Lexical-Functional Grammar 9783110838695, 9027930406, 9027930888, 9027932999, 9027931089, 902797778X, 9027934886, 9789027931696


195 89 11MB

English Pages 404 [408] Year 1983

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Preface
List of abbreviations and symbols
Introduction
1 Bounded processes
1.1 Passives
1.2 Overgeneration
1.3 Extensions
1.4 Argument reduction
1.5 Raising, equi and reflexive/reciprocal interpretation as operations on complex functional structures
1.6 Notes
2 Unbounded processes
2.1 The treatment of wh-words
2.2 Adjective phrase ‘deletion’
2.3 The role of too and enough
2.4 Notes
3 Constraints on [e]-anaphora
3.1 The Noun Phrase Constraint
3.2 The Single Gap Constraint
3.3 The NP/Gap Order Condition
3.4 The applicability of the constraints to Polish
3.5 Notes
4 Functional structures and interpretation
4.1 Functional structures – formal characteristics
4.2 Extensions
4.3 Functional structures and interpretation
4.4 Notes
5 Extensions and conclusions
5.1 General outline of the model
5.2 Rule application
5.3 Extensions and modifications
5.4 Alternative analyses
5.5 Notes
6 References
Recommend Papers

Lexical-Functional Grammar
 9783110838695, 9027930406, 9027930888, 9027932999, 9027931089, 902797778X, 9027934886, 9789027931696

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

Lexical-Functional Grammar

Trends in Linguistics Studies and Monographs 21

Editor

Werner Winter

Mouton Publishers Berlin · New York · Amsterdam

Lexical-Functional Grammar

George M. Horn

Mouton Publishers Berlin · New York · Amsterdam

Library of Congress Cataloging in Publication Data Horn, George M. Lexical-functional grammar. (Trends in linguistics. Studies and monographs; 21) Bibliography: p. 1. Generative grammar. I. Title. II. Series. P158.H59 1983 415 8 3 - 2 3 1 3 ISBN 9 0 - 2 7 9 - 3 1 6 9 - 0 (Netherlands)

© Copyright 1983 by Walter de Gruyter & Co., Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form - by photoprint, microfilm, or any other means - nor transmitted nor translated into a machine language without written permission from the publisher. Typesetting: Copo Typesetting and Artwork Thailand Co. Ltd, Bangkok. - Printing: Druckerei Hildebrand, Berlin. - Binding: Lüderitz & Bauer Buchgewerbe GmbH, Berlin. Printed in Germany.

Preface

No analysis or proposal stands in isolation, and the one outlined in this monograph should be viewed in the larger context of the major developments in linguistic theory over the past decade, stemming from the proposal of the so-called Lexicalist Hypothesis in Chomsky 1970, "Remarks on nominalization" (although the roots of this hypothesis extend back to Aspects of the theory of syntax (Chomsky 1965). The most significant product of linguistic research during this period has been the development and expansion of the lexical component and consequent reorganization and reformulation of the rules of the transformational-generative model, in which this component has been assigned many of the tasks formerly associated with the syntactic component. The decline in the inventory of syntactic rules is clearly seen, beginning with Chomsky 1973, "Conditions on transformations". This analysis has been refined and developed in Chomsky 1976, "Conditions on rules of grammar", Chomsky 1977, "On w/i-movement", and Chomsky 1978, "On binding". More recently, various counterproposals to Chomsky's analysis have been suggested. Perhaps the most important of these is that which was first unveiled in Bresnan 1978, "A realistic transformational grammar", and further developed in Bresnan 1980, "The passive in lexical theory", and a number of very recent papers. The key feature of this alternative is the further, virtually complete, reduction of the syntactic component. The framework which I propose is an attempt to extend and reformulate certain of Bresnan's and Chomsky's ideas, combining the basic organization of Chomsky's model, in which lexical and non-lexical operations are clearly distinguished, with a non-syntactic account of bound anaphora, control, and NP movement phenomena. Consequently, I assume familiarity on the part of the reader with the abovementioned papers. Other helpful background reading which may contribute to a better understanding of this monograph is found in the bibliography. Of particular relevance are the following: Anderson 1977; Bach 1976;

vi

Preface

Brame 1976a und 1976b, which raise interesting objections to Chomsky's analysis; Jackendoff 1972, especially the opening sections discussing lexical entries and thematic relations; Jacobson—Neubauer 1976, which provides background for the discussion of gerundive/infinitive phrase subject interpretation; Roeper—Siegal 1978; and Wasow 1977, which provides the basis for the discussion of passivization. The ideas of many others have contributed to the proposed analysis. I would particularly like to thank Ray Cattell, Emmon Bach, Bob Borsley, and Frank Heny, who provided helpful comments on earlier versions, and the students of the Department of Linguistics, University of Newcastle, who participated in seminars conducted during 1979 and 1980, in which much of the analysis was presented. None of these people necessarily agree with any of the conclusions reached herein, and, of course, any errors or misrepresentations of their ideas or those of others are my responsibility. Finally, I would like to thank Mrs. Joyce Bennett, who worked long hours typing and proofreading the original manuscript.

Contents

Preface List of abbreviations and symbols Introduction 1 Bounded processes 1.1 Passives 1.2 Overgeneration 1.3 Extensions 1.4 Argument reduction 1.4.1 Concluding remarks 1.5 Raising, equi and reflexive/reciprocal interpretation as operations on complex functional structures 1.5.1 Raising 1.5.2 Equi 1.5.3 Raqui 1.5.4 Reflexives and reciprocals 1.6 Notes

1* 1' 3 4 6: 7

8 8 8 9 10 10

2 Unbounded processes 2.1 The treatment of wh-words 2.2 Adjective phrase 'deletion' 2.3 The role of too and enough 2.4 Notes 3 Constraints on [e]-anaphora 3.1 The Noun Phrase Constraint 3.2 The Single Gap Constraint 3.3 The NP/Gap Order Condition 3.4 The applicability of the constraints to Polish 3.5 Notes

1 1 1 1 1 1

viii

Contents

4 Functional structures and interpretation 4.1 Functional structures — formal characteristics 4.2 Extensions 4.3 Functional structures and interpretation 4.4 Notes

202 202 237 265 281

5 Extensions and conclusions 297 5.1 General outline of the model 297 5.2 Rule application 313 5.3 Extensions and modifications 328 5.3.1 The analysis of BE 328 5.3.2 Cleft sentences 335 5.3.3 Another look at easy-adjectives 337 5.3.4 Gerundive/infinitive phrase subject interpretation .339 5.3.5 Tenseless indirect questions 363 5.3.6 Well-formedness Condition II 368 5.3.7 WFRs and lexical transformations 375 5.4 Alternative analyses 377 5.5 Notes 385 6 References

390

List of symbols and abbreviations

(lower case Greek letters) — represent semantic relations 0 (zero) — used to represent phonologically empty morphemes + (plus sign) - used to indicate morpheme boundaries; also used to represent features (i.e. + A = adjectival, + V = verbal, etc.) (double-headed arrow) — used in formulations of lexical transformations => (double-shafted arrow)—used in formulations of operations on functional structures ->• — used in formulations of word formation rules α, β,Ύ,δ,.

NPj NP 2 NP 3 NP p nom gen dat acc instr

. .

— subject — direct object — indirect object — object of preposition — nominative — genitive — dative — accusative — instrumental

E-A GRD NPGOC NPC PNR

— [e]-anaphora rule — Grammatical relations deletion - Noun Phrase/Gap Order Condition - Noun Phrase Constraint — picture noun reflexivization

PU SGC SI SRA WFR

- w[i]/[xi] Pair u n i o n - Single Gap Constraint — Subject interpretation — Semantic relations assignment — Word formation rules

Introduction

Since the Lexicalist Hypothesis was first proposed, the question of where to draw the line between processes that should be treated in the lexical component and those that should be treated in the syntactic component has been debated. In Chomsky's current framework, the inventory of syntactic processes (that is, operations on syntactic structures) has been reduced to NP movement and wft-movement (formally united as α-movement), and bound anaphora rules that assign controllers to reflexive and PRO forms, among others. In Bresnan's counterproposal (as presented in Bresnan 1978 and 1980, for example), NP movement and the bound anaphora rules have been eliminated from the syntactic component and reformulated as lexical processes. She claims that her analysis provides a simpler (more 'realistic') account of certain facts. Her proposals, as they stand, are intuitively quite appealing. However, in terms of its overall organization, the Bresnan framework is in effect virtually a 'mirror image' of preLexicalist models in which all movement rules, control rules, and bound anaphora rules were formulated as syntactic rules. For her, all such rules (with the possible exception of unbounded movement and deletion rules) are formulated as lexical rules. As a result, most of the distinctions that Chomsky first drew between lexical and non-lexical processes, which are reflected in the organization of his model, require some additional statement in Bresnan's model. One consequence of this is that the analyses differ in their account of the verbal/adjectival passive participle distinction, discussed in Wasow 1977. For Chomsky, as well as Bresnan, adjectival participles must be derived by means of a word-formation rule and a lexical transformation. In Chomsky's framework, this transformation can be formulated as shown below (more or less following the proposal in Anderson 1977): (0.1)

[VERB]

NP,1

NP 2, theme

«*· [V-enl.A

NP,1 theme

2

Introduction

Verbal participles, in Chomsky's analysis, are derived non-lexically, and sentences containing them are produced by, among other rules, one or more applications of NP movement. The mechanics of this, as well as the source of the suffix -EN, which is attached to verbal participles, are not important to the present discussion. It should be noted, however, that in the context of Chomsky's framework, verbal passive (and perfect) participles can be produced by a word-formation rule that is unaccompanied by a lexical transformation. Such a wordformation rule is, in effect, a non-syntactic analog of affix-hopping. What is important is that in Chomsky's analysis, verbal participles are inflectional variants of their active verb counterparts, and, as such, are necessarily of the same category and have the same meaning as these. In Bresnan's analysis, in which all participles are derived by means of a lexical transformation (and word-formation process), it does not follow that verbal participles will have the same meaning as their active verb counterparts while only adjectival participles can differ in meaning from their active verb counterparts. Chomsky's framework automatically provides two sources for participles, and need not be further elaborated to accommodate these facts. Bresnan's (1980) arguments that all passive participles must be derived lexically are based on the fact that various lexical rules, such as M/j-prefixation and compounding, apply to these forms. What her arguments actually show is that adjectival (+ A) participles must be derived lexically, but, as I have stated above, this must be done in Chomsky's framework anyway. Moreover, verbal participles can also be produced by word-formation rules in Chomsky's framework, and are therefore also available as the input to further word-formation processes. Consequently, the strongest argument for her purely lexical position is that it provides a more elegant account of most of the data than Chomsky's analysis, even though this is partially offset by its failure to automatically distinguish + A and + V participles. (Bresnan's arguments will be discussed in Chapter 1, section 4.) Let us now turn to the question of control. For Bresnan, whether a verb is a 'raising' or 'equi' verb is indicated by the form of its functional structure, and the term that controls the verb in the embedded complement (that is, the term that is interpreted as the subject of this verb) is indicated in the lexical entry of the matrix verb. Partial lexical entries for try and tend are shown below: (0.2) a try:

V;

try

((SUBJ), (VCOMP)); (SUBJ) = (VCOMP SUBJ)

Introduction

3

b tend: V ;

tend ((VCOMP)); (SUBJ) = (VCOMP SUBJ) Try, an equi verb in former analyses, is represented as taking two terms at the functional level, a subject and a verb complement, while tend, a raising verb in former analyses, is represented as taking a single term, a verb complement. In each case, the syntactic subject of the verb is interpreted as the subject of the verb in the complement. There is, however, little discussion of rules of control or the properties of complex functional structures. Because no rules are explicitly formulated to produce the functional structures of sentences containing verbs like try and tend (or other verb types) using information like that in (0.2) as their input, it is difficult to determine, for example, what sorts of conditions and constraints must be placed on such rules or the structures themselves in order to fully evaluate this analysis and compare it with Chomsky's more developed framework. In Chomsky's framework, the logical forms of examples containing such verbs are derived from S-structures to which PRO interpretation and NP movement, respectively, have applied. These rules are clearly formulated and shown to have certain well-defined properties. They are, for example, non-lexical rules, and are clearly different from lexical processes. Again, such a distinction must be explicitly stated in Bresnan's framework. In this monograph, I will propose and develop an analysis that is, in a sense, a compromise between Chomsky's and Bresnan's analyses. It is closer, organizationally, to the former in that it is 'compartmentalized' in such a way that it maintains Chomsky's division between lexical and non-lexical processes, and draws the line between the two at approximately the same point. At the same time, it is much more closely related, formally (and in spirit), to the latter. Chomsky's operations on syntactic structures, with the exception of wh -movement, are, in effect, replaced by a set of operations on functional structures which apply at a level that is analogous to his level of logical form. In the place of Chomsky's w/i-movement rule, the proposed analysis contains a rule of [e]-anaphora, which applies to surface structures to bind an empty node to some lexical antecedent in the structure. One consequence of the organizational similarities between Chomsky's analysis and the proposed one is that in both there are two sources for passive participles, and Wasow's + A/+ V participle distinctions are automatically accounted for. Furthermore, both analyses utilize nonlexical rules to account for control and bound anaphora facts, as well as raising facts, and in both analyses a unified account is given for whmovement, topicalization, and adjective complement deletion data.

4

Introduction

The proposed analysis, however, is preferable to Chomsky's analysis as it stands on grounds of formal simplicity. The operations on functional structures are formulated in such a way as to eliminate the overgeneration problem associated with Chomsky's syntactic rules, thus eliminating the need for his various constraints and additional mechanisms, and the rule of [e]-anaphora entails fewer and simpler conditions than Chomsky's w/z-movement rule. Moreover, because the rules of control and bound anaphora are formulated on functional, rather than syntactic structures, the same rules will account for these facts in languages that are syntactically quite different from English. The differences between the proposed analysis and Bresnan's are analogous to the organizational differences between Chomsky's analysis and hers. The proposed analysis is preferable to Bresnan's because it can make all of the linguistically significant generalizations that she can make, as well as some that she cannot make so easily, and because, in certain respects, it is more fully formally developed than her analysis in its present form. Comparisons and contrasts with Bresnan's and Chomsky's analyses illustrating these points will be made in appropriate places throughout the monograph, and in Chapter 5, the most important of these will be summarized. The proposed model is organized as shown in the following diagram:

(0.3)

Lexical Component Word-formation rules Lexical transformations

-*•

Lexical items (and entries)

Syntactic Component PS rules Lexical insertion rules [e]-anaphora Pronominal anaphora

Surface structures

Functional Component Operations on functional -*• (Complex) functional structures structures The rules of the lexical component are more or less the same as those proposed for all analyses within the context of the Lexicalist Hypothesis. WFRs perform morphological operations, and lexical transformations relate the functional structures of derived lexical items to those of the items that they are derived from. Inflectional variants of

Introduction

5

lexical items are produced by WFRs that are unaccompanied by lexical transformations, and consequently, all have the same syntactic category and meaning. PS rules and lexical insertion rules apply to create syntactic structures. Lexical insertion is optional, and if it does not apply, the relevant node in the structure will be empty, [e]. The primary operation on syntactic structures is the [e]-anaphora rule. Constituents such as wfc-words (and phrases containing w/i-words) and topicalized NPs are generated in COMP position, rather than moved there from some other position in the structure. One of the tasks of the [e]-anaphora rule is to bind empty nodes elsewhere in the structure to constituents such as these in COMP. The [e]-anaphora rule is unbounded, does not apply cyclically, and is unaffected by command and precedence relations between the empty node and its antecedent. In fact, the domain of this rule is not limited to constituents in COMP, and it can apply to bind empty nodes to constituents that are not located in COMP. The only constraints on the rule are the Noun Phrase Constraint (NPC), the Single Gap Constraint (SGC), and the NP/Gap Order Condition (NPGOC). The role of the [e]-anaphora rule in a unified account of w/i-questions, topicalized sentences, sentences containing eager and easy adjectives, and sentences containing too-ADJ and AOJ-enough constructions will be discussed in Chapter 2, and the constraints on this rule, as they apply to English and Polish, are discussed in Chapter 3. The rules of the syntactic component produce surface structures. The surface structure level is the only level relevant for interpretation. The lexical entries of verbs and verblike constituents contain, among other information, functional structures, in which NP terms are identified according to their grammatical relations and are assigned semantic relations. A typical entry is that for the verb see, shown below (its functional structure is represented in the final line). (0.4)

SEE +V SEE

NP, α 1

NP r

The terms NPj and NP 2 represent the syntactically defined grammatical relations subject and direct object respectively. For English, the subject is the circled NP in the configuration shown in (0.5) a, and the direct object is the circled NP in the configuration shown in (0.5) b (following Bresnan 1978 and Chomsky 1965):

6

Introduction (0.5) a b

s

[@

W[V

VP] © ]

Grammatical relations are defined on syntactic surface structures. In languages like English, they are defined configurationally, in terms of tree structures as shown above. In highly inflected languages like Polish, they are defined morphologically, in terms of case-marking. (Other morphologically defined systems of grammatical relations distinct from case systems are possible.) Grammatical relations may also be defined in terms of word order. (I leave open the question whether this is equivalent in all situations to defining them in terms of hierarchical structure.) The particular definitions of grammatical relations can vary from language to language. In Polish, for example, subjects are defined as NPs marked for nominative case (by means of a particular suffix), and the various types of objects are defined as NPs marked (by particular suffixes) for other cases such as accusative, locative, and dative. In Ergative languages, in which grammatical relations are also defined morphologically, in terms of case-marking, the particular set of grammatical relations is different. Subjects of intransitive verbs and objects of transitive verbs bear the same grammatical relation, while subjects of transitive verbs bear a different grammatical relation. In configurational VSO and VOS languages (an example of the latter is Malagasy), the specific definitions of subject and direct object will necessarily differ from the definitions of these relations (or rather, their analogs) in English. In the proposed analysis, the function of grammatical relations is to identify and distinguish NPs in surface structures in order that their indices can be inserted into the correct positions in functional structures. All languages must have some mechanism for accomplishing this task, and there need not be (and probably is not) a universal set of grammatical relations. (This will be discussed in Chapter 4.) Returning to the lexical entry for see, the terms a and β represent semantic relations. Semantic relations are distinct from grammatical relations, and are assigned by verbs and verblike constituents to the NP terms that appear in their functional structures. There is not a one-toone correspondence between semantic relations and grammatical relations. In English, for example, subjects, NPj, can bear the semantic relations agent with verbs like cause, location with verbs like know, goal with verbs like receive, and so forth. Similarly, in Polish, NPs marked for nominative case can bear the same semantic relations with the corresponding verbs (agent with spowodowac 'cause', location with

Introduction

7

wiedziec 'know', goal with otrzymac 'receive', and so on.) Similarly, NPs bearing other grammatical relations (NP2 or NP 3 in English, NPs marked for accusative, instrumental, or genitive case, for example, in Polish) can be assigned different semantic relations by different verbs. The particular semantic relation that a particular syntactically or morphologically defined term bears depends upon the individual verb. This information is indicated in the functional structure in the lexical entry of the verb, as shown in (0.4). Although I will use labels like theme, agent, goal, and so on, to 'name' semantic relations, none of the non-lexical operations in the analysis are sensitive to particular semantic relations, and these can be represented simply as α, β, γ, δ, . . . Consequently, there need not be a universal, or even a finite, set of semantic relations. I assume that verbs (and predicates in general) function to relate (NP) terms, or to assign some property to a single term, and that this is formally represented in the model by assigning distinct semantic roles to these terms. Because any given functional structure must contain a finite number of terms, every verb or verblike constituent in a particular language could, in theory, assign different specific semantic relations to its NP terms. (For reasons that need not concern us here, this situation, of course, does not exist.) Functional structures can be categorized into four basic types, which are universal. (These will be discussed at length in Chapter 4.) All are of the form shown in (0.6): (0.6)

PRED

NP α

NP β

...

(Y)

...

(W)

...

PRED represents the set of all verblike constituents (that is, those constituents that function as predicates). In English, not only verbs, but adjectives, nominals, picture nouns, other nouns in certain cases, prepositions, and probably certain quantifiers and adverbials can function as predicates, and have functional structures of this general form. The terms Υ and W represent non-NP terms. These can be complements of various types as well as embedded clauses, S, PPs, and terms of other categories. For any given syntactic surface structure, a (complex) functional structure is produced by combining the functional structures of the predicates that appear in that syntactic structure. Various operations on functional structures apply to produce the (complex) functional structure in its final form. Operations on functional structures are formulated in terms of grammatical and semantic relations, and can apply freely to functional

8

Introduction

(sub)structures whenever their structural descriptions are satisfied. They need not be ordered, need not be made obligatory, and do not apply cyclically. Their output (complex functional structures) must conform to two well-formedness conditions. Operations on functional structures are distinct from lexical rules (WFRs and lexical transformations). Only lexical rules operate to create lexical items and their entries, and these rules, but not operations on functional structures, can relate items of different syntactic categories. Only lexical rules, and not functional level operations, can be sensitive to specific semantic relations (that is, apply to semantically definable subsets of lexical items). Operations on functional structures need only be sensitive to the order in which semantic relations are assigned to the NP terms in the relevant functional structure, and be able to distinguish any given semantic relation, a, from any non-α. Operations on functional structures, as non-lexical rules, have the properties associated with non-lexical rules in the context of the Lexicalist Hypothesis. Because functional structure types are universal, the operations on functional structures are universal. The major operations on functional structures will be introduced in Chapter 1 and their universal applicability will be demonstrated there and in Chapter 4. The (complex) functional structures produced by the various operations provide the basis for the interpretations of the syntactic structures from which they are derived if and only if they are well-formed. A grammatical string is one from whose syntactic structure a well-formed functional structure can be derived. An ungrammatical string is one from whose syntactic structure no well-formed functional structure can be derived. An ambiguous string is one from whose syntactic structure more , than one well-formed functional structure can be derived, and two strings are synonymous if equivalent (well-formed) functional structures can be derived from their syntactic structures. The general rule is that an appropriate well-formed functional structure can be derived to provide the basis for all and only the possible interpretations of a given grammatical string. Some sample derivations of English and Polish sentences are shown below. Consider the following examples: (0.7) a John saw Bill b Who did John see The surface structures of these examples are shown in (0.8):

Introduction

9

S

(0.8) α NP

VP

John

V

NP

saw

Bill

b

S COMP

S

NP

who

VP

NP

John

V

NP

see

[e]

In structure a, John is identified as NPj and is indexed as, say, /'. Bill is identified as NP 2 and is indexed as, say,/. In structure b, John is NPj and can be indexed as i, NP 2 is empty, and who, the NP in COMP, is indexed as/. In structure b, the rule of [e]-anaphora applies to bind the empty node in NP 2 position to the wft-word in COMP. Functional structures for these examples are produced from the functional structure of see, which is indicated in its lexical entry (and shown in (0.4), above). To derive the functional structure of example (0.7) a, we begin with the structure shown in (0.9) a, and to derive that of example (0.7) b, we begin with the structure shown in (0.9) b: (0.9) a SEE

NPj

NP 2

b ( „ [ + wh]) [SEE

NPj

NP 2 ]

The indices of the NP terms are inserted into these structures in the proper positions to produce the following: (0.10) α SEE

t

[i]

1

[j]

*(npB [+wh]])[SEE

„[i]

„[X.]]

At this point, semantic relations are assigned to indexed terms to produce the structures shown in (0.11):

10

Introduction (0.11)β SEE b

__ [i] NP α

NP Ü]

(npÜ [ + w h l l ) t S E E

N P N P . 2 [X,]] α '

β

These structures provide the basis for the interpretations of these examples. Now consider the following Polish examples: (0.12) a Jan widziat Marka John (nom) saw Mark (acc) 'John saw Mark' b Kogo Jan widziat who (acc) John (nom) see 'Who did John see' The surface structures of these examples are shown in (0.13): (0.13)a

S acc

-widziat

Marka

COMP NP

acc

kogo

NP

NP Jan

widziat

[e]

In structure a,' Jan is identified as NP nom and can be indexed as i, and Marka is identified as NP___ and can be indexed as /. In structure acc b,' kogo ° is identified as NP acc'. and can be indexed as i,' and Jan is identified as NP nom and can be indexed as /'. The lexical entryJ for widziec ' 'see' is shown below: (0.14)

WIDZIEC +V WIDZIEC

NP

NPacc

In structure (0.13) b, the [e]-anaphora rule applies to bind the empty node to the constituent (kogo) in COMP.

Introduction

11

The functional structures of examples (0.12) a and b are derived from the functional structure of widziec, as in English, by inserting the indices of the surface structure NP terms in the proper positions and assigning semantic relations to these indexed terms. These functional structures are shown in (0.15): (0.15)α WIDZIEC α b (w

acc

nom

[i]

„acc

[j]

β

[i [+ wh]]) [WIDZIEC

w

α

nom

Ül

w

β

acc

Μ

1

These functional structures provide the basis for the interpretation of examples (0.12) α and b respectively. These examples illustrate some of the types of rules and properties of language that are universal, and some of the types of properties that are language-particular in the context of the proposed model. Verbs, as predicates, in both Polish and English have functional structures of the same general form. The verbs see and widziec, in fact, have virtually identical, transitive, functional structures. The [e]-anaphora rule applies to appropriate surface structures in both languages, and the operations of index insertion and semantic relations assignment also apply in the same way in both languages. The primary difference between English and Polish that emerges in these examples lies in the way that grammatical relations are defined in each language (as discussed above). This is reflected in the functional structures of the verbs see and widziec. A consequence of this is the greater degree of freedom of word order in Polish. The following are variants of (0.12) a: (0.16) a b c d

Widziat Jan Marka Marka Jan widziat Jan Marka widziat Widziat Marka Jan

Because grammatical relations are morphologically defined in Polish, the NPs Jan and Marka can be identified as and NP„„„ renom acc spectively, regardless of their positions in the surface structure, and their indices can be inserted into the NPnom and NPacc positions in the functional structure of widziec. Consequently, the functional structure shown in (0.15) a can be derived for all of these examples. In contrast, the following English examples are ungrammatical: (0.17) a »Saw Bill John

12

Introduction b *Bill John saw (as a non-topicalized sentence) Their surface structures must be something like those shown in (0.18): (0.18)e

S

Bill

John

saw

Either (0.18) a or (0.18) b is a possible surface structure for example (0.17) a. In structure (0.18) a, neither Bill nor John satisfies the definition of NPj or NP 2 . In structure (0.18) b, Bill can be identified as NP 2 , but John, again, cannot be identified as N P j . In structure (0.18)c, neither Bill nor John can be identified as either NPj or NP 2 . Consequently, in no case can the indices of John and Bill be inserted into the proper positions in the functional structure of see, and as a result, no well-formed functional structure can be derived for the examples in (0.17), accounting for their ungrammaticality. The correlation between case-marking (morphologically defined grammatical relations) and relative freedom of word order is predicted in the context of the proposed analysis. Other language-particular properties, which are not illustrated in these examples, but which are accommodated in the model, relate to the distribution of functional structure types, the syntactic categories of constituents which have a given function, and the sets of morphemes that trigger given operations on functional structures, all of which can, and do, vary from language to language. In Polish, for example, there are very few verbs which have functional structures of the same form as the functional structures of English 'raising' verbs. Thus Polish verbs

Introduction

13

like wierzyc 'believe' only occur with tensed-S complements as in the following example: (0.19)

Jan wierzy I ze Ewa uderzy la Mari§ 'John believed that Eve hit Mary'

In Polish, nominals function as complements with verbs like zmuszac 'force', as shown below: (0.20)

Marek zmuszal Jana do uderzenia Marii Mark forced John to hitting (nominal form) (of) Mary 'Mark forced John to hit Mary'

In English, only VPs function as 'equi' verb complements. Finally, in English, the verb be functions as an auxiliary verb (SpVP), while the corresponding form in Polish, bye, does not. These points and others will be discussed in Chapters 1 and 4. The proposed model provides a framework in which universal generalizations can be captured, and language variation can be accounted for without the complex machinery of Chomsky's current analysis, at the same time maintaining distinctions that are obscured in Bresnan's purely lexical analysis.

1

Bounded processes

In this chapter, I will propose and discuss an analysis that eliminates the major cyclic transformations: passive, raising, and equi, from the syntactic component. The bulk of the discussion concerns passive sentences, which will be derived from two sources, one which involves the application of a lexical passive transformation, and the other which does not. The latter source, however, does not require a syntactic movement rule or the coindexing of empty nodes, or traces, with lexical antecedents in the syntactic structure. It thus provides a non-transformational account of the data in Wasow 1977 which is preferable in some aspects to existing purely-lexical analyses (such as that proposed in Bresnan 1978 and 1980 for example). Having justified the analysis for English, I will look at passives and certain reflexive sentences in Polish and Italian and relate passivization to a general process of 'argument reduction'. Raising, equi, and reflexive/reciprocal interpretation are reformulated as operations which apply to (complex) functional structures similar to those proposed in Bresnan 1978.

1.1 PASSIVES In the lexical interpretive model, as mentioned in the introduction, cyclic transformations, or NP movement in Chomsky's framework, are eliminated. Consequently, any syntactic analysis of passives, whether it utilizes a passive transformation as such or treats passivization as a subcase of NP movement, should be highly suspect within this framework. Indeed, it would be extremely odd if the model required an NP movement rule whose only function were to derive passive sentences (or some subset of passive sentences). Bresnan (1978 and 1980) and others have argued that all passives can be derived lexically, and that some, in fact, must be so derived. However, we must still account for the data

1.1 Passives

15

that led Wasow (1977) to derive some passives lexically and others transformationally. Specifically, Wasow demonstrated that some passive participles are adjectives and some are verbs, and concluded that English must have two sources for passives: a lexical source for those containing adjectival participles and a transformational source for those containing verbal participles. Some examples of lexically derived passives are the following: (1.1) a The box was broken b The bottle was filled c John was frightened and so on. Examples of transformationally derived passives, which contain non-adjectival participles, are shown in (1.2) below. These are passives of raising verbs like believe, certain idioms, double object verbs like give and verbs that take NP complements like elect, and passives of sentences containing help and thank: (1.2) a b c d

Bill was believed to have hit Mary Advantage was taken of Bill Fred was given a Christmas present Nixon was elected president ί helped ] e John was Γ , , , / by Bill {thanked J

The examples in (1.1) differ from those in (1.2) in that the surface subjects of (1.1) are related to the participles as theme, while the surface subjects of (1.2) bear different semantic relations (goal in c, for example) and need not be related to the participles at all (as in a). In any case, the surface subjects in (1.2) are not related to the participles as theme. The fact that the participles in (1.1) display adjectival behavior while those in (1.2) do not can be accounted for by formulating the lexical passive rule so that it only applies to verbs that occur with a lone object, NP 2 , that functions as theme. A possible formulation is the following: (1.3)

VERB

NPj

NP 2 theme

~

[VERB + EN] A

NPj theme

This rule will not apply to the functional structures of the active verbs corresponding to the participles in (1.2) with the result that there is no lexical source for these examples. Consequently, only the participles in (1.1) can be + A since only lexical rules can relate items of different categories, and the participles in (1.2) must be + V. 1

16

1 Bounded processes

Bresnan 1980 utilizes two lexical rules like those shown in (1.4) and (1.5) to derive the functional structures of participles: (1.4)

VERB

NP, NP W a β α,β = semantic relations

(1.5)

[VERB + EN] V

NP,1

ο

~

theme

[VERB + EN] V

[[VERB + E N ] y ] A

NP. ß

W

NPj theme

Rule (1.4)appliestothefunctionalstructures of break,fill, andfrighten, as well as to the functional structures of believe, give, help, and thank, to produce the functional structures of the verbal participles broken, filled, believed, helped, and so forth. Rule (1.5) applies to the functional structures of those participles with theme subjects to produce the functional structures of adjectival participles. 2 The partial lexical entries for frightened, believed, and helped are the following: (1.6) α

["FRIGHTENED • · · FRIGHTENED

NP, theme

BELIEVED • · · BELIEVED

[(NP,)VP]

HELPED • · · HELPED

NPj goal

Because the functional structure of the verbal participle frightened, but not that of believed or helped, contains an NPj term that functions as theme, only frightened, of these three participles, has an adjectival (+ A) counterpart. In this analysis, however elegant it might be, certain characteristics of verbal and adjectival participles are not automatically accounted for as they were in the Wasow 1977 analysis. For example, there is no reason in principle why + V participles must have the same meaning as their active verb counterparts, since they are produced by a lexical rule, and lexical rules can relate items with different meanings. Only + A parti-

1.1 Passives

17

ciples, in fact, differ in meaning, either systematically (the 'stative' vs. 'active' distinction) or idiosyncratically (worn, as in Those clothes were worn by all of his older brothers, vs. worn (+ A) in His clothes were worn and dirty, for example), from the verbs that they are related to. (This will be discussed in more detail below.) Nor is there any reason in principle why participles of verbs whose objects, NP 2 , do not bear the semantic relation theme cannot be + A since all participles are produced by lexical transformations, and lexical transformations can relate items of different syntactic categories. More significant is the following correlation of properties: the transformation in (1.4) that relates items (verbs and participles in this case) of the same syntactic category relates items with the same meaning and is not sensitive to semantic relations (such as theme)', and contrasts with (1.5) in these respects. Bresnan cites Wasow's 1978, 1980 hypothesis that there are two types of lexical rules: major and minor. The former, of which (1.4) is one example, appear to have all of the attributes formerly associated with non-lexical processes, and (1.4) is, in effect, a lexical reformulation of the syntactic passive transformation. So Wasow's data are mechanically accounted for, but no principled explanation is offered. The analysis which I will propose utilizes only a lexical transformation similar to that formulated in (1.3), but modified as discussed below, to derive adjectival participles. Passive sentences containing verbal (+ V) participles are derived without the involvement of a lexical transformation. There is, however, no specific non-lexical passive rule, or any operation which is peculiar to passive sentences. Rather, the operations that apply to derive these also apply in the derivation of non-passive sentences. Consequently, Bresnan's generalizations can be captured, and the analysis provides a principled account of Wasow's observations and others which will be mentioned when appropriate. The proposed analysis, in a sense, represents a compromise position between that of Wasow 1977 and that of Bresnan 1980. Before discussing the alternative analysis, I will make a further assumption: that lexical insertion is optional, and consequently, NPs (and other syntactic nodes) need not dominate lexical material at the surface structure level. This assumption is not new, and in fact is needed in current NP-movement analyses. In this context, 'non-lexical' passive sentences can be base-generated with their subjects (NPj) in subject position, and with their objects (NP 2 ), or some other postverbal NP, as we shall see, empty. Empty nodes will be represented as NULL in the following discussion. NULL, however, should not be construed as a special terminal symbol. It has

18

1 Bounded processes

no other function than to represent an 'empty' NP (or other category). Empty nodes do not appear in (complex) functional structures, and play no role in interpretation. They thus differ significantly in function from empty nodes, or traces, in Chomsky's analysis. Let us now look at the derivation of these passives in more detail. Consider the following examples: (1.7) a b c d e / g

Bill was hit yesterday John was willingly helped Ralph was thanked for his efforts Bill was given a book Fred was elected president John was believed to have done that Bill was taken advantage of

I will discuss these in turn. The verb hit has a lexical entry something like the following:

(1.8)

HIT +V HIT V

NPj

NP.

agent

theme

The underlying structure of (1.7) a is shown below: (1.9)

S

Adv NULL

yesterday

Verbal participles such as hit in the above structure are derived by a word formation rule (WFR) that attaches the suffix -EN to the verb stem. This rule can be formulated as follows: (1.10)

[XL

[X - ENL

Such Inflectional' WFRs, unlike those that produce adjectival participles, nominals, and so on, are not accompanied by a lexical transformation such as (1.3) or (1.4). Rule (1.10), in fact, can be thought of as

LI Passives

19

the lexical analog of affix-hopping. The rule applies to hit, whose lexical entry is shown in (1.8), to produce the verbal participle hit + 0, whose lexical entry is shown below: (1.11)

|~HIT + 0 +V • · · HIT V

NPj

NP2

agent

theme

The lexical entries of the verb hit and the participle hit differ only in their morphological representations. The important point here is that their functional structures are identical since no lexical transformation has applied. Consequently, the meaning of the verb and its participle, represented as HIT y in both cases, are identical, as is their syntactic category. Inflectional WFRs of this type are independently needed to produce other verb forms in English (such as -ing forms in examples like: John is hitting Bill) as well as the various verb forms in languages like Polish, Italian, and many others, which are inflected for such things as person, number, gender, and so on (depending upon the particular language).3 To interpret (1.7) a, we begin with its functional structure, which is constructed from the functional structure of hit by combining it with S p V p [BE]. This structure is shown below: (1.12)

[ S p V P [BE]

V [HIT]]

NP 1

NP2

Bill is identified as NPj of hit, and is indexed as, say, i. NP 2 of hit is NULL (empty), and is not indexed. The indices, in this case, only i, are inserted into (1.12) to produce (1.13): (1.13)

[ S p V P [BE]

V [HIT]]

NPi

[i]

At this point, semantic relations are assigned to the NP terms in the functional structure. The semantic relations assignment rule is formulated as shown below: (1.14)

Semantic Relations Assignment (SRA): (a) In a (complex) functional structure of the form: PRED

Χ

Υ

Ζ

...

assign the semantic relations a, ß, y, . . . , to the NP terms in the order in which they appear in the lexical entry of the relevant PRED.

20

1 Bounded processes (b) In (complex) functional structures of the form: LSpVP v p [BE][PRED]]

Χ

Υ

Ζ

...

assign β to N P j , and the remaining γ, δ , . . . , to the NP terms to the right of NPj as in (a), where β is the semantic relation assigned to the second NP term in the lexical entry of the relevant PRED. As discussed in the introduction, the term PRED represents the set of lexical items which have functional structures. This set consists of verbs, adjectives, nominals, picture nouns, prepositions, and other nouns under circumstances to be discussed later. In this chapter, I will be mainly concerned with verbs and adjectives. Although BE is the only SpVP element mentioned in this formulation of the rule, I will show in a later section that other SpVP elements (e.g. get) trigger SRA (b). The particular elements that function in this way vaiy from language to language. Finally, Χ, Υ, Ζ are NP and non-NP terms. Let us now return to the example being discussed. According to the lexical entry of hit, (1.11) above, a = agent and β = theme. Thus SRA (b) applies to (1.13) to produce (1.15) below: (1.15)

[SpVp[BE]

V [HIT]]

Wi

[i]

theme

This completes the relevant part of the interpretation of (1.7) a. Examples (1.7) b and c are similarly handled. Help, for example, has a lexical entry something like the following: (1.16)

HELP +V HELP y

NPj

NP 2

agent

goal

The underlying structure of (1.7) b is the following: (1.17)

S NP

VP SpVP

John

BE

VP V

NP

helped

NULL

1.1 Passives

21

The verbal participle helped is derived by WFR (1.10). Beginning with the functional structure of helped, which is identical to that of the verb help, construction of the complex functional structure of the sentence, and insertion of indices produces the following: (1.18) [spVPIBE] r [HELP]] NP, [i] (As above, John has been identified as NPj of helped and indexed as i, NP 2 is NULL and is not indexed.) SRA (b) now applies to assign the semantic relation β (goal in this case) to the indexed term in (1.18) to produce (1.19): ( 1 1 9 ) [ S p V P [BE] V [HELP]] [i] goal

This completes the relevant part of the interpretation of (1.7) b. The same rules apply in the interpretation of (1.7)c. The functional structure of this example is shown in (1.20) below (/' = Ralph): (1 -20) [ S p V P [BE] v [THANK]] m [i] goal

Next, consider example (1.7) d, repeated below: (1.7) d Bill was given a book Give has a lexical entry something like that shown in (1.21): GIVE 0.21) +V • · · GIVE^ GIVE,

NPj

np2

np 3

agent

goal

theme

NPj

np2

PP[TO N P ]

agent

theme

(a) (b)

goal

This verb occurs in (at least) two syntactic environments, and for each one, there is a functional structure. These are labelled (a) and (b). 4 The structure underlying (1.7) d is the following: (1.22)

given

NULL

a book

22

1 Bounded processes

In this structure, Bill is identified as NPi. NP2 is NULL, and NP 3 is a book. NP t is indexed as / and NP 3 is indexed as/'. The relevant functional structure for this sentence is (a) in (1.21). (The functional structures of given (+ V) are identical to those of give.) The complex functional structure of (1.7) d is the following: (1.23)

[ S p V P [BE]

v [GIVE]

ΝΡ χ

NP,

NP 3 ]

Insertion of the indices into (1.23) produces (1.24) below: (1-24)

[ S p V P [BE]

y [GIVE]

^

[i]

^

[j]]

Next, the semantic relations goal and theme (β and γ) are assigned, in that order, to the indexed terms in (1.24) to produce (1.25): (1.25)

[ S p V P [BE]

v [GIVE]

[i] goal

^[j]] theme

This completes the relevant part of the interpretation of (1.7) d. Now suppose that the rightmost NP (NP 3 ) in structure (1.22) were empty. The result would be the following: (1.26) S

χ

BE

given

y

NULL

(χ and y represent lexical NPs.) The NPs, χ and y, are identified as NPj and NP 2 respectively, and are indexed as i and j. As above, NULL NPs are not assigned indices. Insertion of these indices in (1.23) produces the following: (1.27)

[ S p V P [BE]

v [GIVE]

NP[

[i]

Np

[j]]

As before, the semantic relations goal and theme are assigned in that order to the indexed terms in (1.27) to produce (1.28): 0-28)

IspVpIBE]

v

[GIVE]

Np

goal

[i]

NP^ [j]] theine

Structure (1.25) and structure (1.28) are equivalent since each contains two NP terms which bear the same semantic relations (goal and theme)

LI Passives

23

to the same verb (GIVE). Regardless of which postverbal NP is NULL, NPj in the derived functional structure will always be interpreted as goal. This accounts for the ungrammatically or semantic deviance of examples like (1.29) a below, which are bad for the same reason as (1.29) b. In both cases, a book is interpreted as goal:5 (1.29) a ?*A book was given John b ?*John was given to a book Examples like (1.30) have structures like (1.31): (1.30)

A book was given to John

(1.31)

S

a book

BE

given

NULL to John

A book is identified as NPj and indexed as i. NP 2 is NULL and is not indexed. The NP John is indexed as j. The relevant functional structure for this sentence is derived from (b) in (1.21) above: (1.32)

[ S p V P [BE]

y [GIVE]

NP t

NP 2

p p [TONP]]

Insertion of indices into (1.32) produces (1.33): (1.33)

[ S p V P [BE]

v

[GIVE]

m

\i]

p p [TO

N p [j]]]

SRA applies to assign semantic relations to the indexed terms to produce (1.34), completing the relevant part of the interpretation of example (1.30): d·34)

[SpvptBE]

vtGIVE]

np.M theme

P P [TO

NP Ü]] goal

The verb elect is superficially similar to give in that it occurs with either two postverbal NPs, or a postverbal NP and PP. Consider example (1.7) e, repeated below: (1.7) e Fred was elected president The lexical entry for elect is something like the following:

24

I Bounded processes (1.35)

ELECT +V ELECT y

ELECT V

NP,

NP 2

agent

theme goal

NP,

NP 2

agent

theme

(a)

NP 3

pp[T0

NP]

(b)

goal

(Functional structure (b) is required to account for examples like: (1.36) a They elected Nixon to the presidency b Nixon was elected to the presidency.) Example (1.7) e is derived from the following structure:

(1.37)

Fred

BE

elected

NULL president

Fred is NP} of elected and is indexed as i, and president is NP3 of elected and is indexed as/'. NP 2 is NULL and is not indexed. The complex functional structure of this example is derived from the following, which is constructed from the structure in (1.36) α (the indices i and / have been inserted and the structure itself has been abbreviated somewhat): (1.38)

BE

ELECT y

^

[i]

^[j]

Now the semantic relations, theme and goal, are assigned in that order to the indexed terms to produce (1.39): (1.39)

BE

ELECT y

^[i]

^[j]

theme

goal

This completes the relevant part of the interpretation of (1.7) e.

1.1 Passives 25 If the rightmost NP in a structure like (1.37) is empty, the result is the following (x and >> represent lexical NPs):

(1.40)

BE

elected

y

NULL

The lexical NPs, χ and y, are identified as NPj and NP2 of elected, and are indexed as i and / respectively. Insertion of these indices into the functional structure constructed from (1.36) a produces the following: (1.41)

BE

ELECT.

NP, [i]

„ρ D]

As above, the semantic relations theme and goal are assigned to NP. and NP. respectively. Thus NPj is interpreted as theme, and NP 2 is interpreted as goal. Consequently, sentences like the following are semantically deviant: (1.42)

?*President was elected Nixon

Here, president is interpreted as theme, and Nixon, as goal. Since there is no elective office called 'Nixon', example (1.42) makes no sense. This analysis extends to all three-place verbs with no modification. 6 Now consider example (1.7) f , repeated below: (1.7) /

John was believed to have done that

The lexical entry for believe is something like the following: (1.43)

BELIEVE +V • · · BELIEVE y

NPj loc.

[(NP 2 )

VP]

The direct object of believe, NP , bears no semantic relation.

26

1 Bounded processes The structure underlying example (1.7)/is the following: (1.44)

S

John

believed NULL to have done that

John is identified as NPj of believed and is indexed as i. NP2 of believed is NULL and is not indexed. That is identified as NP2 of do and is indexed as / . The complex functional structure of example (1.7) /, constructed by combining the functional structures of believed and do, is shown below: (1.45)

BE

BELIEVEv

NPj

[(NP 2 )

DO

ΝΡχ

NPJ

Insertion of the above indices into (1.45) produces the following: (1.46)

BE

BELIEVEy

^[i]

[DO

NP,

^[j]]

Assignment of semantic relations produces (1.47) below: (1.47)

BE

BELIEVEv

^[i]

[DO

NP, agent

w

U]]

theme

At this point, the functional analog of raising applies to (1.47) to produce (1.48) below, completing the relevant part of the interpretation of (1.7) / (this rule will be discussed in detail in a later section of this chapter): (1.48)

BE

BELIEVE

[DO

NP, [i]

ΝΡ,ϋ]]

agent

theme

Finally, consider examples like (1.7)g, repeated below: (1.7) g Bill was taken advantage of. Assuming that idioms of this type are entered in the lexicon as units, the entry for take advantage of is something like the following: (1.49)

TAKE ADVANTAGE TAKE ADVANTAGE

NP,1

NP of,

agent

theme

1.1 Passives

27

(A more accurate lexical entry for this idiom and others of this class will be proposed in a later section.) Examples like (1.7) g are derived from structures like the following: (1.50)

taken advantage

of N P [NULL]

As above, Bill is identified as NPj in this structure and is indexed as i. NPof- is NULL and is not indexed. Insertion of the index into the functio nal structure of (1.49) produces the following: (1.51)

BE

[TAKE

ADVANTAGE]

NP, [i]

Assignment of the semantic relation, theme, to the indexed term produces (1.52), completing the relevant part of the interpretation of (1.7 )*• (1.52)

BE

[TAKEV

ADVANTAGE]

^

[i]

theme

Idioms of this class (which includes keep tabs on, keep track of, etc.) have internal structure and can be analyzed as follows: (1.53)

V p [V

NP

PP]

In structures like (1.53), NP can be empty. This provides a source for examples like the following: (1.54)

Advantage was taken of Bill.

We see then that Wasow's 'transformational' passives can be derived as shown in the above discussion without the need for a movement rule. His 'lexical' passives, containing adjectival participles like those in (1.1), repeated below, are derived by means of a lexical passive rule similar to rule (1.3), which is formulated in (1.55) below: (1.1) a The box was broken. b The bottle was filled. c John was frightened.

28

1 Bounded processes (1.55)

VERB

NP.

[VERB + EN],

NP2

NP theme

theme

In (1.55), the lone NP term in the structural change of the rule is not identified as N P j , but rather is not assigned a grammatical relation by this rule. I will return to this point shortly. Participles like broken, filled, frightened, inhabited, and so on have lexical entries something like the following: (1.56) a

BROKEN +A • *· [BREAK + EN],

NP theme

FRIGHTENED +A • · · [FRIGHTEN + EN],

NP theme

INHABITED +A • · · [INHABIT + EN],

NP theme

and so on. The word formation rule that produces adjectival participal forms operates on verbal participles, and may be formulated as follows: 7 (1.57)

[X — en]

[[X — e n V ] vJ]A

This WFR is accompanied by the lexical transformation in (1.3), and thus only applies to verbs whose NP 2 term functions as theme. The derivations of frightened (+ V) and frightened (+ A) proceed as follows. We begin with the verb frighten, whose lexical entry is shown below: (1.58)

FRIGHTEN +V FRIGHTEN Γ

NPJ

NP 2 theme

(The particular semantic relation assigned to NP t is not important to the present discussion).

1.1 Passives 29 WFR (1.10) can apply to produce the verbal participle frightened, whose lexical entry is something like the following: (1.59)

FRIGHTENED +V FRIGHTEN V

N? { a

NP 2 theme

Now, WFR (1.57), along with lexical transformation (1.3), applies to derive frightened (+ A), whose lexical entry is something like the following:

(1.60)

FRIGHTENED +A «· · [FRIGHTENED].

NP theme

If WFR (1.57) applies to produce an adjectival participle from the verbal participle of a verb like help or give, the resulting word cannot be assigned a well-formed lexical entry because lexical transformation (1.3) cannot apply to the functional structures of these verbs (and adjectival participles must have adjectival functional structures as ip (1.60)). Consequently such forms are not possible lexical items. Adjectival participles such as unthanked, unhelped, and so on, are derived from verbs whose NP 2 terms do not function as theme. However, these participles are derived from the verbal participles by the UN-WFR, which forms adjectives by prefixing un- to certain stems. The UN-WFR is accompanied by a different lexical transformation, and is not limited to verbs whose NP 2 terms function as theme. In fact, this rule can apply to adjectival forms as well, to produce adjectives like: unlikely, unlovely, unthinkable, and so on. This rule produces words of the following form: [un — X ] A , where X = Ven, A , . . . Intransitive verbs such as descend have -edj-en forms (John has descended to the bottom of the ocean), which are produced by WFR (1.10), and these forms can provide the input to the UN-WFR which can then produce adjectival forms like undescended (in undescended

testicle). This analysis accounts for the fact that verbal and adjectival participles (as well as the verb forms that occur with the auxiliary have) exhibit exactly the same allomorphy.

30

1 Bounded processes

Returning to passive sentences that contain adjectival participles, examples like (1.1) and the following are derived from structures like those shown in (1.62): (1.61)

Antarctica is inhabited (in spite of its climate)

(1.62)a (Structure of (1.1) β): S

The box

broken

b (Structure of (1.1) c): S

John

frightened

c (Structure of (1.61)): S

Antarctica BE inhabited A complex functional structure must be constructed for each example. As mentioned previously, the NP term in the functional structures of adjectives is not identified by a grammatical relation. However, adjectives occur in syntactic structures like (1.62) with the main verb be, whose functional structure contains an NPi term:

1.1 Passives

(1.63)

(AP | BE |NP J

31

NP t

Combining the functional structure of be with that of the relevant adjective , frightened, for example, produces a structure like the following: 8 (1.64)

[BE]

FRIGHTENEDA

NPJ

From structure (1.62) b, John is identified as NPj of be and is indexed as i. This index is inserted into (1.64). SRA (a) applies to assign α (itheme) from the functional structure of frightened to NPj. (The verb be assigns no semantic relation to its NPj term.) These rules thus produce (1.65), completing the relevant part of the interpretation of example (1.1) c: (1.65)

[BE]

FRIGHTENEDa

^

[i]

theme

The complex functional structures of the other examples in (1.1), as well as that of (1.61), are derived in the same way. It is easy to see that Wasow's predictions are automatically accounted for in the proposed analysis. Because the lexical passive transformation, stated in (1.55), only applies to the functional structures of verbs whose NP2 terms function as theme, and because only lexical rules can relate items of different categories, only these verbs will have corresponding lexical adjectival participles (without the prefix un-). The only source for passive sentences containing the participial forms of verbs like help, thank, give, elect, raising verbs (believe) and idiom chunks is the non-lexical (NULL) source described above. Because this source does not involve the application of a lexical transformation like (1.4) or (1.55), the participial forms of these verbs will necessarily be + V. Thus participles like frightened (whose surface subjects function as theme) occur in the lexicon as + A and display adjectival behavior; while the participles of verbs like helped, given and so on (whose surface subjects bear different semantic relations) can only be + V, and only display verb-like behavior. In fact, sentences containing 'adjectival' participles are ambiguous. Consider the following: (1.66) a The door was locked. b John was frightened. c Antarctica is inhabited. These examples can be interpreted as describing the state of John, the door, and Antarctica, or as describing the activity of locking the door,

32

1 Bounded processes

frightening John, or inhabiting Antarctica. (This interpretation of c is clearer with past tense: Antarctica was inhabited.) In general, Wasow's adjectival participles can behave like verbal participles. The only test of Wasow's that involves a + V-only environment (rather than + A-only environments) is the 4very much' test. The selectional restrictions with very much are complex, but note that many adjectival participles can occur in this environment. Some examples are shown below: (1.67) a John was very much frightened (by the monster). b The ancestral castle was very much cherished by the whole family. c ?That treasure chest was very much locked and unlocked over the centuries. The ambiguity of examples like (1.66) and the fact that adjectival participles can behave like verbs (but not necessarily vice versa) can be accounted for, as in Wasow 1977, by analyzing participles like frightened and inhabited as either + A or + V. 9 Because lexical insertion is optional, it need not apply to the objects of sentences containing verbs like frighten, lock, inhabit, and so on, just as it need not apply to the objects of sentences containing verbs like give and help. Consequently, examples like (1.66) will be automatically derivable from two sources. Consider the following: (1.68) a Antarctica is inhabited (+ A), which is surprising considering its climate. b Antarctica is inhabited (+ V) every summer by migrating penguins. The source of a is lexical. That is, it has a structure like that shown in (1.62) c above, and contains the adjectival participle inhabited, whose lexical entry is shown in (1.55) c. The source of b is 'non-lexical'. The structure underlying this example is the following (partially represented): (1.69)

S NP

VP VP

SpVP

Antarctica BE

V

NP

inhabited

NULL

1.1 Passives 33 The verb inhabit has a lexical entry something like the following: (1.70)

INHABIT +V • · · INHABIT,,

NP.

NP 2 theme

(I represent the semantic function of NPj of inhabit as a, leaving open the question of whether it is agent or some other relation. This does not affect the present discussion.) In structure (1.69), Antarctica is identified as NPj of inhabited and is indexed as /'. NP 2 of inhabited is empty and not indexed. Insertion of the index into the functional structure of the sentence produces the following: (1.71)

SpVP

[BE]

INHABIT.

NP, [i]

Assignment of the semantic relation, β, (theme), to the indexed term produces (1.72) to complete the relevant part of the interpretation of this example: (1.72)

SpVP

[BE]

INHABIT.

NP.W theme

The difference in meaning between the adjectival reading of (1.68) a and verbal reading of (1.68) b is due to the difference in meaning between the verb inhabited (represented as INHABIT^) and the adjective inhabited, which is represented as [INHABIT + EN] A , as shown in the lexical entry in (1.56) c. The meanings of all adjectival participles can differ from the meanings of their corresponding verbs. In some cases, this meaning difference is significant and non-systematic. Some examples are lose/lost (+ A). The latter occurs in (1.73) a, but the verb lost does not have the same meaning, and b is semantically odd: (1.73) a Their cause was lost (before the battle began). b *They lost their cause (before the battle began). A similar pair is wear/worn, as in: The old clothes were worn and dirty. In contrast, the meanings of verbal participles are the same, for all intents and purposes, as those of their corresponding active verbs. This follows automatically from the proposed analysis in which adjectival participles are derived by a lexical transformation, which can involve a change in meaning, while lexical transformations play no role in the derivation of verbal participles and hence no change in meaning from

34

1 Bounded processes

that of their active verb counterparts is possible. Neither the ambiguity of sentences containing participles whose surface subjects function as theme, nor the fact that adjectival, but not verbal participles, can differ in meaning from their active verb counterparts is automatically accounted for in a 'purely-lexical' analysis of passives in which all participles are derived by a lexical transformation like (1.4). Actually, it is not true that participles like believed and known never display adjectival behavior. They can, in fact, occur in + A environments just in case their surface subjects (or understood subjects) function as theme. This is illustrated below: (1.74) a b c d

?His fate became known after years of mystery. That story seemed/became widely believed. A known quantity... A widely believed s t o r y . . .

These contrast with (1.75) below: (1.75) α *His fate became known to have been discovered by the police. b *That story seemed/became widely believed to have upset Sue. This is accounted for in the proposed analysis as follows: in addition to structures that contain sentential and VP complements, believe and know occur with lone NP objects (NP 2 ) that function as theme, as in: He believed the story, John knew the answer, and so on. Their functional structures in these cases are something like those shown below: (1.76) a BELIEVE

b KNOW

NPj

NP 2

loc.

theme

NPj

NP 2

loc.

theme

The lexical passive rule in (1.3) above can apply to relate these functional structures to those of corresponding lexical adjectival participles. Consequently, examples like (1.74) can be generated. In contrast, the only source for believed and known with VP complements, as discussed above, is the non-lexical source, which cannot involve a change in category from + V to + A. Thus (1.75) are blocked. In the framework of the proposed analysis, a stronger statement of the distribution of NPs functioning as theme can be made:

1.2 Overgeneration (1.77)

35

Functionally intransitive verbs, participles, and adjectives (i.e. those whose functional structures contain no NP 2 , NP 3 terms) have subjects, N P j , that function as theme.

This is true of adjectives like angry, tall, etc. as well as intransitive verbs like die, and lexical participles (+ A) like broken, frightened, and so forth, all of which have intransitive functional structures by the above definition. Participles for which statement (1.77) is not true, such as helped, thanked, and so on, have transitive functional structures identical to those of their active verb counterparts. In an analysis in which participles like helped, thanked, and so on, have intransitive functional structures, the generalization in (1.77) cannot be made. 10 This completes the basic discussion of, and justification for, the proposed two-source analysis of passives. In the next section, I will discuss the overgeneration question.

1.2 OVERGENERATION Because lexical insertion is optional, there are two primary possibilities for overgeneration: first, structures can be generated which contain be and a passive participle along with too many lexical NPs, and secondly, structures can be generated (with or without be and a participial form) which contain too few lexical NPs. Consequently, there must be some means of filtering out all of the garbage that can be derived. It should be noted that analogous problems exist for analyses in which the derivation of passives involves freely applying NP movement, and various conditions have been proposed to constrain this process and others. See, for example, Chomsky 1975, 1976, and 1978. None of the conditions proposed by Chomsky are required for any other reason in the proposed model, so I will suggest another method of blocking ungrammatical passives. Now consider the examples in (2.1): (2.1) a * John was hit Bill b *Fred was given Bill a book c * Antarctica was inhabited John Such examples can be derived by inserting SflV p[BE] into the structure and at the same time inserting lexical material into all of the NP positions. They have structures like the following:

36

1 Bounded processes

Fred

BE

given

Bill

a book

The functional structures of hit, given, and inhabited (+ V) are repeated below: (2.3) α HIT V

b GIVE y

NPj

NP 2

agent

theme

NPj

NP 2

agent goal (source)

c INHABIT V

NP 3 theme

NPj

NP 2

agent

theme

1.2 Overgeneration

37

In structure (2.2) a, John is identified as NPj and indexed as i, and Bill is NP2 and is indexed as /. Insertion of these indices into the complex functional structure of this example produces the following: (2.4)

[ S p V p [BE]

V [HIT]]

^[i]

Wi\j)

At this point, SRA (b), (1.14), applies to assign semantic relations to the indexed terms in the functional structure. For hit, a = agent and β = theme, and the rule produces (2.5) from (2.4): (2.5)

[ S p V p [BE]

V [HIT]]

Wi[i]

WtU]

theme

The relation a cannot be assigned to NP 2 by SRA as formulated, and consequently, ^ [j], an indexed term, does not bear a semantic relation. 11 Therefore\he functional structure in (2.5) violates the following condition: (2.6)

Well-formedness Condition: In a well-formed functional structure, all NP terms must be indexed and assigned a semantic relation.

Functional structures, as pointed out in the introduction, that are not well-formed do not form the basis for a possible interpretation of the surface structures that they are derived from, and if no well-formed functional structure can be derived from a particular surface structure, then the sentence that that surface structure underlies isungrammatical. This is the case with (2.1) a. The same explanation extends to (2.1) b and c, whose complex functional structures are the following, after index insertion: (2.7) a

BE

GIVE v

i = Fred; b

^

[i]

j = Bill;

BE INHABIT v i = Antarctica;

^

^

[j]

[k]

k = a book [i]

^[j]

j = John

For given, a = agent, β = goal, and γ = theme, and for inhabited (+ V), α = agent (in at least one sense of inhabit), and β = theme. Actually, the particular semantic relations involved are not important to the present discussion. The application of SRA (b) to the structures in (2.7) produces the following:

38

1 Bounded processes (2.8) α BE

GIVE v

[i]

Np

Np

goal

b BE

INHABIT V

[j]

Np

[k]

theme w 1,1

1

Np

[j]

theme

Both of these structures contain an indexed term that bears no semantic relation and are therefore not well-formed. Consequently, examples (2.1) b and c are ungrammatical for the reason given above. Note that lexical passives present no problems of overgeneration. Examples like the following are not derivable in the proposed analysis because the lexical transformation (1.55) eliminates the 'extra' NP 2 term: (2.9)

*Antarctica was inhabited (+ A) John (in spite of its climate)

The functional structure of the adjectival participle inhabited is the following: (2.10)

[INHABITED]A

NP theme

Since this structure, like the functional structures of all adjectival participles, contains a single NP term, there is no source for examples like (2.9). 1 2 ' 1 3 This account extends with no modification to other cases. Consider the following ungrammatical examples: (2.11) a b c d

*John was given a book to *Fred was made dinner for *John was hit and Bill *A book was given and a magazine to Fred

In these examples, some term not directly related to the verb is empty in their syntactic structures. The structure for (2.11) a is the following: S

(2.12) NP

VP VP

SpVP V

John

BE

given

NP

a book

PP Ρ

NP

to

NULL

1.2 Overgeneration

39

The complex functional structure of this example is derived from the following: (2.13)

BE

GIVE v

NP,

N?2

p p [TO

NP p ]

In (2.13), the entire PP, and not its NP object alone, is related to the verb given. In structure (2.12), John is NPj and is indexed as i, a book is NP 2 and is indexed as/. Insertion of these indices into (2.13) produces the following (NP p is empty and is not represented in the functional structure): (2.14)

BE

GIVE v

NP

[i]

NP

Lj3

PP

[TO]

The application of SRA (b) produces (2.15) below: (2.15)

BE

GIVE v

Np

[i]

^[j]

p p [TO]

theme

The structure in (2.15) is not well-formed. NP 2 does not bear a semantic relation because there is no indexed term immediately to its right in the functional structure of give, whose semantic relation can be assigned to it. (Notice too that the object of the preposition, NP p , is not represented in (2.15), so there is no term to which to assign the semantic relation which this preposition imposes on its object, precluding full interpretation of structure (2.12) in any case.) This accounts for the ungrammaticality of (2.1 l ) a . Example (2.11) b is accounted for similarly. Now consider (2.11) c, repeated below: (2.11) c *John was hit and Bill This example has a syntactic structure something like the following:

(2.16)

S VP

NP

VP

SpVP

NP

V NP John

BE

hit

NULL

NP and

Bill

In this structure, John is identified as NP, of hit and is indexed as /. NP 2 of hit is NULL and Bill, which is indexed as /'. Insertion of these indices

40

1 Bounded processes

into the complex functional structure of the sentence, and subsequent application of SRA (b), produces the following: (2.17)

BE

HIT V

^[i]

„ J ]

theme

For the reasons given above, this functional structure is not well-formed. Additionally, NP 2 contains an empty conjunct and cannot be fully interpreted. Thus example (2.11) c is ungrammatical. Example (2.11) d is accounted for in the same way. The functional structure of (2.11) cf is shown below: (2.18) BE GIVE v [i] [j] p p [TO [kl] theme

i = abook;

go£l

'

j = NULL and a magazine;

k = Fred

The functional structures of examples like the following also violate the Well-Formedness Condition in (2.6): (2.19)

*Bill was believed John to have done that

The syntactic structure of the example is shown in (2.20), and its complex functional structure, in (2.21):

Bill

(2.21)

BE

BE

BELIEVEy

/ = Bill;

believed

^

j = John;

John

[i] [ ( ^ [ j ] )

to have done that

[DO NP,

^

[k]]]

k = that

Since NP 2 of believed bears no semantic relation, β, no semantic relation can be assigned by SRA (b) to NPj of believed. The following structure is produced by the application of SRA (a) to the NP terms of do: (2.22)

BE

BELIEVE v

[i] [ ( w [j]) [DO NP,

w *

^

ft

^ Λ

[k]]] ^

1.2 Overgeneration 41 At this point, the functional analog of raising applies to produce (2.23): (2.23)

BE

BELIEVE v

^

[i]

[DO

^ [j] α1

^

β

[k]] 2

This functional structure is not well-formed because NPj of believed bears no semantic relation. Thus, example (2.19) is ungrammatical. Examples like (2.24), below, are derivable: (2.24)

John was believed to have done that

The functional structure of (2.24) after index insertion is shown below. NP 2 of believed is empty and does not appear: (2.25)

BE

BELIEVE y

i = John;

NPj

[i]

[DO

NP,

^

[j]]

j = that

As above, since there is no second NP bearing the semantic relation β, SRA (b) cannot assign a semantic relation to NP, of believed. SRA (a) applies to the NP terms of do, to produce the following: (2.26)

BE

BELIEVE v

[i]

[DO

NP, α

^[j]] β

2

Now, the functional analog of raising can apply to produce (2.27): (2.27)

BE

BELIEVE v

[DO

^[i]

a

'

^Jj]]

β

2

Structure (2.27) is well-formed, and example (2.24) is grammatical. Finally, consider the following example: (2.28)

*Bill was believed went to the store

This example must be derived from a structure like the following:

Bill

BE

believed

NULL went to the store

42

1 Bounded processes

In structure (2.29), Bill is NPj of believed, and is indexed as /'. NPj of go is NULL and is not indexed, and the store is NP p and is indexed as/. The relevant functional structure of believed is shown in (2.30): (2.30)

BELIEVEv

NPj

S

The complex functional structure of (2.28), after index insertion, is the following: (2.31)

BE

BELIEVE y

^

[i]

s[GO

p p [TO

^[j]]

The application of SRA (b) to NPj of believed and SRA (a) to the PP occuring with go produces the structure shown in (2.32): (2.32)

BE

BELIEVE.

NP, [i]

«[GO

PP [TO

OTüJJ]

goal

This functional structure is not well-formed because NPj bears no semantic relation. Moreover, S has no lexical subject, which is required in English. This latter point will be discussed next. In the following example, the subject, N P j , is empty: (2.33)

*Was given Bill a book

This example has a syntactic structure something like the following: (2.34)

S

NULL

given

a book

In this structure, NP t is NULL and is not indexed. NP 2 is Bill and NP 3 is a book. These are indexed as i and / respectively. Insertion of these indices into the complex functional structure associated with (2.34) produces the following: (2.35)

BE

GIVE.

NP [i]

NP [j]

This structure contains no NP } term. SRA (b) can be easily modified so that it assigns β to the leftmost NP term in the structure. The application of SRA (b) produces the following:

1.2 Overgeneration (2.36)

BE

GIVE v

Np

gozä

[i]

43

NPj[j]

theme

Structure (2.36) is well-formed even though example (2.33) is ungrammatical. Such examples are semantically well-formed, but syntactically illformed. The reason for this is that English requires clauses (S) to have lexical subjects at the surface structure level. Example (2.33) is ungrammatical for the same reason as: *Went to the store yesterday and so on. This condition also accounts for the ungrammatically of the following, in which the embedded clauses, of category S, have no lexical subjects: (2.37) a *Bill believed that went to the store b *John would prefer for (INDEF) to do that Example b is grammatical, with no for, when John is interpreted as the subject of do : John would prefer to do that. This, however, is a different situation. In the proposed analysis, subjectless (and tenseless) complements are analyzed as VP rather than S. Thus the above condition on subjects can be formulated. This condition, as expected, is not universal, and does not apply, for example, in Polish. Thus subjectless clauses like the following are not ungrammatical: (2.38) α Dano ksiqzk