a language like Tzeltal, with massive NP ellipsis, a heavy verb strategy is perhaps better. Expressing this proposal from the point of view of a child, it’s as if a child learning English could be thinking: “Verbs are tricky, so I’ll stick with a handful that are general enough to be most useful and let the noun provide the O object reference.” Thus the young English-speaking child often says things like: “do my hair”, “do my belt”, “do my dress”, “do my toy”. The Tzeltal child, however, could be thinking more along these lines: “I’ll stick to verbs that have clear well-defined contexts of use.” So she says things like: “braid” (my hair), “tie” (my belt), “put-on-clothing” (my shoes), “insert-tightly” (my toy), and resists generalizing these verbs across contexts until she hears positive evidence in the input for each verb. The findings (for adult input and for these four children) are consistent with the idea that the semantic specificity in Tzeltal verbs helps children to learn verb meanings, or at least to be conservative about generalizing verb meanings beyond the contexts where they are heard in the input. We might even want to consider the more radical claim that in Tzeltal, transitive verbs are easier to learn than in other languages because they are semantically “nounier” (more informationally rich about the object properties of arguments).11 4.2. The Influence of Typology There is also something about the nature of Mayan noun and verb semantics more broadly of interest for theories of child word learning. Although, from an Indo-European perspective, nouns canonically label individuable things, whereas verbs mostly label activities, processes, or states, in Tzeltal and other Mayan languages neither verbs nor nouns fit this picture very well. On the one hand, it is at least arguable (cf. Lucy, 1992) that inanimate concrete nouns in

182

BROWN

Tzeltal are predicate-like, and label unindividuated “stuff” or material properties of objects—for example, the same word lo’bal can mean ‘banana fruit’, ‘banana tree’, ‘banana leaf’, and so on. Thus noun semantics—at least for such inanimate nouns—omits individuating features; these nouns have to be individuated by a numeral classifier. On the other hand, as we have just seen, many verbs incorporate in their semantics specific features of the nominal arguments that can go with them. Like the specificity provided in numeral classifiers for nouns, such that the classifier can stand on its own for the whole NP, these Tzeltal verbs are like classifiers for actions because they apply only to actions with respect to specific kinds of objects (e.g., specific kinds of foods, places of carrying, positions, shapes, or orientations of O). This is a different kind of specificity from that provided, for example, by a manner component in verbs of other languages (e.g., English, German, and Dutch): the manner component qualifies/specifies the nature of the action/ motion/state, whereas verb specificity in Tzeltal qualifies/specifies what kinds of things the action can apply to (although it may implicitly thereby indicate manner). Hence its relevance for argument realization. A similar argument has recently been made for another classifier language, Mandarin (Tardif, 2006). Mandarin and English differ in how nouns and verbs are lexicalized with respect to which ones are general purpose and which are specific. As Tardif (2006:494) puts it: The main difference is that, in English, adults (as well as children) tend to use more general purpose verbs to approximate one’s meaning and then use prepositions, nouns, and other parts of speech to more fully specify one’s meaning. In Mandarin, verbs are used for very specific meanings (without the addition of distinguishing prepositions). In contrast, Mandarin nouns tend to be general, whereas they tend to be highly specific in English.

Tardif concludes (p. 478) that word-learning theories need to consider not only “cross-linguistic differences in specific features such as syntactic markings and inflections, word order, and differences in the extent to which manner and path are lexicalized with the verb itself,…[but] we need also to consider the nature of the words themselves and how they are organized into a coherent noun and verb lexicon in a particular language”. In short, we need a deeper linguistic understanding of what nouns and verbs are, and of crosslinguistic variability in where the referential load of a language is concentrated. The Tzeltal—and other Mayan—data make the same point (see also Brown, 2001; Bickel, 2002; Danziger, this volume; Rispoli, 1992, 1995; UzielKarl & Berman, 2000). 4.3. Other Influences on Verb Learning in Tzeltal I am not, however, suggesting that semantic specificity is the only factor promoting early verb learning in Tzeltal. As I have discussed elsewhere (Brown, 1998a, 1998b, 1998c), there are several other factors contributing to the prominence of

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

183

verbs in child speech in this language, factors specific to Tzeltal and to the language-learning context in this community that have a bearing on how easy or difficult verbs are to learn, in comparison with nouns. In addition to the semantic facts (verb specificity) that we have been discussing, there are structural facts of Tzeltal, especially the verb initial position, morphological regularity, and morphological distinctiveness of verbs, which surely make verbs easier than in languages with more verbal irregularity. Then there are discourse pragmatic facts, like verb prominence in general due to frequent argument ellipsis, regardless of the semantics of the verb. In Tzeltal (as indeed in many languages), an utterance frequently consists of just a verb alone. Certain interactional facts also contribute to verb prominence in Tzeltal, in particular the prevalence of turn-adjacent conversational repeats—for example, Mother: “Did you feed the chickens?” Child: “(I) fed (them)”—where the response frequently repeats just the verb (Brown, 1998c).12 Finally, there are cultural facts that favor verbs—for example, the absence of any practice of object-labeling for small children, as well as an emphasis on activities as opposed to objects in children’s early socialization (cf. Gaskins, 2005, for a related Mayan society), are both likely to promote verbs in input utterances at the expense of nouns. All of these cumulatively could go against any cognitive bias in favor of early nouns, and contribute to the relative ease of verb learning in this speech community. 5. THEORETICAL IMPLICATIONS What are the implications for theories of semantic acquisition of a language like Tzeltal, a language with a lot of concrete verbs in Tzeltal child language, and with more concrete/more nouny semantics for verbs than in other languages (like English)? Minimally, I suggest that we should add this to the possible learning strategies that have to be incorporated into theories of how children learn verbs. The evidence adduced here that patterns of ellipsis reflect the nature of verb semantics in Tzeltal supports the view that the language being learned can influence how the child proceeds in the word-learning task. We therefore need explanations for word-learning patterns that show how whatever is universal is quickly tuned to expectations concerning a specific kind of language—a phenomenon that has been called (Slobin, 2001) “typological bootstrapping”. In short, typological properties make certain things easier to learn (Slobin, 2001); there is evidence that children’s expectations can be quickly tuned in response to what has already been learned (Smith, 2001). This view is consistent with Gentner’s position on children’s word-learning processes: you do not need to postulate innate constraints on word-learning strategies, just a rapid system-seeking learner and a large role for the input language, especially for verbs (Gentner, 1982: Gentner & Boroditsky, 2001). The argument that less overt O realization tells you that O is recoverable from the context, therefore that (other things being equal) the verb semantics is likely to be concrete—the idea that a transitive verb alone without nouns around it could be thereby easier to learn than one that has nouns around it—might seem to be

184

BROWN

contrary to the spirit of syntactic bootstrapping. The bootstrapping argument is that noun semantics helps you get the verb semantics precisely because the syntactic frame indicates what elements of the scene are important for that verb—what participants are relevant to the scene. Yet, as Adele Goldberg (personal communication) points out, the child doesn’t need the overt nouns to bootstrap the argument structure; she just needs to know what participants are relevant to the scene. This information can be provided either by overt nouns representing referents, or (in less detail) by pronouns or demonstratives or cross-referencing on the verb, or indeed by contextual knowledge (e.g., the O argument may be expressed sequentially, across turns). And in Tzeltal, once the child has the crossreferencing system mastered, ergative and absolutive markers also indicate how 13 many arguments, and of what kind, a verb takes. The properties of Tzeltal child and caregiver speech do not really seem to conflict with syntactic bootstrapping. The learner needs the nouns to be overtly expressed more when there is less elaborate semantics in the verb (i.e., with “general” verbs), and that is precisely where you get them in Tzeltal, so the patterns of argument ellipsis visible in the input have something to help both specific and general verbs get learned. Specific verbs are thus not disadvantaged in relation to general ones; this may explain the early presence of many different semantically specific verbs in Tzeltal children’s speech, along with the much more frequent (in the input) general verbs like ‘give/put’ and ‘do/make’. The typological nature of Tzeltal, in contrast with a language like English, leads to a hypothesis to be explored in further research: the appropriate strategy for tackling a language like Tzeltal may be different from that for a language like English. In Tzeltal, an appropriate strategy is this: pay attention to the lexical semantics of verbs, as that will tell you what the arguments are. Hence, Tzeltal children launch early into verb learning. In English, however, the strategy is (arguably) this: pay attention to the argument structure, as that will tell you what the verbs mean. Hence, syntactic bootstrapping is a plausible starting point. Finally, the Tzeltal data also have implications for theories of how argument realization patterns are acquired. Children’s argument realization has been studied to date in terms of their developing sensitivity to pragmatic constraints (e.g., preferred argument structure, PAS). The Tzeltal data suggest that, at least for some languages, we also need to consider children’s sensitivity to semantic constraints: namely, the preemption of explicit argument realization by specificity in the verb. ACKNOWLEDGMENTS I am grateful to Jürgen Bohnemeyer, Melissa Bowerman, Adele Goldberg, Stephen Levinson, and David Wilkins for comments on a first version of this chapter presented at the MPI argument structure workshop in 1998. Later versions were presented at seminars at the Universidad Nacional Autónoma de México (UNAM), Mexico City, in March 1999, at the MPI for Evolutionary Anthropology in Leipzig

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

185

in April 1999, and at the 8th IASCL meeting in San Sebastian, Spain, in July 1999. Thanks are also due to the participants in these meetings who provided critical feedback and helped me to hone the argument. NOTES 1

In this study, samples coded for new vocabulary totaled about 30 hours (2,100 utterances) for one child and 50 hours (12,600 utterances) for the other (the great majority of these utterances were agreement tokens and other nonword communicative acts). A cumulative vocabulary list was compiled for each child listing every new word the child produced, either by parental report (for the first few words) or in the taped production data samples, until the point at which the child had produced 500 multimorpheme utterance types (age 1;3 to 2;3 for one child, 1;5 to 2;5 for the other). These data were drawn from a larger database collected over four and a half years in a rural hamlet of the highland community of Tenejapa, in Chiapas, Mexico. The data were videotaped every 6 weeks by me, and/or audiotaped monthly by the child’s parents, in five extended families. 2 In contrast, Bowerman (2005: 210) found that an American child overextended the relatively specific verb kick to actions that were kick-like in some respects but not done with a foot. Of course, it is possible that my production data samples and observations of Tzeltal child speech have missed the crucial moments when a child produces overextensions; it is also possible that overextensions of manner verbs like ‘kick’ are more likely than of theme-specific verbs like ‘eatsoft-things’. 3 Because, in Tzeltal, NPs are often ellipsed and arguments are cross-referenced on the verb, the comparison between Ninio’s data and my data is not exact. The Tzeltal combinations in question include all morpheme combinations (e.g., participant cross-referencing morphemes, possessive markers, aspect, etc.), not just full-word combinations. 4 Not only do the relevant constructions differ in Tzeltal and English due to massive argument ellipsis allowed by Tzeltal but not by English, but also some light verbs occur in many different constructions. Note that, for example, English get occurs in quite different constructions (get + O (get the ball), get + location (get into bed), get + attribute (get tired/hungry/excited). Similarly, Tzeltal ak’ ‘give/put’ occurs in different constructions: transitive (‘give/put’ + O), ditransitive (‘give/put’ + REC + DO, ‘give O to me’), ‘give/put’ + V (‘let/make V happen’), and so on. Which construction is supposed to achieve priority in helping the child assign meaning to the verb? 5 The sample consisted of all the input by four adults (two parents, two aunts) to two children during four sessions; approximately 800 verbal utterances were coded. Note that in this community children are raised in extended households, and child care (and hence input speech to small children) is contributed by a number of adults, as well as by elder siblings. 6 Two clarifications are in order. First, although Tzeltal has massive argument ellipsis, core arguments are cross-referenced on the verb. I talk about these Tzeltal NPs as ‘arguments’; I think they probably share properties of argumenthood

186

BROWN

with the cross-referencing markers. This is a grammatical issue not yet resolved for Tzeltal, but nothing rests on this for my discussion here. Second, by ellipsis I mean omission of arguments that are recoverable in the context, and where the argument could have been expressed. This excludes arguments omitted for purely syntactic reasons (as in, e.g., I want to go). 7 That is, the same set of person markers cross-references the subject (S) of intransitive verbs and the object (O) of transitives; a different set cross-references the subjects (A) of transitive verbs. 8 Interlinear glosses use the following abbreviations: ART, definite article; ICP, incompletive; CMP, completive; ERG, ergative; ABS, absolutive; CL, clitic. 9 The data for this study come from a large longitudinal corpus of audio and/or video-taped naturally occurring interaction in five Tzeltal families, collected in the rural hamlet of Majosik’, in Tenejapa, Chiapas, Mexico (see note 1). The community consists of mostly nonliterate Mayan corn farmers. Some people are partially bilingual in Spanish and Tzeltal, but Tzeltal is overwhelmingly the language of the home and almost the only language children hear until they go to school. 10 All verbal utterances were coded. For this analysis, the following principles were followed: (a) I excluded from the counts utterances that are exact repetitions of the previous utterance, or formulae (idioms with fixed argument realization), and (b) I counted all codeable utterances with transitive verbs, except for those with ditransitives or complex verb constructions (these involve distinct constructions and probably different argument ellipsis patterns). Actually, there are only a few semantically general ditransitive verbs (‘give/put’, ‘do/make’) in the data examined, but in principle almost any transitive verb can be construed ditransitively in this language (Brown, 2004, 2007). In my samples there are almost no lexically expressed recipient arguments with ditransitives; crossreferencing on the verb is mostly sufficient for this argument. Complex verbs were excluded because many general verbs occur with complements that are ellipsed according to different constraints: for example, ‘want’ + V, ‘give’[=cause] + V, ‘be able to’ + V. 11 They are not only semantically nounier; they are also in some respects morphologically nounier than verbs in other languages. For example, the same ergative/ absolutive morphology is used with both nouns and verbs (see note 13 and chapter 2, this volume). 12 For analogous facts of interactional style with a bearing on verb learning, see Clancy (1996) on referential strategies for Korean, and Kuntay and Slobin (1997) on “variation sets” in Turkish. 13 It is not at all clear from the literature whether cross-referencing markers on the verb could be used in syntactic bootstrapping. To the extent that they are available to the child as markers of participants, presumably they could be, although like Tzeltal pronouns they indicate only the person and number of the argument, not its semantics. But in this respect, Tzeltal verbal cross-referencing of participants is not very helpful. Although it is obligatory and therefore frequent in the input, and it is learned relatively early, not only is it phonologically minimal (mostly nonsyllabic), but the same markers are used both for arguments of verbs

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

187

and possessors of nouns (if ergative) or predication with nouns (if absolutive). This double dose helps children learn the meanings of the affixes (first, second, third person, and plurals), but it won’t help with syntactic bootstrapping, I think, at least not until the child has the complete paradigm, including aspect markers, around the age of 4 (Brown, 1998b). REFERENCES Allen, S. E. M. (2000). A discourse-pragmatic explanation for argument representation in child Inuktitut. Linguistics, 38, 483–521. Allen, S. E. M., & Schröder, H. (2003). Preferred Argument Structure in early Inuktitut spontaneous speech data. In J. W. Du Bois, L. E. Kumpf, & W. J. Ashby (Eds.), Preferred Argument Structure: Grammar as architecture for function (pp. 301–338). Amsterdam: John Benjamins. Bickel, B. (2002). Referential density in discourse and syntactic typology. Language, 79, 708–736. Bowerman, M. (2005). Why can’t you “open” a nut or “break” a cooked noodle? Learning covert object categories in action word meanings. In C. Gershkoff-Stowe & D. Rakison (Eds.), Building object categories in developmental time (pp. 209–243). Mahwah, NJ: Lawrence Erlbaum Associates. Brown, P. (1998a). Early Tzeltal verbs: Argument structure and argument representation. In E. V. Clark (Ed.), Proceedings of the 29th Stanford Child Language Research Forum (pp. 129–140). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Brown, P. (1998b). Children’s first verbs in Tzeltal: Evidence for an early verb category. Linguistics, 36, 713–753. Brown, P. (1998c). Conversational structure and language acquisition: The role of repetition in Tzeltal adult and child speech. Journal of Linguistic Anthropology, 8, 197–221. Brown, P. (2001). Learning to talk about motion UP and DOWN in Tzeltal: Is there a language-specific bias for verb learning? In M. Bowerman & S. C. Levinson (Eds.), Language acquisition and conceptual development (pp. 512–543). Cambridge: Cambridge University Press. Brown, P. (2004). Learning to express three-participant events in Tzeltal. Poster presented at the 32nd Stanford Child Language Research Forum (CLRF), April. Brown, P. (2007). Culture-specific influences on semantic development: Acquiring the Tzeltal “benefactive” construction. In B. Pfeiler (Ed.), Learning indigenous languages: Child language acquisition in Mesoamerica and among the Basques (pp. 119–154). Berlin: Mouton de Gruyter. Choi, S. (1997). Language-specific input and early semantic development: Evidence from children learning Korean. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 5, pp. 41–133). Mahwah, NJ: Lawrence Erlbaum Associates. Choi, S. (1998). Verbs in early lexical and syntactic development in Korean. Linguistics, 36, 755–780. Clancy, P. M. (1993). Preferred argument structure in Korean acquisition. In E. V. Clark (Ed.), Proceedings of the 25th Annual Child Language Research Forum (pp. 307–314). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Clancy, P. M. (1996). Referential strategies and the co-construction of argument structure in Korean acquisition. In B. Fox (Ed.), Studies in Anaphora (pp. 33–68). Amsterdam: John Benjamins. Clancy, P. M. (2003). The lexicon in interaction: Developmental origins of Preferred Argument Structure in Korean. In J. W. Du Bois, L. E. Kumpf, & W. J. Ashby (Eds.), Preferred

188

BROWN

Argument Structure: Grammar as architecture for function (pp. 81–108). Amsterdam: John Benjamins. Clark, E. V. (1993). The lexicon in acquisition. Cambridge: Cambridge University Press. De León, L. (1999a). Verb roots and caregiver speech in early Tzotzil acquisition. In L. Michaelis & B. Fox (Eds.), Cognition, discourse, and function (pp. 99–119). Stanford University: Center for the Study of Language and Information (CSLI) Publications. De León, L. (1999b). Verbs in Tzotzil early syntax. International Journal of Bilingualism, 3, 219–240. De León, L. (2001). Why Tzotzil (Mayan) children prefer verbs: The role of linguistic and cultural factors. In M. Almgren, A. Barreña, M.-J. Ezeizabarrena, I. Idiazabal, & B. MacWhinney (Eds.), Research on language acquisition: Proceedings of the 8th Conference of the International Association for the Study of Child Language (pp. 947–969). Somerville, MA: Cascadilla Press. Du Bois, J. W. (1987). The discourse basis of ergativity. Language, 63, 805–855. Du Bois, J. W. (2000). Discourse and grammar. In M. Tomasello (Ed.), The new psychology of language (Vol. 2, pp. 47–88). Mahwah, NJ: Lawrence Erlbaum Associates. Du Bois, J. W., Kumpf, L. E., & Ashby, W. J. (Eds.). (2003). Preferred Argument Structure: Grammar as architecture for function. Amsterdam: John Benjamins. Gaskins, S. (2005). Cultural perspectives on infant-caregiver interaction. In N. Enfield & S. C. Levinson (Eds.), Roots of human sociality (pp. 279–298). Oxford: Berg. Gentner, D. (1982). Why nouns are learned before verbs: Linguistic relativity versus natural partitioning. In S. A. Kuczaj (Ed.), Language development, Vol. 2: Language, thought, and culture (pp. 301–334). Hillsdale, NJ: Lawrence Erlbaum Associates. Gentner, D., & Boroditsky, L. (2001). Individuation, relativity, and early word learning. In M. Bowerman & S. C. Levinson (Eds.), Language acquisition and conceptual development (pp. 215–256). Cambridge: Cambridge University Press. Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Goldberg, A. E. (2005). Argument realization: The role of constructions, lexical semantics and discourse factors. In J.-O. Östman & M. Fried (Eds.), Construction grammars: Cognitive grounding and theoretical extensions (pp. 17–43). Amsterdam: John Benjamins. Goldberg, A. E., Casenhiser, D. M., & Sethuraman, N. (2004). Learning argument structure generalizations. Cognitive Linguistics, 15, 289–316. Hollebrandse, B., & van Hout, A. (1994). Light verb learning in Dutch. In M. Verrips & F. Wijnen (Eds.), Papers from the Dutch-German Colloquium on Language Acquisition (pp. 65–89). Amsterdam Series in Child Language Development 3. Amsterdam: University of Amsterdam. Hollebrandse, B., & van Hout, A. (1998). Learning light verbs: Aspectual bootstrapping. In N. Dittmar & Z. Penner (Eds.), Issues in the theory of language acquisition: Essays in honor of Jiirgen Weissenborn (pp. 113–134). Berlin: Peter Lang Verlag. Kuntay, A., & Slobin, D. I. (1997). Listening to a Turkish mother: Some puzzles for acquisition. In D. I. Slobin, J. Gerhardt, A. Kyratzis, & J. Guo (Eds.), Social interaction, social context, and language: Essays in honor of Susan Ervin-Tripp (pp. 265–286). Mahwah, NJ: Lawrence Erlbaum Associates. Lucy, J. A. (1992). Grammatical categories and cognition: A case study of the linguistic relativity hypothesis. Cambridge: Cambridge University Press. Narasimhan, B., Budwig, N., & Murty, L. (2005). Argument realization in Hindi caregiverchild discourse. Journal of Pragmatics, 37, 461–495.

8.

VERB SPECIFICITY AND ARGUMENT REALIZATION

189

Ninio, A. (1996). Pathbreaking verbs in syntactic development. Paper presented at the 7th Conference of the International Association for the Study of Child Language, Istanbul, Turkey, July. Ninio, A. (1999). Pathbreaking verbs in syntactic development and the question of prototypical transitivity. Journal of Child Language, 26, 619–653. Pfeiler, B. (2003a). Noun and verb acquisition in Yucatec Maya. In M. D. Voeikova & W. U. Dressler (Eds.), Pre- and protomorphology: Early phases of morphological development in nouns and verbs (pp. 75–82). Munich: Lincom Studies in Theoretical Linguistics. Pfeiler, B. (2003b). Early acquisition of the verbal complex in Yucatec Maya. In D. Bittner, W. U. Dressler, & M. Kilani-Schoch (Eds.), Development of verb inflection in first language acquisition: A cross-linguistic perspective (pp. 379–399). Berlin: Mouton de Gruyter. Rispoli, M. (1992). Discourse and the acquisition of eat. Journal of Child Language, 19, 581–595. Rispoli, M. (1995). Missing arguments and the acquisition of predicate meaning. In M. Tomasello & W. Merriman (Eds.), Beyond names for things (pp. 331–354). Hillsdale, NJ: Lawrence Erlbaum Associates. Slobin, D. I. (2001). Form–function relations: How do children find out what they are? In M. Bowerman & S. C. Levinson (Eds.), Language acquisition and conceptual development (pp. 406–449). Cambridge: Cambridge University Press. Smith, L. B. (2001). How domain-general processes may create domain-specific biases. In M. Bowerman & S. C. Levinson (Eds.), Language acquisition and conceptual development (pp. 101–131). Cambridge: Cambridge University Press. Tardif, T. (1996). Nouns are not always learned before verbs: Evidence from Mandarin speakers’ early vocabularies. Developmental Psychology, 32, 492–504. Tardif, T. (2006). But are they really verbs? Chinese words for actions. In K. Hirsh-Pasek & R. Golinkoff (Eds.), Action meets word: How children learn verbs (pp. 477–498). Oxford: Oxford University Press. Uziel-Karl, S., & Berman, R. (2000). Where’s ellipsis? Whether and why there are missing arguments in Hebrew child language. Linguistics, 38, 457–482.

CHAPTER 9

Interacting Pragmatic Influences on Children’s Argument Realization Shanley E. M. Allen Boston University

1. INTRODUCTION The syntactic bootstrapping theory proposes that children learn the meanings of verbs from their argument structure (Gleitman, 1990). Thus, children are able to infer from the number and arrangement of arguments expressed with a verb whether the verb communicates a motion event, a transfer event, or a change of state, among other meanings. Languages that allow frequent ellipsis of arguments present a substantial challenge to this theory: if arguments are few, it seems obvious at first look that relatively little information about verb meaning can be deduced from them. However, through case studies in frequent-ellipsis languages (Arrernte and Tzeltal, respectively), the chapters by Wilkins and Brown in this volume elucidate situations in which patterns of argument ellipsis actually could provide information about verb meaning because they are different for different types of verbs. The present chapter draws on data from another frequent-ellipsis language, Inuktitut (Eskimo), to investigate whether children are in fact sensitive to the complex factors that, in adult speech, determine whether an argument is overtly realized in or omitted from a particular utterance. If children are able to retrieve information about omitted arguments from the situational and linguistic context of utterances in the caregiver input, then perhaps they can also infer that those arguments are present in some sense in the utterance, and thus gain far more information about verb meaning from omitted arguments than might have been thought possible. It is not necessarily true, then, that the absence of overt arguments means that syntactic bootstrapping is impossible. Speakers produce arguments in several possible forms in their speech, including full lexical noun phrases, pronouns, and omissions. Of particular interest for the present chapter are omitted arguments, because the sentence contains no overt information about the identity of their referents. Many languages allow omitted subjects and 191

192

ALLEN

objects as a grammatical option (e.g., Inuktitut, Tzeltal, Japanese, Korean), whereas others typically do not except in constrained contexts (e.g., English, German, French). However, even in languages that do allow argument omission, arguments are not freely omitted across the board, but only when the identity of the referent can be retrieved from the situational or linguistic context of the utterance. Retrieval failure typically leads to a breakdown in communication between the speaker and the hearer. Literature on argument realization in adult speech has explored many discourse-pragmatic factors that affect the retrievability of referents (e.g., Ariel, 1990; Chafe, 1987; Clancy, 1980; Du Bois, 1987; Fretheim & Gundel, 1996; Givón, 1983). This work shows quite convincingly that adults are very sensitive to discourse pragmatics in the choices they make about how to realize arguments in their own speech. One can assume, then, that they exercise this knowledge when interpreting the utterances of others to identify the referents of omitted arguments. Are children also sensitive to the discourse-pragmatic factors that govern argument ellipsis, so that they could infer the identity of omitted arguments in the speech of others (i.e., in the caregiver input)? As with adults, an obvious way to begin answering this question is to examine whether children are sensitive to discourse-pragmatic factors in choosing how to realize arguments in their own productions. A small but growing body of literature has addressed sensitivity to discourse pragmatics in the speech of young children learning three different frequent-ellipsis languages—Korean, Inuktitut, and Italian. This literature, reviewed in the next section, shows clearly that children are sensitive to a variety of discourse-pragmatic factors in languages of different typologies and with different morphosyntactic patterns conditioning argument realization. However, this sensitivity is probabilistic rather than absolute. Rather than realizing an argument overtly 100% of the time when it is newly introduced into the discourse, for example, and 0% of the time when it is given, children’s percentages are more like 50% versus 10%.1 Does this mean that children are not sufficiently sensitive to discourse pragmatics, or that researchers are not adequately identifying their sensitivity? It is obviously not a trivial task to discern which of these alternatives is the correct one, and in fact both are likely to be true to some extent. However, before concluding the former, a number of lines of investigation can be followed up on the latter. In this chapter, two of these lines of investigation are addressed with data from Inuktitut. The first picks up on the frequent acknowledgment in the literature that discourse-pragmatic factors do not operate in isolation from each other, and explores the hypothesis that the factors have a cumulative effect, such that children will be much more likely to realize an argument overtly if it is associated with two or three factors rather than with just one. The second investigation looks closely at arguments in children’s speech that violate the predictions assumed to this point about argument realization, to determine whether there are other less obvious discourse-pragmatic factors to which children are also sensitive. All three lines of investigation reported here—discourse-pragmatic factors taken one at a time, multiple interacting factors, and proposed factors for which

9.

INTERACTING PRAGMATIC INFLUENCES

193

only preliminary results are available—show that young children are indeed extremely sensitive in their own speech to factors that influence argument ellipsis. This makes it plausible that, even though arguments are often omitted in frequent-ellipsis languages like Inuktitut, children may nonetheless be able to infer them, and so, after all, to use a verb’s argument structure as a clue to its meaning. 2. THE INFLUENCE OF DISCOURSE-PRAGMATIC FACTORS TAKEN ONE AT A TIME Research on discourse-pragmatic factors in child speech dates from the seminal work of Greenfield and Smith (1976). This work focused primarily on child speech at the one-word stage, asking which word the child chose to realize from a presumably multiword target utterance. The child’s choice tended to be the most informative element in the target utterance (as far as it could be determined)—that component of the target utterance that was most uncertain or least presupposable. The omitted elements, in contrast, could be relatively more easily determined from the surrounding linguistic and situational context. This study analyzed all parts of speech, including verbs and prepositions, but it is clearly relevant to the narrower domain of argument realization and to slightly later stages of development. Since then, many child language researchers have looked at discourse-pragmatic factors specifically in relation to argument realization, focusing on whether an argument is newly introduced to the discourse or not (i.e., new vs. given).2 Several studies have found that children tend to overtly realize those arguments that are new more often than those that are given (e.g., Bloom, 1990; Campbell, Brooks & Tomasello, 2000; Guerriero, Cooper, Oshima-Takane & Kuriyama, 2001; Hyams & Wexler, 1993; Valian, 1991; but see Hamann & Plunkett, 1998, for a different view). This appears to be a relatively robust phenomenon in child language. At least three studies have gone on to investigate the effect on argument realization of discourse-pragmatic factors other than newness. The pioneering work in this area was done by Clancy (1993, 1997, 2003), who looked at 26 hours of data from two children learning Korean, aged 1;8 –2;10, audiotaped in naturalistic communication situations. Korean is a primarily nominative-accusative language with basic SOV word order. Arguments may be overtly realized as lexical noun phrases, demonstratives (third person), or pronouns (first and second person). Subject and object ellipsis is freely allowed in contexts in which the identity of the argument can be determined from the discourse. There are no affixes on the verb that license this ellipsis, or that give any information about the identity of the referent of an elided argument (e.g., person, number, or gender of arguments). Clancy investigated the six discourse-pragmatic factors listed here in (1). She categorized each argument according to its discourse prominence, with each factor having two potential values: discourse-prominent and non-discourse-prominent. Because the other research discussed in this chapter follows Greenfield and Smith (1976) in using the terms “informative” and “uninformative,” these terms rather

194

ALLEN

than Clancy’s are used for the remainder of the chapter (there are subtle differences between these two terminologies, but these are not relevant for the present discussion). The factors and their binary values are as follows: 1. a. NEWNESS: Informative arguments represent referents that are newly introduced into the discourse; uninformative arguments represent referents that have been previously introduced into the discourse.3 b. CONTRAST: Informative arguments introduce or actively maintain explicit contrasts between arguments in the discourse; uninformative arguments are neutral as to contrast. c. QUERY: Informative arguments represent referents that are the focus of or response to a question; uninformative arguments do not represent queried referents. d. ABSENCE: Informative arguments represent referents that are absent from the physical context of the discourse; uninformative arguments represent referents that are present in the physical context. e. PERSON: Informative arguments represent referents that are not speech act participants (i.e., third-person referents); uninformative arguments represent referents that are speech act participants (i.e., first- and second-person referents). f. ANIMACY: Informative arguments represent referents that are inanimate; uninformative arguments represent referents that are animate (i.e., human, animal).

Arguments that were informative with respect to each factor were realized overtly 4 more often than arguments that were not in Clancy’s Korean data. Allen (2000; Allen & Schröder, 2003) investigated the effect of discourse pragmatics on argument realization in 23 hours of data from four Inuktitut-speaking children aged 2;0–3;6, videotaped in naturalistic communication situations. Inuktitut is a polysynthetic and morphologically ergative language of the Eskimo–Aleut family with basic SOV word order. Third-person arguments may take the form of lexical noun phrases or demonstratives, or may be omitted freely given the appropriate discourse-pragmatic context; first- and second-person arguments are always omitted except in rare situations of extreme emphasis.5 All verb stems carry cross-referencing affixes that give information about the person and 6 number of both the subject and the object of the verb. Thus, some information about the identity of the argument is present in the utterance even when the argument itself is missing, but other important information (e.g., animacy, gender, agency) is not present. Allen looked at eight factors of informativeness in her analysis—the same six as Clancy’s, plus the two new factors presented in (2). 2. a. DIFFERENTIATION IN DISCOURSE (DIFFDISC): Informative arguments represent referents that could be confused with other referents already established in the five preceding utterances of the discourse (i.e., another established referent could fit the verb semantics, person, number, etc. of the argument in

9.

INTERACTING PRAGMATIC INFLUENCES

195

question); uninformative arguments represent referents that could not be confused with other preceding referents in this manner. b. DIFFERENTIATION IN CONTEXT (DIFFCONT): Informative arguments represent referents that could be confused with other referents in the immediate physical context (i.e., another referent in the portion of the room where the child is directing his or her eye gaze could fit the verb semantics, person, number, etc. of the argument in question); uninformative arguments represent referents that could not be confused with other preceding referents in this manner.

Allen found that arguments that were informative for each factor were more likely to be realized overtly than arguments that were not. To take into account possible interdependence between the factors, she then investigated the eight factors together in a logistic regression model. A test of the full model with all eight predictors against a constant-only model (i.e., a model with no predictors) showed that the predictors, as a set, distinguished reliably between overt and omitted arguments (χ2 (df = 8, n = 3,168) = 1067.697, p < .001). Prediction success was good, with 95% of omitted arguments and 47% of overt arguments correctly predicted, for an overall success rate of 87% (compared to 100% of omitted arguments, 0% of overt arguments, and 85% of total arguments correctly predicted by a constant-only model). This increase is particularly impressive for overt arguments: a constant-only model predicts that none of them would be overt, but Allen’s model correctly predicts the overtness of 47% of them. Prediction success for omitted arguments is almost at ceiling with the constant-only model, so there is little room for improvement there. Further analysis revealed that five of the eight factors contributed significantly to the effectiveness of Allen’s model: NEWNESS, CONTRAST, ABSENCE, DIFFCONT, and PERSON; QUERY, DIFFDISC, and ANIMACY did not. 7 Finally, Serratrice (2002) investigated discourse pragmatics and argument realization in about 50 hours of naturalistic data from six children aged 1;7–3;4 learning Italian as a native language. Italian is a nominative-accusative language with basic SVO word order, although the subject often appears postverbally. Arguments may be overtly realized as lexical noun phrases, as demonstratives (third person only), or as pronouns (all persons). Subject ellipsis is very common, licensed by obligatory verbal affixes that provide unambiguous information about the person and number of the referent. However, object ellipsis is not normally permitted. Serratrice used five of Allen’s and Clancy’s discourse-pragmatic factors (NEWNESS, CONTRAST, QUERY, ABSENCE, PERSON), and merged two of Allen’s factors, DIFFDISC and DIFFCONT, into a new factor termed DISAMBIGUATION. Finally, following the work of Chafe (1994, 1996), she added an additional factor, ACTIVATION: 3. ACTIVATION: Informative arguments represent referents that are not salient for the hearer or are only in the hearer’s peripheral consciousness; uninformative arguments represent referents that are salient for the hearer at the moment of the utterance, and are a current focus of interest for both speaker and hearer.8

196

ALLEN

To determine how well her data conformed to Du Bois’s (1987) Preferred Argument Structure (PAS), Serratrice also kept track of whether the argument was the subject of an intransitive verb or of a transitive verb. Although this is not a discourse-pragmatic factor per se, it is mentioned here because it was included in her statistical analyses along with the other factors already mentioned.9 Serratrice assessed the effect of these eight variables using logistic regression. A test of the full model with all eight predictors against a constant-only model (a model with no predictors) showed that the predictors, as a set, reliably distin2 guished between overt and omitted arguments (χ (df = 9, n = 3,068) = 2323.417, p < .001). Prediction success was good, with 91% of omitted arguments and 78% of overt arguments correctly predicted, for an overall success rate of 89% (compared to 100% of omitted arguments, 0% of overt arguments, and 68% of total arguments correctly predicted by a constant-only model).10 Six of Serratrice’s eight factors contributed significantly to this result: ACTIVATION, CONTRAST, ABSENCE, 11 DISAMBIGUATION, PERSON, and TRANSITIVITY; NEWNESS and QUERY did not. In sum, there is a strong predictive relationship between the informativeness value of an argument and the form in which that argument is realized; this is found in all three studies of the relationship between argument realization and various discourse-pragmatic factors in child language. However, this relationship is probabilistic rather than absolute. Although the strictest theoretical prediction would be that 100% of informative arguments and 0% of uninformative arguments would be realized overtly, this pattern does not hold. As an example, the results for the feature NEWNESS across the three studies just mentioned are reproduced in Table 9.1. Across the three studies, new arguments are realized overtly from 55% to 72% of the time, and given arguments from 9% to 29%. The lack of a close match between the prediction and the results has at least two possible explanations. First, children may not yet be fully sensitive to discourse-pragmatic factors.12 Alternatively, research has not revealed the true extent of this sensitivity. In the remainder TABLE 9.1 Proportion of New and Given Arguments Realized Overtly by Children in Three Languages Proportion of New Arguments Realized Overtly

Proportion of given Arguments Realized Overtly

Korean (Clancy, 1997)

.72

.29

Inuktitut (Allen, 2000)

.55

.09

.66

.20

Language a

b

Italian (Serratrice, 2002) a

Clancy (1997) gives results for her two subjects separately; the figures here are averaged across the two subjects. She also separated three degrees of newness in her analysis—new, accessible, and given, following Chafe (1987). Only results for new and given are reproduced here. b Serratrice (2002) gives results for four stages of development; the figures here are averaged across the stages. In addition, these figures are for subjects only, and do not include objects.

9.

INTERACTING PRAGMATIC INFLUENCES

197

of this chapter, we explore the second possibility in two ways. First we examine interactions between discourse-pragmatic factors in the Inuktitut data reported in Allen (2000) to see whether informativeness is incremental—do factors reinforce each other in the sense that each successive factor associated with an argument increases its degree of informativeness, and so its likelihood of being overtly realized? Then we ask whether there are any systematic patterns to arguments that violate the hypothesis investigated. Several groups of nonconforming arguments are looked at in detail, and potential explanatory factors are proposed for future research. 3. THE INTERACTION OF DISCOURSE-PRAGMATIC FACTORS Although several studies mention that a complex interaction of discourse-pragmatic factors is likely to be at play in argument realization, only one study to date has reported on the actual interactions between the various factors and on their potential additive influence in child language. Allen (2000) divided the arguments in her data into two groups: those with two or more informativeness factors associated with them, and those with one or fewer informativeness factors. In a first comparison, she found that the odds of an argument’s being overt were about four times as B large for arguments in the higher than in the lower informativeness group (e = 13 3.9519, p < .001). In a second comparison restricted to third-person arguments (recall that first- and second-person arguments are almost always omitted in adult speech), the odds of an argument’s being overt were about twice as large for arguments in the higher than in the lower informativeness group (eB = 1.7658, p < .001 B for 0 vs. 1+; e = 2.1724, p < .001 for 4.0), such that they use significantly more overt forms when the argument is informative for the factor in question. In addition, significant development in sensitivity is seen between stages I (MLUW 1.5–2.0) and II (MLUW 2.0–3.0) for both person and activation. 8 Serratrice did not provide detailed information about how ACTIVATION was coded in her data. This information would obviously be important for a full understanding of the impact of this feature on the success of Serratrice’s model. 9 Clancy (1993, 1997, 2003) and Allen and Schröder (2003) also assessed argument ellipsis in light of verb transitivity. Because they did not include this in their analysis of the discourse-pragmatic factors, it is not mentioned further here. 10 A detailed discussion of the differences in predictability between Serratrice’s and Allen’s models is beyond the scope of this chapter. Before any comparison is made, however, the factor TRANSITIVITY should be removed from Serratrice’s model because it is not a discourse- pragmatic factor, and coding criteria should be presented for the factor ACTIVATION. 11 Note that the results for individual variables are somewhat different from those obtained by Allen (2000), especially with respect to the factor NEWNESS. Serratrice conducted a further logistic regression analysis excluding both PERSON and TRANSITIVITY, and found all six remaining factors, including NEWNESS, to be significant. She hypothesizes that the substantial relationship between NEWNESS and TRANSITIVITY is a confounding factor. It is also likely that there is substantial overlap between the factors ACTIVATION and NEWNESS, although Serratrice does not provide a detailed account of how either of these factors were coded for in her data. The relationships between these variables clearly require further analysis. 12 None of the studies noted in Table 9.1 has reported on adults’ sensitivity to NEWNESS. An earlier paper by Clancy (1993), however, reports on the correlation between discourse prominence (i.e., QUERY, CONTRAST, NEWNESS, and ABSENCE taken together) and argument form in data of both caregivers and children, showing that the patterns are very similar for the two mothers and two children she studied. This suggests that children may indeed be sensitive to NEWNESS alone in a similar way to adults. 13 Note that only the factors ABSENCE, CONTRAST, DIFFCONT, DIFFDISC, NEWNESS, and QUERY were included in this analysis. The factors ANIMACY and PERSON were not considered relevant because they are not discourse-pragmatic in the same sense as the other factors. 14 The following grammatical abbreviations are used in glosses for Inuktitut: Nominal Case: ABS = absolutive; ALL = allative. Verbal Modality: CTG = contingent; DUB = dubitative; IMP = imperative; PAR = participial (functionally equivalent to indicative in the dialect of Inuktitut discussed here). Verbal Cross-Referencing Affixes (e.g., PAR.3sS): 1 = first person; 2 = second person; 3 = third person (disjoint); s = singular; S = subject; O = object. Nominal Inflection (e.g., ABS.SG): SG = singular. Possessive Nominal Inflection (e.g., ABS.2Ssg): 2S = second-person singular possessor; 3S = third-person singular possessor; sg = singular possessum.

9.

INTERACTING PRAGMATIC INFLUENCES

209

Word-Internal Affixes: DIM = diminutive; EMPH = emphatic; FUT = future; HAB = habitual; INCP = inceptive aspect; NEG = negative; PEJ = pejorative; PERF = perfective aspect; POL = politeness (preceding imperative inflection). 15

Items in parentheses are not overt in the utterance, but are implied. Eye gaze is obviously a rough measure of joint attention. More detailed analysis requires a more precise measurement along the lines of that undertaken by Skarabela and Allen (2002a) and reported later in this chapter. 17 Change of topic was coded with admittedly general and rough criteria. Our goal here was only to determine whether looking at topic change might be a fruitful direction for further research, so we focused on a rather quick treatment of the data. Detailed analysis would at minimum require careful delineation of precisely what is meant by the terms “topic” and “change.” 18 Allen (2000) used visual physical context as the domain for coding ABSENCE. However, a broader interpretation of this factor could include the auditory context as well. 19 Further analysis shows that this pattern holds more broadly. For example, deictic gestures are used significantly more often with new arguments than with given 2 arguments for the 1406 third person arguments in the data (c (df = 1, n = 1,279) = 31.507, p < .001; note that 127 of the total 1,406 arguments were excluded from analysis because they were not codable for gesture). 20 The number of arguments treated here does not exactly match the number in the corresponding cells in Table 9.3 because not all the relevant situations on the videotape were clear enough to code the relevant arguments for SPLIT ATTENTION. 21 Since this chapter went to press, Barbora Skarabela has completed a doctoral dissertation on the effect of joint attention on argument realization in child Inuktitut (Skarabela, 2006) which goes far beyond what is reported here. 16

REFERENCES Allen, S. E. M. (2000). A discourse-pragmatic explanation for argument representation in child Inuktitut. Linguistics, 38, 483–521. Allen, S. E. M., & Schröder, H. (2003). Preferred Argument Structure in early Inuktitut spontaneous speech data. In J. W. Du Bois, L. E. Kumpf, & W. J. Ashby (Eds.), Preferred Argument Structure: Grammar as architecture for function (pp. 301–338). Amsterdam: John Benjamins. Ariel, M. (1990). Accessing noun-phrase antecedents. London: Routledge. Baldwin, D. (1991). Infants’contribution to the achievement of joint reference. Child Development, 62, 875–890. Baldwin, D. (1993). Infants’ ability to consult the speaker for clues to word reference. Journal of Child Language, 20, 395–418. Bloom, P. (1990). Subjectless sentences in child language. Linguistic Inquiry, 21, 491–504. Campbell, A. L., Brooks, P., & Tomasello, M. (2000). Factors affecting young children’s use of pronouns as referring expressions. Journal of Speech, Language, and Hearing Research, 43, 1337–1349. Chafe, W. (1987). Cognitive constraints on information flow. In R. Tomlin (Ed.), Coherence and grounding in discourse (pp. 21–51). Amsterdam: John Benjamins. Chafe, W. (1994). Discourse, consciousness, and time: The flow and displacement of conscious experience in speaking and writing. Chicago: University of Chicago Press. Chafe, W. (1996). Inferring identifiability and accessibility. In T. Fretheim & J. Gundel (Eds.), Reference and referent accessibility (pp. 37–46). Amsterdam: John Benjamins.

210

ALLEN

Clancy, P. M. (1980). Referential choice in English and Japanese narrative discourse. In W. L. Chafe (Ed.), The pear stories: Cognitive, cultural, and linguistic aspects of narrative production (pp. 127–202). Norwood, NJ: Ablex. Clancy, P. M. (1993). Preferred argument structure in Korean acquisition. In E. V. Clark (Ed.), Proceedings of the 25th Annual Child Language Research Forum (pp. 307–314). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Clancy, P. M. (1997). Discourse motivations for referential choice in Korean acquisition. In H. Sohn & J. Haig (Eds.), Japanese/Korean linguistics 6 (pp. 639–659). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Clancy, P. M. (2003). The lexicon in interaction: Developmental origins of Preferred Argument Structure in Korean. In J. W. Du Bois, L. E. Kumpf, & W. J. Ashby (Eds.), Preferred Argument Structure: Grammar as architecture for function (pp. 81–108). Amsterdam: John Benjamins. Du Bois, J. W. (1987). The discourse basis of ergativity. Language, 63, 805–855. Fretheim, T., & Gundel, J. (Eds.). (1996). Reference and referent accessibility. Amsterdam: John Benjamins. Givón, T. (Ed.). (1983). Topic continuity in discourse: A quantitative cross-language study. Amsterdam: John Benjamins. Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55. Greenfield, P., & Smith, J. H. (1976). The structure of communication in early language development. New York: Academic Press. Guerriero, S., Cooper, A., Oshima-Takane, Y., & Kuriyama, Y. (2001). A discourse-pragmatic explanation for argument realization and omission in English and Japanese children’s speech. In A. H. J. Do, L. Dominguez, & A. Johansen (Eds.), Proceedings of the 25th Annual Boston University Conference on Language Development (pp. 319–330). Somerville, MA: Cascadilla Press. Hamann, C., & Plunkett, K. (1998). Subjectless sentences in child Danish. Cognition, 69, 35–72. Hyams, N., & Wexler, K. (1993). On the grammatical basis of null subjects in child language. Linguistic Inquiry, 24, 421–459. Serratrice, L. (2002). Syntax and pragmatics in the acquisition of Italian subjects. Paper presented at the 9th Conference of the International Association for the Study of Child Language, Madison, WI, July. Serratrice, L. (2005). The role of discourse pragmatics in the acquisition of subjects in Italian. Applied Psycholinguistics, 26, 437–462. Skarabela, B. (2006). The role of social cognition in early syntax: The case of joint attention in argument realization in child Inuktitut. PhD dissertation, Boston University, Boston, MA. Skarabela, B., & Allen, S. E. M. (2002a). The role of joint attention in argument realization in child Inuktitut. In B. Skarabela, S. Fish, & A. H. J. Do (Eds.), Proceedings of the 26h Annual Boston University Conference on Language Development (pp. 620–630). Somerville, MA: Cascadilla Press. Skarabela, B., & Allen, S. E. M. (2002b). Correlations between newness, joint attention, and argument realization in child Inuktitut. Paper presented at the 9th Conference of the International Association for the Study of Child Language, Madison, WI, July. Tomasello, M. (2001). Perceiving intentions and learning words in the second year of life. In M. Bowerman & S. C. Levinson (Eds.), Language acquisition and conceptual development (pp. 132–158). Cambridge: Cambridge University Press. Tomasello, M., & Farrar, J. (1986). Joint attention and early language. Child Development, 57, 1454–1463. Tomasello, M., & Kruger, A. (1992). Joint attention on actions: Acquiring verbs in ostensive and non-ostensive contexts. Journal of Child Language, 19, 311–334. Valian, V. (1991). Syntactic subjects in the early speech of American and Italian children. Cognition, 40, 21–81.

PART III

Transitivity, Intransitivity, and Their Associated Meanings: A Complex Work-Space for Learnability

CHAPTER 10

Intransitive Verbs in Ewe and the Unaccusativity Hypothesis James Essegbey University of Florida

1. INTRODUCTION The Unaccusativity Hypothesis (UH) of Perlmutter (1978) claims that there are two classes of intransitive verbs, unergatives and unaccusatives. In this chapter, I argue that “one-place verbs” that occur in the intransitive clause in Ewe, a Kwa language spoken in Ghana, Togo, and Benin, cannot be divided into the two distinct classes. I claim that the verbs can, instead, be placed in a single class with a commonality that is the lack of an element of meaning that I refer to as “Cause.” I propose, therefore, that the intransitive clause in Ewe constitutes an argument structure construction with its own semantics à la Goldberg (1995). The chapter is organized as follows: in section 2, I review the UH, noting areas of agreement and disagreement in the literature. In section 3, I introduce the issue of obligatory complement verbs (OCVs). It is the nature of these verbs that calls for an investigation into intransitive verbs. In section 4, I discuss the types of verbs that occur in the intransitive clause in Ewe. I show that they can be divided into three distinct classes based on their formal argument realization properties. Section 5 discusses why the three classes cannot be further divided into two distinct unergative–unaccusative classes, and section 6 proposes an argument structure construction analysis for the intransitive clause in Ewe. In section 7, I conclude the chapter by taking a brief look at the implications for learnability. 2. THE UNACCUSATIVITY HYPOTHESIS (UH) The UH was born out of the realization that different constructions within and across languages seem to suggest a division of intransitive verbs into two classes, with the arguments of one class often displaying properties that are similar to the objects of transitive clauses. A construction that illustrates the division is the resultative 213

214

ESSEGBEY

construction in English: the resultative phrase can be predicated of the direct object but not the subject. Thus, in Mary beat John black and blue, ‘black and blue’ can be predicated only of John. It turns out that a resultative phrase can also be predicated of the subject of a subset of intransitive verbs. Thus, John dropped dead is acceptable but Mary cried silly is not. We see from these examples that the object of beat and the subject of intransitive drop share some properties. Another kind of construction that divides intransitives into two classes is the auxiliary selection identified in languages like Dutch and Italian. In Dutch, one class of intransitive verbs selects the auxiliary zijn ‘be’ (e.g., hij is gevallen ‘he has fallen’) whereas another class selects hebben ‘have’ (e.g., hij heeft gesproken ‘he has spoken’). A third kind of construction is the morphological case marking identified in Active–Inactive languages (I use the term “construction” to include case marking in the sense of Goldberg, 1995). These are languages in which the argument of some intransitive verbs receives the same casemarking as the subject of transitive verbs while that of others receives that of objects. The examples that follow are from Caddo, a Caddoan language spoken in West Oklahoma (cf. Mithun, 1991: 525): ?

?

?

1a. ci:widahšahyah ‘I jumped.’

2a. hákkut náw uhsa ‘I’m sick.’

1b. cihahyún
ah ‘I’m going home.’

2b. ku:táy ayah ‘I’m tired, disgusted, fed up.’

3a. ci:wida:kuhnah ‘I grabbed him.’

4a. ku:wida:kuhnah ‘he grabbed me.’

?

3b. ci:kí:
ah ‘I’m going to kill him.’ 4b. ku:kí
ah ‘he’s going to kill me.’ ?

?

The morpheme ci- marks the first-person singular subject of the intransitive verbs glossed as ‘jump’ and ‘go’, as well as the subject of transitive verbs glossed as ‘grab’ and ‘kill’. In contrast, ku- is used to represent the same personal pronoun when it occurs as the subject of the intransitive verbs glossed as ‘sick’ and ‘tired, disgusted, fed up’, as well as the object of the transitive verbs translated as ‘grab’ and ‘kill’. In order to account for the parallels in constructions like these, which serve to distinguish between two classes of intransitives and which are referred to as unergative/unaccusative diagnostics, Perlmutter (1978:160) promulgates the Unaccusitivity Hypothesis (UH). The hypothesis, which is within the relational grammar (RG) framework, states: Certain intransitives have an initial 2 but no initial 1.

In non-RG terms, this means that the single argument of some intransitive verbs occurs in the same underlying position as the object of a monotransitive clause. It is these verbs that are classified as unaccusative verbs. Contrasted with the unaccusative verb is the unergative verb, which has an initial 1, that is, its argument has the same underlying position as the subject of a transitive clause. Although the unaccusative verb has an underlying object, this argument surfaces as a subject due to the “Final 1 Law,” which, in RG, essentially requires that predicates have subjects.

10.

INTRANSITIVE VERBS IN EWE

215

Proponents of the government and binding (GB) framework adopt the same division of intransitive verbs into unaccusative and unergative verbs and assign similar underlying positions to the arguments. Following Burzio (1986), they assume that the unaccusative verb is not able to assign structural (i.e., accusative) case to its underlying object, and that this explains why the argument ends up in subject position in surface structure. An alternative characterization provided by Levin and Rappaport Hovav (1995) is that the unaccusative verb is not able to assign a theta role to its subject. Although RG and GB accounts of unaccusativity have a lot in common, they differ in that the latter allows for the unergative verb to take an object in some cases (cf. Levin & Rappaport Hovav, 1995: 40), whereas the RG definition of unergativity excludes this possibility. The UH seeks to account for the mapping between syntax and semantics. A general version of the mapping that Perlmutter and Postal (1984: 97) refer to as the universal alignment hypothesis states: There exist principles of universal grammar that predict the initial relation borne by each nominal in a given clause from the meaning of the clause.

The hypothesis, which is a strong representation of the relation, suggests that the unaccusative/unergative distinction is determined by the semantics of the verbs in intransitive clauses. Perlmutter (1978) and Perlmutter and Postal (1984) propose that unaccusatives comprise predicates expressed by adjectives in English, predicates whose initial term is a semantic patient, predicates of existing and happening, and predicates of nonvoluntary emission of stimuli that impinge on the senses. Unergative verbs, by contrast, comprise predicates describing willed or volitional acts (including manner of speaking verbs and sounds made by animals) as well as those describing certain involuntary bodily processes like ‘cough’, ‘sneeze’, and ‘weep’. Opinions are divided on the issue of whether unaccusativity is semantically determined. For instance, Rosen (1984) argues that this is not the case. Her reason is that despite the fair amount of correlation between a verb’s semantics and its syntactic behavior, there are also several cases that are expected to fall into one class but end up in the other. An example is ‘die’, which is expected to be unaccusative across languages but is unergative in Choctow. Moreover, not only do equivalent verbs in different languages fall into different classes, but the same verb in some languages can display properties of both classes. Rosen concludes that the ability of a verb to occur in a class is not predictable and, therefore, that the phenomenon is purely syntactic. Rosen’s position is contested by Van Valin (1987, 1990). He argues that the distinction is best accounted for in semantic terms, but that crosslinguistically the phenomenon is not determined by a single semantic property. In some languages, such as Italian, the split is determined by the inherent aspect of the lexical items: activities are unergative whereas achievements and states are unaccusative. In

216

ESSEGBEY

other languages, such as Tsova-Tush and Acehnese, the distinction is determined by agentiveness or volitionality. The determining factor for agentiveness varies depending on the language. Thus DeLancey (1985: 3), writing on the agentive criteria for split intransitives, observes: The semantic distinction coded in such systems is usually described in terms of the presence or absence of “volitionality” or “control” on the part of the actor, based either on the prototypical volitionality value of the predicate or the actual degree of volitionality imputed by the speaker to the actor in the event reported.

Thus where agentiveness is the determining factor, it could in turn be determined by volition or control. Van Valin also argues that in some languages the criteria for determining unaccusativity may be split between agentiveness and inherent lexical aspect. He claims that a semantic analysis based on both parameters obviates the need to assign syntactic underlying forms to unergative and unaccusative verbs. Levin and Rappaport Hovav (1995) also argue that unaccusativity is semantically determined. However, unlike Van Valin, they do not think that this removes the need for a syntactic account. For them, the phenomenon is semantically determined and syntactically encoded. That is to say that the semantics of the verb determines an underlying syntactic position for its single argument. On the semantic issue, they note that the inconsistencies that Rosen and others discuss arise because the semantic distinctions that were used are not necessarily the ones that are syntactically relevant. For example, they argue against Perlmutter’s class of “bodily process” predicates in favor of different classifications that they consider to be more fine-grained. As an example, they point out that while the bodily-process verbs snore and blush in English can also be classified as activity verbs (when the classification is based on inherent aspect), arossire, the Italian equivalent of ‘blush’, literally means ‘become red’ and therefore is a change-of-state verb (cf. Levin & Rappaport Hovav, 1995: 9). This, they suggest, is evidence that seemingly equivalent predicates may, across languages, differ in the crucial element of meaning that is syntactically relevant. Levin and Rappaport Hovav also discuss other semantic notions that are different from Van Valin’s. For the purposes of this chapter, I concentrate on three, namely, internal causation, external causation, and directed change. Internal causation means that inherent properties of the referent of an argument are responsible for bringing about the state of affairs coded in a verb. According to Levin and Rappaport Hovav, arguments of such verbs are realized as underlying subjects. Where external causation is concerned, it is the referent of an external argument that brings about the state of affairs. The argument that represents the entity thus acted on is realized as an underlying object. Verbs of directed change also have their arguments realized in underlying object position. An important aspect of Levin and Rappaport Hovav’s proposal is that directed change supercedes internal causation in determining the underlying position of an argument. Thus, if an internally caused

10.

INTRANSITIVE VERBS IN EWE

217

verb occurs in a construction where it further expresses directed change (e.g., ‘run down a tunnel’), its argument will be realized in underlying object position. This issue is discussed in more detail in section 5.2, where I show that it accounts neatly for the otherwise puzzling behavior of some intransitive verbs in Ewe. In the next section, I discuss obligatory complement verbs. 3. OBLIGATORY COMPLEMENT VERBS (OCVS) Ewe, like several other Kwa languages, has a large number of verbs that take obligatory complements (OCVs). Some of these verbs have been described as inherent complement verbs (ICVs), which Nwachukwu (1987: 22) defines as verbs “whose citation form is followed by a meaning-specifying complement.” The definition is intended to capture verbs whose semantics are difficult to determine without first taking into consideration the contribution of the semantics of their obligatory complements. Intuitively, one could say that the Ewe verb ƒú in (5) is an ICV because speakers are not able to provide an off-the-cuff meaning for it without first knowing the complement it occurs with. 5. Kofi Kofi

1

ƒú

tsi

ICV

water

‘Kofi swam.’

The verb can also occur with the complement du ‘course/race’ to mean ‘to run’. In addition to canonical ICVs like ƒú, which supposedly take obligatory complements because they do not have any “concrete” meaning, there are also verbs with relatively specific meanings that cannot occur without a complement. These verbs can be separated into two classes based on the type of obligatory complement with which they occur. In one class are verbs that take the generic complement nú ‘thing’. Examples are êu nú ‘eat’and Nl nú ‘write’. I refer to the complement as generic because it does not contribute to the semantics of the verb and complement sequence. The other class consists of verbs that take morphological or semantic cognates as obligatory complements (à la Austin, 1982). An example of the former is fi fi ‘steal’, and an example of the latter is êú Äe ‘dance’. It can be observed that all the ICVs encode states of affairs that are encoded by simple verbs on their own in languages like English and French. Moreover, most ICVs occur as intransitives in these languages. This has led researchers to ask whether such noun-complement sequences are intransitive, and, for Nwachukwu (1985, 1987) and Avolonto (1995), to propose that ICVs in Igbo and Fon, respectively, are indeed intransitive. There is clear evidence from the syntax of Ewe that ICVs and other OCVs are not only transitive, but that they cannot be distinguished from canonical transitive verbs without recourse to translation. I discuss two properties that point to the syntactic independence of the complements of the ICV subset (note that this applies to

218

ESSEGBEY

all OCVs). To begin with, just like canonical transitive complements, inherent complements (ICs) can be focused and thereby placed in sentence initial position. This is illustrated in (6): 6. tsi water

má-é

Kofí

ƒú

that-FOC

Kofi

ICV

‘Kofi swam that water.’

Here, the complement not only occurs separated from the verb, but it takes a demonstrative determiner. The next property involves the ability of the verb to occur in the nyá construction. This construction is structurally similar to the passive construction in English in that the unmarked complement occurs in subject position, a modal is introduced, and the logical subject is adjoined to the clause via the preposition ná ‘to/for’. The semantics of the construction, however, differs from that of the passive construction. Sentence (7a) shows how this applies to a canonical transitive verb: 7a. xeví-á bird-DEF

nyá

wu

ná

Kofí

MOD

kill

for

Kofi

‘Kofi was able to kill that bird.’ 7b. tsi water

má

nyá

ƒú

ná

Kofí

that

MOD

ICV

for

Kofi

‘Kofi was able to swim that water.’

In both cases, the underlying complement occurs in subject position. Note that the existence of these constructions rules out the possible suggestion that the inherent complement is incorporated into the verb. This discussion shows that there is no class of complement-taking verbs that can be opposed to one-place intransitive verbs in Ewe, even if their equivalents in other languages happen to be intransitive. In Ewe, all OCVs, of which ICVs constitute a subset, are just as transitive as canonical transitive verbs. In the next section I discuss intransitive verbs, and show that there is no criterion that enables them to be divided into two distinct classes. 4. INTRANSITIVE VERBS Intransitive verbs in Ewe can be divided into three classes based on their formal argument realization properties. In this section, I discuss the three classes and explore the semantics of some of the verbs that occur in them. My semantic classification of the verbs is based on native-speaker intuitions and does not have any formal basis. The terminology I adopt, on the other hand, is taken from the literature on

10.

INTRANSITIVE VERBS IN EWE

219

unaccusativity (cf. Danziger, 1996; DeLancey, 1985; Levin & Rappaport Hovav, 1995; Mithun, 1991; and Van Valin, 1987, inter alia). 4.1. The “Specified-Complement” Class The verbs in this class can occur with or without complements. As the name of the class suggests, the complements they can occur with are specified and hence limited (although some take more complements than others; cf. Essegbey, 1999). When they do take a complement, there is a shift in interpretation, and this places a restriction on the kind of argument that they can be predicated of. The verbs that occur in this class are inherently directed motion verbs (e.g., yi ‘go’, do ‘move out’), involuntary bodily process verbs (e.g., gb ‘breathe’, gb ya ‘blow air’, nye ‘entity exudes from the body’, nye N ‘fart’), movement verbs (e.g., z ‘an entity that is mobile undergoes change of location’, z azli ‘human walk [on two feet]’, dzo ‘body of an entity moves upward, fly [of birds]’, dzo kpó ‘jump’), and crying verbs (e.g., fa ‘mourn, sounds made by trees’, fa avi ‘cry’). Let us consider do ‘move out’ in detail. Do expresses a state of affairs in which any kind of entity comes out of an enclosure. This is why I gloss it simply as ‘move out’. It is also why I classify it as an inherently directed motion verb. Consider the following examples, which show the verb occurring with an animate and an inanimate argument: 8a. Kofí Kofi

le

do-do-m´

le

xɔ-a

me

be.at

RED-move.out-PROG

at

room-DEF

in

‘Kofi is leaving the room.’ 8b. u blood

le

do-do-m´

le

Kofí

ŋú

be.at

RED-move.out-PROG

at

Kofi

skin

‘Blood is dripping from Kofi.’

Sentence (8b) shows that do can be classified in some contexts as a substance emission verb. Note that this is a function not only of the verb, but also of the argument it takes, as well as the construction in which it occurs. The important thing, however, is that there is a ‘move out’ meaning component in the two sentences. Do can occur with the complement go ‘out’,2 which, as the gloss suggests, is semantically connected to the meaning of the verb (the result of ‘exiting’ a place is that the one doing so is ‘out’). When it takes this complement, the verb can only be predicated of animate entities; thus: 9. Kofí/*euu Kofi/*blood

le

go

do-m´

be.at

out

move.out-PROG

‘Kofi/*blood is going out.’

220

ESSEGBEY

The reason that do go can be predicated only of animate arguments is that the twoplace construction refers to a state of affairs that is under the control of the participant. In section 6, I argue that this is the property that determines whether the verbs in this group occur with a complement in a particular sentence. 4.2. The “Causative-Inchoative Alternation” Class The name for this class, taken from Levin and Rappaport Hovav (1995), refers to verbs whose subject argument in the intransitive clause occurs as object in the transitive clause. The argument encoded as the subject of the transitive clause is usually the causer of the state of the argument encoded as the object. Consider the following examples: 10a. atí-á stick-DEF

ŋé

10b. Kofí

break

‘The stick broke/is broken.’

Kofi

ŋé

atí-á

break

stick-DEF

‘Kofi broke the stick.’

In (10a), the single argument atíá ‘the stick’ occurs in subject position. In (10b), in contrast, where a second argument is introduced, it occurs in object position. The verbs in this group are inchoatives (e.g., gl˜ ‘become crooked’, vúvú ‘become torn’), verbs of substance emission (e.g., êuêu ‘trickle’, tsy ‘drip’) and verbs of (manner of) movement (e.g., mli ‘roll’, tró ‘spin’). 4.3. The “Only-Intransitive” Class Only intransitive, as the name suggests, means that the verbs in this class cannot take any complement. We might therefore want to say that they are the true intransitive verbs in the language. The verbs that occur in the class are inchoatives (e.g., tsi ‘become old, grown’, bí ‘become cooked’), verbs of sound and smell emission (e.g., êi ‘resound’, múmú ‘smell fishy’), involuntary bodily process (e.g., háhá ‘yawn’), cries (e.g., xl˜ ‘animal cry, involuntary human cry’, wó ‘bark’), and (manner of) movement (e.g., t ‘limp’, tá ‘crawl’). 4.4. Summary I have identified three classes of intransitive verbs in Ewe based on their argument realization properties. Verbs of the first class can occur with or without a complement; that is, they can occur transitively as well as intransitively. Those in the second class can also occur both transitively and intransitively, albeit with the difference that the intransitive subject corresponds to the transitive object. The final class consists of verbs that can only occur intransitively. Table 10.1 summarizes the semantic classes of verbs that occur in each group.

10.

INTRANSITIVE VERBS IN EWE

221

TABLE 10.1 Semantic Classes of Intransitive Verbs in Ewe Causative-Inchoative Alternation

Only Intransitive

1. Inchoative

X

X

2. Emission

X

X

Semantic Classes

Specified Complement

3. Bodily process

X

4. Directed motion

X

5. Crying

X

6. Movement

X

X

X X

X

In the next section, I consider whether these three classes can be regrouped into unergative/unaccusative classes. 5. DOES EWE HAVE UNACCUSATIVES/UNERGATIVES? As I discussed under the review of the UH, there are three main positions on unaccusativity. One states that the phenomenon is purely syntactic, another that it is purely semantic, and yet another that it is semantically determined and syntactically encoded. In the present section, I consider whether the syntactic criterion proposed by Rosen or the semantic criteria proposed by Van Valin or Levin and Rappaport Hovav enable us to regroup the three classes of intransitive verbs discussed in section 4 into two classes. 5.1. The Syntactic Criterion According to proponents of the syntactic criterion, some intransitive verbs have an underlying subject whereas others have an underlying object. Proponents of the syntactic criterion base their position on different types of constructions in the various languages (the unergative–unaccusative diagnostics), which, according to them, determine the placing of arguments. Some of these constructions—for example, the resultative construction in English and auxiliary selection in Dutch—were briefly discussed in section 2. Ewe does not have any construction that one could use as an unergative–unaccusative diagnostic. We therefore have to base our discussion on the realization of the arguments of the intransitive verbs in the language. If we adopted the syntactic position, we could classify the intransitive verbs in the specified-complement class as unergatives, because when they take an additional argument their single argument in the intransitive clause remains in subject position. One could therefore say that these arguments occur in underlying subject position. Consider the following examples with the verb z ‘move’:

222

ESSEGBEY

11a. Kofí Kofi

11b. Kofí

zɔ move

‘Kofi walked.’

Kofi

zɔ

azɔli

move

walk.on.two.feet

‘Kofi walked.’

Kofí occurs in subject position in both (11a), where z does not take a complement, and (11b), where it does. The syntactic criterion would further classify the verbs in the causative-inchoative alternation group as unaccusatives. This is because the single argument of the intransitive form of this verb occurs in object position when the verb takes an extra argument. Sentences (10a) and (10b) are repeated here: 10a. atí-á stick-DEF

ŋé break

‘The stick broke/is broken.’

10b. Kofí Kofi

ŋé

atí-á

break

stick-DEF

‘Kofi broke the stick.’

It could be said that the single argument of verbs like Né ‘break’ occurs in underlying object position and that it only moves to subject position when there is no other argument for this position. Although it is possible to place the verbs in the specified-complement class and causative-inchoative alternation class in the unergative and unaccusative classes, respectively, on the basis of their argument realization properties, the same cannot be done for those in the only-intransitive class. We could place them together with the specified-complement class because, being always intransitive, their arguments always occur in the subject position. Note, however, that one could equally claim that only-intransitive verbs are unaccusatives with underlying objects that surface as subjects. In section 6, I give reasons why such a claim is not unreasonable for Ewe. However, the reason I provide is semantic, and applies to all the intransitive verbs. I therefore conclude that the syntactic criterion fails to regroup our three classes noncontroversially into two distinct classes. I now consider whether the semantic criteria do a better job of subdividing the classes. 5.2. Semantic Criteria As I stated in section 2, Van Valin proposes that lexical aspect or agentiveness, or a combination of the two, accounts for the division between intransitive verbs. A distinction based on lexical aspect will predict that activity predicates possess a property that is distinct from achievement and stative predicates. However, this property does not distinguish any of the three classes from the others in Ewe. In fact, the standard tests for activity verbs apply to verbs from all three classes. One such test—the ability to occur with the adverbial phrase ‘for an hour’—shows that verbs of emission, which belong to the causative-inchoative alternation and only-intransitive classes, are activity verbs, as are verbs of crying and movement, which belong to the specified-complement and only-intransitive classes. Here are some examples:

10.

INTRANSITIVE VERBS IN EWE

223

Specified complement: 12a. u-a vehicle-DEF

zɔ

uu

gaƒoƒo

eká

háfí

t ɔ´

move

continuously

hour

one

before

stop

‘The vehicle moved continuously for an hour before stopping.’

Causative-inchoative alternation: 12b. kpé-a stone-DEF

mli

uu

gaƒoƒo

eká

háfí

t ɔ´

roll

continuously

hour

one

before

stop

‘The stone rolled for an hour before stopping.’

Only intransitive: 12c. tsi-a water-DEF

dza

uu

gaƒoƒo

eká

háfí

t ɔ´

fall

continuously

hour

one

before

stop

‘It rained continuously for an hour before stopping.’

Agentiveness is a better criterion, because no matter which definition is adopted (i.e., volitionality or control), it classifies the specified-complement and the causative-inchoative alternation verbs as unaccusatives. However, depending on whether we adopt a volitionality or control definition of agentiveness, we get different results within the onlyintransitive class: volitionality classifies inchoatives like bí ‘become cooked’and biá ‘become red’ as unaccusatives, and crying (in a broad sense) and movement verbs like kúk ‘stammer’ and t ‘limp’ as unergatives. Control, on the other hand, classifies all the verbs as unaccusatives. The control criterion would therefore group all intransitive verbs in Ewe into a single class. Because there would be no defined unergative class with which to contrast such a class, we have to assume this criterion fails as well. As I stated in section 2, Levin and Rappaport Hovav (1995) propose internal/external causation and directed change among the semantic properties that determine unaccusativity. The internal versus external causation criteria predict that inchoatives like vó ‘become rotten’ and tsi ‘become old’, which belong to the only-intransitive class, have underlying subjects. This is because it is properties inherent to the referents of the verbs’ arguments that bring about the state of affairs expressed by the verbs. By contrast, inchoatives like vúvú ‘become torn’ and báká ‘become mixed’ will have underlying objects, since they are externally caused. As I noted, Levin and Rappaport Hovav’s directed change supercedes internal causation; hence, according to their analysis, an argument that should be in underlying subject position because of a verb’s semantics could be realized in object position if the clause in which it occurs expresses directed change. This proposal accounts for the otherwise problematic behavior of some only-intransitive verbs, like the light emission verb kle~´ ‘shine’in (13b, 13c, and 13d):

224

ESSEGBEY

13a. Kofí Kofi

si

akaí-á

kplé

mátsési

light

lamp

with

matches

‘Kofi lit the lamp with a match.’ kl ~e´

13b. Akaí-á Lamp-DEF

shine

‘The lamp shone.’ 13c. *Kofí Kofi

kl ~e´

akaí-á

kplé

mátsési

shine

lamp-DEF

with

matches

‘Kofi shone the lamp with a match.’ 13d. Kofí Kofi

kl ~e´

akaí

é

xɔ-á

me

shine

lamp

ALL

room-DEF

in

‘Kofi shone the lamp in the room.’

Si is the Ewe word meaning to cause an object to burn brightly. As (13a) shows, it allows the entity that is “lit” to occur in object position. By contrast, kl~e´ refers to the inherent shining property of entities like lamps and stars. As such, it cannot express states of affairs in which an entity is caused to shine. This verb is therefore basically intransitive. However, the brightness expressed by the verb can be caused to shine in a certain direction, giving rise to a directed-change interpretation. As (13d) shows, this additional sense allows the argument of the verb that cannot occur in object position in a “normal” transitive clause to be realized thus. The internal and external causation criteria distinguish between the causativeinchoative alternation and only-intransitive verbs, classifying the former as unaccusatives and the latter as unergatives. The problem comes with the specified-complement verbs. According to the internal causation criterion, these verbs are unergatives, being internally caused. This seems problematic, however, because verbs of inherently directed motion, which also occur in this group, are usually considered to be the exemplary unaccusative candidates. This problem is resolved by appealing to the directed change criterion that classifies the verbs as unaccusatives. Note, however, that this gives rise to another problem: one criterion classifies a subset of the verbs in this class as unaccusatives and the other classifies another subset as unergatives. It is clear from the preceding discussion that neither syntactic nor semantic approaches enable us to divide the verbs that occur in intransitive clauses into two distinct classes. In the next section, I consider an alternative proposal that groups all the verbs into one class.

10.

INTRANSITIVE VERBS IN EWE

225

6. THE ARGUMENT STRUCTURE CONSTRUCTION ANALYSIS Recall that when specified-complement verbs take complements, they have particular interpretations such that they cannot be predicated of inanimate entities. In section 4.1, I showed that when do ‘move out’ occurs without a complement, it can be predicated of inanimate arguments like Vu ‘blood’. However, when combined with the complement go ‘out’, it can only take animate arguments. The same applies to z ‘move’, exemplified in sentences (11a) and (11b). Without a complement, this verb can be predicated of any entity that is inherently mobile. However, when it takes a complement, it can be predicated only of human entities. Although all the verbs discussed so far can take inanimate arguments when they are intransitive, there are also verbs that require only animate arguments even when they are intransitive. The interpretation that these verbs receive when they are intransitive is, however, different from that when they are transitive. This can be illustrated with nye, an involuntary bodily process verb: 14a. *Kofí Kofi

nye

bé

‘exude’ that

ye-adó

dzikú

ná

ye

velíá

3SG-cause

anger

for

3SG

friend

‘Kofi sneezed in order to annoy his friend.’ 14b. Kofí Kofi

nye

ŋ ɔ´

bé

ye-adó

dzikú

ná

ye

velíá

‘exude’

fart

that

3SG-cause anger

for

3 SG

friend

‘Kofi farted in order to annoy his friend.’

Nye ‘exude’ occurs only with animate entities. In (14a), it occurs without a complement and the purpose clause is unacceptable. In (14b), in contrast, where it takes a complement, it allows for the purpose clause. In order to express an ‘exuding’ similar to that in (14b), but where control is not asserted, an intransitive verb is used, as illustrated in (15): 15.

ŋ´ɔ

gbɔ

le

Kofí

ŋú

fart

arrive

at

Kofi

skin

‘Fart escaped Kofi.’

What these examples show is that specified-complement verbs only assert control when they take a complement. There appear to be problems with the claim that specified-complement verbs assert control only when they take a complement, however. One such problem is raised by the verb dzo, which I translate as ‘entity move upward’. When this verb

226

ESSEGBEY

occurs without a complement, it refers to an involuntary jump as well as to the flight of birds. Do I then intend to claim that the expression of the flight of birds does not involve the assertion of control? Following this proposal by Langdon, I do: “the cries and motions of animals are spontaneous, and probably not subject to conscious control” (Langdon, 1977: 6). This explains why the flight of birds is expressed by a verb without a complement. When dzo takes a complement (e.g., ‘dzo high jump’—jump a high jump), it refers only to purposeful jumps and is, therefore, predicated of humans. Another problem involves the verb gb, which without a complement refers to breathing, and is predicated of animate entities including humans. To account for this, I extend Langdon’s proposal to states of affairs that are considered intrinsic to all animate entities, including humans. These would all be construed as spontaneous and therefore not subject to conscious control. I claim that breathing falls into this category. This would also explain why z ‘move’ occurs in this class. Recall that without a complement, this verb can be predicated of all mobile entities. In contrast, when it takes a complement, it refers only to human movement. This means that without a complement, z ‘move’ simply refers to the intrinsic motion of the mobile entities it is predicated of. Among such entities are humans and inanimate entities like vehicles and boats. At this level, control is not asserted. When one wants to refer to a purposeful walk, which, presumably, involves the assertion of control, azli ‘walk’ occurs as a complement to the verb, thus yielding a transitive construction. To conclude, when the specified-complement verbs express states of affairs in which no control is asserted, they do not take complements. When control is asserted, they do. I now turn to the causative-inchoative alternation verbs. Like the specified-complement verbs, when these verbs occur with a single argument they express states of affairs in which control is not asserted. Unlike them, however, they do not assert control when they take a second argument. Instead, the subject argument is construed as having caused the state of the second verb. Thus, while the transitive version of the specified-complement verbs expresses control, that of the causativeinchoative alternation verbs expresses cause. What is the relation between cause and control? To answer this, I draw on a proposal by DeLancey (1990). DeLancey (1990: 314) proposes the following chain of causation as a cognitive model of action in Lhasa Tibetan: Volition ¡ Act ¡ Event ¡ Result In this chain, elements on the left are supposed to constitute cause while those on the right represent effect. Even though DeLancey’s proposal is designed as a cognitive model of action in Lhasa Tibetan, he notes that the schema represents more or less what a commonsense analysis of our understanding of events might come up with. We know that although the transitive clause prototypically expresses a state of affairs in which one participant intentionally brings about a change in the state or

10.

INTRANSITIVE VERBS IN EWE

227

location of another, there are deviations from this prototype. Thus, the same clause can simply express a state of affairs in which an entity unintentionally brings about a change in another. The following two sentences represent this: 16a. John broke the window. 16b. The stone broke the window.

In (16a), we have a default interpretation that John intentionally performed an action, the result of which is a change in the state of the window. We could consider this as a series of causal relations: John performs an act that acts on the window and causes it to change its state. In contrast, (16b) simply states that the stone brought about a change in the state of the window. It does not include any initial activity on the part of the stone. Wilkins and Van Valin (1996) argue that the default attribution of intention to John in (16a) is due to the combined semantics of the verb, the NPs that function as its arguments, and the construction in which they occur. Crucially, intentionality is not entailed and John’s role might not be different from that of the stone. Supposing then that we said that the transitive clause in Ewe encodes a single causal relation, with additional information left to other contextual or pragmatic factors. DeLancey’s model of action would then be modified for Ewe thus: Control ¡ Act ¡ Event ¡ Result What this means is that, in Ewe, a participant who is in control is taken to be responsible for, and therefore causing, an event in the same way as a participant who brings about a change in the state of other entities. Control is therefore a subcomponent of “Cause.” This explains why clauses in which specified-complement and causative-inchoative alternation verbs occur without a complement have something in common: they do not assert Cause. In order to express this causal element, the verbs have to take a complement. It should be clear that, like DeLancey, I take Cause to be an abstract notion. An entity causes a state of affairs if it brings about a change in the state or location of another, or if it is assumed to be capable of exercising some degree of control. This proposal also accounts for the behavior of the only-intransitive verbs as well as that of so-called ICVs: the former do not assert Cause and so cannot occur with a complement, whereas the latter express autonomous states of affairs, which, according to Ravin (1990), involve protagonist control and so must occur with a complement. The notion of control explains the occurrence of some verbs without complements even though they could be considered as agentive. These are t ‘limp’ and kúkÛ ‘stammer’. To the extent that these verbs express states of affairs that involve defect, they do not assert control. I should stress that it is the semantic representation of the verbs that denotes a lack of control. Thus it does not matter whether a person is only pretending to be stammering or limping; in either case the person is construed as

228

ESSEGBEY

acting out an uncontrolled state of affairs. In this sense, Ewe is like Lakhota and Mohawk, active–inactive languages in which these states of affairs are encoded with a patient case (call it unaccusative case) because the referents of the arguments of which the verbs are predicated are not perceived to be in control of the state of affairs encoded by the verb (cf. Mithun, 1991). There is independent evidence that a certain perspective on physical defects is encoded in the morphosyntax of Ewe. Ameka (1991) discusses the phenomenon in relation to the Ewe possessive construction. In identifying the motivation for the use of the suffix -n to signify possession of a property (e.g., d-n ‘sick one/ patient’, tsukú-n‘mad one’) as opposed to tÛ (e.g., in kale~-tÛ ‘courageous one’ and aso-tÛ ‘fool’), Ameka (1991: 39) explains that n “occurs on nominals that denote adverse states, for example, a disease, an infirmity or a handicap, to indicate that someone is the undergoer of the unpleasant condition.” In his semantic characterization of the affix, therefore, Ameka (1991: 38) includes this sentence: “Because of that this person (X) cannot do some things like other people.” This is contrasted with his characterization of tÛ, which includes the specification that “This person (X) could do some things because of that.” What this suggests, in effect, is that the choice of affix is dependent on whether the entity that possesses the property is rendered incapable of controlling a state of affairs as a result of it. It is in this light that one can understand the syntactic distinction between verbs that refer to speaking and walking in general versus stammering and limping. A logical conclusion of the preceding proposal is that the intransitive clause in Ewe is incompatible with the expression of the semantic element that I have referred to here as Cause (i.e., bringing about a change or engaging in a state of affairs in which control is asserted). To express this notion, a clause must be transitive. To capture the distinction, I refer to the intransitive clause as a one-place construction (following Goldberg, 1995) and the transitive clause as a causal two-place construction. The term “causal” is included to distinguish this from another transitive construction in which cause is not asserted, that is, the noncausal two-place construction (see Essegbey, 1999). The noncausal two-place construction, unlike the causal construction, does not occur in the nyá construction. Recall that this is the construction in which the undergoer is realized as subject. There is the temptation to couch the claim I have made here in terms of the UH. I now discuss the implications of such a step. For instance, within Kwa linguistic circles it has been informally stated that all Kwa verbs are transitive. This has only been formally argued for Igbo, however (see references in Manfredi, 1991). To say that all verbs in Ewe are transitive will first presuppose that ICVs are transitive. Next, the one-place verbs will be analyzed as unaccusatives, because they could be said to lack agentive arguments (i.e., if agentiveness is based on control). This would mean that they have underlying objects. Thus the claim that all Kwa verbs are transitive will be theory-internal. It should be noted that, in strictly theoretical terms, to claim that all verbs are transitive would be to deny the relevance

10.

INTRANSITIVE VERBS IN EWE

229

of the UH to Ewe, because such a claim does not contrast the unaccusatives with any unergative class, there being no way to distinguish between canonical transitive clauses and “unergative transitive” clauses. But what would be the justification for such a claim? From a distributional perspective, it would be acceptable for the causative-inchoative alternation verbs, because the argument that occurs in the subject position of their one-place construction actually ends up in the object position in their two-place construction. And, as I speculated in section 5.1, we might even push things further and claim by fiat that the only-intransitive verbs also have their single arguments generated in object position even if they always end up in subject position. But the claim becomes tenuous for the specified-complement verbs, where the same participant occurs in subject position in both constructions. 7. IMPLICATIONS FOR LANGUAGE ACQUISITION It has been shown that verbs that occur in intransitive clauses in Ewe cannot be divided into the two distinct classes of unaccusatives and unergatives. Instead, they possess a common property, which is their inability to express a causal meaning. This situation poses problems for both the semantic and syntactic bootstrapping theories of acquisition, to the extent that they require universal principles to determine the mapping from meaning to form. Semantic bootstrapping cannot explain how the child learning Ewe, unlike the child learning English, for instance, is able to determine that the argument structure of, for example, ‘crawl’ is different from that of ‘run’ (an ICV). And syntactic bootstrapping cannot explain how Ewe learners, using universal semantic mapping principles, determine the meaning of ICVs from their argument structure, because these verbs, as we have seen, have a different argument structure from their equivalents in other languages. Having said that, it should be pointed out that a syntactic bootstrapping account could explain how—once children have already learned something about argument structure in Ewe—they can distinguish between such minimal pairs as z ‘move’ and z azli ‘walk, by human’. Because all transitive constructions involve the semantic element that I have referred to as Cause, the child should infer that the addition of a complement to the verb means the addition of a causal component of meaning. NOTES 1

In all the Ewe examples, only the high tone is marked. I adopt the following abbreviations: 1 = first person, 2 = second person, 3 = third person, ALL = allative, DEF = definite, FOC = focus, IC = inherent complement, ICV = inherent complement verb, MOD = modal, OCV = obligatory complement verb, PL = plural, PROG = progressive, RED = reduplicative segment, SG = singular. 2 Although glossed as ‘out’, go is a nominal which functions as an argument of the verb (cf. Essegbey, 1999).

230

ESSEGBEY

REFERENCES Ameka, F. (1991). Ewe: Its grammatical constructions and illocutionary devices. PhD dissertation, Australian National University. Austin, P. (1982). Transitivity and cognate objects in Australian languages. In P. Hopper & S. A. Thompson (Eds.), Syntax and semantics, Vol. 15: Studies in transitivity (pp. 37–47). New York: Academic Press. Avolonto, A. (1995). Pour une approche minimaliste des verbes à objets inhérents en Fngbe. PhD dissertation, University of Québec, Montréal. Burzio, L. (1986). Italian syntax: A government-binding approach. Dordrecht, The Netherlands: D. Reidel. Danziger, E. (1996). Split-intransitivity and active-inactive patterning in Mopan Maya. International Journal of American Linguistics, 62, 379–414. DeLancey, S. (1985). Agentivity and syntax. In W. H. Eilfort, P. D. Kroeber, & K. L. Peterson (Eds.), Papers from the Parasession on Causatives and Agentivity, CLS 21 (pp. 1–12). Chicago: Chicago Linguistic Society. DeLancey, S. (1990). Ergativity and the cognitive model of event structure in Lhasa Tibetan. Cognitive Linguistics, 1, 289–321. Essegbey, J. (1999). Inherent complement verbs revisited: Toward an account of argument structure in Ewe. PhD dissertation, Leiden University. Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Langdon, M. (1977). Semantics and syntax of expressive “say” constructions in Yuman. In K. Whistler (Ed.), Proceedings of the 3rd Annual Meeting of the Berkeley Linguistics Society (pp. 1–11). Berkeley, CA: Berkeley Linguistics Society, University of California. Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax–lexical semantics interface. Cambridge, MA: MIT Press. Manfredi, V. (1991). Agbo and Ehugbo: Igbo linguistic consciousness, its origins and limits. PhD dissertation, Harvard University. Mithun, M. (1991). Active/Agentive case marking and its motivations. Language, 67, 510–546. Nwachukwu, P. A. (1985). Inherent complement verbs in Igbo. Journal of the Linguistics Association of Nigeria, 3, 61–74. Nwachukwu, P. A. (1987). The argument structure of Igbo verbs. Lexicon Project Working Papers 18. Center for Cognitive Science, Cambridge, MA: MIT Press. Perlmutter, D. M. (1978). Impersonal passives and the unaccusative hypothesis. In J. J. Jaeger, A. C. Woodbury, F. Ackerman, C. Chiarello, O. D. Gensler, J. Kingston, E. E. Sweetser, H. Thompson, & W. Kenneth (Eds.), Proceedings of the 4th Annual Meeting of the Berkeley Linguistics Society (pp. 157–189). Berkeley, CA: Berkeley Linguistics Society, University of California. Perlmutter, D. M., & Postal, P. M. (1984). The I-Advancement Exclusiveness law. In D. M. Perlmutter & C. Rosen (Eds.), Studies in relational grammar 2 (pp. 81–125). Chicago: University of Chicago Press. Ravin, Y. (1990). Lexical semantics without thematic roles. Oxford: Clarendon Press. Rosen, C. (1984). The interface between semantic roles and initial grammatical relations. In D. M. Perlmutter & C. Rosen (Eds.), Studies in relational grammar 2 (pp. 38–77). Chicago: University of Chicago Press. Van Valin, R. D., Jr. (1987). The unaccusative hypothesis vs. lexical semantics: Syntactic vs. semantic approaches to verb classification. North Eastern Linguistic Society, 17, 641–661. Van Valin, R. D., Jr. (1990). Semantic parameters of split intransitivity. Language, 66, 221–260. Van Valin, R. D., Jr., & Wilkins, D. P. (1996). The case for ‘effector’: Case roles, agents, and agency revisited. In M. Shibatani & S. A. Thomas (Eds.), Grammatical constructions: Their form and meaning (pp. 289–322). Oxford: Clarendon Press.

CHAPTER 11

He died old dying to be dead right: Transitivity and Semantic Shifts of ‘Die’ in Ewe in Crosslinguistic Perspective Felix K. Ameka Leiden University

1. INTRODUCTION Mortality is a fact of life, and dying would appear to be a universal experience common to all peoples. Murdoch (1945) lists funeral rites as one of his 73 cultural universals, and D. Brown (1991: 59) reports that the treatment of the dead has been proposed as belonging to the cultural scheme of universal patterns. He suggests that people everywhere have ideas about death (Brown, 1991: 139). Although the experience of dying and death-related practices are subject to cultural variation, one would expect the concept of dying, as a universal experience, to be codified in all languages. Wierzbicka (1996, 1998) has hypothesized that this is indeed the case and that every language has an expression that embodies the concept of DIE. Hence it is a lexical universal, and is one of the “fundamental innate elementary meanings” that are “the same” across all languages (Wierzbicka, 1998: 114). However, there is no perfect equivalence across languages in terms of the use of ‘die’ predicates or in the properties of the referents that they are predicated of. This means that learners have to acquire the conventions and properties of ‘die’ predicates from the people around them and not from universal patterns. The concept of DIE has to be “acquired via ‘the cultural tool of language’ ” (Wierzbicka, 1996: 19) rather than being a universal innate concept. An indication of this comes from universals of language use, where the dying of humans is characteristically talked about in euphemisms. There are two dimensions of this that point to language- or culture-specific patterns that have to be learned. First, the euphemistic expressions and the “bald” expression may be in complementary distribution. For instance, in English, it appears that the dying of people we know tends to be talked about euphemistically whereas 231

232

AMEKA

more distant humans are talked about using the bald ‘die’ verb. Second, and more significantly, there are cultures in which the equivalent of ‘die’ is generally tabooed for talking about humans dying. Central Australian languages such as Arrernte and Warlpiri are a case in point. In Arrernte, the term ilwe-, which all bilingual speakers identify as equivalent to ‘die’, and which is used, for instance, of animals dying, story characters dying, and of general dying, is more often used with respect to real humans to mean ‘become unconscious, faint’ (David Wilkins, personal communication; see Ameka & Wilkins, 1999). Gropen, Epstein, and Schumacher (1997:168) propose that verb learning is context sensitive in the sense that “contextual information about participants is an important part of the product and process of verb learning and that verb learners are…predisposed to associate types of participants with the arguments of a verb.” If so, it suggests that Arrernte children should have some problems in acquiring the form ilwe- because they would be more likely to learn and experience the conditions for ilwe- in the sense of ‘dying’ with respect to nonhumans, animals, and plants first, and in the sense of ‘unconsciousness’ with humans. Death and dying may be a human universal that is sociobiologically driven, but it is constrained by cultural practices. Thus cultural practices and other language-specific factors will affect the learning of ‘die’-related elements. In this chapter I leave aside the problems of culture, and examine the structure and semantics of the equivalent of ‘die’ in Ewe, a Kwa (Niger–Congo) West African language, against the backdrop of universalist expectations about such a verb. I also discuss the potential problems the Ewe verb may pose to the learner and to theories of verb learning. In particular, I demonstrate that the Ewe verb kú ‘die’ is not just a one-participant verb, as is universally expected of verbs that represent a dying situation, but it is also a two-place verb, that is, a transitive verb. This latter feature is not predicted by the universal principles that are usually assumed for the linking of arguments and predicates. It is rather rare for ‘die’ predicates to be transitive across the languages of the world. Hence the Ewe feature calls for an explanation, and it raises the question of what syntax/semantic alignment information can be said to be universal with respect to ‘die’ predicates. Even though the argument structure properties of kú ‘die’ in Ewe are peculiar, the multiple senses it has developed have crosslinguistic parallels. I argue that in addition to the meaning of ‘die’, the verb kú has a second meaning of ‘intensity’ and in some uses an interpretation of ‘not want’—a negative desiderative reading. These meanings of kú are similar to the meanings of the English verb die in sentences like I am dying of thirst, that is, I am very thirsty, and I am dying to see you, that is, I want very much to see you, respectively. In the case of English, die has a positive desiderative reading, whereas in Ewe kú has a negative desiderative meaning. I use ‘desiderative’ to cover both unless a distinction is required. I survey these types of semantic shifts of ‘die’ elements across a selection of unrelated languages in section 4 to show that the semantic structure of the Ewe verb is not that unusual. Nevertheless, the polysemous structure and the semantic shifts associated with the Ewe verb kú mean that its acqui-

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

233

sition must demand considerable attention to context (cf. Clark, 1993; Gropen et al., 1997). That is, the child should not only pay attention to participant types that will generate contextual readings of the ‘die’ meaning, but she should also relate the elements that generate the other meanings and readings, for example, the presence of two surface NPs. Thus, learning this verb does not just require lexical knowledge, it also requires knowledge of constructions which are language specific and which are not derivable from universal principles. In section 2 I present some syntactic and semantic properties of ‘die’ predicates from a crosslinguistic perspective. I demonstrate that the variation in behavior that we find means that there is very little about ‘die’ predicates that can be attributed to innate knowledge. I then focus on the argument structure properties of the verb kú ‘die’in Ewe (section 3). In section 4, I review the semantic shift that ‘die’predicates undergo from ‘die’ to ‘intensity’ and ‘desiderative’ (which can be positive or negative), and the attendant syntactic ramifications. Such shifts provide further evidence that the language learner cannot be guided by a simple form–meaning pairing. The chapter concludes with potential learnability problems. 2. ‘DIE’ VERBS ACROSS LANGUAGES The expectation from the semantics of situations that ‘die’ predicates primarily denote is that they involve one participant. This participant undergoes an experience and attains an end state. A logical paraphrase of English die is ‘BECOME dead (x)’ (cf., e.g., Van Valin, 1990). From a bootstrapping perspective, such a meaning is universal and the syntax of the verb should follow from it. But across the languages of the world there are various mismatches. First, if the single participant of the ‘die’ verb is an undergoer, as suggested earlier, the prediction from some theories of grammar (e.g., the Unaccusativity Hypothesis) is that it should not occur with or be followed by an NP that recapitulates its content. Such an NP is called a cognate object. However, ‘die’verbs in many languages occur in such constructions, including Ewe (shown later) and English, as illustrated in (1) and (2). 1. He died a horrible death. 2. Opera stars don’t die normal deaths, they die spectacular deaths. (P. Wells, Montreal Gazette; cited in MacFarland, 1995:198)

In (1) and (2) the ‘die’ verb occurs with a cognate object NP headed by the noun ‘death’, which is semantically a recapitulation of the semantics of the verb. It is verbs whose single participants are viewed as actors, for example, run, as opposed to those that have undergoers, like ‘die’, that are predicted from these theories to occur in such a syntactic frame (see Levin & Rappaport Hovav, 1995; MacFarland, 1995). In addition, languages differ with respect to how the single participant of the ‘die’ verb is treated grammatically. In Acehnese (Austronesian) it can take either the undergoer form, as in (3a) or the actor form, as in (3b).1

234

AMEKA

3. Acehnese (Durie, 1987) a. gopnyan

ka-matê PERF-die

3SG ‘He died.’ b. rila ready

ji-matê 3SG[+Actor]-die

‘He’s ready to go to his death.’

In other languages, such as English, the single participant of the ‘die’verb is consistently treated as an undergoer, and in yet others it is consistently treated as an actor. This variability in the treatment of the single participant means that one cannot assume any universal pattern for the single participant as part of the innate knowledge from which to predict the syntax of the ‘die’verb à la semantic bootstrapping, or the semantics à la syntactic bootstrapping. Another crosslinguistic feature of ‘die’ verbs is that they may occur with two surface noun phrases, where the referent of one of the NPs is the central semantic participant (i.e., the actor or the undergoer, depending on the language) and the other NP is not a cognate object but a secondary predicate. A secondary predicate, roughly speaking, is an expression that specifies the condition or state of the participant engaging in the state of affairs designated by the main predicate (see Schultze-Berndt & Himmelmann, 2004). In the English sentences in (4), the postverbal noun phrases are secondary predicates. They can be interpreted either as the condition or the manner in which the participant died. 4. He died a hero/a pauper/a martyr.

In Ewe, the ‘die’ verb kú can surface with two NPs. There are two types of such structures. In one type, the structure corresponds to a transitive construction (see later discussion). The other structure corresponds to the use of the verb with a secondary predicate nominal. In Ewe such secondary predicates are optionally marked by a special morpheme -i ‘ATTRIButive’. In English, however, this is not the case. The occurrence of ‘die’ predicates in structures with two surface NPs are problematic, I think, for theories that suggest that the surface syntactic form can guide a child to verb semantics. In the case of the secondary predicate structures, there is a mismatch between the number of NPs on the surface and the number of semantic participants the verb has. It appears that the specificities of ‘die’ verbs in each language need to be examined in order to understand the issues that such a verb can engender for learnability hypotheses. 3. THE EWE VERB KÚ ‘DIE’ In this section, I focus on the Ewe verb kú ‘die’. I point out its characteristics pertaining to the properties of participant types of the verb and the grammatical

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

235

constructions in which the verb appears. My aim is to highlight the peculiar properties of the verb and the way it deviates from some of the universalist expectations about such verbs. I suggest that the peculiarity of the verb kú ‘die’ in Ewe means that its learning might not conform to the patterns expected from bootstrapping theories. Before turning to the characteristics of the verb, I introduce the language and its typological features relevant for the verb lexicon and grammar. Ewe is a member of the Gbe language cluster (Capo, 1991; Duthie, 1996). It is spoken by about 2.5 million people in the southeastern part of the Volta Region of Ghana across to parts of southern Togo as far as and just across the Togo–Benin border in West Africa. Ewe is a tone language and a grammatical word order language with basic SVO syntax. Apart from constituent order, which distinguishes between the core arguments in a basic clause, the subject is distinguished from the nonsubject relations by the form of pronominal clitics that may be used to express such arguments. It is also an isolating language with agglutinative features (Westermann, 1930). As such, most morphosemantic features are expressed by lexical items or markers and by morphosyntactic periphrasis. Some implications of this typological profile for the verb lexicon are, first, that “there are no productive morphological processes for the formation of new verbs” (Ameka, 1994: 57). Second, “[V]erbal specialisation [i.e., specifications for semantic functions, FKA], largely a matter of derivational morphology in many languages, is primarily a syntactic phenomenon” (Clements, 1972: 240). Thus, a verb can occur in several argument constructions where there could be a mismatch between the number of NP referents in the construction and the number of semantic participants coded in the verb. In addition, Ewe is a strict transitive language. That is, all relevant core arguments, even if recoverable, must be overtly expressed in the clause. Some notions, such as ‘run’ and ‘swim’, which seem to be widely expressed in intransitive clauses across languages, are expressed in transitive constructions in Ewe (cf. Essegbey, 1999, this volume). From a syntactic bootstrapping perspective, the fact that arguments are not left implicit should be an advantage. However, because the arguments expressed on the surface often do not correspond to the number of semantic participants, a syntactic bootstrapping approach cannot help the learner to deduce verb meaning. I return to these issues later with specific reference to the ‘die’ verb in Ewe. The Ewe data used in this chapter come from both spoken and written texts. The sources for the written texts are indicated either by the author, date, and page system (e.g., Agbodza, 1962: 15) or by author and text line number in concordance database system (e.g., Obianim, 2312). Some sentences are overheard from conversations or adopted from radio and TV performances and are indicated as such. Where a sentence is not explicitly attributed, then it is constructed by myself, based on my nativespeaker intuitions but cross-checked with other speakers for acceptability. 3.1. The Place of Kú ‘die’ in the Ewe Verb Lexicon The Ewe verb kú ‘die’ has some properties that set it apart from the majority of Ewe verbs. First, it is one of a very small number of verbs where the same root form can

236

AMEKA

function both as a verb and as a noun. Out of about 600 verb roots only 5 or 6, in2 cluding kú ‘die’, behave like this. 5. a. (Ayeke 2899 slightly adapted)
eví-á-wó

mé-nyá

kú

o

child-DEF-PL

NEG-know

death

NEG

‘The children do not know (i.e., understand) death.’ b.
eví-á-wó child-DEF-PL

mé-nyá

kú

o

NEG-MOD

die

NEG

‘The children certainly did not die.’

The sentences in (5a) and (5b) are identical in surface form but they differ in terms of structure and meaning. In (5a) kú ‘death’is a noun and it is the complement of the verb nyá ‘to know’. In (5b) we have the same forms but they have different functions: kú ‘die’ in this case is a verb and nyá ‘certainty modal’ functions as a preverb marker and is a grammaticalized form of the verb nyá ‘know’, which occurs in (5a).3 Such same-form–multiple-function situations are potentially problematic for language learners and users, because they cannot always rely on formal cues for interpretation. Second, the verb kú ‘die’ is unique in the language in the following way: it is the only intransitive verb whose central participant is linked to the subject relation in a one-place construction and is also linked to the subject in the two-place construction when the verb alternates its syntactic frames. For other intransitive verbs that can also occur in transitive constructions, such as change-of-state verbs like gba ‘break’ (Essegbey, 1999, this volume), the participant linked to the subject role in the intransitive is linked to the object role in the transitive construction, as in The pot broke vs. He broke the pot. Although kú ‘die’is semantically a change-of-state verb, it does not pattern like other change-of-state verbs in the language. The language learner would have difficulty abstracting generalizations that would be applicable to it. Thus, it would seem to be hard for a language learner to make abstract generalizations about the paradigmatic relations between the constructions in which the verb kú ‘die’ occurs. This is because its pattern of argument structure alternation is less frequent than other verb alternations within the patterns of relations between one-place and two-place constructions. The verb kú ‘die’ does, however, behave like other inchoative verbs in Ewe in some respects. Thus, it has a variable contextual interpretation as a stative verb, that is, ‘come to V’ or ‘be V’, in its unmarked aspect (aorist) form, as in (6).4 6.
eví-á child-DEF

kú die

‘The child died.’/‘The child is dead.’

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

237

Ewe inchoative verbs have the interpretation of ‘getting to V’ or ‘about to V’ in the progressive, as exemplified in (7). 7.
eví-á child-DEF

le

kù-kú- m ´

be.at:PRES

RED-die-PROG

‘The child is dying.’/‘The child is about to die.’/‘The child is sick.’

5

In contrast to inchoative verbs like kú ‘die’, ‘active’ verbs like sí ‘flee, escape’ have a past interpretation in unmarked aspect and an “ongoing situation” interpretation in the progressive, as shown in (8). 8. a.
eví-á child-DEF

sí escape

‘The child escaped.’ b.
eví-á child-DEF

le

sì-sí- m ´

be-at:PRES

RED-escape-PROG

‘The child is running away.’

To sum up, the verb kú ‘die’in Ewe does not conform to expectations from a universalist perspective. From the point of view of its place in the Ewe verb lexicon, it is also peculiar: it is one of the very few verbs that has an identical nominal root counterpart and it is unique among intransitive verbs with respect to its linking behavior in transitive constructions; that is, the participant that is linked to the subject position in the intransitive construction also functions as the subject in the alternate transitive construction. In the next section, I discuss the properties of the verb in relation to the grammatical constructions in which it participates and the semantic interpretations that are generated. 3.2. The Verb Kú ‘die’ in Grammatical Constructions The verb kú ‘die’ occurs in a variety of constructions, most of which can be viewed as involving combinations of specific constructions with the major argument structure constructions in Ewe. Argument structure constructions are form–meaning pairings used to encode basic clauses (Goldberg, 1995, 2002). Three of the Ewe argument structure constructions identified by Essegbey (1999; this volume: 228) are relevant for our purposes. These are the “one-place construction,” the “causal two-place construction,” and the “noncausal two-place construction.” A one-place construction is the realization of an intransitive clause; it has only one core syntactic argument, as in sentences like Máng gé ‘A mango dropped’ or Áma kú ‘Ama died’. Essegbey (1999; this volume) argues that the semantics of this construction is “lack of cause.” That is, the single syntactic argument is not viewed

238

AMEKA

as being in control of bringing about the state of affairs characterized in the construction. It can be seen as undergoing a change of state or being in a state. In the causal two-place construction there are two core syntactic arguments, one with more Actor-like properties and the other with more Patient-like properties. It is the realization of the prototypical transitive clause. For example, if Kofi tackled and brought Yawo down to the ground in a football match, one could say Kofí gé Yawo ‘Kofi felled (lit.: dropped) Yawo’. In this construction, the Actor-like argument is linked to the subject role and is viewed as instigating the state of affairs characterized in the rest of the clause. In the noncausal two-place construction, the two arguments are in a kind of figure–ground relation. Essegbey (1999) calls the construction a “Theme-Locative construction,” where the theme is understood as the argument whose location or state is at issue and the location is broadly conceived to include states, properties, and experiences (cf. DeLancey, 1995, 1997). Neither argument is construed as controlling or initiating the realization of the state of affairs characterized in the construction. An example of such a construction is shown in (9). 9. Awu garment

lá

ƒo


i

DEF

hit

dirt

‘The garment is dirty.’

‘The garment’, which is linked to the subject role in example (9), is in a state and that state is characterized by the postverbal noun ‘dirt’, which is linked to the object grammatical relation. As the earlier illustrations with the verb gé ‘drop’ show, a verb can enter any of the constructions as long as its semantics and its construal with the other members of the construction are compatible. The compatibility of arguments with constructions can come from the semantics of the verb or be contributed by the construction. Thus, a single verb can occur in several of the argument structure constructions. The verb kú ‘die’ is one of the verbs that can surface in all three of the argument structure constructions outlined already. It can also surface in the cognate object and external possessor constructions. It can also occur in serial verb constructions, but since there are no other changes in semantics, these are not further discussed.6 In the rest of this section we take a look at instantiations of these argument structure constructions, paying attention to the properties of the referents of the central argument and the semantic interpretations that are derived. These properties are features that the language learner must attend to. 3.2.1. Kú ‘die’ as a One-Participant Verb In this section, I argue that when the verb kú ‘die, be dead’ occurs in structures where there is one central participant, it has a meaning that can be roughly para-

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

239

phrased as follows, along the lines of the Natural Semantic Metalanguage mode of explications (see, e.g., Wierzbicka, 1988, 1998): 10. NP (=X) kú before t: entity X can be thought of as a K (= genus, e.g., person) because there are things that one can say about X at time t: something happened to X after this: one cannot say that X is a K

The basic idea is that there is a cluster of properties that defines a class of entities (cf. Wierzbicka, 1988). Some of these may be salient such that when they are lost, an entity can no longer be characterized as that thing. For various classes of entities, the properties of the referents coerce particular interpretations of the verb. Thus, for living things such as people, animals, and things that grow out of the ground, that is, plants, the salient property that is understood to be lost is ‘life’, as illustrated in (11). 11. (Obianim 1581) Vi-nye

kú

child-1SG

die

‘My child is dead.’ (i.e., after something happened to my child, my child no longer lives)

Another class of referents is made up of inanimate entities that have a periodic or delimited time of existence, such as a sore or the moon. When the verb is predicated of these, it is used to indicate that they do not exist any more. Thus, one can say Dzinú lá kú (lit.: ‘The moon died’, i.e., the month has ended). The crucial point about this class of referents is that there is a time when they cannot be seen anymore, and being seen is the property that is lost. A further class of referents that can fill the central participant slot of the verb kú ‘die’ is made up of things that are made by people for the purpose of using them to do certain things. For this class of referents, there is some property that ensures that they can be effective for their prototypical function. When they lose this property and therefore cannot be used in the performance of their function, then they can be talked about as ‘dead’. For instance, a cutting implement such as a machete, an axe, or a knife has to be sharp for it to be a good instrument in the performance of cutting actions. When the property of being sharp is lost, it can be said to be dead, as in Yí-á kú ‘The cutlass is dead’, that is, the cutlass has become blunt. Similarly, other inanimate things that lose their salient property and therefore cannot function effectively in the prototypical way are said to have died. A piece of land that has lost its potential to make things grow out of it can be said to have died: Anyígbá lá kú [land the die] ‘The land is dead’. That is, the land has become infertile.

240

AMEKA

Another class of referents whose dysfunctional nature can be talked about using the verb kú ‘die’is person or body parts. Thus a person’s ear or head may be literally said to be dead as in (12a,b), which mean that the person is deaf or is a blockhead, respectively. 12. a. É-ƒé 3SG-poss

tó

kú

ear

die

‘His/her ear is dead.’ (i.e., S/he is deaf.) b. É-ƒé 3SG-poss

ta

kú

head

die

‘His/her head is dead.’ (i.e., S/he is a blockhead.)

When the referents involved are person parts, the verb kú ‘die’ may also be used in external possessor constructions to signal the experiential or affectedness status of the possessor of the person part in the situation. There are two kinds of external possessor constructions. One is a dative experiencer external possessor construction in which the person part functions as the central participant and the possessor is coded as the object of the dative preposition, as in (13). 13. (Overheard utterance) Tó

kú

nε

ear

die

to:3SG

‘Ear is dead to him/her.’ (i.e., S/he is deaf.)

The second is an allative (goal) external possessor construction where the possessor is presented as the experiencer and is realized as the dependent of a postposition ú ‘side’. The postpositional phrase is then marked as the target by its function as the complement of the allative preposition, as in (14). 14. Abɔ/afɔ

kú


é

ŋú-

nye

arm/leg

die

ALL

side-

1SG

‘Arm/leg is weary on me.’ (i.e., My arm/leg is weary/numb.)’

Essentially the external dative possessor construction implies that the condition is permanent or at least cannot be easily restored. Thus, deafness can be talked about using this construction. The allative external possessor construction, on the other hand, is used to express temporary physical conditions such as the temporary numbness of an arm or a leg. To sum up, the verb kú ‘die’ occurs in various structures where there is one core or central participant, that is, in one-place constructions and in external possessor constructions. To the extent that the semantics of the verb in these structures is invariant, the language learner might be helped by syntactic bootstrapping to infer the

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

241

meaning. In all the cases discussed in this subsection, the verb occurs in a one-place construction. However, the interpretation of the surface forms that the child will encounter depends on the type of referent the single participant is. If it is a living thing, the child might interpret the verb as denoting loss of life. If it is an inanimate referent, the child might interpret the verb as indicating loss of function. However this is not all, because the child soon encounters situations where the verb occurs with two surface NPs. In some cases the semantics discussed so far is still applicable, while in others the child is confronted with a meaning change. 3.2.2. Kú ‘die’ with Two Surface NPs The contexts of use of the verb kú described in the previous section all involve structures in which there is only one NP filling the core argument position in the clause in which the verb occurs. In the case of the external possessor constructions, the adjunct position in the clause is filled by a prepositional phrase. In this section we are concerned with situations in which there are at least two NP slots in the surface structure of the clauses in which the verb occurs. There are four specific constructions of this kind: the cognate object construction, as exemplified in (15), the secondary predicate construction, illustrated in (16), the Theme-Locative construction, exemplified in 7 (17), and the Causal two-place construction, shown in (18). 15. (Folktale Yiyi line 7) Yiyi

kú

anyrá-kú

Spider

die

wicked-death

‘Spider died a wicked death.’ (i.e., pretended to die)’ 16. É-kú 3SG-die

vi-ma-dzi-ma-dzi(-i) child-NEG-bear-NEG-bear-ATTRIB

‘S/he died childless.’ 17. (A mother complaining about the clothes of a child) Awu

lá

kú


i

garment

DEF

die

dirt

‘The garment is dead dirty.’ 18. É-dzilá-wó 3SG-parent-PL

kú

tó


é

nya

lá

die

ear

ALL

matter DEF

me containing.region.of

‘Her parents turned a deaf ear to the matter.’(lit.: ‘Her parents died ear to the matter.’)

The fact that two NPs appear on the surface with the verb kú ‘die’ is odd, and raises the question of how a learner would come to acquire this verb and its syntax and semantics. From a syntactic point of view, one question that these constructions

242

AMEKA

raise is whether the second NP, the postverbal NP, is a syntactic object argument of the verb kú ‘die’. A related question is whether the interpretation of such structures can be straightforwardly derived from the semantics of the verb kú ‘die’ proposed earlier, or whether one needs to posit a second meaning for the verb. I address these questions in turn. An NP bearing the object grammatical relation in Ewe can be identified among other things on the basis of its postverbal position and its permutation with its verb head under nominalization (Ameka, 2002). According to these criteria, the second NPs in the cognate object construction (15), the Theme-Locative construction (17), and the Causal two-place construction (18) function as grammatical objects. The secondary predicate NP does not satisfy these criteria (see (19)). For instance, an adjunct temporal noun like ets ‘one day removed from today’ and a speed adverb such as kábá ‘quickly’ cannot occur between a verb and its object, as demonstrated by the unacceptability of (20b) (compare with the acceptable version in (20a)). However such words can occur between the verb and the secondary predicate, as illustrated in (19). 19. a. É-kú

vi-ma-dzi-ma-dzi-i

kábá

3SG-die child-NEG-bear-NEG-bear-ATTRIB

quickly

‘S/he died childless quite quickly, i.e., early in life.’ b. É-kú 3SG-die

kábá

vi-ma-dzi-ma-dzi-i

quickly

child-NEG-bear-NEG-bear-ATTRIB

‘S/he died childless quite quickly, i.e., early in life.’ 20. a. É-kú 3SG-die

nyágã

kábá

old person

quickly

‘She became a very old person quite quickly.’ b. *É-kú 3SG-die

kábá

nyágã

quickly

old person

‘S/he died early an old person.’

Another piece of evidence that the secondary predicate nominal is not a grammatical object in Ewe is that it cannot be permuted with or preposed to the verb to form a nominalized VP in a progressive or prospective aspect construction (see Essegbey, 1999: 120). In contrast, the cognate object NP, and the locative NP and the patient NP in the other constructions, can be preposed, as illustrated in (21) and (22). In (21) the cognate object in the nominalized VP of the prospective aspect construction is made up of the noun kú ‘death’modified by kpo ‘sudden’. In (22) the NP ƒenyí ‘swoon’ is preposed to the verb kú ‘die’, showing its objecthood.

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

243

21. (Fiawoo 1838) Ví-nye,

è-le

kú

kpo

kú

gé

child-1SG

2SG-be.at:PRES

death

sudden

die

PROSP

‘My child, you are going to die a very sudden death.’ 22. (Obianim 3512) Amedzró

lá

le

ƒenyí

´ kú-m

guest

DEF

be.at:PRES

swoon

die-PROG

‘The guest is in a coma (and he is almost dead).’

Thus, in three out of the four structures in which the verb kú ‘die’ occurs with two surface NPs, the second NP can be shown to be a grammatical object. The secondary predicate nominal construction stands out. The language learner could be helped to identify this secondary predicate structure on the basis of the optional marker /-i/ that sometimes appears on the surface. Such an overt form, from a syntactic bootstrapping perspective, might guide the learner to the meaning of such a construction, namely, the state or condition the central participant was in when it lost the salient property. Perhaps the form of the cognate object NP, containing a copy, as it were, of the verb, might also provide a cue to the learner that there is only one semantic participant whose manner of losing a salient property is specified by the second surface NP. It is not clear, however, how a syntactic or semantic bootstrapping approach would guide the child to the right interpretation of the verb kú ‘die’ in the ThemeLocative and Causal two-place constructions. In these contexts, there is a shift in the meaning of the verb. These shifts are discussed in the next section. 3.2.3. The Semantic Shifts of Kú ‘die’ In this section I justify the claim made earlier that in the Theme-Locative (two-place) construction, the verb kú ‘die’ has a feature of intensity, whereas in the Causal two-place construction it triggers a desiderative reading. In section 4, I provide examples from other languages showing that these meaning shifts are systematic, and occur across a variety of unrelated languages. However, these meanings are by no means universal. Furthermore, in Ewe there seems to be an added evaluative connotation of ‘this is bad’ associated with these structures. The intensity sense of kú ‘die’ in Theme-Locative constructions can be drawn out if the interpretations of such sentences are compared with expressions for similar situations that codify a more neutral state. The expression in (23a) instantiates the copular construction for assigning an entity to a class or for specifying an entity’s relation. This is the unmarked construction. The utterance in (23b), however, is more specific. It signals that the person being talked about is a very old person;

244

AMEKA

that is, it expresses an intensity meaning. Given that the only contrast between the two sentences is in the verb, we can conclude that intensity is a semantic component of the verb kú ‘die’ in the Theme-Locative construction. 23. a. É-nyé 3SG-COP

nyág㐠old.person

‘S/he is an old person.’ b. É-kú 3SG-die

nyág㐠old.person

S/he is a very old person. (lit.:‘S/he died an old person.’)

Similarly, a pair of expressions exists for talking about things that are dirty given the norm and those that are very dirty. The expression for the very dirty things is a ThemeLocative construction involving the verb kú ‘die’. Compare the sentences in (24). 24. a. Awu garment

lá

ƒo


i

DEF

hit

dirt

‘The garment is dirty.’ (= 9 shown earlier) b. Awu garment

lá

kú


i

DEF

die

dirt

‘The garment is dead dirty.’ (= 17 shown earlier)

The nouns that fill the object slot in the Theme-Locative construction all describe states or conditions of entities. Some of the common collocations include kú aha [die alcohol] ‘be dead drunk’, kú alâ [die sleep] ‘be dead asleep’, kú tsu ‘be dead mad’, and kú ameg㏠[die old person] ‘be dead/very old’. One striking thing about the Ewe combinations is that they tend to be intense states that are thought to have adverse effects on the entities that are in that state. Thus, there is a negative connotation in Ewe, which, as we see later, seems to be absent from the intensity reading of ‘die’ predicates in other languages. Such a feature is an Ewe-specific property that the language learner must acquire. Apart from its intensity (‘very’) meaning with its evaluative component (‘This is bad’), the verb kú ‘die’ has a further negative-desiderative (‘not want’) reading when it occurs in the Causal two-place construction, as in example (18) shown earlier. For this reading, the subject role must be filled by an NP whose referent is a person (or part of a person), not a thing. This follows from the desiderative aspect of the reading and also suggests that it may be a default reading rather than being part of the entailments of such structures. For language learners, however, this means that they must not only cope with the fact that the verb kú ‘die’ has two arguments, and probably therefore a different meaning than that of single-participant constructions, but they should also pay attention to the participant-type properties of the ar-

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

245

guments in order to generate the right contextual interpretation. They also have to take account of the properties of the second NP, because the referent of the subject argument can be a person in the Theme-Locative construction as well. In short, the semantic shifts triggered by the occurrence of kú ‘die’in two-place constructions in Ewe give rise to a complex system of mapping between forms and meaning, and additionally to conventional interpretations that need to be learned. To sum up thus far, I have shown the different types of constructions in which the verb kú ‘die’ occurs in Ewe: the one-place construction, cognate object construction, Theme-Locative (two-place, intensity) construction, and Causal two-place (negative-desiderative) construction. I have also indicated that the verb can occur in a secondary predicate structure as well. Various semantic shifts of the verb correspond to the different syntactic constructions in which the verb occurs. In the one-place, the secondary predicate, and the cognate object constructions, the verb kú ‘die’ carries the meaning of the loss of a salient property that is central for characterizing the referent of the subject argument as a member of the category of entities that it belongs to. In the Theme-Locative construction, the verb has the sense of intensity ‘very’ and is used to indicate that an entity is deeply in a state and this is evaluated as bad. In the Causal two-place construction where the subject NP argument refers to a person, a negative desiderative interpretation is triggered. Thus the verb kú ‘die’ has at least two senses, with a conventional implicature associated with one of the senses. The shifts of the meaning of the ‘die’ verb in Ewe just outlined are not unique. In many languages of the world (e.g., Dutch, Arrernte [Australia], Oluta Popoluca [Mexico], English, Japanese) a ‘die’/‘dead’ element shows a corresponding meaning shift from DIE to INTENSITY and/or from DIE to DESIDERATIVE when the verb takes a second argument. In fact the three readings of die in English are reflected in the first part of the title of this chapter. In he died old, die occurs with an adjectival secondary predicate and has the normal sense which can be roughly glossed as ‘become not alive’, whereas the interpretation of dying to be dead right is ‘wanting to be very right’. Thus the reading of die in dying is desiderative and of dead is intensity. Even though the meaning shifts correlate with different syntactic patterns, the patterns are not the same across the two languages. We cannot therefore account for these features from universal principles. 4. SEMANTIC SHIFTS OF ‘DIE’ TERMS ACROSS LANGUAGES In this section, I adduce evidence to show that the semantic shifts associated with the Ewe kú ‘die’verb noted in section 3 are attested in typologically and genetically diverse languages. Nevertheless, the similar senses are associated with different syntactic configurations. In some languages, the meanings are associated with a single lexical item belonging to one category, as in Ewe or Japanese. In other languages, such as Oluta, the meaning changes are in part associated with category shift involving grammaticalization. Thus, in Oluta there is a ‘die’verb and an auxiliary form that is grammaticalized from it, as shown in Table 11.1. Such semantic

TABLE 11.1

Die: Intensity

V NP [STATE] (Theme-Locative construction) É-kú aha ‘He died alcohol.’ i.e., he was dead drunk. V [+PROG] of phrase I am dying of thirst.

Verb satellite modifying V root; ‘V too much, a lot’ ?i-mot-?o:k-nüwak ERG3-salt-die-ASPECT ‘He salted it a lot.’ V hodo (Adverbial) ‘to the degree of dying’ i.e., very much Taro wa shinu hodo okot-teiru Taro TOP die degree get.angry-ASPECT ‘Taro is very angry.’

Die: Primary Sense

kú V (+NP cognate object/ secondary predicate)

die V (+NP cognate object/ secondary predicate)

?o:k V (+ secondary predicate)

shinu V

Language

Ewe

English

Oluta Popoluca

Japanese

Patterns of Semantic Shift of ‘Die’ Verbs across Languages Die: Desiderative

[no examples]

V INF Aux ‘want to V’ ta-?o:k-i-?o:k-pa ABS1-die-NOM-die-ASPECT ‘I am dying to die.’

V[+PROG] V INF I am dying to see you.

V[+PROG] for phrase I am dying for a drink.

V NP (Causal two-place construction) É-kú tó ‘He died [his] ear.’ i.e., he did not want to listen.

246 AMEKA

V van ‘of’ phrase Ik sterf van de dorst ‘I am dying of thirst.’

sterven V (+NP cognate object/secondary predicate)

sterben V (+NP cognate object/secondary predicate)

Dutch

German

V vor ‘for’ phrase Ich sterbe vor Hunger / Kält / Angst 1SG die for hunger / cold / fear ‘I am dying of hunger / cold / fear.’

V [infinitive] + Adj (compound Adj) stervenskoud stervensdruk ‘freezing cold’ ‘dead busy’

V van ‘of’ phrase (impersonal) ‘Het sterft er van … ’ Het sterft er van de vliegen It die-PRES here of the flies ‘It is swarming with flies.’

[no examples]

met V (+NP cognate object)

Hebrew

[no examples]

[no examples]

V + INF / V + DAT Prep phrase hu met la lexet le seret he die.PRES.SG:MASC to go to movie ‘He is dying to go to the movies.’

11. TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE 247

248

AMEKA

shifts have been observed by Whorf (1935: 605), who notes that “throughout the daughter tongues [of Uto-Aztecan languages (FKA)], the verb ‘die’…is used as the second member of a compound with an entirely different meaning, that of feeling a sensation or being in a certain bodily state.” Wichmann (1993: 55–56) makes a similar observation concerning Proto-Oaxaca Mixean. He notes that one can reconstruct an intransitive compound verb, *yu:?o:?k ‘be hungry’, which comprises yu:? ‘be hungry’ and ?o:?k ‘die’. He adds that ?o:?k has a specialized meaning and that “intensity is certainly part of the meaning it conveys. In that sense it is somewhat reminiscent of expressions like I’m dying to see you.” In other languages, such as Dutch and English, it is not only the ‘die’ verb element for which the semantic shifts are observed; nonverb ‘die’elements such as adjectives or adverbs (as in Arrernte) also are associated with the same semantic shifts (see Table 11.2). Tables 11.1 and 11.2 display data from selected languages for ‘die’ verbs and nonverbs, respectively. The three broad readings identified for Ewe kú ‘die’, namely, ‘die’, ‘intensity’, and ‘(negative) desiderative’, head the columns. It is evi-

TABLE 11.2 Patterns of Semantic Shift of Nonverb ‘Die’ Elements across Languages

Language

‘Death’/‘Dead’: Basic form

Arrernte tnye Nominal (Nominals function as both N and Adj)

‘Death’/‘Dead’: Intensity

tny-ante (Adverb) lit.: ‘dead-only’ ‘do something too much/a lot’

‘Death’/‘Dead’: Desiderative

[no examples]

Ampe yanhe tny-ante angkeme child that dead-only speak/say ‘That kid is screaming its lungs out.’ tnye-ke-ante (Adverb) lit.: ‘dead-DAT-only’ ‘do something too much/a lot’ The tnye-ke-ante arlkwe-ke 1SG:ERG dead-DAT-only eat-PAST ‘I ate far too much.’

Dutch

dood N/Adj

N/Adj + Adj (compound Adj) Doodziek doodmoe doodnormaal ‘dead sick’ ‘dead tired’ ‘very normal’

[no examples]

English

dead Adj

Adj + Adj (compound Adj) dead good, dead right dead calm, dead easy dead straight, dead set

[no examples]

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

249

dent from the tables that the semantic shifts have different grammatical consequences in the individual languages. In Ewe, the shifts occur when the verb occurs with a noncognate direct object. In other languages, such as English, the shift occurs when the ‘die’ verb occurs in a specific form and takes a sentential complement. In spite of the broad crosslinguistic similarities, there are still differences between languages. In particular, there are differences in the range of application of these readings, reflecting the language-specific semantic differences between the forms. These are not completely reflected in the tables. For instance, in some cases when the same referent is the expressed argument of the verb in two different languges, it triggers one interpretation in one language and another interpretation in the other. In Ewe, for example, when the verb is predicated of a river, the interpretation is that the river has dried up. In Arrernte (Australian), on the other hand, when the ‘die’ element is predicated of a river it means it has lost its vital force, not dried up but perhaps poisoned or muddied. Such limits on the range of application are consistent with the language-specific semantics of the terms. 5. CONCLUDING REMARKS The analysis of the Ewe verb kú ‘die’ that emerges is one in which there are at least three readings of the verb linked to the specific construction types in which it occurs. There is one sense for the one-place construction where the single participant undergoer loses a salient property and therefore can no longer be identified as a member of its category. It is this same sense that is modulated contextually when the verb occurs in cognate object constructions or with nominal secondary predicates or in some types of external possessor constructions. A second sense, that of intensity interpretation, is linked to the Theme-Locative (two-place) construction. The theme participant is understood as having the property represented by the second NP to an extreme degree. The third reading is the one that is triggered by the verb’s occurrence in a Causal two-place construction. There is a default negative desiderative interpretation associated with this usage. As a verb with variable meaning interpretation, the acquisition of kú ‘die’requires considerable attention to context (cf. Clark, 1993, for children’s difficulties with variable-meaning words). In addition, the type of participant that is central in each of the constructions must also be attended to, because that also contributes to the contextual interpretation of the verb (cf. Gropen et al., 1997). Although the shift from ‘die’ to ‘intensity’ to ‘(positive or negative) desiderative’ is crosslinguistically widespread, the formal correlates of these meanings vary considerably. In the case of Ewe, the possibility for the verb kú ‘die’ to take syntactic objects and to occur in the causal transitive construction is not something that would be predicted by the universal principles that are usually assumed for the linking of arguments and predicates, and certainly not for a ‘die’ predicate. My conjecture is that when the verb occurs with two surface NPs, this syntactic pattern would be a signal to the language learner that there is more to the meaning than when there is one NP.

250

AMEKA

Thus, some syntactic bootstrapping might help the learner. The problem, however, is that the same surface structure of two NPs with the verb can have any of the three readings noted earlier. It is not clear how any universalist principle would help the language learner to factor out the distinctions. It seems the learner just has to learn the constructions. However, the Ewe learner also has to learn some specific collocations, not only those pertaining to intensity like kú aha ‘dead drunk’ but also the types of entities that ‘die’in Ewe can be predicated of, like sores, the moon, and various body parts. It is intriguing that a universal experience such as dying is so highly constrained culturally and linguistically that expressions pertaining to it might have to be learned on an item-by-item basis rather than from any universalist claims or expectations about them. ACKNOWLEDGMENTS I thank several colleagues and friends who shared their knowledge about various languages with me and commented on earlier drafts: Caroline Angenent, Anneke Breedveld and Carlien de Witte (Dutch), Jidong Chen (Mandarin Chinese), James Essegbey (Ewe), Sotaro Kita (Japanese), C. Karl (Hebrew), David Wilkins (Arrernte and English), Eva Schultze-Berndt (German), and Roberto Zavala (Spanish and Oluta). I am also grateful to audiences at Trondheim University, Leiden University, Leipzig University, and the Argument Structure Workshop, MPI Nijmegen, especially Lars Hellan, Lawrence Boadi, Ekkehard Wolff, Orin Gensler, Kim Hak-Soo, and Beth Levin. Special thanks to Eva Schultze-Berndt, Penny Brown, and Melissa Bowerman for their incisive comments on an earlier version. NOTES 1

The following abbreviations are used in the interlinear glosses:

ALL

allative preposition

poss

possessive linker

ATTRIB

attributive marker

Postp

postposition

CONN

connective

PostpP

postpositional phrase

CQ

content question marker

Prep

preposition

DEF

definiteness marker

PRES

present

DIM

diminutive

PROG

progressive aspect marker

FEM

feminine

PROSP

prospective aspect marker

aFOC

argument focus marker

RED

reduplicative formative

HAB

habitual aspect marker

REL

relative clause introducer

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

LOC

locative preposition

REP

repetitive

MASC

masculine

SBJV

subjunctive

MOD

modal

SER

serial attributive marker

NP

nominal phrase

SG

singular

NEG

negative

UFP

utterance-final particle

NPRES

non-present

PAST

past tense

1

first person

PERF

perfective aspect marker

2

second person

PL

plural marker

3

third person

251

For Ewe examples, high tones are marked throughout with an acute accent in addition to the low tones that are customarily marked in the traditional orthography with a grave accent. The hacek marks a rising tone. Ewe orthographic ƒ and  are the voiceless and voiced bilabial fricatives respectively. 2 The other forms that behave like this include: fi V ‘to steal’, N ‘theft’; nó V ‘to suckle’, N ‘breast’; dzu V ‘to insult’, N ‘insult’. 3 One crucial distinction between the two forms is that the nyá ‘certainty modal’ cannot take the habitual marker, whereas, as should be expected, the verb form does. 4 After this chapter was finalized, Robert Botne’s (2003) crosslinguistic study of the lexical aspect and temporal structure of ‘die’ verbs appeared. He distinguishes four temporal structures of basic ‘die’: ‘acute’, where the ‘die’ verb codes only a punctual nucleus as in Yoruba or Norwegian; ‘inceptive’, where the verb codes both an onset phase and a punctual nucleus as in English, French, or Hausa; ‘resultative’, where the ‘die’ verb codes the nucleus plus a coda as in Japanese, Thai, Korean, and Finnish; and ‘transitional’, where the verb entails an onset, a nucleus, and a coda. The last type occurred only in the Niger Congo languages of Africa in Botne’s sample, namely, Akan (Kwa, West Africa) and Cindali (Bantu, East Africa). The Ewe verb kú ‘die’, as I demonstrate, also has the transitional temporal structure like its genetic and areal relative Akan. 5 This expression does not imply that the child is fatally sick. In fact, Me-le ku-kúm, lit.: ‘I am dying’, can be used in response to a ‘how are you’ question just to signal that one is not feeling well. 6 A serial verb construction (SVC) is a monoclausal construction in which two or more verb phrases including complements occur. All the verb phrases in the series share the same subject, which is expressed only once on the first verb phrase. An example of such a construction involving the verb kú ‘die’ is: É-no tsi kú [3SG-drink water die] ‘S/he drowned’ (see Ameka, 2002, for further details). 7 There are two types of two-place (transitive) constructions in Ewe. The semantics of one indicates that it is construed as involving cause, and the participant that occurs in the privileged syntactic position is an agentive Actor. The second type is a noncausal two-place construction whose central participant is a

252

AMEKA

Theme. To distinguish the two types, we refer to the former as the Causal twoplace construction. REFERENCES Ameka, F. K. (1994). Ewe. In C. Goddard & A. Wierzbicka (Eds.), Semantic and lexical universals: Theory and empirical findings (pp. 57–86). Amsterdam: John Benjamins. Ameka, F. K. (2002). Constituent order and grammatical relations in Ewe in typological perspective. In K. Davidse & B. Lamiroy (Eds.), Nominative and Accusative and their counterparts (pp. 319–352). Amsterdam: John Benjamins. Ameka, F. K., & Wilkins, D. P. (1999). Is ‘DIE’ a semantic prime? Evidence from Ewe and Arrernte. Presented at 4th Rasmus Rask Colloquium on Language and Communication (The Anna Wierzbicka Festival), Odense University, Denmark, March. Botne, R. (2003). To die across languages: Toward a typology of achievement verbs. Linguistic Typology, 7, 233–278. Brown, D. J. (1991). Human universals. New York: McGraw Hill. Capo, H. B. C. (1991). A comparative phonology of Gbe. Berlin and Garome: de Gruyter (Foris) and LABOGBE. Clark, E. V. (1993). The lexicon in acquisition. Cambridge: Cambridge University Press. Clements, G. N. (1972). The verbal syntax of Ewe. PhD thesis, University of London. DeLancey, S. (1995). Verbal case frames in English and Tibetan. Available at http:// darkwing.uoregon.edu/~delancey/papers/caseframes.html. DeLancey, S. (1997). What an innatist argument should look like. In T. Hankoja, M.-L. Helasuvo, & M. Miestamo (Eds.), SKY 1997 (1997 Yearbook of the Linguistics Association of Finland) (pp. 7–24). Helsinki: Linguistics Association of Finland. Durie, M. (1987). Grammatical relations in Acehnese. Studies in Language, 11, 365–399. Duthie, A. S. (1996). Introducing Ewe linguistic patterns. Accra: Ghana Universities Press. Essegbey, J. (1999). Inherent complement verbs revisited: Towards an understanding of argument structure in Ewe. PhD dissertation, Leiden University. Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Goldberg, A. E. (2002). Surface generalizations: An alternative to alternations. Cognitive Linguistics, 13, 327–356. Gropen, J., Epstein,T., & Schumacher, L. (1997). Context-sensitive verb learning: Children’s ability to associate contextual information with the argument of a verb. Cognitive Linguistics, 8, 137–182. Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity. Cambridge, MA: MIT Press. MacFarland, T. (1995). Cognate objects and the argument/adjunct distinction in English. PhD thesis, Northwestern University. Murdoch, G. P. (1945). The common denominator of cultures. In R. Linton (Ed.), The science of man in the world crisis (pp. 123–142). New York: Columbia University Press. Schultze-Berndt, E., & Himmelmann, N. P. (2004). Depictive secondary predicates in crosslinguistic perspective. Linguistic Typology, 8, 59–131. Van Valin, R. D., Jr. (1990). Semantic parameters of split intransitivity. Language, 66, 221–260. Westermann, D. (1930). A study of the Ewe language. Oxford: Oxford University Press. Whorf, B. L. (1935). The comparative linguistics of Uto-Aztecan. American Anthropologist, 37, 600–608. Wichmann, S. (1993). Grammaticalization in Mixe-Zoquean languages. Sprachtypologie und Universalien Forschung (STUF), 46, 45–60.

11.

TRANSITIVITY AND SEMANTIC SHIFTS OF ‘DIE’ IN EWE

253

Wierzbicka, A. (1988). The semantics of grammar. Amsterdam: John Benjamins. Wierzbicka, A. (1996). Semantics: Primes and universals. Oxford: Oxford University Press. Wierzbicka, A. (1998). Anchoring linguistic typology in universal semantic primes. Linguistic Typology, 2, 141–194.

Ewe Texts Agbodza, M. K. (1962). Kesinnu kple yayra (Riches and blessings). London/Ho: Thomas Nelson and Sons Ltd./EP Church Book Depot. Ayeke, K. (1974). Hl~biabia (Taking revenge). Accra: Bureau of Ghana Languages. Fiawoo, F. K. (1990). Tk At~lia (The Fifth Landing Stage). Accra: Sedco. Obianim, S. J. (1990). Agbezuge. Accra: Sedco.

CHAPTER 12

Acquiring Telicity Crosslinguistically: On the Acquisition of Telicity Entailments Associated with Transitivity Angeliek van Hout University of Groningen

1. INTRODUCTION In many languages event descriptions come in two semantic classes: some predicates are telic, others are atelic. Write a book, perform a play, and bake a cheesecake are telic predicates; the events they denote culminate at a specific moment. Atelic event descriptions, such as carry a book, look at a play, and love cheesecake, do not involve such a moment of culmination. This property of whole verb phrases (the verb plus its internal arguments, goal or resultative phrases, and relevant particles and morphology) is called “lexical” aspect (also referred to as Aktionsart and situation-type aspect). Telicity is one of the lexical aspectual features; along with the additional features stative/dynamic and durative/punctual, four aspectual classes can be defined: states, activities, accomplishments, and achievements (Vendler, 1957). Lexical aspect is to be distinguished from grammatical aspect. The latter refers to the actual beginning and final boundaries of an event, whether they are implied or not. A perfective reading presents the event as an unanalyzed whole, including its initial and final boundaries, whereas an imperfective reading zooms in on the event in progress without reference to the time when it started or ended (Comrie, 1976; Smith, 1991). Lexical aspect is a property of verb phrases, whereas grammatical aspect is carried by tense and aspect morphology.1 Lexical and grammatical aspect are thus two separate categories of aspect. Grammatical aspect is “built” on top of lexical aspect, in the sense that, in most languages, both telic and atelic verb phrases can be combined with perfective or imperfective aspect (cf. de Swart, 1998, for an analysis of the interaction of the aspectual tenses of English and French with 255

256

VAN HOUT

(a)telicity). Imperfective aspect yields the same reading for telic and atelic VPs, namely, an event in progress with no entailment of completion or termination. Perfective aspect, in contrast, gives rise to subtly different readings: with a telic predicate it entails completion, whereas with an atelic predicate it presents an event as simply terminating (i.e., at an arbitrary moment). This chapter discusses children’s acquisition of the lexical aspect notion of telicity across three language families—Germanic, Finno-Ugric, and Slavic—that encode telicity differently. Raising an old theme from the 1970s—the relative learnability of different languages—I ask whether the way one language marks telicity is easier to learn than how another language marks it. Although it is generally accepted that there are universal processes involved in the process of language development and that languages are generally similar overall in ease of acquisition, crosslinguistic studies have revealed differences in the pace of development of particular aspects of the grammars under consideration. One poignant example is the expression of underlying grammatical relations such as agent and theme through word order rules or various types of inflectional systems. Slobin (1982) and Slobin and Bever (1982) have found that learners of free word order grammars that encode these relations through inflections (Turkish, in their study) master these relations at a younger age than learners of grammars that rely exclusively on word order (English and Italian, in their study). Slobin explains the relative ease of mastery in Turkish by referring to what he calls “local cues” (e.g., the accusative case marker on the object in Turkish), which “can be interpreted without taking the entire sentence into account. The cue is local because it operates in a localized sentence element” (Slobin, 1982: 162). This explanation adopts the well-known concept of “surface cues to underlying structure” from sentence processing theories (e.g., Fodor, Bever, & Garrett, 1974), and applies it to the process of language development as an acquisitional strategy that leads to early discovery of form–meaning relations. I argue that the locality of the cue also plays a role in the acquisition of telicity: it is easier to learn telicity in languages or constructions that mark this property on the verb itself, for example, perfective marking in the Slavic languages and resultative verb particles in the Germanic languages, than in constructions in which telicity has to be construed from the properties of the verb and its object together, as in the Germanic languages and Finnish. Section 2 lays out the different syntaxes of telicity in English, Dutch, Finnish, Russian, and Polish, arguing that the first three languages have what I call “compositional telicity” whereas the latter two have “predicate telicity.” Section 3 presents results from aspect comprehension studies in Dutch, English, and Finnish that show that learners of these languages who are as old even as 5 or 6 years still have problems establishing telicity entailments. A review of aspect comprehension studies in Polish and Russian in section 4 shows that young learners of these languages, in contrast, do not have trouble with telicity entailments. In section 5 I draw the conclusion that it is easier to learn predicate telicity than compositional telicity, and speculate on why this may be so.

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY

257

2. TELICITY SYNTAX IN ENGLISH, DUTCH, FINNISH, RUSSIAN, AND POLISH 2.1. Properties of Telic and Atelic Event Descriptions In Krifka’s (1986, 1992) semantic framework, the formal properties of atelic event descriptions, but not telic ones, include cumulativity and homogeneity. Cumulativity is related to the addition of two objects or events to each other. If one adds water to water, the result can still be referred to as water. But if one adds one apple to another apple, the result is two apples; it cannot be referred to as an apple. The mass term water is a cumulative predicate, whereas the count term an apple is non-cumulative: it denotes a specific amount of apple (Verkuyl’s 1972 terminology) and is quantized (Krifka’s terminology). In the same way, write a book is noncumulative. If one writes a book, and then later writes another book, these are two distinct book-writing events (and two books have come into existence as a result). So write a book is a quantized predicate. But if one loves cheesecake on one occasion and still loves it on another, we usually refer to this as one single state of affairs of loving cheesecake, not two distinct ones. Hence, love cheesecake is cumulative and non-quantized. The property of homogeneity can be determined by examining the time spans of events. A predicate is homogeneous if one can refer to part of that time span or an extension of it with the same predicate. If it is true that one carried a book from 1 o’clock to 3 o’clock, then we may conclude that one also carried the book from 1 to 2. But if one wrote a book from 1 to 3, there is no similar entailment that one also wrote a book from 1 to 2. Thus, carry a book is a homogeneous predicate, and write a book is non-homogeneous. Telicity is partly determined by the lexical semantics of the verb. The difference between carry a book and write a book comes from the verbs write versus carry. Writing implies culmination: the moment when the book is finished. Carrying does not involve culmination: an event of carrying a book will at some point stop, but this moment is arbitrary. Write, perform, and bake all have a property that distinguishes them from verbs like carry, look at, and love: their direct objects are so-called “incremental themes” (Dowty, 1979), in the sense that the progress of the event can be tracked by the incremental involvement of the object undergoing a particular change of state.2 Activity verbs like carry and look at, along with stative verbs like love, hate, and know, do not have incremental theme objects: an activity like carry or look at does not progress along a specific scale associated with the object. And a state is not dynamic, so there is no sense in which one can measure the progress of a state by examining what happens to the object. 2.2. Telicity Syntax in English and Dutch The aspectual meaning of the verb is important in determining telicity, but the realization of the verb’s arguments matters too. Although a verb may be specified in the

258

VAN HOUT

lexicon with two arguments, they do not both always appear overtly; for example, the English transitive verb eat may turn up in an intransitive or passive frame (John ate, the cookie was eaten). Moreover, a theme argument may appear as direct object or as the object of a preposition. When there is an overt direct object, its quantificational semantics affects the telicity of the verb phrase. The details of argument realization thus play a crucial part in the computation of telicity. In the 1990s it became clear that there are close links between argument structure and event structure: a full theory of argument structure should incorporate the aspectual effects associated with different argument structures and, conversely, a full theory of event structure should be linked to a verb’s argument structure (Borer, 1994; Grimshaw, 1990; Ramchand, 1997; Schmitt, 1996; Tenny, 1987, 1994; van Hout, 1996). Consider the possibilities for the two-argument verb write in the argument alternation patterns in (1), with their different telicity values. Temporal modifiers are added to bring out (a)telicity: in one summer is a time frame adverbial and combines with telic predicates; for months is a durative adverbial and combines with 3 atelic predicates. In order to focus on lexical aspect without interference from the tense and aspect markers that contribute toward grammatical aspect, I use infinitive forms. 1. a. …to write a film script in one summer/*for months. b. …to write film scripts *in one summer/for months. c. …to write at a film script *in one summer/for months. d. …to write *in one summer/for months.

To write a film script may involve stopping and continuing again, but if one gives up the whole project at some point, we cannot refer to this writing event as write a film script. For this predicate, a time frame adverbial is appropriate, because it measures the event plus the culmination (1a). Note, however, that a verb phrase with a bare plural object is atelic (1b). The semantics of the direct object thus affects telicity: only count term direct objects may give telicity; bare plurals and mass terms do not (Verkuyl, 1972). Unlike transitive write in (1a), to write at a film script in (1c) may refer to a writing project that was abandoned at some point. With the theme argument in a prepositional phrase (in what I call an “oblique” verb frame), there is no development along a clear path to culmination; hence a time frame adverbial is impossible. Like write at a film script, intransitive write without any overt object at all is atelic (1d). When one leaves the theme argument of writing unspecified, there is no inherent culmination, so only a durative adverbial may be used. As the three alternants in these examples show, the syntactic and semantic realization of the theme argument matters for telicity in English and Dutch: only when it is realized as a direct object is the predicate telic.4 This pattern is robust and holds

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY

259

for all incremental theme verbs. There is a strong correlation between telicity and transitivity: in order to create a telic predicate, an incremental theme verb must appear in a transitive verb frame. Moreover, its direct object must be a count term of a specified quantity. I refer to this composition of verb and object as “compositional telicity.” Notice that the implication does not hold in the opposite direction. That is, transitivity does not imply telicity, because not all transitive predicates are telic; in particular, when stative and activity verbs appear in a transitive frame, the predicate is atelic. In (1), I am reporting the typical judgments found in the aspect literature. Not all speakers of English, including a reviewer of this chapter, agree that for months (the atelic reading) in (1a) is ungrammatical; in fact, the experimental results with adults to be reported shortly also suggest that transitives such as (1a) are ambiguous between a telic and atelic reading. I will come back to this point in the discussion of the results. The correlation between transitivity and telicity holds up nevertheless as a one-way implication.5 Many verbs with an incremental theme, like write, are flexible in their argument projection patterns, and so in their aspect, but some are not. One such inflexible verb is write up, and its Dutch counterpart opschrijven, literally ‘up-write’: 2. a. Elena heeft de review *urenlang/in een uur opgeschreven. Elena has the review *hours-long/in an hour up-written. ‘Elena wrote up the review *for hours/in an hour.’ b. *Elena heeft opgeschreven. Elena has up-written. c. *Elena heeft aan de review opgeschreven. Elena has at the review up-written.

Opschrijven ‘write up’ can only appear with a direct object in a transitive frame, (2a); it cannot appear in an intransitive or oblique verb frame (2b, 2c). Unlike schrijven ‘write’, it does not have the option to appear in either telic and atelic predicates; rather, opschrijven ‘write up’ is inherently telic. This shows again the close correlation between telicity and transitivity. Telicity in the case of particle verbs like opschrijven ‘write up’ comes from the resultative particle (Filip, 1993; van Hout, 1996). This is an instance of what I call “predicate telicity”: there is an overt marker on the verbal predicate that carries telicity. Summarizing, although telicity is often characterized in terms of the aspectual meaning inherent in the verb’s semantics, the telicity of a verb phrase is not a purely lexical affair, but rather a compositional process between the verb and its object. Thus, a verb’s lexical property of taking an incremental theme or not makes the difference between telic write a film script, perform a play, and bake a cheesecake and atelic carry a film script, look at a play, and love cheesecake, but telicity is further constrained by syntactic–semantic factors in the verb phrase. Importantly, the

260

VAN HOUT

computation of telicity depends only on direct objects, not oblique objects or non-overt, understood objects. Moreover, the quantificational semantics of the direct object noun phrase (mass or count term) is implicated; only count terms may yield telicity. Compositional telicity involves these joint syntactic–semantic effects. This means that for a learner of a language with compositional telicity, the verbal syntax of argument realization as well as the nominal syntax of count/mass must be learned in order to be able to compute telicity correctly. 2.3. Telicity Syntax in Finnish Finnish also has compositional telicity, but the telicity of verb phrases is syntactically encoded in a different way. Unlike English and Dutch, Finnish noun phrases do not have articles that determine the count/mass distinction; instead, the morphological case of the direct object is crucial. Finnish transitive verbs that take an incremental theme are flexible case-assigners: they can assign either accusative or partitive case to their objects. The two different cases align with telicity: an accusative object makes the predicate telic, whereas a partitive object makes it atelic. Compare the alternation in (3), from de Hoop (1992: 105).6 3. a. Ammuin shot

karhu-n. bear-Acc

‘I shot the (a) bear.’ b. Ammuin

karhu-a.

shot

bear-Part

‘I shot at the (a) bear.’

With an accusative object the verb phrase is telic and the implication is that the bear was hit as a result of shooting (3a). With a partitive object the predicate is atelic and refers to a non-culminating activity of shooting; it can be translated into English with a prepositional object (‘shoot at the bear’) (3b). In Finnish, then, the combination of verb and object determines the telicity of the verb phrase, so compositional telicity is at play here too, albeit in a different way. In English and Dutch the presence or absence of determiners yields the mass/count distinction between noun phrases, which in turn affects a predicate’s telicity. The partitive–accusative case distinction in (3) does not encode the mass/count distinction of the noun phrase (both are count nouns), but partitive case marks the verb phrase as a whole as atelic and accusative case as telic. The aspectual effect of the case distinction can be illustrated with temporal modifiers, as shown in (4). A durative adverbial is infelicitous with an accusative object (4a), but is fine with a partitive object (4b); a time frame adverbial gives the opposite results.

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY 4. a. …kirjoittaa …write

kirja-n

yhdessä

kesässä/*kuukausia.

book-Acc

one

summer/months

261

‘…to write a book in one summer/*for months.’ b. …kirjoittaa …write

kirja-a book-Part

*yhdessä one

kesässä/kuukausia. summer/months

‘…to write at a book *in one summer/for months.’

To summarize, in Finnish as in English and Dutch, the nominal syntax–semantics of direct objects is critical in the computation of telicity. In English and Dutch it is the count/mass syntax of nominals that marks the countability of nominals, and hence the telicity of the verb phrase. In Finnish it is the case distinction between accusative and partitive case that makes a similar telicity contrast. Unlike the count/mass encoding on Dutch and English objects carried out by articles, the case distinction in Finnish does not directly mark the count/mass status of noun phrases, but it has direct effects on the telicity of the verb phrase. Accusative case in Finnish thus seems to be the formal counterpart to count noun syntax in English and Dutch (both yield a quantized reading of the object), but partitive case is not equivalent to mass noun syntax, because it is also compatible with a count noun interpretation of the object (as in (3b) and (4b)). (See Kiparsky, 1998, Krifka, 1992, and de Swart and Verkuyl, 1999, for formal semantic analyses of the aspectual effects of Finnish case.) Despite these differences, the common factor across English, Dutch, and Finnish is that the morphosyntax of the direct object is relevant for learning telicity. 2.4. Telicity Syntax in Russian and Polish In Russian and Polish there is no similar role for the direct object in establishing telicity. Russian and Polish do not have articles that mark the count/mass distinction, as in English or Dutch, nor is there an object case distinction associated with telicity, as in Finnish.7 Instead, telicity is indirectly marked on the verb itself, as a strong correlate of morphological perfectivity. Verbs in the Slavic languages are always aspectually marked as perfective or imperfective. This morphological paradigm is mapped onto the grammatical aspect readings perfective and imperfective, respectively.8 For the class of verbs under consideration in this chapter—transitives with an incremental theme—morphological aspect interacts with telicity: a telic reading arises only if the verb is morphologically perfective (Filip, 9 1993; Krifka, 1986, 1992; M»ynarczyk, 1998; Schoorlemmer, 1995). Because telicity is encoded on the verb itself, I refer to this as “predicate telicity,” in contrast to the system described for English and Dutch—which I termed “compositional telicity”—in which the combination of verb and direct object determines telicity. I illustrate the telicity effect of perfective aspect with a mixture of Russian and Polish examples, as verbs with an incremental theme behave alike in these two languages.

262

VAN HOUT

Consider the Russian perfective–imperfective pair for pisat’ ‘write’ in (5), taken from van Hout (2000b: 409–410) and discussed in detail by Schoorlemmer (1995). 5. a. Vanja Vanja

pisal

(pis’mo).

wrote (Imperf) (letter)

‘Vanja was writing (the/a letter).’ b. Vanja Vanja

na-pisal

*(pis’mo).

Perf-wrote

letter

‘Vanja wrote the/a letter.’

The aspectual paradigm affects the realization of the arguments: when pisat’ ‘write’ is imperfective it can appear as either intransitive or transitive—the object is optional (as in 5a). The perfective variant na-pisat’ ‘Perf-write’, in contrast, is obligatorily transitive (as in 5b). Recall that, similarly, inherently telic incremental theme verbs in English and Dutch are obligatorily transitive (see example (2)). The meaning difference between (5a) and (5b) involves grammatical aspect, as suggested by the translations with a progressive for the imperfective and a simple past for the perfective. But is there also a telic–atelic difference similar to the direct object–oblique object alternation in English/Dutch in (1), or the Finnish accusative–partitive case alternation in (3)–(4)? The temporal modifier test with durative and time-frame modifiers that was used for the other languages is suggestive but not conclusive in determining (a)telicity in Slavic languages, as these modifiers turn out to be sensitive to grammatical aspect. Imperfectives always combine with ‘for’-adverbials, despite the quantized nature of the object, but perfectives combine with ‘in’-adverbials (Borik, 2002; Schoorlemmer, 1995), which suggests that perfectives are telic. Example (6) illustrates this with Polish perfective and imperfective infinitives. 6. a. …pisa´c

scenariusz

…write(Imperf) script

filmowy film

*w dwie godziny/miesia¸c-ami. in two hours/months-Instr

‘…to be writing a film script *in two hours/for months.’ b. …na-pisa´c …Perf-write

scenariusz

filmowy

w dwie godziny/*miesia¸c-ami.

script

film

in two hours/months-Instr

‘…to write (and complete) a film script in two hours/*for months.’

Another effect of the perfective–telic association can be seen when perfective aspect puts a requirement of quantization on the verb’s direct object, which is an indication that perfectives are telic. The example in (7) is from Polish (M»ynarczyk, 1998). Similar effects have been observed for Czech (Filip, 1993).

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY 7. a. Maria Maria

pi»a

mleko.

drank (Imperf)

milk

263

‘Maria was drinking (the) milk.’ b. Maria Maria

wy-pi»a

mleko.

Perf-drank

milk

‘Maria drank the milk.’

Mleko ‘milk’ is interpreted as either a mass term or count term with an imperfective verb, as in (7a), but it must be interpreted as a countable unit of milk when the verb 10 is perfective, as in (7b). Given the necessarily quantized interpretation of the direct object, perfective incremental theme verbs are telic. There is an ongoing discussion in the Slavic aspect literature whether imperfectives can be telic or are always atelic. Filip (1993), Schoorlemmer (1995), and Borik (2002) all argue that imperfectives like (5a) and (6a) (but not (7a)) are indeed telic, derived by compositional telicity between the verb and the count term object. Thus, perfectivity cannot be fully equated with telicity, and imperfectivity with atelicity. The point I want to stress here is that Slavic languages have a formal way of encoding telicity on the predicate through perfective aspect that is formally equivalent to telic particles like up/op in English and Dutch: both ways of encoding yield predicate telicity. To summarize, I have argued that, for the class of verbs with incremental themes, there are two ways in which syntax encodes telicity: compositional telicity and predicate telicity. In languages with compositional telicity—including English, Dutch, and Finnish—the morphosyntax of the direct object affects the telicity of the verb phrase. In English and Dutch it is count versus mass noun syntax, that is, the presence or absence of articles in the noun phrase, that determines the telicity of the event description. In Finnish it is accusative versus partitive object case marking that determines telicity. In contrast, in languages with predicate telicity such as Russian and Polish, and also for English and Dutch constructions with particle verbs, the morphosyntax of the object does not matter. It is the verb and its morphology that are critical. Morphological aspect—the perfective/imperfective paradigm—not only determines grammatical aspect but also has effects on lexical aspect: perfective marking triggers telicity for verbs with an incremental theme. 3. LEARNING COMPOSITIONAL TELICITY Is the way one language marks telicity easier to learn than the way another does it? On the basis of comprehension studies from the acquisition literature on aspect, I argue that it is easier to learn telicity in languages with predicate telicity than in lan-

264

VAN HOUT

guages with compositional telicity. I now present results from Dutch, English, and Finnish comprehension studies on the acquisition of compositional telicity. 3.1. Telicity Acquisition in Dutch and English In a comprehension study, I tested Dutch and English children’s knowledge of telicity syntax, looking in particular at their aspectual interpretation of transitive and intransitive sentences (van Hout, 1998a, 1998b). Children as well as adults were presented with stories about characters involved in eating or drinking situations. In each story there were two characters: one who finished his food or his drink completely (completed situation) and another who ate or drank only part of it (incomplete situation). For each character three pictures accompanied the story: these showed the start of the event, the event in progress, and the final situation in which no more eating or drinking is going on. An example story is recounted in (8) and (9) and illustrated in Figures 12.1a and 12.1b. 8. Excerpt from the Dutch protocol: [incomplete situation] Hier is een witte muis. Hij heeft net een stuk kaas gevonden. Kijk, hier is hij aan het eten. Hij knabbelt er een beetje af, maar dit stuk is veel te groot voor hem. Hij laat nog wat over voor later. [completed situation] En hier is een rode muis. Hij heeft ook een stuk kaas gevonden. Kijk, hier is hij aan het eten. De rode muis vindt zijn kaasje erg lekker. Dat kan je wel zien ook: er blijft niets van over. 9. Excerpt from the English protocol: [incomplete situation] Here’s a white mouse. He just found a piece of cheese. Look, here he’s eating. He takes a couple of bites, but his cheese is too big for him for now. He leaves a piece for later. [completed situation] And here’s a red mouse. He also found a piece of cheese. Look, here he’s eating. The red mouse likes his cheese very much. You can see that here: his cheese is all gone.

After hearing a story, participants were asked one “yes/no” question about each character. Test questions were interspersed with three filler questions, one of them requiring a “yes” answer and two a “no”; the latter were included to discourage a potential “yes” bias. There were four kinds of questions, which were similar except for the form of the predicate; these are exemplified in (10) to (13):

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY

265

FIGURE 12.1. (a) Mouse story—completed situation. (b) Mouse story—incomplete situation.

10. Intransitives: a. Heeft de witte muis gegeten? Heeft de rode muis gegeten? b. Did the white mouse eat? Did the red mouse eat? 11. Transitives with a bare noun object: a. Heeft de witte muis kaas gegeten? Heeft de rode muis kaas gegeten? b. Did the white mouse eat cheese? Did the red mouse eat cheese? 12. Transitives with a full noun phrase object: a. Heeft de witte muis zijn kaasje gegeten? Heeft de rode muis zijn kaasje gegeten? b. Did the white mouse eat his cheese? Did the red mouse eat his cheese? 13. Transitives with the resultative particle op/up: a. Heeft de witte muis zijn kaasje opgegeten? Heeft de rode muis zijn kaasje opgegeten? b. Did the white mouse eat up his cheese? Did the red mouse eat up his cheese?

Participants were asked two questions of each kind, for a total of eight stories. Which situation type (completed or incomplete) was presented as the first substory and which as the second was counterbalanced across participants, as was which character was asked about first (the one in the completed or the incomplete situation). The different question types were randomized across the eight stories.

266

VAN HOUT

The direct object nouns were used bare (i.e., as a mass noun) or in a noun phrase with a possessive pronoun in the full noun phrase and particle verb conditions (e.g., coke/his coke, milk/his milk). In Dutch, the diminutive suffix -je was added in some cases, along with the possessive pronoun; this is an additional quantizer indicating a small portion of the denotation of the noun (e.g., ijs/zijn ijsje ‘ice cream/his (portion of) ice cream’, kaas/zijn kaasje ‘cheese/his (bit of) cheese’). A possessive pronoun by itself does not necessarily quantize the NP (his ice cream may refer to an unbounded amount of ice cream). The objects here always involved a specific and bounded quantity that was explicitly specified as such in the stories and pictures. For the purposes of this study then, the possessive pronoun can be taken as 11 quantizing the object. The “yes/no” questions were phrased in a perfective tense: the present perfect in Dutch and the simple past in English (for discussion of the semantics of these tenses, see Boogaart, 1999; Dowty, 1979; and Smith, 1991, among others). A perfective tense presents an event with its final boundary. Sentences with perfective tense differ in whether their final moment entails completion—for telic predicates—or termination—for atelic predicates. Only a perfective tense can bring out the difference between telic and atelic predicates. Had the questions been asked in an imperfective tense (such as the progressive in English, e.g., Was the red mouse eating his cheese?), the completion of the situation would not have mattered and the answers would be “yes” for both characters in the story, because both were involved in eating or drinking. Thus, telic predicates with a perfective tense are felicitous only for completed situations, not for incomplete ones. Atelic predicates are felicitous with both completed and incomplete situations. Forty-five Dutch children participated in the Dutch study, 15 in each of three age groups (3-, 4-, and 5-year-olds), and 15 adults. Forty-six American children participated in the English study, nineteen 3-year-olds, seventeen 4-year-olds, and eleven 5-year-olds, and 16 adults. All participants were native speakers of Dutch or English. The children were tested individually at their kindergarten or day care center; these sessions were recorded on audiotape. The adults were tested at the experimenter’s home or at the University of Pennsylvania.12 Telic predicates (with a perfective tense) are sensitive to completion versus lack of completion; atelic predicates are not. For the completed situations, participants can felicitously answer “yes” to all the questions regardless of the form of their predicate (Did the white mouse EAT/EAT CHEESE/EAT HIS CHEESE/EAT UP HIS CHEESE?); this is because a completed situation can be described with a telic or an atelic predicate. But for the incomplete situations, participants should reject questions with predicates that they interpret as telic (e.g., eat up his cheese), because the event was not “completed,” but only terminated. An atelic form—unlike a telic form—can be used felicitously even though the situation is not completed but only terminated. Acceptance of a question (a “yes” response) for an incomplete situation is thus taken to mean that the participant interprets its predicate atelically, and a

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY

267

negative answer (a “no” response) is a sign of a telic interpretation. Because only the responses to the incomplete situations are crucial in determining whether speakers interpret a particular predicate type as telic (“no” responses) or atelic (“yes” re13 sponses), I focus on these in reporting the results of the experiment. Figures 12.2 and 12.3 show the percentage of “yes” responses to the four question types about the incomplete situations for Dutch and English, respectively.

FIGURE 12.2. Dutch: Percent “yes” responses to the four types of questions about the incomplete situations (suggesting an atelic interpretation—lack of a completion entailment).

FIGURE 12.3. English: Percent “yes” responses to the four types of questions about the incomplete situation (suggesting an atelic interpretation—lack of a completion entailment).

268

VAN HOUT

Learners of Dutch and English of all ages treat questions with particle verbs differently from all the other question types, rejecting them significantly more often as descriptions of the incomplete situations. This shows that they correctly give these sentences a telic reading. But the children do not give systematically different treatments to questions with an intransitive verb, a transitive verb with a bare noun, and a transitive verb with a full NP object. Their high proportion of “yes” responses (i.e., atelic interpretations) to all these question types indicates that they have not yet picked up on the implications for telicity of the semantics of direct objects (no object or bare object—both atelic; full noun phrase object—in principle telic). Only when telicity is overtly marked on the verb by a particle do children reject the predicate as a description of an incomplete event. For adult speakers of Dutch and English, particle-verb questions also triggered (relatively) unambiguous telic readings (i.e., “no” responses). Op/up is a clear telicity marker in both languages, categorically so in Dutch and to a somewhat lesser extent in English. But unlike the children, adult speakers of both languages also distinguished among the remaining question types, giving a telic reading (“no” response) to transitive questions with a full NP object (Did the white mouse eat his cheese?) significantly more often than to intransitive questions (Did the white mouse eat?) or to transitive questions with a bare NP (Did the mouse eat cheese?). Transitive questions with a full NP object modified by a possessive pronoun were, however, by no means always given a telic reading (“no” response), and their preferred interpretation by adults was in fact different in the two languages: For Dutch speakers, the preferred reading was telic (only 20% “yes” responses), whereas for English speakers it was atelic (fully 75% “yes” responses)—a significant difference. Transitive questions with a full NP object can be interpreted telically in both languages (and are likely to be in Dutch), but in both languages they can also be interpreted atelically (and are likely to be in English), as if they were equivalent to partitives like Did the white mouse eat of his cheese? Children, in contrast, are much less aware that these sentences can be interpreted telically. I conclude that Dutch and English children as old as 5 do not know the syntax of compositional telicity. 3.2. Telicity Acquisition in Finnish Weist, Wysocka, and Lyytinen (1991) present a crosslinguistic study of the development of temporal systems in children learning English, Polish, and Finnish. Here I summarize the results from the Finnish children on an aspect comprehension task (those of the Polish children are discussed in section 4).14 Weist et al. used a sentence-picture matching task with two picture alternatives and two sentence alternatives. The pictures showed an ongoing situation versus a completed situation, for example, a girl drawing a flower versus a girl happily smiling next to an already-drawn flower. After describing the pictures in some detail, the experimenter read the two sentence alternatives and then asked: Which one

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY

269

15

shows…?, repeating one of the alternatives. The Finnish sentence alternatives differed in the accusative versus partitive case of the object. For example: 14. a. Tyttö girl

piirsi

kukka-a.

drew

flower-Part

‘A/The girl was drawing a/the flower/drew at a/the flower.’ b. Tyttö

piirsi

kuka-n.

girl

drew

flower-Acc

‘A/The girl drew a/the flower.’

Weist et al. tested 36 children—12 in each of three age groups (2;6, 4;6, and 6;5). The youngest children scored at chance (50% correct). The 4;6-year-olds scored significantly better than chance, but the scores of the 6;5-year-olds were not significantly better than chance, so one may conclude that even by age 6 Finnish children cannot reliably compute the telicity entailments associated with partitive versus ac16 cusative object case. It appears that Finnish children, just like their Dutch and English counterparts, find compositional telicity hard. 4. LEARNING PREDICATE TELICITY In this section I review results from three different aspect comprehension experiments reported in the literature: Weist et al. (1991) on Polish, and Stoll (1998) and Vinnitskaya and Wexler (2001) on Russian. Although the experimental setups in these studies differ, the results are strikingly similar: children as young as 2;6 years are able to compute telicity entailments associated with the perfective/imperfective paradigm. 4.1. Telicity Acquisition in Polish The Weist et al. (1991) study, discussed earlier for Finnish, also included Polish. The children—12 in each of five age groups (2;6, 3;6, etc., up to 6;6) —were presented with six pairs of pictures showing completed versus ongoing situations, along with a matching pair of sentences that differed in their morphological aspect. For example, for the flower-drawing pictures, they would hear the following pair of sentences: 15. a. Dziewczynka girl

rysowa»a

kwiatek.

drew (Imperf)

flower

‘A/The girl was drawing a/the flower/The girl drew at a/the flower.’ b. Dziewczynka girl

na-rysowa»a

kwiatek.

Perf-drew

flower

‘A/The girl drew a/the flower.’

270

VAN HOUT

Even the youngest children did well on this task: they scored about 76% correct, which was reliably different from chance (at a confidence level of p < .01 in a Kolmogorov–Smirnov two-sample test). This suggests that by age 2;6 Polish children have acquired the aspectual semantics of the perfective/imperfective paradigm. 4.2. Telicity Acquisition in Russian Also using a sentence-matching task, Vinnitskaya and Wexler (2001) carried out an aspect comprehension study with Russian children. Participants—12 in each of three age groups (3, 5, 6;6)—heard one sentence, with either perfective or imperfective aspect, and had to choose one picture out of three. Two pictures differed in whether what was otherwise the same action was completed or ongoing; the third was a distractor (depicting the wrong verb). For example, one set of pictures showed three girls: one reading a book, the second finished reading a book, with the book closed, and the third playing in the yard. For this picture set, participants were presented with an imperfective sentence: 16. Devo…ka girl

…itala

knigu.

read (Imperf)

book

‘A/The girl was reading a/the book.’ ‘A/The girl read at a/the book.’

Across picture sets, participants got four imperfective and four perfective verbs, all transitive. All the children performed well on this task, including the youngest: the 3-yearolds scored 75% correct for the imperfectives and 79% for the perfectives, which was significantly above chance. These results are similar to the Weist et al. findings for Polish. Stoll (1998) tested Russian children using a different experimental setup with pairs of short video stimuli. Videos constitute an excellent format for testing comprehension of temporal contrasts, because the temporal progress of events can be shown. With pictures, in contrast, temporal progress must be inferred, which may be much harder. Each of Stoll’s video pairs showed two puppets involved in a similar action. In video pairs depicting what Stoll termed “telic actions” (described with incremental theme verbs), one puppet completed her action, whereas the other was continuously engaged in her action.17 A typical trial first showed one of the puppets on one side of a split screen, for example, continuously reading a book. This side then went black and the other puppet appeared on the other side, in this case reading the book all the way through and closing it when she was finished. After a brief moment of blackness both members of the video pair were then shown simultaneously. At a given moment the video stopped, freezing an image of each puppet, and the experimenter asked a “who” question, for example:

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY 17. Kto who

pro-…ital

knigu?

Perf-read

book

271

‘Who read a book?’

The child’s task was to point to the matching video. All the verbs tested—14 in total—were morphologically perfective, so the adult choice would always be the completed event. One hundred children took part in the study, 20 in each of five age groups (2, 3, etc., up to 6). The youngest group of children did not perform significantly better than chance (58% correct), but from 3 years on they did well (75% correct or better).18 This outcome is exactly what Vinnitskaya and Wexler (2001) found. Taking these two studies together, especially in view of their different methodologies (picture selection versus video selection), one may conclude that 3-year-old Russian learners know the telic aspectual semantics associated with morphologically perfective transitive verbs with incremental themes. 4.3. Discussion of the Aspect Comprehension Studies The experiments summarized from the Slavic literature plus the Finnish one differ in various respects from the study of Dutch and English presented in section 3.1. To what extent do the Slavic and Finnish aspect studies test knowledge of telicity (lexical aspect) rather than knowledge of grammatical aspect? Can we indeed compare them all and draw conclusions about the acquisition of compositional telicity versus predicate telicity? The first difference concerns methodology. The “who/which” questions in the studies by Vinnatskaya and Wexler and by Stoll required the child to match one sentence to exactly one picture or movie. This approach can only reveal children’s preferences, because it is possible that children would also accept the sentence as a description of the other picture or movie. The same is true of the Weist et al. task. The pragmatics of wh-questions suggests that only one match is correct. But Stoll remarked that children sometimes chose both movies or said they could not choose; other children might also have responded in this way if they had been explicitly offered the option to do so. Asking “yes/no” questions about both story characters, as was done in the Dutch/English study, eliminates this problem and can reveal ambiguity (many subjects answered “yes” to the same question asked about a completed 19 story and its incomplete counterpart). This issue is crucial for interpreting perfectives in Russian and Polish, and accusative-marked objects in Finnish: are children able not only to accept these sentences as descriptions of completed situations, but also to reject them for ongoing situations? It is possible, then, that the Slavic and Finnish studies give an “inflated” picture of the child’s competence. Results might have been poorer if children had also been asked to judge whether the test sentence or question could be applied to the other picture or video of a pair.

272

VAN HOUT

Another important difference between the various studies was the nature of the contrast in the presentation of the situations: between a completed and an incomplete situation in the past in the Dutch/English study, and between a completed and a presently ongoing situation in the other studies. When Dutch or English speakers reject a telic predicate in the Dutch present perfect or English simple past as a description of an incomplete situation (e.g., heeft opgegeten, ate up), it must be because they know that the predicate is telic and that telicity entails completion. But with the completed/ongoing contrast, the notions of incompleteness and ongoingness are confounded. The choice of a completed situation as the scene described by a perfective verb (Polish, Russian) or a verb with an accusative object (Finnish) may signal an understanding of telicity entailments, but it is equally compatible with a belief that these forms are inappropriate for ongoing situations. Lack of ongoingness, as in the completed situations, does not distinguish between telic and atelic predicates, but rather between progressive and perfective aspect, a grammatical aspect distinction. So the contrast tested in the Slavic and Finnish experiments was not optimal for exploring knowledge of telicity entailments. The Slavic perfective and the Finnish accusative object require the event to culminate to completion. That is, there must be a lack of ongoingness and the theme participant must be fully involved (e.g., fully eaten, fully drawn); hence, the event is bounded in two respects (both in the verb and in the object). Because young Polish and Russian children differentiate the perfective and imperfective, reliably choosing the completed picture or video more often for a perfective, they apparently know this entailment of (double) boundedness, whereas Finnish children do not. Arguably, then, the experiments compared in this article are not exactly comparable because they used different methods and designs. Further comparative research should bring these more into line with each other.20 5. CONCLUSIONS A review of the aspect comprehension literature across some Germanic and Slavic languages and Finnish has brought a surprising result: young Polish and Russian children acquire knowledge of aspectual entailments much more quickly than their Dutch, English, and Finnish peers do. Recall that Polish and Russian have what I have called “predicate telicity”: the perfective/imperfective paradigm of the verb carries telicity entailments—in particular, the perfective form of transitives with incremental themes requires boundedness of both event and object participant. Learners of Polish and Russian as young as 2;6 or 3 years old are able to correctly differentiate situations for which a perfective is an appropriate description (completed situations) from those for which it is not (ongoing situations). Dutch, English, and Finnish, in contrast, have “compositional telicity”: the telicity of the verb phrase is computed on the basis of the joint properties of verb and object. The relevant morphosyntactic feature is the

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY

273

quantization of the object noun phrase, as marked by the presence/absence of an article in Dutch and English and the partitive/accusative object case paradigm in Finnish. Dutch and English also have some constructions with predicate telicity: particle verbs for which the particle establishes the telicity the verb phrase. By 3 years of age learners of Dutch and English recognize that particle verbs like eat up and drink up are telic, but even as old as 5 they still do not understand the marking on the object (presence or absence of determiner) as having implications for telicity. And Finnish children even as old as 6 years do not reliably select between completed and ongoing situations on the basis of the distinction between accusative and partitive case marking on the object, whereas adults do. The data show that it is easier to learn predicate telicity than compositional telicity. Why would this be so? At this point I can only speculate and raise questions. Is there an innate bias toward expecting aspectual distinctions to be marked on the verb rather than on the object—a bias that must be overcome in the acquisition of languages with compositional telicity? Bybee (1985) introduces the term “relevance” to refer to affinities that she finds in her typological studies between certain meanings and certain forms. Positing that “verbal inflections differ with respect to the extent to which they are relevant to the verb, that is, the extent to which their meanings directly affect the lexical content of the verb stem,” she argues that “[t]he more relevant a morphological category is to the verb, the more likely it is to occur in a synthetic or bound construction with the verb” (Bybee, 1985: 11–12). This principle of relevance has been used by Slobin (1985) as a potential explanation for the relative ease or difficulty of acquiring various constructions across languages. One of his examples is that it is difficult to learn to mark the gender and definiteness of the object on verbs, but easy to acquire tense/aspect marking on verbs; clearly, tense/aspect is more “relevant” to verbs than information about the object is. Bybee’s relevance principle can be directly applied to telicity encoding: telicity is relevant to the verb in that it specifies the temporal contour of the event introduced by the verb more precisely than is done by the verb’s inherent lexical aspect alone. Under relevance, then, it can be expected that children will have less difficulty learning predicate telicity than compositional telicity. On this approach, one can predict, for example, that quantization-marking of (object) noun phrases is acquired earlier when it occurs together with the object itself, as with articles in Dutch and English or cases in Finnish, than when it is marked on the verb, as with perfective marking in Slavic, because it is relevant to the object, not the verb. The question remains, however, what the status is in the learner’s grammar of the form–meaning affinities predicted by the principle of relevance. Are they (in some way logical) subset options that the learner first considers (as the Subset Principle would have it), or are they operating principles specifically geared to guide learners in their discovery of form–meaning relations (cf. Slobin, 1985)? Still another possibility is that a syntactic–semantic theory of telicity could be developed so that close and less close form–meaning affinities fall out from it in a principled way.

274

VAN HOUT

Another approach to explaining why predicate telicity is easier to learn than compositional telicity is to consider the grammatical operations involved. Can it be a matter of semantic complexity, with the composition of verb and prefix or particle less complex to compute than the composition of verb and object? Alternatively, is it a matter of morphosyntactic complexity, with predicate telicity less complex because it is a lexical property of the prefixed verb, and compositional telicity more complex because it requires an additional syntactic operation (e.g., telicity checking, as in van Hout’s 1996 theory, or movement to the Specifier of an appropriate Aspect phrase, as in Borer, 1994)? An explanation for relative ease of acquisition based on quantifying the complexity of constructions could be tested by formulating analogous predictions about development in other domains. To conclude, the surprising acquisition facts that came together in this chapter may inspire theoreticians as well as acquisitionists to further develop the theory of the syntax of telicity and examine its development in children. ACKNOWLEDGMENTS I thank Lila and Henry Gleitman, Bart Hollebrandse, Tom Roeper, and Laura Wagner for insightful and helpful discussion of the acquisition work presented here. I thank Olga Borik, Hana Filip, Patrycja Jablonska, Anna M»ynarczyk, Boóena Rozwadowska, and Maaike Schoorlemmer for discussing my questions about Slavic aspect with me, and Saara Jantunen for the Finnish data. I am very grateful to Melissa Bowerman for her careful reading and commenting on a first draft of this chapter. I have further benefited from questions and comments from many people on many occasions at which I have presented parts of this research, including the 1996 and 1997 BU conferences, the 1997 New Perspectives conference at the University of Massachusetts, the 1997 LSA workshop on Event Structure, the 1998 Max Planck workshop on the acquisition of argument structure, the 2001 Perspectives on Aspect conference at Utrecht University, the baby-lab meetings at IRCS, University of Pennsylvania, the acquisition lab meetings at UiL OTS at Utrecht University, and colloquiums at various universities. I acknowledge the support from NWO (Netherlands Organization for Scientific Research) for my project on the acquisition of aspect (number 300-75-025). NOTES 1

The term “lexical” aspect is unfortunate, as it suggests that aspect is determined by the verb’s lexical properties. It is well known in the aspectual literature, and the examples that follow illustrate this, that other elements in the verb phrase also affect lexical aspect class. In this article I follow common usage and use the term “lexical aspect” to refer to those aspectual properties of whole verb phrases that determine aspectual class. 2 Others have used different terms for the same property. Krifka (1992) calls it “mapping-to-objects,” which means that each part of the consumed or created object is

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY

275

mapped to a part of the temporal development. For Tenny (1994), who views the lexicon–syntax interface as one in which verbs assign aspectual roles rather than thematic ones, these verbs assign a so-called “measuring-out” role. Verkuyl (1993) characterizes these verbs as running an “event odometer” along their objects. All assign this effect to a lexical property of a class of transitive verbs. 3 A durative adverbial is not ungrammatical with telic predicates, but rather gives another reading: to perform a play for months implies an iteration of performing a play again and again. The temporal adverbials test must therefore be used to check single-event readings. A single event of performing a play can be done in one evening, but not for months. 4 Dutch works just like English in the examples above and below. 5 For the incremental theme verbs under consideration the implication goes: if telic, then transitive. If one-argument verbs are also taken into account, it becomes: if telic, then transitive (for two-argument verbs) or unaccusative (for one-argument verbs). See van Hout (1996, 2000a, 2004) for further discussion. 6 In addition to the lexical aspect function, the Finnish accusative/partitive distinction can also mark the count/mass property of a noun phrase: accusative case gives a count term and partitive case a mass term. Moreover, the case distinction may affect grammatical aspect and create a distinction between perfective and imperfective. See Kiparksy (1998). 7 Russian and Polish do have two object cases, accusative and genitive, but these play a role in negation rather than telicity: negated transitive sentences take a genitive object in Polish and optionally a genitive or accusative in Russian, whereas positive sentences take accusative objects in both languages. 8 The Slavic aspectual paradigm also encodes other lexical and grammatical aspects, including inchoative, iterative, habitual, and others. 9 The entailment does not hold in the opposite direction; that is, it is not the case that, if perfective, then necessarily telic. 10 These effects of Polish perfective aspect on the direct object are mirrored in particle verbs in English and Dutch, where the particle makes the predicate telic. With inherently telic drink up a mass term is impossible; only count terms can appear: Maria drank up *milk. Inherently telic verbs in English and Dutch thus involve predicate telicity: telic particles, like perfective prefixes in Russian and Polish, make the verbal predicate telic. 11 Further studies are needed to look at the aspectual effects of both articles (definite and indefinite) and measure phrases. A follow-up study on Dutch using objects with definite articles (het water ‘the water’) and measure phrases (een emmer water ‘a bucket of water’) shows that, for these question types, the answers did not differ from the condition with possessive pronouns. 12 I thank the following places for their hospitality in allowing me to test their children: pre-school St. Jan Baptist in Oerle, day care ‘t Sterretje in Veldhoven, and day care centers Bright Start, Discovery, Green/Byrne, Magic Years, and Rocking Horse in Philadelphia. I thank Inge Hoeks for helping me test the Dutch adults. 13 Both Dutch- and English-speaking participants of all ages overwhelmingly said “yes” to all four question types for completed situations. Only rarely did participants say “no” to questions about an incomplete situation.

276

VAN HOUT

14

I describe here the results from their so-called “Aspect-small” task in the second experiment, which varied object case. 15 It is unclear from the article whether the experimenters counterbalanced the order of the test sentences. 16 Weist et al. do not present the percentages, but judging from the graph showing mean percentage correct, they are as follows: 50% correct at 2;6 (chance), 81% at 4;6, and 73% at 6;5 years. The authors note that the result for the oldest children does not exceed what would be expected by chance, but the result for the 4;6-year-olds does. 17 Stoll also tested other aspectual classes (delimitatives, ingressives, and semelfactives), but these are not considered further here. 18 Stoll does not report tests of significance, but I assume that 75% correct was indeed significantly different from chance (50%), and 58% was not. 19 This response pattern cannot be ascribed to a general “yes” bias. Subjects regularly answered “no” to questions about the incomplete situation, as the low bars in Figures 12.2 and 12.3 show, as well as to the filler questions requiring a “no.” 20 In a new experiment, I presented Polish subjects with a contrast between completed and incomplete situations as well as contrast pairs with completed versus ongoing and incomplete versus ongoing situations. When given a perfective verb, young children indeed prefer the completed situation and reject ongoing and incomplete situations, even the 2;6-year-olds (van Hout, 2005), showing full aspectual knowledge of perfective aspect as carrying telicity and perfectivity. Their behavior on imperfectives is not adult-like, however. REFERENCES Boogaart, R. (1999). Aspect and temporal ordering. PhD dissertation, Free University of Amsterdam. Borer, H. (1994). The projection of arguments. In E. Benedicto & J. Runner (Eds.), Functional projections, University of Massachusetts Occasional Papers 17 (pp. 19–47). Amherst, MA: GSLA. Borik, O. (2002). Aspect and reference time. PhD dissertation, Utrecht University, The Netherlands. Bybee, J. (1985). Morphology: A study of the relation between meaning and form. Amsterdam: John Benjamins. Comrie, B. (1976). Aspect. Cambridge: Cambridge University Press. de Hoop, H. (1992). Case configuration and noun phrase interpretation. PhD dissertation, University of Groningen, The Netherlands. Published in 1994. New York: Garland Publishing. de Swart, H. (1998). Aspect shift and coercion. Natural Languages and Linguistic Theory, 16, 347–385. de Swart, H., & Verkuyl, H. (1999). Tense and aspect in sentence and discourse. Reader for the 11th ESSLLI Summer School. Unpublished manuscript, Utrecht University, Utrecht, The Netherlands. Dowty, D. R. (1979). Word meaning and Montague grammar. Dordrecht, The Netherlands: D. Reidel.

12.

ACQUIRING TELICITY CROSSLINGUISTICALLY

277

Filip, H. (1993). Aspect, situation types and nominal reference. PhD dissertation, University of California, Berkeley. Published in 1999 as Aspect, eventuality types and noun phrase semantics. New York: Garland Publishing. Fodor, J. A., Bever, T. G., & Garrett, M. F. (1974). The psychology of language: An introduction to psycholinguistics and generative grammar. New York: McGraw-Hill. Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press. Kiparsky, P. (1998). Partitive case and aspect. In M. Butt & W. Geuder (Eds.), The projection of arguments (pp. 265–308). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Krifka, M. (1986). Nominalreferenz und Zeitkonstitution: Zur Semantik von Massentermen, Pluraltermen und Aspektklassen. PhD dissertation, Munich University, Germany. Published in 1989. Munich, Germany: Fink. Krifka, M. (1992). Thematic relations as links between nominal reference and temporal constitution. In I. Sag & A. Szabolcsi (Eds.), Lexical matters (pp. 29–53). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. M»ynarczyk, A. (1998). Aspektualiteit in het Nederlands en in het Pools. Unpublished MA thesis, Warsaw University. Ramchand, G. (1997). Aspect and predication: The semantics of argument structure. Oxford: Clarendon Press. Schmitt, C. (1996). Aspect and the syntax of noun phrases. PhD dissertation, University of Maryland. Schoorlemmer, M. (1995). Participial passive and aspect in Russian. PhD dissertation, Utrecht University, The Netherlands. Smith, C. (1991). The parameter of aspect. Dordrecht, The Netherlands: Kluwer. Slobin, D. I. (1982). Universal and particular in the acquisition of language. In E. Wanner & L. Gleitman (Eds.), Language acquisition: The state of the art (pp. 128–170). Cambridge: Cambridge University Press. Slobin, D. I. (1985). Crosslinguistic evidence for the language-making capacity. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition, Volume 2: Theoretical issues (pp. 1157–1249). Hillsdale, NJ: Lawrence Erlbaum Associates. Slobin, D. I., & Bever, T. G. (1982). Children use canonical sentence schemas: A crosslinguistic study of word order and inflections. Cognition, 12, 229–265. Stoll, S. (1998). Acquisition of Russian aspect. First Language, 18, 351–377. Tenny, C. L. (1987). Grammaticalizing aspect and affectedness. PhD dissertation, MIT, Cambridge, MA. Tenny, C. L. (1994). Aspectual roles and the syntax–semantics interface. Dordrecht, The Netherlands: Kluwer. van Hout, A. (1996). Event semantics of verb frame alternations: A case study of Dutch and its acquisition. PhD dissertation, Tilburg University, The Netherlands. Published in 1998. New York: Garland Publishing. van Hout, A. (1998a). On the role of direct objects and particles in learning telicity in Dutch and English. In A. Greenhill, M. Hughes, H. Littlefield, & H. Walsh (Eds.), Proceedings of the 22nd Annual Boston University Conference on Language Development (pp. 397–408). Somerville, MA: Cascadilla Press. van Hout, A. (1998b). On learning the role of direct objects for telicity in Dutch and English. In B. Hollebrandse (Ed.), New Perspectives on Language Acquisition, University of Massachusetts Occasional Papers 22 (pp. 87–104). Amherst, MA: GLSA. van Hout, A. (2000a). Event semantics in the lexicon–syntax interface: Verb frame alternations in Dutch and their acquisition. In C. Tenny & J. Pustejovsky (Eds.), Events as grammatical objects (pp. 239–282). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications.

278

VAN HOUT

van Hout, A. (2000b). Projection based on event structure. In P. Coopmans, M. Everaert, & J. Grimshaw (Eds.), Lexical specification and insertion (pp. 397–422). Amsterdam: John Benjamins. van Hout, A. (2004). Unaccusativity as telicity checking. In A. Alexiadou, E. Anagnostopoulou, & M. Everaert (Eds.), The unaccusativity puzzle: Studies in the syntax– lexicon interface (pp. 60–83). Oxford: Oxford University Press. van Hout, A. (2005). Imperfect imperfectives: On the acquisition of aspect in Polish. In P. Kempchinsky & R. Slabakova (Eds.), Aspectual inquiries (pp. 317–343). Berlin: Springer. Vendler, Z. (1957). Verbs and times. Philosophical Review, 66, 143–160. Verkuyl, H. (1972). On the compositional nature of the aspects. Dordrecht, The Netherlands: D. Reidel. Verkuyl, H. (1993). A theory of aspectuality. Cambridge: Cambridge University Press. Vinnitskaya, I., & Wexler, K. (2001). The role of pragmatics in the development of Russian aspect. First Language, 21, 143–186. Weist, R. M., Wysocka, H., & Lyytinen, P. (1991). A cross-linguistic perspective on the development of temporal systems. Journal of Child Language, 18, 67–92.

CHAPTER 13

The Acquisition of the English Causative Alternation Melissa Bowerman Max Planck Institute for Psycholinguistics

William Croft University of New Mexico

1. INTRODUCTION Languages are riddled with partial regularities—patterns that are productive, but not completely so. Such patterns create a challenge for theories of language acquisition: how can children discover the underlying regularities and use them creatively, without at the same time overshooting the bounds of what fluent speakers consider normal and acceptable? Explaining this process would be straightforward if fluent speakers corrected children when they overgeneralized patterns (and if children paid attention). But it is widely accepted that explicit correction is rare and unsystematic (Baker, 1979; Bowerman, 1988; Braine, 1971; Brown & Hanlon, 1970; Pinker, 1989). Some researchers urge that even if there is little overt correction, there are interaction patterns in adult–child discourse that provide indirect negative evidence—for example, adult reformulations of children’s erroneous utterances (e.g., Bohannon & Stanowicz, 1988; Demetras, Post, & Snow, 1986; Hirsh-Pasek, Treiman, & Schneiderman, 1984). But it is controversial whether this kind of evidence is widely available, whether it has the logical power to correct the child, and whether children are in fact even sensitive to it (see Marcus, 1993; Morgan & Travis, 1989; Pinker, 1989: 9ff., for critiques). This state of affairs—often termed the No Negative Evidence problem—has led many researchers to conclude that models of grammar learning cannot depend on learners’ receiving information about what is not a possible sentence. Children must be able to arrive at the adult state on the basis of positive evidence alone—hearing how other speakers talk about things. 279

280

BOWERMAN AND CROFT

The challenge of explaining how this takes place has attracted much attention. For researchers of a nativist bent, at least part of the solution is sought in inborn grammatical knowledge and mechanisms that block undesirable generalizations from the outset, or enable children to identify and reject incorrect grammatical hypotheses without recourse to negative evidence (e.g., Baker, 1979; Gropen, Pinker, Hollander, & Goldberg, 1991; Pinker, 1989; Randall, 1990). Learning-minded researchers, in contrast, urge that correct generalizations can be built up, and overgeneralizations pruned back where necessary, through general cognitive mechanisms, for example, the effects of type and token frequency on schema induction and on the activation strength and entrenchment of forms (Brooks & Tomasello, 1999; Brooks, Tomasello, Dodson, & Lewis, 1999; Croft, 2001; Croft & Cruse, 2004; Goldberg, 1993, 1995, 2003; MacWhinney, 1987; Tomasello, 2003). Children’s overgeneralizations clearly constitute a fertile testing ground for the clash between alternative theories of language acquisition. Much of the debate has revolved around errors of a particular genre: overgeneralizations of argument structure alternations. An argument structure alternation is a pattern in which a set of verbs systematically appears in two different syntactic frames (Hale & Keyser, 1987; Levin, 1985, 1993; Pinker 1989). English has several important alternations that give rise to errors in children’s speech. Three of these—the causative/inchoative alternation (henceforth simply “causative alternation”), dative alternation, and locative alternation—are illustrated in Table 13.1. These alternations are all productive in adult English, and can be applied to novel verbs. But not every verb can undergo a given alternation, even if it seems semantically and syntactically similar to a verb that can. This is evident from examples like (1c-i), (2c-g), and (3d-g) in Table 13.1 from learners of English: these utterances are readily understandable, but they seem strange to fluent adult speakers of English. In this chapter we evaluate two proposals for how children arrive at an adult understanding of which argument structure frames verbs can appear in: Pinker’s (1989) tightly structured nativist model and a looser constructivist scenario based on a cluster of usage-based learning mechanisms. We do this by testing the predictions of the models against a large corpus of spontaneous argument structure errors collected over many years from two learners of English, Bowerman’s daughters, C and E (cf. examples in Table 13.1). These children’s language development was followed closely, through audio-taping and diary notes, from about 1 to 3 years of age, with continuing attention to certain forms up through the teenage years. C’s and E’s argument structure errors, as presented in Bowerman (1974, 1982a,1982b, 1988, 1996), have constituted a jumping-off point for much of the discussion in the literature of the No Negative Evidence problem (e.g., Pinker, 1989). Here, we focus on a particular error type—the causativization of an intransitive verb or adjective (as in (1c–g) of Table 13.1). This error type was selected because it was by far the most frequent and persistent in the children’s speech.

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

281

TABLE 13.1 Three Argument Structure Alternations of English, and Their Overgeneralization in Children’s Speech (1) CAUSATIVE/INCHOATIVE ALTERNATION

a. (i) The chocolate melted. (ii) Mom melted the chocolate. b. (i) The ball rolled down the hill. (ii) Linda rolled the ball down the hill. Children’s errors (c-g: noncausative to causative; h-i: causative to noncausative): c. C 3;5

How come you had a little troubling going it? (M [Mother] couldn’t start car.)

d. C 7;5

But he disappeared the green one and he disappeared the blue one! (Watching magican do tricks with scarves on TV.)

e. C 12;3 Salt clings it together. (As C mixes playdough.) f. E 4;3

Can I glow him? (Wants to play with a monster toy that glows after being held under a light.)

g. C 5;0

OK. If you want it to die. E’s gonna die it. She’s gonna make it die. (C’s sister E is about to touch a moth.)

h. C 2;11 Bert knocked down. (After sees Bert topple over on TV.) i. C 4;5

But the parts might lose. (Concerned about taking a game to school with her.)

(2) LOCATIVE ALTERNATION

a. (i) Harry loaded books onto the cart. (ii) Harry loaded the cart with books. b. (i) The cook sprinkled powdered sugar onto the cake. (ii) The cook sprinkled the cake with powdered sugar. Children’s errors (c-d: require ‘with’; e-g: require locative preposition): c. E 5;0

Can I fill some salt into the bear? (=bear-shaped salt shaker.)

d. E 7;11 I’m going to decorate them on the edge. (Putting a row of thumbtacks along edge of new bulletin board.) e. E 4;5

I’m gonna cover a screen over me. (Child is pretending to do a magic trick with a blanket.)

f. E 2;11

I poured you. [M: you poured me?] Yeah, with water. (Pretending, waving an empty cup near M.)

g. E 4;11 I don’t want it [=toast] because I spilled it of orange juice. (After spills orange juice on her toast.) (3) DATIVE ALTERNATION

a. (i) Sarah gave some books to the orphanage. (ii) Sarah gave the orphanage some books. b. (i) I told the whole story to my parents. (ii) I told my parents the whole story. c. (i) Linda baked a cake for John. (ii) Linda baked John a cake.

(continued)

282

BOWERMAN AND CROFT TABLE 13.1

(continued)

Children’s errors: d. C 2;6

Don’t say me that or you’ll make me cry.

e. L 7;8

Shall I whisper you something?

f. C 3;4

Button me the rest. (Most of her pyjama snaps are closed, wants M to fasten the remaining ones.)

g. M 5+

Choose me the ones that I can have.

From Bowerman (1974, 1982a,b, 1988, unpublished records). Age in years; months.

2. PINKER: A NATIVIST APPROACH TO LEARNING ARGUMENT STRUCTURE ALTERNATIONS In an early exploration of the learning puzzle posed by argument structure alternations, Pinker (1984) assumed that errors like those in Table 13.1 mean that English-learning children’s initial rules for argument structure alternations are too general; the rules must somehow be cut back. Linguists had noted that argument structure alternations are often subject to specifiable semantic and sometimes morphosyntactic conditions on the verbs to which they apply. For example, to undergo the dative alternation and enter into the double-object construction, an English verb must have a dative argument that refers to a “prospective possessor” of the theme argument (Green, 1974; Mazurkewich & White, 1984; Oehrle, 1976). This means that Mary baked John a cake/poured John a drink/faxed John a message are acceptable, because Mary intends for the cake, drink, and message to end up in John’s possession, but *Jim washed Susan the dishes/opened Susan the door are strange because the dishes and door do not change hands as a result of the action. To correct an argument structure alternation rule that is initially too general, proposed Pinker (1984), a child must, over time, annotate the rule with the appropriate conditions on the verbs to which it can apply. When annotation is complete, errors will cease. This hypothesis captured some important constraints on argument structure alternations, but further work showed that it could not stand as an adequate account of acquisition (Bowerman, 1988; Pinker, 1989). One critical flaw was that for each alternation, there were verbs that seemed to satisfy all the proposed criteria but still did not undergo the alternation. For example, “saying” and “whispering” seem to be perfectly good ways of getting something (information) into someone’s possession, and yet—unlike semantically similar verbs such as tell—say and whisper do not undergo the dative alternation (cf. errors (3d, e) in Table 13.1). How will the child learn this? The No Negative Evidence problem reappears in full force. Further mysteries were why an argument structure rule should have these seemingly arbitrary annotations in the first place, and why children should bother to identify them, given that their initial rule, being overly general, can already parse and interpret any

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

283

input it receives. To solve these problems, Pinker (1989) proposed a new and more intricate acquisition theory. 2.1. Argument Structure Alternations as Lexical Rules for Changing Verb Meaning In the model just discussed, rules for changing a verb’s argument structure were seen as having the purely syntactic effect of rearranging the verb’s arguments (and, in the case of the causative, also introducing an argument). Following work by Levin and Rappaport (Levin, 1985; Levin & Rappaport, 1986), Pinker now proposed that rules for argument structure alternations are, instead, lexical rules that create a new verb from an old one by changing the verb’s semantic structure. For example, the rule for dative alternation takes a predicate that means roughly “X cause Y to go to Z” (as in give1 a book to John) and converts it into a predicate that means “X cause Z to have Y” (give2 John a book) (Pinker, 1989: 82). In this new formulation, the syntactic rearrangement of the arguments does not have to be simply stipulated; it can fall out naturally from the meaning of the verb, through linking rules that map arguments in certain positions in the (compositional) semantic representation of the verb to particular syntactic roles. One such linking rule, according to Pinker (1989; Gropen et al., 1991), states that an entity that is specified to be causally affected is mapped to the grammatical role of direct object. For give1 it is the theme argument (i.e., the object given) that is specified to be causally affected (it is the second argument of CAUSE in “X cause Y to go to Z”). For give2 , in contrast, it is the dative argument (“X cause Z to have Y”). Because rules for argument structure alternations are, on this new account, basically semantic operations rather than purely syntactic ones, it is not surprising that they are sensitive to the semantic properties of verbs. In particular, the meaning of a verb must be compatible with the semantic change that is brought about by the rule. It is understandable, for instance, why Wash Susan the dishes and Open Susan the door are ungrammatical—Susan is not caused to “have” the dishes or door as a result of the actions. The rule for the causative alternation, in Pinker’s account, takes a predicate that specifies a “change” (an event of acting or moving in some way) and converts it into a predicate that means “by acting on, cause to change (in the specified way).” (Or vice versa: the rule is bidirectional and can run in either direction.) Linking rules specify that the first argument of CAUSE, the agent, is mapped to the subject role and the second argument, the affected entity, to the object role. Just as for the dative alternation, a number of verbs are immediately rendered outside the scope of the causative rule because they are incompatible with the basic semantic operation the rule brings about. For example, stative intransitives such as be and ache cannot be causativized because they do not specify a change. As presented so far, Pinker’s revised model of the acquisition of English argument structure alternations has the advantage over his earlier model that it provides a prin-

284

BOWERMAN AND CROFT

cipled explanation of major semantic restrictions on which verbs can participate in an alternation. Learning these restrictions would be part and parcel of learning the lexical rules for an alternation in the first place, not conditions to be arbitrarily tacked onto a more general rule. But there is still a critical weakness: just as in the earlier model, there are verbs that satisfy these restrictions and yet do not alternate. For example, why can’t disappear undergo the causative alternation (as in (1d) of Table 13.1), given that it satisfies the requirement that the verb specify a change? 2.2. Broad-Range Rules and Narrow-Range Rules To tackle this problem, Pinker (1989) proposed analyzing each rule for alternation into two levels: a broad-range rule and one or more narrow-range rules, which are semantically more specific versions of the broad-range rule. The broad-range rule provides the necessary conditions for a verb to alternate, but does not specify whether or not it actually does alternate. The narrow-range rules, in contrast, provide the sufficient conditions. 2.2.1. Broad-Range Rules A broad-range rule relates two “thematic cores,” which are conflations of semantic elements that define a kind of possible verb meaning. Such rules are more formal specifications of the kind of information already described in section 2.1 for the dative and causative alternations; they capture what all the verbs that undergo the alternation have in common. The broad range rules for these two alternations are shown in (1) and (2) (the arrows indicate that the rules are bidirectional): 1. Dative alternation: a. X CAUSE [Y GO TO Z ] (e.g., Mary gave a book to John)

£

b. X CAUSE [Z HAVE Y [by means of CAUS ing [Y GO TO Z]]] (e.g., Mary gave John a book) 2. Causative alternation: a. Y event: ACT/GO (e.g., The ball rolls)

£

b. X ACT on Y, thereby CAUS ing Y ACT/GO (e.g., John rolled the ball)

The broad-range rule for an alternation insures that no verb can participate in the alternation unless it can be represented in terms of both thematic cores, and the rule specifies what the new verb would mean if the rule were applied. This rule provides an initial semantic filter that excludes a large number of verbs from the alternation. For example, the specification in the thematic core of (2a) captures the generalization that the caused situation must be an event (i.e., a predicate built around ACT or GO); put differently, the causative alternation cannot be applied to in-

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

285

transitive verbs with BE or HAVE in their semantic representation (e.g., be, exist, stay, wait, have) (Pinker, 1989: 223). The thematic core of (2b), the transitive causative, has in its main clause “X ACT on Y.” This core is responsible for the reading of “direct” or “unmediated” causation associated with lexical causatives: “direct,” proposes Pinker (1989), is the default interpretation of “ACT on”. 2.2.2. Narrow-Range Rules Some verbs meet the specifications of a broad-range rule, but still do not alternate. For example, the intransitive English verbs go, fall, and disappear are , as the broad-range rule for the causative alternation requires, but they do not have a morphologically identical transitive, causative counterpart. Drawing on work by Laughren, Levin, and Rappaport (1986), Pinker proposed that each broad-range rule is paired with one or more narrow-range rules: from the candidate alternators admitted by the broad-range rule, the narrow range rules provide a more delicate filter by picking out semantically coherent subclasses of verbs that do in fact alternate. For the causative alternation, there are narrow-range rules that pick out two important classes of verbs that alternate, as shown in (3) (Pinker, 1989: 130; Levin & Rappaport Hovav, 1995: 93). Classes of verbs that lack a narrow-range rule and so do not alternate are shown in (4). 3. Classes with narrow-range rules for the causative alternation (alternators) a. Verbs of EXTERNALLY-CAUSED CHANGE OF PHYSICAL STATE: melt, open, break, shrink, shatter… b. Verbs of MOTION TAKING PLACE IN A PARTICULAR MANNER : slide, skid, float, roll, bounce… 4. Classes without a narrow-range rule for the causative alternation (nonalternators) a. Verbs of MOTION IN A LEXICALLY SPECIFIED DIRECTION : go, come, rise, fall, exit, ascend, leave, arrive… b. Verbs of COMING INTO OR GOING OUT OF EXISTENCE: die, appear, disappear, expire, vanish … c. Most verbs of EMISSION OF LIGHTS, SOUNDS, SUBSTANCES, AND SMELLS: glow, glisten, sparkle, blaze, shriek, buzz, bubble, leak, ooze, smell… d. Verbs of INTERNALLY-CAUSED STATE CHANGE: bloom, blossom, decay, blush, wax, wane… e. Verbs of VOLITIONALLY OR INTERNALLY-CAUSED ACTIONS: jump, walk, talk, climb, drink, sing…) (apparent exceptions like gallop/walk/jump a horse belong to a different alternation, according to Levin & Rappaport Hovav, 1995) f. Verbs of PSYCHOLOGICAL ACTIVITY: think, hope, wish, hesitate, refrain from… g. Most verbs of EMOTIONAL EXPRESSION: smile, cry, laugh, frown, blink…

286

BOWERMAN AND CROFT

For any particular alternation, it is not possible to predict, a priori, which verb classes sanctioned by the broad-range rule have an associated narrow-range rule. For example, it would be possible for English to causativize verbs of “motion in a lexically specified direction” or verbs of “coming into or going out of existence”; English simply lacks narrow-range rules for these classes. But Pinker (1989: 133) points outs that several of the noncausativizing verb classes shown earlier—particularly (4c–g)—probably do not causativize for a principled reason: because they specify internally-caused events and so resist the “directness” interpretation required by the broad-range rule (see Levin & Rappaport Hovav, 1995, on the distinction between internally- and externally-caused events). In other words, these verbs are not causativizable for reasons that are central to the semantic structure of the causative rule itself, not simply because they lack an associated narrow-range rule. But whether a verb specifies an internally-caused event is often not obvious, and in ambiguous cases—for example, especially classes (4c) and (4d)—different languages may take different stances (Pinker, 1989: 302). 2.3. Learning In Pinker’s (1989) account of language acquisition, children approach the task with inborn knowledge of the primitive semantic elements out of which verb meanings are composed (e.g., CAUSE, GO, BE, ACT), as well as of the linking rules associated with them. This knowledge insures that if a child represents the meaning of a verb correctly, she will know how to link its arguments. But even if learners can formulate their broad-range alternation rules correctly, they must still determine which narrow-range rules are associated with them. If there is not some water-tight procedure for identifying these rules accurately from the very beginning, the No Negative Evidence problem reasserts itself in full force: the child’s rules will be too general, and it is unclear how she can discover the exceptions to them. To solve this problem, Pinker proposes that children develop the broad-range rule and the narrow-range rules for a particular alternation in tandem. This insures that there is never a time when a broad-range rule operates unconstrained by one or more narrow-range rules. Children formulate the broad-range rule through a topdown process of abstraction over verbs that have been observed to display the alternation. Simultaneously, they formulate narrow-range rules through a conservative bottom-up process in which the privilege of alternating generalizes, but only out to the boundary of each semantic class for which an instance of an alternating verb has been encountered. (What constitutes a relevant semantic class is highly constrained by innate mechanisms; see Pinker, 1989: 273–280, on this key feature of his model.) The crucial claim of Pinker’s model is, then, that the child’s rules are correctly constrained from the start, so there is no need to explain how retreat can take place in the absence of negative evidence. But if the child’s grammar develops so accurately, why do errors like those in Table 13.1 occur? Pinker (1989: 292ff., 350) offers two explanations:

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

287

5. Causes of children’s argument structure errors: a.

One-shot innovations. By hypothesis, speakers of all ages sometimes use broad-range rules creatively on-line to produce forms that are not licensed by any of the narrow-range rules associated with them. This may occur in children more often than in adults for several reasons; for example, children may innovate to extend their communicative resources when they don’t yet know a more appropriate verb, or cannot access it at the moment. One-shot innovations are not actually licensed by the speaker’s grammar, so they don’t require any specific unlearning. (See Braine & Brooks, 1995, for a similar proposal.) b. Erroneous verb meanings. Some argument structure errors arise, Pinker hypothesizes, because children have associated a verb with an incorrect meaning. If a child’s semantic representation for a verb is wrong, the appropriate application of linking rules to this representation might result in errors from the adult point of view. For example, suppose a child associates the verb fill (roughly “cause X [e.g., a cup] to become full [of Y, e.g., water]”) with a meaning more similar to that of pour (“cause Y to move in a certain way”). In this case the “affected object” linking rule will assign Y, as the affected object, to the role of direct object, resulting in errors like (2c) in Table 13.1 (“fill Y into X”). Repeatedly observing the situations to which adults apply the verb—for example, hearing fill for events where there is “becoming full” but no “pouring”—will lead the child to reanalyze the verb’s meaning (e.g., which argument is taken to be the “affected” one), and errors will automatically cease.

Crucially, both of these explanations for errors are compatible with Pinker’s claim that the child’s rules for argument structure alternations are basically correct from the beginning. 2.4. Evaluating Pinker’s Model Pinker’s model is explicit, coherent, and based on a well-developed theory of lexico-semantic structure. There is much to admire about it. But is its account of the acquisition of argument structure alternations correct? 2.4.1. Innate Linking The success of the theory depends on the accuracy of many interacting assumptions, some of them highly controversial. For example, the theory requires knowledge of linking rules to be innate. This is because correct linking must follow automatically from meaning: as long as children have represented the meaning of a verb correctly, they must be able to link its arguments correctly. There is by now a good deal of literature debating whether children in fact show evidence of innate knowledge of linking, and it is fair to say that there is as yet little consensus on this. (See section 2.1 of chapter 1, this volume, for an overview of this literature, with references.)

288

BOWERMAN AND CROFT

2.4.2. Narrow-Range Rules The theory also requires children to be sensitive to syntactically relevant semantic subclasses of verbs from the beginning, because they must generate narrow-range rules for these from the ground up, never generalizing too far. Initial explicit tests of the hypothesis proved negative: there was no evidence for this sensitivity (Braine & Brooks, 1995; Pye & Loeb, 1995; see also Ingham, 1992). In a more recent experiment, Brooks and Tomasello (1999) did find evidence for sensitivity to two semantic classes of verbs relevant for the causative alternation. These researchers taught children (age 2;6, 4;6, and 6-7) a novel verb in either an intransitive or a transitive, causative frame, and then tried to induce them to use it in the other, as yet unattested frame. Children were more willing to use the verb in the unattested frame if its apparent meaning was something like “spin” than if it was something like “ascend.” Recall that “motion taking place in a particular manner” (such as spinning) is a narrow-range class for the causative alternation (cf. (3b) above), whereas “motion in a lexically specified direction” (such as ascending) is not (4a). Strikingly, though, this result was found only in children over 4;6 years old. Children of 2;6 years used both kinds of novel verbs in the unattested frame with equal probability. This outcome is incompatible with Pinker’s claim that children’s rules for argument structure alternations are appropriately constrained from the beginning. It suggests instead—as Brooks and Tomasello indeed argue—that the needed semantic constraints are discovered only gradually over time. In the present study, we find remarkably little evidence that children are constrained by the narrow-range semantic categories that are relevant for the causative alternation, either early or late in development. We come back to this issue presently. 2.4.3. Accounting for Errors An aspect of Pinker’s theory that has so far received little attention is whether children’s argument structure errors, such as those shown in Table 13.1, can really be “explained away,” as the theory requires, either as one-shot innovations licensed by the broad-range rule or as casualties of incorrect verb meanings. How well does this claim hold up against the novel lexical causatives produced by our two language learners, C and E? Proliferation of Errors. Novel causatives followed a very similar course in the children’s speech: they appeared around age 2, flourished—especially for C— between about 3 and 5, and then continued on at a lower level until about age 12, after which they essentially disappeared (total number of recorded errors: C 225 tokens, 79 types; E 92 tokens, 54 types). The children made many errors with verbs from all the noncausativizable narrow-range semantic classes listed earlier in (4). Their errors are summarized in the Appendix, broken down by semantic class.1

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

289

The very quantity, variety, and persistence of these novel causatives over a long period of time seems at odds with Pinker’s theory; such a profusion of errors does not really square with the view that the children’s grammars were perfectly adult-like except for one-shot innovations and erroneous verb meanings. The presence of multiple errors in the “externally-caused change of physical state” class is also troubling. This class is supposed to causativize (see (3a) earlier), so how can the child determine that intransitive verbs like overflow do not? (cf. You’re gonna overflow the spoon with medicine, C 6;7. See also Braine & Brooks, 1995, for a more general discussion of negative exceptions to Pinker’s causativizable subclasses, i.e., verbs that do not causativize even though they supposedly fall into a 2 causativizable narrow-range class.) Incorrect Verb Meanings? Pinker hints that at least some of children’s novel causatives are caused by incorrect verb meanings (1989: 325), but he makes no concrete suggestions about this; most of his evidence for this source of errors revolves around a different alternation, the locative (see section 1.1 of chapter 1, this volume). It is indeed not clear what could be wrong with the meaning of most of the words shown in the Appendix that would make them susceptible to causativization. Especially resistant to this interpretation are novel lexical causatives created from highly frequent verbs like come, go, disappear, and stay. These errors persisted over a period of many years even though the children used their intransitive base forms in an entirely adult-like way. One-shot innovations? This puts the burden of explanation for novel causatives on Pinker’s “one-shot innovation” hypothesis, which posits that many of children’s errors reflect the creative online use of the broad-range rule, perhaps especially under communicative pressure when the child doesn’t know or can’t remember a better verb. The persistence of many of the errors argues against this explanation (as Pinker, 1989: 325, also recognizes). For example, C causativized stay (e.g., stay the door open) at least 43 times between the ages of 2;4 and 10;4, long after she knew—and usually used—the more appropriate verbs keep and leave. She causativized go at least 28 times between the ages of 2;8 and 7;11, long after she knew verbs like send and take. Often the children did not even begin to causativize a verb erroneously until well after an appropriate counterpart for it was already well established in their speech (e.g., causative come vs. bring). Even so, the novel form was sometimes powerful enough to temporarily almost supplant the correct form (Bowerman, 1974). Also problematic for the “one-shot innovation” hypothesis is that many of the children’s errors fall outside the scope of the broad-range rule that is supposed to constrain them. Recall that Pinker’s strategy for solving the learnability problem associated with argument structure alternations is to insure that the child never generalizes too broadly to begin with, and so has nothing to repair later. With this goal

290

BOWERMAN AND CROFT

in mind, Pinker formulates the broad-range rule for the causative alternation (cf. (2), shown earlier) as restrictively as the facts of adult English will allow: First, the caused event must be (i.e., the verb must have ACT or GO in its semantic representation); second, the causing event must involve an ACT whereby an agent impinges on a patient; and third, this act must bring about the caused event directly. C and E violated all three constraints repeatedly, as illustrated in Table 13.2. They causativized verbs that lack ACT or GO in their representation (Table 13.2, examples a–f); they causativized when the causing situation cannot be conceptualized as an “act” on a patient by any stretch of the imagination, not even a TABLE 13.2 Violations of Pinker’s (1989) Broad-Range Rule for Causativization

1. Caused event: Counter to the broad-range rule, in many errors the caused situation is not a event (i.e., the underlying predicate does not contain GO or ACT), but instead is static, e.g., with BE, STAY, or HAVE: a. C 5;5

I meant to be it like this. (=have it be. Showing with her hand how she had intended an unsuccessful styrofoam Christmas tree to turn out.)

b. C 4;5

(C making drawings to bind as a book; upset with a poor picture.) This one is yukky! Be it for a picture. (=let/have be [only] a picture) (M: Hmm?) C: Be it for a picture, I don’t need a book.

c. E 7;11 I was used to turning it [TV] on a channel and being it on a channel. (= keeping it, letting it continue to be …) d. C 2;11 Maybe they had a cold and the cold stayed them awake. (=kept.) e. E 6;7

Now I’m going to have you a lesson. (=give.)

f. E 5;3

This is aching my legs. (As climbs stairs.)

2. Causing event: sometimes there’s no “act”, not even a metaphorical “impingement” of an actor on a patient: e.g., (a)-(d) above, and: g. C 3;1

Is this to climb her up? (=enable her to climb up. C looking at picture of a hippo at the bottom of a ramp leading into a truck, pointing to the ramp.)

3. Violations of the “directness” constraint on lexical causatives (according to Pinker, directness is an automatic consequence of the fact that in the broad-range rule, the causee is a patient): h. E 3;3

Will you climb me up there and hold me? (Wants help climbing a pole.)

i. C 10;5 (C doing a trick; explains that the magician must first make everyone feel a marble hidden under a scarf:) First you have it, and you feel it to everybody. (=make/ have everybody feel it.) j. C 4;3

Andrea, I want to watch you this book! (Trying to get a friend to look at a book she is holding.)

k. C 3;3

(C has drawn a puzzle.) M: Do you think Daddy can guess that one? C: I’m gonna guess it to him! (=have him guess it. Runs off to find F.)

l. E 3;2

Everybody makes me cry. (F: I didn’t make you cry.) Yes, you did, you just cried me.

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

291

purely metaphorical impingement (examples a–d, f, g); and they causativized when the causation was clearly indirect; that is, when a physically or psychologically active animate causee mediated between the agent’s act and the resulting event (ex3 amples h–l) (see also Bowerman, 1982a: 46–47). Causatives with truly animate causees, such as (h)–(l) in Table 13.2, were relatively infrequent; most errors with verbs of volitional or semivolitional events, like climb, walk, swim, eat, and cry, involved dolls and other toys that could not really carry out the action independently. Noting this, Pinker (1989: 302ff.) argues that this shows that children are sensitive to the “directness” constraint of the broad-range rule for causativization. If they were not, he suggests, they should produce many more errors with volitional or semivolitional verbs than they do. After all, “opportunities for producing such errors are rampant: parents forcing, threatening, inducing, preventing, or allowing children to do things, and children enticing or badgering their parents or siblings to do things, have to be among the most common events involving some notion of causation that children are likely to think about or comment on” (Pinker, 1989: 302). But Pinker’s argument is valid only if children actually do talk frequently about the causation of volitional or semivolitional events. If they seldom do, even using periphrastic causatives (e.g., She made me sing), then the relatively low numbers of novel lexical causatives (e.g., She sang me) of this semantic category would reflect only the low number of opportunities to make such errors—that is, children’s relative conversational neglect of the causation of (semi)volitional actions. It would tell us nothing about children’s sensitivity to “directness” in lexical causatives. To explore this issue, we calculated, for three semantic classes of noncausativizable verbs, the total number of opportunities to make an error: that is, the sum obtained by adding together the number of novel lexical causatives (e.g., You just cried me) and the number of periphrastic causatives with verbs of the same semantic class (e.g., Everyone makes me cry—cf. Table 13.2, example l). The classes were: (a) verbs of VOLITIONAL AND SEMIVOLITIONAL EVENTS, a composite of verbs of “volitionally or internally-caused actions,” “psychological activity,” and “emotional expression,” for example, crawl, guess, giggle, cf. class 7 in the Appendix (we included in the calculation only utterances referring to events with a truly animate, active causee, i.e., not a doll or other inanimate); (b) verbs of MOTION IN A TABLE 13.3 Proportion of Novel Lexical Causatives out of All Causatives (Novel Lexical Plus Periphrastic) Belonging to Three Semantic Classes in C’s and E’s Speech C

E

1. VOLITIONAL AND SEMIVOLITIONAL EVENTS

70%

(14/20)

55%

(6/11)

2. MOTION IN A LEXICALLY SPECIFIED DIRECTION

74%

(45/61)

76%

(26/34)

3. COMING INTO/GOING OUT OF EXISTENCE

58%

(14/24)

63%

(5/8)

292

BOWERMAN AND CROFT

LEXICALLY SPECIFIED DIRECTION, such as go, fall, rise—class 2 in the Appendix; and (c) verbs of COMING INTO OR GOING OUT OF EXISTENCE, such as disappear,

die—class 3 in the Appendix. If Pinker’s argument is correct, the proportion of novel lexical causatives to all opportunities to produce a novel lexical causatives should be significantly lower for verbs of VOLITIONAL AND SEMIVOLITIONAL EVENTS, which seriously violate Pinker’s “directness” constraint, than for verbs of MOTION IN A LEXICALLY SPECIFIED DIRECTION and verbs of COMING INTO OR GOING OUT OF EXISTENCE, which 4 do not. The proportions are shown in Table 13.3. This table shows that the children talked relatively infrequently about the causation of volitional and semi-volitional actions, but when they did, they used novel lexical causatives no less often (C) or only slightly less often (E) than when they talked about events that do not violate “directness.” Differences among the three proportions were not significant (one-way analysis of variance [ANOVA] for the child C, F(2, 102) = .096, p = .39; for the child E, F(2, 50) = 1.05; p = .36). Consistent with this finding, children in an elicited production study (Pye & Loeb, 1995) were just as willing to causativize English volitional action verbs as state-change verbs and verbs of motion in a lexically specified direction. Contrary to Pinker, then, children’s rule for the causative alternation is by no means restricted to events involving “direct” causation. To summarize, Pinker’s explanation cannot account adequately for the error data from C and E. The children causativized prolifically for many years across a broad range of verbs (see Appendix), respecting neither the distinction between causativizable and noncausativizable narrow-range verb classes nor between verbs that fall inside or outside of the scope of Pinker’s broad-range rule for causativization. For learners, causativizing an intransitive predicate seems to require little more than that the predicate describe a situation that can be conceptualized as being “caused” (Bowerman, 1974, 1982a; see also Gergely & Bever, 1986, for the same conclusion based on Bowerman’s data). But if this is true, then explaining how and why children eventually stop producing novel causatives does, after all, require— counter to Pinker’s (1989) nativist model and in accord with usage-based assumptions—explaining how they retreat from a causativizing operation that is overly general. 3. A USAGE-BASED SOLUTION TO THE ACQUISITION OF THE CAUSATIVE ALTERNATION In the last decade there has been a surge of interest in constructivist, usage-based models of language and language acquisition. These explain the representation of language structures not by reference to highly abstract, perhaps innate grammatical constructs and principles, but by invoking properties of the use of utterances in communication (e.g., type and token frequency of word forms and constructions, competition among forms), in interaction with the mental processes involved in

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

293

representing such properties (e.g., activation, schema formation, entrenchment, decay) (Barlow & Kemmer, 2001; Bybee & Hopper, 2001; Croft & Cruse, 2004; Goldberg, 1995; Langacker, 1987, chapter 10; MacWhinney, 1987; Regier, 1996; Tomasello, 1998/2002, 2003). This framework is attractive to researchers who find it more plausible that grammar is acquired through general cognitive mechanisms than through innate knowledge of language-specific categories and principles. 3.1. Usage-Based Mechanisms for Grammar Induction Within this general framework, several mechanisms have been singled out as critic a l to ex p la in in g w hy ch ild r e n s to p m a k in g arg u m e n t s tr u c tu r e overgeneralizations: preemption, the induction of semantic categories, and the entrenchment of verbs in particular syntactic frames. 3.1.1. Preemption For some of children’s argument structure errors, adult speech provides a conventional verb that expresses the same meaning as the child’s form. For example, the child’s causative use of die in (1g) of Table 13.1 (E’s gonna die it [a moth]) is perfectly matched by the adult word kill (E’s gonna kill it). The relationship between kill and causativized die is loosely analogous to the relationship between irregular inflectional forms and their regularized counterparts in child speech, for example, ran and runned, feet and foots. Following this analogy, we will say for convenience that kill is “suppletive” for causativized die, just as ran is suppletive for runned, although for reasons mentioned by Pinker (1989: 293) this label is not quite accurate; we come back to this in section 4. Other “suppletive” causatives include bring for come (e.g., I came it closer so it won’t fall—pulling bowl on counter toward herself), keep for stay (Mommy, can you stay this open?—having trouble with refrigerator door), drop for fall (I’m just gonna fall this on her—dropping piece of paper on her sister), and remind for remember (Will you please remember me what I came in for?) (Bowerman, 1982a). In virtually every theory of language acquisition, it is assumed that the consistent clash between a child’s error and the conventional adult form for this meaning will eventually bring the child into line with adult speech; that is, the adult form comes to preempt the child’s erroneous form (e.g., Clark, 1987; MacWhinney, 1987; Pinker, 1984; Pye & Loeb, 1995). Pinker (1989:293–294) also assigns an important role to lexical preemption, arguing that once forms like kill and bring have been strengthened enough, there will be no need for the child to make one-shot innovations (e.g., causativized die and come) to plug the gaps associated with their absence. Direct preemption of one word by another cannot be the whole solution to the problem, because by no means all of children’s erroneous lexical causatives are matched by a conventional lexical causative in adult speech. For instance, there is no lexical causative in English that means what the child’s causativized form of dis-

294

BOWERMAN AND CROFT

appear means (cf. (1d) in Table 13.1). For verbs like disappear, it has been suggested that the child’s erroneous lexical causative form is preempted by the corresponding periphrastic causative, for example, make disappear (e.g., Clark, 1987; MacWhinney, 1987). The logical case for extending the notion of preemption to periphrastic causatives is weaker than for forms like kill, because lexical and periphrastic causatives are, as constructions, systematically associated with different meanings (Bowerman, 1988). But a child might notice when adults do not use the verb the child has predicted, especially if they use a more marked construction, for example, make disappear instead of causativized disappear (Goldberg, 1995; see also Regier & Gahl, 2004). Brooks and Tomasello (1999) found some evidence for this process in children older than 4;6: When children were taught a novel verb in an intransitive frame (e.g., it’s tamming, see section 2.4.2), they were less likely to use it as a lexical causative (he’s tamming it) if they had been exposed to a periphrastic causative alternative (he’s making it tam) than if they had not. 3.1.2. Induction of the Relevant Semantic Subclasses of Verbs In his initial proposal for how argument structure alternations are learned, Pinker (1984) suggested that children home in only gradually on the semantic categories of verbs relevant for a particular alternation. In his later model, Pinker (1989) discarded this hypothesis as both implausible and unfeasible, opting instead to capture some of the relevant semantic constraints as inherent properties of the (broad-range) rule itself, while postulating that others fall out automatically from how the rule generalizes—that is, only to other verbs in the same semantic class as verbs that have been observed to alternate (see section 2.3). In more recent research, the notion that semantic schemas can be learned gradually through induction has staged a strong comeback (e.g., Brooks & Tomasello, 1999; Goldberg, 1993, 1995). The hypothesis is supported by the success of recent connectionist simulations of category induction (see Regier, 1996; Schütze, 1994; Ping & MacWhinney, 1996, for studies relevant to word meaning and to verb syntax and morphology). It is also consistent with growing interest in construction grammar (e.g., Croft, 2001; Goldberg, 1995, 2003; Tomasello, 2003) and networkstyle theoretical approaches to morphology (Bybee 1985)—frameworks that stress input-driven learning. 3.1.3. Entrenchment The idea behind “entrenchment” in the domain of argument structure is straightforward: repeated experience with a verb that is always heard in the same syntactic frame (e.g., as an intransitive) strengthens the association between verb and frame to the point where the correct frame consistently wins out over the incorrect frame generated by the child’s too-broad alternation schema (Braine, 1971; Braine & Brooks, 1995; MacWhinney, 1987). Entrenchment and preemption often go together; for ex-

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

295

ample, every encounter with a periphrastic causative like make disappear simultaneously both exemplifies the verb disappear in an intransitive frame (yet again) and offers an alternative to causativized disappear. There is some experimental evidence for entrenchment in the domain of learning argument structure alternations. In the face of adult questions aimed at eliciting overgeneralizations of fixed-transitivity verbs, young learners of English were less likely to overgeneralize early-learned (hence presumably more entrenched) verbs than later-learned verbs; for example, they were less likely to produce I comed it than I arrived it (Brooks et al., 1999). 3.2. A USAGE-BASED MODEL FOR ACQUISITION OF THE CAUSATIVE ALTERNATION 3.2.1. The Model Let us draw on the mechanisms just discussed to construct and test aspects of a straightforward usage-based scenario for how children master the causative alternation. The model runs as follows (we give the prediction first, and then, where needed, the reasoning behind the prediction): Step 1. First, individual verbs are learned with (a subset of) their correct argument structures (transitive, intransitive, or both). (This step is documented in Bowerman, 1974, 1982a, and Tomasello, 1992, 2003.) Step 2. Next, the lexical causative is (over)generalized across a wide range of forms and semantic classes. (The child has observed a high enough type frequency of low enough token frequency forms that alternate to merit building a schema for the alternation. This schema—which varies in strength [i.e., productivity] across children [Maratsos et al., 1987]—is broader than Pinker’s [1989] broad-range rule for the causative alternation, because it also generates lexical causatives for indirect causation [as in Table 13.2].) Step 3. Errors abate or cease with verbs that have high-frequency lexical causative counterparts (e.g., kill for causativized die). (Frequency in the input strengthens the entrenchment of these forms at the expense of their child-generated competitors, resulting in preemption. The removal of the preempted forms from the abstract schema for causativization also “bleeds” (weakens) the schema’s overall strength.) Step 4. Semantic subclasses of causativizable verbs begin to develop (i.e., Pinker’s [1989] narrow-range classes), and fewer and fewer errors occur outside these classes. (The input has begun to more densely populate narrowly semantically specified areas of semantic space. Within those areas, lower-level subschemas become increasingly entrenched, and this—like preemption—bleeds

296

BOWERMAN AND CROFT

the more abstract schema.) Somewhere around this time, less frequent suppletive causatives also become entrenched, for example, remind comes to replace causative remember. Step 5. The last errors to fade out are causativizations of noncausativizable predicates that are in the right semantic ballpark and have no suppletive counterparts (e.g., disappear, small). (These should be the last to go because the main mechanisms working against them are the overall weakening of the abstract schema [steps 3 and 4] and the strengthening of the association between the verb and its intransitive frame through repeated exposure [entrenchment]. Preemption by periphrastic causatives [make disappear] may also contribute to the demise of errors with these forms, but this type of preemption should not be as powerful as preemption by a suppletive causative counterpart [e.g., kill for causativized die, as in Step 2], because the competition between error and candidate preemptor is less direct.) 3.2.2. Testing the Model Is it indeed true—as virtually everyone has supposed—that errors fade out earlier for verbs with suppletive causative counterparts than for verbs without them (Step 3)? And do errors abate earlier for verbs that are semantically distant from the core classes of causativizable verbs than for those that are semantically closer (Step 4)? We tested these two predictions against our corpora of novel lexical causatives, collected longitudinally from C and E over a period of more than a decade. Suppletion. To analyze the role of suppletion, we divided each year of the child’s life between age 2 and 12 years into three 4-month periods, and calculated for each period the number of tokens of novel causatives formed from predicates (verbs and adjectives) that do, versus do not, have straightforward suppletive counterparts.5 The frequencies of errors with verbs of these two kinds are shown in Figures 13.1 and 13.2. If the existence of a suppletive lexical causative works preemptively against a child’s tendency to erroneously causativize an intransitive verb, the line representing errors with predicates that have suppletive counterparts should decline more rapidly than the line representing predicates that do not. This was roughly true for E (although in fact she simply made fewer errors overall on verbs with suppletives), but not at all for C: for this child, forms with and without suppletive counterparts declined in parallel. Semantic Classes. To examine the role of semantic class, we plotted for each child the frequency of novel causatives of various classes during each of several successive time periods (Figures 13.3 and 13.4). A point on the x axis (time line) such as “3;0” means errors produced between ages 2;6 and 3;6. The errors are assigned to five different semantic classes, as follows (see also legend in Figure 13.4):

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

297

FIGURE 13.1. Frequency (in tokens) over time of novel lexical causatives with and without suppletive counterparts in C’s speech.

FIGURE 13.2. Frequency (in tokens) over time of novel lexical causatives with and without suppletive counterparts in E’s speech.

• The first three bars at each time period represent predicate classes (verbs and adjectives) that are semantically close to the core causativizable verb classes. The first bar in fact represents idiosyncratically noncausativizable members of the two core causativizable classes: EXTERNALLY-CAUSED CHANGE OF PHYSICAL STATE (e.g., overflow, bigger) and MOTION TAKING PLACE IN A PARTICULAR MANNER (e.g., slip [in the sense of ‘make someone slip’, not ‘slip your shoes on’]) (collapsed together in the Appendix as class 1, where

298

BOWERMAN AND CROFT

FIGURE 13.3. Frequency in tokens over time of novel lexical causatives in different semantic classes in C’s speech.

FIGURE 13.4. Frequency (in tokens) over time of novel lexical causatives in different semantic classes in E’s speech.

the reader can see which predicates are counted). The second and third bars represent verbs that are not causativizable, but are similar to core causativizable verbs in that they are unaccusative (Levin & Rappaport Hovav, 1995) and involve external causation: verbs of MOTION IN A LEXICALLY SPECIFIED DIRECTION (e.g., go, fall, rise—class 2 in the Appendix) and verbs of COM-

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

299

ING INTO OR GOING OUT OF EXISTENCE and of EXISTING OR BEING IN A PLACE/STATE (e.g., disappear, die—class 3 in the Appendix, and be, stay—

class 4, collapsed together and shown as a single bar). • The last two bars at each time period represent verbs that are semantically distant from the core classes because they involve internal causation (see Levin & Rappaport Hovav, 1995; Pinker, 1989): verbs of EMISSION of light, fluid, sound, etc. (e.g., glow, sweat, squeak) and of INTERNALLY-CAUSED STATE CHANGE (e.g., bloom, grow [feet], stick [=make adhere](classes 5 and 6 in the Appendix, collapsed together) and verbs of VOLITIONAL ACTION, EMOTIONAL EXPRESSION, and PSYCHOLOGICAL EVENTS (e.g., ride, laugh, remember) (class 7). (For this last bar, the frequencies encompass utterances with both truly animate causees and “pretend” animate causees like dolls and stuffed animals; see section 2.4.3.) If the induction of semantic categories is important in children’s retreat from causative overgeneralizations, the last two bars (semantically distant from core causativizable classes) should decline faster than the first three bars (semantically close). But this pattern is not found. Verbs of EMISSION and INTERNALLY-CAUSED STATE CHANGE (fourth bar), never very frequent to begin with, do tend to abate early. But errors with verbs of VOLITIONAL ACTION, PSYCHOLOGICAL EVENTS, and EMOTIONAL EXPRESSION (fifth bar)—internally-caused events that violate the presumed semantic constraints most egregiously—hold their own remarkably well over time against the first three classes, and fade out at the same time, about age 12. In this data set, then, there is no evidence for the hypothesized role of semantic class induction in the decline and disappearance of causative errors. 4. DISCUSSION In this study we have focused on the causative alternation to evaluate two proposals for how children master argument structure alternations: Pinker’s nativist proposal and our own usage-based proposal. In our longitudinal spontaneous speech data from two learners of English, there is remarkably little support for either proposal. The predictions of Pinker’s model were violated repeatedly: the children respected neither the broad-range rule nor the narrow-range rules hypothesized for the causative, and their errors cannot easily be dismissed as one-shot innovations or due to faulty verb meanings. Both Pinker and constructivist theorists posit an important role for lexical preemption, with proposals often extending this mechanism to less precise competitors to children’s errors, such as periphrastic causatives. But the effect of lexical preemption is visible only very mildly in the data from one of our two children (E, Figure 13.2), and not at all in the data from the other (C, Figure 13.1). Finally, both Pinker and constructivists stress the role of semantically defined subclasses of predicates, although their predictions differ on when these effects

300

BOWERMAN AND CROFT

should appear: for Pinker, the relevant subclasses should constrain children’s generalizations from the very beginning, whereas for constructivists the classes are induced gradually over time. Under either scenario, there is little evidence in our data for sensitivity to semantic classes. In particular, the children causativized verbs expressing animate, internally-caused events (severe violators of semantic constraints on the causative alternation) just as robustly as unaccusative verbs expressing externally-caused events (far less severe violators). Errors of both types continued over a period of many years, declined in parallel, and faded out entirely at about the same time. Why do these widely invoked mechanisms play so little role in our data? With respect to preemption, we can think of two potential explanations. First, hearing the adult counterpart to a child’s causative overgeneralization (e.g., kill for causative die) might act not only to weaken the child’s form by repeatedly displaying an alternative way to express the same meaning, but also, ironically, to strengthen it by reinforcing the semantic niche it occupies. Thus, whenever the child understands an instance of kill in the input to mean what she herself would mean when she uses die causatively, she is reminded that the verb die indeed has a transitive lexical causative counterpart. If she remembers this, but forgets the specific form kill, she may be more likely to use die causatively. Overgeneralizations that lack suppletive counterparts, such as causative disappear, are not weakened by lexical preemption, but neither are they strengthened by evidence for the existence of a lexical causative with the same meaning. If the two hypothesized influences of lexical preemption—one eroding the tendency to use a verb causatively and the other promoting it—are approximately in balance, the net effect would be little overall difference in the rate at which children causativize verbs that are, versus are not, matched by adult lexical counterparts. A second factor that could detract from the effectiveness of lexical preemption in the domain of causatives is that few pairings between a child’s erroneous form and a competing adult form constitute perfect one-to-one matches from the semantic point of view. (This is one reason why it is not really accurate to speak of “suppletion” in this domain.) In some cases, a child’s causative error has several possible adult counterparts, each with a different nuance. For example, causativized stay corresponds sometimes to keep (“Mommy, can you stay this [a door] open?”) and sometimes to leave ( “[she] won’t stay things where I want them to be” [angry at meddling sister]). Causativized fall corresponds to both drop (“I’m just gonna fall this on her”) and knock (“you fell me down”). Causativized go corresponds to take (“go me to the bathroom”), put (“go it over there”), send (“Do you have anything else you’d like to go to China?”), and a variety of manner verbs (see note 5). In other cases several different child errors may correspond to a single adult form. For example, give is the most natural rendering of causative uses of both have (“will you have me a lesson?”) and take (“we took him a bath yesterday”; cf. “take a bath”).

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

301

Given this complex, many-to-many semantic mapping, it cannot be easy for children to work out which competing adult form is the one needed on a particular occasion. Using the intransitive form causatively is safe and accurate, because the resulting transitive verb will convey exactly the same information as the intransitive, plus “cause,” and nothing more. Turning now to semantic class, how can we explain our failure to find that children become sensitive, at least over time, to a verb’s semantic class membership, such that that their errors are increasingly restricted to verbs with the right kinds of meanings? This mechanism, although widely presupposed to be important in the recovery from argument structure errors, has received little empirical investigation. The best evidence for it comes from Brooks and Tomasello’s (1999) novel-verb experiment (see section 2.4.2). These authors found that after age 4;6 children were more likely to causativize a new verb that had been modeled only intransitively if it seemed to refer to an event of spinning (manner of motion: a causativizable class) than if it referred to an event of upward motion (motion in a lexically specified direction: noncausativizable). The authors take this as evidence for the gradual induction of the relevant semantic classes, but the findings are limited (only two verbs) and other interpretations are equally plausible. One major concern is that there was no test of whether the children understood the meanings of the novel verbs as intended. The authors infer that they did, because when the children talked about the actions, they often referred to them with real English verbs of the right semantic class, for example, spin or swing (manner of motion) versus go or come (motion in a lexically specified direction). But this defense introduces its own source of doubt: to the extent that the children equated the novel verbs with real verbs of English, their tendency to use them causatively or not may have been influenced not, as intended, by the novel verbs’ abstract semantic class membership, but by the syntax, already at least partially learned, of these specific real verbs. Finally, there is a complete confounding in the experiment, as the authors also recognize, between semantic class and directness of causation. For the “manner of motion” event the agent pushed on an object hanging on a rope, making it spin (direct), whereas for the “motion in a lexically specified direction” event the agent did not touch the patient, but pulled on a rope attached to a container it was in, thereby causing it to move up a ramp (indirect). Because using a verb causatively is often possible and preferred for events of direct causation, but impossible or dispreferred for events of indirect causation (McCawley, 1978; Pinker, 1989; see also section 2.2.1), children’s greater willingness to causativize the “spin”-type verb than the “go up”-type verb may have been influenced by sensitivity to this constructional distinction rather than by the semantic-class membership of the verbs.

302

BOWERMAN AND CROFT

In sum, experimental evidence is weak that children’s recovery from causative overgeneralizations has anything to do with semantic class induction. To our knowledge, our study is the first to test this hypothesized recovery mechanism against a longitudinal corpus, and our failure to find support for the hypothesis in children’s real-life spontaneous speech is sobering. Of course, it does not mean that speakers never identify the implicit semantic categories associated with verbs that can be causativized, but it does suggest that recovery from causative overgeneralization can and does proceed without this mechanism. When lexical preemption and semantic class induction fall by the wayside, the main mechanism we are left with is entrenchment: repeatedly hearing verbs like fall, disappear, and go only in intransitive syntactic frames, until the association between verb and frame becomes so strong that it consistently prevails in the child’s production. This mechanism played a relatively modest role in our proposed usage-based model of the acquisition of the causative alternation, serving primarily to clean up stragglers left over after preemption and semantic category induction have done their job (see Step 5 earlier). But our findings suggest that it should be treated with new respect (see also Braine & Brooks, 1995, and Brooks et al., 1999): after all, it may turn out to be the most powerful force available to counteract children’s causative overgeneralizations. ACKNOWLEDGMENTS We are grateful to Cynthia Fisher and Adele Goldberg for incisive comments in their role as discussants. NOTES 1

The children also sometimes intransitivized transitive causative verbs, as is shown in 1h–i of Table 13.1 (see Bowerman, 1982a, for discussion). However, there were far fewer of these errors than of lexical causatives derived from intransitive verbs or adjectives (see also Brooks & Tomasello, 1999, who could elicit far fewer of them in an experimental setting). 2 Most of the errors in this category are derived from adjectives rather than intransitive verbs (see Appendix). Since adjectives are , they do not directly qualify for causativization under the broad-range rule (though they can sometimes be formed from deadjectival intransitives, as in The clothes dried/Betty dried the clothes or The milk warmed slowly/Mom warmed the milk slowly. (See Levin & Rappaport Hovav, 1995: 95–96, for more on the causativization of adjectives.) In any event, Pinker gives no account of how children determine which adjectives can be used to express a caused state change and which cannot. 3 Examples e and i–k in Table 13.2 illustrate still another way in which the children often violated the broad-range rule for the causative: the causativized verb is transitive, and already has an agent or experiencer subject argument (see Bowerman, 1982a).

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

303

4

Novel lexical causatives, being errors, were more likely to be noticed and noted down than periphrasic causatives. This means that the proportion of novel lexical causatives to all causatives of each semantic class is probably higher in this corpus than in the children’s “real life” speech. But it is the relative value of the proportions in the three verb classes that is of interest, not the absolute value, and there is no reason to think that the sampling bias in favor of novel lexical causatives over periphrastic causatives affects the three classes differentially. 5 Causativized forms considered to have a suppletive counterpart included: die, dead (kill), come (bring, take [e.g., for “come me over there”]), stay (keep, leave), fall (drop, knock down), go (take, put, send), eat (feed), full (fill), remember (remind), learn (teach), higher, rise, go up (raise), go down (lower), round (rotate, turn), have (get, take [a bath/nap], give), be (put, make, keep), hot (heat), happy up (cheer up), broken (break), sharp (sharpen), flat (flatten), straight (straighten), tight (tighten), stable (stabilize) (see Bowerman, 1982a, Table 1, for errors with these predicates). Excluded from the calculation are causative uses of go and come where adults would use a manner-of-motion or state-change verb (e.g., “you go [=push] it in” [of a chair at the table]; “go [=pull] it up to the cloth” [of a diaper around the rubber ankles of a doll with cloth torso]; “go [=turn] on the bathtub”. Here the adult form adds so much information that is missing from the child’s simple go that it seems inappropriate to speak even loosely of “suppletion”. APPENDIX: VERBS AND ADJECTIVES USED BY C AND E AS NOVEL LEXICAL CAUSATIVES 1. EXTERNALLY-CAUSED STATE CHANGE/ MANNER OF MOTION

C (37 errors, age 2;0–10;3) full (6), flat, dirty, stuck [= make clogged], unstuck [= make unclogged] (2), sharp, straight, unstraight, stable, round (5), yellow, stick [= make stuck, jammed], fasten [= make go fast] (2), bigger, smaller, smallen, largen, longen, sour, colder, separate (adjective pronunciation), face, overflow (2), slip [= make someone slip] (2) E (11 errors, age 2;3–7;8) tight, untight, broken, full (2), round (2), bumpy, hot, smallen, largen 2. MOTION IN A LEXICALLY SPECIFIED DIRECTION

C (45 errors, age 2;0–9;8) go (28), come (7), fall (5), rise, cross (3), higher E (26 errors, age 1;10–7;8) go (12), come (4), fall (7), cross (2), higher 3. COMING INTO OR GOING OUT OF EXISTENCE

C (13 errors, age 2;8–12;4) peek out, spell [make letters on a spelling toy spell “X”], die (2), disappear (6), vanish (2), lose turn E (6 errors, age 3;7–11;11) spell [cf. above], dead, disappear (2), subside (2)

304

BOWERMAN AND CROFT

4. EXISTING, BEING IN A PLACE OR STATE

C (59 errors, age 2;1–11;3) be (9), have (5), stay (43), take too long, lie around E (8 errors, age 3;7–11;7) be (2), stay (3), have, wait, lie around 5. EMISSION

C (10 errors, age 3;0–6;7) bleed, sweat (3), sing [of music box] (2), squeak, squeaky, whistle (2) E (9 errors, age 2;11–10;2) bleed (2), water [eyes], sing [of musical instruments] (2), talk [of music box], glow, bubble, leak 6. INTERNALLY-CAUSED STATE CHANGE OR SITUATION

(cf. Levin & Rappaport, 1995: 90ff.) C (5 errors, age 3;6–12;3) bloom (2), grow [feet], cling together, soak in E (1 error, age 3;8) stick [= make adhere] 7. VOLITIONAL (AGENTIVE) ACTION, EMOTIONAL EXPRESSION, AND PSYCHOLOGICAL EVENT

C (57 errors, age 2;3–11;3) climb, crawl, jump (8), skate, ride (3), walk, drink (2), eat (3), guess, laugh, learn, play [= make act a part], remember (4), watch, feel, touch (2), turn a somersault (3), do a trick, take a bath, take little bites, take a ride (3), take a quiet time, take a walk, get [= cause to receive] (2), lie down (3), sit (3), itch, feel better (4) E (32 errors, age 1;11–10;11) ride, swim, climb, stagger, cry (3), drink, giggle, talk (4), walk, watch, take a ride, take a walk (2), lag, bow down, sit down, perform, remember, recognize, learn, itch, ache (2), sore, happy, comfy

REFERENCES Baker, C. L. (1979). Syntactic theory and the projection problem. Linguistic Inquiry, 10, 533–581. Barlow, M., & Kemmer, S. (Eds.). (2001). Usage-based models of language. Stanford, CA: Center for Research in Language Development (CSLI) Publications. Bohannon, J. N., & Stanowicz, L. (1988). The issue of negative evidence: Adult responses to children’s language errors. Developmental Psychology, 24, 684–689. Bowerman, M. (1974). Learning the structure of causative verbs: A study in the relationship of cognitive, semantic, and syntactic development. Papers and Reports on Child Language Development, 8, 142–178. Bowerman, M. (1978). Systematizing semantic knowledge: Changes over time in the child’s organization of word meaning. Child Development, 49, 977–987. Bowerman, M. (1982a). Evaluating competing linguistic models with language acquisition data: Implications of developmental errors with causative verbs. Quaderni di Semantica, 3, 5–66. Bowerman, M. (1982b). Reorganizational processes in lexical and syntactic development. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art (pp. 319–346). Cambridge: Cambridge University Press.

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

305

Bowerman, M. (1988). The “no negative evidence” problem: How do children avoid constructing an overly general grammar? In J. A. Hawkins (Ed.), Explaining language universals (pp. 73–101). Oxford: Basil Blackwell. Bowerman, M. (1996). Argument structure and learnability: Is a solution in sight? In J. Johnson, M. J. Juge, & J. L. Moxley (Eds.), Proceedings of the 22nd Annual Meeting of the Berkeley Linguistics Society (pp. 454–468). Berkeley, CA: Berkeley Linguistics Society, University of California. Braine, M. D. S. (1971). On two types of models of the internalization of grammars. In D. I. Slobin (Ed.), The ontogenesis of grammar (pp. 153–186). New York: Academic Press. Braine, M. D. S., Brody, R. E., Fisch, S. M., & Weisberger, M. J. (1990). Can children use a verb without exposure to its argument structure? Journal of Child Language, 17, 313–342. Braine, M. D. S., & Brooks, P. J. (1995). Verb argument structure and the problem of avoiding an overgeneral grammar. In M. Tomasello & W. E. Merriman (Eds.), Beyond names for things: Children’s acquisition of verbs (pp. 353–376). Hillsdale, NJ: Lawrence Erlbaum Associates. Brooks, P. J., & Tomasello, M. (1999). How children constrain their argument structure constructions. Language, 75, 720–738. Brooks, P. J., Tomasello, M., Dodson, K., & Lewis, L. B. (1999). Young children’s overgeneralizations with fixed transitivity verbs: The entrenchment hypothesis. Child Development, 70, 1325–1337. Brown, R., & Hanlon, C. (1970). Derivational complexity and order of acquisition in child speech. In J. R. Hayes (Ed.), Cognition and the development of language (pp. 11–53). New York: John Wiley & Sons. Bybee, J. (1985). Morphology: A study of the relation between meaning and form. Amsterdam: John Benjamins. Bybee, J., & Hopper, P. J. (Eds.). (2001). Frequency and the emergence of linguistic structure. Amsterdam: John Benjamins. Clark, E. V. (1987). The principle of contrast: A constraint on language acquisition. In B. MacWhinney (Ed.), Mechanisms of language acquisition (pp. 1–33). Hillsdale, NJ: Lawrence Erlbaum Associates. Croft, W. (2001). Radical construction grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press. Croft, W., & Cruse, D. A. (2004). Cognitive linguistics. Cambridge: Cambridge University Press. Demetras, M. J., Post, K. N., & Snow, C. E. (1986). Feedback to first language learners: The role of repetitions and clarification questions. Journal of Child Language, 13, 275–292. Gergely, G., & Bever, T. G. (1986). Related intuitions and the mental representation of causative verbs in adults and children. Cognition, 23, 211–277. Goldberg, A. E. (1993). Another look at some learnability paradoxes. In E. V. Clark (Ed.), Proceedings of the 25th Annual Child Language Research Forum (pp. 60–75). Stanford, CA: Center for Research in Language Development (CSLI) Publications. Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Goldberg, A. E. (2003). Constructions: A new theoretical approach to language. Trends in Cognitive Science, 7, 219–224. Goldberg, A. E., Casenhiser, D. M., & Sethuraman, N. (2004). Learning argument structure generalizations. Cognitive Linguistics, 15, 289–316. Green, G. M. (1974). Semantic and syntactic regularity. Bloomington: Indiana University Press. Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991). Affectedness and direct objects: The role of lexical semantics in the acquisition of verb argument structure. Cognition, 41, 153–195.

306

BOWERMAN AND CROFT

Hale, K., & Keyser, S. J. (1987). A view from the middle. Lexicon Project Working Papers 10. Cambridge, MA: MIT Center for Cognitive Science. Hirsh-Pasek, K., Treiman, R., & Schneiderman, M. (1984). Brown and Hanlon revisited: Mothers’ sensitivity to ungrammatical forms. Journal of Child Language, 11, 81–88. Ingham, R. (1992). Review of Learnability and cognition: The acquisition of argument structure, by S. Pinker. Journal of Child Language, 19, 205–211. Langacker, R. W. (1987). Foundations of cognitive grammar, Vol. 1: Theoretical prerequisites. Stanford, CA: Stanford University Press. Laughren, M., Levin, B., & Rappaport, M. (1986). What’s behind theta-roles: What syntax tells us about lexical representation. Unpublished manuscript, Center for Cognitive Science, MIT. Levin, B. (1985). Lexical semantics in review: An introduction. In B. Levin (Ed.), Lexical semantics in review (pp. 1–62). Lexicon Project Working Papers 1. Cambridge, MA: MIT Center for Cognitive Science. Levin, B. (1993). English verb class alternations: A preliminary investigation. Chicago: University of Chicago Press. Levin, B., & Rappaport, M. (1986). The formation of adjectival passives. Linguistic Inquiry, 17, 623–661. Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax–lexical semantics interface. Cambridge, MA: MIT Press. McCawley, J. (1978). Logic and the lexicon. In D. Farkas, W. Jacobson, & K. Todrys (Eds.), Papers from the parasession on the lexicon (pp. 261–277). Chicago Linguistic Society. Chicago: University of Chicago Press. MacWhinney, B. (1987). The competition model. In B. MacWhinney (Ed.), Mechanisms of language acquisition (pp. 249–308). Hillsdale, NJ: Lawrence Erlbaum Associates. Maratsos, M., Gudeman, R., Gerard-Ngo, P., & DeHart, G. (1987). A study in novel word learning: The productivity of the causative. In B. MacWhinney (Ed.), Mechanisms of language acquisition (pp. 89–113). Hillsdale, NJ: Lawrence Erlbaum Associates. Marcus, G. F. (1993). Negative evidence in language acquisition. Cognition, 46, 53–85. Mazurkewich, I., & White, L. (1984). The acquisition of the dative alternation: Unlearning overgeneralizations. Cognition, 16, 261–283. Morgan, J. L., & Travis, L. L. (1989). Limits on negative information in language input. Journal of Child Language, 16, 531–552. Oehrle, R. T. (1976). The grammatical status of the English dative alternation. PhD dissertation, MIT, Cambridge, MA. Ping, L., & MacWhinney, B. (1996). Cryptotype, overgeneralization, and competition: A connectionist model of the learning of English reversative prefixes. Connection Science, 8, 1–28. Pinker, S. (1984). Language learnability and language development. Cambridge, MA: Harvard University Press. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Pye, C., & Loeb, D. (1995). Experimenting with the causative alternation. Unpublished manuscript, University of Kansas. Randall, J. (1990). Catapults and pendulums: The mechanics of language acquisition. Linguistics, 28, 1381–1406. Regier, T. (1996). The human semantic potential: Spatial language and constrained connectionism. Cambridge, MA: MIT Press. Regier, T., & Gahl, S. (2004). Learning the unlearnable: The role of missing evidence. Cognition, 93, 147–155.

13.

ACQUISITION OF THE ENGLISH CAUSATIVE ALTERNATION

307

Schütze, H. (1994). A connectionist model of verb subcategorization. In A. Ram & K. Eiselt (Eds.), Proceedings of the 16th Annual Conference of the Cognitive Science Society (pp. 74–78). Hillsdale, NJ: Lawrence Erlbaum Associates. Tomasello, M. (1992). First verbs: A case study of early grammatical development. New York: Cambridge University Press. Tomasello, M. (Ed.). (1998, 2002). The new psychology of language, Vols. 1 and 2. Mahwah, NJ: Lawrence Erlbaum Associates. Tomasello, M. (2000). Do young children have adult syntactic competence? Cognition, 74, 209–253. Tomasello, M. (2003). Constructing a language: A usage-based theory of language acquisition. Cambridge, MA: Harvard University Press.

CHAPTER 14

What Adverbs Have to Do with Learning the Meaning of Verbs Angelika Wittek ETH (Swiss Federal Institute of Technology), Zurich

“…and the prince woke Sleeping Beauty again.” 1. INTRODUCTION Acquiring verb meanings is certainly not an easy task for a child. It has been shown that the process is quite complicated; for example, children tend to pick up certain meaning components earlier than others, and often seem to have underspecified or overspecified verb meanings at first (Pinker, 1989). This chapter focuses on the acquisition of transitive verbs expressing a change of state, like fill or wake and their German counterparts. These verbs have a complex semantic structure. For a verb like wake, as in example (1), 1. The prince woke Sleeping Beauty. [prince] CAUSE [BECOME (AWAKE (Sleeping Beauty))]

at least three meaning components should be distinguished (see, e.g., Dowty, 1979): (a) an agent or force that typically does something, as indicated by the first bracketed constituent in (1), prince; (b) a component that represents the changed endstate of the referent of the object NP, such as AWAKE, as part of the second bracketed constituent, and (c) a component expressing a causal relationship between (a) and (b): CAUSE. Languages differ as to how this complex meaning is mapped onto linguistic form. English typically uses morphologically simple verbs (wake). German also uses morphologically simple verbs (wecken ‘wake’), but the preferred way to express causative state changes is with morphologically complex verbs with a separate particle that specifies the changed endstate (e.g., wachmachen ‘awake-make’, i.e., wake). 309

310

WITTEK

Languages also differ as to whether an event has to be fully realized for a change-of-state verb to apply. The “realization of an event” is the normal final state of the affected entity/ies of an event described by a verb, that is, the changed endstate (cf. Pederson, this volume). In languages like English or German, a changed endstate is entailed by the meaning of a causative change-of-state verb. For example, the verb wake, as in (1), necessarily entails that the patient first was not awake and then is awake. In contrast, in other languages like Tamil (cf. Pederson, this volume), causative verbs are by default used to refer to state changes, but they can also felicitously be applied to situations in which no changed endstate comes about. In this chapter I am mainly concerned with the learning task of English- and German-speaking children. Adult speakers of English and German know that an endstate is part of the meaning of causative change-of-state verbs, but do children know it as well? There is evidence that children learning English do not have an adult understanding of the meaning of morphologically simple change-of-state verbs like mix and fill (Gentner, 1978; Gropen, Pinker, Hollander & Goldberg, 1991). Specifically, children often do not view the endstate as a necessary part of the meaning. How general is this problem? Do children learning German, in which state changes are expressed with morphologically complex verbs, also neglect the endstate? If so, how do children learn the full meaning of change-of-state verbs? The chapter is organized as follows. First, I summarize the results of the studies that have been concerned with English-speaking children. I then present the results of a series of comprehension studies that I conducted with German-speaking children. On the basis of these results I formulate the learning problem and place it in a wider crosslinguistic perspective (section 2). Then I move on to the question of how children arrive at an adult understanding of the meaning of causative change-of-state verbs. I first examine whether existing proposals on how verb meaning is learned (Gleitman, 1990; Pinker, 1989) can provide an answer (section 3). Then I introduce a new hypothesis (the adverbial modification cue hypothesis, section 4) and present a comprehension experiment with novel verbs that tests this hypothesis (section 5). The main findings of the chapter are summarized in section 6. 2. A PROBLEM IN CHILDREN’S UNDERSTANDING OF THE MEANING OF CHANGE-OF-STATE VERBS 2.1. Children’s Interpretation of Change-of-State Verbs: The Weak Endstate In a study of the acquisition of common cooking terms, Gentner (1978) observed that children aged 5–7 tend to interpret the change-of-state verb mix as if it specified a certain manner of motion, but no particular endstate. For example, they accept as an instance of mix a scene showing somebody stirring or beating cream, a substance that cannot increase in homogeneity. Similarly, Gropen et al. (1991) found that

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

311

children aged roughly 4 to 6 tend to treat a change-of-state verb like fill as if it specified only that an agent performs a particular action that normally leads to the endstate. For example, as an instance of fill they choose a scene of somebody pouring liquid from a pitcher into a glass, even though the glass does not become full, in preference to a scene showing a glass indeed becoming full when somebody lets water drip into it from a faucet; that is, they interpret fill as if it means pour. Both studies show, then, that in interpreting change-of-state verbs children learning English tend to neglect the endstate component and assume that a certain manner of motion performed by the agent is more important. This phenomenon has been termed the “manner bias” or the “action-over-result” preference. Do children learning languages other than English also neglect the endstate? In particular, how do children learning a language like German interpret causative change-of-state verbs? In German, the preferred way to express state changes is with morphologically complex verbs with a particle that specifies the endstate. Importantly, in these verbs, the changed endstate is displayed transparently on the linguistic surface. Some have argued that speakers make use of transparency in interpreting the meaning of words (e.g., Clark, 1993: 115ff). We might therefore hypothesize that children pay more attention to the endstate in interpreting verbs that are transparent with respect to the endstate (wachmachen ‘awake-make’, i.e., wake) than verbs that are not transparent (wecken ‘wake’). To test this hypothesis, I carried out a series of comprehension experiments with 4- and 5-year-old children (for details see Wittek, 1999a, 1999b, 2002). Subjects watched short video clips. In the critical test films, an action was shown that might be expected to result in a particular endstate, but did not (= “no-change” condition). For example, a film testing knowledge of the transitive change-of-state verbs wecken ‘wake’ and wachmachen ‘awake-make’ (i.e., wake) showed a girl holding up a ringing alarm clock right in front of a sleeping man. After the alarm clock has done its job, the man is still sleeping. After watching each video clip, the child had to answer a question that contained one verb of the pair to be tested in that trial, e.g., (2) or (3): 2. Hat das Mädchen den Mann wachgemacht? ‘Has the girl the man awake-made?’ (Did the girl wake the man?) 3. Hat das Mädchen den Mann geweckt? ‘Has the girl the man woken?’ (Did the girl wake the man?)

The correct answer to this question in the no-change condition is “no.” Most of the children’s answers in this condition were the correct “no,” but an average of 30% were the incorrect “yes,” independent of the morphological form of the test verbs. How should this be interpreted? I would hesitate to conclude that these children completely neglected the endstate of the change-of-state verbs that I tested, for example, thought that only the action per-

312

WITTEK

formed by the agent was relevant to the verb. Some of the answers that children gave in the no-change condition demonstrate that they are in fact sensitive to the endstate: 4. Hat das Mädchen den Mann wachgemacht? (Experiment 1) ‘Has the girl the man awake-made?’ (made the man awake) N 4;5: Ja, aber der tut nich aufwachen. ‘Yes, but he does not up-wake.’ (wake up.) 5 Hat das Mädchen den Mann geweckt? (Experiment 1) ‘Has the girl woken the man?’ J 4;9: Ja, aber der hat’s gar nicht gehört. ‘Yes, but he did not hear it at all.’ 6. Stimmt es, dass das Mädchen den Mann weckte? (Experiment 3) ‘Is it right that the girl the man woke?’ (that the girl woke the man?) M 4;5: [Nods] Aber der stand nich auf. ‘But he did not stand up.’ 7. Stimmt es, dass das Mädchen den Apfel pflückte? (Experiment 3) ‘Is it right that the girl the apple picked?’ (that the girl picked the apple?) a. T 4;0: Ja, aber der is noch dran. ‘Yes, but it is still on there [i.e., on the twig].’ b. D 5;2: Ja, aber’s kann nich, weil der so festhängt. ‘Yes, but it [i.e., she] cannot, because it hangs so tight.’

These children give incorrect “yes” answers to questions containing both morphologically complex and simple change-of-state verbs, but they also elaborate on their answers in ways that suggest a different, more adult-like understanding. These elaborations are interesting in two respects. First, children sometimes comment on the outcomes of the scenes depicted in the video clips. Some children make comments about the object that was manipulated by the girl. For example, in answer to the question about whether the girl has picked the apple, children comment that the apple is (still) on the tree (examples 7a and b); that is, they describe what can be seen on the screen at the end of the video clip. Or, in answer to the question about whether the girl has woken the man, they state that the man did not wake up (4) or stand up (6), or that he did not hear the alarm clock (5); that is, they explain that the object did not undergo a certain change. Second, children often connected these descriptions to their “yes” answers with aber ‘but’. Aber ‘but’ implies a contrast between two conjuncts (Grice, 1961: 126ff; for evidence on children’s use and understanding of aber ‘but’, see also Kail

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

313

& Weissenborn, 1984). We can interpret these “but” clauses as describing something that violated children’s expectations. In particular, for the scenes to match with the test verbs, the children might have expected that the girl’s actions would lead to particular results, but the actual results did not correspond to their expectations. In sum, a qualitative analysis of the data shows that even though the endstate might not constitute a critical meaning component in young German-speaking children’s understanding of change-of-state verbs, they do not completely neglect it, as was hypothesized for English-speaking children (Gentner, 1978; Gropen et al., 1991). In particular, German children seem to interpret these verbs as if they specify that a particular endstate might well be expected to come about, but need not. I term this the WEAK ENDSTATE interpretation (Wittek, 1999b, 2002). Is such an interpretation plausible from a linguistic perspective? In the following, I will show it is by no means exotic: there are verbs in the adult language that closely correspond to the children’s weak endstate interpretation. 2.2. “Weak Endstates” in the Adult Language Consider the English verb wash. The default interpretation of this verb, and of its German equivalent, is that it refers to a state change of the object acted on; for example, on hearing a sentence like He washed the shirt we assume that the shirt ended up clean (Talmy, 1991:32). However, the endstate “clean” is defeasible, as adding a “but” clause shows: 8. I washed the shirt, but it came out dirty. (Talmy, 1991: 32)

Hence, wash is used for situations in which a particular endstate (“clean”) is likely to come about, but need not. In this respect, the way adults interpret wash corresponds closely to children’s weak endstate interpretation of change-of-state verbs. To capture children’s weak endstate interpretation in a more precise way, let us now have a look at what wash actually means. Specifically, is the endstate “clean” part of its meaning or not? Talmy argues that the meaning component we are dealing with here (“clean”) is a kind of “implicature” which is associated with a particular lexical item (wash); this meaning component is therefore probably part of the lexical content of wash. However, it is not “conventionally implicated,” because a conventional implicature—unlike the implicature associated with wash—is not defeasible. For example, the conventional implicature associated with but, which is contrast, cannot be defeated (Grice, 1961: 126ff). Talmy therefore introduces a new kind of implicature to capture the phenomenon: “lexicalized implicature.” According to Talmy (1991:29), there are only a few verbs in English that behave like wash in that they imply, but do not assert, a changed endstate. For example, even though speakers of English associate the

314

WITTEK

endstate “dead” with the verb strangle, some speakers can apply this verb to situations where a person does not actually die (Talmy, 1991: 31). Talmy’s account of verbs like wash is problematic, however. In particular, what should the semantic representation of verbs like wash look like? Under the view of a “universal semantics” (see Jackendoff, 1983; Pinker, 1984, 1989), the component “state” is either present in the meaning of a verb or it is not. How can we integrate an implicated endstate into the semantic representation of a verb in this framework? Would we need to introduce a new primitive? A more elegant solution, in my view, is the one proposed by Brisson (1994) (see Levin & Rappaport Hovav, 1991, and Rappaport Hovav & Levin, 1998, for the same analysis). Brisson (1994: 94) provides an Aktionsart analysis for the verb sweep (which is similar to wash in the relevant respect), on the basis of which she classifies sweep as describing only a manner of surface contact. She argues that sweep can be modified with time-span adverbials like for an hour, as in Jack swept the floor for an hour. This is evidence that “sweeping” can go on homogeneously in time, so sweep should be analyzed as an activity predicate (see Vendler, 1967). But sweep can also be modified by time-frame adverbials like in an hour, as in Jack swept the floor in an hour. This is normally taken as evidence that the verb can specify a natural endpoint or logical culmination, and so should be classified as an accomplishment (Vendler, 1967).1 However, Brisson argues that sweep is not a real accomplishment predicate, because the verb does not specify a natural endpoint. In her view, the endstate “clean” is only pragmatically favored, not lexically specified, because—as Talmy (1991) noted for wash—it can be disclaimed with a “but” clause. More generally, Brisson proposes that a distinction should be made between telic, or completive, readings and true accomplishment readings. According to this distinction, the sentence Jack swept the floor in an hour has only a completive reading, that is, “the job is done,” not an accomplishment reading (“the floor is clean”). Brisson (1994: 97ff.) points out that verbs that allow a “job is done” reading are typically verbs of cleaning or agriculture that have “location” arguments, such as plow a field, mow a lawn. She suggests that the reason we can get the completive reading for sentences like He swept the floor in ten minutes is perhaps that there is a specific location whose extent delimits the activity. Borrowing Brisson’s terminology, I term verbs like wash and sweep PRAGMATICALLY FAVORED ENDSTATE verbs (Wittek, 1999b, 2002). However, although Brisson (1994) hypothesizes that pragmatically favored endstates are limited to verbs that take location arguments, I would argue that there are many pragmatically favored endstate verbs, and they fall into several semantic classes. For example, verbs like sweep, wipe, and wash (wipe verbs, Levin, 1993:125), have the endstate “clean” associated with them. Verbs like sew, weld (shake verbs, Levin, 1993: 161), and darn—when they are used to refer to repairing something— suggest the endstate “fixed” or “being whole.” Verbs like shoot, strangle, hang, crucify, and poison (poison verbs, Levin, 1993: 232) suggest “dead.” Verbs like bite, punch, stab, and swat (swat verbs, Levin, 1993: 150) suggest a change in the physical state of a patient (“hurt”). All these verbs belong to a subtype of activity

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

315

verbs in that they describe a manner or means by which a certain endstate is typically achieved. Recall that I argued that German children sometimes interpret change-of-state verbs as if a certain endstate might well come about, but need not (weak endstate interpretation). For example, children interpret wake as if it meant ‘perform an action that typically causes someone to wake up’. Children’s interpretation of change-of-state verbs is thus similar to adults’ interpretation of pragmatically favored endstate verbs. This suggests that the meanings children attribute to change-of-state verbs are not in themselves exotic. I intend this characterization only as a rough approximation to children’s interpretation of change-of-state verbs. I do not claim that children interpret the verb clean as if it meant, for example, 2 sweep or wash, or kill as if it meant shoot or stab. If this were the case, it would boil down to saying that children interpret change-of-state verbs as if they were manner verbs (as has indeed been argued for children learning English), because real pragmatically favored endstate verbs do in fact specify particular manners. One of the reasons it has been argued for the acquisition of English that children interpret change-of-state verbs like manner verbs is that, in linguistic theory, a fundamental two-way distinction has been made between action/manner and result verbs (see Gropen et al., 1991; Levin & Rappaport, 1991; Pinker, 1989; Rappaport Hovav & Levin, 1998). Given the stress on these two categories, it has been natural to assume that if children neglect the endstate of a change-of-state verb, they must assume that the verb specifies a particular manner. But when a child has a weak endstate interpretation of a change-of-state verb this does not necessarily mean that she interprets the verb as if it specified a specific manner or means— she may simply associate it with some action that would normally bring about the endstate. This kind of interpretation is not implausible: some languages have verbs that revolve entirely around a certain endstate and do not specify a particular manner, but that can still be used felicitously to refer to situations in which the endstate does not come about. One such example is causative change-of-state verbs in Tamil (see Pederson, this volume, for a detailed analysis of these verbs). In Tamil, monomorphemic causative verbs that are by default used to refer to state changes do not necessarily entail a state change of the patient. Pederson tested Tamil subjects on their interpretation of change-of-state verbs (using the video stimuli that I had created to test German children’s interpretation of change-of-state verbs), and his subjects gave answers similar to those of the German children (i.e., sentences along the lines of “Yes, he broke it, but it didn’t break”). How do children correct weak endstate interpretations when they are inappropriate? For children learning German and English, the existence of pragmatically favored endstate verbs as well as change-of-state verbs in the linguistic environment presents a subtle learning problem: how can a child who hears a verb applied to a change-of-state situation determine whether the verb entails or only pragmatically favors the endstate? For example, how can a child who hears “You VERBED the shirt”, after a shirt got washed and ended up clean, determine whether the verb

316

WITTEK

referred only to the action performed on the shirt (wash) or also to the resulting state 3 (clean)? In the next section, I evaluate what existing proposals about the acquisition of verb meaning might have to say about this. 3. EXISTING PROPOSALS ABOUT HOW VERB MEANING IS LEARNED THAT CAN BE APPLIED TO THE PROBLEM OF THE “WEAK ENDSTATE” Major existing proposals about the acquisition of verb meaning assume that verb meanings are internally structured and that there is a close relation between a verb’s meaning and its syntactic properties. The proposals can roughly be classified into two categories with respect to the perspective they take on a child’s learning task. First, children can use the linguistic context a verb is heard in to learn something about its meaning (“syntactic bootstrapping”). And second, children come equipped with certain constraints about verb meanings in general and can to a great extent make use of the nonlinguistic context to determine a particular verb’s meaning (“semantic structure hypothesis testing”). 3.1. The Syntactic Bootstrapping Hypothesis Language learners constrain their hypotheses about what a verb could mean by systematically taking into account the syntactic surface structures in which it appears. The most prominent approach of this type is the syntactic bootstrapping hypothesis (Gleitman, 1990; Naigles, 1990). The basic idea behind syntactic bootstrapping is that children can use information about universal syntactic–semantic correspondences to home in on the meaning of a verb. In particular, children can draw information from the set of subcategorization frames in which the verb occurs. Thus, there should be a set of subcategorization frames that distinguishes change-of-state verbs from other verbs. Critically, for our purposes, this set should be different from the set in which pragmatically favored endstate verbs like wash appear. Is this the case? In both English and German, the basic syntactic frame for change-of-state verbs is the transitive frame (e.g., She cleaned the shirt). However, the transitive frame is basic for many kinds of other verbs as well, including contact verbs (e.g., She stroked the cat; see Levin, 1993:148ff. and 155ff., for English; and Rapp, 1997, for German) and—critically—pragmatically favored endstate verbs (e.g., She washed the shirt). Even many stative verbs like include or surround (as in The mountains surround the village) are transitive. Children would thus not be able to distinguish change-of-state verbs from verbs in other semantic classes simply by hearing them in transitive frames (for experimental evidence that children indeed do not make such a distinction, see Naigles & Kako, 1993). Transitivity alone, then, is not enough to distinguish change-of-state verbs from verbs in other semantic classes. But are there other syntactic frames that might do the job?

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

317

One possibility is the unspecified object alternation. This is a transitivity alternation in which a basically transitive verb can appear both with and without a direct object; the subject remains the same in both versions (John reads a book/John reads) (see the references in Levin, 1993: 33, for English; see Ehrich, 1997, and Rapp, 1997, for German). Many English and German manner verbs participate in the alternation, including pragmatically favored endstate verbs like sweep (John swept the floor/John swept). But causal change-of-state verbs typically do not (*John killed). Could children use the participation of manner verbs in the unspecified object alternation to distinguish them from change-of-state verbs? Crucially, learners would have to infer from a verb’s nonparticipation in the unspecified object alternation that it expresses a causal state change. But there are other kinds of transitive verbs that also require a direct object, such as contact verbs (She stroked the cat/*She stroked) and stative verbs (Mountains surround the village/*Mountains surround). A transitive verb that does not participate in the unspecified object alternation is not, then, necessarily a change-of-state verb. There are other plausible candidate syntactic frames (the resultative, the causative–inchoative alternation, and the locative alternation with verbs of removal; see Wittek, 1999b, 2002, for a detailed discussion). However, none of these syntactic frames straightforwardly distinguishes causative change-of-state verbs from verbs in other semantic classes. I conclude that the syntactic bootstrapping proposal does not, at least at present, offer a straightforward solution to the weak endstate problem. 3.2. Semantic Structure Hypothesis Testing Verb meanings are learned to a great extent by observing the real-world scenes the verbs are used to refer to, with certain additional factors constraining learners’ hypotheses. This approach is known as semantic structure hypothesis testing (Pinker, 1989). According to Pinker, children come equipped with a finite set of universal conceptual categories that are relevant for the meaning of verbs; these include, for example, ‘path’, ‘manner’, and—most critically—‘state’. On hearing a verb applied to a certain situation, children test which of these categories might play a role in the verb’s meaning (semantic structure hypothesis testing). According to Pinker’s (1989) model, then, a child would learn the meaning of a change-of-state verb, like that of any other verb, by hearing it used across situations. In a situation where a shirt is washed and cleanliness is achieved, a child might hear either the change-of-state verb clean or the pragmatically favored endstate verb wash. A scenario of shirt-washing can therefore be ambiguous between cleaning and washing (see Pinker, 1989: 254, on a similar ambiguity with respect to a scenario of filling a cup, which can be ambiguous between filling and pouring). The child might now assume that a particular way of manipulating an object is critical for the meaning of clean. But suppose the child also hears clean when a floor is swept or a counter is wiped. The child could infer from this that what is at stake is not a particular way of manipulating objects, but a certain endstate, that is, “clean.”

318

WITTEK

She would then revise her assumptions about the meaning of clean, correctly assigning it a change-of-state meaning. This sounds straightforward, but there are some hidden complications. First, there is a large overlap in the situations to which both change-of-state verbs and pragmatically favored endstate verbs can be applied. As noted in the previous section, pragmatically favored endstate verbs can be used to refer to situations in which a certain endstate comes about. For example, wash and sweep are usually used in situations in which cleanliness is achieved, and, for example, strangle and shoot when death comes about. The child might not, then, so readily reject the incorrect assumption that manner plays a role in the meaning of clean. She could, for example, revise her original assumption that clean means ‘wash’ not by switching to an entailed endstate meaning, but by hypothesizing that clean expresses a more general type of manner that encompasses “cleansing” actions on surfaces in general.4 But let us suppose the child solves this problem and recognizes that clean is not about manner but about the endstate “clean.” Now an even more critical problem arises: how can the child infer that this endstate is entailed, and not just pragmatically favored? Recall that many children treat change-of-state verbs like pragmatically favored endstate verbs, and that other languages—like Tamil—have nonmanner verbs that are otherwise comparable to clean, break, and kill but that do not require the endstate to be realized for felicitous use. To solve this problem, Pinker (1989: 256ff) might resort to one of the “practical constraints” that he posits play a role in learning verb meanings. Two such constraints are the Principle of Contrast (Clark, 1987), and attention to the discourse context. The Principle of Contrast states that no two words have exactly the same meaning. Following this principle, a child would not assign a new word a meaning that is identical to the meaning of an existing word in her lexicon. This would prevent her from assigning the same meaning to clean as, for example, to wash. But this does not ensure that she will conclude that clean has an entailed endstate. Instead, she might assume that clean expresses a more abstract type of manner that encompasses actions on surfaces in general. Alternatively, she might realize that clean does not express any particular manner at all, but still think that the endstate “clean” is only pragmatically favored, as, for example, in Tamil. Thus, even though the Principle of Contrast might help the child to distinguish between clean and wash, it does not solve the weak endstate problem. Pinker’s second relevant constraint is the existence of discourse contexts in which adults make explicit statements about the applicability of a verb (Pinker, 1989: 257). For example, imagine a situation in which an adult insists that for the verb clean to apply, a shirt must end up clean (This shirt is still dirty, so you did not clean it.). This could alert the child to the fact that the endstate “clean” is a necessary component of the meaning of clean. Although such statements presumably occur occasionally, it is unclear that they would be frequent or systematic enough—covering all the relevant verbs—for children to rely on them to distinguish state-change verbs from pragmatically favored endstate verbs.5

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

319

In sum, hearing a change-of-state verb across multiple situations, as is called for by the semantic structure hypothesis-testing proposal, does not guarantee that a child will abandon an inappropriate weak endstate interpretation. In particular, the presence of pragmatically favored endstate verbs in a child’s linguistic input might support these interpretations because the situations in which these verbs and change-of-state verbs are used overlap considerably. Neither syntactic bootstrapping nor semantic structure hypothesis testing provides a straightforward solution to how children could learn a subtle meaning difference like the distinction between entailing and only implicating a certain endstate. Ideally, a child would need something in the input that is more discriminating than subcategorization frames and more frequent than definitional discourse contexts. In the next sections, I show that certain adverbials in the linguistic input do indeed satisfy these requirements: they are frequent and reliable cues, and children are capable of using at least one such cue, the adverb wieder ‘again’, to determine that a verb’s meaning entails a state change. I term this new approach to solving the weak endstate problem the ADVERBIAL MODIFICATION CUE HYPOTHESIS. 4. A NEW PROPOSAL: THE ADVERBIAL MODIFICATION CUE HYPOTHESIS Modifying a verb with certain adverbials means giving additional information about parts of that verb’s meaning (Dowty, 1979; Morgan, 1969). The basic idea of my proposal is that children can use this information to learn about the verb’s meaning. When we provide additional information about something, we make it more salient. We know that children pay special attention to highlighted or salient elements in a speech unit (Slobin, 1985: 1164ff.). For example, children are attentive to phonological saliency (e.g., Peters, 1985: 1034ff.) and positional saliency (e.g., Slobin, 1985: 1166; Wittek, 1999, 2002). As another saliency factor we can now consider the relative weight placed on a particular meaning component of a verb, depending on whether it is modified by an adverbial. I refer to this as “informational saliency.” The adverbial modification cue hypothesis can be stated as follows (Wittek, 1998, 1999b, 2002): Adverbials apply to particular parts of the meaning of the verb they modify. In being picked out, these meaning components are highlighted and made informationally more salient. Because children pay special attention to salient elements in a speech unit, their attention should be drawn to these components. This in turn could lead learners to infer that these components are critical for the verb’s meaning.

The adverbial modification cue hypothesis provides a candidate answer to the question of how children solve the the weak endstate problem. If a child has a weak endstate interpretation of a change-of-state verb, an adverbial that modifies the endstate might draw her attention to the importance of this meaning component. One such adverb is again.

320

WITTEK

The adverb again, as in example (9) (adapted from Dowty, 1979: 260ff), specifies that something occurs more than once. This “something” can be everything that is expressed by the verb (the repetitive reading, schematized in 9a): it is again the case that the prince did something that caused Sleeping Beauty to be awake. But it can also be just the endstate component of the verb’s meaning (the restitutive reading, shown in 9b): the prince did something that caused Sleeping Beauty to again be in a state of being awake. The same holds true for German wieder ‘again’: 9. The prince woke Sleeping Beauty again. a. Repetitive reading: again [prince CAUSE [BECOME (AWAKE (Sleeping Beauty))]] b. Restitutive reading: [prince] CAUSE [BECOME again (AWAKE (Sleeping Beauty))]

The restitutive reading is the interpretation one should get on reading the fairy tale. It is called a situation of “reversal”: Sleeping Beauty is awake, she falls asleep, then the prince comes and kisses her for the first time, which causes her to be awake again. What justifies the presence of again is the recurrence of the former state, “awake.”6 The restitutive reading is not possible with pragmatically favored endstate verbs, such as shoot (examples 10 and 11 are adapted from Cruse, 1991: 229): 10. Frankenstein brought the monster to life, then shot it again. ¡ pragmatically favored endstate verbs: only repetitive reading possible (Frankenstein shot the monster twice). 11. Frankenstein brought the monster to life, then killed it again. ¡ change-of-state verbs: both repetitive (Frankenstein brought the monster to death twice) and restitutive (Frankenstein brought the monster to life and then brought it back to death) readings possible.

Recall that children can have a weak endstate interpretation of change-of-state verbs; that is, they don’t see a changed endstate as being an entailed part of the verb’s meaning. But it is exactly this endstate that the adverb again modifies in the restitutive reading. How could a child use the presence of again in the linguistic input as a clue to a verb’s meaning? The idea goes as follows. When a child hears, for example, The prince woke Sleeping Beauty again in the context of this story, it is clear from the story that the prince is not performing this action for a second time; rather, a certain state is coming about for a second time: Sleeping Beauty is awake again. The child can thus infer that again modifies this endstate, and that the verb therefore encodes the endstate. Is the restitutive reading of again or German wieder available to young children? Existing work shows that children have mastered the reading by at least age 3 or 4.

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

321

For example, in a production study conducted by Clark, Carpenter, and Deutsch (1995), German children had to command the experimenter to put objects back into their original states. Even the youngest children (mean age 3;7) added wieder ‘again’ to their requests for reversal about 40% of the time (for further evidence see Bamberg, 1994, on the German frog stories). Analysis of spontaneous speech supports this conclusion. A search of the CHILDES corpora (MacWhinney, 1985) that I conducted (Wittek, 1999b, 2002) revealed that children around their third birthday as well as adults use the adverb in its critical reading in their everyday speech. 5. COMPREHENSION EXPERIMENT WITH NOVEL VERBS TESTING RESTITUTIVE WIEDER ‘AGAIN’ AS A CUE THAT A VERB ENTAILS AN ENDSTATE To test whether children can use restitutive wieder ‘again’ in the linguistic input to learn that an endstate is part of the meaning of a verb, I carried out a comprehension study in which I presented novel verbs to German-speaking children. 5.1. Method Subjects. Forty-eight children (28 girls, 20 boys), age 4;0–6;10 (mean age 5;6), were recruited from two kindergartens in Kleve, in the German Niederrhein area. Materials. The stimuli for the experiment were eight video clips showing novel events. Each video clip consisted of two parts, a TEACHING PHASE and a TEST PHASE. The events were novel in that, in the teaching phase, an unusual action caused a certain endstate to come about. The TEACHING PHASE of each video clip first showed an object in a certain state (e.g., a lamp which was off). The state of the object was then changed by a cleaning woman (the lamp was switched on). Then an actor gave a single demonstration of a novel event: he manipulated the object in an unusual way, which caused it to change back to its original state (he rotated the lamp once by pushing it, which resulted in the lamp being again off). Half the novel events depicted a change in a property of the object at the end of the teaching phase, the other half a change of location; both can be considered a kind of state change. In the TEST PHASE a similar object was shown which had already undergone a change of state (e.g., a lamp which was already on). An actress manipulated the object in the same way as the actor had, but the object did not change back to its original state (e.g., the lamp stayed on). The teaching phases are described in Table 14.1, along with the novel verbs that were used to label them (which event was labelled with which verb was randomized across subjects). The stimuli, preceded by four warm-up items and interleaved with eight filler items, were presented by means of a Hi-8 video player on a color monitor.

322

WITTEK TABLE 14.1 Novel Events Used in the Comprehension Experiment with Novel Verbs (description of the teaching phases)

State changes: 1. die Lampe prillern ‘priller the lamp’* A lamp, which is off, is switched on by a woman (dressed as a cleaning woman). The actor rotates the arm of the lamp once by pushing on it. At the end of the rotation, the lamp clicks off. 2. den Korb moffeln ‘moffel the basket’ An empty basket is filled with scarves by the cleaning woman. The actor pulls on twisted cords which are fastened at either end of the basket, making it revolve upside down a few times. The scarves fall out and the basket is empty again. 3. die Dose zuppern ‘zupper the box’ An open box is closed by the cleaning woman. The actor picks up the box and shakes it a few times. The lid comes up and the box is open again. 4. die Kette bauseln ‘bausel the chain’ The two loose ends of a chain draped across the back of a chair are joined by the cleaning woman. The actor pounds on the chain a few times with a metal spoon. The ends separate so that the chain is in two pieces again. Location changes: 5. den Ball luschern ‘luscher the ball’ A ball is taken out of a cardboard box and put on a table by the cleaning woman. The actor dribbles the ball a couple of times towards the edge of the table with a big book. The ball falls off the table and back into the box. 6. das Tierchen klanscheln ‘klanschel the little animal’ A sweater is on the wall with a velcro animal stuck to it. The animal is taken off by the cleaning woman. The actor uses a big wooden spoon as a catapult to flip the animal through the air toward the sweater. In the end, the velcro animal is attached to the sweater again. 7. den Holzmann flimpern ‘flimper the wooden man’ A wooden man lying on a table is set upright by the cleaning woman. The actor pushes the wooden man across the table with a pair of tongs. The wooden man topples over and ends up lying on the table again. 8. den Vogel totzeln ‘totzel the bird’ A toy bird lying on the floor is hung from a horizontal stick by the cleaning woman. The actor makes the bird swing back and forth by blowing on it a couple of times. The bird falls off and ends up lying on the floor again. *All the novel verbs were suffixed by -eln or -ern, because these suffixes promote an activity interpretation for adults (Steinitz 1981), and, as pilot work has shown, for children as well. By using these suffixes I increased the likelihood that, in the absence of wieder ‘again’, subjects would assign the novel verbs an activity interpretation. The particular novel verbs that were used to label the novel events were randomized across the events; see the design of the experiment in Table 14.2.

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

323

Procedure. All subjects were tested individually in a separate room in their kindergarten. The task was introduced as a game between the child and a little bear puppet. The actress in the test phase was introduced as the bear’s friend. The bear wants to know whether his friend is doing the same as the actor in the teaching phase, but he is still so little that he cannot judge for himself. Therefore, the child is asked to answer the experimenter’s questions for him. Before the actor gave his single demonstration of the novel event in the teaching phase of each video clip, subjects heard the novel verb for the first time (Der Peter macht jetzt was ganz Witziges. Er zeigt uns jetzt, was _____ ist ‘Peter is going to do something really funny now. He will show us what _____ is’, with the blank filled in by the novel verb in the infinitival form, e.g., bauseln). After the child viewed the teaching event, a teaching sentence was used to describe what had just occurred. Subjects in the wieder condition (wieder group, about half of the children: 16 girls and 10 boys) always heard the novel verb together with restitutive wieder ‘again’, for example, Peter hat die Lampe wieder gebauselt ‘Peter bauseled the lamp again’. Because Peter had not done anything to the object before, wieder ‘again’ can only refer to the object’s return to its original state. Subjects in the no-wieder condition (no-wieder group, 12 girls and 10 boys) heard a teaching sentence that was identical except that it lacked wieder ‘again’, for example, Peter hat die Lampe gebauselt ‘Peter bauseled the lamp’. After hearing each teaching sentence, subjects were asked to repeat it. Then the test part of each video clip was shown, and afterward children were asked the test question, which was the same for subjects in both the

TABLE 14.2 Design of Experiment Testing Wieder as an Adverbial Modification Cue Teaching Phase

Scene in the video clip depicts a reversal (cleaning woman changes the state of an object; actor manipulates object, thereby changing it back to its original state) Teaching Sentence

no-wieder group: Teaching sentence does not contain wieder Peter hat den Korb gemoffelt. ‘Peter moffeled the basket.’

wieder group: Teaching sentence contains wieder Peter hat den Korb wieder gemoffelt. ‘Peter moffeled the basket again.’

Test Phase

Scene in the video clip does not depict a reversal (actress manipulates object - which has already undergone the first state change - in the same way as actor did, but object does not change back to its original state) Test Question

Hat Anna den Korb gemoffelt? “Did Anna moffel the basket?’

324

WITTEK

wieder and no-wieder groups, for example, Hat Anna die Lampe gebauselt? ‘Did Anna bausel the lamp?’. The design of the experiment is sketched in Table 14.2. 5.2. Hypothesis Subjects in the wieder group should infer that the novel verb entails a change of state. However, no change of state comes about in the test phase. Therefore, they should give more “no” answers to the questions following the test phase than subjects in the no-wieder group. 5.3. Results Subjects in the wieder group (n = 26) gave “no” answers 62.5% of the time (130 out of 208 answers), while subjects in the no-wieder group (n = 22) gave “no” answers only 36% of the time (64 out of 176 answers). A t-test shows that the difference is significant (df = 46; t = -2.48; p < .05). The experimental prediction is therefore confirmed. The results are even more striking when, instead of comparing the responses of the two groups across all items taken together, as in the simple t-test, we compare them on each item individually in a t-test for paired samples (df = 7; t = -7.94; p < .0005). No significant difference was observed in the responses to the property-change events versus the location-change events: wieder ‘again’ thus alerted children to an entailed endstate for verbs of both kinds. 5.4. Discussion This experiment tested restitutive wieder ‘again’ as an Adverbial Modification Cue. According to the experimental prediction, subjects in the wieder group should infer that the novel verbs entail a state change. Because no state change came about in the test phase of each trial, even though an action had been carried out that might be expected to lead to this state change, these subjects should give significantly more “no” answers than the subjects in the no-wieder group. The results confirm the hypothesis. Could children in the wieder group have given more “no” responses than children in the no-wieder group not because they took the meaning of wieder ‘again’ into account, but simply because they heard some adverb in the sentence? It is unclear why the mere presence of an adverb per se would lead to more “no” responses, but perhaps children for some reason view any adverb as a state change marker.7 To determine whether the effects of exposure to wieder ‘again’ were based on the semantics of this adverb or were due simply to the presence of any adverb, I carried out a control experiment in which I tested a different adverbial, namely, the durational (time-span) adverbial ganz lange ‘very long’. Ganz lange ‘very long’ specifies the duration of an activity or state (e.g., Er streichelte die Katze ganz lange ‘He stroked the cat very long’; cf. Vendler, 1967). Although it is probably more frequent with activity verbs, it is also compatible with

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

325

change-of-state verbs, for which it specifies only the duration of the endstate. For example, the sentence Wir haben das Fenster ganz lange geöffnet ‘We opened the window for a long time’can mean that we caused the window to stay open for a long time (see McCawley, 1971, for a parallel interpretation of the English counterparts). Children’s performance on ganz lange ‘very long’ (the ganz lange group) in the control experiment was compared with their performance on wieder ‘again’ in the main experiment. Twenty-four children (9 girls, 15 boys), aged 3;11–6;11 (mean age 5;4), participated in the study. The experiment was carried out in a kindergarten in Kleve. The stimulus material consisted of six of the eight video clips used in the first experi8 ment. The procedure was the same as in the main experiment, except for the adverbial: in the teaching sentences of the control experiment, subjects always heard the novel verb together with the adverbial ganz lange ‘very long’, for example, Peter hat den Korb ganz lange gemoffelt ‘Peter moffeled the basket very long’. The prediction was that the percentage of “no” answers should be higher for subjects in the wieder group than for subjects in the ganz lange group. The results are as follows. Subjects in the wieder group gave “no” answers to the test questions 62.5% of the time (130 out of 208 answers), compared to 36% (52 out of 114) for subjects in the ganz lange group. The results of a t-test confirm that this difference is significant (df = 48; t = 2.52; p < .05). Taken together, the wieder and ganz lange experiments show that children do not view all adverbs as state-change markers. We can therefore conclude that in the main experiment the adverbial modification cue worked as it was hypothesized to work: children made use of the meaning of the adverb wieder ‘again’ to infer that the novel verbs entail a state change. The proposal that children can use adverbials as cues to verb meaning has clear advantages over major recent proposals about the acquisition of verb meaning (Gleitman, 1990; Pinker, 1989). First, the adverbial modification cue hypothesis explains how children can learn that specific components are critical for a verb, namely, they are entailed by its meaning. In contrast, Gleitman’s (1990) syntactic bootstrapping hypothesis does not have fine-grained enough machinery to predict how children make inferences about specific meaning components; it only makes predictions about how children can use syntactic frames to get a general picture of the meaning of a verb, that is, to narrow down the search space for its meaning.9 Pinker’s (1989) proposal for semantic structure hypothesis testing does predict how children make inferences about specific meaning components. But if a child views the endstate of a change-of-state verb like clean only as pragmatically favored, Pinker cannot easily explain how she can learn that the endstate plays a more critical role than this—that it is entailed by the verb’s meaning. Second, the adverbial modification cue seems to be quite powerful: hearing restitutive wieder ‘again’ only once together with a particular verb can already guide the child to a change-of-state interpretation. Other proposals about the acquisition of verb meaning, in contrast, would require children to hear a change-of-state

326

WITTEK

verb in multiple syntactic (Gleitman, 1990) or situational (Pinker, 1989) environments to be able to home in on the change-of-state component of its meaning—if, in fact, they are able to do so at all with these procedures. 6. CONCLUSIONS This chapter adds two new findings to research on the acquisition of verb meaning. First, it demonstrates that even though children might not view the endstate component of change-of-state verbs as entailed, they sometimes interpret it as specifying only a weak endstate: there is a particular endstate that they associate with the verb, and this endstate might well come about, but even if it does not, the verb can still be used felicitously. This interpretation is by no means exotic: in adult German and English there are verbs that are typically used to describe situations in which particular endstates usually come about even though these endstates are not entailed; for example, wash and sweep are associated with cleanliness, whereas sew and darn are associated with repairing. These “pragmatically favored endstate” verbs specify a certain kind of action that might be expected to lead to the associated endstate. But some languages, notably Tamil, allow verbs otherwise comparable to break and kill to be extended to situations in which no state change comes about (see Pederson, this volume). Second, the chapter shows evidence supporting a new proposal about how children could arrive at an adult understanding of the meaning of change-of-state verbs. According to the ADVERBIAL MODIFICATION CUE HYPOTHESIS, children can use an adverb like German wieder or its English counterpart again as a cue to learn that certain verbs entail an endstate. Unlike distinctive subcategorization frames (Gleitman, 1990) or definitional discourse contexts (Pinker, 1989), the adverb wieder ‘again’ used in a restitutive context is a frequent and reliable cue to the presence of an entailed endstate. But wieder ‘again’ is just one example of an adverbial modification cue. Adverbs may reveal something about verb meaning to learners much more generally. For example, linguists use adverbials like for 5 hours versus in 5 hours as diagnostics to determine a verb’s temporal properties (Vendler, 1967). Perhaps children can use them to learn these properties in the first place. ACKNOWLEDGMENTS The research reported in this chapter is part of my dissertation project, which was supported by a grant from the Max Planck Society for the Advancement of Science, Munich. I owe special thanks to Melissa Bowerman and Wolfgang Klein for discussing ‘again’ over and over again; thanks also to the members of the Acquisition group and the Institute’s Argument Structure and Scope Project groups and to the participants and the invited discussants of the workshop “Crosslinguistic perspectives on argument structure: Implications for learnability” for critical discussion of the paper. I am also very grateful to Henriette Hendriks for her skill as an actress in all of my

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

327

video clips, and to Eric van Loosbroek for acting in the novel verb experiment. I am especially indebted to kindergartens “Sonnenblume,” “Purzelbaum,” “St. Elisabeth,” and “St. Marien” in Kleve for their invaluable cooperation. Last but not least, thanks also to Jos van Berkum for help with the statistics. NOTES 1

See Ehrich (1997) for the observation that the German verb fegen ‘sweep’—like its English counterpart—can be combined with both time-span and time-frame adverbials. Although Ehrich concludes that the verb is ambiguous between an activity and an accomplishment interpretation, I favor Brisson’s (1994) analysis for the reasons already given. 2 In fact, we do not know how children interpret the legitimate pragmatically favored endstate verbs of their language; this has never been tested. 3 Naigles (1996: 223) formulated a similar learning problem for contact verbs: “How should a child who hears only ‘You’re X-ing the cat’ while stroking a responsive cat determine whether the verb referred specifically to the action performed on the cat (stroke), or to the resulting action that the cat produced …?” 4 Naigles (1991: 74) points out a similar problem for a child’s hypotheses about the meaning of bring. On hearing bring applied to situations that involve carrying, the child might assume that it expresses a particular manner, carrying. On hearing it applied to situations that involve pulling, she might not reject the hypothesis that a particular manner plays a role in the meaning of bring, but instead assume that the manner for bring is of a more general/coarser grain. 5 In the next section I argue that discourse contexts indeed can convey information about the meaning of a verb. But I present a new approach to solving the weak endstate problem, according to which a child who does not realize that a particular change-of-state verb entails a caused endstate does not need to hear an adult making an explicit statement about the verb’s applicability, but instead can discover this by using the information provided by optional elements (adverbials like again) in the adult’s utterance. 6 It is still an open issue in linguistic theory how the scopal properties of adverbs like again should be captured in more formal terms (see, e.g., Dowty, 1979, and von Stechow, 1996, and their references). However, this does not affect the argument that I want to make here, which is that the restitutive reading of again is only possible when the adverb modifies a change-of-state expression. 7 One way to give such a hypothesis a theoretical basis might be to follow up on Keyser and Roeper’s (1992) proposals about abstract clitics. Keyser and Roeper assume for English (although they give their idea a universal flavor by taking into account crosslinguistic data) that each verb is associated with a categoryneutral abstract clitic position. This position can be occupied by verbal prefixes like re-, particles like up, a dative object, and so on. The clitic position is linked to the aspect of a sentence. For example, the particle up, as in John ate the apples up, marks “delimitedness” (Keyser & Roeper, 1992: 117). Perhaps children

328

WITTEK

know that such an abstract clitic position exists, they assume that adverbials like again can occupy such a position, and for some reason they associate the position with a “delimitedness” or endstate interpretation. 8 Two video clips were omitted because the novel actions they depicted were too brief to be modified appropriately with the durational adverbial ganz lange ‘very long’. These were items number 1 (rotating the lamp), and 6 (catapulting the animal) (see Table 14.1). 9 Gleitman and her collegues recognize this: “What most people think of as the ‘meaning’(that open concerns being ajar while close concerns being shut) is nowhere to be found in the syntax of sentences. Rather, we have shown that the initial narrowing of the search-space for that meaning, by attention to argument structure as revealed by the syntax, is the precondition for using the scene information efficiently to derive the meaning” (Fisher, Hall, Rakowitz, & Gleitman, 1994: 367). REFERENCES Bamberg, M. (1994). Development of linguistic forms: German. In R. Berman & D. I. Slobin (Eds.), Relating events in narrative: A crosslinguistic developmental study (pp. 192–211). Hillsdale, NJ: Lawrence Erlbaum Associates. Brisson, C. (1994). The licensing of unexpressed objects in English verbs. In K. Beals, J. Denton, R. Knippen, L. Melnar, H. Suzuki, & E. Zeinfeld (Eds.), Papers from the 30th Annual Regional Meeting of the Chicago Linguistic Society (CLS), Volume 1: The main session (pp. 90–102). Chicago: Chicago Linguistic Society. Clark, E. V. (1993). The lexicon in acquisition. Cambridge: Cambridge University Press. Clark, E. V., Carpenter, K. L., & Deutsch, W. (1995). Reference states and reversals: Undoing actions with verbs. Journal of Child Language, 22, 633–662. Cruse, D. (1991). Lexical semantics. Cambridge: Cambridge University Press. Dowty, D. R. (1979). Word meaning and Montague grammar. Dordrecht, The Netherlands: D. Reidel. Ehrich, V. (1997). Wertsteigerung und Wertverlust—Die Veränderung der Valenz. In C. Dürscheid, K. H. Ramers, & M. Schwarz (Eds.), Sprache im Fokus. Festschrift für Heinz Vater zum 65. Geburtstag (pp. 259–276). Tübingen: Niemeyer. Fisher, C., Hall, D. G., Rakowitz, S., & Gleitman, L. (1994). When it is better to receive than to give: Syntactic and conceptual constraints on vocabulary growth. Lingua, 92, 333–375. Gentner, D. (1978). On relational meaning: The acquisition of verb meaning. Child Development, 49, 988–998. Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55. Grice, H. P. (1961). The causal theory of perception. Proceedings of the Aristotelian Society, Suppl. Vol. 35 (pp. 121–152). Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991). Syntax and semantics in the acquisition of locative verbs. Journal of Child Language, 18, 115–151. Jackendoff, R. (1983). Semantics and cognition. Cambridge, MA: MIT Press. Kail, M., & Weissenborn, J. (1984). A developmental cross-linguistic study of adversative connectives: French ‘mais’ and German ‘aber/ sondern’. Journal of Child Language, 11, 143–158. Keyser, S., & Roeper, T. (1992). Re: The abstract clitic hypothesis. Linguistic Inquiry, 23, 89–125. Levin, B. (1993). English verb classes and alternations: A preliminary investigation. Chicago: University of Chicago Press.

14.

ADVERBS AND LEARNING THE MEANING OF VERBS

329

Levin, B., & Rappaport Hovav, M. (1991). Wiping the slate clean: A lexical semantic exploration. Cognition, 41, 123–151. MacWhinney, B. (1985). The CHILDES project: Tools for analyzing talk. Hillsdale, NJ: Lawrence Erlbaum Associates. Morgan, J. (1969). On arguing about semantics. Papers in Linguistics, 1, 49–70. Naigles, L. R. (1990). Children use syntax to learn verb meanings. Journal of Child Language, 17, 357–374. Naigles, L. R. (1991). Review article on ‘Learnability and cognition: The acquisition of argument structure’ by Steven Pinker. Language and Speech, 34, 63–79. Naigles, L. R. (1996). The use of multiple frames in verb learning via syntactic bootstrapping. Cognition, 58, 221–251. Naigles, L. R., & Kako, E. T. (1993). First contact in verb acquisition: Defining a role for syntax. Child Development, 64, 1665–1687. Peters, A. (1985). Language segmentation: Operating principles for the perception and analysis of language. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition, Volume 2: Theoretical issues (pp. 1029–1067). Hillsdale, NJ: Lawrence Erlbaum Associates. Pinker, S. (1984). Language learnability and language development. Cambridge, MA: Harvard Uinversity Press. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Rapp, I. (1997). Fakultativität von Verbargumenten als Reflex der semantischen Struktur. Linguistische Berichte, 172, 490–529. Rappaport Hovav, M., & Levin, B. (1998). Building verb meanings. In M. Butt & W. Geuder (Eds.), The projection of arguments (pp. 97–134). Stanford, CA: Center for the Study of Language and Information (CSLI) Publications. Slobin, D. I. (1985). Crosslinguistic evidence for the language-making capacity. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition, Volume 2: Theoretical issues (pp. 1157–1249). Hillsdale, NJ: Lawrence Erlbaum Associates. Steinitz, R. (1981). Der Status der Kategorie “Aktionsart” in der Grammatik. Linguistische Studien A, 76. Berlin: Akademie der Wissenschaften der DDR. Talmy, L. (1991). Path to realization: A typology of event conflation. In L. A. Sutton, C. Johnson, & R. Shields (Eds.), Proceedings of the 17th Annual Meeting of the Berkeley Linguistics Society (pp. 480–519). Berkeley, CA; Berkeley Linguistics Society, University of California. Vendler, Z. (1967). Verbs and times. In Z. Vendler (Ed.), Linguistics and philosophy (pp. 97–121). Ithaca, NY: Cornell University Press. von Stechow, A. (1996). The different readings of wieder ‘again’: A structural account. Journal of Semantics, 13, 87–138. Wittek, A. (1998). Learning verb meaning via adverbial modification: Change-of-state verbs in German and the adverb wieder ‘again’. In A. Greenhill, M. Hughes, H. Littlefield, & H. Walsh (Eds.), Proceedings of the 22nd Annual Boston University Conference on Language Development (pp. 779–790). Somerville, MA: Cascadilla. Wittek, A. (1999a). Zustandsveränderungsverben im Deutschen—wie lernt das Kind die komplexe Semantik? In J. Meibauer & M. Rothweiler (Eds.), Das Lexikon im Spracherwerb (pp. 278–295). Tübingen: Francke. Wittek, A. (1999b). “…and the prince woke Sleeping Beauty again.” Learning the meaning of change-of-state verbs: A case study of German. PhD dissertation, University of Tübingen. Wittek, A. (2002). Learning the meaning of change-of-state verbs: A case study of German child language. Studies on language acquisition, 17. Berlin: Mouton de Gruyter.

CHAPTER 15

Event Realization in Tamil Eric Pederson University of Oregon

1. INTRODUCTION The following utterance (from Herring, 1988: 282) is perfectly sensible in Tamil, although the English translation reads as self-contradictory: 1. aiyar

teengkaay-ai uTai-tt-aar.

aanaal teengkaay uTai-ya-villai.

brahmin coconut-acc break(tr)-Ps-3sResp but

coconut

1

break(intr)-Inf-Neg

‘The brahmin broke the coconut. But the coconut didn’t break.’

This chapter explores how one language encodes the realization of an event. The “realization of an event” is understood here as the normal final state of the affected entity/ies of an event described by a minimal verb, that is, a simple lexical root plus all obligatory co(n)-verbs, inflections, and so on.2 For example, in English, The Brahmin broke the coconut necessarily entails that the coconut is afterward broken—the change of state of the patient is the full realization of a ‘breaking’event. In the case of break, the realization is typically expressed with the participial form of the same root (broken). However, with other verbs a different root may be used for the resultant state. For example, John killed the lizard entails that the lizard be subsequently dead. As we show, the strength of these entailments varies not only from verb to verb, but also across languages for translation-equivalent verbs. Examples of less than full entailment of realization have been discussed for Japanese (Ikegami, 1985), Mandarin (Klein, Li Ping, & Hendriks, 2000; Talmy, 1991), and Tamil (Herring, 1988; Pederson, 1990; Talmy, 1991), although the terminology and analyses differ significantly. The term “realization” has been used by Pederson and Talmy. In some accounts, realization is straightforwardly a type of temporal profiling, that is, a property of either Aktionsart or aspect (see Ikegami, 1985; Klein et al., 331

332

PEDERSON

2000). I argue instead that realization should be analyzed as notionally distinct from temporal profiling. Of course, there may often be a conflation of notions, with a single marking indicating both a specific temporal profile and realization (or perhaps lack of realization). As an analogy, tense and aspect are notionally distinct for the analyst, but the semantics of both are commonly incorporated into a single morpheme or construction. Realization may also be entangled with tense or aspect marking in some languages, so we need to carefully separate these semantic categories. Even if a language allows realization to be defeated (i.e., sensibly denied in subsequent clauses/utterances, as in the ‘killed lizard’ not being dead), languages vary as to which features of the event or degrees of realization are entailed as having occurred. Must the ‘killed’lizard be at least badly injured or affected? Or could it actually not be affected at all? Assume for the moment that the typical agent–patient verb has something like 1DO 2CAUSE/AFFECT 3STATE/EFFECT as part of its se3 mantic representation. Most so-called “lexical causative” verbs are of this type in English. For the English verb kill, all three of these components are entailed. For an event to be characterized as ‘killing’ in English, an agent must not only 1DO something, but this DOing must 2CAUSE a consequent or simultaneous (sub-)event, that is, an effect on the patient. The patient must then be in a consequent 3STATE that holds for some interval of time (typically indefinitely). To continue our example, John raises a stick and brings it down forcefully on our hapless lizard (DO); this quite literally impacts on the lizard (CAUSE), such that the lizard is now in a new and specific STATE (dead). If there is a different resultant state, kill is an inappropriate lexical representation (instead, a verb like bruise might have been called for). One cannot say in English John killed the lizard, but it did not die even if there was full intent to kill. If there is no CAUSE component in the referenced event, then the expression needs a verb that does not entail any causal effect on the patient (e.g., John hit the lizard), or an activity verb that does not even implicate a causal component (e.g., John waved at the lizard). English breaktrans, like kill, requires that the subject/agent DO something that CAUSEs an effect that results in a particular STATE—broken, in this case. A number of verbs in English have a clear entailment of DO and CAUSE, but the STATE is only strongly implicated, not entailed. For example, Talmy (1991) cites choke as differing from strangle in that the former only implicates the resultant death of the asphyxiatee, whereas the latter—at least for many speakers—entails a resultant death. However, many parallel examples in Tamil may entail as far as 1DO but not even necessarily as far as 2CAUSE. That is, ‘agent breaks patient’ (in Tamil) may only entail that the agent 1DO something (breaking-like), but does not entail that there be any 2CAUSAL EFFECT on the patient. Example (1) does not require that the coconut is even temporarily damaged. For certain classes of verbs, for example, causative forms of mental/social/physical change-of-state verbs, realization can be quite systematically denied:

15.

EVENT REALIZATION IN TAMIL 2. naan pooliic-kaaran-ai I

police-man-Acc

333

koopa-ppaTu-tt-in-een,

aanaal avar

anger-suffer-Caus-Ps-1s

but

3smResp

koopa-ppaT-a-villai. anger-suffer-Inf-Neg ‘I made the policeman angry, but he didn’t get angry.’ 3. naan avan I

3sm.Obl

vaarkai-yai

kaRai-ppaTu-tt-in-een,

aanaal

avan

life-Acc

soil-suffer- Caus-Ps-Is

but

3sm.Obl

vaarkai-yil kaRai-ppaT-a-villai. life-Loc

soil-suffer-Inf-Neg

‘I damaged his reputation, but his reputation was not damaged.’

In contrast, the occasional English verbs that do not entail realization (e.g., choke, as mentioned earlier) typically entail that the agent 1DO something (place hands or instrument around patient’s neck) such that this DOing 2CAUSEs something with respect to the patient (caused to have impaired breathing), but the patient need not reach the typical resultant 3STATE (dead by asphyxiation). Many Manda4 rin verbs appear to be of this pattern: 4. Tal mai3 shu1 he buy

book

ke3shi4 mei2

mai3-dao4

but

buy-achieve

not

‘He bought a book but didn’t get it.’ (Klein & Hendriks, 1995)

It might be tempting to assume that the core verb in examples like (1) essentially has the notion of ‘try to X’ built into its semantic description, such that, without explicit marking of resultant state, the verb can be understood as indicating only intention. However, this explanation cannot account for why constructions as in (5) are instead used when the speaker indeed intends to communicate that the subject ‘tries to VERB’.5 5. aiyar brahmin

teengkaay-ai

uTai-kka-ppaar-tt-aar.

aanaal

coconut-Acc

break(tr)-Inf-see/try-Ps-3sResp

but

teengkaay uTai-ya-villai. coconut

break(intr)-Inf-Neg

‘The brahmin tried to break the coconut. But the coconut didn’t break.’

Note also that in examples (1) and (5), the verb root, uTai ‘break’, is the same. The first clause in the sentences uses the transitive inflectional paradigm and the second clause uses the intransitive paradigm. If a verb root meant something like ‘try to break/be.broken’, then the translation ‘The brahmin tried to break the coconut. But the coconut didn’t try to break’ would be even more absurd.

334

PEDERSON

More importantly for this counterargument, many sentences with defeated realization explicitly deny intentionality on the part of the subject. This is clearly not consistent with an interpretation of the semantics as having a TRY component: 6. naan I

katti-yaal knife-Inst

avan-uTaiya kai-yai his-Gen

hand-Acc

tappippooy accidentally

veTT-in-een. cut-Ps-1s

aanaal nallaveelai aTi-paT-a-villai. but

fortunately

blow-suffer-Inf-Neg

‘I accidently cut his hand with the knife. But fortunately, he wasn’t cut.’

Clearly then, neither the verb roots nor the simple transitive verb construction invokes a notion of intentionality. Yet these examples suggest that these Tamil verbs can differ strikingly from their English counterparts. How does a child learn that defeated realization is an appropriate use of much vocabulary in Tamil? Conversely, how does a child speaking a language like English learn that this is an inappropriate use of most vocabulary? The straightforward answer for the Tamil-speaking child might seem simple enough: she simply hears utterances like (1) and notes them. However, such utterances in both adult and child spontaneous speech are actually quite rare. Although all the speakers I have consulted largely agree that these ‘Vtr, but not Vintr’ sen6 tences are acceptable, they seldom produce them spontaneously. This is an odd fact in and of itself. If such sentences are as rare as they appear to be, we might expect only speakers with fully developed language to share adult judgments. I have not tested these sentences with young children, but one 6-year-old, who was ostensibly playing and ignoring our discussions in the next room, repeatedly called out the same judgments as his consultant mother with whom I was working. The converse questions seem equally interesting: do young English- (or German-) speaking children accept or produce utterances with defeated realization? If so, how do they learn that this is inappropriate? If not, what is it about learning English that precludes defeasible realization, while Tamil fosters it? 2. EXPERIMENTAL EVIDENCE 2.1. The Original German Study Wittek (see chapter 14 of this volume) ran an experiment that addresses this issue with 4- to 5-year-old German-speaking children and German-speaking adults as controls. In each trial, a puppet predicted that a particular event would occur (e.g., a door would be closed or an animal killed) in a videotaped segment that the child was about to view. The child was then shown a videotaped event in which the expected realization either did or did not occur, and was asked whether the puppet had predicted correctly. The experiment was designed to determine whether German-

15.

EVENT REALIZATION IN TAMIL

335

speaking children understand various change-of-state verbs (either lexical causatives like füllen ‘to fill’ or explicit causatives like vollmachen ‘make full’) as refer7 ring to activities or to events with realized states. Unsurprisingly, German-speaking adults never accepted video clips of unrealized events as correctly predicted. However, children accepted unrealized events as appropriate and correct roughly 30% of the time with these change-of-state verbs. Indeed, some children responded with explicit statements analogous to the Tamil examples shown earlier, for example, saying that ‘The hunter killed the animal but the animal didn’t die!’ Superficially, at least, the German-speaking children in this experiment pattern somewhat like Tamil adults in elicitation sessions. Because many German-speaking children and Tamil-speaking adults accept change-of-state verbs for unrealized events, this suggests that there may be a universal tendency to interpret change-of-state verbs as allowing reference to unrealized events. Something in the process of becoming an adult speaker of German, but not of Tamil, suppresses this tendency. For direct comparison between the German and Tamil speakers, I took a copy of Wittek’s materials to South India and ran essentially the same study with fluent native speakers of Tamil (older children and adults). 2.2. Materials as Used in South India Video clips were presented on a portable (ca. 12 cm) video screen under nonglare conditions to participants with apparently normal vision. Unfortunately, Wittek’s videotaped stimuli were more appropriate for a European population than a largely rural South Indian population. One scene (breaking open a walnut with a nutcracker) from Wittek’s original tape was deleted because of cultural unfamiliarity. In other scenes, certain elements needed to be redefined. Two participants remarked that the video clip of the actor closing a door showed the door momentarily closed (for about two frames), and then it reopened by itself. Thus it was appropriate for at least these participants to say that ‘He closed the door, but it opened.’ Accordingly, I do not consider responses to this video clip to be particularly revealing. 2.3. Translatability of the German Questions Another problem was to decide on the best translation8 of the German expressions. The most serious problem was with Kerze ausmachen/löschen (‘blow out the candle’), which must be translated with a verb sequence uuti aNai (blow-extinguish). All other expressions could be translated with simplex Tamil verbs. Even so, some problems with the exact semantic profiles of simplex verbs remain. For example, translating in Schrank klettern (‘climb into a closet’), would typically require the use of a simplex verb in Tamil, but this gives a somewhat different profile of the event: nuzai ‘enter’ essentially refers to the crossing of a relevant boundary to a

336

PEDERSON

container (e.g., a doorway to a house or a gate to a temple). As a simplex verb, nuzai 9 does not specify that the subject progresses into the interior of a container. The video clip of this event shows a scene where the potential entry into a closet is barred by shelves. Because nuzai potentially could refer to threshold crossing without consequent containment within the interior, this cannot be a good test of an unrealized event. However, this scene was used with German participants only as a control trial, so the comparison with the Tamil is less critical. Last, one of the scenes that was retained for analysis (Apfel abmachen/ pflücken ‘pick apple’) was quite translatable as pazam paRi ‘pluck fruit’. However, some consultants complained that the protagonist was not actually engaged in paRi behavior because the action was too gentle to remove a fruit from the tree. A similar complaint was made about Teller zerbrechen/ kaputtmachen ‘break plate’ (Tamil: taTTu uTai). All of the other trials seemed unproblematic in terms of results and comments from consultants and participants. Note that all of the verbs used in the questioning were unambiguously transitive. Many have intransitive counterparts built from the same root, but the tense morphology makes it quite clear that it is the transitive verb stem that is used. The Appendix contains the complete text of what was said to the participants. Note that all forms used in the prediction about each video clip are in the simple present tense. The simple present is normally used in Tamil for definitive future time reference as well as present time reference (as is common in many languages). 2.4. Method Prior to the experiment, participants were not given information about the specific purpose of the experiment, but they had enough information to understand the research as part of my efforts to better learn Tamil as a second language. Because I was working with older children and adults, I used an adult native speaker rather than a puppet to predict what the protagonist would do. (Wittek also did this with her adult controls.) After seeing each video clip, the participant was given the same verb in the simple past tense with question marking. See the Appendix for this text. Participants were to give a simple yes/no response, which in Tamil most naturally takes the form of repeating the verb in the same tense with either the person marking or the negation marking. That is, a “yes” response will actually have the explicit form of, for example, ‘he broke [it]’. A “no” response will have the form ‘not break [it]’. There were several control trials in which the protagonist engaged in an entirely different activity than the one predicted. The only correct response to these trials is a “no” response. The critical trials were presented with both a realized and unrealized endstate in the video clips (although each participant saw only one version). Participants were divided into two groups: one group saw three of the critical trials (the first set below) as realized clips and the other four as unrealized clips; the other group saw the reverse realized/unrealized presentations. Both groups saw the same control trial clips, and the trial orders were the same for both groups.

15.

EVENT REALIZATION IN TAMIL

337

These scenes consisted of: Filling a glass (Trial code 1, see Appendix) Breaking a plate (Trial code 4) Picking a fruit (Trial code 7) Killing a demon (Trial code 2) Waking a man (Trial code 3) Closing a door (Trial code 5) Extinguishing a candle (Trial code 8) As mentioned earlier, closing a door, breaking a plate, and picking a fruit all received occasional comments suggesting that these depictions were imperfect for our research question for these Tamil speakers. 2.5. Participants The 51 participants were mostly monolingual and ranged in age from 9 to 66 years. Many were nonliterate and unschooled. Nine participants were excluded from the analysis because they responded “yes” (i.e., the prediction was evaluated as correct) to all or most trials—including the control trials that called for a “no” response.10 All other participants gave “yes” responses for those trials that showed the predicted realization and “no” for the control trials, indicating that they understood and were correctly engaged with the task. 2.6. Results I focus here only on responses to trials showing the unrealized scenes. For the first set of these (fill glass, break plate, pick fruit) there were responses from 22 (nonexcluded) participants, and for the second set (kill demon, wake man, close door, extinguish candle) there were reponses from 20. The results are reasonably consistent: the unrealized scenes were accepted with a description using the simple transitive verb between 20% and 55% of the time (Table 15.1). If we focus just on the unproblematic video clips (filling, killing, waking) and exclude ‘extinguishing a candle’, which requires a complex verb construction rather than a simple verb, we see that in roughly half of the responses, the simple past tense of a transitive change-of-state verb was accepted as a description of an unrealized event. This proportion is conservatively low because many participants first gave responses on the order of ‘Yes, he broke it, but it didn’t break’; then, when asked to give a single-clause yes/no response, they replied ‘No, he didn’t break it.’ These responses were coded as “no” responses. I estimate that about 6–7 participants gave this type of response at some point. (I do not have tape recordings of most participants and the written notes may be incomplete as to what was said beyond the

338

PEDERSON

TABLE 15.1 Responses on Unrealized Event Trials Unrealized Events

“Yes” Responses

“No” Responses

10

12

Breaking a plate

6

16

Picking a fruit

5

17

Filling a glass

Killing a demon

9

11

Waking a man

10

10

Closing a door

11

9

4

16

Extinguishing a candle

yes/no response.) The participants who responded once in this way never gave another “yes” response to an unrealized event. I take this as indicating that referring to unrealized events with simple past transitive verbs is sufficiently inappropriate, in the context of this task, that participants decided that the most “proper” response is “no” for unrealized events despite their initial “yes” responses. Even if we consider only the clearest “yes” responses of the table, we see that the Tamils accept these verbs for unrealized events far more often not only than German-speaking adults (who reject them completely), but also than German-speaking children (who accept them roughly 30% of the time). In other words, the phenomenon of using simplex verbs to refer to unrealized events is robust in Tamil. This needs an explanation, especially vis-à-vis languages like English and German. However, we cannot accept any account that simply states that the Tamil verbs have distinctively different semantics from English or German verbs (e.g., ‘try to VERB’, as discussed earlier), because more than half the time Tamil speakers rejected their use for unrealized events. Having experimentally verified the difference between Tamil and German verb use, let us more carefully examine the role of realization in the semantics of Tamil verbs. 3. DIFFERENT POSSIBLE ACCOUNTS There are various possible accounts of why Tamil often allows realization to be defeasible. In this section I consider five of these, before going on, in section 4, to propose my own analysis. 3.1. Referentiality of the Object Argument Objects that are incorporated into a verb (e.g., berry-picking vs. picking berries) are often described as having reduced referentiality. That is, the ‘berry’ of berry-picking is generic for the activity, and does not refer to any specific berries to be picked.

15.

EVENT REALIZATION IN TAMIL

339

Consistent with this, when one incorporates an object in English, the realization of an event can often be defeated; for example, We went berry-picking yesterday, but we couldn’t find any berries anywhere and We fished for hours without catching any fish are both sensible. If the incorporated noun is nonreferential—that is, there is no entailment of an actual patient—then there is no semantic entailment that such a patient would have been affected. Such examples may seem somewhat parallel to the Tamil examples with defeated realization. Indeed, the absence of a definite article in examples like (1) may superficially appear to suggest object-incorporation. However, Tamil does in fact mark a definite/indefinite distinction with inanimate objects. When inanimate objects are indefinite, they do not take the accusative marking and when they are definite, such marking is obligatory. Accordingly, in examples like (1), where there is an explicit accusative marker on the inanimate object, there can be no doubt that the coconut is a specific (definite) coconut, making this sentence directly parallel to breaking the coconut in English. Because the degree of referentiality is comparable for both English and Tamil, an explanation of the difference with respect to event realization must lie elsewhere. 3.2. Different Unmarked Aspect If the difference in entailed realization between Tamil and a language like English or German does not lie with the marking of the patient, then presumably it lies in some systematic difference between the verbal forms. Perhaps the simple past in English entails the completion of an event (and hence its full realization), whereas the simple past in Tamil is actually semantically incompletive. As we show later, the tense marking in Tamil is reasonably neutral with regard to realization, and re11 fers just to the time of occurrence of the event. But first let us consider tense and aspect marking with respect to realization in some other languages. Realization in English can be somewhat defeasible when the past progressive form of the verb is used: 7. *I broke the coconut, but the priest stopped me and it didn’t break. 8. ?I was breaking the coconut, but the priest stopped me and it didn’t break.

Indeed, Klein et al. (2000) provide a similar argument for Mandarin. Like Tamil, Mandarin appears often to sanction the denial of realization. The basic facts are similar to the Tamil case, in that the normal resultative complement associated with a verb can be denied in a subsequent clause.12 9. Mao1 pa2-shang4 cat

shu4, ta1

climb-ascend tree

he

pa2

shu4 ke3shi4 mei2 pa2-shang4.

climb tree

but

not

climb up

‘The cat climbed up the tree, he climbed the tree but did not climb up.’

340

PEDERSON

That is, the use of an unmodified verb in a declarative sentence does not entail that there be a second state (for the affected entity). Examples such as this assert in effect that the cat was engaging in “climbing” (analogous to progressive aspect in English), but that the final (second) state of having climbed up is not explicit. However, certain traditionally recognized aspectual markers (such as the “perfective” le) have as their primary function to “assert” the actuality of the second state. That is, the function of Mandarin aspectual particles is to assert the truth value of certain segments of the temporal sequence of the event depicted by the verb. For example, the use of le asserts all states of the (simplex or compound) verb marked with le; the final state cannot be denied: 10. Ta1 He

sha4-le

ren2.

kill-le

man

‘He killed the man.’ 11. *Ta1 He

sha4-le

ren2

ke3shi4

mei2

sha4-si3.

kill-le

man

but

not

kill-dead

‘He killed the man but did not kill him dead.’

Although this theory of temporal particles may prove adequate for Mandarin, it is difficult to apply the same analysis to Tamil. Many languages use simple tenses for both progressive and non-progressive aspect. However, Tamil has an explicit progressive auxiliary verb koNTiru. Yet it is actually more difficult to defeat the realization of a verb if it is marked with progressive koNTiru than if it is a simple verb alone:13 12. ??/*kuppai-yai eri-ttu-kkoNTiru-nt-een

aanaal kuppai eri-ya-villai.

trash-Acc burn(tr)-CoV-koNTiru-Ps-1s but

trash

burn(intr)-Inf-Neg

‘I was burning the trash, but it didn’t burn.’ 13. kuppai-yai trash-Acc

eri-tt-een

aanaal

kuppai

eri-ya-villai.

burn(tr)-Ps-1s

but

trash

burn(intr)-Inf-Neg

‘I burned the trash, but it didn’t burn.’

Even if we ignore the entailment of realization with explicit progressive marking, there is a problem with the hypothesis that progressive aspect allows defeasibility of realization. Under this hypothesis, we should expect that all languages with a simple past tense marking that is neutral with respect to progressivity will allow realization to be defeated for verbs marked with this form. German, for example, has a simple past form that is unmarked for progressivity. However, German examples with denied realization were unequivocally rejected by my consultants:

15.

EVENT REALIZATION IN TAMIL 14. *Ich töt-et-e I

kill-Ps-1s

341

ihn,

aber

er

starb

nicht.

3sm.Acc

but

3s.Nom

die.Ps.3s

Neg

‘I killed/was killing him, but he didn’t die.’

Whatever the marking, to the extent that progressives in languages like English allow defeasible realization, defeasibility cannot further extend through the penultimate subevent. English minimally requires DO CAUSE and probably also DO CAUSE BE.STATE, where the STATE is allowed to be, for example, NOT.BROKEN. That is, ? I was breaking the coconut, but it didn’t break (if acceptable at all) requires that some activity have impacted the coconut, although the acceptability of this sentence is so marginal that interpretations among those consulted were inconsistent. In contrast to these decidedly odd English readings, Tamil allows every subpart of the depicted event beyond the initial DOing to be defeated: 15. naan maaTT-ai I

kooTaali-yaal veTT-in-een aanaal kooTaali veTT-a-villai.

cow.Obl-Acc axe-Inst

cut-Ps-1s

but

axe

cut-Inf-Neg

‘I cut the cow with the axe, but the axe wouldn’t cut (e.g., it was too dull to have any effect whatsoever on the cow).’

Even if one were to defend the argument that the difference between Tamil and English is one of aspectual marking, this last difference in the scope of what must be entailed would not be predicted. 3.3. Different Verbal Semantics—Aktionsarten Because of these differences concerning which semantic features can be subsequently denied, some researchers have hypothesized that verbs vary more systematically across languages in their core semantics than previously suspected. That is, “translation-equivalent” verbs in different languages make very different commitments about what occurs during the referent event. In the terms we have been using so far, most English agent–patient verbs (e.g., kill, break) entail as far as STATE, but occasional English examples (e.g., shoot, choke) seem to entail only as far as CAUSE, leaving the outcome ambiguous. Mandarin and Japanese, on the other hand, may more commonly have verbs that entail only as far as CAUSE. Some researchers (Ikegami, 1985, for Japanese; Tai, 1984, for Mandarin) posit that different languages have different general patterns of Aktionsarten than what is found for English agent–patient verbs. Ikegami notes that Japanese and English differ in the acceptability of sentences such as the following14: 16. musubime-o

yurumeta

keredo,

mada

kata-katta.

knot-Obj

loosened

though

still

tight-was

*‘I loosened the knot, but it was still tight.’

342

PEDERSON

17. sara-o kawakashita keredo, mada dish-Obj dried though still ?‘I dried the dishes, but they are still wet.’

nureteiru. is-wet

18. boku-no kamera-de Mary-no shashi-o totta keredo, utsutte inakatta 1s-Gen camera-Inst Mary-Gen picture-Obj took though taken wasn’t ?‘I photographed Mary with my camera, but no picture turned out.’

Ikegami attributes the differences in realization entailments to the general Aktionsart patterns characteristic of each language. For example, English call is an accomplishment verb (in the Vendlerian sense15), in that it cannot readily be combined with certain markers of duration such as how long? when the duration refers to the length of conversation (19). How long? can, however, be used to refer to the duration of the activity of attempting to reach the point of the accomplishment, as in (20). In contrast to this, the Japanese translation equivalent denwa o suru can readily combine with nampun hodo in both senses. 19. *How long did you call John? (talking time) John-ni nampun hodo denwa-o shimashita ka? John-Loc how.many.minutes about telephone-Obj did Question 20. How long did it take to call John? (time to reach/talking time) John-ni denwa-o suru no ni nampun hodo John-Loc telephone-Obj do Nml Loc how.many.minutes about kakarimashita ka? took Question

From such examples, Ikegami concludes that agent–patient verbs are typically accomplishments in English but activities in Japanese. Accomplishments inherently require entailment of a final STATE, but activities do not. Let us suppose that Ikegami’s analysis is also correct for other languages that readily allow event realization to be defeated, including Mandarin and Tamil. But then it is not clear why perfective marking would create an entailment of realization. One would expect the perfective marking to temporally bound the activity, but why should that entail the typical end result of that activity? To repeat the Mandarin example given earlier, defeasibility is impossible when the perfective le is appended to the first verb. Clearly we must analyze the function of le as involving more than simple temporal boundedness. 21. *Ta1 He

sha4-le

ren2

ke3shi4

mei2

sha4-si3.

kill-le

man

but

not

kill-dead

‘He killed the man but did not kill him dead.’

15.

EVENT REALIZATION IN TAMIL

343

Recall also that, as we have seen in (12), even the progressive marking of Tamil verbs entails realization. This is certainly inconsistent with an account that appeals to a different default Aktionsart of Tamil agent–patient verbs. However well Ikegami’s argument may work for Japanese, it makes the wrong predictions for Tamil. Ikegami’s analysis also makes the wrong predictions for English. Many activity or stative verbs in English (i.e., nonaccomplishments) cannot be defeated. For example, sleep is an activity verb that requires actual sleeping success: *I slept last night, but couldn’t sleep. It is simply nonsensical to speak of such an activity as having any realization independent of the activity DO. Now consider the corresponding Tamil verb tuungku. This verb is better translated as ‘fall asleep’—that is, as an accomplishment, not an activity. With tuungku/fall asleep we have a clearly distinguishable DO (transitioning away from wakefulness) and a realized STATE (being asleep). Strikingly, tuungku does not entail realization, although the usual implication of tuungku is that there is a consequent sleep.16 English fall asleep clearly disallows denial of a realized state. But in (22), the tuungku in the first sentence sets up reference to the attempted accomplishment of falling asleep and then in the second sentence the same root is negated to deny that this accomplishment is realized. This example also shows that realizations can be defeated in Tamil with verbs other than the agent–patient verbs discussed previously: 22. neeRRu

raattiri-yil tuungk-in-een aanaal mika cattam iru-nt-at-aal

yesterday night-Loc

sleep-Ps-1s

but

much noise be-Ps-Nml-Inst

tuungk-a muTi-ya-villai. sleep-Inf manage-Inf-Neg *‘I slept [fell asleep] last night, but because it was so noisy, I couldn’t [fall a-]sleep.’

In summary, in Tamil even those verbs that are most clearly not simple activity or stative verbs allow their realization to be defeated. This is counter to Ikegami’s account for Japanese, and suggests that a purely Aktionsart solution to our puzzle is incomplete at best. 3.4. Different Verbal Semantics—Event Type Talmy (1991) takes a different approach to the problem of defeasible event realization. He analyzes Tamil data (from Herring, 1988, and Pederson, 1990, personal communication) and Mandarin data (from Jian-Sheng Guo, personal communication) as exemplifying a fundamentally different pattern of verbal semantics from the pattern of languages like English. That is, Tamil and Mandarin verb roots translating as ‘cause-state’ are taken to actually mean something like ‘engage in behavior that is cause-state-like’. Depending on the general typological marking pattern of the language, full realization is marked with a separate verb (Tamil) or a satellite (Mandarin). Talmy explains:

344

PEDERSON

Of course, the English verbs used to gloss the Mandarin verbs here, e.g., open, kill, kick, do not really correspond in meaning, hence…they can be misleading. For example, a sentence gloss like ‘I killed him but he didn’t die’ is genuinely paradoxical in English but thus incorrectly represents the non-paradoxical Mandarin original, which would be more closely rendered as ‘I assaulted him with intent to kill (and with what would otherwise have been the presumption of killing), but he didn’t die’. …the unitary referent of an English verb often has as a counterpart in Mandarin a two-part conceptualization expressed by a verb plus a satellite. Thus, we have already seen the counterpart of ‘kick’as “‘propel the foot so as to impact with’+ ‘into impact’”, of ‘kill’ as “‘assault so as to kill’ + ‘to death’”, and of ‘open’ as “‘work on so as to open’ + ‘ajar’”. In the same way, we observe the counterpart of ‘cure’ as “‘treat so as to cure’ + ‘to health (lit: good)’”, of ‘break’ (e.g., snap a stick) as “‘squeeze circumpivotally in upon so as to break’ + ‘broken’”, and of ‘select’ as “‘deliberate over so as to choose among’ + ‘into choice’. (Talmy, 1991:34)

Because other lexical semantic features are clearly variable language-internally, such an approach might predict variability within a single language. Consistent with his analysis, Talmy finds parallels to the Tamil and Mandarin cases in some English verbs. To explicitly mark a resultant state (with a resultative complement) is only really acceptable when the verb need not entail the result: 23. The stranger choked/stabbed/strangled/drowned him. 24. The stranger choked/stabbed/?strangled/*drowned him to death. 25. I soaked/?washed/*cleaned the shirt clean.

Like Ikegami, then, Talmy claims that transitive agent–patient verbs in Japanese, Mandarin, and Tamil have fundamentally different meanings from their typical English counterparts, although he characterizes this meaning difference in a different way. However, even if true, why should this be the case? To simply state that the verbs have different semantic features fails to explain why certain languages pattern in one way and other languages in another. Talmy’s account seems most reasonable only when verbs are considered one at a time. Perhaps ‘kill’ in Tamil, Mandarin, and so on really means ‘attack with murderous intent’? On a verb-by-verb basis, such an account seems reasonable, but the sweeping claim that such an ‘exotic’ characterization holds for all or most agent–patient verbs of the language needs demonstration. It is not clear in any event that Talmy’s translations (‘with murderous intent’, etc.) appropriately characterize the relevant semantics. We have already seen a Tamil example (6) where intentionality is not a relevant factor. Indeed, the only candidate semantic factor that is consistently relevant to these examples of defeated realization is simply that the event may be indicated as unrealized. Of course, Tamil verbs may be used most frequently to imply DO CAUSE STATE, but they need not have STATE as part of their semantic specification.

15.

EVENT REALIZATION IN TAMIL

345

Finally, and perhaps most importantly, we again need to consider why perfective marking should entail realization (as in the Mandarin le in example (10) earlier, and as with almost any aspectual marking in Tamil, as we show later). If a verb root meant ‘attack with murderous intent’, wouldn’t a perfective version of that verb mean ‘attack with murderous intent (seen as a completed activity in the past)’, without a strict entailment that the ‘attacking with murderous intent’ specifically ended in a death? 3.5. Different Pragmatics Across the Languages As an alternative approach to the difference between English and Tamil, we might simply assume that the nature of defeasibility is different across languages (or language communities). The semantics of verbs would be essentially the same, but for some reason the Tamil speaker is more comfortable denying what is undeniable for the English speaker. That is, perhaps semantic entailments are generally less com17 pelling for Tamil speakers than for English speakers. But this would be an appeal to a vague difference across language communities for which we have no further explanation. I favor treating the basic mechanisms of pragmatic inference as moderately universal unless we are forced to do otherwise by clear evidence and an explanation of why each speech community has its own preferences. Another argument against this variable pragmatic strength argument comes from the occasional seemingly parallel examples in English (and other languages). English speakers can use choke to implicate ‘dead’, but the final state is certainly defeasible. Generally, we take the difference between strangle and choke to be semantic rather than pragmatic—that is, the entailments of the verbs are simply different as a result of their lexical semantics. Similarly, I cleaned the shirt, but it didn’t get clean seems more a reflection on the meaning of transitive clean than on an English speaker’s notion of cleanliness (this is the line of Talmy, 1991, discussed earlier). Although I see no argument for different pragmatic operations in English versus Tamil, there does appear to be something quite general about the entailment of realization in the grammar of English. English has a number of constructions that modify a verb so as to allow denial of final realization, for example, I shot at the soldier (but missed) versus I shot the soldier. Tamil has relatively few such constructions, but a wealth of means for asserting realization, one of which we see in the next section. 4. A SOLUTION An alternative to the preceding accounts begins from the premise that the core meanings expressed by typical “translation equivalent” verbs in English and Tamil are the same. However, the Tamil verbs are simply more “flexible” in the boundaries of their extensional range, i.e., they are semantically more general just with respect to realization. Thus, Tamil ‘kill’ really does mean ‘kill’ in its basic use, but it can be used

346

PEDERSON

extensionally in ways prohibited by English. This different range of usage needs, of course, to be motivated by other general characteristics of the language. One striking difference between Tamil and English is that verbs usually do not occur as simple verbs alone in a sentence in Tamil, whereas they often do in English. The most basic clause-linking strategy in Tamil is the converb construction (a.k.a. the adverbial participle), in which the converb suffix (abbreviated CoV in the glosses) is affixed to a nonfinite verb stem, creating a formally subordinate clause in sequence with a following clause. Many of the preceding examples contained converb constructions, even though for simplicity I sought to minimize the number of converbs. Typically the converb construction has the function of indicating that events stand in a coherent sequence (e.g., 26) or take place simultaneously (e.g., manner plus path verbs, as in 27).18 26. naan

poo-y-viT-Tu-varu-kiR-een. 19

I

go-CoV-viTu -CoV-come-Pr-1s

‘I am going and afterward I’ll come (back).’ (i.e.,’Goodbye’) 27. naan

naTa-ntu-poo-N-een.

I

walk-CoV-go-Ps-1s

‘I walked away.’

Only the final verb is inflected and, with a few regular exceptions, these inflections have scope over the preceding (syntactically subordinate) verbs. Although the system is fully productive, there are also many heavily grammaticized uses of specific verbs in converb constructions. Most relevant to the current discussion, the realization of the first verb in a converb construction provides the semantic input to the following verb. This is routinely apparent with the auxiliary verb constructions: here, the verb that gives the main semantic information is marked with the converb marker, and then followed by the auxiliary verb, which will be finite if it is sentence final. Consider first the perfective function of viTu: 28. naan

avan-ai

konR-u-viT-T-een…

I

he-Acc

kill-CoV-leave/Prtv-Ps-1s

‘I killed him dead’ (lit.: I killed him and left) *aanaal cak-a-villai. but

die-Inf-Neg

‘but (he) didn’t die.’

Because the verb viTu follows the verb ‘kill’ in a converb construction, the realization of ‘kill’ cannot be subsequently denied. One might think from such examples

15.

EVENT REALIZATION IN TAMIL

347

that the entailment of realization is a function of the perfective semantics of viTu rather than of the converb construction in general (this would point to an analogy between viTu and the Mandarin le). However, consider auxiliaries with a clearly 20 nonaspectual function, such as the reflexive use of the auxiliary koL. Although the simple form of veTTu ‘to cut’ in (6) allows realization to be denied, when veTTu takes converb marking to allow the reflexive auxiliary, realization is strictly entailed. 29. *naan katti-yaal I

knife-Inst nallaveelai

kai-yai

tappippooy

hand-Acc accidently

veTT-i-kko-NT-een.

aanaal

cut-CoV-Ref-Ps-1s

but

aTi-paT-a-villai.

fortunately blow-suffer-Inf-Neg ‘I accidently cut my hand with the knife. But fortunately, I wasn’t cut.’

There is nothing inherent to the reflexive semantics of the Adv+koL construction which would require the realization of the preceding verb as “input”. Rather, entailment of realization of the preceding verb appears to be a general property of the converb construction. Klein et al. (2000) make a similar proposal about Mandarin: that certain aspectual particles explicitly mark the realization of the verb with which they are associated. Only some of the particles have this effect, so Klein et al. consider it a consequence of their lexical semantics. In Tamil, any second verb has the effect of asserting the result of the preceding verb in a converb-marked clause. That is, the strict entailment of realization is fully independent of lexical semantics. Importantly, on this account the realization of the first verb should not be understood as part of the semantics of the second verb. Rather, it is a requirement for the use of the converb construction that a second verb can be used only if the first verb is understood as fully realized. Because Tamil verbs so often occur in the converb construction and the converb construction definitively entails realization of the first verb, it is tempting for a speaker to assume that, when a verb is not used in a converb construction, it might not entail realization. In other words, the regular association of the verb with a construction that entails realization may reduce the need to interpret the verb as also having the same entailment of realization when it occurs alone. Thus, we find a grammatical motivation for why Tamil simple verbs might come to be used without the entailment of realization. This is consistent with the observation that although Tamil consistently uses the converb construction to entail realization, it has few efficient means for explicitly indicating that realization does not occur (e.g., Tamil has a scarcity of words like English almost). The assumption of realization is the norm for English verbs, and—quite the opposite of Tamil—English has a wealth of vocabulary items and constructions that explicitly deny realization, such as almost, nearly, shoot AT the soldier. The English

348

PEDERSON

resultative complement construction (see Goldberg, 1995), as in worry oneself sick, beat the conductor into submission, and freeze solid, might at first glance look like a device for confirming realization, but in most cases it only further specifies the nature of the realization that the verb alone already entails: notice that I cooked the sausage WELL done is fine, but I cooked the sausage done is distinctly odd. Importantly, we need not assume that the meanings of the verbs are dramatically different in Tamil than in English or German. Rather, a Tamil speaker can choose to avoid a construction that entails realization, and this straightforwardly sanctions an interpretation of the verb as not entailing realization. Processes of this general sort are actually familiar in other domains under the rubric of metonomy. All languages (speakers?) can probably be assumed to have productive processes of metonymy, that is, indirect reference to one thing by direct reference to something associated with it (e.g., a capital city standing for a government or nation). Without fairly automatic metonymic processing, the number of distinct lexical items that would be needed would be enormous. For example, piano can refer to the music played on a piano, the musical instrument, the player of that instrument, or—with appropriate context—even the impact crater left by a catapulted piano. Metonymy allows for a semiotically efficient lexical system. Some common forms of metonymy (more traditionally these are termed synecdoche) are the use of a term that normally refers to a part to stand for the whole (e.g., brain for ‘smart person’), or a term for a whole to refer to one of its parts (society for ‘upper class society’). Let us apply this notion of whole-for-part metonymy to verbal events. A verb like ‘break’normally refers to an entire event from DO to final STATE (broken). Now what should speakers do if they need to refer to an event that did not actually contain the final realized state (and they do not wish to specify a different endstate such as ‘bent’)? English speakers can simply add almost, as in I almost broke my arm. The availability of this simple marking strategy in English reduces the motivation to use event verbs in a whole-for-part metonymy. Indeed, a verb used in this way seems pragmatically quite odd—whole-for-part metonymy is almost entirely limited to nouns in English (e.g., the government might be used to refer to ‘the representatives in Congress’, but govern seems unlikely to be used for ‘passing a bill in Congress’). The systematic use of explicit constructions to deny realization in English reinforces the interpretation of basic verbs as fundamentally entailing realization. So even if an English speaker were to use a whole-for-part metonymy with a verb, the final realization of an event is the component perhaps least likely to be dropped. For Tamil speakers, use of a whole-for-part metonymy with verbs is functionally and grammatically well motivated. When Tamil lacks a lexical item meaning exactly the first part(s) of a certain kind of event, such as breaking, but without the final STATE that is associated with it (broken), speakers can still conveniently apply a verb (e.g., transitive ‘break’) to refer to this sequence. Of course, they cannot apply this strategy when they use a construction (such as the converb construction) that entails realization, as this would be contradictory. When whole-for-part

15.

EVENT REALIZATION IN TAMIL

349

metonymy is a potential interpretation, the speaker can explicitly indicate that this interpretation is indeed intended by denying the realization (e.g., adding ‘but it did not break’). Without such denial, the hearer is free to assume that the verb is probably being used to refer to the entire event. Because there is unlikely to be a lexical item for ‘doing the breaking/killing/etc. behavior without the actual broken/dead/etc. final state’, this whole-for-part metonymy is 21 a useful strategy when quickly communicating about these atypical ‘partial’events. In careful speech, one might avoid such a metonymy because it relies on the absence of marking to indicate the less typical event circumstance. A safer communicative strategy would be to specify explicitly which components of an event occur. Under this account, Tamil and English retain essentially the same verbal semantics. What varies is the motivation to use language in a specific way. To explain this variation, we can appeal to the general patterns of verbal clause structure in each language. I have argued that the explicit denial of realization in Tamil is associated with speakers’ use of a metonymic process whereby a verb is used to refer to only a subpart of the verb’s complete semantic representation. If this argument is correct, this process should in principle be available in all languages, albeit to different degrees. Speakers of a language like English would use this metonymy only when there is a more extreme reason to do so, because—given the availability of devices like almost and started to—there is little grammatical motivation for it. But some marginal examples can be imagined; for example, with great furrowing of the brow, some speakers will accept He threw the ball to her, but didn’t exactly let go of the ball at the last second, especially in situations where the speaker repeats someone else’s verb, for example: 30. A: I hope your son threw the ball to her! B: Well, he did throw the ball to her, but [pause, scrunched eyes] didn’t exaaaactly let go of the ball at the last second.22

As a further example, an exasperated colleague was recently overheard to say I kept sending that email, but it wouldn’t send, in which the first instance of send presumably refers essentially to the clicking of a “send” button that is a salient and controlled DO-portion of the entire sending of the email message. 5. LEARNABILITY REVISITED Children obviously must learn the general features of the verbal system of their language at a reasonably early age. Wittek’s data suggest that 4- to 5-year-old children do not necessarily presume that verbs entail realization. I suggest that there may be a universal tendency to sometimes use change-of-state verbs for unrealized events when the speaker does not have a ready alternative expression. German-speaking children learn to suppress this tendency as they become increasingly proficient in German. Because Tamil adults need not presume that realization is entailed, we

350

PEDERSON

must assume that there is something specific to learning German (or English) that teaches learners of these languages that realization is indeed entailed. If we were to hypothesize that what is being learned is simply the correct meaning of each verb, then we should expect children to arrive at adult judgments of appropriate and inappropriate denial of realization on a verb-by-verb basis. But if, as I suggest, there is an overarching systematicity to the adult patterns, then children learning German, English, Tamil, Mandarin, or Japanese should reveal changing sensitivity to general facts about verb usage in their particular languages. As the German- and English-speaking children learn the explicit and productive markers of denial of realization (almost, etc.) that occur in the same clause as the verb, they are encouraged to presume that verb meanings generally entail realization unless this is explicitly canceled. Accordingly, however plausible it might initially have seemed to refer to an event of near-breaking with the verb break, the child learns that this is grammatically inappropriate. Learning the verbal system in German or English includes learning that lack of realization will be explicitly marked in the clause. This makes denial of realization parallel to negation. Tamil-speaking children learn just the opposite. As they acquire the basic converb clause, they learn that it entails realization of the syntactically subordinate verb; when a verb occurs as a simple main verb, in contrast, it is not necessarily realized. If realization is denied, it is denied in a separate finite sentence, not in the same clause as the verb in question. That is, it is a particular interpretation of the previous sentence that is denied, not—as with almost or negation—the simple verb’s semantics. As a result, children learning Tamil may actually use simple verbs for unrealized events increasingly often over time. German- and Tamil-speaking children need not acquire different understandings of events or of the semantics of the simple verbs. Rather, they gain a languagesensitive appreciation of what is an allowable extended use of the verbs and what is not. With respect to realization, a basic pattern of verbal clauses teaches Tamilspeaking children that they can be flexible with their verbs in a way that is grammatically denied to German-speaking children. NOTES 1

The Tamil data were elicited from various speakers of the main (Broadcast standard) dialect of the Madurai District in South India. All examples were checked with at least two native Tamil speakers. I have presented only the more consistently accepted examples, though there may still be individual variation in acceptability. In the phonemic transcription of the Tamil examples, the capital letter ‘T’ represents a retroflexed oral stop /
/. ‘N’represents the corresponding retroflexed nasal stop. ‘R’ represents a (typically alveolar) trill / r / which is not retroflexed. ‘z’ represents a retroflexed approximant /  /. ‘L’ represents a retroflexed //. The following glossing conventions are followed:

15. -

EVENT REALIZATION IN TAMIL morpheme boundary

.

gloss separator for indicating grammatical category of a morpheme

1/2/3

First/Second/Third person

?

Question marker

Acc

Accusative

Adj

Adjectivizer

Caus

Causative

CoV

Converb

Dat

Dative

Dist

Distal

Gen

Genitive

Inf

Infinitive form

Inst

instrument

intr

intransitive

Loc

Locative

m

masculine

Neg

Negative

Nml

Nominalizer

Nom

Nominative

Obj

Object

Obl

Oblique

Pl

Plural

Pr

Present

Prtv

Perfective

Ps

Past

Ref

Reflexive

Resp

Respective form

s

singular

tr

transitive

2

351

Complex events and complex verbs may have their own (complex) realizations, but in this chapter, I will only worry about simple verbs. Both Talmy (1991) and I use the term “realization” for this phenomenon, although we provide different analyses, as discussed below. 3 Here I follow McCawley (1973), who gives the example of almost having scope over either DO CAUSE or STATE in Bill almost killed Harry. There are many variants on the semantic notation used here, but these need not be discussed in this context.

352

PEDERSON

4

While translation equivalents of defeated realization are readily found in other languages (notably Mandarin and Japanese), I do not wish to suggest that the account I give for Tamil is necessarily correct for these languages as well. It is not even certain that such translation equivalents exemplify the same phenomenon in each language. 5 Also, if the general simple verb construction of Tamil simply meant ‘try to verb’ (or some variant of that notion), we would expect fairly consistent interpretations by all speakers that all verbs do not entail realization. Instead, we find that different verbs allow differing degrees of realization, with speakers not necessarily agreeing in their judgments. I will only consider less controversial examples below. 6 For example, I have found none in my small (under 50,000 words) database of spontaneous narratives and conversations. 7 Gropen, Pinker, Hollander, and Goldberg (1991) established that children of roughly 4 to 6 years old understand fill as referring to a manner of moving fluids rather than to an event resulting in a full container. 8 All translations were arrived at in consultation with native speakers: V. Krishnaswami (monolingual Tamil), J. Rajasekaran (Tamil/English bilingual), and Venkatapathy A. Vidya (Tamil/English bilingual). 9 In contrast, English enter and come in are not felicitously used if the subject steps through the doorway of a room but stops without a further step. 10 These tended to be older participants who were apparently overwhelmed by the experimental setting, the urbanized Tamil assistant, and the foreign researcher bearing exotic electronic equipment. A common cultural pattern is to answer almost any question with a “yes” unless you are confident that “no” is appropriate in the context. 11 The negative form of the verb is tenseless for past, present, and definitive future time. Since the tense marking only indicates occurrence relative to the present (and not aspect or realization), the absence of tense marking for the negative (a non-event) is quite consistent. 12 The Mandarin facts are considerably more complicated than can be mentioned in this chapter on Tamil. 13 The simple progressive use of koNTiru is not as common in colloquial Tamil as the progressive aspect is in English. koNTiru is more commonly used, along with a temporal marker, to indicate that events described by different clauses are simultaneous. Realization is entailed in such cases as well: *avan en-kal

viiTTu-kku

va-nt-u

3sm 1-Pl.Obl house.Obl-Dat come-Ps-CoV

potu when

naan

avan-uTaiya

pustakan-kaL-ai

eri-ttu-kkoNTiru-nt-een

I

3sm-Gen

book-Pl-Acc

burn(tr)-CoV-koNTiru-Ps-1s

aanaal

anta

pustakan-kaL

eri-ya-villai.

but

that

book-Pl

burn(intr)-Inf-Neg

*‘When he came to my house, I was burning his books, but the books didn’t burn/light.’

15.

EVENT REALIZATION IN TAMIL

353

14

I have had difficulty finding Japanese speakers who concur with Ikegami’s judgments. 15 Vendler’s (1967) four classes of verbs are Stative (unbounded durative), Activity (unbounded durative), Accomplishment (bounded durative), and Achievement (bounded punctual). 16 Many stative verbs in English (e.g., sit) are actually change-of-state verbs in Tamil, which require a stative auxiliary verb to semantically specify ‘being in the state of having sat down’ or ‘sitting’. Of course, such a state can commonly be inferred from the simple use of the change-of-state verb. Analogously, I seated myself down in the office need not be followed by the phrase and I remained sitting. 17 Formal semantics and pragmatics have long relied on a strict distinction between entailment (with no notion of variable strength) and implicature (with variable strength), but the entailment and implicature facts must be specified in the lexical entry for each individual verb. 18 This converb construction differs from co-verb constructions in languages which typically apply to restricted lexical sets (more commonly referred to as serial verb constructions). 19 The verb glossed with the Tamil viTu in (26) derives from a verb meaning ‘leave’, which can be sensible literally in this example. However, in many other examples viTu has an unambiguously auxiliary verb function. (See Herring, 1988, for a discussion of the textual function of viTu.) Many of the second (or third) verbs in these verbal sequences have auxiliary functions, as we shall see. 20 The reflexive derives its meaning historically from a verb meaning ‘to take’. That is, the first verb becomes syntactically subordinate to ‘take’ and the result of the first verb ‘is taken’ by the subject of the main verb. 21 While I do not claim competence with the Mandarin facts, we might analyze Mandarin similarly. Mandarin has a resultative verb construction to mark (even different types of) realization for the verb. These function similarly to the converb construction in Tamil. This stock of constructions in Mandarin grammar may give Mandarin speakers greater pragmatic freedom to also use verbs metonymically for unrealized or partially realized actions/events for which there is no lexical item. 22 Example courtesy of David Wilkins (p.c.). APPENDIX: THE TAMIL QUESTIONS FOR VIDEO PRESENTATION (TAPE 1 ORDER) The Tamil is given here in written standard dialect, but was pronounced in (Madurai) spoken dialect. FORMAT: Code Number.*

Pre-video statement … Post-video question.

13. a-nta paiyan oru peeppar maTi-kkiR-aan. … peeppar maTi-tt-aan-aa? Dist-Adj boy a paper fold.tr-Pr-3sm … paper fold.tr-Ps-3sm-? ‘The boy (will) fold a piece of paper. … Did he fold the paper?’

354

PEDERSON

14. paiyan meecai-yaic cuttam paNNu-kiR-aan. … cuttam paNN-in-aan-aa? boy table-Acc clean do-Pr-3sm… clean do-Ps-3sm-? ‘The boy (will) clean the table. … Did he clean the table?’ 11. paiyan naaRkaali-yaic cuRRu-kiR-aan. … cuRR-in-aan-aa? boy chair-Acc turn.tr-Pr-3sm … turn.tr-Ps-3sm-? ‘The boy (will) turn the chair around. … Did he turn (it) around?’ 8. paiyan oru mezukuvartti-yai uut-i aNai-kkiR-aan. … uut-i aNai-tt-aan-aa? boy a candle-Acc blow-CoV extinguish.tr-Pr-3sm … blow-CoV extinguish.tr-Ps3-sm-? ‘The boy (will) blow out a candle. … Did he blow it out?’ 1. paiyan oru klaas-il nirappu-kiR-aan. … niRapp-in-aan-aa? boy a glass-Loc fill-Pr-3sm… fill-Ps-3sm-? ‘The boy (will) fill a glass. … Did he fill it?’ 9. oru laari-pommai-yai cuttam paNNi-kiR-aan. … cuttam paNN-in-aan-aa? a lorry-doll-Acc clean do-Pr-3sm …clean do-Ps-3sm-? ‘The boy (will) clean the toy truck. … Did he clean it?’ 5. paiyan katav-ai aTai-kkiR-aan. … aTai-tt-aan-aa? boy door-Acc close.tr-Pr-3sm … close.tr-Ps-3sm-? ‘The boy (will) close the door. … Did he close it?’ 2. paiyan oru peey-yaik kol-kiR-aan. … peey-yai ko-nR-aan-aa? boy a demon-Acc kill-Pr-3sm … demon-Acc kill-Ps-3sm-? ‘The boy (will) kill a demon. … Did he kill the demon?’ 4. paiyan oru taTTu uTai-kkiR-aan. … uTai-tt-aan-aa? boy a plate break.tr-Pr-3sm … break.tr-Ps-3sm-? ‘The boy (will) break a plate. … Did he break (it)?’ 12. paiyan oru laari-pommai-yait taLLu-kiR-aan. … taLL-in-aan-aa? boy a lorry-doll-Acc push-Pr-s3m … push-Ps-3sm-? ‘The boy (will) push a toy truck. … Did he push (it)?’ 7. paiyan oru pazam paRi-kkiR-aan. … paRi-tt-aan-aa? boy a fruit pick-Pr-3sm… pick-Ps-3sm-?

15.

EVENT REALIZATION IN TAMIL

355

‘The boy (will) pick a fruit. … Did he pick (it)?’ 10. paiyan alamaari-kk-uuL nuzai-kiR-aan. … nuzai-nt-aan-aa? boy closet-Dat-Inside enter-Pr-3sm … enter-Ps-3sm-? ‘The boy (will) enter the closet. … Did he enter (it)?’ 3. oru aaL-ai ezupp-iR-aan. … ezupp-in-aan-aa? a person-Acc wake.tr-Pr-3sm … wake.tr-Ps-3sm-? ‘The boy (will) wake someone. … Did he wake (him)?’ *Note that the trial code number refers to the original numbering of Wittek’s stimuli. Trial 6 was not used with Tamil participants.

REFERENCES Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991). Syntax and semantics in the acquisition of locative verbs. Journal of Child Language, 18, 115–151. Herring, S.C. (1988). Aspect as a discourse category in Tamil. In S. Axmaker, A. Jaisser, & H. Singmaster (Eds.), Proceedings of the 14th Annual Meeting of the Berkeley Linguistics Society (pp. 280–294). Berkeley, CA: Berkeley Linguistics Society. Ikegami, Y. (1985). ‘Activity’—‘accomplishment’—‘achievement’—a language that can’t say ‘I burned it but it did not burn’ and one that can. In A. Makkai & A.K. Melby (Eds.), Linguistics and philosophy: Essays in honor of Rulon S. Wells (pp. 265–304). Amsterdam: John Benjamins. Klein, W., & Hendriks, H. (1995). Assertion marking, temporal particles, and resultative verb constructions in Chinese. Unpublished manuscript, Max Planck Institute for Psycholinguistics, Nijmegen. Klein, W., Li Ping, & Hendriks, H. (2000). Aspect and assertion in Mandarin Chinese. Natural Language and Linguistic Theory, 18, 723–770. McCawley, J. D. (1973). Syntactic and logical arguments for semantic structures. In O. Fujimura (Ed.), Three dimensions of linguistic theory (pp. 259–376). Tokyo: Institute for Advanced Studies of Language. Pederson, E. (1990). The Tamil auxiliary kol: A study in polysemy. Paper presented at the 12th Annual South Asian Languages Analysis Roundtable, Berkeley. Tai, J. H.-Y. (1984). Verbs and times in Chinese: Vendler’s four categories. In D. Testen, V. Mishra, & J. Drogo (Eds.), Proceedings of the 20th Annual Meeting of the Chicago Linguistic Society (Parasession on lexical semantics), 289–296. Talmy, L. (1991). Path to realization: A typology of event conflation. In L. A. Sutton, C. Johnson, & R. Shields (Eds.), Proceedings of the 17th Annual Meeting of the Berkeley Linguistics Society (pp. 480–519). Berkeley, CA: Berkeley Linguistics Society, University of California. Vendler, Z. (Ed.). (1967). Linguistics in philosophy. Ithaca, NY: Cornell University Press.

Author Index

A Abbot-Smith, K., 9 Ackerman, F., 74 Agbodza, M. K., 235 Akhtar, N., 8, 21 Allen, S. E. M., 11, 173, 175, 194, 195, 196, 197, 204, 205, 207, 208, 209 Ameka, F. K., 105, 228, 232, 235, 242, 251 Anderson, S. R., 91 Ariel, M., 192 Ashby, W. J., 173 Asher, R. E., 136 Austin, P., 217 Avolonto, A., 217 Ayeke, K., 236 B Baayen, H., 12, 16 Baker, C. L., 279, 280 Baldwin, D., 8, 205 Bamberg, M., 321 Barlow, M., 293 Bassano, D., 45 Bavin, E., 11 Berman, R., 156, 182 Bever, T. G., 256, 292 Bickel, B., 182 Bloom, P., 193 Bloomfield, L., 74

Bohannon, J. N., 279 Bohnemeyer, J., 32, 45, 51, 52, 53, 62, 65, 66, 103, 104, 108 Boogaart, R., 266 Borer, H., 258, 274 Borik, O., 262, 263 Boroditsky, L., 32, 183 Botne, R., 251 Bowerman, M., 4, 5, 8, 12, 29, 154, 163, 181, 185, 279, 280, 282, 289, 291, 292, 293, 294, 295, 302, 303 Braine, M. D. S., 5, 279, 287, 288, 289, 294, 302 Bricker, V. R., 32, 45 Brinkmann, U., 8 Brisson, C., 314, 327 Brooks, P. J., 5, 12, 193, 280, 287, 288, 289, 294, 295, 301, 302 Brown, D. J., 231 Brown, P., 11, 32, 45, 52, 89, 102, 103, 108, 168, 170, 171, 174, 182, 183, 186, 187, 191 Brown, R., 31, 43, 279 Bruce, R. D., 32 Budwig, N., 10, 11, 174 Burzio, L., 215 Bybee, J., 113, 114, 273, 293, 294 C Campbell, A. L., 193 357

358

AUTHOR INDEX

Capo, H. B. C., 235 Carpenter, K. L., 320 Carstairs-McCarthy, A., 113 Casenhiser, D. M., 5, 18, 170 Cassidy, K., 11, 19 Chafe, W., 192, 195, 196 Chen, J., 22 Choi, S., 11, 45, 168 Chomsky, N., 2, 3, 29 Clancy, P. M., 10, 11, 16, 173, 186, 192, 193, 194, 195, 196, 207, 208 Clark, E. V., 168, 169, 233, 249, 293, 294, 311, 318, 320 Clements, G. N., 235 Comrie, B., 136, 147, 255 Cooper, A., 193 Croft, W., 1, 12, 20, 33, 45, 280, 293, 294 Cruse, D. A., 280, 293, 320 D Danziger, E., 35, 36, 38, 43, 44, 45, 182, 219 de Hoop, H., 260 de Swart, H., 255, 261 DeHart, G., 295 DeLancey, S., 43, 216, 219, 226, 227, 238 De León, L., 168 Demetras, M. J., 279 Demuth, K., 11 Deutsch, W., 320 Di Nubila, J. A., 8 Dixon, R. M. W., 43, 111, 114, 115 Dobson, V., 143 Dodson, K., 280, 295, 302 Dowty, D. R., 1, 43, 52, 66, 257, 266, 309, 319, 327 Dryer, M. S., 76, 119 Du Bois, J. W., 173, 192, 196 Durie, M., 234 Duthie, A. S., 235 E Ehrich, V., 317, 327 Eisengart, J., 8 Elliott, S. R., 44 England, N. C., 44 Epstein, T., 12, 17, 232, 233, 249 Erteschik-Shir, N., 1, 12 Essegbey, J., 219, 228, 229, 235, 236, 237, 238, 242

F Farrar, J., 205 Fernandes, K. J., 8 Fiawoo, F. K., 243 Filip, H., 259, 261, 262, 263 Fillmore, C. J., 74, 96 Fisher, C., 7, 8, 9, 10, 17, 21, 32, 69, 153, 328 Fodor, J. A., 256 Foley, W. A., 32, 43 Frei, H., 74 Fretheim, T., 192 G Gahl, S., 294 Garrett, M. F., 256 Gaskins, S., 183 Gentner, D., 32, 34, 45, 69, 183, 310, 313 Gerard-Ngo, P., 295 Gergely, G., 292 Gertner, Y., 8 Geurriero, S., 193 Gillette, J., 6 Givón, T., 192, 207 Gleitman, H., 6, 12 Gleitman, L. R., 6, 7, 8, 9, 11, 12, 14, 15, 17, 19, 21, 30, 31, 32, 49, 50, 51, 58, 69, 74, 84, 112, 132, 141, 143, 147, 151, 152, 154, 158, 159, 160, 191, 310, 316, 325, 326, 328 Goldap, C., 56 Goldberg, A. E., 1, 5, 18, 20, 74, 80, 89, 91, 96, 97, 99, 100, 108, 169, 170, 184, 213, 214, 228, 237, 280, 293, 294, 348 Goldberg, R., 5, 6, 8, 18, 280, 283, 310, 313, 352 Goldin-Meadow, S., 8 Gopnik, A., 45 Green, G. M., 282 Green, J., 143 Greenfield, P., 193 Grice, H. P., 312, 313 Grimshaw, J., 1, 3, 11, 20, 21, 33, 64, 65, 258 Gropen, J., 5, 6, 8, 12, 17, 18, 232, 233, 249, 280, 283, 310, 313, 315, 352 Gudeman, R., 295

AUTHOR INDEX Gundel, J., 192 Guo, J. S., 343 H Haiman, J., 74 Hale, K., 1, 86, 280 Hall, D. G., 7, 9, 17, 21, 32, 69, 328 Hamann, C., 193 Hampe, B., 74 Hampson, J., 45 Hanlon, C., 279 Harkins, J., 143 Henderson, J., 142, 143 Hendriks, H., 331, 333, 339, 347 Herring, S. C., 331, 343, 353 Hill, D., 136, 143 Himmelmann, N. P., 234 Hirsh-Pasek, K., 279 Hockett, C. F., 74 Hoddinott, W. G., 86 Hoff-Ginsberg, E., 7, 69 Hofling, A., 32 Hollander, M., 5, 6, 8, 18, 280, 283, 310, 313, 352 Hollebrandse, B., 170 Hopper, P. J., 33, 293 Hyams, N., 193 I Ikegami, Y., 18, 331, 341, 342, 343, 344, 353 Ingham, R., 288 J Jaakkola, R., 8, 21 Jackendoff, R., 1, 5, 54, 62, 66, 314 K Kail, M., 312 Kako, E., 316 Kaufman, T., 55, 66 Kay, P., 74, 96 Keenan, E., 1 Kemmer, S., 293 Kessler Shaw, L., 45 Keyser, S. J., 1, 280, 327 Kiparsky, P., 261, 275 Kirsner, R. S., 74

359 Kita, S., 90, 104, 108 Klein, W., 331, 333, 339, 347 Kofod, F. M., 86 Krämer, M., 65 Krifka, M., 62, 257, 261, 274 Kruger, A., 205 Kumari, T. C., 136 Kumpf, L. E., 173 Kuntay, A., 186 Kuriyama, Y., 193 L Lambrecht, K., 74 Landau, B., 6, 49, 50, 51, 58, 69, 75, 316 Langacker, R. W., 33, 74, 293 Langdon, M., 226 LaPolla, R. J., 1 Laughren, M., 86, 285 Leach, E., 136 Leach, J., 136 Lederer, A., 6 Lee, J. N., 11, 160 Lehmann, C., 45, 55, 56, 65, 66 LeSourd, P., 74 Levin, B., 1, 5, 7, 12, 43, 52, 54, 84, 91, 100, 215, 216, 219, 220, 221, 223, 233, 280, 283, 285, 286, 298, 299, 302, 304, 314, 315, 316, 317 Levinson, S. C., 95 Lewis, L. B., 280, 295, 302 Li, P., 12, 331, 339, 347 Lidz, J., 12 Lieven, E., 8 Loeb, D., 288, 292, 293 Lois, X., 32 Lucy, J. A., 44, 45, 65, 66, 67, 181 Lyytinen, P., 268, 269, 270, 271, 276 M MacFarland, T., 233 Machobane, M., 11 MacWhinney, B., 280, 293, 294, 321 Manfredi, V., 228 Marantz, A., 21 Maratsos, M., 32, 34, 43, 295 Marcus, G. F., 8, 279 Margetts, A., 119, 123, 128, 131, 136 Martin, S., 114 Mazurkewich, I., 282

360

AUTHOR INDEX

McCawley, J., 301, 325, 351 McGregor, W. B., 86 Mel’
uk, I., 113 Merlan, F., 86 Michaelis, L., 74 Mithun, M., 214, 219, 228 M»ynarczyk, A., 261, 262 Mohanan, T., 74, 80 Moloi, F., 11 Morgan, James L., 279 Morgan, Jerry L., 319 Murdoch, G. P., 231 Murty, L., 10, 11, 174 Mylander, C., 8 N Naigles, L. R., 7, 8, 11, 12, 69, 160, 316, 327 Nappa, R., 11, 19 Narasimhan, B., 10, 11, 174 Nash, D., 86 Nelson, K., 45 Newman, J., 111, 136 Ninio, A., 169, 170, 185 Nwachukwu, P. A., 217 O Obianim, S. J., 235, 239, 243 O’Connor, M. C., 74, 96 Oehrle, R. T., 282 Oetzel, R., 135 Oetzel, S., 135 Oshima-Takane, Y., 193 P Papafragou, A., 11, 19 Pawley, A., 74 Peck, C., 44, 45 Pederson, E., 310, 315, 326, 331, 343 Perlmutter, D. M., 1, 44, 213, 214, 215, 216 Peters, A., 319 Pfeiler, B., 168 Pine, J., 8 Ping, L., 294, 331, 347 Pinker, S., 1, 3, 4, 5, 6, 8, 11, 12, 14, 18, 21, 30, 31, 34, 49, 50, 58, 59, 60, 62, 63, 64, 66, 84, 91, 100, 106, 112, 132, 147, 152, 158, 159, 279, 280, 282, 283, 284, 285,

286, 287, 288, 289, 290, 291, 292, 293, 294, 295, 299, 300, 301, 302, 309, 310, 313, 314, 315, 317, 318, 325, 326, 352 Plunkett, K., 193 Post, K. N., 279 Postal, P. M., 215 Pustejovsky, J., 1, 12 Pye, C., 21, 288, 292, 293 Q Quine, W. V., 31 R Rakowitz, S., 7, 9, 17, 21, 32, 69, 328 Ramchand, G., 258 Randall, J., 5, 12, 16, 21, 280 Rapoport, T., 1, 12 Rapp, I., 316, 317 Rappaport (Hovav), M., 1, 5, 7, 12, 43, 52, 54, 84, 91, 100, 215, 216, 219, 220, 221, 223, 233, 283, 285, 286, 298, 299, 302, 304, 314, 315 Ravin, Y., 227 Regier, T., 293, 294 Rispoli, M., 10, 11, 154, 160, 182 Robertson, J. S., 45 Roeper, T., 327 Rosen, C., 1, 215, 216, 221 Ross, M., 111 Ruhl, C., 55, 74 S Sapir, E., 32 Sasse, H. J., 32, 33 Schmitt, C., 258 Schneiderman, M., 279 Schoorlemmer, M., 261, 262, 263 Schröder, H., 173, 194, 208 Schultze-Berndt, E., 70, 71, 75, 83, 84, 86, 234 Schumacher, L., 12, 17, 232, 233, 249 Schutze, H., 294 Serratrice, L., 195, 196, 206, 207, 208 Sethuraman, N., 5, 18, 170 Shannon, T., 51 Shirai, Y., 12 Silverstein, M., 143

AUTHOR INDEX Simpson, J., 86 Skarabela, B., 205, 207, 209 Slobin, D. I., 12, 156, 183, 186, 256, 273, 319 Smailus, O., 32 Smith, C., 255, 266 Smith, J. H., 193 Smith, L. B., 183 Snow, C. E., 279 Stanowicz, L., 279 Stoll, S., 269, 270, 271, 276 Stolz, C., 66 Straight, S. H., 45 Swadesh, M., 32 Syder, F. H., 74 T Tai, J. H.-Y., 341 Talmy, L., 18, 50, 54, 55, 58, 60, 89, 91, 94, 104, 313, 314, 331, 332, 343, 344, 345, 351 Tardif, T., 45, 168, 182 Taylor, J. R., 74 Tenny, C. L., 1, 12, 258, 275 Tesucún, F. F., 32 Thompson, S. A., 33 Tomasello, M., 5, 8, 9, 12, 18, 134, 154, 193, 205, 280, 288, 293, 294, 295, 301, 302 Travis, L. L., 279 Treiman, R., 279 Trueswell, J. C., 11, 19 Tsunoda, T., 86 Turner, M.-M., 143 U Ulrich, M., 35, 36, 38, 44, 45 Ulrich, R., 35, 36, 38, 44, 45 Uziel-Karl, S., 182

361 V Valian, V., 193 van Hout, A., 12, 16, 170, 258, 259, 262, 264, 274, 275, 276 Van Valin, R. D., Jr., 1, 21, 43, 51, 215, 216, 219, 221, 222, 227, 233 Vapnarsky, V., 32 Vendler, Z., 43, 52, 255, 314, 324, 326, 342, 353 Verbeeck, L., 45 Verkuyl, H., 257, 258, 261, 275 Vinnitskaya, I., 269, 270, 271 von Stechow, A., 327 Vouloumanos, A., 8 W Weissenborn, J., 12, 16, 313 Weist, R. M., 12, 268, 269, 270, 271, 276 Westermann, D., 235 Wexler, K., 193, 269, 270, 271 White, L., 282 Whorf, B. L., 29, 30, 33, 34, 40, 41, 43, 44, 248 Wichmann, S., 248 Wierzbicka, A., 231, 239 Wilkins, D. P., 11, 136, 143, 163, 191, 227, 232, 353 Wilson, S., 86 Wittek, A., 11, 22, 311, 313, 314, 317, 319, 321, 334, 335, 336, 349, 355 Wunderlich, D., 65 Wysocka, H., 268, 269, 271, 276 Z Zaenen, A., 51 Zavala, R., 66 Zwicky, A., 74

Language Index

A Acehnese, 216, 233–4 Akan, 251 Arrernte, 15, 20, 136, 141–162, 191, 232, 245, 248, 249 B Barai, 136 C Caddo, 214 Chinese, 168, see also Mandarin Choctow, 215 Cindali, 251 Czech, 262 D Dutch, 17, 51, 182, 214, 221, 245, 247, 248, 256–75 passim E English, 2, 4, 7–19 passim, 30–1, 34, 36, 40, 45, 50, 52–5 passim, 59–66 passim, 70, 74, 78, 84–5, 91, 96, 97, 103, 111–112, 113, 119, 134, 143, 147, 151, 152, 154, 161, 167, 169, 170, 171, 181–5 passim, 192, 214–8 passim, 221, 229, 231–4 passim, 245, 246,

248, 249, 251, 255–69 passim, 271–3 passim, 275, 280–302, 309–11, 313, 315–17, 325, 326, 327, 331–4, 338–50, 352, 353 Ewe, 17, 213, 217–29, 231–46, 248–251 F Finnish, 17, 251, 256–7, 260–4, 268–9, 271–3, 275 Fon, 217 French, 192, 217, 251, 255 G German, 8, 18, 22, 51, 182, 192, 309–13, 315–17, 320–7, 334–6, 338–40, 348, 349–50 Germanic, vii, 256, 272, see also Dutch, English, German H Hausa, 251 Hebrew, 169, 171, 247 Hindi, 10, 11, 174 I Igbo, 217, 228 Inuktitut, 11, 16, 104, 173, 191–3, 194, 196–9, 204–9 Italian, 51, 192, 195–6, 214, 215, 216, 256 363

364

LANGUAGE INDEX J

Jaminjung, 14, 69–73, 75–86 Japanese, vii, 10, 11, 14, 89–104, 106–8, 114, 192, 245, 247, 251, 331, 341–4, 350, 352, 353

P Polish, 17, 256–7, 261–3, 268–72, 275, 276 Proto-Oaxaca Mixean, 248 R

K Korean, 10–11, 16, 168, 173, 186, 192, 193–4, 196, 251 Kriol (northern Australia), 73 Kwa, 228, see also esp. Ewe

Romance, 51, see also French, Italian, Spanish Russian, 17, 256–7, 261–2, 263, 269, 270–2, 275 S

L Lakhota, 228 Latin, 51 Lhasa Tibetan, 226 Likpe, 14, 105, 108 Longgu, 136

Saliba, 14–15, 111–36 Sesotho, 11 Slavic, 256, 261–3, 271–3, 275, see also Polish, Russian Spanish, 52, 59–60, 62, 65, 66, 186 T

M Malayalam, 136 Mandarin, 11, 18, 22, 182, 331, 333, 339–45, 347, 350, 352, 353 Mayan, 11, 32, 45, 55, 66, 168, 180–2, 183, see also Mopan Maya, Tzeltal Maya, Yukatek Maya Mohawk, 228 Mopan Maya, 11, 13, 33–45 Mparntwe Arrernte, see Arrernte N Ngaliwurru, 86 Norwegian, 251

Tamil, 18–19, 310, 315, 318, 326, 331–50, 352–5 Thai, 251 Tsez, 136 Tsova-Tush, 216 Turkish, 186, 256 Tzeltal Maya, 11, 14, 16, 45,102–3, 167–87, 191, 192 U Uto–Aztecan, 248 W Warlpiri, 86, 232

O Y Oceanic, vii, 114, 122, see also Longgu, Saliba Oluta Popoluca, 245, 246

Yoruba, 251 Yukatek Maya, 13, 44, 49–65, 66, 67, 103–4, 106, 107, 108

Subject Index

A Adjective and attribute as universal linkage, 30, 32, 34 causativization of in acquisition of English, 280, 296, 297, 302n.2, 303 and ‘die’ verbs, 248 in Mopan Maya, 36–37 as problematic category, 43.n3 and unaccusativity, 215 inYukatek Maya, 52 See also Form class universals Adverbial modification cue hypothesis, 310, 319–325, 327nn.5,6 Aktionsart, see Aspect, lexical Arguments vs. adjuncts in Arrernte, 147–150, 159–160 in language acquisition, 21n.5 Argument ellipsis in Arrernte, 146 and language acquisition, 10–11, 13, 15–16, 20, 21, 160, 184, 191, 206 and verb specificity in Tzeltal, 167, 171–175, 179–181, 183–184 See also Argument realization; Preferred argument structure (PAS) Argument realization and accessibility, 317n.2

and deictic gestures, 200, 204–205, 209n.19 and discourse-pragmatic factors in child Inuktitut, 192–206, 207n.7, 208nn.8,11,12,13 and eye gaze, 195, 200, 209n.16 statistical patterns and language acquisition, 11, 19–20 in Arrernte, 152–155, 159–161, 163n.9 and verb specificity in child Tzeltal Maya, 16, 167–187 See also Argument ellipsis; Preferred argument structure (PAS) Argument sharing in Jaminjung complex verbs, 70, 74–75, 79, 80, 81, 82 Argument structure alternations, 4–6 See also Broad-range rules; Causative(-inchoative) alternation; Dative alternation; Locative alternation; Narrow-range rules Argument structure, defined, 1 Argument structure and event structure, 12–13 Argument structure and learnability: controversies and questions, 7–12 Aspect acquisition of, 39, 255–256, 263–274, 275n.11, 276n.20 and event structure, 12 grammatical, 255 lexical (Aktionsart), 12, 43, 45n.19, 215–216, 222, 251, 255–256, 365

366

SUBJECT INDEX

257, 258, 263, 271, 273, 274n.1, 275n.6, 314, 317, 318, 327n.1, 331, 339–343, 353n.15 See also Telicity Atelic verbs, see Telicity B Broad-range rules (Pinker’s model), 284–285, 286, 289, 290–292, 302nn.2,3 violations of in causative alternation, 289–292 C Causative(-inchoative) alternation acquisition of in English, 4, 18, 279–302 children’s errors, 18, 280–281, 288–304, 303nn.4,5, 304nn.2,3 and direct vs. indirect causation, 285, 290, 291, 292, 295, 301 in Ewe, 220–224 and internal vs. external causation, 84, 216–229, 289, 290–292, 303, 304 in Jaminjung, 83 See also Broad-range rules; Narrowrange rules Caused locative change construction in Japanese, 97–100 Caused motion construction, 96–98 See also Caused locative change construction in Japanese Change-of-location verbs in Yukatek Maya, 49, 54–55, 58, 60–65, 66n.7, 67n.12 Change-of-state verbs causative transitive, 83, 236 acquisition of, 18–19, 309–326; See also Adverbial modification cue hypothesis and entailed endstate, 18, 310, 315, 318, 319, 321–326 and pragmatically favored endstate, 314–319, 325, 326, 327n.2 in resultative verb compounds in Mandarin, 22n.7, 353n.21

in Tamil, 310, 315, 326, 331–350, 352n.4, 353n.16 and weak endstate interpretation, 310–316, 325, 326, 327n.5 See also Caused locative change construction in Japanese; Event realization intransitive, 49, 216; See also Changeof-location verbs in Yukatek Maya; Verb meanings, ‘die’ verbs Complex predicates, 74 in Jaminjung, 21, 69–85 particle verbs, 74; See also under Telicity resultative verb compounds in Mandarin, 22n.7, 353n.21 Converbs in Tamil, 346–347, 348, 353n.18 Concrete object semantics and nouns, 32–33, 34, 37, 40, 41, 42, 43, 45n.19 Construction grammar approach, 20, 89, 294 in Ewe, 228, 235–245, 249, 250, 251nn.6,7 to Figure–ground indeterminacy in Japanese, 96–100 to complex verbs in Jaminjung, 70, 74 and light verbs in language acquisition, 170 to motion event expressions inYukatek Maya, 51, 66nn.4,7 Count/mass syntax, 17, 257, 260, 261, 263, 275nn.6,10 Coverbs, see Complex predicates in Jaminjung Cultural factors in adult and child language, 17, 20, 21, 159, 161, 183, 231–232, 252n.10 D Dative alternation, 119, 280, 281, 282, 283 children’s errors, 280, 281–282 lexical rule for, 284 E Ellipsis, see Argument ellipsis; Argument realization Endstate, see under Change-of-state verbs, causative transitive

SUBJECT INDEX Entrenchment, 18, 280, 293, 294–295, 296, 302 Ergativity/ergative markers, 15, 21n.4, 72, 142, 143, 147, 149, 154, 194, 172, 174, 186n.11, 187n.13 Errors and predicted errors in child speech in Arrernte, 157–158, 160–161, 163nn.8,10 in English, 4–5, 18, 279–282, 287, 288–304, 303nn.4,5, 304nn.2,3 in Mopan Maya, 39–40, 42, 45n.21 in Saliba, 112, 132, 137n.18 Event category labeling, 58–60 Event realization, 310, 331–334, 339–349, 352nn.4,5,13, 353n.21 and language acquisition, 349–350 and metonymy, 348–349, 353n.21 See also Change-of-state verbs, entailed endstate; Converbs in Tamil Event structure, 1, 12–13, 16, 258 F Figure–ground indeterminacy, 89–107 Form class universals (noun, verb, adjective), 30–31, 32–33, 43n.3 and Mopan Maya, 34–38 G Gesture, 133, 200, 204–205, 209n.19 H Home-sign, 8 I Implicature, 50, 62, 245, 313, 319, 353n.17 Innate knowledge, 2–9, 18, 29–30, 31, 32, 40, 42, 43, 50, 84, 107, 183, 231, 233, 234, 273, 280, 282–287, 292–293 See also Form class universals; Semantic bootstrapping; Semantic and syntactic bootstrapping, challenges, universality of alignments; Semantic-syntactic linking; Syntactic bootstrapping

367 Intransitive verbs active–inactive languages, 214 in Ewe, 216–229 and internal/external causation, 84, 216, 223–224, 299 split intransitives, 216 See also Causative(-inchoative) alternation; Unaccusative/ unergative verbs J Joint attention, 203, 205, 206, 209n.16 L Lexical rules for changing verb meaning, 283–286 Light verbs, 74 and verb learning, 169–170, 171, 176–178, 180, 181, 184, 185n.4 Linking rules; see Form class universals; Semantic bootstrapping; Semantic–syntactic linking; Syntactic bootstrapping Locative alternation, 4–5, 91, 280, 281, 317 children’s errors, 45, 280, 281 See also Caused locative change construction in Japanese M Manner bias, 311 Motion event encoding, 19 and semantic/syntactic bootstrapping, 50–51, 58–60, 63–65 in Yukatek Maya, 49–58, 61–63, 67nn.11,12 N Narrow-range rules (Pinker’s model), 284–286, 288, 289, 292, 293, 295, 296–299, 301, 302; See also Schema/semantic class induction Negative evidence, 106, 279–280, 282, 286, 289 Noun, see Concrete object semantics and nouns; Count/mass syntax; Form class universals

368

SUBJECT INDEX P

Particle verbs, see under Complex predicates Perfectivity and imperfectivity, 255–256 and case marking, 275n.6 and event realization, 344–345, 347 See also under Telicity Pragmatically favored endstate verbs, see under Change-of-state verbs Preemption, 293–294, 295–296, 299, 300, 302 Preferred argument structure (PAS) in child language, 173–175, 184, 196 R Resultatives, 22n.7, 213–214, 221, 256, 317, 339, 344, 348, 353n.21 S Schema/semantic class induction, 5, 18, 280, 293, 295, 299, 300, 301, 302 Semantic bootstrapping, 3–6, 7, 13, 14, 19, 20, 21n.2, 31–33; See also Form class universals; Semantic and syntactic bootstrapping, challenges; Semantic–syntactic linking Semantic structure hypothesis testing, 316, 317–319, 325 Semantic and syntactic bootstrapping, challenges action semantics and form class in Mopan Maya, 13, 34, 38, 39–40, 41–42 argument ellipsis in adult speech, 10–11, 15–16, 20, 21, 160, 184, 191, 206 complex verbs in Jaminjung, 75–76, 78, 84–85 ‘give’ verbs in Saliba, 14–15, 132, 137n.18 mismatch between number of participants and number of arguments, 19 motion event encoding in Yukatek Maya, 13–14, 49–51, 57–60, 63–64 multiple frames, 7, 11–12

one-participant verbs in Ewe, 16–17, 229, 232–235, 240–241, 243, 249–250 ‘put’ and ‘look’ in Arrernte, 15, 19, 21n.5, 141–161 universality of alignments, 9–10, 141, 158–161 “weak endstate” problem, 316–317, 319, 325 Semantic–syntactic linking, universals of, 1–2, 3, 4, 5, 6, 8, 9–10, 14, 18, 21.n2, 32–38, 43, 43n.3, 49, 50, 58, 84, 132, 141, 158–161, 232, 283, 286, 287; See also Form class universals; Semantic bootstrapping; Semantic and syntactic bootstrapping, challenges; Syntactic bootstrapping Stem suppletion, Saliba ‘give’ verbs, 111, 112, 113–117, 131, 136n.5 Syntactic bootstrapping, 3, 6–7, 9, 19, 20, 21n.2, 31–33, 69, 141, 184, 186n.13, 328n.9 See also Form class universals; Semantic and syntactic bootstrapping, challenges; Semantic–syntactic linking T Telicity, 13, 17–18, 20, 43, 255–263, 275nn.3,5,10 acquisition of, 256, 263–274, 276n.20 and argument realization, 258 compositional vs. predicate telicity, 18, 256, 259–261, 263–274 and count-mass noun syntax, 17, 260, 261, 275n.6 and (im)perfectivity, 17–18, 255–256, 261–263, 266, 269, 270, 271, 272, 273, 275, 275nn.6,9,10, 276n.20 and incremental theme, 258–260, 263, 275n.5 and particle verbs in acquisition, 259, 263, 268, 273, 275n.10 and relevance (Bybee), 273 temporal adverbials as diagnostics for, 222, 258, 275n.3, 314, 326, 327n.1 and transitivity, 259, 275n.5 Temporal adverbials, see under Telicity

SUBJECT INDEX Transitivity and ‘give’ verbs in Saliba, 111–134 in language acquisition, 7, 11, 12, 16, 17, 18–19, 21n.2 and number of arguments, 1, 11, 17, 160 and one-participant verbs in Ewe, 213–229 and telicity, 17–18, 259–260, 261, 268, 275n.4 See also Causative(-inchoative) alternation; Change-of-state verbs, causative transitive; Event realization; Semantic– syntactic linking; Syntactic bootstrapping; Verb meanings; ‘die’ verbs Typological bootstrapping, 183 U Unaccusative/unergative verbs, 12, 17, 44n.14, 213–217, 221–224, 229, 233, 300 Universality, see Form class universals; Innate knowledge; Semantic and syntactic bootstrapping, challenges; Semantic–syntactic linking Usage-based approach to language acquisition, 18, 154 and acquisition of the causative (-inchoative) alternation, 292–302 See also Entrenchment; Preemption; Schema/semantic class induction V Verb, see Verb classes; Form class universals; Verb learning; Verb meanings Verb classes of intransitives in Ewe, 217–229 in Jaminjung, 84 in Mopan Maya, 44n.14 in Yukatek Maya, 51–55, 65n.1 See also Aspect, lexical (Aktionsart); Change-of-location verbs; Change-of-state verbs; Intransitive verbs; Transitivity; Unaccusative/ergative verbs; Verb meanings

369 Verb learning and frequency, 15, 63, 142, 144, 153–154, 160, 163n.7, 170–171, 295, 296–298 and multiple cues, 19, 20 and semantic specificity, 167, 171–175, 179–181, 183–184 structure mapping as early strategy, 9–10 See also Adverbial modification cue hypothesis; Syntactic bootstrapping Verb meanings bodily process verbs, 52, 215, 216, 219, 220, 221, 225 contact verbs, 7, 83, 105, 314, 316, 317, 327n.3 ‘die’ verbs, 17, 49, 53, 55, 83, 215, 231–253 ‘eating’ verbs in Tzeltal Maya, 16, 168–169, 171–172, 174, 176, 177–178 ‘enter’ verbs, 53, 56, 58, 62, 100, 103–104, 106, 107, 335–336, 352n.9 ‘give’ verbs, 14–15, 19, 76, 82, 111–134, 144, 168–169, 184, 185n.4, 186n.10 ‘look/see’ verbs in Arrernte adult language and language acquisition, 15, 19, 141–161 manner verbs in language acquisition, 6–7, 21n.6, 185n.2, 288, 297, 300, 301, 303, 310–311, 314–315, 318, 327n.4, 352n.7 in various languages, 44n.14, 52, 54–55, 59, 60, 63, 66n.4, 73, 115, 144, 177, 182, 185n.2, 215, 220, 234, 243, 285, 300, 314, 315, 317, 318, 346 motion verbs, see Motion event encoding perception verbs, 7, 143, 151, 152; See also ‘look/see’ verbs positional verbs, 44n.13, 45n.18, 52, 53, 102–103, 168, 169, 170 posture verbs, 52, 53 psychological activity verbs, 285, 292, 299, 304

370 ‘put’ verbs, 19, 70, 141–164 in Arrernte adult language and language acquisition, 15, 19, 141–161 and figure–ground indeterminacy, 90, 91, 97, 98, 100, 101, 103, 106 in Jaminjung, 70, 75–81, 83, 85, 86n.7 ‘run’ and related verbs, 13, 30–31, 38–42

SUBJECT INDEX transfer verbs, see Verb meanings, ‘give’ verbs, ‘put’ verbs W Weak endstate interpretation, see under Change-of-state verbs Whorfian hypothesis, 29–30, 33, 34, 38, 40–41, 43, 44n.4 Wieder/again, 19; See also Adverbial modification cue hypothesis