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 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, 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 feature, inflected verbs raise when is strong, but as the infinitive is and , 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 , as in the early L1 acquisition analysis, is not yet a feature of Jose's Interlanguage but strong is. According to Eubank, this makes a prediction: verb raising should occur when marked for inflection, but when it has the feature , 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. ) 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 either in INFL, as in English, or in COMP, as in German. Meisel & Müller assume that German children (and all children) initially place the 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 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 comple ment, while think selects a 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
![Prosodic Phonology (Studies in Generative Grammar [Sgg])](https://epdf.pub/img/300x300/prosodic-phonology-studies-in-generative-grammar-s_5ab29a4eb7d7bc917376f758.jpg)
