david lebeaux language acquisition and the form of the grammar 2000
TRANSCRIPT
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
1/308
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
2/308
LANGUAGE ACQUISITION AND THE FORM OF THE GRAMMAR
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
3/308
This book was originally selected and revised to be included in the World Theses Series
(Holland Academic Graphics, The Hague), edited by Lisa L.-S. Cheng.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
4/308
LANGUAGE ACQUISITION
AND THE FORMOF THE GRAMMAR
JOHN BENJAMINS PUBLISHING COMPANY
PHILADELPHIA/AMSTERDAM
DAVID LEBEAUX
NEC Research Institute
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
5/308
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 Z39.48-1984.8
TM
Library of Congress Cataloging-in-Publication Data
Lebeaux, David.
Language acquisition and the form of the grammar / David Lebeaux
p. cm.
Includes bibliographical references and index.
1. Language acquisition. 2. Generative grammar. I. Title.
P118.L38995 2000
401.93--dc21 00-039775
ISBN 90 272 2565 6 (Eur.) / 1 55619 858 2 (US)
2000 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 P.O.Box 27519 Philadelphia PA 19118-0519 USA
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
6/308
Table of Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
C1
A Re-Definition of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1 The Pivot/Open Distinction and the Government Relation . . . . . . . . 7
1.1.1 Braines Distinction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 The Government Relation . . . . . . . . . . . . . . . . . . . . . . . . . 91.2 The Open/Closed Class Distinction . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Finiteness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.2 The Question of Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Triggers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 A Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Determining the base order of German . . . . . . . . . . . . . . . . 17
1.3.2.1 The Movement of NEG (syntax) . . . . . . . . . . . . . . . 24
1.3.2.2 The Placement of NEG (Acquisition) . . . . . . . . . . . . 26
C2
Project-, Argument-Linking,
and Telegraphic Speech . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1 Parametric variation in Phrase Structure . . . . . . . . . . . . . . . . . . . . . 31
2.1.1 Phrase Structure Articulation . . . . . . . . . . . . . . . . . . . . . . . 31
2.1.2 Building Phrase Structure (Pinker 1984) . . . . . . . . . . . . . . . 32
2.2 Argument-linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2.1 An ergative subsystem: English nominals . . . . . . . . . . . . . . 41
2.2.2 Argument-linking and Phrase Structure: Summary . . . . . . . . 45
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
7/308
vi TABLE OF CONTENTS
2.3 The Projection of Lexical Structure . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.1 The Nature of Projection . . . . . . . . . . . . . . . . . . . . . . . . . . 512.3.2 Pre-Project- representations (acquisition) . . . . . . . . . . . . . . 56
2.3.3 Pre-Project- representations and the Segmentation Problem . 60
2.3.4 The Initial Induction: Summary . . . . . . . . . . . . . . . . . . . . . 65
2.3.5 The Early Phrase Marker (continued) . . . . . . . . . . . . . . . . . 66
2.3.6 From the Lexical to the Phrasal Syntax . . . . . . . . . . . . . . . . 75
2.3.7 Licensing of Determiners . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.3.8 Submaximal Projections . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
C3Adjoin-and Relative Clauses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.2 Some general considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.3 The Argument/Adjunct Distinction, Derivationally Considered . . . . . 94
3.3.1 RCs and the Argument/Adjunct Distinction . . . . . . . . . . . . . 94
3.3.2 Adjunctual Structure and the Structure of the Base . . . . . . . . 98
3.3.3 Anti-Reconstruction Effects . . . . . . . . . . . . . . . . . . . . . . . . 102
3.3.4 In the Derivational Mode: Adjoin- . . . . . . . . . . . . . . . . . . 104
3.3.5 A Conceptual Argument . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.4 An Account of Parametric Variation . . . . . . . . . . . . . . . . . . . . . . . 112
3.5 Relative Clause Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.6 The Fine Structure of the Grammar, with Correspondences: The
General Congruence Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.7 What the Relation of the Grammar to the Parser Might Be . . . . . . . 136
C4
Agreement and Merger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1454.1 The Complement of Operations . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.2 Agreement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.3 Merger or Project- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.3.1 Relation to Psycholinguistic Evidence . . . . . . . . . . . . . . . . . 154
4.3.2 Reduced Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.3.3 Merger, or Project- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.3.4 Idioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
8/308
TABLE OF CONTENTS vii
C5
The Abrogation of DS Functions:Dislocated Constituents and Indexing Relations . . . . . . . . . . . . . . . . . 183
5.1 Shallow Analyses vs. the Derivational Theory of Complexity . . . . 184
5.2 Computational Complexity and The Notion of Anchoring . . . . . . . . 188
5.3 Levels of Representation and Learnability . . . . . . . . . . . . . . . . . . . 192
5.4 Equipollence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.5 Case Study I: Tavakolians results and the Early Nature of Control . . 203
5.5.1 Tavakolians Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
5.5.2 Two Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5.5.3 PRO as Pro, or as a Neutralized Element . . . . . . . . . . . . . . . 2085.5.4 The Control Rule, Syntactic Considerations: The Question of
C-command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.5.5 The Abrogation of DS functions . . . . . . . . . . . . . . . . . . . . . 220
5.6 Case Study II: Condition C and Dislocated Constituents . . . . . . . . . 224
5.6.1 The Abrogation of DS Functions: Condition C . . . . . . . . . . . 226
5.6.2 The Application of Indexing . . . . . . . . . . . . . . . . . . . . . . . 229
5.6.3 Distinguishing Accounts . . . . . . . . . . . . . . . . . . . . . . . . . . 234
5.7 Case Study III: Wh-Questions and Strong Crossover . . . . . . . . . . . . 239
5.7.1 Wh-questions: Barriers framework . . . . . . . . . . . . . . . . . . . 240
5.7.2 Strong Crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
5.7.3 Acquisition Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
5.7.4 Two possibilities of explanation . . . . . . . . . . . . . . . . . . . . . 248
5.7.5 A Representational Account . . . . . . . . . . . . . . . . . . . . . . . . 249
5.7.6 A Derivational Account, and a Possible Compromise . . . . . . 251
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
9/308
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
10/308
There are two ways of painting two trees together. Draw
a large tree and add a small one; this is called fu lao
(carrying the old on the back). Draw a small tree and add
a large one; this is called hsieh yu (leading the young by
the hand). Old trees should show a grave dignity and an
air of compassion. Young trees should appear modest and
retiring. They should stand together gazing at each other.
Mai-mai Sze
The Way of Chinese Painting
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
11/308
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
12/308
Acknowledgments
This book had its origins as a linguistics thesis at the University of Massachu-setts. First of all, I would like to thank my committee: Tom Roeper, for scores
of hours of talk, for encouragement, and for his unflagging conviction of the
importance of work in language acquisition; Edwin Williams, for the example of
his work; Lyn Frazier, for an acute and creative reading; and Chuck Clifton, for
a psychologists view. More generally, I would like to thank the faculty and
students of the University of Massachusetts, for making it a place where creative
thinking is valued. The concerns and orientation of this book are very much
molded by the training that I received there.
Further back, I would like to thank the people who got me interested in allof this in the first place: Steve Pinker, Jorge Hankamer, Jane Grimshaw, Annie
Zaenen, Merrill Garrett and Susan Carey. I would also like to thank Noam
Chomsky for encouragement throughout the years.
Since the writing of the thesis, I have had the encouragement and advice of
many fine colleagues. I would especially like to thank Susan Powers, Alan
Munn, Cristina Schmitt, Juan Uriagereka, Anne Vainikka, Ann Farmer, and Ana-
Teresa Perez-Leroux. I am also indebted to Sandiway Fong, as well as Bob
Krovetz, Christiane Fellbaum, Kiyoshi Yamabana, Piroska Csuri, and the NEC
Research Institute for a remarkable environment in which to pursue the researchfurther.
I would also like to thank Mamoru Saito, Hajime Hoji, Peggy Speas,
Juergen Weissenborn, Clare Voss, Keiko Muromatsu, Eloise Jelinek, Emmon
Bach, Jan Koster, and Ray Jackendoff.
Finally, I would like to thank my parents, Charles and Lillian Lebeaux, my
sister, Debbie Lebeaux, and my sons, Mark and Theo. Most of all, I would like
to thank my wife Pam, without whom this book would have been done badly, if
at all. This book is dedicated to her, with love.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
13/308
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
14/308
Preface
What is the best way to structure a grammar? This is the question that I startedout with in the writing of my thesis in 1988. I believe that the thesis had a
marked effect in its answering of this question, particularly in the creation of the
Minimalist Program by Chomsky (1993) a few years later.
I attempted real answers to the question of how to structure a grammar, and
the answers were these:
(i) In acquisition, the grammar is arranged along the lines of subgrammars.
These grammars are arranged so that the child passes from one to the next,
and each succeeding grammar contains the last. I shall make this clearer
below.
(ii) In addition, in acquisition, the child proceeds to construct his/her grammar
from derivational endpoints (Chapter 5). From the derivational endpoints,
the child proceeds to construct the entire grammar. This may be forward or
backward, depending on what the derivational endpoint is. If the derivation
endpoint, or anchorpoint, is DS, then the construction is forward; if the
derivational endpoint or anchorpoint is S-structure or the surface, then the
construction proceeds backwards.
The above two proposals were the main proposals made about theacquisition sequence. There were many proposals made about the syntax. Of
these, the main architectural proposals were the following.
(iii) The acquisition sequence and the syntax in particular, the syntactic
derivation are not to be considered in isolation from each other, but
rather are tightly yoked. The acquisition sequence can be seen as the result
of derivational steps or subsequences (as can be seen in Chapter 2, 3, and
4). This means that the acquisition sequence gives unique purchase onto the
derivation itself, including the adult derivation.
(iv) Phrase structure is not given as is, nor is derived top-down, but rather iscomposed (Speas 1990). This phrase structure composition (Lebeaux 1988),
is not strictly bottom up, as in Chomskys (1995) Merge, but rather involves
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
15/308
xiv PREFACE
(a) the intermingling or units, (b) is grammatically licensed, and not simply
geometrical (bottom-up) in character (in a way which will become clearerbelow), and (c) involves, among other transformations, the transformation
Project- (Chapter 4).
(v) Two specific composition operations (and the beginnings of a third) are
proposed. Adjoin- (Chapter 3) is proposed, adding adjuncts to the basic
nuclear clause structure (Conjoin- is also suggested in that chapter). In
further work, this is quite similar to the Adjunction operation of Joshi and
Kroch, and the Tree Adjoining Grammars (Joshi 1985; Joshi and Kroch
1985; Frank 1992), though the proposals are independent and the proposals
are not exactly the same. The second new composition operation isProject- (Chapter 4), which is an absolutely new operation in the field. It
projects open class structure into a closed class frame, and constitutes the
single most radical syntactic proposal of this book.
(vi) Finally, composition operations, and the variance in the grammar as a
whole, are linked to the closed class set elements like the, a, to, of, etc.
In particular, each composition operation requires the satisfaction of a
closed class element; as well as a closed class element being implicated in
each parameter.
These constitute some of the major proposals that are made in the course of this
thesis. In this preface I would like to both lay out these proposals in more detail,
and compare them with some of the other proposals that have been made since
the publication of this thesis in 1988. While this thesis played a major role in the
coming of the Minimalist Program (Chomsky 1993, 1995), the ideas of the thesis
warrant a renewed look by researchers in the field, for they have provocative
implications for the treatment of language acquisition and the composition of
phrase structure.
Let us start to outline the differences of this thesis with respect to laterproposals, not with respect to language acquisition, but with respect to syntax. In
particular, let us start with parts (iv) and (v) above: that the phrase marker is
composed from smaller units.
A similar proposal is made with Chomskys (1995) Merge. However, here,
unlike Merge:
(1) The composition is not simply bottom-up, but involves the possible
intermingling of units.
(2) The composition is syntactically triggered in that all phrase structurecomposition involves the satisfaction of closed class elements
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
16/308
PREFACE xv
(Chapters 3 and 4), and is not simply the geometric putting together
of two units, as in Merge, and(3) The composition consists of two operations among others (these are
the only two that are developed in this thesis), Adjoin- and
Project-.
With respect to the idea that all composition operations are syntactically triggered
by features, let us take the operation Adjoin-. This takes two structures and
adjoins the second into the first.
(1) s1:
s2:
the man met the woman
who loved him
the man met the woman
who loved him
Adjoin-
This shows the intermingling of units, as the second is intermeshed with the first.
However, I argue here (Chapter 4), that it also shows the satisfaction of closed
class elements, in an interesting way. Let us call the wh-element of the relative
clause, who here, the relative clause linker.
It is a proposal of this thesis that the adjunction operation itself involves the
satisfaction of the relative clause linker (who), by the relative clause head (the
woman), and it is this relation, which is the relation of Agreement, which composesthe phrase marker. The relative clause linker is part of the closed class set. This
relative clause linker is satisfied in the course of Agreement, thus the composi-
tion operation is put into a 1-to-1 relation with the satisfaction of a closed class
head. (This proposal, so far as I know, is brand new in the literature).
(2) Agree Relative head/relativizer Adjoin-
This goes along with the proposal (Chapter 4), which was taken up in the
Minimalist literature (Chomsky 1992, 1995), that movement involves thesatisfaction of closed class features. The proposal here, however, is that composi-
tion, as well as movement, involves the satisfaction of a closed class feature (in
particular, Agreement). In the position here, taken up in the Minimalist literature,
the movement of an element to the subject position is put into a 1-to-1 corre-
spondence with agreement (Chapter 4 again).
(3) Agree Subject/Predicate Move NP (Chapter 4)
The proposal here is thus more thoroughgoing than that in the minimalist
literature, in thatboththe composition operation, and the movement operation aretriggered by Agreement, and the satisfaction of closed class features. In the
minimalist literature, it is simply movement which is triggered by the satisfaction
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
17/308
xvi PREFACE
of closed class elements (features); phrase structure composition is done simply
geometrically (bottom-up). Here, both are done through the satisfaction ofAgreement. This is shown below.
(4) Minimalism Lebeaux (1988)
Movement syntactic (satisfaction
of features)
syntactic (satisfaction
of features)
Phrase Structure
Composition
asyntactic (geometric) syntactic (satisfaction
of features)
This proposal (Lebeaux 1988) links the entire grammar to the closed class set
both the movement operations and the composition operations are linked to
this set.
The set of composition operations discussed in this thesis is not intended to
be exhaustive, merely representative. Along with Adjoin- which Chomsky-
adjoins elements into the representation (Chapter 3), let us take the second, yet
more radical phrase structure composition operation, Project-. This is not
equivalent to Speas (1990) Project-, but rather projects an open class structure
into a closed class frame. The open class structure also represents pure thematic
structure, and the closed class structure, pure Case structure.
This operation, for a simple partial sentence, looks like (5) (see Lebeaux
1988, 1991, 1997, 1998 for further extensive discussion).
The operation projects the open class elements into the closed class (Case)
frame. It also projects up the Case information from Determiner to DP, and
unifies the theta information, from the theta subtree, into the Case Frame, so that
it appears on the DP node.
The Project- operation was motivated in part by the postulation of asubgrammar in acquisition (Chapters 2, 3, and 4), in part by the remarkable
speech error data of Garrett (Chapter 4, Garrett 1975), and in part by idioms
(Chapter 4). This operation is discussed at much greater length in further
developments by myself (Lebeaux 1991, 1997, 1998).
I will discuss in more detail about the subgrammar underpinnings of the
Project-approach later in this preface. For now, I would simply like to point to
the remarkable speech error data collected by Merrill Garrett (1975, 1980), the
MIT corpus, which anchors this approach.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
18/308
PREFACE xvii
(5)
V
N
agent
V
V N
patient
womanseeman
man
VP
VP
DP
+nom
DP
+agent
+nom
Det
+nom
V
V
Det
+nom
the
the
NP
e
V
V
DP
+acc
DP
+patient
+acc
NPDet
+acc
a esee
Theta subtree (open class) Case Frame (closed class)
Project-
NP
+agent
see
Det
+acc
NP
+patient
a woman
Garrett and Shattuck-Hufnagel collected a sample of 3400 speech errors. Of
these, by far the most interesting class is the so-called morpheme-stranding
errors. These are absolutely remarkable in that they show the insertion of open
class elements into a closed class frame. Thus, empirically, the apparent impor-
tance of open class and closed class items is reversed rather than open class
items being paramount, closed class items are paramount, and guide the deriva-
tion. Open class elements are put into slots provided by closed class elements, in
Garretts remarkable work. A small sample of Garretts set is shown below.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
19/308
xviii PREFACE
(6) Speech errors (stranded morpheme errors), Garrett (personal commu-
nication) (permuted elements underlined)Error Target
my frozers are shoulden my shoulders are frozen
that just a back trucking out a truck backing out
McGovern favors pushing busters favors busting pushers
but the cleans twoer twos cleaner
his sink is shipping ship is sinking
the cancel has been practiced the practice has been
cancelled
shes got her sets sight sights set a puncture tiring device tire puncturing device
As can be seen, these errors can only arise at a level where open class elements
are inserted into a closed class frame. The insertion does not take place correctly
a speech error so that the open class elements end up in permuted slots
(e.g. a puncture tiring device).
Garrett summarizes this as follows:
why should the presence of a syntactically active bound morpheme be
associated with an error at the level described in [(6)]? Precisely because theattachment of a syntactic morpheme to a particular lexical stem reflects a
mapping from a functional level [i.e. grammatical functional, i.e. my theta
subtree, D. L.] to a positional level of sentence planning
This summarizes the two phrase structure composition operations that I propose
in this thesis: Adjoin-and Project-. As can be seen, these involve (1) the inter-
mingling of structures (and are not simply bottom up), and (2) satisfaction of
closed class elements. Let us now turn to the general acquisition side of the
problem.
It was said above that this thesis was unique in that the acquisition sequence
and the syntax in particular, the syntactic derivation were not considered
in isolation, but rather in tandem. The acquisition sequence can be viewed as the
output of derivational processes. Therefore, to the extent to which the derivation
is partial, the corresponding stage of the acquisition sequence can be seen as a
subgrammar of the full grammar. The yoking of the acquisition sequence and the
syntax is therefore the following:
(7) A subgrammar approach
S phrase structure composition from smaller units
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
20/308
PREFACE xix
The subgrammar approach means that children literally have a smaller grammar
than the adult. The grammar increases over time by adding new structures (e.g.relative clauses, conjunctions), and by adding new primitives of the representa-
tional vocabulary, as in the change from pure theta composed speech, to theta
and Case composed speech.
The addition of new structures e.g. relative clauses and conjunctions
may be thought of as follows. A complex sentence like that in (8) may be
thought of as a triple: the two units, and the operation composing them (8b).
(8) a. The man saw the woman who loved him.
b. (the man saw the woman (rooted), who loved him, Adjoin-)
Therefore a subgrammar, if it is lacking the operation joining the units may be
thought of as simply taking one of the units let us say the rooted one and
letting go of the other unit (plus letting go of the operation itself). This is
possible and necessary because it is the operation itself which joins the units: if
the operation is not present, one or the other of the units must be chosen. The
subgrammar behind (8a), but lacking the Adjoin-operation, will therefore generate
the structure in (9) (assuming that it is the rooted structure which is chosen).
(9) The man saw the woman.This is what is wanted.
Note that the subgrammar approach (in acquisition), and the phrase structure
composition approach (in syntax itself) are in perfect parity. The phrase structure
composition approach gives the actual operation dividing the subgrammar from
the supergrammar. That is, with respect to this operation (Adjoin-), the
grammars are arranged in two circles: Grammar 1 containing the grammar itself,
but without Adjoin-, and Grammar 2 containing the grammar including Adjoin-.
(10)
Grammar 2
(w/ Adjoin- )
Grammar 1
The above is a case of adding a new operation.
The case of adding another representational primitive is yet more interesting.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
21/308
xx PREFACE
Let us assume that the initial grammar is a pure representation of theta relations.
At a later stage, Case comes in. This hypothesis is of the layering of vocabu-lary: one type of representational vocabulary comes in, and does not displace,
but rather is added to, another.
(11) theta theta + Case
Stage I Stage II
The natural lines along which this representational addition takes place is
precisely given by the operation Project-. The derivation may again be thought
of as a triple: the two composing structures, one a pure representation of theta
relations, and one a pure representation of Case, and the operation composing them.
(12) ((man (see woman)), (the __ (see (a __))), Project-)
the sees in theta tree and Case frame each contain partial informa-
tion which is unified in the Project- operation.
The subgrammar is one of the two representational units: in this case, the unit
(man(see woman)). That is a sort of theta representation or telegraphic speech.
The sequence from Grammar 0 to Grammar 1 is therefore given by the addition
of Project-.
(13)
Grammar 1
(w/ Project- )
Grammar 0
The full pattern of stage-like growth is shown in the chart below:
(14) A: Subgrammar Approach
Add construction operations Relative clauses,
to simplified tree Conjunction (not discussed here)
Add primitives to Theta Theta + Case
representational vocabulary
As can be seen, the acquisition sequence and the syntax syntactic derivation are tightly yoked.
Another way of putting the arguments above is in terms of distinguishing
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
22/308
PREFACE xxi
accounts. I wish to distinguish the phrase structure operations here from Merge;
and the acquisition subgrammar approach here from the alternative, which is theFull Tree, or Full Competence, Approach (the full tree approach holds that the
child does not start out with a substructure, but rather has the full tree, at all
stages of development.) Let us see how the accounts are distinguished, in turn.
Let us start with Chomskys Merge. According to Merge, the (adult) phrase
structure tree, as in Montague (1974), is built up bottom-up, taking individual
units and joining them together, and so on. The chief property of Merge is that
it isstrictlybottom-up. Thus, for example, in a right-branching structure like see
the big man, Merge would first take big and man and Merge them together,
then add the to big man, and then add see to the resultant.
(15) Application of Merge:
V Det Adj N
see the big man
N
Adj
big
N
man
DP
Det NP
Adj N
man
bigthe
V
see
VP
DP
Det NP
the Adj N
manbig
The proposal assayed in this thesis (Lebeaux 1988) would, however, have a
radically different derivation. It would take the basic structure as being the basic
government relation: (see man). This is the primitive unit (unlike with Merge).
To this, thethe and thebig may be added, by separate transformations, Project-
and Adjoin-, respectively.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
23/308
xxii PREFACE
(16)
V
N
man
man
V
see
V
V
V
V
DP
DP
NP
NP
Det
Det
e
(see)
(see)Case Frame
Project-
Theta subtree
the
the
a. Project-
( see ( the man))
( big)V DP
ADJ
b. Adjoin-
Adjoin- ( see ( the ( big man)))V DP NP
How can these radically distinct accounts (Lebeaux 1988 and Merge) be
empirically distinguished? I would suggest in two ways. First, conceptually the
proposal here (as in Chomsky 19751955, 1957, and Tree Adjoining Grammars,
Kroch and Joshi 1985) takes information nuclei as its input structures, notarbitrary pieces of string. For example, for the structure The man saw the
photograph that was taken by Stieglitz, the representation here would take the
two clausal nuclear structures, shown in (17) below, and adjoin them. This is not
true for Merge which does not deal in nuclear units.
(17) s1:
s2:
the man saw the photograph
that was by Stieglitz
the man saw the photograph
that was by Stieglitz
Adjoin-
Even more interesting nuclear units are implicated in the transformationProject-, where the full sentence is decomposed into a nuclear unit which is the
theta subtree, and the Case Frame.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
24/308
PREFACE xxiii
(18)
The man saw the woman
(man (see woman))
(the _(see a_))
The structure in (18), the man saw the woman, is composed of a basic nuclear
unit, (man (see woman)), which is telegraphic speech (as argued for in Chap-
ter 2). No such nuclear unit exists in the Merge derivation of the man saw the
woman: that is, in the Merge derivation, (man (see woman)) does not exist as
a substructure of ((the man) (saw (the woman)).
This is the conceptual argument for preferring the composition operation
here over Merge. In addition, there are two simplicity arguments, of which I willgive just one here.
The simplicity argument has to do with a set of structures that children
produce which are calledreplacement sequences(Braine 1976). In these sequenc-
es, the child is trying to reach (output) some structure which is somewhat too
difficult for him/her. To make it, therefore, he or she first outputs a substructure,
and then the whole structure. Examples are given below: the first line is the first
outputted structure, and the second line is the second outputted structure, as the
child attempts to reach the target (which is the second line).
(19) see ball (first output)
see big ball (second output and target)
(20) see ball (first output)
see the ball (second output and target)
What is striking about these replacement sequences is that the child does not
simply first output random substrings of the final target, but rather that the first
output is an organized part of the second. Thus in both (19) and (20), what the
child has done is first isolate out the basic government relation, (see ball), andthen added to it: with big and the, respectively.
The particular simplifications chosen are precisely what we would expect with
the substructure approach outlined here, and crucially not with Merge. With the
substructure approach outlined here (Chapter 2, 4), what the child (or adult) first
has in the derivation is precisely the structure (see ball), shown in example (21).
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
25/308
xxiv PREFACE
(21) V
V
see
N
+patient
ball
To this structure is then added other elements, by Project- or Adjoin-. Thus,
crucially, the first structure in (19) and (20) actually exists as a literal substruc-
ture of the final form line 2 and thus could help the child in deriving the
final form. It literally goes into the derivation.By contrast, with Merge, the first line in (19) and (20) never underlies the
second line. It is easy to see why. Merge is simply bottom-up it extends the
phrase marker. Therefore, the phrase structure composition derivation underlying
(20) line 2, is simply the following (Merge derivation).
(22) Merge derivation underlying (20) line 2
(N ball)
(DP (D the) (N ball))
(see (DP (D the) (N ball)))
However, this derivation crucially does not have the first line of (20) (see (ball))
as a subcomponent. That is, (see (ball)) does not go into the making of (see
(the ball)), in the Merge derivation, but it does in the substructure derivation.
But this is a strong argument against Merge. For the first line of the
outputted sequence of (20), (see ball), is presumably helping the child in
reaching the ultimate target (see (the ball)). But this is impossible with Merge,
for the first line in (20) does not go into the making of the second line, accord-
ing to the Merge derivation.That is, Merge cannot explain why (see ball) would help the child get to the
target (see (the ball)), since (see ball) is not part of the derivation of (see (the
ball)), in the Merge derivation. It is part of the sub-derivation in the substructure
approach outlined here, because of the operation Project-.
The above (see Chapters 2, 3, and 4) differentiates the sort of phrase
structure composition operations found here from Merge. This is in the domain
of syntax though I have used language acquisition argumentation. In the
domain of language acquisition proper, the proposal of this thesis the
hypothesis of substructures must be contrasted with the alternative, whichholds that the child is outputting the full tree, even when the child is potentially
just in the one word stage: this may be called the Full Tree Hypothesis. These
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
26/308
PREFACE xxv
differential possibilities are shown below. (For much additional discussion, see
Lebeaux 1991, 1997, 1998, in preparation.)
(23) Lebeaux (1988) Distinguished From
Syntax phrase structure com-
position
Both:
(1) no composition
(2) Merge
Language Acquisition subgrammar approach Full Tree Approach
Let us now briefly distinguish the proposals here from the Full Tree Approach.In the Full Tree Approach, the structure underlying a child sentence like ball
or see ball might be the following in (24). In contrast, the substructure
approach (Lebeaux, 1988) would assign the radically different representation,
given in (25).
(24) Full Tree Approach
IP
TP
AgrSP
AgrOP
VP
V
DP
NPD
V
DP
AgrO
AgrS
T
DP
D NP
balleeeeeeee
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
27/308
xxvi PREFACE
(25)
V
V N
+patient
ball
Substructure Approach
How can these approaches be distinguished? That is, how can a choice be made
between (25), the substructure approach, and (24), the Full Tree approach? I
would suggest briefly at least four ways (to see full argumentation, consultLebeaux 1997, to appear; Powers and Lebeaux 1998).
First, the subgrammar approach, but not the full tree approach, has some
notion ofsimplicity in representation and derivation. Simplicity is a much used
notion in science, for example deciding between two equally empirically
adequate theories. The Full Tree Approach has no notion of simplicity: in
particular, it has no idea of how the child would proceed from simpler structures
to more complex ones. On the other hand, the substructure theory has a strong
proposal to make: the child proceeds from simpler structures over time to those
which are more complex. Thus the subgrammar point of view makes a strong
proposal linked to simplicity, while the Full Tree hypothesis makes none.
A second argument has to do with the closed class elements, and may be
broken up into two subarguments. The first of these arguments is that, in the Full
Tree Approach, there is no principledreason for the exclusion of closed class
elements in early speech (telegraphic speech). That is, both the open class and
closed class nodes exist, according to the Full Tree Hypothesis, and there is no
principled reason why initial speech would simply be open class, as it is. That is,
given the Full Tree Hypothesis, since the full tree is present, lexical insertioncould take place just as easily in the closed class nodes as the open class nodes.
The fact that it doesnt leaves the Full Tree approach with no principled reason
why closed class items are lacking in early speech.
A second reason having to do with closed class items, has to do with the
special role that they have in structuring an utterance, as shown by the work of
Garrett (1975, 1980), and Gleitman (1990). Since the Full Tree Approach gives
open and closed class items the same status, it has no explanation for why closed
class items play a special role in processing and acquisition. The substructure
approach, with Project-, on the other hand, faithfully models the difference, byhaving open class and closed class elements initially on different representations,
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
28/308
PREFACE xxvii
which are then fused (for additional discussion, see Chapter 4, and Lebeaux
1991, 1997, to appear).A third argument against the Full Tree Approach has to do with structures
like see ball (natural) vs. see big (unnatural) given below.
(26) see ball (natural and common)
see big (unnatural and uncommon)
Why would an utterance like see ball be natural and common for the child
maintaining the government relation while see big is unnatural and uncom-
mon? There is a common sense explanation for this: see ball maintains the
government relation (between a verb and a complement), while see and bighave no natural relation. While this fact is obvious, it cannot be accounted for
with the Full Tree Approach. The reason is that the Full Tree Approach has all
nodes potentially available for use: including the adjectival ones. Thus there
would be no constraint on lexically inserting see and big (rather than see
and ball). On the substructure approach, on the other hand, there is a marked
difference: see and ball are on a single primitive substructure the theta
tree while see and big are not.
A fourth argument against the Full Tree Approach and for the substructure
approach comes from a paper by Laporte-Grimes and Lebeaux (1993). In this
paper, the authors show that the acquisition sequence proceeds almost sequential-
ly in terms of the geometric complexity of the phrase marker. This is, children
first output binary branching structures, then double binary branching, then triply
binary branching, and so on. This complexity result would be unexpected with
the Full Tree Approach, where the full tree is always available.
This concludes the four arguments against the Full Tree Approach, and for
the substructure approach in acquisition. The substructure approach (in acquisi-
tion) and the composition of the phrase marker (in syntax) form the two main
proposals of this thesis.
Aside from the main lines of argumentation, which I have just given, there
are a number of other proposals in this thesis. I just list them here.
(1) One main proposal which I take up in all of Chapter 5 is that the acquisition
sequence is built up from derivational endpoints. In particular, for some purpos-
es, the childs derivation is anchored in the surface, and only goes part of the
way back to DS. The main example of this can be seen with dislocated constitu-
ents. In examples like (27a) and (b), exemplifying Strong Crossover and a
Condition C violation respectively, the adult would not allow these constructions,
while the child does.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
29/308
xxviii PREFACE
(27) a. *Which mani did hei see t? (OK for child)
b. *In Johnsi house, hei put a book t. (OK for child)It cannot be simply said, as in (27b), that Condition C does not apply in the
childs grammar, because it does, in nondislocated structures (Carden 1986b).
The solution to this puzzle and there exist a large number of similar puzzles
in the acquisition literature, see Chapter 5is that Condition C in general applies
over direct c-command relations, including at D-Structure (Lebeaux 1988, 1991,
1998), and that the child analyzes structures like (27b) as if they were dislocated
at all levels of representation, thus never triggering Condition C (a similar
analysis holds of Strong Crossover, construed as a Condition C type constraint,
at DS, van Riemsdijk and Williams 1981). That is, the child derivation, unlike
the adult, does not have movement, but starts out with the element in a dislocat-
ed position, and indexes it to the trace. This explains the lack of Condition C and
Crossover constraints (shown in Chapter 5). It does so by saying that the childs
derivation is shallow: anchored at SS or the surface, and the dislocated item is
never treated as if it were fully back in the DS position.
This is the shallowness of the derivation, anchored in SS (discussed in
Chapter 5).
(2) A number of proposals are made in Chapter 2. One main proposal concernsthe theta tree. In order to construct the tree, one takes a lexical entry, and does
lexical insertion of open class items directly into that. This is shown in (28).
(28) V
N V
V
see
N
patientwoman
man
This means that the sequence between the lexicon and the syntax is in fact a
continuum: the theta subtree constitutes an intermediate structure between those
usually thought to be in the lexicon, and those in the syntax. This is a radical
proposal.
A second proposal made in Chapter 2 is that Xprojections project up as far
as they need to. Thus if one assumed the X-theory of Jackendoff (1977) (as I
did in this thesis) recall that Jackendoff had 3 X levels then an elementmight project up to the single bar level, double bar level, or all the way up to the
triple bar level, as needed.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
30/308
PREFACE xxix
(29) N
N
N
N
This was called the hypothesis of submaximal projections.
A final proposal of Chapter 2 is that the English nominal system is ergative.
That is, a simple intransitive noun phrase like that in (29), with the subject in thesubject position (of the noun phrase) is always derived from a DS in which the
subject is a DS object. Crucially, this includes not simply unaccusative verbs (i.e.
nominals from unaccusative verbs) but unergative verbs as well (such as sleeping
and swimming).
(30) a. Johns sleeping
derived from: the sleeping of John (subject internal)
b. Johns swimming
derived from: the swimming of John (subject internal)This means that the English nominal system is actually ergative in character
a startling result.
Some final editorial comments. For space reasons in this series, Chapter 5 in the
original thesis has been deleted, and Chapter 6 has been re-numbered Chapter 5.
Second, I have maintained the phrase structure nodes of the original trees, rather
than trying to update them with the more recent nodes. The current IP is
therefore generally labelled S (sentence), the current DP is generally labelled NP
(noun phrase), and the current CP is sometimes labelled S (S-bar, the old namefor CP). Finally, the term dislocation in Chapter 5 is intended to be neutral by
itself between moved and base-generated. The argument of that section is that
wh-elements which are moved by the adult, are base generated in dislocated
positions by the child. Finally, I would like to thank Lisa Cheng and Anke de
Looper for helpful editorial assistance.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
31/308
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
32/308
Introduction
This work arose out of an attempt to answer three questions:
I. Is there a way in which the Government-Binding theory of Chomsky (1981)
can be formulated so that the leveling in it is more essential than in thecurrent version of the theory?
II. What is the relation between the sequence of grammars that the child adopts,
and the basic formation of the grammar, and is there such a relation?
III. Is there a way to anchor Chomskys (1981) finiteness claim that the set of
possible human grammars is finite, so that it becomes a central explanatory
factor in the grammar itself?
The work attempts to accomplish the following:
I. To provide for an essentially leveled theory, in two ways: by showing that
DS and SS are clearly demarcated by positing operations additional to
Move- which relate them, and by suggesting that there is a ordering in
addition by vocabulary, the vocabulary of description (in particular, Case
and theta theory) accumulating over the derivation.
II. To relate this syntactically argued for leveling to the acquisition theory,
again in two ways: by arguing that the external levels (DS, the Surface, PF)
may precede S-structure with respect to the induction of structure, and by
positing a general principle, the General Congruence Principle, which relatesacquisition stages and syntactic levels.
III. To give the closed class elements a crucial role to play: with respect to
parametric variation, they are the locus of the specification of parametric
difference, and with respect to the composition of the phrase marker: it is
the need for closed class (CC) elements to be satisfied which gives rise to
phrase marker composition from more primitive units, and which initiates
Move- as well.
In terms of syntactic content, Chapters 24 deal with phrase structure both the
acquisition and the syntactic analysis thereof and Chapter 5 deals with the
interaction of indexing functions, Control and Binding Theory, with levels of
representation, particularly as it is displayed in the acquisition sequence.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
33/308
2 INTRODUCTION
Thematically, a number of concerns emerge throughout. A major concern is
with closed class elements and finiteness. With respect to parametric variation,I suggest that closed class elements are the locus of parametric variation. This
guarantees finiteness of possible grammars in UG, since the set of possible
closed class elements is finite.1 With respect to phrase structure composition, it
is the closed class elements, and the necessity for their satisfaction, which require
the phrase marker to be composed, and initiate movement as well (e.g. Move-wh
is in a 1-to-1 correspondence with the lexical necessity: Satisfy +wh feature).
The phrase marker composition has some relation to the traditional generalized
transformations of Chomsky (1957), and they may apply (in the case of
Adjoin-) after movement. But the composition that occurs is of a strictlylimited sort, where the units are demarcated according to the principles of GB.
Finally, closed class elements form a fixed frame into which the open class (OC)
elements are projected (Chapters 1, 2, and 4). More exactly, they form a Case
frame into which a theta sub-tree is projected (Chapter 4). This rule, I call
Merger (or Project-).
A second theme is the relation of stages in acquisition to levels of grammat-
ical representation. Since the apparent difficulty of any theory which involves
the learning of transformations,2 the precise nature of the relation of the acquisi-
tion sequence to the structure of the grammar has remained murky, without a
theory of how the grammatical acquisition sequence interacts with, or displays
the structure of the grammar, and with, perhaps, many theoreticians believing
that any correspondence is otiose. Yet there is considerable reason to believe that
there should be such a correspondence. On theoretical grounds, this would be
expected for the following reason: The child in his/her induction of the grammar
is not handed information from all levels in the grammar at once, but rather from
particular picked out levels; the external levels of Chomsky (class lectures, 1985)
DS, LF, and PF or the surface.These are contrasted to the internal level, S-structure. Briefly, information
from the external levels are available to the child; about LF because of the paired
meaning interpretation, from the surface in the obvious fashion, and from DS,
construed here simply as the format of lexical forms, which are presumably
given by UG. As such, the childs task (still!) involves the interpolation of
operations and levels between these relatively fixed points. But, this then means
1. Modulo the comments in Chapter 1, footnote 1.
2. Because individual transformations are no longer sanctioned in the grammar. I do not believe,
however, that the jury is yet in on the type of theory that Wexler and Culicover (1980) envisage.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
34/308
INTRODUCTION 3
that the acquisition sequence must build on these external levels, and display the
structure of the levels, perhaps in a complex fashion.A numerical argument leads in the same direction: namely, that the acquisi-
tion theory, in addition to being a parametric theory, should contain some
essential reference to, and reflect, the structure of the grammar. Suppose that, as
above, the closed class elements and their values are identified with the possible
parameters. Let us (somewhat fancifully) set the number at 25, and assume that
they are binary. This would then give 225 target grammars in UG (=30 million),
a really quite small finite system. But, consider the range of acquisition sequenc-
es involved. If parameters are independent a common assumption then any
of these 25 parameters could be set first, then any of the remaining 24, and soon. This gives 25! possible acquisition sequences for the learning of a single
language (=1.51025), a truly gigantic number. That is, the range of acquisition
sequences would be much larger than the range of possible grammars, and
children might be expected to display widely divergent intermediate grammars in
their path to the final common target, given independence. Yet they do nothing
of the sort; acquisition sequences in a given language look remarkably similar.
All children pass through a stage of telegraphic speech, and similar sorts of
errors are made in structures of complementation, in the acquisition of Control,
and so on. There is no wide fecundity in the display of intermediate grammars.
The way that has been broached in the acquisition literature to handle this
has been the so-called linking of parameters, where the setting of a single
parameter leads to another being set. This could restrict the range of acquisition
sequences. But, the theories embodying this idea have tended to have a rather
idiosyncratic and fragmentary character, and have not been numerous. The
suggestion in this work is that there is substructuring, but this is not in the
lexical-parametric domain itself (conceived of as the set of values for the closed
class (CC) elements), but in the operational domain with which this lexicaldomain is associated. An example of this association was given above with the
relation of the wh-movement to the satisfaction of the +wh feature; another
example would be with satisfaction of the relative clause linker (the wh-element
itself), which either needs or does not need to be satisfied in the syntax. This
gives rise to either language in which the relative forms a constituent with the
head (English-type languages), or languages in which it is splayed out after the
main proposition, correlative languages.(1) Lexical Domain
+wh must be satisfied by SS+wh may not be satisfied by SS
Operational Domain
Move-wh applies in syntaxMove-wh applies at LF
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
35/308
4 INTRODUCTION
Lexical Domain
Relative Clause linker must be
satisfied by SS
Relative Clause linker may not
be satisfied by SS
Operational Domain
English-type language
Correlative language
The theory of this work suggests that all operations are dually specified in the
lexical domain (requiring satisfaction of a CC lexical element) and in the
operational domain.
The acquisition sequence reflects the structure of the grammar in two ways:
via the General Congruence Principle, which states that the stages in acquisition
are in a congruence relation with the structure of parameters (see Chapter 3 for
discussion), and via the use of the external levels (DS, PF, LF) as anchoring
levels for the analysis essentially, as the inductive basis. The General
Congruence Principle is discussed in Chapter 24, the possibility of broader
anchoring levels, in Chapter 5. The latter point of view is somewhat distinct
from the former, and (to be frank) the exact relation between them is not yet
clear to the author. It may be that the General Congruence Principle is a special
case, when the anchoring level is DS, or it may be that these are autonomousprinciples. I leave this question open.
The third theme of this work has to do with levels or precedence relations
in the grammar. In particular, with respect to two issues: (a) Is it possible to
make an argument that the grammar is essentially derivational in character, rather
than in the representational mode (cf. Chomskys 1981 discussion of Move-)?
(b) Is there any evidence of intermediate levels, of the sort postulated in van
Riemsdijk and Williams (1981)? I believe that considering a wider range of
operations than Move- may move this debate forward. In particular, I propose
two additional operations of phrase structure composition: Adjoin-, whichadjoins adjuncts in the course of the derivation, and Project-, which relates the
lexical syntax to the phrasal. With respect to these operations, two types of
precedence relations do seem to hold. First, operation/default organization holds
within an operation type. In the case of Adjoin- and its corresponding default,
Conjoin- (i.e., two of the types of generalized transformations in Chomsky
1957, are organized as a single operation type, with an operation/default relation
between them). The other precedence relation is vocabulary layering and this
hold between different operations, for example, Case and theta theory (see
Chapter 2, 3, and 4 for discussion). Further, operations like Adjoin-may follow
Move-, and this explains the anti-Reconstruction facts of van Riemsdijk and
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
36/308
INTRODUCTION 5
Williams (1981); such facts cannot be easily explained in the representational
mode (see Chapter 3).In general, throughout this work I will interleave acquisition data and theory
with pure syntactic theory, since I do not really differentiate between them.
Thus, the proposal having to do with Adjoin- was motivated by pure syntactic
concerns (the anti-Reconstruction facts, and the attempt to get a simple descrip-
tion of licensing), but was then carried over into the acquisition sphere. The
proposal having to do with the operation of Project-(or Merger) was formulat-
ed first in order to give a succinct account of telegraphic speech (and, to a lesser
degree, to account for speech error data), and was then carried over into the
syntactic domain. To the extent to which this type of work is successful, the twoareas, pure syntactic theory and acquisition theory may be brought much closer,
perhaps identified.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
37/308
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
38/308
C 1
A Re-Definition of the Problem
1.1 The Pivot/Open Distinction and the Government Relation
For many years language acquisition research has been a sort of weak sister in
grammatical research. The reason for this, I believe, lies not so much in its own
intrinsic weakness (for a theoretical tour de force, see Wexler and Culicover
1980, see also Pinker 1984), but rather, as in other unequal sibships, in relation.
This relation has not been a close one; moreover the lionizing of the theoretical
importance of language acquisition as the conceptual ground of linguistic
theorizing has existed in uneasy conscience alongside a real practical lack of
interest. Nor is the fault purely on the side of theoretical linguistics: the acquisi-tion literature, especially on the psychological side, is notorious for having
drifted further and further from the original goal of explaining acquisition, i.e.
the sequence of mappings which take the child from G0to the terminal grammar
Gn, to the study of a different sort of creature altogether, Child Language (see
Pinker 1984, for discussion and a diagnostic).
1.1.1 Braines Distinction
Nonetheless, even in the psychological literature, especially early on, there werea number of proposals of quite far-reaching importance which would, or could,
have (had) a direct bearing on linguistic theory, and which pointed the way to
theories far more advanced than those available at the time. For example,
Braines (1963a) postulation of pivot-open structures in early grammars. Braine
essentially noticed and isolated three properties of early speech: for a large
number of children, the vocabulary divided into two classes, which he called
pivot and open. The pivot class was closed class, partly in the sense that it
applies in the adult grammar (e.g., containing prepositions, pronouns, etc.) but
partly also in the broader sense: it was a class that contained a small set of
words which couldnt be added on to, even though these words corresponded to
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
39/308
8 LANGUAGE ACQUISITION AND THE FORM OF THE GRAMMAR
those which would ordinarily be thought of as open class (e.g. come); these
words operated on a comparatively large number of open class elements. Anexample of the Braine data is given below.
(1) Stevens word combinations
want baby see record
want car see Stevie
want do
want get whoa cards
want glasses whoa jeep
want head
want high more ball
want horsie more book
want jeep
want more there ball
want page there book
want pon there doggie
want purse there doll
want ride there high
want up there mommawant byebye car there record
there trunk
it ball there byebye car
it bang there daddy truck
it checker there momma truck
it daddy
it Dennis that box
it X etc. that Dennis
that X etc.get ball
get Betty here bed
get doll here checker
here doll
see ball here truck
see doll
bunny do
daddy do
momma do
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
40/308
A RE-DEFINITION OF THE PROBLEM 9
The second property of the pivot/open distinction noticed by Braine was that
pivot and open are positional classes, occurring in a specified position withrespect to each other, though the positional ordering was specific to the pivot
element itself (P1 Open, Open P2, etc.) and hence not to be captured by a
general phrase structure rewrite rule: S Pivot Open. This latter fact was used
by critical studies of the time (Fodor, Bever, and Garrett 1974, for example) to
argue that Braines distinction was somehow incoherent, since the one means of
capturing such a distinction, phrase structure rules, required a general collapse
across elements in the pivot class which was simply not available in the data.
The third property of the pivot/open distinction was that the open class
elements were generally optional, while the pivot elements were not.
1.1.2 The Government Relation
What is interesting from the perspective of current theory is just how closely
Braine managed to isolate analogs not to the phrase structure rule descriptions
popular at that time, but to the central relation primitives of the current theory.
Thus the relation of pivot to open classes may be thought of as that between
governor and governed element, or perhaps more generally that of head to
complement; something like a primitive prediction or small clause structure (in
the extended sense of Kayne 1984) appears to be in evidence in these early
structures as well:
(2) Steven word utterances:
it ball that box
it bang that Dennis
it checker that doll
it X, etc. that Tommy
that truckthere ball here bed
there book here checker
there doggie here doll
there X, etc. here X, etc.
Andrew word combinations:
boot off airplane all gone
light off Calico all gone
pants off Calico all done
shirt off salt all shut
shoe off all done milk
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
41/308
10 LANGUAGE ACQUISITION AND THE FORM OF THE GRAMMAR
water off all done now
all gone juiceclock on there all gone outside
up on there all gone pacifier
hot in there
X in/on there, etc.
Gregory word combinations:
byebye plane allgone shoe
byebye man allgone vitamins
byebye hot allgone eggallgone lettuce
allgone watch
etc.
The third property that Braine notes, the optionality of the open constituent with
respect to the pivot, may also be regularized to current theory: it is simply the
idea that heads are generally obligatory while complements are not.
The idea that the child, very early on, is trying to determine the general
properties of the government relation in the language (remaining neutral for nowabout whether this is case or theta government) is supported by two other facts
as well: the presence of what Braine calls groping patterns in the early data,
and the presence of what he calls formulas of limited scope. The former can be
seen in the presence of the allgone constructions in Andrews speech. The
latter refers simply to the fact that in the very early two-word grammars, the set
of relations possible between the two words appears limited in terms of the
semantic relation which hold between them. This may be thought of as showing
that the initial government relation is learned with respect to specific lexical
items, or cognitively specified subclasses, and is then collapsed between them.See also later discussion. The presence of groping patterns, i.e. the presence,
in two word utterances of patterns in which the order of elements is not fixed for
lexically specific elements corresponds to the original experimentation in
determining the directionality of government (Chomsky 1981, Stowell 1981). The
presence of groping patterns is problematic for any theory of grammar which
gives a prominent role to phrase structure rules in early speech, since the order
of elements must be fixed for all elements in a class. See, e.g., the discussion in
Pinker (1984), which attempts, unsuccessfully I believe, to naturalize this set of
data. To the extent to which phrase structure order is considered to be a deriva-
tive notion, and the government-of relation the primitive one, the presence of
lexically specific order difference is not particularly problematic, as long as the
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
42/308
A RE-DEFINITION OF THE PROBLEM 11
directionality of government is assumed to be determined at first on a word-by-
word basis.
1.2 The Open/Closed Class Distinction
Braines prescient analysis was attacked in the psychological literature on both
empirical and especially theoretical grounds; it was ignored in the linguistic
literature. The basis of the theoretical attack was that the pivot/open distinction,
being lexically specific with respect to distribution, would not be accommodated
in a general theory of phrase structure rules (as already mentioned above);moreover, the particular form of the theory adopted by Braine posited a radical
discontinuity in the form of the grammar as it changed from a pivot/open
grammar to a standard Aspects-style PS grammar. This latter charge we may
partly diffuse by noting that there is no need to suppose a radical discontinuity
in the form of the grammar as it changed over time, the pivot/open grammar is
simply contained as a subgrammar in all the later stages. However, we wish to
remain neutral, for now, on the general issue of whether such radical discontinu-
ities are possible. The proponents of such a view, especially the holders of the
view that the original grammar was essentially semantic (i.e. thematically
organized), held the view in either a more or less radical form. The more
extreme advocates (Schlesinger 1971) held not simply that there was a radical
discontinuity, but that the primitives of later stages syntactic primitives like
case and syntactic categories like noun or noun phrase were constructedout
of the primitives of the earlier stages: a position one may emphatically reject.
Other theoreticians, however, in particular Melissa Bowerman (Bowerman 1973,
1974) held that there was such a discontinuity, but without supposing any
construction of the primitives of the later stages from those of the earlier. Wereturn, in detail, to this possibility below.
More generally, however, the charge that the pivot/open class stage presents
a problem for grammatical description appears to dissolve once the government-
of relation is taken to be the primitive, rather than the learning of a collection of
(internally coherent) phrase structure rules.
However, more still needs to be said about Braines data. For it is not
simply the case that a rudimentary government relation is being established, but
that this is overlaid, in a mysterious way, with the open/closed class distinction.
Thus it is not simply that the child is determining the government-of andpredicate of relations in his or her language, but also that the class of governing
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
43/308
12 LANGUAGE ACQUISITION AND THE FORM OF THE GRAMMAR
elements is, in some peculiar way, associated with a distributional class: namely,
that of closed class elements.While the central place of the government-of relation in current theory gives
us insight into one-half of Braines data, the role of the closed class/open class
distinction, though absolutely pervasive in both Braines work and in all the
psycholinguistic literature (see Garrett 1975, Shattuck-Hufnagel 1974, Bradley
1979, for a small sample) has remained totally untouched. Indeed, even the
semantic literature, which has in general paid much more attention to the
specifier relation than transformational-generative linguistics, does not appear to
have anything to say that would account for the acquisition facts.
What could we say about the initial overlay of the elements closed class andthe set of governors? The minimal assumption would be something like this:
(3) The set of canonical governors is closed class.
While this is an interesting possibility, it would involve, for example, including
prepositions and auxiliary verbs in the class of canonical governors, but not main
verbs. Suppose that we strengthen (3), nonetheless.
(4) Only closed class elements may govern.
What about verbs? Interestingly, a solution already exists in the literature: in fact,two of them. Stowell (1981) suggests that it is not the verb per se which governs
its complements, but rather the theta grid associated with it. Thus the comple-
ments are theta governed under coindexing with positions in the theta grid. And
while the class of verbs in a language is clearly open class and potentially
infinite, the class of theta grids is equally clearly finite: a member of a closed,
finite set of elements. Along the same lines, Koopman (1984) makes the
interesting, though at first glance odd, suggestion that it is not the verb which
Case-governs its complements, but Case-assigning features associated with the
verb. She does this in the context of a discussion of Stowells Case adjacency
requirement for case assignment; a proposal which appears to be immediately
falsified by the existence of Dutch, a language in which the verb is VP final, but
the accusative marked object is at the left periphery of the VP. Koopman saves
Stowells proposal by supposing that the Case-assigning features of the verb are
at the left periphery, though the verb itself is at the right. This idea that the two
aspects of the verb are separable in this fashion will be returned to, and support-
ed, below. What is crucial for present purposes is simply to note that Case-
governing properties of the verb are themselves closed class, though the set ofverbs is not. Thus both the Case-assigning and theta-assigning properties of the
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
44/308
A RE-DEFINITION OF THE PROBLEM 13
verb are closed class, and we may assume that these, rather than some property
of the open class itself enters into the government relation.There is a second possibility, less theory-dependant. This is simply that, as
has often been noted, there is within the open part of the vocabulary of
language a subset which is potentially closed: this is the so-called basic vocabu-
lary of the language, used in the teaching of basic English, and other languages.
The verb say would presumably be part of this closed subset, but not the verb
mutter, as would their translations. The child task may be viewed as centering on
the closed class elements in the less abstract sense of lexical items, if these are
included in the set.
1.2.1 Finiteness
While the syntactic conjecture that the Case features on the verb are governing
its object has been often enough made, the theoretical potential of such a
proposal has not been realized. In essence, this proposal reduces a property of an
open class of elements, namely verbs, to a property of a closed class of elements
(the Case features on verbs). Insofar as direction of government is treated as a
parameter of variation across languages, by reducing government directionality
to a property of a closed class set, the two sorts of finiteness, lexical and
syntactic, are joined together. The finiteness of syntactic variation (Chomsky
1981) is tied, in the closest possible way, to the necessary finiteness of a lexical
class (and the specifications associated with it).
Let us take another example. English allows wh-movement in the syntax;
Chinese, apparently, apportions it into LF (Huang 1982). This is a parametric
difference in the level of derivation at which a particular operation applies.
However, this may well be reducible to a parametric difference in a closed class
element. Let us suppose, following Chomsky (1986), that wh-movement ismovement into the specifier position of C.
Ordinarily it is assumed that lexical selection (of the complement-taking
verb) is of the head. Let us assume likewise the matrix verb must select for
a +/ wh feature in Comp. This, in turn, must regulate the possible appearance
of the wh-word appearing in the specifier position of C. We may assume that
some agreement relation holds between these two positions, in direct analog to
the agreement relation which exists generally between specifier and head
positions, e.g. with respect to case. Thus the presence of the overt wh-element in
Spec C is necessary to agree with, or saturate the +wh feature which is base-generatedin Comp. What then is the difference between English and Chinese? Just
this: the agreeing element in Comp must be satisfied at S-structure in English,
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
45/308
14 LANGUAGE ACQUISITION AND THE FORM OF THE GRAMMAR
while it needs only be satisfied at LF in Chinese. This difference, in turn, may
be traced to some intrinsic property of agreement in the two languages, we mighthope.
(5) wonder C
C
I
I
VP
NP
e
V
saw
I
NP
John
Comp
who
If this sketch of an analysis is correct or something like it is then the
parametric difference between English and Chinese with respect to wh-move-ment is reduced to a difference in the lexical specification of a closed class
element.1 Since the possible set of universal specifications associated with a
closed class set of elements is of necessity finite, the finiteness conjecture of
Chomsky (1981) would be vindicated in the strongest possible way. Namely, the
finiteness in parametric variation would be tied, and perhaps only tied, to the
finiteness of a group of necessarily finite lexical elements, and the information
associated with them.
1.2.2 The Question of Levels
There is a different aspect of this which requires note. The difference between
Chinese and English with respect to wh-movement is perhaps associated with
features on the closed class morpheme, but this shows up as a difference in the
appearance of the structure at a representational level. I believe that this is in
1. I should note that the term closed class elementhere is being used in a somewhat broader sensethan usual, to encompass elements like the +whfeature. The finiteness in the closed class set cannot
be that of the actual lexical items themselves, since these may vary from language to language, but
in the schema which defines them (e.g. definite determiner, indefinite determiner, Infl, etc.).
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
46/308
A RE-DEFINITION OF THE PROBLEM 15
general the case: namely, that while information associated with a closed class
element is at the root of some aspect of parametric variation, this differenceoften evidences itself in the grammar by a difference in the representational level at
which a particular operation applies. We may put this in the form of a proposal:
(6) The theory of UG is the theory of the parametric variation in the
specifications of closed class elements, filtered through a theory of
levels.
I will return throughout this work to more specific ways in which the conjecture
in (6) may be fleshed out, but I would like to return at this point to two aspects
which seem relevant. First is the observation made repeatedly by Chomsky(1981, 1986a), that while the set of possible human languages is (at least
conjecturally) finite, they appear to have a wide scatter in terms of surface
features. Why, we might ask, should this be the case? If the above conjecture (6)
is correct, it is precisely because of the interaction of the finite set of specifica-
tions associated with the closed class elements, and the rather huge surface
differences which would follow from having different operations apply at
different levels. The information associated with the former would determine the
latter; the latter would give rise to the apparent huge differences in the descrip-
tion of the worlds languages, but would itself be tied to a parametric variation
in a small, necessarily finite set.
How does language acquisition proceed under these circumstances? Briefly,
it must proceed in two ways: by determining the properties of lexical specifica-
tions associated with the closed class set the child determines the structure of the
levels; by determining the structure of the levels he or she determines the
properties of the closed class morphemes. The proposal that the discovery of
properties associated with closed class lexical items is central obviously owes a
lot to Borers (1985) lexical learning hypothesis, that what the child learns, and
all that he/she learns is associated with properties of lexical elements. It consti-
tutes, in fact, a (fairly radical) strengthening of that proposal, in the direction of
finiteness. Thus while the original lexical learning hypothesis would not guaran-
tee finiteness in parametric variation, the version adopted in (6) would, and thus
may be viewed as providing a particular sort of grounding for Chomskys
finiteness claim.
However, the proposal in (6) contains an additional claim as well: that the
difference in the specifications of closed class elements cashes in as a difference
in the level that various operations apply. Thus it provides an outline of the waythat the gross scatter of languages may be associated with a finite range.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
47/308
16 LANGUAGE ACQUISITION AND THE FORM OF THE GRAMMAR
1.3 Triggers
1.3.1 A Constraint
The theory of parametric variation or grammatical determination has often been
linked with a different theory: that oftriggers (Roeper 1978b, 1982, Roeper and
Williams 1986). A trigger may be thought of, in the most general case, as a
piece of information, on the surface string, which allows the child to determine
some aspect of grammatical realization. The idea is an attractive one, in that it
suggests a direct connection between a piece of surface data and the underlying
projected grammar; it is also in danger, if left further undefined, of becomingnearly vacuous as a means of grammatical description. A trigger, as it is
commonly used, may apply to virtually any property of the surface string which
allows the child to make some determination about his or her grammar.
There is, as is usual in linguistic theory, a way to make an idea more
theoretically valuable: that is, by constraining it. This constraint may be either
right or wrong, but it should, in either case, sharpen the theoretical issues involved.
In line with the discussion earlier in the chapter, let us limit the content of
trigger in the following way:
(7) A trigger is a determination in the property of a closed class element.
Given the previous discussion, the differences in the look of the output grammar
may be large, given that a trigger has been set. The trigger-setting itself, however, is
aligned with the setting of the specification of a closed class element.
There are a number of instances of triggers in the input which must be re-
examined given (7) above, there are however, at least two very good instances
of triggers in the above sense which have been proposed in the literature. The
first is Hyams (1985, 1986, 1987) analysis of the early dropping of subjects inEnglish. Hyams suggests that children start off with a grammar which is
essentially pro-drop, and that English-speaking children then move to an English-
type grammar, which is not. These correspond to developmental stages in which
children initially allow subjects to drop, filter out auxiliaries, and so on (as a first
step), to one in which they do not so do (as the second step). The means by
which children pass from the first grammar to the second, Hyams suggests, is by
means of the detection of expletives in the input. Such elements are generally
assumed not to exist in pro-drop languages; the presence of such elements would
thus allow the child to determine the type of the language that he or she was facing.
-
8/13/2019 David Lebeaux Language Acquisition and the Form of the Grammar 2000
48/308
A RE-DEFINITION OF THE PROBLEM 17
1.3.2 Determining the base order of German
The other example of a trigger, in the sense of (7) above, is found in Roepers
(1978b) analysis of German. While German sentences are underlyingly verb-final
(see Bierwisch 1963, Bach 1962, Koster 1975, and many others), the verb may
show up in either the second or final position.
(8) a. Ich sah ihn.
I saw him.
b. Ich glaube dass ich ihn gesehen habe.
I believe that I him seen have
Roepers empirical data suggests that the child analyses German as verb-final at
a very early stage. However, this leaves the acquisition question open: how does
the child know that German is verb final?
Roeper proposes two possible answers:
(9) i. Children pay attention to the word order in embedded, not
matrix clauses.
ii. Children isolate the deep structure position of the verb by
reference to the placement of the word not, which is alwaysat the end of the sentence.
At first, it appears that the solution (i) is far preferable. It is much more general,
for one thing, and it also allows a natural tie-in with theory namely, Emonds
(1975) conception, that various transformations apply in root clauses which are
barred from applying in embedded contexts. However,