Language acquisition studies in generative grammar: papers in honor of Kenneth Wexler from the 1991 GLOW workshops

LANGUAGE ACQUISITION STUDIES IN GENERATIVE GRAMMAR

LANGUAGE ACQUISITION & LANGUAGE DISORDERS

EDITORS

Harald Clahsen University of Essex

William Rutherford University of Southern California

EDITORIAL BOARD

Melissa Bowerman (Max Planck Institut für Psycholinguistik, Nijmegen) Patricia Clancy (University of California at Santa Barbara) Werner Deutsch (Universität Braunschweig) Kenji Hakuta (Stanford University) Kenneth Hyltenstam (University of Stockholm) Peter Jordens (Free University, Amsterdam) Barry McLaughlin (University of California at Santa Cruz) Jürgen Meisel (Universität Hamburg) Anne Mills (University of Amsterdam) Csaba Pleh (University of Budapest) Michael Sharwood Smith (University of Utrecht) Catherine Snow (Harvard University) Jürgen Weissenborn (Max Planck Institut für Psycholinguistik, Nijmegen) Lydia White (McGill University) Helmut Zobl (Carleton University, Ottawa)

Volume 8

Teun Hoekstra & Bonnie D. Schwartz (eds) Language Acquisition Studies in Generative Grammar

LANGUAGE ACQUISITION STUDIES IN GENERATIVE GRAMMAR PAPERS IN HONOR OF KENNETH WEXLER FROM THE 1991 GLOW WORKSHOPS

Edited by

TEUN HOEKSTRA Leiden University

BONNIE D. SCHWARTZ University of Durham

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA 1994

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data Language acquisition studies in generative grammar : papers in honor of Kenneth Wexler from the 1991 G L O W workshops / edited by Teun Hoekstra, Bonnie D . Schwartz. p. cm. -- (Language acquisition & language disorders : ISSN 0925-0123; v. 8) Based on papers presented at workshops on syntactic acquisition organized as part of the 14th G L O W Conference which was held Mar. 1991, Leiden University. Includes bibliographical references and indexes. Contents: Introduction : on the initial stages of language acquisition / Teun Hoekstra & Bonnie Schwartz - VP, null arguments, and C O M P projections / Nina Hyams -Crosslinguistic evidence for functional projections in early child grammar / Vivian Deprez & Amy Pierce ~ The seeds of structure : a syntactic analysis of the acquisition of case marking / Harald Clahsen, Sonja Eisenbeiss & Anne Vainikka — From adjunct to head / Teun Hoekstra & Peter Jordens — Early null subjects and root null subjects / Luigi Rizzi — Asking questions without CPs? : on the acquisition of root wh-questions in Bernese Swiss German and Standard German / Zvi Penner ~ Successful cyclic move­ ment / Rosalind Thornton & Stephen Crain - Early acquisition of scrambling in Japanese / Yukio Otsu - Direct access to X'-theory : evidence from Korean and Turkish adults learning German / Anne Vainikka & Martha Young-Scholten ~ Word order and nominative case in non-native language acquisition : a longitudinal study of (L1 Turkish) German interlanguage / Bonnie D . Schwartz & Rex A. Sprouse ~ Optionality and the initial state in L2 development / Lynn Eubank. 1. Language acquisition. 2. Generative grammar. 3. Grammar, Comparative and gen­ eral-Syntax. I. Hoekstra, Teun. II. Schwartz, Bonnie D . III. Wexler, Kenneth. IV. G L O W Conference (14th : 1992 : Leiden University) V. Series. P118.L2544 1993 401'.93~dc20 93-43090 ISBN 90 272 2472 2 (Eur.) / 1-55619-244-4 (US) (Hb; alk. paper) CIP ISBN 90 272 2473 0 (Eur.) / 1-55619-245-2 (US) (Pb; alk. paper) © Copyright 1994 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O. Box 75577 • 1070 AN Amsterdam • The Netherlands John Benjamins North America • 821 Bethlehem Pike • Philadelphia, PA 19118 • USA

Contents Tables and Figures

vii

Abbreviations

ix

Contributors

xi

Introduction: On the initial states of language acquisition Teun Hoekstra & Bonnie Schwartz VP, Null Arguments and COMP Projections Nina Hyams Crosslinguistic Evidence for Functional Projections in Early Child Grammar Vivian Deprez & Amy Pierce The Seeds of Structure: A syntactic analysis of the acquisition of Case marking Harald Clahsen, Sonja Eisenbeiss & Anne Vainikka From Adjunct to Head Teun Hoekstra & Peter Jordens

1 21

57

85 119

Early Null Subjects and Root Null Subjects Luigi Rizzi Asking Questions without CPs? On the Acquisition of Root wh-questions in Bernese Swiss German and Standard German Zvi Penner

151

Succesful Cyclic Movement Rosalind Thornton & Stephen Crain Early Acquisition of Scrambling in Japanese Yukio Otsu

215

Direct Access to X'-Theory: Evidence from Korean and Turkish adults learning German Anne Vainikka & Martha Young-Scholten Word Order and Nominative Case in Non-Native Language Acquisition: A longitudinal study of (L1 Turkish) German Interlanguage Bonnie D. Schwartz & Rex A. Sprouse Optionality and the Initial State in L2 Development Lynn Eubank Index of Languages Index of Names Index of Subjects

177

253

265

317 369 389 391 397

Tables and Figures

Tables Deprez & Pierce 1. Early Distribution of Negatives in Three Children 2. Percentage of finite forms prior to V2

1 71

Clahsen, Eisenbeiss & Vainikka 1. Suffixes for Case Marking in German 2. Omissions of Determiners 3. Noun Phrases with Adjectives 4. The Acquisition of Structural Case in Finnish

92 99 100 111

Hoekstra & Jordens 1. Distribution of Modal Expressions

131

Thornton & Grain 1. Production of Argument and Adjunct Questions by Individual Subjects 2. Comparison of Downstairs Responses versus Production of Medial-wh Responses 3. Double Dissociation Downstairs Answers versus Medial-wh 4. Comparison of Responses Experiments 2 and 4 Otsu 1. 2. 3. 4.

Reliance on NNV Strategy Various Word Orders with Five Selected Verbs OSV Order in Utterances of Five Two- and Three Year Olds Test Results

235 240 240 247

256 257 258 260

viii

TABLES AND FIGURES

Vainikka & Young-Scholten 1. Implicational Table (to Be Revised) 2. Overt Subjects at the AGRP Stage 3. Revised Implicational Table A. B. C. D. E. F. G-l. G-2. H-l. H-2. I-1. I-2.

Biographic Information and Source of Data (Alphabetical Order) Breakdown of Analyzed and Unanalyzed Utterances with Verbs (Alphabetical Order) Position of the Verb in the VP* (Developmental Order) Bare-VP Sentences and Sentences with More than a VP (Developmental Order) Empty Subjects (Developmental Order) Agreement on Raised Main Verbs (Developmental Order) Distribution of -n and -0 on All Verbs (Early FP Stage) Distribution of -e (ISG) and -t (3SG) on Main Vs (Early FP Stage) Distribution of sein 'to be' (Early FP Stage) Distribution of Modals (Early FP Stage) Proportion of -n and Other Suffixes on Raised Main Verbs (AGRP Stage) Agreement Suffixes (Except for -n) on Raised Main Verbs (AgrP Stage)

Figures 1. Empty Subjects in the Two Types of Sentences 2. Verb Raising and the Specifier Position Tables Schwartz & Sprouse 1. Parametric Differences between Turkish and German 2. Stage 1 3. Stage 2 4. Stage 3 5. Declarative Main Clauses with Two or More Non-Verbal Constitu­ ents 6. Summary of the Parametric Values for Turkish and German Eubank 1. Main Clause Agreement by Affix for Jose 2. Thematic Verb Agreement and Infinitives in SV and VS orders for Jose

278 289 296 300 301 302 303 304 305 306 306 307 307 308 308 280 294

334 338 339 341 341 360

383 385

Abbreviations

Object language is italicized throughout, with a translation in English following in single quotation marks. Additionaly, italics are used for emphasis. Parts of examples may be highlighted by underscoring. An overscore must be read as negating the overscored (e.g., A vs. A), whereas a prime indicates a projection level (e.g., X'). The word case begins with capital C when it refers to grammatical Case. Technical terms are set in small capitals on their first occurrence only. Grammatical categories are printed in full capitals, usually abbreviated; features are represented between angled brackets. Indications of children's ages have the format (years;months;days). In the interlinear glosses, a hyphen is used only if the corresponding English gloss carries a hyphen. When one word in the example corresponds to two words in the gloss, a period separates the two. Abbreviations are printed in small capitals. When an abbreviation corresponds to an affix, the plus-sign is used. If an abbreviation merely specifies the preceding element, it is connected to that element with a colon. Finally, if an abbreviation itself consists of two parts, these parts are separated by a hairline space. Abbreviations Ө  A(D) ADJ ADV AGR ARG

theta (= thematic) phi (= person, number, and gender features) adjective adjunct adverb agreement argument

ASL AUX bv C(OMP) D(ET) DB DO ECP

American Sign Language auxiliary bound variable complementizer determiner discourse binder direct object empty category principle

x

ABBREVIATIONS

FCH full clause hypothesis fin finite FP finite phrase GB government and binding GF grammatical function HMC head movement constraint I(NFL) inflection IO indirect object L1 (A) first language (acquisition) L2(A) second language (acquisition) LC licensing condition LD licensing device LF logical form LLH lexical learning hypothesis N noun nc null constant NEG negation NUM number O(BJ) object

OP P PC PF PLD QR ref S(UBJ) SCH SCP ShCH Spec t TL UG V V2 XP

operator preposition parametric choice phonological form primary linguistic data quantifier raising referential subject small clause hypothesis structural correspondence principle short clause hypothesis specifier trace target language universal grammar verb verb second X phrase (X = N, V, A, D, etc.)

Abbreviations used in interlinear glosses ABIL ACC AGR AOR

ability accusative agreement aorist

NOMIN

nominalizer

OB C L

object clitic

COP DAT DECL DIR ELA FEM GEN IMP INF MASC NEG

copula dative declarative direction elative feminine genitive imperative infinitive masculine negative

NEUT

neuter

NOM

nominative

PART

partitive

PAST

past tense

PBY PL POL PRES PROG PRT

possibility plural polite present tense progressive particle

RED REL SG SUB C L SUBJ SBJU TOP

reduced relative marker singular subject clitic subject subjunctive topic

Contributors

Harald Clahsen University of Essex Dept. of Language and Linguistics Wivenhoe Park Colchester C04 3SQ, United Kingdom e-mail: [email protected]

Lynn Eubank University of North Texas Dept. of English Denton, TX 76203-3827, USA e-mail: [email protected]

Stephen Crain University of Connecticut/ Haskins Laboratories Dept. of Linguistics 341 Mansfield Road Storrs, CT 06259, USA e-mail; crain@uconnvm

Teun Hoekstra Leiden University Dept. of General Linguistics/ Holland Institute of Generative Linguistics PO Box 9515 NL-2300 RA Leiden, The Netherlands e-mail: [email protected]

Viviane Deprez Rutgers University Dept. of Linguistics 18 Seminary Place New Brunswick, NJ 08903, USA e-mail : [email protected]

Nina Hyams University of California Dept. of Linguistics 405 Hilgard Avenue Los Angeles, CA 90024, USA e-mail: [email protected]

Sonja Eisenbeiss University of Düsseldorf General Linguistics Department Universitätsstrasse 1 D-4000 Düsseldorf, Germany

Peter Jordens Free University Amsterdam Dept. of Linguistics PO Box 7161 NL-1007 MC Amsterdam The Netherlands

xii

CONTRIBUTORS

Yukio Otsu Keio University Institute of Cultural and Linguistics Studies 2-15-45 Mita Minato-ku Tokyo 143, Japan e-mail: [email protected] Zvi Penner University of Berne Linguistic Institute Länggasstrasse 49 CH-3000 Bern 9, Switzerland e-mail: [email protected] Amy Pierce 77 Randall Road Princeton, NJ 08540, USA e-mail: [email protected] Luigi Rizzi University of Geneva Dept. of General Linguistics Boulevard des Philosophes, 22 CH-1205 Geneva, Switzerland e-mail: rizzi®uni2a.unig.ch Bonnie D. Schwartz University of Durham School of English and Linguistics Elvet Riverside II, New Elvet Durham DH1 3JT, United Kingdom e-mail: [email protected]

Rex A. Sprouse Harvard University Dept. of Germanic Languages Boyston Hall Cambridge, MA 02138, USA e-mail: [email protected] Rosalind Thornton Massachusetts Institute of Technology Dept. of Brain and Cognitive Science E10-246 Cambridge, MA 02139, USA e-mail: thornton@uconnvm Anne Vainikka University of Massachusetts at Amherst Dept. of Linguistics South College Amherst, MA 01003, USA e-mail: vainikk@umass Martha Young-Scholten University of Durham School of English and Linguistics Elvet Riverside II, New Elvet Durham DH1 3JT, United Kingdom e-mail: [email protected]

Introduction On the Initial States of Language Acquisition Teun Hoekstra Leiden UniversitylNIAS

1.

Bonnie D. Schwartz University of Durham

Introductory Remarks

Generative linguistics has consistently maintained that explanatory adequacy in linguistics be tied to language acquisition. The core of the puzzle for which the notion Universal Grammar provides the conceptual solution has aptly been named the "logical problem of language acquisition" (Hornstein & Lightfoot 1981): with only limited linguistic experience, the child somehow constructs a language-particular knowledge system that essentially converges on the grammar of the adult. Generative grammarians duly acknowledged the logic of the conceptual problem, and in response they primarily concerned themselves — through in-depth (and often comparative) linguistic examinations — with the precise specifications of UG. Yet, while the tradition of this philosophical linking between adult knowledge states and linguistic growth has been espoused for nearly thirty years, it is only relatively recently that we have begun to see active reciprocity between what linguists do and what acquisitionists do. Among the few acquisitionists who have long and steadfastedly sought to secure the ties between formal linguistics and language acquisition stands Ken Wexler. His career is in itself a testimonial to the insights that can be gained from the productive exchange between theoreticians of language and of language development. The range of topics he has addressed, topics which are so often at the heart of both conceptual and empirical questions, speaks to the influence his research has had on the evolution of the field: in linguistic domains from binding theory (e.g., Wexler & Manzini 1987; Chien & Wexler 1990; Avrutin & Wexler 1992), to null subjects (Hyams & Wexler 1993) to functional categories (e.g., Poeppel & Wexler 1993; Wexler 1992); from issues of learnability (e.g., Wexler

2

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

& Culicover 1980; Wexler 1982; Wexler 1993) to the issues of continuity vs. maturation (e.g., Borer & Wexler 1987) — to name but a few. For nearly two decades now, Ken has continually called for well-informed interchange between linguistic theory and language development. And finally, not only has that call been heeded in a serious fashion, but the theoretical perspective may now even dominate language acquisition study. It was thus fitting that Ken be asked to organize the first GLOW workshops devoted to syntactic development. Indeed, one might say that as this was to be the 14th annual GLOW — the leading meeting for generative linguistics in Europe — the time to showcase generative studies on language acquisition was even overdue. The unexpectedly large attendance (if not just the stamina of the participants) might also be taken as evidence for this charge, and it is largely thanks to all who were there that the two workshops, "The Development of Movement and Inflection" and "Wh-movement, ECP and Bound Variable Development", were both met with 'GLOW-ing' success. It was more than unfortunate that after months of preparation, Ken himself was unable to go to Leiden. But of course the intellectual legacy of Ken's many years of research was apparent throughout the presentation and discussion of the 16 papers and 5 commentaries. In light of the encouraging response from attendees, it was decided that there would be no better way to show appreciation of Ken, both for his short-term contribution to structuring the workshops and for his long-term contribution to developing the field, than to put together a volume of papers from those two days. It is thus with much admiration and warmth that we dedicate Language Acquisition Studies in Generative Grammar to Professor Kenneth Wexler. This book, then, is the product of the workshops on syntactic acquisition that were organized as part of the 14th GLOW conference, held in March 1991 at Leiden University. The conference was organized by the members of the "Werkgroep Grammatica", now part of the recently initiated Holland Institute of Generative Linguistics (HIL). It was sponsored by Leiden University, The Foundation for Linguistic Research of the Dutch Organization for Scientific Research (NWO), The Royal Academy of Sciences (KNAW) and the Ministry of Education; their contributions are hereby gratefully acknowledged. We are also especially indebted to the 27 people who conscientiously reviewed the submissions, but whom, to retain their anonymity, we unfortunately cannot thank by name. A word of gratitude should also go to René Mulder, who gave this volume its beautiful appearance.

INTRODUCTION

3

The theme of both the GLOW conference and the associated workshops was "The Role of Heads". The local organizing committee invited Bonnie Schwartz and Ken Wexler to join Teun Hoekstra in the organization of the workshops on syntactic acquisition. This volume represents a selection of the papers that were first presented in those workshops and have now been reviewed, rethought and revised. A fairly broad range of topics and languages is covered in this collection, but the papers are all syntactic studies within the Principles and Parameters framework and tend to fall into two main categories: the development of inflection and movement in relation to functional categories (for both native and non-native language) and the development of XP (A-)movement. In what follows, we briefly highlight the main contribution that each paper offers.

2.

Overview of Papers

Perhaps the most controversial issue in the development of native-language grammars concerns whether grammatical knowledge is available through all stages of development or whether such knowledge increases over time. The first position is sometimes referred to as the CONTINUITY HYPOTHESIS (Hyams 1983; Pinker 1984). Development under this view results from increase in other domains; specifically, lexical elements and values associated with these elements need to be acquired, but grammatical principles as well as grammatical categories as such are antecedently available. Acquisition of grammar, then, is mainly the instantiation of available abstract knowledge with lexical elements, including the setting of parameters associated with these items. The second view, on the other hand, holds that the grammar may grow not only through the acquisition of the lexicon but also by the additions of categories or principles themselves. This view may of course be realized in various, distinct ways. One popular version is the theory of UG-CONSTRAINED MATURATION, proposed by Borer & Wexler (1987), which imposes a set of stages as a result of genetically determined maturational steps. In the papers in this volume, these two opposing views, as well as a variety of intermediate perspectives, are contrasted in regard to the development of inflection and phenomena associated with it, such as various movement processes. A strong maturation view proposes that this development is guided by the maturation of functional categories: initially, these categories are unavailable to the child, who therefore is unable to structure sentences in terms of them. A continuity approach contests such claims. It will be evident that

4

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

several papers directly bear on this question of how much knowledge should be seen as representing the initial state of language development. The papers by Hyams and by Deprez & Pierce advocate the strongest form of the Continuity Hypothesis with respect to the availability of function­ al categories. Hyams first makes the conceptual claim that the Continuity Hypothesis is in fact the null hypothesis, as it allows for the least degree of freedom for the child's grammar to diverge from the adult grammar. She then presents empirical and learning-theoretic evidence in favor of the claim that children have both a CP and an IP from the earliest stages of acquisition onwards. If the CP-system were lacking, it would lead one to expect that German children would generalize verb second to embedded clauses, which is not the case. Secondly, Dutch and German children produce verb-initial clauses when topics are dropped, i.e., in cases of missing arguments. To account for this without a CP-level would require that the child's licensing mechanisms deviate from those of the target grammar, indeed from the requirements imposed by UG. The phenomenon of topic drop leads Hyams to reassess the null subject phenomenon in early stages of acquisition. Given that a null subject in an adult grammar may result from either topic drop or pro-drop, the question arises as to how children know what type of language they are in. In order to answer this question, she proposes a null argument parameter of UG, formulated as "pro may be licensed under Spec-head agreement in an A/A-position," where the A/A-options constitute the parameter. Apart from being licensed, pro must also be identified. This identification may be either by 'rich' AGR (in pro-drop languages) or by a discourse topic (in topic-drop languages). The latter should be the default value, if English children's null subjects indeed involve topic identification. The most natural assumption would furthermore be that the Ä-option of licensing pro correlates with topic-identification, just as the A-option of licensing pro correlates with AGR-identification. The problem for this assumption is the fact that children's early English, unlike Dutch and German, appears to allow only null subjects, not objects; this is unexpected if the A-system is involved. Hyams therefore argues that the A-value of the null argument parameter is the default, bringing children in a pro-drop language immediately to the correct value, as AGR is a suitable identifier in these languages. In English, then, pro is licensed in Spec-IP as well, but it cannot be identified, unless it moves to Spec-CP where it may be topic-identified. Under this scenario, null objects are not allowed, as they lack licensing. Dutch/German children, on the other hand, have more robust evidence for the A-option of the null argument parameter, because of the verb second

INTRODUCTION

5

character, which the child is aware of at a very early age. Indirectly, then, the discussion of null subjects corroborates the early availability of both IP and CP. A further argument for the availability of CP is built on the absence of specific types of mistakes in early Icelandic which are predicted under an alternative scenario in which there is no CP/IP distinction at the earlier stages. Hyams therefore concludes that the premise that functional categories are missing if functional items are missing is difficult to maintain given the fact that children have syntactic operations involving functional categories at the point at which they fail to reliably produce functional elements. What needs to be learned, under this perspective, is the missing lexical items. Deprez & Pierce adduce a different kind of evidence, from early English, French and German, for the STRONG CONTINUITY HYPOTHESIS. They look at the position of the verb vis-à-vis other elements in the sentence, specifically the subject and negation. They contend that subjects may remain in VP-internal position, which should not be taken as the absence of an IP-projection: rather, the position of the verb, as well as its position relative to negation, indicates that VP-external categories must be available to allow the formation of head-chains. Early French and early English differ in this regard along the same lines as the adult languages: inflected main verbs in early French precede negation as well as the VP-internal subject, showing that IP and NEGP dominate VP, with the verb raising to INFL. In early English, the negation precedes both the VPinternal subject and the verb, but — as in the adult language — modals consistently precede negation, indicating again that the INFL position, dominat­ ing NEG, is available in the grammar of these English children. Their interpreta­ tion of the available German data contrasts with Hyams': they conclude that the children make a distinction between finite and nonfinite verbs, evidencing the availability of (clause-final) INFL in German, but that movement of the finite verb to COMP is optional at first. This lack of movement to COMP is extended to English subject-AUX inversion, which they reinterpret as resulting from non-movement of the subject to Spec-IP, leaving the AUX in INFL, rather than having raised to COMP. Deprez & Pierce move from the uniform availability of both CP and IP to the problem of the optionality of head movement to COMP and of subject raising to Spec-IP — which should be contrasted with the fact that both wh-movement and movement of the finite verb to INFL are targetlike from the beginning. They argue that this difference reflects the difference in the nature of UG principles. According to Deprez & Pierce, some principles have a unique way of being satisfied, while others allow distinct ways of satisfaction. So, V-to-I movement, either at S-structure or at LF, is the only way in which a variable

6

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

becomes available for Tense. Case assignment, on the other hand, may involve either Spec-head agreement or government. Thus, optionality is linked either to parameters which permit UG principles to be satisfied in distinct ways or to language-particular constraints; these appear to be set after some period of delay in which the child's grammar manifests simultaneously the concurrent options made available by UG. If, in contrast, a unique resolution is dictated by UG, the child converges immediately on the correct value for the target grammar, and therefore optionality is excluded. Less strict adherence to the Continuity Hypothesis is advocated in the papers by Clahsen, Eisenbeiss & Vainikka, Hoekstra & Jordens and Rizzi. Clahsen et al. investigate the acquisition of Case marking in German and Finnish. They use their findings to advocate a WEAK CONTINUITY approach to the development of phrase structure, specifically as instantiated by functional categories. They argue that positions of phrase structure are constructed gradually, based on X'-theory and the input data. For the domain studied in the paper, i.e., Case marking, they assume this process is guided by two UG principles: first, Case assignment under either government or Spec-head agreement; second, the STRUCTURAL CORRESPONDENCE PRINCIPLE (Vainikka 1989), according to which there is a one-to-one correspondence between structural positions and structural Cases. As for the acquisition of Case markings, they assume that the child may be guided via semantic bootstrapping. The specific assumptions they make regarding phrase structure positions are that the complement position is automatically posited together with the corresponding head, whereas specifier and adjunct positions as well as functional heads and their projections are established only through positive evidence. With respect to the order of development of accusative, dative and genitive Case in German, their assumptions lead to the prediction that they will be acquired in that order, as accusative is the Case of the unmarked verb complement, dative is also a complement Case (but one that is semantically restricted), while genitive is assigned to the specifier of DP. Their findings for German do not support the prediction, as the genitive is acquired prior to accusative and dative. This unexpected result can be explained, they contend, by closer examination of the realization of Case marking in German. There is a fundamental difference between genitive and the other Case markings: genitive Case is expressed by means of an -s suffix on the noun; accusative and dative are expressed via the determiner as well as through agreement on adjectives. It is then claimed that the acquisition of genitive -s is the trigger for constructing the determiner system. Prior to the postulation of this determiner system, determiners, genitives and

INTRODUCTION

7

adjectives occur in complementary distribution. Moreover, determiners and adjectives show a high percentage of agreement errors in this period. Clahsen et al. account for these facts by the assumption that all three occupy the same structural position, viz. Spec-NP. Once genitive -s triggers the postulation of the determiner system, determiners become obligatory, combinations of determiners and prenominal adjectives occur, and agreement errors drop. It is only then that accusative Case can be realized, as its realization depends on the availability of the determiner system. To assess whether this account of the unexpected acquisition order is correct, Clahsen et al. next turn to the acquisition of Case marking in Finnish, where the different Cases are not marked in different ways. They take the position that in Finnish, partitive is the complement Case, genitive the structural Case of specifiers, and elative the structural Case of adjuncts. Their prediction, then, is that partitive will be acquired first (since complements are automatically given phrase structure options), genitives arise if specifier positions arise, and elatives will be acquired later still. This prediction is borne out. Hoekstra & Jordens argue that while functional categories may be available in principle, their presence may be optional at stages of linguistic development. The specific claim they advocate is that elements instantiating functional categories in the adult language do not necessarily instantiate these categories in the grammar of the child; rather, various elements which will ultimately find phrase structure positions in projections of functional categories may at first be misrepresented before these categories become obligatory. In particular, they argue that such elements may initially be combined with lexical projections in terms of a general adjunction schema. They thus challenge two claims: first, that evidence that a category X is available at stage S implies that the category is always present; second, that an element instantiating a phrase structure position in the projection of X in the target grammar instantiates that same position in the child's. This proposal is in some sense an intermediate position between the Strong Continuity Hypothesis (as advocated by Hyams and by Deprez & Pierce) and the Weak Continuity Hypothesis of Clahsen et al. As one piece of evidence for their hypothesis, Hoekstra & Jordens discuss the acquisition and interaction of modals and negation in Dutch before the age of 2. They argue that at that stage, the elements expressing modality and negation do not instantiate the categories COMP and NEG but rather are adjoined to infinitival projections, even though COMP (as well as INFL) is available in the child's grammar. A second piece of evidence involves nominal phrases at the age of 2 to 3. They essentially agree with Clahsen et al. concerning the lack of a determiner system at the relevant age. However, instead of assuming that

8

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

elements combining with N occupy Spec-NP, they argue that these elements adjoin to the N-projection. The motivation for these analyses is based on the acquisition of SCRAMBLING in Dutch, which is the correct placement of objects in what is called the MIDDLE FIELD. Scrambling interacts with the determiner system, which develops gradually. During this development, a phenomenon of SUBSCRAMBLING is observed, via which parts of the nominal object are scrambled out of it. This subscrambling is neither in accordance with the target grammar nor permitted by principles of UG if the structure of nominal phrases were taken to uniformly implicate the DP-system. Similar subextraction phenomena are found with adjectival modifiers, involving elements which in the adult system are placed in positions made available through functional categories, specifically Degree Phrases. The paper offers some speculations on the factors that lead the child to revise the initial hypotheses so as to converge on the target grammar. In that respect, they follow a proposal made in Rizzi's contribution. Rizzi likewise maintains that both the Strong Continuity Hypothesis and a Maturation Hypothesis of functional categories are too strong. He instead argues for a relaxation of the obligatoriness of functional categories in earlier stages, thereby allowing the absence of a particular level of functional categories at points in which they nevertheless are in evidence. This proposal thus provides an alternative to the Deprez & Pierce hypothesis concerning variation in UG principles as a way to account for optionality in children's grammars with respect to phenomena where the target grammar allows no optionality. The principal empirical domain Rizzi investigates concerns the types of null subjects also discussed in Hyams' contribution. He retains the idea that early null subjects manifest a genuine grammatical option, as Hyams (1986) proposed, but provides an alternative interpretation of this option when compared to standard pro-drop as manifested in languages such as Italian. In essence, Rizzi is in agreement with Hyams' idea of topic-licensed null arguments, but his analysis is ultimately rather different. According to Rizzi, the major configurational property of the early null subject in, e.g., English is its occurrence in the specifier of the root. This configurational constraint is not specific to transitional systems in acquisition, as it is equally operative in certain styles in the adult language, such as diary English. In this respect, English early null subjects are fundamentally different from those in Italian as well as from those in early Italian (which match adult Italian). To explain the early null subjects (as well as their diary counterparts), Rizzi extends the class of empty categories, following Lasnik & Stowell (1991), to include a NULL CONSTANT (nc). The nc is like a variable, in being Ā-bound, but differs from it in that it is not a variable but a null definite description, which

INTRODUCTION

9

can only be bound by a non-quantificational empty operator. As such, it features in various empty operator constructions, such as easy-to-please and parasitic gap constructions. Postulating this nc in the grammar of acquirers of English raises the question of why this option is not available in normal adult English. Rizzi proposes that the nc is possible only if the specifier of the root is an A-position. This option can be realized only if IP is the root, as then its specifier is an A-position. In adult English, this is impossible, as CP is obligatorily present; hence its specifier is present, which is not an A-position. The only thing specific to child/diary English under this perspective is that CP is not obligatory. This accounts for the fact that early English allows no null subjects in wh-preposing constructions or in embedded clauses. Optionality, then, is accounted for in terms of the optional presence of functional categories, specifically COMP. Rizzi suggests that in the relevant systems other functional categories may likewise be optionally absent, yielding the root small clauses typical of early stages of acquisition. Yet, their occurrence cannot be taken as evidence for the unavailabil­ ity of functional categories, as adherents to a Maturation Theory would maintain. Penner also assumes that CP is initially available to German-speaking children, but he claims that no elements are allowed to surface in either the head position or its specifier. Starting from this premise, he addresses the question of how children go about satisfying universal licensing requirements which involve these positions in the adult grammar, specifically in constituent questions. The paper focuses on root constituent questions, which in the adult languages are locally licensed at the CP-level in terms of Spec-head agreement. He argues that the child is forced to resort to compensatory strategies both to satisfy the communicative need to ask questions and to circumvent the problems posed by the unavailability of satisfying the licensing requirements in CP. Children acquiring German and Bernese Swiss German use two alternative routes in asking questions. One pattern, the INVERSION pattern, either involves no wh-marking at all or displays a particle-like question marker attached to a copular verb in clause-initial position. Penner claims the structure does not involve an operator at all. The interrogative chain is wh-marked on a head in a scope position and involves a slot for the null constant (Rizzi, this volume, discussed above). This, perhaps prototypical question form, is probably inter­ preted as "replace the nc by one x, such that y." What must be licensed in this pattern is the nc. Given the lack of an operator, the licensing conditions of operators are not activated at all. The second pattern of question formation is what Penner calls the "V-end" pattern. This pattern is analyzed as a relativeclause-like structure, involving in fact a root question in the guise of an

10

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

embedded clause. The conclusion of the paper is that no default strategy is available to the child to circumvent the licensing requirements. The child is aware of the licensing requirements and extends available mechanisms so as to satisfy them. Two other papers report on experimental studies but address much the same issue, i.e., whether grammatical knowledge is continuous or is made available only at a later stage. Thornton & Crain advocate Continuity in regard to the availability of the apparatus involved in long wh-movement in early English, taking issue with de Villiers, Roeper & Vainikka (1990). In previous work (Crain & Thornton 1991; Thornton 1990), also based on production experiments, Crain and Thornton concluded that successive cyclic movement is continuously available in the grammar of children. In addition to targetlike question forms, English children also produce other varieties of long-distance questions, utilizing either the PARTIAL MOVEMENT strategy or the COPYING strategy, which are found in other languages. The use of these alternative strategies was not taken as an indication that the grammars lack the successive-cyclic option, since targetlike questions are found as well. Their conclusion was challenged by de Villiers et al., who, on the basis of children's non-adult responses to long-distance questions in comprehension experiments, took the position that the acquisition of wh-questions involves at least three distinct stages, the first two of which lack the option of successive cyclic movement. At stage 1, wh-questions involve a mechanism of co-indexation, with the empty category taken as pro rather than a trace resulting from movement. At the next stage movement can only be local, leading to partial movement structures typical of Romani and certain varieties of German (cf. McDaniel 1989). Only at the third stage do children have access to the mechanism of successive movement to generate long-distance questions. The first stage is motivated by two observations: First, children exhibit a lack of inversion in their initial w/z-questions, which would follow if the wh-constituent is basically like a topic. Second, children assign coreferential interpretations to Who does he think has a hat?, which, if a variable were involved, would be a Principle C violation; taking the empty category to be pro, rather than wh-trace, avoids this problem. A key result of the comprehension experiment is the occurrence of the WRONG-QUESTION-DOWN responses given to How did Kermit ask who to paint?, where children answered the who rather than the how. De Villiers et al. interpret this finding as the result of partial movement of who, with how serving as a scope marker. They extend this account from these comprehen­ sion results to the partial movements observed by Crain and Thornton in their

INTRODUCTION

11

experimental production data; each of these results is said to be explained if the child can move wh-words only locally at the second stage. This constitutes the background to the Thornton & Crain contribution here in which they report on four experiments carried out to settle the issues that arise if the two positions are compared. The first experiment demonstrates that at least 4-year-olds may produce long-distance adjunct questions. They take this as an indication that children at that age can apply wh-movement successive cyclically, as there is no other way to satisfy the ECP for adjunct extractions. The second experiment investigated the validity of the claim that children who give the wrong-question-down responses are the ones producing partial movement structures. The findings show that this correlation does not hold. The third experiment was designed to test the generality of the wrong-question-down phenomenon. Children were asked to answer questions like What did he say was in the box?. If what is a mere scope marker and the real wh-element is moved no further than the embedded clause, they should answer what was in the box rather than what he said was in the box (the child knew both what was in the box and what the he-person said was in the box). Performance on this task was error-free, a result that is again construed as evidence for successive movement. The children in this experiment were 3-year-olds who had indeed given the wrong-question-down answers in the second experiment. The fourth experiment was set up to show that wrong-question-down answers to questions of the type How did he say what Gummy Bear was eating? might have resulted from the lack of an adequate answer to the how in the pragmatic context. This turned out to indeed be the case: if the context provides a suitable answer to the how question, the number of wrong-question-down answers decreases dramatically. This is unexpected if the wrong-question-down answers were the only option made available by the child's grammar at the relevant stage. The results of all four experiments thus support the Continuity Hypothesis. Otsu investigates whether scrambling is part of the grammar of 3-year-old acquirers of Japanese. Examination of naturalistic data indicates that children are indeed able to apply scrambling, ordering the arguments in correspondence with the given-new distinction. This might seem surprising in light of the results of Hayashibe's (1975) influential experiment, which were taken to show that Japanese children — in apparent disregard of Case markings, and hence of scrambling — interpret sentences by making use of a perceptual strategy. The strategy would be that thematic relations are determined by word order, interpreting NNV (Japanese being an SOV language) as agent-patient-verb, independent of the Case marking on the nouns. This might be construed as an

12

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

indication that scrambling is not part of their grammar. Otsu demonstrates that the results are in fact an artifact of the experiment. A more careful design of essentially the same experiment demonstrates that the children by and large exhibit error-free performance. This result, then, most clearly indicates that scrambling is indeed part of the grammatical knowledge of very young Japanese children. This is an important finding, especially since Hayashibe's results have been used to argue in favor of non-syntactic accounts of early grammars, such as Pinker's (1984) theory of semantic bootstrapping. There is of course a common ground in acquisition studies on native language (L1) and non-native language (L2). Both seek to document and explain the expansion of linguistic systems. Nonetheless, the foundational premise for L1 acquisition research is still at issue in the field of L2 acquisition: this concerns whether Universal Grammar can be 'reactivated' in (particularly adult) Interlanguage development. The three papers here on non-native language acquisition all address this issue by concentrating on the acquisition of German word order, with special attention given to verb placement. One way or another, previous research on the L2 acquisition of German motivates each study. In these earlier L2 German investigations, the native language of the acquirers was Romance (Italian, Portuguese or Spanish). The data indicate that these (un­ tutored) acquirers initially posit a head-medial VP (VO) — contrary to the headedness of VP in German (OV) as well as contrary to any stage that children pass through in their acquisition of German as a native language. The question arose as to the source of this Interlanguage VO system: from L1 influence (e.g., duPlessis, Solin, Travis & White 1987) or from some sort of 'canonical word order strategy', as proposed by Clahsen & Muysken (1986, 1989). The answer various researchers gave to this question also lined up with their conclusion about the possibility of 'UG accessibility': those claiming no L1 influence also hypothesized that UG was inoperative in (adult) L2 acquisition; those maintain­ ing that the VO system was the result of L1 'transfer' also argued that UG was available, as evidenced by the possibility to reset the headedness parameter for VP to OV at a later stage. This is the backdrop against which the following set of papers should be placed. So, each study here, while ultimately in accord with the claim that UG is operant in L2 acquisition, seeks to specify its role, especially in view of the properties of the L1 grammar. Again, all three papers concur that the L1 plays a definite role in the creation of Interlanguage systems. Where they differ is in the claimed extent of L1 influence. Thus despite the consistency of topic across these papers, each looks at very different data and each ends up representing a distinct theoretical point of view.

INTRODUCTION

13

The Vainikka & Young-Scholten paper is a cross-sectional study of 17 adult native speakers of Korean and Turkish, and their focus is on the develop­ ment of phrase structure at the earliest stages. Their investigation was born from the desire to compare the previous research on German L2 acquisition by L1 Romance speakers with what speakers of OV languages do in the acquisition of German. The canonical word order strategy of Clahsen & Muysken (see above) predicts that in early stages these L2 acquirers, like the L1 Romance speakers, should exhibit a head-medial VR What Vainikka & Young-Scholten find, contrary to the Clahsen & Muysken prediction, is that the earliest data exhibit an OV order. Only at the next stage does the verb begin to precede the object, and this happens before the full range of verbal agreement is acquired at the third stage, which is also the point at which null subjects are consistently absent. The analysis they provide of these three stages combines a traditional approach to L1 transfer with recent hypothesizing on L1 development of phrase structure (Clahsen, Eisenbeiss & Vainikka, this volume). Their main claim is that L1 influence is restricted to lexical projections, and so the earliest stage is characterized as a bare VP whose OV order carriers over from the L1. At stage 2, a functional projection above the VP is built. Similar to certain proposals on L1 acquisition of German (Clahsen 1991), this functional projection is underspecified and head-medial, and its head is able to serve as the landing site for (optional) verb movement. This functional projection is specified as AGRP at the third stage, which is where Vainikka & Young-Scholten attempt to argue that such an account ties together the three characteristics of this stage: that verbal morphology is consistently targetlike, that V regularly precedes the rest of the material in the VP, and that null subjects are now so rare. They further suggest that even at this stage, there is little (overt) evidence for the existence of a CP. At the theoretical level, the Vainikka & Young-Scholten paper should also be of interest to the fields of both syntax and native language development. If their analysis is right, it points to dissociating verb movement from verbal morphology. On the other hand, they claim that the connection between verb movement and raising an NP follows from what they term the FULL HOUSE PRINCIPLE (cf. Vainikka 1989), which states that movement to a head position licenses the specifier position. In regard to implications for L1 acquisition, Vainikka & Young-Scholten claim that their findings add support to non-maturational accounts of L1 development: as maturation cannot be responsible for the development of phrase structure in adult L2 acquisition, the similarities of phrase structure development among L1 and adult L2 acquirers considerably weaken

14

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

explanations of native-language growth that are maturationally based (cf. e.g., Radford 1990). The interest of the Schwartz & Sprouse paper also lies in determining the effect of the structure of the L1 grammar on L2 acquisition, but theirs is a longitudinal study, tracking the development of German word order by an adult native speaker of Turkish over a 26-month period. Contrary to leading paradigms in L2 acquisition research, they suggest that it is inappropriate to analyze L2 systems from the perspective of 'match' or 'mismatch' with the target language. Rather, their goal is to determine whether Interlanguage reflects an internal consistency that is formally statable in the language of Universal Grammar. To this end, the inquiry focuses on the position of the verb at three stages. At stage 1 (perhaps equivalent to the second or even third stage in Vainikka & YoungScholten), Cevdet produces only (ADV)SVO orders; that is, there are no instances of SUBJECT-VERB INVERSION, even when a non-subject is found in clause-initial position. Subject-verb inversion marks the second stage — however, this XVS pattern is limited to when the subject is a pronoun; when the subject is non-pronominal, the order is XSV. It is only at the third stage that subject-verb inversion occurs with both types of subjects. Yet even at this point, Cevdet continues to produce what are superficially 'verb third' and 'verb first' patterns, in contrast to the verb second input to which he is exposed. In approaching these data, Schwartz & Sprouse defend a theoretical position which sharply contrasts with that of Vainikka & Young-Scholten: they assume a model in which the L1 grammar in its entirety characterizes the INITIAL STATE OF L2 ACQUISITION; as UG itself is also available, input which at one point is irreconcilable will later force the adoption of other options of UG. Specifically, regarding the three stages of verb placement, they analyze the data by appealing to the different mechanisms UG makes available for checking nominative Case: head-government, specifier-head agreement, and subject-clitic incorporation. In accordance with their theoretical model, they assume that the complete structure of the Turkish clause is available from the beginning, and this includes a head-medial CP, for which they claim there is evidence in Turkish. Capitalizing on this structure, their analysis of the early SVO pattern consists of three parts: first, the verb moves to COMP; second, because the only way to satisfy the Case Filter at this point is via the Spec-head agreement relation, the subject is forced to move to Spec-CP; third, to account for the XSV order, XP elements can adjoin to CP, in line with the proposal of Hoekstra & Jordens (this volume) in which they argue that adjunction is the first option taken in creating phrase structure in L1 acquisition. A different mechanism for nominative Case, not

INTRODUCTION

15

found in Turkish, is added at the second stage, namely, subject-clitic incorpora­ tion. As incorporation of non-pronominal subjects is disallowed (Baker 1988), this account captures the asymmetry of verb placement at this stage (possible subject-verb inversion with pronominal subjects; no subject-verb inversion with non-pronominal subjects). At stage 3, the head-government option for nominative Case enters the Interlanguage: only at this point can the subject be checked for Case from the verb that has moved to COMP and hence governs the Spec-IP position. A final topic taken up by Schwartz & Sprouse is the etiology of Interlanguage systems, that is, the reasons for this particular course of develop­ ment. Exploring relations among the L1, intermediate systems, input data and learnability considerations, they suggest that one of the main differences between L1 and L2 acquisition may be that only the former is a deterministically driven process, which they call the DETERMINACY PROBLEM IN L2 ACQUISITION. What this means is that due to the structure of the L1 grammar, there may be cases in which it is impossible for the L2 system to converge on the grammar of the target language because there exist no data that would be able to force retraction from a previous system. They argue, nevertheless, that such a predicament may arise precisely because the only hypotheses an L2 acquirer can posit are those that UG makes available (in this instance, the general operation of adjunction, utilized at stage 1). In short, to the extent their analysis correctly captures the data, they claim it supports the idea that L2 linguistic knowledge can be of the same type of knowledge as native-speaker competence — even if not identical to it (Schwartz 1989). Eubank, too, looks at longitudinal data on the L2 acquisition of German, but he returns to one of the native Spanish-speaking subjects from the original ZISA project (Clahsen, Meisel & Pienemann 1983). His study focuses on adverb-initial utterances, which have often been assumed to fall into two discrete stages: ADV-S-V (non-inversion) followed by ADV-V-S (inversion). Closer examination of Jose's adverb-initial data reveals overlap of the two patterns for quite some time. The problems Eubank addresses is to account, first, for the 'optional' inversion stage, and second, for the later stage in which non-inversion is abandoned. Thus, as in the papers by Deprez & Pierce, Hoekstra & Jordens and Rizzi, the issue of optionality in early stages of development is confronted; the tack Eubank takes, however, is distinct. The impetus for his analysis comes from Wexler's (1991) attempt to explain a phenomenon which appears to be crosslinguistically common in L1 development, namely, the production of the infinitive instead of the required finite verb forms. Wexler's idea is that <+tense> is not yet uniformly present in this early period of the child's grammar, and so

16

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

economy considerations (Chomsky 1991) bring about affix lowering on to the verb (rather than verb raising), the result of which is the infinitival verb form. At a later point, <+tense> is obligatorily selected and so inflected forms of the verb are found as required. While agreeing with the intuition of Wexler's analysis, Eubank notes certain problems for L1 English acquisition. He amends Wexler's syntactic framework for verb movement, utilizing features of Agreement (<-> or <+>, the latter able to be 'weak' or 'strong') as well as Lasnik's (1992) features for both inflection and the verb, ranging from 'strong' to 'weak'. In Lasnik's proposal, a head may never adjoin to another head in which a feature value of the adjoiner exceeds a feature value of the adjoinee. In brief, with the <+tense> feature, inflected verbs raise when <+agr> is strong, but as the infinitive is <-agr> and <-tense>, then affix lowering results. After showing how this adaptation can successfully accommodate the problematic L1 English data, Eubank goes on to apply the framework to the L2 adverb-initial data. His hypothesis is that <+tense>, as in the early L1 acquisition analysis, is not yet a feature of Jose's Interlanguage but strong <+agr> is. According to Eubank, this makes a prediction: verb raising should occur when marked for inflection, but when it has the feature <-agr>, there should be lowering and hence non-inflected verb forms should result. Noting that ADV-S-V utterances may be ambiguous in regard to a lowering or a raising analysis, he develops four tests to determine whether the patterns of inversion and non-inversion can be seen as analogous to presence or absence of inflection and tense. The data do pass all four tests; yet, Eubank goes on to discuss a further (rather surprising) consequence that arises from this account. It should be noted, moreover, that there is in fact an important theoretical claim at stake, since the reasons for the L2 optional inversion as well as the demise of non-inversion are similar to the explanation for the L1 facts (i.e., optional infinitives which are then supplanted by obligatorily inflected forms): in order to prevent raising the verb, Eubank claims that there are syntactic features of the L1 grammar (viz. <+tense>) which are not present as part of the initial state of L2 acquisition. In other words, in contrast to the position of Vainikka & Young-Scholten, in which only lexical projections transfer, and in contrast to the position of Schwartz & Sprouse, in which the whole of the L1 grammar is posited as the initial state of L2 development, the analysis of Eubank takes him to an intermediate theoretical position: while both lexical and functional projections from the L1 constitute the initial state of L2 acquisition, not all the features associated with these categories transfer.

INTRODUCTION 3.

17

Concluding Remarks

As we hope to have sufficiently foreshadowed in the summaries above, the contributions in this volume address many of the issues that continue to be debated in the fields of native and non-native language acquisition. While the studies involve a range of languages — Bernese, Dutch, English, Finnish, French, German, Icelandic, Italian, Japanese, Korean, Spanish, Turkish — and while a spectrum of theoretical positions is advanced, in a certain sense there is a single thread uniting all the papers, namely: the extent of the knowledge that forms the initial state of language acquisition: in native language, the issue is the extent to which Universal Grammar is available from the onset of development; in non-native language, the issue is the extent to which knowledge of the native-language grammar determines subsequent development. Crucial to honing proposals directed at these issues is the specifics of linguistic theorizing, as is amply illustrated in the following papers. And so as linguistic theory evolves, so too will the study of language acquisition. The development of linguistic theorizing has profited greatly from the comparative research which the Principles and Parameters framework invited. Widening the database places severe restrictions on proposals that inherently make universal claims. The variety of languages investigated in the papers in this volume is illustrative of a similar comparative angle emerging in language acquisition studies. The variation with respect to e.g., the absence of pronominal subjects in early stages of native language clearly shows how intimately the phenomena are linked to specific structural properties of the languages to be acquired, therefore also showing how sensitive the acquisitional mechanism should be for these properties. This surely constitutes major methodological progress. Though many issues are still unresolved, the growing ties between formal linguistics and language acquisition research provide the foundation for approaching them, asking the right questions, and putting them to empirical test. The papers in this volume are aimed at furthering these developments.

References Avrutin, Sergej & Kenneth Wexler. 1992. "Development of Principle B in Russian: Coindexation at LF and coreference." Language Acquisition 2.259-306. Baker, Mark. 1988. Incorporation: A theory ofgrammaticalfunction changing. Chicago: The University of Chicago Press.

18

TEUN HOEKSTRA & BONNIE D. SCHWARTZ

Borer, Hagit & Kenneth Wexler. 1987. "The Maturation of Syntax." Parameter Setting, ed. by Thomas Roeper & Edwin Williams, 123-172. Dordrecht: Reidel. Chien, Yu-Chin & Kenneth Wexler. 1990. "Children's Knowledge of Locality Conditions in Binding as Evidence for the Modularity of Syntax and Pragmatics." Language Acquisition 1.225-295. Chomsky, Noam. 1991. "Some Notes on Economy of Derivation and Representation." Principles and Parameters in Comparative Grammar, ed. by Robert Freidin, 417-454. Cambridge, Mass.: MIT Press. Clahsen, Harald. 1991. "Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language." Language Acquisition 1.361-391. Clahsen, Harald, Sonja Eisenbeiss & Anne Vainikka. This volume. "The Seeds of Structure: A syntactic analysis of the acquisition of Case marking." Clahsen, Harald, Jürgen Meisel & Manfred Pienemann. 1983. Deutsch als Zweitsprache. Der Spracherwerb ausländischer Arbeiter. Tübingen: Gunter Narr. Clahsen, Harald & Pieter Muysken. 1986. "The Availability of Universal Grammar to Adult and Child Learners: A study of the acquisition of German word order." Second Language Research 2.93-119. . 1989. "The UG Paradox in L2 Acquisition." Second Language Research 5.1-29. Crain, Stephen & Rosalind Thornton. 1991. "Recharting the Course of Language Acquisition: Studies in elicited production." Biobehavioral Foundations of Language Development, ed. by Norman Krasnegor, Duane Rumbaugh, Richard Schiefelbusch & Michael Studdert-Kennedy, 321-338. Hillsdale, N.J.: Lawrence Erlbaum. duPlessis, Jean, Doreen Solin, Lisa Travis & Lydia White. 1987. "UG or not UG, that is the Question: A reply to Clahsen and Muysken." Second Language Research 3.56-75. Hayashibe, Hideo. 1975. "Word Order and Particles: A developmental study in Japanese." Descriptive and Applied Linguistics 8.1-18. Hoekstra, Teun & Peter Jordens. This volume. "From Adjunct to Head." Hornstein, Norbert & David Lightfoot. 1981. "Introduction." Explanation in Linguistics: The logical problem of language acquisition, ed. by Norbert Hornstein & David Lightfoot, 9-31. New York: Longman. Hyams, Nina. 1983. "The Acquisition of Parameterized Grammars." Diss., City University of New York. Hyams, Nina. 1986. Language Acquisition and the Theory of Parameters. Dordrecht: Reidel. Hyams, Nina & Kenneth Wexler. 1993. "On the Grammatical Basis of Null Subjects in Child Language." Linguistic Inquiry 24.421-459.

INTRODUCTION

19

Lasnik, Howard. 1992. "Case and Expletives: Notes toward a parametric account." Linguistic Inquiry 23.381-405. Lasnik, Howard & Tim Stowell. 1991. "Weakest Crossover." Linguistic Inquiry 22.687-720. McDaniel, Dana. 1989. "Multiple Movement and Partial Movement." Natural Language and Linguistic Theory 7.565-604. Pinker, Steven. 1984. Language Learnability and Language Development. Cambridge, Mass.: Harvard University Press. Poeppel, David & Kenneth Wexler. 1993. "The Full Competence Hypothesis of Clause Structure in Early German." Language 69.1-33. Radford, Andrew. 1990. Syntactic Theory and the Acquisition ofEnglish Syntax. Oxford: Basil Blackwell. Rizzi, Luigi. This volume. "Early Null Subjects and Root Null Subjects." Schwartz, Bonnie D. 1989. "L2 Knowledge: What is the null hypothesis?" Paper presented at the 14th Boston University Conference on Language Development, Boston, 15 October. Ms., Boston University. Thornton, Rosalind. 1990. "Adventures in Long-Distance Moving: The acquisition of complex wh-questions." Diss., University of Connecticut, Storrs. Vainikka, Anne. 1989. "Deriving Syntactic Representations in Finnish." Diss., University of Massachusetts, Amherst, de Villiers, Jill, Thomas Roeper & Anne Vainikka. 1990. "The Acquisition of Long-Distance Rules." Language Processing and Language Acquisition, ed. by Lyn Frazier & Jill de Villiers, 257-297. Dordrecht: Kluwer. Wexler, Kenneth. 1982. "A Principle Theory for Language Acquisition." Language Acquisition: The state of the art, ed. by Eric Wanner & Lila Gleitman, 288-315. New York: Cambridge University Press. . 1991. "Optional Infinitives, Head Movement and the Economy of Derivation in Child Grammar." Paper presented at the Conference on Verb Movement, University of Maryland, College Park, 13 October. Ms., Masssachusetts Institute of Technology. . 1992. "Optional Infinitives, Head Movement and the Economy of Derivation in Child Grammar." MIT Center for Cognitive Science Occasional Paper, No. 45. Massachusetts Institute of Technology, Cambridge. . 1993. "The Subset Principle is an Intensional Principle." Knowledge and Language: Issues in representation and acquisition, ed. by Eric Reuland & Werner Abraham, 217-239. Dordrecht: Kluwer. Wexler, Kenneth & Peter Culicover. 1980. Formal Principles of Language Acquisition. Cambridge, Mass.: MIT Press. Wexler, Kenneth & Maria Rita Manzini. 1987. "Parameters and Learnability in Binding Theory." Parameter Setting, ed. by Thomas Roeper & Edwin Williams, 41-76. Dordrecht: Reidel.

V2, Null Arguments and COMP Projections

Nina Hyams University of California, Los Angeles

1.

Introduction

The question of how children acquire functional heads has a long history within language acquisition research. Brown (1973) attempted to describe and explain what appears to be an invariant order in the development of the '14 grammatical morphemes,' following what he dubbed THE TELEGRAPHIC STAGE, a stage during which various functional elements and closed-class items such as inflections, determiners, auxiliaries, etc. were either completely absent or not reliably produced in the child's language. This observation led ultimately to the hypothesis that children's early grammars were semantically-based (Bowerman 1973; Schlesinger 1971), which is to say that they expressed semantic or thematic relations, but not structural ones. A more recent hypothesis concerning the nature of telegraphic speech is Radford's (1990) SMALL CLAUSE HYPOTHESIS (henceforth SCH), also proposed under somewhat different assumptions by Guilfoyle & Noonan (1989) and Lebeaux (1988). The central idea is that children start out with maximal projections of lexical heads, that is, small clauses, and gradually add the functional categories DET, INFL, COMP and their projections during the course of development. The SCH, while not implausible a priori, faces both conceptual and empirical problems. First, there is the problem of explaining how the functional categories are acquired if they are initially absent. Most proponents of the SCH would suggest a maturational account (cf. Radford 1990), that is, the functional cate­ gories are biologically determined to emerge at specific points in development some time after the emergence of lexical categories. While maturation of functional

22

NINA HYAMS

categories is in principle possible, it is not the optimal working hypothesis. A much stronger claim would be that children's grammars have the same basic form as adult grammars and are constrained by the same principles of grammar (Klein 1982; Hyams 1983, and many others). This is often referred to as the CONTINUITY HYPOTHESIS (Pinker 1984). The Continuity Hypothesis has far fewer degrees of freedom. If the assumption of continuity cannot be maintained for some specific aspect of grammar, then maturation becomes a plausible account. In short, I take continuity to be the null hypothesis concerning development. There are also empirical problems with the SCH. Thus far, the main empirical basis for the SCH has been English child language.1 However, there is a broad range of acquisition data from other languages which cannot be readily accounted for under the assumption that early grammars lack all functional heads. In fact, it is fair to say that the characterization of early language as 'telegraphic', then 'asyntactic', and finally 'small-clause-like' is in large measure a historical accident. Had the grammatical study of early child language begun with German and Dutch, for example, rather than English, a very different picture would have emerged. In these languages, as well as many others, we see that children acquire certain inflectional elements at a very early age, from the beginning of their multi-word utterances. Moreover, they control syntactic operations such as verb raising and verb second (V2), which are dependent on the presence of functional head positions (cf. Clahsen 1991 ; Hoekstra & Jordens, this volume; Hyams 1992a; Meisel & Müller 1992; Pierce 1989; Pierce & Deprez 1993; Sigurjónsdóttir 1987; Weissenborn 1992; Poeppel & Wexler 1993, and others). We return to this in Section 2. There is another proposal concerning early phrase structure, one which I will focus on in this paper. This proposal is that children have one or more INFL projections, but do not have COMP or its projections (Meisel & Müller 1992; Clahsen 1991; Gawlitzek-Maiwald, Tracy & Fritzenschaft 1992; Penner 1992). I will refer to this as the SHORT CLAUSE HYPOTHESIS (ShCH). The ShCH also suffers some conceptual problems. First, given the close relationship between INFL and COMP in natural language, it seems particularly unnatural for the child to project lexical categories and INFL (or AGR, TENSE, etc.), but not COMP. Second, we are still faced with the problem of explaining how COMP develops. The ShCH also raises questions of an empirical nature. There is a range of child language phenomena related to the verb second requirement in Germanic and residual V2 (see Section 2.1, p. 25) in languages like English, and also to the child's use of null arguments, which are most naturally accounted for by assuming that

V2, NULL ARGUMENTS AND COMP PROJECTIONS

23

the child has a full clausal structure. Moreover, I believe that the assumption that children begin with a COMP-less grammar raises a significant learnablity problem. In this paper I would like to focus on these two issues — the empirical and learnability-theoretic motivation for what I will call the FULL CLAUSE HYPOTHESIS (FCH). The model I will propose is one in which children begin with a minimal as well as maximal functional projection, that is, IP and CP. INFL may ultimately split into two or more functional heads depending on the particular language being acquired, but I will have little to say here about this process. Thus, I am proposing that the child begins with roughly the structure in (1).

This chapter is organized as follows. In Sections 2 and 3 I will discuss two sorts of empirical problems with the ShCH. As noted above, the first of these concerns V2-related phenomena in early language. This is discussed in Section 2. The second problem is related to the null argument phenomenon in child language. I turn to this in Section 3, where I also offer a new account of the null argument parameter intended to explain the crosslinguistic variation associated with null arguments in early language. In Section 4 I will discuss the learnability issue. 2.

Verb Second

The verb second phenomenon, which exists in all the Germanic languages (though only marginally in English, cf. Rizzi 1991a), provides a direct route for exploring the question of whether young children have functional categories, and specifically, a COMP system. The standard analysis of V2 within current theory, based on the original insights of Koster (1975) and den Besten (1983), involves movement of a finite verb to the head of CP. Subsequent movement of an XP to the specifier of COMP results in the verb appearing in second position.2 V2 is restricted to main clauses, except in Icelandic and Yiddish where it also holds in subordinate clauses. The German sentences in (2) illustrate standard V2 effects.

24

NINA HYAMS (2)

a. b. c.

Ich kenne den Mann. 'I know the man.' Hans sagt daß er den Mann kennt. 'Hans says that he the man knows.' Hans behauptet den Mann zu kennen. 'Hans claims the man to know.'

The question for us is: Do children show knowledge of the V2 requirement? If they do, then we have good reason to believe that they have the functional categories involved in the V2 effect. There has been a good deal of crosslinguistic research devoted to this question and all of the available evidence shows that from the earliest multi-word utterances, children distinguish finite from non-finite verb forms and raise finite verbs to some functional head outside the VP. Let us just briefly review some of the relevant facts. Clahsen (1991) notes that in early German modals and verbs with a -t affix appear in second position, while infinitives (-n affix) and unmarked verbs (0 affix) only occur in final position. He also notes that postverbal negation typically occurs with verbs inflected with the -t affix, while preverbal negation is used with the infinitive form and the unmarked verb form (0 affix). Similar results are reported in Meisel & Müller (1992), Weissenborn (1990) and Pierce & Deprez (1993). In a closely related language, Dutch, de Haan (1986), Frijn & de Haan (1991) and Weverink (1989) note that at around age 2, the Dutch child uses finite verb forms in first and second position (first position when the subject is dropped), and nonfinite verb forms in final position. Platzack (1992) calculates the rate of verb movement in finite vs. non-finite sentences in early Swedish. Swedish is underlyingly SVO and hence evidence for movement is somewhat less robust than in Dutch or German. We need to look for non-subject initial V2 structures. Platzack's data show that there are many such cases in finite utterances, while there are no cases of movement in utterances with non-finite verbs. Thus, the pattern that we observed in the other Germanic languages is repeated in Swedish. The distribution of finite and nonfinite forms in Dutch, German and Swedish follows straightforwardly if we assume that the finite verb form undergoes movement to a functional position preceding NEGP (recall the German negation facts), arguably to COMP, though we return to the question of which functional position the finite verb moves to in the following section.3

V2, NULL ARGUMENTS AND COMP PROJECTIONS

25

2.1 The Position of V in V2 and Residual V2 Language While there is substantial evidence for verb raising in early child language, there is controversy as to the landing site of the moved verb. We saw that in adult German and Dutch, the finite verb moves to COMP and thus we might take movement to COMP as the null hypothesis, in accordance with the Continuity Hypothesis. However, the Germanic child language data present us with an apparent paradox. While there are clear V2 effects in the V2 languages, we see little other direct evidence of COMP. For example, Meisel & Müller (1992) note that the two French-German bilingual children they studied used no complementizers before the ages of 2;6 and 3;0. Radford (1990) and others have made the same observation about English-speaking children. Meisel & Müller observe, moreover, that the children in their study fail to invert the auxiliary and subject clitic in French, as in (3). In the adult grammar the auxiliary would raise to COMP in these cases (Emonds 1978; Rizzi & Roberts 1990), as illustrated by the sentences in parentheses — a phenomenon which Rizzi (1991a) refers to as "residual V2" (examples from Meisel & Müller). (3)

a. b. c.

Où il est le papa? 'Where he is the Daddy?' Où l'est le papier? 'Where it is the paper?' Où il est baguette? 'Where it is (the) stick?'

(cf. Où est-il le papa?) (cf. Où est-il le papier?) (cf. Où est-elle la baguette?)

To reconcile the V2 facts with the lack of inversion in wh-questions and the absence of complementizers, Meisel & Müller adopt Pollock's (1989) ' split INFL hypothesis', in which INFL consists of (at least) two projections — TP and AGRP, and they propose that in the early grammar, the verb raises only as far as TENSE, the higher of the two projections. The COMP system is unavailable. The subject (or other XP) raises from a VP-internal position to Spec-TP giving rise to an apparent V2 effect. Thus, on Meisel & Müller's analysis, children have 'short clauses'. I have referred to this as the Short Clause Hypothesis (ShCH).4 Meisel & Müller's structure for German is given in (4).

26

NINA HYAMS

On the ShCH, the development of V2 in child language arises from the setting and resetting of a number of parameters, including the head parameter and Platzack & Holmberg's (1989) finiteness parameter, which specifies the placement of <±finite> either in INFL, as in English, or in COMP, as in German. Meisel & Müller assume that German children (and all children) initially place the <+F> feature in INFL (= TENSE). Moreover, they assume that while AGRP and TP are both head-final in adult German, children assume a mixed system in which AGRP is head-final and TP is head-initial. Thus, TP is head-initial and contains the <+F> operator making it a landing site for the raised verb. Although the ShCH has a good deal of intuitive appeal, it runs into a number of problems, which we will discuss in the following sections. 2.2

V'-above-Subject

First, it is unclear how the ShCH can handle verb raising in languages such as English and French (residual V2). Klima & Bellugi (1967) were the first to observe that English-speaking children go through a stage (their Stage C) in the development of questions in which they front the wh-phrase but fail to invert the auxiliary, as in (5). (5)

a. What he can ride in? b. Which way they should go? c. Where the other Joe will drive? (Klima & Bellugi 1967 — Stage C)

The sentences in (5) bear a striking similarity to the French sentences in (3). Recall that Meisel & Müller take the lack of inversion in (3) as evidence that children do not have a COMP position. What is curious about this stage is that the English-speaking children who are producing the sentences in (5) do invert the auxiliary systematically in yes/no questions, as in (6).5

V2, NULL ARGUMENTS AND COMP PROJECTIONS

(6)

27

a. Does the kitty stand up? b. Is Mommy talking to Robins grandmother? c. Will you help me? d. Can I have a piece of paper? (Klima & Bellugi 1967 — Stage C)

The inverted structure in (6) provides prima facie evidence that children do have a functional projection above the subject in Spec-IP, namely COMP. However, there is an alternative hypothesis, proposed by Guilfoyle & Noonan (1989). Guilfoyle & Noonan adopt the VP-internal subject hypothesis (Koopman & Sportiche 1991; Kitagawa 1986), according to which subjects are basegenerated inside some projection of V, say Spec-VP. They further propose that children producing inverted sentences such as those in (6) actually raise the verb to I but leave the subject in its VP-internal position. The relevant structure would be as in (7). On this analysis, then, there is no COMP projection.

As discussed by Pierce (1989), Lebeaux (1988) and Guilfoyle & Noonan, there is some evidence from the early placement of negation that children do, in fact, go through a stage in which the subject fails to raise from its base-generated VP-internal position. As Klima & Bellugi note, the sentences in (8) are quite typical in the earliest stages of development. (8)

a. Not the sun shining. b. No Fraser drink all tea. c. No I see truck. (Klima & Bellugi 1967 — Stage A negation) If we assume that there is a NEGP between the VP and INFL, then the 'external negation' in (8) would result from a failure to raise the subject to Spec-IP, as in (9).

28

NINA HYAMS

However, it is important to note that children who are producing inverted yes/no questions are well beyond this VP-internal subject stage. The sentences in (8) occur during Klima & Bellugi's Stage A (mean age 2;1), while the inverted yes/no questions occur during Stage C (mean age 3;0), at which point negative sentences are by and large well formed. Examples of Stage C negation are given in (10). (10) a. You didn't eat supper with us. b. I can t see it. c. It's not cold. d. I gave him some so he won't cry. e. I not crying. (Klima & Bellugi 1967 — Stage C) Thus, the hypothesis that children fail to raise subjects out of their VPinternal position fails to account for the correct placement of negation during Stage C. Moreover, the Guilfoyle & Noonan proposal raises a further question, namely, if children can have the subject internal to VP in yes/no questions, producing apparent inversion as in (6), why don't they also allow the subject to remain inside the VP in wh-questions, in which case we would find apparent inversion in w/z-questions as well, rather than the uninverted questions in (5). Thus, the facts in (5) and (6) are more compatible with the hypothesis that the subject is in Spec-IP. If this is so, then the auxiliaries in (6) must be in a functional position above Spec-IP, presumably COMP. Thus, the evidence from the development of questions in English argues that children at this stage do have a COMP system.6

V2, NULL ARGUMENTS AND COMP PROJECTIONS

29

2.3 Verb Raising in German Child Language A second argument against the ShCH concerns the V2 phenomenon in early German and Dutch. Recall that in the adult German and Dutch, finite verbs raise to COMP in main clauses, as in (11a); verb raising to COMP in tensed subordinate clauses is blocked by the presence of the complementizer in COMP, as illustrated by the contrast (llb,c). (11) a. b. c.

[CP Ichj [C kennej ] [IP tj [VP den Mann tj ] [I t'j ]]] 'I know the man.' Hans sagt [CP [C daß] [IP er den Mann tj [kennti ]]] 'Hans says that he the man knows.' *Hans sagt daß kennt er den Mann.

Recall from Section 2, page 24, that German- and Dutch-speaking children produce V2 structures from the earliest stage. It is thus striking that these children rarely make the mistake of overgeneralizing V2 to subordinate clauses. Clahsen (1986, 1991), Meisel & Müller (1992), Weissenborn (1990) and others report that as soon as subordinate clauses develop, they are always verb-final. This result is inexplicable on the ShCH, which holds that children raise verbs to some INFL projection (either AGR or TENSE). Under standard assumptions, the INFL positions are equally available in both main and subordinate clauses. Thus, if V2 effects in child language are the result of movement to INFL (or TENSE), then we should find verb raising in both embedded and main clauses, contrary to fact.7 Note that the fact that German-speaking children go through a stage, roughly from age 2;6 to 3;0, in which they produce verb-final subordinate clauses without overt complementizers (Harald Clahsen, pers. comm.; Penner 1992; Weissenborn 1990) suggests that children have a null complementizer in COMP which blocks inversion. I will assume that this is the case. I return to this point in Section 4. To sum up, then, the fact that children generally fail to overgeneralize V2 to embedded clauses in V2 languages argues against the ShCH and in favor of an adultlike system. Note finally that the ShCH fails to explain why the child who has V-to-I would give up such an analysis in favor of V-to-C once COMP matures. We run into a classic subset problem since all of the available V2 data are compatible with a V-to-I analysis. As we will discuss in more detail in Section 4, the short clause child would require negative evidence to move from a V-to-I grammar to a V-to-C grammar.

30 3.

NINA HYAMS Null Arguments

Another rich source of evidence that children have a COMP-projection at the very earliest stages of development comes from the child's use of null arguments, in particular null subjects. In Hyams (1983, 1986), I argued that the early grammar is a pro-drop grammar, comparable to the grammar of adult prodrop languages such as Italian and Spanish, in which a null pronominal (pro) is licensed in Spec-IP. This hypothesis amounts to the claim that subjectless sentences such as those in (12), which appear systematically at the earliest stages of language development, represent a grammatical option for the child and are not simply the result of a production constraint limiting the length of children's output.8 (English examples from Bloom, Miller & Hood 1975; Dutch examples from de Haan & Tuijnman 1988; French examples from Pierce 1989): (12) Show Mommy that. No open wallet. Want look a man. Kan niet slapen op een schaap. 'Cannot sleep on a sheep.' Is een trein hè? 'Is a train huh?' Veux pas lolo. '(I) want not water.' Est casse. 'Is broken.' Marche pas. '(S/he) walks not.' While null subjects appear to be a universal property of early language, the structural conditions under which arguments are omitted vary across child languages. Thus, de Haan & Tuijnman (1988) note that Dutch-speaking children at the relevant age (roughly 2 years) omit subjects predominantly when the verb occurs in first position. This is illustrated in the Dutch examples in (12). Sentences with null subjects stand in contrast to sentences with overt subjects, where the verb may surface in either first or second position, as in (13) (from de Haan & Tuijnman 1988). (13) a. b.

Hij doet het. he does it Moeten wij het weggooien? must we it away.throw

V2, NULL ARGUMENTS AND COMP PROJECTIONS

31

De Haan & Tuijnman note, moreover, that children also omit objects, but again, only when the verb is in first position, as in (14). (14) a. b. c.

Heeft Papa voorgelezen. has daddy read Moet ik maken. must I make Heeft Anna weer opgeplakt. has Anna again on.glued

3.1 Topic drop Based on the facts in (12) through (14), de Haan & Tuijnman argue that Dutch children have a process of topic drop, a process which also exists in the adult language. We assume that topic drop, like pro-drop (on Rizzi's 1986 account) is subject to both a LICENSING and IDENTIFICATION requirement. (We return to this below.) Thus, under topic drop, a null argument is licensed in topic position. We take topic position to be Spec-CP. This explains the co-occurrence of V1 and null arguments, i.e., (12) through (14) above; the null argument is in Spec-CP, while the (fronted) verb is in COMP. We further assume, following de Haan & Tuijnman, that the null argument in Dutch is identified by a discourse topic, an option available in so-called discourse-oriented or topic-prominent languages (Huang 1984). The distribution of null and overt arguments in early Dutch provides rather straightforward evidence that the early grammar projects to a CP, since this is the position in which topic drop is licensed. And the same pattern shows up in early German. Thus, Poeppel & Wexler (1993) found that the German-speaking child whose language they examined omitted subjects when the verb appears in first position, and otherwise not. In (15) we list the different sentence types found by Poeppel & Wexler for this child and their frequency of occurrence (from Poeppel & Wexler 1993). (15) OVS AdvVS SVO proY0 OVpro AdvVpro

19 31 129 19 0 0

32

NINA HYAMS

The crucial point is that subjects in postverbal position, Spec-IP by our assumption, must be phonologically realized, while subjects preceding the fronted verb, i.e., in Spec-CP, may be null. Similarly, Penner (1992), in a study of the acquisition of Bernese Swiss German, notes that initially the verb most often appears in first position; i.e., the verb is in COMP but there is no raising of an XP to Spec-CP. The occurrence of V2 increases developmentally. Penner notes a similar increase in null subjects over time (until approximately age 2;6 when the children stop omitting subjects). If Bernese-speaking children have a topicdrop grammar, we expect a direct relation between these two phenomena. For whatever reason, Bernese children initially do not have (or fail to use) movement to Spec-CP — precisely the position in which null arguments are licensed. As XP-topicalization increases, so should null arguments. Thus, the distribution of null subjects in early German further supports the topic-drop hypothesis and a fortiori the claim that the early grammar projects to CP. In the section that follows we will provide an analysis of the crosslinguistic variation in child language with respect to null argument use. We will argue that null subjects in early English also occur in topic position (though in structures which are distinct from early German and Dutch). The analysis of English null subjects will thus provide further support for the FCH. 3.2 The Null Argument Parameter We see that in child grammars, as in adult grammars, null arguments may be licensed in two contexts, in A-position, e.g., Spec-IP, and in Spec-CP, which is canonically an A-position. We find the former in pro-drop languages such as Italian and Spanish and the latter in languages such as German and Dutch. Whenever we find crosslinguistic differences of this sort, the question arises as to how such knowledge develops in the child. How does the child know that he is in a pro-drop vs. topic-drop language? Assuming that the options are available in UG, is one or the other a default option? Finally, what is the status of the child's null subjects in a language like English, which appears to be neither prodrop nor topic drop in its adult state? In this section, we will sketch a theory of the setting of the NULL ARGUMENT PARAMETER which addresses these issues. The analysis is not meant to be comprehensive (being tangential to the main point of this paper), but rather offers a promising direction in which to proceed.9 As a point of departure, let us assume as a general principle that pro is licensed under Spec-head agreement. As noted above, however, languages vary with respect to the status of the specifier position in which pro is licensed. In

V2, NULL ARGUMENTS AND COMP PROJECTIONS

33

pro-drop languages, pro is licensed in A-position, namely Spec-IP. By our assump­ tion it is licensed under Spec-head agreement with AGR. In German and Dutch, on the other hand, pro is licensed in Ā-position, that is, Spec-CP. (We return to the status of Spec-CP shortly.) In this instance, we assume that pro is licensed under Spec-head agreement with the features on the fronted verb in COMP. Null subjects in Spec-CP are licensed by agreement with the INFL features that the verb has picked up from AGR and null objects are licensed by agreement with the verb's selectional/subcategorization features which, we may assume, are inherited by COMP. Thus, in contrast to the pro-drop situation, in topic-drop languages both subjects and objects are licensed in Spec-CP. Note that the hypothesis that object pro is licensed under agreement with the verb's selectional/subcategorization features derives the result, noted by Cardinaletti (1990), that null objects in topic-drop languages are always 3rd person (in contrast to null subjects, which may be any person). This follows since V does not select for any particular person or number, while these features are specified on AGR. We may formulate the licensing requirement as a parameter, as in (16). (16) The Null Argument Parameter Pro is licensed under Spec-head agreement in A/Ā-position. As noted earlier, we assume that null arguments must be both licensed and identified (Rizzi 1986; Jaeggli & Safir 1989). As just outlined, the licensing requirement for pro is Spec-head agreement, and languages 'choose' between the A- and Ä-options in (16). However, pro must also be identified. We make the standard assumption that in pro-drop languages, pro is identified by a 'rich' AGR. In German and Dutch, in contrast, pro is identified by a discourse topic, as discussed earlier. In this respect it is similar to Chinese, as described in Huang (1984); the null element in Spec-CP is associated with and identified by a discourse topic through a process of 'topic-chaining'.10 We will further assume that null non-arguments in topic position such as adverbials need not be licensed, but only identified by an appropriate identifier. Thus, in topic-drop languages, we find sequences of the following sort, with an adverbial null topic which is identified by the overt adverb in prior discourse. (17) A:

B:

Ik weet dat je vandaag hard gewerkt hebt, maar wat heb je gisteravond gedaan? 'I know that you have worked hard today, but what have you done yesterday evening?' Ben ik naar de film geweest. 'Am I to the movies been.'

34

NINA HYAMS

In light of the foregoing discussion, let us re-examine null subjects in early English. Hyams (1983, 1986) argued that young English-speaking children, and in fact all children, begin with a pro-drop grammar. There were a number of empirical problems with the particular pro-drop analysis I proposed, which are discussed in Hyams (1992b). Moreover, there is a particular conceptual problem, which is that it is unclear how the identification requirement on pro could be satisfied by the early grammar of English, or by the adult grammar for that matter, since the language has a very impoverished morphology. This is especially true for young children in the pro-drop stage for whom verbal morphology is largely absent. This renders implausible the claim that Englishspeaking children have an Italian-like grammar. For this reason and others, Jaeggli & Hyams (1988) proposed that young English-speaking children have topic identification and thus are closer to Chinese-speaking children than to Italian-speaking children. Hyams & Wexler (1993) make a similar proposal, namely, that English-speaking children have a topic-drop grammar. We will have more to say about the identification requirement below. The idea that English-speaking children have a topic-drop grammar amounts to the claim that this is a default (or initial) grammatical option since there is no evidence in the input which would lead the English-speaking child to such an analysis. This would imply that the default option along the null argument parameter in (16) is that pro is licensed in an Ä-position. However, there is an obvious problem with this suggestion. In the typical topic-drop language both subjects and objects may be dropped, as discussed above. And in fact de Haan & Tuijnman show that this is the case for Dutch-speaking children and adults alike, and Wang, Lillo-Martin, Best & Levitt (1992) show that this is also the case for Chinese-speaking children. However, in English there is a strong asymmetry; young children drop subjects to a significant degree, but object drop is exceedingly rare. (See Hyams & Wexler 1993; and Wang et al 1992 for the statistics associated with subject and object drop.) Thus, children will allow null subjects without any evidence that this is a grammatical possibility in the language, for example young English-speaking children. However, they will not assume that objects may be null unless there is positive evidence that this is the case, as in German, Dutch and Chinese. How can we explain this subject/object asymmetry, within the framework we are developing? Let us begin by noting that while Spec-CP is canonically an Ä-position, it is sometimes an A-position. Rizzi (1991b) proposes the following definition of A-position:

V2, NULL ARGUMENTS AND COMP PROJECTIONS

35

(18) a is an A-position if: (i) a is assigned a thematic role, or (ii) a is construed with agreement. According to (18) Spec-CP is an A-position just in case the subject (bearing the index of AGR) moves to Spec-CP. By transitivity, Spec-CP is thus construed with agreement (assuming, for example, that CP inherits the index of its specifier) as illustrated in (19). (19) [ CPi John i [C [IP ti AGR¡ ... ]]] Notice that this induces a basic subject/object asymmetry since Spec-CP cannot be an A-position by virtue of the object moving to it, as illustrated in (20). (20) [CP Johni [C' [IP Maryj AGRj loves ti]] Let us return now to our original problem, namely, why the early grammar of English licenses null subjects but not null objects. Let us assume that the default option along the null argument parameter in (16) is that pro is licensed in A-position. This means that, barring evidence to the contrary, the child assumes that a null argument is licensed in Spec-IP, which is a canonical A-position. However, pro must also be identified, in this instance by AGR. For the Italian (and Spanish) child, the identification requirement will be satisfied since these languages have a 'rich' enough AGR. Thus, Italian/Spanish-speaking children will have an early grammar which is just like the adult grammar in all relevant respects. English AGR, however, is not a possible identifier in this sense. Thus, for the English-speaking child, pro in Spec-IP is licensed (given the default parameter setting), though not identified and hence, not a grammatical option. There is, however, another grammatical option. By hypothesis, the movement of the subject to Spec-CP creates an A-position and hence a licensing context for pro; thus, pro is licensed just in case it moves from Spec-IP to Spec-CP, but crucially not if it moves from the object position. We thus derive the basic subject/object asymmetry of null arguments in early English. A null subject will be licensed in Spec-CP under the default parameter setting; no such option is available for null objects. We assume that pro in SpecCP is identified via topic-identification. We return to the question of identifica­ tion below. Thus, early English represents a kind of hybrid — the identification requirement is met via topic-identification as in German and Dutch, while the licensing requirement is satisfied via Spec-head agreement with AGR roughly in the manner of a pro-drop language (though we will have more to say about this below).

36

NINA HYAMS

There are a number of interesting questions which arise. First, note that the null argument parameter as formulated in (16) does not have a non-null argument option. All that is required to license pro is Spec-head agreement, potentially available in all languages. Thus, by our hypothesis all languages are null subject languages as regards the licensing requirement. They will vary with respect to whether pro is licensed in A- or Ä-position and we assume that each language realizes either one or the other option.11 The situation with respect to identifi­ cation is different, however. While only languages with rich AGR can identify pro in Spec-IP, topic identification seems less categorical. There are languages such as Chinese which are strongly discourse-oriented and hence virtually any argument can be null provided it has a discourse antecedent. At the other extreme, there are languages such as English which allow topic drop under rather restricted circumstances. Thus, in English second person subjects can be omitted in questions such as (21a), non-thematic subjects may also sometimes be omitted as in (21b), and first person subjects may be dropped in diary contexts (Haegeman 1990). English is thus a RESIDUAL TOPIC-DROP LANGUAGE. (21) a. b. c.

Wanna leave? Seems like it's gonna rain. Had a wonderful day today.

Moreover, there is a great deal of dialect variation with respect to which topics can be null in the Germanic topic-drop languages. For example, several Southern German dialects and Swiss German allow null second person singular subjects, while Northern German does not (cf. Penner 1990). Thus, the determination of an appropriate identifier is dependent to a large degree on language-specific factors (though there are clearly general principles operating as well). It seems natural, then, that the locus of variation in the null argument phenomenon should be identification and not licensing. In this respect, the hypothesis being proposed here seems to have the right general character, i.e., all languages license null subjects, but they vary with respect to what counts as an appropriate identifier. To the extent that this is true, the identification mechanisms require learning or at least fine-tuning. This would explain why young children acquiring English can drop subjects (i.e., topics) in contexts in which this is not possible for adult speakers, i.e., contexts outside those in (21). At the earliest stages, they have not yet appropriately restricted the class of identifiers. How they do this is not an easy question to answer and we will not attempt to do so here. The analysis being proposed, specifically, the claim that Spec-CP is an

V2, NULL ARGUMENTS AND COMP PROJECTIONS

37

A-position and hence a licensing context for pro, has a number of empirical consequences. First, as noted above, we derive the asymmetry in argument drop that we find in early English. Null objects are excluded in English since pro must be licensed in an A-position. This default parameter setting may not be abandoned without positive evidence and as we will see shortly, there is no trigger in English which would induce the child to reset to the A-option, as there is in German and Dutch. (We return to these languages shortly.) Second, since we assume that English-speaking children do not differ from adults as regards the licensing of null subjects, i.e., the parameter in (16) (though the identifiers differ for children and adults), we do not expect to find a discrete shift in development from null subject to non-null subject use since this is not the result of a parameter resetting. Rather, the change will be more gradual as the child determines the proper discourse conditions for topic identification in English. Again, this seems a particular natural treatment since we know independently that the development of discourse and pragmatic knowledge can be quite protracted (Chien & Wexler 1990; Hamburger & Crain 1982; Karmiloff-Smith 1980). We take the English situation to be a marked one. Recall that the child's initial hypothesis is that pro is licensed in an A-position under Spec-head agreement with AGR. In the unmarked case, this refers to a canonical A-position as represented, for example, by Italian. In English, pro is licensed in a noncanonical A-position, derived by movement, and thus the licensing Spec-head relation is non-local, i.e., Spec-CP and AGR, as illustrated in (19) above. We assume that there is a certain cost associated with this. This might ultimately explain the fact that topic drop is a rather marginal phenomenon in English. Moreover, with respect to development, if there is a cost associated with the construction, we might expect that null subjects in English would be less frequent than in a language such as Italian with 'canonical' pro-drop. This latter prediction is confirmed in Valian (1990), who compares the rate of null subject use by young English- and Italian-speaking children. Her results show that Italian children omit lexical subjects at a rate of approximately 70 percent while English-speaking children at a comparable developmental stage omit subjects at a rate of 30 percent.12 A final prediction which follows from our analysis is that English-speaking children should omit subjects only from first position, i.e., Spec-CR This prediction is supported. Valian (1991) notes that children rarely, if ever, drop subjects in wh-questions (cf. also Rizzi, this volume).13 Before concluding this section, let us turn to the question of null arguments in early German and Dutch. We have argued that the default option for the

38

NINA HYAMS

parameter in (16) is that pro is licensed in A-position. In German/Dutch, however, as in English, AGR is not a possible identifier, and hence pro in SpecIP is not a grammatical option for the German/Dutch child. There are two possible scenarios, then. The German/Dutch child might assume the marked English analysis, in which a null subject pro is licensed in Spec-CP. This would predict an initial subject/object asymmetry in null arguments for German and Dutch children, as is the case for English-speaking children. Weissenborn (1990) reports such a stage at a very early point in the acquisition of German, although it is unclear how general a phenomenon this is. Alternatively, the evidence for the Ā-value along the null argument para­ meter may be so robust that children immediately set the parameter to that value. One likely trigger for the Ā-option is the realization that the language is V2 since it is the presence of V in COMP which licenses pro in Spec-CP. As dis­ cussed earlier, children learn the V2 requirement at a very early age. If this is the case, then we would predict that German and Dutch children would have both null subjects and objects from the earliest point. It is also possible that there is individual variation with respect to this aspect of grammar; some children might opt for the marked English analysis, while others might analyze the data more efficiently, by-passing such a stage, and moving directly to the A-parameter setting. On the assumption that German/Dutch children do go through the marked English-type analysis, there would be pressure to reset to the other value of the parameter, in order to reduce the markedness of the system. The data which would trigger a resetting to an unmarked value in Dutch/German are not available for English-speaking children, who therefore stay with the marked analysis.14 To conclude this section, let us just note that to the extent that the Spec-CP analysis of null arguments in early English is empirically supported, it provides further evidence for the main hypothesis of this paper, which is that children project to a CP from the earliest stages.

4.

A Learnability Argument for COMP

In the previous sections we considered several conceptual and empirical advantages of the FCH. In this section we will show that considerations of learnability also require that children start out with a COMP-system. We noted earlier that in languages such as German and Dutch, the verb raises to COMP in tensed clauses. The evidence for this is the complementarity which exists between complementizers and fronted verbs, illustrated in (22). In (22a) daß occupies COMP and hence the verb is blocked from raising to that position.

V2, NULL ARGUMENTS AND COMP PROJECTIONS

(22) a. *Ich I b. Ich I

weiß know weiß know

daß that daß that

die the die the

Kinder children Kinder children

39

haben den Film gesehen. have the film seen den Film gesehen haben the film seen have

There are, however, other V2 languages in which in we find verb raising and topicalization in tensed subordinate clauses in the presence of a complemen­ tizer. Icelandic is such a language. Thus, in Icelandic we do not find the complementarity between verb raising and complementizers that we find in German and Dutch. Consider in this regard the Icelandic sentence in (23). (23)

Jón segir að pessum hring hafi Páll lofad Maríu. John says that this ring has Paul promised Mary

Thráinsson (1986), Rögnvaldsson (1984), and more recently, Rögnvaldsson and Thráinsson (1990), argue that in Icelandic the verb does not raise all the way to COMP, but rather only as far as INFL, (or assuming a split INFL system, it raises to the higher INFL position). Diesing (1990) argues that Yiddish, another V2 language, also has V-to-I.15 According to the raising-to-INFL-analysis, in (23), the tensed auxiliary raises to the head of IP while the object, this ring, raises to Spec-IP. The relevant structure is as in (24). (I will assume, following Rögnvaldsson & Thráinsson 1990, that the subject Páll is in a base-generated VP-internal position, though it is possible that it occupies a lower IP-internal position. This is irrelevant to the analysis.) (24)

[IP Jón segir [CP [C að] [IP pessum hringi [j haftij] [VP Páll [tj] lofad [ti] Maríu]]]] 'John says that this ring has Paul promised Mary.'

If it is the case that V2 languages vary with respect to the position the verb raises to, as Icelandic (and Yiddish) seem to show, then the question arises as to how this parameter might be fixed by the child.16 In particular, what would the default hypothesis be? A priori, there are two possibilities, either the child assumes the V moves to INFL or she or he assumes V-(to I)-to COMP. The short clause analysis discussed earlier is equivalent to the claim that the default option is V-to-I (since there is no COMP position). As we will see, however, this raises a rather vexing learnability problem. So let us assume that V-to-I is the default setting. This is roughly the hypothesis being proposed by the ShCH. In this case the Icelandic child will have no further work to do. The German/Dutch child, in contrast, will have to reset the parameter to V-to-C. Recall, however, that the evidence that V moves

40

NINA HYAMS

to COMP in German and Dutch is provided by the fact that the verb cannot raise in the presence of a complementizer, for example in data such as that in (22a). In other words, the child would require negative evidence, namely, information that certain strings or structures are not possible in the language. It is a fact of acquisition that children do not have systematic access to negative data of this sort. Given this, the German- or Dutch-speaking child who had a V-to-I analysis would have no way of revising this incorrect setting on the basis of positive evidence alone.17 Thus, the proposal that children begin with V-to-I is suspect on grounds of learnability. Moreover, as discussed in Section 2, this hypothesis predicts, contrary to fact, that German and Dutch children will front verbs in embedded clauses just as Icelandic adults do. Let us now consider the second scenario in which V-to-C is the default setting. This will be the correct analysis for the Dutch and German child and hence no parameter resetting will be necessary in these languages. On the other hand, the Icelandic child will need to reset the parameter to V-to-I and she or he will need positive evidence to do this. Such evidence is immediately available. Sentences such as (24), in which the verb raises in the presence of a complemen­ tizer, provide clear evidence that V does not raise to COMP in Icelandic, but to some lower functional head. Proponents of the ShCH might argue at this point that the learnability problem is not a problem for their analysis since, strictly-speaking, on this hypothesis there is no COMP position for the verb to raise to and thus V-to-I is the only option. On this view, then, the default is 'Raise to the highest functional head there is.' 18 During the short clause stage, the child raises to INFL, as is proposed within the ShCH, and it is only during the subsequent stage, with the maturation of COMP, that the (new default) option to raise to COMP is instantiated. While this proposal would reconcile the ShCH with the necessity for a default V-to-C analysis, it faces an empirical problem. The 'Raise to the highest head' hypothesis predicts that Icelandic children will initially assume V-to-I (during the short clause stage), then adopt the default V-to-C when COMP matures, and then finally move to the (correct) V-to-I analysis. As we will see in the following section, the empirical data from Icelandic do not seem to support this hypothesized 3-stage developmental sequence. To sum up, learnability considerations require that V-to-C be the default setting and hence that COMP be available as a landing site for verb raising in the initial grammar. In the following section, we review some interesting data from the acquisition of Icelandic which provide empirical confirmation for this logical argument.

V2, NULL ARGUMENTS AND COMP PROJECTIONS

41

4.1 V-to-C in Icelandic Child Language The analysis just outlined predicts that Icelandic children will initially assume the default parameter setting in which V raises to COMP. If this is the case, the child's language will differ in predictable ways from the adult language and there is a specific pattern of inversion errors which should occur. The first is an 'error of omission'. We predict that the Icelandic child will not allow topicalization in embedded clauses. That is, we should not find sentences such as (24). Recall that (24) is possible in the adult language because topicalization moves an XP into Spec-IP, not Spec-CP. By our hypothesis, topicalization in the child's language is to Spec-CP. Sigurjónsdóttir (1987) reports that Ice­ landic children between the ages of 2;0 and 3;6 systematically fail to topicalize in embedded clauses. This result is particularly striking in that these same children topicalize freely in matrix clauses, as illustrated in (25), and their topicalization sentences are correct about 90 percent of the time (examples from Sigurjónsdóttir 1987).19 (25) a. b. c.

Nú bakkar hún. now backs she Essa á mamma. this one owns mama Kannski kemur hann. maybe comes him

This difference between main and subordinate clauses in child language follows from our analysis since in the matrix clause the topic can move to Spec-CP and the fronted verb to COMP, which is empty. There is no question that by age 2;0, Icelandic children have verb inversion. According to Sigurjónsdóttir, inverted sentences including OVS sentences such as (25b), and VSO sentences, which are interrogative or imperative, as in (26a), comprise about 30 to 40 percent of their utterances;20 the other 60 to 70 percent are SVO, as in (26b). Finally, Icelandic children always correctly invert the verb in wh-questions, as in (26c). (26) a. b. c.

Kann pessi stúlker ekki að labba? knows this girl not to walk? Eg vil fa kex. I want get crackers Huad fékk ég i skóinn? what got I in the shoe?'

42

NINA HYAMS

Sigurjónsdóttir points out that Icelandic children in the age range she studied make virtually no word order errors. One error is an occasional failure to front the verb in topicalized sentences (cf. note 19). A more interesting error given our analysis involves an overgeneralization of verb fronting which occurs in embedded interrogatives. Sigurjónsdóttir reports that approximately 10 percent of their embedded questions appear with the verb incorrectly inverted, as in (27a). This should be compared to the correct adult sentence in (27b). (27) a. b.

Sérdu á hverju stendur klukkan? see you on what stands the clock? Sérdu á hverju klukkan stendur? (correct adult form)

This particular word order error follows from our hypothesis that the children are moving V-to-C, while in the adult language, the movement is V-to-I. The structures corresponding to (27a, b) are given in (28a, b), respectively.

What is striking about the Icelandic data is the selective pattern of omission and error. These children have productive control of verb raising and they exhibit the full range of inverted structures — except for topicalization in embedded clauses. They make very few word order errors, but they incorrectly invert the verb in embedded interrogatives. Both of these results follow from the analysis proposed here, namely, that the default setting along the verb movement parameter is V-to-C, an analysis which is otherwise forced by considerations of learnability. This analysis leads to a number of questions and problems. One of these is that, as we noted earlier, German-speaking children never overgeneralize verb raising in embedded contexts. They do not raise verbs in embedded declaratives, such as (22a) and, as it turns out, they do not raise the verb in embedded

V2, NULL ARGUMENTS AND COMP PROJECTIONS

43

interrogatives either (Harald Clahsen, pers. comm.). Thus, they differ from Icelandic children in this respect. This would seem to be a problem for our analysis in that by our hypothesis German and Icelandic children both have V-to-C, and hence, all else being equal, they should behave the same. So the question is: Why do German (and Dutch) children fail to move the verb to COMP in embedded interrogatives, while Icelandic children do? Our answer to this question is that although both German and Icelandic children have V-to-C, all else is not equal. In particular, in standard adult German complementizers may not delete, while in Icelandic complementizers may delete in certain contexts.21 Let us assume that Icelandic and German children know the relevant property in their respective languages. The Icelandic child will assume that complementizers are optional and that whenever a complementizer fails to occur, for example in embedded questions, verb raising is possible. The German child, in contrast, will assume that the COMP position must be filled in embedded clauses.22 As noted earlier (see also Hyams 1985), I am assuming that COMP may contain a null complementizer, which though it has no phonological matrix nevertheless has sufficient syntactic content to block movement of V-to-C.

5.

Concluding Remarks

In this paper we have argued on both logical and empirical grounds that the early phrase structure (by which I mean by age 2, if not earlier) must contain, in addition to some minimal INFL system, a COMP projection. I refer to this as the Full Clause Hypothesis (FCH). This runs directly counter to the Small Clause Hypothesis, which holds that the early grammar projects only lexical categories, and the Short Clause Hypothesis, according to which the early grammar lacks a COMP system (though it may have one or more INFL projections). We have presented a number of empirical arguments for the FCH. First, children acquiring V2 languages such as Dutch and German, as well as children acquiring residual V2 languages like English, raise finite verbs (or auxiliaries in the case of English) to a functional position above IP, arguably COMP. Moreover, interesting differences in the distribution of null arguments in the acquisition of German/Dutch vs. English vs. Italian follow under the related hypotheses (i) that German/Dutch licenses null arguments (pro) in Spec-CP, and (ii) that English is a residual topic-drop language, which also licenses a subject pro in topic position (as a marked extension of the grammar). The V2 and null

44

NINA HYAMS

subject properties follow from the hypothesis that the early grammar projects to a CP. Finally, we provided a logical argument for the CP analysis; to the extent that V2 languages vary with respect to the position that the verb raises to, V-to-I versus V-to-C, learnability considerations require that V-to-C be the default option. If, as we are proposing, children have full clauses, what accounts for the 'telegraphic look' of early language? That is to say, why would children who have functional projections fail to reliably produce functional elements? The basic premise of the small clause and short clause analyses is that missing functional elements reflect missing categories and the primary empirical basis for these hypotheses is the absence of inflectional elements, complementizers, etc. However, functional items are often missing during a period of time in which children have syntactic operations which require functional categories. For example, young English-speaking children typically drop 3rd person singular -s and auxiliaries. By the assumptions of the SCH these children project only as high as VP. But we know from the position of subjects and negation that children who drop -s and auxiliaries have subject raising to Spec-IP, as illustrated by the examples such as those in (29) (from Bellugi 1967): (29)

He not bite you. That no blast off. I not get it dirty. I not bending them.

Moreover, crosslinguistic studies show that children acquiring more richly inflected languages do reliably supply verbal inflection at an age when Englishspeaking children are still omitting -s. (See Hyams 1992a for review of relevant studies and languages.) Thus, the problem would seem to have more to do with properties of the inflectional system of English than with any deep-seated developmental principle such as the SCH. In Hyams (1992a) I suggested that in assessing the development of inflection in English-speaking children, we should focus on the acquisition of -ing rather than on -s, since the progressive morpheme is used to refer to ongoing activity and is hence referential, while 3rd person singular -s has only a generic, non-referential reading with non-stative verbs (Enç 1987; Campbell 1991). In this respect, the English present tense system differs markedly from the other languages which have been the focus of acquisition studies, e.g., German, Dutch, Italian, etc. Thus, the delay of -s is arguably due to semantic difficulties and the 'real' present tense affix is -ing, which is acquired relatively early; -ing is the first of Brown's (1973) fourteen

V2, NULL ARGUMENTS AND COMP PROJECTIONS

45

morphemes to be acquired. This assumption, coupled with Wexler's (1992) hypo­ thesis that children at this stage have a free use of infinitives in root contexts, accounts for the telegraphic quality of early English, at least as regards inflection. Similarly, we saw earlier that children acquiring V2 languages and residual V2 languages give evidence of verb movement to COMP at a point at which they are not yet using lexical complementizers, again calling into question the basic premise of 'missing elements = missing category'.23 With respect to the question of missing complementizers, Hyams (1985) proposed that this is an effect of lexical learning. The choice of complementizer depends on the selectional properties of particular verbs. Thus, try selects a <-tense> comple­ ment, while think selects a <+tense> complement, etc. We know that selectional properties must be learned and so must the complementizers themselves and the lack of semantic content makes them not particularly salient or interesting. It is therefore not surprising that such lexical items are acquired relatively late. As proposed in Hyams (1985), we need to distinguish the acquisition of complemen­ tation as a semantic/syntactic phenomenon from the acquisition of complemen­ tizers which is a kind of lexical learning. Thus, we can imagine a structural realization principle such as that in (30) (from Hyams 1985) which will allow the child to project a CP complement based on her knowledge of the semantics of the verb, i.e., the knowledge that it selects a propositional argument.24 (30)

In order for a constituent to be interpreted as propositional it must be syntactically realized as CP.

Since the knowledge of complementation (principle (30)) is logically separate from the selection of the tense of the complement of particular verbs and hence the choice of complementizer, it follows that these should develop independently. This is what we expect within a modular system. Thus, a child may have knowledge of the CP structure of complements while still not producing lexical complementizers. To conclude, the premise that missing functional items = missing functional categories is difficult to maintain given the fact that children have syntactic operations involving functional categories at the point at which they fail to reliably produce functional elements. A simpler explanation for the missing lexical items is just that the lexical items are missing, essentially because they have properties which make them difficult to learn, for example, lack of referentiality or meaning, etc. Since on everyone's account these lexical elements (words and affixes) must be learned, the most parsimonious explanation is that this is all that the child must acquire.25 This view will likely mean that there are

46

NINA HYAMS

different reasons for the absence of different functional elements. This is in contrast to the SCH and ShCh which propose a unified explanation for all missing items. But, in fact, the crosslinguistic variation associated with the acquisition of functional elements, as well as the staggered development of different items within a particular language, fully supports a lexical learning approach over a structural one.

Acknowledgements I would like thank a number of people with whom I have had the opportunity to discuss the material presented in this paper and who have shared their insights with me — Harald Clahsen, Teun Hoekstra, Kyle Johnson, Peter Jordens, Bonnie Schwartz, Charlotte Reinholtz, Tom Roeper, and Ken Wexler. I would also like to express my appreciation to the participants of the Workshop on The Development of Movement and Inflection at the 1991 GLOW Colloquium, where a shorter version of this paper was presented, and to the Netherlands Institute for Advanced Studies (NIAS), where I was able to write a first draft of this paper. My thanks also to the participants of my UCLA seminars during the Winter 1991 and 1992 quarters and to three anonymous reviewers for their many helpful comments. Finally, my appreciation to Jeannette Schaeffer for editorial assistance and patience. All errors and idiocies are my own. Notes 1.

One exception to this is Platzack (1992), who argues for a small clause analysis of Swedish child language. However, there are a number of problems with Platzack's analysis of the Swedish data, which are discussed in detail in Hyams (1992a). When properly interpreted we believe that these data argue against a SCH and in favor of the full clause analysis proposed in this chapter. See Hyams (1992a) and also Wexler (1992) for discussion.

2.

Following standard analyses, I assume, moreover, that the verb first raises to one or more inflectional head positions and then to COMP. Whether or not this is the case in the child's grammar is discussed shortly.

3.

See Hyams (1992a) for detailed discussion of the German, Dutch and Swedish acquisition data. See also Wexler (1992) and the various papers in Meisel (1992).

4.

Clahsen (1991) also argues that children do not raise the verb all the way to COMP but rather to an intermediate INFL position. A similar proposal is made in Jaeggli & Hyams (1988) in order to account for pro-drop effects in early German. We claimed that tense

V2, NULL ARGUMENTS AND COMP PROJECTIONS

47

features are initially located in INFL rather than COMP. However, our analysis (like Clahsen's) did not assume that the child has a split INFL system, as Meisel & Müller do. 5.

Meisel & Müller do not discuss yes/no questions in French and thus we do not know whether French children exhibit the same curious pattern as English children who invert only in yes/no questions.

6.

Nothing we have said thus far explains why English-speaking (and French-speaking) children fail to raise the auxiliary in wh-questions, as shown in (3) and (5). If children have a CP and hence a COMP position for the finite auxiliary to raise into, this result is unexpected. Klein (1982) (within an earlier (pre-CP) framework) deals with this issue at length. She proposes that children assume AUX inversion is a substitution into COMP As a result, inversion is blocked when COMP is occupied by a wh-phrase. Translating Klein's proposal into the current framework, in which CP contains a head and specifier position (Chomsky 1986), one possible explanation is that children misanalyze whelements as heads — possibly base-generated in COMP (as Klein assumed). The whelement thus blocks AUX from raising to COMP. This hypothesis leads to at least two empirical predictions. First, we predict that children will have more difficulties with complex wh-phrases such as what boy, which dog, etc. since these cannot easily be analyzed as heads. Second, wh-movement should be strictly local, being an instance of head movement. We do not know at present whether these predictions are confirmed or not. An analysis along these lines has been independently proposed in van Kampen (1989).

7.

Meisel & Müller (1992) report that one of the three children in their study did make V2 errors in subordinate clauses; Gawlitzek-Maiwald et al. (1992) also report on one German monolingual child who made such errors. Though these cases require explanation, they are the exception rather than the rule. Moreover, almost all of the V2 errors cited occur in adjunct clauses introduced by wenn 'if/when' (eg., ...wenn da komm andere schiffe dann gehn die dagegen 'when there comes other boats then go they it-against') or conjoined clauses. It is possible that in these cases, the child has misanalyzed the dependent clauses as a main clause introduced by an adverbial when (Meisel & Müller suggest a similar idea). V2 would then apply as in main clauses. More compelling evidence of V2 in subordinate clauses would be errors in subcategorized complements introduced by true complementizers. Only one such error is reported.

8.

Bloom (1990) and Valian (1990) argue against the grammatical account of subjectless sentences and propose instead that children drop subjects because of performance limitations. See their papers for further discussion and Hyams & Wexler (1993), who show that processing models completely fail to account for the basic statistical and theoretical data associated with pro-drop in early language.

9.

Rizzi (this volume) independently arrives at an analysis of the null subject phenomenon in early English similar to the one proposed in this section. In particular, Rizzi and I

48

NINA HYAMS both argue (contra Hyams 1983, 1986) that null subjects in early English differ from null subjects in pro-drop languages such as Italian, and are most closely related to the 'diary-drop' phenomenon discussed in Haegeman (1990). The two analyses differ in other respects and a detailed comparison is beyond the scope of this paper. The ideas proposed in Section 3.2 are quite tentative. Hyams (in prep.) further develops and substantially modifies the analysis by eliminating the null argument parameter as an independent parameter and deriving the crosslinguistic variation in null arguments from independent principles of grammar, notably verb raising.

10.

We should note that the analysis of topic drop proposed here is restricted to the V2 languages and not intended to cover Chinese-type topic-drop languages.

11.

We make the assumption that a language must choose a single option though we recog­ nize that this may be too strong a requirement. American Sign Language (ASL) seems to allow both options. See Lillo-Martin (1992) for discussion. For the purposes of the discussion, we leave the case of ASL, and other 'mixed' systems (if such exist) aside.

12.

Valian argues that the frequency differences are due to the fact that English-speaking children omit subject for performance reasons (processing overload), while Italian children do so for grammatical reasons, that is, they have a pro-drop grammar. Note, however, that there is no theoretical reason why performance-induced null subjects should occur at a lower frequency than grammatically-induced null subjects. See Hyams & Wexler (1993) for a detailed critique of performance accounts of the null subject phenomenon.

13.

Roeper (1991) notes that Adam (CHILDES, MacWhinney & Snow 1985) does omit subjects in wh-questions, as in (i): (i) Where going? We will assume that sentences such as in (i), which are apparently rare, are derived via adjunction of the wh-phrase to CP.

14.

Notice that there is an important distinction between a 'default' parameter setting, a 'marked' option, and an 'initial' setting. The default setting is that which is assumed in the absence of evidence to the contrary. By our hypothesis, pro in A-position is a default analysis. It also may be the initial setting. This is an empirical matter and depends on whether German/Dutch children go through an initial English-like stage, despite the evidence that German/Dutch licenses pro in Ā-position. The English acquisi­ tion facts cannot decide the question since in English the default and initial options converge in the absence of evidence for an A-analysis. Finally, with respect to the issue of markedness, we take the licensing of pro in a non-canonical A-position to be marked, for the reasons outlined in the text. However, we maintain that neither of the antecedently available options along the null argument parameter is marked with respect to the other. In other words, we reject the position that the unmarked value is necessarily identified with either the initial or default option and more generally, the idea that markedness is a developmental notion, though it has obvious implications for development, as discussed in the text. See Hyams (1986, chapter 6) for further discussion of this issue.

V2, NULL ARGUMENTS AND COMP PROJECTIONS 15.

49

Vikner & Schwartz (forthcoming) argue that V2 always involves raising to a position outside IP, thus disputing the IP analysis for Icelandic and Yiddish. The details of this debate are outside the scope of this paper. Suffice it to say that if it turns out that V2 always involves movement to the COMP position, then this renders even less plausible the claim that children do not have a COMP projection since verb raising is a very early development, as discussed in the text.

16. If V2 languages vary with respect to the position of the raised verb, then it cannot be the case that the defining characteristic of V2 languages is V-to-C movement (cf. Diesing 1990; Rögnvaldsson & Thráinsson 1990). Thus, we do not follow Platzack & Holmberg (1989) in assuming that theirfinitenessparameter (V-to-I vs. V-to-C) separates V2 from non-V2 languages. Rather, we assume that this parameter separates V2 languages such as German from V2 languages like Icelandic and that the parameter is therefore set only by children acquiring V2 languages. Thus, although we will propose that V-to-C must be the default setting of this parameter, this will have no effect for children acquiring non-V2 languages such as English and French, for whom the parameter is not accessible. With regard to the latter languages, we assume, following Chomsky (1992), that there is an independent parameter which determines whether Vto-I happens in the syntax, as in French, or at LF, as in English. This parameter may interact with the V2 parameter but is distinct from it. 17.

One might argue that since German is head-final and INFL is on the right periphery, the child could deduce V-to-C simply by noting that the finite verb is not in final position (i.e., INFL), but rather in some position to the left. There are two problems with this solution, however. First, there is nothing to prevent the child from assuming that the position on the left which contains the finite verb is some other functional head, distinct from both IP and CP, as in (i): (i) [FP Ichi [F kennej [IP ti [VP den Mann tj ]]]] FP≠IP, CP Second, although German and Dutch are 'mixed' systems, i.e., CP is head-initial and IP is head-final, there are V2 languages with V-to-C in which both INFL and COMP are on the left, for example Swedish. Thus, the logical problem discussed in the text remains even if not in the specific case of German.

18.

I am grateful to two anonymous reviewers for pointing out this possibility to me.

19.

The errors in topicalized structures typically involve a failure to front the verb, i.e., the verb occurs in third position. Thus we have examples such as those in (i): (i) Nú hann er búin. 'Now he is done.' Hérna stóllinn sér míg eki. 'Here the chair sees me not.' Pa ég kom á sjá bilinn. 'Then I came to see the car.'

20.

Included in the 30-40 percent are wh-questions.

50

NINA HYAMS

21.

Specifically, complementizers under bridge verbs may delete when the embedded subject is pronominal (Sigurjónsdóttir, pers. comm.). Consider the following contrasts: Ég tel (ad) hún fan á morgun. (i) 'I believe (that) she leaves:SUBJ tomorrow.' Ég tel (*að) Jón fari á morgun. (ii) 'I believe (that) John leaves tomorrow.' (iii) Eg veit (að) hún fer á morgun. 'I know (that) she leaves:IND tomorrow.' (iv) Eg veit (*að) Jón fer á morgun. 'I know (that) John leaves tomorrow.' The reason for the pronominal restriction is not clear to us. Recall, however, that Icelandic has V2 and fronting of an XP in embedded clauses as well as root clauses. Thus, one possible explanation for the optionality of complementizers with pronouns is that the pronominal subjects are optionally clitics. When they are realized as clitics, they raise to COMP (rather than Spec-IP, the normal position for the subject or other constituent to raise to). When the pronoun occupies COMP, no complementizer may occur there.

22.

I am indebted to Kyle Johnson for pointing out to me this syntactic difference between German and Icelandic and for suggesting how it might influence the development of inversion in subordinate clauses in the two languages.

23.

See Hyams (1992a), where I argue that this premise is also conceptually flawed.

24.

For discussion of structural realization principles (otherwise referred to as boots-trapping), see Grimshaw (1979, 1981) and Pesetsky (1982).

25.

If functional elements are processed or accessed differently from lexical elements, as has been argued by Garrett (1975) and others, this might also contribute to their relative inaccessibility.

References Bellugi, Ursula. 1967. "The Acquisition of Negation". Diss., Harvard University, Cambridge, Mass. den Besten, Hans. 1983. "On the Interaction of Root Transformations and Lexical Deletive Rules." On the Formal Syntax of the Westgermania: Papers from the "3rd Groningen Grammar Talks", Groningen, January 1981. (= Linguistik Aktuell, 3), ed. by Werner Abraham, 47-131. Amsterdam: John Benjamins. Bloom, Lois, Peggy Miller & Lois Hood. 1975. "Variation and Reduction as Aspects of Competence in Language Development." The 1974 Minnesota Symposia on Child Psychology, ed. by A. Pick, 3-55. Minneapolis, Minn.: University of Minnesota Press.

V2, NULL ARGUMENTS AND COMP PROJECTIONS

51

Bloom, Paul. 1990. "Subjectless Sentences in Child Language." Linguistic Inquiry 21.491-504. Bowerman, Melissa. 1973. Early Syntactic Development. Cambridge: Cambridge University Press. Brown, Roger. 1973. A First Language: The early stages. Cambridge, Mass.: Harvard University Press. Campbell, Richard. 1991. 'Tense and Agreement in Different Tenses." Ms., University of Pennsylvania. Cardinaletti, Anna. 1990. "Pronomi nulli e pleonastici nelle lingue germaniche e romanze." Diss. Venice University. Chien, Yu-Chin & Kenneth Wexler. 1990. "Children's Knowledge of Locality Conditions in Binding as Evidence for the Modularity of Syntax and Pragmatics." Language Acquisition 1.225-295. Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press. . 1992. "A Minimalist Program for Linguistic Theory." MIT Occasional Papers in Linguistics 1. Department of Linguistics and Philosophy, Massachusetts Institute of Technology, Cambridge. Clahsen, Harald. 1986. "Verb Inflections in German Child Language: Acquisition of agreement markings and the functions they encode." Linguistics 24.79-121. . 1991. "Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language." Language Acquisition 1.361-391. Diesing, Molly. 1990. "Verb Movement and the Subject Position in Yiddish." Natural Language and Linguistic Theory 8.41-79. Emonds, Joseph. 1978. "The Verbal Complex V'-V in French." Linguistic Inquiry 9.151-175. Enç, Murvet. 1987. "Anchoring Conditions for Tense." Linguistic Inquiry 18.633-657. Frijn, Jacqueline & Ger de Haan. 1991. "The Acquisition of Finiteness in Dutch." Paper presented at the 14th GLOW Colloquium (Workshop on the Development of Movement and Inflection), Leiden, March. Garrett, Merrill F. 1975. "The Analysis of Sentence Production." The Psychology of Learning and Motivation. Volume 9, ed. by G. H. Bauer. New York: Academic Press. Gawlitzek-Maiwald, Ira, Rosemary Tracy & Agnes Fritzenschaft. 1992. "Language Acquisition and Competing Linguistic Representations: The child as arbiter." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 139-179. Dordrecht: Kluwer. Grimshaw, Jane. 1979. "Comp Selection and the Lexicon." Linguistic Inquiry 10.

52

NINA HYAMS

Grimshaw, Jane. 1981. "Form Function and the Language Acquisition Device." The Logical Problem of Language Acquisition, ed. by C.L.Baker & John J. McCarthy, 165-182. Cambridge, Mass.: MIT Press. Guilfoyle, Eithne & Maire Noonan. 1989. "Functional Categories in Language Acquisition." Ms., McGill University, Montreal. de Haan, Ger. 1986. "A Theory-bound Approach to the Acquisition of Verb Placement in Dutch." Paper presented at the Workshop on Universals in Child Language, Heidelberg University. de Haan, Ger & Kees Tuijnman. 1988. "Missing Subjects and Objects in Child Gram­ mar." Language Development, ed. by Josien Lalleman & Peter Jordens, 110-122. Dordrecht: Foris. Haegeman, Liliane. 1990. "Non-overt Subjects in Diary Contexts." Grammar in Progress, ed. by Juan Mascaré & Marina Nespor, 167-179. Dordrecht: Foris. Hamburger, Henry & Stephen Crain. 1982. "Relative Acquisition." Language Develop­ ment. Vol 1: Syntax and semantics, ed. by S.Kuczaj II, 245-272. Hillsdale, N.J.: Lawrence Erlbaum. Hoekstra, Teun & Peter Jordens. This volume. "From Adjunct to Head." Hyams, Nina. 1983. "The Acquisition of Parameterized Grammars." Diss., City University of New York. . 1985. "The Acquisition of Clausal Complementation." Proceedings of WCCFL IV, ed. by J.Goldberg, S.Mackaye & M.Wescoat, 99-111. Stanford: The Stanford Linguistics Department, Stanford University. . 1986. Language Acquisition and the Theory of Parameters. Dordrecht: Reidel. . 1992a. "The Genesis of Clausal Structure." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 371-400. Dordrecht: Kluwer. . 1992b. "A Reanalysis of Null Subjects in Child Language." Theoretical Issues in Language Acquisition, ed. by Jürgen Weissenborn, Helen Goodluck & Thomas Roeper. Hillsdale, N.J.: Lawrence Erlbaum. . In prep. "Null Arguments and Verb Raising." Hyams, Nina & Kenneth Wexler. 1993. "On the Grammatical Basis of Null Subjects in Child Language." Linguistic Inquiry 24.421-459. Huang, C.-T. James. 1984. "On the Distribution and Reference of Empty Pronouns." Linguistic Inquiry 15.531-574. Jaeggli, Osvaldo & Nina Hyams. 1988. "Morphological Uniformity and the Setting of the Null Subject Parameter." Proceedings of NELS 18, 1987, ed. by James Blevins & Juli Carter, 238-253. Amherst, Mass.: Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachu­ setts, Amherst.

V2, NULL ARGUMENTS AND COMP PROJECTIONS

53

Jaeggli, Osvaldo & Kenneth Safir. 1989. "The Null Subject Parameter and Parametric Theory." The Null Subject Parameter, ed. by Osvaldo Jaeggli & Kenneth Safir, 215-238. Dordrecht: Kluwer. van Kampen, J. 1989. "De verwerving van wh-vraagzinnen." Master's thesis, Utrecht University. Karmiloff-Smith, A. 1980. "Psychological Processes Underlying Pronominalization and Non-Pronominalization in Children's Connected Discourse." Papers from the Parasession on Pronouns and Anaphora. Chicago: Chicago Linguistic Society. Kitagawa, Yoshihisa. 1986. "Subjects in Japanese and English." Diss., University of Massachusetts, Amherst. Klein, Sharon. 1982. "Syntactic Theory and the Developing Grammar." Diss., University of California, Los Angeles. Klima, Ed & Ursula Bellugi. 1966. "Syntactic Regularities in the Speech of Children." Psycholinguistics Papers, ed. by John Lyons and Roger Wales, 183-207. Edinburgh: Edinburgh University Press. Koopman, Hilda & Dominique Sportiche. 1991. "The Position of Subjects." Lingua 85.211-258. Koster, Jan. 1975. "Dutch as an SOV Language." Linguistic Analysis 1.111-136. Lebeaux, David. 1988. "Language Acquisition and the Form of the Grammar." Diss., University of Massachusetts, Amherst. Lillo-Martin, Diane. 1992. Universal Grammar and American Sign Language. Dordrecht: Kluwer. MacWhinney, Brian & Catherine Snow. 1985. "The Child Language Data Exchange System". Journal of Child Language 12.271-296. Meisel, Jürgen, ed. 1992. Acquisition of Verb Placement: Functional categories and V2 phenomena in language development. Dordrecht: Kluwer. Meisel, Jürgen & Natascha Müller. 1992. "Finiteness and Verb Placement in Early Child Grammars: Evidence from simultaneous acquisition of two first languages: French and German." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 109-138. Dordrecht: Kluwer. Penner, Zvi. 1990. "The Acquisition of the Syntax of Bernese Swiss German: The role of functional elements in restructuring early grammars." Paper presented at the 15th Annual Boston University Conference on Language Development, Boston, October 19-21. Ms., University of Berne, 1991. . 1992. "The Ban on Parameter Resetting, Default Mechanisms, and the Acquisition of V2 in Bernese Swiss German." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 245-282. Dordrecht: Kluwer.

54

NINA HYAMS

Pesetsky, David. 1982. "Paths and Categories." Diss., Massachusetts Institute of Technology, Cambridge. Pierce, Amy. 1989. "On the Emergence of Syntax: A crosslinguistic study." Diss., Massachusetts Institute of Technology, Cambridge. Pierce, Amy & Viviane Deprez. 1993. "Negation and Functional Projections in Early Child Grammar." Linguistic Inquiry 24.25-67. Pinker, Steven. 1984. Language Learnability and Language Development. Cambridge, Mass.: Harvard University Press. Platzack, Christer. 1992. "Word Order and the Finite Verb in Early Swedish." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 66-82. Dordrecht: Kluwer. Platzack, Christer & Anders Holmberg. 1989. "The Role of AGR and Finiteness in Germanic VO Languages." Working Papers in Scandinavian Syntax 43.51-76. Lund: Department of Scandinavian Linguistics, University of Lund. Poeppel, David & Kenneth Wexler. 1993. "The Full Competence Hypothesis of Clause Structure in Early German." Language 69.1-33. Pollock, Jean-Yves. 1989. "Verb Movement, Universal Grammar, and the Structure of IP." Linguistic Inquiry 20.365-424. Radford, Andrew. 1990. Syntactic Theory and the Acquisition ofEnglish Syntax. Oxford: Basil Blackwell. Rizzi, Luigi. 1986. "Null Objects in Italian and the Theory of pro." Linguistic Inquiry 17.501-557. . 1991a. "Residual Verb Second and the WH-Criterion." Technical Reports in Formal and Computational Linguistics 3, Geneva University. . 1991b. "Proper head government and the Definition of A-positions." Paper presented at the 14th GLOW Colloquium, Leiden, March. Abstract in GLOW Newsletter 26.46-47. . This volume. "Early Null Subjects and Root Null Subjects." Rizzi, Luigi & Ian Roberts. 1990. "Complex Inversion in French." Probus 1.1-30. Roeper, Thomas. 1991. "Why a Theory of Triggers supports the pro-drop Analysis." Ms., University of Massachusetts, Amherst. Rögnvaldsson, Eiríkur. 1984. "Icelandic Word Order and það-insertion." Working Papers in Scandinavian Syntax 8. Trondheim: Linguistics Department, University of Trondheim. Rögnvaldsson, Eiríkur & Höskuldur Thráinsson. 1990. "On Icelandic Word-order once More." Modern Icelandic Syntax (=Syntax and Semantics, 24), ed. by Joan Maling & Annie Zaenen. New York: Academic Press. Schlesinger, I. 1971. "Production of Utterances and Language Acquisition." The Ontogenesis of Grammar, ed. by Dan I.Slobin. New York: Academic Press.

V2, NULL ARGUMENTS AND COMP PROJECTIONS

55

Sigurjónsdóttir, Sigríður. 1986. "Spurnarsetningar i mali tveggja islenskra barna." Master's thesis, University of Iceland, Reykjavik. (Published in Series of Selected Master's Theses, The Linguistics Institute, University of Iceland, Reykjavik 1991.) . 1987. "The Development of Subject-Verb Inversion Phenomenon in the Early Grammars of Icelandic German and English Speaking Children." Ms., University of California, Los Angeles. Thráinsson, Höskuldur. 1986. "V1, V2 and V3 in Icelandic." Verb Second Phenomena in Germanic Languages, ed. by Hubert Haider & Martin Prinzhorn. Dordrecht: Foris. Valían, Virginia. 1990. "Null subjects: A problem for parameter setting models of language acquisition." Cognition 35.105-122. . 1991. "Syntactic Subjects in the Early Speech of American and Italian Children." Cognition 40.21-81. Vikner, Sten & Bonnie D. Schwartz. Forthcoming. "The Verb Always Leaves IP in V2 Clauses." To appear in Parameters and Functional Heads: Essays in comparative syntax, ed. by Adriana Belletti & Luigi Rizzi. Oxford: Oxford University Press. Wang, Qi, Diane Lillo-Martin, Catherine T.Best & Andrea Levitt. 1992. "Null Subject Versus Null Objects: Some evidence from the acquisition of Chinese and English." Language Acquisition 2.221-254. Weissenborn, Jürgen. 1990. "Functional Categories and Verb Movement: The ac­ quisition of German syntax reconsidered." Spracherwerb und Grammatik: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie (= Linguistische Berichte, Special Issue 3), ed. by Monika Rothweiler, 190-224. Opladen: Westdeutscher Verlag. Weissenborn, Jürgen. 1992. "Functional Categories and the Form of the Grammar in Early French and German." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel. Dordrecht: Kluwer. Weverink, Meike. 1989. "The Subject in Relation to Inflection in Child Language." Master's thesis, Utrecht University. Wexler, Kenneth. 1992. "Optional Infinitives, Head Movement and the Economy of Derivation in Child Grammar." MIT Center for Cognitive Science Occasional Paper, No. 45. Massachusetts Institute of Technology, Cambridge.

Crosslinguistic Evidence for Functional Projections in Early Child Grammar Viviane Deprez Rutgers University

1.

Amy Pierce 1

Background

In recent years, research in linguistic theory has produced new proposals regarding the basic structure of the sentence. It has been proposed that subjects are generated internal to the VP projection at D-structure (i.e., in Spec-VP) (Kitagawa 1986; Contreras 1987; Koopman & Sportiche 1991, among others), that negation heads its own projection under IP (Pollock 1989; Laka 1989; Zanuttini 1990), and that crosslinguistic variation in the ordering of verbs with respect to negators results from the parametric option to raise the verb to INFL or to lower INFL onto the verb (Emonds 1978; Pollock 1989). This combination of assumptions results in the D-structures in (1) for English and French, and (2) for German, assuming that the headedness of German INFL parallels that of German verbal heads. Koopman & Sportiche (1991) have proposed a parameter distinguishing two ways of assigning nominative Case to the subject. In their view, subjects may receive Case either under agreement with INFL or under government by INFL. When nominative Case is assigned under agreement, VP-internal subjects raise to the Spec of IP; when nominative Case is assigned under government, subjects are not forced to raise and can therefore remain in VP-internal position at S-structure. Koopman & Sportiche (1991) have also argued that canonical-direct­ ional government by INFL licenses a pro subject in VP-internal position (see also Adams 1987).2 These proposals offer a new perspective on the correlated phenomena of subject inversion and subject drop. As discussed at length in Deprez & Pierce (1993), crosslinguistic acquisition data on the position of subjects, negators and verbs serve to confirm these

58

VIVIANE DEPREZ & AMY PIERCE

theoretical proposals. There are, on the one hand, systematic errors in the placement of the subject with respect to negation. These errors are as expected if subjects are generated VP-internally and if the option to assign Case under government is available in the early grammar. Errors in the placement of negation relative to the verb, on the other hand, are systematically absent. Placement of negation with respect to the verb is as expected if X°-chains are present and if the verb movement parameter is correctly set near the start of

FUNCTIONAL PROJECTIONS

59

grammatical development. In addition to providing support for certain theoretical hypotheses, our analysis challenges the view that the telegraphic nature of early speech reflects a phase during which the child's grammar lacks functional projections (Radford 1990). The early acquisition of verb movement, along with the early placement of subjects and negation, strongly imply that functional projections are represented. Before turning to the data, let us outline the developmental predictions resulting from the theoretical proposals summarized above. First, if children fail to raise the subject consistently and the VP-internal subject hypothesis holds, we expect — holding all other factors constant — to see the word order NEGSUBJ-V in early English, French and German. We establish in our previous work that such evidence clearly exists in the development of all three languages. Next, given main verb raising in French and early setting of the relevant parameter, we expect to find postverbal subjects and postverbal negation in French children's speech (i.e., Vfin-NEG-SUBJ). Given the absence of main verb raising in English, on the other hand, we do not expect to find postverbal subjects and negation in English children's speech. Nevertheless, subjects and negation are expected to be able to follow overt auxiliaries and modals, which occur in INFL, even in languages with no main verb raising. Here again, Deprez & Pierce (1993) show that this expectation is fulfilled, demonstrating the early presence of the functional category INFL. Looking at German, where INFL is a right-branching head, with the result that movement of the verb to INFL is string vacuous, the main indication of V-to-I having occurred in the absence of V-to-C is the placement of finite verbs in sentence-final position in utterances of three or more words. If verb movement to COMP is mastered early on, we would expect to find postverbal subjects and negation in two-year old German. As we reported, although sometimes instantiated, these constructions are found to be used inconsistently. The apparent lack of obligatory verb movement to COMP in German leads us to examine this type of movement from a crosslinguistic perspective. On the basis of our findings, we argue that verb movement to COMP appears to 'stabilize' later than verb movement to INFL in the acquisition of English, French, German and Swedish. The rest of this paper is organized as follows. We first review the compa­ rative data, presented in more detail in Deprez & Pierce (1993), on the position of negation and the subject in English and French acquisition. Next, we extend the survey and analysis to German acquisition. We then discuss the crosslinguistic acquisition of verb movement to COMP, considering Swedish child language as well. Note that our analysis of the acquisition evidence in Sections

60

VIVIANE DEPREZ & AMY PIERCE

2 and 3 is drawn from Deprez & Pierce (1993), and that it is presented here expressly in summary form. Finally, in Section 4, we discuss possible explana­ tions for the optional nature of verb movement to COMP in child language.

2.

The Comparative Acquisition Data

2.1. Negation in English and French Acquisition In (3), some example utterances are presented from Bellugi's (1967) wellknown study of the acquisition of English negation by three children: (3)

Examples from Bellugi's (1967) data a. not have coffee b. no singing song c. not Fraser read it d. no I see truck

Note the placement of overt subjects to the right of negation in (3c,d). While Bellugi took these to be instances of true sentential negation, there has since been some debate about the intended meaning of these child negatives. Bloom (1970) claims that young children only produce sentence-initial negatives with so-called anaphoric interpretation. In anaphoric negation, the sentence-initial negative negates a prior utterance and is structurally external to the sentence. Sentences with anaphoric negative are thus interpreted as affirmative assertions. Examination of contextual information like that in (4), however, readily distinguishes the anaphoric from the non-anaphoric cases, the latter being the only negative type we consider here. (4)

Context of utterance (3d) Mother: Did you see the truck? Child: No I see the truck. Mother: No, you didn't see it? There goes one.

As shown by the mother's question, the child's utterance is clearly not affirmative. Based on data like that in (3), Bellugi characterized a first stage in the acquisition of negation as having no subjects to the left of the negative element. In order to evaluate Bellugi's characterization of negation in early English, Deprez & Pierce (1993) examined the spontaneous speech of three children using the CHILDES computerized database (MacWhinney & Snow 1985). One of these children, Eve, was among those originally studied by Bellugi. And, as can be seen from Table 1, she has the least clear-cut NEG-initial stage of our sample.

FUNCTIONAL PROJECTIONS

61

Table 1. Early Distribution of Negatives in Three Children

Eve* Peter† Nina‡ * † ‡

18-21 months 23-25 months 23-25 months

NEG-initial

NEG-medial

12 (86%) 23 (96%) 36 (100%)

2 1 0

(14%) (4%)

data from Brown (1973) data from Bloom (1970) data from Suppes, Smith & Leveille (1973)

Table 1 shows that, in line with what Bellugi claimed, there is in fact an early period in which subject-initial negatives are rare. During this period, children usually produce null subject negatives and negatives with subjects to the right of the negative element.3 (5) contains a few of the many subject-internal utterances produced by one child, Nina, where consideration of conversational context was used to determine that each such utterance is indeed a non-anaphoric negative. (5)

Nina a. b. c. d.

(Suppes et al. 1973): some non-anaphoric negatives4 no my play my puppet, play my toys no Mommy doing. David turn no dog stay in the room, dont dog stay in the room never Mommy touch it

(6;0;2) (7;0;2) (8; 1 ;2) (9;1;2)

In short, we have evidence for early use of the NEG-SUBJ-V word order that was predicted jointly by the VP-internal subject hypothesis and inconsistent subject raising by the young child. There are, in principle, two possible lines of explanation for the position of subjects with respect to negation in these child productions. It could be assumed either that the position of the child's negative differs from that of the adult, or that the observed ordering results from the differing placement of subjects, as opposed to negatives. Previous analyses of NEG-initial constructions (Bloom 1970; Wode 1977) pursued the first alternative, proposing that the child's grammar generates the negative in COMP or, equivalently, in some position external to S/IP. The fact that children often use no rather than not was taken as further evidence for a difference between the structures of child and adult negation. Although the NEG-in-COMP analysis may appear to fit with NEG-initial data such as that in (3) and (5), it clearly fails to account for other known facts about the development of English negation. We argue in previous work that there are two types of relevant evidence. First, at a somewhat older stage (Bellugi's stage two), no itself appears in non-initial

62

VIVIANE DEPREZ & AMY PIERCE

position (e.g., he no bite you). Second, we fail to observe child utterances in which NEG occurs to the left of an overt auxiliary. That is, the form [CP No(t) [IP (subject) INFL [VP (subject) V]]] is, to our knowledge, never instantiated in children's productions. Both of these observations are inconsistent with the claim that NEG is generated in COMP in the early grammar. If, on the other hand, negation in the adult's and the child's grammar forms part of the inflectional complex, these data indicate that the early grammar, in contrast to the mature grammar, manifests the option to leave the subject in VP-internal position at S-structure. In accordance with the theoretical assump­ tions outlined above, we propose the structure in (10) for child negatives like those in (3) and (5).

While we are not claiming that the child's grammar systematically excludes subject raising during this early period, we maintain that the subject NP can be Case marked VP-internally and is therefore licensed to remain in VP-internal position. In addition to being consistent with the observed English data, as well as with the crosslinguistic data described shortly, this analysis implies a 'continuity' of structure between the adult and the child grammar. It suggests, furthermore, that the child's placement of subjects is not a syntactic innovation of the immature grammar, but rather a manifestation of a UG option for Case assignment which has been independently argued for on the basis of comparative syntactic analysis (Koopman & Sportiche 1991). This analysis is also consistent with the child's positioning of negation with respect to verbal elements. (5c) above is but one example showing that negation is correctly placed after auxiliary and modal verbs in the earliest productions. Negators in the English data are never found after a main verb. In other words,

FUNCTIONAL PROJECTIONS

63

young children never produce negatives of the type Mommy sleep no(t). Stroms wold (1990) notes that children also appear to distinguish homophonous occurrences of main verb do and have from their auxiliary counterparts, never erroneously placing negation after the former. In view of the theoretical framework adopted here, these data suggest an early setting of the verb movement parameter. That is, children correctly raise auxiliaries to INFL, but they never raise main verbs. Turning to the French data, we find that NEG is frequently positioned to the right of the finite verb. In seeming contrast to English child language, that is, negatives are often not sentence-initial.5 Consider the child utterances in (7): (7)

Some French data showing that <± finito determines location of NEG a.

b.

c.

<-finite> pas la poupee dormir not the doll sleep N(l;9;3) pas casser not break D(l;8;l) pas chercher les voitures not look for the cars P(2;l;3)

<+finite> d. elle a pas la bouche she has not the mouth N(l;10;2) e. marche pas works not D(l;8;3) f. est pas mort is not dead P(2;2;0)

Looking at (7), the position of the negative pas relative to the verb clearly depends on whether or not the verb is finite (Weissenborn 1988). This is statistically significant for a number of children measured individually (see Pierce 1989). Note that the contrast between English and French in the positioning of negation in early language is the result of verb movement in the latter case, and not a result of differences in D-structure. Within the adopted model, these data constitute strong evidence that the child's early grammar instantiates X°-chains involving verb raising to the projection of INFL. Note further that the frequent non-initial position of the negative in French is inconsistent with the view that NEG is generally generated sentence-externally in child grammar, leaving little doubt that NEG in child French, like NEG in adult French, is projected below INFL. For the nonfinite cases in (7), we assume the S-structure in (7), with the subject generated either to the left or to the right of V-bar, a possibility that Koopman & Sportiche (1991) raise for adult French.6 But for the finite cases, which involve verb raising to INFL, we propose the analysis in (8):

64

VIVIANE DEPREZ & AMY PIERCE

Acquisition data like that in (7) show quite convincingly that children have knowledge of the finite-nonfinite distinction. Paralleling the adult grammar, this distinction is expressed in terms of V-to-I. According to this analysis, data like those in (7) show that the functional projection INFL is represented in the early grammar and serves as a landing site for the verb. The observation that there are few lexical subjects appearing in sentence-initial position in early French and English negatives is worth exploring further. We turn here to a comparative look at the position of subjects in affirmative utterances in French and English. 2.2. Subjects in English and French Acquisition Another finding consistent with the VP-internal subject approach is that postverbal lexical subjects are clearly preferred to preverbal lexical subjects in early French child language, and not only in negative sentences (Pierce 1989, 1992). In (9), a few examples are provided. (9)

Some postverbal subjects in French child language

a.

b.

<+finite> lit maman reads mommy N(2;0;l) pleure clown cries clown D(l;8;3)

e. fait du bruit la voiture makes noise the car P(2;2;l) f. veut encore Adrien du pain wants more A.:SUBJ bread G(2;l;3)

FUNCTIONAL PROJECTIONS (9)

65

Some postverbal subjects in French child language (Continued)

c.

d.

<-finite> assis la poupee sit the doll N(2;2;l) dormir la Michel sleep there Michel P(2;2;l)

g. manger salade Adrien eat salad Adrien G(l;9;2) h. plus jouer tracteur bébé no more play tractor baby D(l;8;l)

The data in (9) strongly suggest that subjects are licensed in the child's grammar to remain in VP-internal position. Note that (9) includes instances of both VSO and VOS order in transitive constructions. This raises the possibility that there are two sources for postverbal subjects in French child language, one being verb raising and the other being either underlying V-SUBJ order or right dislocation. Right dislocation constructions are indeed very frequent in the informal adult language, as well as in child language. We will not consider these alternatives further here. Suffice it to say that at least some postverbal subjects must be attributed to verb raising (i.e., those with VSO word order), providing additional evidence that verb raising becomes available very early on in the acquisition of French syntax. This contrasts in a revealing way with English development, where postverbal subjects are observed very rarely. We take this to suggest that Spec-VP is generated exclusively to the left of V' in English (contrary to Kitagawa 1986). The vast majority of declarative postverbal subjects in English child language, in fact, occur with unaccusatives and be. A few examples with unaccusatives are shown in (10). (10) Unaccusatives with postverbal subjects a. going (re)corder b. came a man c. fall down lady d. broken the light

Naomi( 1 ; 10; 1 ) Eve(l;10;0) Nina( 1 ; 11 ;2) Peter(2;2;0)

Unaccusatives with postverbal subjects account for 90 percent of declarative VS order in English child language (Pierce 1992). This is exactly what we expect if the young child sometimes fails to raise the D-structure object to the Spec-IP position. At the same time, it provides evidence for the unaccusative analysis of these particular verbs. The absence of postverbal subjects in unergative contexts leads us to conclude that children acquiring English never incorrectly raise main

66

VIVIANE DEPREZ & AMY PIERCE

verbs to INFL. French two-year olds, on the other hand, widely employ the V-to-I rule. This is consistent with Emond's (1978) and Pollock's (1989) characterization of the relevant parameter distinguishing English from French, and shows that this parameter is set very early on. Not surprisingly, the input necessary to set this parameter correctly appears to be available in unambiguous form in the two languages. Prior to extending this review to German child language, let us summarize and interpret the comparative English-French results. The comparative data on the acquisition of negation show that NEG is situated below IP in early grammar. Evidence for this comes from the observation that the English child's negative never surfaces to the left of an auxiliary and that the French child's negative generally occurs to the right of the raised verb. While we do not see raising of main verbs in early English, there is strong evidence for V-to-I in early French based on the contingency between NEG-placement and the finite-nonfinite distinction. Turning to the position of the subject, note that in both languages the child, in contrast to the adult, optionally maintains subjects in VP-internal position at S-structure. Furthermore, while postverbal subjects in English child declaratives are limited to unaccusative contexts, some of those in child French arguably result from V-to-I in conjunction with unraised VP-internal subjects. The use of V-to-I in early grammar constitutes evidence for functional projections, since INFL is necessarily available as a landing site for the raised verb. We now compare these findings to some facts of German acquisition. 2.3. Comparative German Acquisition In the previous sections we have shown that the language of very young French and English speakers manifests overt VP-internal subjects and early fixation of the verb movement parameter. In this section, we turn to German and show that the data from the acquisition of this language are consistent with the findings discussed above. We will first look at the position of subjects in relation to negation and then turn to verb movement. Note that, because of its status as a verb-second language, German raises an issue not yet considered in the previous sections, namely that of the development of verb movement to COMP, a functional projection distinct from INFL. The French data discussed above confirm the existence of V-to-I movement in early grammar. The question arises whether the acquisition of V-to-C is simultaneous with that of verb movement to INFL. As we will see, the German data suggest that consistent V-to-C is, in

FUNCTIONAL PROJECTIONS

67

fact, a later development. This finding is supported when we consider the crosslinguistic evidence in a later section. Recall from the introduction that the child's option to Case mark subjects within VP under INFL government leads us to expect instances of post-negation subjects in German, just as in English and French. This expectation is fulfilled, as shown by the examples in (ll): 7 (11) Initial negation with overt subjects a. nein ich putt mache no I broken make 'I won't break it.' b. nein Auto kaput no car broken c. nein diese Messer Auau no this knife hurting 'This knife is not hurting.' d. nicht da Mama not here mommy 'Mommy is not here.' e. nein Pferd gelauf no horse run8

S(26 months)

K(24—26 months) K(25-26 months)

K(25-26 months)

H(28 months)

Park (1977) claimed that overt subjects never occur before the negation at the two-year old stage, something that has not been noted by other researchers. Recall, however, that we are not claiming that (11) reflects the only subject position available in immature German. Subjects can also precede negation, as in (12), or be null, as in (13): (12) Pre-negation subjects a. baby nich Nuckel habe baby not pacifier has 'The baby does not have a pacifier.' b. ich nein schlafen I no sleep 'I don't sleep'

S(24 months)

(from Felix 1987)

(13) Initial (non-anaphoric) negation with empty subjects a. nein Kata helfen K(24-26 months) no Kata help b. nich hause gehn M(22 months) no home go

68

VIVIANE DEPREZ & AMY PIERCE

(13) c. d.

e. f.

nichaua mache S(23 months) (don't/doesn't hurt) nein auch hause gehn M(22 months) no also house go 'I am not also going home' neine sach L(27 months) not said nei will hebe hebe no want hold (from Boser, Lust, Santelmann & Whitman 1992)

The existence of initial negation is well attested throughout the German acquisition literature (Wode 1977; Park 1977; Clahsen 1983; Felix 1987). Clahsen (1983), for one, describes an early stage with negation in initial position 70-100 percent of the time. There is also contextual evidence that these NEG-initial utterances have non-anaphoric interpretations, as discussed in Deprez & Pierce (1993). It appears, then, that German data are consistent with our previous findings with respect to the child's option to maintain the subject in VP-internal position at S-structure. Let's turn now to verb movement. Recall the structures expected under the adopted model. If the child raises the verb to COMP, negation should surface in postverbal position. If the verb raises to INFL but not to COMP, negation should occur preverbally and, given our assumption of head-final INFL, finite verbal forms should appear in sentence-final position. The examples given in (13) show that preverbal negation occurs. As a consequence, movement of the verb to COMP is decidedly not obligatory in the child's early grammar. There are nevertheless a number of cases where the negation surfaces postverbally. A few examples of this type are shown in (14): (14) Postverbal negation a. de-de-des geht nicht this goes not 'This does not work' b. paßt nicht fits not c. mag nicht like not d. pass nein fit no

1(28 months)

M(22 months) S(21 months) K(24 months)

FUNCTIONAL PROJECTIONS

69

The examples in (14) show that verb raising to COMP, although not obligatory, is nonetheless possible. There is a consensus in the German acquisition literature, however, that children have a clear preference for verb-final position during an extended period of development (Miller 1979; Mills 1985; Clahsen & Smolka 1986). Miller calculates that 90 percent of the three-word utterances of one 22-month old child (Meike) and 78 percent of another (Simone) are verb-final. As noted by Clahsen & Smolka (1986), the preference for verb-final position reflects the head-final structure of German. At the same time, it suggests a delay in the acquisition of verb movement to COMP. The German acquisition data seem to indicate that at least one instance of X°-movement, namely V-to-C, may not be systematically available at the start of grammatical development. Note that this observation does not conflict with our findings concerning the availability of verb movement in French if the final position of the verb in German reflects the movement of the verb to a finite head-final INFL. Under this view, the delay in verb movement to COMP in German does not imply lack of knowledge of the finite-nonfinite distinction, but a delay in a head movement process involving a functional projection distinct from INFL. Many verbs in sentence-final position in child German occur with apparent infinitival morphology. Examples of this type are given in (12b) and (13a). For this reason, the sentence-final position has been argued to reflect overtly the underlying position of verbs and to be strictly limited to nonfinite forms (Wexler 1992; Boser et al. 1992). It is important to note, however, that the German -en ending is not simply a marker of the infinitival form; it can also represent imperative and present tense plural forms. In fact, the -en ending is the most ambiguous form in the German conjugation. It is therefore far from clear that children's final -en forms should be systematically interpreted as nonfinite. Casting further doubt on this interpretation is the fact that -en verbal forms also occur in positions analyzed as resulting from verb-second, involving the movement of the verb to INFL and then to COMP. The child utterances in (15) below involve either a postverbal object or a postverbal subject: (15) Some -en verbs in non-D-structure position a. Julia schieben zuch (Clahsen & Smolka 1986) Julia push train b. ranstecken an zuch (Clahsen & Smolka 1986) put at train c. sitzen Puppa (Miller 1979) sit doll

70

VIVIANE DEPREZ & AMY PIERCE

(15) d.

schlafen Max sleep Max

(Miller 1979)

If, as is generally assumed, verb movement to COMP involves a prior passage through INFL, and if, as we argue on the basis of crosslinguistic evidence, verb movement to INFL in child language is limited to finite forms, then the data in (15) provide evidence that the -en ending may be interpreted by the child as a finite form. Note, furthermore, this form is by no means the only verbal ending used at this stage of development. Two-year old children are also using verbal forms that can only be analyzed as finite. According to Clahsen & Smolka (1986), young children use some finite endings correctly well before they have mastered the verb-second rule. While still in the verb-final 'stage', children are said to make finite-nonfinite distinctions by using root, third person singular and infinitival affixes (Mills 1985; Clahsen & Smolka 1986). In their study of the development of verb placement, Clahsen & Smolka (1986) explicitly distinguish a stage (their stage II) which includes finite endings in verb-final position. This suggests that the use of finite verb forms in early German has been underestimat­ ed and that children can distinguish finite and nonfinite verbal forms prior to consistent instantiation of V-to-C. Further evidence to this effect is shown in (16). Since two-word utterances are structurally ambiguous (among a structure in which the verb has remained in its base position, one in which the verb has raised to COMP, and one in which the verb has raised to a head-final INFL), it is crucial to consider three or more word utterances with sentence-final tensed verbs, such as these: (16) Final non-infinitival verbs in three-word utterances a. mone auch schlaft Simone also sleep+3SG b. ander auch geht other also go+3SG c. das auch paßt this also fit+3SG d. mama auch kam mother also came:PAST e. da Bela kuche-backe macht here Bela cake.baking does+3SG

S(22 months) M(22 months) M(22 months) M(22 months) K(26 months)

Utterances of this type are predicted to occur only if early German manifests verb movement to a head-final INFL. In addition, this phenomenon provides support for two-step movement of the verb (Vikner & Schwartz 1990).

FUNCTIONAL PROJECTIONS

71

To sum up, we have seen that the verb occurs frequently in sentence-final position in two-year old German speech, yet that the child appears to make finite-nonfinite distinctions. These findings are consistent with the assumption that, in German as in French, V-to-I raising is an early acquisition. The com­ bined findings of German, French and English acquisition provide strong support for an early setting of the verb movement parameter. As discussed above, Ger­ man acquisition data regarding the subject position are also consistent with the French and English findings, further supporting the VP-internal subject approach.

3. Verb Movement to COMP The observation of both frequent verb-final positioning and frequent NEG-initial positioning in German child language suggests the possibility of a lag in the acquisition of obligatory V-to-C movement. That is, obligatory V-to-C may not be acquired in the same way or at the same time as V-to-I. In order to substantiate this observation, we turn to other languages and structures in which V-to-C plays a role. As it turns out, crosslinguistic evidence suggests a universal delay in the acquisition of consistent V-to-C movement. In this section, we survey four types of additional evidence. 3.1. Evidence from Swedish In a recent paper by Platzack (1990), findings similar to German acquisition are reported for Swedish acquisition. Swedish, like German, manifests the V2 effect, but it differs from German in that it is underlyingly SVO. V-to-C movement is reflected in the existence of (XP)-V-S order, as well as in the placement of the verb with respect to negation and sentential adverbs. Table 2, adapted from Platzack (1990), shows that V-to-C is not fully acquired right away in Swedish. Although present throughout the earlier period, V2 is seemingly not obligatory. These facts are similar to the German facts noted above. Importantly, table 2 also shows that the Swedish children have a high number of finite forms well before the V2 constraint is acquired (roughly 80 percent), suggesting that the distinction between finite and nonfinite may be acquired earlier than the V2 rule itself. Again, this parallels the German facts as we have analyzed them.9

72

VIVIANE DEPREZ & AMY PIERCE

Table 2. Percentage of finite forms prior to V2* age

Embla finite V2 stable

22 24 26 28 30 32 34

53% 56% 85% 90% 99%

*

no no no yes yes

Freja finite V2 stable

Tor finite V2 stable

61% 68% 45% 83% 95%

39% 78% 79% 93%

no no no no yes

no no no yes

from Platzack (1990)

3.2. Evidence from French In adult French, V-to-C movement is most obviously manifested in subject clitic inversion in interrogatives. Children acquiring French consistently fail to include this construction among their early interrogatives. Some examples of non-inversion are shown in (17): (17) No clitic inversion a. Où il est? where he is b. Où ils sont? where they are c. Comment on fait pour casser les maisons? how one does to break houses 'How do you break houses?' d. Où elle est? where she is

P(2;l;3) P(2;2;0) P(2;2;0)

G(2;0;5)

This fact, although suggestive, does not provide strong evidence for the absence of consistent V-to-C raising in early French, since subject clitic inversion is not uniform in adult spoken French. Another observation providing a somewhat stronger indication is exemplified in (18). The inversion of the clitic ce is obligatory in the adult language, even in informal speech. It is therefore noteworthy that young speakers produce strictly ungrammatical utterances of the type in (18), plausibly because of the lack of obligatory V-to-C movement:

FUNCTIONAL PROJECTIONS (18) No ce-inversion a. que c'est ça dedans? what this.is this in b. et ça que c'est? and that what this.is c. que c'est ça fait du bruit what this.is that makes noise d. maintenant qu'on fait? now what.we do?

73

P(2;2;l) P(2;2;l) P(2;3;l) P(2;3;2)

3.3. Evidence from English Turning to English, SUBJECT-AUX inversion in questions is thought to be an early acquisition. This finding was replicated in an extensive study by Stromswold (1990). Stromswold reports that 93 percent of young children's wh-questions are correctly inverted, as are 94 percent of their yes/no questions. This finding might at first be thought to contradict the delay in V-to-C movement that we have observed in a number of languages. But the contradiction is only apparent. Within the model we have adopted, AUX-SUBJECT order is ambiguous between two underlying structures. It reflects either the movement of the verb to COMP over a raised subject, as in the structure [CP AUX [IP S ... ]], or the occurrence of an AUX in INFL with a VP internal subject, as in the structure [IP AUX [VP S ... ]]. A look at the contexts in which children appear to overgeneralize SUBJECT-AUX inversion points to the second alternative. According to Stromswold, children overgeneralize AUX-SUBJECT order to two contexts in which the adult grammar does not permit it. First, young children are reported to produce SUBJECT-AUX inversion in 11 percent of their 'how comequestions' (e.g., how come is that?). Second, and perhaps more interesting, children overgeneralize inverted structures to embedded contexts. As calculated by Stromswold, inversion occurs in 10 percent of embedded wh-questions. A few examples from children aged two to four years are shown in (19): (19) Inversion in embedded interrogatives a. I don't know who is she. b. I don't know what is his name. c. Do you know what's my best story? d. I don't know what is that bunny called.

(Stromswold 1990)

74

VIVIANE DEPREZ & AMY PIERCE

Finally, as discussed in Pierce (1989), children under three years may also overgeneralize AUX-inversion to declarative contexts. A few of the examples given there are shown in (20): (20) AUX-SUBJECT order in declarative contexts (Pierce 1989) a. Adult: But what I like best were the monkeys. Nina: ...Was monkeys climb on that balloon. b. Adam: Do you drink it. [comment in transcript: seems to be a statement] c. Adam: D'you take off. [comment: no question intonation in transcript] All of these errors have a common feature: they contain AUX-SUBJECT word order in contexts in which the adult grammar precludes movement of the auxiliary to COMP. If we assume that the child's inversion structures involve movement of the verb to COMP, we must conclude that possible universal constraints on V-to-C are absent from the child's early grammar. If, on the other hand, these inverted structures reflect unraised subjects and the lack of verb movement to COMP, these errors are just as predicted by our model. The latter alternative is independently supported by crosslinguistic evidence on the delayed acquisition of consistent V-to-C movement. 3.4. U-shaped Development There is yet another observation supporting our claim that the acquisition facts discussed here are accounted for in part by the VP-internal subject hypothesis. As analyzed by Stromswold (1990), the development of inversion in wh-interrogatives conforms to a U-shaped curve. That is, there is an initial period of relatively consistent AUX-SUBJECT order followed by a drop in inversion during the three-to-four year old period. It appears that children pass through a phase of confusion where the rate of correct inversion descends before climbing again to stability. Although both Stromswold and Labov & Labov (1978) note this U-shaped drop in the inversion rate, neither suggest an explanation for it. If, as we suggest, early success in inversion reflects unmoved subjects and auxiliaries in INFL, the drop in inversion may be accounted for by the onset of obligatory subject raising being prior to the onset of obligatory V-to-C movement in interrogatives. See Deprez & Pierce (1993) for further discussion.

FUNCTIONAL PROJECTIONS

75

To summarize this section, the crosslinguistic facts suggest a delay in the mastery of the V-to-C rule. Verb movement to COMP appears to be possible in the child's grammar, but its obligatory character in V2 languages and in interrogative structures is not in evidence during the early stage we have examined. The question arises, then, as to why the child's grammar fails to manifest consistency with respect to this instance of head movement. It has been suggested in the literature that this lack of stability indicates a failure to distinguish finite from nonfinite forms (Platzack 1990). This hypothesis, however, is not in line with the crosslinguistic findings we have reviewed. There is little doubt that this distinction is present at an early stage in French, and there are strong indications that it is also present in other child languages. It has also been suggested that the delay in V-to-C acquisition stems from the lack of functional projections in the child's grammar (Guilfoyle & Noonan 1988; Radford 1990, among others). As we have mentioned, very early use of V-to-I movement suggests that INFL is present from the start. The possibility of V-to-C implies that the delay in consistency or full mastery does not result from the overall absence of the COMP projection. If COMP were simply absent, we would expect V-to-C to be impossible, since there would be no appropriate landing site for the verbal head. Note, moreover, that the view that the COMP projection is not present in the immature grammar leaves the well-known early acquisition of wh-movement unexplained.10 What, then, might account for the optional character of V-to-C in early child language? In the next section, we turn to some possible explanations. After critically reviewing two recent proposals, we suggest that this delay in mastery results from the distinctive character of verb movement to COMP.

4.

The Optionality of V-to-C

4.1. An Economy Account Optionality shows up in aspects of child grammar besides V-to-C. As Wexler (1992) states, it has long been observed that the present third person singular form (e.g., he runs) is used inconsistently for a certain period by children acquiring English, alternating with the root or uninflected form of the verb. Generalizing from the crosslinguistic child language data, in particular from French and German, Wexler argues that the observed alternation in English child language is between finite and infinitival forms of the verb, and that the

76

VIVIANE DEPREZ & AMY PIERCE

optionality of V-to-C in German and that of verbal inflection in English reduce to the same thing. That is, both reflect the optionality of finiteness, or the child's lack of the knowledge that sentences require tense. Realizing finiteness is a matter of moving to TENSE, which in the case of German is in COMP. Wexler derives the optionality of finite forms in the following way. First he establishes, after Chomsky (1989), that there can only be optionality where the principles of economy of derivation determine that two derivations are equivalent in terms of cost. Since there is optionality of finiteness in child language, it must be that the derivations of finite and nonfinite forms are cost-equivalent in the child's grammar. Assuming that all finite forms in child language are derived via verb movement to INFL or TENSE (and nonfinite forms from INFL lowering to V), it must be that child grammar does not force INFL to have scope over VP because, if it did, then INFL-lowering (the nonfinite case) would entail two steps rather than one, following Pollock (1989). He then proposes that the requirement that INFL have scope over VP is absent because the young child does not distinguish values of TENSE (i.e., past and nonpast). Wexler's approach leads to the prediction that there is no past tense during an early stage in child grammar. Furthermore, it predicts that the optional infinitive stage will end only once the child has acquired the past tense, a correlation which has yet to be substantiated. Problematically, this analysis implies that main verbs raise in English child grammar. That is, according to this theory, finite verbs in early English have undergone raising to INFL. As we have discussed above with respect to the placement of negation, young English speakers demonstrate knowledge that main verbs (as opposed to modals and auxiliaries) do not raise in the target grammar. Thus, while the optional tense approach elegantly captures the French and Germanic acquisition facts, some problems remain in the case of English acquisition. 4.2. Empty Auxiliaries Another explanation for the apparent optionality of V-to-C in German early child language, which has been advanced by Boser et al. (1992), is the EMPTY AUXILIARY HYPOTHESIS. Under this view, the sentences which exhibit verb-final order are not taken to manifest optionality of finiteness or lowering of the in­ flectional features, but rather the nonlexical realization of a modal or an auxil­ iary. Verbs appear in sentence-final position in adult German whenever an overt auxiliary is present. In view of this fact, Boser et al. propose that the early child grammar is structurally identical to the adult grammar, the only difference being

FUNCTIONAL PROJECTIONS

77

that the lexicalization of the auxiliary is optional. In their study, Boser et al. observed a much greater correlation between finite forms and verb-second order than has previously been noted in the literature. That is, in their corpus, finite forms tend overwhelmingly to occur in verb-second position, whereas nonfinite forms are found mostly in verb-final position. In keeping with our findings summarized above, this result confirms that children's early syntax incorporates a knowledge of the finite-nonfinite distinction, which is expressed by the movement of the verb. In contrast to Wexler, however, it asserts that finiteness is not optional. Rather, it is always realized either visibly with the movement of the verb or invisibly through the insertion of an empty auxiliary in COMP. 4.3. Distinctions in Parameter Setting A common feature of the two analyses summarized above is the assumption that in German, tense is realized in COMP and not in head-final INFL. Yet, within a theoretical model in which affixation is the result of head movement, the realization of finiteness in embedded sentences causes problems for this assumption. V-to-C movement in adult German is impossible in embedded contexts, but this does not prevent the manifestation of tense on embedded verbs. Clearly, then, the adult grammar of German must be assumed to distinguish the projection of COMP from the projection of INFL and to permit the realization of tense in the latter. It might be argued that the child's grammar is distinct from the adult gram­ mar in this respect. In other words, since early utterances do not contain em­ bedded sentences, the child might erroneously construe COMP as the only loca­ tion of tense. The development of embedding would then reset the initial option, permitting tense realization in a head-final INFL. Note, however, that such an approach is not substantially different from the claim that COMP (but not INFL) is missing in early German grammar, since it amounts to a fusion of the role of COMP and INFL. If further research supports the rather strong correlation between finite forms and verb-second position found in the Boser et al. corpus, this hypothesis would be reinforced. But a number of questions would remain. First, if Spec-CP is the landing site of wh-movement, as proposed by Chomsky (1986), then the early acquisition of overt wh-movement in child language suggests that COMP is available as a landing site — and at the earliest stage — in those languages in which it is distinct from the projection of tense (e.g., French and English). There are thus at least some child grammars in which COMP and INFL are distinguished. Why would early German differ in this

78

VIVIANE DEPREZ & AMY PIERCE

respect? Second, there is the timing difference between the regularity of verb movement to INFL in French and that of verb-second in German. The question arises as to why this should be so. Finally, accounts which assume a fusion in the roles of COMP and INFL pose a problem for a unified approach to the acquisition of functional projections. On the one hand, verb movement in French provides strong evidence for the presence of the functional projection INFL in the early grammar, suggesting that the child's functional structure is parallel to that of the adult (the continuity hypothesis). On the other hand, an assumed fusion between COMP and INFL suggests that functional structure may not be realized in the child's grammar as in the adult's (the maturation hypothesis). To put it differently, the assumption that V-to-I and V-to-C are crosslinguistically parallel in child languages is not compatible with the assumption of an identity between child and adult functional structures, nor with the fact that these two types of verb movement manifest clear differences in the adult grammar. Although it is possible that these problems will find satisfactory answers in future research, it is important to note they would not arise if it could be maintained that COMP and INFL in the early grammar, as in the adult grammar, were both realized and distinct. This approach is thus worth pursuing, as it offers the possibility of a unified account of the acquisition of functional projections, even if the acquisition of V-to-I and of V-to-C are not fully parallel. We offer the sketch of such an approach here. Recent syntactic analysis has distinguished verb movement to INFL from verb movement to COMP in a number of respects. Rizzi & Roberts (1989) argue that V-to-I contrasts with V-to-C in that the former involves adjunction and the latter substitution. This analysis provides a natural explanation for why an overtly realized COMP can block verb movement. A further distinction is that V-to-C, in contrast to V-to-I, is never directly motivated by morphological considerations. If V-to-C does not occur, no affix, whether abstract or phonologically realized, remains stranded. Finally, as proposed by Pollock (1989), V-to-I movement involves an operator-variable relation between a tense operator in INFL and the verb. No such relation is posited with respect to V-to-C. As Pollock (1989) notes, the proposal that the I-V relation is an operator-variable relation establishes an interesting parallel between V-to-I and wh-movement. On this view, their obligatory character follows from the ban on vacuous quantification, an inviolable principle of UG. A further parallel between V-to-I and wh-movement concerns the way they have been assumed to vary crosslinguistically. Parametric variation in these movements have been argued to involve the level at which they occur, namely

FUNCTIONAL PROJECTIONS

79

S-structure or LF. As proposed by Chomsky (1981), languages which do not manifest wh-movement at S-structure manifest it at LF. Similarly, Chomsky (1989) proposes that languages which do not manifest V-to-I movement at S-structure manifest it at LF. Under this view, there are in fact no languages which fail to manifest either of these types of movement; the variation concerns only the level at which they are manifested. In such a model, the task facing the child is to determine the level at which the movement occurs. V-to-C does not entail an operator-variable relation. In this respect it is similar to NP-movement. Further parallels between these two types of movement have been suggested. Both have been argued to involve substitution, into the head of COMP and into Spec-IP respectively. Both have been argued to show parametric variation involving alternative well-formed structures at the level of S-structure. Structures with both raised and unraised NPs are assumed to satisfy the Case filter at the same level but in different ways. As proposed by Koopman & Sportiche (1991), they involve Case assignment under Spec-head agreement or under government by INFL, respectively. In the latter case, further NP movement is not required at LF.11 Similarly, in non-V2 languages, structures which do not involve V-to-C are considered to be well formed and not to require further LF satisfaction. That is, in non-V2 languages, V-to-C is not forced to occur subsequently at LF. When we confront these theoretical assumptions with the acquisition data described above, the following hypothesis emerges. Parameters involving a unique way of satisfying some UG principle, either overtly or covertly, appear to be set at the onset of grammatical development. Parameters involving distinct ways of satisfying UG principles or language particular constraints appear to be set after some period of delay in which the child's grammar manifests simulta­ neously the concurrent options made available by UG. If it is correct to view wh-movement and V-to-I as involving the first type of parametric variation, as proposed by Pollock (1989), then early acquisition is expected. This proposal is confirmed by the evidence for early V-to-I in the French data, as well as by the early acquisition of wh-movement (Stromswold 1990). We suggest that the relative delay in the acquisition of NP raising and V-to-C, resulting in the optionality described here, illustrates the second type of parameter setting. Although this proposal provides an interesting account for the acquisition data, it is clear that further research is needed. As with the other accounts outlined above, this proposal is at present tentative. Nonetheless, it permits us to explain the acquisition of V-to-I and V-to-C without presupposing that C is absent or indistinct from I in the child's grammar, or that verb movement operates

80

VIVIANE DEPREZ & AMY PIERCE

differently in child and adult grammars. It is important to note that, in the view we have sketched here, when nonfinite sentences are produced no UG principles are violated. That is, tenseless sentences may be CP and IP structures in which INFL is weak in Pollock's (1989) sense, entailing that V-to-I movement occurs neither at S-structure nor at LF. Tenseless sentences, in this view, are not syntactically deviant. Their occurrence in child language arises from the child's lack of knowledge of appropriate contexts of use. The child acquiring the languages we have surveyed may not know that tenseless sentences are restricted to embedded contexts. Untensed sentences are then plausibly understood in terms of a lack of knowledge of appropriate selectional restrictions. What the child fails to know is that tenseless sentences occur only in contexts in which they are selected for. The child's production of tenseless sentences is thus not incompatible with the assumption that the child's functional structure and grammatical principles governing verb movement are identical to that of the adult.

5.

Summary

This paper has been intended as a review of the kinds of data that have been brought to bear on the status of functional projections in child grammar. The evidence strongly suggests that functional structure is in place at the start of language development. We ended by outlining three possible accounts for what has been observed to be the optional nature of V-to-C movement in the acquisition of a variety of languages.

Notes 1.

Authors are listed in alphabetical order.

2.

Since pro subjects must also be identified (Rizzi 1986; Adams 1987), only those languages that include both licensing and identification of pro in Spec-VP position have the null subject property. In accordance with Rizzi (1986) and Huang (1989), we assume that there are essentially two ways in which a null subject can be identified — syntactically, when it is supported by 'rich' overt morphological agreement (e.g., Italian), or discursively, via discourse context (e.g., Chinese).

3.

NEG-initial includes null subject data, because, according to the adopted framework, null subjects are only licensed VP-internally.

FUNCTIONAL PROJECTIONS

81

4.

Ages here are given in the form: year;month;week.

5.

As described in Deprez & Pierce (1993), the data for two of the children, Nathalie and Daniel, were made available to us directly by Patsy Lightbown. The data for the other two children, Philippe and Gregoire, are available through the CHILDES database (MacWhinney & Snow 1985).

6.

We do not exclude the possibility that some of the nonfinite examples with postverbal subjects have undergone 'short' movement of the verb — whether the verb is an infinitive or a past participle — to an AGR-object position under INFL (Pollock 1989; Deprez 1990). In this case, the linear order NEG-V is maintained.

7.

The German examples come from a variety of sources. Simone and Meike are described in Miller (1979), Kathrin in Park (1981), Lukas in Weissenborn & Verrips (1989), Ivar and Caroline in Meisel & Müller (1992), and Holger in Boser et al. (1992). The names of the children are sometimes coded by first letter in the text.

8.

The context here is telling. The adult asks Was macht er? 'What is he doing?', and (11e) is how the child responds.

9.

A cursory examination of the on-line Danish data in the MacWhinney & Snow (1985) database reveals that obligatory V-to-C movement also appears to be delayed in the acquisition of this language.

10.

If, as proposed by Chomsky (1986), wh-movement is to Spec-CP, in the absence of this landing site we would expect children to produce wh-m-sku questions for a time — something that, to the best of our knowledge, they do not do.

11.

Note, furthermore, that there is evidence that the Case filter cannot be satisfied by further LF NP-movement. Consider the sentence, *There seems a man to be in danger, which clearly violates the Case filter. If as Chomsky (1986) proposes, expletive replacement takes place at LF, then it must be assumed that LF movement of NP to the Spec of the matrix IP cannot repair the S-structure Case violation. This suggests that NP movement at LF, if it ever occurs, cannot be motivated by the satisfaction of the Case filter.

References Adams, Marianne. 1987. "From Old French to the Theory of pro-drop." Natural Language and Linguistic Theory 5.1-32. Bellugi, Ursula. 1967. 'The Acquisition of Negation." Diss., Harvard University, Cambridge, Mass. Bloom, Lois. 1970. Language Development: Form andfunction in emerging grammars. Cambridge, Mass.: MIT Press.

82

VIVIANE DEPREZ & AMY PIERCE

Boser, Katharina, Barbara Lust, Lynn Santelmann & John Whitman. 1992. "The Syntax of V2 in Early German Grammar: The strong continuity hypothesis." Proceedings of the North Eastern Linguistic Society 22. Amherst, Mass.: Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachusetts, Amherst. Brown, Roger. 1973. A First Language: The early stages. Cambridge, Mass.: Harvard University Press. Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press. Clahsen, Harald. 1983. "Some Remarks on the Acquisition of German Negation." Journal of Child Language 10.465-469. Clahsen, Harald & Klaus Smolka. 1986. "Psycholinguistic Evidence and the Descrip­ tion of V-second Phenomena in German." V2 Phenomena in Germanic Languages, ed. by Hubert Haider & Martin Prinzhorn, 137-166. Dordrecht: Foris. Contreras, Heles. 1987. "Small Clauses in English and Spanish." Natural Language and Linguistic Theory 5.225-243. Deprez, Viviane. 1990. "Two Ways of Moving the Verb in French." Papers on Whmovement (= MIT Working Papers, 12), ed. by Lisa L. S. Cheng, 47-85. Cambridge, Mass.: Department of Linguistics and Philosophy, Massachusetts Institute of Technology. Deprez, Viviane & Amy Pierce. 1993. "Negation and Functional Projections in Early Grammar." Linguistic Inquiry 24.47-85. Emonds, Joseph. 1978. "The Verbal Complex V'-V in French." Linguistic Inquiry 9.151-175. Felix, Sascha. 1987. Cognition and Language Growth. Dordrecht: Foris. Guilfoyle, Eithne & Maire Noonan. 1988. "Functional Categories and Language Acquisition." Paper presented at the 13th Annual Boston University Conference on Language Development, Boston, 13 October. Huang, C.-T. James. 1989. "Pro-drop in Chinese: A generalized control theory." The Null Subject Parameter, ed. by Osvaldo Jaeggli & Kenneth Safir, 185-214. Dordrecht: Kluwer. Kitagawa, Yoshihisa. 1986. "Subjects in Japanese and English." Diss., University of Massachusetts, Amherst. Koopman, Hilda & Dominique Sportiche. 1991. "The Position of Subjects." Lingua 85.211-258. Laka, Itziar. 1989. "Constraints on Sentence Negation." Functional Heads and Clause Structure (=MIT Working Papers in Linguistics, 10), ed. by Itziar Laka & Anoop Mahajan, 199-216. Cambridge, Mass.: Massachusetts Institute of Technology. Lightbown, Patsy. 1977. "Consistency and Variation in the Acquisition of French." Diss., Columbia University.

FUNCTIONAL PROJECTIONS

83

MacWhinney, Brian & Catherine Snow. 1985. "The Child Language Data Exchange System." Journal of Child Language 12.271-296. Meisel, Jürgen. 1990. "Inflection: Subjects and subject-verb agreement." Two First Languages: Early grammatical development in bilingual children, ed. by Jürgen Meisel. Dordrecht: Foris. Meisel, Jürgen & Natascha Müller. 1992. "Finiteness and Verb Placement in Early Child Grammars: Evidence from simultaneous acquisition of two first languages: French and German." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 109-138. Dordrecht: Kluwer. Miller, Max. 1979. The Logic of Language Development in Early Childhood. Berlin: Springer. Mills, Anne. 1985. "The acquisition of German." The Crosslinguistic Study of Language Acquisition. Volume 1: The Data, ed. by Dan I. Slobin, 141-254. Hillsdale, N.J.: Lawrence Erlbaum. Park, Tschang-Zin. 1979. "Some Facts on Negation: Wode's four-stage developmental theory of negation revisited." Journal of Child Language 6.147-151. . 1981. The Development of Syntax in the Child with Special Reference to German. Innsbruck: Institut für Sprachwissenschaftder Universität Innsbruck. Pierce, Amy. 1989. "On the Emergence of Syntax: A crosslinguistic study." Diss., Massachusetts Institute of Technology, Cambridge. . 1992. Language Acquisition and Syntactic Theory: A comparative analysis of French and English child grammars. Dordrecht: Kluwer. Platzack, Christer. 1990. "A Grammar without Functional Categories: A syntactic study of early Swedish Child language." Working Papers in Scandinavian Syntax 45.13-34. Lund: Linguistics Department, University of Lund. (Published in Nordic Journal of Linguistics 13.107-126.) Pollock, Jean-Yves. 1989. "Verb Movement, Universal Grammar, and the Structure of IP." Linguistic Inquiry 20.365-424. Radford, Andrew. 1990. Syntactic Theory and the Acquisition ofEnglish Syntax. Oxford: Basil Blackwell. Rizzi, Luigi. 1986. "Null Objects in Italian and the Theory of pro " Linguistic Inquiry 17.501-557. Rizzi, Luigi & Ian Roberts. 1989. "Complex Inversion in French." Probus 1.1-30. Sportiche, Dominique. 1988. "A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure." Linguistic Inquiry 19.425-449. Stromswold, Karin. 1990. "Learnability and the Acquisition of Auxiliaries." Diss., Massachusetts Institute of Technology, Cambridge. Suppes, Patrick, R. Smith & Madeleine Leveille. 1973. "The French Syntax of a Child's Noun Phrases." Archives de Psychologie 42.207-269.

84

VIVIANE DEPREZ & AMY PIERCE

Verrips, Maaike & Jürgen Weissenborn. 1990. "Finite Agr as an Innate Category." Ms., Max Planck Institute, Nijmegen. Vikner, Sten & Bonnie Schwartz. 1988. "All Verb-second Clauses are CPs." Working Papers in Scandinavian Syntax 43.27-50. Lund: Linguistics Department, Lund University. Weissenborn, Jürgen. 1988. "The Acquisition of Clitic Object Pronouns and Word Order in French: Syntax or morphology?" Ms., Max Planck Institute, Nijmegen. Weissenborn, Jürgen & Maaike Verrips. 1989. "Negation as a Window to the Structure of Early Child Language." Ms., Max Planck Institute, Nijmegen. Wexler, Kenneth. 1992. "Optional Infinitives, Head Movement and the Economy of Derivation in Child Grammar." MIT Center for Cognitive Science Occasional Paper, No. 45. Massachusetts Institute of Technology, Cambridge. Wode, Henning. 1977. "Four Early Stages in the Development of L1 Negation." Journal of Child Language 4.87-102. Zanuttini, Raffaella. 1990. "Two Types of Negative Markers." Proceedings of NELS 20, 19[89], ed. by Juli Carter, Rose-Marie Déchaine, Bill Philip & Tim Sherer, 517-530. Amherst, Mass.: Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachusetts, Amherst.

The Seeds of Structure A Syntactic Analysis of the Acquisition of Case Marking Harald Clahsen University of Essex & University of Düsseldorf

1.

Sonja Eisenbeiss University of Düsseldorf

Anne Vainikka University of Massachusetts at Amherst

Introduction

Much recent work in linguistic theory suggests that syntactic differences among languages are to a large extent determined by properties of (languagespecific) lexical and morphological items (cf. Rizzi 1986; Platzack & Holmberg 1989). Chomsky (1989:44), taking up previous work by Borer (1984), has proposed that parameters of Universal Grammar (UG) should only relate to the lexicon, and not to syntax. Similarly, Rizzi (1989) argued that only heads (=X°) or properties of heads may enter into a UG parameter. In the field of language acquisition research, these approaches lead to the so-called LEXICAL LEARNING HYPOTHESIS (LLH), namely, that the syntax of a particular language can be determined by the acquisition of the lexicon (cf. Borer 1984; Clahsen 1991). In its strongest version, the LLH claims that the acquisition of grammar is in essence nothing but the learning of lexical and morphological items and their associated properties. The acquisition of Case morphology and the role it might play in the development of phrase structure is an interesting test case for the LLH. In natural language, morphological Case markers are associated with specific phrase structure positions, e.g., the accusative Case with the direct object position in German. This holds for so-called 'structural Case' which involves Case as­ signment based on the syntactic structure of the construction, but not for 'inherent Case', which is specified in the lexicon for a particular lexical item (e.g., German helfen +dative 'help'). Thus, in child language development we expect to find a close connection between the emergence of phrase structure positions and the acquisition of structural Case morphology. Given the LLH, we

86

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

hypothesize that the acquisition of structural Case markers contributes to the instantiation of language-specific phrase structure positions. Against this background, we will study in this paper the acquisition of Case markers in German and in Finnish. In German, there are differences between the various Cases in terms of their morphological realization: the possessive genitive -s is marked on the noun denoting the possessor, while accusative and dative in full noun phrases are marked on determiners. A priori, we do not know what effect this difference in morphological realization has on acquisition. To isolate this variable, we have also studied acquisition data from a language, namely Finnish, in which all Cases are marked via suffixation on the noun.

2.

Theoretical Considerations

2.1. Theories of Phrase Structure Development The question of how syntactic phrase structure emerges in child grammar is one of the most interesting puzzles in language acquisition research. Basically, three proposals have been made: the strong continuity approach the maturational approach the weak continuity approach The Strong Continuity Approach Under this view, all functional projections are present in the child's grammar from the beginning of syntactic development. The fact that certain positions are not overtly realized is attributed to peripheral factors such as the child's limited vocabulary. An analysis of German child language using the strong continuity approach is given in Weissenborn (1990) and in Boser, Lust, Santelmann & Whitman (1992). According to these analyses, the German child has an IP and a CP already at the two-word stage. The fact that the child does not use the full CP-structure is attributed, for example, to the complexity of morphological paradigms, including the partly irregular inflection of finite verbs in German. A conceptual problem we find with this approach is that it is unclear where the full-blown phrase structure tree might come from. The German child, for example, seems to have to know that German has a head-initial CP before she

THE SEEDS OF STRUCTURE

87

or he has acquired any complementizers or wh-elements. Note that this knowledge cannot be taken directly from UG, since UG does not contain phrase structure trees for the grammars of particular languages. One might perhaps assume that dominance relations between different maximal projections could be fixed in UG (cf. Roeper 1992), but, clearly, this holds neither for the internal feature specifications of functional categories, nor for directionality, i.e., for the linear order of the positions to one another. Therefore, language-specific phrase structure trees cannot be taken directly from UG. The other two approaches involve gradual development of phrase structure trees; discrete stages based on the development of individual nodes or projections are expected to be found under both approaches. The Maturational Approach Under this view, the development of grammar is taken to result from processes of maturation. The influential theory of UG-constrained maturation advocated in Borer & Wexler (1987, 1990), at least in its present form, focuses on the development of A-chains, and does not make predictions about phrasestructure development. Ouhalla (1991) speculates that functional categories mature based on a genetically determined order of development. This claim, however, has not yet been empirically verified. For example, one prediction would be that across languages, functional categories enter into the child's grammar at a particular age. It remains to be seen whether this is the case.1 The Weak Continuity Approach As an alternative to the maturational approach, it has been proposed that principles of X'-theory are available to the child from the beginning of language development, and that phrase structure positions gradually emerge in children's grammars based on the interaction of X'-principles and the input. Pinker (1984) and Lebeaux (1988) have presented the most detailed theory of phrase structure development in this framework. They suggested a set of procedures that take a string of words plus a semantic representation as input and yield a hierarchical phrase structure representation as output. Principles of X'-theory are assumed to constrain the outputs of the tree-building procedures from the beginning of phrase structure development. The child is said to start out with a grammar that contains only lexical X°-categories and to enter into phrasal syntax by projecting these categories into the corre­ sponding maximal projections (Xmax) via X'-principles. Pinker's (1982, 1984) original proposal was made in the framework of Lexical Functional Grammar (cf.

88

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

Bresnan 1982). Thus, it involved just one level of phrase structure representation and no syntactic movement. Pinker's account also involved mechanisms for assigning grammatical function indices to syntactic phrase structure positions. In contrast, Lebeaux (1988) adopted the multi-leveled framework of GovernmentBinding theory (cf. Chomsky 1981,1986). Consequently, Lebeaux, in his account of the acquisition of phrase structure, distinguished between procedures for establishing D-structure trees ('Project a') and procedures that take D-structure trees as input and yield S-structure representations ('Move a'). Moreover, Lebeaux adopted the GB-view of grammatical relations, i.e., configurationally defined rather than as theoretical primitives (as in LFG). Another important component of the Pinker-Lebeaux approach is the SEMANTIC BOOTSTRAPPING HYPOTHESIS, according to which the child initially uses semantic notions as evidence for the presence of grammatical entities in the input. In our analysis of the acquisition of Case marking, we will adopt the semantic bootstrapping hypothesis. Pinker (1984) has pointed out that it is not sufficient for the child to know that there exist nouns, verbs, agreement affixes, etc. in natural languages. Rather, the child must have a way of finding the nouns, verbs, agreement affixes, etc. in his or her particular language. The semantic bootstrapping hypothesis provides a solution to this problem which relies on two assumptions: (i) that canonical correspondences between semantic notions and syntactic categories are available in the input; (ii) that a syntactic category once posited is henceforth autonomous and freed from its prior reliance on correspondences with semantic notions. According to the semantic bootstrapping hypothesis, the child has access to a set of (conceptually based) semantic notions, such as thing, action, property, etc.; these semantic notions correspond to the set of syntactic categories such as N, V, A, etc. Pinker (1984:41) has provided a comprehensive list of such correspondences which are relevant for acquisition. Crucially, given semantic bootstrapping, the syntax-semantics correspondences are not assumed to hold bidirectionally; rather, they are only used as a way to enter into the formal system of a particular language. Thus, once a formal grammatical category is constructed based on its canonical realization, it is integrated into the grammar and is available to the child independent of its particular instantiating instance; in this way, the formal system itself is freed from its earlier reliance on semantic or conceptual notions.

THE SEEDS OF STRUCTURE

89

2.2. Assumptions on Phrase Structure Positions We adopt the standard version of X'-theory (cf. Chomsky 1986; Abney 1987), according to which (i) lexical and functional heads (Xo) project to maximal phrases (XP = X2), (ii) the complement is defined as the sister of the head, and (iii) the specifier is the sister of X1. Following Pinker (1984) and Lebeaux (1988), we assume that in language acquisition, phrasal categories (Xmax) are projected from lexical categories (Xo) based solely on positive evidence. However, we want to prevent this mechanism from automatically projecting a specifier position in each phrasal category. Let us therefore briefly consider the amount of information provided in UG (via X'-theory) concerning heads, complements and specifiers. Heads We assume that X'-theory provides information on the possible heads of maximal projections, but does not specify that all heads are obligatorily instantiated in any language. It might be that certain lexical categories (V, N, P, A, etc.) or rather their categorial feature specifications are universal. With respect to the overt realization of functional heads (and their maximal projections), however, there is substantial crosslinguistic variation. Some languages, e.g., German, have overt determiners (=DET), some languages do not, e.g., Korean (cf. Ahn 1988). Therefore, we suggest that functional heads are made available by UG as potential heads and that in acquisition, these heads (and their projections) are only posited based on evidence from the input. That is, a phrase structure position Xn is only posited if Xn is phonetically overt or if Xn is involved in some syntactic process, e.g., as a landing site for movement, in binding or in control structures. Complements We assume that XPs universally contain a complement position, i.e., a sister of Xo. This means that once the child has identified a lexical or functional head, she or he expands the lexical or functional category to the X 1 level. In this way, the complement position (YP) is established. Specifiers and Adjuncts With respect to specifier and adjunct positions, it is not clear whether all XPs may have specifiers and/or adjuncts, and whether this holds across languages. Thus, we assume that X'-theory does not provide an obligatory

90

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

specifier or adjunct position as a sister of the X1- or X2-projection, but rather an optional one. This allows certain lexical or functional XPs, e.g. PPs (cf. Bierwisch 1988) or NEG-Ps, not to have a specifier position, and it implies that the child can only posit a specifier position for a particular head based on positive evidence. To summarize, we adopt the weak continuity approach of phrase structure development according to which phrase structure positions are constructed gradually, based on X'-theory and input data. More specifically, we assume that the complement position is automatically posited together with the corresponding head, whereas specifier and adjunct positions as well as functional heads and their projections are only established through positive evidence. 2.3. Assumptions on the Developmental Relation between Case Marking and Phrase Structure In languages with overt Case marking, we suggest that there are two ways in which the acquisition of Case morphology is related to the development of phrase structure: Case-after-Position Morphological Case markers are acquired successively for each newly created phrase structure position. Case-prior-to-Position The acquisition of morphological Case markers provides evidence for the creation of new phrase structure positions. Case-after-Position Vainikka (1989) has proposed the STRUCTURAL CORRESPONDENCE PRINCIPLE (SCP) according to which there is a one-to-one correspondence between structural positions and overt structural Cases. We think that this principle is also relevant for language acquisition. Suppose that the child at some point posits a new phrase structure position (XP). Given the SCP, the child subsequently realizes (if the respective language shows overt Case morphology) that XP occurs with a specific overt structural Case (CXP). Later, the child creates a new position YP. Based on the SCP, the child notices that YP occurs with a specific structural Case CYP. Thus, under these conditions the development of structural Cases CXP, CYP, CZP,... mimics the gradual development of the syntactic tree.

THE SEEDS OF STRUCTURE

91

We hypothesized that the complement position is universally given (cf. Section 2.2) and should therefore be earlier in acquisition than the specifier and the adjunct positions. Thus, our prediction with respect to Case marking is that Case markers corresponding to the complement position should be acquired before Case markers of the specifier and adjunct positions. Case-prior-to-Position Suppose the child possesses some mechanism to acquire Case markers which does not rely on phrase structure. Then, once the child has acquired some Case marker (C), this could provide the child with the evidence needed for positing a new phrase structure position XP which was not available before the acquisition of C. In this way, the acquisition of Case markers would lead to the expansion of the syntactic tree. This scenario raises two questions: (i) If the child does not have the requisite syntactic structure, how can she or he acquire a novel Case marker? (ii) Through what mechanism does the presence of Case markers cause the child's syntactic tree to expand? With respect to the first question, we assume (following Pinker 1984) that Case markers can be acquired via semantic bootstrapping. Pinker stressed that in the early input grammatical functions (SUBJECT, OBJECT, etc.) and thematic roles are correlated in restricted ways. Thus, the child may use the thematic role of a Case-marked noun phrase as evidence that it encodes a certain grammatical function. For example, the thematic role RECIPIENT correlates with the grammatical function INDIRECT OBJECT. In a language with overt Case marking, this grammatical function might be indicated by dative Case marking. The child may exploit this correlation and determine that the indirect object is marked with dative Case. Given the semantic bootstrapping strategy, it is possible that the child acquires certain Case markers prior to the corresponding syntactic structure. Suppose, for example that in German the dative in three-place predicates is a structural Case associated with a phrase structure position, the indirect object position, which is structurally distinct from the complement position for direct objects (cf. Czepluch 1988; Wegener 1990 and the discussion in Section 3.2). Let us further assume that there is a close correspondence between the dative and the thematic role RECIPIENT. Thus, in language acquisition we may find that — as a result of semantic bootstrapping — dative Case is acquired before the unique indirect object position has been posited.

92

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

Concerning the syntactic consequences of newly acquired morphological Cases, we assume the following principle to be part of UG (cf. Vikner 1991): (1)

Case Assignment Principle Case is assigned from the head in either of the following two ways: (i) head-government (ii) Spec-head agreement

The Case Assignment Principle ensures that, once the child has acquired a new Case marker, she or he will try to establish a structural relationship in which the new Case marker can be assigned from the phrasal head to its specifier or to a governee. If this is not possible in the syntactic tree available at that point in development, the tree will be expanded. In this way, we expect that the acqui­ sition of Case markers may lead to the creation of new phrase structure positions.

3.

Case Marking and Phrase Structure in Adult German

3.1. Case

Morphology

German has three morphological Cases: genitive, accusative and dative. We will not discuss nominative Case, as it is identical to the citation form. Concern­ ing the morphological realization of German Cases, a distinction should be made between the prenominal possessive genitive which is marked on the noun denoting the possessor (as in Marias auto 'Maria's car') 2 and the other Cases. Accusative and dative are marked either on personal pronouns (mich vs. mir 'me:ACC' vs. 'me:DAT'; dich vs. dir 'you:ACC:ACC' vs. 'you:DAT' etc.) or on determiners and adjectives. 3 While Case-marked personal pronouns are suppletive forms, Case marking on determiners, possessive pronouns and adjectives involves regular suffixation. The suffixes, which combine Case, gender and number specifications, are represented in Table 1 (cf. Wurzel 1970:56): Table 1. Suffixes for Case Marking in German number

Case

nom acc dat

masc -r -n -m

singular gender fem -e -e -r

plural neut -s -s -m

-e -e -n

THE SEEDS OF STRUCTURE

93

This paradigm applies (apart from minor differences) to articles, possessive pronouns and demonstrative pronouns (Wurzel 1970:55). In addition, it applies to attributive adjectives when there is no determiner in the noun phrase that bears one of the suffixes in Table 1, e.g. dein miserabler Wein, 'your terrible wine'. That is, each noun phrase must have at least one element which has a Case suffix drawn from the paradigm in Table 1. 3.2. Phrase Structure With respect to the syntactic structure associated with the morphological Cases, we assume that accusative, dative and genitive can occur as structural Cases in German. Accusative is the structural Case associated with the direct object position, i.e., the complement of V:

Concerning the status of dative Case in German, Czepluch (1988) and Wegener (1990) have argued that at least certain occurrences of dative are instances of structural Case, namely those with three-place predicates like geben 'give'. They give several reasons for analyzing dative as a structural Case. In passive constructions corresponding to the English get-passive, for example, the RECIPIENT argument, which has dative Case in the active sentence, carries nominative Case in German. This is parallel to the accusative-nominative alternation under passivization, which is taken to be a result of structural Case assignment. Thus, based on Czepluch and Wegener, the dative of a double object verb would have to be assigned to a position that is a sister of a higher projection of the verb (e.g. V1), as shown in the following tree:4

We assume that the prenominal genitive is the structural Case of the Spec-DP position. According to Abney (1987), the genitive is assigned to the

94

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

possessor phrase in the DP in the same fashion as the nominative is assigned to the subject in the IP (cf. (4)). This suggests that the prenominal genitive is a structural Case, just like nominative Case:

Haider (1988) and Olsen (1990) have presented analyses of German noun phrases in this framework. According to these analyses, DET can be lexically realized by a personal, demonstrative or anaphoric pronoun, by a definite article or by some other non-definite determiner. One further lexical realization of DET is the genitive suffix -s which is treated as a structural Case marker that is assigned from the feature in DET in a leftward direction to the head of the phrase located in Spec-DP. It licenses the specifier by Case assignment and assigns it a thematic role. The position of adjectives within the DP is controversial. Olsen (1991) assumes that APs occupy the specifier position of NP. However, this analysis is not com­ patible with the assumption that possessor nouns are base-generated in Spec-NP. Therefore, Bhatt (1991) suggests that D° may f-select an AP which dominates NP. We will come back to this controversy in our analysis of the child data.

4.

Results on German Child Language

4.1. Hypotheses Given our assumptions on phrase structure development and Case mor­ phology, we make the following predictions:

THE SEEDS OF STRUCTURE (5)

a. b. c. d.

95

Accusative Case should be acquired prior to the prenominal genitive. As soon as the prenominal genitive has been acquired, the DP is established. Accusative Case should be acquired prior to dative. As soon as dative Case has been acquired, the indirect object position is posited. Before the acquisition of dative Case, there should be no unique phrase structure position for indirect objects.

The direct object position should occur early in acquisition, because it is universally given via UG, whereas other phrase structure positions (specifiers and adjuncts) require language-particular evidence to be posited. Furthermore, according to the SCP, phrase structure positions should have unique structural Case markers. Based on these assumptions, we predict that in German, accusative is acquired before the other structural Cases (cf. (5a) and (5c)). In German, the dative of the indirect object and the prenominal genitive are each canonically associated with a unique thematic role: dative with RECI­ PIENT, and genitive with POSSESSOR. Thus the child may identify these two Cases via semantic bootstrapping, even at a point at which she or he does not yet have the corresponding phrase structure positions. Furthermore, given the Case Assignment Principle, we expect changes in the syntactic tree, once the child has acquired a new Case marker. In this way, the acquisition of the prenominal genitive would result in the creation of the DP (cf. (5b)), and the acquisition of the dative would lead to the creation of the indirect object position (cf. (5d)). 4.2. Data There are several recent empirical studies on Case morphology in German child language: Tracy (1984, 1986) and Clahsen (1984) studied monolingual German children, Meisel (1986) and Parodi (1990) were concerned with bilingual children (German/French), and Mills (1985) provided a summary of the results from the early diary studies on German child language (Stern & Stern 1928; Preyer 1882, among others). As will be shown in the following, the descriptive results from these different sets of data are very similar with respect to the acquisition of Case morphology. In addition to the previous empirical studies, we have been able to analyze the development of Case marking in the extensive longitudinal corpus of the monolingual German-speaking child Simone. This corpus has been collected by Max Miller and parts of it have been analyzed in Miller (1976). Simone's speech

96

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

was recorded between the age of 1;7 and 3;5 at regular intervals of 1 to 14 days; the recording sessions lasted for about 2 to 4 hours. This resulted in a large corpus of about 1000 to more than 2000 utterances per month. The data we have analyzed so far with respect to Case marking range from the age of 1;10;20 to 2;09;10. During this period, Simone acquires most of the Case morphology of German. 4.3. Overview The following descriptive results support our hypotheses: Accusative forms occur earlier than dative forms (cf. (5c)). The first instances of dative Case occur at the same time as the first double-object constructions (cf. (5d)). The acquisition of the genitive -s suffix correlates with the development of nominal agreement and the elaboration of the syntactic structure of noun phrases (cf. (5b)). In addition, we observed that, contrary to the prediction in (5a), the genitive suffix -s is acquired earlier than dative and accusative suffixes. We will argue, however, that this contradiction is only apparent, and that the comparatively late acquisition of accusative suffixes is due to their morphological realization in German: these suffixes occur on determiners and thus require a carrier system (= DP) which is not available to the child before the acquisition of the genitive. 4.4. The Acquisition of the Prenominal Genitive Results from Previous Studies The -s suffix of the possessive genitive is acquired earlier than regular (non-suppletive) accusative and dative affixation. Tracy (1984:285) notes that in her data genitives are used in the first quarter of the second year, whereas accusatives occur at the beginning of the third year, and datives even later. In a later study, Tracy (1986:53) reports two very early cases of possessive genitive, given in (6): (6)

a.

b.

wode where-the 'Where is wode where-the 'Where is

Stephanies Stephanie's Stephanie's stephanies Stephanie's Stephanie's

ball? ball ball?' auto? car car?'

St( 1 ; 11 )

St(l;ll)

THE SEEDS OF STRUCTURE

97

In the data reported in Clahsen (1982, 1984), the -s suffix is missing from the possessor-possessed constructions until 2;10 (Daniel) and 2;5 (Mathias); cf. (7). From the age of 3;1 on, neither Daniel nor Mathias omits the -s suffix in the relevant contexts (cf. (8)): (7)

a. b.

(8)

a. b.

julia Julia julia Julia

zimmer auch room too' schere weg scissors gone

und julias stuhl and Julia's chair das is doch julias schijf that is Julia's ship

D(2;8) M(2;5) D(3;6) M(3;l)

For these two children, accusative-marked articles and genitives occur at the same time. However, there are many mistakes with accusatives whereas, after the occurrence of the -s suffix, the possessive genitive constructions are always correct, suggesting that the possessive -s is mastered earlier than accusative inflection (cf. Clahsen 1984 for details). Results from the Simone-corpus Recall that in adult German, the genitive suffix -s occurs in possessive N+N constructions and on proper nouns denoting possessors. In the data from Simone, the acquisition of the possessive -s takes place according to the following sequence:

•

•

non-occurrence (l;10;20-2;0;23) During this age period, Simone does not produce any -s suffix, although there are 15 N+N constructions that require the possessive genitive (cf. (9)). development (2;0;25-2;02;21) During this period, the first -s suffixes occur, but the system has not yet stabilized: Simone has 49 obligatory contexts for the possessive genitive, 33 with the -s suffix (cf. (10a, b)), and 16 without -s (cf. (10c)). mastery (2;04;17-2;09;10):5 There is a total of 41 obligatory contexts, all of them with the -s suffix (cf. (11)). (9)

maxe bauch Max belly

(10) a.

das is mones boot that is Simone's boat'

S(l;10) (H;2)

98

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

(10) b. c. (11)

mones is das Simone's is that is mone ei is Simone egg

S(2;2)

nur mones is da only Simone's is there'

S(2;4)

S(2;2)

With respect to the developmental relation between the prenominal genitive and accusative/dative, we found that before 2;0;25, i.e., before the appearance of the genitive -s, Simone has not yet acquired accusative Case. The earliest instances of accusative occur at 2;1, i.e., later than the first -s genitives. Furthermore, there are still Case errors with accusatives (cf. Section 4.6.), even after the age of 2;4, suggesting that accusative Case marking is mastered later than the possessive genitive. The results presented in this section seem to go against our claim that accusative is the earliest morphological Case in development. We suggest that the delay in the acquisition of the accusative suffixes results from the morpholog­ ical spell-out of this Case: in contrast to genitive marking, accusative marking (within noun phrases) involves regular affixation and a carrier system — the DP. The creation of the DP, however, is induced by the acquisition of the prenominal genitive (cf. Section 4.5). Therefore, the occurrence of regular accusative affixation has to await the acquisition of the prenominal genitive. 4.5. The Prenominal Genitive and the Structure of Noun Phrases In the following, we will test the claim that the acquisition of the genitive suffix leads to the creation of the DP in German child language (cf. (5b)). This hypothesis has not previously been investigated. Based on an analysis of the Simonecorpus, we will show that the acquisition of the prenominal genitive correlates with the development of nominal inflection and the structure of noun phrases. We analyzed three aspects of noun-phrase development in the data from Simone: (i) omissions of determiners, (ii) the distribution of determiners and prenominal adjectives and (iii) nominal agreement. The quantitative results for (i) are shown in Table 2: Table 2 shows that before the appearance of the genitive suffix -s, i.e., up to the age of 2;0;23, determiners are often missing in obligatory contexts. Proper nouns never occur with a determiner, although German adult language allows such combinations.

THE SEEDS OF STRUCTURE

99

Table 2. Omissions of Determiners D missing age in years;months;days

total obligatory contexts

n

in%

1000 1257 1933

654 517 223

65 41 12

1;10;20 - 2;00;23 2;00;25 - 2;02;21 2;04;17 -2;09;10

The first column provides Simone's age. The division of the corpora is based on the occurrence of the -s suffix (cf. Section 4.4). The second column contains the total number of obligatory contexts for determiners, i.e., anaphorically used determiners, (DET-)N, (DET-)AD and (DET-)AD-N structures in which determiners are required. Pronouns, one-word utterances consisting of a single noun, onomatopoeia such as bumm 'bang' and the deictic element das were not taken into account. The third column presents the total number of determiners that are missing in obligatory contexts. The fourth column shows the percentages of missing determiners calculated in relation to the total number of obligatory contexts.

The determiners occurring most frequently (within noun phrases or as pronominals) are neutralized, invariant forms without Case and gender distinctions and invariant quantifiers (cf. (12a-d)). There are also some determiners with the suffix -e (die, eine, meine etc.) which are used irrespective of the required Case and gender marking (cf. (12e)). Finally, there are some scattered examples of der and das with nouns, often with gender errors (cf. (13)). (12) a. b. c. d. e. (13) a. b. c.

wo's de lala where's the:RED pacifier da is e kugel there is a:RED ball mehr kaffee more coffee noch mehr still more auch eine also a:FEM bed:NEUT das geht nich it doesn't work das geht nich it doesn't work das geht nich it doesn't work

S(l;10) S(l;10) S(l;10) S(l;10) bett

das wecker the:NEUT alarm clock:MASC der wecker the:MASC alarm clock:MASC das ticktack theiNEUT alarm clock:FEM

S(l;10) S(2;0) S(2;0) S(2;0)

100

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

As soon as the -s suffix appears, i.e., from 2;0;25 onwards, the number of missing determiners decreases from 65 percent to 41 percent. This difference is significant; z= 11.52, p < 0.01, two-tailed. Furthermore, from 2;1 on, proper nouns often occur with definite articles, e.g. der maxe 'the Max'. When the possessive genitive is mastered (2;04;17), the proportion of empty determiners is reduced down to 12 percent. This change is also significant; z= 19.31, p < 0.01. The low omission rate for the recordings between 2;04;17 and 2;09;10 is not due to averaging effects. The mean omission rate, weighing each recording equally, is 9.5 percent, s.d. = 6.8 percent. Thus, Simone acquires obligatory determiners at the same time as the genitive -s in the possessive construction. Table 3 shows the quantitative results for nominal agreement and noun phrases with adjectives. Table 3. Noun Phrases with Adjectives noun phrases with adjectives

age in years;months;days

total

AD +N

l;10;20-2;00;23 2;00;25 - 2;02;21 2;04;17-2;09;10

116 89 95

113 65 25

agreement errors

D +AD

D+ AD+ N

n

in %

—

3 14 34

93 50 26

80 56 27

10 34

The first column provides Simone's age. The division of the corpus is based on the occurrence of the -s suffix (cf. Table 2). The second column contains the total number of noun phrases with adjectives. The following three columns contain the absolute frequencies of the following kinds of noun phrases: AD+N, D+AD, D+AD+N. The sixth column provides the total number of agreement errors in noun phrases with adjectives. The last column presents the percentages of agreement errors calculated in relation to the total number of noun phrases with adjectives.

Before the first occurrence of the genitive -s (2;0;25), adjectives and determiners occur in nearly complementary distribution: AD-N combinations occur early and frequently, but there are no DET-AD combinations and only three instances of DET-AD-N, of which at least two cases seem to be formulaic (de ande laffellalla 'the other bottle/lollie'). This suggests that the child's grammar at this stage provides only one structural position for prenominal modifiers which can be filled either by a determiner or by an adjective. After the acquisition of the possessive genitive, however, adjectives and determiners can be freely combined

THE SEEDS OF STRUCTURE

101

to yield DET-AD and DET-AD-N patterns, indicating that at this point the child has more than one prenominal position in noun phrases. Finally, the system of nominal agreement has not been established before the acquisition of -s. As shown in Table 3, 80 percent of the attributive adjectives are incorrectly inflected with respect to agreement. Consider the examples in (14): (14) a. b. c.

kleines baila small:NEUT ball:MASC kleiner baila small:MASC ball:MASC große baila

S(l;10) S(l;10) S(l;10)

big:FEM ball:MASC

d.

großes baila big:NEUT ball:MASC

S(l;10)

From 2;0;25 onwards, nominal agreement gradually develops. Table 3 shows that the number of agreement errors in noun phrases decreases significantly with the appearance of -s: from 80 percent to 56 percent (z = 3.70, p < 0.01, two-tailed), and from 56 percent to 28 percent (z = 4.0, p < 0.01, two-tailed). All suffixes occurring in adult language (-0, -e, -r, -s) are used. Thus, after the appearance of -s, Simone seems to have an agreement paradigm for determiners and adjectives. In sum, two clearly distinguishable stages of noun-phrase development have been observed. Initially, the genitive suffix -s and nominal agreement inflections are absent; determiners are optional and occur in complementary distribution with adjectives. Later, the genitive suffix -s is mastered; at the same time, the number of missing determiners decreases significantly, determiners and adjectives are inflected for gender/number and can be freely combined to yield DET-AD and DET-AD-N constructions. The Childs Grammar before the Acquisition of the Genitive In the following, we will try to represent the two above-mentioned stages of noun-phrase development in terms of grammars. Consider, first, the phrase structure tree in (15):

102

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

The tree in (15) provides only one structural position for elements modifying the noun. This prenominal position can be occupied by all kinds of adnominal modifiers, irrespective of their status in the adult grammar. Thus, it may contain determiners, quantifiers, adjectives and nouns denoting a possessor. A unique position for determiners and nominal agreement features has not yet been established. This analysis is based on Olsen's assumption that APs are generated in the Spec-NP position (cf. Olsen 1991). If one assumes that AP dominates NP (cf. Bhatt 1991), adjectives and the determiner-like elements de, e, eine etc. that were found in the earliest stage would have to be analyzed as heads of APs, whereas possessor nouns would appear in Spec-NP. In both analyses we would have a noun phrase structure without DP and only one structural position for adjectives and 'determiners'. Thus, in both syntactic analyses it is possible to account for the observed complementary distribution of these elements in the acquisition data. Moreover, the functional category DET (including its feature) and its maximal projection are absent from the tree in (15). This has several conse­ quences for the child's grammar. First, determiners are often missing at this stage, since the tree in (15) does not contain the functional category DET with features that have to be realized by appropriate lexical material (i.e., words or suffixes). In our view, the forms de, e, ein, etc. that occasionally occur in the early utterances, are not lexical realizations of DET, but optional adnominal modifiers which are generated under Spec-NP and occur only if the child wants to convey some additional information, e.g. quantification or deixis. This is consistent with the observation that, in contrast to the adult language, proper nouns never occur with a definite article at this stage. The second consequence of the absence of the DP in the early stage is that the child's grammar does not contain an element capable of assigning the genitive suffix. This accounts for the absence of the -s suffix from the relevant contexts at this stage.6 Finally, the tree in (15) does not contain agreement features. Therefore, the child can only use uninflected determiners and adjectives or default forms of these elements. In this way, the tree in (15) accounts for all the relevant properties of noun phrases in the early stage. The Childs Grammar after the Acquisition of the Genitive For the later stage, we propose the phrase structure tree in (16) overleaf. Compared to the earlier tree in (15), a DP has been introduced above the NP. This accounts for the observed changes. First, determiners are categorized as lexical realizations of DET, just like in the adult language, and occur as heads

THE SEEDS OF STRUCTURE

103

of the DP. Thus, they are no longer optional modifiers, and the proportion of missing determiners decreases. Second, the introduction of the DP removes the limitation to one structural modifier position. Determiners are categorially and structurally distinguished from other prenominal elements. Therefore, they can be freely combined with these elements, for example with adjectives. Third, the feature inside DET is now available and can be realized in terms of gender and number inflections on determiners and adjectives. Finally, the -s suffix can now be used as a Case marker in possessor-possessed constructions because the tree in (16) contains the feature , which assigns genitive Case from the head of the DP to its specifier position. Thus, we conclude that the differences between the two observed developmental stages are related to the creation of the DP. Finally, we want to argue that in German child language the genitive suffix -s leads to the creation of the DP in child German. There are several reasons for that. First, the -s genitive is easily accessible from the input: it canonically encodes the concept of possession in the input and can thus be acquired via semantic bootstrapping. The -s genitive is also easy to isolate because it is only realized on nouns and not on functional elements. Moreover, our data indicate that the acquisition of the prenominal genitive is correlated with other develop­ ments in the inflection and the syntax of noun phrases. We suggest that this correlation follows from the Case Assignment Principle. This principle requires a Case-assigning head (or feature of the head) and a position to which Case can be assigned. According to Olsen (1990), the genitive suffix is assigned from inside the functional category DET. In the earlier stage (cf. (15)), however, the feature is absent from the child's noun phrases. Since the possessor is not a complement, it cannot receive Case from the head noun. Thus, in the earlier tree in (15), there is no way of assigning genitive Case to the first noun of a possessor-possessed construction. This could lead to the creation of a new functional category with Case-assigning features and a position for Case-marked elements, as soon as the child (based on semantic bootstrapping) has identified the genitive suffix as a Case marker.

104

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

4.6. The Acquisition of Accusative and Dative In this section, we will test our hypotheses that accusative is acquired before dative, and that the acquisition of the dative leads to the creation of the indirect object position (cf. (5c, d)). We will first briefly summarize results from previous studies and then examine the data from the Simone-corpus. Results from Previous Studies The results of previous studies on monolingual and bilingual children (Clahsen 1984; Mills 1985; Tracy 1986; Meisel 1986; Parodi 1990) converge on the following six observations: First, noun phrases initially occur in the citation (= nominative) form without overt Case marking, even in contexts in which accusatives or datives would have been necessary in the adult language (cf. Clahsen 1984:9; Mills 1985:178; Tracy 1986:54f.). Even after the appearance of accusative and dative forms, Case-neutral forms are found in contexts in which non-nominative Case marking is required; consider the following examples taken from Clahsen (1984):7 (17) a.

b.

hab der keubeu noch nich auf D(3;0) have the:NOM cowboy yet not on (correct: den:ACC; D. says that he has not yet put on the cowboy hat.) noch ein fisch malen J(2;4) another:NOM fish draw (correct: einen:xcc; J. wants her mother to draw another fish.)

Second, dative- and accusative-marked personal pronouns occur earlier than regular accusative and dative suffixes on determiners and adjectives (Mills 1985:181; Tracy 1990:392f. and Parodi 1990). In the data from Clahsen (1984), for example, the first Case-marked forms to occur were mich 'me:ACC' and dich 'you:ACC' (Julia 2;4, Mathias 3;0) and the invariant definite article form den 'the:ACC'.8 With the third child, Daniel, Case-marked pronouns appeared somewhat later, at the age of 3;3. Furthermore, Parodi (1990:28) observed that in her data Case-marked personal pronouns are used correctly at a point at which full noun phrases are not yet inflected for Case. Third, accusative forms occur at a younger age than datives. For example, one of the three children studied in Clahsen (1984), Julia, produced only accusative forms; the second child, Daniel, used accusative forms earlier (3;1)

THE SEEDS OF STRUCTURE

105

than datives (3;4); with the third child, Mathias, both occur at the same time; similar results are reported in Tracy (1986:59). Fourth, accusative markers are overgeneralized to contexts where dative would have been required in the adult language, but not vice versa9 (Mills 1985:178 ff.; Tracy 1984: 289; Clahsen 1984: 12); consider the following examples from Clahsen (1984:11): (18) a.

b.

da freß ich dich die zuckerkluntje auf then eat I you:ACC the lollies up (correct: dir.DAT; Then, I will eat up your lollies.) ich such mich eins aus I choose me:ACC one out (correct: mir.DAT; I will choose one for me.)

M(3;6)

D(3;5)

Finally, it was found in studies on monolingual children that accusatives and datives are not overgeneralized for subjects; cf. Clahsen (1984) and Tracy (1984, 1986).10 In order to verify the findings from previous studies, we will now analyze the data from the Simone-corpus with respect to accusatives and datives. Results from the Simone-corpus First, Simone uses Case-neutral forms before she has accusative or dative Case marking; this is similar to the children from previous studies. The first unambiguous instances of non-nominative Case markers occur at the age of 2;1 (cf. (19)). n Before that age, there are only Case-neutral forms, but even after 2;1 we find many Case-neutral forms (cf. (20)). (19) a. b. (20) a.

b.

hase fängt den wauwau hare catches the:ACC doggie mone hält den Simone holds him:ACC

S(2;l) S(2;l)

nich der hut auf S(2;l) not the:NOM hat on (correct: den:ACC; S. takes off the hat.) der mußde größer machen S(2;4) he:NOM must-you bigger make (correct: den'.ACC; you should make it bigger, (the stable).)

Second, Simone produces Case-marked personal pronouns before accusative or dative suffixes. The two sentences in (21) contain the earliest instances of

106

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

Case-marked personal pronouns in Simone's data. The only accusative form that also occurs at this age is the definite article den (cf. (19) above). (21) a.

b.

mir schmeckt nich me:DAT please not (=I don't like (the meal).) gib mir no mehr give me:DAT more

S(2;l)

S(2;2)

Even after the first occurrences of accusative-marked definite articles (2;1), indefinite articles, possessive and demonstrative pronouns still appear as Case-neutral forms (cf. (22)). The earliest instances of the suffix -n on deter­ miners other than den can be found at 2;4 (cf. (23)). (22) mone macht ein pfirsich makes a:NOM peach S. (correct: einen:ACC)

S(2;l)

(23) ich mal einen mund I draw a:ACC mouth

S(2;4)

Thus, the data from the Simone-corpus and from previous studies provide evidence for developmental differences between Case-marked pronouns, the accusative form den, and other Case-marked determiners. We argue that this observation can be attributed to differences in the morphological realization of these elements: while the child initially has fixed, word-specific entries for Case-marked pronouns (and den) which can be directly inserted from the lexicon, regular Case affixation is not yet possible. Since inflectional affixes cannot be used in isolation, a carrier system is required, such as a set of stems or roots to which affixes from the general paradigm can be added. The carriers of German Case inflections are determiners. Therefore, as long as the DP has not been established, regular accusative Case affixation is not possible. Third, the first clear instances of datives in Simone's data12 are lexical datives and cliticized dative forms on prepositions (cf. (24)). These two types of datives and the first unambiguous accusative forms appear at the same time (2;1). Thus, with respect to lexical datives and datives in PPs, it is not possible to observe a developmental difference between accusatives and datives in Simone's data. However, our hypothesis for the developmental order of accusative and dative only pertains to direct accusative objects versus indirect dative objects. For these elements, we find a developmental difference: The first dative markers on indirect objects appear at 2;2 (cf. (25)), i.e., later than accusative markers on direct objects (2;1).

THE SEEDS OF STRUCTURE (24) a.

b.

c. (25)

107

gehört dem maxe belongs the:DAT Max ((Something) belongs to Max.) mir schmeckt nicht me:DAT likes not (I don't like (it).) hier is de häsi im bett here is the hare in+the:DAT bed

S(2;l)

gib mir no mehr give me:DAT more

S(2;2)

S(2;l)

S(2;l)

Fourth, Simone's data contain 11 overgeneralizations of the accusative suffix -n in dative contexts (cf. (26a, b)). In addition to that, there are two examples of the cliticized form im (= in+the:DAT) in cases in which accusative is required (cf. (26c, d)). Most probably, these forms are not internally analyzed by Simone. (26) a.

b.

c.

d.

den wauwau auch S(2;l) the:ACC doggie also (correct: dem:DAT; S. wants the dog to also get some food.) gib en Tommy milch S(2;5) give the:ACC Tommy milk (correct: dem:DAT) der nimmt im mund S(2;5) he takes in+the:DAT mouth (correct: in den:ACC Mund) der pappa ist im wasser gefalln S(2;5) the daddy is in+the:DAT water fallen (correct: ins Wasser)

Fifth, overgeneralizations of non-nominative forms to subjects do not occur in the Simone-corpus. Finally, the data show that the first instances of datives coincide with the first occurrences of indirect objects. The earliest double-object constructions are found at 2;1 (cf. (27a)), i.e., immediately after the first instances of lexical datives. As the indirect objects in these utterances are proper names that do not require determiners, dative Case is not realized overtly. The first overtly Case-marked indirect objects appear at 2;2 (cf. (25)). Before 2;1, the indirect objects of three-place predicates are always left out (cf. e.g., (27b)). On the whole, the data suggest that the acquisition of dative Case leads to the creation

108

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

of the indirect object position and that this position is not available before dative Case has been acquired. (27) a.

b.

die flasche gebe bias the bottle give Tobias 'give the bottle to Tobias' maxe windel anziehn Max diaper put on 'Max puts the diaper on (the child).'

S(2;l)

S(2;0)

4.7. Preliminary Summary The -s suffix of the possessive genitive is acquired earlier than the dative and accusative suffixes. We have argued that the comparatively late appearance of accusative suffixes is due to differences in the morphological realization of the three structural Cases in German: accusative marking requires a carrier system (= DP) which at the time the direct object position emerges (presumably at the two-word stage) is not yet available to the child. The acquisition of the genitive -s suffix correlates with other developments, particularly with the acquisition of nominal agreement and the structural elaboration of noun phrases. This correlation indicates that the possessive genitive leads to the creation of DP in German child language. Accusative forms occur earlier in development than dative forms. Thus the Case of the universally-given direct object position occurs before the Case of the later-developing indirect object position. The first instances of datives coincide with the first occurrences of double-object constructions, indicating that the indirect object position is posited, as soon as dative Case markers appear. These results support our predictions in (5) suggesting that there is a close connection in child language development between the acquisition of structural Case markers and the creation of new phrase structure positions.

5.

Case Marking and Phrase Structure in Adult Finnish

In terms of the morphological realization of Cases, adult Finnish is more homogeneous than German. In Finnish all Cases are marked on the noun. Thus, comparing Finnish and German will enable us to determine in which way the morphological realization of Case markers influences their acquisition.

THE SEEDS OF STRUCTURE

109

We adopt Vainikka's (1989) analysis of adult Finnish according to which three of the 15 Cases in Finnish are structural: partitive, genitive and elative. There are two objective Cases in Finnish, accusative and partitive. Vainikka argued that partitive is the basic Case of the object (cf. the tree in (28)) because partitive has a much wider distribution than accusative. Whereas partitive occurs on verbal complements in many affirmative sentences and all negative sentences, the occurrence of accusative is narrowly restricted: accusative is used to imply completed action (cf. (29a) vs. (29b)). Furthermore, partitive is also the Case of complements of prepositions (cf. (29c)), while accusative never occurs in PPs.

(29) a.

b.

c.

Söin omenaa I-ate apple:PART 'I was eating the apple.' Söin omenan I-ate apple:ACC 'I ate the apple (completely).' Liisa lähti ilman Maijaa NOM left Without PARTT

'Liisa left without Maija.' The genitive Case in Finnish occurs in the specifier position of all lexical heads (i.e., N, A, V, and P; cf. (31)). This gives us the tree in (30).

(31) a.

b.

c.

Näin Maijan koiran I-saw GEN dog:ACC 'I saw Maija's dog.' Liisa on Maijan näköinen NOM is GEN similar-in-appearance (ADJ) 'Liisa resembles Maija.' Liisa on Maijan takana NOM is GEN behind 'Liisa is behind Maija.'

110

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA (31) d.

Liisa käski Maijan tulla ulos NOM asked GEN come+INF out 'Liisa asked Maija to come out.'

Finally, Vainikka observed that the complements of nouns and adjectives as well as optional complements of verbs and prepositions (unless they bear any lexical or semantic Case) typically occur in elative Case (cf. (32)). On the basis of these observations, Vainikka has argued that the elative Case is the structural Case of the adjunct position of all lexical heads; cf. Vainikka (forthcoming) for details on the elative. (32) a.

b.

c.

d.

e.

Liisa on ylpeä Maijasta NOM is proud Maija:ELA 'Liisa is proud of Maija.' Maija on vihreä kateudesta NOM is green envy:ELA 'Maija is green with envy.' Tämä on kertomus suomalaisista this is story Finns.ELA This is a story of the Finns.' Liisa sai sakot ylinopeudesta NOM got charges speeding:ELA 'Liisa got a ticket for speeding.' Minusta Liisa on mukava I:ELA NOM is nice 'I think Liisa is nice.'

In sum, according to the analysis we adopt from Vainikka (1989), partitive is the structural Case of the complement position, genitive is the structural Case of the specifier position (of lexical heads) and elative is the structural Case of the adjunct position.

6.

Results on Finnish Child Language

6.1. Hypotheses Given our general approach on the acquisition of Case and phrase structure development and our assumptions on adult Finnish, we can make the following predictions for Finnish child language:13

THE SEEDS OF STRUCTURE (33) a. b.

111

Partitive should be acquired prior to genitive, Partitive should be acquired prior to elative.

Similar to accusative in German, partitive Case is associated with the (universally given) complement position and should therefore emerge earlier than the two other structural Cases which correspond to later-developing phrase structure positions. 6.2. Results We reanalyzed the child language data presented in Toivainen (1980). Toivainen studied the acquisition of Finnish inflectional affixes by 18 mono­ lingual children. The corpus involved longitudinal diary data. Altogether 32 different inflectional suffixes were used by most of the children by the age of 3;10. In Table 4, the children's ages at the time of the first occurrence of the three relevant Cases are given. Table 4. The Acquisition of Structural Case in Finnish* name of the child Arja Harri I1po Katja Kirsti Kyösti Leevi Maila Marja Marko Mika Niina Saana Saila Sanna Teppo Ville Ulla mean: * t

partitive 2;4 1;10 1;10 1;3 1;8 l;8 2;0 l;8 2;2 1;11 2;4 l;9 2;2 2;2 1;8 1;9 1;5 2;l 1;10

genitive 2;4 2;4 2;2 1;10 2;9 1;5 2;0 1;8 2;5 l;ll 1;10 2;3 2;2 2;7 2;2 2;6 l;8 2;1 2;1

elative 2;10 2;4 3;1 2;5 2;9 2;3 3;4 3;9 3;4 2;4 2;7 2;3 3;1 3;8 2;2 2;6 2;3

— † 2;7

Based on tables 46, 47 and 52 in Toivainen (1980). As the data of one child, Ulla, do not contain any instances of elative inflection, we excluded this child from our statistical analysis of the developmental relation between partitive and elative.

112

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

We found that partitive occurs significantly earlier than genitive Case (t (34) = 2.253, p < 0.05, two-tailed) and significantly earlier than elative Case (t (32) = 6.132, p < 0.01, two-tailed). This confirms our hypotheses in (33). Table 4 shows that 17 children (out of 18) produce elative forms after the acquisition of partitive; one child (= Ulla) does not have elative at all. Table 4 also shows that partitive occurs before genitive Case for 10 of the 18 children. With 6 children (Arja, Leevi, Maila, Marko, Saana and Ulla), partitive and genitive appear at the same time. However, the recording in which the two Cases occur for the first time is preceded by a break of 3-9 months (cf. Appendix 2 in Toivainen (1980)); that is, the two Cases have been acquired between the two recordings, and we have no way of telling which occurred first. The two remaining children (Kyösti and Mika) have genitive Case before partitive Case. Studying the actual utterances from Kyösti (Appendix 2 and Table 52, p. 141) reveals that there is only one instance of genitive in a recording in which there are no instances of partitive Case (1;5). In the subsequent recording (1;8), both partitive and genitive appear. The data from Mika, however, are clearly different from the other data: Mika's first recording (1;10) contains instances of genitive, but no partitive; both Cases occur in the next recording (2;4). In sum, the available data on Finnish child language support our hypothesis that partitive Case is acquired before genitive and elative. We suggest that this is due to the fact that partitive is the structural Case associated with the (universally given) complement position, whereas genitive and elative Case are structural Cases of later-developing positions.

7.

Summary and Conclusion

The theoretical approach to the development of Case marking and phrase structure advocated in the present study involves three major assumptions: (i) children construct phrase structure trees in a gradual fashion, based on X'-theory and language-particular evidence from the input; (ii) children are guided by the Case Assignment Principle and the Structural Correspondence Hypothesis which state a close relationship between Case morphology and syntactic positions; (iii) children may acquire Case markers via semantic bootstrapping. Based on these assumptions, we hypothesized two ways in which the acquisition of Case morphology and the development of phrase structure could be related to each other in language development: (i) Case-prior-to-position,

THE SEEDS OF STRUCTURE

113

(ii) Case-after-position. According to (i), the acquisition of morphological Case markers leads to the creation of new phrase structure positions. We have argued that this holds for the acquisition of the prenominal genitive and the dative in German child language. The second scenario, 'Case-after-position', means that whenever the child has established a new syntactic position, she or he will also realize that this position is associated with a unique structural Case (if the language has one corresponding to this position). This predicts that the emergence of structural Cases will mimic the development of the syntactic tree. We found that in German child language, the structural Cases do not appear in the expected order: regular accusative Case marking is acquired later than the prenominal genitive. We argued that the comparatively late acquisition of accusative Case results from its morphological realization in German. Regular accusative marking requires a DP as a carrier system for Case marking which the child has not yet established in the two-word stage. Once the DP has been created, accusative is acquired as well. In a language where all morphological Cases are marked on the noun, we would expect to witness the acquisition of objective Case before other structural Cases. This has been shown to hold for Finnish child language: the partitive, the structural Case of the complement position, is acquired before the other structural Cases (genitive and elative).

Acknowledgments This research is supported by two grants of the German Science Foundation to the first author (grant-No C1 97/1-1, 1-2). We are grateful for specific comments from Ken Drozd, Nigel Duffield, Peter Indefrey, Gary Marcus, Teresa Parodi, Claudia Schmidt, Bonnie Schwartz, Dieter Wunderlich, and from two anonymous reviewers. The data from the Miller corpus were kindly made available to us by Jürgen Weissenborn. Notes 1.

Felix (1987) provided an analysis of child German in which the maturation of X'-principles at the age of roughly two years was considered to be responsible for the development of syntactic phrase structure. Felix allowed for arbitrary rule systems in early child language which are later expunged through the maturation of X'-principles. We think that this approach is theoretically unrestrictive and should only be taken if the data force us to do so. Such data, however, have not been presented by Felix; cf. Clahsen (1992) for a reanalyis of Felix' account of German child language without maturation.

114

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

2.

The postnominal genitive (e.g. die Blumen der Frau 'the flowers of the:GEN woman') as well as the genitive selected by certain verbs such as gedenken (e.g. Er gedenkt des Mannes 'he thinks of the+GEN man') involves Case marking on determiners and adjectives. These types of genitive are comparatively rare in the input and were not attested in the data under study. Mills (1985:185 ff.), for example, reported that none of the children studied in the early diary studies used the postnominal genitive before the age of six years. In contrast to the English genitive -s, the prenominal -s genitive in German is restricted to proper nouns.

3.

Besides the prenominal genitive, only dative plural is marked on the head noun, with the suffix -n (e.g. den Kindern 'the:DAT children:DAT'); this Case marker is, however, very rare, and does not occur in the relevant period of development.

4.

Exactly where the indirect object occurs in the VP is problematic. As represented in the tree in (3), the indirect object position would be identical to the Spec-VP position; this would present a problem for the view that subjects are base-generated in Spec-VP. We will leave this question open.

5.

Note that there are no data at (2;3).

6.

The N-N constructions that appear in the early stage seem to violate the Case Filter, because according to our analysis the prenominal possessor nouns cannot be Case-marked. Notice however, that the Case Filter applies to DPs and not to NPs. Thus, as long as the DP has not been established, it operates vacuously.

7.

Recall from the paradigm in Table 1 that the -n suffix is the only accusative suffix that is distinct from its nominative counterpart, whereas all dative suffixes are distinct from both the nominative and the accusative.

8.

In previous studies (cf. Clahsen 1984 and Tracy 1986) it has been observed that the accusative form den is used as a definite article and as a pronoun and that it occurs earlier in development than the accusative affix -n on indefinite articles, possessive and demonstrative pronouns. This developmental difference between Case-marked definite articles and Case markers on indefinite articles and other determiners suggests that the early instances of den do not result from regular accusative Case affixation; rather, they seem to be suppletives for the child.

9.

In an elicitation experiment, Eisenbeiss (1991) found that some children (3;8 to 5;2) used dative forms for direct objects when the direct object preceded the indirect object in the VP (cf. (i)). When the direct object occurred in other positions, the same children produced correct accusative forms (cf. (ii)). (i) gib der maus der Schildkröte VP 1(4;2) give the:DAT/FEM mouse the:DAI/FEM turtle' 'Give the mouse to the turtle.' und die wolle seb ich der_ katze VP 1(4;2) (ii) and the:ACC/FEM wool give I the:DAT/FEM cat 'and I give the wool to the cat.'

THE SEEDS OF STRUCTURE

115

These data suggest that dative is structurally assigned in double-object constructions, since the error does not depend on a specific verb, but only on the structure of the sentence. One might speculate that the use of dative in such examples is related to the development of scrambling. 10.

Rosemary Tracy (pers. comm.) reports that in her recent studies on older children one child (4;5) sometimes uses accusative markers for the subject. The significance of these data for the acquisition of Case morphology is not clear, since, under normal conditions, the acquisition of accusative and dative Case in monolingual children takes place much earlier than 4;5. In her study on bilingual children, Parodi (1990) found three overgeneralizations of accusative forms to subjects at the age of 3;0 and 3;8 (cf. for example (i)). Mills (1985:179) reported similar examples from Leopold's bilingual child Hildegard (5;0). These data are in contrast to monolingual children. (i) zur feier is den paputt gegangen C(3;0) to the fire has it:ACC broken 'It got broken in the fire.'

11.

The first instances of den occur at the age of 2;0; these, however, are just isolated one-word or two-word utterances without a verb. The situational contexts suggest that Simone considers den to be a variant of the deictic pronoun das, cf. for example (i): (i) den uhr S(2;0) the:ACC/MASC watch (correct: die:FEM uhr; context: 'Welche uhr ist schöner? Maxe hat zwei uhren! (= 'Which watch is nicer? Max has two watches.) At 2; 1, den is used in sentences with verbs and can no longer be regarded as a deictic pronoun.

12. The earliest dative markers occur in PPs, simultaneously with the accusative form den, namely at 2;0. However, similar to the initial occurrences of den (cf. footnote 11 ), these early instances of datives are hard to interpret, since they occur in incomplete or partially intelligible sentences. It seems premature to draw any strong conclusions from these data. (i) im kücheschrank S(2;0) in-the:DAT cupboard' (ii) hier im mund S(2;0) here in-the:DAT mouth 13.

Furthermore, we would predict that the acquisition of genitive should lead to the creation of the specifier position of lexical heads, and that the adjunct position is posited as soon as the elative is acquired. However, the child Finnish data available to us (cf. Toivainen 1980) do not allow us to carry out a phrase structure analysis.

116

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA References

Abney, Stephen. 1987. "The English Noun Phrase in its Sentential Aspect." Diss., Massachusetts Institute of Technology, Cambridge. Ahn, Hee-Don. 1988. "Preliminary Remarks on Korean NP." Papers from the Sixth International Conference in Korean Linguistics, ed. by Eung-Jin Baek, 1-15. Seoul: Hanshin. Bhatt, Christa. 1991. "Deg(ree) in Adjektivphrasen." Ms., Cologne University. Bierwisch, Manfred. 1988. "On the Grammar of Local Prepositions." Untersuchungen zur Semantik, ed. by Manfred Bierwisch, Wolfgang Motsch & Ilse Zimmermann, 61-99. Berlin: Akademie-Verlag. Borer, Hagit. 1984. Parametric Syntax. Dordrecht: Foris. Borer, Hagit & Kenneth Wexler. 1987. "The Maturation of Syntax." Parameter Setting, ed. by Thomas Roeper & Edwin Williams, 123-172. Dordrecht: Reidel. . 1990. "A Principle-based Theory of the Structure and Growth of Passive." Ms., University of Massachusetts, Amherst and Massachusetts Institute of Technology, Cambridge. Boser, Katharina, Barbara Lust, Lynn Santelmann & John Whitman. 1992. "The Syntax of V2 in Early German Grammar: The strong continuity hypothesis." Proceedings of the North Eastern Linguistic Society 22. Amherst, Mass.: Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachusetts, Amherst. Bresnan, Joan. 1982. "Passive in Lexical Theory." The Mental Representation of Grammatical Relations, ed. by Joan Bresnan, 3-86. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1981. Lectures on Government and Binding: The Pisa lectures. (= Studies in Generative Grammar, 9.). Dordrecht: Foris. . 1986. Barriers. Cambridge, Mass.: MIT Press. . 1989. "Some Notes on Economy of Derivation and Representation." Functional Heads and Clause Structure (=MIT Working Papers in Linguistics, 10), ed. by Itziar Laka & Anoop Mahajan, 43-74. Cambridge, Mass.: Massachu­ setts Institute of Technology. (Published in Principles and Parameters in Comparative Grammar, ed. by Robert Freidin, 417-454. Cambridge, Mass.: MIT Press, 1991.) Clahsen, Harald. 1982. Spracherwerb in der Kindheit. Tübingen: Günther Narr. . 1984. "Der Erwerb von Kasusmarkierungen in der deutschen Kindersprache." Linguistische Berichte 89.1-31. . 1986. Die Profilanalyse. Berlin: Marhold. . 1989. "The Grammatical Characterization of Developmental Dysphasia." Linguistics 27.987-920.

THE SEEDS OF STRUCTURE

117

Clahsen, Harald. 1991. "Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language. Language Acquisition 1.361-391. . 1992. "Learnability Theory and the Problem of Development in Language Acquisition." Theoretical Issues in Language Acquisition, ed. by Jürgen Weissenborn, Helen Goodluck & Thomas Roeper, 53-76. Hillsdale, N.J.: Lawrence Erlbaum. Czepluch, Hartmut. 1988. "Kasusmorphologie und Kasusrelationen: Überlegungen zur Kasustheorie am Beispiel des Deutschen." Linguistische Berichte 116.275-310. Eisenbeiss, Sonja. 1991. Passiv und Kasus—Evidenz aus der deutschen Kinder sprache. Master's thesis, Cologne University. Felix, Sascha. 1987. Cognition and Language Growth. Dordrecht: Foris. Haider, Hubert. 1988. "Die Struktur der deutschen Nominalphrase." Zeitschrift für Sprachwissenschaft 7.32-59. Lebeaux, David. 1988. "Language Acquisition and the Form of the Grammar." Diss., University of Massachusetts, Amherst. Meisel, Jürgen. 1986. "Word Order and Case Marking in Early Child Language: Evidence from simultaneous acquisition of two first languages: French and German." Linguistics 24.123-183. Miller, Max. 1976. Zur Logik der frühkindlichen Sprachentwicklung. Stuttgart: Klett. Mills, Anne. 1985. "The Acquisition of German." The Crosslinguistic Study of Language Acquisition. Volume 1: The data, ed. by Dan I. Slobin, 141-254. Hillsdale, N.J.: Lawrence Erlbaum. Olsen, Susan. 1990. "AGR in the German Noun Phrase." Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences, ed. by Christa Bhatt, Elisabet Löbel & Claudia Schmidt, 39-49. Amsterdam: John Benjamins. . 1991. "Die deutsche Nominalphrase als Determinansphrase." DET, COMP und INFL: Zur Syntax funktionaler Kategorien und grammatischer Funktionen, ed. by Susan Olsen & Gisbert Fanselow, 35-56. Tübingen: Max Niemeyer. Ouhalla, Jamal. 1991. "Functional Categories and the Head Parameter." Paper presented at the 14th GLOW Colloquium, Leiden, March. Parodi, Teresa. 1990. "The Acquisition of Word Order Regularities and Case Mor­ phology." Two First Languages: Early grammatical development in bilingual children, ed. by Jürgen Meisel, 157-192. Dordrecht: Foris. Pinker, Steven. 1982. "A Theory of the Acquisition of Lexical Interpretive Grammars." The Mental Representation of Grammatical Relations, ed. by Joan Bresnan, 655-726. Cambridge, Mass.: MIT Press. . 1984. Language Learnability and Language Development. Cambridge, Mass.: Harvard University Press.

118

HARALD CLAHSEN, SONJA EISSENBEISS & ANNE VAINIKKA

Platzack, Christer & Anders Holmberg. 1989. "The Role of AGR and Finiteness." Working Papers in Scandinavian Syntax 43.51-76. Lund: Linguistics Department, University of Lund. Preyer, William. 1882. Die Seele des Kindes. Leipzig: Grieben. Rizzi, Luigi. 1986. "Null Objects and the Theory of pro." Linguistic Inquiry 17.501-557. . 1989. "On the Format for Parameters." The Behavioral and Brain Sciences 12.355-356. Roeper, Thomas. 1992. "From the Initial State to V2: Acquisition principles in action." The Acquisition of Verb Placement, ed. by Jürgen Meisel, 333-370. Dordrecht: Kluwer. Stern, Clara & Wilhelm Stern. 1928. Die Kindersprache. Leipzig: Barth. Toivainen, Jorma. 1980. "Inflectional Affixes Used by Finnish-Speaking Children Aged 1-3 years. Suomalaisen Kirjallisuuden Seuran Toimituksia [Publications of the Association of Finnish Literature] 274. Hämeenlinna: Suomalaisen Kirjallisuuden Seuran [Association of Finnish Literature]. Tracy, Rosemarie. 1984. "Fallstudien: Überlegungen zum Erwerb von Kasuskategorie und Kasusmarkierung." Syntaktische Struktur und Kasusrelationen, ed. by Hartmut Czepluch & Hero Janssen, 271-313. Tübingen: Günther Narr. . 1986. "The Acquisition of Case Morphology in German." Linguistics 24.47-78. . 1990. "Sprachliche Strukturentwicklung." Diss., University of Göttingen. Vainikka, Anne. 1989. "Deriving Syntactic Representations in Finnish." Diss., University of Massachusetts, Amherst. . Forthcoming. "The Three Structural Cases in Finnish." Forthcoming in Case and Other Topics in Finnish Syntax, ed. by Anders Holmberg & Urpo Nikanne. Vikner, Sten. 1991. "Verb Movement and the Licensing of NP-Positions in the Germanic Languages." Diss., Geneva University. (Revised version forthcoming in Oxford Studies in Comparative Syntax, Oxford University Press.) Wegener, Heide. 1990. "Komplemente in der Dependenzgrammatik und in der Rektions- und Bindungstheorie: Die Verwendung der Kasus in Deutschen." Zeitschrift für germanistische Linguistik 18.150-184. Weissenborn, Jürgen. 1990. "Functional Categories and Verb Movement: The ac­ quisition of German syntax reconsidered." Spracherwerb und Grammatik. Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie (= Linguistische Berichte, Special Issue 3), ed. by Monika Rothweiler, 190-224. Opladen: Westdeutscher Verlag. Wurzel, Wolfgang. 1970. "Studien zur deutschen Lautstruktur." Studia Grammatica VIII. Berlin: Akademie-Verlag.

From Adjunct to Head

Teun Hoekstra Leiden UniversitylNIAS

1.

Peter Jordens Free University Amsterdam/NIAS

Introduction

1.1. Three Types of Categories Current generative theory makes a distinction between LEXICAL and other categories, including FUNCTIONAL categories. This distinction broadly corre­ sponds to the traditional concepts of open and closed class elements. So, nouns, verbs, and adjectives, and potentially prepositions are considered lexical categories. Apart from these (major) lexical categories, there is a variety of other expressions. The latter again can be divided into two classes, according to current theory. On the one hand, there are elements which are claimed to have the status of heads. Like lexical elements, they function as the head of a phrasal category, taking a complement and a specifier, in accordance with X'-theory (cf. Chomsky 1986). Instances of such functional categories are the category COMP of comple­ mentizer elements, the category DET of determiner elements, as well as several classes of inflectional heads, including TENSE, AGR(eement) and NEG(ation) (cf. Pollock 1989). To briefly illustrate, consider the example in (1). The nominal phrase the book is considered to be a DP, headed by the functional category DET, to which the determiner the belongs. DET takes an NP complement, headed by the lexical category N, to which book belongs. The D-element the is in complementary distribution with other D-elements, among which the genitival -s. This element differs from the in requiring a DP in its specifier, as shown in (lc) (cf. Abney 1987; Szabolcsi 1983).1

120

TEUN HOEKSTRA & PETER JORDENS

This leaves us with a heterogeneous set of elements which belong neither to the set of lexical categories, nor to the set of functional categories. This set comprises all sorts of modifying elements, such as adverbs of various sorts. They are not considered to be heads projecting their own phrasal structure, but rather are generally taken to be adjoined to some other projection. As an example, con­ sider the adverb often in (2a). Simplifying somewhat, the structure of (2a) can be represented as in (2b), where often is adjoined to the lexical projection VP.

We shall refer to this third category as ADJUNCT EXPRESSIONS. It is not always easy to establish whether a particular expression instantiates a functional category or an adjunct category, given our current understanding of linguistic categories. One criterion for the status of functional category is that the element is in complementary distribution with other elements belonging to this category.

FROM ADJUNCT TO HEAD

121

Another criterion is based on intervention effects. To briefly illustrate, consider the English negative adverbs not and never. As is illustrated in (3), not blocks the linking of I(nflection) and V, while the negative adverb never does not. This difference can be explained under the following assumptions: INFL must be associated to V. Both INFL and V are heads. Not is also a head, belonging to the functional category NEG. NEG is generated in between INFL and V. The HEAD MOVEMENT CONSTRAINT (HMC) Or the EMPTY CATEGORY PRINCIPLE (ECP)

excludes a relationship between heads across another head (cf. Pollock 1989; Chomsky 1991). Therefore, do-support is called for to host the INFL-features, linking of these to V being excluded by the intervening head NEG. Never, on the other hand, has no such blocking effect. This would follow if never is not represented as an intervening head, but rather as an adjunct. (3)

a. b. c. d.

John John John John

does not read a book. never reads a book. INFL NEG [VP read a book] INFL [VP never [VP read a book]]

The difference vis-à-vis the HMC therefore motivates a distinct categorization of not and never, the former instantiating a functional category, the latter belonging to an adjunct category. 1.2. Functional Categories: The acquisitional debate As is well known, children's utterances typically lack closed-class expressions in positions where they are obligatory in the adult grammar. Given their abundant availability in the input, their absence is rather surprising, and an explanation is called for. The issue is much debated in the current acquisitional literature under the heading of the acquisition of functional categories. One position is that the initial absence of closed class items can be explained on the assumption that functional categories are not available in the initial stage, as they are subject to maturation (cf. Radford 1990a; Lebeaux 1988; Guilfoyle & Noonan 1988). An important point, which is often overlooked in discussions about the acquisition of functional structure, is that a distinction needs to be made between a functional category F and elements instantiating this category. Several properties make it rather unattractive to assume that the category F itself has to be learned in the process of acquisition. Typically, functional categories are nondescriptive, i.e., there is no direct relation between anything in the nonlinguistic situation that F can be said to describe. Moreover, more often than not elements

122

TEUN HOEKSTRA & PETER JORDENS

instantiating F are phonologically weak, e.g., an unstressed affix. Thirdly, the morphological paradigms expressing F often show syncretism with other affixes — e.g., TENSE and NUM(ber) are fused — as well as full suppletions with irregular forms. The very fact that F is acquired at all, even at the ages at which this occurs, very strongly suggests that learning is not at all an adequate concept. Rather, it seems more attractive to assume that F is given by UG, either initially or at some later point as a result of maturation. Learning, then, is restricted to identifying the elements which instantiate F. The important question, however, is whether the identification of some expression E is sufficient to establish that E instantiates category F. There are two possibilities to consider: (i) F is available, as it is part of the initial state; (ii) F is not available in the relevant period, as it matures at a later point. Clearly, if F is not yet available in the grammar, such identification is excluded in principle. However, if F is absent, this does not preclude the identification of E per se; only the categorization of E as belonging to category F is impossible. It is conceivable in principle that E is identified and categorized as instantiating some category other than F. Next, consider the alternative under which F is available. There are reasons to believe that even then identification of E does not necessarily trigger its categorization as F, at least not in general. To see why this is so, let us consider some examples. Our first example concerns pronominal possessives: in some languages we find possessive adjectives, as in Italian, while in other languages they have the status of determiner elements, i.e., a functional head which selects NP as its complement. The syntactic status of pronominal possessives thus varies between an adjunct within the DP-structure or the head of the DP-structure. Both options are apparently consistent with UG. As another example, consider modality. In English, as in many other languages, there are modal elements in the inflectional system, i.e., may, can, must, etc., which are considered to be generated as instances of a functional head. However, modality is categorially also expressed by adverbs and adjectives, syntactically represented as adjuncts on some or other projection. As a final example, consider negation. As argued by Zanuttini (1990), negation may be expressed by an element heading a functional projection NEGP, or by an adverb which is adjoined to some other projection, as well as by a combination of these. Again, both options are apparently allowed by UG. From this it follows that the mapping of an element to a functional category is not semantically driven. We would like to formulate this as our first maxim: (4)

There is in general no unique mapping from meaning to a syntactic status of functional head.

FROM ADJUNCT TO HEAD

123

A consequence of this is that the absence of an expression E before a certain age is not explained by the theory of maturation of functional categories: E may be identified and represented in a manner other than as the head of a functional projection. Similarly, presence of some expression E of category F in the adult grammar at a certain point of development does not by itself constitute sufficient evidence that F is part of the grammar at that stage. It is clear that this signifi­ cantly complicates the debate over the status of functional categories in acquisition.2 Let us consider the issue somewhat more closely. We are assuming now that there are two ways in which some element X can combine with some projection YP: either X is adjoined to YP as in (5a), or X is a functional head, taking YP as its complement and projecting its own phrase, as in (5b). (5)

a.

YP

b.

X'

X YP X YP Both structural options are available on the basis of UG, and, given (4), the child has to choose on some other basis than the semantics of X with respect to YP. Of these two options, adjunction requires less information, and is also less restricted, i.e., for X to relate to YP under a head-complement relation, X must categorially select YP, X must be restricted to zero-level expressions, X must be in complementary distribution with other members of its category and X may trigger intervention effects on relationships that the head of YP may entertain with other heads (cf. the discussion of English not in Section 1.1). None of these restrictions hold if X adjoins to YP. If the child is driven by set-theoretic

124

TEUN HOEKSTRA & PETER JORDENS

which also illustrate this path of acquisition, i.e., those elements that are later accommodated inside a functional projection are initially treated as adjuncts. The first case involves degree expressions, modifying adjectives; the second involves the acquisition of the functional category C(omplementizer). 2.

Negation and Modality

2.1. Some Properties of Dutch Syntax Before entering the discussion of our first example, it is perhaps useful to briefly sketch the structure of Dutch clauses. Dutch is an SOV language, displaying Verb Second in root clauses. We shall assume, following more or less standard analyses, that the SOV character is due to the head-final nature of the functional category INFL.3 The verb moves from its base position in VP to INFL. In root clauses, the inflected verb moves to the head of the CP, where it is preceded by maximally one XP. This accounts for the rigid verb-second order, as well as for the asymmetry between root and embedded clauses (cf. den Besten 1983). This description is illustrated by the examples in (6a)-(6b) overleaf, which both derive from the structure in (6c). The complementizer dat of the embedded clause is absent in the structure underlying (6b), but its position serves as the landing site of the inflected verb. Two further points require mentioning. The first concerns modal verbs in Dutch. While the English modals can, may etc. might be given a special status in the grammar, i.e., as instantiations of some inflectional functional category, there are no special properties of Dutch modal verbs that set them apart from other verbs (cf. Evers 1975), apart from their modal meaning. We may therefore treat them as regular verbs syntactically, i.e., as heading a VP. The second point concerns the status of negation. Sentential negation is normally expressed by niet, but we find all sorts of cases in which the negation finds expression in some fused form, such as nooit 'never', or geen (which may be considered a fusion of a negation and an indefinite determiner). We shall assume that niet is an adverb, adjoined to VP, and disregard other negative constituents, as they play no role in the discussion below. Note that this means that both the status of modal verbs and the overt expression of sentential negation in adult Dutch differs from that in adult English.4

FROM ADJUNCT TO HEAD

125

2.2. Negative Modals in Early Dutch 2.2.1. Two Negatives and Their Distribution With this background, let us now turn to a discussion of the acquisition of negation and modality in Dutch in the very early stages, i.e., from 1;7 to 1; 11 months. The discussion is based on an investigation of longitudinal diary data, ranging from 1;3 years of age until 6;0 years, collected by Peter Jordens from his child, Jasmijn. This empirical domain in other languages is recently subject of extensive study within acquisition studies in the framework of GB-theory (cf. Clahsen 1988, 1991; Meisel & Müller 1992; Deprez & Pierce, this volume; Weissenborn 1990; Roeper 1990; Wexler 1991 and many others). In Jasmijn's data two negative elements are used in this period: nee 'no' and niet 'not', with nee dying out at 1;11. Before 1;11, however, both are used, and it would be misleading to suggest, therefore, that the use of the anaphoric nee constitutes a

126

TEUN HOEKSTRA & PETER JORDENS

stage prior to the use of niet (contrary to the claim about German, made by Wode 1977). Nee and niet also occur in adult Dutch, but nee is used exclusively as an anaphoric negation in the adult grammar (i.e., in negative answers to questions). Jasmijn uses nee both as an anaphoric negation and as a sentential negation (cf. Bellugi 1967, who notes the same about English no and not). Given the existence of two elements, we face the question of their relative distribution. The following is a fair summary: (7)

Relative distribution of niet and nee a. nee never occurs with modal verbs such as kan and mag: here we find niet, or rather nie exclusively; b. nee never occurs with finite verbs; c. niet never occurs with nonfinite verbs; d. in nonverbal sentences, both niet and nee are used.

Clearly, then, the relative distribution of nee and niet is not random, but results from some system. What is this system? We claim that the distribution follows from the following characterization of the meaning of nee and niet: (8)

a. b.

niet is a nonmodal negation: it negates a description pertaining to the here-and-now; nee is a modal negation, dominantly expressing boulemaeic negation, i.e., meaning 'I do not want'.

It is not surprising that the adult anaphoric negation is extended to a modal negation with the meaning 'I don't want'. From an anaphoric negative response {Do you want to sit here? No, I don't want to sit here.) it is generalized to other negative wishes (nee poes hier zitten 'I don't want the pussy cat to sit here'). 2.2.2. Finite and Nonfinite Clauses Let us now consider how (7) follows from (8). (7b) and (7c) express a complementarity of niet and nee in the domain of finiteness. As is well known, at the relevant stage we find both infinitival and finite constructions, even though the distribution of finite constructions is rather limited. We exclude constructions with modal verbs at this point, returning to these later on. The analysis of finite constructions in early Dutch and German is subject of considerable debate. On the one hand, people argue that these constructions should be analyzed in the same way as their adult counterparts, basically in the way we analyzed (6b) above (cf. Wexler 1991; Verrips & Weissenborn 1992; Poeppel & Wexler 1993; Hyams, this volume). Others (e.g., Clahsen 1991; Meisel & Müller 1992; Penner

FROM ADJUNCT TO HEAD

127

1990; Whitman, Lee & Lust 1990, among others) take intermediate positions, the precise nature of which does not concern us here, but which all hold that one or more functional categories found in adult grammar are not yet available. We shall not take a stance in this discussion, as it is irrelevant for the point under investigation. Another issue that divides the minds at present, is whether the finite verb is indeed generated in some position lower in the structure, or rather base-generated in their surface position. We shall adopt a movement approach here, without intending to commit ourselves to any particular position in this debate. For ease of exposition, we give the following representation of a finite construction in early Dutch:

In infinitival sentences, the verb does not move to the F-position, but rather stays in VP (or in INFL), i.e., in final position. The occurrence of independent infinitival clauses is not limited to Dutch, but seems to be a general feature of early stages of acquisition. The proper analysis of such infinitival sentences is not clear: what is it that allows such clauses in early grammars, but not in later grammars? There are two hypotheses that come to mind. The first hypothesis would be to assume that such clauses lack certain functional projections which are required in the adult grammar, and which are responsible for the obligatoriness of a finite verb. One might for instance assume that adult main clauses are CPs, while children allow a choice of either CP or IP. If the finiteness requirement resides in COMP, infinitival clauses would be allowed under the choice of IP instead of CP (cf. Rizzi, this volume, for such a hypothesis). Alternatively, it might be hypothesized that the full CP-structure is available, as in adult grammar. Infinitival sentences could then be analyzed as finite sentences from which a finite auxiliary has been dropped (the modal-drop analysis). Wexler (1991) makes an alternative proposal. He argues that infinitival clauses occur up until the child distinguishes two tense values (past-present). The infinitival verb does not move up for reasons of economy in the sense of Chomsky (1991). We may summarize these three hypotheses by reference to the structure in (9). If the position labeled +F is equated with COMP, this position might be obligatory in the adult system, but optional in the child's grammar. If +F is equated with Tense, movement to this position is optional for the child as long as no Tense values are distinguished. Alternatively, a finite auxiliary occupies the +F position, but is dropped. So, under either of these proposals, the +F-position is not available for the infinitival verb, which hence remains in a position lower in the structure. Finite and nonfinite constructions show two patterns of negation: in finite

128

TEUN HOEKSTRA & PETER JORDENS

contexts, we find postverbal negation, whereas nonfinite constructions feature preverbal negation. This is not only true for Dutch, but appears to be a rather general phenomenon, reported for Swedish (Lundin & Platzack 1989; Plunkett & Strömqvist 1990), French (Weissenborn, Verrips & Berman 1989; Pierce 1989) and German (Clahsen 1988; Weissenborn, Verrips & Berman 1989; Wexler 1991; Poeppel & Wexler 1993). These two patterns can be explained in terms of the representation in (9), if we assume that the position of the negation is in between the +F-position and the V-position (or I-position). As stated in (7b, c), there is a strong correlation between the use of nee vs. niet and the choice between a finite and a nonfinite construction. What explains this correlation? We would like to maintain the statements in (10): (10) a. b.

Finite constructions describe the here-and-now. Nonfinite constructions do not describe the here-and-now.

The claim in (10a) squares well with the hypotheses mentioned above concerning the licitness of infinitival structures, certainly if we exclude the utterances with modal verbs, as we did. Under the modal-drop analysis, as well as under Rizzi's hypothesis, the use of the +F-position indeed signals nonmodality. If the claim in (10a) is correct, the use of niet follows from the characterization of its meaning in (8a), just as the non-occurrence of nee follows from its meaning. If (10a) is correct, the correctness of (10b) would appear to follow straightforwardly: absence of finiteness would signal absence of a here-and-now description. This would in turn imply that nonfinite constructions have a modal interpretation, as absence of reference to the here-and-now requires that whatever is being said relates to some other world-time coordinate, to which the speaker relates the proposition in some modality or other. This concurs with the intuition behind the modal-drop analysis. Such a distinct meaning of finite and nonfinite constructions would appear to be optimal, as it provides a raison d'être for the distinction. However, the claim is not uncontroversial. Lightbown (1977) points out that in early French, infinitival verbs are also used to describe an ongoing activity, along with finite forms. Wexler (1991) and Poeppel & Wexler (1993) also argue that the choice between finite and nonfinite forms is just optional, with no difference of meaning being attached to it. Indeed, the claim that nonfinite constructions are always modal in interpretation may be too strong. In fact, this is consistent with a proper interpretation of the modal-drop analysis, mentioned above, which should perhaps be referred to as auxiliary drop. Jordens (1990) notes that after the relevant period (i.e., after 1;11), there is a rapid increase of auxiliaries in what we have called the +F-position. Often, such sentences with finite auxiliaries are intended to refer to ongoing activity, even if the sentence

FROM ADJUNCT TO HEAD

129

does not have such an interpretation in the adult language. An example is given in (11), which would refer to a future event of eating in adult Dutch, but which was clearly intended to refer to an ongoing activity by the child: (11)

Gaat hij eten? goes he eat

Let us therefore conclude by saying that while (10a) is correct for finite sen­ tences without modal verbs, (10b) is too strong, and should be changed to (12): (12)

Nonfinite sentences do not necessarily refer to the here-and-now.

Nee, being incompatible with finite constructions due to its inherent modality, is perfectly suitable in nonfinite constructions, which may have a modal interpreta­ tion. This brings us close to the complementarity described in (7b, c). The fact that niet is not found in nonfinite constructions would follow from (10b), but not from (12). However, it might be reasonable to assume that negative non-infini­ tival constructions are necessarily used in modal contexts, even though positive infinitivals may refer to ongoing activity. We leave this as an open question. 2.2.3. Modals Let us now turn to sentences with modal verbs. A significant observation concerning the modal elements that we find in the relevant period, basically kan 'can' and mag 'may', is that they only occur in negative sentences, i.e., we find the forms kannie and magnie without kan and mag occurring by themselves. The same is observed for the first occurrences of modal elements in English by Klima & Bellugi (1966). They claim that at the relevant stage the combinations of modal and nt constitute unanalyzed units, for which the absence of positive counterparts of modals as well as the absence of inversion are presented as evidence. We follow this suggestion. Under this construal, then, nee, kannie and magnie fall in a single category of negative modals. These are in complementary distribution within the domain of infinitival constructions, while this entire context is itself in complementary distribution with niet, which occurs in finite environments. This distribution is summarized in (13). (13)

Distribution of negation in verbal contexts: finite nonfinite niet

kan/mag + nie vs. nee

If these claims are correct, the fact that niet and nee have an overlapping distribution in nonverbal contexts hardly comes as a surprise. Our analysis

130

TEUN HOEKSTRA & PETER JORDENS

entails, however, that nonverbal utterances with niet differ in interpretation from such utterances with nee. Some examples are given in (14)—(15): (14) a. b. c. d. (15) a. b. c.

die niet that not die niet goed that not good Peter niet oppe hand Peter not on.his hand niet voor poes not for pussycat

1;10

nee poes vlees no pussy meat nee thee no tea nee Tom Poes buiten. Cynthia ook niet no Tom Poes outside. Cynthia also not

1;10

1;10 1;11 1;11

1;10 1;10

It is hard to go beyond the impressionistic level, but our interpretation is that the utterances in (15) do not describe the current situation, e.g., in (15a) the situation that the pussy cat has no meat, but rather the negative desire with respect to that situation: 'I don't want the pussy cat to have meat'. The utterances in (14), on the other hand, give just an assessment of the situation. We see, then, that the semantic difference we claim in (8) captures the distribution described in (7). The system of modality we have set up can be further supported by also taking the non-negative utterances into account. Via (10) we impute a modal distinction to hold in non-negative sentences as well, i.e., modality comes in two polarities. Interestingly, the positive polarity of boulemaeic modality is also overtly expressed in the utterances of the child, by the element minne/unne/hunne, meaning something like 'I want'. Like nee, its negative congener, minne occurs both in nonfinite verbal environments and in nonverbal sentences. Some examples are given in (16)—(17): (16) a.

minne hoene uit

1;7

MINNE shoes out

b.

'I want my shoes out.' unne die open MINNE that open ' I want that open.'

1;7

FROM ADJUNCT TO HEAD (17) a.

131

hunne Mijnie sijfe

1;7

MINNE Mijnie write

b.

'Mijnie wants to write.' hunne opbase papa

1;7

MINNE inflate daddy

'I want daddy to inflate.' (a balloon) This leads to the following analysis: (18)

System of negation and modality in verbal utterances until 1;10

positive negative

finite

finite + niet

(kan/mag) kan/mag + niet

minne nee

The brackets around kanlmag under positive epistemic modality are intended to indicate that these forms are virtually absent, as mentioned above, contrary to the negative counterparts. The claim that nee and minne are correlated can be shown by inspecting some quantitative data. Table 1.Distribution of Modal Expressions negative age 1;10 1;1l 2;0

kan/mag-nie + VP 7 16 7

16% 80% 88%

positive nee + VP 38 4 1

84% 20% 12%

kanlmag + VP 2 3% 19 73% 7 88%

minne + VP 73 97% 7 27% 1 12%

The two columns under each modal expression represent the total number and the percentage.

We see in these figures that while both nee and minne have a large distribution at 1;10, they disappear in a period of two months. At the same time, the use of modal verbs expands. This is particularly noticeable for the use of modal verbs in positive environments: while kanlmag hardly occurs at 1;10, they show a rapid increase at 1;11. Also, other modal verbs, such as moeten 'must' and willen make their appearance. From a semantic point of view, these figures are slightly misleading, as they suggest that nee and minne give way to kanlmag(+niet). However, kanlmag are in the epistemic section of the modal system, while nee and minne are in the

132

TEUN HOEKSTRA & PETER JORDENS

boulemaeic section. It is in itself quite remarkable that there are no verbs of desire at 1;10. At 1;11, however, the verb willen is acquired, and it is this verb that replaces minne. Negative desires are not yet rendered by a combination of willen + niet: rather, in contexts where this might be expected, the negative epistemics are used, which is a quite sensible functional replacement. Instead of ik wil niet S, the child says S mag/kan niet. So, at 1;11 the system of modality is as in (19): (19)

Modal system at 1 ; 11

negative positive

kan/mag-nie kan/mag wel

nee, *(wil niet) MINNE, wil

Some examples with willen are given in (20): (20) a.

b.

poes il mij vinger happe pussycat want me finger bite 'Pussy cat wants to bite my finger.' ik wil mellek pakke I want milk take 'I want to take the milk.'

1;11

1;11

2.4. The Syntactic Representation of Negative Modals How can we interpret this acquisitional process? Sofar we have not been very specific about the syntactic analysis of the modal system, but one may have wondered why we put the negative modals kan/mag + niet in a separate category from the finite verbs. The claim we want to make is that until 1;10 there is no syntactic difference between these negative modals and nee. In adult grammar, nee is an adverb, while kanlmag is a finite verb, but we propose that at the relevant stage, there is no syntactic distinction. Instead, we propose that both nee and kanlmag + niet are adjoined to the projection of which the nonfinite verb is the lexical head. We shall use the label VP for this projection, although it would not be impossible to argue that this constituent is labeled IP instead, depending on the precise analysis of infinitivals. So, the structure we propose is as in (21). The modal negations are treated as adjuncts on a verbal projection, rather than treated as heads selecting the verbal projection as their complement.

FROM ADJUNCT TO HEAD

133

The main motivation for this proposal concerns a rather drastic developmental change which takes place around 1;11: (i)

(ii) (iii)

(iv)

The intonational pattern changes from the typical doubly stressed adjunction pattern to an integrative pattern with a single stress on the infinitive: kánnie zitten becomes kannie zitten. This integrative accentuation is typical of a head-complement structure, where stress is placed on the lexical head of the complement. The nonverbal modals nee and minne disappear. Whereas earlier kanlmag + nie can either precede or follow the nonfinite construction, it now precedes it. If kanlmag were treated as finite verbs, this positional freedom is hard to account for, as finite verbs invariably occur in first or second position. Other modal elements make their appearance, and kanlmag start coming in use without nie.

This change can be represented in the grammar if we assume that instead of the adjunctional structure in (21), the grammar has a head-complement structure, with a requirement on the head that it be verbal, i.e., the structure is as in (22). Although the hierarchical relation remains the same, the functional relation as well as the categorial labelling has changed, kanlmag are now analyzed as finite verbs, the nie-part being identified with the nonmodal negation niet, and nee becomes obsolete as it is not a finite verb.

134

TEUN HOEKSTRA & PETER JORDENS

2.5. The Triggering of Reanalysis The final question is what triggers this restructuring. This question is not easy to answer, as it involves various assumptions about the child's grammar at the relevant point. Two important aspects will play a role: first, the child may have found a morphological analysis of kan/mag-nie, identifying nie with niet; secondly, the freedom to use nonfinite sentences may become excluded. The precise interpretation of this depends on which hypothesis about this use is preferred. Following Rizzi's (this volume) proposal, we might formulate this acquisitional step as involving the obligatoriness of CP, taking our +F-position as COMP. Kan/mag would then be interpreted as instantiating +F, in accordance with the adult system. If Wexler's (1991) hypothesis is correct, we would predict that the child would start making tense-distinctions around the same time. We leave this as a matter for future research. The shift from (21) to (22) thus constitutes our first example of the development from modifying adjunct to governing functional head. We have argued that before the age of about 1;11, the child's grammar has a set of modal negations, which have the syntactic status of adjunct on a verbal projection. Around that time, a change takes place, after which these modal negations have disappeared from the grammar: the element nee disappears, the elements magnie and kannie are reanalyzed as consisting of the negation niet and the modal verbs maglkan, which occupy a functional head position, which we referred to as +F. We now turn to our second example.

3.

Scrambling and Subscrambling

In this section we discuss the acquisition of SCRAMBLING. The section is organized as follows. In 3.1 the basics of the adult system as relevant to the phenomenon of scrambling are introduced and briefly discussed. In 3.2 the development of scrambling in the period of 2;1 to 2;8 is examined. The final section is devoted to a particular type of scrambling, in which part of a nominal expression is extracted, something which is not allowed in the adult system. We explain this phenomenon of subscrambling in terms of our general hypothesis that the child may provide an adjunction structure for material which is later hosted by a functional category.

FROM ADJUNCT TO HEAD

135

3.1. Scrambling in the Adult System We now move on to a later stage of development, starting at 2;1, and investigate the acquisition of scrambling. Before entering into a discussion of this aspect of acquisition, it will be useful to briefly sketch the adult system of Dutch as relevant in this respect. We continue the analysis which we started in (6), limiting ourselves to embedded order so as to be able to abstract away from the Verb Second effect of main clauses. As we stated earlier, we consider niet to be an adverb, adjoined to the VP (but cf. note 5). If nothing more is said, this would lead to the expectation that VP-constituents would occur in between the negation and the finite verb (see (23)): (23)

C [ IP NP

[ V P niet[ V P

ti] INFL + Vi

Yet, this expectation is not borne out. As is well known, the word order within the clause in Dutch and German allows quite some freedom. The example in (24) illustrates that the direct object and locative PP may occur in two different orders. (24) a. b.

dat Jan in Amsterdam z'n vriendin ontmoette. that John in Amsterdam his girlfriend met dat Jan z'n vriendin in Amsterdam ontmoette.

In Hoekstra (1984) it was proposed that (24b) is derived from the order in (24a) by the rule of scrambling, moving the DP z'n vriendin to the left of the PP in Amsterdam, adjoining it to the VP. This implies that the moved DP now is in a non-argument or Ä-position. Bennis & Hoekstra (1984) support this analysis by showing that scrambling licenses parasitic gaps. It was furthermore assumed that the rule of scrambling is optional, as is in accordance with the alternation between (24a) and (24b). However, the situation is much more complex than this rather sketchy description would suggest. First of all, scrambling gives rise to a kind of antiweak crossover effect. To illustrate, while zijn in (25a) cannot be bound by the DP iedere student, such binding is allowed in (25b). This is not at all what one would expect if scrambling were movement to an Ā-position. Rather, such effects are normally found with A-movement, as in passives and raising. (25) a. b.

dat Jan op zijn kamer iedere student bezocht. that John in his room every student visited dat Jan iedere student op zijn kamer bezocht.

Vanden Wyngaerd (1989) indeed proposes to analyze scrambling as involving movement to an A-position, more particularly to the specifier of a functional

136

TEUN HOEKSTRA & PETER JORDENS

projection OBJ-agreement, which is in between INFL and VP (cf. also Mahajan 1990; Koopman & Sportiche 1991). This movement would also bring the object to a VP-external position, but, unlike the alleged rule of scrambling to an adjoined position, this A-movement would be obligatory for reasons of Case: the object must move to the specifier of OBJ-agreement in order to receive Case. (26) provides the basic structure of this analysis:5 (26) C [IP DP [OBJ-AGRP-AGRPIi DPi OBJ-AGR [VP .... ti tj] INFL + Vj]] Such an approach to scrambling not only would capture the anti-weak-crossover effect, it would in principle also make scrambling obligatory in its effect, i.e., in order to satisfy the Case requirement on the object DP. It is precisely here that the situation becomes rather complicated. First of all, as the alternation in (24) shows, not every variation in word order can be accounted for in terms of an obligatory movement of the object DP. Secondly, and more important for the present discussion is the interaction of scrambling with negation. Consider the examples in (27): (27) a. b. c.

dat Jan that John dat Jan that John dat Jan that John

het boek niet leest. the book not reads geen boek leest. no book reads een boek niet leest. a book not reads

Although there are other possibilities if focus is also taken into account, (27) shows the basic distribution: definite objects DPs occur to the left of niet, unspecific indefinite DPs trigger the occurrence of the fused negation geen, while specific indefinites also precede niet. Geen in the adult system occurs in combination with count nouns, both singular and plural, as well as with mass terms. The OBJ-agreement approach may account for this if we assume that unspecific indefinite DPs may lack Case (e.g., because they meet the Case filter by being assigned partitive Case, as proposed by Belletti 1988), while other DPs have to move to OBJ-agreement, in order to receive Case. If niet is indeed adjoined to VP, the order object-niet-verb is explained. We shall not discuss these issues any further (cf. den Dikken & Mulder 1991 for some recent discussion), but now turn to the relevant stage of acquisition, where we investigate the interaction of negation and scrambling.

FROM ADJUNCT TO HEAD

137

3.2. The Development of Scrambling At the age of 2;1, the only negator is niet: nee has disappeared, niet itself is separated from the finite verb in second position in main clauses by intervening subjects, both pronouns and lexical NPs. The system of negation differs from the adult system in two respects: (i) (ii)

The fused negator geen is not available and it remains absent until 2;9. We shall not discuss the acquisition of geen here. The position of the object DP vis-à-vis the negative element niet is incorrect at this point, at least as far as definite DPs are concerned, as they follow rather than precede niet, as in the adult system.

In conclusion, then, at 2; 1 the interaction of semantic/syntactic properties of the object DP and negation are not yet acquired. Scrambling of DPs is acquired gradually. Pronouns are the first to scramble, followed by proper names. These are definite inherently, i.e., their definiteness is not a function of a determiner element. Typically, determiners are very frequently absent at this stage. To the extent that they do occur, their occurrence is mainly restricted to DPs in the complement of prepositions. As soon as determiners are acquired the DPs they are part of behave as expected. Some illustrative examples of transitive constructions at 2;1 are given in (28), while (29) illustrates the system at 2;8: (28) a.

b.

c.

(29) a.

b.

ik mag niet modewijzer I may not fashion.designer 'I may not have {the/a} fashion designer.' ik kan niet Maria zoeken I can not Mary search 'I can't find Mary.' ik wil niet dit I want not this 'I don't want this.'

2;1

dat jij niet lolly heb that you not lollipop have '... that you don't have a lollipop.' ik vin dat jongetje niet lief I find that boy not sweet 'I don't find that boy sweet.'

2;8

2;1

2;1

2;8

138

TEUN HOEKSTRA & PETER JORDENS

(29) c.

ik kan dat niet I can that not 'I can't do that.'

2;8

In (29b) the full DP object is correctly scrambled to the left of niet, as is the pronominal object in (29c). There are no mistakes in the positioning of pronominal objects, or proper name objects, nor are there any mistakes in the position of DPs with a definite determiner. There still is one type of mistake, from an adult perspective, which is displayed by (29a). Instead of niet we should have the fused negator geen. The reason for the delay in the acquisition of scrambling seems to be the delay in the acquisition of the determiner system. In itself this delay is something of a mystery. MacNamara (1977) has demonstrated that very young children, before the age of two, are sensitive to the distinction between definite and indefinite determiners. So, there is no problem of comprehension or perception involved. A perception based answer would certainly be rather difficult to maintain in the case of the late acquisition of geen, which should perceptually be clear enough. Yet, not a single occurrence of geen is found in the data before 2;9. It would seem, therefore, that a syntactic explanation is called for. Several venues towards an explanation are open at this point, but we limit ourselves to just mentioning a number of these, as it is not relevant to our point. If scrambling is indeed Case-driven movement to an A-position, two alternatives come to mind. The first is to say that A-movement is not yet available, as it is subject to maturation, as argued by Borer & Wexler (1987). However, we think that this explanation is difficult to maintain in the light of the VP-internal subject hypothesis (cf. Koopman & Sportiche 1991, among many others), according to which the subject is base-generated inside VP and moved to the specifier of IP by Case-driven A-movement. Alternatively, the absence of such Case-driven movement might be due to the absence of a Case requirement at the relevant stage. There are again several interpretations of why there is no such Case requirement. One interpretation would be to say that certain modules of UG are subject to maturation, as suggested by Lebeaux (1988): in this case, one would have to argue that the Case module has not yet matured. An alternative would be to say that DPs are subject to the Case requirement: if determiners are not yet available, nominal arguments may fail the Case filter because there are not DPs, but e.g., NPs, which are not subject to a Case requirement. This line of expla­ nation would fit in best with our finding that scrambling always takes place if an undisputable determiner is present. It raises questions with respect to the status of proper names and pronouns, however, which should not be DPs in the period in which they do not scramble. We return to this issue at the end of the next section.

FROM ADJUNCT TO HEAD

139

3.3. The Phenomenon of Subscrambling The previous sections have set the stage for the phenomenon that is relevant to the general point of this paper, viz. the development from adjunct structure to head-complement structure. We have noted that during the development of scrambling, there are quite a few of apparent violations of the so-called left branch condition. Relevant examples are given in (30). (30) a.

b.

c.

d.

e.

f.

g.

h.

die heb ik niet sok aan that have I not sock on 'I am not wearing that sock.' jou niet pyama you not pyjamas That is not your pyjamas.' Cynthia is dat niet pyama Cynthia is that not pyjamas 'That is not Cynthia's pyjamas.' zien Cynthia mag mij niet navel Cynthia may me not bellybutton see 'Cynthia may not see my bellybutton.' jij ziet mij niet osen you see me not eyes 'You don't see my eyes.' ik vin Cynthia niet tekening leuk I find Cynthia not drawing nice 'I don't like Cynthia's picture.' ik kan' die niet vis vange I can that not fish catch 'I cannot catch that fish.' Cynthia mas mij niet pyama zien Cynthia may me not pyjamas see 'Cynthia may not see my pyjamas.'

2;3

2;5

2;5

2;7

2;7

2;8

2;8

2;8

In all these examples, the underlined parts would form a single constituent in the adult grammar. From the point of view of argument structure, it would seem reasonable that the underlined parts also form a constituent at some level of representation in the grammar of the child. The splitting results from topicalization in (30a)-(30c), and from scrambling in the other examples. We shall refer to this splitting as SUBSCRAMBLING. Under a DP-analysis of nominal arguments (Szabolcsi 1983; Abney 1987), some of the examples involve extraction of the specifier ((30c) and (30f)), others

140

TEUN HOEKSTRA & PETER JORDENS

involve extraction of the head of DP ((30a) and (30g)), while the other cases are not clear, as they depend on the analysis of these pronouns, either as personal pronouns in Spec-D or as possessive pronouns in head position. Clearly, in adult grammar neither movement of the head of DP or of its specifier, while stranding the NP-part of the DP, is allowed. The relevant generalization to make, here, is that all the leftward scrambled parts can be independent nominal arguments. So, next to (30e), jij mag mij niet zien would also be a well-formed sentence, not only in adult Dutch, but also with respect to the apparent control of the rule of scrambling on the part of the child, i.e., a pronominal object like mij would be scrambled to the left of niet. The same holds for the other examples. There are no instances in which a determiner which cannot occur by itself, such as de 'the' or een 'a', is subscrambled. So, in terms of surface syntactic distribution, the object may be split up in two parts if each of the two parts can be used as an independent nominal argument, and each part also occupies the position expected for its type. So, the pattern is as in (31) where NP2 is a bare common noun, and NP1 either is a pronoun or a proper name: (31)

... NP1 niet NP2 V

The question we now face is what the grammatical analysis of such a string should be. One way to think of the relationship between NP1 and NP2 would be to draw an analogy with datives of possession of the type discussed by Guéron (1984). Under such an approach, NP1 and NP2 do not constitute a single constituent at any level of analysis. However, apart from the fact that these constructions do not involve inalienable possession, this might be a possible analysis only for cases where NP1 is a proper name or a personal pronoun, but not for those cases where NP1 is a demonstrative. A second possibility might be to assume that the common noun NP2 is not an NP, but rather part of a compound verb. So, (30g) might be analyzed as involving the compound verb vis vangen taking die as its object. Yet, this also does not provide a unified analysis, as it is inapplicable in most other cases. We therefore claim that NP1 and NP2 do form a single constituent, in accordance with the interpretative fact that they jointly represent a single argument. The surface discontinuity of this argument then results from movement of NP1, i.e., subscrambling. The possibility of such subscrambling is not at all expected from the point of view of UG, if these nominal arguments had the structure of DP. If that were the case, these subscramblings either involve extraction of the specifier or of the head. Let us look at the specifier cases first. The relevant structure, if we assume DP-status, is given in (32): the proper name or pronoun occupies the specifier position of the DP. Extracting the specifier used to be excluded by an appeal to the left branch condition or some similar

FROM ADJUNCT TO HEAD

141

constraint. From a more recent theoretical perspective (e.g., Chomsky 1986), however, it is not immediately clear why the specifier of a complement DP could not be extracted: it does not yield an ECP violation, nor a subjacency violation. Nor would it be possible to claim that specifiers in general cannot be extracted, given that subjects are allowed to move from the Spec-IP position, which, under the DP-analysis in (32), is completely parallel (cf. Corver 1990 for discussion). It is important to observe, however, that the genitive marker, which would occupy the head-position of DP, is generally lacking at the relevant stage. This is similar to what has been widely reported for English genitival -s (cf. Bloom 1970; Brown 1973, and many others). So, instead of papa's horloge 'daddy's watch', the child says papa loozje. Similarly, the child uses personal pronouns instead of possessive pronouns. If (32) were the right structure, then, the D-head position is empty.

Corver (1990) makes a specific proposal to explain the unavailability of possessor extraction in terms of the ECP. He also discusses the fact that some languages, notably Polish, do allow possessor extraction. He explains this by arguing that nominal arguments in Polish lack the DP-structure, and therefore allow possessor extraction. Let us turn to the subscrambling of demonstratives next. Again assuming the structure in (32), with the demonstrative in head position, this subscrambling would be an instance of head movement, and as such in principle not be allowed by structure preservation, unless the position of NP1 in (31) is some kind of head position, for which there is no indication, nor would it allow for a unified analysis. In conclusion, then, if the structure of nominal phrases in this stage were identical to the adult model, the subscramblings illustrated in (30) would be excluded by UG principles, basically the same principles that would supposedly lead the child to analyze the demonstrative as the head of a DP. We therefore would like to follow up on Corver's proposal that in Polish and other languages which allow possessor extraction, DP is lacking, and suggest an alternative to the structure in (32) for the child, viz. the structure in (33), with the proper names and demonstratives in adjoined, rather than in head position.6

142

TEUN HOEKSTRA & PETER JORDENS

Again we face the question as to what induces the change from (33) to (32). Clearly, this is the question about the acquisition of determiners, including the genitive -s morpheme. The scenario may be similar to Rizzi's (this volume) hypothesis about the optional infinitive stage. The basic tenet of that hypothesis is that while a particular functional category may be part of the child's grammar at a particular stage, its presence may be optional, even when its presence is obligatory in the adult system. In the case at hand, while DP may be available, it is not present in the absence of determiners. This interpretation is consistent with the gradual development of scrambling that we saw in Section 3.2. The relevant generalization was that during the period from 2;1 to 2;8 NPs with overt D-elements are properly scrambled. The rule of scrambling may then be said to apply to DPs, but not to NPs, the grammar at that stage allowing both as instantiations of a nominal argument. If that line of reasoning is chosen, the cases of subscrambling should be analyzed as in (34), rather than as in (33):

Note that in (34), a DP is adjoined to an NP. Its adjoined status allows its extraction, while its DP status forces its scrambling. The acquisitional step then involves recognition of the obligatory nature of DP.

4.

Two Speculative Other Examples

In this section, we discuss two other examples which we take to illustrate the same point. The first concerns subextraction from adjectival phrases, while the second concerns the development of embedded clauses.

FROM ADJUNCT TO HEAD

143

4.1. Degree Extractions The discussion in this section is necessarily somewhat brief and sketchy. First of all, the case to be discussed is not documented in the data of Jasmijn, but relevant examples were observed by various people independently. An experiment has been conducted to get a better grasp of the phenomenon (cf. Hoekstra, Koster & Roeper 1992). The results clearly support the impressionistic data. Consider the example in (35): (35)

kijk eens hoe ik groot ben look how I big am 'See how big I am.'

Again, the phrase hoe groot has been split in a way which is not allowed in the adult grammar. If the structure of the phrase hoe groot were as in (36), it would also violate the left-branch constraint, or whatever explains the effects of this condition.

We may again follow Corver (1990), who proposes that the structure of such adjectival phrases as 'how big' are not simple APs, as in (36a), but rather that they should be analyzed as projections of the DEG(ree), a functional category in his view, taking AP as its complement. The structure of the phrase hoe groot under this proposal would be as in (37), rather than as in (36a), and the lack of movement in the adult grammar can then be seen as a consequence of the absence of a suitable landing site for movement of the head of the DEG-phrase.

However, if in the child's grammar the structure is as in (36b), with hoe adjoined to the AP rather than in the specifier position as in (36a), the extraction illustrated in (35) is no longer problematic. We would then have yet a further instance of the acquisition schema proposed in this paper.

144

TEUN HOEKSTRA & PETER JORDENS

4.2. Embedded Clauses As a final example, consider the acquisition of embedded clauses. The earliest embeddings are found at 2;1, but lexical complementizers as well as wh-phrases are lacking at this point (cf. Rothweiler 1990, who reports the same for German; cf. also Weissenborn 1990): (38) a.

b.

ik weet niet koekmonster is I know not Cookie.Monster is 'I don't know where Cookie Monster is.' vin je niet leuk is find you not funny is 'Don't you think that that is funny?'

wh-expressions in embedded questions are found before overt complementizers show up. The first complete embedded clause with a complementizer is found at 2;6, given in (39): (39)

Cynthia zegt dat ik mee mag Cynthia says that I along may 'Cynthia says that I can come as well.'

Before that, quite a few cases occur of embedded questions with an overt wh-element. The wh-complementizer of makes its first appearance in 2;9, eight months after the first embedded question. It is perhaps worth noting, although not directly relevant to this point, that all instances of embedded clauses feature finite verbs (i.e., the optionality of infinitives does not seem to extend to embedded clauses), and that these finite verbs consistently occur clause-finally. We might interpret this as evidence that these embeddings are initially IPs, to which the wh-expressions are initially adjoined,7 rather than generated as projections of the functional category COMP. After the acquisition of comple­ mentizers, this situation is reanalyzed, and wh-expressions occur in Spec-CP. This development would then also instantiate the proposed schema. There is one piece of additional evidence in favor of this tentative proposal. Even though the nominative forms of subject pronouns are already available, the earliest examples of embeddings with pronominal subjects feature accusative rather than nominative pronouns, as in the examples in (40): (40) a.

b.

weet ik niet jou is know I not you:ACC is 'I didn't know where you were.' dan weet Papa nietwaar mij is then knows daddy not where me is

FROM ADJUNCT TO HEAD

145

Under an analysis of the embedding as IPs instead of CPs, there would be no barrier to protect the embedded subject from government by the matrix verb, which may hence assign accusative case. As the example in (39) shows, the pronominal subject is in the nominative form if a lexical complementizer is available. Whether this is a true generalization has to be left to future research.

5.

Concluding Remarks

We are aware that a number of questions have remained unanswered in this paper. Our proposal, though consistent with a view on acquisition as growth of the system, meets a learning theoretic problem if the notion of functional category is available throughout. We do not at this point want to base any further conclusions on this state of affairs. Also, the proposals we have made will have to be supported by further evidence, both in terms of similar phenomena in other languages, and in terms of a larger number of Dutch subjects.

Acknowledgements This paper was written during our stay at NIAS (the Netherlands Institute for Advanced Studies in the Humanities and Social Sciences), where we participated in a project on language acquisition. We thank the other members of this group (Toni Borowsky, Harry van der Hulst, Alan James, Charlotte Koster, Ann Mills, David Lebeaux, Rita Manzini, Tom Roeper, Herman Wekker) for discussions we had. Some of the ideas in this paper are shared by or inspired by ideas which other participants in our group have discussed with us. Notes 1.

The D-elements the and -s differ with respect to their specifier: while -s requires a nominal phrase, such as John, in its specifier, this is impossible for the and other D-elements. It is outside the scope of the present paper to go into the question of how to account for such differences.

2.

The maxim in (4) is indeed intended as a general claim. It is conceivable that there is no variation with respect to a particular functional category. As an example, consider AGR, which is currently assumed to be an functional category. As far as we know, there are no languages in which agreement is instantiated by elements which could belong to the adjunct category, and it seems even quite unlikely that any such language exists. This is not incompatible with the claim in (4).

146

TEUN HOEKSTRA & PETER JORDENS

3.

Hence we abstract away from more recent discussions in which INFL is split up in various functional categories, including Tense and Agreement (Pollock 1989), or subject-agreement and object agreement (cf. Chomsky 1991), as these involve issues which are not relevant to the problems under discussion.

4.

Again, we suppress several questions of detail which are not relevant to our discussion. For more extensive discussion of negation in West-Flemish, which might extend to standard Dutch, cf. Haegeman & Zanuttini (1991).

5.

The relative position of OBJ-AGR vis-à-vis the VP is irrelevant at this point. Also, the verb will move through this position on its way to INFL, a step we have not represented in this structure.

6.

A different alternative would be to generate these elements in the Spec-NP position. We take it that this option is excluded by UG, on the assumption that phrases generated internal to a lexical projection must be in a theta-relationship to the head. This is not in line with Radford (1990a, ch.4), who suggests that possessors do occupy the SpecNP, although Radford (1990b) makes the suggestion (cf. his footnote 3) that "possessor nominais are adjuncts to some projection of N" (1990b: 203).

7.

De Villiers (1991) argues that English children initially adjoin wh-phrases to the left of matrix IPs, by which she accounts for the lack of inversion at the relevant stage. She furthermore claims that placing wh-phrases in Spec-CP in matrix clauses is developed once embedded wh-expressions turn up. This doesn't exclude the possibility that English children too initially adjoin wh-phrases to IP in embedded questions, then put them in Spec-CP in embedded clauses, and finally do the same in matrix clauses, now corectly yielding inversion.

References Abney, Stephen. 1987. "The English Noun Phrase in its Sentential Aspect." Diss., Massachusetts Institute of Technology, Cambridge. Bennis, Hans & Teun Hoekstra. 1984. "Gaps and Parasitic Gaps." The Linguistic Review 4.29-89. Belletti, Adriana. 1988. "The Case of Unaccusatives." Linguistic Inquiry 19.1-34. Bellugi, Ursula. 1967. "The Acquisition of Negation." Diss., Harvard University, Cambridge, Mass. den Besten, Hans. 1983. "On the Interaction of Root Transformations and Lexical Deletive Rules." On the Formal Syntax of the Westgermania: Papers from the "3rd Groningen Grammar Talks", Groningen, January 1881. (=Linguistik Aktuell, 3), ed. by Werner Abraham, 47-131. Amsterdam: John Benjamins. Bloom, Lois. 1970. Language Development: Form andfunction in emerging grammars. Cambridge, Mass.: MIT Press. Borer, Hagit & Kenneth Wexler. 1987. "The Maturation of Syntax." Parameter Setting, ed. by Thomas Roeper & Edwin Williams, 123-172. Dordrecht: Reidel.

FROM ADJUNCT TO HEAD

147

Brown, Roger. 1973. A First Language: The early stages. Cambridge, Mass.: Harvard University Press. Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press. . 1991. "Some Notes on Economy of Derivation and Representation." Principles and Parameters in Comparative Grammar, ed. by Robert Freidin, 417-454. Cambridge, Mass.: MIT Press. Clahsen, Harald. 1988. "Critical Phases of Grammar Development: A study of the acquisition of negation in children and adults." Language Development, ed. by Peter Jordens & Josien Lalleman, 123-148. Dordrecht: Foris. . 1991. "Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language." Language Acquisition 1.361-391. Corver, Norbert. 1990. "The Syntax of Left Branch Extractions." Diss., Catholic University of Brabant, Tilburg. Deprez, Vivian & Amy Pierce. This volume. "Crosslinguistic Evidence for Functional Projections in Early Child Grammar." den Dikken, Marcel & René Mulder. 1991. "Double Object Scrambling." Papers from the Third Student Conference in Linguistics, 1991 (= MIT Working Papers in Linguistics, 10.), ed. by Jonathan D. Bobaljik & Tony Bures, 67-82. Cambridge, Mass.: Department of Linguistics and Philosophy, Massachusetts Institute of Technology. Evers, Arnold. 1975. "The Transformational Cycle in Dutch and German." Diss., Utrecht University. Distributed by the Indiana University Linguistics Club, Bloomington. Guilfoyle, Eithne & Maire Noonan. 1988. "Functional Categories and Language Acquisition." Paper presented at the 13th Annual Boston University Conference on Language Acquisition, Boston, 13 October. Guéron, Jacqueline. 1984. "Inalienable Possession, PRO-inclusion and Lexical Chains." Grammatical Representation, ed. by Jacqueline Guéron, Hans-Georg Obenhauer & Jean-Yves Pollock, 43-86. Dordrecht: Foris. Haegeman, Liliane & Raffaella Zanuttini. 1991. "Negative Heads and the NEG Criterion." The Linguistic Review 8.233-251. Hoekstra, Teun 1984. Transitivity: Grammatical relations in Government-Binding theory. Dordrecht: Foris. Hoekstra, Teun, Charlotte Koster & Thomas Roeper. 1992. "Left Branch Violations in Child Grammar." Paper presented at the 17th Annual Boston University Conference on Language Development, Boston, October. Hyams, Nina. This volume. "V2, Null Arguments and COMP Projections." Jordens, Peter. 1990. "The Acquisition of Verb Placement in Dutch and German." Linguistics 28.1407-1448.

148

TEUN HOEKSTRA & PETER JORDENS

Klima, Ed & Ursula Bellugi. 1966. "Syntactic Regularities in the Speech of Children." Psycholinguistic Papers, ed. by John Lyons & Roger Wales, 183-207. Edinburgh: Edinburgh University Press. Koopman, Hilda & Dominique Sportiche. 1991. "The Position of Subjects." Lingua 85.211-258. Lebeaux, David 1988. "Language Acquisition and the Form of the Grammar." Diss., University of Massachusetts, Amherst. Lightbown, Patsy. 1977. "Consistency and Variation in the Acquisition of French." Diss., Columbia University. Lundin, Barbro & Christer Platzack. 1989. "The Acquisition of Verb Inflection, Verb Second and Subordinate Clauses in Swedish. Working Papers in Scandinavian Syntax 42.43-55. Lund: Linguistics Department, Lund University. MacNamara, John. 1977. Language Learning and Thought. New York, Academic Press. Mahajan, Anoop. 1990. "The A/A-bar Distinction and Movement Theory." Diss., Massachusetts Institute of Technology, Cambridge. Meisel, Jürgen & Natascha Müller. 1992. "Finiteness and Verb Placement in Early Child Grammars: Evidence from simultaneous acquisition of two first languages: French and German." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 109-138. Dordrecht: Kluwer. Penner, Zvi. 1990. "The Acquisition of Syntax in Bernese Swiss German: The role of functional elements in restructuring of early grammars." Paper presented at the 15th Annual Boston University Conference on Language development. Boston, 19-21 October. Ms., University of Berne, 1991. Pierce, Amy. 1989. "On the Emergence of Syntax: A crosslinguistic study." Diss., Massachusetts Institute of Technology, Cambridge. Plunkett, Kim & Sven Strömqvist. 1990. "The Acquisition of Scandinavian Lan­ guages." Gothenburg Papers in Theoretical Linguistics. Pollock, Jean-Yves. 1989. "Verb Movement, Universal Grammar and the Structure of IP." Linguistic Inquiry 20.365-424. Poeppel, David & Kenneth Wexler. 1993. "The Full Competence Hypothesis of Clause Structure in Early German." Language 69.1-33. Radford, Andrew. 1990a. Syntactic Theory and the Acquisition of English Syntax. Oxford: Basil Blackwell. . 1990b. "The Syntax of Nominal Arguments in Early Child English. Lan­ guage Acquisition 1.195-223. Rizzi, Luigi. This volume. "Early Null Subjects and Root Null Subjects." Roeper, Thomas. 1990. "How a Marked Parameter is Chosen: Adverbs and do-insertion in the IP of child grammar." Ms., University of Massachusetts, Amherst. Rothweiler, Monika. 1990. "Nebensatzerwerb im Deutschen: Eine empirische Untersuchung zur Primärspracherwerb." Diss., University of Tübingen.

FROM ADJUNCT TO HEAD

149

Szabolcsi, Anna. 1983. "The Possessor that Run Away from Home." The Linguistic Review 3.89-102. Verrips, Maaike & Jürgen Weissenborn. 1992. "Routes to Verb Placement in Early German and French: The independence of finiteness and agreement." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 283-331. Dordrecht: Kluwer. de Villiers, Jill. 1991. " 'Why' Questions." Papers on the Acquisition of WH (=University of Massachusetts Occasional Papers, Special Edition.), ed. by T. L. Maxfield & Bernadette Plunkett. Distributed by the Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachusetts, Amherst. Weissenborn, Jürgen. 1990. "Functional Categories and Verb Movement in Early German: The acquisition of German syntax reconsidered." Spracherwerb und Grammatik: Linguistische Untersuchungenzum Erwerb von Syntax und Morphologie (= Linguistische berichte, Special Issue 3), ed. by Monika Rothweiler, 190-224. Opladen: Westdeutscher Verlag. Weissenborn, Jürgen, Maaike Verrips & Ruth Berman. 1989. "Negation as a Window to the Structure of Early Child Language." Ms., Max Planck Institute, Nijmegen. Wexler, Kenneth. 1991. "Optional Infinitives, Head Movement and the Economy of Derivations in Child Grammar." Ms., Massachusetts Institute of Technology, Cambridge. Whitman, John, Kwee-Ock Lee & Barbara Lust. 1990. "Continuity of Universal Grammar in First Language Acquisition: The issue of functional categories." Paper presented at the Annual Meeting of the North Eastern Linguistic Society, Montreal, October. Wode, Henning. 1977. "Four Early Stages in the Development of L1 Negation." Journal of Child Language 4.87-102. vanden Wyngaerd, Guido. 1989. "Object Shift as an A-movement Rule. Papers from the First Student Conference in Linguistics (= MIT Working Papers in Linguistics, 11). Cambridge, Mass.: Department of Linguistics and Philosophy, Massachusetts Institute of Technology. Zanuttini, Raffaella. 1990. "Two Types of Negative Markers." Proceedings of NELS 20, 19[89], ed. by Juli Carter, Rose-Marie Déchaine, Bill Philip & Tim Sherer, 517-530. Amherst, Mass.: Generative Linguistics Student Organization, Department of Linguistics, University of Massachusetts, Amherst.

Early Null Subjects and Root Null Subjects

Luigi Rizzi Université

1.

de Genève/SISSA,

Trieste

Introduction

Around the age of two, children freely drop subjects, irrespective of whether or not the target language is a null subject language. For instance, the phenome­ non is typically found in the acquisition of English and French: (1)

(2)

want more find Giorgie is broken

(Hyams 1986)

boit café fait un autre est tombé

(Pierce 1989)

Much recent work on the acquisition of English provides robust evidence for a selective drop of subjects: by and large, learners freely drop subjects, not obligatory objects (Hyams & Wexler 1993; Valian 1991; Wang, Lillo-Martin, Best & Levitt 1992). So, the EARLY NULL SUBJECT is not trivially amenable to some global strategy of structural reduction. Hyams (1986) interpreted this state of affairs in terms of the NULL SUBJECT PARAMETER (Rizzi 1982; 1986; Jaeggli & Safir 1989): the initial setting is the null subject value, learners of English, French, etc. have to reset the parameter on the basis of experience, and this is normally done a few months after the second birthday. I would like to retain of Hyams' seminal approach the idea that the early null subject stage manifests a genuine grammatical option and cannot be reduced to extragrammatical factors (see Hyams & Wexler 1993 for a detailed discussion of this issue). On the other hand, a number of structural properties of

152

LUIGI RIZZI

the early null subjects are emerging which suggest that this phenomenon is quite different from the drop of subjects in an adult grammatical system like Italian. In the first part of this paper I would like to discuss some such properties and identify the major configurational constraint: by and large, the early null subject is possible in the first position of the structure, i.e., in the specifier of the root. I will then claim that such a configurational constraint is not specific to transitional systems in acquisition: instances of root null subjects can be found in adult grammatical systems, hence they represent a genuine option of Universal Grammar. Various questions are raised by the existence of root null subjects in acquisition, in special registers, and in normal adult languages: What is the status of the null element involved? How is it licensed and identified? How does it differ from a discourse-bound null operator (Huang 1984)? Why is this option lost in the course of the acquisition of English and many other languages? I will try to provide a partial answer to these questions by developing an analysis of ROOT NULL SUBJECTS based on the typology of null elements proposed in Lasnik & Stowell (1991).

2.

Some Structural Properties of Early Null Subjects

Valian (1991: 39) points out that null subjects occur very rarely after a preposed wh-element in her corpus (natural production from 21 learners of English ranging from 1;10 to 2;8): only 9 null subjects out of 552 wh-questions in which the wh-element is not a subject; i.e., (3b) does not occur as a regular variant of (3a): (3)

a. b.

Where daddy go? Where go?

This is a significant and surprising finding, especially in view of the fact that a null subject in this environment is perfectly acceptable in a null subject language (e.g., in Italian: Dove va? 'Where (he) goes?', Cosa fai? 'What (you) do?', etc.). As Valian's observation is quite isolated in the acquisition literature, I tried to find confirmation elsewhere. A preliminary check of the standard Brown (1973) corpora confirms the observed tendency (thanks to Rick Kazman for technical help). In the whole of Eve's corpus (20 recordings from 1;6 to 2;4) I counted 12 null subjects out of 191 wh-questions with a non subject wh-element. Roeper (1991), addressing the issue, lists a number of examples with null

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

153

subjects following a wh-element from the Adam corpus. Such cases indeed exist, but they are quite limited: if we look at the first 10 files of the Adam corpus, in which null subjects in declaratives are well over 50 percent (57 percent according to Bloom 1990, see also Hyams & Wexler 1993), we find 21 cases of null subjects out of 158 questions with non-subject wh-elements. The proportion of null subjects in this environment thus drops to a percentage (13 percent) close to the one of null objects in the same corpus (8 percent according to Bloom 1990).1 Another relevant element is provided by the apparent existence of whquestions without movement in early English. The Adam corpus presents quite a few cases of WH IN SITU:

(4)

They are for who? It's a what? He may do what to me?

(Adam 25) (Adam 26) (Adam 33)

The majority of these cases (22 out of 34) involve a null subject; here is a sample: (5)

see what bear? and do what? use dat for what? have what? doing what? cutting what? close what? fighting what? sing what song?

(Adam (Adam (Adam (Adam (Adam (Adam (Adam (Adam (Adam

03) 25) 25) 31) 35) 35) 36) 36) 38)

The existence of a consistent in situ strategy in early English is, of course, far from being established by this small sample of examples (but see Radford 1990; Whitman, Lee & Lust 1991 for supporting evidence and discussion); if confirmed, it would raise various interesting questions, as wh in situ is not a property of the target language (apart from echo-questions): why does it arise? how and when is it delearned? Let me put these questions side for the moment.2 Let me simply notice the high proportion of null subjects with in situ questions, which suggests that early null subjects are not affected by the status of the sentence as a question, but by the preposing of the wh-element. Another crucial property of the early null subject is that it is limited to main clauses. Roeper & Weissenborn (1990) point out that no cases of null subjects are produced by the learner of English in the first finite subordinate clauses. This

154

LUIGI RIZZI

observation is confirmed by Valian (1991), who found no case of null subject out of the 123 examples of tensed subordinate clauses of her corpus. One could observe that this state of affairs may not be very significant for the analysis of the structural properties of early null subjects: after all it could be that the child starts producing finite subordinate clauses after the end of the null subject stage. But this interpretation is strongly disfavored by the fact that we also find occasional cases in which a pronominal subject is dropped in the main clause and not in the embedded clause in the same utterance, or anyhow in immediate succession, e.g.: (6)

a.

went in the basement # that what we do # after supper (Eve 19) know what I maked (Adam 31 )

b.

Again, this observation invites a systematic verification; if confirmed on a larger scale, it strongly supports the hypothesis of a structural incompatibility between early null subjects and embedding, rather than a simple succession of acquisition stages.3 As in the previous case, this is quite different from what we find in a null subject language of the Italian type, in which a zero subject pronoun is equally possible in main and embedded subject position:4 (7) (I)

so che cosa know what (you)

hai detto said

In conclusion, the early null subjects produced by learners of English appear to obey a strong distributional constraint: their natural environment is the first position of the structure, they tend not to appear after a preposed element and they do not appear in embedded clauses, two properties which are quite different from what we find in adult null subject languages. What happens in the acquisition of Italian? Does the Italian learner around the age of 2 produce null subjects with the structural properties of early English? Or does she already conform to the properties of adult Italian? Even small production samples strongly suggest an early convergence to the properties of adult Italian: we typically find examples like the following, with a null subject following a preposed wh-element (examples taken from the Martina corpus, see Cipriani, Chilosi, Bottari & Pfanner 1992, available on CHILDES, MacWhinney & Snow 1985 — thanks to Teresa Guasti for technical help). (8)

a.

Ov'è? 'Where is?'

(1;8)

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS (8)

b. Cos' è? 'What is?' c. Che vol? 'What (you) want?' d. Pecché piangi? 'Why (you) cry?' e. Quetto cosa fa? 'This what does?'

155 (1;10) (2;3) (2;3) (2;5)

In the Martina corpus (13 recordings roughly each month from 1;7 to 2;7) we find that, out of 35 questions with a non-subject wh-element, 20 have a null pronominal subject. Even though the data base is limited, the indication is quite clear: early null subjects in the acquisition of Italian are not restricted to the first position; in this respect, the early system is just like adult Italian. The emerging picture thus seems to support the hypothesis of an early fixation of the null subject parameter: around the age of two, learners of English and Italian have already converged to the values of the parameter expressed by the adult languages. The early null subject manifested in the acquisition of English is a different phenomenon, structurally characterized by the fact that it is limited to the initial position, the specifier of the root.5 The next question we want to ask is whether such a root null subject is a special property of transitional systems in acquisition, or whether it is actually found in some adult grammatical systems, in which case it would correspond to a genuine option of Universal Grammar.

3.

Subject Drop in Diaries

The closest analogue to the observed properties of the early null subject is found in certain abbreviated varieties of English and other languages. Haegeman (1990) notices the following cluster of properties in the register of diaries: i.

(9)

Subjects can be freely dropped even if the standard register of the language does not allow this, e.g., in English and French; dropped subjects are not necessarily 1st person, as the French example shows: A very sensible day yesterday. saw noone. took the bus to Southwark Bridge. walked along Thames Street... (Virginia Woolf, Diary. Vol. 5, 1936-1941, pp. 203-204)

156

LUIGI RIZZI (10)

m'accompagne au Mercure, puis à la gare ... '(He) takes me to Mercure, then to the station ...' s'est donné souvent l'illusion de l'amour ... '(He) often gave himself the illusion of love ...' me demande si ...je lui eus montré les notes ... '(I) ask myself if ... I would have shown him the notes ...' (Paul Léautaud, Le Fléau, Journal Particulier, 1917-1930, pp. 60-70) Notice that the second and third line of the French example also show that a structurally represented null subject must be postulated, otherwise the anaphoric clitic would not be bound.

ii.

The subject cannot be dropped after a preposed element:

(11) a. b. iii.

was so stupid! *Ho w stupid was!

Main subjects can be dropped, embedded subjects cannot:

(12) a.

can't find the letter that I need.

b. *I can't find the letter that

need.

(see also the third line of (10)). iv.

Subjects can be dropped, objects cannot:

(13) a. saw her at the party. b. *She saw at the party. So, on the basis of Haegeman's (1990) description, subject drop in diaries appears to be ruled by the same structural constraints which characterize early null subjects. Haegeman notices that the root character of subject drop suggests a topic-drop type of analysis, involving a discourse bound null operator in the matrix spec of COMP binding a variable in subject position. Under such an analysis, the null operator would be in competition with an overt preposed operator, whence the ungrammaticality of (11b), etc. But she also notices that under a topic drop analysis the subject-object asymmetry (13) is unexpected: why couldn't the null operator bind a variable in object position? Of course, the same difficulty arises for an unqualified topic drop analysis of the early null subject. On the other hand, the structural conditions on root null subjects and topic drop seem close enough to invite a detailed comparison.

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS 4.

157

Topic Drop

In colloquial German (as well as in most V2 Germanic varieties), it is possible to drop a main clause subject in the specifier of COMP in a V2 configuration (14a). The option disappears in clause-internal position, i.e., when the specifier of COMP is filled by a preposed element, as in (14b), and in embedded clauses, no matter whether V2 (14c) has applied or not (14d): (14) a. b. c. d.

(Ich) habe es gestern gekauft. '(I)have it yesterday bought.' Wann hat *(er) angerufen? 'When has he telephoned?' Hans glaubt *(ich) habe es gestern gekauft. 'Hans believes I have it yesterday bought.' Hans glaubt daß *(ich) es gestern gekauft habe. 'Hans believes that I it yesterday bought have.'

We thus seem to find the same structural restrictions operative on the early null subject. Still, the dropping of arguments extends to preposed objects in colloquial German: (15)

(Das) habe ich gestern gekauft. 'This have I yesterday bought.'

This apparent subject-object symmetry lead researchers to analyze the construc­ tion as involving topic drop (Ross 1982), or, in current terms, movement of a discourse-bound null operator to the specifier of the root CP (Huang 1984), from where it could bind a variable in subject or object position. Under this analysis, (14a) and (15) would have the following parallel representations: (16) a. b.

[OP habe [t es gestern gekauft]] [OP habe [ich t gestern gekauft]]

If this analysis was correct, the analogy with our early null subjects would be partial at best. But Cardinaletti (1990) has pointed out that there remains an important asymmetry between subject and object drop: subject drop can involve pronouns with any person specification, provided that the dropped element is sufficiently salient in the context, whereas object drop is restricted to 3rd person. For instance, a second person object pronoun cannot be dropped even in the most favorable case of contextual saliency, question-answer pairs: (17) a.

Hast du mich gesehen ? 'Have you me seen?'

158

LUIGI RIZZI (17) b. c.

Dich habe ich nicht gesehen. 'You have I not seen' * habe ich nicht gesehen. '(you) have I not seen.'

If in general operators are intrinsically marked for 3rd person (as is the case for interrogative operators, see below) the limitations on the object case follow from the structural analysis (16b); but then, Cardinaletti concludes, subject drop should not involve a null operator, and representation (16a) should be revised. If (14) and (15) are to be dissociated as in Cardinaletti's proposal, then it becomes more plausible to partially assimilate the former case to the early English system.6 If the root null subject does not involve a discourse bound null operator, we are then left with the questions: What is its status with respect to the typology of null elements? How can we express its minimal differences with respect to the null operator constructions? None of the currently assumed empty categories seems to have the right intersection of formal and interpretive properties.

5.

Null Constants

A recent proposal by Lasnik & Stowell (1991), offers a new option which is worth exploring. In the context of a general discussion of the scope of WEAK CROSSOVER effects, they observe that certain null operator constructions (and also some A-dependencies involving overt operators, such as appositive relatives) differ significantly at the interpretive level from ordinary operator variable constructions, e.g., questions: (18) a. b.

John wonders who to please t John is easy OP to please t

The former involves quantification ranging over a possibly non-singleton set, in the latter the null element never ranges over a non-singleton set, rather, it has its reference fixed to that of the antecedent. This interpretive difference correlates to the sensitivity to the weak crossover effect: while both kinds of Ä-binding manifest sensitivity to STRONG CROSSOVER: (19) a. *Whoi did you get himi to talk to ti ? b. *Johni is easy for us OPi to get himi to talk to ti ? Only the former manifests weak crossover effects:

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

159

(20) a. *Whoi did you get hisi mother to talk to ti ? b. Johni is easy for us OPi to get hisi mother to talk to ti ? Lasnik & Stowell introduce a split between the two types of Ā-bound traces in accordance with the semantic intuition. Only the trace bound by a genuine quantifier is a variable; the trace bound by the non-quantificational empty operator is not, is a non-variable R-expression to be assimilated to a null epithet or, more generally, to a null definite description; in order to properly distinguish it from the variable, we will call this new type of null element THE NULL CONSTANT (nc, terminological suggestion due to Cornelia Hamann). Lasnik & Stowell then claim that weak crossover is a property of variables, hence null operator constructions, not involving variables, are exempted from it. On the other hand, both kinds of A-bound traces are R-expressions: as Principle C is a property of all R-expressions, the homogeneous behavior with respect to strong crossover is accounted for. How is the new type of empty category to be characterized? As a straightforward alternative to Lasnik & Stowell's functional definition, let me simply propose that the feature <±v>(ariable) combines with the familiar features <±a>, <±p>, thus giving rise to 8 cases: (21)

1.

2.

3.

4.

<+a> <+p>=* <+v>

<+a> <+p>=PRO <-v>

<+a> <-p>=* <+v>

<+a> <-p>=NP-t <-v>

5.

6.

7.

8.

<-a> <-p>=vbl <+v>

<-a> <-p>=nc <-v>

<-a> <+p>=pro(res) <+p>=pro <+v> <-v>

Of these, 1 and 3 are presumably excluded by the inherent incompatibility of <+a> (requiring A-binding) and <+v> (requiring A-binding): a single element cannot simultaneously belong to the A- and Ä-system.7 The remaining 6 combinations are all attested. 2, 4, 6 and 7 are the familiar types; 5 is pro used as a resumptive pronoun (e.g., as in Rizzi 1982, chapt. 2; Georgopoulos 1991); and 8 is the null constant, a non-variable R-expression. We are now left with the question: What forces Ä-binding of the null constant by a null operator? That is, as overt epithets and other definite descriptions can freely occur and directly pick up their referent in discourse, why can the null variant not do the same?

160

LUIGI RIZZI (22)

I tried to visit John last week, but I was unable to persuade the {guy/*ec} to see me.

Lasnik & Stowell's answer is: The null definite description, as all null elements, must satisfy an identification requirement which is fulfilled by the null operator. We can make this suggestion precise by assuming that the specific identification requirement on null constants is the same one that holds for variables and other types of traces, i.e., the identification component of the EMPTY CATEGORY PRINCIPLE (ECP) (in the sense of Rizzi 1990), which holds for all nonpronominal empty categories: (23)

ECP (Identification) Empty categories <-p> must be chain-connected to an antecedent

where the antecedent can be an A- or A- (or, irrelevantly here, X°-)position, depending on the kind of chain.8 In sum, an empty category <-a, -p> can be <±v>; if it is <+v> it is a variable and must satisfy (23) by being chain connected to a genuine quantifier (e.g., a question operator), assigning it a range; if it is <-v> it is a null constant, it must still be chain connected to an Ā-element to satisfy (23) (it could not be A-bound, as it is an R-expression), but by a non-quantificational Ā-element, typically a null operator. The other combinations (a variable bound by a non-quantificational operator, a null constant bound by a quantificational operator) are excluded by the appropriate version of the BIJECTION PRINCIPLE, the principle barring vacuous quantification, or whatever principle requires quantifiers to bind variables and variables to be bound by quantifiers (Koopman & Sportiche 1982; Chomsky 1986). Moreover, as the null constant is an R-expression, chainconnection to an element in A-position is barred by Principle C. A non-quantifi­ cational operator thus remains as the only possible identifier of the null constant.9

6.

Discourse Identified Null Elements Revisited

Lasnik & Stowell's proposal directly refers to cases of sentence-bound null operator constructions (easy-to-please, parasitic gaps, etc.), but can be immediately extended to the empty elements bound by discourse identified null operators. In fact, if we apply Lasnik & Stowell's diagnostic criterion for null constants, the empty category bound by a discourse-identified null operator in German falls into this class: it is sensitive to strong crossover, but not to weak crossover: *Den Hansi hat eri ti gesehen. 'Hans has he seen.' b. *OPi hat eri ti gesehen.

(24) a.

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

161

(25) a.

Den Hansi hat [seini Vater] ii gesehen. 'Hans has his father sehen.' b. OPi hat [seini Vater] ti gesehen.

We can now go back to Cardinaletti's (1990) asymmetry between the local subject case and all the other cases. In all the other cases, e.g., the object case, the null constant (nc) is bound by a null OP, which as such has intrinsic features of 3rd person singular. (26)

[OP hat [er schon nc gesehen]] (ihn, es, *mich, *dich, ...)

What about the local subject case? If the representation of (27) was (28), it would be hard to distinguish the two cases: (27) (28)

habe es schon gesehen. [OP habe [nc es schon gesehen]]

One could not, for instance, simply claim that (27) can have a 1st person interpretation because the inflection endows OP with such features: if I had this capacity, what could exclude the possibility of non 3rd person question operators in the appropriate contexts?10 (29)

*Chi (di vol) sapete la risposta? who (among you) know+3PL the answer

But there is a possible alternative representation for (27). It has been repeatedly noticed that the Spec-CP position in V2 languages can behave as an A-position when the local subject is moved to it (with the trace in Spec-IP behaving like an NP-trace, cf. Holmberg 1986; Taraldsen 1986; Rizzi 1991b). Suppose then that (27) allows a representation with the null constant in the Spec-CP, binding an NP-trace in the Spec-IP, and involving no null OP at all: (30)

[nc habe [t es schon gesehen]]

As the intrinsic limitation to 3rd person is specific to operators, there is no reason to expect it in (30). Still (30) should violate the identification requirement of the ECP, as the null constant lacks a clause-internal identifier. Can (30) be made consistent with the ECP? A natural possibility is offered by an extention of an idea proposed by Chomsky (1986) in the context of the theory of binding. According to Chomsky's proposal, the governing category of an element is the minimal domain with certain characteristics in which the binding requirements of the element are

162

LUIGI RIZZI

satisfiable in principle. Suppose that we extend and adapt this idea to the ECP by adding the following specification to (23): (23)

ECP (Identification) Empty categories <-p> must be chain-connected to an antecedent

(31)

... if they can

So, an empty element must be identified in the way indicated by (23) if it can, i.e, if there is a potential identifier, a c-commanding maximal projection (possibly, a c-commanding Xo for an empty head). This has the effect of exempting from the identification requirement the specifier of the root, the highest position of the structure, the position that c-commands everything and is not c-commanded by anything. The specifier of the root then is the only position in which an empty element can fail to have a clause internal identification, and is available for discourse identification. Under this interpretation of (23), an unbound null constant can survive in the specifier of the root in structures like (30), and receive its referential value in discourse.11

7.

Cross-linguistic Variation and Developmental Sequence

It is now natural to extend the proposed null constant analysis to all the observed cases of root null subjects, including the early null subject. But why is the option lost in adult standard English and French, for instance? The basic idea that I would like to develop to deal with both the crosslinguistic variation and the observed developmental sequence is that the possibility of a root null subject can arise when the specifier of the root is an A-position, and is lost in the varieties in which the specifier of the root is an A-position, not a suitable host for the null constant. Underlying this approach is the idea that the feature system <±a>, <±p>, (and <±v>) defines empty elements in A-position: null elements belonging to the Ä-system are not classified by this feature system, so the null constant (on a par with the variable, PRO, the NP-trace, etc.) simply cannot be defined in an Ä-position. Let us start from a natural assumption on the nature of the root category. Following Stowell (1981), Radford (1988) and many others, I will assume that the following principle holds: (32)

Root = CP

This principle amounts to saying that we normally speak through propositions, not fragments of propositions: in the unmarked case the root category is the

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

163

canonical structural realization of the proposition, the CP. Let us first consider the case of the Spec-CP in a V2 language such as German. I shall adopt the following definition of A-POSITIONS: (33)

A-positions: theta positions and specifiers construed with AGR.

(See Rizzi 1991b for discussion of the different cases.) Under this definition, the Spec-CP in a V2 language is an A-position when the local subject is moved into it, as in (34):

The subject is construed with the AGR-specification of the highest inflectional head; if the latter is moved to COMP and the local subject is moved to the Spec-CP in V2, the Spec-AGR configuration is reconstituted at the CP level, hence the Spec-CP is an A-position under (33). The Spec-CP can therefore host our null constant and structure (30) is well formed. If the language also possesses the discourse-linked null operator, as is the case in colloquial German, the null constant will also be possible in other structural positions, e.g., the object as in (26), provided that it is bound by the discourse identified null operator in the main Spec-CP, with the interpretive properties that we have discussed in the previous section. On the other hand, in English, French and other non-V2 languages the structure of the root is:

If the language possesses the discourse-identified null operator, a null constant in Spec-IP (or in any other position) will be possible, being bound by the null operator in the spec of C (this may be the case of European Portuguese, according to Raposo 1986). If the language does not have the discourse-identified null operator (as is the case in English or French) the null constant is not permissible in this configuration: the Spec-CP is not a possible host because it

164

LUIGI RIZZI

is an A-position, the Spec of I is not a possible host because it is not the Spec of the root (but see Section 9), there is a higher position which can be a potential antecedent, the Spec-CP, hence an unbound null constant in the Spec of IP is ruled out under (23). Why is the root null subject option available in early English? Given the picture I have proposed, the most natural element to hold responsible for the developmental sequence is principle (32). Suppose that this principle is not operative initially, i.e., the encircled part can be omitted in (35) — note that it does not have to be omitted: in fact, much recent work on the early manifestation of V2, e.g., Boser, Lust, Santelmann & Whitman (1991), Poeppel & Wexler (1991) strongly suggests that the possibility of starting from CP is available from very early on — if the CP layer is omitted, then the Spec-IP, an A-specifier of the root, becomes a suitable host for the null constant, and the root null subject is allowed. No other position is available for the null constant: if the CP layer is present in (35), for instance in a question with wh-movement, the Spec-IP ceases to be the specifier of the root and an unbound null constant in this position is excluded by (23); similarly an unbound null constant in object position (or any other position) is excluded in, for example, (36)

*I

met nc

because there are higher positions acting as potential identifiers — the subject position in (36) — hence (23) is enforced. As early English (like adult English) does not possess the discourse-identified null operator, the null constant is not permissible in any position other than the main subject position (when the CP layer is omitted), whence the observed subject-object asymmetry. As soon as principle (32) becomes operative (perhaps an event triggered by an inner maturational schedule, in the sense of Borer & Wexler 1987) the conditions for the root null constant cease to be met, and the early null subject disappears. The principle in (32) may remain a weak principle, tough, susceptible of being 'turned off' on abbreviated registers and, perhaps, under special contextual conditions (possibly question-answer pairs, etc.).

8.

Speculations on Developmental Correlations

If principle (32) is not operative initially, then root categories different from CPs are adequate starting points for early linguistic expressions. We have shown how the possibility of root null subjects follows from this property: if the CP

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

165

layer can be omitted, then the structural conditions for the null constant in subject position are met. Do we expect other properties to correlate with root null subjects under this analysis? An immediate consequence of our proposal is that, if CP is not the compulsory starting point in early grammars, we would expect children to use a much wider variety of root categories, i.e., simple NPs, PPs, APs, (non-finite) VPs, different kinds of uninflected small clauses, etc. Arguing for such a scenario goes beyond the limits of the present paper; we will just note in passing that the very high proportion of nominal or otherwise non-verbal utterances in children's production around the second birthday, as well as its sharp decline few months later, is what the proposed approach leads us to expect. For instance, the proportion of utterances containing verbs in the production corpus of Valian's (1991) first group of English learners (age range l;10-2;2) is only 0.27, a proportion that raises dramatically over 0.70 in her third and fourth group (age ranges 2;3-2;6 and 2;6-2;8, respectively; the author has excluded imperative sentences from this calculation). In our terms, as soon as (32) becomes operative, the option of using (non-verbal) fragments of CPs as complete utterances ceases to be generally available, and is confined to whatever special discourse contexts allow it in the adult grammar. See also Radford's (1990) comprehensive discussion of the early stages, particularly in connection with the early use of root small clauses.12 A less obvious property that may be related to early null subjects is the apparent in situ stage that was hinted at in the first section. Remember what the problem is: If learners of English go through a stage in which both syntactic wh-movement and wh in situ are possible, how does she 'delearn' the in situ option, thus moving from a superset system to a subset system? Following a suggestion due to Michal Starke, we could think that the real option that the child has is not between choosing syntactic or LF wh-movement, but rather between choosing CP or a different category as the root: if CP is taken, then syntactic movement is obligatory as in the adult English grammar; if a different category is selected, then wh-movement cannot take place, there being no appropriate host, so wh in situ is the only option (presumably with QR applying at LF to the wh-element as a salvaging strategy to create the necessary operator-variable structure). Under this interpretation, when principle (32) becomes operative in English, then the (parametrized) principle requiring obligatory movement in this language applies in full force, and wh in situ is ruled out.13 Should one expect other co-occurrence relations between the root null subject and other properties of the early grammar? One could speculate that the

166

LUIGI RIZZI

delay in the operativity of (32) is related to the general parsimony of functional elements that the child's initial production manifests. Much recent work has shown that it is too radical to assume that the child's linguistic representations are purely lexical around the age of 2: functional heads must be postulated in the child's grammar to account both for morphological analysis, the very presence of functional elements, and various word order phenomena (for instance, see Deprez & Pierce 1990 on French, Poeppel & Wexler 1991, and much other work on the acquisition of V2 in German, Guasti 1992 on Italian, etc.). Still, a residue of the idea that initial syntactic representations are purely lexical (Guilfoyle & Noonan 1991; Lebeaux 1988; Platzack 1990; Radford 1990) seems to remain valid: the child's production around the age of 2 manifests a liberty of omission of functional heads that is not found in the target languages (see also Kazman 1990), e.g.: i.

Determinerless NPs appear to freely alternate with full DPs in the early stages of languages in which the determiner is obligatory, e.g., French (examples from Pierce 1989; Friedemann 1992 and Radford 1992, who relate this property to the freer distribution of nominais in the early stages):

(37) a. b. ii.

The use of main clause participial sentences involves (at least) the omission of the functional head auxiliary (Bottari, Cipriani & Chilosi 1991):

(38) a.

b.

iii.

Pas pousser chaisse papa. not push chair papa Fini café Madeleine. finished coffee Madeleine

Fatto Diana done Diana 'Diana has done it.' Che fatto la bimba? what done the little girl 'What has the little girl done?'

Diana(l;ll)

Diana(2;0)

The use of main clause infinitives found in many languages may involve the omission of tense (and/or of AGR), see Friedemann (1992), Guasti (1992), Wexler (1991) and references cited therein for recent discussion (examples from Pierce 1989):

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

(39) a. b.

167

Monsieur conduire. man (to) drive Tracteur casser maison. tractor break house

Some fundamental questions remain open in these cases, in particular the question of whether the (partial) lack of phonetically realized functional elements in production is to be expressed as blocking of spell-out for otherwise present categories, or it manifests the radical absence of the category. It is conceivable that different answers will be appropriate for different kinds of missing functional elements. If at least some cases of missing phonetic realization are to be analyzed as involving radical absence of the category, then the non-operativity of (32) could be viewed as a special case of the overall parsimony of functional elements in the early grammar.14

9.

Root Expletive Subjects

As is well known, learners of English drop non-referential subjects as well as referential subjects: (40)

yes,

is toys in there

(from Hyams 1986)

Can the null constant analysis be extended to cover this case? Notice first of all that some adult languages clearly extend the root null subject option to non-referential elements, e.g., Swedish (examples due to Christer Platzack): (41) a. b.

(Det) verkar som om ... it seems as if ... (Det) telefonerades mycket igaar. it was.telephoned a.lot yesterday

These appear to be genuine cases of root null subjects in that the zero variant of the expletive is only possible in initial position: when the Spec-CP is filled by a different element in a V2 construction, the expletive cannot be dropped: (42)

Igaar telefonerades *(det) mycket Yesterday was.telephoned it a.lot

So, UG apparently allows the null constant to be an expletive. If the unmarked case for an expletive is to be null, then it is natural that the child will take the null constant option for the expletive.15 Some natural languages allow root null subjects only with the non-referen­ tial interpretation. This is the case of colloquial French, and may be true of some

168

LUIGI RIZZI

colloquial registers of English, even though the latter are less easy to tease apart from other registers in which also referential subjects can be dropped: (43) a. b. (44) a. b.

{semble/paraît/s'avère} que Marie est malade. {seems/appear s/turns} out that Mary is sick. * *

{dit/sait/pense} que Marie est malade. {says/knows/thinks} that Mary is sick.

We find the root null subject properties again: if a wh-element is preposed, or if the structure is embedded, the expletive cannot be dropped. (45) a. b.

Pourquoi semble *(t-il) que ... Why does *(it) seem that...

(46) a. b.

Jean dit que *(il) semble que ... John says that *(it) seems that...

Why is it that colloquial French and English retain the option of root null subjects but restricted to certain types of expletives (here we gloss over the fact that not all types of expletives can be naturally dropped in the relevant context)? The hypothesis that these systems also possess a structural device to suspend the application of (32) (say, on a par with the abbreviated registers) would require an independent parametrization of the referential interpretation (diary French would take the option of discourse-identified referential interpretation, colloquial French would not), not a very appealing possibility for many reasons. A somewhat more interesting possibility is the following. Suppose that specifiers are optional in general, unless required by some special principle such as the EXTENDED PROJECTION PRINCIPLE (EPP). Suppose then that the specifier of the root COMP can be missing, i.e., that (35) can have the following shape:

If this is correct, then an unbound null constant in the Spec-IP becomes possible again under (23), as there is no c-commanding maximal projection which may act as its antecedent. So, the non-referential nc is possible here, while it remains excluded from embedded contexts such as (46), or main contexts in which the

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

169

Spec-CP is present, such as (45). But this now leaves us without our structural explanation for the impossibility of the referential nc. We can observe that, over and above the constraints discussed so far, there are strong restrictions on discourse identification, illustrated by the fact that a discourse-linked null operator in German must be in the root Spec-CP: if it sits in an embedded Spec-CP (even in a V2 structure), the sentence is excluded: (48) (49)

OP habe ich t schon gesehen. (this) have I already seen *Hans glaubt OP habe ich t schon gesehen. Hans believes that (this) have I already seen

It appears that discourse identification of an empty category is restricted to the root, cannot look more deeply inside the structure apart from the immediate domain of the root. An intuitively plausible way to express this constraint is to generalize the idea we borrowed from Chomsky's binding theory and state it also as a con­ straint on discourse identification (while keeping it as a relaxing condition on principles of sentence grammar such as the ECP): (50)

A null element can be discourse-identified only if it is not c-commanded sentence-internally by a potential identifier.

So, the null constant in Spec-IP in (47) is formally licit (if no Spec-CP is selected, (23) is vacuous), but it cannot be discourse-identified because it is c-commanded by COMP, a potential identifier (typically, heads function as identifiers at least for one type of null element, pro). As is always the case for formally licit but unidentified null elements (cf. much current literature on pro), this instance of the null constant can only function as a non-argumental expletive.

Acknowledgements Versions of this paper at different stages of elaboration were presented at the GLOW Workshop The Development ofMovement and Inflection (Leiden, March 28,1991), at the symposium Syntactic Theory and First Language Acquisition: Cross Linguistic Perspectives (Cornell, April 25, 1992), at the Certificat de spécialisation of the University of Geneva and at the Language Acquisition Seminar of SISSA, Trieste. I am grateful to the audiences of these events and to Adriana Belletti, Guglielmo Cinque, Marc-Ariel Friedemann, Teresa Guasti, Liliane Haegeman and an anonymous reviewer for helpful comments and suggestions.

170

LUIGI RIZZI Notes

1.

2.

3.

4.

5.

We leave the question open of determining the nature of the latter kind of omission. See also Hamann (1992), who found cases of non-root (middle-field) null subjects in special contexts at a later stage of the acquisition of German. Nevertheless, the general trend in the acquisition of V2 languages seems to involve null subjects located in the specifier of the root (see, in particular, de Haan & Tuijnman (1988) on Dutch; on early null subjects in German see Clahsen (1991), Meisel (1990), Weissenborn (1992) and references cited there). It remains to be determined whether French early null subjects manifest similar occurrence restrictions (see Pierce 1989; Friedemann 1992, for relevant discussion; preliminary results from Crisma 1992 seem to strongly support an extension of our analysis to early French). The delearning problem arises because the child apparently shifts from a system allowing both syntactic movement and in situ (like, say, French in main clauses) to a system requiring syntactic movement, apparently a move from a superset system to a subset system. Of course, the problem arises only if a substantive number of the in situ examples are genuine questions. If they turned out to be analyzable as echo questions in the general case, there would be no delearning problem (even though a different question would presumably arise: Why are there so many echo questions in early English?). We will go back to this issue in the concluding remarks. It also rules out for this case the possibility that the child may be interchangeably using two grammars for some time (one with null subjects and no subordination and the other with subordination and no null subjects) a possibility suggested in a somewhat different context by Anthony Kroch. If anything, a zero pronominal subject is even favored in embedded environments under coreference with the main subject, and compulsory when the embedded clause is adverbial: in the following: (i) Gianni canta quando (lui) è contento. Gianni sings when (he) is happy the zero pronoun must be selected to express coreference with the main subject. Another observation of Valian's (op. cit.) is potentially relevant in this context. She notes that early null subjects in Italian are about twice as frequent as early null subjects in English, given comparable age groups. Even if relative frequencies, as such, do not immediately bear on the hypothesis of a structural difference (no direct frequency predictions are made by a structural hypothesis), if anything, the observed difference goes in the direction expected under a structural analysis of the sort advocated in the text: the null subject of Italian is possible in a larger set of structural environments, hence, all other things being equal, we would expect it to be more frequent than the early null subject in English. Do Italian learners possess the root null subject option, on top of the early positive setting of the null subject parameter? I know of no empirical reason to exclude or confirm this possibility. On the other hand, if the suggestion of note 15 is correct, the availability of a pro subject may block the root null subject option in early Italian. I will leave this question open.

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

171

6.

Another reason for dissociating the dropping of local subjects in the specifier of the root from cases of topic drop comes from Swedish: in this language, the dropping of the local subject can involve an expletive (i), an element which cannot be topicalized, e.g., from the subject position of an embedded clause (ii), whether or not it is dropped. Genuine cases of topic drop are restricted to referential elements, which can be topicalized (iii): (i) a. (Det) verkar som om ... (it) seems as if... b. (Det) telefonerades mycket igaar. (it) was.telephoned a.lot yesterday (ii) a. *(Det) vet jag [ t verkar som om ...] it know I seems as if b. *(Det) visste jag [ t telefonerades ... ] it knew I was.telephoned (iii) (Det) visste jag [ t skulle haenda ] it knew I should happen Cases of root null subjects in adult languages are occasionally reported in the literature (see Kenstowicz 1989:264 on Levantine Arabic and Somali, Solà 1992 on Corsican). The question of whether such cases are amenable to an analysis along the lines proposed in the text will be left open here.

7.

This conclusion is also supported by the fact that languages using resumptive elements (overt variables) always use pronouns, never anaphors: the feature <+v> can mix with <+p>, not with <+a>. Guglielmo Cinque and Robin Clark have pointed out that PROs giving rise to so-called PRO gate effects, i.e., alleviating weak crossover violations in examples like Who did PRO visiting his relatives annoy [t] ? (Higginbotham 1980), may be analyzable as cases of <+v> PRO. If this is correct, then the incompatibility of <+v> with <+a> should be restricted to the pure anaphor, i.e., the case of <+a> which is assigned a governing category.

8.

See Rizzi (1990); in the terms of that system, the chain-connection can be established via antecedent government or via binding, depending on the availability of referential indices. Guglielmo Cinque suggests that our null constant may be assimilated to a variable bearing a referential index in the sense of Rizzi (1990) as sharpened by Cinque (1990), as opposed to a variable unable to bear such an index, hence available for antecedent government connections only. If this reduction is correct, the <±v> feature simply is a descriptive label for an independently needed distinction (at least as far as <-a -p> null elements are concerned).

9.

A consequence of this system is worth noticing. It has been occasionally observed that, while the other three canonical types of empty categories can (must) be the heads of Achains, the purely anaphoric empty category <+a, -p>, can only be an A-trace, a non-head of chain; in other words, natural languages do not seem to allow null variants of overt anaphors, which typically head their own chains: (i)

John saw {himself/*ec}

172

LUIGI RIZZI (while typically allowing null R-expressions and null pronominals that head their A-chains, on a par with their overt counterparts). See, e.g., Brody (1985) for discussion. This gap follows from the proposed system, in fact from the interplay of the identification part of the ECP and the feature specification of the null elements: a <+a, -p> empty category, qua <-p> is in the scope of (23), hence it must be chain connected to an antecedent; qua <+a> it must be locally bound by an A-antecedent; since the A- and A-systems do not mix (see previous discussion), it follows that the empty category must be chain-connected to an A-antecedent, hence it never is the head of an A-chain, QED. This conclusion does not hold for other types of empty category: <+a, +p>, <-a, +p> are not in the scope of (23), hence chain-formation with a higher element is not enforced; <-a, -p> is in the scope of (23), must be chain-connected to an antecedent, but this antecedent can be an Ä-position, as the empty category is <-a> (in fact the antecedent that the empty category is chain-connected to must be an A-position, because of principle C).

10.

Overt and null operators can inherit features different from 3rd, but only from a binding antecedent, not from AGR: (i) Voi, OP che t sapete la risposta ... you who know+3PL the answer ... (ii) You are easy OP to please.

11.

(30) also violates the formal licensing part of the ECP, proper head government (Rizzi 1990). If, following Moro (1992), we restrict the proper head government requirement to the case in which a head intervenes between a null element and its identifier, the requirement will not apply to (30). Note that our interpretation of (23) also provides an explanation of the fact that embedded C°s generally are overt, while main C°s can (must) be null (assuming roots to be CPs in the general case, see below): the head of the root lacks a potential identifier (a c-commanding head), hence it can be left empty. Consider also the possibility of dropping semantically empty fillers of INFL {do, have, be) in questions, but not in declaratives (Schmerling 1973; Akmajian, Demers & Harnish 1984): (i) (are) you going to lunch? (ii) (have) you ever been to Chicago? (iii) (does) she like her new house? I-to-C movement raises the AUX to the root head, where it can be null under our interpretation of (23). (The possibility of dropping a second person pronoun in this environment may involve incorporation of the pronoun into the null auxiliary, and/or be akin to the null subject of imperatives.)

12.

Diaries and other special 'abbreviated' registers allowing root null subjects could now be characterized as retaining the non-operativity of (32). We would then expect, among other things, that on these registers discourse or textual units do not have to be full propositions, can be fragments of propositions. A quick perusal of Haegeman's corpus supports this hypothesis, as we typically find in diaries chunks of propositions corresponding to maximal projections different from CP, e.g. (i) Après-midi à discuter, puis agréable. afternoon to discuss, then pleasant (P. Léautaud, op. cit.)

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

13.

173

Notice that this approach requires that the principle in question, the wh-Criterion of Rizzi (1991a), should also be interpreted as 'obligatory if satisfiable in principle'. For the sake of the argument, I have assumed that (4)-(5) are genuine questions. If a generalized echo question analysis is tenable, then the delearning problem does not arise.

14. All other things being equal, we could then expect root null subjects to disappear concomitantly with the loss of main clause infinitives, etc., even though one cannot a priori exclude the possibility that different principles enforcing the presence of different types of functional elements do not emerge simultaneously, but follow a maturational schedule. 15.

Some languages allowing null constants (and null operators) do not seem to like nonreferential null constants, e.g., standard Dutch: (i) a. (Hij) praatte erover. he talked there-about b. *?(Het) regent. 'It is raining.' c. *(Er) werd lang gedanst. there was long danced This may be an irreducible property of the language, to be learned through indirect negative evidence, or may be related to the fact that Dutch, contrary to Swedish, allows some cases of expletive pro: it could be that natural languages do not like to have two different types of null elements functioning as expletives (Icelandic may be problematic for this conjecture, see Sigurðsson 1989). References

Akmajian, Adrian, Richard A. Demers & Robert M. Harnish. 1984. Linguistics. Cambridge, Mass.: MIT Press. Bloom, Paul. 1990. "Subjectless Sentences in Child Language." Linguistic Inquiry 21.491-504. Borer, Hagit & Kenneth Wexler. 1987. "The Maturation of Syntax." Parameter Setting, ed. by Thomas Roeper & Edwin Williams, 123-172. Dordrecht: Reidel. Boser, Katharina, Barbara Lust, Lynn Santelmann & John Whitman. 1991. "The Syntax of CP and V2 in Early German Child Grammar." Paper presented at the 22th Annual Meeting of the North Eastern Linguistic Society, University of Delaware. Bottari, P., P.Cipriani, A.M.Chilosi. 1991. "Pre-syntactic Devices in the Acquisition of Italian Free Morphology." Ms., Istituto Stella Maris, Pisa. Brody, Michael. 1985. "On the Complementary Distribution of Empty Categories." Linguistic Inquiry 16.505-564. Brown, Roger. 1973. A First Language: The early stages. Cambridge, Mass.: Harvard University Press.

174

LUIGI RIZZI

Cardinaletti, Anna. 1990. "Pronomi nulli e pleonastici nelle lingue germaniche e romanze." Diss., Venice University. Chomsky, Noam. 1986. Knowledge of Language. New York: Praeger. Cinque, Guglielmo. 1990. Types of Ā-Dependencies. Cambridge, Mass.: MIT Press. Cipriani, P., A. M. Chilosi, P. Bottari, L. Pfanner. 1992. L'acquisizione della morfosintassi in italiano: Fasi e processi. Padova: UniPress. Clahsen, Harald. 1991. "Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language." Language Acquisition 1.361-391. Crisma, P. 1992. "On the Acquisition of wh-Questions in French." Ms., Geneva University. Deprez, Viviane & Amy Pierce. 1990. "A Crosslinguistic Study of Negation in Early Syntactic Development." Paper presented at the 15th Annual Boston University Conference on Language Development, Boston, October. Ms., Rutgers University, New Brunswick and Massachusetts Institute of Technology, Cambridge. Friedemann, Marc-Ariel. 1992. "The Underlying Position of External Arguments in French: A study in adult and child grammar." Ms., Geneva University. Georgopoulos, Carol. 1991. Syntactic Variables. Dordrecht: Kluwer. Guasti, Maria-Teresa. 1992. "Verb Syntax in Italian Child Grammar." Ms., Geneva University. Guilfoyle, Eithne & Maire Noonan. 1991. "Functional Categories and Language Acquisition." Ms., Massachusetts Institute of Technology, Cambridge and McGill University, Montreal, de Haan, Ger & Kees Tuijnman. 1988. "Missing Subjects and Objects in Child Grammar." Language Development, ed. by Peter Jordens & Josien Lalleman, 101-121. Dordrecht: Foris. Hamann, Cornelia. 1992. "Late Empty Subjects in German Child Language." Technical Reports in Formal and Computational Linguistics 4, Geneva University. Haegeman, Liliane. 1990. "Understood Subjects in English Diaries." Multilingua 9.157-199. Higginbotham, James. 1980. "Pronouns and Bound Variables." Linguistic Inquiry 11.679-708. Holmberg, Anders. 1986. "Word Order and Syntactic Features." Diss., University of Stockholm. Huang, C.-T. James. 1984. "On the Distribution and Reference of Empty Pronouns." Linguistic Inquiry 15.531-574. Hyams, Nina. 1986. Language Acquisition and the Theory of Parameters, Dordrecht: Reidel. Hyams, Nina & Kenneth Wexler. 1993. "On the Grammatical Basis of Null Subjects in Child Language." Linguistic Inquiry 24.421-459.

EARLY NULL SUBJECTS AND ROOT NULL SUBJECTS

175

Jaeggli, Osvaldo & Kenneth Safir, eds. 1989. The Null Subject Parameter. Dordrecht: Kluwer. Kazman, Rick. 1990. "The Acquisition of Functional Categories and the Lexicon: A psychologically plausible model." Ms., Carnegie Mellon University, Pittsburg. Kenstowicz, Michael. 1989. "The Null Subject Parameter in Modern Arabic Dialects." The Null Subject Parameter, ed. by Osvaldo Jaeggli & Kenneth Safir, 263-275. Dordrecht: Kluwer. Koopman, Hilda & Dominique Sportiche. 1982. "Variables and the Bijection Principle." The Linguistic Review 2.139-161. Lasnik, Howard & Tim Stowell. 1991. "Weakest Crossover." Linguistic Inquiry 22.687-720. Lebeaux, David. 1988. "Language Acquisition and the Form of the Grammar." Diss., University of Massachusetts, Amherst. MacWhinney, Brian & Catherine Snow. 1985. "The Child Language Data Exchange System." Journal of Child Language 12.271-296. Meisel, Jürgen. 1990. "Infl-ection; Subjects and Subject-Verb Agreement in Early Child Language: Evidence from simultaneous acquisition of two first languages: German and French." Ms., University of Hamburg. Moro, Andrea. 1992. "Heads as Antecedents: A brief history of the ECP." Forthcoming in Lingua e stile. Bologna: Il Mulino. Pierce, Amy. 1989. "On the Emergence of Syntax: A crosslinguistic study." Diss., Massachusetts Institute of Technology, Cambridge. Platzack, Christer. 1990. "A Grammar without Functional Categories: A syntactic study of early Swedish child language." Working Papers in Scandinavian Syntax 45.13-34. Lund: Department of Scandinavian Linguistics, University of Lund. (Published in the Nordic Journal of Linguistics 13.107-126.) Poeppel, David & Kenneth Wexler. 1991. "The Status of Functional Categories in Early German Grammar." Ms., Massachusetts Institute of Technology, Cambridge. Radford, Andrew. 1988. Transformational Grammar. Cambridge: Cambridge University Press. . 1990. Syntactic Theory and the Acquisition of English Syntax. Oxford: Basil Blackwell. Raposo, Eduard. 1986. "The Null Object in European Portuguese." Studies in Romance Linguistics, ed. by Osvaldo Jaeggli & Carmen Silva-Corvalan, 373-390. Dordrecht: Foris. Rizzi, Luigi. 1982. Issues in Italian Syntax. Dordrecht: Foris. . 1986. "Null Objects in Italian and the Theory of pro." Linguistic Inquiry 17.501-557. . 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. . 1991a. "Residual Verb Second and the wh-Criterion." Technical Reports in Formal and Computational Linguistics 3, Geneva University.

176

LUIGI RIZZI

Rizzi, Luigi. 1991b. "Proper Head Government and the Definition of A-Positions." GLOW Newsletter 26.46-47. Roeper, Thomas. 1991. "Why a Theory of Triggers Supports the pro-drop Analysis." Ms., University of Massachusetts, Amherst. Roeper, Thomas & Jürgen Weissenborn. 1990. "How to Make Parameters Work." Language Processing and Language Acquisition, ed. by Lynn Frazier & Jill de Villiers, 147-162. Dordrecht: Kluwer. Ross, John Robert. 1982. "Pronoun Deleting Processes in German." Paper presented at the Annual Meeting of the Linguistic Society of America, San Diego, California. Schmerling, S. 1973. "Subjectless Sentences and the Notion of Surface Structure." Chicago Linguistic Society 9. Chicago: University of Chicago. SigurÖsson, Halldór Á. 1989. "Verbal Syntax and the Case in Icelandic." Diss., University of Lund. Solà, Joan. 1992. "Agreement and Subjects." Diss., Universitat Autònoma de Barcelona. Stowell, Tim. 1981. "Origins of Phrase Structure." Diss., Massachusetts Institute of Technology, Cambridge. Taraldsen, Knut Tarald. 1986. "Som and the Binding Theory." Topics in Scandinavian Syntax, ed. by Lars Hellan & Kirsti Koch-Christensen, 149-184. Dordrecht: Reidel. Valian, Virginia. 1991. "Syntactic Subjects in the Early Speech of American and Italian Children." Cognition 40.21-81. Wang, Qi, Diane Lillo-Martin, Catherine T.Best & Andrea Levitt. 1991. "Null Subject Versus Null Object: Some evidence from the acquisition of Chinese and English." Language Acquisition 2.221-254. Weissenborn, Jürgen. 1992. "Null Subjects in Early Grammars: Implications for parameter setting theories." Theoretical Issues in Language Acquisition: Continuity and change in development, ed. by Jürgen Weissenborn, Helen Goodluck & Thomas Roeper, 269-299. Hillsdale, N.J.: Lawrence Erlbaum. Wexler, Kenneth. 1991. "Optional Infinitives, Head Movement and the Economy of Derivation in Child Grammar." Paper presented at the Conference on Verb Movement, University of Maryland, College Park, 13 October. Ms., Masssachusetts Institute of Technology. Whitman, John., Kwee-Ock Lee & Barbara Lust. 1991. "Continuity of the Principles of Universal Grammar in First Language Acquisition: The issue of functional categories." Proceedings of the North Eastern Linguistic Society 21, ed. by Tim Sherer, 383-397. Amherst, Mass.: Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachusetts, Amherst.

Asking Questions without CPs? On the Acquisition of Root wh-Questions in Bernese Swiss German and Standard German Zvi Penner University of Berne

1.

Preliminaries

The observation that early grammars deviate from the target language is a trivial one. A close inspection of the data, however, reveals that the differences between the initial and final stages are by no means random. As has been shown in recent work on language acquisition, it seems that the central aspects of the IP-internal syntax are accessible to the child from early on. Early grammars are consistent with the target grammar with regard to: (1)

a. b. c. d. e.

VP directionality (Roeper 1992) Verb movement to INFL (Roeper 1991) Local scrambling (Weissenborn 1990; Penner, Tracy & Weissenborn 1992) Verb Projection Raising (Penner 1990a,b) No genuine violations of the pro-drop parameter (Weissenborn 1988, 1992a; Weverink 1990; Rizzi, this volume)

While the data concerning the internal structure of IP seem to be consistent with the target grammar, this is not the case with the COMP system. Confining ourselves to examples from German and French, the initial stage can be characterized as follows: (2)

a. b. c.

Overt complementizers are not attested in subordinate clauses (Clahsen 1991; Müller 1991, 1992; Penner 1992a) Object topicalization is absent (Weissenborn (1990) Clitic placement is inconsistent with the target grammar (Haverkort & Weissenborn 1991; Penner 1992a)

178

ZVI PENNER

(2)

d. e.

Subject (topic) drop displays the liberal 'diary' value (Penner 1992a; Rizzi, this volume) Absence of complex inversion in French (Müller 1991; Weissenborn 1992b)

This body of data has been taken as evidence for the assumption that early grammars lack a C-projection. Although the status of the COMP system at the initial stage is rather obscure, it seems that children correctly grasp the V2 effect in root/nonroot clauses very early. This observation is true both for High German and Bernese Swiss German. For a detailed discussion, cf. Rothweiler (1989), Weissenborn (1990), Müller (1991), Gawlitzek-Maiwald, Tracy & Fritzenschaft (1992), Penner (1992a), and Müller & Penner (1992). This is best seen from the fact that, at the stage preceding the acquisition of V2 and complementizers, children produce the so-called PRECONJUNCTIONAL (i.e., complementizer-less) EMBEDDED CLAUSES which regularly display the 'standard' verb-final pattern of subordinate clauses. A typical example would be the comp-less relative clause in (3) (Weissenborn 1990:204): (3)

rutschsocken Mami waschen hat socks mother washed has 'the socks mother washed'

H(2;l)

By placing the verb clause-finally, the child treats the embedded relative clause as if it were introduced by an overt conjunction, although there is no overt complementizer in this structure. Given these observations, the overall impression is that the 'knowledge' of the COMP system at the initial stage is in some sense incomplete. At first glance, it seems that the COMP system is available at the level of feature instantiation, but is inoperative at the level of spell out. In other words, the COMP projection is part of the grammatical representation, but neither COMP nor Spec-CP are allowed to surface. How can this asymmetry be captured in terms of parameter setting? Let us follow Roeper & Weissenborn (1990) in assuming that the core of the target parametric system is accessible to the child from early on by virtue of the SYNTACTIC BOOTSTRAPPING DEVICE, that is, the child sets the parameters of subject pro-drop, scrambling, VP- and IP-directionality, etc. on the basis of root/nonroot asymmetries. Suppose that the child is also aware of root/nonroot alternations of post- and preverbal negation and related matters at the initial stage (cf. Weissenborn, Verrips & Berman 1989). If this hypothesis is basically correct,

ACQUISITION OF ROOT

WH-QUESTIONS

179

we might contend that the child, due to the Syntactic Bootstrapping Device, will be capable of assessing that the input belongs to a V2 language as soon as the root/nonroot asymmetry with regard to NEG-placement is registered. This line of argumentation relies on the assumption that the learning procedure refers to the so-called Degree-1 input. Note that similar conclusions can be drawn if we adopt the Degree-0 procedure of David Lightfoot. In their (1992) paper, Evers & van Kampen argue for a bottom-up learning procedure by virtue of which the child may deduce the V2 value at the CP level on the basis of root information. Adopting Roeper's (1992) thesis that the CP > IP > VP hierarchy is given a priori, they show that the directionality parameter of the VP can be correctly triggered in periphrastic structures. They further assume a default rule which says that, if V is head-final the IP must be head-final as well. Since in root sentences the child is exposed to the finite verb usually precedes its complement, the child must draw the conclusion that the INFL node is occupied by a trace, and that the landing site of the finite verb is the higher head, namely CP. Let us assume that there is a procedure which guarantees that the child is capable of deducing <+V2> at the initial stage. Note, however, that this is a 'universal' decision. In fact, V2 is only a cover term for the multiple occurrence of INFL at the clause level, i.e., INFL is taken to be instantiated both clauseinternally on INFL and clause-externally on COMP. Let us refer to the former instance of INFL as INFLa and to the latter as INFLa'. The mechanism known as V2 unifies (or conflates) INFLa and INFLa' by moving the INFL carrier from INFL to COMP. The statement the child deduces V2 amounts to saying that she knows that INFLa/INFLa' conflation takes place. This piece of information is, however, only one prerequisite for the V2 mechanism. What remains beyond the detection capacity of both the Syntactic Bootstrapping Device and the Degree-0 procedure is the language-specific value of V2. As recent work on V2 has shown, the relationship between INFLa and INFLa' unfolds in an idiosyncratic and language-specific manner, depending, inter alia, on how tense and agreement are distributed across COMP and INFL and how these features are spelled out. So, for instance, the following options are manifested in the various types of V2 languages:1 (4)

Root V2 a. The finite verb displays two different endings, depending on whether it occurs in INFL or COMP.2 b. V2 languages may opt for the complex inversion pattern.

180

ZVI PENNER

(4)

c. d.

e. (5)

V2 languages may vary with regard to whether or not the SpecCP position in root declaratives must be realized. V2 languages may vary with regard to whether or not they license complementizers to be placed in root COMPs (cf. for instance, (45) below). V2 languages may vary with regard to whether or not move­ ment to COMP involves agreement.

Nonroot clauses a. COMP can be overtly inflected for agreement (as in, e.g., Bavarian or Frisian). b. V2 languages might have different complementizers, depending on whether or not operators occur in the Spec-CP position (e.g., the att/som-alternation in Swedish). c. The articulation of I-in-C might be multiple-headed (as in Scandinavian) or single-headed (as in High German). d. V2 languages might vary with respect to whether or not they license doubly-filled complementizers.

Given this variation, it is clear that the discovery of the V2 mechanism per se is still insufficient for assessing the language-specific mode of implementation of this rule. Recall that generalized V2 can be deduced on the basis of alternations which take place inside the IP. Conversely, the information concerning the language-specific content and the internal structure of COMP is encoded in the COMP system itself, hence cannot be inferred in this way. The mode of instantiation of INFLa' and especially of INFLa'/INFLa conflation bears a different relation to the input evidence than plain V2 and probably requires more extensive exposure to language. In particular, it seems that the child gains access into the language-specific mechanism of V2 on the basis of the knowledge of the idiosyncratic properties of inflectional paradigms. This kind of information seems not to be available prior to the age of about 2;6. I assume that the fact that the language-specific value of INFLa' cannot be detected at the initial stage narrows down the conditions under which C may be overtly spelled out.3 Based on these assumptions, we may now account for the 'V2 paradox' in the following way: At the initial stage, the feature <+V2> is accessible to the child in its underspecified or canonical form. The delayed integration of INFLa' into the operative system, however, is responsible for the child's incapability of implementing INFLa/INFLa' conflation in a language-specific way. This has as a consequence that the child can articulate her knowledge of V2 as long as

ACQUISITION OF ROOT

WH-QUESTIONS

181

INFLa' remains uninstantiated. The only context where this requirement is met is probably the preconjunctional embedded clause construction (cf. (3) above) where COMP is empty, and the root/nonroot distinction is expressed by the V-end pattern.4 Contrarily, the whole range of root V2 phenomena, which directly involve activation of INFLa', is ruled out in early grammar. This accounts for why genuine movement to COMP as well as related cliticization phenomena are absent at the initial stage. In the light of these facts, we might say that, at the initial stage, the child possesses partial knowledge of her native language. How does she navigate her way through the conflicts brought about by this kind of partial competence? The question of how to represent incomplete knowledge in intermediate grammars is a central issue of the research program in language acquisition. Selective knowledge can be captured in different ways. As has been proposed in recent work, syntactic knowledge can be grasped in terms of licensing conditions. Following Chomsky (1992), one might want to argue that grammar is in fact a system of licensing conditions which operate on Spec-head configurations. Some of these licensing conditions (e.g., the Case Filter) are IP-internal. The licensing conditions on operators, however, are generally taken to be associated with the COMP system. Returning to the problem of incomplete knowledge, the following question arises: how are the licensing conditions met at a stage where the formal devices for satisfying them are only partially available? So, for instance, what happens when the target grammar comprises a licensing condition LC which says that: Spec-XP is well-formed, iff it agrees with Xo with regard to a feature F.

(6)

and Xo, for some reason, is not allowed to be overtly spelled out? Let us assume the following scenario for an intermediate grammar G¡: (7)

a. b.

Due to indirect evidence the child is aware of LC. Since Xo cannot be spelled out, it is inoperative for the purpose o f LC.

There are three possible routes for the child to take in order to resolve this conflict: (i)

(ii)

The null hypothesis might be that the child 'freezes' all the syntactic operations and structures which presuppose the full availability of Xo. Let us refer to this route as the FREEZING STRATEGY. Another logical option would be to evade LC by resorting to a DEFAULT STRATEGY. Following Lebeaux (1988), Roeper & de Villiers (1991) and Roeper (1992), the notion of 'default' in the case of LC could be defined as:

182

ZVI PENNER (8)

(iii)

(9)

Default: Adjoin new material to the highest node possible. That is, whenever the application of LC exceeds the boundaries of the Spec-head configuration available to the current grammar, new material is automatically treated as an adjunct being attached to the highest node of the available structure. This is, of course, a universal operation which does not necessarily converge on the parametric system of the target grammar, The third option would be to circumvent the conflict between (7a) and (7b) by means of resorting to compensatory strategies which are consistent with the target grammar. The notion of 'consistent' will be defined below. We will refer to this option as the CONSISTENCY HYPOTHESIS. Consider the following scenario: a.

b. c.

Given (i) and (ii): (i) The Parametric Choice PC: A grammatical function GF (GF = interrogativity, sentential predication, subordination, etc.) is associated in the target gram­ mar Gt with the licensing condition LC. The licensing device LD in LC involves Spec-XP and Xo. (ii) In an intermediate grammar G¡, there is a procedure PR by virtue of which the child becomes aware of LC. However, Xo in its target form is still unavailable for LD. In order to express GF, the child resorts to PC' under the condition that PC' is LC-related to PC. PC' is LC-related to PC iff both refer to the same type of licensing configuration.

As for the notion of 'the same type of licensing configuration', I will assume that wh-operators can be licensed in two different types of structure: an adjunction configuration and a Spec-head configuration. In the latter case, the licensing device is standardly taken to be feature sharing. As for the adjunction configura­ tion, it is not entirely clear at the present stage of investigation whether wh-adjuncts are simply exempted from strict licensing in the sense of the Spec-head structure or whether they require a special specifier-adjunct relationship (cf. Chomsky 1992). The purpose of this paper is to examine the validity of these three hypotheses against the empirical data from the acquisition of root constituent questions in High German and Bernese Swiss German. The domain of root

ACQUISITION OF ROOT

WH-QUESTIONS

183

constituent questions has been chosen as an arena for the competing assumptions discussed here, since there are good reasons to believe that the child becomes aware of the licensing conditions on questions at a very early stage. The communis opinio is that wh-operators in root constituent questions in full and residual V2 languages are licensed in Spec-head configurations in COMP which involve feature sharing. In the unmarked case, the wh-operator is licensed in Spec-CP by moving the inflected verb to COMP.5 I will further assume that this licensing condition is part of the architecture of UG, so that the child does not need special evidence to assess it. In other words, the fact that wh-operators are licensed in Spec-head configurations is a universal property of the V2 grammar. If the assumption that the child can identify her native language as <+V2> at the outset is basically correct, we also expect her to be aware of the licensing conditions on wh-operators V-to-C requirement from early on. Notice, however, that this leads the child into a contradictory situation: on one hand, the child knows that the well-formedness of root constituent questions hinges upon the correct application of the V-to-C movement. On the other hand, COMP cannot be activated as the landing site for the head-moved verb. How does the child resolve this conflict? The hypotheses discussed above make clear predictions in this regard: (i) (ii) (iii)

The Freezing Hypothesis predicts that the child will not form questions at all. The Default Hypothesis predicts that the child will attach the wh-word to the highest XP. The Consistency Hypothesis says that the child will not violate the specifier requirement.

In the remainder of this chapter I shall attempt to show that children are guided by the LANGUAGE ACQUISITION DEVICE to choose the third route. In Section 2, I will review the relevant developmental data from High German and Bernese Swiss German. In Section 3, I will analyze the various patterns the child employs in root constituent questions. I will conclude the paper with a discussion of how the various hypotheses are related to the analysis in Section 3.

2.

The Data: Early Root Constituent Questions in German and Bernese

Root constituent questions begin to be productive at about 2;0. In the period prior to the acquisition of V2, three main patterns can be distinguished. I will

184

ZVI PENNER

refer to them as zero questions, particle questions, and V-end questions. All three patterns occur in both High German and Bernese Swiss German. Before we go on outlining the developmental scheme, let me roughly characterize each pattern: (i)

Zero questions are of the type:

(10) a.

(ii)

ist das? is this '{Where/what} is it?' b. das ist? In this pattern there is no overt wh-operator. The verb is either clause-initial or clause-final, and it belongs to the class of copular verbs: sein 'be', gehören 'belong', existential haben 'have', kommen 'come' (in the sense of 'belong'), etc. Particle questions. Here we can distinguish two subtypes: (A) the clause-initial word is a short form of a wh-pronoun (mostly wo 'where') attached to a cliticized copular verb (wo-dә, wo-de, wo-zd, wo-s, wo-za, wo-tha, vddd, etc.):

(11)

wo-de Zunge? where+cop tongue 'Where is the tongue.' (B) the initial element is not wh-marked. In fact, it is a particle-like word which as such does not exist in the target language:6

(12)

(iii)

(13)

d'

ist das? PRT is this '{Where/what} is it?'

V-end Questions display the inflected verb at the clause-final position as in subordinate clauses. Sometimes, a parasitic copula may occur in the second position: wo (ist) Mutter ist? where is mother is 'Where is mother?'

Let us now look at how these patterns are distributed across developmental stages. Starting with Bernese Swiss German, I will draw upon data from two longitudinal studies. Both children begin to produce root constituent questions at 2;0. Until the acquisition of the full range of CP effects at the age of 2;7 (cf. Penner 1992a), S. uses root constituent questions of two types only: zero questions and particle questions:

ACQUISITION OF ROOT (14) a.

isch das? is that '(What) is it?' d' ischdas?

b.

PRT

is

-QUESTIONS

185 S(2;0)

S(2;3)

this

'What is it?' Interestingly enough, the gap position may be occupied by an overt demonstra­ tive pronoun as in (15):7 s' isch das abegfaue? PRT is this down.fallen 'What fell down?'

(15)

S(2;3)

In general, M.'s data are very much the same. The prevailing patterns are zero and particle questions: (16) a.

chut da (gehört das)? belongs that To whom does it belong?' k' hett's Hammer? PRT has.it hammer 'Where is the hammer?'

b.

M(2;2)

M(2;l)

Note that M., in contrast to S., also produces V-end questions. They are, however, rather exceptional: (17)

u das macht? PRT that does 'What is it doing?'

Note that the interrogative particle is by no means restricted to constituent questions. In fact, we also find this element in yes/no questions as in: (18)

d'

isch da Guggi (Joghurt) PRT is here yoghurt 'Is here yoghurt?'

S(2;l)

We now turn to High German. The present review is based on data from Wode (1971), Grimm (1973), Park (1974), Felix (1980), Fritzenschaft, GawlitzekMaiwald, Tracy & Winkler (1990); Weissenborn (1990); Müller (1991); Tracy (1991a, b) and Gawlitzek-Maiwald et al. (1992). I will pay special attention to 10 longitudinal studies which focus on word order in early questions (Wode 1971; Fritzenschaft et al. 1990; Weissenborn 1990; Müller 1991; Tracy 1991a, b;

186

ZVI PENNER

and Gawlitzek-Maiwald et al. 1992).8 In the light of these data, two types of learner can be distinguished. Some children seem to opt for one pattern using it throughout the pre-CP period. Other children use more than one pattern. We will refer to the former as Type 1 and to the latter as Type 2. As for Type 1, it seems that children are free to choose between the two main options. So, for instance, the child I. (Wode 1971) opts for the V-end pattern. Consider, for example, (19), from Wode (1971:276): (19)

wo Björn wohnt? where Björn lives 'Where does Björn live?

I(2;8)

Grimm (1973:104) reports that the child M. exclusively produces questions like: (20)

und wer des ist? and who that is 'And who is that?'

Conversely, the children studied in Weissenborn (1990) and Müller (1991) seem to uniformly take the inversion (i.e., zero or particle questions) route. Note that the absolutely prevailing structure is the wo is(t) 'where is' particle pattern: In addition, let me point out that these authors do not mention any cases of zero questions. The examples in (21a, b) are from Müller (1991: 172) and Weissenborn (1990:204), respectively. (21) a.

b.

wo is der Auto? where is the car 'Where is the car?' wo bin du? (sic) where am you? 'Where are you?'

I(2;8)

H(2;0)

The group of Type 2 learners is studied in detail in Fritzenschaft et al. (1990), Tracy (1991a,b), and Gawlitzek-Maiwald et al. (1992). Tracy (1991b) proposes a three-stage model for the development of root constituent questions in V.: Stage 1: Zero Questions and V-end Questions:9 (22)

der Flöte is? the flute (recorder) is? '(Where) is the flute?'

V(l;ll)

ACQUISITION OF ROOT

WH-QUESTIONS

187

Stage 2: (2;0-2;2): V-end Questions: (23)

wo der Eimer ist? where the bucket is 'Where is the bucket?'

Stage 3: (until 2;3): V-end alternating with 'inverted' particle questions (24) a. b. c.

was is das? what is that was das is? was der gerne will? what he like wants 'What does he want?'

Stage 4: Regular target inversion No such scheme is provided in Gawlitzek-Maiwald et al. (1992). The authors state that at the age of 2;11 the child B. makes use of both inverted and V-end patterns. So we find in the representative sample of Gawlitzek-Maiwald et al. (1992) structures like (25a, b): (25) a.

b.

warum der auch ein Keks ißt? why he also a biscuit eats 'Why does he eat a biscuit too?' was macht (sic) du? what makes you 'What are you doing?'

Fritzenschaft et ai, reporting on w- structures attested in the corpora of the child L., note that the absolute minority of root constituent questions displays the inverted form. The overwhelming majority of her root interrogatives are of the V-end pattern, with or without an overt w- pronoun. This is exemplified in (26), from Fritzenschaft et al. (1990:40): (26) a. b.

das sind? that are was du machst? what you do 'What are you doing?'

L(2;l)

Tracy (1991a), describing the development of interrogatives in the children J. and S., notes that they produce both particle and V-end questions. In contrast to L.'s data, V-end questions are in the minority here (data from Tracy 1991a: 219,299):

188

ZVI PENNER (27) a.

b.

wo-de Badewanne where+cop bath.tub 'Where is the bath tub?' wo-dd Bobo? where+cop Bobo 'Where is Bobo?'

J(2;l)

S(l;9, 2;1)

Tracy also observes that the girl S. frequently uses the zero pattern Stephanie macht? '(What) is Stephanie doing?' until the age of 1;11. At this stage, S. replaces the zero pattern by the more adult-like was Stephanie macht?. Note that the verb-final word order remains constant. We will see below that the linkage between the verb-final pattern and object questioning is not random. Interestingly enough, the V-end pattern is intimately associated with the copying phenomena mentioned above. That is, we find a contracted copula in the second position and an inflected verb in the clause-final position (Tracy (1991a: 311,240): (28) a.

b.

wo-de Stephanie holt? wh+COP Stephanie brings 'What does Stephanie bring?' wo is das Junge is? where is the boy is

S(l;ll)

J(2;3)

The third child studied in Tracy (1991a), namely M., produces both V-end and 'inverted' patterns. Root constituent questions with differentiated wh-pronouns (was 'what', wo 'where', and wer 'who') begin to be productive at 2;8. This is also the stage at which overt complementizers begin to manifest themselves in subordinate clauses. During the period prior to this stage, the initial position is usually occupied by the 'discourse particle' oh, which also occurs in declaratives. In questions, oh is followed by a contracted form of the copula. Note that the latter does not necessarily express subject agreement, cf., for instance, (29), from Tracy (1991a: 337): (29) a. b.

c.

oh izd da Hühner drin? PRT COP here chickens inside oh izd dd Baecker heißt? PRT COP the baker called 'What is the name of the baker?' oh isn das? PRT COP that 'What is that?'

M(2;l) M(2;l)

M(2;l)

ACQUISITION OF ROOT

WH-QUESTIONS

189

The following general remarks on early questions are in order here. Note first that copulae and copular verbs are the only verbal elements which occur in the inverted patterns of the pre-CP stage. In general, they do not necessarily articulate subject agreement. Thematic verbs like kommen 'come' or kaufen 'buy' seem to occur very infrequently prior to the age of 2;6. I found only one example (Weissenborn 1990:240, with the verb kaufen at the age of 2;2) where a thematic verb appears in the left periphery of a root constituent question. We also have to pay attention to the fact that, as in Bernese, the wh-element in inverted structures may also introduce yes/no questions. According to Fritzenschaft et al. (1990: 30), this is attested in M.'s corpus: (30) was müssen wir gehen? what must we go 'Must we go?'

M(3;2)

This phenomenon goes hand in hand with the observation that children at the initial stage have considerable difficulties with the semantics of root constituent questions both at the level of processing and production. So, for instance, Müller (1991:172) reports that P(2;5) uses the was-is(t) pattern to question directionality (cf. also Tracy 1991a: 311): (31) was is das, legen? what is that put 'Where should I put that?' Grimm (1973), Felix (1980), Weissenborn (1990), Tracy (1991a:210ff.,252) and Weissenborn, Roeper & de Villiers (1991) observe that children often fail to comprehend the meaning of the wh-word as, for instance, in (32) (Felix 1980): (32) F:

wo sucht denn die Mami überall where looks for mother everywhere 'Where does mother look for everywhere?' Ch: Salz salt

Another particularity of the semantics of early questions is the lack of the <+variable> feature, i.e., young children fail to provide a 'distributive' (or exhaustive) answer when presented with questions like who has a hat? in a context where several people wear hats. In other words, they interpret open constituent questions not as denoting sets of pairs of things, individuals, places, etc., and the corresponding descriptions, but rather as a demand to 'name one X such that Y'. See also Roeper & de Villiers (1992a) and the references therein.

190

ZVI PENNER

Let us conclude this overview with a brief remark on the distribution of the overt expression of wh-in early grammars. As observed by Müller (1991, 1992), overt wh-marking in early grammars is a pure root phenomenon. Indirect questions display the pattern I have referred to above as preconjunctional embedded clause, i.e., indirect questions are generally complementizerless. Consider, for instance, (33), taken from Müller (1991:173):10 (33)

i muß i muß fragen Mama ha-hat hat die K[j]lebe I(2;10) I must I must ask Mami has has the adhesive.tape

Given these observations, the following descriptive statements can be made: (i)

(ii)

(iii)

(iv)

(v) (vi)

Taking 'inversion' and V-end to be the main patterns of early root constituent questions, it seems that there are no compelling correla­ tions between the interrogative patterns and specific developmental stages in the pre-CP period. There is no one to one correlation between question patterns and semantic functions. However, a close inspection of the data reveals that there is clear a preference in the choice of word order patterns. In those cases, in which children make use of both inverted and V-end root constituent questions, there is a strong bias for object questions to be realized as verb-final. The data show that there is no regularity in the preference for a certain pattern at the level of individual choice. Given the above facts, we may say that both patterns are equally productive. Neither the inversion pattern nor the V-end pattern can be considered 'exceptional'. As depicted above, some children opt for the V-end pattern, while others take the opposite route. The same happens with Type 2 children: some have a strong preference for the inverted pattern, while others prefer the V-end structure. In other words, it seems that, in the absence of COMP, both patterns are equally available. As for the inversion pattern, it seems that the clause-initial wh-elements are nonreferential. They are particle-like in the sense that they mark interrogatives as such, sometimes crossing the boundaries between constituent and yes/no questions. They are not used to denote subjects, objects, certain adverbials, etc. The bound variable reading is not available in early grammars. One striking fact about early inverted root constituent questions is the systematic exclusion of thematic verbs. In other words, taking the copula to be a bare feature carrier, we may say that inverted root

ACQUISITION OF ROOT

WH-QUESTIONS

191

constituent questions involve only 'dummy' verbs. A dummy copula may co-occur with a clause-final verb, (vii) There is no overt wh-marking in nonroot positions.

3.

The Analysis

3.1. Preliminaries We now have the ingredients to enter into the theoretical discussion. It has become clear that the first strategy to avoid inconsistency between the target and the intermediate grammar, namely 'freezing', is not the route the child takes. Interrogative structures are a productive sentential mode in early grammars. We are now left with the choice between the default and the language-specific option. From the point of view that the explanandum is the entire range of phenomena mentioned in the previous section, we will have to exclude any analysis which is not designed in this way. For convenience, let us briefly formulate the factors a theory of root constituent questions in early grammars has to account for: (34) a. b. c. d. e. f. g. h. i. j.

Alternation of inverted and V-end patterns Nonreferentiality of the wh-element Initial wh-elements occur in both yes/no and constituent questions Constraints on the verb type in inverted structures: only copular verbs may occur. The bound variable reading is unavailable. Overt wh-marking is a pure root phenomenon. In some cases the gap (or the operator) position is occupied by an overt demonstrative pronoun. If the child adopts a mixed strategy, object questions are likely to be spelled out as V-end patterns, V-end structures may involve a dummy copula in the second position, Correlation with the absence of productive V2 and other COMP related phenomena.

In fact, most of the recent analyses of early root constituent questions I know of do not fulfill this condition. This is, for instance, the case for Müller (1991), who proposes that inverted root constituent questions are created by placing the

192

ZVI PENNER

operator in Spec-IP which can figure as an A-position,11 or for Weissenborn (1990), who assumes that inverted questions are CPs of the kind found in residual V2 languages like English and French.12 Roeper (1992) suggests that particle questions can be captured as a special case of adjunction in the sense of the default rule in (8) above. Again, this analysis must also be extended in order to account for the non-inverted patterns. The only analysis I know of where a serious attempt is made to derive the two competing patterns from one and the same grammar is Tracy (1991b). She assumes that V.'s developmental scheme (cf. (22) and (24) above) is amenable to the fact that there is an intermediate stage at which the grammar is allowed to 'swing' between a head-initial and head-final IP. Apart from the fact that this analysis is incapable of straightfor­ wardly accounting for (34b)-(34j), it also raises the question of whether the child is not guided into an irreversible grammar. That is, from the point of view of learnability, if the child grasps the input in this way, there is no reason why this intermediate stage should ever be abandoned in favor of the target value. 3.2. The 'Inverted Structures Let us assume, as a first approximation, that the inverted structures are XPs, Xo is the highest position of the inflected verb, with the wh-marking as part of verbal inflection. The latter is spelled out as a copular verb. In other words, the overt wh-morpheme is not the question operator as in the adult language, but the expression of the wh-feature instantiated on the head of the highest projection. Slightly extending the core idea of Roeper & de Villiers (1992b), let us further assume that our inverted questions are NULL CONSTANT CONSTRUCTIONS in the sense of Lasnik & Stowell (1991) and Rizzi (this volume). The main idea behind the null constant approach is that semantically nonquantificational operators bind a <-variable> empty category which can be characterized as a null R-expression or null constant. The <-variable> feature should account for the characteristic lack of weak crossover effects and related phenomena in operator-involving construction, such as sentential adjuncts. This is illustrated in (35): (35) whoi did you see ti [OPi before his mother saw nci] In Rizzi (this volume), this theory is extended to account for topic drop constructions in German as well. Thus, for instance, object topic drop can be assigned the structure in (36): (36) ... DBi OPi [IP nci habe [ich schon gesehen]] have I already seen

ACQUISITION OF ROOT

WH-QUESTIONS

193

In (36), a null constant chain which consists of a DISCOURSE BINDER (DB), an empty operator, and a null constant in the gap position. We will return to this issue below. Suppose now that early inverted root constituent questions are instances of the null constant construction as well, assigning them the underlying structure depicted in (37): (37) DBi OPi [[+wh]-X° ... nci....] That is, the gap position is occupied by a null constant bound by a clauseexternal null operator. The wh-feature is spelled out on a copular head. There is no wh-operator in a specifier position, as in the adult language. This accounts for the lack of the bound variable reading. This analysis, which will be refined as we proceed, has the following implications: (i)

(ii)

The child treats interrogative marking on a par with languages like Vicentino and related Romance languages in which wh is spelled out as part of the inflectional complex of the verb.13 There is no overt wh-operator in a specifier position. This kind of wh-spell-out expresses interrogativity as a sentential mode. In this way, the lack of referentiality in the above sense (34b) is accounted for. In the same vein, we may say that the child, using inverted questions, does not strictly distinguish between constituent and yes/no questions. This explains the occurrence of wh-elements in yes/no questions (34c). We will return to this issue below. The constraint on the occurrence of thematic verbs in early questions (cf. (34d)) can also be couched within this analysis. Copulae are intrinsically specified as pure spell-out forms of inflectional features. As such, they are predestinated to serve as index carriers. This is what we find in predication structures as the expression of subjectpredicate coindexation, or in case of discourse linking at the clauselevel.14 On the hypothesis that question formation is accomplished by means of null constant chains, and that the wh-marked element participates in the null constant chain, it would not be implausible to assume that the feature <+wh> will be spelled out on an index carrier. This accounts for why inverted questions at the initial stage tend to attach the wh-marker to a copular rather than to a main verb. Support in behalf of the hypothesis that the copula in inverted questions is a dummy index carrier comes from root constituent questions of the type (28) above. We will refine this analysis below.

194

ZVI PENNER (iii)

(iv)

The null constant hypothesis also accounts for (34g): the fact that an overt demonstrative pronoun may show up in the gap position indicates that this position is not occupied by a <+variable> trace. This goes hand in hand with the semantics of early questions (34e). The restriction of overt wh-marking to root contexts can be seen as a side-effect of the underlying structure (37). Topic drop is strictly illicit in subordinate clauses. This is amenable to the fact that the discourse antecedent fails to govern into the embedded clause, in order to transfer its index to the null operator. Given (37), the structure of our root constituent questions is the interrogative counterpart of the topic drop pattern. By analogy, (37) in nonroot position is ruled out. This accounts for (34f).

So far, we have been able to account for (34b)-(34f) without having settled the question of the identity of Xo in (37). Turning now to the question of whether Xo in our scheme is an IP-internal head or COMP, it seems that, from a theoretical point of view, there is no a priori reason to prefer one of these options over the other. Weissenborn (1990) proposes that early inverted questions involve V-to-C movement. This is an appealing assumption, since, in some cases, the inverted pattern mimics the target V2 structure. Note, however, that in order to maintain this approach, one has to stipulate that questions, but not declaratives, involve a COMP projection. This asymmetry hypothesis has, of course, one conspicuous advantage: it immediately accounts for the clause-initial position of the inflected verb. Obviously, this account is tenable only if we assume that the child proceeds in a structure-specific way or has a dual grammatical system. Contrarily, if we assume V-to-C movement both in declaratives and interrogatives, as well as clitic movement to COMP (Haverkort & Weissenborn 1991), the introduction of overt complementizers, and the transition from an un­ constrained to a severely constrained applicability domain for topic drop are all by-products of the fact that the COMP position becomes available, then we expect correlations with related phenomena to entangle the structural identity of early questions. As discussed in detail in Penner (1992a) there is indeed such a correlation between the following factors at the age of 2;6-3;0: (i) (ii) (iii)

the acquisition of V2; the 'explosive' reduction of topic drop: the transition from 15 SUBJECT DROP to the target value; the disappearance of zero and particle questions.

DIARY

ACQUISITION OF ROOT

WH-QUESTIONS

195

These correlations can be attributed to the fact that the introduction of CP in V2 structures narrows down the occurrence of null constant in some way. Let us first turn to subject omission in early grammar. Recent work on this topic (cf. Weissenborn 1988, 1992a; Weverink 1990; and Rizzi, this volume, among others) has suggested that subject omission in early German is an instance of topic drop. A closer inspection of the data shows that we have to distinguish two stages in the development of null subjects in the language. Subject omission at the earlier stage seems to be consistent with what Haegeman (1990) dubs "nonovert subjects in diary contexts," i.e., the subject may be freely dropped, regardless of whether it has an intermediate explicit discourse antecedent or not. Thus, for instance, the child regularly omits the subject in utterances like (38): (38) habe Apfel have:ISG apple This omission occurs independently of whether or not the subject has an overt antecedent within the same discourse sequence. At the descriptive level, we would like to claim that there is no 'formai' topic chain in this case. The second stage is consistent with the target grammar, which allows only restricted topic drop (or 'pronoun zap'). The kind of topic drop we find in the adult language usually takes place in short question-answer sequences of the type in (39): (39) A: B:

Was machte Hans, als du ihn sahst? what did Hans when you him saw las ein Buch read:PAST a book

More precisely, we may say that topic drop in the adult language is licensed when the conditions in (40) are met: (40) a. b.

The antecedent is overt. The antecedent is placed in the utterance immediately preceding the CP which contains the topicalized element.

The major difference between non-overt subjects in diary contexts and topic drop is that only the latter requires a topic chain as defined above, while the former can be arbitrarily bound to any explicit or implicit element in the discourse. On the assumption that both constructions involve a null constant, we might say that null constants in topic drop, but not in diary contexts are obligatorily discoursebound by an immediate antecedent. This can be captured in structural terms as follows: recall that topic drop is intimately associated with the Spec-CP position, which hosts the null operator. In other words, we may say that the empty subject

196

ZVI PENNER

position is assigned an index through the COMP system. This assumption goes hand in hand with the hypothesis that COMP is the host of the feature (erentiality).16 The latter is a cover term for factivity, discourse linking, and sentential predication. In German, the assignment of <+ref> to COMP is a local process. In the unmarked case (factive complements, relative clauses, etc.) the index has its source in the superordinate clause. Topic chain formation is not exceptional in this regard: the discourse binder must be locally accessible to the null constant chain. Pursuing this line of argumentation, we would like to argue that non-overt subjects in diary contexts are assigned their 'diary value' under more liberal conditions. In general, we would like to say that non-overt subjects in diary contexts do not necessarily involve chains headed by an explicit and local discourse binder. Put in our terminology: the linkage between the null constant and a potential antecedent does not involve assignment through COMP. This assumption is compatible with the hypothesis that diary contexts with a nonovert subject operate on IPs rather than on full-fledged CPs. This implies that the transition from constructions with a non-overt subject in a diary context to topic drop constructions is a side-effect of the change from an IP- to a CP-oriented grammar. Given that null subjects, as opposed to null objects (cf. Rizzi, this volume), are indifferent with regard to -features, we may think of the underlying structure of constructions with a non-overt subject in diary context as comprising a null constant linked to the discourse binder through the inflected verb: (41) DBi ... [IP nci [I' Vfinitei ... Returning now to root constituent questions, the question our analysis raises is whether they are derived on a par with constructions involving a non-overt subject in a diary context. In the light of the above considerations, we may say that the reference of the null constant chain in root constituent questions is recovered like in diary contexts, rather than like in topic drop constructions. Recall that early particle (and zero) questions are nonreferential in the sense that they do not contain fully specified wh-pronouns which denote place, time, modality, individuals, things, etc. The identity of the questionee must be recovered in much the same way as the missing subject in the diary context con­ structions. By analogy, we could contend that this kind of nonreferential interpretation in questions is possible as long as the relevant features do not percolate through the COMP system. Note that this assumption fits the obser­ vation that wh-can be overtly marked in root, but not in subordinate position. On the view that the diary contexts with a non-overt subject and nonreferential questions are merely two different manifestations of one and the same

ACQUISITION OF ROOT

WH-QUESTIONS

197

underlying phenomenon, namely the absence of a mediating COMP, it would not be implausible to assume that early inverted questions are in fact bare IPs. Recall now that the identification chain in questions comprises a 'dummy' copula which serves as an index carrier. Given these factors, we may now replace (37) by (42) as the underlying structure of zero/particle questions: (42) DBi [IP [+w]-copulai ... nci ...] This line of argumentation is an attempt to account for the semantic properties of early inverted questions on the basis of syntax-semantics correspondences. One unexpected fact is the word order pattern itself: why do we have inversion at all if V-to-C movement should not apply? The tension between the assumption that early grammars lack a COMP projection and the existence of V2-like inversion patterns has been extensively thematized in recent literature. The general assumption is that there should exist a right-governing IP-internal node to which the finite verb might be raised. To name just a few proposals, Clahsen (1991) refers to this node as F°(initeness), Meisel & Müller (1992) assume movement to TENSE0, Gawlitzek-Maiwald et al. (1992) argue for I-final/I-initial alternation, and Penner (1992a) demonstrates that the finite verb in Bernese Swiss German moves to the so-called Wackernagel position. There is, however, no direct morphological evidence in favor of this hypothesis. In contrast with certain varieties of Dutch (cf. Zwart 1992), in which the ending of the verb in COMP is morphologically distinct from its IP-internal counterpart, IP-internal verb movement in German is not overtly marked. It is thus rather difficult to find morphological evidence for the assumption that the verb does not move beyond the IP.17 Fortunately, the 'intermediate movement hypothesis' is buttressed by the acquisition data of those languages which express interrogation by means of complex inversion. One such language is French. According to Rizzi & Roberts (1989), verb-clitic inversion of the type exemplified in (43) emerge from movement to COMP, combined with right-dislocation of the thematic subject: (43)

où est-il, le papa where is.he the daddy

As shown in Meisel & Müller (1992) and Müller (1991), children fail to invert the verb and the subject clitic in early grammar. The same is shown by Haverkort & Weissenborn (1991) for object clitics in imperatives. However, if we carefully inspect the data of early questions, we find out that children do not use the declarative pattern, but, in fact, build interrogatives with a partial inversion. So, for instance, all noncleft questions in the corpora of Müller (1991) are of the type in (44), with a fronted subject clitic and a right-dislocated subject:

198

ZVI PENNER (44)

où (i)l-est, le papa

This pattern is reminiscent of the clitic inversion found in, for example, indirect questions in Northern Italian dialects (see (46) below). As shown in detail in Haverkort & Weissenborn (1991), clitic placement in early French is determined by the fact that the verb cannot move to COMP. I will therefore assume that the inflected verb is in AGR° and that the subject clitic is extracted from Spec-VP and lands on the left side of AGR. Subject dislocation is triggered due to Case absorption by the clitic. Notice that this pattern mimics the early inverted questions in German, the main difference being that the former is morphological­ ly more transparent. This is due to the fact that the French child can express inflectional features in a more articulated form, namely by breaking up AGR into a verbal ending and a subject clitic. It is not unreasonable to assume that the subject clitic plays the same role as the 'dummy' copula in German question, namely to express the linkage between the null constant and the discourse as well as to host the wh-marker. As can be clearly seen from the data, the wh-marker is always hosted by the fronted subject clitic. I have not found cases where the wh-element is attached to the verb. This indicates that, in order to be properly interpreted, the wh-marker must be moved to its focus position in the syntax. This is the leftmost position of the IP.18 3.3. The V-end Structures We now turn to the residue of our explanandum, namely (34a) and (34h). So far, we have assumed that the peculiarities of the inverted patterns are amenable to the interaction between the unavailability of COMP and the feature <+variable> in early grammars. It would be natural to assume that the V-end pattern of early questions is another by-product of this effect. Intuitively, the V-end pattern of root constituent questions is reminiscent of the preconjunctional embedded clauses of the type exemplified in (3) above. If we assume that V-end questions are an instance of 'subordinate clauses in root position', we have to account for how such a shift may emerge. In fact, root questions in subordinate disguise are not seldom at all, crosslinguistically. On the one hand, we find in Vicentino, Colloquial French, and related Romance dialects root questions with a doubly filled COMP of the type in (45) (see Penner & Bader, in prep.): (45) chi che venga? who that came: SBJU 'Who came?'

ACQUISITION OF ROOT

WH-QUESTIONS

199

On the other hand, we find cleft constructions. Cleft questions constitute a very wide-spread interrogative pattern across languages. Wh-clefts are obligatory in many VSO and VOS languages (e.g., Palauan, cf. Georgopoulos 1991; Berber, cf. Calabrese 1987 and Ouhalla 1988; Irish, Stenson 1981; McCloskey 1979; Sells 1984, and Siadhail 1989), in subject questions in certain Romance dialects (e.g., Véneto, cf. Poletto 1990 and Penner & Wegmüller 1991) and Quebec French, cf. Noonan 1989 and Lefevre 1982), and in subject questions in Bantu dialects (cf. Demuth 1992a, 1992b).19 In languages like Bernese Swiss German and High German, cleft questions are entirely optional. There are many ways to mark cleft questions. In some languages, the clefted element is embedded inside a copular structure. This is what we find in Vicentino, for instance (46a). In other languages (e.g., Irish, cf. (46b,c)) we find a relative-clause-like construction with the wh-pvonoun figuring as a head noun. Consider the examples in (46): (46) a.

b.

c.

chi xe che-'l ga visto, Giani whom isthat+SUBCLhas seen Giani 'Who has Giani seen?' cé aL dhiól an domhan? who REL sold the world 'Who sold the world?' an fear aL dhiól an domhan the man REL sold the world 'the man who sold the world'

In High German and Bernese Swiss German, cleft questions, as the marked option for question formation, fulfill a special semantic function.20 So, for instance, a Bernese cleft question of the type in (47), with a subordinate clause introduced by the relativizer wo, is assigned the 'exhaustive listing' or individual interpretation: (47)

wär isch es,wo du mit-im geschter gschpiut hesch? who is it, REL you with.him yesterday played have

In other words, the reference of the relativized element is part of the background knowledge of both speaker and addressee. This comes very close to our null constant reading. Returning now to the early stages of language development, we know that the child has all the prerequisites to mimic cleft questions in the adult language, namely the null constant device and the preconjunctional embedded clause mechanism. The V-end question of the type in (16c), repeated here as (48), is translated into the cleft pattern in (49):

200

ZVI PENNER (48)

was der gerne will? what he like wants 'What does he want?'

We may say that the V-end structure is in fact the preconjunctional embedded clause version of a relative clause (cf. (3) above) with the wh-pronoun in the head noun position followed by a conjunctionless subordinate clause: (49)

[NP wasi [CP [JP der [VP nci gerne tj] willj]]]

From a functional point of view, the main difference between V-end and inverted patterns is that only the former structure is compatible with a 'referential' reading of the wh-element. As an occupant of the head noun position, the wh-element is more likely to fulfill its pronominal function (i.e., wer for subjects, was for objects, etc.). This analysis accounts for our observation in (34h): for those children who make use of both inverted and V-end patterns, object questions tend to be of the latter type. Similar distribution can also be observed in early French. As alluded to above, the children studied by Müller (1991) use both (semi-)inverted and cleft patterns. A close inspection of the distribution of these patterns reveals that (semi-)inverted questions are overwhelmingly où questions, while clefts of the type in (50) tend to co-occur with argument q-pronouns: (50)

qu'est-ce qu'il fait lui, avec ça what.is.it that.he does him, with this

This parallelism indicates that the coexistence of the inverted and the V-end patterns in early questions is not random: the former structure is reserved for the universal interrogative reading, while the latter is made use of whenever the referential interpretation requires more salient expression.21 3.4. On Dummy Copula Constructions Let me conclude this section with a brief remark on the 'parasitic' copula in V-end questions (point (34i) above). As shown in (28) and (29) above, a dummy copula may co-occur with another inflected verb in the clause-final position. This is by no means a particularity of early German grammar. Parasitic copulae of this kind are found in early English as well. They are documented in Ingram (1988) and references therein, Roeper (1990, 1991) and Stromswold (1990).22 The main property of the dummy copula both in German and English is that it might, but need not, match the -features of the sentential subject. Thus, both questions of the type (51a) and (51b) are attested:

ACQUISITION OF ROOT

WH-QUESTIONS

201

(51)

a. are you don't know Sharon's name is? b. where is dey're inside dǝ house? Let us refer to these patterns as DUMMY COPULA CONSTRUCTIONS. Dummy copula constructions occur already at a fairly early stage. As can be deduced from (49a), they appear prior to the acquisition of complementizers. At first glance, dummy copula constructions in German seem to display the familiar ' V2' effect: the dummy copula occurs clause-initially, while the inflected verb remains in its clause-final position. Is it an in-situ implementation of I-to-C movement? Probably not, since under a closer inspection the copula 'copying' turns out to be a 'fake' one: the ϕ-features of the clause-final and the clause-initial inflection do not necessarily match. Given this basic fact, we can accommodate the dummy copula constructions within the theory developed here in the following way. Dummy copula construction questions are IPs, which implies that the child may opt for the verb-final pattern. The wh-marking, however, must be clause-initial, in order to meet the scope requirement. In order for the wh-morpheme not to strand, an expletive dummy copula can be generated in the clause-initial position. This dummy copula is now part of the null constant chain, but not of an I/C chain. This is why the ϕ-features of the copula do not necessarily match those on the main verb.23 This analysis also accounts for the fact that there is no fullfledged wh-operator, at least in (28a) and (29). In the former wo-de denotes a target 'what' (rather than 'where is'), while the particle oh-izd in the latter should be paraphrased as 'how'. 24

4.

Relating the Analysis to the Proposed Solutions

The question this paper addresses is how the child treats syntactic operations which exceed the boundaries of her current grammar. Three potential solutions were mentioned in this regard: (i) (ii) (iii)

The Freezing Hypothesis The Default Hypothesis The Consistency Hypothesis

This paper checks these three learning-theoretical assumptions against the data from the acquisition of root constituent questions. This domain was chosen as the arena for these competing solutions, since we have good reasons to assume that the child should be aware of the licensing requirements on questions in German from the outset.

202

ZVI PENNER

I have shown that the child makes extensive use of the sentential mode 'interrogative', arguing that question formation in early grammar is accomplished as null constant chains either by means of wh-marking on the inflected verb or by means of cleft questions. The corresponding configurations are: (52) a. DBi [IP [+w]-copulai ... nc¡ ...] b. [NP w pronouni [CP [IP ... nci ... V-fin]]] Recall that the <+w> marking on the copula in (52a) may be covert. The data show that the hypothesis of the Freezing Hypothesis is simply invalid, since from 2;0 on, questions are produced although the child still does not have all the ingredients the target grammar requires. Notice also that neither (52a) nor (52b) is an adjunction structure in the sense of (8) above, i.e., the predictions of the Default Hypothesis are not borne out by the data either. It should also be pointed out that the application of the adjunction rule in (8) would not only run afoul of the X'-theoretical status of wh-elements in V2 language, but would also violate a basic rule in German which says that wh-operators are always placed in a specifier position, regardless of whether they are in situ or fronted. For a detailed discussion of this point, cf. Rizzi (1991) and Penner & Schönenberger (1992) and references cited therein. It is not clear to me how the child would ever be able to get rid of this overgeneration. The same constraint also holds for negated nominals and related elements. This implies that the misor default analysis of wh-at the initial stage can contaminate further parts of the system.25 What about the predictions made by the Consistency Hypothesis? It seems that the condition associated with this hypothesis, namely the continuity of the licensing configuration is met in a somewhat abstract way. As far as the inverted pattern in (52a) is concerned, we would like to say that it does not involve an operator at all. The interrogative chain is wh-marked on a head in a scope position and involves a slot for a null constant. This, perhaps proto-typical question form, is probably interpreted as 'replace the null constant by one x, such that y'. What must be licensed in this pattern is the null constant. Given the lack of an operator, the licensing conditions of the latter are not activated at all. In other words, the child 'freezes' potential conflicts with the well-formedness conditions on operators by creating operator-less configurations. In this way, the learner avoids any change of licensing configuration of the wh-pronoun. I leave open the question whether the operator-less pattern is just another type of a default mechanism. The emergence of V2 structures, which obligatorily open an operator position in Spec-CP puts an end to this kind of question formation.

ACQUISITION OF ROOT

WH-QUESTIONS

203

With the structure (52b), the child mimics the cleft pattern. My proposal has been that V-end questions are a special type of relative clause with a wh-pronoun in the position of the head noun. As alluded to at the outset, the licensing mechanism of wh-operators in V2 languages involves feature sharing in a Spec-CP-head-CP configuration. We can also put it this way: an unsaturated element, namely the head (and its complement), is linked to a specifier element to create a saturated structure by virtue of sharing a certain feature. Is it plausible to assume that the relative clause structure in (52b) fulfills a similar function? Note first that the wh-chain in (52b) is spelled out as a. predica­ tion configuration. In the spirit of predication theories of the kind proposed in Rothstein (1983) and subsequent work, let us assume that predication configura­ tions consist of a predicate, which is the unsaturated element, and a subject. It is the rule of Predicate Linking (i.e., coindexation of subject and predicate under c-command) due to which saturation is achieved. That is, the open predication configuration is 'closed up' by virtue of index sharing. Given this analogy, we may say that predication structures are 'extended Spec-head configurations'. In the same vein, one might argue that in (52b) the wh-pronoun is the subject of the predication configuration with the relative clause predicated of it. That is, they yield an extended Spec-head configuration in the above sense. If this analysis is basically correct, we may say that the wh-pronoun in V-end root questions figures as an operator in a specifier position. The licensing requirements on wh-operators are not met by means of local Spec-head agreement, but rather by virtue of the rule of Predicate Linking which ensures that the wh-operator and the relative clause are coindexed. From this perspective, we can now define the term 'the same type of licensing configuration' (cf. (9c) above) as follows. Putting aside the gap position, we would say that the licensing configuration of the target grammar involves local feature sharing between the head of CP and its specifier. Contrarily, in the licensing configuration of V-end questions a nonlocal process of feature sharing between the head of a subordinate CP and the w/z-pronoun, which figures as the subject of the subordinate CP, takes place. Both licensing configurations are of the 'same type' in the sense that in both cases the licensing device is feature sharing. The difference between the two licensing configurations boils down to the local/nonlocal distinction. It is tempting to say that the formation of V-end questions is the 'last resort strategy'. One might want to assume that the child first attempts to create a local feature sharing configuration for the w/z-operator. This attempt fails, since the head of the feature sharing configuration, namely COMP, is not available yet.

204

ZVI PENNER

The child then applies the alternative strategy, namely the mechanism of nonlocal feature sharing. Again, the success of this step depends on whether the prerequisites for this strategy are provided by the initial grammar. This time, however, the licensing conditions are fulfilled: the mechanism of nonlocal feature sharing can be spelled out as a preconjunctional embedded clause structure, the latter being independently available in the grammar. The question this analysis raises is why the child should ever abandon this question pattern at all, given that it is 'tolerated' by the target grammar. The basic intuition is that the creation of predication chains with wh-antecedents is licensed only as long as the subordinate clause is an instance of preconjunctional embedded clause, but becomes ill-formed as a side-effect of the introduction of overt complementizers. A detailed discussion of the syntax of preconjunctional embedded clauses is found in Müller & Penner (1992). A review of this analysis would take us too far afield. Confining ourselves to the most essential ingredients of Müller & Penner's approach, their main claim is that the COMP position of the subordinate clause is occupied by a 'universal' index (i.e., generalized <+ref> which is assigned either by factive verbs or by coindexation in relative clauses and adverbial adjuncts). The presence of this universal index is reconstructed by the clause-final placement of the finite verb, i.e., the fact that the finite verb in its final position marks COMP as being non-empty. We would thus say that preconjunctional embedded clauses are licensed as long as: (i) (ii)

The embedded COMP is assigned an index by a matrix element, COMP is marked as being non-empty.

By contrast, the licensing conditions on overt COMPs are less liberal. To give just one example, the COMP of relative clauses in Bernese comprises a pro operator in the specifier position of the relativizer wo. In addition to the formal licensing of the relativizer wo and the pro operator, the feature content of the latter must be identified. As proposed by Penner & Bader (in prep.), the identifying feature is Case. Note, however, that this feature cannot be supplied by the wh-operator in the head noun position, since it has no source of Case assignment itself. Put differently, as soon as the target COMP is integrated into the subordinate clause, replacing the preconjunctional embedded clause system, V-end questions of the type (52b) become incapable of fulfilling the licensing requirements. They thus have to disappear from the child language, giving way to V2 root questions. In sum, I have argued that, at least as far as root questions are concerned, the acquisition process does not involve a default mechanism of the 'universal

ACQUISITION OF ROOT

WH-QUESTIONS

205

adjunction' type. The data indicate that the child tends not to violate the target system of licensing conditions. If an utterance contains an overt wh-operator, and the required licensing configuration is inoperative, the child extends the latter, keeping the licensing principle unchanged. The application of this mechanism to other acquisition domains awaits future research.

Acknowledgements This paper grew out of my talk at the Workshop on the Development of Movement and Inflection, 1991 GLOW Colloquium, Leiden. I thank the partici­ pants of this workshop for their valuable comments. My thanks also go to Thomas Bader, Teun Hoekstra, Natascha Müller, Janet Randall, Thomas Roeper, M. Schönenberger, Rosemary Tracy, Jürgen Weissenborn, and two anonymous reviewers for helpful comments. Notes 1.

For a detailed overview of language variation in this domain, cf. Penner & Bader (in prep.).

2.

This is what Zwart ( 1992) proposes for Dutch.

3.

I assume that this is the modular expression of what is hidden behind Clahsen & Penke's (1992) approach, which says that the acquisition of subject-verb agreement is an indispensable prerequisite for the integration of COMP into the operative system of the child's grammar and the theory of Verrips & Weissenborn (1992) which relies on the assumption that verb movement is triggered by 'finiteness', regardless of whether strict agreement has been acquired or not. It might be the case that finiteness suffices to triggering verb movement across NEG and related elements, which is a canonical V2 effect. It is, however, not unreasonable to assume that the integration of the languagespecific COMP system refers to idiosyncratic properties of inflectional paradigms. In Penner (1992b) it is assumed that the unavailability of COMP is just a special manifestation of a more general acquisition procedure due to which the unification of Ä- and A-domains is inoperative at the initial stage. This should account for parallelisms in the acquisition of both DET and COMP.

4.

Müller & Penner (1992) assume that COMP may be realized at the initial stage only if it bears an abstract index (assigned, e.g., by the rule of Predication in relative clauses) whose presence is reconstructed by the verb-final structure. We shall return to this issue in Section 4 below.

5.

See Noonan (1989) and Rizzi (1991) for a detailed discussion of this point. Rizzi's basic assumption is that the inflected verb agrees with the operator in Spec-CP with regard to <+wh>, which is an intrinsic feature of the former.

206

ZVI PENNER

6.

A similar phenomenon is also attested in Swedish. Lange & Larsson (1973) report that the child E. used the particle heja as a universal interrogative marker. This element does not exist in the target language.

7.

Unpublished English material made available to me by Rosemary Tracy shows that this is by no means an exceptional phenomenon. The data contain a series of 'anomalous' wh-questions of the type: (i) who is it ball? Such patterns seem to be productive at about 2:0.

8.

Unfortunately, the corpora are not uniformly transcribed. Only in Wode (1971) and Tracy (1991a) the examples phonetically express 'particlehood', contraction, and related effects.

9.

Tracy (1991b) also reports on uncertainty with respect to verb placement. This can be seen in utterances of the type: (i) wo is noch eine rote ... die kleine Kette is? V(l;ll) where is another a red the small chain is

10.

I. is a French-German bilingual child. He is the only child reported on in the acquisition literature who produces V3 preconjunctional embedded clauses.

11.

A similar proposal is made in Vainikka (1992).

12.

Weissenborn (1992b) also discusses the V3 pattern, which he explains as a special manifestation of the split-COMP analysis of Müller & Sternefeld (1990).

13.

A northern Italian dialect of the Veneto group. As shown in detail in Penner & Bader (in prep.), this dialect has a special AGR (subject clitic) paradigm which occurs only in open interrogatives. Another morpheme which is traditionally taken as an interroga­ tive marker in Io is ti in many French varieties (cf. Noonan 1989 and Roberts 1990, among others). The case of ti is, however, not entirely clear. Penner & Bader (in prep.) show that at least in certain French dialects ti is part of the NEG complex. Its function is to mark questions as 'closed up' (anti bound variable effect).

14. This is what we find in the so-called nominal sentences in the Semitic languages. Penner & Bader (in prep.) show that the copula in Hebrew V2 moves to COMP in order to express discourse linking of the whole proposition. This is a special case of index spell-out on the copula. 15.

As shown in Penner (1992a), the rate of (nontarget) missing subject drastically decreases at the age of 2;6 (from 20 percent to 35 percent at the age between 2;1 and 2;6 to practically zero). The data in Weissenborn (1992a) point to a more gradual transition.

16. For a detailed discussion of this issue, cf. Penner & Bader (in prep.). 17.

In Penner (1992a) it is shown that movement to the Wackernagel position in early Bernese Swiss German is marked by a 'parasitic' subject clitic. This kind of evidence is restricted to declaratives.

ACQUISITION OF ROOT

WH-QUESTIONS

207

18. Natascha Müller (pers. comm.) points out to me that French children do not produce in situ wh-questions. This is a bit surprising, since the in situ pattern is wide-spread in the target language. This observation is compatible with the assumption that in early questions the wh-element is part of a head, rather than an operator. In situ wh-operators are IP-internal XPs. As such, they cannot be licensed at the initial stage. 19. According to Noonan (1989) and Poletto (1990), cleft questions in Romance emerge as a side-effect of the incapability of the verb to move to COMP. 20.

The grammar of cleft questions in Bernese Swiss German is discussed in Penner & Bader (in prep.).

21.

In addition to the inverted and V-end patterns, we also find 'V3' w/z-questions of the type in (i): (i) warum Papa geht weg? B(2;7) why Daddy goes away? (Felix 1980) (ii) warum weiter geht nicht? B(2;ll) why further goes not? (Fritzenschaft et al. 1990) V3 questions are quite rare. They occur after the acquisition of V2. V3 subordinate clauses are discussed in great detail in Weissenborn (1992b) and Penner (1992b). In that paper I propose that they are instances of recursive CPs of the kind found in weil clauses in German (i.e., a V2-CP embedded under a COMP). The same can now be claimed for the above examples: the V3-root-constituent question is a preconjunctional embedded-clause pattern with a recursive instead of a simple subordination.

22.

These observations are confirmed by unpublished material made available to me by Rosemary Tracy.

23.

There are some particularities of early yes/no questions which are not really covered by this account. Note first that questions with the second person have are in the initial position although the theory predicts both are and is. This generally holds for the English material I examined. As for German, example (21b) shows that bin 'am' may wrongly occur with the 2SG in w/z-questions. Interestingly enough, the are you ... V-finite questions are found only in yes/no questions, while AGR mismatching is observed only in constituent questions. It has been proposed above that at the initial stage inverted root constituent questions and yes/no question are treated alike. In the light of these findings, this account must be made more precise. It is well known from the acquisition literature on English that some children have inversion in yes/no questions prior to w/z-questions. The relevant data are summarized in Weinberg (1990). The opposite is claimed by Valian, Lasser & Mandelbaum (1992). As for German, it has recently been claimed by Rosemary Tracy (1991b and pers. comm) that German children invert in w/z-questions prior to w/z-questions. The same is true for my Bernese data. While inverted patterns of root constituent questions already occur at the age of 2;0, the first inverted patterns of yes/no questions are observed at the age of 2;2-3. Prior

208

ZVI PENNER to this stage, yes/no questions display the tag pattern. Translating this observation into our system, it seems that it is easier for the child to establish a well-formed nc chain for constituent than for yes/no questions. Following the theory developed in Penner & Bader (in prep.), my basic assumption is that the nc in yes/no questions is part of INFL. In other words, in root constituent questions the nc in the 'gap position' and the wh-marked copula form a chain. Conversely, in yes/no questions, the nc is part either of the wh-carrier itself or clause-final INFL-node. This implies that the child has to form two distinct types of nc chains, a fact which accounts for the chronological discrepancy between the two question types. Another consequence of this fact is that in the case of DCCs the dummy copula must not disagree with the clause-final verb with regard to ϕ-features. The observation that the child may fail to distinguish between wh- and yes/no questions is to be attributed to the fact that she assumes wh-marking on the head rather than on an operator in both types of question.

24.

Jürgen Weissenborn points out to me that the question of why null subjects are ruled out in what we have referred to as (early) inverted questions (cf. Weissenborn 1988 and subsequent work on this topic) is not explicitly accounted for by the analysis proposed here. If my approach is on the right track, the restriction on null subjects in inverted questions must be a side effect of the nc-chain. More precisely, the nc-chain with its tail in the gap position should for some reason be incompatible with the nc-chain of the subject. My assumption is that the nc-chain is well formed only if the nc is exposed to discourse government. In configurational terms, the nc must either be placed in the left most position of the clause or be coindexed with the clause-initial element. In our case, the clause-initial element by which the index is mediated is the copula. However, if the intuition that the copula in inverted questions does not carry subject agreement, but articulates the index of the interrogative chain is basically correct, the condition on being accessible to discourse binding is not met.

25.

Penner & Schönenberger (1992) show that the adjunct/specifier distinction is crucial for the correct application of the rule of Verb Projection Raising.

References Calabrese, Andrea. 1987. "Focus Structure in Berber: A comparative analysis with Italian." Lexicon Project Working Papers 14.103-121. Department of Linguistics and Philosophy, Massachusetts Institute of Technology, Cambridge. Chomsky, Noam. 1992. "A Minimalist Program for Linguistic Theory." MIT Occasional Papers 1. Department of Linguistics and Philosophy, Massachusetts Institute of Technology, Cambridge. Clahsen, Harald. 1991. "Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language." Language Acquisition 1.361-391.

ACQUISITION OF ROOT

WH-QUESTIONS

209

Clahsen, Harald & Martina Penke. 1992. "The Acquisition of Agreement Morphology and its Syntactic Consequences: New evidence on German child language from Simone-corpus." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 181-224. Dordrecht: Kluwer. Demuth, Katherine. 1992a. "Accessing Functional Categories in Sesotho: Interaction at the morpho-syntax interface." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 83-108. Dordrecht: Kluwer. . 1992b. "The Acquisition of Sesotho." The Crosslinguistic Study of Language Acquisition, ed. by Dan I. Slobin, 557-638. Hillsdale, N.J.: Lawrence Erlbaum. Evers, Arnold & J. van Kampen. 1992. "An Acquisition Procedure for the I-projection in Dutch." Ms., Utrecht University. Felix, Sascha. 1980. "Cognition and Language Development: A German child's acqui­ sition of question words." Studies in Language Acquisition, ed. by D. Nehls, 91-109. Berlin: Julius Groos. Fritzenschaft, Agnes, Ira Gawlitzek-Maiwald, Rosemary Tracy & S. Winkler. 1990. "Der Erwerb der komplexen Syntax im Kinderalter." Research Report, University of Tübingen. Gawlitzek-Maiwald, Ira, Rosemary Tracy & Agnes Fritzenschaft. 1992. "Language Acquisition and Competing Linguistic Representations: The child as arbiter." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 139-179. Dordrecht: Kluwer. Georgopoulos, Carol. 1991. Syntactic Variables. Dordrecht: Kluwer. Grimm, H. 1973. Struktur-analytische Untersuchung der Kindersprache. Bern: Hans Huber. Haegeman, Liliane. 1990. "Non-overt Subjects in Diary Contexts." Grammar in Progress, ed. by Juán Mascaré & Marina Nespor, 167-174. Dordrecht: Foris. Haverkort, Marco & Jürgen Weissenborn. 1991. "Clitic and Affix Interaction in Early Romance." Paper presented at the 16th Annual Boston University Conference on Language Development, Boston, October. Ingram, David. 1988. First Language Acquisition. Cambridge: Cambridge University Press. Lange S. & K. Larsson. 1973. "Syntactic Development of a Swedish Girl Embla, between 20 and 42 Months of Age." Project Child Language Syntax, University of Stockholm. Lasnik, Howard & Tim Stowell. 1991. "Weakest Crossover." Linguistic Inquiry 22.687-720.

210

ZVI PENNER

Lebeaux, David. 1988. "Language Acquisition and the Form of the Grammar." Diss., University of Massachusetts, Amherst. Lefevre, Claire. 1982. "Qui qui vient? ou Qui vient? Voilà la question." La syntaxe comparée du française standart et populaire: Approches formelle et fonctionelle. Office de langue française, gouvernement du Québec. McCloskey, James. 1979. Transformational Syntax and Model Theoretic Semantics: A case study in Modern Irish. Dordrecht: Reidel. Meisel, Jürgen & Natascha Müller. 1992. "Finiteness and Verb Placement in Early Child Grammars: Evidence from simultaneous acquisition of two first languages: French and German." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 109-138. Dordrecht: Kluwer. Müller, Gereon & Wolfgang Sternefeld. 1990. "Improper movement". Arbeitspapiere 26, Linguistics Department, Konstanz University. Müller, Natascha. 1991. "Komplexe Sätze: Der Erwerb von COMP und Wortstel­ lungsmustern bei bilingualen Kindern." Diss., Hamburg University. . 1992. "Parameters Cannot Be Reset: Evidence from the development of COMP." Ms., Hamburg University. Müller, Natascha & Zvi Penner. 1992. "On the Early Stages in the Acquisition of Finite Subordinate Clauses: The syntax of the so-called preconjunctional embedded clauses." Paper presented at the 17th Annual Boston University Con­ ference on Language Development, Boston, October. Noonan, Maire. 1989. "The [+WH]-CP/IP Parameter: A comparison of Standard French and Québec French." Paper presented at the GLOW Colloquium, Utrecht, March. Ouhalla, Jamal. 1988. "The Syntax of Head Movement." Diss., University of London. Park, Tschang-Zin. 1979. "A Study of German Language Development." Diss., Psychological Institute, University of Berne. Penner, Zvi. 1990a. "On the Acquisition of Verb Placement and Verb Projection Raising in Bernese Swiss German." Spracherwerb und Grammatik. Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie (= Linguistische Berichte, Special Issue 3), ed. by Monika Rothweiler. Opladen, Westdeutscher Verlag. . 1990b. "The Acquisition of the Syntax of Bernese Swiss German: The role of functional elements in restructuring early grammars." Paper presented at the 15th Annual Boston University Conference on Language Development, Boston, October. . 1992a. "The Ban on Parameter Resetting, Default Mechanisms, and the Acquisition of V2 in Bernese Swiss German." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 245-282. Dordrecht: Kluwer. . 1992b. "Possible Domains for Individual Variation in Early developmental Stages." Ms., Berne University.

ACQUISITION OF ROOT

WH-QUESTIONS

211

Penner, Zvi & Thomas Bader. In prep. Issues in the Syntax of Subordination: A comparative study of the complementizer system in Germanic, Romance, and Semitic with special reference to Bernese Swiss German. Penner, Zvi & M. Schönenberger. 1992. "Cross-dialectal Variation in Swiss German: Doubling verbs, verb projection raising, barrierhood, and LF movement." Paper presented at the 7th Workshop on Comparative Germanic Syntax, Stuttgart. Penner, Zvi & U. Wegmüller. 1991. "Constituent Questions in Vicentino." Generative Grammatik im Süden Meeting, Berne. Penner, Zvi, Rosemary Tracy & Jürgen Weissenborn. 1992. "Scrambling in Early Developmental Stages in Standard and Swiss German." Paper presented at the Deutsche Forschungs Gemeinschaft Meeting, Düsseldorf. Poletto, C. 1990. "Subject Clitic/Verb Inversion in North Italian Dialects." Ms., Venice University. Rizzi, Luigi. 1991. "Residual Verb Second and the Wh Criterion." Technical Reports in Formal and Computational Linguistics, 3. Geneva University. . This volume. "Early Null Subjects and Root Null Subjects." Rizzi, Luigi & Ian Roberts. 1989. "Complex Inversion in French." Probus 1.1-30. Roberts, Ian. 1990. "Inversion and Subject Clitic in Valdôtain." Parametric variation in Germanic and Romance (Proceedings from the DYANA Workshop, University of Edinburgh), ed. by Elisabeth Engdahl, Milce Reape, Martin Mellor & Richard Cooper, 155-168. Centre of Cognitive Science, University of Edinburgh. Roeper, Thomas. 1990. "How the Least Effort Concept Applies to the Acquisition of Head Movement, Copying, and Cyclic wh-movement." Ms., University of Massachusetts, Amherst. . 1991. "Parametric Variation in IP and the Shift from Adjunction to C P " Commentary presented at the 14th GLOW Colloquium (Workshop on Wh-movement, ECP, and Bound Variable Development), Leiden, March. . 1992. "From the initial State to V2: Acquisition principles in action." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 333-370. Dordrecht: Kluwer. Roeper, Thomas & Jill de Villiers. 1991. "The Emergence of Bound Variables." Papers on the Acquisition of WH (= University of Massachusetts Occasional Papers, Special Edition.), ed. by T.Maxfield & Bernadette Plunkett, 225-256. Distributed by the Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachusetts, Amherst. . 1992a. "Lexical Links in the Wh-chain." Ms., University of Massachusetts, Amherst and Smith College, Northampton, Mass. . 1992b. "The One Feature Hypothesis." Ms., University of Massachusetts, Amherst and Smith College, Northampton, Mass.

212

ZVI PENNER

Roeper, Thomas & Jürgen Weissenborn. 1990. "How to Make Parameters Work." Language Processing and Language Acquisition, ed. by Lynn Frazier & Jill de Villiers, 147-162. Dordrecht: Kluwer. Rothstein, Susan. 1983. "The Syntactic Forms of Predication." Diss., Massachusetts Institute of Technology, Cambridge. Rothweiler, Monika. 1989. "Nebensatzerwerb im Deutschen." Diss. Tübingen University. Siadhail, M. 1989. Modern Irish. Cambridge: Cambridge University Press. Stenson, N. 1981. Studies in Irish Syntax. Tübingen: Günther Narr. Sternefeld, Wolfgang. 1990. "Beheaded Barriers." Arbeitspapier nr. 14, Linguistics Department, Konstanz University. Stromswold, Karin. 1990. "Learnability and the Acquisition of Auxiliaries." Diss., Massachusetts Institute of Technology, Cambridge. Tracy, Rosemary. 1991a. Sprachliche Entwicklung. Tübingen: Gunter Narr. . 1991b. "Restructuring the Left Periphery: The acquisition of Wh-questions in German." Paper presented at the 14th GLOW Colloquium (Workshop on Wh-movement, ECP, and Bound Variable Development), Leiden, March. Vainikka, Anne. 1992. "Why not Me Sleeping? Case in the child CR" Paper presented at the Second Workshop on the Acquisition of Wh-Movement, University of Massachusetts, Amherst. Valian, Virginia, Ingeborg Lasser & Deborah Mandelbaum. 1992. "(Where) Do Children Invert?" Paper presented at the Second Workshop on the Acquisition of Wh-Movement. University of Massachusetts, Amherst. Verrips, Maaike & Jürgen Weissenborn. 1992. "Routes to Verb Placement in Early German and French: The independence of finiteness and agreement." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language acquisition, ed. by Jürgen Meisel, 283-331. Dordrecht: Kluwer. de Villiers, Jill, Thomas Roeper & Anne Vainikka. 1990. "The Acquisition of Long-Distance Rules." Language Processing and Language Acquisition, ed. by Lynn Frazier & Jill de Villiers, 257-297. Dordrecht: Kluwer. Weinberg, Amy. 1990. "Markedness Versus Maturation: The case of subject-auxiliary inversion." Language Acquisition 1.165-194. Weissenborn, Jürgen. 1988. "Null Subjects in Early Grammars: Implications for parameter setting theories." Paper presented at the 13th Annual Boston University Conference on Language Development, Boston, October. . 1990. "Functional Categories and Verb Movement: The acquisition of German syntax reconsidered." Spracherwerb und Grammatik: Linguistische Unter­ suchungen von Syntax und Morphologie (= Linguistische Berichte, Special Issue 3), ed. by Monika Rothweiler, 190-224. Opladen, Westdeutscher Verlag.

ACQUISITION OF ROOT

WH-QUESTIONS

213

Weissenborn, Jürgen. 1992a. "Null Subjects in Early Grammars: Implications for parameter setting theory." Theoretical Issues in Language Acquisition: Continuity and change in development, ed. by Jürgen Weissenborn, Helen Goodluck, & Thomas Roeper. Hillsdale, N.J.: Erlbaum. Weissenborn, Jürgen. 1992b. "Constraining the Child's Grammar: The development of verb movement in German and French." Paper presented at the Conference on Syntactic Theory and First Language Acquisition: Crosslinguistic Pespective, Cornell University. Weissenborn, Jürgen, Thomas Roeper & Jill de Villiers. 1991. "The Acquisition of Wh-movement in German and French." Papers on the Acquisition of WH (= University of Massachusetts Occasional Papers, Special Edition.), ed. by T. Maxfield & Bernadette Plunkett, 43-79. Distributed by the Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachusetts, Amherst. Weissenborn, Jürgen, Maaike Verrips & Ruth Berman. 1989. "Negation as a Window to the Structure of Early Child Language." Ms., Max Planck Institute, Nijmegen. Weverink, Meike. 1990. "What's Missing in Dutch?." Papers and Reports on Child Language Development 29.125-132. Wode, Henning. 1971. "Some Stages in the Acquisition of Questions by Monolingual Children." Word 27.261-310. Zwart, Jan Wouter. 1992. "Verb Movement and Complementizer Agreement." Ms., Groningen University and Massachusetts Institute of Technology, Cambridge.

Successful Cyclic Movement

Rosalind Thornton MIT

1.

&

Stephen Crain University of Connecticut/ Haskins Laboratories

Introduction

This paper contrasts two competing applications of principles of Universal Grammar to explain children's emerging knowledge of successive cyclic movement. The results of four experiments are presented to help decide between the competing accounts. In this respect, the present paper contributes to the tradition of trying to use both linguistic theory and empirical studies of child language as a basis for understanding children's universal mastery of syntax. One set of claims about the development of successive cyclic movement is our own (Crain 1991; Thornton 1990; Thornton & Crain, forthcoming). Based on a series of experimental investigations of children's (elicited) productions and comprehension of wh-questions, we concluded that children as young as three successfully produce and understand long-distance questions and abide by the constraints that govern them. A few children in our studies were found to meander somewhat in the course of development, however. These children consistently produced non-adult questions with an extra wh-phrase, e.g. What do you think what Cookie Monster likes to eat? We analyzed these exceptional questions as conforming to wh-movement options made available within the theory of Universal Grammar, and expressed in dialects of languages such as Romani and German. These languages generate several question forms, as described and analyzed by McDaniel (1986). One is a 'partial movement' structure involving strictly local movement of a w/z-phrase to the Spec-CP of the embedded clause, with a 'scope marker' inserted in the matrix Spec-CR In addition, there are two long-distance movement structures. One of these is called 'wh-copying' by McDaniel, because a copy of the w/z-phrase is left behind as it

216

ROSALIND THORNTON & STEPHEN CRAIN

moves through the intermediate Spec-CP to the matrix Spec-CP. The other is long-distance movement, as in English, where the intermediate Spec-CP contains no overt material. Following McDaniel, we analyzed some of the exceptional questions produced by children as instances of partial movement, and others as instances of wh-copying. Since every child who produced partial movement structures also produced wh-copying structures, we concluded that successive cyclic movement was a component of the grammars of the children whose productions differed from those of adults. A different set of claims has been advanced by de Villiers, Roeper and Vainikka (Roeper 1990; de Villiers, Roeper & Vainikka 1990). They concluded from a series of comprehension studies that children's non-adult responses to complex two-clause wh-questions are evidence of the late emergence of successive cyclic movement in language acquisition. On their account, the appearance of successive cyclic movement is preceded by two stages. These stages are regarded as ontogenetic reflections of parametric differences observed across languages. The first stage corresponds to languages without (syntactic) movement; that is, children's initial grammars are taken to lack wh-movement (see de Villiers et al.: 282-283, 285, 293). The contention is that children at this stage use coindexation, not movement, to interpret long-distance questions, and that they postulate the empty category pro, not wh-trace, at the site of 'extraction'. (Cf. Cinque 1990; Koster 1986, who postulate pro for other constructions.) A second stage precedes children's mastery of successive cyclic movement on this account. This stage is seen to correspond to languages with local movement, but not long-distance movement. De Villiers et al. invoke the partial movement structures exhibited in Romani and German to explain some children's non-adult responses in their study. Children are seen to take two steps in passing from the initial no-movement stage, to the adult grammar. At the first stage, children can only move a wh-phrase within an embedded clause. This stage is characteristic of the younger children, according to de Villiers et al. By contrast, the older children are at a later point in the transition to the adult grammar. They can move a w/z-phrase within the matrix clause, but not within the embedded clause. Both groups of children at this stage are hypothesized to postulate whtrace, not pro, as the empty category left behind by movement of a w/z-phrase from the extraction site. This brings the grammars of these children more in line with the target grammar, but not totally in line with it. The authors state: "The difference between the child and adult grammars in this system would be that children can only move one w/z-phrase per sentence, while adults can move more

SUCCESSFUL CYCLIC MOVEMENT

217

than one (this would presumably be more of a restriction on the parser than on the grammar [...])" (de Villiers et al 1990:290). Finally, children converge on the adult grammar, and begin to interpret long-distance questions via successive cyclic movement of wh-phrases. It is important to note that the alternative positions are based on children's responses to different linguistic structures. Still, the competing accounts of children's knowledge of successive cyclic movement, or lack of it, cannot both be right. Roeper (1990) argues in favor of the de Villiers et al. account on the grounds that it provides a unitary explanation of both sets of results, whereas our account does not. He proposes that the exceptional questions that some children in our studies produced (with an 'extra' wh-phrase) are the expressive counter­ parts to the interpretive errors made by the youngest children in their studies. According to Roeper, both groups of children are at the local-movement stage of their account. We disagree. In our view, the comprehension errors children made in their studies should not be given the same analysis as the exceptional questions we elicited in our studies. Since parsimony would seem to be against us, we feel obliged to take up this challenge to our conclusion that successive cyclic movement is available in early child grammar. In this paper, we marshal empi­ rical support for our account. We present four new studies of young children's knowledge of successive cyclic movement. The experiments are modeled in part upon those of de Villiers et al. (1990), as well as upon our own earlier work. Crucially all four experiments were conducted with the same group of three- and four-year-old children. The findings from the experiments reported here argue against there being a unitary explanation of the data from previous research. To motivate the experiments, the next section reviews the research and argumenta­ tion behind each account, beginning with our own. 1.1. Evidence in Favor of Successive Cyclic Movement Our proposal rests in large part on the findings from an experiment designed to elicit wh-questions from young children. We used the methodology of elicited production because of the dearth of relevant data in the transcriptions of children's spontaneous productions. Previous studies of elicited production offer abundant evidence that children are capable of producing many seemingly complex sentence structures that occur rarely, if at all, in their spontaneous speech (Crain 1991; Crain & Fodor 1993; Thornton & Crain, forthcoming). This is probably a reflection of their tendency to use alternative structures that also

218

ROSALIND THORNTON & STEPHEN CRAIN

suffice for the communicative purposes at hand. Circumventing the problem of optional forms of expression calls for experimental contexts in which the targeted structure is uniquely felicitous. In the experiment we conducted, we devised situations in which the most appropriate responses by the child were long­ distance questions. We elicited questions involving extraction from tensed embedded clauses and from embedded infinitival clauses. To elicit the target questions, one experimenter acted out short vignettes with toys in front of the child. The experimenter then asked the child to find out from a puppet, 'Ratty' (being manipulated by another experimenter), what he thought had happened, or what he wanted to have happen next. The input to the child consisted of a brief description of the situation, followed by an indirect question; for example, We know there is a marble under the box. Ask Ratty what he thinks or One of these guys gets to take a walk. Ask Ratty who he wants. Crucially, we did not provide the child subjects with overt clues about the structural properties of the complementizer or the embedded clause in long­ distance questions. 1 The main elicitation experiment tested 21 three- and four-year-old children. Roughly two-thirds of the children produced well-formed questions, such as What do you think is in the box?, What do you think Cookie Monster eats?', and Who do you want to take a walk? One-third of the children produced non-adult questions, however. Most of their exceptional questions were ones like (1) and (2), with matching wh-phrases at the front of both the matrix and the embedded clause. As the examples show, the 'medial-wh' appeared both when children were extracting from subject and object position of a tensed embedded clause. (1) (2)

Who do you think who is in the box? Who do you think who Cookie Monster likes?

In addition to questions like these, some of the same children produced ones like (3) and (4). Here, the true wh-phrase appears to have moved only as far as the Spec-CP of the embedded clause, and the wh-word what appears to have been inserted in sentence initial position. (3)

What do you think who ate this?

(4)

What do you think where this froggy lives?

Children's non-adult questions such as (l)-(4) occurred only when they attempted to extract from a tensed embedded clause. No child ever produced a question with an added wh-phrase in extracting from an infinitival clause. That is, children produced correct question forms on those trials, such as Who do you

SUCCESSFUL CYCLIC MOVEMENT

219

want to take a walk?, but never ones like Who do you want who to take a walk? The results of our elicitation experiment clearly demonstrate that some children at age three and four have more movement options available to them than do adult speakers of English. We argued that the different paths these children follow towards the adult grammar correspond to parametric options of Universal Grammar. The medial-wh questions in (1) and (2) parallel certain question forms that appear in dialects of German and Romani. These are called 'wh-copying' questions by McDaniel (1986), who gives (5) as an example from German. (5)

Weri glaubst du wer¡ nach Hause geht? 'Who do you think who goes home?'

McDaniel argues that questions like (5) are instances of long-distance movement. This construction is contrasted with a 'partial movement' option, illustrated in (6), which is also found in dialects of German and Romani. In the partial movement construction, the true wh-phrase moves only short-distance to the embedded Spec-CP, and a scope-marker is generated in the matrix Spec-CP. We maintain that children's questions like (3) and (4) are also instances of partial movement. (6)

Wasi glaubt Hans mit wemi Jakob jetzt spricht? 'What does Hans believe with whom Jakob is now talking?'

We gave two main arguments for analyzing children's medial-wh questions such as (1) and (2) as wh-copying structures, and not as partial movement structures. One argument is based on an observation by McDaniel concerning the extraction of complex wh-phrases in languages with wh-copying. McDaniel notes that it is ungrammatical in these structures to copy a complex wh-phrase in the intermedi­ ate Spec-CP position. This is shown in (7). (7)

*Wessen Buch¡ glaubst du wessen Buch¡ Hans liest? 'Whose book do you think whose book Hans is reading?

We hypothesized that if children's medial-wh questions correspond to wh-copying in German, then children, too, should refrain from putting complex wh-phrases in medial position. This was investigated in a longitudinal study of two children who produced medial wh-questions. Again, short vignettes were acted out in front of the child, who then asked the rat particular questions about what had happened. In one scenario, for example, three Smurfs climbed under a large blanket to take a nap. Two of them wore shoes, but one wore roller skates to bed. The child was instructed to ask the rat which Smurf was wearing the roller skates. By providing a set of objects of the same kind, in this case Smurfs, it was appropriate to give the child the lead-in: "One of the Smurfs is wearing roller

220

ROSALIND THORNTON & STEPHEN CRAIN

skates. Ask Ratty which Smurf he thinks." If children's medial-wh questions correspond to wh-copying in German, we would expect them to use a complex wh-phrase in sentence initial position, but not in medial position. This is exactly what we found. Neither child produced questions like (8), with a complex wh-phrase in medial position. (8) #Which Smurf do you think which Smurf is wearing roller skates? In attempting to extract complex wh-phrases, one of the children produced adult questions without a medial wh-phrase at all. The other child pursued two options. Sometimes she used an agreeing, but reduced medial wh-phrase, as in (9). On other occasions she produced partial movement questions, as in (10). (9) Which Smurf do you think who is wearing roller skates? (10) What did you think which animal says "woof woof' ? We offered a second argument for analyzing medial-wh questions and partial movement questions differently. The argument appealed to a finding from the longitudinal study. It turned out that the two sentence structures followed separate courses in the grammatical development of one child. For this child, medial-wh questions were eventually supplanted by adult long-distance questions while, at the same time, partial movement questions continued to appear. This could only have happened, we reasoned, if medial-wh questions and partial movement questions were distinct grammatical structures in this child's grammar. The parallels between children's questions and the questions of other languages also extends to infinitival clauses. As noted, children never used medial w/z-phrases in extracting from infinitival clauses. In German, too, whcopying and partial movement are not grammatical options in extracting from an infinitival clause.2 A long-distance movement structure must be used, as in (11). (11)

Wen¡ versucht Hans anzurufen ? 'Whom is Hans trying to call?'

(infinitival)

There was another interesting finding from our studies of children's medialwh questions. In longitudinal studies of several children, we found that a medial w/z-phrase initially appeared in both subject and object extraction questions, as in (1) and (2): (1)

Who do you think who is in the box?

(2)

What do you think what babies drink?

After a time, however, the medial w/z-phrase occurred only in subject extraction questions. In extracting from object position, these children began to use adult question forms, without a medial w/z-phrase.

SUCCESSFUL CYCLIC MOVEMENT

221

This ends the summary of our elicitation studies. To interpret the data, we appealed to the Empty Category Principle (ECP) as it is formulated in the Relativized Minimality framework of Rizzi (1990). Adopting this version of the ECP, we were able to explain both the existence of medial-wh questions, and their course of development in child grammar. As in other systems, Rizzi's version of the ECP requires an empty category, such as the trace of whmovement, to be in a position where it is properly governed. In Rizzi's system, however, proper government is satisfied by heads. The legitimate head governors are A, N, P, V, AGR and T(ense). In long-distance questions, a trace in object position of an embedded clause is properly head governed by the verb. An adjunct trace will have either the verb or Tense as its head governor. For traces in subject position, AGR comes to the rescue, as the result of Spec-head agreement. In English, Spec-head agreement transfers the features of AGR onto the null complementizer (0). Having the features of AGR, the null comple­ mentizer becomes an 'agreeing complementizer' which is licensed to headgovern a trace in subject position of an embedded clause. This is illustrated in (12). (12)

Who do you think [cp t [c, 0 A G R [t is in the box]]] Spec Head

In English, only a null complementizer can bear AGR features. If the complementizer is overt, Spec-head agreement is blocked, preventing proper head government of the subject trace. This accounts for 'that-trace' effects in English. In other languages, by contrast, there is Spec-head agreement between a trace and an overt complementizer. A familiar example is the alternation of que/qui in French (See Kayne 1976; Pesetsky 1982, for details). Extraction of a subject is possible in French only if the overt complementizer que changes to qui. This change reflects the process of Spec-head agreement, which transfers the features of AGR to COMP. In both English and French, Spec-head agreement is necessary only for proper head government of subject traces. Object and adjunct traces are properly head governed in other ways, as noted earlier. In Irish, however, the process of Spec-head agreement is quite general, triggering observable changes in the features of an overt complementizer in all cases of extraction from tensed embedded clauses (See McCloskey 1979; Chung & McCloskey 1987). We proposed to analyze children's medial-wh questions using the natural seams of natural language, as just outlined. We observed that the stages of development correspond to alternative instantiations of the ECP, as it is manifested in Irish and French, for example. We suggested that some children

222

ROSALIND THORNTON & STEPHEN CRAIN

initially use the medial-wh as an overt complementizer, reflecting Spec-head agreement, for all cases of extraction. The first stage of development for a child with a medial-wh grammar, then, uses Spec-head agreement across-the-board, as in Irish.3 This is why a medial-wh appears in all of their extractions from tensed embedded clauses. Subsequently, children encounter evidence that initiates grammatical change. One source of input informs them that Spec-head agreement is not obligatory in extracting from the object position of an embedded clause. The evidence is simply adult questions of this kind with no overt complementizer, or with no agreeing complementizer (i.e., with that). Encountering these data, medial-wh questions would be expected to appear only when the child attempted to extract the subject of a tensed embedded clause. This would make their grammars like French, with overt Spec-head agreement for subject extraction, but not for object or argument extraction. Other positive evidence would force medial-wh children to abandon overt Spec-head agreement altogether. The evidence consists of adult questions involving extraction from the subject of an embedded clause, where an overt complementizer is obligatorily absent. If children encounter this kind of input first, then they would not be expected to pass through an intermediate French­ like stage. Instead, they should make an immediate transition to the adult grammar in which the null complementizer carries AGR features. Notice that it should not be possible, however, for a child go through these two possible stages in reverse order. That is, no child should use medial wh-questions only in extracting subjects at an earlier stage than she uses them across-the-board. We conclude this retelling of our account with two further remarks. First we would note that our account makes use of a set of UG parameters, including ones proposed in Rizzi (1990). Another debt we owe to Rizzi is the observation that Spec-head agreement is not utilized in extractions from infinitival complement clauses. As we noted earlier, children's medial wh-questions were limited to extractions from tensed embedded clauses, and never occurred in their productions involving infinitivals. This lends additional weight to Rizzi's theory. For additional theoretical details and empirical justification, see Thornton (1990). 1.2. Evidence against Successive Cyclic Movement Quite different conclusions from ours were drawn by de Villiers et al. They postulate a sequence of stages in the course of the development of long-distance wh-questions. At the first stage, movement is not an option. The second stage is

SUCCESSFUL CYCLIC MOVEMENT

223

divided into two substages, each being characterized by a different form of strictly local movement. At all of these stages, they contend, children's grammars lack successive cyclic movement. The next two sections will consider their arguments and evidence.4 1.2.1. The No Movement Stage De Villiers et al. claim that children's grammars lack the mechanisms of successive cyclic movement until children are three and four, and possibly as old as seven or eight. This conclusion was reached mainly on the basis of findings from two studies of children's comprehension of wh-questions. The first study provides the underpinnings for the conclusion that children's initial stage of development is one at which wh-movement is missing entirely. The authors begin their argument by confronting some potential counter-evidence to this claim; namely, the fact that young children produce wh-questions. The example they cite is (13). ( 13)

What he can drive ?

The authors propose to reinterpret questions such as (13) as having underlying nonmovement structures. This interpretation requires them to make two adjustments to early child grammars. One adjustment requires children, at least initially, to use mechanisms "equivalent to topicalization" in forming wh-questions. At this stage, their questions "could be formed without a complementizer" (de Villiers et al.: 283). The proposal is that children's wh-questions have the same structure that underlies sentences like (14). ( 14)

Trucks he can drive.

The authors remark that "this would provide part of the explanation for why inversion is commonly absent with these structures". The second adjustment to child grammars concerns the empty category following verbs which are missing an argument, e.g. drive in (13). The proposal is that children fill the position after the verb with pro, not wh-trace, as in (15). (15)

Whati he can drive proi

As evidence for the view that children use pro, the authors cite the findings of a study by Roeper, Rooth, Mallis & Akiyama (1985), who found in an experiment with over 100 children that strong crossover was not present, in other words they gave bound variable readings for BOTH:

224

ROSALIND THORNTON & STEPHEN CRAIN (58)

a. Who does he think has a hat? b. Who thinks he has a hat? Roeper et al. proposed that the initial who might be connected via a small pro rather than a trace in (a): (59) Who¡ does he think proi has a hat? in which the bv reading could occur for these subjects. It will be helpful to unpack the argument that children's confusion between crossover questions and bound variable questions could reflect their assignment of pro, rather than wh-trace, as the empty category in wh-questions. The argument turns on the observation that if the empty category is pro, then Principle B, rather than Principle C, would govern the coreference relations between the empty category and any potential antecedents. Accordingly, coreference between the sentence-initial wh-phrase, the pronoun, and the empty category pro would be permitted in crossover questions like (58a), as indicated in (16). Notice that the proi in (16) is free in its governing category. So, coindexation (and hence coreference) between pro{ and the pronoun is allowed. (16)

Who{ does hex think proi has a hat?

This analysis of crossover questions puts them on a par with bound variable questions with respect to coreference possibilities. The crossover question in (16) could be interpreted in exactly the same way as the bound variable question represented in (17). (17)

Whox thinks hex has a hat?

Both the crossover question (16) and the bound variable question (17) would therefore be asking the same question, i.e., to name the set of individuals such that each of them thinks that he has a hat. The associated semantic representation for both questions would be as given in (18).5 (18)

WHO ⋏,(x) [x think x has a hat]

In sum, the No Movement stage at which children are seen to postulate the empty category pro explains their apparent errors in responding to crossover questions. This summarizes the argument that children's grammars initially lack whmovement. Our response begins with the observation that the authors do not cite an example of a natural language that shares the basic properties they attribute to the grammars of young learners. Specifically, the language in question must handle topicalization and wh-questions in the same way, but must not allow

SUCCESSFUL CYCLIC MOVEMENT

225

wh in situ questions. Putting aside the issue of whether or not there is a possible language with these properties, there is reason to doubt that these properties are characteristic of children learning English. The reason is that children learning English never produce wh in situ questions. As the authors note, English-speaking children at this stage produce the same questions as adults. This observation leads immediately to a learnability dilemma on the de Villiers et al. account: there is apparently no evidence to prompt children to reject their erroneous topicalization analysis in favor of one with wh-movement. The same learnability dilemma must be faced in explaining the one difference between the child and adult grammar that is attested in the literature. Recall that de Villiers et al. claim that children misinterpret crossover questions. However, just as it is unlikely that children are informed about their production errors, it seems unlikely that children are informed about semantic disparities between their grammars and those of adults. But without such evidence it is difficult to see how children achieve the adult grammatical system in which crossover questions are interpreted as such. There is another aspect of the de Villiers et al. proposal that does not conform with the natural seams of natural languages. At the local movement stage children are taken to be responding solely to the wh-phrase in the embedded Spec-CP. This is why they call this response the 'wrong Q down' response. It is important to note that this is not the interpretation that is assigned to partial movement structures in languages like German and Romani. In these languages, the wh-phrase which appears in the embedded Spec-CP at S-structure has scope over the entire sentence at the level of semantic representation. That is, a partial movement structure, such as (6), asks the question With whom does Hans believe that Jacob is now talking?, not With whom is Jacob now talking? (6)

Wasi glaubt Hans mit wemi Jakob jetzt spricht? 'What does Hans believe with whom Jakob is now talking?'

Despite the position of the true wh-phrase, the interpretation of a partial movement question crosslinguistically is that of a long-distance movement question. It seems clear that de Villiers et al do not mean this to be the interpretation that children assign who give the 'wrong Q down' response. This difference in semantic interpretation casts doubt on the claim that children's responses are based on a partial movement analysis. We also wish to raise several empirical problems with their account. We note, first, that no child has ever been observed to systematically fail to invert the

226

ROSALIND THORNTON & STEPHEN CRAIN

Subject and Auxiliary in wh-questions, like (13). In fact, Stromswold (1990) calculates on the basis of the wh-questions produced by children on the CHILDES database that they invert 93 percent of the time. A similar conclusion is reached by Pinker (1984). If inversion is the rule, as these researchers maintain, then the proposal that children's w/z-questions utilize the same mechanisms as topicalized sentences faces abundant apparent counterexamples. Another empirical concern is with the crossover findings that are cited in support of the de Villiers et al. account. Elsewhere we consider in detail the evidence that children understand crossover questions and bound variable questions in the same way (Thornton & Crain, forthcoming). As Roeper et al. note, the only truly convincing evidence that children interpret crossover questions as bound variable questions would be the finding that they can assign multiple referents to the pronoun in crossover questions. As (19) shows, if a single referent is assigned to the pronoun, it is possible to give an accidental coreference answer. (19)

Speaker: Who does he think has a hat? Hearer: Why, he thinks HE does.

(20)

Hearer: *Why, he thinks he does and he thinks he does.

The possibility of an accidental coreference interpretation is permitted for a single referent of the pronoun, but not for multiple referents, as (20) indicates. Our objection to their conclusion that children interpret crossover and bound variable questions in the same way hinges on this observation.6 The Roeper et al. study used a comprehension methodology in which the child's task was to indicate the answers to the experimenter's questions by pointing to the relevant cartoon characters in a picture. In the picture corre­ sponding to the question in (19), several characters would have been depicted as thinking that they, themselves, had a hat (i.e., they appeared with a hat in a 'thought bubble' above their heads). If children had responded by pointing to several characters who each thought that they had a hat, then this clearly would constitute evidence that they had interpreted the crossover questions as Bound Variable questions. Inspection of the data indicates, however, that only about 12 of the 100-plus children gave multiple responses to crossover questions. The majority of children who gave coreferential responses indicated a single (as opposed to multiple) referent as the semantic value of the pronoun in crossover questions. Therefore, they could have been assigning an accidental coreference interpretation to the crossover questions. We did not think it enough to raise criticisms of the Roeper et al.

SUCCESSFUL CYCLIC MOVEMENT

227

experiment and let the matter drop there. As a next step, we attempted to implement the substance of our criticisms in an experimental investigation of children's knowledge of the strong crossover constraint. Our study directly confronted children with a multiple referent response to both crossover and bound variable questions. As the examples in (21) illustrate, this was ac­ complished by providing answers to embedded questions. (21) a.

I know who he thinks has the best hat.

b.

/ know who thinks he has the best hat.

Grover and Big Bird Grover and Big Bird In response to question-answer pairs such as these, children correctly rejected the multiple response interpretation of the crossover sentences, such as (21a), 86 percent of the time. By contrast, they accepted the multiple response interpreta­ tion of the bound variable sentences, such as (21b), 91 percent of the time. As a final comment, it is worth noting that the children in our study were considerably younger than those children in the Roeper et al. study, where there were some who apparently misunderstood crossover questions. The children in our study ranged in age from 3;7 to 4;8 with an average age of 4;2. The children who erred in the Roeper et al. study were 6- to 8-years-old. This brings out a final criticism we would make of the claim that children initially lack w/z-movement. The main evidence cited in favor of a stage of No Movement is based on findings from children who are much older than those claimed to be at more advanced stages. 1.2.2.

The Local Movement Stage

We now turn to the evidence for the more advanced stages, at which young children are seen to have only local movement. The stages of local movement are hypothesized in order to explain children's apparent misunderstanding of questions with two w/z-phrases, as in (22). (22) a. b. c. d.

Who did the girl ask what to throw? Who did Big Bird ask how to paint? How did Kermit ask who to paint? When did the boy know how he hurt himself?

ARG + ARG ARG + ADJ ADJ + ARG ADJ + ADJ

The notation to the right of each question indicates the type of extraction in the questions, Argument or Adjunct, of the matrix and embedded S, in that order. Two further properties of these questions are worth noting. First, note that none

228

ROSALIND THORNTON & STEPHEN CRAIN

of the questions that children misunderstood involve successive cyclic movement. That is, the sentence-initial w/z-phrase does not move successive cyclically through the intermediate Spec-CP on its way to the matrix Spec-CP position. Instead, these questions exhibit two instances of local w/z-movement (at least on the interpretation that is highly preferred for adults). A second property of note is that the questions children misinterpreted contain both infinitival complements (as in (22a-c)) and tensed complements (as in (22d)). We return to the import of this observation shortly. The experiment involved 25 children between the ages of 3;7 and 6;11. The task was to answer questions about a sequence of pictures. The pictures accompanied a short story read by the experimenter. At the end of each story, the child was asked specific questions. One of the main findings was that children gave incorrect responses to the questions in (22). Consider (22c): (22) c.

How did Kermit ask who to paint?

ADJ + ARG

This type of question has an adjunct in sentence initial position, and an argument in the intermediate Spec-CP position. Many of the three- and four-year-old children failed to answer the /zöw-question and, instead, responded to the lower who. These children responded in the same way to questions with either a tensed or an infinitival embedded clause, as we mentioned earlier. De Villiers et al. interpret children's incorrect responses as indicating that they can only move "one wh-phrase per sentence". In other words, they claim that successive cyclic movement is not yet available to these children. Assuming that children can only move one wh-phrase per sentence, the authors propose that in questions like (22c), children interpret 'how' as a scope marker for the true wh-phrase, 'who': (23)

How did Kermit ask who to paint t? I I scope true marker wh-phrase

The older children at this stage give another type of incorrect response. They interpret question (22c) to be about who Kermit asked, not who he asked to paint. These 'wrong Q upstairs' answers are analyzed as representing a second stage of strictly local movement at which children can move wh-phrases only within the matrix clause. Notice, however, that it is the wh-phrase of the lower clause that undergoes movement within the higher clause. This type of error will be ignored for the remainder of the paper, for two reasons. First, responses of this type never occurred in our replication of the de Villiers et al. experiment,

SUCCESSFUL CYCLIC MOVEMENT

229

which is reported in the next section of the paper. But more importantly, this error by older children is not at issue in assessing Roeper's contention that there is a common cause underlying both the comprehension errors that children made in their study and the exceptional questions children produced in our study. Rather, the common cause is supposed to explain children's 'wrong Q down' responses, which they analyze as in (23). Roeper (1990) proposed to extend the analysis given in (23) to the non-adult questions that we elicited from some children, as illustrated in (1) and (2). Recall that these exceptional questions had matching wh-phrases at the front of the matrix clause and the tensed embedded clause. (1)

Who do you think who is in the box?

(2)

What do you think what babies drink?

The proposal is that the 'medial-wh' is the true wh-phrase, and that the initial wh-word is a scope marker. In this way, Roeper attempts to bring both sets of findings under the same roof, as reflecting the first stage of strictly local movement. In the next section we present a series of experiments that were undertaken specifically to evaluate this claim. Before turning to the experiments, however, we would like to summarize the case we have made so far against a unitary explanation of the structures we elicited and the questions they examined. As noted earlier, the two studies offer disparate results. One difference was in how children responded to questions involving extraction from infinitival clauses. Recall that in our study children never produced an extra wh-phrase in the embedded Spec-CP when extraction was from an infinitival clause. In the de Villiers et al. study, on the other hand, the 'wrong Q down' response occurred in both tensed and infinitival contexts. The fact that children in the de Villiers et al study treated extraction from tensed and infinitivals in the same way is problematic, it seems to us. We know of no languages in which partial movement questions occur with extractions from infinitival clauses. If such languages would stand in violation of principles of UG, then this would seriously undermine their analysis of children's responses to questions like (22a-c) as reflecting partial movement.7 On learnability grounds, we challenged their proposal that children learning English pass through stages of no movement and strictly local movement. Since the authors do not confront the issue of learnability from positive data, we simply wish to raise this point as a potential problem with their analysis. To end our critique, we wish to point out an inconsistency in the data from

230

ROSALIND THORNTON & STEPHEN CRAIN

the de Villiers et al. study. As the authors note, the partial movement analysis they give for ADJ + ARG questions (i.e., ones with an adjunct in sentence initial position and an argument in the intermediate Spec-CP) logically extends to other question types with 2 wh-phrases that they tested, including the following combinations: Argument-Argument (ARG + ARG) and Adjunct-Adjunct (ADJ + ADJ). As they, themselves, remark: For both sentence types ARG + ARG and ADJ + ADJ the younger children move the element during Cycle 1 only, resulting in the medial-wh phrase being answered with respect to the downstairs verb (wrong Q down), just as with the sentence type ADJ + ARG. Inspection of the data reveals, however, that the 'wrong Q down' response does not occur with equal regularity for all three question types. Specifically, de Villiers et al. report 31 downstairs responses out of a possible 99 for the ADJ + ARG combination. By contrast, children gave only 5 downstairs responses to the ADJ + ADJ combination. And not a single response of this kind occurred with the ARG + ARG combination. The data are given below (extracted from their Table 5). Question Type ' wrong Q down' (across children) ADJ + ARG 31/99 ADJ + ADJ 5/48 ARG + ARG 0/50 Recall that the authors' basic claim is that younger children must interpret all of these question forms in the same way, because there is only one grammatical structure available to them. Hence, the children who account for the 31 percent error rate for ADJ + ARG questions should have responded incorrectly at a similar rate to all relevant question types. This concludes our critique of the acquisition scenario advanced by de Villiers et al. The next section reports the findings of our recent experimental attempts to adjudicate between the competing accounts.

2.

The Experiments: Testing Between the Competing Accounts

This section presents four experiments designed to investigate the emergence of successive cyclic movement. The experiments were modeled after our own earlier work as well as that of de Villiers et al. (1990). Crucially, they were conducted

SUCCESSFUL CYCLIC MOVEMENT

231

on the same fifteen three- and four-year-old children. We briefly review the motivation behind the experimental designs before turning to the details. The first experiment was an elicitation study designed to elicit long-distance extraction of adjuncts. In current theory, adjunct extraction is uncontroversially assumed to involve successive cyclic movement of the wh-phrase (e.g. Rizzi 1990; Cinque 1990). Therefore, children's correct productions of these questions would support our claim that they command successive cyclic movement. The second experiment comprised 2 tasks. One task investigated children's compre­ hension of the ADJ + ARG structure that had evoked the highest number of misinterpretations in the de Villiers et al study. Our task was a modified version of their task. In addition, a range of long-distance questions was elicited from children. In this way, Experiment 2 was designed to test their prediction that those children who give the 'wrong Q down' response also produce medial-wh questions. This outcome would be expected if partial movement was the only structure available to these children. The third experiment also tested long­ distance questions, this time using a comprehension methodology. This experiment was designed to determine if those children who give downstairs answers to questions with 2 wh-phrases can nevertheless correctly interpret long­ distance questions. The final experiment was designed to further investigate the source of children's mistaken answers. We wanted to explore the possibility that pragmatic difficulties, rather than lack of syntactic knowledge, were responsible for children's misconstrual of ADJ + ARG questions like How did he ask who to paint? New contexts were contrived to make how-questions more salient. If children produced fewer erroneous responses in these situations, then this would argue that their errors in earlier experiments could have been due in part to their inability to come up with the adult response in situations that are less felicitous for the questions at hand. 2.1. Experiment 1 : Elicitation of Adjunct Extraction Questions In this experiment, we attempted to elicit wh-questions involving extraction from adjunct position. In these questions, the wh-phrase is moved from its basegenerated adjunct position successive cyclically through the intermediate Spec-CP position and then to the matrix Spec-CR Consider the adjunct extraction question in (24). (24)

[ cp How [c do [IP you think [CP t [IP the Smurf went to town t ]]]]]

232

ROSALIND THORNTON & STEPHEN CRAIN

The movement of the wh-phrase how from the base-generated adjunct position to the intermediate Spec-CP to the matrix Spec-CP yields a well-formed chain that preserves government relations. In this structure, the trace at the tail of the chain (the adjunct's position of origin) is antecedent-governed by the intermedi­ ate trace, and this in turn, is antecedent-governed by the wh-phrase how. If there was no intermediate trace, then a legitimate chain would not be formed and the derivation would be ruled out (Rizzi 1990). To our knowledge, theoreticians are in general agreement that adjunct extraction involves a derivation by successive cyclic movement. Thus, children's production of adjunct extraction questions would constitute a convincing demonstration of their knowledge of successive cyclic movement. Questions of this kind would be difficult to explain on the de Villiers et al. account.8 The adjunct wh-phrases we attempted to elicit were how and what way. We included complex wh-phrases such as what way to be sure that there would be no doubt that the wh-phrase is in Spec-CP position, the position it must occupy to derive a well-formed chain. 'Complex' wh-phrases such as what way are maximal projections and therefore must be in the specifier position. They cannot appear in COMP (which is available only for movement of head categories) without violating X'-theory (Chomsky 1986). Questions involving extraction from argument positions were also elicited. This was to see whether there were children who could produce questions involving argument extraction but not questions involving adjunct extraction. These questions were also needed for Experiment 2. 2.1.1. Method 2.1.1.1. Subjects The subjects were 15 English-speaking children between the ages of 3;0 and 4;8 (mean age = 3;ll). The children were all enrolled in daycare centers in Storrs, Connecticut or Arlington, Massachusetts. 2.1.1.2. Materials This experiment elicited long-distance questions using the technique reported in Crain & Thornton (1991), Thornton (1990) and Thornton & Crain (forthcoming). On a typical trial, the child witnessed one experimenter act-out a short vignette using toys and small props. While each vignette was being acted out, the rat puppet, played by a second experimenter, covered his eyes so that he could not see what happened. At the close of the vignette, the rat uncovered his

SUCCESSFUL CYCLIC MOVEMENT

233

eyes, and was asked to make a guess about something that had happened in the enacted scenario. For example, in the sample protocol given below, the rat was asked how he thought the Smurf got to the donut store. (25)

Protocol for Elicitation of Adjunct Questions

Experimenter:

Child: Experimenter:

Experimenter: Child: Experimenter:

Child: Rat: Child:

In this story, there's a Smurf who's very hungry. "Oh, I wish I could have a donut to eat," he says. "I think I'll go to the store to buy one. I want to get to the store really fast. Let's see, shall I walk, or shall I go on my bike, or shall I go on my skateboard. (To child:) Do you have an idea?" The skateboard! (As the Smurf) "Good idea! My skateboard is very fast. Here I go!" And the Smurf buys a big donut. "Yum, yum, yum," the Smurf says, as he eats his donut. (To child) Tell the rat he can come out now. You can come out now! Are you ready, Ratty? (To child, in a low voice so the rat can't hear) We know that the Smurf went to the donut store on his skateboard, right? Ask the rat what way he thinks. What way do you think the Smurf went to the donut store? Urn, did he go on his bike? No, he went on the skateboard...

As noted by one reviewer, the lead-in designed to evoke long-distance questions from the child consists of an indirect question which takes an embedded clause. That is, it is a long-distance question in which the embedded clause is elided. In our view, a child who knows the grammar should be able to reconstruct the elided clause so that they can ask the rat the appropriate question. By contrast, a child who does not have the structure for producing long-distance questions will not be able to make sense of the ellipsis in the lead-in. The same reviewer also observes that the lead-in sounds somewhat unnatural. It improves sig­ nificantly, however, if the pronoun is stressed, and this was how the lead-in was presented to the child (as can be seen from the underlining on the pronoun in the protocol in (25). We attempted to elicit four adjunct extraction questions from each child, two with how, and two with what way. The target questions were variations on the questions listed in (26).

234

ROSALIND THORNTON & STEPHEN CRAIN

(26) Target Adjunct Extraction Questions a. How/what way do you think he got to the store? (on his skateboard or bike) b. How/ what way do you think he put out the fire? (with the fire extinguisher or baseball bat) c. How/ what way do you think he hides? (disguised in Barbie s bathing suit or in a mask) d. How/ what way do you think he tried to fix the truck? (by putting the wheels back on or hitting it) The stimulus materials for the argument extraction questions are given in the materials section for Experiment 2. 2.1.1.3. Procedures In this experiment, and those that follow, children were tested individually in a research room near their classroom. All sessions were taperecorded for later transcription and data analysis. Long-distance questions involving adjunct extraction and ones involving argument extraction were elicited in a single session. The results concerning argument extraction questions will be presented in Experiment 2. Those children who proved to be facile with long-distance questions had 4 trials with adjunct questions, and sometimes more, if they wanted to continue playing the game. If a child was having difficulty coming up with the structure, however, we stopped the game before completing all of the trials, rather than running the risk of frustrating the child. 2.1.2. Results The results from the experiment are summarized in Table 1 overleaf. As can be seen in the table, seven of the 15 children, mostly the four-year-olds, successfully produced long-distance adjunct extraction questions. This is indi­ cated by a check in the column marked 'Adj' in the table. All these children also produced argument extraction questions. All children who produced adjunct extraction questions used how or what way as the questioned wh-phrase on at least 3 of the 4 trials. The number of adjunct extraction questions produced by each child per total trials is given in parentheses in the 'Adj' column. Each story counted as one trial. Two sample questions from each of the children who success­ fully produced questions with successive cyclic movement are given in (27). On the trials in which children did not produce a long-distance adjunct extraction question, they usually produced a matrix question or a yeslno question instead.

SUCCESSFUL CYCLIC MOVEMENT

235

Table 1. Production of Argument and Adjunct Questions by Individual Subjects

(27)

Kelly(4;8) How do you think the baby went to the store? What way do you think how did he put out the fire? Leah(4;7) How do you think he hid, in Barbies bathing suit or the mask? What way do you think she went to the donut store, this or this? Noah 1(4;6) Why do you think she fell off the bike, 'aaaK like that? How do you think he can see with these eyeglasses, this mask? Alex(4;4) How do you think the man tried to fix the truck? How do you think the Smurf got to the peanut store? Dano(4;3) What way do you think he hides? How do you think this girl went on, to the donut store to eat this donut? Aisha(4;l) What way do you think he went to the cookie store? How do you think that the girl went to the b, urn the donut store? Bryan(3;10) How do you think the Smurf got to the peanut store? What way do you think the fireman put out the fire?

236

ROSALIND THORNTON & STEPHEN CRAIN

We were less successful in eliciting questions from the 3-year-old group. Five of eight children in this group had an ask/tell problem. That is, when these children were requested to ask the puppet a question, they told the puppet an answer to the question instead. We have found that, with time, we can usually overcome this problem, but since we attempted only one session with each child, we are left with incomplete data from these five children. In addition, two of the 3-year-olds (Daniel and Tanya) were too shy to produce questions in the elicitation part of the experiment, although they participated in the sessions using comprehension methodology. These results lead us to conclude that, at least by age four, children successfully demonstrate competence with successive cyclic movement. We cannot infer that the 3-year-olds lack the ability to form long-distance questions, but, likewise, we cannot infer that they have competence on the basis of the present experiment. The possibility remains, then, that three-year-old children lack successive cyclic movement. Therefore, additional experimental maneuvers were implemented.

2.2. Experiment 2. Comparison of Comprehension and Production Data This experiment tested the prediction in de Villiers et al. that children who give 'downstairs' responses in comprehension experiments testing ADJ + ARG questions should be the same children who produce questions with a medial-wh in experiments eliciting long-distance questions. This follows, on their analysis, because children's 'errors' on both tasks are supposed to stem from the same source. They occur in both cases because children generate partial movement representations instead of the appropriate adult structures. (The adult structure for the ADJ + ARG question would involve moving each wh-phrase locally, and long-distance questions would involve successive cyclic movement of one whphrase). To examine this prediction, we tested a group of children on both tasks. Children were given a version of the comprehension task used by de Villiers et ai, and the task used for elicitation of long-distance questions described in Experiment 1, and reported in Crain & Thornton (1991) and Thornton (1990). 2.2.1. Method 22.1.1. Subjects The same 15 children (3;0-4;8, mean age = 3; 11) who participated in Experiment 1 also participated in this experiment.

SUCCESSFUL CYCLIC MOVEMENT

237

2.2.1.2. Materials Different materials were used in the comprehension part of the experiment and the elicitation part. We begin with the materials from the comprehension task. This task focused on the questions that had produced the majority of 'wrong Q down' responses in the de Villiers etal. comprehension study. These questions were the ADJ + ARG combination, such as in (28): (28) a.

How could Bert see what Gummy Bear is eating?

The comprehension task differed from the de Villiers et al. task in several respects. The de Villiers et al. task had the child listen to a story whose events were depicted with pictures. In our experiment, instead of illustrating events with pictures, we acted out stories with toy characters. Another difference was that the question was not posed directly to the child. Instead, a puppet, Kermit the Frog, was required to answer the test question for each story. The question was always too hard for the puppet, who deferred to the child. The purpose of this experimental manipulation was to reduce the child's feeling of being tested. We assumed that these differences would not influence the basic pattern of results obtained in our task as compared to the de Villiers et al. study. We chose to use two different verbs in the matrix clause of the target sentences, find out, and see. These verbs were used instead of ask, which was used by de Villiers etal., because sentences with ask introduce a potential garden path that we wished to avoid. In interpreting a question like How did Kermit ask who to paint?, for example, it seems possible that children could use the whphrase who on-line to fill an extra NP position associated with ask but not with the verbs find out or see. The extra filler is overt in sentences like How did Kermit ask Grover who to paint?. A sample protocol from one of the stories acted-out with toys is given in (29). (29) Experimenter:

Protocol for Comprehension of ADJ + ARG Questions In this story, Gummy Bear is very hungry, and he says "Mm, let me see, shall I eat the pizza or this hamburger? The pizza looks good, I think I' 11 eat this. Yum yum." Then, way over here is Bert. And he says, "It's lunch time and I'm hungry for something good to eat. Is that Gummy Bear over there? He's too far away for me to see what he's eating. Maybe if I look through these glasses, I' 11 be able to see over there. Oh, these glasses don't help. They must be for the beach. Then, I'll try this telescope. Oh yes, I see, Gummy Bear is eating a pizza, not the hamburger."

238

ROSALIND THORNTON & STEPHEN CRAIN

(29) Continued Kermit:

Experimenter: Kermit: Child:

OK, there s Gummy Bear and some pizza, a hamburger and Bert and some glasses and a telescope. OK, what do you want to know? How could Bert see what Gummy Bear was eating? That's a tough one. How could Bert see what Gummy Bear was eating? With the telescope (upstairs) OR The pizza (downstairs)

There are several points to note about the protocol. In each story presented to the child, we attempted to make the questions represented by the upper and lower wh-phrases equally salient. To do this, objects appropriate for answering either question were present in the workspace. In the story in (29), for example, the toy figure Bert was in possession of a telescope and some glasses, and was separated some distance from Gummy Bear, who had a pizza and a hamburger in front of him. Bert wanted to see what Gummy Bear was eating. He attempts to use the glasses first, but could only find out by using the telescope, which he tries next. The four target sentences used in our comprehension task are given in (30). Below each target sentence, the choices for upstairs and downstairs responses are given. (30) a. b. c. d.

How could Bert see what Gummy Bear was eating? (with glasses/telescope pizza/hamburger) How could the Ghostbuster see who the troll was talking to? (by chair/ladder ghost/baby) How did Wonderwoman find out who the witch was scolding? (with walkie talkie/loudspeaker pig/horse) How did Spidermanfind out what the Incredible Hulk stepped on ? (by radio/messenger bug hook/tomato)

The elicited production portion of the experiment investigated the form of children's questions produced in situations where an adult would produce a long­ distance question. In particular, we were interested in eliciting questions from children involving subject extraction, because extraction from subject position had proven most successful in evoking medial-wh questions from children (see Thornton 1990 for details). By comparing children's subject extraction questions with their performance on the comprehension task, we had the greatest chance of correlating the two results and thus confirming or disconfirming the de Villiers et al. prediction that children who asked medial wh-questions would also give 'wrong Q down' responses in the comprehension task.

SUCCESSFUL CYCLIC MOVEMENT

239

The wh-questions involving subject extraction were elicited in a 'guessing game' in which the child asks a puppet a question. The situations were designed to be appropriate to production of a long-distance question rather than a simple matrix question. A typical protocol for subject extraction is given in (31). (31) Protocol for Subject Extraction Questions Experimenter:

Child: OR:

Lef s play a guessing game with the rat (puppet) ... Let's hide a strawberry in this box, and the teddy bear under the can. Lef s do the box first. We know there s a strawberry in the box, right? Ask the rat what he_ thinks. What do you think is in the box? What do you think what is in the box?

2.2.1.3. Procedures The comprehension part of the experiment was tested in a separate session from the elicitation experiment. The two sessions were about one week apart, with the comprehension session preceding the elicitation portion of the experiment. 2.2.2. Results The results of both tasks used in this experiment are summarized in Table 2 overleaf. The table shows that many children produced the 'wrong Q down' response reported by de Villiers et al. In fact, the 'wrong Q down' response was highly favored among the three-year-old group of children. Notice, however, that the 'wrong Q down' response did not correlate with production of medial-wh questions in the elicitation experiment. In fact, there was a double dissociation between the 'wrong Q down' response and the use of medial-wh questions. Bryan and Anne both gave the 'wrong Q down' answer three out of four times in the comprehension experiment, but both children produced adult long-distance questions. The opposite trend appeared in Noah 1 's data. Noah 1 gave no 'wrong Q down' answers, correctly answering the sentence-initial adjunct question in each case. In the elicitation experiment, however, he used a medial-wh in all of his subject extraction questions. The double dissociation is illustrated in Table 3 overleaf.9

240

ROSALIND THORNTON & STEPHEN CRAIN

Table 2. Comparison of Downstairs Responses versus Production of Medial-wh Responses subject

age

Kelly Leah Noah 1 Anne Alex 1 Dano Aisha Bryan Aaron Daniel Noah 2 Emma Alex 2 Tanya Rachel

4;8 4;7 4;6 4;4 4;4 4;3 4;1 3; 11 3;8 3;5 3;4 3;4 3;4 3;3 3;0

downstairs responses

medial wh-questions

2/4 0/4 0/4 3/4 1/4 1/4 4/4 3/4 1/4 4/4 4/4 3/4 4/4 4/4 4/4

no no yes no no no yes no no — — — — — —

Table 3. Double Dissociation of Downstairs Answers versus Medial-wh child Bryan, Ann Noah 1

downstairs error + -

medial-wh

question +

Some examples from the sessions with Bryan (3; 11) and Noah 1. (4;6) on both tasks are given in (32) and (33). (32)

Bryan 3; 11 a. 'Wrong Q Down' Error Kermit: Bryan: b.

How did the Ghostbuster see who the troll was talking to? The baby

Adult Questions in Elicitation Task (sub/obj/adj) What do you think's in this box? What do you think babies drink? Kermit, how do you think the Smurf got to the peanut store?

SUCCESSFUL CYCLIC MOVEMENT

241

(33) Noah 1 4;6 a.

No 'Wrong Q Down Error

Kermit: How did Wonderwoman find out who the wicked witch was scolding? Well, gee, I dont know. Do you know? Noah: From a loudspeaker! Kermit: Oh, wow! b.

Medial-wh Questions (subjects only) What do you think what's under the box? What do you think babies drink? What way do you think the Smurf got to the donut store fast?

To conclude, the results from the experiment do not support the de Villiers et al. prediction that the 'wrong Q down' in comprehension goes hand-in-hand with the medial-wh in production. To the contrary, we found a double dissociation. In short, children who give the 'wrong Q down' responses are not necessarily the ones who ask medial-wh questions, and children who produce medial-wh questions do not necessarily give the downstairs response. It is, of course, logically possible that the two errors may surface in the same child, but in view of the above results, this would be a coincidence, and not because the two errors have the same grammatical source. What is the source of the two errors? De Villiers, Roeper and Vainikka proposed that the two errors might arise because, at this point in their gram­ matical development, children are only able to produce partial movement structures and not questions involving successive cyclic movement (or two instances of local movement). Let us suppose that de Villiers et al. are correct in claiming that the wrong Q down response reflects the generation of a partial movement structure. Given the double dissociation in the errors noted above, the logical conclusion would be that if the 'wrong Q down' response reflects partial movement, then the medial-wh structure is not a partial movement structure. In fact, this is our claim. We have argued that the questions with a medial-wh involve long-distance movement of the wh-phrase. If the 'wrong Q down' answer does not reflect generation of a partial movement wh-question, however, it remains possible that it occurs for reasons unrelated to the child's ability to produce structures involving successive cyclic movement. We explore this possibility in our final experiment.

242

ROSALIND THORNTON & STEPHEN CRAIN

2.3. Experiment 3. Interpretation of Long-Distance Questions As discussed earlier, de Villiers et al. discovered that children respond with the 'wrong Q down' answer to ADJ + ARG questions. On their analysis, what children lack is the ability to conjure up a structure in which two wh-phrases undergo local movement, or two successive applications of movement of any kind. They go on to suggest that children misanalyze the structure of ADJ+ARG questions, such that they respond only to the wh-phrase in the embedded SpecCP. By the same reasoning, we should expect children at this stage to provide the 'wrong Q down' response to ARG + 0 questions. Although these questions involve successive cyclic movement in the adult grammar, as we just saw this grammatical option is supposedly unavailable to young children. Consider the ARG + 0 question in (34). (34) What did he say was in the box? In this case, we need to assume that there is a null wh-phrase in the embedded Spec-CP. On this assumption, children should respond to the question by answering what actually was in the box, not what someone said was in the box. The experiment described in this section investigates this prediction. Before describing the experiment, it will be useful to recall a point made earlier about partial movement structures in languages like German and Romani. In these languages, although the true wh-phrase appears in the embedded Spec-CP, it maintains scope over entire sentence. Returning to (34), suppose children interpret this as if it were a partial movement structure, as in (35): (35) What did he say 0 was in the box

↑↑ (scope marker) (true wh-phrase) In languages like German and Romani this would be asking what someone said was in the box, not what was in the box. That is, the interpretation is that of a long-distance question. By contrast, children who give a 'wrong Q down' response to such questions would be giving them a different interpretation, as if they asking what was in the box. Therefore, if children answer the 'wrong Q down' to the ARG + 0 questions in the present study, we should not conclude at they are generating partial movement structures. On the other hand, if children respond as adults do to (34), we will not be able to tell whether they are generating a partial movement structure, or a long-distance movement structure, since the interpretation of the two structures is the same.

SUCCESSFUL CYCLIC MOVEMENT

243

23.1. Method 2.3.1.1. Subjects The children who participated in this experiment were the nine youngest children from the group (N= 15) who had participated in Experiments 1 and 2. These children ranged in age from 3;0 to 4;1 years of age. 2.3.1.2. Materials To test whether children invariantly answer the 'wrong Q down', or whether they have the adult interpretation for ARG + 0 questions, children were engaged in another guessing game with Kermit the Frog. Two objects were laid out for the child and Kermit to see. Kermit then was asked to cover his eyes. The child was invited to choose one of the two objects to be hidden in a box. Kermit then guessed which one had been hidden. Kermit always made the wrong guess, and was corrected by the child. A sample protocol for the game is given in (36). (36) Experimenter: Kermit: Experimenter: Child: Experimenter: Child: OR

Protocol for Testing Long-Distance Interpretation (To child) Shall we hide the baby or the bear? Bear? OK. (To puppet) You can come out now. (To child) I think you probably hid the baby under there. What did he say's under there? The baby! What is under there? A bear! The baby!

The trials for this experiment were tested in one short session. Each child had three trials. The procedures were the same as in the previous experiments. 2.3.2. Results The results were clear-cut. The nine children had 3 trials each, giving a total of 27 opportunities for them to err. Not a single error was made, however. The finding of this study, then, was that long-distance movement interpretations of ARG + 0 questions were available even to children who consistently gave downstairs answers in response to ADJ + ARG questions. What we cannot tell is whether children generate a derivation involving successive cyclic movement, as in a long-distance movement question, or whether they generate a partial movement question, that inserts a scope marker in the matrix Spec-CR

244

ROSALIND THORNTON & STEPHEN CRAIN

Up to this point, we have shown that four-year-olds, at least, demonstrate knowledge of successive cyclic movement in their productions. In addition, we have seen that children do not treat long-distance movement structures and ADJ +ARG questions (which are instances of local movement) in the same way. The remaining question is: Do children who consistently give the downstairs ADJ + ARG response actually lack the grammatical analysis for this structure, or is the correct structure available to them, but difficult to access for some reason? The final experiment was designed to address this question. 2.4. Experiment 4. An Experiment in the Manipulation of Pragmatics The final study manipulated the pragmatic context associated with ADJ + ARG questions like How did he see what Gummy Bear was eating? Our goal was to determine whether children's erroneous responses to the questions could be influenced by contextual factors. In particular, we sought to provide a supporting context for the correct responses; that is, we sought to enhance the salience of the upper clause question. By making the correct, 'upstairs' answer more accessible, we hoped to eliminate or at least greatly reduce children's proclivity to give 'downstairs' answers. Of course, if children's grammars precluded the correct answer, then changing the context could have little effect. 2 4.1. Method 2.4.1.1. Subjects We selected the nine children who had given two or more 'wrong Q downstairs' answers out of four trials in Experiment 2. We also included one more child, Katherine (4; 11) who participated in the comprehension sessions, but missed the elicitation due to absence from school on that day. The ten children covered a wide span in age — from 3;0 to 4;11. 2.4.1.2. Materials The same comprehension methodology was used as in Experiment 2. Scenarios were acted out with toys and a puppet, Kermit the Frog, was asked a question about what had happened in the story. Once again, the questions proved too hard for the puppet, so he asked the child to help supply some of the answers for him. Experiment 2 tested questions of the ADJ + ARG type. There, we en­ deavored, like de Villiers et ai, to make both the 'upstairs' and 'downstairs' answers equally accessible. However, it could be that in those contexts the upstairs answer was not entirely felicitous. In the contexts devised for Experi-

SUCCESSFUL CYCLIC MOVEMENT

245

ment 4, then, we attempted to focus the situation more on the correct, 'upstairs' answers. As before, though, children were free to answer the 'downstairs' wh-phrase, if that was the only interpretation that their grammars made available. The stories were about two contestants, each of whom used a different method to achieve her goal. Only one of them was successful, however. Kermit first asked the child who won the contest. This question was of the form ARG+ARG. Kermit's second question was how the winner had won. This was the question of most interest to us, being of type ADJ + ARG. Finally, Kermit asked the child about the goal of the contest, a question of type ARG + 0 . Therefore, the child had three opportunities to give 'wrong Q down' answers on each trial, although we knew already that this was only a likely response for the ADJ + ARG question. Our hope was that by focusing on the way each of the competitors sought to achieve her goal, children would correctly answer questions of this type. A sample protocol is given in (37). (37) Experimenter:

Child: Experimenter: Child: Experimenter:

Kermit: Child: Kermit: Child: Kermit: Child:

Protocol with Focus for ADJ + ARG Questions In this story, this bird has an egg, and these two guys, Robocop and Snuffy have to try tod climb this wall to find out who the egg belongs to. (To child) Do you think Snuffy should use the straw or the ladder? The straw OK, so Robocop should use the ladder? Which guy do you think will win? Robocop OK, so Snuffy says "Let me try to find out who the egg belongs to. My straw is very long, so I think I can climb the wall. Here I go! I'm getting high. Oh no, I must be too heavy, my straw is bending. Oh, I fell down — this is no good." Then Robocop says "Let me try to find out. I bet with my ladder I can climb the wall and see who the egg belongs to. Here I go! One step, two steps, three steps, yes, I got to the top. I found out who the egg belongs to — the bird." Which guy found out who the egg belongs to? Robocop What way did he find out who the egg belongs to? With the ladder Who did he find out the egg belongs to? The bird

246

ROSALIND THORNTON & STEPHEN CRAIN

There were four trials, each consisting of 3 questions as in the protocol above, and in (38a), below. The 4 target situations were as follows: (38) a.

b. c. d.

Contest between little bear and rabbit to find a way over a wall to find out who has the marble (i) Which guy found out who has the marble? (ii) What way did he find out who has the marble? (iii) Who did he find out has the marble? Contest between Bert and a Kiwi Bird to find a way of figuring out when they're blindfolded, who has the cheese Contest between Snuffy and Robocop to find a way of seeing who the egg belongs to Contest between Kiwi and Bear to find a way of seeing what the baby was eating

2.4.1.3. Procedures The trials for Experiment 4 made up the final session in the series of experiments. For some children, we supplemented the four trials described above with some extra trials. We did this if we weren't quite sure how to interpret the child's responses. 2.4.2. Results We are most interested in children's responses to the second of the three questions asked on each trial, so we will give the results by individual subjects for these questions. First, we should look at children's responses to the first and third questions, which we will report in percentages. In response to the ARG + ARG questions (e.g. Which guy found out who has the marble?), children gave the correct answer 92 percent of the time, and responded to every question (except for one child, Alex 2, who didn't complete the game). In response to the ARG + 0 questions, children gave the correct answer 78 percent of the time, with 14 percent 'I don't know' responses. To summarize, children gave appropriate answers the vast majority of the time to the questions surrounding the targeted ADJ + ARG structures. This brings us to the questions of interest, the ADJ + ARG questions. Table 4 places the results from questions of this kind alongside those of Experiment 2, in which ADJ + ARG questions were tested without focusing on the event described in the matrix clause. Comparing the two studies, we found that children gave significantly more correct responses in the present experiment. Their overall performance increased from 18 percent correct to 68 percent correct. Just two of the very youngest children did not improve.10

SUCCESSFUL CYCLIC MOVEMENT

247

Table 4. Comparison of Responses Experiments 2 and 4 subject

age

Katherine Kelly Aisha Bryan Daniel Noah 2 Emma Alex 2 Tanya Rachel

4;11 4;8 4;1 3;11 3;5 3;4 3;4 3;4 3;3 3;0

upstairs responses experiment 2 2/4 2/4 0/4 1/4 0/4 0/4 1/4 0/4 0/4 1/4

upstairs responses experiment 4 4/4 4/4 3/4 4/5 3/4 4/5 3/6 0/2 3/4 1/4

In short, comparison of findings from Experiment 2 and Experiment 4 show that even children who are highly prone making 'wrong Q down' responses in some contexts are able to assign the alternative interpretation, if the salience of the upstairs question is heightened. Here is a sample of the protocol at work for one child. (39) Noah 2 3;4 (a) 'Wrong Q Down' Response in Experiment 2 Kermit: What do you wanna know? Let's see, there's Bert over there, and there's a telescope and glasses and there's Gummy Bear. Experimenter: How could Bert see what Gummy Bear was eating? Kermit: So how could Bert see what Gummy Bear was eating. How did Bert see what Gummy Bear was eating? What's the answer, Noah? Noah: Pizza (b) Adult Response in Experiment 4 Kermit: Noah, which guy found out who the egg belongs to? (Noah points) Robocop? What way did he find out who the egg belongs to? Noah: with his ladder Kermit: And who did he find out the egg belongs to? Noah: the bird

248

3.

ROSALIND THORNTON & STEPHEN CRAIN

Conclusion

Real advances have been made recently in developmental psycholinguistics; in particular, in relating linguistic theory and experimental investigations of children's emerging linguistic knowledge. Appealing to principles of Universal Grammar, researchers have been able to explain several cases in which children systematically make errant productions (e.g. Hyams 1986) and assign non-adult interpretations to sentences (e.g. Chien & Wexler 1990). Within this tradition, two accounts have been offered for the findings concerning children's 'errors' in the acquisition of wh-movement structures. One account argues for the early emergence of successive cyclic movement. On this account, there can be several stages of development prior to convergence on the adult grammar. Most children seem to converge directly, but some children pass through stages at which they ask questions using non-adult-like forms, although their questions receive adult interpretations. Several options co­ exist, running along the seams of natural language. Children's 'long-distance' questions are formed by 'partial movement' or by 'wh-copying'. Since whcopying is a successive cyclic movement structure, and since every child who uses partial movement also produces questions with this structure, there is no evidence of a stage of development at which successive cyclic movement is unavailable. The competing account argues that successive cyclic movement emerges late in the course of language development, preceded by several nonadult stages. First, there is a 'no movement' stage, and then a stage at which movement is strictly 'local'. We see the results of the studies summarized in this paper as confirmation for the former account of the acquisition of children's questions. The experi­ mental findings do not sit well with the hypothesis that children pass through a stage in which they are restricted to partial movement. To the contrary, the findings lend additional weight to the view that young children have the mechanisms of long-distance movement available to them. One thing remains unclear, however. We have no real understanding of the source of children's erroneous responses to the ADJ + ARG combination in the de Villiers et al study and replicated in our Experiment 2. That is, why do children make the 'wrong Q down' error? Although the error is greatly diminished when the salience of the adjunct question in the matrix clause is heightened, it is not eliminated altogether, even when the pragmatic context favors the adult interpretation. One possibility is that the source of the error is grammatical. Another possibility exists, however. It may be that the question of

SUCCESSFUL CYCLIC MOVEMENT

249

how someone asked something is a highly abstract concept for children. If so, their errors could simply be the result of conceptual difficulties that lie outside the syntax module. In any event, the results of the experiments designed to disentangle the two viewpoints favor the conclusion that the results from the de Villiers et al. study and those from our own studies should not be conflated. Finally, it seems clear that even young children are able to express and understand long-distance questions, using the same mechanism as adults: successive cyclic movement.

Acknowledgements We would like to thank Peter Culicover, John Frampton, Teun Hoekstra, Howard Lasnik, Bonnie Schwartz and Ken Wexler for discussions of the data and analysis presented here, as well as three anonymous reviewers who provided extensive useful comments on an earlier draft of the paper. Thanks also to Carrie O'Leary and Julie Solomon for assisting with the experiments. Finally, we wish to acknowledge the directors, teachers and children at Child Labs, The University of Connecticut, Storrs, Conn., and Another Place to Grow, Arlington, Mass. Notes 1.

Questions in which extraction was from a tensed embedded clause contained the verb think in the matrix clause, for the most part. The verbs guess and say were also used, but to a lesser extent. Questions in which the extraction was from an infinitival clause were confined to the verb want in the experiments reported here, but we have used the verb have in other work, with similar results.

2.

According to McDaniel (1986), Romani does not have a structure equivalent to infinitival clauses, so our observations are restricted to German. In German, though, as McDaniel (1986, 1989) points out, wh-phrases are prohibited in the Spec-CP of embedded infinitival clauses, such as in embedded questions. It is hardly surprising, therefore, that German does not permit a wh-phrase in medial position of wh-copying and partial movement structures. It is our position, however, that the appearance of whphrases is prohibited by principles of Universal Grammar.

3.

In this respect, our analysis departs from McDaniel's (1986) analysis of whcopying. McDaniel suggests that the wh-copy is in the intermediate Spec-CP.

4.

To avoid misconstrual of our position, we wish to note two claims that we are not making. First, we are not arguing against the view that language

250

ROSALIND THORNTON & STEPHEN CRAIN

acquisition occurs in stages. It is clear from the data obtained in our own studies that different children take different paths to a target grammar. Some children appear to advance through discrete stages in learning that Spec-head agreement involves a null complementizer, as outlined above. Second, we do not wish to claim that children have wh-movement from the earliest stages of acquisition. Hamburger (1981) documented the acquisition of relative clauses by a 2-year-old child and argued that her grammar initially formed 'proto-relative' clauses without wh-movement, resulting in 'gapless' modifying clauses. Shortly thereafter, there was a grammatical shift to relatives with a trace of movement. Having learned the properties of wh-movement in relative clauses, she was prevented by principles of learnability from using wh in situ questions, according to Hamburger, because this would have involved relearning the properties of wh-movement. If Hamburger is correct, then, there may be a stage in acquisition for some children, at around 24 months, at which their grammars lacks wh-movement. 5.

In (18) the lambda operator is used to bind a variable, x, which ranges over individuals. The lambda expression has as its extension a set of individuals who think that they, themselves, have a hat. The initial wh-phrase can be interpreted as a generalized quantifier referring to the family of properties which people within the domain of discourse have. Therefore, (18) is asking the hearer to identify those people who have the property of thinking that they have a hat among their family of properties.

6.

We would add also that an accidental coreference interpretation of crossover questions is limited to hearers, and is not ordinarily given by speakers. That is, the person who asks (19) does not intend the interpretation according to which the referent of the pronoun thinks that he, himself, has a hat. Accordingly, it has been suggested that the term NONCOREFERENCE is, properly speaking, short-hand for intended noncoreference of speakers (e.g., Lasnik 1989).

7.

According to the syntactic analysis offered by Rizzi (1990), the complemen­ tizer in the head position of the intermediate CP cannot bear AGR features when the embedded clause is infinitival, and therefore cannot participate in Spec-head agreement. Frampton (1990) also proposes that the head position of an infinitival CP is "syntactically inactive." Since we have argued that the medial-wh in children's questions does bear agreement features, the expectation is that these wh-phrases should not appear in extractions from infinitival clauses in any natural language.

SUCCESSFUL CYCLIC MOVEMENT

251

8.

In contrast to adjunct extraction, opinions diverge about how to represent wh-questions involving argument extraction. It has recently been argued that the wh-phrase in argument questions is base-generated in the matrix SpecCP, and that this wh-phrase is linked to its trace by a binding relation. It remains unclear whether this binding relation is relevant only in cases of 'long' movement out of weak wh-islands, or applies more generally. In any event, the representation assigned to long-distance argument questions is not seen to implicate an intermediate trace on this analysis (see Rizzi 1990 and Cinque 1990 for details).

9.

Noah 1 only used a medial-wh in subject extraction questions, and not in questions extracting a wh-phrase from object or adjunct position. This is in keeping with the findings in and Thornton (1990) and Thornton & Crain (forthcoming). The medial-wh is analyzed as appearing only in subject extraction questions because it is only in this case that it is needed to satisfy the ECP.

10. At 3;0 and 3;3, these children were younger than any of the children who participated in the de Villiers, Roeper & Vainikka experiment.

References Chien, Yu-Chin & Kenneth Wexler. 1990. "Children's Knowledge of Locality Conditions in Binding as Evidence for the Modularity of Syntax and Pragmatics." Language Acquisition 1.225-295. Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press. Chung, Sandra & James McCloskey. 1987. "Government, Barriers and Small Clauses in Irish." Linguistic Inquiry, 18.173-237. Cinque, Guglielmo. 1990. Types of Ä-Dependencies. Cambridge, Mass.: MIT Press. Crain, Stephen. 1991. "Language Acquisition in the Absence of Experience." The Behavioral and Brain Sciences 14.597-650. Crain, Stephen & Janet D. Fodor. 1993. "Competence and Performance in Child Language." Language and Cognition: A developmental perspective, ed. by Esther Dromi. Norwood, N.J.: Ablex. Crain, Stephen & Rosalind Thornton. 1991. "Recharting the Course of Language Acquisition: Studies in elicited production." Biobehavioral Foundations of Language Development, ed. by Norman Krasnegor, Duane Rumbaugh, Richard Schiefelbusch & Michael Studdert-Kennedy, 321-338. Hillsdale, N.J.: Lawrence Erlbaum.

252

ROSALIND THORNTON & STEPHEN CRAIN

Frampton, John. 1990. "Parasitic Gaps and the Theory of Wh-chàins" Linguistic Inquiry 21.49-77. Hamburger, Henry. 1981. "A Deletion ahead of Its Time." Cognition 8.389-416. Hyams, Nina. 1986. Language Acquisition and the Theory of Parameters. Dordrecht: Reidel. Kayne, Richard. 1976. "French Relative Que." Current Studies in Romance Linguistics, ed. by Frank Hensey & Martha Lujan, 255-299. Washington D.C.: Georgetown University Press. Koster, Jan. 1986. Domains and Dynasties. Dordrecht: Foris. Lasnik, Howard. 1989. Essays on Anaphora. Dordrecht: Kluwer. McDaniel, Dana. 1986. "Conditions on Wh-chains." Diss., City University of New York. . 1989. "Multiple Movement and Partial Movement." Natural Language and Linguistic Theory 7.565-604. McCloskey, James 1979. "Questions and Relative Clauses in Modern Irish." Diss., University of Texas at Austin. Pesetsky, David. 1982. "Complementizer-Trace Phenomena and the Nominative Island Condition." The Linguistic Review 1.297-343. Pinker, Steven. 1984. Language Learnobility and Language Development. Cambridge, Mass.: Harvard University Press. Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. Roeper, Thomas. 1986. "How Children Acquire Bound Variables." Studies in the Acquisition ofAnaphora. Volume 1: Defining the Constraints, ed. by Barbara Lust, 191-200. Dordrecht: Reidel. . 1990. "How the Least Effort Concept Applies to Partial Grammars: Copying in the acquisition of inversion and cyclic movement." Ms., University of Massachusetts, Amherst. Roeper, Thomas, Mats Rooth, Lourdes Mallis & Satoshio Akiyama. 1985. "The Problem of Empty Categories and Bound Variables in Language Acquisition." Ms., University of Massachusetts, Amherst. Stromswold, Karin. 1990. "Learnability and the Acquisition of Auxiliaries." Diss., Massachusetts Institute of Technology, Cambridge. Thornton, Rosalind. 1990. "Adventures in Long-Distance Moving: The acquisition of complex wh-questions." Diss., University of Connecticut, Storrs. Thornton Rosalind & Stephen Crain. Forthcoming. "Levels of Representation in Child Grammar." Forthcoming in Levels, Principles and Processes: The structure of grammatical representations, ed. by Wynn Chao & Geoffrey Horrocks. Berlin: Mouton de Gruyter. de Villiers, Jill, Thomas Roeper & Anne Vainikka. 1990. "The Acquisition of Long­ distance Rules." Language Processing and Language Acquisition, ed. by Lynn Frazier & Jill de Villiers, 257-297. Dordrecht: Kluwer.

Early Acquisition of Scrambling in Japanese

Yukio Otsu Keio University, Tokyo

1.

Introduction

This paper is an attempt to show that local (i.e., intra-clausal) scrambling constitutes part of the grammatical knowledge of Japanese three-year-olds. I would like to begin by reviewing Hayashibe (1975), which is often cited in conjunction with the acquisition of word order in Japanese.1 Hayashibe (1975) is an experimental study which was carried out within the framework of Bever's theory of perceptual strategies (e.g., Bever 1970). Perceptual strategies are a set of heuristics allegedly used by hearers during comprehension. One of those strategies is a word order strategy, the basic idea being that hearers attempt to assign thematic roles based on word order. Thus, in the case of English, when the hearer recognizes a sequence of N-V-N, he assigns the agent role of the action denoted by V to the first N and the patient role to the second N. This strategy is commonly called the NVN strategy. It has been claimed that the NVN strategy explains why it takes more time to process a passive sentence than to process its active counterpart. Bever and his associates also claimed that there is a period in development during which children heavily rely on perceptual strategies. Thus, for those who are in that period, passive sentence (lb) is often understood as its active counterpart (la). (1)

a. b.

The dog kissed the cat. The dog was kissed by the cat.

Hayashibe (1975) applied this idea to the acquisition of Japanese. He attempted to test whether there is a developmental period during which children heavily rely on the NNV strategy, the Japanese counterpart of the NVN strategy.

254

YUKIO OTSU

Hayashibe tested 23 children between 3;1 and 5;8 using an act-out method. Each subject is presented sentences like (2a) and (2b), among others, and is asked to act-out what the sentences mean by manipulating the toy animals put on the table in front of her.2 (2)

a.

b.

Kame-san ga ahiru-san o osimasita. turtle NOM duck ACC push+POL+PAST 'A/The turtle pushed a/the duck.' Ahiru-san o kame-san ga osimasita. duck ACC turtle NOM PUSH+POL+PAST 4 A/The turtle pushed a/the duck.'

Notice that (2b) is a scrambled version of (2a). If the child interprets (2b) as an adult would, she should have the turtle push the duck; if she interprets it according to the NNV strategy, she should have the duck push the turtle. Hayashibe's results are summarized in Table 1. For the present purposes, it is the existence of Group B subjects that is important. These 10 subjects, ranging from 3;4 to 5; 10, are those who apparently used the NNV strategy in executing the task. Table 1. Reliance on NNV Strategy Group

average age

A B C D

5;0 4;9 4;0 3;7

Group A: Group B: Group C: Group D:

youngest 4;5 3;4 3;1 3;1

oldest

n

5;8 5;10 5; 10 4;2

7 10 6 5

Those who interpreted the sentences in agreement with the adult grammar. Those who interpreted the sentences relying on the NNV strategy. Those who randomly interpreted both scrambled and nonscrambled sentences. Those who showed 'egocentric responses.' See Section 3.3.

Essentially the same results have been replicated by other researchers, most notably in a well-designed experimental study by Sano (1977). These results have been taken to indicate that there is a period in develop­ ment where young children tend to interpret scrambled sentences like (2b) as if they were nonscrambled sentences like (3). (3)

Ahiru-san ga kame-san o osimasita. duck NOM turtle ACC push+POL+PAST 'A/The duck pushed a/the turtle.'

SCRAMBLING IN JAPANESE

255

The question to be asked at this point is whether these results are reflections of the subjects' grammatical knowledge, i.e., whether their grammar lacks scrambling, or merely reflections of their overreliance on a word order comprehension heuristic, i.e., the NNV strategy. The latter interpretation would maintain that scrambling is already in their grammar, but that the heuristic overrides it. I assume that this is Hayashibe's (1975) original position. However, there is not much discussion of this matter in his paper, and hence I believe the interpretation of his results as well as the results of the various replications is an open question. The position I would like to defend in the remainder of this paper is that this interpretation is in fact the correct one.

2.

Production Data

First of all, I would like to present evidence from production in naturalistic contexts. The data have been taken from Iwatate (1981). The first part of his paper is a report of an examination of more than 48 hours of recorded speech of a two-year-old subject. Table 2 shows the child's use of various word orders with five selected verbs. The second row represents the number of utterances that had the object-subject-verb order with appropriate Case-marking particles attached. Clearly, scrambled forms were uttered by the subject. Table 2. Various Word Orders with Five Selected Verbs word order S 0 V 0 S V 0 V S

kaku 'draw'

taberu 'eat'

tukuru 'make'

11

3

kobosu 'spill'

6 2

1

kau 'buy'

total

2 1

16 7 3

The second part of the same paper reports on an examination of his recordings of utterances of 5 two- and three-year olds. Table 3 is a summary. Again, the second row represents the number of utterances that had the objectsubject-verb order with appropriate Case-marking particles attached. These two tables clearly show that even two-year olds do produce forms with apparent scrambling in naturalistic contexts. In the next section, I will report experimental evidence.

256

YUKIO OTSU

Table 3. OSV Order in Utterances of Five Two- and Three Year Olds word order with galo

Yuko (2;5) f

Jiro (2;5) m

Akiko (2;7) f

Jiro (3;6) m

Fumie (3;7) f

Taro (3;9) m

total

SOV O SV

24 2 1 2

6 4

11 1 3

11

12 2 3 1

15 1

79 10 7 4 0 1

svo OVS

1

vso v os (hours)

3.

1 8

6

8

6

8

6

Experiment

3.1. Linguistic Background As mentioned at the outset, the primary purpose of this paper is to show that scrambling is part of children's grammar at least by the age of three. One thing I must do first, then, is to look for reasons why Hayashibe's (1975) subjects were not able to make use of their knowledge of scrambling. While the NNV strategy is a possible candidate, let's reexamine Hayashibe's test sentences a little more carefully. Take a look at (2b), repeated below. (2) b. Ahiru-san o kame-san ga osimasita. duck ACC turtle NOM push+POL+PAST 'A/The turtle pushed a/the duck.' Notice that the child subjects are given (2b) in isolation, i.e., without discourse context. As has sometimes been pointed out, scrambled sentences like (2b) sound somewhat unnatural without discourse context. On the other hand, if we add a sentence prior to (2b) with a minimal change in (2b) as well, it sounds perfectly natural: (4) a. Kooen ni ahiru-san ga imasita. park in duck NOM ÍS-POL-PAST b. Sono ahiru-san o kame-san ga osimasita. the duck ACC turtle NOM push-POL-PAST 'There was a duck in a park. A turtle pushed the duck.' In (4), ahiru-san 'duck' was introduced into a discourse context as a discourse topic by the first sentence. Once it becomes a discourse topic, it is not only okay but even more natural to begin the second sentence with ahiru-san

SCRAMBLING IN JAPANESE

257

fronted by scrambling. Thus, there must be a discourse contextual reason to use scrambled sentences. In fact, there is some discussion about this point in Masunaga (1983). She claims that scrambling is possible when the scrambled element performs a BRIDGING FUNCTION, which she defines as "the function of providing a 'bridge' between the preceding discourse and the rest of the present sentence" (p. 456). I suspect use of scrambled sentences without any discourse context in Hayashibe's (1975) experiment is the reason why his subjects produced so many NNV interpretations of the stimulus sentences. At this point, it might be objected that, even if there is a bridging function at work, its effect is very subtle, and that it is difficult to imagine that the subjects' performance would be so grossly affected by such a factor. However, young children are in fact very sensitive to such a subtle discourse factor. For an excellent demonstration of this point, see Hamburger & Crain (1982), where it is shown that various discourse factors greatly affect children's comprehension of restrictive relative clauses in English. 3.2. Method Subjects: 24 subjects were tested. Twelve of them were three-year olds, and 12 were four-year olds. They are all mono-lingual native speakers of Japanese living in Tokyo. Procedure: Half of the children in each age group were assigned to the experimental group and half to the control group. Each child was tested individually. The task is essentially the same as that of Hayashibe's (1975), i.e., act-out. All subjects received four practice sentences, two of them intransitive sentences, and two nonscrambled transitive sentences. There were four test items, all of which contained scrambled sentences. The test items given to the control group are essentially the same as those of Hayashibe's (1975), i.e., scrambled sentences in isolation such as (2b). On the other hand, the test items given to the experimental group each consisted of two sentences like (4), the first of which serves to introduce an animal as a discourse topic, and the second of which is a scrambled sentence with the expression referring to the discourse topic scrambled to the front. Note again that scrambled sentences in isolation sound somewhat unnatural. While trained linguists are able to assign well-formed interpretations to scrambled sentences by supplying their own discourse context, this can hardly be expected of young child subjects.

258

YUKIO OTSU

3.3. Results The results are summarized in Table 4. 'C' represents a correct response from the viewpoint of the adult grammar. 'W' represents a wrong response. 'E' stands for 'egocentric responses/ by which we mean the child acted as an agent and the two animals mentioned in the sentence were chosen as patients. Such a response pattern has also been observed in Hayashibe (1975) as well as in other experimental studies. Although there were a few wrong and egocentric responses in the experimental group, the overwhelming majority of responses (43 out of 48) were correct ones. On the other hand, there were a fairly large number of wrong responses (22 out of 48) in the control group with a tendency that older subjects made fewer errors. Table 4. Test Results practice 1 experimental group 1 3;1 F F 2 3;3 M 3 3;5 M 4 3;7 F 5 3;9 6 3;10 M F 7 4;0 8 4;2 M M 9 4;4 F 10 4;7 F 11 4;9 12 4;11 M total

C E C C C C C C C C C C

control group 13 3;1 F M 14 3;2 M 15 3;4 F 16 3;6 M 17 3;9 18 3;11 F 19 4;1 M M 20 4;3 21 4;4 F M 22 4;7 F 23 4;8 24 4;11 F total

E C C C C C C C C C C C

C

c c c c c c c c c c c E

w w w w c w w c w c c

total W E

test items 2 3

4

c

C E C W C C C C C W C C

C E C C C C C C C C C C

C

c c c w c c c c c c c

4 2 4 3 3 4 4 4 4 3 4 4 43

0 0 0 1 1 0 0 0 0 1 0 0 3

0 2 0 0 0 0 0 0 0 0 0 0 2

E W W C W C C W C

E W W C W C C W C

c c c

c w c

0 0 0 2 0 4 3 0 4 2 3 4 22

0 4 4 2 4 0 1 4 0 2 1 0 22

4 0 0 0 0 0 0 0 0 0 0 0 4

E

w w w w c c w c w c c

SCRAMBLING IN JAPANESE

259

3.4. Discussion As is clear from Table 4, there is a sharp contrast between the responses of the two groups. The control group responded essentially in the same way as the Hayashibe subjects. On the other hand, NNV responses were almost entirely lacking in the experimental group. This clear contrast is undoubtedly due to the addition of the discourse context, which has apparently made the use of scrambled sentences quite natural. Thus, the results have shown that three-year olds can handle scrambled sentences using their grammatical knowledge of scrambling if given an appropriate discourse context. It should be noted here that sono 'the' is inserted at the beginning of the second sentence of the experimental group test items because the initial NP in this sentence becomes discourse presupposed following presentation of the first sentence. Sono plays no role whatsoever in marking the agent or patient of the sentence.

4.

Young Children's Case Marking

Children do, however, make mistakes in performance with scrambling. There is some evidence that children tend to use ga in sentence-initial position. Sentences (5) through (8) have been taken from Ito's (1990) book, which is based on his production corpus. (5)

Raazi ga akatyan miteru yo. radio baby look at The baby is looking at the radio.' correct adult form: Raazi o akatyan ga miteru yo.

(6)

Okaasan ga moratta no. mother

get

(2;2)3

(2;5)

PAST

4

1 got it from Mom.' correct adult form: Okaasan ni moratta no. There is also some evidence that children tend to use o in the position im­ mediately before the verb. Thus, in an elicited imitation study with three to sixyear olds, Sano (1977:225) reports that in simple active sentences, regardless of the particle that is missing, children display a remarkable tendency to place o in the second NP, that is, in the position immediately preceding the verb.4 A closer examination of these production data suggests that children tend

260

YUKIO OTSU

to use o or no particle at all in the position just before the verb, and that they tend to use ga in any other position. This accounts for (7) and (8) as well as (5). (7)

Gakko o iku no, gakkoo ga. school go 'I will go to school.' correct adult form: Gakkoo (e) ikuno, gakkoo e.

(2;5)

(8)

Arautte, epuron ga wash+will apron 'I will wash the apron.' correct adult form: Arautte, epuron o.

(4;1)

Note, incidentally, that o can be dropped in the adult grammar as well if it immediately precedes the verb. I conjecture at this point that this is a reflection of a property of Japanese grammar: namely, that whereas nominative Case is inherent in Japanese in the sense that it is not assigned by any element such as INFL, and thus, that it is freely base-generated, accusative Case is assigned by the verb under proper government. Thus, ga is an unmarked Case in some sense. See, for example, Saito (1982). What the word order strategy we discussed earlier attempts to capture seems to be these performance tendencies. The results of my experiments, however, along with Iwatate's (1981) production data, have shown that scrambling is there in the children's grammar and can be used in performance if appropriate care is taken.5

5.

Conclusion

What we have considered so far is limited to o-ga (accusative-nominative) scrambling. What about o-ni (accusative-dative) scrambling? We have some evidence that children do produce such forms. (9) and (10) are taken from Okubo's (1981-1983) fairly large production corpus. (9)

a.

b.

(Referring to the data cards of his utterances) Kore (o) otoosan ni misette agemasyoo. these ACC Daddy DAT show let's 'Let's show these to Daddy.' (Referring to the band-aid on his finger) Kore (o) otoosan ni hatte moratta no. this ACC Daddy DAT puton get PAST 'Daddy put this on.'

(3;0)6

(3;0)

SCRAMBLING IN JAPANESE (10) a.

b.

261

(Referring to Winnie-the-Pooh) Puu-san ni wan-piisu (o) kise. Pooh DAT dress ACC put.on+iMP 'Dress Pooh.' Aya-tyan ni, Aya-tyan ni obake (o) osie tai. Aya-tyan DAT Aya-tyan DAT ghosts ACC teach want 'I want to tell Aya (little sister's name) about ghosts.'

(3;0)

(3;0)

Notice that all these examples seem to follow the given-new information pattern. This strongly suggests that three-year-olds already know this in­ formational regularity, and can be taken as additional support for the claim that the reason why Hayashibe's (1975) subjects did poorly on the given task is that the stimulus sentences apparently violate this regularity. To sum up, we have seen that at least some kinds of local scrambling are in the grammar of three-year-olds, possibly earlier. If this is correct, it is one of the earliest movement rules, clearly earlier than any kind of passive, in the acquisition of Japanese grammar. The theoretical implications are to be worked out, but this is, I believe, an important first step toward understanding the nature of Japanese acquisition, and, more generally, the nature of grammar acquisition.7

Appendix In (i), a typical transitive sentence of Japanese is provided. (i)

Taroo ga Hanako ni hon o ageta. Taroo NOM Hanako DAT book ACC give+PAST Taro gave a book to Hanako.'

A generally accepted S-structure representation of (i) is given in (iii). As illustrated, Japanese is a postpositional language in which Case-marking particles as well as non-Case-marking particles are attached to NPs. Japanese is also a verb-final language. Word order among preverbal phrases is relatively free. Thus, the five sentences in (ii) are all acceptable variants of (i). (ii)

a. b. c. d. e.

Taw ga hon o Hanako ni Tara ga Hanako ni hon o hon o Taw ga hon o Hanako ni

Hanako ni hon o Taw ga Hanako ni Taw ga

gave. gave. gave. gave. gave.

262

YUKIO OTSU

It is generally assumed that a movement operation called scrambling is involved in the generation of these variants. It is also commonly assumed that the scrambled element is adjoined to IP, thus creating a structure like (iv), for example.

So-called long-distance scrambling is also possible in Japanese, but we will not deal with those cases in this paper.

Acknowledgements I have presented portions of this paper on various occasions, including colloquium talks at the University of Connecticut, the University of Ottawa, and

SCRAMBLING IN JAPANESE

263

McGill University. I have also discussed the experiment that is reported in this paper at the 1991 GLOW workshop on The Development of Movement and Inflection, and the 1991 CUNY sentence processing conference. I would like to thank those who gave me invaluable comments on those occasions. I wish to thank Kevin Gregg, John Whitman, and Bonnie Schwartz for their comments both on content and style. I am also grateful to anonymous reviewers for this volume for their invaluable comments. The usual disclaimers apply. This work was supported in part by the Monbusho (the Ministry of Education, Science and Culture of Japan) Grant-in-Aid for General Scientific Research 04831008 and the Inamori Foundation Grant for Scientific Research. Notes 1.

Readers with no or little familiarity with Japanese syntax are referred to the Appendix, where a brief description of Japanese phrase structure is provided.

2.

Sentences (2a) and (2b) differ from the exact sentences that were used in Hayashibe (1975) on the following two points. First, Hayashibe's sentences do not contain -san after the agent and patient nouns. Second, verbs in Hayashibe's sentences do not contain the polite marker -mas and are in the present tense. I have made these changes in order to make (2a) and (2b) in conformity with the sentences I used in my experiment that is reported below.

3.

Raazi should be razio in adult speech. Whether raazi in (5) is an ideolect of the child subject or a performance error is not clear.

4.

An anonymous reviewer pointed out the following possibility. There is some evidence that ga is used as a focus marker early in child speech. If we assume that IP-adjunction scrambling such as the one we are discussing in this paper is a focus operation, as claimed, for example, by Ueyama (1991), it would make sense for children to use ga to mark scrambled elements which receive a focus interpretation. The reviewer also pointed out that if ga is used as a focus marker in child speech, it would also account for the use of ga in (7) and (8).

5.

An anonymous reviewer pointed out that this performance tendency might be related to the fact that the Case markers, as isolated, phonologically reduced, and often deleted morphological markers, are highly susceptible to noise in spoken contexts. If true, we would be led to the following principle that was also pointed out by the same reviewer: Resort to a canonical sentoid strategy (e.g., NNV strategy) in noisy contexts.

6.

Misette should be misete in adult speech. Whether misette in (9a) is an ideolect of the child subject or a performance error is not clear.

264 7.

YUKIO OTSU It has been claimed (e.g., Borer & Wexler 1987) that formation of A-chains is maturationally controlled and that it explains why verbal passives in English which involve A-movement are acquired late, while wh-questions which involve an A-movement are acquired early. If we take the view that local scrambling in Japanese is an A-movement as we did in the Appendix, our conclusion is in conformity with such a claim.

References Bever, Thomas G. 1970. "The Cognitive Basis for Linguistic Structure." Cognition and the Development of Language, ed. by John R. Hayes, 279-352. New York: John Wiley. Borer, Hagit & Kenneth Wexler. 1987. "The Maturation of Syntax." Parameter Setting, ed. by Thomas Roeper & Edwin Williams, 123-172. Dordrecht: Reidel. Hamburger, Henry & Stephen Crain. 1982. "Relative Acquisition." Language Development. Volume 1 : Syntax and Semantics, ed. by Stan A. Kuczaj IL, 245-274. Hillsdale, N.J.: Lawrence Erlbaum. Hayashibe, Hideo. 1975. "Word Order and Particles: A developmental study in Japanese." Descriptive and Applied Linguistics 8.1-18. Ito, Katsutoshi. 1990. Kodomo no Kotoba. Tokyo: Keiso Shobo. Iwatate, Shizuo. 1981. "Nihongo-ji no Shoki Hatsuwa ni okeru Gojun." Japanese Journal of Education and Psychology 29.11-17. Masunaga, Kiyoko. 1983. "Bridging." Proceedings of the XIIIth International Congress of Linguists, ed. by Shiro Hattori & Kazuko Inoue, 455-460. Tokyo: Proceedings Publishing Committee. Okubo, Ai. 1981-83. Yoji no Kotoba Shiryo. 6 Volumes. Tokyo: Shu-ei Shuppan. Saito, Mamoru. 1982. "Case Marking in Japanese: A preliminary study." Ms., Massachusetts Institute of Technology, Cambridge. Sano, Keiko. 1977. "An Experimental Study on the Acquisition of Japanese Simple Sentences and Cleft Sentences." Descriptive and Applied Linguistics 10.213-233. Ueyama, Ayumi. 1991. "Scrambling and the FOCUS Interpretation." Paper given at the Rochester Workshop on Japanese Linguistics, University of Rochester.

Direct Access to X'-Theory Evidence from Korean and Turkish Adults Learning German Anne Vainikka University of Massachusetts at Amherst

1.

Martha Young-Scholten University of Durham

Introduction

Research on verb placement and X'-theory in generative syntax has provided a way to account for interesting stages in first language (L1) acquisition. In this paper we present and analyze data from Turkish and Korean adults acquiring German. We will show that these speakers straightforwardly acquire the German head-final VP, contrary to previous reports in the literature. Furthermore, we propose that these speakers acquire functional projections in a manner similar to what has been proposed for children, whereby they apply X'-theory in a gradual fashion. This process results in interesting developmental stages: an early stage without functional projections, an intermediate stage with an underspecified functional projection — resulting in the acquisition of verb raising and a version of non-pro-drop — and a more advanced stage with a more specified functional projection, resulting in the acquisition of the agreement paradigm, a more refined version of non-pro-drop and obligatory verb raising. 1.1. L2 Acquisition Compared to L1 Acquisition 1.1.1. Adult Access to UG An issue of current concern in language acquisition research is whether the acquisition of syntax by adults acquiring a second language is fundamentally different from children acquiring their first language. Recently various researchers have argued that acquisition by adults and children is qualitatively similar and thus that adults have access to Universal Grammar (UG). The proponents of this

266

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

view include Flynn (1984, 1987), White (1985, 1989), Schwartz (1986), duPlessis, Solin, Travis & White (1987), Eubank (this volume) and Schwartz & Sprouse (this volume). Under this view, the obvious differences between the first and second language acquisition of syntax can be accounted for by one of the following: (i) (ii) (iii)

(initial) parameter setting at the L1 value, parameter setting at a value available in UG (but neither an L1 nor an L2 value), or setting of various parameters at a schedule different from that which obtains in the L1.

Clahsen & Muysken (1986, 1989), Schachter (1988) and Bley-Vroman (1989) adopt the position, known as the FUNDAMENTAL DIFFERENCE HYPOTHESIS, that adults have limited or no access to UG, (see also Felix 1985). To the extent that adults have access to UG, it would then either be through the invariant principles of UG or possibly through the parameter settings of the learner's native language, i.e., indirectly. These researchers attribute second language (L2) acquisition to general cognitive mechanisms. Based on their analysis of naturalistic learners of German, Clahsen & Muysken (op. cit.) have argued that neither native-language parameter settings nor UG guides the L2 acquisition of word order. Rather, learners — irrespective of their native language background — apply a canonical word order strategy in their acquisition of German, which invariably yields an SVO order. In our research we initially sought to reinvestigate whether Turkish learners rely on this canonical word order strategy, under which no transfer occurs, or whether such learners transfer their Turkish head-final VP in their acquisition of German (i.e., transfer a parameter setting). The data which we will examine in this study provide evidence not only that learners transfer the L1 value of their parameter, but also that they adopt parameter values different from those in their L1. We will argue in this paper that our data support the idea that first and second language acquisition are very similar, at least with respect to the development of phrase structure and related phenomena (such as verb raising and the acquisition of inflectional morphology). However, such a claim cannot be made independently of a particular theory of first language acquisition. Since the acquisition of phrase structure is a controversial topic in L1 acquisition, we will briefly review the phrase structure debate here (cf. Clahsen, Eisenbeiss & Vainikka, this volume, for a more detailed discussion).

DIRECT ACCESS TO X'-THEORY

1.1.2.

267

The Weak Continuity Approach

Under the so-called STRONG CONTINUITY APPROACH, all functional projections are present at the outset of L1 acquisition (cf. e.g., Weissenborn 1990; Boser, Lust, Santelmann & Whitman 1992; Roeper 1992). That is, the full CP-tree is available to the child at the earliest stages of syntactic development. According to an alternative view, the WEAK CONTINUITY APPROACH, components of UG such as the X'-theory are available to the child at the outset of acquisi­ tion, but the actual syntactic tree has to be posited based on the input (cf. Pinker 1984; Lebeaux 1988; Vainikka 1992; Clahsen, Eisenbeiss & Vainikka, this volume). Although the Weak Continuity Approach resembles the maturational approach of Radford (1988, 1990) and Ouhalla (1991) in that both approaches allow for an initial stage without functional projections, the Weak Continuity Approach attempts not to use maturation as an explanation for developmental stages. If the analyses proposed in this paper for L2 acquisition are on the right track, this would provide strong evidence against the maturational approach for L1 acquisition: the explanation for the development of functional projections in adult L2 acquisition cannot rely on a maturational schedule. Therefore, whatever the explanation turns out to be for adults, this suggests that maturation is not necessary in order to account for similar acquisition facts from L1 acquisition (cf. Vainikka 1992 and Clahsen, Eisenbeiss & Vainikka, this volume). In previous work on the L2 acquisition of phrase structure, the equivalent of the Strong Continuity Hypothesis has been assumed (see, for example, Schwartz & Sprouse, this volume). We wish to propose that adopting the Weak Continuity Approach in L2 acquisition can explain the developmental stages proposed in this paper. Under this approach, we posit what we will term 'minimal trees' to account for the development of phrase structure; i.e., at any given stage of development, as few positions and projections are posited as are needed to analyze the data, and no more. If we assume that the language learner holds to such a minimalist principle, then projections and positions are only posited based on the learner's analysis of the input data. In its strongest form this approach leads to the following formulation of X'-theory (Chomsky 1986; Abney 1987): (i) (ii) (iii)

a head, once identified, projects a maximal projection; a complement position is posited based on positive evidence; and a specifier position is posited based on positive evidence.

268

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

The most crucial task of the language learner seems to be to identify a potential head; once a head is identified, the learner will posit argument positions (provided by X'-theory) if positive evidence is found. Our claim is that utilizing X'-theory in this fashion does not involve or require maturation, and therefore both children and adults are able to posit new maximal projections based on the input data. 1.2. Syntactic Background If adults have no access to UG, we would not expect to find those developmental connections between the acquisition of certain phenomena which have been attested for children acquiring German as a first language (cf. Clahsen 1986, 1988a). Clahsen's data on the first language acquisition of German reveals that with the acquisition of agreement, the main verb obligatorily raises to a higher functional projection from its head-final position in the VP and null subjects become infrequent. Accordingly, in the analysis of our L2 data, we are concerned with the acquisition of four specific syntactic phenomena. As a first step, we will look at the word order in the VP, with the aim of determining whether the headedness of the VP is transferred from the speakers' first language. We will then discuss phenomena associated with higher functional projections of the sentence, most notably IP and AGRP, concentrating on verb raising, subject-verb agreement and null subjects. By studying these phenomena we will attempt to determine whether an IP/AGRP is present in the speaker's grammar at various points of development. Before turning to the L2 data, an overview of the relevant syntactic processes in German, Turkish and Korean is necessary. According to the standard analysis of verb second (V2) (cf. Koster 1975; Safir 1981; den Besten 1983 and Platzack 1986), matrix and embedded clauses in German share the tree in (1) overleaf; following Clahsen (1990), we adopt Pollock's (1989) AGRP projection (replacing the traditional IP projection). 1.2.1. Word Order in the VP While German superficially exhibits a variety of word orders with respect to the finite and non-finite verb, the VP is generally assumed to be head-final; consider the following examples in (2):

DIRECT ACCESS TO X'-THEORY

(2)

a.

b.

269

Ich habe [ vp Klaus heute in Grafenberg gesehen] I have Klaus today in Grafenberg seen '{I have seen/I saw} Klaus today in Grafenberg.' Bitte Ihre Fahrkarte hier entwerten! please your ticket here validate 'Please validate your ticket here!'

Turkish and Korean also have a head-final VP, since all of the non-verbal material in the VP precedes the verb: (3)

Turkish a. Helmut [vp simdi Istanbul-da Türkce ögren-iyor] Helmut now Istanbul+LOC Turkish learn+PROG 'Helmut is learning Turkish in Istanbul now.' b. Bu kitab-i ev-de oku-ma! this book+ACC home+LOC read+NEG 'Don't read this book at home!'

(4)

Korean a. Helmut-ka [vp Peter-eke chaek-ul chu -oss -ta] Helmut+NOM Peter+DAT book +ACC give+PAST+DECL 'Helmut gave the book to Peter.' b. / chaek-ul chaeksang-e noh-ara! this book +ACC desk +DIR put +IMP Tut this book on the table!'

Thus, the word order in the VP in Turkish and Korean is similar to that of German.

270

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

1.2.2. Word Order in the IP Assuming that Turkish and Korean have an IP (cf. Choe 1988 for Korean), it would presumably have to be one where INFL follows the VP, since whatever tense/aspect/agreement marking occurs is found as a suffix on the verb. Consider the preceding examples in (4) and the following examples in (5): (5)

Korean a. Peter-nun umsik-ul mok -kosip-cian-ass -ta. Peter+TOP food +ACC would-like +NEG+PAST+DECL 'Peter would not have wanted to eat food.' b. Helmut-nun ttona-yaha-n -ta. Helmut+TOP leave-must+PRES+DECL 'Helmut must leave.'

The situation in German is more complicated, however. In matrix clauses the finite verb occurs in the second position, occupying the COMP position. In embedded clauses, where COMP is filled by a complementizer, the finite verb occurs in the AGR position resulting in a verb-final word order. (6)

a.

b.

Sie wollte Oskar erstechen. she wanted Oskar stab 'She wanted to stab Oskar.' Ich lese gerade, daß sie Oskar erstechen wollte. I read just that she Oskar stab wanted 'I'm just reading that she wanted to stab Oskar.'

1.2.3. Empty Subjects Both Korean and Turkish allow empty (thematic) subjects, whereas German does not: (7)

Turkish §imdi Istanbul-da Türkce ögren-iyor -um. now Istanbul+LOC Turkish learn +PROG+ISG 'Now I am learning Turkish in Istanbul.'

(8)

Korean Naeil

Frankfurt-e

ga-n

-ta.

tomorrow Frankfurt+DIR GO+PRES+DECL

'Tomorrow {I/you/he/she/we/you/they} go to Frankfurt.'

DIRECT ACCESS TO X'-THEORY (9)

271

German *Morgen fährt nach Frankfurt. tomorrow travel to Frankfurt

1.2.4. Subject-Verb Agreement German has a full subject-verb agreement paradigm consisting of suffixes which mark person and number. Like German, Turkish has subject-verb agreement, as can be observed in the example in (7);1 however, Korean does not mark subject-verb agreement.2 Thus, under standard analyses all three languages have a head-final VP projection and a head-final IP/AGRP projection. However, unlike in Turkish and Korean, the finite verb of the German matrix sentence does not occur at the end of the sentence. Rather, it is analyzed as having raised to a head-initial COMP position (resulting in the verb-second word order). German further differs from Turkish and Korean in that omitting a thematic (non-expletive) subject is ungrammatical. Korean, on the other hand, differs from Turkish and German in that Korean does not exhibit a subject-verb agreement paradigm. In the following section we will describe in more detail our test subjects and the techniques used in data collection. In Section 3 we will show that our cross-sectional data can be used to investigate developmental stages. In Section 4 the developmental stages are discussed, together with the characteristic syntactic properties of each stage. Finally, in Section 5 broader theoretical implications will be considered.

2.

Methods and Materials

2.1. Data Collection Our findings are based on cross-sectional data from 11 Turkish and 6 Korean adults acquiring German.3 We collected the data from 7 of the Turkish speakers and from all 6 Korean speakers using a variety of techniques which will be detailed below. Testing sessions took place either at the test subject's home, place of work, or a cultural center. The data for each speaker was collected during one or two sessions of 45 to 90 minutes each in duration. The complete session for each speaker was tape recorded, all utterances throughout the session

272

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

transcribed and the transcriptions checked against the tape. The data from 4 of the Turkish speakers in our study (Memduh, Sevinc, Kemal and Kadir) represents the least advanced speakers in von Stutterheim's cross-sectional corpus for which interviewing and narrative techniques were used (cf. von Stutterheim 1987).4 As we had access to von Stutterheim's tapes, we were also able to utilize information from the speaker's intonation and from the context in analyzing the data. The age of the learners from whom our data comes ranged from 28 to 60 years and their length of residence in Germany from IV2 to 24 years. All speakers had arrived in Germany after age 20. We found no correlation between length of residence and stage of syntactic development. For example, the least advanced Turkish learner had lived in Germany for 11 years at the time of testing, while one of the most advanced Turkish learners had been in the country only 6 years. Table A in the Appendix provides detailed biographic information. All of the speakers had received very little or no formal instruction in German. Using such speakers enables us to make more independent claims about the acquisition mechanism in adults, as it is not clearly understood how the variable of formal instruction affects acquisition.5 The testing sessions consisted of several elicitation tasks, the purpose of which was to elicit a variety of sentence types in order to examine verb placement and the usage of subjects, and to elicit different grammatical persons and numbers to examine subject-verb agreement. Above all, we wanted to avoid the type of elliptical utterances which can occur when interviewing techniques are used and we wanted to provide clear extralinguistic contexts in order to better interpret the learners' utterances. In two of the tasks, the test subjects were asked to tell the stories in comic strips which contained minimal or no text. One set of comic strips was designed to provide singular and plural contexts in order to test agreement, while another set was chosen to provide the possibility of using an empty subject in connection with a topicalized object. In addition, the comic strips provided contexts for different word order possibilities, depending on whether a subject or an object was topicalized. In all of the picture description tasks, the test subject was asked questions such as What's going on here? rather than What's Snoopy doing?, in order to preclude the possibility of elliptic answers. In a third task, the test subjects were asked to describe hand-drawn picture sets depicting everyday actions with transitive verbs in the third person singular. In order to elicit second person singular verbs, we asked the test subjects in a fourth task to describe an action that the test administrator had just carried out.

DIRECT ACCESS TO X'-THEORY

273

Finally, in a fifth task, we provided the test subject with two types of contexts: one in which the main verb would normally be raised (i.e., present tense), and one in which the main verb would typically remain in the VP (i.e., present perfect tense in German). The actual task involved the test subject describing the step-by-step preparation of tea in two different ways, based on pictures of the relevant objects. In the description of one type of tea preparation, the test subject was asked to use the present tense. In the second series, the test subject was instructed to use the present perfect tense. Speakers were only corrected at the beginning of each part of this experiment if they answered elliptically, if they omitted the agent or if they did not use some reference to past time in the second part of the experiment. Utterances which were repetitions (or near repetitions) of the test administrator's corrections were not included in our data. Since this test was administered last, it did not skew the data from the other tasks in forcing the speakers to use particular syntactic constructions. In addition to the five tasks described above, the sessions included several experiments involving grammaticality judgements constructed as cloze tests.6 Spontaneous comments made during these experiments were included in our database as were the answers to the interview-type questions asked prior to the administration of the tests and all the informal conversation which took place before, during and after administration of the experiments. 2.2. Preliminary Analysis of the Data Several hundred utterances were collected from each speaker, but only a proportion of these were useful to us in investigating verb placement and related phenomena. We excluded all utterances without a verb, as well as sentences which consisted of a verb and nothing else. In addition, sentences which were imitations or clear idiomatic phrases were not included. For example, Aysel's data represents two 45-minute sessions during which she uttered only 138 sentences containing a verb together with some other material. Table B in the Appendix provides the total number of sentences containing verbs and other material for each speaker. The number of such sentences ranges from 84 to 482 per speaker. We then attempted to analyze each sentence with respect to verb placement, basing our analysis solely on word order. That is, we wanted to find out whether the sentence contained a finite verb that had been raised to a higher functional head (e.g., INFL or COMP) or whether the verb remained in the VP (as non-finite verbs in German do).7 We did not consider agreement marking as a criterion. Word order turned out to be a more reliable property, since many of

274

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

our speakers had not acquired the full German agreement paradigm, and since it was not clear in many instances whether or not the verb was inflected. Thus, the sentences were divided into three groups, based on word order: (i) (ii) (iii)

sentences with the verb in the VP, sentences with a raised verb, and sentences for which it was not possible to decide whether the verb was in the VP or had been raised.8

We attempted to make as strict as possible our criteria for deciding whether the verb was in the VP or whether it had been raised. The verb was analyzed as being in the VP if it was preceded by one or more of the following: a direct object, an indirect object, a locative adverb/PP, other PP arguments or adjuncts of the verb, or a predicate noun or adjective. Some examples are given in (10): (10) a.

b.

c.

Haar schön machen. (Changsu #124) hair pretty make '(She)'s making (her) hair (look) pretty.' Er mit Schnee spielen. (Dosik #165) he with snow play 'He's playing with snow.' Unten Frau ganz nix Deutsch sprechen. (Mine #27) below woman at.all not German speak '(The) woman downstairs doesn't speak German at all.'

We assumed that the verb had been raised if all of the material in the VP — as given above — followed the verb, or if the subject followed the verb.9 Consider the following examples of what we analyze as raised verbs: (11) a.

b.

c.

Ich kaufen Brot so türkische Geschäft. I buy bread so Turkish store 'I buy bread (at a) Turkish store.' Wir kaufen hier so Kaffee extra. we buy here so coffee extra 'We buy coffee here extra.' Trinkst du Cola? drink you cola 'Are you drinking cola?'

(Mine #187)

(Mehmet #121)

(Samran#113)

When a verb was followed only by one of the following, we did not consider this to be a reliable indication of a raised verb: time adverbs and other sentential adverbs, sentential particles, nonfinite verb forms and negation.10

DIRECT ACCESS TO X'-THEORY

275

The examples in (12) illustrate sentences for which we felt that we were unable to determine whether the finite verb had been raised or not, given our strict word-order-based criteria: (12) a.

b.

c.

Weiter halbedrei Stunden dauert? (Samran #199) further half.three hours take '(Will it) take another three-quarters of an hour?' Immer gucken oh, oh. (Mehmet #57) always look, oh oh '(I'm) always looking, oh oh.' Ich diese Kessel machen heiße setzen. (Mine #73) I this kettle make hot set 'I set this kettle (on the stove and) heat (it).'

The sentences of the type in (12) will not be considered in the remainder of the paper. In the following discussion, we will maintain the strict division between sentences that appear to consist of a bare VP projection (since they contain no raised verb) and sentences in which there seems to be at least one functional projection (since one of the verbs has been raised to the left of VP). Making this distinction allows us to discover generalizations that hold within each of the two groups of sentences, but that do not hold if the two groups are combined.

3.

Crosslinguistic Data as Developmental Data

Since our data is cross-sectional, it was not obvious how to determine whether developmental stages could be established based on the data. The goal of this section is twofold: first, to show that the data we obtained can indeed be used to make developmental claims, and second, to introduce the developmental stages. The data arguing for the head-final VP (cf. Section 1) are also presented in this section. 3.1. The Implicational Table Meisel, Clahsen and Pienemann (1981) developed a technique of implica­ tional scaling for the purpose of determining developmental stages in L2 acquisition; in addition, the technique has been successfully used for L1 acquisition (cf. also Hatch & Farhady 1982). Using this technique, we looked for implications that might hold between the four syntactic phenomena under

276

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

investigation. The results are summarized in Table 1. A cut-off point of 60 percent was used as a general criterion for acquisition; i.e., we judged a construction to have been acquired if it was used in at least 60 percent of the obligatory contexts.11,12 The test subjects are ordered here according to their level of development, as determined by us based on the four syntactic phenomena. All subsequent tables, including those in the Appendix, follow the same order of test subjects given in this table. Table 1. Implicational Table (to Be Revised) name Aysel Memduh Changsu Sevinc Kemal Kadir Ahmet Mehmet Dosik Park Özgül Mine Gabho Samran Emine Harva Ensook

level I I I I I II II II II III III IV IV IV IV IV IV

language T T K T T T T T K K T T K K T T K

head-final VP + + + + + + + + + + + + + + + + +

no null subjects

verb raising

-

-

+ + + + + + + + + + + +

+ + + + + + + +

agreement paradigm

+ + + + + +

The table provides information for each of our informants concerning whether or not they have acquired the following: the German head-final VP; the fact that German does not allow empty (thematic) subjects — unlike Turkish and Korean; verb raising to the left of the VP (which, to the extent that it occurs in Turkish and Korean, applies string-vacuously to the right); and the German subject-verb agreement paradigm. The level of each speaker is given in the left-most column; this is determined based on the number of relevant phenomena that the speaker has acquired.

DIRECT ACCESS TO X'-THEORY

277

The 'head-final VP' column in Table 1 shows that all of our speakers have a head-final VP. To arrive at this result, we considered two groups of sentences: (i) (ii)

sentences with VX/XV word order, but no subject;13 and sentences in which the finite verb preceded the VP and a second (infinitival or participial) verb occurred in the VP.

We assumed that these types of sentences would accurately reflect the word order in the VP, independently of whether the test subject had acquired verb raising. In their analysis of similar Turkish data, Clahsen & Muysken (1986, 1989) conclude that these learners do not adopt a head-final VP for German. However, of the sentences we considered in our data, over 60 percent were clearly verb-final and were verb-final for all of the speakers regardless of their overall level of competence.14 For example, of Memduh's 129 relevant sentences, 127 were verb-final, and of Samran's 73 such sentences, 67 were verb-final. Table C in the Appendix provides the detailed figures for each speaker. Consider now the 'verb raising' column (we will return to the second column shortly). This column is based on the proportion of sentences with a raised verb in the speaker's corpus, i.e., sentences in which the finite verb precedes the VP material (according to the strict criteria outlined in Section 2). For the speakers who have a ' - ' value in the 'verb raising' column, fewer than 60 percent of their sentences contain a raised verb, while more than 40 percent consist of what looks like just a bare VP-projection (or an infinitival clause). The speakers with a '+' value in this column produced sentences with a raised verb over 60 percent of the time. For the 3 least advanced speakers, more than 80 percent of their sentences were analyzable as bare VPs, while for the two most advanced speakers, more than 80 percent of their sentences contained a raised verb outside of the VP. Table D in the Appendix provides the individual values. The 'no null subjects' column of Table 1 refers to the distribution of subjects. We found that while the distribution of subjects in bare-VP sentences was not a reliable indicator of the level of development (cf. discussion below), the usage of overt subjects in sentences with a raised verb was very predictable, depending on the level of development. This is not particularly surprising, if we assume that the null subject parameter is associated with properties of INFL; we will return to this question in Section 4. Therefore, given all the sentences with a raised verb, we have calculated the proportions of empty and overt subjects. The speakers with a '+' value fulfill two criteria:

278

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN (i) (ii)

they have an overt subject in over 60 percent of the sentences with a raised verb and they have a raised verb in over 60 percent of the sentences containing an overt subject.15 (See Table E in the Appendix for the individual values.)

In the data we find a clear difference in the occurrence of subjects in sentences with a raised verb and in sentences without a raised verb. Figure 1 shows the proportion of empty subjects in both types of sentences.16

Figure 1. Empty Subjects in the Two Types of Sentences Figure 1 shows that, in general, there is an overt subject in sentences with a raised verb much more often than in VP-sentences. It is also clear that the increase of overt subjects in sentences with a raised verb (shown by the decrease in empty subjects in the graph) is not accompanied by a comparable increase of overt subjects in bare-VP sentences — for all speakers. That is, it appears that these speakers are not acquiring a general rule which states that a German sentence must have a subject, but rather that sentences with a raised verb must have an overt subject.17 Consider now the rightmost column in the implicational table; in this column the results of our investigation of the productive acquisition of the

DIRECT ACCESS TO X'-THEORY

279

agreement paradigm are summarized. Recall that word order was our sole criterion for determining whether a verb was raised or not. This final column gives us information about the agreement morphology on verbs which we had previously established as having been raised from the VP.18 In order to assess our speakers' acquisition of the agreement paradigm, we took all sentences containing a raised main verb and calculated the proportion of correct agreement suffixes occurring on the verb. We excluded auxiliary verbs and modals, as they do not reflect the productive paradigm in German, at least in the present tense.19 The speakers indicated as having acquired the agreement paradigm had a correct agreement suffix over 60 percent of the time on raised main verbs. Moreover, in order to ensure that the paradigm had been acquired, we included the additional criterion that there be at least two correct instances of four different agreement suffixes.20 Speakers who did not meet these two criteria were considered not to have acquired agreement. The six most advanced speakers had a correct agreement suffix on the raised main verb at least 80 percent of the time, and they used at least four of the five suffixes correctly. All other speakers produced at most two different agreement suffixes. For many of the less advanced speakers, about half of their raised verbs carried the infinitival/ plural suffix -n, which we assume to be the default form that occurs in the bareVP sentences, both in first and second language acquisition. The individual proportions are given in Table F in the Appendix. 3.2. The Implicational Scale and the Developmental Stages Table 1 gives rise to an implicational scale of development, given in (13): (13)

Agreement - Verb Raising - No Null Subjects - Head-final VP

That is, if a particular speaker has acquired the agreement paradigm, that speaker has also acquired verb raising, non-pro-drop, and a head-final VP; however, the acquisition of verb raising does not necessarily mean that the agreement paradigm has been mastered. Because verb raising is acquired before the agree­ ment paradigm, raised verbs may not be consistently marked for agreement.21 The scale in (13) suggests four developmental stages, each of which is characterized by the acquisition of a syntactic phenomenon. It thus appears that our cross-sectional data can indeed be used to study the developmental stages of the L2 acquisition of German. As we take into account further syntactic phenomena which are not included in the implicational table, we find that minor adjustment is called for with respect to the stages to which some of the speakers

280

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

belong. In the following section we end up with three developmental stages rather than four: at the first stage, basically only a VP-projection has been acquired; at the second stage, one functional projection has been acquired; at the third stage, the features of the functional projection have been specified. How­ ever, even the speakers at the third stage cannot be considered very advanced, as it appears that even they do not have a full mastery of the CP-projection.

4.

Stages of Development

4.1. The First Stage: The VP Stage The grammar of the least advanced speakers appears to involve just a VP without any further functional projections; we will refer to this stage as the 'VP stage'. The implicational table indicates that the major portion of the utterances of Aysel, Memduh, Changsu, Sevinc and Kemal do not have a raised verb. We assume that such utterances consist of a bare VP. In this section we will concen­ trate on Aysel and Memduh (Turkish speakers)22 and Changsu, a Korean speaker.23 Typical examples of sentences produced by Aysel, Memduh, and Changsu are given in (14). (14) a.

b.

c.

d.

e.

f.

Oya Zigarette trinken. Oya cigarette drink 'Oya smokes cigarette(s).' Teekanne die Ofen setzen. teapot the oven put '(I) put the teapot (on) the stove.' Ama ich zwei Jahre Berlin bleiben. but[Turkish] I two years Berlin stay 'But I stayed (in) Berlin two years.' Ja alles hier kaufen. yes everything here buy 'Yes (I) buy everything here.' Eine Katze Fisch alle essen. a cat fish entire eat 'A cat ate the entire fish.' Hier Jacke ausmachen. here jacket off.make 'Here (you) took (your) jacket off.'

(Aysel #11)

(Aysel #24)

(Memduh #190)

(Memduh #069)

(Changsu #033)

(Changsu #150)

DIRECT ACCESS TO X'-THEORY

281

Recall that according to our criteria, all of our speakers have a head-final VP. We take this to mean that the word order in the VP is transferred from Turkish/Korean. This is contrary to the position held by Clahsen & Muysken (1986, 1989), who offer the absence of transfer of the head-final VP from Turkish to German as further support that adult learners use general cognitive strategies in their acquisition of a second language.24 If presence of a head-final VP for the Turkish (and Korean) learners no longer necessitates an explanation involving the use of a canonical word order strategy by second language learners, then data from the Spanish, Italian and Portuguese speakers discussed by Clahsen & Muysken (op. cit.) can no longer be explained in this way either; the SVO order these learners adopt at the earliest stages of acquisition must also be the result of transfer of the headedness of their (head-initial) VP (cf. Young-Scholten & Vainikka 1992 for discussion; duPlessis etal. 1987 and Schwartz & Tomaselli 1990 also maintain that these speakers start out with the head-initial VP of their native languages). We propose that the majority of utterances from the three least advanced speakers, Aysel, Memduh and Changsu, be analyzed as bare VPs, without any further functional projections. If their phrase structure grammar for a German sentence consists of just a VP, we would not expect to find any evidence for, e.g., an IP or a CP We list the predictions that such an analysis would make in (15) (cf. also Radford 1988, for a similar discussion regarding L1 acquisition). (15) Predictions for the VP stage a. Verbs occur at the end of the sentence, since the German VP is head-final, and since there is no position for the verb to raise to. b. There is no agreement paradigm, since there is no INFL (or AGR) position for base-generating agreement suffixes; verbs should either carry no suffix, or carry a non-agreement suffix. c. There are no overt complementizers, since there is no COMP position for base-generating the complementizer. d. The distribution of overt subjects may differ, since they occur in the Spec-VP and not in the Spec-IP.25 Furthermore, depending on the analysis of modals and auxiliaries (which behave like main verbs in German), we might not expect to find any modals or auxiliaries (associated with INFL) or questions with a moved wh-phrase or a raised verb (associated with CP). In order to investigate these predictions in the data of the three relevant speakers, we will first concentrate on their VP-sentences, ignoring the sentences

282

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

with a raised verb for now (about 15 percent of the total), unless otherwise stated. By definition, (15a) holds of this set of sentences: bare-VP sentences contain a verb in the VP, at the end of the utterance. Concerning the agreement data for the 3 speakers, we find the following. All of Aysel's VP-sentences contain a verb with the (default) suffix -n regardless of the person/number referred to,26 suggesting she does not have any notion of the agreement paradigm.27 About 71 percent of Memduh's and Changsu's VPsentences contain a verb with the default suffix -n, regardless of the person or number referred to. All but one of Memduh's remaining verbs have no suffix at all.28 In addition to the default suffix -n Changsu produces some potential agreement suffixes; however, several of them are incorrect.29 We conclude that none of these speakers use the German agreement paradigm productively in their VP-sentences. Recall from the implicational table that these speakers have not acquired the agreement paradigm for raised verbs either. Table F (Appendix) shows that (apart from -n) Aysel and Memduh produce no correct instances of any agreement suffix on a raised verb, while Changsu produces only two correct instances of such suffixes. The fact that Turkish has a subject-verb agreement paradigm does not seem to give Aysel and Memduh an advantage over the Korean speaker Changsu (whose native Korean lacks subject-verb agreement, potentially putting her at a disadvantage). In addition, neither Changsu nor Aysel produced any instances of modals or auxiliaries; if modals and auxiliaries are base-generated in INFL, their occurrence is unexpected at this stage. Memduh's VP-sentences contain no auxiliaries, but we find five instances of the modal muß 'must', shown in (16). Since there is no infinitival verb in these constructions, it is possible that muß is treated as a main verb by Memduh (as it would be in native German).30 (16)

Ja Frau Türkei muß yes wife Turkey must 'Yes (my) wife must (go to) Turkey.'

(Memduh #39)

As is predicted by the absence of a CP, we find no instances of embedded clauses with an overt complementizer in the data of Aysel, Memduh or Changsu (cf. 15c), neither in the VP-sentences nor in the sentences with a raised verb. In addition, none of these speakers produced any wh-questions with a fronted wh-phrase, or any yes/no questions with a fronted verb. Instead of verb raising, yes/no questions with a typical VP-pattern were produced, as shown in (17): (17)

Morgen wiederkomm oder vielleicht? tomorrow again-come or perhaps '(You will) come again tomorrow perhaps?'

(Memduh #507)

DIRECT ACCESS TO X'-THEORY

283

Finally, we turn to the distribution of overt subjects (prediction (15d)). Aysel produced 66 VP-sentences; 55 (or 83 percent) of these have an empty subject. Similarly, 82 percent of Memduh's VP-sentences have an empty subject (124 of 151), and 59 percent of Changsu's VP-sentences contain an empty subject (40 of 68). Clearly, for these speakers subjects are optional in the bare-VP sentences. Recall, in fact, that subjects are optional in VP-sentences for all 17 of our speakers (as illustrated in Figure 1, Section 3). This is true even for the speakers for whom subjects seem to be obligatory in sentences with a raised verb. We propose tentatively that the null subject parameter can only be set with respect to a functional projection above the VP; we will address this in detail below. In the Spec-VP position, a subject is treated just like any other argument; at this early stage, arguments are frequently left out.31 Thus, all of the predictions listed in (15) above hold for the VP-sentences produced by Aysel, Memduh and Changsu. That is, about 85 percent of their utterances can be analyzed as just consisting of a VP. Turning to the remaining 15 percent for each speaker, we find some indication, especially for Memduh and Changsu, that these speakers might be in the process of entering the next stage, with a higher functional projection. Aysel's remaining nine sentences differ from the bare-VP sentences only in terms of word order; i.e., the verb precedes the object in these sentences.32 We assume that these sentences are the first indications of the subsequent stage at which the verb is raised to a head-initial functional projection. This suggests that verb raising may be a trigger for a higher projection, rather than, for example, agreement morphology or the presence of auxiliaries/modals (none of which Aysel seems to have). Fourteen of Memduh's remaining 24 sentences involve a main verb with the default suffix -n (or no suffix):33 (18) a.

b.

Meine Vater nicht rauchen Sigara. (Memduh #270) my father not smoke cigarette[Turkish] 'My father doesn't smoke cigarette(s).' Sechs Mann hier arbeit Farbe eine Schicht. (Memduh #275) six men here work paint one shift 'Six men work (on) painting here (during) one shift.'

As in Aysel's case, the examples from Memduh in (18) may involve verb raising to a higher functional head (INFL or AGR). In addition, three of Memduh's relevant sentences involve the modal muß 'must' as shown in (19a). Seven of his sentences with a verb to the left of VP are copular sentences with either ist 'is', bin 'am' or sind 'are', correctly used, cf. (19b):

284

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

(19) a.

b.

AmaÇT) diese Junge so muß Deutsch lernen. (Memduh #194) but this boy so must German learn 'But this boy must learn German.' Das ist Atatürk Bild ja. (Memduh #137) that is Atatiirk picture yes 4 Yes, that's (a) picture (of) Atatiirk.'

Altogether, 10 of Memduh's 175 sentences (or 6 percent) involve an agreement form of the copula or a modal, occurring to the left of the VP. We would like to suggest that these examples are precursors of the next stage of acquisition. The distribution of Changsu's 11 sentences with a raised verb is as follows. There is one example of a modal, with the correct agreement suffix (20a), and 4 examples with a correct form of the copula, as in (20b). In addition, one instance of the correct form of the auxiliary is found (20c). The remaining 5 examples involve a raised main verb, 2 of them with the correct inflection -t, as exemplified in (20d). (20) a.

b.

c.

d.

Ich möchte nicht gut da. I want not good there '1 don't want (to go) there.' Schnee ist da. snow is there '(There) is snow there.' Ich bin Frankfurt Zug fahren. I am Frankfurt train go 'I have gone (to) Frankfurt (by) train.' Die Frau alleine trinkt Saft. the woman alone drinks juice 'Only the woman is drinking juice.'

(Changsu #125b)

(Changsu #98)

(Changsu #15)

(Changsu #146)

As with Memduh, we would like to suggest that these examples are precursors of the next stage. The only instances of correctly used productive agreement suffixes on the main verb produced by Changsu involved the -t suffix (some of which seem to occur in the VP and some on raised verbs). No such suffixes were produced by Aysel or Memduh. We conclude that, by all the criteria considered, about 85 percent of the sentences (i.e., those sentences containing a verb and some other material) produced by these three speakers look like bare VPs. The structure of these sentences is then the one given in (21).

DIRECT ACCESS TO X'-THEORY

285

The subject is (optionally) base-generated in the Spec-VP, following suggestions in syntactic theory (cf. e.g., Zagona 1982; Kitagawa 1986; Sportiche 1988). The remaining 15 percent of the sentences produced by these learners may involve an emerging functional projection. In describing the data from these speakers, we have followed the approach discussed in Section 1.1.2., according to which projections are posited only as required, based on the data. Given that it is difficult to empirically distinguish this point of view from the alternative approach, it may be that these speakers have — in addition to a VP — both an IP and a CP projection.34 The occurrence of sentences with a raised verb (15 percent of total) could thus be explained by positing an IP projection. However, if there is an IP projection in all of their VP sentences, we would then have to explain why in most sentences there is no evidence at all for an IP, i.e., why the verb does not always raise to INFL. The corresponding problem for our approach is why a speaker might sometimes use a bare VP and sometimes an IP; we are assuming that for the 3 speakers at the first stage of syntactic acquisition, an IP is not yet a productively available projection. Given the explanatory power of an approach which assumes minimal trees in accounting for developmental stages, we will continue to use this approach. 4.2. The Third Stage: The AGRP Stage To facilitate presentation we will now consider the most advanced group of our informants. We will return to the remaining, intermediate, speakers in Section 4.3. Our most advanced group consists of 6 speakers: the 3 Turkish speakers Mine, Emine and Harva, and the 3 Korean speakers Gabho, Samran and Ensook.35 These speakers are the only ones who — according to the implicational table in Section 3 — have acquired a head-final VP, non-pro-drop, verb raising and the agreement paradigm. Consider the following typical examples uttered by these six speakers: (22) a.

Ich liebe diese so. I love:ISGthis so 4 1 really love this.'

(Mine #156)

286

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN (22) b.

Ich kaufe dich Eis. (Gabho #254) I buy:ISGyou ice-cream 'I (will) buy you ice-cream.' c. Trinkst du Cola? (Samran # 275) drink: 2SG you cola? 'Are you drinking cola?' d. Er hat gesagt, nimmst du Lokomotive? (Emine #62) he has said, take:2SG you train? 'He said, (will) you take (the) train?' e. Der kleine geht Kindergarten. (Harva #21) the small.one goes:3SG kindergarten 'The young one goes (to) kindergarten.' f. Sie kommt zu Hause. (Ensook #131) she comes:3SG at home 'She's coming home.' The speakers in this group are our only informants who have acquired the German agreement paradigm, as shown in Tables (F) and (1-1,1-2) in the Appendix. Each of the speakers used at least four of the five agreement suffixes in German correctly, and apart from some persistent usage of the default suffix -n, these speakers almost always inflected the raised main verb correctly.36 We wish to propose that these 6 speakers have acquired an AGRP projection, as shown in (23).

Given the tree in (23), subject-verb agreement is expected to occur productively and correctly; this is the case for the 6 speakers at this stage. Given the higher functional head AGR, we now have a position for verbs to raise to; as we saw in Section 3, these 6 speakers have acquired verb raising (to the left of the VP).37 Table 2 provides the proportion of overt subjects in sentences with a raised verb.

DIRECT ACCESS TO X'-THEORY

287

Table 2. Overt Subjects at the AGRP Stage name

language

Mine Gabho Samran Emine Harva Ensook * t

T K K T T K

sentences w/ V in AGR* 173 108 127 109 76 62

overt subjectst 145 87 122 96 65 54

84% 81% 96% 88% 86% 87%

The sentences in this table do not include yes/no questions, imperatives or pro-drop sentences which would be acceptable in colloquial German. This figure includes subjects both preceding and following the verb in AGR.

Given the frequent occurrence of overt subjects with a raised verb, we can say that although the native languages of these speakers allow empty subjects, they have learned (without instruction) that German generally does not allow empty subjects. Recall that in the VP-sentences produced by these speakers, subjects are frequently omitted (cf. Figure 1 in Section 3). This suggests that the distribution of the overt subject is tied to the presence or absence of a functional projection. We will propose in Section 4.3 that the relevant functional projection is the AGRP projection. We claim, then, that the 6 speakers at this stage have acquired the full-fledged AGRP, together with subject-verb agreement, verb raising to AGR, and non-pro-drop. However, except for Harva and Ensook, it appears that the bare-VP structure is still available to some degree for the rest of the speakers at this stage.38 For all 6 speakers at this stage, we find some evidence for a CP projection. They each produce some wh-questions and yes/no questions,39 but very few or no embedded clauses with overt complementizers.40 Thus it is possible that in addition to the AGRP, these speakers have acquired the CP projection. However, their German would differ from native German in that both of their functional projections at the sentence level are head-initial; recall that embedded sentences in German have the finite verb at the end, and no such examples were found in our data. Further research is required on speakers at this stage to determine the status of the CP. The acquisition of the functional projections appears to differ for children and adults. The agreement paradigm has been acquired at our AGRP stage, together with a head-initial AGRP, but at a comparable stage for children, a head-final functional projection is found (cf. Clahsen 1990), demonstrated by the

288

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

finding that children's embedded clauses are head-final as soon as they appear (cf. Rothweiler 1990).41 In our data we do not find evidence for a head-final AGRP associated with embedded clauses; this does not rule out the possibility of a head-final AGRP at a stage later than those we investigate in this study (cf. Schwartz & Sprouse, this volume). Let us now turn to the 8 intermediate speakers in our study who have not yet acquired the agreement paradigm but who have some notion of verb raising. The data from these speakers may shed some light on how the AGRP develops in L2 acquisition. 4.3. The Second Stage: The FP Stage The remaining eight speakers fall between the VP stage and the AGRP Stage in terms of development. As we saw in Section 3, none of these speakers have acquired the agreement paradigm. In this section we will first concentrate on the 5 least advanced speakers in this group: the Turkish speakers Sevinc, Kemal, Kadir, Ahmet and Mehmet. 4.3.1. Verb Raising and Agreement A considerable proportion of the sentences uttered by these 5 speakers is still of the bare-VP type, as is the case at the earliest VP stage. Consider the following examples: (24) a.

b.

c.

d.

e.

Und dann hier Wohnungsamt anrufen. (Sevinc #69) and then here housing.authority up.call 'And then (I) call(ed) up (the) housing authority here.' Der deutsche Buch lesen. (Kemal #344) the German book read ' (I) read German book(s).' Immer jeden Tag fünfhundert Stück machen. (Kadir #342) always every day five.hundred unit make '(I) always make five hundred units every day.' Jetzt Hauptschule achter Klass Oberkassel weiter machen. now main.school eighth class Oberkassel further make '(She) is now continuing eighth grade (in) Oberkassel middle school/ (Ahmet #8) Diese auch Ofen ausschalten. (Mehmet #40) this also oven off.turn '(I) also turn this oven off.'

DIRECT ACCESS TO X'-THEORY

289

Although these speakers produce many VP sentences of the type exemplified in (24), they also produce a good number of sentences with a raised verb: Sevinc has a raised verb in 37 percent of his sentences, and Kemal, Kadir, Ahmet and Mehmet raise the verb 45-53 percent of the time (cf. Table D, Appendix). Consider the following examples of sentences with a raised verb: (25) a.

b.

c.

d.

e.

Jetzt brau Wohnungsamt fragen. (Sevinc #111) now need housing.authority ask 'Now (I) need to ask (the) housing authority.' Ich sehen Schleier. (Kemal #604) I see veil 'I see (the) veil.' Polizei komm zu mir ja. (Kadir #523) police come to me yes 'Yes, (the) police came to me.' Und dann mitnehmen vielleicht Wohnung. (Ahmet #61) And then with.take maybe apartment 'and then (I) maybe take (it) along (to my) apartment.' Mir machen nichts mehr. (Mehmet #103) me make nothing more '(They will) do nothing more (to) me.'

We conclude that since about half of the relevant sentences uttered by these speakers contain a verb in the VP (and no additional verb) and the other half contain a verb in the raised position, verb raising is optional for these speakers. Moreover, these speakers have clearly not acquired the full agreement paradigm. According to Table F (Appendix), none of these speakers used more than two of the five agreement suffixes in German (cf. Section 3 for a discussion of the criteria for acquiring agreement). Tables (G-l) and (G-2) in the Appendix provide information on the individual suffixes; most of the time these speakers used either the suffix -n or no suffix. When other suffixes are used, they are usually used incorrectly. While it appears that these speakers may be in the process of acquiring the agreement paradigm, they have not yet mastered it. Since these speakers have not acquired the full agreement paradigm and verb raising is possible but not obligatory for them, we propose that they have acquired a functional projection without agreement features. We will refer to this functional projection as FP (for 'Finite Phrase'), following Clahsen (1991) who proposes such an underspecified, head-initial functional projection for a comparable stage in the first language acquisition of German. The German FP tree is shown in (26).

290

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

The head of FP provides a position for verbs to raise to, including modals and auxiliaries (cf. Tables (H-l) and (H-2) for listings of modals and auxiliaries in these speakers' data). Since the agreement paradigm has not been mastered, the verbs that raise to F are often not correctly marked for agreement; i.e., the notion of 'finiteness' cannot be based solely on agreement marking for these speakers. Note that the L2 data points to a difference from first language acquisition in this respect, since in German children's data, verbs which are not marked for agreement do not raise (cf. Clahsen & Penke 1991). Since the functional projections in Turkish would have to be head-final (cf. Introduction) and since the corresponding projection in German is assumed to be head-final (in order to account for the verb-final order in embedded clauses), it is not immediately obvious how the head-initial FP is acquired by these speakers. We might want to assume that these speakers have the tree in (27), where the verb is raised to COMP (as in the standard analysis of German).

However, we find a problem with this tree: namely, that we are not at all convinced by the data from these speakers that they have acquired a full-fledged CP structure. None of the 5 speakers produce any instances of embedded clauses with an overt complementizer.42 However, a good number of examples of the following coordinating conjunctions were found: und 'and', aber 'but', oder, 'or'

DIRECT ACCESS TO X'-THEORY

291

and ama (Turkish) 'but'. Furthermore, none of these speakers produced any yes/no questions or wh-questions involving a raised verb and an overt subject which cannot be construed as formulaic. There is thus no support for the existence of a CP at this stage — assuming that once structural positions have been acquired, their existence is reflected in the production data. We will therefore continue to assume that the functional projection acquired by these 5 Turkish speakers is an underspecified FP which — once agreement features have been specified — turns into an AGRP, resulting in the AGRP stage (cf. Section 4.2 above). 4.3.2. Empty Subjects The remaining 3 speakers to be analyzed are Özgül (a Turkish speaker), and Park and Dosik (Korean speakers). These speakers pattern with the 5 speakers discussed in the previous section in terms of the agreement paradigm, which they clearly have not acquired (cf. Table F in the Appendix). Furthermore, the pattern of verb raising for these 3 speakers can still be considered to represent optional verb raising; the verb remains in the VP 37—41 percent of the time for them (cf. Table D in the Appendix). We will therefore assume that Özgül, Park and Dosik are still at the FP stage, rather than at the AGRP stage (which is represented by a productive agreement paradigm and obligatory verb raising; cf. Section 4.2). The data from the distribution of null subjects, to be discussed in the following, also supports the assumption that these speakers are at the FP stage.43 Recall that based on the Implicational Table in Section 3, most of our speakers use an overt subject in sentences with a raised verb at least 60 percent of the time (while overt subjects are much less frequent in VP sentences, for all speakers; cf. Figure 1, in Section 3.1). At the AGRP stage, as discussed in Section 4.2, the proportion of overt subjects in the relevant sentences is even higher (over 80 percent). When studying the data from the FP stage in more detail, we observe an interesting phenomenon: whenever the verb is raised, there is a strong tendency for the position preceding the raised verb (i.e., in the Spec-FP position) to be filled, whether with a subject, a locative phrase, or with some other argument or adjunct.44 The generalization holding at the FP stage seems to be the requirement that some XP fill the Spec-FP position, rather than the requirement that a subject occur in the sentence; the latter, on the other hand, holds at the AGRP stage. The graph in Figure 2 gives the proportion of XPs preceding the raised verb, and the

292

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

proportion of overt subjects in sentences with a raised verb (refer back to Table 1 for speakers' numbers). Except for Dosik (#9), the graph in Figure 2 shows that the speakers at the FP stage (speakers 4—11) typically have an overt XP in the Spec-FP position. Since the XP is most often the subject, this indirectly results in the fairly high proportion of overt subjects shown on the graph, even though the null subject parameter presumably has not yet been set. We therefore propose that the null subject parameter is only set in connection with the acquisition of the AGRP projection; this projection is then responsible for the acquisition of the agreement paradigm, for the setting of the null subject parameter — resulting in obligatory overt subjects (over 80 percent of the time) — and in very frequent verb raising. As we have seen, this pattern is found in the data from the 6 most advanced speakers we studied (cf. Section 4.2).

Figure 2. Verb Raising and the Specifier Position While Dosik's distribution of overt subjects suggests that he is already at the AGRP stage, he has not yet acquired the agreement paradigm (cf. Table F; Appendix), and only 59 percent of his sentences contain a verb outside of the VP. Given this, we must assume that he is still at the FP stage, and that his

DIRECT ACCESS TO X'-THEORY

293

proportion of overt subjects is an anomaly, possibly due to the small amount of data collected from him. If the null subject parameter is only set at the AGRP stage, why should the Spec-FP position already be filled by some XP at the FP stage? We propose that this is due to a general requirement on the licensing of positions, which we have termed the 'Full House Principle'.45 (Vainikka 1989; cf. also Zwart 1991). According to the Full House Principle, a syntactic position needs to be licensed by the occurrence of an element in it (at some level of representation). This principle holds in UG, and does not need to be acquired either by adults or children. At the FP stage, an F-position is licensed if a verb occurs in F; a Spec-FP position is licensed if an XP occurs in that position. At the AGRP stage, an AGR-position is again licensed by a verb in AGR, and the Spec-AGRP position is licensed by the subject-NP.46 Given the above, we find the following sequence of development with respect to empty subjects. At the earliest stage (the VP stage), the Spec-VP position is optional (cf. discussion in Section 4.1), and therefore overt subjects are optional (whenever a subject occurs, it licenses a Spec-VP position; otherwise the Spec-VP position is not projected). At the FP stage, overt subjects frequently co-occur with raised verbs, but we have suggested that this is an indirect result of the requirement that the Spec-FP position be filled by an XP. At the AGRP stage, the null subject parameter is set at the non-pro-drop value, and subjects are obligatory with raised verbs.

5.

Discussion

The evidence we have presented in this paper reveals that adults who are learning German without formal instruction transfer the headedness of the lexical projection VP from their native language. We have argued that at the earliest stage the grammar of our L2 speakers consists solely of this transferred lexical projection. The data further reveals that adults do not transfer the headedness of higher functional projections, such as IP or AGRP (and possibly CP). What our data indicates is that the learners determine the headedness of functional projections and the characteristics of the specifier position based on the input and on syntactic principles such as X'-theory and the Full House Principle. To the extent that headedness of functional projections involves a parameter, the adults at our FP stage and AGRP stage have reset a parameter to a value not found in their first language. Furthermore, the speakers at the AGRP stage have acquired obligatory verb raising and non-pro-drop in German, even though their native

294

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

languages, Turkish and Korean, differ from German in both respects. Both the Korean speakers and the Turkish speakers at the AGRP stage were shown to have acquired the agreement paradigm. For the Korean speakers this involves a parameter setting different from that in their native language, which does not mark subject-verb agreement. Moreover, even though Turkish does mark subject-verb agreement, this gives the Turkish speakers no apparent advantage in acquiring agreement in German; we find no differences between the Turkish data and the Korean data in this respect. As a summary of the data and the stages presented in Section 4, we provide here a revised implicational table, based on the preliminary table given in Section 3. Table 3. Revised Implicational Table name

stage

Aysel Memduh Changsu Sevinc Kemal Kadir Ahmet Mehmet Dosik Park Özgül Mine Gabho Samran Emine Harva Ensook

language

VP VP VP FPI FPI FPI FPI FPI FPU FPU FPU AGRP AGRP AGRP AGRP AGRP AGRP

T T K T T T T T K K T T T K T T K

head-final VP + + + + + + + + + + + + + + + + +

no null subjects (+)* (+) (+) (+) (+) + (+) (+) + + + + + +

verb raising -

(+)t (+) (+) (+) (+) (+) (+) (+) + + + + + +

*

Parentheses around the *+* indicate that the Spec(FP) is typically filled.

t

Parentheses around the '+' indicate that verb raising is optional.

agreement paradigm

+ + + + + +

As Table 3 shows, the acquisition of the agreement paradigm and the setting of the null subject parameter coincide in our data; furthermore, at the same point an increase in verb raising is witnessed. The table also shows that the acquisition of (optional) verb raising coincides with the requirement that an XP occupy the

DIRECT ACCESS TO X'-THEORY

295

Spec-FP position, the precursor of non-pro-drop in our analysis. The revised implicational table yields the following implicational scale, apart from one of Dosik's values (cf. Section 4.3.2): (28)

Agreement/Obligatory V-Raising/Non-pro-drop -> Optional V-Raising/Filled Spec-FP -> Head-final VP

That is, if any of the syntactic processes studied here have been acquired by a speaker, then at least the German head-final VP has been acquired. Furthermore, the acquisition of the agreement paradigm (and the setting of the null subject parameter) implies that (optional) verb raising has been acquired. We have observed that the development of phrase structure in second language acquisition follows a pattern noted in first language acquisition, where the language learner appears to start off with a bare lexical projection, then posits an underspecified functional projection, and finally specifies the features of the functional projection. That is, in both first and second language acquisition, the learner posits minimal trees based on the input, using principles of UG as a guideline. In addition, for both groups of learners, the acquisition of agreement, non-pro-drop and obligatory verb raising coincide. The parallelism between the first and second language development of phrase structure provides evidence against the position that adults have no access to the parameters of UG, or that they only have indirect access through the parameter settings of their first language. However, we also find some differences between the acquisition of the functional projections by adults and by children. Adults and children seem to arrive at different conclusions about the headedness of AGRP; unlike children (cf. Clahsen 1991), our adult second language learners posit a head-initial AGRP, not a head-final one. This is especially surprising for the Turkish speakers, given the possibility that they could transfer a head-final IP/AGRP from Turkish, and given that the German input (i.e., embedded clauses with a finite verb at the end) provides further confirming evidence that this functional projection is head-final in German. While it is possible that German children also undergo a brief stage with a head-initial AGRP before the acquisition of the CP, further research on the first language acquisition of the AGRP and CP in German is required to determine whether there really is a difference between first and second language acquisition in this respect. A clearer difference between adults and children is that while children never raise verbs with the default suffix -n (according to Clahsen & Penke 1991), adults frequently raise verbs with this default suffix to F and even — to a lesser degree — to AGR. But perhaps the most salient difference between adults and

296

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

children is that to a large degree, the adults in our study retain the 'old' trees in their grammar; for any given speaker at the FP stage, a large proportion of the sentences still consist of only the bare VPs from the previous stage. While this phenomenon has also been attested in children's data, it is not as predominant (cf. e.g., Clahsen & Penke 1991). At this point we cannot offer any reasons as to why the phenomenon of what we will term 'stage seepage' should be much more typical of (adult) second language acquisition than first language acquisition.47 What are the triggers in the input which motivate our learners' progression from one stage of acquisition to the next? The answer to this question may provide answers to why adults raise verbs with an -n and why they acquire a head-initial rather than head-final AGRP. For children learning German as their first language, the trigger for the acquisition of AGRP is claimed to be agreement (cf. Clahsen 1991); however, it is not clear whether agreement operates in a similar fashion for adults. We suggest that the trigger for the acquisition of FP by adults is word order in terms of the head-complement parameter (Chomsky 1986), whereby if the parameter has been set for the head-final value in VP, as we argue, then any other position of the verb must involve a verb outside of the VP. Conversely, the trigger for children's acquisition of the FP has been proposed to be a feature (possibly agreement-related) connected with the suffix -t (cf. Clahsen & Penke 1991). If for adults word order rather than agreement features is involved in the acquisition of the FP, it is not surprising that verbs with -n are raised. Thus, if we assume that word order is, in fact, the trigger, then we have an explanation for the greater frequency with which adults raise verbs with -n. Once the FP is created, some notion of agreement features clearly starts to come in for the adults, but the causal link may well be the reverse from what has been observed for children. If the trigger for the AGRP is not agreement, what might be responsible? We suggest that for adults the trigger for AGRP are auxiliaries; since the various forms of the auxiliary sein 'be' are particularly distinct from each other (as in English), this may provide a handle for the acquisition of the agreement paradigm without recourse to the less salient agree­ ment suffixes. Note that even at the AGRP stage many of the speakers do not raise main verbs very frequently, with only the most advanced raising main verbs more than half the time (cf. Section 4.2). If agreement were a trigger for the acquisition of AGRP, we would expect all main verbs to be raised at this stage. Finally, why do adults acquire a head-initial AGRP rather than a head-final AGRP as children presumably do? If we take the trigger for AGRP to be auxiliaries for adults, these triggers would occur more often at the beginning of

DIRECT ACCESS TO X'-THEORY

297

the sentence, in matrix clauses, rather than at the end, in embedded clauses. Even at the FP stage children have to analyze a sentence-final verb with a -st suffix (as in daß du kommst 'that you come') as a sentence-final finite verb, providing evidence for a head-final functional projection, i.e., the head-final AGRP. However, for adults at the FP stage who have acquired verb raising, but no agreement features, such examples would be analyzed as having a non-raised verb. Thus there is less evidence available for the adults than for the children for a head-final AGRP. Once the agreement paradigm has been acquired by the adults, we would expect an eventual reanalysis of the headedness of the AGRP with the discovery that the final verbs in embedded clauses are marked for agreement. We can summarize our proposal for the elements which serve as triggers for adults by excluding those elements which have been observed to act as triggers for children, namely functional elements realized as affixes (in this particular instance, as suffixes). In other words, elements which can be classified as free morphemes appear to operate as triggers for adults. One avenue of investigation which may shed further light on the nature of triggers adults use is that of the relationship between the acquisition of functional elements and prosodic structure (cf. Demuth 1992 on L1 acquisition). For example, the elements we have proposed as triggers in German for adults would generally constitute at least a metrical foot, while those proposed for children are elements with less prosodic weight. It is uncontroversial that the L2 acquisition of phonological structure is considerably more problematic than the acquisition of syntactic structure; the explanations we seek may lie in the extent to which problems with certain aspects of phonological structure impinge on the acquisition of various functional elements. Assuming that adults have access to the principles and parameters of UG does not entail the assumption that they have available to them the same triggers as children do. However, adults still must find a way to set the parameters through positive evidence. From our data, it is clear that adults do find a way; the triggers chidren employ are not the only possible set of triggers. That this may lead to slight differences in the progression from one stage to the next is aptly illustrated by our Korean and Turkish learners of German.

298

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

Appendix Table A. Biographic Information and Source of Data (Alphabetical Order) name

language

sex

age

length of res.

data source

Ahmet Aysel Emine Harva Kadir Kemal Mehmet Memduh Mine Özgül Sevinc Changsu Dosik Ensook Gabho Park Samran

Turkish Turkish Turkish Turkish Turkish Turkish Turkish Turkish Turkish Turkish Turkish Korean Korean Korean Korean Korean Korean

M F F F M M M M F F M F M F M M F

52 43 28 36 36 37 55 47 42 45 34 60 34 41 38 38 35

22 years 11 years 6 years 6 years 11 years 11 years 24 years 9 years 22 years 17 years 9 years 6 years VA years 4 years 13 years 13 years 3 years

Lexlern Lexlern Lexlern Lexlern von Stutterheim* von Stutterheim Lexlern von Stutterheim Lexlern Lexlern von Stutterheim Lexlern Lexlern Lexlern Lexlern Lexlern Lexlern

* Some of the names of the learners from the von Stutterheim corpus have been changed for ease of presentation.

DIRECT ACCESS TO X'-THEORY

299

Table B. Breakdown of Analyzed and Unanalyzed Utterances with Verbs (Alphabetical Order) name

lang.

Ahmet Aysel Emine Harva Kadir Kemal Mehmet Memduh Mine Özgül Sevinc Changsu Dosik Ensook Gabho Park Samran

T T T T T T T T T T T K K K K K K

total # w/verbs 97 138 198 106 342 482 157 259 351 136 204 103 116 84 185 133 256

SV sentences 12 *20 27 12 65 101 14 51 46 15 30 13 15 3 13 15 31

12% 14% 14% 11% 19% 21% 9% 20% 13% 11% 15% 13% 13% 4% 7% 11% 12%

other unclear utterances 13 13% 43* 31% 12 6% 3 3% 25 7% 77 16% 25 16% 33 13% 38 11% 4 3% 24 12% 9% 9 10 9% 4% 3 13 7% 9 7% 24 9%

total sentences w/V, analyzed 72 74% 54% 75 80% 159 86% 91 74% 252 304 63% 118 75% 175 68% 76% 267 117 86% 74% 150 81 79% 78% 91 78 93% 159 86% 82% 109 201 79%

* Of Aysel's 63 unclear utterances with a verb, 59 are verb-final; 27 of the utterances under 'other unclear' have the form 'die + V'.

300

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

Table C. Position of the Verb in the VP* (Developmental Order) name Aysel Memduh Changsu Sevinc Kemal Kadir Ahmet Mehmet Dosik Park Özgül Mine Gabho Samran Emine Harva Ensook

level I I I I I II II II II III III IV IV IV IV IV IV

language Turkish Turkish Korean Turkish Turkish Turkish Turkish Turkish Korean Korean Turkish Turkish Korean Korean Turkish Turkish Korean

total XV and VX 55 129 43 75 146 138 33 58 39 55 53 119 71 73 68 16 17

XV (% total) 55 127 41 62 118 117 30 56 37 45 45 103 60 67 57 14 16

100% 98% 95% 83% 81% 85% 91% 97% 95% 82% 85% 87% 85% 92% 84% 88% 94%

VX (% total) 0 2 2 13 28 21 3 2 2 10 8 16 11 6 11 2 1

0% 2% 5% 17% 19% 15% 9% 3% 5% 18% 15% 13% 15% 8% 16% 12%

6%t

* Thefiguresin this table include 'bare VPs' — e.g. Milch kaufen 'milk buy' — and sentences with a verb to the left of the VP followed by a VP containing a verb — e.g. Jetzt möchte Milch kaufen 'Now want milk buy' — they do not include SXV or SVX types with just one verb. t Ensook also produced 10 sentences with a sentence-initial copula is(t) 'is' which were not included in the VX count.

DIRECT ACCESS TO X'-THEORY

301

Table D. Bare-VP Sentences and Sentences with More than a VP (Developmental Order) name Aysel Memduh Changsu Sevinc Kemal Kadir Ahmet Mehmet Dosik Park Özgül Mine Gabho Samran Emine Harva Ensook

level I I I I I II II II II

m m IV IV IV IV IV IV

language total sentences with verbs Turkish Turkish Korean Turkish Turkish Turkish Turkish Turkish Korean Korean Turkish Turkish Korean Korean Turkish Turkish Korean

75 175 81 150 304 252 72 118 91 109 117 267 159 201 159 91 78

verb in the VP 66 151 68 95 143 139 39 61 37 44 43 90 41 64 39 15 8

88% 86% 84% 63% 47% 55% 54% 52% 41% 40% 37% 34% 26% 32% 25% 16% 10%

verb outside of the VP 9 24 13 55 161 113 33 57 54 65 74 177 118 137 120 76 70

12% 14% 16% 37% 53% 45% 46% 48% 59% 60% 63% 66% 74% 68% 75% 84% 90%

302

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

Table E. Empty Subjects (Developmental Order) name

level

Aysel Memduh Changsu Sevinc Kemal Kadir Ahmet Mehmet Dosik Park Özgül Mine Gabho Samran Emine Harva Ensook * † ‡

I I I I I II II II II III III IV IV IV IV IV IV

lang.

T T K T T T T T K K T T K K T T K

total sent. with raised verb 9 24 12 51 139 103 33 56 50 62 72 173 108 127 109 76 62

overt subj. (of raised V sent.)t 4 44% 18 75% 11 92% 24 47% 78 56% 74 72% 24 73% 41 73% 45 90% 47 76% 55 76% 84% 145 87 81% 122 96% 88% 96 65 86% 54 87%

total sent. with overt subject 15 45 39 58 112 109 36 48 52 56 62 168 92 136 103 75 57

raised V (of) subj. sent.)‡ 4 18 11 24 78 74 24 41 45 47 55 145 87 122 96 65 54

27% 40% 28% 41% 70% 68% 67% 85% 87% 84% 89% 86% 95% 90% 93% 87% 95%

The sentences in this table do not include yes/no questions, imperatives or pro-drop sentences which would be acceptable in colloquial German. This figure includes subjects both preceding and following the verb. This column shows, given all sentences with an overt subject, how many of these also have a verb outside of the VP.

DIRECT ACCESS TO X'-THEORY

303

Table F. Agreement on Raised Main Verbs (Developmental Order) name

level

Aysel Memduh Changsu Sevinc Kemal Kadir Ahmet Mehmet Dosik§ Park§ Özgül Mine Gabho Samran Emine Harva Ensook

I I I I I II II II II III III IV IV IV IV IV IV

lang.

total raised suffix -n main verbs (% correct)

T T K T T T T T K K T T K K T T K

9 14 5 46 100 51 26 35 40 16 33 106 43 25 35 40 29

5 8 1 29 57 37 20 13 5 3 16 50 13 5 6 8 9

[3]† 13% [0] 14% 9% 16% 20% 8% [0] [1] 6% 22% 15% [4] [3] [4] [6]

default -n of total* [2] 50% [1] 54% 52% 61% 62% 34% 13% 13% 45% 37% 26% 4% 9% 10% 10%

# of agr. other suffixes suffixes (correct) used 4 [0] 0 6 [0] 0 4 [2] 1‡ 17 53% 2 2 43 65% 2 16 75% 6 [3] 1 22 82% 2 35 2 88% 13 92% 1 2 17 71% 56 4 88% 30 4 97% 20 95% 5 5 29 97% 32 4 97% 20 95% 5

*

These figures represent the proportion of incorrect or unclear -n's..

‡

Whenever the numbers involved were smaller than 10, we did not calculate a percentage; the raw numbers are given in brackets.

‡

This figure indicates how many different agreement suffixes we found at least 2 correct instances of for each speaker.

§

Although the percentages for Park and Dosik were quite good, there was not enough evidence that the agreement paradigm had been acquired, since there were only 1 or 2 suffixes that they produced at least 2 correct instances of.

304

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

Table G-1. Distribution of '-n' and -Ø' on All Verbs (Early FP Stage) name

total # of verbs

Sevinc Kemal Kadir Ahmet Mehmet

150 304 252 72 118

correct 7+10 12+23 15+24 8+1 1+5

total

-n + Ø wrong

unclear

45+4 139+6 69+2 44+2 45+1

51+12 40+10 75+7 4+0 23+2

129 230 192 59 77

86% 76% 76% 82% 65%

Table G-2. Distribution of-e (ISG) and -t (3SG) on Main Vs (Early FP Stage) name

total # of main verbs

Sevinc* Kemal* Kadir Ahmet Mehmet

140 257 197 64 100

correct 3+5 3+8 2+6 1+2 l+17t

-e + t wrong

unclear

total

0+2 0 0 2+0 1+0

0+4 0+5 0+4 0 0+3

8 11 8 3 18

*

In addition to the suffixes listed in (G-2), Sevinc and Kemal each produced 1 correct and 1 unclear instance of the 2SG -st on main verbs; the other 3 speakers produce no instances of this suffix on main verbs.

t

Mehmet's 17 correct -t's consist of the following: 12 kommt 'comes', 2 geht 'goes', 1 bleibt 'stays', 1 passiert 'happens', 1 macht 'makes'. According to our criteria, 2 of these are in the VP and 15 are raised.

DIRECT ACCESS TO X'-THEORY

305

Table H-1. Distribution of 'sein' (to be) (Early FP Stage) name Sevinc Kemal Kadir Ahmet Mehmet

ist

bin

bist

warlwär

sind

3SG

ISG

2SG

I/3SG PAST

1/3PL

8 27

3

1 1

6

1

5 2 1 6

total 5 14 31 1 14

3

1

Table H-2. Distribution of Modals (Early FP Stage) name Sevinc Kemal Kadir Ahmet Mehmet * t

muss* 5 4 8

möchte(n) 12 5

1

wolle(n)

OTHERt 1 1

3 2

total 6 17 8 8 3

müssen 'must'; möchten 'would like'; wollen 'want' Sevinc has 1 instance of brau (brauch) 'need'; Kemal has l instance of müssen 'must' (plural); Mehmet has 1 instance of sollen 'should' (plural) and 1 of können 'can' (plural).

306

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

Table I-1. Proportion of-n and Other Suffixes on Raised Main Verbs (AGRP Stage) name Mine Gabho Samran Emine Harva Ensook

total raised main verbs 106 43 25 35 40 29

-n

% correct

50 13 5 6 8 9

22% 15% 80% 50% 50% 67%

other

% correct

56 30 20 29 32 20

88% 97% 95% 97% 97% 95%

TableI-2.Agreement Suffixes (Except for -n) on Raised Main Verbs (AgrP Stage) name

Mine Gabho Samran Emine Harva Ensook

total raised main verbs 106 43 25 35 40 29

corr/wrong+ unclear -t

corr/wrong+ unclear -e

corr/wrong+ unclear -st

corr/wrong+ unclear

11/3 8/0 5/0 13/0 9/1 13/0

17/2 18/0 9/0 5/0 7/0 2/0

1/0 0/1 2/0 3/1 0/0 2/0

20/2 3/0 3/1 7/0 15/0 2/1

0

DIRECT ACCESS TO X'-THEORY

307

Acknowledgements The results reported in this paper were obtained in the Lexlern Project at the University of Düsseldorf, DFG Grant C197/1-1 (directed by Harald Clahsen) while the authors were working for the project. We wish to thank everyone on the project, as well as Monika Rothweiler, Bonnie D. Schwartz, two anonymous reviewers and the Psycholinguistics Group at the University of Massachusetts, Amherst for comments on earlier versions of the paper.

Notes 1.

All other examples given for Turkish contain the zero allomorph for the third person singular.

2.

Strictly speaking it is true that Korean generally lacks subject-verb agreement; however, the honorific form of the verb could be considered to mark subject-verb agreement.

3.

Although for the purposes of determining developmental stages it would be preferable to have longitudinal data, such data was not available to us. Longitudinal data does exist (from the ESF Project) from 4 Turkish speakers learning German; however, the data was not useful to us since these learners had received some instruction and were also too advanced for our purposes.

4.

We wish to thank Christiane von Stutterheim for making her tapes and transcripts available to us. (For ease of presentation we used different names for several of von Stutterheim's test subjects.) The comparability of von Stutterheim's data with the data we collected ourselves might be questioned on the basis of use of different elicitation tasks. While we employed a greater variety of techniques than did von Stutterheim, like her data our data also consists of a good number of utterances in an interview context. Furthermore, because our test subjects showed little evidence of sophisticated metalinguistic awareness, the tasks (i.e., the grammaticality judgement cloze tasks) which had the potential to elicit qualitatively different data failed to do so. What the elicitation techniques we used enabled us most effectively to do was to clarify the position of the verb in the VP as head-final. Von Stutterheim's data exhibits no qualitative differences in this respect.

5.

One way in which formal instruction might affect acquisition is through the provision of explicit evidence in the form of corrective feedback for parameters which have been incorrectly set (cf. White 1991; Schwartz 1993). Since the theory of parameter setting states that parameters are set by positive evidence alone, the possibility that learners receive and make use of negative evidence adds a variable which could — if negative evidence actually has an influence on competence — render the context of acquisition very different from that for children.

308

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

6.

Some of the grammaticality judgement tasks were designed as pilot studies for actual quantifiable experiments. However, too few speakers from each stage underwent the experiments to allow quantification of the results. To the extent that it is possible to interpret the results, they appear to support our analysis of the elicited production data and the spontaneous data.

7.

Under the standard analysis of German, the finite verb raises to COMP. However, we do not wish to assume that these speakers have the target language analysis of verb raising from the beginning of acquisition. We will return to the question of which functional head the verb raises to.

8.

Unanalyzable sentences, or group (iii), constituted roughly one-fourth of the relevant sentences. This includes all the sentences consisting of just a subject and a verb (SVsentences). Of all the sentences 4-21 percent were of this type; the individual proportions are given in Table B (Appendix). The remaining unanalyzable sentences constituted 4-16 percent of the total, except for Aysel (cf. Table B for details).

9.

We assumed, following work in syntactic theory (cf., e.g., Zagona 1982; Kitagawa 1986 and Sportiche 1988) and in L1 acquisition (Deprez & Pierce, this volume; Vainikka 1992) that subjects may be base-generated in the VP. We will propose that, as in L1 acquisition, the position of the subject at the earliest stage of L2 development is the Spec-VP position.

10. Although negation has been used as an indicator of verb placement in the L1 acquisition of German (cf. Clahsen 1988b and Lebeaux 1988) it is not clear that negation is analyzed consistently across the various stages of L2 acquisition. Due to potential variability in negative placement, we considered negation an unreliable test of verb placement in the L2 data. 11.

Using 60 percent as the cut-off point is not crucial, but it allows us to make distinctions regarding speakers whose German is not very advanced. This in turn enables us to argue for the validity of using our data as developmental data, while providing an initial idea of which of the speakers belong to which developmental stages (to be refined in Section 4). In Table 1, the results would be identical for the columns 'head-final VP' and 'agreement paradigm' if a higher cut-off point such as 80 percent had been used (cf. Tables C and F in the Appendix). If the 80 percent cut-off point had been used for verb raising, only the two most advanced speakers would qualify as having acquired this (cf. table D in the Appendix); however, we will argue in Section 4 that even speakers with less raising have already acquired a higher functional projection into which the verb can raise. The 60 percent cut-off point distinguishes between speakers who clearly prefer to raise the verb and speakers who show no preference. Similarly, for null subjects this cut-off point distinguishes speakers who prefer to use an overt subject from those for whom having an overt subject is clearly optional (cf. table E in the Appendix).

DIRECT ACCESS TO X'-THEORY

309

12.

For this implicational table, the coefficient of scalability is 1.0 (cf. Hatch & Farhady 1982) since there are no errors; i.e., our data is highly scalable: Crep = 1 - 0/(17)(4) = 1.0 MMrep = (4)(17)/(4)(17) = 1.0 coefficient of scalability = 0/1-1 = 1.0

13.

Given the grammar of German, including the 'VX' types biases the data somewhat against us, since some of the VX sentences could result from verb raising. However, such examples were too few to skew the results.

14.

In fact, over 80 percent of the VPs were head-final, for each speaker.

15.

The second criterion ensures that we are only including speakers for whom there is a correlation between an overt subject and a raised verb (i.e., overt subjects in VP-sentences are not relevant for this measure). In addition, using the first criterion alone would include Changsu and Memduh, who have so few examples of raised verbs that their percentages may not be reliable; for example, 11 of Changsu's 12 sentences of this type contain both a raised verb and a subject, but we did not consider this a sufficient indication that Changsu has acquired the non-pro-drop value of the null subject parameter.

16.

Due to the small number of sentences with a raised verb, the first three speakers are excluded from this graph. Furthermore, given the small number of bare-VP sentences for the two most advanced speakers, their proportions of empty subjects in VP sentences are not reliable (and are therefore not given in the graph). See tables D and E in the Appendix for the individual values.

17. This suggests that the null subject parameter and the Extended Projection Principle should be stated in terms of a functional projection, such as an IP (i.e., 'there must be a subject in the Spec(IP) position') rather than in terms of whether or not the maximal projection contains a predicate (cf. Chomsky 1986:116). 18.

We will return to the question of whether agreement morphology occurs on the verbs in the VP.

19. Although modals and auxiliaries in German exhibit more agreement than they do in English, we excluded modals on the grounds that ISG and 3SG forms of modals are identical in German (e.g., kann 'can' is both the ISG and the 3SG form). Regarding our exclusion of auxiliaries, we do not find L2 speakers using the infinitival forms, especially sein 'be', unlike with main verbs. Given this, the lexical entry of 'be' may be ist 'is' and use of this 3SG form would not reflect the acquisition of the productive agreement system. 20.

German has the following five subject-verb agreement suffixes in the present tense: -e (ISG), 0 (ISG colloquial), -st (2SG), -t (3SG and 2PL) and -n (IPL and 3PL).

21.

This suggests that something other than the stranding of an agreement suffix must be responsible for verb raising. In order to account for NP movement and verb raising in Finnish, Vainikka (1989) proposes that there is a general motivation for the raising of

310

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN both heads and arguments: in order to be licensed, each syntactic position has to be filled at some level of representation; this results in raising from the VP into functional projections. We will return to this in Section 4.

22.

Aysel's data was collected by us; Memduh's data comes from the von Stutterheim corpus.

23.

The very high proportion of VP sentences in the data for Aysel, Memduh and Changsu — about 85 percent — as opposed to the figures for Sevinc and Kemal (63 percent and 47 percent, respectively) suggests that only the first three speakers can truly be considered to be at the earliest stage (cf. Table D in the Appendix). In addition to the higher proportion of raised verbs, their acquisition of modals and auxiliaries also suggests that Sevinc and Kemal should be considered to be at a more advanced stage; we will return to a discussion of these speakers.

24.

Although Clahsen & Muysken claim not to have found SOV orders in the Turkish data they examined, they do allow that "Turkish learners may perhaps postulate an OV order in early stages" adding that "surprisingly enough more advanced Turkish speakers show the [SVO] pattern (...) as well." (Clahsen & Muysken 1986:104). The SVO patterns were attributed to the operation of the canonical word order strategy and the instances of OV as involving a topicalized object. However, since we found a good number of clearly SOV structures in our data, Clahsen & Muysken's analysis is no longer descriptively adequate. Clahsen (1990) has since modified his position to include the possibility of access to UG through the learner's L1, which would allow for the transfer of headedness of the VP from the L1 to the L2.

25.

Case assignment to subject and licensing of a null subject are typically stated in terms of INFL (cf., e.g., Rizzi 1986); if INFL is lacking, it is not clear what the predicted consequences are for the form and distribution of subjects; we will return to this.

26.

The suffix -n is the infinitival suffix in German as well as the IPL and 3PL suffix; we assume that verbs occur in this form in the lexicon, at least for the L2 learners.

27.

We are assuming that the two instances of komm 'come' which were found in Aysel's data represent forms in which the -n suffix has undergone progressive nasal assimilation. Such assimilation is common in colloquial German (cf. Hall 1992).

28.

Memduh produces one incorrect -e suffix (ISG). Since the colloquial form of the ISG involves a zero allomorph, we might think that Memduh has acquired the ISG form of the verb. However, the distribution of his -0 forms suggests that this is not an agreement form for him: of the 41 such forms, 12 are used correctly, 7 incorrectly and 22 are unclear (i.e., we were unable to determine based on the context which person/number he was referring to).

29.

Changsu produced 8 -e suffixes (all correct) 5 0 suffixes (4 incorrect), and 7 -t suffixes (6 correct) in her VP-sentences.

30.

Since Memduh also produced a lot of -0 forms for main verbs, the lack of the -n suffix in (16) is not a sufficient argument for claiming that muß 'must' occurs in a headfinal INFL.

DIRECT ACCESS TO X'-THEORY

311

31.

Consider Lebeaux's (1988) proposal that the L1 learner needs to make a parametric decision about whether or not arguments are optional in the language she or he is acquiring; this proposal is based on the analysis of adult Warlpiri by Jelinek (1984), according to which full-NP arguments are optional in this language. We suggest that the unmarked value of this parameter involves the optional omission of arguments.

32.

Five of the nine sentences have an empty subject. All of the nine verbs have either the -n suffix or no suffix. There are no modals, auxiliaries, complementizers, wh-questions or yes/no questions.

33.

The verbs that occur without a suffix in Memduh's sentences with a raised verb are komm 'come' and arbeit 'work'. However, the morphological absence of the default suffix may only be apparent, being due to phonological factors. In addition to the likelihood that the -n undergoes assimilation after the [m] in komm 'come' (see note 27), there may be prosodic constraints involved for L2 learners in the early stages of acquisition which result in the deletion of an unstressed syllable (i.e., either syllabic [n] or [n] preceded by a schwa) when it follows a syllable receiving secondary stress, as in ar.beit.en 'work' (cf. Demuth 1992, who addresses this in first language acquisition). In addition to -n and -0, Memduh also produced one example with a raised verb bearing the -t suffix; this example, given in (i), is potentially idiomatic in that it is completely well formed: (i) Alles hier tut weh. (Memduh #397) everything here makes hurt 'Everything here hurts.'

34.

Following the approach of Schwartz & Sprouse (this volume), we might assume that these speakers have transferred a head-final IP and CP from Turkish/Korean. This would be a more plausible idea if we found some indication of INFL material or COMP material at the end of the sentence, but there does not seem to be much available. Changsu's verbs with the -t suffix in the VP might be examples of a head-final IP, but since we also found instances of the -t suffix in her sentences with a verb preceding the VP, this is not very convincing evidence for a head-final IP at this early stage.

35.

All of the data from these speakers were collected by us in the LEXLERN Project.

36.

In addition to showing that they have acquired the paradigm, the data from Samran, Emine, Harva and Ensook suggest that these four speakers no longer use the default -n (cf. Table I-1 in the Appendix).

37.

That is, over 60 percent of their sentences have a raised verb (cf. Table D in the Appendix). Except for Mine and Samran, verb raising occurs at least 74 percent of the time.

38.

In Ensook's and Harva's data, verbs are raised at least 84 percent of the time, as opposed to 75 percent or less for the other speakers. Furthermore, Harva and Ensook raise the main verb (as opposed to modals or auxiliaries) to AGR more often than the other speakers: 77 percent (40 of 52) of Harva's main verbs occur in AGR; the

312

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN corresponding figure for Ensook is 78 percent (29 of 37). For all other speakers beyond the VP stage, only 28-57 percent of the main verbs have been raised; modals and auxiliaries occur in the raised position more frequently, possibly because they may have been base-generated in this position.

39.

They each produce 1-7 examples of wh-questions, as exemplified here: (i) a. Warum de ich gehen so heute später so Schule? why I go so today later so school 'Why am I going to school later today?' (Mine #210) b. Was ist er denn? what is he then [i.e., boy or man] 'What is he then?' (Gabho #192) c. Wie gefällt mir oder? how like me or 'How do you like my (German?)' (Samran #106) d. Warum du hast mir viele gefragt? why you have:2SG me much asked 'Why have you asked me so many (questions?)' (Emine #78) e. Wohin kommen Sie? where.to come you 'Where do you (polite) come from?' (Harva #58) f. Ah, was machst du? ah, what do:2SG you 'Ah, what are you doing?' (Ensook #122)

40.

There were no occurrences of embedded clauses in the data from Gabho, Emine and Ensook. Samran had one instance of an embedded clause with weil 'because' with matrix clause word order (i.e., unlike in German, the finite verb was not in final position) and Mine had two such instances. Harva had three instances of wenn (the complementizer 'when'), all with matrix clause word order.

41.

It is an open question whether children's early head-final embedded clauses — without an overt complementizer — have a CP or not.

42.

Although we cannot necessarily draw conclusions from the general absence of embedded clauses, the fact that none occur in these speakers' data is a good indication that such sentence types do not fall within their competence. The absence of embedded clauses cannot be the result of the type of data collected; a good deal involved narrative accounts and description of sequential pictures in which embedded sentences can be expected to occur.

43.

The data from Dosik are somewhat problematic, however, as we will see.

44.

See Section 4.3.1. for arguments against a CP projection at the FP stage; given that we assume that the Spec-CP position is not yet available, the Spec-FP position would have to function as an A-position.

45.

We thank Sonja Eisenbeiss for suggesting this term to us.

DIRECT ACCESS TO X'-THEORY

313

46.

The Full House Principle appears to be too strong in that empty categories do not have to satisfy this principle — except in the case of a trace of movement, where two positions 'share' a licenser (one position is licensed at D-structure, one at S-structure). We will assume, however, that in the cases under discussion anything that might license empty heads or empty specifier positions (such as government by a matrix verb) is not available in the structures posited by these early learners of German (since there is no embedding yet); therefore, the effects of the Full House Principle can be observed.

47.

Note that the observation that 'stage seepage' occurs is not a new one. The appearance of phenomena from a previous stage of acquisition is included, for example, in Selinker's (1972) description of fossilization.

References Abney, Stephen. 1987. "The English Noun Phrase in its Sentential Aspect." Diss., Massachusetts Institute of Technology, Cambridge. den Besten, Hans. 1983. "On the Interaction of Root Transformations and Lexical Deletive Rules." On the Formal Syntax of the Westgermania: Papers from the '3rd Groningen Grammar Talks' Groningen, January 1981 (=Linguistik Aktuell, 3), ed. by Werner Abraham, 47-131. Amsterdam: John Benjamins. Bley-Vroman, Robert. 1989. "What is the Logical Problem of Foreign Language Learning?" Linguistic Perspectives on Second Language Learning, ed. by Susan Gass and Jacquelyn Schachter, 41-68. Cambridge: Cambridge University Press. Boser, Katharina, Barbara Lust, Lynn Santelmann & John Whitman. 1992. "The Syntax of V2 in Early German Grammar: The strong continuity hypothesis." Proceedings of NELS, 1992. Amherst, Mass.: Graduate Linguistic Student Organization, Department of Linguistics, South College, University of Massachusetts, Amherst. Choe, Hyon Sook. 1988. "Restructuring Parameters and Complex Predicates: A transformational approach." Diss., Massachusetts Institute of Technology, Cambridge. Chomsky, Noam. 1986. Knowledge of Language: Its nature, origin, and use. Praeger: New York. Clahsen, Harald. 1986. "Verb Inflections in German Child Language: Acquisition of agreement markings and the functions they encode." Linguistics 24.79-121. . 1988a. "Parameterized Grammatical Theory and Language Acquisition: A study of the acquisition of verb placement and inflection by children and adults." Linguistic Theory in Second Language Acquisition, ed. by Susanne Flynn and Wayne O'Neil, 47-75. Dordrecht: Kluwer.

314

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

Clahsen, Harald. 1988b. "Kritische Phasen der Grammatikenwicklung. Eine Untersuchung zum Negationserwerb bei Kindern and Erwachsenen." Zeitschrift für Sprachwissenschaft 7.3-31. . 1990. "Learnability Theory and the Problem of Development in Language Acquisition." Theoretical Studies of Language Acquisition, ed. by Thomas Roeper & Jürgen Weissenborn, 53-76. Hillsdale, N.J.: Lawrence Erlbaum. . 1991. "Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language." Language Acquisition 1.361-391. Clahsen, Harald, Sonja Eisenbeiss & Anne Vainikka. This volume. "The Seeds of Structure: A syntactic analysis of the acquisition of Case marking." Clahsen, Harald & Pieter Muysken. 1986. "The Availability of Universal Grammar to Adult and Child Learners: A study of the acquisition of German word order." Second Language Research 2.93-119. . 1989. "The UG Paradox in L2 Acquisiton." Second Language Research 5.1-29. Clahsen, Harald & Martina Penke. 1991. "The Acquisition of Agreement Morphology and its Syntactic Consequences: New evidence on German child language from the Simone-corpus." Studies on the Acquisition of Agreement and Case Marking (= Theorie des Lexikons. Arbeiten des Sonderforschungsbereiches 282, 2.). Demuth, Katherine 1992. "Competence or Performance? What phonology shows about children's emerging syntax." Paper presented at the 17th Annual Boston University Conference on Language Development, Boston, October. Deprez, Viviane & Amy Pierce. This volume. "Crosslinguistic Evidence for Functional Projections in Early Child Grammar." duPlessis, Jean, Doreen Solin, Lisa Travis & Lydia White. 1987 "UG or not UG, that is the Question: A reply to Clahsen and Muysken." Second Language Research 3.56-75. Eubank, Lynn. This volume. "Optionality and the Initial State." Felix, Sascha. 1985. "More Evidence on Competing Cognitive Systems." Second Language Research 1.47-72. Flynn, Suzanne. 1984. "A Universial in L2 Acquisition Based on a PBD Typology." Universals of Second Language Acquisition, ed. by F.R.Eckman, L.H.Bell & D. Nelson. Rowley, Mass.: Newbury House. . 1987. A Parameter Setting Model of L2 Acquisition: Experimental studies in anaphora. Dordrecht: Reidel. Hall, Tracy A. 1992. Syllable Structure and Syllable-Related Processes in German. Tübingen: Max Niemeyer. Hatch, Evelyn & Hossein Farhady. 1982. Research Design and Statistics for Applied Linguistics. Rowley, Mass.: Newbury House.

DIRECT ACCESS TO X'-THEORY

315

Jelinek, Eloise. 1984. "Empty Categories, Case and Configurationality." Natural Language and Linguistic Theory 2.39-76. Kitagawa, Yoshihisa. 1986. "Subjects in Japanese and English." Diss., University of Massachusetts, Amherst. Koster, Jan. 1975. "Dutch as an SOV Language." Linguistic Analysis 1.111-136. Lebeaux, David. 1988. "Language Acquisition and the Form of the Grammar." Diss., University of Massachusetts, Amherst. Meisel, Jürgen, Harald Clahsen & Manfred Pienemann. 1981. "On Determining Developmental Stages in L2 Acquisition." Studies in Second Language Acquisition 3.109-135. Ouhalla, Jamal. 1991. "Functional Categories and the Head Parameter." Paper presented at the 14th GLOW Colloquium, Leiden, March. Pinker, Steven. 1984. Language Learnability and Language Development. Cambridge, Mass.: Harvard University Press. Platzack, Christer. 1986. "COMP, INFL and Germanic Word Order." Topics in Scandanavian Syntax, ed. by Lars Hellan & Kirsti Koch-Christensen. Dordrecht: Reidel. . 1990. "A Grammar without Functional Categories: A syntactic study of early child language." Working Papers in Scandinavian Syntax 45.13-34. Pollock, Jean-Yves. 1989. "Verb movement, Universal Grammar and the Structure of IP." Linguistic Inquiry 20.365=424. Radford, Andrew. 1988. "Small Children's Small Clauses." Transactions of the Philological Society 86(1).1-43. . 1990. Syntactic Theory and the Acquisition of English Syntax. Oxford: Basil Blackwell. Rizzi, Luigi. 1986. "Null Objects and the Theory of pro.'" Linguistic Inquiry 17.501-557. Roeper, Thomas. 1992. "From the Initial Stage to V2: Acquisition principles in action." The Acquisition of Verb Placement, ed. by Jürgen Meisel, 333-370. Dordrecht: Kluwer. Rothweiler, Monika. 1990. "Nebensatzerwerb im Deutschen: Eine empirische Untersuchung zur Primärspracherwerb." Diss., University of Tübingen. Safir, Kenneth. 1981. "Inflection-government and Inversion." The Linguistic Review 1.417-467. Schachter, Jacquelyn. 1988. "Second Language Acquisition and its Relationship to Universal Grammar." Applied Linguistics 9.219-235. Schwartz, Bonnie D. 1986. "The Epistemological Status of Second Language Acquisi­ tion." Second Language Research 2.120-159. . 1993. "On Explicit and Negative Data Effecting and Affecting 'Competence' and 'Linguistic Behavior'. Studies in Second Language Acquisition 15.147-163.

316

ANNE VAINIKKA & MARTHA YOUNG-SCHOLTEN

Schwartz, Bonnie D. & Rex Sprouse. This volume. "Word Order and Nominative Case in Nonnative Language Acquisition: A longitudinal study of L1 Turkish German interlanguage." Schwartz, Bonnie D. & Alessandra Tomaselli. 1990. "Analysing the Acquisition Stages of Negation in L2 German: Support for UG in adult SLA." Second Language Research 6.1-38. Selinker, Larry. 1972 "Interlanguage." International Review of Applied Linguistics. 10.209-231. Sportiche, Dominique. 1988. "A Theory of Floating Quantifiers and its Corollaries for Constituent Structure." Linguistic Inquiry 19.425-449. von Stutterheim, Christiane. 1987. Temporalität in der Zweitsprache. Berlin: Walter de Gruyter. Vainikka, Anne. 1989. "Deriving Syntactic Representations in Finnish." Diss., University of Massachusetts, Amherst. . 1992. "Case in the Development of English Syntax." Ms., University of Massachusetts, Amherst. Weissenborn, Jürgen. 1990. "Functional Categories and Verb Movement: The acquisition of German syntax reconsidered." Spracherwerb und Grammatik: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie, ed. by Monika Rothweiler, 190-224. Opladen: Westdeutscher Verlag. White, Lydia. 1985. "The 'pro-drop' Parameter in Adult Second Language Learning." Language Learning 35.47-62. . 1989. Universal Grammar and Second Language Acquisition. Amsterdam: John Benjamins. . 1991. "Adverb Placement in Second Language Acquisition: Some effects of positive and negative evidence in the classroom." Second Language Research 7.133-161. Young-Scholten, Martha & Anne Vainikka. 1992. "The Development of Functional Projections in L2 Syntax." Paper presented at the Linguistic Society of America Annual Meeting. January 9-12, Philadelphia. Zagona, Karen. 1982. "Government and Proper Government of Verbal Projections." Diss., University of Washington. Zwart, Jan-Wouter. 1991. "Clitics in Dutch: Evidence for the position of INFL." Groninger Arbeiten zur Germanistischen Linguistik, ed. by Werner Abraham, 33.71-92. Amsterdam: John Benjamins.

Word Order and Nominative Case in Non-Native Language Acquisition A Longitudinal Study of (L1 Turkish) German Interlanguage Bonnie D. Schwartz University of Durham 1.

Rex A. Sprouse Harvard University

Introduction

From a certain perspective, the study of NON-NATIVE LANGUAGE ACQUISI­ (L2A) is quite simply more difficult than that of NATIVE LANGUAGE ACQUISITION (L1A). The central problem to be solved in L1A, within the tradition of generative grammar, is explicitly known in advance. Not so with L2A. In L1A the researcher can implement linguistic theory and learnability theories without reservation in order to (attempt to) explain the child's growth of competence; required of a theory ofL1A is the explanation of the intermediate phases of development as well as the means by which the child moves from one phase to the next and arrives ultimately at the final, steady-state grammar, which in all relevant aspects is shared by the native speakers in that language community. Moreover, the data source available in L1A is confined to PRIMARY LINGUISTIC DATA (PLD), and thus the logic of the acquisition problem can limit itself to the interaction of PLD and Universal Grammar (UG) (and a learning procedure) in view of the steady state to be attained. Finally, theories of L1A have little reason even to consider, as self-contained explanations for grammar development, other cognitive capacities that fall outside linguistic-specific functions, for children are quite immature in those domains. In brief, the L1A researcher interested in the development of grammatical knowledge is able to presuppose the relevance of linguistic theory and learning theory, the form of UG, the form of the input, the restricted relevance of non-linguistic-specific cognitive capacities, and the form of the steady-state grammar — and from these the L1A researcher can, in essence, work backwards from the steady state in order to devise explanations for linguistic development. The L2A researcher has TION

318

BONNIE D. SCHWARTZ & REX A. SPROUSE

none of these to presuppose: the 'final' state of L2 knowledge is not uniformly shared by all non-native acquirers (henceforth 'L2ers'); input to L2ers may consist of PLD (from native speakers and non-native speakers, where in the case of the latter, the form may not match native-speaker input), but the PLD are often supplemented by explanations about the TARGET LANGUAGE (TL) grammar and by negative data, both of whose precise effects are still largely unknown; and in regard to especially adult L2ers, the nonlinguistic cognitive functions are fully developed. Add to this the as-yet-to-be-resolved issues of a 'critical period' for language development and of (the extent of) L1 influence on L2A as well as the possibility — unlikely to be characteristic of native language — that any given piece of overt L2 linguistic behavior may well reflect a combination of knowledge sources,1 and the difficulties of L2A investigation seem to multiply those of L1A. As a result of these features of the L2A context, the L2A researcher cannot — given the current state of the field — presuppose the relevance of linguistic theory to explaining L2 development, for the central question in fact is whether the knowledge gained in L2A is representative of human language grammars. This is not to say, of course, that linguistic theory is irrelevant to L2A research. Indeed, it has been argued that not only can the L2A researcher fruitfully implement linguistic theory, but in fact that this is the sole way to begin to investigate the nature of the knowledge system(s) that L2ers create (Gregg 1989; Schwartz 1986). The main conceptual difference, then, between L1A and L2A research is that in L1A, we know (more or less) what the answer to the logical problem of language acquisition should look like, but in (adult) L2A it is not yet clear whether the same (kind of) answer is appropriately demanded — for in fact there is a question as to whether the same (kind of) logical problem even exists. In the face of these (seemingly pessimistic) introductory remarks, we wish to emphasize the overall purpose of this paper: to exemplify one way in which linguistic theory can be employed in L2A research to produce theoretically revealing results. The 'Interlanguage' data come from a longitudinal study of an adult native Turkish speaker acquiring German (see Section 4). In spite of the fact that the 'end' state of knowledge is unknown, we argue that the displayed linguistic development follows a path that is rather intriguing in light of recent work in syntax related to word order and nominative Case checking. Specifically, what we propose is that certain aspects of the Interlanguage development that concern word order in general and verb placement in particular seem to depend crucially on the mechanisms for nominative Case checking (as innately given by

WORD ORDER AND NOMINATIVE CASE

319

UG), as the Interlanguage 'grammar'2 attempts to reconcile progressively larger amounts of German PLD. Descriptively, this development is unlike the documented development of German as a native language (see e.g., Clahsen 1991; Gawlitzek-Maiwald, Tracy & Fritzenschaft 1992; Roeper 1992; Weissenborn 1991); moreover, other work on German L2A, in which the L1s were (null subject) Romance, also traces an entirely distinct developmental route (Clahsen & Muysken 1986, 1989; Meisel, Clahsen & Pienemann 1981). At first blush, then, such differences speak to the influence the L1 grammar may have on non-native linguistic development. As we shall attempt to show, in fact, the data and subsequent analyses here support a model of L2A in which the L1 parameter values serve as part of the INITIAL STATE (Schwartz 1987; White 1985, 1989b); the system of L2 knowledge changes as more and more PLD are perceived to be in need of accommodation, forcing the parameter values to be revised (and perhaps re-revised) along the way. If it is correct that L2ers bring the already fully specified set of parameter values associated with the L1 grammar to the task of L2A, it is in principle possible that certain hypotheses that 'need' to be revised cannot be revised or will in practice only seldom be revised because of the absence or rarity of certain pieces of PLD. This in itself does not imply that UG is inoperant in L2A. Our aim is not dissimilar from what is needed in an explanatory theory of L1 development which assumes the CONTINUITY HYPOTHESIS (Hyams 1986; Pinker 1984), namely: to determine whether it is possible to capture each intermediate phase of the Interlanguage system via (independently motivated) principles and parameters of UG (duPlessis, Solin, Travis & White 1987). (Clearly, such a model explicitly rejects any strong version of the critical period hypothesis of Lenneberg 1967.) To the extent that this goal is met — in particular by tying the changes in the Interlanguage system to changes in parameter values3 — there is evidence for the hypothesis that the L2ers' linguistic knowledge is comparable (in terms of knowledge type) to that of native-speaker competence. Indeed, a secondary goal of this study is to suggest that it is inappropriate to analyze L2 systems merely from the perspective of a 'match' or a 'mismatch' with the TL grammar (Bley-Vroman 1983). To show that the L2 system departs — even perhaps radically — from the TL grammar in no way provides evidence that the system underlying Interlanguage departs from human language grammars (Schwartz 1989, 1992). In short, this work is a call for studying the properties of Interlanguage systems in their own right — not from the perspective of the TL — in order to determine whether there is an internal consistency that is formally explicable in the language of Universal

320

BONNIE D. SCHWARTZ & REX A. SPROUSE

Grammar. Such a position contrasts with (much of the) more traditional L2A research which focuses on the fact that (adult) L2A so often leads to less than native-like 'mastery' of the TL, even after years of seemingly abundant exposure to PLD.4 This also relates to a second subsidiary goal in this study: to attempt to address the issue of the etiology of Interlanguage systems, a topic that seems to be virtually absent from current work on L2A within the principles and parameters framework. As a first approximation, we will suggest that one of the main differences between L1A and L2A may be that only L1A is a deterministicallydriven process. We call this the DETERMINACY PROBLEM IN L2A, which we explore more fully below. A few words on methodology: our interest is the course of development in the Interlanguage system, and so longitudinal data are best suited to the pursuit of tracing the growth of L2 knowledge in an individual. In addition, the data that have been made available to us have been elicited (generally) by asking the subject to retell the plot of a silent movie or by casual conversation. We are aware that such production data could well be skewed and the extent of knowledge under-represented, especially since 'ungrammaticality' can only be inferred from non-occurrence. Nevertheless, it is our belief that the drawbacks of such production data are more than offset by some of the benefits. Of traditional concern in (adult) L2A (see, e.g., Krashen 1981) is, somewhat simplified, whether rote memorization of facts about the TL should be seen as reflective of the systematic knowledge that an L2er creates on the basis of PLD. (Note that this is something that generally does not plague the study of L1A.) Conceptually, this has been hypothesized to be more or less equivalent to the linguist's need to separate prescriptive knowledge from 'cognized' (Chomsky 1975) knowledge of grammar.5 Thus, many in the field of L2A are inclined towards using data gathered by techniques that somehow minimize the potential influence of metalinguistic knowledge about the TL. On-line, casual production tasks would seem to tend towards this, for subjects are highly geared to communicate their thoughts rather than being focused on the form of what they say. We agree that a more complete database would include direct information about what is not generable by the subject's system of linguistic knowledge at any particular point in development; nonetheless, the fact that certain syntactic patterns are completely missing at one stage but then turn up at a later stage — particularly when the specific elicitation task is held constant — does seem to point to a definite change in linguistic knowledge.6 The organization of the paper is as follows: Given the general model of

WORD ORDER AND NOMINATIVE CASE

321

L2A that we will be defending, viz., that the L1 grammar serves as the point of departure in L2A, we start, in Section 2, with a brief description of the relevant features of Turkish and German clause structure, adopting the verb movement hypothesis of Emonds (1978) extended to multiple verbally-related functional heads by Pollock (1989) and adapting it somewhat to characterize some basic facts of Turkish. In Section 3 we review the different ways in which nominative Case checking can be satisfied, following proposals of Koopman & Sportiche (1991) and Rizzi & Roberts (1989). We turn to the Interlanguage data in Section 4, where we first provide some background information on the subject as well as the particulars of the data elicitation technique and corpus, and we next describe, in general terms, the observed L2 development. The analysis of each stage is presented in Section 5. In Section 6 we discuss the possible reasons underlying this particular path of development. We conclude the paper by con­ sidering some of the theoretical implications with respect to both L1A and L2A.

2.

Clause Structure in Turkish and German

At this point it will be useful to establish some basic facts about clausal architecture and the morphosyntax of verbs in Turkish and German. 2.1. Turkish Clause Structure Turkish is an SOV language, and it is a textbook case of an agglutinating language.7 Consider sentence (1): (1)

Siz

ev

-e

gid-e

-me -yebil-ir

-sin-iz.

YOU+PL home+DAT go +PBY+NEG+ABIL+AOR+2 +PL

'It is possible that you (p1) may not be able to go home.' In (1) the finite verb (gid-e-me-yebil-ir-sin-iz) is the last word of the clause. This is true of both main and embedded clauses in Turkish.8 The verb is easily segmentable into a set of bound morphemes which must occur in a fixed order. If they contain phonologically underspecified segments, these bound morphemes undergo rules of Turkish vowel harmony and consonant assimilation, which operate at the word level. Contrast (1) with the sentence in (2): (2)

Siz

bunu yap-a -ma -yabil-ir -sin-iz. YOU+PL that do +PBY+NEG+ABIL+AOR+2 +PL 'It is possible that you (p1) may not be able to do that.'

322

BONNIE D. SCHWARTZ & REX A. SPROUSE

In (1) and (2) the suffixes -E 'possibility', -mE 'negative' and -(y)Ebil 'ability' contain low unrounded vowels unspecified for <±back>. They harmonize with the <+back> feature of the root vowel /a/ of yap- 'do' in (2) and with the <-back> feature of the root vowel /i/ of gid- 'go' in (l). 9 We will analyze the syntax of Turkish verbs making the following assumptions. VP is head-final, and the specifier position to which the subject ultimately moves (for nominative Case checking — see Section 3) necessarily precedes the VP. Following the general lines of Pollock (1989), each bound morpheme corresponds to exactly one (functional) head that projects its own phrase. The finite verb must raise (à la Emonds 1978) into each of the c-commanding, verbally-related functional heads (either by adjunction or substitution), in order to meet a morphophonological well-formedness condition. This movement occurs in a step-by-step fashion, subject to conditions such as the Head Movement Constraint (Travis 1984). We assume that the order of suffixes mirrors the hierarchy of the functional heads (à la Baker 1985), and given the general head-final and suffixal nature of Turkish grammar, verb movement in Turkish is string vacuous. In Turkish, the right-most inflectional suffix on the verb is generally person/number agreement. This indicates that AGR° is the highest of the verbally-related functional heads. As the subject precedes the VP in 'unscram­ bled' clauses (see below), it is reasonable to assume that the subject NP in Turkish moves to Spec-AGRP at S-structure, where its nominative Case feature is checked under a Spec-head (agreement) relation (see Section 3). Following Kornfilt (1984:271-277) we assume the existence of C° and the C-system in Turkish on general theoretical grounds (contra Kennelly 1990). To find strong unambiguous evidence for the position of C° in Turkish is somewhat more difficult, however. Most instances of clausal complements and clausal adjuncts in Turkish do not resemble their counterparts in Indo-European languages: they are not marked by a lexical complementizer, have 'defective' tense/aspect verbal morphology, and have the external syntax of NPs. An example is given in (3b) (from Underhill 1976:322): (3)

a.

b.

Halil gel -di -Ø /gel -ir -Ø /gel -iyor -Ø. Halil:NOM come+PAST+3SG/come+AOR+3SG/come+PRES+3SG 'Halil came/comes/is coming.' [Halil in gel -dig -in] -i bil -iyor -urn. Halil

+GEN come+NOMIN+3SG+ACC know+PRES+ISG

'I know that Halil came/comes/is coming.'

WORD ORDER AND NOMINATIVE CASE

323

Finite sentential subordination also exists in Turkish; here, it is the lexical element filling the head of CP which determines the ordering between the embedded sentence and the superordinate verb. Consider the examples in (4): (4)

a.

b.

Left-headed C': [CP [C COMP AGRP]] with ki 'that' Duydum [CP [C' ki [AGRP sen gel -ecek-sin]]] I.heard that you+SG come+FUT+2SG 'I heard that you will come.' Right-headed C': [CP [C' AGRP COMP]] with diye 'that'/'because' Ben [CP [C' [AGRP Hasan ne ye -di -Ø] diye]] I Hasan what eat+PAST+3SG diye merak ettim. curiosity did 'I wondered what Hasan ate.'

Sentence (4a) exemplifies a left-headed C' with ki 'that', while (4b) exemplifies a right-headed C' with diye 'that'/'because'. There is a limited number of verbs that allow sentential complements of this kind; nevertheless, ki-clauses are rather common in certain varieties of colloquial Turkish (Underhill 1976:433). We will assume that Turkish grammar allows embedded CPs to either the left or the right of the verb. The relative order of C° and its complement is dependent on the particular superordinate predicate and on the particular lexical item heading the subordinate CP. (See Bayer 1991 for similar alternations in Bengali.) While SOV is a very common order, it is by no means the only one possible, as illustrated in the following paradigm from Kural (1992:1): (5)

a.

b. c. d. e. f.

Ahmet [bu kitabi] Berna'ya vermiş. Ahmet this book to.Berna gave 'Ahmet gave this book to Berna.' Ahmet Berna!ya [bu kitabi] vermiş. [bu kitabi] Ahmet Berna'ya vermiş. Berna ya Ahmet [bu kitabi] vermiş. [bu kitabi] Berna' ya Ahmet vermiş. Berna ya [bu kitabi] Ahmet vermiş.

S DO IO V

S IO DO V DO S IO V IO S DO V DO IO S V IO DO S V

As the sentences in (5) suggest, while the position of the verb within a Turkish clause is fixed, phrasal categories within a clause exhibit much variability in their order. This can be accounted for by assuming an SOV base and a scrambling operation that adjoins phrases to higher phrases containing them. On the basis

324

BONNIE D. SCHWARTZ & REX A. SPROUSE

of his analysis of scope facts, Kural (1992) argues that scrambling in Turkish may adjoin phrases to AGRP, but not to CP. 2.2. German Clause Structure According to standard assumptions, German clause structure shares with Turkish clause structure the property of being underlyingly verb-final. However, in German this essential property is often obscured by the fact that the finite verb must move to the C° position in clauses in which no other material occupies this position, i.e., in main clauses and in complement clauses not introduced by a lexical complementizer or abstract wh-element. This type of analysis (den Besten 1983; Koster 1975; Thiersch 1978) provides for a straightforward characterization of the following basic word order facts. Let us first consider the placement of the verb in embedded clauses: (6)

a.

b. (7)

a.

b. (8)

a. b.

Hans sagt, daß er heute den Roman liest. John says that he today the novel reads 'John says that he is reading the novel today.' *Hans sagt, daß er liest den Roman. John says that he reads the novel Hans sagt, daß er den Roman mitbringt. John says that he the novel along-brings 'John says that he is bringing the novel along.' *Hans sagt, daß er bringt den Roman mit. John says that he brings the novel along Hans sagt, daß John says that *Hans sagt, daß John says that

er he er he

den Roman gelesen hat. the novel read has hat den Roman gelesen. has the novel read

In (6a), a simple finite verb is in embedded clause-final position. In (7a), a verb with a particle ('separable prefix') likewise occurs in this position with the order particle-verb. In (8a), a periphrastic verb form appears at the end of the clause with the nonfinite verb preceding the finite verb. In the corresponding (b)-examples, any other placement of the finite verb results in ungrammaticality. Let us turn now to the placement of the verb in main clauses: (9)

a.

Der Mann liest heute den Roman. the man reads today the novel 'The man is reading the novel today.'

WORD ORDER AND NOMINATIVE CASE

(9)

b. c. d.

(10) a.

b. c. d. (11) a.

b. c.

325

Den Roman liest der Mann heute. the novel reads the man today Heute liest der Mann den Roman. today reads the man the novel *Heute der Mann liest den Roman. today the man reads the novel Der Mann bringt den Roman mit. the man brings the novel along 'The man is bringing the novel along.' Den Roman bringt der Mann mit. the novel brings the man along *Der Mann bringt mit den Roman. the man brings along the novel *Der Mann mitbringt den Roman. the man along.brings the novel Der Mann hat den Roman gelesen. the man has the novel read 'The man (has) read the novel.' Den Roman hat der Mann gelesen. the novel has the man read *Der Mann hat gelesen den Roman. the man has read the novel

Examples (9)-(11) are the main-clause analogues of (6)-(8). In (9a)-(9c) the finite verb appears in the second position following the subject, the direct object, and adverbial. However, example (9d), in which two constituents precede the finite verb, is ungrammatical. The examples in (10) and (11) show that the verbsecond V2 property of main clauses refers solely to the finite verb. Nonfinite verbal material, such as particles, as in (10), and participles, as in (11), must remain in clause-final position.

3.

Mechanisms for Nominative Case Checking

We assume that Universal Grammar includes a licensing condition, i.e., the Case Filter (Chomsky 1981), to which every (overt) NP is subject. Regardless of whether it is a primitive condition on syntactic representations or a theorem derivable from other principles of grammar, the Case Filter requires that the Case

326

BONNIE D. SCHWARTZ & REX A. SPROUSE

feature of every overt NP (inter alia) be checked. In the spirit of Koopman & Sportiche (1991) and Rizzi & Roberts (1989), we assume that UG makes available three mechanisms for the checking of Case features: (12) a.

b. c.

Spec-head agreement The checker is the head (H) of a phrase (HP) and the NP whose Case feature it checks is in the specifier of that phrase (Spec-HP). Government The checker governs the NP whose Case feature it checks. Incorporation The head of the NP is incorporated into the checker.

Here we assume a version of X'-theory in which phrases have unique head and specifier positions. The definition of GOVERNMENT we utilize is one in which strict c-command is a requirement for government and in which a head that governs YP also governs the Spec-YP position (see Roberts 1993:18-19). In this paper we will be concerned with the checking of Case features solely on the subjects of finite clauses, and thus, part of our focus will be on the means by which the subject NP satisfies the Case Filter. We will further assume that AGR° crosslinguistically is the prototypical nominative Case checker, and the analyses of the other languages pertinent to this study conform to this statement (see below). Following Koopman & Sportiche (1991), we take it to be a simple matter of crosslinguistic parameterization which option or options in (12) a given grammar happens to implicate. A grammar will have to utilize at least one of the three options if it is to generate any well-formed finite clauses. Moreover, of (12a) and (12b), a given grammar will have to instantiate at least one, if it is to generate any well-formed finite clauses with a nonclitic subject.10 However, there is no relationship of mutual exclusivity among any of the options listed in (12): an individual grammar might exhibit one, two, or all three. Assuming the VP-internal subject hypothesis of, inter alia, Zagona (1982), Kitagawa (1986), Kuroda (1988) and Sportiche (1988) the choice of mechanism(s) for nominative Case checking has striking consequences for the surface landscape of a language. Let us consider first the Spec-head agreement option of (12a). This is sketched in (13), where we leave out the other functional projections which may intervene between AGRP and VP. In a language with the Spec-head agreement mechanism, the subject will have to leave its D-structure position within the VP (here, to simplify exposition, in the specifier position which is left of V') and raise to the Spec-AGRP

WORD ORDER AND NOMINATIVE CASE

327

position. In a periphrastic construction, a grammar with this mechanism and with overt verb movement to AGR° will generate SV<+fin>V<-fin>O if the VP is head-medial and SV<+fin>OV<_fin> if the VP is head-final. Using X to stand for a topicalized nonsubject phrase, it may also be possible to derive the order XSV<+fin>V<-fin>O or XSV<+fin>OV<-fin>, if the language allows (leftward) adjunction to AGRP. Assuming Spec-AGRP to be to the left of AGR', it will not be possible to derive verb-initial sentences in a language with only option (12a), because the finite verb (contained in AGR°) and the subject would necessarily not be in a Spec-head relation. Nor would it be possible to generate XV<+fin>S ... sentences for precisely the same reason. Turning to the government option of (12b), we see that this mechanism offers more possibilities for word order variation within a language. Consider the following structures in (14):

328

BONNIE D. SCHWARTZ & REX A. SPROUSE

WORD ORDER AND NOMINATIVE CASE

329

(14a) is a sketch of a (surface) V<+fin>SV<-fin>O language, such as Welsh, with government from AGR°. In such a grammar, the finite verb moves to AGR° and then AGR° can govern the subject (here in Spec-VP), checking its nominative Case feature. (14b) and (14c) are representations of possible structures in (underlyingly) SOV<-fin>V<+fin> languages, such as German, with the government option, but in this instance from C° rather than from AGR°. In such a grammar, the finite verb can move from Vo to AGR° to C°; from C°, AGR° is able to govern — and hence check the nominative Case feature of — the subject in Spec-AGRP. In (14b) the Case Filter is fulfilled through the trace of the subject in Spec-AGRP. In (14c) the subject stays in Spec-AGRP and another XP (e.g., the direct object, as depicted in (14c)) moves to Spec-CP. Following standard practice, we will use the (perhaps, misleading) term INVERSION to refer to such occurrences of post-verbal subjects. In languages such as German, where the verb moves to C° in main clauses, the question arises as to how the subject NP satisfies the Case Filter in embedded clauses with an overt complementizer in C°. Two alternatives have been suggested. One possibility is that lexical complementizers in such languages are themselves Case checkers (Koopman 1984; Platzack 1986). Another possibility is that the Spec-head agreement mechanism (at the AGRP level) is also available in such a grammar, operative in embedded clauses. For the sake of concreteness, we will assume the latter possibility, but we refer the reader to Vikner (1991) for detailed review. Finally, let us turn to the incorporation option of (12c). Consider the contrasts in (15), exemplifying the well-known facts of French inversion, which is permitted with pronominal subjects, but not with nonpronominal subjects: (15) a.

b.

c.

d.

Est-il allé? is he gone 'Has he left?' Est Jean allé? is John gone 'Has John left?' Où est-il allé? where is he gone 'Where has he gone?' *Où est Jean allé? where is John gone 'Where has John gone?'

330

BONNIE D. SCHWARTZ & REX A. SPROUSE

To account for the above asymmetry between pronominal and nonpronominal subjects, Rizzi & Roberts (1989) propose an analysis according to which the subject in (15a) and (15c) is a clitic that incorporates into the (finite) verb that has raised (through AGR°) to C°, as depicted in (16):11

Following Baker (1988), Rizzi & Roberts motivate their analysis by assuming that the Case Filter should be viewed as a general requirement that nominais be associated with some Case feature; this requirement can be accomplished in two ways: either by assigning the Case feature from a head to the nominal (Spec-head agreement, (12a), or government, (12b)) or by incorporating the nominal into a Case-feature bearing head. According to Rizzi & Roberts (see in particular p. 25, note 3), the grammar of French instantiates option (12a) for realizing the nominative Case feature on the subject in SVO sentences; if option (12b) were also available in French, one would expect (15b) and (15d) to be grammatical. Without the government option, an explanation is needed for the grammaticality of (15a) and (15c). Arguing that cliticization is a form of incorporation, and following Kayne (1983), Rizzi & Roberts propose that a subject pronoun (in Spec-AGRP) is allowed to cliticize to an inflected verb that has already raised into C° and hence c-commands it. Incorporation thus provides a means for the subject pronoun in subject-clitic inversion to satisfy the Case Filter. The Rizzi & Roberts analysis thus captures the fact that inversion in French with a nonpronominal subject is ungrammatical; only clitic pronouns can

WORD ORDER AND NOMINATIVE CASE

331

satisfy the Case Filter through the incorporation option. Let us now address some specific questions in an analysis of Turkish and German employing the set of options outlined in (12). In a verb-final language with scrambling and without a process of verb-fronting, such as Turkish, it is not immediately obvious whether Case is checked under the agreement option or under the government option. Let us assume that the verb necessarily raises to the highest verbally-related functional head position at S-structure. In a surface SOV clause, it is possible that the subject NP is in Spec-AGRP position and has its Case feature checked under the agreement option or that the subject is in a lower specifier position and has its Case feature checked under the government option. Nevertheless, we will assume that within this framework Turkish exhibits only the agreement option. Given the fact that Turkish has instances of C° to the left of AGRP, (4a), positing the government option in Turkish would seem to predict that verb-initial or verb-second surface clauses should be well formed. Since this is not the case, we conclude that the government option is unavailable. Furthermore, the incorporation option would seem to be unavailable, since Turkish does not have nominative clitics. A different question arises for the analysis of German nominative pronouns. Following Tomaselli (1990), we assume that atonic nominative pronouns in German are clitics. This assumption allows Tomaselli to account for the paradigm in (17): (17) a.

b. c. d. e. f. g.

Ich glaube, I believe daß der Mann den Roman gelesen hat. that the man+NOM the novel+ACC read has 'that the man has read the novel.' daß der Mann ihn gelesen hat. that the man+NOM it+ACC read has daß ihn der Mann gelesen hat. that it+ACC the man+NOM read has daß er den Roman gelesen hat. that he+NOM the novel+ACC read has *daß den Roman er gelesen hat. that the novel+ACC he+NOM read has daß er ihn gelesen hat. that he+NOM it+ACC read has *daß ihn er gelesen hat. that it+ACC he+NOM read has

332

BONNIE D. SCHWARTZ & REX A. SPROUSE

The generalization is that while nonsubject pronouns can be adjacent to the complementizer, as in (17c), they need not be, as in (17b); in contrast, (un­ stressed) subject pronouns must be adjacent to the complementizer (compare the grammatical (17d) and (17f) with the ungrammatical (17e) and (17g)). Tomaselli argues that as clitics, German atonic nominative pronouns must cliticize to the lexical material in C°. If this cliticization process must occur at S-structure, then it would seem that the incorporation option would be implicated. However, it is unclear whether this cliticization takes place at S-structure or at PF: if only at PF, then these pronouns presumably have their Case feature checked under the same mechanisms as nonpronominal subjects. In Table 1 we summarize the parametric differences between Turkish and German that we henceforth will assume: Table I. Parametric Differences between Turkish and German Tur Ger SOV VP-AGR C-AGRP as possible order Nominative checked under agreement Obligatory movement of finite verb to empty COMP Left-Adjunction to CP possible Spec-CP as landing site for topics Nominative clitics Nominative checked under incorporation Nominative checked under government

4.

+ + + + -

+ + + + + + + ? +

The L2 Data

4.1. The Subject Cevdet (a pseudonym) was born in Turkey in 1966, and he grew up as a monolingual Turkish speaker in Se, an Anatolian village east of Ankara. He received nine years of formal education in Turkey before going to the Federal Republic of Germany in October 1981 to seek work. The European Science Foundation (ESF) transcripts provided to us comprise interviews with Cevdet held over the period from July 1982 through August 1984. The ESF researchers characterized his oral/aural levels in German at the first encounter as 'little' and

WORD ORDER AND NOMINATIVE CASE

333

his knowledge of written German as 'not determined'. The ESF researchers supplied us with further biographical information on Cevdet, which we summarize below. From the summer of 1982 through the summer of 1983, Cevdet received 10 hours per week of instruction in German language and 20 hours per week of vocational and social instruction through Maßnahmen zur beruflichen und sozialen Eingliederung, a program sponsored by the Federal German governmental employment office whose purpose is to assist young non-German nationals in vocational training and placement and in acculturation into German society. In addition, he had meetings with a social worker for approximately 3 hours per week for 3 months. From September 1983 through June 1984, Cevdet worked for a firm that cleaned and refurbished furniture. German was the language of the work place. From June 1984 onward he worked as a roofer, again with German as the language of the work place. Outside of work, Cevdet belonged to integrated sports teams. During this period Cevdet continued to live in an apartment with his parents and his brother. They spoke Turkish at home. This information gives a sense of the quality and quantity of his exposure to German in naturalistic settings. It may seem odd that a study such as the present one would base itself on developmental data from a subject who has had a year of instruction in the TL, since the precise effects of the knowledge gained either in terms of metalinguistic knowledge about German or in terms of the organization of the underlying Interlanguage system are difficult to sort out. While Cevdet's formal language training no doubt had an effect on certain aspects of his knowledge of German, particularly with respect to the lexicon and verbal morphology, we would like to point out that the time during which Cevdet exhibited the most dramatic linguistic developments does not coincide with his German language course (summer 1982 to summer 1983), as will be shown in more detail below. Indeed, upon his entry into a German-speaking work environment and well after the end-point of his formal instruction in German, his Interlanguage continued to undergo significant syntactic changes. 4.2. The Corpus The ESF protocols serve as the sole corpus for this study. They are transcriptions of tape recordings (to which we have not had direct access) of interviews between Cevdet and one or more ESF researchers over a 26-month period, in which there are 11 different interview dates.12 It is clear from the content of the interviews that at least some of the ESF researchers could speak

334

BONNIE D. SCHWARTZ & REX A. SPROUSE

(at least some) Turkish. However, after the early interviews only German was spoken. Cevdet's utterances were transcribed in a broad phonetic system, with gaps, overlaps, pauses, false starts, and other incomplete utterances all included. In addition, the rising intonation of questions was indicated. In these interviews the ESF researchers never pose questions to Cevdet about his metalinguistic knowledge of German. Indeed, the topic of the German language per se never arises. While the ESF researchers occasionally fill in a lexical gap for Cevdet, they never correct his speech; in fact, they uniformly praise him for his ability to speak German, thereby apparently creating an environment in which the interviews could proceed comfortably. A few of the interviews include discussions of Cevdet's background, family, and current living situation. There is also a limited amount of role-playing, where Cevdet and the interviewer play 'store', for example. The vast bulk of the data, however, comes from the following task: Cevdet views a silent film (Harold Lloyd, Charlie Chaplin, etc.) and then retells the plot of the film to the interviewer(s). In the process, the interviewer(s) may ask for further information or clarification; however, it is Cevdet who does most of the talking. It is apparent from the transcripts that Cevdet sees a couple of films many times over the two-year period. Not surprisingly, he is able to recount the plots in greater detail and with greater linguistic sophistication over time. 4.3. The Interlanguage Developmental Path Since this study is concerned with Cevdet's Interlanguage development with respect to the basics of sentential word order, we carried out counts of the following utterance types: subject-initial main clauses, non-subject-initial main clauses with pre-verbal subjects, and non-subject-initial main clauses with post-verbal subjects. In doing so we have distinguished between sentences with (unstressed) pronominal subjects and sentences with nonpronominal subjects. In our counts, we include only sentences with sufficient syntactic material to allow for the possibility in principle within Cevdet's system for non-subject-initial clauses.13 In the very first interview (2 July 1982) Cevdet produces only one-word and two-word utterances apart from a few pat phrases (e.g., Ich weiß nicht 'I don't know'). In the subsequent interviews, however, Cevdet begins to produce Verbal' sentences, although occasionally with missing arguments.14 Taking the first interview to represent a Stage 0 for which we lack adequate data, we have identified three successive developmental stages relevant to our interests.

WORD ORDER AND NOMINATIVE CASE

335

4.3.1. Stage 1 (22 October 1982 to 10 February 1983) This stage provides clear evidence of an SOV system. Particles ('separable prefixes') and nonfinite verb forms are consistently in clause-final position, as illustrated in (18): (18) a.

b.

c.

der Mann seine Frau geküßt the man his wife kissed The man kissed his wife.' falsches Wagen ein-gesteige wrong car in+climbed 'gotten into the wrong car' der ist aussteigen he is out+climb(ed) 'He got out.'

(10 February 1983)

(10 February 1983)

(10 February 1983)

Examples (18a) and (18b) lack a finite verb altogether. Here the perfect participle is in clause-final position. In examples (18b) and (18c) there is a particle with a perfect participle. The particle always precedes the clause-final participle. At this stage it is not the case that finite verbs occur in clause-final position, as shown in (18c); rather, they systematically occur 'earlier' in the clause, immediately after the subject, as in the example (19): (19)

jetzt er hat Gesicht [das is falsches Wagen] now he has face that is wrong car 'Now he makes a face (that) that is the wrong car.' (10 February 1983)

At this stage subject-initial sentences are by far the most common; however, fronting of nonsubjects occurs. When the first constituent is a non-argument, as in (19), the subject follows immediately or is occasionally null. There is not a single instance of a post-verbal subject at this stage. An additional feature of Stage 1 is the absence of syntactically identified embedding. There are, however, clear examples which are understood as sentential complementation, such as (19) or (20). Yet, none of Cevdet's 'functionally' embedded utterances exhibits a lexical complementizer. (20)

der Chef hat gesag [der Zug fährt ab] the boss has said the train goes away

Table 2 summarizes the data for Stage 1:

(10 February 1983)

336

BONNIE D. SCHWARTZ & REX A. SPROUSE

Table 2. Stage 1 S = -pronoun

S =+pronoun

Total

svx... xsv... xvs...

11 1 0

18 3 0

29 4 0

Total

12

21

33

Thus, at Stage 1, of the 33 clauses of sufficient syntactic complexity to allow inversion with a constituent other than a nonfinite verb form, not a single one of them exhibits inversion. 4.3.2. Stage 2 (9 June 1983 through 16 May 1984) In this stage Cevdet's Interlanguage exhibits two significant modifications as compared to the earlier data. He acquires inversion with pronominal subjects, and he begins to lexically mark (complement and adjunct) subordinate clauses. Consider first the examples in (21): (21) a. wenn du so wie ein Holz gefunden hast, when you such as a wood found have dann bringst de mir (9 June 1983) then bring you me 'When you have found a piece of wood like this, then you will bring it to me.' b. hier gibt es nicht viel (?)der Bäume (14 July 1983) here gives it not many (of.)the trees 'Here there are not many trees.' c. dann trinken wir bis neun Uhr (1 November 1983) then drink we until nine o'clock Then we will drink until nine o'clock.' d. Q: Kannst du's lesen? A: ja, kann ich lesen (14 July 1983) can you+it read? yes can I read 'Q: Can you read it? A: yes, I can read (it).' e. Q: Cevdet, wie ist es bei euch? Kann man auch alte Cevdet how is it chez you? can one also old Sachen kaufen? things buy? A: ja, kann man kaufen (1 November 1983) yes can one buy 'Q: Cevdet, how are things in your country? Can one buy old things too? A: yes, one can buy (them).'

WORD ORDER AND NOMINATIVE CASE

337

In each of the examples in (21), a pronominal subject immediately follows the finite verb. This holds over a variety of clause types: a then-clause (the apodosis of a conditional (21a)), simple declaratives ((21b) and (21c)), and answers to questions, including answers apparently exhibiting the 'topic drop' pattern typical of colloquial German ((21d) and (2le)). 15 Note that even in the presence of a nonsubject constituent in sentence-initial position, non-inversion is a possibility, both with pronominal subjects (e.g., ich T ) , as shown in (22a)-(22b), and with nonpronominal subjects (e.g., der Lehrer 'the teacher'), as in (22c): (22) a.

Ankara ich kenne Ankara I know 'I know Ankara.'

(14 July 1983)

b.

in mein Dorf ich habe fünf Jahre in die Schule gegangen in my village I have five years in the school gone 'In my village I went to school for five years.' (14 July 1983)

c.

in der Türkei der Lehrer kann den Schüler schlagen in the Turkey the teacher can the pupil beat 'In Turkey the teacher can hit the pupil.' (14 July 1983)

In Table 3 we summarize the data for Stage 2: Table 3. Stage 2 S = -pronoun

svx... xsv... xvs... Total

S = +pronoun

Total

93 26 1

109 38 69

202 64 70

120

216

336

We interpret these numbers as indicating a grammatical system which allows post-verbal pronominal subjects, but not post-verbal nonpronominal subjects. Stage 2 is also the point at which lexically identified syntactic embedding first emerges. Example (21a) above provides an early example of an adjunct clause introduced by wenn 'when'/'if'. Argument clauses introduced by daß 'that' start somewhat later in this stage, the first occurrences being in the

338

BONNIE D. SCHWARTZ & REX A. SPROUSE

interview of 27 January 1984. The first clause introduced by ob 'whether' appears in the transcript of 19 April 1984 (23f). Note that all of the examples in (23) exhibit (finite) verb-final word order.16 (23) a.

b.

c.

d.

e.

f.

g.

daß ich mit Brot war that I with bread was 'that I had some bread' daß sie ein Brot geklaut hat that she a bread stolen has 'that she stole a loaf of bread' daß er eine Wagen brauchte that he a car needed 'that he needed a car' daß der Wolf Schafe ißt that the wolf sheep eats 'that the wolf eats sheep' daß du mein Kind genommen hast that you my child taken have 'that you took my child' ob der Zug noch da ist whether the train still there is 'whether the train is still there' daß sie so schnell komme that she so quickly comes 'that she comes so quickly'

(27 January 1984)

(27 January 1984)

(27 January 1984)

(29 February 1984)

(19 April 1984)

(19 April 1984)

(16 May 1984)

4.3.3. Stage 3 (5 August 1984) The third stage corresponds to a single interview; however, the data for this date are quite extensive. Recall that in Stage 2 Cevdet produced 69 examples of inversion with pronominal subjects but only one example of inversion with a nonpronominal subject; in contrast, at Stage 3 he begins to exhibit inversion with nonpronominal subjects as well, as illustrated in (24) and (25). (24) a.

das hat eine andere Frau gesehen that has an other woman seen 'Another woman saw that.'

WORD ORDER AND NOMINATIVE CASE

339

(24) b.

draußen hatte die Polizei eine Wagen brauchen sollten outside had the police a car need should 'Outside the police should have (used) a vehicle.'

(25) a.

hat viele Menschen zu ihm gehorcht has many people to him obeyed 'Many people obeyed him.' [e] hat jemand gestohlen has someone stolen 'Someone has stolen (it).'

b.

Note that in (25), the finite verb is the sentence-initial constituent. We summarize the data for Stage 3 in Table 4: Table 4. Stage 3 S = -pronoun SVX...

xsv... xvs... Total

S =+pronoun

Total

46 7 8

48 2 67

94 9 75

61

117

178

4.3.4. Summary of the Stages The findings of our tabulations for all three stages are summarized in Table 5: Table 5. Declarative Main Clauses with Two or More Nonverbal Constituents pronominal subjects pre-verbal stage

svx

1 2 3

18 86% 109 50% 48 41%

stage 1 2 3

xsv 3 38 2

14% 18% 2%

post-verbal ...VS 0 69 67

0% 32% 57%

total 21 216 117

nonpronominal subjects pre--verbal post-verbal XSV SVX ...VS 11 93 46

92% 78% 75%

1 26 7

8% 21% 11%

0 1 8

0% 1% 13%

100% 100% 100%

total 12 120 61

100% 100% 100%

340

BONNIE D. SCHWARTZ & REX A. SPROUSE

We interpret these findings in the following way: At Stage 1, Cevdet's grammar does not allow for post-verbal subjects at all. At Stage 2, post-verbal subjects are permitted, but only if they are pronominal. At Stage 3, post-verbal subjects may be either pronominal or nonpronominal. At all three stages, the fronting of a nonsubject does not require that the subject occur post-verbally, although in Stage 3 there are only two cases of pre-verbal pronominal subjects that are preceded by another fronted constituent.

5.

The Analysis

In this section we present our analysis of the three stages we have identified in Cevdet's data. We treat each of these stages as a grammatical system to be described 'on its own terms' in the sense that — while our description will be guided by the restrictive version of UG we assume — we are not primarily concerned with the question of whether the syntactic properties of any of these stages matches those of (any of the stages in L1 acquisition of) German. In other words, in the comparison between Cevdet's Interlanguage and German, we explicitly reject the following equation: matching of (certain aspects of) surface patterns equals isomorphy of grammatical analysis. At each point in the discussion below, the grammatical properties of the immediately antecedent system will play a substantially more important role than the parametric values attributed to German in Section 3. The direct role of German lies in the German PLD to which Cevdet is exposed. Cevdet's Interlanguage changes in order to accommodate more German PLD, not in order to match the parametric values of native German. 5.1. Stage 1 All of the data of our Stage 1 are compatible with an analysis that posits an underlying SOV order (see also Vainikka & Young-Scholten, this volume), as in L1 Turkish. Both indirect and direct evidence supports this hypothesis. First, we find no utterances in which a nonfinite verb precedes its object.17 On the other hand, there are a number of sentences in which either a separable prefix precedes a nonfinite verb or the direct object precedes a nonfinite thematic verb. For finite verb (V<+fin>) forms, the word order patterns are quite different. Unlike Turkish, utterances in which V<+fin> follows its direct object are unattested, even in the earliest data we have. By far the most common sentence pattern (again,

WORD ORDER AND NOMINATIVE CASE

341

as opposed to what is possible in Turkish) is SV<+fin>(X); yet, if one assumes that the finite verb moves from its base position, these are syntactically ambiguous with respect to underlying order. In addition, one kind of non-subject-initial sentence occurs, XSV<+fin> ..., in which the X is usually an adverbial (the order XSV<+fin> ... being ungrammatical in German). Finally, there is no evidence yet for syntactic embedding (i.e., no overt marking of subordination). On the basis of the word order patterns that both are and are not attested, and in line with our general hypothesis that the L1 grammar serves as the starting point for creating the Interlanguage system, we propose the analysis in (26):

In (26) the general phrase structure made available in Turkish is maintained: VP and the (verbally-related) non-COMP functional projections (which we reduce here without consequence to AGRP) are head-final. CP, in contrast, is head-medial, as indicated by the analysis of Turkish embedded clauses utilizing the complementizer ki 'that' (see Section 2.1).18 In addition, we follow work arguing that the subject NP originates inside the VP (Section 3). As (26) schematically shows, it is hypothesized that the finite verb vacates its base position, ultimately substituting into C° by movement through the intervening functional heads, and that in addition the subject raises first to Spec-AGRP and then on to Spec-CP. Finally, in order to capture the XSV<+fin> ... pattern, we

342

BONNIE D. SCHWARTZ & REX A. SPROUSE

propose that as one means to effect topicalization, the XP element adjoins to CP. While all of this taken together will account for the attested patterns and rule out the unattested ones, there are still two major questions that need to be addressed: Why does the verb move? And why does the subject — and only the subject (contrary to V2 German) — appear in immediate pre-verbal position? By attributing a general property of conservatism to the process of language acquisition, it is not unreasonable to place the bulk of the answer to both questions on the PLD that Cevdet receives. A major (and evidently salient) difference between the surface orders in Turkish and German is the possibility of SV<+fin>X(V<_fin>) in German main clauses, which in native German is also the most frequently occurring main-clause order. The finite verb must be placed somewhere to the left of the VP, and the pre-existing C° is in fact the only (functional) head position available. The minimal change that would need to be posited, then, without revising the basic phrase structure of the clause, is to move the verb to C°. For our purposes, we will treat the obligatoriness of this movement as a parametric stipulation. To explain why the subject occurs in pre-verbal position, and hence in Spec-CP, we follow an idea of Rizzi & Roberts (1989: 25, note 3), which has been elaborated in Roberts (1993). Roberts argues that there are consequences which derive from the intrinsic differences between the two types of structural relation relevant for Case: whereas Case checking under the government option is not affected by movement of the Case-checking head (e.g., a transitive verb), Case checking under the agreement option is; in short, moving the head out of a Spec-head configuration invalidates the Spec-head agreement relation for Case checking (and hence the trace of the head is always insufficient to satisfy a Spec-head relation).19 In particular, raising the AGR° complex after the verb has moved into it ([AGRo V°+AGR0]) destroys the environment for nominative Case checking of an NP in Spec-AGRP through the agreement option. Returning now to (26), the hypothesis that the AGR° complex substitutes into C° thus means that the subject cannot remain in Spec-AGRP and be checked for Case under the agreement option. If we further hypothesize that Case checking under the government option is not available, then the result is that the subject is forced to move from Spec-AGRP to Spec-CP in order to be in an appropriate Spec-head configuration with the inflected verb, which must move to C°. Therefore, raising the subject to Spec-CP is in fact the only way for it to satisfy the Case Filter. (See Rizzi & Roberts 1989:7 on nominative Case and the nonpronominal subject in French complex inversion.) Notice, also, that a consequence of limiting nominative Case checking to the agreement option at the

WORD ORDER AND NOMINATIVE CASE

343

CP level is that all V2 orders other than SV<+fin> ... (since the inflected verb inside AGR° occupies C°) are ruled out, for the subject would be unable to be checked for nominative Case in any other position. We build on this analysis to account for the XSV<+fin> ... data: as one means to realize topicalization (see also below), a (non-argumental) constituent adjoins to CP (cf. the English sentence Yesterday what did John see?). Two reasons lead us to this proposal: First, there is the fact that this pre-subject position is restricted to adjuncts (in contrast with the kind of topicalization that emerges in the next stage). The second reason stems from our analysis of SV<+fin>(X)... which forces the subject to be in Spec-CP. The XP topicalization of the XSV<+fin> ... pattern must be to the left of the subject — hence, adjoined to CP.20 A further consequence of this analysis of the SV<+fin>(X) ... pattern is that the absence of syntactic embedding with an overt complementizer is not surprising; if finite verb movement to C° is obligatory in Cevdet's Stage 1 system, then this must also obtain in what is functionally a sentential com­ plement of a higher verb; given the fact that at this stage agreement within CP is the only option for nominative Case checking, the embedded subject must also raise to Spec-CP in order to be checked for nominative Case. As a result, no position is available to house a grammatical marker of subordination (without the creation of an additional CP layer). While we leave the many issues of etiology aside until Section 6, we would like to point out the following: the small number of changes attributed to the Interlanguage system on this analysis in actuality captures a large amount of the data to which Cevdet is exposed, viz., the preponderance of German PLD SV<+fin>0(V<_fin>) (main clause) orders as well as part of the input indicating that subjects need not be the initial element of a German main clause. 5.2. Stage 2 Two innovations characterize the Interlanguage of Stage 2: Embedded sentences, whose sole order is SOV<+fin>,21 start to appear (first with wenn 'when'/'if', a bit later with daß 'that' and ob 'whether'). As for main clauses, the word order patterns of Stage 1 ((X)SV<+fin> ...) continue. In addition, however, a different kind of non-subject-initial sentence regularly occurs: XV<+fin>S ..., viz., inversion. The 'X' in this kind of sentence is usually an ADVP or PP but sometimes a direct object; interestingly, the subject is always a pronoun (S<+pron>). Our analysis begins with the Interlanguage data found in embedded clauses. Consider the tree in (27).

344

BONNIE D. SCHWARTZ & REX A. SPROUSE

In these subordinate clauses, the verb raises only as far as AGR°, because C° is filled by the complementizer. The subject's nominative Case feature must be checked under the agreement option at the AGRP level. In addition to the appearance of lexical complementizers, Stage 2 is characterized by the introduction of post-verbal pronominal subjects in main clauses. This is interesting, because it is only post-verbal subjects that display such an asymmetry between pronominal and nonpronominal subjects. These observations have led us to propose the analysis schematized in (28), based on the analysis of French inversion in Rizzi & Roberts (1989) (see Section 3) — in addition to maintaining the analysis given in Section 5.1 for SV <+fin> (X)... and XSV<+fin> .... Regarding the XV<+fin>S<+pron> ... order, there are two parts to our analysis. First, as an additional way to achieve topicalization, a nonsubject constituent is fronted to Spec-CR Second, we propose that subject pronouns in the Interlanguage are able to be analyzed as clitics. (See Section 6 for discussion of why this option might be taken.) Under the assumptions that Spec-CP is filled and that V<+fin> continues to move to C°, the only way a pronominal subject can have its Case feature checked by AGR° after the verb has raised to C° (in the absence of the government option) is through the incorporation option. (This is because, as discussed in Section 5.1, once the AGR° complex — containing the inflected verb — substitutes into C°, the Spec-head configuration at the AGRP level is destroyed.) The Rizzi & Roberts analysis can thus be used to explain the noted asymmetry, with respect to inversion, between pronominal and nonpronominal

WORD ORDER AND NOMINATIVE CASE

345

subjects. Hence, although much of the Interlanguage data of this stage appear to conform to the V2 constraint, the system that has been created still lacks the ability to structurally check the Case on a subject in Spec-AGRP if the verb has raised to C°. The XSV<+fin> ... order is of course further evidence that the Interlanguage does not yet have the characteristics of a full V2 system. Note, however, that the initial constituent in the XSV<+fin> ... pattern is no longer restricted to adjuncts. Yet, there is a reason for thinking that the two nonsubjectinitial patterns should still be kept structurally distinct: the subject in the XSV<+fin> ... pattern can be either <+> or <-> pronominal; this is simultaneously similar to what occurs in SV<+fin>(X) ... sentences but also distinct from the restriction to pronominal subjects in the inversion pattern. (X)SV<+fin>(X) ... orders thus receive a unitary account, separate from the one for inversion. Potential problems arise from this account of the Stage 2 Interlanguage data. Let us first consider possible implications from the analysis of inversion: why is the incorporation option not relied upon in the embedded context? In other words, it might seem that this analysis predicts the order COMP ... V<+fin>S<+pron>, with the subject cliticized on the V<+fin> that has raised to (right-peripheral) AGR0.22 As noted, subject clitic inversion in French is restricted to main clauses. The Rizzi & Roberts analysis of inversion has movement of V<+fin> to C° as a precondition (p. 11ff.); consequently, their explanation for the lack of inversion in nonroot contexts hinges on two generalizations (p. 18): Inversion is possible only if (i) CP is not selected and (ii) C° is not filled.23 In a clause with a lexical

346

BONNIE D. SCHWARTZ & REX A. SPROUSE

complementizer, then, condition (ii) is not met. Note that by adding the incorporation option much more main-clause PLD can be fitted into the developing Interlanguage system without effect on what is permitted in embedded clauses. Having examined the implications of the main-clause analysis for embedded clauses, we need to determine whether any effects in main clauses follow from the analysis given for embedded contexts. Specifically, how can (X)SOV<+fin> (with the inflected verb in AGR°) be ruled out if nominative Case can be checked in Spec-AGRP through the agreement option? Under the analysis that the finite verb always raises to C°, no consequences for main-clause word orders are expected. We recognize that this still begs the question as to why the finite verb is forced to move — but this is a problem for virtually all accounts of V2 (see Vikner 1991 for critical overview), which we do not intend to resolve here. On the assumption that the finite verb must move to C°, there are no reper­ cussions in the root context from the analysis of embedded clauses, and hence main-clause utterances in which the finite verb appears clause-finally are not expected. One final observation before turning to Stage 3. Although the analyses for main and embedded clauses differ, from other perspectives the analyses of the data are similar: with respect to all the non-inversion data, nominative Case can be checked under the agreement option; likewise, the data in both main and embedded contexts point to the unavailability of the government option for nominative Case. 5.3. Stage 3 The new data characterizing Stage 3 concern main-clause word orders only: inversion with nonpronominal subjects as well as with verb-first (declarative) utterances. All the other Interlanguage data described so far also continue to occur. The analysis that we propose for this stage in (29) again does not require revising the basic architecture of the clause. Comparing (29) with the earlier analyses of main-clause data, the nomina­ tive Case feature of post-verbal (nonpronominal) subjects is now checked under the government option by the finite verb (contained in the AGR° complex) that has moved to C° (cf. Koopman 1984; Platzack 1986). In this way, nonpronominal subjects (raised from their base position inside VP) are able to remain in Spec-AGRP to be checked for Case. (Note that (29) is irrelevant to the analysis of embedded clauses, and so we say nothing more about them here.)

WORD ORDER AND NOMINATIVE CASE

347

More than a year has elapsed since the first appearance of post-verbal pronominal subjects; this gap supports the hypothesis that the earlier inversion phenomenon was not to be assimilated with the inversion here. Thus, adoption of (rightward) government as a mechanism for nominative Case checking can be viewed as the ultimate step towards accommodating all the V2 data of main clauses. Note, however, that despite its greater data coverage, the created Interlanguage system is still far from being a syntactic equivalent of native German. Not only do XSV<+fin> ... orders still occur, but so do verb-first declaratives; both are (generally) ungrammatical in German. Nevertheless, the existence of neither type is evidence against the analysis in (29); indeed, while verb-first utterances with pronominal subjects are occasionally found at the earlier stages, at Stage 3 they become much more frequent and now also occur with nonpronominal subjects, as would be expected on this analysis. It is known that declarative verb-first is allowed in German under certain pragmatic conditions (typically in narrative or journalistic contexts). Reasons for the rise in verb-first utterances at Stage 3 and the continued existence of other 'non-German' sentences — as well as other learnability-related issues — are the next topics we explore.

348

6.

BONNIE D. SCHWARTZ & REX A. SPROUSE

Speculations on Etiology and Learnability in Adult L2A

As this longitudinal study investigates the acquisition of German by only one Turkish speaker, it can only be offered as preliminary. Attention has been focused on some of the word order phenomena that characterize the differences among the stages of development. There are many aspects of Cevdet's Interlanguage that fall way short of being 'target-like' (for example, the morphosyntax of nominals). Nonetheless, in addition to the facts that have been discussed, at the macro-level there is a definite expansion in Cevdet's Inter­ language over the 26 months that the transcripts cover. In the first transcripts Cevdet frequently resorts to Turkish, and it is obvious that he is at a loss to communicate in German. By the last transcript, he exhibits much more facility and control in German. Despite the ostensible growth in knowledge that underlies this more fluent linguistic behavior, the precise nature of the system that has been constructed is still unclear. It is certainly not German. Yet, we have tried to argue that at each particular stage of development, something new character­ izes the knowledge system, and moreover that each of these 'somethings' appears to be drawn from the apparatus of natural language grammars, at a level of abstraction far removed from the raw input data. In this sense, then, one goal of this study has been met: the aspects of Interlanguage we have examined are explicable (or at least statable) in terms of Universal Grammar. Important questions remain, however. For instance, on the assumption that our description and analysis of each stage are correct, why is this the exhibited order? From a practical perspective, Cevdet would have been better off had he immediately latched onto the 'correct' analysis of German. That this did not occur, however, is of course not a conundrum unique to L2A — for if first language acquisition were any different, there would be no need to talk of the 'idealization' of instantaneous acquisition. Nevertheless, one should still ask whether Cevdet's developmental path is an arbitrary one, a fully deterministic one, or one of a small number of possible developmental paths. While we have no comparative data to bring to bear on this global issue, in what follows we will attempt to argue that not only is this particular path of development not arbitrary, but that given the existence of these three stages, no other order would have been possible, and moreover that the relationship within each pair of successive stages (i.e., Turkish to (Stage 0 to) Stage 1, Stage 1 to Stage 2, and Stage 2 to Stage 3) is one that learnability considerations can illuminate to a significant degree. In an attempt to address the issue of etiology in L2 development, we start with a simple observation, one that of course Cevdet cannot 'know' before he

WORD ORDER AND NOMINATIVE CASE

349

starts to acquire German: while the architecture of the sentence in Turkish and German share many basic structural similarities, the two grammatical systems are, nevertheless, very different. Assuming our model of L2A, the only thing that must quickly be apparent is that the grammar that has previously been constructed is unable to assign appropriate structural representations to the German PLD. Thus, unlike the child acquiring L1 German, the non-native acquirer approaches the language acquisition task with a fully fleshed out grammar, which means that certain parametric options are necessarily already instantiated. This includes, for example, headedness values for lexical and functional categories as well as the mechanisms for nominative Case checking. On the other hand, similar to L1A, input that is not analyzable by the current grammar should force new hypotheses to be adopted. Potential solutions to the developmental problem in L1A differ, and some are seemingly inapplicable to adult L2A (maturation of UG principles, for example — cf. Borer & Wexler 1987). Whatever the answer, the process of L1A is deterministic in that the L1 child eventually hits upon one solution which necessarily accommodates all the input data: 'the' German grammar.24 The question is whether L2A can be deterministic in this sense as well, in spite of the fact that the starting points of L1A and L2A differ — and if not, why not. In L1A, then, the end-state comes from 'input plus (maturing/activated) UG'. Under a model that claims the L1 grammar to be instrumental in L2 development, in contrast, there are several ways to conceive of the changes that need to be made: (i) Part of the machinery of the L1 grammar is relinquished; (ii) A new property or mechanism of grammar is appended to the current system; (iii) Part of the machinery of the L1 grammar is replaced by a new property or mechanism. Moreover, if UG is indeed operating in L2A, one would expect there to be changes that arise for purely grammar-internal reasons, that is, a certain chain of repercussions that follow from (i), (ii) or (iii) (for instance, the impossibility of logical contradictions, e.g., V' cannot be both head-initial and head-final). Thus, in order to move directly (and instantaneously) from the L1 grammar to 'the' German grammar, numerous simultaneous relinquishments, additions and replacements would be entailed, some of which may in fact not be possible on the basis of positive data alone (unlike in L1A — see below). Which set of input data, together with (unspecified) UG and the L1 grammar, would be necessary (and sufficient) in order to create 'the' German grammar? Let us bear in mind two more factors: (a) It is unlikely that a store of actual input sentences is retained on the part of the acquirer (Wexler & Culicover 1980); and (b) Taken individually, many (if not in fact all) of the input sentences to which Cevdet is

350

BONNIE D. SCHWARTZ & REX A. SPROUSE

exposed are multiply ambiguous as to possible analysis (Clark & Roberts 1993). In sum, from a computational perspective, the number of possibilities that would need to be computed in order to move straight from the L1 grammar to 'the' German grammar would seem to be enormous; in this sense, smaller adaptations made to the existing system can be seen as more highly valued, for the precise effects of small and/or localized, incremental changes can be more easily computed with respect to the incoming PLD. The extent to which (i), (ii) and (iii) lead or can lead to a real 'restructuring' of the Interlanguage system would seem to be an important issue in the etiology of non-native language acquisition. 6.1. The Etiology of Stage 1 We began our analysis of Stage 1 by claiming that one aspect of the input data that is very quickly acted upon is that in German — unlike in Turkish — the finite verbal element in main clauses does not surface in clause-final position (except for 'short' SV<+fin> sentences). Intuitive reasons for thinking that this must be 'salient' to the acquirer include the fact that it is entirely distinct from what occurs in Turkish, that it happens (virtually) without exception in main clauses (and thus is extremely frequent) and that the finite verb (of main clauses) is then of course closer to the start of the sentence and thus even partial attempts to parse the incoming data will be made on (sub)strings generally containing a finite verb. Moreover, assuming that at least with respect to thematic verbs, the verb is the nucleus of the sentence, it is not surprising that the position of the finite verb is one of the first aspects of the new PLD to be fitted into the underlying system. Recall that SV<+fin>X(V<_fin>) sentences are by far the most frequent mainclause order in the input. Unless the initial state of the Interlanguage is revised, the vast bulk of the PLD will not be parsable. The minimal change that can be effected to accommodate a significant proportion of the data is to move the finite verb to the C° position that the L1 grammar makes available. The pre-existing structure of the sentence can readily incorporate this hypothesis; nevertheless, there are certain consequences, as already discussed. With V<+fin> in C°, how can the subject have its Case checked? The (nonpronominal) subject NP must either move to Spec-CP to be in an appropriate configuration for Spec-head agreement or it can remain in Spec-AGRP and the government option would have to be added; if neither occurs, a grammatical violation would result. On the one hand, the government option (which includes government of the trace of an NP subject that has raised to Spec-CP) would cover more data. On the other hand, whereas

WORD ORDER AND NOMINATIVE CASE

351

nominative Case checking under the government option is not implicated in the L1 grammar, the agreement option is. Based on the distribution of occurring and non-occurring sentence patterns at the first stage, we claimed that it is the agreement option (at the CP level) that is instantiated. What might this tell us about the process of non-native language acquisition? In Section 2.1 our brief sketch of Turkish had Case checked on the subject through the agreement option at the AGRP level. On this assumption, then, the data of Stage 1 seem to indicate that the possibility of making use of a pre-existing mechanism (the agreement option) is apparently more highly valued than adding a new one (the government option), even if data coverage suffers. Other implications come from comparing the SV<+fin>X(V<_fin>) pattern with the other sentence pattern included in this first stage: XSV<+fin> .... We point out that partial parses of (main clause) XV<+fin>S ... PLD will include this initial X constituent. We have speculated that in an attempt to locally approximate the placement of the initial constituent in non-subject-initial sentences, the adjuncts are adjoined to CP — unlike what the German grammar permits and thus unlike what is included in the PLD. Under the analysis moving the verb to C° with the contingent consequence of moving the subject to Spec-CP, no other possibility exists (except to ignore these non-subject-initial data). Viewing the acquisition data this way indicates that creation of the knowledge system cannot be solely dependent on the PLD — for otherwise, with respect to the TL, one would not expect any such 'errors of commission'. This then suggests two other aspects of the non-native language acquisition process: one overriding organizational 'strategy' is that violations of grammatical principles cannot be tolerated. Note, however, that the addition of this type of (optional) adjunction does not produce a grammatical violation. Indeed, research in L1A has recently argued that the process of adjunction is an essential component in the creation of grammars (Hoekstra & Jordens, this volume; Roeper 1992, following Lebeaux 1990 and Tavakolian 1981); the claim is that in building the architecture of the grammar, lexical elements that cannot be reconciled into the pre-existing (nominal or clausal) structure always first adjoin to the pre-existing structure and then may (but need not) later force the projection of further syntactic structure. Extending this hypothesis to L2A, it seems, then, that these XSV<+fin> ... Interlanguage 'errors of commission' stem from an operation that not only (crucially) does not violate any principle of grammar but also simultaneously has little effect on the rest of the internal workings of the system. Finally, and perhaps most notably,

352

BONNIE D. SCHWARTZ & REX A. SPROUSE

adjunction in general has been singled out as being fundamental to the process of L1A. All in all, the first stage of Interlanguage seems to depend in large part on two aspects of the input data: the frequency of SV<+fin>X(V<_fin>) (which high­ lights the non-sentence-final position of finite verbs in main clauses) and nonsubject-initial sentences. Regarding the latter, the analysis points to adding the possibility of adjunction, implicating a mechanism claimed to be essential to L1 grammar construction. However, we will later pursue the ultimate effect of this particular instantiation of adjunction. As for the former, the phrase structure of Turkish provides the landing site for the verb to raise into. Once verb raising to an empty C° becomes obligatory in the Interlanguage grammar, nominative Case checking will involve Spec-head agreement not at the AGRP level, but at the CP level. 6.2. The Etiology of Stage 2 Ideally, one would like to find a single cause to explain the simultaneous appearance of the two seemingly unrelated new phenomena in Stage 2: syntactically identified subordinate clauses and matrix-clause inversion with (pronominal) subjects. In this case, however, a unified explanation seems unlikely. Our analysis of the embedded-clause data relied on the identification of lexical items as subordination devices and the Spec-head relation at the AGRP level for Case checking. The main-clause phenomenon, in contrast, was claimed to be due to an analysis under which subject pronouns are analyzed as clitics and nominative Case is checked through the incorporation option. Assuming our analysis to be correct, again we should ask what it might tell us about the process of L2A. Recall that in Stage 1, we found only one kind of functional embedding and this mirrored the main-clause data.25 Given the (superficial) similarities between embedding in German and at least one kind of embedding in Turkish, one wonders why no syntactically identified embedding occurs earlier. Part of the answer may stem from Cevdet's simply not having yet acquired the necessary lexical items. On the other hand, the account given for Stage 1 (V<+fin> raising to C°; the subject NP moving to Spec-CP) may suggest a complementary reason. At this point it is perhaps worth mentioning that the language acquirer's perspective must be differentiated from the linguist's perspective. In particular, the fact that the (epi)phenomenon of V2 is (generally) restricted to root sentences is presumably not a descriptive distinction relevant to the creation of the

WORD ORDER AND NOMINATIVE CASE

353

underlying system. This is to say, if UG operates in L2A, then the properties of the knowledge system that is created must interact in such a way as to derive the asymmetry between main and subordinate clauses. With this in mind, then (under our analysis), it is not surprising that the word order of the (functional) embeddings of Stage 1 matches the word order of main clauses. What would be unexpected, however, is an asymmetry between main and embedded clauses of the following type: SVO-SOV, respectively — but without the subordinating lexical items. The fact that this does not occur suggests that Cevdet is not merely implementing some sort of 'strategy' to simply match the embedded-clause input that he hears. Why might there not be such an asymmetry, especially since Turkish does instantiate the agreement option at the AGRP level for nominative Case checking? While the hypothetical possibility of checking nominative Case through the agreement option at the AGRP level is in principle never lost from the Interlanguage system, it is only upon the identification of lexical items which need to be fitted into the embedded CP level that checking at this level is 're-activated'. At Stage 1, Cevdet acquires verb movement to C° and this movement occurs categorically. The exceptionless nature of this pattern suggests that it is a reflex of the same parametric choice which is ultimately responsible for verb movement to C° in 'real' V2 grammars. At Stage 2, Cevdet acquires lexical complementizers. In clauses with a lexical complementizer in C°, finite verb movement to C° is precluded. The acquisition of these items thus forces the requirement on verb movement to be relaxed in such a way that the finite verb must move to C° only in the absence of material in C°. Now, in those clauses in which material is base-generated in C°, Spec-head agreement can check Case at the AGRP level without the violation of any independent requirements of the system. In sum, at Stage 2, the learning of particular lexical items feeds into the (re-)implementation of Spec-head agreement for Case checking at the AGRP level, which was not possible in Stage 1. One final observation regarding Stage 2: if the government option for nominative Case checking were available, the fact that inversion (with nonsubject topicalization to Spec-CP) is restricted to subject pronouns in main clauses would be left without an explanation. Note that again data coverage for main clauses suffers by not selecting the government option. Nonetheless, much more input (compared to Stage 1 ) can be captured by analyzing subject pronouns as clitics and adding the incorporation option — with no effect on embedded clauses. The intuition underlying the Subset Principle (Berwick 1985) may provide an explanation for why the incorporation option is chosen over the

354

BONNIE D. SCHWARTZ & REX A. SPROUSE

government option. Consider the diagram in (30), which represents the set of all non-subject-initial sentences to which Cevdet is exposed: (30)

Main-clause input with post-verbal subjects

As illustrated in (30), the set of non-subject-initial V2 sentences comprises those with pronominal and those with nonpronominal post-verbal subjects, but the set containing pronominal subjects and the (non-intersecting) set containing nonpronominal subjects are each (necessarily) subsets of the entire set (of nonsubject-initial V2 sentences). Regarding non-subject-initial V2 sentences, the incorporation option allows the set of sentences with pronominal subjects to be separated out from the others, whereas there is no nominative Case-checking mechanism applicable to only the nonpronominal post-verbal subject counterpart. In other words, while there exists no mechanism of grammar to separate XVS<-pron> from the set of main clauses with post-verbal pronominal subjects in (30), the incorporation option will allow for only XVS<+pron> to be generated. Conservatism on the part of the acquirer, then, might lead to the adoption of the incorporation option, for while it enlarges the amount of data that can be accommodated, it does not lead to an overly encompassing hypothesis from which it would be impossible to retract.26 Summarizing: In Stage 2, several different causes for the move from Stage 1 to Stage 2 in the Interlanguage have been proposed. First, the acquisition of particular lexical items that need to be fitted into the system had the consequence of 'activating' another mechanism for checking nominative Case. Second, a different sort of topicalization which placed a nonsubject element in Spec-CP was added. And third, we claimed that a learning procedure respecting conservatism was the cause underlying the acquirer's hypothesis that subject pronouns are clitics which can satisfy the Case Filter by incorporating into a c-commanding finite verb. Whether all intermediate Interlanguage hypotheses are such that they allow retraction is an issue that we will consider further in what follows. Finally, recall that in Section 5.2, we hypothesized that with respect to inversion, the new type of topicalization to Spec-CP is what forces the addition of a new means for (pronominal) subjects to satisfy the Case Filter. Logically, of course, it could be that analyzing subject pronouns as clitics, which then feeds into the incorporation option, frees up Spec-CP as a possible landing site for topicalization. For the moment, we put aside these two possibilities, for it is the analysis in Stage 3 that provides the clue to sorting them out.

WORD ORDER AND NOMINATIVE CASE

355

6.3. The Etiology of Stage 3 We turn now to the final stage of the data. Recall that here we have analyzed the Interlanguage as adopting the government option for nominative Case checking, thus accommodating sentences exhibiting inversion with nonpronominal subjects in the PLD. While all the V2 patterns are generable at this stage, it is not the case, however, that there is a concomitant disappearance of XSV<+fin> ... orders. The latter suggests that whatever the underlying explanation for the phenomenon of V2, it has not been acquired; in other words, although the system (as we have described it) has XP movement to Spec-CP and verb movement to C°, this is perhaps nevertheless not equivalent to 'real' V2. If it were, one might expect adjunction to CP to be relinquished at this stage, but it is not.27 This highlights an intriguing learnability situation, one that has recently been noticed and addressed in other L2A contexts, namely, the problem of 'delearning' (Gubala-Ryzak 1992; Schwartz & Gubala-Ryzak 1992; White 1992).28 Specifically, the issue here is how it will be possible, without negative evidence, for adjunction to CP to be abandoned. What PLD can force Cevdet to construct a system that would rule out what are epiphenomenally V3 sentences? Logic seems to dictate that there are no such (positive) data; nonetheless, it would be important to determine empirically whether a non-native speaker like Cevdet ever does relinquish adjunction to CP. If it ultimately is possible, then several nontrivial implications for L2A, which need to be distinguished and which have not yet been investigated in any detail, could follow: (i) Negative data are able to cause the abandonment; (ii) Indirect negative data (i.e., the noticing of the non-occurrence of V3 in the input) are able to impinge upon the process of L2A; or (iii) The internal apparatus of the system somehow conspires to effect the change (on the third possibility, as it relates to L1A and diachronic change, see Clark & Roberts 1993). The simplest hypothesis at this point, it seems, is to predict that since there are no PLD that could force the abandonment of adjunction to CP, this adjunction is not delearnable. Note that this hypothesis suggests, in a fairly precise manner, a way in which L1A and L2A differ: L1A is in all (nonpathological) cases deterministic; it is never the case that native-language acquirers find themselves in a learnability predicament from which it is impossible to retract and hence unable to build a grammar that is the strong generative equivalent to that of the input providers. The type of analysis given here, coupled with the extension of the 'no negative evidence' axiom driving L1A research, offers a way to explain the 'lack of success' in (adult) L2A. It is not that UG is inoperative

356

BONNIE D. SCHWARTZ & REX A. SPROUSE

in L2A (nor is it the case that learning principles like the Subset Principle are unavailable); on the contrary, the lack of succes may arise exactly because the sole hypotheses that an L2er can employ in the construction of an L2 grammar are those that UG makes available, but as they apply in conjunction with the L1 grammar, these hypotheses will in certain circumstances miss the mark and without (the necessary) negative data, the L2er will be unable to retract.29 Notably, the specific cause of this particular delearning problem lies in adjunction, an operation implicated as one of the fundamental means by which Llers attempt to build the architecture of the grammar (Hoekstra & Jordens, this volume; Roeper 1992). In short, in the process of creating the L2 knowledge system, (UG) mechanisms that are adopted in attempts to incrementally and locally accommodate PLD may be such that no further PLD will be able to engender retraction. In this sense there is a Determinacy Problem in L2A. Our intuition is that the number of individual L2A situations that instantiate this logical predicament will not be very large, for in most cases there will be positive data to overcome intermediate hypotheses. Nevertheless, the precise input data needed may have to be of a more exotic kind (and will hence be less typically robust) than needed for L1A. Notice that there is a consequence of this too: what works as 'triggering data' in L1A may not necessarily be triggering data in L2A (Schwartz, forthcoming). In fact, what counts as triggering data is never absolute.30 While these speculations certainly go far beyond what can be empirically substantiated on the basis of this small study, they can perhaps suggest some insight into the long-known problems of fossilization and 'lack of success' in L2A. Returning to other aspects of etiology in our Stage 3, recall that there is additional evidence for saying that the German instantiation of V2 is not yet part of the Interlanguage system. This is shown by the opposite violation of V2, namely, non-topic-drop, declarative verb-first utterances (V1), such as in (25a) above, repeated here as (31): (31)

hat viele Menschen zu ihm gehorcht has many people to him obeyed 'Many people obeyed him.'

In verb-initial examples such as (31), all of the arguments of the verb are overtly realized, so there is no null pronoun in Spec-CP. The increased usage of such V1 main clauses at this stage may shed light on the possible interdependence relations outlined earlier for Stage 2 regarding topicalization to Spec-CP and the adoption of subject-clitic incorporation.

WORD ORDER AND NOMINATIVE CASE

357

We suggest the following chain of dependencies: At Stage 2, nonsubject topicalization to Spec-CP forces a new way for subjects to satisfy the Case Filter; conservatism leads to the choice of the incorporation option, because it generates the smaller language. Yet, inversion with nonpronominal subjects falls outside this system, and thus the failure to assign a well-formed representation to the other non-subject-initial V2 input (viz., with nonpronominal subjects) forces the adoption of the government option at Stage 3; it is not until this point, upon movement of the verb to C° in main clauses, that nominative Case can be checked in Spec-AGRP, thereby freeing up Spec-CP, which at the two previous stages had been the sole position licensed for nonpronominal subjects (when C° has no material in it). In Cevdet's system, when Case checking through the government option allows the subject to remain in Spec-AGRP, (overt) topicalization is not forced. Hence, the rise in V1 main clauses turns out to support our analysis that nominative Case checking under the government option has been added to the system, since V1 with nonpronominal subjects reduces to a special Case of the government option. Why should the familiar requirement that Spec-CP be filled be relaxed in Cevdet's system, when there is little or no direct evidence for V1 in the PLD? We believe that the explanation lies in the fact that Cevdet, unlike native acquirers of German, posits adjunction to CP. Once Cevdet has adopted the government option for nominative Case checking, he can successfully parse main-clause input in which there is inversion with a nonpronominal subject. Given that the Interlanguage in Stage 3 has inherited both adjunction to CP and substitution into Spec-CP, the system will now be able to assign a structural representation to any XV<+fin>S ... sentence in one of two ways: either with the XP adjoined to CP, as in (32a), or with the XP in Spec-CP, as in (32b). (32) a. b.

[ C P XP[ C P Ø [C'V<+fin>[AGRPS...]]]] [CP XP [C' V<+fin> [AGRP S ...]]]

In (32a), Spec-CP is 'radically' empty. In so far as adjunction to CP is always optional, it follows that V1 main clauses should be well-formed. In sum, given the pre-existence of certain parameter values in Cevdet's system, at this point the PLD are analyzed in such a way that they end up being a kind of 'direct positive evidence' for the possibility of radically empty Spec-CP and thus, by deduction, for V1, although such clauses per se are generally absent from the PLD. The observation that adjunction to CP has not been abandoned at Stage 3 prompts us to ask analogous questions regarding two further aspects of our

358

BONNIE D. SCHWARTZ & REX A. SPROUSE

proposals for earlier stages: (i) the analysis of subject pronouns as clitics and incorporation; and (ii) nominative Case checking through the agreement option at the CP level in main clauses. Concerning the status of atonic nominative pronouns as clitics and of incorporation, learnability considerations may seem to suggest that since the government option offers a unified analysis of all non-subject-initial V2 main clauses, the incorporation option would be relinquished, because it has become superfluous. Empirically, it is unclear whether the incorporation option has in fact been forced out at Stage 3. However, as we saw in Section 3, it is equally unclear whether the incorporation option for nominative Case checking is implicated in the grammar of native German, since it is unclear whether the encliticization of post-verbal atonic nominative pronouns obtains at S-structure or 'only' at PR Thus, we must leave this matter unresolved. Likewise, one might wonder about the status of nominative Case checking under the agreement option at the CP level at Stage 3. Empirically, there is no way to test between this and what government of Spec-AGRP allows: all of the sentences that can be generated under the agreement option are subsumed under the government option. Similarly, as far as we can tell, there is no way to argue on grounds of learnability (except uniformity of analysis, perhaps) that the government option has forced out the agreement option at the CP level. Thus, this matter must also be left open. In Table 6 we repeat the summary of the parametric values for Turkish and German given at the end of Section 3. Here we fill in the values for Cevdet's three Interlanguage Stages (S1, S2, S3). Table 6. Parametric Values for Turkish and German; Cevdet's Three Interlanguage Stages

SOV VP-AGR C-AGRP as possible order Nominative checked under agreement Obligatory movement of finite verb to empty COMP Left-Adjunction to CP possible Spec-CP as landing site for topics Nominative clitics Nominative checked under incorporation Nominative checked under government

Tur

S1

S2

S3

Ger

+ + + +

+ + + + + + -

+ + + + + + + + + -

+ + + + + + + + ? +

+ + + + + + + ? +

-

WORD ORDER AND NOMINATIVE CASE

7.

359

Conclusion

As Section 6 underscores, in regard to precisely what does and what does not exist in the system underlying the Interlanguage at Stage 3, much remains inconclusive. Notice, however, that without the descriptions and analyses of the data from Stages 1 and 2, no such discussion would have been conceivable. This highlights the import of looking at longitudinal L2 data, for had we examined only the last stage in Cevdet's development, these issues would not have even arisen. Yet, the known differences between L1A and (adult) L2A make such reflections about the succession of Interlanguage stages indispensable to L2A research interested in the nature of Interlanguage systems, especially when considered in light of the Continuity Hypothesis, for the following reason. By definition, the final state of L1 A is a natural language grammar; unlike the result of L1 development, the 'final state' of L2A is not uniform across L2 acquirers (even holding the L1 constant) and hence is not known. Note that in the context of native language, Continuity functions as a working hypothesis that constrains the types of analysis that can be attributed to each developmental stage. The question of fact concerning Continuity and L1 development notwithstanding, Continuity itself has no bearing on the relationship between UG and the final-state grammar in native languages. In L2 development, in contrast, since what constitutes the 'final state' differs nontrivially among acquirers, Continuity must function as more than just a working hypothesis in L2A research; in so far as any particular point in L2 development could be some L2er's 'final state', then the Continuity Hypothesis becomes a necessary mediator in the attempt to determine the relationship between Interlanguage and UG {contra Hong 1991).31 In other words, in order to ask whether the nature of Interlanguage(s) is the same as that of natural languages, one needs to show that each stage of L2 development in fact conforms to what UG allows. For example, had Stage 2 not been amenable to an analysis utilizing the principles of UG but Stage 3 had been, it would be difficult to make positive claims about the relationship between UG and Interlanguage in general. And it is here that considering the succession of stages in Interlanguage might prove to be crucial: our discussion in Section 6 tried to illustrate that knowing what the earlier states are in Interlanguage turns out to be essential in interpreting later states. In our introduction, we stated that the overall goal of this study was to illustrate one way in which linguistic theory could be implemented to shed light on research in non-native language acquisition. We believe that through the specifics of syntactic analyses we have come some distance toward that goal.

360

BONNIE D. SCHWARTZ & REX A. SPROUSE

While we have explored only a very limited area in the L2 development of only a single subject, all of the principles and mechanisms we have invoked to capture the data of each stage are independently motivated from the syntax of natural language grammars. Nevertheless, it should go without saying that one should be cautious about what to conclude. We realize that much remains unknown. Would all native speakers of Turkish follow this same path? Would some skip a stage? Has Cevdet skipped a stage that a different native Turkish speaker would more unambiguously exhibit? To what extent does the L1 grammar impinge upon the latter stages in the L2A process? Can adjunction to CP ever be abandoned? Has nominative clitic incorporation been abandoned? If so, on the basis of what kind of data and/or which learning principle do properties of earlier grammars get forced out? Can this Interlanguage 'grammar' ever become the strong generative equivalent of the German grammar? To what extent does the Determinacy Problem characterize (adult) L2A? That we are left with perhaps more questions than answers is needless to say not in itself a bad sign for L2A research. It is our hope that this study has also made some headway with respect to the other two goals we originally enumerated. While we crucially employed certain features of the L1 Turkish grammar in our explanation and while we also considered the facts of the German PLD that Cevdet was most likely to have encountered with some frequency, we tried at all phases to consider the system underlying the Interlanguage in terms of its own internal consistency rather than from the perspective of the linguistic analysis of the TL. Regarding the etiology of L2 development, despite the general inconclusiveness, we feel that questions of learnability should be given much more prominence in L2A research and we hope to have made a move in that direction. We close by making explicit what we feel this study may suggest in regard to the differences and similarities between the process of L1A and the process of L2A. Several differences have been proposed throughout the paper: the influence of the L1 grammar; the problem of relinquishing 'wrong' hypotheses; the Determinacy Problem caused by the absence of necessary data. Nevertheless, our claim has been that the overall processes of L1 A and L2A are very similar: Intermediate systems are constructed in attempts to reconcile progressively larger amounts of PLD; the mechanisms that are utilized in these systems are restricted to those that Universal Grammar makes available; the general operation of adjunction as found in constructing the architecture of L1 grammars is implicated; and the adoption of one potential hypothesis over another follows from local approximations based on partial parses of the PLD, from learning lexical items, and from constraints within the language learning procedure (such

WORD ORDER AND NOMINATIVE CASE

361

as the Subset Criterion). The aim of trying to show that each successive state in Interlanguage emerges on the basis of the interaction of the L1 grammar, (positive) input, principles of UG and aspects of a (language) learning procedure is, we feel, one certainly worth pursuing, because such a research program should prove capable, to a considerable extent, of illuminating our understanding of the non-native language acquisition process.

Acknowledgements This publication is based on data collected within the European Science Foundation Additional Activity 'Second Language Acquisition by Adult Immigrants' by the team of Rainer Dietrich. The data were made available to us by the Central Data Archives of this project, located at the Max Planck Institut für Psycholinguistik, Nijmegen, The Netherlands. We wish to express our gratitude to all those concerned for generously making the interview transcripts available to us. This is a significantly revised version of talks given in 1991-1992 at the following venues: Princeton, Harvard, the GLOW Workshop on The Develop­ ment of Movement and Inflection (Leiden), Theory Construction and Methodolo­ gy in Second Language Acquisition Research (Michigan State University), the Boston University Conference on Language Development, and the University of Düsseldorf. We wish to thank our audiences for helpful questions and comments. Much appreciation is extended to the following people, who may have helped us more than they know: Harald Clahsen, Robin Clark, Lynn Eubank, Maria-Teresa Guasti, Gabriella Hermon, Jaklin Kornfilt, Elaine MacNulty, Ian Roberts, Monika Rothweiler, Engin Sezer, Alessandra Tomaselli, Tom-Tom, Sten Vikner, Martha Young-Scholten, and (C. J.) Wouter Zwart. Notes 1.

See the brief remarks on methodology below.

2.

In using the word grammar we do not intend to imply that one knows in advance that the knowledge system an L2er creates necessarily conforms to the formal properties that are characteristic of either full-fledged natural language grammars or even developing L1 grammars. Thus, to avoid bias we will generally use the term system.

3.

Alternatively, one might tie changes to the invariant principles of UG operating in the Interlanguage system in a way that, for independent reasons, does not arise in the L1 grammar. See Schwartz (1992) for an example of this kind of analysis.

362

BONNIE D. SCHWARTZ & REX A. SPROUSE

4.

The ultimate purpose for this is, reasonably, to understand the discrepancies so that they can be 'treated' and 'cured'.

5.

For example, no linguist would want to base an analysis of English on the prohibition against preposition stranding, even though if one were to ask many 'naive' native English speakers who have had a rather traditional education in English, they might well say something like 'A preposition is a word that you can never end an English sentence with'.

6.

In a certain sense, such a methodology relying on argumentio ex silencio is reminiscent of the constraints under which diachronic (generative) syntax must proceed.

7.

We wish to thank Jaklin Kornfilt and Engin Sezer for discussion, help with data, and the many insights they shared with us about Turkish. We refer the reader to Kornfilt (1990) for further background on the syntax of Turkish.

8.

Turkish allows extraposition of 'old information' to the right of the verb. Hankamer (1979: 162, 197) has shown that this corresponds to material which could also be optionally deleted within the same discourse context. We will abstract away from this possibility here.

9.

Here we follow standard Turkological practice of representing underspecified segments ('archiphonemes') with upper case letters. For detailed discussion of underspecification in Turkish, see Clements & Sezer (1982).

10.

Following Baker (1988), we assume that incorporation of nonpronominal subjects is excluded.

11.

The details of the Rizzi & Roberts analysis need not concern us, as we rely only on the part that pertains to subject pronouns in 'subject-clitic inversion', as in (15a) and (15c), which they further extend to French complex inversion, as in Jean est-il allé?

12.

The dates of the interviews are the following: 2 July 1982; 22 October 1982; 10 February 1983; 9 June 1983; 14 July 1983; 1 November 1983; 27 January 1984; 29 February 1984; 19 April 1984; 16 May 1984; 5 August 1984.

13.

Thus, our counts include only clauses with at least two nonverbal constituents. Since no examples in our corpus exhibit the fronting of nonfinite verbs or of the negation element nicht 'not', we assume that these are not possible targets of fronting in Cevdet's Interlanguage systems; thus, neither of these elements was counted toward the two non­ verbal constituents required for inclusion in our tabulation. Furthermore, we exclude copular and existential-presentative sentences from consider­ ation, since the Case-assignment mechanism relevant for them is unclear (but see Maling & Sprouse 1992 on the Case of predicate NPs). In principle, we would also exclude pat phrases, such as ich weiß nicht 'I don't know'; however, in general, such phrases independently fail to meet our criteria for inclusion.

14.

This may be attributable to the well-formedness of null topics in Turkish or simply to Interlanguage performance limitations.

WORD ORDER AND NOMINATIVE CASE

363

15.

The standard analysis of this pattern in German takes the topic to be a null pronoun in clause-initial position, immediately followed by the finite verb. Thus, the clause is only superficially verb-initial.

16.

In the entire corpus we have found only one exception: (i) daß der Zug fährt ab that the train goes away 'that the train is departing'

17.

(19 April 1984)

This should be contrasted with the uncontested acceptance of SVO as the underlying order that native (null subject) Romance speakers initially adopt in their acquisition of German. (See Clahsen & Muysken 1986, 1989; duPlessis et al. 1987; Schwartz & Tomaselli 1990.) Thus, early stages in the L1 Romance, TL German corpus include sentences like the following: (i) mein Vater hat gekaufen ein Buch my father has bought a book 'My father bought a book.' (from Jordens (1988:151, ex. 8)

18. We assume that head-medial rather than head-final CP is adopted in Cevdet's Stage 1 grammar as a result of German PLD. 19. More precisely, Roberts argues that it is the inherent difference between the two types of structural relations that can explain why there is a 'Government Transparency Corollary' (Baker 1988), but no analogous 'Agreement Transparency Corollary'. In turn, as Rizzi & Roberts (1989:25, note 3) point out, due to the Government Transparency Corollary, a raised verb can still govern its object (and, hence, check its Case features), but since there is no Agreement Transparency Corollary, moving a head destroys the Spec-head configuration, ruling out Case checking through Spec-head agreement. 20.

In a sense, this may be seen as an extension of the rather free adjunction to IP in Turkish to adjunction to CP. See below.

21.

This is unlike what has been found in the L2A of German by (null subject) Romance speakers, whose placement of the finite verb in embedded clauses begins in immediate post-subject position and continues as such until a much more advanced stage. (See Clahsen & Muysken 1986, 1989; Meisel, Clahsen & Pienemann 1981.)

22.

We thank Alessandra Tomaselli (pers. comm.) for bringing this to our attention.

23.

See Rizzi & Roberts (1989:18-20) for their attempt to derive generalization (i) from the Projection Principle.

24.

This abstracts away from 'imperfect' acquisition as a source for diachronic change. See Clark & Roberts (1993) for an attempt to deal with the inherent tension between language acquisition and historical development with a precisely defined approach to learnability and parameter setting. Much of the discussion in this section was indirectly inspired by that paper.

25.

With the emergence of overtly identified subordination, the other kind of embedding does not completely disappear. In German, without a lexical complementizer, embedded V2 is the sole possible word order.

364

BONNIE D. SCHWARTZ & REX A. SPROUSE

26.

Such an explanation, it should be noted, is contrary to White's (1989a, 1989b) hypothesis that the Subset Principle, as a learning principle independent of UG, does not operate in L2A.

27.

A similar conclusion concerning the reluctance to de-hypothesize adjunction (in this case, to AGRP) has been found by duPlessis et al. (1987) in their study of English and French native speakers acquiring German in a tutored L2 context.

28.

See Bley-Vroman (1986) for early suggestions on potential problems of retraction in L2A.

29.

This is not to suggest that negative data are actually usable in L2 'grammar construction'. (See Schwartz 1993, Schwartz & Gubula-Ryzak 1992; for a different position, see White 1987, 1991, 1992.)

30.

Clark & Roberts (1993:341) come to a similar conclusion in their discussion of input and native language acquisition: "[E]xactly the same string Si can successfully trigger a parameter setting P(V1) in one grammatical system Gi but fail to trigger P(V1) in system Gj ≠ Gi."

31.

Stipulating that Interlanguages are natural languages does, of course, circumvent the problem, but this begs the question at hand. Consider, for example, the discussion of the possible relations between linguistic theory and L2A data in Gass & Schachter (1989), the strongest of which states: "[L]inguistic theory, because it is a theory of natural language, must be tested against second language data to be validated. Thus, any theory of language would be false if it failed to account for second language data" (p. 5). In a similar, albeit slightly weaker vein, Flynn & O'Neil (1988: 19), as they outline future directions for UG-related L2A research, state: "We must (...) make sure that the questions asked are theoretically motivated. They must reflect what we know to be true about language and learning in general. They must also be articulated within frameworks that will allow us to generate unified theories of language acquisition in general."

References Baker, Mark. 1985. "The Mirror Principle and Morphosyntactic Explanation." Linguistic Inquiry 16.373-415. . 1988. Incorporation: A theory of grammaticalfunction changing. Chicago: The University of Chicago Press. Bayer, Josef. 1991. "On the Origin of Sentential Arguments in German and Bavarian." Ms., University of Düsseldorf. Berwick, Robert. 1985. The Acquisition of Syntactic Knowledge. Cambridge, Mass.: MIT Press.

WORD ORDER AND NOMINATIVE CASE

365

Bley-Vroman, Robert. 1983. "The Comparative Fallacy in Interlanguage Studies: The case of systematicity." Language Learning 33.1-17. . 1986. "Hypothesis Testing in Second Language Acquisition Theory." Language Learning 36.353-376. Borer, Hagit & Kenneth Wexler. 1987. "The Maturation of Syntax." Parameter Setting, ed. by Thomas Roeper & Edwin Williams, 123-172. Dordrecht: Reidel. Chomsky, Noam. 1975. Reflections on Language. New York: Pantheon. . 1981. Lectures on Government and Binding: The Pisa lectures (= Studies in Generative Grammar, 9.) Dordrecht: Foris. Clahsen, Harald. 1991. "Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language." Language Acquisition 1.361-391. Clahsen, Harald & Pieter Muysken. 1986. "The Availability of Universal Grammar to Adult and Child Learners: A study of the acquisition of German word order." Second Language Research 2.93-119. . 1989. "The UG Paradox in L2 Acquisition." Second Language Research 5.1-29. Clark, Robin & Ian Roberts. 1993. "A Computational Model of Language Learnability and Language Change." Linguistic Inquiry 24.299-345. Clements, George N. & Engin Sezer. 1982. "Vowel and Consonant Disharmony in Turkish." The Structure of Phonological Representations (Part II), ed. by Harry van der Hulst & Norval Smith, 213-255. Dordrecht: Foris. den Besten, Hans. 1983. "On the Interaction of Root Transformations and Lexical Deletive Rules." On the Formal Syntax of the Westgermania: Papers from the '3rd Groningen Grammar Talks' Groningen, January 1981 (=Linguistik Aktuell, 3), ed. by Werner Abraham, 47-131. Amsterdam: John Benjamins. duPlessis, Jean, Doreen Solin, Lisa Travis & Lydia White. 1987. "UG or not UG, That is the Question: A reply to Clahsen and Muysken." Second Language Research 3.56-75. Emonds, Joseph. 1978. "The Verbal Complex V'-V in French." Linguistic Inquiry 9.151-175. Flynn, Suzanne & Wayne O'Neil. 1988. "Introduction." Linguistic Theory in Second Language Acquisition, ed. by Suzanne Flynn & Wayne O'Neil, 1-24. Dordrecht: Kluwer. Gass, Susan & Jacquelyn Schachter. 1989. "Introduction." Linguistic Perspectives on Second Language Acquisition, ed. by Susan Gass & Jacquelyn Schachter, 1-9. New York: Cambridge University Press. Gawlitzek-Maiwald, Ira, Rosemarie Tracy & Agnes Fritzenschaft. 1992. "Language Acquisition and Competing Representations: The child as arbiter." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 139-179. Dordrecht: Kluwer.

366

BONNIE D. SCHWARTZ & REX A. SPROUSE

Gregg, Kevin. 1989. "Second Language Acquisition Theory: The case for a generative perspective." Linguistic Perspectives on Second Language Acquisition, ed. by Susan Gass & Jacquelyn Schachter, 15-40. New York: Cambridge University Press. Gubala-Ryzak, Magda. 1992. "On Non-Unlearnable Adverb Positions: The case of French learners learning English." Ms., Boston University. Hankamer, Jorge. 1979. Deletion in Coordinate Structures. New York: Garland. Hong, Upyong. 1991. "External Constraints and the Role of Universal Grammar in Second Language Acquisition." Universal Grammar in Second Language Acquisition (= Theorie des Lexikons, 4), 67-80. Düsseldorf: Heinrich Heine Universität Düsseldorf. Hoekstra, Teun & Peter Jordens. This volume. "From Adjunct to Head." Hyams, Nina. 1986. Language Acquisition and the Theory of Parameters. Dordrecht: Reidel. Jordens, Peter. 1988. "The Acquisition of Word Order in L2 Dutch and German." Language Development, ed. by Peter Jordens & Josine Lalleman, 149-180. Dordrecht: Foris. Kayne, Richard. 1983. "Chains, Categories External to S, and French Complex Inversion." Natural Language and Linguistic Theory 1.109-137. Kennelly, Sarah. 1990. "Theta Government in Turkish." Ms., University of Siena. Kitagawa, Yoshihisa. 1986. "Subject in Japanese and English." Diss., University of Massachusetts, Amherst. Koopman, Hilda. 1984. The Syntax of Verbs: From verb movement rules in the Kru languages to Universal Grammar. Dordrecht: Foris. Koopman, Hilda & Dominique Sportiche. 1991. "The Position of Subjects." Lingua 85.211-258. Kornfilt, Jaklin. 1984. "Case Marking, Agreement, and Empty Categories in Turkish." Diss., Harvard University, Cambridge, Mass. . 1990. "Turkish and the Turkic Languages." The Worlds Major Languages, ed. by Bernard Comrie, 619-644. New York: Oxford University Press. Koster, Jan. 1975. "Dutch as an SOV Language." Linguistic Analysis 1.111-136. Krashen, Stephen. 1981. Second Language Acquisition and Second Language Learning. Oxford: Pergamon Press. Kural, Murat. 1992. "Properties of Scrambling in Turkish." Ms., University of California, Los Angeles. Kuroda, S.-Y. 1988. "Whether We Agree or Not: A comparative syntax of English and Japanese." Lingvisticae Investigationes 12.1-47. Lebeaux, David. 1990. "The Grammatical Nature of the Acquisition Process: Adjoinalpha and the formation of relative clauses." Language Processing and Language Acquisition, ed. by Lyn Frazier & Jill de Villiers, 13-82. Dordrecht: Kluwer.

WORD ORDER AND NOMINATIVE CASE

367

Lenneberg, Eric. 1967. The Biological Foundations of Language. New York: Wiley. Maling, Joan & Rex A.Sprouse. 1992. "The Case of Predicate NPs." Ms., Brandeis University, Waltham, Mass. and Harvard University, Cambridge, Mass. Meisel, Jürgen, Harald Clahsen & Manfred Pienemann. 1981. "On Determining Developmental Stages in Natural Second Language Acquisition." Studies in Second Language Acquisition 3.109-135. Pinker, Steven. 1984. Language Learnability and Language Development. Cambridge, Mass.: Harvard University Press. Platzack, Christen 1986. "COMP, INFL, and Germanic Word Order." Topics in Scandinavian Syntax, ed. by Lars Hellan & Kirsti Koch-Christensen, 185-234. Dordrecht: Reidel. Pollock, Jean-Yves. 1989. "Verb Movement, Universal Grammar, and the Structure of IP." Linguistic Inquiry 20.365-424. Rizzi, Luigi & Ian Roberts. 1989. "Complex Inversion in French." Probus 1.1-30. Roberts, Ian. 1993. Verbs andDiachronicSyntax: A comparative history of English and French. Dordrecht: Kluwer. Roeper, Thomas. 1992. "From the Initial State to V2: Acquisition principles in action." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 333-370. Dordrecht: Kluwer. Schwartz, Bonnie D. 1986. "The Epistemological Status of Second Language Acquisi­ tion." Second Language Research 2.120-159. . 1987. "The Modular Basis of Second Language Acquisition." Diss., University of Southern California, Los Angeles. . 1989. "L2 Knowledge: What is the null hypothesis?" Paper presented at the 14th Annual Boston University Conference on Language Development, Boston, 15 October. Ms., Boston University. . 1992. "An Alternative Account of Apparent Inaccessibility to UG in L2A." Ms., University of Durham. . 1993. "On Explicit and Negative Data Effecting and Affecting Competence and Linguistic Behavior." Studies in Second Language Acquisition 15.147-163. . Forthcoming. "Parameters in Non-native Language Acquisition." Forthcoming in Issues in Second Language Acquisition: An introduction, ed. by Peter Jordens & Josine Lalleman. Berlin: Mouton de Gruyter. Schwartz, Bonnie D. & Magda Gubala-Ryzak. 1992. "Learnability and Grammar Re-organization in L2A: Against negative evidence causing the unlearning of verb movement." Second Language Research 8.1-38. Schwartz, Bonnie D. & Alessandra Tomaselli. 1990. "Some Implications from an Analysis of German Word Order." Issues in Germanic Syntax, ed. by Werner Abraham, Wim Kosmeijer & Eric Reuland, 251-214. Berlin: Mouton de Gruyter.

368

BONNIE D. SCHWARTZ & REX A. SPROUSE

Sportiche, Dominique. 1988. "A Theory of Floating Quantifiers and its Corollaries for Constituent Structure." Linguistic Inquiry 19.425-449. Tavakolian, Susan. 1981. "The Conjoined-clause Analysis of Relative Clauses." Language Acquisition and Linguistic Theory, ed. by Susan Tavakolian, 167-187. Cambridge, Mass.: MIT Press. Thiersch, Craig. 1978. "Topics in German Syntax." Diss., Massachusetts Institute of Technology, Cambridge. Tomaselli, Alessandra. 1990. La Sintassi del Verbo Finito nelle Lingue Germaniche. Padua: Unipress. Travis, Lisa. 1984. "Parameters and Effects of Word Order Variation." Diss., Massachusetts Institute of Technology, Cambridge. Underhill, Robert. 1976. Turkish Grammar. Cambridge, Mass.: MIT Press. Vikner, Sten. 1991. "Verb Movement and the Licensing of NP-positions in the Germanic Languages." Ms., University of Stuttgart. (Forthcoming in Oxford Studies in Comparative Syntax, Oxford: Oxford University Press.) Weissenborn, Jürgen. 1991. "Functional Categories and Verb Movement: The acquisition of German syntax reconsidered." Spracherwerb und Grammatik: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie (= Linguistische Berichte, Special Issue 3), ed. by Monika Rothweiler, 190-224. Opladen: Westdeutscher Verlag. Wexler, Kenneth & Peter Culicover. 1980. Formal Principles of Language Acquisition. Cambridge, Mass.: MIT Press. White, Lydia. 1985. "The pro-drop Parameter in Adult Second Language Acquisition." Language Learning 35.47-62. . 1987. "Against Comprehensible Input: The input hypothesis and the develop­ ment of L2 competence." Applied Linguistics 8.95-110. . 1989a. "The Adjacency Condition on Case Assignment: Do L2 learners observe the Subset Principle?" Linguistic Perspectives on Second Language Acquisition, ed. by Susan Gass & Jacquelyn Schachter, 134-158. New York: Cambridge University Press. . 1989b. Universal Grammar and Second Language Acquisition. Amsterdam: John Benjamins. . 1991. "Adverb Placement in Second Language Acquisition: Some effects of positive and negative evidence in the classroom." Second Language Research 7.133-161. . 1992. "On Triggering Data in L2 Acquisition: A reply to Schwartz and Gubala-Ryzak." Second Language Research 8.120-137. Zagona, Karen. 1982. "Government and Proper Government of Verbal Projections." Diss., University of Washington.

Optionality and the Initial State in L2 Development Lynn Eubank University of North Texas

1.

Introduction

For nearly a decade now, the so-called ZISA data of Clahsen, Meisel & Pienemann (1983) have been the focus of intensive research on the role of Universal Grammar (UG) in second-language (L2) development.1 For intermedi­ ate stages, perhaps the most well-known data comprise adverb-initial utterances in which the subject and verb may or may not appear in the inverted order (i.e., VS) required in standard German. Examples of these data are shown in (la,b) below. (1)

a. b.

vielleicht ich kommen auch perhaps I come too hier haus hab ich eine here home have I one

(79) (79)

DuPlessis, Solin, Travis & White (1987), Tomaselli & Schwartz (1990) and others have proposed analyses of such utterances that are argued to be UG constrained. Importantly, these analyses also assume such utterances to belong to discrete stages of development, with uninverted utterances like (la) appearing before inverted utterances like (lb). However, while uninverted utterances do, on the whole, predominate before inverted utterances, the ZISA transcripts tabulated in Clahsen (1984) confirm that the two types overlap. In other words, there is a significant period during which this kind of inversion appears to be optional. In Jose's data, this period extends from week 69 through week 95, when inversion appears to become obligatory. The purpose here, then, is to attempt to explain why main clause inversion

370

LYNN EUBANK

appears to be optional during the intermediate period of development and how the L2 learner discovers that it is obligatory. Note at the outset that purely learningtheoretic considerations cannot be evoked here: on the standard assumption that only positive data are relevant to acquisition, exposure to data in which some linguistic phenomenon consistently appears cannot disprove optionality. Thus, the solution that I will suggest here, which is based on Wexler's (1991) analysis of somewhat similar phenomena in first-language (L1) development, derives from grammar-internal relations related to those discussed by Chomsky (1991). What makes this solution interesting is that, as we shall see, it calls into question certain common assumptions about the initial state of L2 development. The discussion will begin with a sketch of Wexler's solution for the apparent optionality of nonfinite verb forms in main clauses during early L1 development. What we will find is that Wexler's analysis offers a very promising means of explaining the optionality of main clause finiteness, yet the L1 data still mandate a certain amount of modification to this means. The next section is thus devoted to sorting out just how it is that main clause infinitives can be explained. The solution offered here involves making relations among lexical features explicit by means of labeled values and subvalues. Then come the ZISA data, for a relevant question is whether the explanation for the optionality of finiteness in L1 development also predicts optional inversion in the ZISA data. To this end, I develop a series of tests and examine Jose's data for evidence. However, if this analysis is supported by the data at hand, then certain parts of previous analyses of the ZISA data (see above) will require a certain amount of modification; this I turn to at the end of the section.

2.

"If the Verb Can Move, It Must Move"

Wexler (1991) sets out to explain the apparent optionality of verb finiteness in L1 development. Utterances illustrating this situation are shown in (2a,b), from Brown (1973). (2)

a. b.

Cromer come on Wednesday. Fraser comes on Saturday.

As Wexler notes, it is difficult to understand how UG might sanction the appearance of forms like (2a) in main clauses. He then points out that the situation is not unique to the acquisition of English; the German examples in (3a,b), for instance, also show the appearance of nonfinite verbs in main clauses (data from Clahsen 1982).

OPTIONALITY AND THE INITIAL STATE (3)

a. b.

371

ich schaufei haben I shovel have+INF ich hah auch kuchen I have+AGR also cake

Importantly, the numerous non-English data Wexler treats also bring out two other facets of his analysis. First, the non-English data show that the nonfinite verbs are in fact infinitives rather than simple verb stems; as a result, Wexler is able to maintain that the nonfinite forms in the English data are not verb stems, but infinitives, analogous to those in the non-English data. Second, the nonEnglish data often show a placement asymmetry; this is illustrated by the German data in (3a,b), where the agreeing verb shows up consistently on the left of the complement while the nonagreeing, infinitival verb is on the right. Wexler relates this asymmetry to the widely held assumptions that these languages involve raising of finite verbs, sometimes from head-final VPs. Why, then, do children permit infinitives to appear as main clause verbs? The framework Wexler employs to answer this question derives principally from the analyses of verb movement in Pollock (1989) and Chomsky (1991). Consider the tree in (4), which may be taken to represent D-structural relations in both English and French.

In this framework, the finite verb raises obligatorily to I in French main clauses, thereby correctly locating the adverb between the finite verb and its complement at S-structure. In English, by contrast, I lowers to the finite (thematic) verb, which remains in situ in the VP at S-structure; as a result, the adverb correctly remains in preverbal position. What is important for Wexler is how Chomsky motivates the obligatoriness of verb movement. In Chomsky's analysis, the crucial mechanics include a parameter involving abstract subject-verb agreement, a universal requirement for interpretation at Logical Form (LF), and a principle of Economy, by which

372

LYNN EUBANK

derivations with less movement are preferred over those with more movement. The agreement parameter derives from the difference between the agreement paradigms of English-like and French-like languages and determines whether or not thematic roles can be transmitted from the raised position. French agreement involves a relatively robust set of person/number affixes and is thus 'strong'. In this condition, verb movement is permitted because the strong agreement configuration allows the raised verb to transmit its thematic roles. English agreement is, by contrast, impoverished and thus 'weak'. In this condition, verb movement is prohibited due to the impossibility of thematic role transmission. Note, however, that strong agreement does not make verb movement obligatory. For this, the assumption is that because the operator indicating finiteness, <+tns>, must have scope over the VP at LF, the finite verb must be able to govern the VP at LF, hence either from COMP (verb-second, henceforth: V2 languages) from INFL (non-V2 languages). In this framework, then, French finite verbs raise to INFL in the syntax and remain there at LF because the alternative combina­ tion of lowering INFL to V at S-structure and then raising V to INFL at LF would be a more costly derivation. Obligatory raising of the French finite verb is thus motivated without stipulation: if the finite verb can move, it must move. In English, however, because of weak agreement, the more costly derivation involving INFL lowering to V and subsequent LF raising of V to INFL is allowed, there being no other option. Wexler does not employ the agreement parameter in his analysis of apparent optionality, but he does argue that the tense feature has not been acquired yet. What is particularly important is the absence of the positively specified tense feature, <+tns>: because it is this feature specification that necessitates LF raising for interpretation, its absence obviates the requirement. In other words, Wexler effectively eliminates the Economy-derived obligatoriness of verb raising in raising languages, hence allowing I-to-V lowering as an option.2 In addition, Wexler also suggests that the output of V-to-I raising is interpreted as a finite form while the output of I-to-V lowering is interpreted as nonfinite. The result is as desired: If the verb optionally raises, then inflection occurs. If the verb remains in situ, then one finds an infinitive. Moreover, once the features of tense are acquired (or, as Wexler tentatively suggests, once tense matures), then LF-raising will automatically come into play, hence eliminating the apparent optionality of main clause infinitives.3 This analysis seems attractive enough, but there are nonetheless difficulties that require attention. For example, as Wexler points, out the analysis implies a difficulty with, e.g., adult English: if V-to-I raising is interpreted as finite, then how

OPTIONALITY AND THE INITIAL STATE

373

can one explain the obviousfinitenessof English verbs, which, following Chomsky's assumptions, do not raise to INFL in overt syntax? Connected to this problem is another that Wexler does not discuss. Again, the working assumption is that I-to-V lowering is interpreted as nonfinite while V-to-I raising is finite. This analysis thus suggests that finite sentences like (2b) involve verb raising, the only means by which finiteness can be interpreted. The problem here is that the English acquisitional data do not suggest a period during which finite thematic verbs precede adverbs or negators, as they should if raising were involved. Indeed, what the data suggest to me is that verbs with agreement inflection alternate freely with infinitives in the English data while appearing only in raising contexts in the French and Germanic data that Wexler reviews. Put differently, on the likely assumption that I-to-V lowering is the only option open to the English learners, what one finds is that lowering results in either infinitives or agreementinflected verbs in the English data while permitting only infinitives in the French and Germanic data, which all involve raising of finite verbs.

3.

Features and Values

As far as I can see, Wexler's assumption that the features of tense are not acquired is on the right track; the root of the problem seems to be the idea that verb finiteness is interpreted directly from V-to-I raising or I-to-V lowering alone. However, perhaps a somewhat more differentiated analysis of the situation can preserve something of the outlines of Wexler's analysis. For the sake of clarity, though, let us state beforehand exactly what needs explanation. The assumptions underlying the statements in (5) below are that the strength of agreement plays an active role in verb raising and that the learners have already acquired the appropriate agreement configurations, strong or weak; hence, the weak agreement learners know that V-to-I raising is ruled out. (5)

For learners of strong agreement, raising languages, a. V-to-I raising allows only agreement; b. I-to-V lowering allows only infinitives. For learners of weak agreement, lowering languages, c. V-to-I raising is not an option; d. I-to-V lowering allows either agreement or infinitives.

To explain this situation, it may be of value to reconsider the notion of 'finiteness'. Assume for languages in which verbal inflection alternates with an

374

LYNN EUBANK

infinitival form that inflection involves one or more prime features (e.g., agreement, tense, etc.) with associated values that themselves may range over several possible subvalues. Hence, tense has the values <-tns> and <+tns>, with <+tns> having the possible subvalues <present> or <past>. Similarly, agreement has the values <-agr> and <+agr>, the latter ranging over weak or strong. In this view, saying that some verbal element is finite reduces to an epiphenomenon that results from the presence of particular value specifications; in the same way, the infinitive simply involves a set of different value specifications. Hence, finite verbs in, for example, German involve the features <+agr>, <+tns> while the infinitive shows <-agr>, <-tns>. That features tend to be equal (all negative or all positive) seems to be common trait, yet it is also possible for features to be different. As is well known, for instance, Portuguese has so-called (subjectagreement) inflected infinitives (Rouveret 1980). Here one might assume tense and agreement with the features <-tns> and <+agr>. In addition, particular features have auxiliary consequences for syntax. In this regard, the thematic nature of verbs plays an important role, as Pollock (1989) has shown: those that do not assign theta(0)-roles (e.g., French être) raise to INFL broadly,4 while thematic verbs only raise under certain circumstances. Here, then, it suffices to focus only on thematic (i.e., main) verbs. Negatively specified features (e.g., <-agr> or <-tns>) prohibit main verb raising to the head with that feature, likely due to a raised infinitive's inability to transmit thematic roles.5 Following Chomsky (1991), we also assume not only that the positive tense feature requires LF interpretation, but also that strong agreement permits verb raising while weak agreement prohibits it, both cases reducible to Ө-role transmission. (For agreement, note that we are dealing with three distinct values: weak <+agr>, strong <+agr>, and <-agr>.) To simplify matters below, it will be of value simply to list these prelimi­ nary conditions individually; this is shown in (6). (6)

a. b. c. d. e. f.

Positive features result in inflection. Negative features result in infinitival forms. Negative features prevent main verb raising. <+tns> requires LF raising. weak agreement prohibits main verb raising. strong agreement permits main verb raising.

There would appear to be a curious redundancy between two of the conditions in (6): both (6c) and (6e) prevent the raising of main verbs, and both have, in a vaguely defined sense, negative specifications. Here one might imagine, for

OPTIONALITY AND THE INITIAL STATE

375

instance, that 'weak' and 'negative' are equivalent terms. For the moment, though, I leave the conditions in (6) as they stand, pending further modification below. Importantly, all of the conditions in (6) are assumed to be given antecedently by UG. What of acquisition? As Iatridou (1990) points out, it seems somewhat unreasonable to expect every humanly accessible prime (tense, agreement, etc.) to appear in all languages. Thus, for German, French or English, one might well expect the child to wind up with an agreement feature, but not for Swedish, Gbadi or Afrikaans, which do not instantiate agreement. What one might consider, then, is that particular features (and values) given by UG are selected only as required by the primary data.6 Now to the acquisitional data at hand, those resulting in (5). For the English data Wexler reviews, agreement has been selected, with the associated value, weak; tense has not been selected. Accordingly, the infinitive results from <-agr> and agreement inflection results from <+agr>; because of the weak value of <+agr>, there is no placement asymmetry in the English data. Later on, tense morphology is acquired and tense selected, but the appearance of the feature <+tns> produces no overt syntactic effect on the English data because weak agreement still prevents verb raising. For the German and French data, however, a problem occurs. Here the assumption is that only agreement has been selected (i.e., no tense) and that the strong value has been determined. Hence, <-agr> will allow only lowering and the infinitive while <+agr> optionally allows verb raising and agreement. Here lies the problem: If <+agr> allows raising only as an option, then one might also expect <+agr> to allow lowering, given that <+tns> is not present. In other words, the condition in (6f) incorrectly predicts that learners of German will produce agreement inflection on both the left and the right of the complement rather than only on the left as shown in (3b) above. How do we escape the quandary? On the one hand, we want to follow Wexler in avoiding <+tns> so that infinitives can appear in main clauses. On the other, if <+tns> is removed, then there is no way to prevent agreement inflection from appearing on the right for German. One way to escape the problem is, of course, to return to Wexler's proposal, where verb finiteness is read directly from raising and lowering; as we have seen, however, this proposal incurs problems of its own. Another solution is to follow the general idea in Lasnik (1992), who suggests a structural condition governing, among other things, the transmission of thematic roles. His suggestion is that a head adjoined to a second head must never have a feature value that exceeds that of the second head. What is important is that Lasnik includes in this idea not only values of inflection (e.g.,

376

LYNN EUBANK

strong versus weak), but also values for the verb itself. We might assume that uninflected verbs (i.e., verb stems) take a value that is somewhere between weak and strong, perhaps 'moderate'.7 The result of this idea is the configurations in (7a-d), where V denotes a verb and I an inflection.

Both (7a) and (7b) represent I-to-V lowering, the former adjoining strong inflection to the verb and the latter, an adjunction of weak inflection. By Lasnik's condition, (7a) is not permitted because the value of inflection, <strong>, exceeds the value of the verb, <moderate>. The configuration in (7b) is permitted by the same condition. Both (7c) and (7d) are created by adjunction of the verb to inflection in V-to-I raising. However, (7c) is ruled out because the moderate verb exceeds weak inflection; by the same condition, (7d) is permitted since the moderate verb does not exceed strong inflection. Above we noted a curious redundancy between (6c) and (6e), where <-agr>, weak agreement, and <-tns> seem to prevent verb raising to a head with that feature or value. At this point, it would seem natural to think that these facts can be handled by means of the adaptation of Lasnik's idea in (7) above. Assume that the strength of negatively specified features is weak, making these features not unlike weak agreement. If so, then by the condition in (7c), raising of the verb to weak or negatively specified inflection is disallowed. From the general idea that Lasnik proposes, we thus shed the more stipulative formulation in (6c). Note, however, that the reverse of (6c) must also be examined: can negatively specified features themselves raise? To locate a contrast, consider positive specification first. Suppose that weak <+agr> lowers to the moderate verb, creating the adjoined complex [V V [<+agr>]]. If tense is <+tns>, requiring LF interpretation, then the adjoined complex [V V [<+agr>]] must raise to tense, creating the complex [T [V V [<+agr>]] <+tns>]. It is this complex that is interpreted at LF. Now consider negative specification. If <-agr>, now weak,

OPTIONALITY AND THE INITIAL STATE

377

lowers to the moderate verb, the nonfinite complex [V V [<-agr>]] is created. If tense is <+tns>, then this complex must raise, creating the new, mixed complex [T [V V [<-agr>]] <+tns>]. Is this complex interpretable at LF? Apparently not, for this particular combination is, to my knowledge, not attested among languages that instantiate bound agreement and tense morphology. Explaining why this may be so is, however, more elusive. A likely candidate is, however, the required inclusion of the negatively specified feature <-agr> into the LF­ interpreted <+tns> complex. Assume that this is so; we might now say that <+tns> must be interpreted at LF and that the complex containing <+tns> may not include feature conflict. Assume, then, that these considerations can be maintained: Lasnik's idea supplies the structural reason underlying the possibility of transmitting Ө-roles under raising and lowering; the strength of inflection extends to include negative specifications (<-tns>, <-agr>); and the <+tns> complex interpreted at LF cannot include feature conflict. For his insight, let us name Lasnik's idea after Lasnik himself, LASNIK'S CONDITION, and substitute it for (6c), (6e) and (6f), as shown in (8). (8)

a. b. c. d.

Positive features result in inflection. Negative features result in infinitival forms. <+tns> requires LF raising without feature conflict. Lasnik's Condition must be observed.

Given (8), one may now ask if the quandary observed above for strong agreement learners still persists. Once again, the assumption is that tense has not been selected, but agreement has, and that agreement is strong. By (8b,d), <-agr> will always result in I-to-V lowering and the infinitive. The problem occurred when we considered <+agr> for strong agreement. Under (6), <+agr> permits verb raising and agreement, but does not disallow lowering, given that tense is not present. Under Lasnik's Condition , however, the desired result falls out: if strong, <+agr> cannot lower to V because it would exceed the value of the moderate verb, but it does permit V-to-I raising. In other words, (8) correctly predicts that learners of German will produce agreement inflection only on the left as shown in (3b) above. What happens when tense appears? Again, before this point, the child has two options: generate with <-agr> or generate with <+agr>; either raising or lowering may be involved. However, <+tns> requires LF interpretation, hence verb raising. Thus, when tense is selected and <+tns> generated, the equality of derivations between <-agr> and <+agr> disappears. The child learning English

378

LYNN EUBANK

discovers that the complex created by lowering <-agr> to the verb cannot be employed to lexicalize <+tns> for LF interpretation due to the feature conflict it creates. The only remaining option is to generate <+agr>, which can be lowered to the verb and subsequently raised to <+tns> without difficulty. Now to the child learning German. If <-agr> is generated, then we again find the feature conflict problem when tense is selected. Once again, the only available option for the child is to generate <+agr>. In other words, just as Wexler envisions, the acquisition of tense brings to an end the apparent optionality of infinitival forms in main clauses. The general picture is, then, that the optionality Wexler observes may derive ultimately from the type of agreement (strong versus weak) and the selection of tense, both interacting with antecedent requirements in (8). Note, though, the assumption throughout that the French and Germanic learners have acquired strong agreement, thus promoting verb raising, and that the learners of English have acquired weak agreement, thus prohibiting verb raising. Before turning to the L2 data, it will be necessary to clarify matters here further. Harald Clahsen was among the first to note that verb raising and the acquisition of subject-verb agreement relate to one another in L1 development. As early as 1985, for instance, Clahsen argued that verb raising in L1 German comes about because the agreement paradigm is completed with the addition of the unique person affix -st. In other words, acquisition of what we now call the strong agreement paradigm in some sense requires verb raising. This view was challenged, however, by the survey in Verrips & Weissenborn (1992), who find that the acquisition of the agreement paradigm is not a prerequisite to verb raising. The two views seem to contrast sharply, yet in another way they turn out to be very similar: in both cases, discussion centers around the complete mastery of the subject-verb agreement paradigm. For German, then, if agreement plays any role at all in verb movement, then one would expect raising to come about only when all four agreement affixes are fully acquired.8 Importantly, however, Platzack & Holmberg's (1989) analysis of the adult state suggests that this assumption may be too strong; their view is that person agreement alone may be the crucial key underlying what we have called strong. Subsequent work appears to confirm this view. For example, Rohrbacher (1992), examining diachronic as well as synchronic facts of raising and nonraising languages, finds the crucial distinction to lie in the presence of person agreement, specifically, a morphologi­ cal distinctness of at least two persons.9 In addition, Meisel (1992) finds for acquisition that the onset of verb raising coincides with the acquisition of only two of the three person affixes of German, all of this occurring before tense

OPTIONALITY AND THE INITIAL STATE

379

morphology is acquired. In other words, this research suggests an important role for agreement in verb movement, yet not one requiring the full subject-verb agreement paradigm. Part of the problem here is that the earlier and more recent studies above have dealt with two different phenomena. It is one thing to claim that verb raising can occur, yet quite another to say that it is obligatory, as we have seen. Indeed, if the present modification of Wexler's analysis is correct, then even the acquisition of the full agreement paradigm for languages like German should not result in obligatory raising if tense is not yet acquired. For present purposes, I will continue to maintain the relation between agreement and raising, though, following the lead of Rohrbacher and Meisel, I will also assume that strong agreement reduces to agreement on as few as two persons.

4.

'Optional Inversion' in L2

Do the L2 inversion phenomena in (la,b) admit in principle to the kind of analysis proposed above? To determine this, it is worthwhile to consider the grammar underlying (la-b). Previous literature on this stage of development for speakers of Romance (duPlessis et al., 1987; Tomaselli & Schwartz 1990; Eubank 1992) suggests for intermediate stages a VP not completely unlike the tree in (4). The idea they propose is that uninverted utterances like (la) have the verb in situ in VP. As noted in the introduction, inverted utterances are assumed to appear in a subsequent stage of development, and for this stage the idea is that CP is added: the initial adverb is in Spec-CP, the verb is in COMP, and the subject in Spec-IP.10 What unites these analyses is thus that verb movement does not occur in SVO utterances like (la), but does in inverted utterances like (lb). What would it take for the explanation of optional infinitives in L1 development to apply to the apparent optionality of inversion? Note once again that inverted and uninverted utterances do appear simultaneously during the lengthy, 'optional inversion' period after the SVO-only stage and before the inversion-only stage. Now, for the analysis presented above to work, we might assume, pending confirmation below, that tense has not been acquired, but that strong agreement has. Hence, if strong <+agr> appears, then we expect inflection, and if <-agr> appears, then the infinitive should turn up. Now, because of the nature of inversion, the only plausible analysis for utterances like (lb) is verb raising. Uninverted utterances like (la) are, however, amenable to either a raising analysis or a lowering analysis. However, if raising occurs, then we again expect

380

LYNN EUBANK

to find agreement inflection; if lowering occurs, then only the infinitive is expected. Generally, then, we will at some point require an analysis that differs from previous work in the assumption that both raising and lowering are at work. Specific tests of predictions are now in order. In particular, we will want to have answers to each of the questions in (9). (9)

a. b. c. d.

Does the acquisition of strong agreement coincide with the beginning of the 'optional inversion' period? Does tense not appear during the 'optional inversion' period? Does the acquisition of tense coincide with obligatory raising? Is there a predictable asymmetry between agreement and infinitives during optional inversion?

If the explanation for L1 optionality above is to obtain here, then the analyses below must yield an affirmative answer for each of the questions in (9).

5.

Strong agreement and the optional inversion period?

Does the acquisition of strong agreement coincide with the beginning of the optional inversion period? Eubank (1988) and Tomaselli & Schwartz (1990) have pointed out that verb raising predicts the appearance in the ZISA data of V-ADV-NP orders. Their prediction is confirmed by Clahsen's (1984) tabulation of the ZISA data. For Jose in particular, Clahsen's tabulation indicates that raising has begun as early as week 69. Our question above is now this: does the onset of verb raising in week 69 occur just when strong agreement is acquired? Recall from above that strong agreement in German is to be identified with the appearance of two person affixes. Table 1 provides the necessary agreement information for Jose. An estimate based on the figures in the table is that strong agreement is present by week 69. In other words, the data confirm our first test: the relevant type of agreement is present just when verb raising appears in the data.11

6.

The acquisition of tense

Does tense not appear during the optional inversion period? Does the acquisition of tense coincide with obligatory raising?

OPTIONALITY AND THE INITIAL STATE

381

Table 1. Main Clause Agreement by Affix for Jose

weeks of exposure

-e/-Ø 54 63 69 73 79 83 85 95 102

2/2 (100%) 60% 6/10 10/15 67% 80% 12/15 40/41 98% 32/34 94% 96% 27/28 27/30 90% 92% 23/25

-St

-t

0/0 (0%) 1/1 i(100%) 1/3 33% 3/3 100% 6/6 100% 8/8 100% 9/9 100% 15/15 100% 3/3 100%

0/0 (0%) 0/0 (0%) 2/3 67% 7/7 100% 12/12100% 12/12 100% 11/13 85% 4/6 67% 7/7 100%

-n 0/4 0/2 1/14 1/4 3/17 2/10 3/6 2/4 10/12

0% (0%) 7% 25% 18% 20% 50% 50% 83%

The two columns under each of the four agreement affixes represent a ratio and a percentage; the rows are determined by the weeks of exposure to German, when recording sessions took place. The ratio represents the number of correct uses of the affix over the total uses of the affix for that recording session. Percentages are calculated from the ratios. Following Clahsen (1982), percentages are shown in parentheses when the affix is present on fewer than three occasions during the recording session. The figures in the cells include correct agreement with unambiguous null subjects.

As these two questions can be answered with the same data, they will be handled together. The only thorough analysis of tense marking in the ZISA data is Meisel (1987), and, as it turns out, Jose is one of the very subjects that Meisel examines. Meisel finds that during the initial weeks Jose only employs adverbial elements to indicate reference to time; there is no indication of grammatical tense. Soon thereafter Jose begins to employ the German perfect (auxiliary + participle forms) to refer to past events. Finally, his analysis shows that Jose begins to employ the preterite in week 95. Meisel's examination thus provides affirmative answers to both questions: during the optional inversion period, Jose apparently does not control tense, and the acquisition of tense, signaled by the appearance of the preterite, coincides with obligatory raising.

7.

Agreement

Is there a predictable asymmetry between agreement and infinitives during optional inversion? Native-language learners of German establish a head-final VP from the beginning. This is important because the L1 data thus show an obvious placement-finiteness asymmetry: finite forms are on the left while nonfinite forms

382

LYNN EUBANK

are on the right. How, then, is one to locate a relation between verb placement and finiteness in the L2 data? After all, if SV(X) orders common to this period are amenable to either a raising or a lowering analysis, then locating some mixture of finite and nonfinite forms will not be revealing. Note, however, that we can take advantage of orders that unambiguously require a raising analysis, specifically, VS orders, which, fortunately, happen to be abundant in Jose's transcripts. The idea is to compare utterances with unambiguously raised verbs to utterances that may involve either raising or lowering. If the modification of Wexler's proposal is correct, then utterances with unambiguous raising should show high figures for agreement and low figures for infinitives; by comparison, utterances amenable to either raising or lowering should show lower figures for agreement and higher figures for infinitives. The relevant data are in Table 2, which presents figures for agreement inflection and infinitives in unambiguous raising contexts and in contexts where either raising or lowering may be involved. Only thematic verbs are taken into account here since the free raising of nonthematic verbs may distort the picture. The figures for VS orders suggest part of the parallel we are looking for: if verb raising is involved, then agreement inflection prevails and infinitives are rare. (Indeed, two of the three infinitives that do appear in the raising context both occur in the same recording session, in week 79.) Note, too, that while the agreement figure for SV contexts is relatively high, these contexts are ambiguous in that they may involve either raising or lowering. More revealing is to take the figures for both agreement and infinitives into consideration: only when lowering is predicted to be possible, namely, in SV contexts, do we find both a lower figure for agreement and a higher figure for infinitives. We thus seem to have an affirmative answer to our question: there is a predictable asymmetry between agreement and infinitives, the former occurring overwhelmingly in unambiguous raising contexts and the latter more common just when a lowering analysis is plausible. At this point, it would appear that the modification of Wexler's (1991) ideas may indeed apply to Jose's development as well. However, as Lydia White (pers. comm.) has pointed out, one problem with this analysis is the headedness of VP. Once again, the general assumption in the L2 literature is that the verb precedes its complement within VP during the early part of development. However, as Clahsen's (1984) tabulation of the ZISA data shows, the following, intermediate period of development is characterized not only by the presence of VS orders, but also by the appearance in clause-final position of nonfinite elements like particles and participles. To deal with this, duPlessis et al. (1987), Tomaselli &

OPTIONALITY AND THE INITIAL STATE

383

Table 2. Thematic Verb Agreement and Infinitives in SV and VS orders for Jose, Weeks 69-95 agreement VS orders (verb raising) SV orders (ambiguous)

48/53 66/98

90.5% 67.3%

infinitive 3/53 22/98

5.6% 23.4%

The two columns each under agreement and infinitive represent a ratio and a percentage. The ratio represents the number of agreeing or infinitival verbs over the total number of verbs. Percentages are calculated from the ratios. Note that Jose also makes a number of agreement errors. Because of this, incorrect agreement does not always mandate the presence of an infinitive, so adding together the numerators of the ratios does not necessarily equal the denominators.

Schwartz (1990), and Eubank (1992) all assume that the headedness of VP has been switched: The verb now follows its complement. And so here lies the problem: if the verb sometimes does not raise, as maintained here, then it should appear as a main-clause infinitive in clause-final position. The data from Jose do not support this prediction. How, then, are we to understand this predicament? On the one hand, the agreement facts and placement asymmetries presented above seem to bear out the explanation proposed above, but this explanation now also falsely predicts the appearance of clause-final infinitival main verbs. On the other hand, if we abandon this explanation and assume the proposals made heretofore (see above), then criticisms of these proposals would appear to have clear force. Meisel (1991) argues, for instance, that the very appearance of infinitives as main verbs excludes the possibility that Universal Grammar constrains nonnative syntax. Perhaps a more discriminating analysis of the VP is called for. The data relevant to the VP-headedness question comprise utterances in which not just one, but two verbal elements appear: One is the main verb, the other, some nonfinite element (particle, participle, or infinitive). As noted above, the nonfinite element appears in clause-final position. The question is, if the clause has only one verb, should we necessarily expect it to appear in clause-final position? Perhaps not. To deal with phrasal structure that involves two (or more) verbal elements, it has long been maintained that two (or more) different projections are involved (see, e.g., von Stechow & Sternefeld 1988). What is important about this idea is that it may well be plausible to maintain that these two projections, however they are labeled, have different headedness characteristics. In particular, we might maintain something like the tree in (10) below, where the X-head is a nonfinite verbal element.

384

LYNN EUBANK

The different kinds of nonfinite elements that appear in clause-final position will likely lead to certain modifications of (10); see, for instance, the discussion in Keyser & Roeper (1992) on what particle constructions may involve. Such considerations aside, if the nonnative grammar instantiates something along the lines of (10) during the optional inversion period, then we should find nonfinite elements appearing in final position, but only when they are accompanied by some second verbal element. If, however, only one verb is projected, then it is possible, without contradiction, to maintain the optional-raising analysis, which presupposes that the (higher) verb precedes its complement underlyingly. In other words, main verbs may appear as unraised infinitives during the period in question without predicting that they must appear in clause-final position.

8.

Conclusion

And so we have an explanation for the apparent optionality of inversion in the ZISA data: agreement is strong, but tense is absent. Hence, if learners generate strong <+agr>, then we find inflection; if <-agr> is generated, then the infinitive appears. The optionality of inversion thus comes about because strong <+agr> requires verb raising while <-agr> allows only lowering. When features of tense do appear, LF-interpretation of <+tns> comes into play: the only feature that does not conflict with <+tns> in the <+tns> complex at LF is <+agr>. At the most general level, this analysis indicates that lexical-driven information like the status of abstract agreement and tense are acquired, not transferred from the native language into the L2 initial state. This turns out to be of importance, for it contrasts sharply with other work on what comprises the L2 initial state. For example, work by White (see, e.g., 1989, 1991) and Schwartz (e.g., 1992) suggests that this state may comprise a complete representation of the native language, including values such as strong agreement. At the other end of the scale, we also find Vainikka & Young-Scholten (this volume), who argue that the native language contributes only lexical projections to the L2 initial state. L2 acquisition thus proceeds in a way not unlike that envisioned by Clahsen,

OPTIONALITY AND THE INITIAL STATE

385

Eissenbeiss & Vainikka (this volume) for L1 development. In contrast to these more extreme views, the present analysis suggests a middle ground: lexical and functional projections transfer from the native language, and so do the headedness characteristics of those projections, but lexically driven values under functional heads are not transferred. Note, interestingly, that an initial state so configured may well turn out to be unique to L2 development. Assume that L1 acquisition is indeed characterized by the early absence of functional projections (e.g., CP, IP, DP). If we further assume, following the general outlines of the Lexical Parameterization Hypothesis, that it is lexical information that ultimately drives the acquisition of these projections, then one can predict for L1 development the absence of functional projections that stand without some kind of lexical value under their respective heads. According to the L2 analysis proposed here, however, this prediction would not obtain in the case of L2 development.

Acknowledgements I would like to thank Jürgen Meisel, who kindly provided me with Jose's transcripts. I would also like to thank the many people who have discussed these matters with me over the last year of so, especially Jürgen Meisel, Bonnie Schwartz, and Lydia White.

Notes 1.

ZISA stands for 'Zweitsprachenerwerb italienischer und spanischer Arbeiter' (second language acquisition by Italian and Spanish workers). The L2 data cited below are from one of the ZISA subjects, Jose, a native speaker of Spanish. Numbers in parentheses following the utterances indicate weeks of exposure to the L2, German. See Clahsen (1984) for biographical details.

2.

Although not made explicit, Wexler appears to follow Chomsky's (1991) suggestion that the trace of the lowered INFL is deleted prior to LF in order to avoid an ECP violation.

3.

Wexler surveys a satisfyingly broad range of data to support this analysis. One inter­ esting finding that Wexler does not include, however, is the fact that regular past tense and third person agreement are nearly coincident in Brown's (1973) morpheme order.

4.

English auxiliaries do not raise to INFL in infinitival clauses, but, as Chomsky (1957, 1991) has maintained, this may be because infinitival tense in English does not involve an affix. See Iatridou (1989) for discussion.

386

LYNN EUBANK

5.

This analysis suggests that the cases of so-called short movement of nonfinite verbs that Pollock (1989) notes do not involve verb raising. See Iatridou (1990) for discussion on alternatives to Pollock's raising analysis.

6.

Taking this view would appear to commit one to the idea that functional categories are not initially present in L1 development (see, e.g., Radford 1990; Meisel & Müller 1992).

7.

Lasnik considers inflection to have three possible values and verbs to have two possible values. Inflection includes 'weak', 'moderate', and 'strong' values. His motivation for the three-way distinction, rather than the two-way distinction that Pollock (1989) and Chomsky (1991) postulate, is to account for the apparent capacity of nonfinite auxiliaries to raise in French, but not in English. However, there may be independent reason for the failure of nonfinite auxiliary raising in English (see note 4). As for verbs, Lasnik includes the values 'strong' for main verbs and 'moderate' for auxiliaries. His motivation here is the different behavior of main verbs and auxiliaries. Note, however, that if the crucial effect of the configurations in (5a-d) is the capacity to transmit 6roles, then verbs that do not have Ө-roles to transmit (e.g., auxiliaries) will be exempt from Lasnik's conditions. For present purposes, then, we will maintain a single value for verbs, in our case, 'moderate'.

8.

As noted, German agreement involves four person/number affixes: -st, -el-0, -n and -t. The distribution of these affixes is shown below: -e/-Ø first singular -st second singular -n first and third plural -t third singular and second plural

9.

Rohrbacher's (1992) particular argument is that the first and second person affixes must differ from one another, from any forms for other persons, and from the infinitival form.

10.

Not all previous analyses agree on just how this works, however. For instance, duPlessis et al. (1987) assume that the adult state of German involves CP only if inversion occurs; for uninverted utterances that have no preposed nonsubject, they assume only IP. This analysis is then carried over to the L2 data. Tomaselli & Schwartz (1990), by contrast, follow Vikner & Schwartz (forthcoming) in assuming that all main-clause finite verbs are located in COMP for the final state. For Tomaselli & Schwartz, then, initial-stage utterances, including those like (la), involve IP while all later stage utterances include CP. For later reference, note that all off these analysts assume sentences such as (lb) to involve a head-final VP.

11.

Clahsen (1988) argues that the development of agreement does not coincide with the appearance of V2 in the ZISA data, hence that UG is not 'available' to adult L2 learners. However, Clahsen's claim relies on Köpcke's (1987) analysis of agreement, which deals with the development of the full person/number paradigm, not strong Agreement as it is more narrowly defined here. In this sense, the analytical means Clahsen employs for his L2 claim are identical to the means he employs in his early work on L1 development, also reported in Clahsen (1988).

OPTIONALITY AND THE INITIAL STATE

387

References Brown, Roger. 1973. A First Language: The early stages. Cambridge, Mass.: Harvard University Press. Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. . 1991. "Some Notes on Economy of Derivation and Representation." Principles and Parameters in Comparative Grammar, ed. by Robert Freidin, 417-454. Cambridge, Mass.: MIT Press. Clahsen, Harald. 1982. Spracherwerb in der Kindheit. Tübingen: Günther Narr. . 1984. "The Acquisition of German Word Order." Second Languages: A crosslinguistic perspective, ed. by R. Andersen, 219-242. Rowley, Mass.: Newbury House. . 1985. "Parameterized Grammatical Theory and Language Acquisition." Paper presented at the Workshop on Linguistic Theory and Second Language Acquisi­ tion, Cambridge, Mass., October. . 1988. "Parameterized Grammatical Theory and Language Acquisition: A study of the acquisition of verb placement and inflection by children and adults." Linguistic Theory in Second Language Acquisition, ed. by Susan Flynn & Wayne O'Neil, 47-75. Dordrecht: Kluwer. Clahsen, Harald, Sonja Eisenbeiss & Anne Vainikka. This volume. "The Seeds of Structure: A syntactic analysis of the acquisition of Case marking." Clahsen, Harald, Jürgen Meisel & Manfred Pienemann 1983. Deutsch als Zweitsprache: Der Spracherwarb ausländischer Arbeiter. Tübingen: Günther Narr. duPlessis, Jean, Doreen Solin, Lisa Travis & Lydia White. 1987. "UG or not UG, That is the Question: A reply to Clahsen and Muysken." Second Language Research 3.56-75. Eubank, Lynn. 1988. "Universal Grammar, German grammar, and L2 Development." Ms., University of Texas, Austin. . 1992. "Verb movement, Agreement, and Tense in L2 Acquisition." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 225-244. Dordrecht: Kluwer. Iatridou, Sabine. 1990. "About Agr(P)." Linguistic Inquiry 21.551-577. Keyser, Samuel Jay & Thomas Roeper. 1992. "Re: The abstract clitic hypothesis." Linguistic Inquiry 23.89-125. Köpcke, K.-M. 1987. "Der Erwerb morphologischer Ausdrucksmittel durch L2-Lerner am Beispiel der Personalflexion." Zeitschrift für Sprachwissenschaft 6.186-205. Lasnik, Howard. 1992. "Case and Expletives: Notes toward a parametric account." Linguistic Inquiry 23.381-405. Meisel, Jürgen. 1987. "Reference to Past Events and Actions in the Development of Natural Second Language Acquisition." First and Second Language Acquisition Processes, ed. by C. Pfaff, 206-224. Cambridge, Mass.: Newbury House.

388

LYNN EUBANK

Meisel, Jürgen. 1991. "Principles of Universal Grammar and Strategies of Language Use: On some similarities and differences between first and second language acquisition." Point Counterpoint, ed. by Lynn Eubank, 231-276. Amsterdam: John Benjamins. . 1992. "Getting FAT: Finiteness, agreement and tense in early grammars." Ms., University of Hamburg. Meisel, Jürgen & Natascha Müller. 1992. "Finiteness and Verb Placement in Early Child Grammars. The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 109-138. Dordrecht: Kluwer. Platzack, Christer & Anders Holmberg. 1989. "The Role of AGR and Finiteness in Germanic VO languages." Working Papers in Scandinavian Syntax 43.51-76. Lund: Linguistics Department, University of Lund. Pollock, Jean-Yves. 1989. "Verb movement, Universal Grammar, and the Structure of IP." Linguistic Inquiry 20.365-424. Rohrbacher, B. 1992. "English AUX^NEG, Mainland Scandinavian AUX^NEG and the Theory of V to I Raising." Ms., University of Massachusetts, Amherst. Rouveret, Alain. 1980. "Sur la notion de proposition finie." Recherches Linguistiques 9. Schwartz, Bonnie D. 1992. "An Alternative Account of Apparent Inaccessibility to UG in L2A." Paper presented at the Annual Boston University Conference on Language Development, Boston, October, von Stechow, Armin & Wolfgang von Sternefeld. 1988. Bausteine syntaktischen Wissens. Opladen: Westdeutscher Verlag. Tomaselli, Alessandra & Bonnie D.Schwartz. 1990. "Analysing the Acquisition Stages of Negation in L2 German: Support for UG in SLA." Second Language Research 6.1-38. Vainikka, Anne & Martha Young-Scholten. This volume. "Direct Access to X'-Theory: Evidence from Korean and Turkish Adults Learning German." Verrips, Maaike & Jürgen Weissenborn. 1992. "Routes to Verb Placement in Early German and French: The independence of finiteness and agreement." The Acquisition of Verb Placement: Functional categories and V2 phenomena in language development, ed. by Jürgen Meisel, 283-331. Dordrecht: Kluwer. Vikner, Sten & Bonnie D. Schwartz. Forthcoming. "The verb always leaves CP in V2 clauses." Parameters and Functional Heads, ed. by Adriana Belletti & Luigi Rizzi. Oxford: Oxford University Press. Wexler, Kenneth. 1991. "Optional Infinitives, Head Movement and the Economy of Derivation in Child Grammar." Paper presented at the Conference on Verb Movement, University of Maryland, October. Ms., Massachusetts Institute of Technology, Cambridge, 1991. White, Lydia. 1989. Universal Grammar and Second Language Acquisition. Amsterdam: John Benjamins. . 1991. "The Verb Movement Parameter in Second Language Acquisition." Language Acquisition 1.337-360.

Index of Languages

Arabic 171, 175 ASL 48 Bantu 199 Bengali 323 Berber 199,208 Bernese 3, 9, 17, 32, 53, 148, 177, 178, 182-184, 189, 197, 199, 204,206,207,210,211 Chinese 33, 34, 36, 48, 55, 80, 82, 176 Corsican 171 Danish 81 Dutch 2, 4, 7, 8, 17, 22, 24, 25, 29-35, 37-40, 43, 44, 46, 48, 49, 51-53, 123-129, 135, 140, 145-147, 170, 173, 197, 205, 209, 213, 315, 316, 366 (see also West-Flemish) English 4, 5, 8-10, 16, 17, 19, 22, 23, 25, 26, 28, 30, 32, 34, 35, 36-38, 43-45, 47-49, 53-55, 57-67, 71, 73, 75-77, 82, 83, 93, 114, 116, 121-124, 126, 129, 141, 146, 148, 151-155, 158, 162, 163-165, 167, 168, 170, 174-176, 192, 200, 206, 207,

216, 219, 221, 225, 229, 232, 253, 257, 264, 296, 309, 313, 315, 316, 343, 362, 364, 366, 367, 370-373, 375, 377, 378, 385, 386, 388 Finnish 6, 7, 17, 19, 86, 108-113, 115, 118, 309,316 French 5, 17, 25, 26, 30, 47, 49, 51, 53-55, 57-60, 63-67, 69, 71, 72, 75, 76, 77-79, 81-84, 95, 117, 128, 148, 149, 151, 155, 156, 162, 163, 166-168, 170, 174, 175, 177, 178, 192, 197-200, 206, 207, 210, 211-213, 221, 222, 252, 329, 330, 342, 344, 345, 362, 364, 365-367, 371-375, 378, 386, 388 (see als Corsican and Quebec French) Frisian 180 German 3, 4-6, 9, 10, 12-15, 17-19, 22-26, 29, 31-40, 42-44, 46, 47, 48-51, 53-55, 57-59, 66-71, 75-78,81-83,85,86,89,91,92, 93-98, 103, 106, 108, 111, 113, 114, 116-118, 126, 128, 135, 144, 147-149, 157, 160, 163,

390

LANGUAGES

German (Cont'd) 166, 169, 170, 173-178, 180, 182, 183-185, 192, 195-202, 206-213, 215, 216, 219, 220, 225, 242, 249, 265, 266, 268-274, 276-279, 281, 282, 284, 286-290, 293-297, 302, 307-310, 312-314, 316-319, 321, 324, 329, 331-334, 337, 340-343, 347-352, 356-358, 360, 363-371, 374, 375, 377-381, 385-388 Germanic 22-25, 36, 54, 55, 76, 82, 118, 157, 211, 315, 367, 368, 373, 378, 388 Icelandic 5, 17, 23, 39-43, 49, 50, 54, 55, 173, 176 Irish 199, 210, 212, 221, 222, 251, 252 Italian 8, 12, 17, 30, 32, 34, 35, 37, 43, 44, 48, 54, 55, 80, 83, 122, 152, 154, 155, 166, 170, 173-176, 198, 206, 208, 211, 281, 385 (see als Veneto and Vicentino) Japanese 3, 11, 12, 17, 18, 53, 82, 253, 257, 260-264, 315, 366 Korean 3, 13, 17, 89, 116, 265, 268-271, 276, 280-282, 285, 291, 294, 297, 298, 300, 301, 307,311,388 Palauan 199

Polish 141 Portuguese 12, 163, 175, 281, 374 Quebec French 199, 210 Romance 12, 13, 175, 193, 198, 199, 207, 209, 211, 252, 319, 363, 379 Romani 10, 215, 216, 219, 225, 242, 249 Russian 17 Somali 171 Spanish 12, 15, 17, 30, 32, 35, 82, 281, 385 Swedish 24, 46, 49, 54, 59, 71, 83, 128, 148, 167, 171, 173, 175, 180, 206, 209, 375 Swiss 3, 9, 32, 36, 53, 148, 177, 178, 182-184, 197, 199, 206, 207, 210, 211 (see also Bernese) Turkish 3, 13-15, 17, 265, 266, 268-272, 274, 276, 277, 280-282, 285, 288, 290, 291, 294, 295, 297, 298, 300, 301, 307, 310, 311, 316-318, 321-324, 331-334, 340-342, 348-353, 358, 360, 362, 363, 365, 366, 368, 388 Veneto 199, 206 Vicentino 193, 198, 199,211 West-Flemish 146 Yiddish 2 3 , 3 9 , 4 9 , 5 1

Index of Names

Abney, Stephen 89, 93, 776, 119, 139, 146, 267, 313 Adams, Marianne 57, 80, 81 Ahn, Hee-Don 89, 776 Akiyama, Satoshio 223, 252 (see also Roeper et al.) Akmajian, Adrian 172, 173 Avrutin, Sergej 1, 77 Bader, Thomas 198, 204-208, 277 Baker, Mark 15, 17, 188, 322, 330, 362, 363, 364 Bayer, Josef 323,564 Belletti, Adriana 136,746 Bellugi, Ursula 26-28, 44,50, 53, 60, 61, 81, 126, 129, 746, 148 Bennis, Hans 135, 746 Berman, Ruth 128, 749, 178, 213 Berwick, Robert 353, 364 Best, Diane 34, 55, 151, 776 (see also Wang et al.) den Besten, Hans 23, 50, 124, 746, 268, 313, 324, 365 Bever, Thomas 253, 264 Bhatt, Christa 94, 102, 776 Bierwisch, Manfred 90, 776 Bley-Vroman, Robert 266, 313, 319,

364, 365 Bloom, Lois 30, 50, 60, 61, 81, 141, 746 Bloom, Paul 47, 57, 153, 775 Borer, Hagit 2, 3, 18, 85, 87, 776, 138, 746, 164, 775, 264, 349, 365 Boser, Katharina 68, 69, 76, 77, 81, 82, 86, 776, 164, 775, 267, 575 Bottari, P. 154, 166, 775, 774 Bowerman, Melissa 21, 57 Bresnan, Joan 88, 776 Brody, Michael 172,175 Brown, Roger 21, 44, 57, 61, 82, 141, 747, 152, 775, 370, 385, 387 Calabrese, Andrea 199,205 Campbell, Richard 44, 57 Cardinaletti, Anna 33, 57, 157, 158, 161, 774 Chien, Yu-Chin 1, 18, 37, 57, 248, 257 Chilosi, A.M. 154, 166, 775, 774 Choe, Hyon Sook 270,575 Chomsky, Noam 16, 18, 47, 49, 57, 76, 77, 79, 81, 82, 85, 88, 89,

392 Chomsky, N. (Cont'd) 116, 119, 121, 127, 141, 146,747, 160, 161, 169, 174, 181, 182, 208, 232, 251, 267, 296, 309, 313, 320, 325, 365, 370, 371, 373, 374, 385-387 Chung, Sandra 221, 251 Cinque, Guglielmo 171, 174, 216, 231, 251 Cipriani, P. 154, 166, 173, 174 Clahsen, Harald 6, 7, 12, 13, 15, 18, 22, 24, 29, 43, 46, 47, 51, 68-70, 82, 85, 95, 97, 104, 105, 113, 114, 116, 117, 125, 126, 128, 147, 170, 174, 177, 197, 205, 208, 209, 266-268, 275, 277, 281, 287, 289, 290, 295, 296, 308, 310, 313-315, 319, 363, 365, 367, 369, 370, 378, 380-382, 384-386, 387 Clark, Robin 171,350,355,363,364, 365 Clements, George N. 362, 365 Contreras, Heles 57, 82 Corver, Norbert 141, 143, 147 Crain, Stephen 10, 11, 18, 37, 52, 215, 217, 226, 232, 236, 251, 252, 257, 264 Crisma, P. 170, 174 Culicover, Peter 2, 19, 349, 368 Czepluch, Hartmut 91,93,777 Demers, Richard A. 172, 173 Demuth, Katherine 199, 209, 297, 311,374 Deprez, Viviane 4, 5, 7, 8, 15, 22, 24, 54, 57, 59, 60, 68, 74, 81, 82, 125, 747, 166, 774, 308, 574 Diesing, Molly 39, 49, 57 den Dikken, Marcel 136, 747 duPlessis, Jean 12,18, 266, 281, 314, 319, 363, 364, 365, 369, 379, 382, 386, 387

NAMES Eisenbeiss, Sonja 6, 13,18, 114, 777, 266, 267, 312, 374, 387 (see also Clahsen et al.) Emonds, Joseph 25, 57, 57, 82, 321, 322, 365 Enç, Murvet 44, 57 Eubank, Lynn 15, 16, 266, 314, 379, 380, 383, 387 Evers, Arnold 124, 747, 179, 209 Farhady, Hossein 275, 309, 574 Felix, Sascha 67, 68, 82, 113, 777, 185, 189, 207, 209, 266, 574 Flynn, Susanne 266, 574, 364, 365 Fodor, Janet D. 217, 257 Frampton, John 250, 252 Friedemann, Marc-Ariel 166, 170, 774 Frijn, Jacqueline 24, 57 Fritzenschaft, Agnes 22, 57, 178, 185-187, 189, 207, 209, 319, 365 (see also Gawlitzek-Maiwald et al.) Garrett, Merrill F. 50,57 Gass, Susan 364, 365 Gawlitsek-Maiwald, Ira 22, 47, 57, 178, 185-187, 197, 209, 319, 365 (see also Fritzenschaft et al.) Georgopoulos, Carol 159, 774, 199, 209 Gregg, Kevin 318, 366 Grimm, H. 185, 186, 189, 209 Grimshaw, Jane 50, 51,52 Guasti, Maria-Teresa 154, 166, 774 Gubala-Ryzak, Magda 355, 366-368 Guéron, Jacqueline 140, 747 Guilfoyle, Eithne 21, 27, 28, 52, 75, 82, 121, 747, 166, 774 de Haan, Ger 24, 30, 31, 34, 57, 52, 170, 774 Haegeman, Liliane 36, 48, 52, 146, 747, 155, 156, 172, 774, 195,209

NAMES

Haider, Hubert 94,777 Hall, Tracy A. 310, 314 Hamann, Cornelia 159, 170, 174 Hamburger, Henry 37, 52, 237, 238, 250, 252, 257, 264 Hankamer, Jorge 362, 366 Harnish, Robert M. 172, 173 Hatch, Evelyn 275, 309, 314 Haverkort, Marco 177, 194, 197, 198, 209 Hayashibe, Hideo 11, 12, 18, 253-259,261,263,264 Higginbotham, James 171, 174 Hoekstra, Teun 6, 7, 14, 15, 18, 22, 52, 135, 143,746,747,351,356, 366 Holmberg, Anders 26, 49, 54, 85, 118, 161, 774, 378, 388 Hong, Upyong 359, 366 Hood, Lois 30,50 Hornstein, Norbert l, 18 Huang, C.-T. James 31, 33, 52, 80, 82, 152, 157, 774 Hyams, Nina 1, 3-5, 7, 8,18, 22, 30, 34, 43-48, 50, 52, 126, 747, 151, 153, 167, 774, 248,252, 319, 366 Iatridou, Sabine 375, 385, 386, 387 Ingram, David 200,209 Ito, Katsotoshi 259, 264 Iwatate, Shizuo 255, 260, 264 Jaeggli, Osvaldo 33, 34, 46, 52, 53, 151,775 Jelinek, Eloise 3 11, 315 Johnson, Kyle 50 Jordens, Peter 6, 7, 14, 15, 18, 22, 52, 125, 128,747,351,356,363, 366 van Kampen, J. 47, 53, 179, 209 Karmiloff-Smith, Annette 37, 53 Kayne, Richard 221, 252, 330, 366

393

Kazman, Rick 152, 166, 775 Kennelly, Sarah 322, 366 Kenstowicz, Michael 171, 175 Keyser, Samuel Jay 384, 387 Kitagawa, Yoshihisa 27, 53, 57, 65, 82, 285, 308, 575, 326, 366 Klein, Sharon 22,47,55 Klima, Ed 26-28, 53, 129, 748 Koopman, Hilda 27, 53, 57, 62, 63, 79, 82, 136, 138, 148, 160, 775, 321, 326, 329, 346, 366 Köpcke, M. 386,557 Kornfilt, Jaklin 322, 362, 366 Koster, Charlotte 143, 747 Koster, Jan 23, 53, 216, 252, 268, 575, 324, 366 Krashen, Stephen 320, 366 Kural, Murat 323, 324, 366 Kuroda, S.-Y. 326,566 Laka, Itziar 57, 82 Lange, S. 206, 209 Larsson, K. 206,209 Lasnik, Howard 8, 16, 79, 152, 158-160, 775, 192, 209, 250, 252, 375-377, 386, 387 Lasser, Ingeborg 207, 272 Lebeaux, David 21, 27, 53, 87-89, 777, 121, 138, 148, 166, 775, 181, 210, 267, 308, 311, 575, 351,566 Lefevre, Claire 199,270 Lenneberg, Eric 319, 367 Leveille, Madeleine 61, 83 (see also Suppes et al.) Levitt, Andrea 34, 55, 151, 776 (see also Wang et al.) Lightbown, Patsy 81, 82, 128, 748 Lightfoot, David 1, 18, 179 Lillo-Martin, Diane 34, 48, 53, 55, 151, 176 (see also Wang et al.)

394 Lundin, Barbro 128, 148 Lust, Barbara 68, 82, 86, 776, 127, 149, 153, 164,173,176, 267, 313 (see also Boser et al.) MacNamara, John 138, 148 MacWhinney, Brian 48, 53, 60, 81, 83, 154, 175 Mahajan, Anoop 136, 148 Maling, Joan 362, 367 Mallis, Lourdes 223, 252 (see also Roeper et al.) Mandelbaum, Deborah 207, 212 Manzini, Maria Rita 1,79 Masunaga, Kyogo 257, 264 McCloskey, James 199, 210, 221, 251, 252 McDaniel, Dana 10, 19, 215, 216, 219, 249, 252 Meisel, Jürgen 15, 18, 22, 24-26, 29, 46, 47, 53, 81, 83, 95, 104, 117, 125, 126,148, 170,175, 197,210, 275, 375, 319, 363, 367, 369, 378, 379, 381,383, 386,357,388 Miller, Peggy, 30 50 Miller, Max 69, 70, 81, 83, 95, 777 Mills, Anne 69, 70, 83, 95, 104, 105, 114, 115, 777 Moro, Andrea 172, 775 Mulder, René 136,747 Müller, Gereon 206, 270 Müller, Natascha 22, 24-26, 29, 47, 53, 81, 83, 125, 126, 148, 177, 178, 185, 186, 189-191, 197, 200, 204, 205, 207,210, 386,388 Muysken, Pieter 12, 13,18, 266, 277, 281, 310,314, 319, 363,365,387 Noonan, Maire 21, 27, 28, 52, 75, 82, 121, 147, 166, 174, 199, 205-207, 270

NAMES O'Neil, Wayne 364,365 Okubo, Ai 260,264 Olsen, Susan 94, 102, 103, 777 Otsu, Yukio 11, 12 Ouhalla, Jamal 87, 777, 199, 270, 267, 375 Park, T. 67, 68, 81, 83, 185, 270 Parodi, Teresa 95, 104, 115, 117 Penke, Martina 205, 209, 290, 295, 296, 374 Penner, Zvi 9, 22, 29, 32, 36, 53, 126, 148, 177, 178, 184, 194, 197-199, 202, 204, 205-208, 270, 277 Pesetsky, David 50, 54, 221, 252 Pfanner, L. 154,174 Pienemann, Manfred 15, 18, 275, 375, 319, 363,367,369,357 Pierce, Amy 4, 5, 7, 8, 15, 22, 24, 27, 30, 54, 57, 59, 60, 63-65, 68, 74, 81, 82, 83, 125, 128, 747, 148, 151, 166, 170, 774, 775, 308, 374 Pinker, Steven 3, 12, 19, 22, 54, 87-89, 91, 777, 226, 252, 267, 375, 319, 367 Platzack, Christer 24, 26, 46, 49, 54, 71,72,75,53,85,775, 128,745, 166, 167, 775, 268, 375, 329, 346, 367, 378, 355 Plunkett, Kim 128,745 Poeppel, David 1, 19, 22, 31, 54, 126, 128, 745, 164, 166, 775 Poletto, C. 199,207,277 Pollock, Jean-Yves 25, 54, 57, 66, 76, 78-81, 53, 119, 121, 146, 745, 268, 375, 321, 322, 367, 371,374,386,355 Preyer, William 95, 775

NAMES Radford, Andrew 1 4 , 7 9 , 2 1 , 2 5 , 5 4 , 59, 75, 83, 121, 146, 148, 153, 162, 165, 166, 175, 267, 281, 315, 386 Raposo, Eduard 163, 175 Rizzi, Luigi 6, 8, 9, 15, 19, 23, 25, 31, 33, 34, 37, 47, 54, 78, 80, 83, 85,775, 127, 128, 134, 142,745, 151, 159-161, 163, 171-173, 775, 776, 177, 178, 192, 195-197, 202, 205, 277, 221, 222, 231, 232, 250, 251, 252, 310, 375, 321, 326, 330, 342, 344, 345, 362, 363, 367 Roberts, Ian 25, 54, 78, 83, 197, 206, 277, 321, 326, 330, 342, 344, 345, 350, 355, 362-364, 365, 367 Roeper, Thomas 10, 19, 48, 54, 87, 118, 125, 143, 747, 148, 152, 153, 776, 177-179, 181, 189, 192, 200, 211-213, 216, 217, 223, 224, 226, 227, 229, 241, 251, 252, 267, 315, 319, 351, 356, 367, 384, 387 (see also de Villiers et al.) Rögnvaldsson, Eiríkur 39, 49, 54 Rohrbacher, B. 378, 379, 386, 388 Rooth, Mats 223, 252 (see also Roeper et al.) Ross, John Robert 157, 776 Rothstein, Susan 203, 272 Rothweiler, Monika 144, 148, 178, 272, 288, 315 Rouveret, Alain 314,388 Safir, Kenneth 53 Saito, Mamuro 260, 264 Sano, Keiko 254,259,264 Santelmann, Lynn 68, 82, 86, 776, 164, 773, 267, 373 (See also Boser et al.)

395

Schachter, Jacquelyn 266, 375, 364, 365 Schlesinger, I. 21, 54 Schmerling, Susan 172, 776 Schönenberger, Manuala 202, 208, 277 Schwartz, Bonnie D. 3, 14-16,19, 49, 55, 70, 84, 266, 267, 281, 288, 307,311,375,376,318,319,355, 356, 361, 363, 364, 367,368, 369, 379, 380, 383, 384, 386, 388 Selinker, Larry 313, 376 Sezer, Engin 362, 365 Siadhail, M. 199,272 Sigurðsson, Halldór 173, 776 Sigurjónsdóttir, Sigríður 22, 41, 42, 50,55 Smith, R. 37, 53, 61, 83, 211 (see also Suppes et al.) Smolka, Klaus 69, 70, 82 Snow, Catherine 48, 53, 60, 81, 83, 154, 175, 274, 284 Solà, Joan 171,776 Solin, Doreen 12,18, 266, 374, 319, 365, 369, 387 (see also duPlessis et al.) Sportiche, Dominique 27, 53, 57, 62, 63, 79, 82, 83, 136, 138, 148, 160, 775, 285, 308, 376, 321, 326, 366, 368 Sprouse, Rex 14-16, 266, 267, 288, 311,376, 362,367 Starke, Michael 165 von Stechow, Armin 383, 388 Stenson, N. 199, 272 Stern, Wilhelm 95,775 Sternefeld, Wolfgang 206, 210, 212, 383, 388 Stowell, Tim 8, 79, 152, 158-160, 162, 775, 776, 192, 209

396

NAMES

Strömqvist, Sven 128, 148 Stromswold, Karin 63, 73, 74, 79, 55, 200, 212, 226, 252 von Stutterheim, Christiane 272, 298, 307, 310, 316 Suppes, Patrick 61, 83 Szabolcsi, Anna 119, 139, 149 Taraldsen, Knut Tarald 161, 776 Tavakolian, Susan 351, 368 Thiersch, Craig 324, 368 Thornton, Rosalind 10, 11, 18, 19, 215, 217, 222, 226, 232, 236, 238, 251, 252 Thráinsson, Höskuldur 39, 49, 54, 55 Toivainen, Jorma 111, 112, 115, 118 Tomaselli, Alessandra 281, 316, 331, 332, 363, 367, 368, 369, 379, 380, 382, 386, 388 Tracy, Rosemary 22, 57, 95, 96, 104, 105, 114, 115, 118, 177, 178, 185-189, 192, 206, 207, 209, 211, 212, 319, 365 (see also Fritzenschaft et al. and GawlitzekMaiwald et al.) Travis, Lisa 12, 18, 266, 314, 319, 322, 365, 368, 369, 387 (see also Solin et al.) Tuijnman, Kees 30, 31, 34, 52, 170, 774 Ueyama, Ayumi 263, 264 Underhill, Robert 322, 323, 368 Vainikka, Anne 6, 10, 13, 14, 16,18, 19, 90, 109, 110, 775, 206,272, 216, 241, 251, 252, 266, 267, 281, 293, 308, 309, 314, 316, 340, 384, 385, 387, 388 (see also Clahsen et al. and de Villiers et al.) Valian, Virginia 37, 47, 48, 55, 151, 152, 154, 165, 170, 776, 207,272

Verrips, Maaike 81, 84, 126, 128, 149, 178, 205, 272, 275, 378, 388 Vikner, Sten 49, 55, 70, 84, 92, 775, 329, 346, 368, 386, 388 de Villiers, Jill 10,19, 146,149, 181, 189, 192, 211-213, 216, 217, 222, 223, 225, 226, 228-232, 236-239, 241, 242, 244, 248, 249, 251, 252 Wang, Qi 34, 55, 151, 776 Wegener, Heide 91, 93, 775 Wegmüller, U. 199,277 Weinberg, Amy 207, 272 Weissenborn, Jürgen 22, 24, 29, 38, 55, 63, 81, 84, 86, 775, 125, 126, 128, 144, 149, 153, 170, 776, 177, 178, 185, 186, 189, 192, 194, 195, 197, 198, 205-208, 209, 211-213, 267, 316, 319, 368, 378, 555 Weverink, Meike 24, 55, 177, 195, 275 Wexler, Kenneth 1-3, 15, 16, 77-79, 22, 31, 34, 37, 45-48,51,52, 54, 55, 69, 75-77, 84, 87, 776, 125-128, 134, 138, 746, 148, 149, 151, 153, 164, 166, 775, 174-176, 248, 257, 264, 349, 365, 368, 370-373, 375, 378, 379, 382, 385, 555 White, Lydia 12, 75, 266, 307, 314, 316, 319, 355, 364, 365, 368, 369, 382, 384, 557, 555 see also Solin et al.) Whitman, John 68, 52, 86, 776, 127, 149, 153, 164, 775, 776, 267, 575 (See also Boser et al.) Winkler, S. 185, 209 (see also Fritzenschaft et al.)

NAMES Wode, Henning 61, 68, 83, 84, 96, 126,749, 185, 186,206,275 Wurzel, Wolfgang 92, 93, 118 Young-Scholten, Martha 13, 14, 16, 281, 316, 340, 384, 388

397

Zagona, Karen 285, 308, 316, 326, 368 Zanuttini, Raffaella 57, 84, 122, 146, 147, 149 Zwart, Jan-Wouter 197, 205, 273, 293, 316

Boser et al (= Boser, Lust, Santelmann & Whitman 1992) 69, 76, 77, 81 Clahsen et al (= Clahsen, Eisenbeiss & Vainikka, this volume) 6, 7 duPlessis et al (= duPlessis, Solin, Travis & White 1987) 281, 363, 364, 379, 382, 386 Fritzenschaft et al (= Fritzenschaft, Gawlitzek-Maiwald, Tracy & Winkler 1990) 185-187, 189, 207 Gawlitzek-Maiwald et al (= Gawlitzek-Maiwald, Tracy, Fritzenschaft 1992) 47, 185-187, 197 Roeper et al (= Roeper, Rooth, Mallis & Akiyama 1985) 224, 226, 227 Suppes et al (= Suppes, Smith & Leveille 1973) 61 de Villiers et al (= de Villiers, Roeper & Vainikka 1990) 10, 216, 217, 222, 223, 225, 226, 228-232, 236-239, 241, 242, 244, 248, 249 Wang et al (= Wang, Lillo-Martin, Best & Levitt 1992) 34

Index of Subjects

Acquisition of agreement 268, 296, 378-380 of Case 85 and phrase structure 90, 92, 112 of determiners 142 of embedded clauses 144 of functional elements 21, 22, 46, 78,89, 121, 123,295,297,375 L2 267 of inflection 44 of modals 123, 125 of negation 60, 66, 125 of NP movement 79 of null arguments 38, 43, 152 of phrase structure 88 in L2 266,267,295 of verb placement 32, 59, 65, 66, 69, 71, 79, 183, 373 of wh-movement 75, 77, 79 of wh-questions 248, 250 of scrambling 134, 135 Bijection Principle 160, 175 Bootstrapping 6, 12, 88, 91, 95, 103, 112, 178, 179

Case Filter 14, 114, 181, 325, 326, 329-331, 342, 354, 357 Complex inversion 54, 83, 178, 179, 197,211, 342,362,366,367 Continuity and long wh-movement 10, 11 hypothesis 3, 4, 22, 78, 319, 359 strong 5, 86, 267 of functional categories 8 vs. maturation 2 weak 6,87,90,267 Copying 10, 188, 201, 211, 215, 216, 219, 220, 248, 249, 252 anti-weak 136 criterion for null constants 160 questions vs. bound variable ques­ tions 224-226 scrambling induces weak 135 strong, absence of 223 children's knowledge of 227 homogenious behavior of R-expressions 159 weak, alleviated in Pro gates 171 as a property of variables 159, 192

400

SUBJECTS

Determinacy 15, 320, 356, 360 Diary subject drop 178, 194, 195 difference with topic drop 195 does not involve chain with discourse binder 196 value for topic drop at initial state 178 Diary contexts IP rather than CP 196 Diary English CP not obligatory in 9 early null subject in 8 first person subject drop in 36, 48 Discourse binder 196 ECP 309 and adjunct extraction 11 excludes possessor extraction 141 explanation for Jo-support 121 explanation for medial-wh questions 221, 251 identification requirement for null constant 160-162, 172 FCH 23, 32, 38, 43 Freezing 181, 183, 191, 201, 202 Full Clause Hypothesis see FCH Full House Principle 13, 293, 313 Fundamental Difference Hypothesis 266 Initial state of L2 acquisition 16, 370, 384, 385 of the Interlanguage 350 Lasnik's condition 376, 377 Lexical Learning Hypothesis see also LLH 45, 46, 85 Licensing device 182, 203 LLH 85

Maturation account, and small clause theory 21 weakened by similarities L1 and L2 14 of A-movement 138,264 of COMP enables movement 40 of functional categories 3, 8, 78, 121, 123 of X'-principles 113 strong 3 UG-constrained 3, 87 vs. continuity 2, 267 Negative evidence 29, 40, 173, 307, 316, 355, 367, 368 Null argument co-occurrence with V1 31 parameter 4, 23, 32-36, 38, 48 phenomenon in child language 22, 23, 30, 32 subject/object asymmetry 35, 38 topic-licensed 8, 31 Null constant 9, 159-165, 167-169, 171, 192-196, 198, 199, 201, 202, 8, 9 Positive evidence 6, 34, 37, 40, 89, 90, 222, 267, 268, 297, 307, 357 Residual verb second 22, 25, 26, 43, 45, 183, 192 ShCH 22, 23, 25, 26, 29, 39, 40, 46 Subject empty non-thematic 36 empty thematic 270, 271, 276 right-dislocation of thematic 197 Subscrambling 8, 134, 139-142 Subset Principle 19, 29, 165, 170, 353, 356, 361, 364, 368

SUBJECTS Thematic relations determined by word order 11, 253 in early grammars 21 Thematic roles in early grammars 91 verb raising and transmission of 375 Thematic verbs agreement 383 finite, preceded by direct object 340 in early grammars 189, 190 in early questions 193 movement of 371, 373

401

Theta -roles, verb raising and transmission of 372, 374, 377, 386 -relation internal to lexical projec­ tion 146 position, in definition of A-position 35, 163 Topic drop 4, 31, 32, 37, 156, 157, 171 in adult language 195 null constant account 192, 195 residual 36 wh in situ 153, 165, 225, 250 yes/no question 207, 234