cover
title: author: publisher: isbn10 | asin: print isbn13: ebook isbn13: language: subject publication date: lcc: ddc...
64 downloads
1115 Views
3MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
cover
title: author: publisher: isbn10 | asin: print isbn13: ebook isbn13: language: subject publication date: lcc: ddc: subject:
next page >
Studies of Passive Clauses SUNY Series in Linguistics Postal, Paul Martin. State University of New York Press 0887060838 9780887060830 9780585068077 English Grammar, Comparative and general--Passive voice, Grammar, Comparative and general--Clauses, Arc pair grammar, Relational grammar. 1986 P281.P53 1986eb 415 Grammar, Comparative and general--Passive voice, Grammar, Comparative and general--Clauses, Arc pair grammar, Relational grammar.
cover If you like this book, buy it!
next page >
< previous page
page_i
next page > Page i
Studies of Passive Clauses
< previous page
page_i If you like this book, buy it!
next page >
< previous page
page_ii
next page > Page ii
SUNY Series in Linguistics Mark Aronoff, Editor
< previous page
page_ii If you like this book, buy it!
next page >
< previous page
page_iii
next page > Page iii
Studies of Passive Clauses Paul M. Postal State University of New York Press
< previous page
page_iii If you like this book, buy it!
next page >
< previous page
page_iv
next page > Page iv
Published by State University of New York Press, Albany © 1986 State University of New York All rights reserved Printed in the United States of America No part of this book may be used or reproduced in any manner whatsoever without written permission except in the case of brief quotations embodied in critical articles and reviews. For information, address State University of New York Press, State University Plaza, Albany, N.Y., 12246 Library of Congress Cataloging in Publication Data Postal, Paul Martin, 1936 Studies of passive clauses. (SUNY series in linguistics) Bibliography: p. 255 Includes index. 1. Grammar, Comparative and generalPassive voice. 2. Grammar, Comparative and generalClauses. 3. Arc pair grammar. 4. Relational grammar. I. Title. II. Series. P281.P53 1985 415 84-26850 ISBN 0-88706-083-8 ISBN 0-88706-084-6 (pbk.)
< previous page
page_iv If you like this book, buy it!
next page >
< previous page
next page >
page_v
Page v Contents Preface
ix
Acknowledgements
xiii
Chapter 1 Introduction
1 1
1.1 Background 4 1.2 Passive Clauses 8 1.3 Impersonal Passive Clauses 11 1.4 Relevant Aspects of Arc Pair Grammar 11 1.4.1 Remarks 12 1.4.2 Nonconstructive Grammars 14 1.4.3 Some Rule Notations 15 1.4.4 Pair Networks 22 1.5 A Theoretical Note 23 Notes Chapter 2 Virtues of a Universal Account of Passives
27 27
2.1 Two Virtues of a Universal View of Passives 27 2.2 Some Lawful Properties of Passive Clauses 28 2.2.1 Double Passives 30 2.2.2 Passives of Unaccusatives 32 2.2.3 Passives of Inversion Clauses
< previous page
page_v If you like this book, buy it!
next page >
< previous page
next page >
page_vi
Page vi 33 2.2.4 Passives of Impersonal Clauses 34 2.2.5 Summary 34 2.3 Facilitating the Statement of Language-Particular Restrictions 34 2.3.1 Remarks 35 2.3.2 Apparently Anomalous French Personal Passives 41 2.3.2 Postscript 43 2.3.3 A Restriction in Ancient Greek 45 2.3.4 A Restriction in Modern Greek 50 2.3.5 A Restriction in Korean 52 2.3.5 Postscript 52 2.3.6 Quiché Instrumental Advancement 56 2.3.7 Agreement and Passives in Chichewa 65 2.3.7 Postscript 66 2.3.8 Agreement and Passives in Chi-Mwi:ni 84 2.3.9 Halkomelem Passives and Raising 87 2.3.10 Summary 88 Notes Chapter 3 The Inadequacy of the Relational Grammar Account of Impersonal Passives 95 95 3.1 The Relational Grammar View 96 3.2 English Personal Passives with Dummies Advanced from 2 to 1 99 3.3 French Personal Passives with Dummies Advanced from 2 to 1
104 3.4 More English Personal Passives Involving Advanced Dummies 104 3.5 Summary 104 Notes Chapter 4 An Arc Pair Grammar Definition of Impersonal Passives
108 108
4.1 The Arc Pair Grammar View of Clause Structure and the Problem of Impersonal Passives 109 4.2 Ghost Arcs and Dummy Nominals 112 4.3 Ghost Sponsors and the Characterization of Impersonal Passives 117 4.4 An Argument Distinguishing French Impersonal Passives from Certain 'Impersonal' Personal Passives 118 4.5 Formalizing the Subjunctive Restriction 122 4.6 Reflexive Passives 125 4.7 Reflexive Impersonal Passives and the Subjunctive Restriction 128 4.7 Postscript
< previous page
page_vi If you like this book, buy it!
next page >
< previous page
next page >
page_vii
Page vii 130 4.8 The Ghost Arc Law and a Fundamental Property of Impersonal Passives 132 4.9 Rules Involving Passive Clauses 136 Notes Chapter 5 Apparent Impersonal Passives of Unaccusative Structures and Possible Violations of the 1AEX
140 140
5.1 Remarks 141 5.2 Turkish 143 5.2 Postscript 157 5.3 Extensions of Passive Morphology 159 5.4 Sanskrit 165 5.5 Nerbonne's Critique 166 5.5.1 Lithuanian 184 5.5.1 Postscript 190 5.5.2 Irish 192 5.5.3 Estonian 192 5.5.4 Lithuanian and German Medio-passives 197 5.6 Summary 198 5.7 A Note on Italian 199 Notes Chapter 6 English Pseudo-Passives
203 203
6.1 Background 209 6.2 An Arc Pair Grammar Analysis of Pseudo-Passives 216 6.3 Arguments for 'Reanalysis'
220 6.4 The Pseudo-Passive Control Rule 222 6.5 An Alternative Arc Pair Grammar View of Pseudo-Passives 222 6.5.1 The Issue 224 6.5.2 Blocking Pseudo-Passives with Any 2 Arc 224 6.5.3 Some Earlier Observations 227 6.5.4 Some Demotion Paradigms 237 6.6 Open Questions 237 6.6.1 Other Constraints Limiting the Class of PseudoPassives 239 6.6.2 Additional Complex Passives 241 Notes Chapter 7 Generalized Passive Clauses
246 246
7.1 Remarks 247 7.2 Theoretical Modification 247 7.3 Weaker Constraints on Prepassive Arcs 253 7.4 The Scope of Non-2 Prepassive Arcs 253 Notes
< previous page
page_vii If you like this book, buy it!
next page >
< previous page
next page >
page_viii
Page viii References
255
Index
265
< previous page
page_viii If you like this book, buy it!
next page >
< previous page
page_ix
next page > Page ix
Preface The present work seeks to introduce, motivate and support major features of the grammatical framework of ARC PAIR GRAMMAR (APG) as developed in Johnson and Postal (1980) and in Postal (1982). The substantive ideas about SENTENCE STRUCTURE of this view are an outgrowth, refinement and extension of work in RELATIONAL GRAMMAR (RG); see Perlmutter (1983a) and Perlmutter and Rosen (1984) for an introduction to this work. However, the APG notations of RULE, GRAMMAR and WELL-FORMEDNESS DETERMINATION have no real precedents in modern grammatical study; see Langendoen and Postal (1984). The goal of justifying APG views about sentences and grammars is approached by applying them to the description of a range of facts involving PASSIVE CLAUSES in a variety of natural languages. It is assumed, contrary to a not insignificant skeptical tradition, that passive clauses are an identifiable and theoretically cross-linguistically characterizable aspect of NLs. One important feature of the current framework is that it permits improvements in the universal characterization of the notion 'passive clause' originated in RG work. These improvements involve, first, an account of the notion 'dummy nominal', which is crucial in characterizing the class of impersonal passive clauses. Second, the APG framework permits refinements needed to yield the major known subtypes of passive clause: impersonal vs. personal and reflexive vs. plain. These cross-classify to determine the four major
< previous page
page_ix If you like this book, buy it!
next page >
< previous page
page_x
next page > Page x
categories of passive, all attested. Third, improvements are also provided via the construction of a precise system of rule concepts for talking about the restrictions on passive clauses in NLs. It is suggested that the universalist RG/APG approach to passive structures can be supported from the passive domain in three distinguishable ways: (i) by permitting the formulation of otherwise apparently unstatable lawful characteristics of passive structures in all NLs possessing them; (ii) by facilitating the statement of various NL-specific constraints relevant to passive constructions in diverse NLs; (iii) by allowing uniform statement in grammars of recurrent constraints on the varied objects rightfully called passives. The approaches in (i) and (ii) extend and develop lines of argument already richly represented in previous RG work. Each mode of support is developed and applied to actual cases. Various problematic data claimed by critics to be inconsistent with the universal approach to passives are treated and argued to have adequate analyses consistent with all relevant universal claims. A pervasive feature of this study is the elaboration and refinement of the conceptual and notational devices available for stating APG grammatical rules. These devices are extensively illustrated and motivated throughout via the construction of several dozen explicit rules for various NLs. Each such notion is explicitly defined on the basis of the formalized system in Johnson and Postal (1980). Such devices are a key aspect of the construction of a universal language for the formulation of NL grammatical rules. They thus amount to part of a theory of the nature of NL grammars. A principle feature of the book is the development of a precise account of the notion 'impersonal passive clause'. This type of structure, though widely attested in the descriptive literature, has not abeen accorded much theoretical attention in most other grammatical frameworks, like transformational grammar. This is perhaps due to the fact that such frameworks have been primarily concerned with the study of English, which lacks impersonal passives. But impersonal passives are central to the clause structure of many NLs. It is argued that the initial account of impersonal passives in RG work, while essentially correct, provides only a necessary condition. A sufficiency condition made possible by the specific character of APG representations is proposed and argued for. Finally, this study offers a number of refinements, modifications and corrections of the APG framework. New laws are suggested, several previous laws are rejected and a number of theoretical and conceptual elaborations and changes are discussed. Problems for a number of earlier APG laws are treated. Notably, it is briefly shown how facts of constituent ordering can be precisely and simply treated without appeal to any special primitives, thereby correcting a serious error in JP (1980). Overall, this is both a descriptive work involving particular factual accounts relevant to theoretical issues and a theoretical study buttressed by
< previous page
page_x If you like this book, buy it!
next page >
< previous page
page_xi
next page > Page xi
attention to grammatical detail. My hope is that it will be of interest to those concerned with all aspects of NL grammatical theory and description. It should appeal especially to those concerned with RG, grammatical relations, the description of passive and impersonal structures, and, more generally, to those drawn to the goal of constructing a substantive theory of NL sentence structure. In addition, this study should have something to say to those working on the construction of explicit grammatical descriptions of specific NLs which combine NL-particular statements with a rich system of principles claimed to be grammatical laws. This work should also be of interest to those investigating NLs in nonrelational frameworks, particularly the many 'surface' approaches to NL grammar now current. Here the appeal lies in the fact that the current study offers many analyzed cases which appear to challenge the basic philosophy of such approaches. Independent of theoretical orientations, the present investigation has something to offer to those working on any of the many particular NLs dealt with, as well as on related and structually similar NLs. The discussion is based on material from English, French, German, Lithuanian, Quiché (Mayan), Welsh, Sanskrit, Modern Greek, Ancient Greek, Halkomelem (Salish, British Columbia), Chichewa (Bantu), Chi-Mwi:ni (Bantu), Quechua (Andean, Ecuador) and Turkish. The work is organized as follows. Chapter 1 provides some conceptual background, discusses the general theoretical approach underlying this study and provides a brief sketch of basic APG ideas relevant to the subsequent discussion. Chapter 2 suggests the virtues of a universal characterization of passive clauses generally. Chapter 3 discusses the limitations of RG accounts of impersonal passive structures. Chapter 4 introduces the proposed APG refinement of these accounts, supporting the innovation with material from French. Chapter 5 considers several recent arguments that some of the lawful properties of passives discussed in Chapter 2 fail for impersonal passives in NLs including Sanskrit, Turkish, and Lithuanian. It is suggested that a careful distinction between impersonal passives in the sense of Chapter 4 and other, distinct impersonal clauses provides a way of keeping the cited data consistent with these regularities. Chapter 6 sketches a general account of English pseudo-passive clauses, showing how these difficult expressions, often treated as pathological, can be naturally characterized in the framework previously developed. Finally, Chapter 7 argues the necessity of generalizing the notion 'passive clause' so that is also includes cases in which the 'passivized' element is not a direct object. The required generalization is essentially trivial.
< previous page
page_xi If you like this book, buy it!
next page >
< previous page
page_i
next page > Page i
Studies of Passive Clauses
< previous page
page_i If you like this book, buy it!
next page >
< previous page
page_xiii
next page > Page xiii
Acknowledgements I would like to thank David Johnson and David Perlmutter, and also several anonymous referees, for many helpful comments on earlier versions of this study. The sections on Turkish in Chapter 5 have benefited greatly from a number of helpful communications by Inci Özkaragöz of the University of California at San Diego. She is, however, not at all responsible for the interpretations I have given, which are incompatible with proposals in several of her papers. The sections on German and Lithuanian in Chapter 5 are improved due to comments by. John Nerbonne of Ohio State University. He is not to blame for current interpretations, which conflict with the positions he has expressed in work considered in Chapter 5. Further, I would like to thank my colleague Eva-Maria Mueckstein for providing me with helpful information about German relevant to the discussion in Chapter 5. Finally, I am greatly indebted to my principal recent French consultant, Marie-Madeleine Saphire, for hours of patient and helpful cooperation. This study would not have been possible without her help. Every French example cited below which is not taken from the literature has been checked with her at least three times on separate occasions, and, in most cases, far more. Nonetheless, I alone am responsible for any errors in this domain and for the particular interpretations of her judgements.
< previous page
page_xiii If you like this book, buy it!
next page >
< previous page
page_1
next page > Page 1
Chapter 1 Introduction 1.1 Background Grammatical investigation is, in my view, concerned with the nature of sentences and with certain collections of sentences forming NLs. In this view, it would not be wrong to consider grammar to be the science of sentences in the same sense in which logic is the science of correct arguments or number theory is the science of numbers. Following Katz (1981, 1984), Langendoen and Postal (1984), I assume that NLs and sentences are abstract objects rather than physical or psychological ones, and that the goal of grammatical theory is to provide a characterization of such objects. Since the ontology of NLs is akin to that of the objects studied in logic and mathematics, this characterization will say nothing directly about human beings or about the nature of any other creature. This view is, of course, almost entirely contrary to the currently dominant conception of linguistic activity which has been promulgated with great effectiveness by N. Chomsky over recent decades. According to this position, a grammar of an NL is a description of human linguistic knowledge and hence a psychological (sometimes biological) object. 1 On this account, grammatical theory is supposed to be a characterization of some innate human endowment which makes first NL acquisition possible. However, as Katz (1981, 1984) argues in detail, the Chomskyan program involves profound confusions and errors (see also Langendoen and
< previous page
page_1 If you like this book, buy it!
next page >
< previous page
page_2
next page > Page 2
Postal (1984)). In particular, it is obvious that for any X, knowledge of X presupposes the existence of X. No one is tempted to confuse, e.g., phone numbers with knowledge of phone numbers, genetic structures with knowledge of genetic structures, etc. NLs must likewise not be confused with knowledge of NLs; thus Choctaw is entirely different from knowledge of Choctaw, the latter presupposing the existence of the former. Therefore, were it true, as the Chomskyan program proclaims, that grammars describe knowledge of NLs, this would leave no aspect of linguistics to describe NLs themselves. Various analogies utilizing algorithms, computer programs which represent these algorithms, and the physical representation of such programs can help clarify matters here; see Langendoen and Postal (1984: Chapter 6). As viewed here then, a specified NL grammar characterizes a certain NL but says nothing directly about knowledge of that NL, just as an account of some algorithm says nothing directly about a program embodying that algorithm. In the same way, grammatical theory is concerned with discovering the laws governing the nature of sentences and their organization into NLs independent of the relation of these laws to any particular creatures who might know them or manifest representations of them. Evidently though, there can hardly be serious study of the nature of linguistic knowledge without a deep understanding of the nature of the objects known: NLs. Hence, the formulation of a theory of grammar as understood here is surely a prerequisite to understanding linguistic knowledge or how it is acquired. Without a serious analysis of NLs there is no independent account of what an NL-learning creature gains knowledge of, just as there can hardly be a study of how knowledge of integers is obtained without some understanding of integers and their properties. 2 Another crucial point, which Katz has stressed in recent work, is that if there is no independent characterization of NLs, there is no meaning or content to questions about knowledge of language. From the point of view adopted here, developed in greater detail in Johnson and Postal (1980) (JP (1980) hereafter), Postal (1982) and Langendoen and Postal (1984), grammatical theory has the character of a definition of the form in [1]: Theoretical inquiry is intended to fill in the dots with a substantial characterization. Such an account is in one sense a simultaneous characterization of both all true linguistic universals and all true linguistic variation. That is, the linguistic universals are those properties which are logically determined by the full characterization that [1] schematizes. Genuine linguistic variation consists of all properties P of objects satisfying the characterization determined by [1] where, however, P are not entailed by the characterization but are nontheless consistent with it. For instance, infor-
< previous page
page_2 If you like this book, buy it!
next page >
< previous page
page_3
next page > Page 3
mally, since there will be no law subsumed by [1] which says that clausal subjects either follow or precede the predicates of their clauses, there exist NL sentences manifesting both order patterns. In other words, in this approach, there is absolutely no need to specify independently of grammatical laws any parameters of variation. The full set of laws will itself determine possible variation in NLs. It is advisable to say a little more about linguistic universals in general before turning to the central concerns of this study. The idea of linguistic universal, involves universal quantification over some collection of NLs. But sensible discussion requires distinguishing at least the following collections:
Let us start with [2c]. These are NLsLatin and Navajo, for examplewhich have historically known speakers. We will ignore the fact that such names designate distinct NLs differing in ways which are small in comparison to those differences separating NLs not ordinarily given a common name. Obviously, all universals must be induced from generalizations about attested NLs. However, grammatical theory is not really interested in this set per se, since its membership is determined in part by historical accident. The collection [2b] subsumes [2c] but also contains undocumented NLs of past linguistic communities, those of future linguistic communities, etc. It includes all NLs actually learnable by humans, 3 with those so far learned as an accidental subset. Grammatical theory is not concerned with this collection per se, either, as the property of being an attestable NL is determined by nonlinguistic principles, namely, linguistically accidental features of human nature. These may include innate linguistic mechanisms of the sort Chomsky has speculated about as well as general limitations of memory, computing power, lifespan, etc. For example, no doubt every attestable NL, like every attested NL, will have a maximum bound on attestable morpheme length of about a dozen or so phonemes. Certainly, none will contain a constrained4 morpheme of phoneme length nine billion. But this fact about attestable NLs will surely follow from a characterization of human nature, and is thus not strictly a linguistic universal in the sense of being a property of NLs per se; NLs involve no bound on morpheme length, just as they manifest none on sentence length. The truly linguistic universals are those independent of the nature of any creature and which follow from the nature of NL itself, just as the truly logical principles are those which follow from the nature of logical objects themselves independent of the nature of human beings. The linguistic universals in this sense are precisely the truths of a5 correct grammatical
< previous page
page_3 If you like this book, buy it!
next page >
< previous page
page_4
next page > Page 4
theory, truths subsuming its laws together with all the theorems these laws determine. The fundamental problem of grammatical study as I see it is to construct a theory which correctly projects from the properties of attested NLs to a characterization of NLs per se. One primary difficulty is to distinguish features of attested NLs which are a reflection of the properties of NL as such from those which are simply a function of the nature of the NL-learner, that is, from properties merely reflecting limitations of the conditions of human knowledge. In these terms, the data of individual attested NLs provide the grammarian with samples of sentences and of nonsentences of a particular NL. From this, one must project characterizations of individual NLs and the notion NL sentence. The goal of grammatical theory is then to provide the richest and narrowest possible projection from all available valid data to a lawful characterization of sentence and NL. For example, if we had no knowledge of the phenomenon of 'noun incorporation', (see Postal (1979)), we would construct a grammatical theory which predicts that no NL sentence manifests this feature. However, since we know that 'noun incorporation' does exist, our goal can only be to determine its nature anti to restrict its occurrence to the maximal extent consistent with attested cases. In what follows, I attempt to flesh out the above remarks not by direct discussion of the issues raised, but by illustrating aspects of an actual attempt to construct, justify and apply to diverse facts in a variety of NLs one tiny part of a grammatical thory of the sort just hinted at. 1.2 Passive Clauses Both traditional and modern linguistic descriptions of the most diverse NLs are filled with discussion of so-called 'passive clauses'. Examples like those in [3] are representative:
< previous page
page_4 If you like this book, buy it!
next page >
< previous page
page_5
next page > Page 5
Usages treating inter alia all of [3a-d ii,iii] as passives imply that there is some notion PASSIVE CLAUSE of which the various clauses so characterized are truly instances. This similarity is intuited despite myriad superficial differences among the structures characterized as passive. French passives generally require an auxiliary, 6 none of the others do; the Indonesian passive nominal marker oleh is optional, which is not true of its analogues in any of the other cases; Tzotzil and Malagasy have subject-final word order, which is not true of the other cases; Tzotzil has ergative person agreement, which is not true of the others, etc. But in spite of these and other contrasts, linguists of diverse backgrounds, views and cultures recognize a common feature, passive, among the divergences. Of course it is conceivable that the multitude of traditional and modern investigators of these matters who assume a general, universal concept are wrong and there simply is no notion of passive clause for NL which either requires or is capable of precise explication. But this seems a pointlessly negative conclusion, for which it is difficult to see any motivation beyond attachment to a theoretical framework unable to provide the characterization in question.7 Clearly though, little can be gained from a priori philosophical discussion of the worthiness of attempting to reconstruct notions like passive clause. Any merit to a program of grammatical theory construction including this goal can only be revealed by at least three sorts of insights into assumed passive structures and related phenomena. If the goal of a universal explication of passive is valid, the resulting characterization should, first, permit the formulation of otherwise unstatable lawful characteristics of passive structures in all NLs possessing them; second, facilitate the statement of various NLspecific constraints relevant to passive constructions in diverse NLs and, third, permit uniform statement in grammars of
< previous page
page_5 If you like this book, buy it!
next page >
< previous page
page_6
next page > Page 6
recurrent constraints on the varied grammatical phenomena called passives, for example, restrictions requiring only 'short' passives, only nonreflexive passives, etc. (see Chapter 4, section 9). A universally valid analysis of passive clause obviously need not require every clause called 'passive' in the literature to turn out to be a passive clause. Given a precise reconstruction covering a reasonable portion of alleged passive structures, it would be appropriate to let it and supporting facts decide in limited cases that certain claimed passive clauses are not passive, provided the framework offers adequate alternative analyses for these. Chapter 5 argues that certain clauses taken as passives in recent work on Turkish, Sanskrit, Lithuanian and other NLs are not true passive clauses, in a sense later clarified. The aim of a universal description of passive structures must be seen as part of the broader goal of providing a precise, well-defined structure for every sentence in every NL. The development of a theoretical notion of passive clause must be part of an overall theory restricting the objects which can be NL sentences. The goal is ultimately to formulate a grammatical theory articulated enough to determine theorems about an individual clause C, of the form: 'If C is not passive, then ...', where the dots cover the restricted class of other possibilities recognized by the theory. If we take clauses to be analogous to chemical compounds, the goal of explicating passive clauses should be understood as part of the broader task of providing a complete 'chemistry' for clauses, such that every clause in every NL is a model of some clause type that the theory allows, with lawfully determined consequences for the properties of each type. Work in RG beginning in the early seventies adopted the goal of formulating a grammatical theory that included a precise reconstruction of the notion passive clause (see Perlmutter and Postal (1977, 1983a, 1984a, 1984b), Perlmutter (1982, 1983b, 1984a, 1984d)). This work provides initial indications that such a goal can justify itself in the ways mentioned above (see Chapter 2). RG adopted the assumption that traditional notions of passivization were essentially valid, needing chiefly to be clarified and embedded in an explicit overall grammatical framework. What is the traditional conception here? Such accounts speak with relative consistency of a passive clause as one in which 'a direct object has become a subject' and with perhaps slightly less consistency as one in which a subject fails to appear or appears in some 'oblique' function. See Jespersen (1924: 164), Grevisse (1969: 561) and Sweet (1891: 112-113). An especially clear traditional account is given in Kurylowicz (1946: 5): The majority of languages have available to them the possibility of using the passive construction, that is, of changing the direct object into the subject, the subject into an oblique case, generally the instrumental, and the active verb into a passive ...
< previous page
page_6 If you like this book, buy it!
next page >
< previous page
page_7
next page > Page 7
The weakness of such characterizations 8 is that all the key notions, 'direct object', 'subject', 'oblique', 'change', 'become', etc., were left unanalyzed. Given the dominance of transformational grammar in recent decades, it would be natural to inquire into the possibility of explicating passive in terms of its notions, specifically in terms of the concept (NP) movement rule. But it is not hard to show that no such characterization has any chance of success (see Perlmutter and Postal (1977, 1984a, 1984b), Perlmutter (1982, 1984b)). This follows in particular since some NLs manifest word orders for corresponding active and passive which are systematically identical for the relevant constituents (see the discussion of Quiché in Chapter 2). Chomsky (1981: 120-121) in effect now admits that the notion passivization cannot be reconstructed in terms of transformational movement.9 RG work showed how an initial reconstruction of the traditional idea of passive clause can be developed and made precise in a straightforward way in terms of this framework's general conception of sentence structure as characterizable in terms of Relational Networks (RNs). These are sets of objects called ARCS, meeting certain conditions. In this framework, a clause is passive if and only if the node defining it is the tail of three arcs in an RN substructure like:
Here the traditional notions of 'subject', 'direct object', etc., are taken as primitive relations represented by a set of primitive R-SIGGS which are their names and which enter into formal linguistic structures as labels on arcs. R-signs relevant for the following discussion include:10
< previous page
page_7 If you like this book, buy it!
next page >
< previous page
page_8
next page > Page 8
An arc involves a pair of primitive NODES representing linguistic elements, an R-sign naming the binary grammatical relation in which the elements stand, and a finite, nonnull continuous sequence of integers called COORDINATES, which represent linguistic levels. In [4] and certain following structures, Greek letters indicate arbitrary coordinate sequences, which are possibly null. The interpretation of an arc is that its head node bears the grammatical relation named by its R-sign to the tail node at all the levels specified by the coordinates. It is convenient to be able to refer to linguistic elements corresponding to the heads of various arcs in terms of the R-signs of those arcs. For instance, I will speak of 'a 1', meaning a substructure corresponding to the head of a 1 arc. Since the same node can head more than one arc, such characterizations are not necessarily unique. For example, the same node can be both a 1 and a. 2 of the same clause; see 'below for many illustrations. The traditional notion of passive clause has then been reconstructed in RG terms as one in which a linguistic element (formally, a node) a is a 2 at a level containing a distinct element b as a 1 and in which a is a 1 at a subsequent level; at this subsequent level, the original 1 is not a 1; in typical cases it is an 8. 11 Basic to this account, and not treated in very early and thereby inadequate RG discussions of this topic, is the notion that a passive clause not only involves advancement of a 2 to a 1, but also the existence of a distinct earlier 1. Clauses involving 2 to 1 advancement (21A) without the latter feature exist, but are not passive in the technical sense being reconstructed. They are said, in the current RG literature, to involve UNACCUSATIVE ADVANCEMENT (see section 4, Chapter 5 and Dubinsky and Rosen (1983)). 1.3 Impersonal Passive Clauses It is not hard to show how the approach of section 1.2 reconstructs the traditional idea of passive clause in the desired way in a wide range of standard and relatively uncontroversial cases, as discussed at some length in the RG literature; see Dubinsky and Rosen (1983) for relevant references. I will not be directly concerned with this here. Rather, my interest in this study is to focus on more difficult cases, where it is not obvious how the suggested characterization can pick out the right clauses as passives while accounting
< previous page
page_8 If you like this book, buy it!
next page >
< previous page
page_9
next page > Page 9
for their properties. In particular, a major challenge to the RG account, indeed to any attempted universal explication of passive structures, is provided by the fact that many NLs, English notably not among them, 12 have structures typically categorized as IMPERSONAL passive clauses. Instances are found in Indo-European, e.g., in Czech, Dutch, French, German, Latin, North Russian and Welsh and in non-Indo-European, e.g., in Ainu, Arabic, Finnish, Maasai and Turkish. Examples include:
The existence of clauses like those in [6ii] indicates that earlier examples like [3] are not only passive clauses but a subtype called PERSONAL PASSIVES. Henceforth, personal and impersonal passives will generally be respectively abbreviated as PPs and IPs. At this point, both of these notions are informal. Later they are made precise in such a way that some clauses initially called 'impersonal passives' turn out not to be IPs in the technical sense. Obviously, the notion IP involves any obscurity of the more general concept passive clause, plus additional unclarities of its own. Roughly though, referring to [6], one can characterize IPs initially as clauses which are in some sense passive, but in which no 'inherent' nominal occurs as superficial 1. Thus IPs share with PPs the property that a nominal which would be 'expected' to occur as 1 does not and either is not visible at all or occurs in some 'oblique' function. This is the case of the 'agent' nominals in [6aii, bii]. But, in contrast to PPs, the closest active (nonpassive) correspondent of an IP does not contain a non-1 corresponding to the superficial 1 of
< previous page
page_9 If you like this book, buy it!
next page >
< previous page
page_10
next page > Page 10
the IP. There are two superficial subtypes of IPs, those having a visible dummy nominal as surface 1, as in Dutch [6bii], and those in which this is not the case, as in Latin [6aii] and Arabic [6cii]. Both subtypes can exist in a single NL, as illustrated by German: 13
The existence of IPs has frequently been cited as a ground for rejecting a universal view of passives like that sketched in (4), for instance in Comrie (1977), Jain (1977) and Keenan (1975). This rejection has been particularly stressed for IPs lacking visible superficial 1s; such an absence has often been taken as self-evident grounds for rejecting the notion that all passives involve an element which is a 2 as well as a 1. However, these criticisms have very little force, as stressed in Perlmutter and Postal (1984b). For the possibility exists that IPs can be subsumed under the characterization in [4] if they involve the advancement of a dummy 2 to 1, in some cases of an invisible dummy nominal. This is the general RG position. Perlmutter and Postal (1984b) show that even for IPs without visible 1s there are good arguments for the existence of a dummy as both 2 and 1. Following RG work, I assume that this is uniformly the case, both for IPs in NLs where no IP has a visible dummy, and for IPs in NLs where no clause of any type has a visible dummy. Perlmutter and Postal (1984b) motivate such a description for Welsh, which never has visible dummies in IPs. The assumption that all IPs involve superficial 1s is, of course, a sine qua non of any attempt to subsume IPs under the RG characterization of passive clause in terms of substructures like [4]. I will return in passing to matters relevant to this issue. An idea sometimes expressed or hinted at (see Bresnan (1982a: 81; fn. 1), Chomsky (1981: 125-126), Marantz (1981: 159)) is that IPs are passives of intransitive structures. While this is certainly true extensionally in certain specific NLs, and while there may be something intensionally valid to be said universally along these lines, taken as a definition, such an account seems to fail in both directions.14 First, there are passives of intransitives which are not IPs. English PSEUDO-PASSIVES like (8b) are typically based on intransitive predicates and yet are PPs; see Chapter 6:
Second, IPs are apparently not lawfully limited to intransitive predicates. At least three different situations seem to be indistinguishable. For example, certain varieties of French truly do not allow IPs of transitives, even those with invisible unspecified 2s:15,16
< previous page
page_10 If you like this book, buy it!
next page >
< previous page
page_11
next page > Page 11
[9c] has all the morphological characteristics of what would be the IP corresponding to [9a]. But it cannot be so interpreted, having only a structure under which il is an anaphoric pronoun corresponding to the initial 2 of the predicate; that is, [9c] is just the PP corresponding to [9b]. While German is like French in not allowing IPs of transitives with expressed 2s, 17 the German analog of [9c] is ambiguous:
That is, [10c] is interpretable either as an IP, (paraphrasing [10a]), or as a PP (paraphrasing [10b]). Finally, NLs like Welsh (see Awbery (1976) and Chapter 2, section 1.2 below) allow IPs of both intransitive and transitive predicates quite generally, including transitives with superficial expressed underlying 2s (see example [2.5c] below). A central goal of this study is to expand and refine the characterization of both PP and IP given in RG terms so far, making use of the richer and more articulated conception of sentence structure provided by the APG framework. It will be indicated in particular why the RG account provides only a necessary condition for IP hood and then be shown how the basic ideas of APG permit formulation of a sufficiency condition. This yields a precise characterization of the notion IP as a subtype of passive clause. PPs are then, trivially, simply passives which are not IPs. 1.4 Relevant Aspects of Arc Pair Grammar 1.4.1. Remarks Because the range of APG ideas is extensive and because these ideas are very different from currently standard grammatical views, I will only be able to
< previous page
page_11 If you like this book, buy it!
next page >
< previous page
page_12
next page > Page 12
sketch some bare outlines here. For fuller discussion, see JP (1980), Postal (1982). The APG framework involves unique conceptions of SENTENCES and GRAMMARS and also of the FUNCTION that any grammatical framework needs to associate with each grammar viewed as a formal syntactic object, a unique NL, namely, that NL the grammar is said to characterize or specify. 18 I begin with matters relevant to the latter two areas, turning subsequently to the APG notion of sentence. 1.4.2. Nonconstructive Grammars The key APG idea is that grammars are NONCONSTRUCTIVE; they do not consist of some Turing-machine analog and are not viewed as devices which 'assemble' sentences. Technically, APG grammars are not GENERATIVE grammars, but characterize the collections of sentences which are NLs in something like the way axiom systems in mathematics and logic characterize collections of objects via the notion of certain objects satisfying the axioms. The technically nongenerative character of APG grammars is a crucial virtue, since Langendoen and Postal (1984) presents a demonstration that no NL has a generative grammar. Conceptually, the nonconstructive specification of an NL involves the three elements in [11]:
Of these, the first two are of course really part of grammatical theory. It does not matter whether one uses 'grammar' in the broad sense to refer to all of [11a,b,c] or narrowly just to refer to [11c]. [11a] defines the collection19 of all linguistic objects in the most general sense; in JP (1980) these are formally reconstructed as objects called PAIR NETWORKS (PNs). The principles in [11b] define a proper subcollection of the collection of all PNs; call it UNIVERSAL SENTENCE (US).20 US is the collection of all PNs satisfying every sentence law. Each member of US is a sentence in some NL, although not necessarily an attested or even attestable one. From this point of view, an NL is simply some subcollection of US. The grammar of a particular NL, K, in the narrow sense is just a device to pick out a subcollection of US by excluding all those members of US which are not in K. Consequently, every rule has one and only one function: to specify certain members of US as not being elements of the NL whose grammar contains the rule in question. An APG nonconstructive grammar is assumed to take the form of a set of STATEMENTS, formulae to which truth values can be assigned. I assume that each is conditional in form. More precisely, each rule can be regarded as a material implication. Put differently, each rule is an object identical in logical character to the sentence laws, only interpreted more narrowly. While sentence laws determine membership in US, and thus help define sentence-
< previous page
page_12 If you like this book, buy it!
next page >
< previous page
page_13
next page > Page 13
hood for all NLs, rules only define sentencehood for particular NLs. One can in fact view a grammar as a single statement instead of a set of such statements, namely, as the logical conjunction of all the rules in the set. From this point of view, the NL defined by such a grammar is composed of all and only those members of US satisfying the (huge) single statement in question. When this last remark is made precise, as in JP (1980: Chapter 14), Postal (1982), Langendoen and Postal (1984), it provides the function to interpret grammars, which was mentioned earlier. This framework essentially reduces the mechanisms whereby a grammar characterizes an NL to the most basic principles of logic and set theory. Within the framework of nonconstructive grammar, there is no motivation for RULE SCHEMATA, almost universally appealed to in versions of constructive grammar. This conceptual complication can be totally eliminated in the APG framework, with grammars of attested NLs displayed directly rather than generated from displayed metagrammars. 21 Evidently, in this nonconstructive approach, none of the grammatical devices normally appealed tobase rules, transformations, interpretive rules, filters, etc.can exist. Similarly, there is no lexicon, lexical rules, lexical entries, etc. Every rule is simply a condition on allowable structures. The discussions that follow specify a number of proposed rules for various NLs. Each is of the character just sketched and should be understood as determining a subcollection of US, as follows. For any rule R, some member of US, M, is in the NL defined by the grammar including R only if R is true of M (M satisfies R). This 'only if' turns into 'if and only if' when one generalizes over all the rules, or, more simply, takes the grammar to be the conjunction of all the rules. Grammatical rules in the APG sense are, syntactically, strings in a vocabulary of logical elements ( , 'Not', etc.) and nonlogical elements. The chief formal constraint on such rules beyond their characterization as conditional formulae is that the nonlogical vocabulary be limited to certain notions defined by the theory. This means that the nonlogical vocabulary will be a subset of APG concepts, that is, various notions referring to arcs and to the primitive and defined relations between them (LOCAL SUCCESSOR, etc.; see below). The need for constraints on APG rules is more limited than one might assume, however, since, in particular, grammatical rules bear absolutely none of the burden of distinguishing NL sentences from objects which are not NL sentences. This task is entirely accomplished by the definition of 'sentence' and by the sentence laws. Thus the fact that bananas, sonatas, arbitrary programming statements, Queen Elizabeth's telephone number, etc., are not sentences of, e.g. English, is specified entirely independently of the rules of English by virtue of the presumed fact, which must be guaranteed by the definition of 'sentence' and/or by the sentence laws, that these objects are not members of US. Put differently, the grammar of English only needs to separate the sentences of English from bona fide sentences in
< previous page
page_13 If you like this book, buy it!
next page >
< previous page
page_14
next page > Page 14
other NLs; it does not have to distinguish them from non-NL sentences. The latter task is fully accomplished by grammatical theory. 1.4.3. Some Rule Notations R-signs fall into classes which define relevant classes of arcs. For instance, I and 2 define the class of NUCLEAR TERM ARCS, 1, 2, 3, the set of TERM ARCS, 1 the set of 1 arcs, etc. It is useful for stating rules to permit such ARCSET NAMES to be affixed to various APG relations between arcs and to properties of arcs. I do this informally by using the sign '=' as a suffix or a prefix. Illustrating with the Local Successor relation, characterized below, which is basic to the substantive discussion that follows, notations like [12] will be used:
With A, B, etc., taken as variables over arcs, the inscriptions in [12] are strictly speaking ill-formed, since the variables are apparently not bound. However, all such pseudo-free variables are interpreted as bound by universal quantifiers with maximally wide scope; this imposes the convention of JP (1980) of SUPPRESSNG wide scope universal quantifiers. It is extremely useful to be able to systematically convert APG binary relations between arcs into properties of one or the other argument. One can do this via the notations '+' and '+ +', which pick out respectively the first and second arguments of relations.
Without these notations, [13a] would have to be given as: Thus the '+' and '+ +' operators permit suppression of many existential quantifiers as well as of the variables they bind. Since these operators define properties, they can combine with the arc class prefixes to yield notations like [15]:
PREDECESSOR is the converse of Successor. The notations and conventions mentioned here jointly permit specification as [16a] of what would otherwise require formulation as [16b], and whose meaning is given verbally in [16c]:
< previous page
page_14 If you like this book, buy it!
next page >
< previous page
page_15
next page > Page 15
Several rules proposed below are parallel in form to [16a]. Some sense of the clumsiness which the notations help to avoid can also be obtained by considering the rules in Postal (1982), which do not use it. 1.4.4 Pair Networks While logically independent of the nonconstructive conception of grammar, the APG analysis of sentences as PNs is particularly suited to such an account since PNs are claimed to provide a mechanism for embedding all of the grammatical information in a sentence in a single well-defined formal object. It is then a conceptually straightforward question whether such objects do or do not satisfy a given nonconstructive grammar. The notion PN is a development of the earlier RG idea of RN as a set of arcs. Arcs are based on three types of primitives, R-signs, COORDINATES, and NODES. Nodes are of two types, NONTERMINAL and TERMINAL. Nonterminal nodes represent all the abstract elements of sentences: clauses, phrases, etc. They have been identified with a set of integers, say those greater than 20. 22 Terminal nodes are of several subtypes and represent the more substantive elements of sentences: phonological primes, grammatical categories, and semantic primitives. In the framework of JP (1980) nonterminal nodes occur only as heads of arcs with the R-sign L, standing for the grammatical relation Labels. It is most often convenient to suppress L arcs notationally, compressing them together with the arcs they are branches of:
< previous page
page_15 If you like this book, buy it!
next page >
< previous page
page_16
next page > Page 16
An RN in the sense of RG work is a definable subpart of each PN. But PNs involve crucially the idea that there are two primitive binary relations between arcs, called SPONSOR and ERASE. Informally, the assumption is that each arc has a 'creator' arc, its sponsor, and that some arcs, namely, those that have no superficial existence, have a 'destroyer' arc, their eraser. Given the constraints on space, I can hardly do better in introducing these concepts than to consider the relevant structure in such terms of an example like [18]: The relevant PN for this is, to a reasonable level of detail, that in [19]:
The following notational conventions are, inter alia, relevant to the interpretation of [19] and of later representations of PNs. The Sponsor relation is represented by a wiggly arrow connecting arcs, with the sponsoring arc at the tail; the Erase relation is represented by a double arrow with the erased arc at the head. A double arrow with internal Xs is a notational
< previous page
page_16 If you like this book, buy it!
next page >
< previous page
page_17
next page > Page 17
device for collapsing one sponsor arrow and one erase arrow. The latter notation, not used in JP (1980) or in Postal (1982), greatly simplifies graphic representations of PNs. It is interpreted to mean that the arc at the head both sponsors and is erased by the arc at the tail. Hence, these notations are to be read as in [20]:
Further, the various triangles in [19] indicate subconstituents whose structures are artificially not represented. Thus the triangle with its top at node 50 indicates that the word Melvin corresponds to a nominal which plays several roles in two different clauses, but whose internal structure is not specified. Triangles are thus an official indication that the structure of some phrase is not given. The PN in [19] specifies that English [18] involves a passive clause embedded as a complement of an initially intransitive V of the 'raising' sort; in fact, speaking informally, 'raising' is obligatory for this V. 23 The representation in [19] artificially ignores the structure of the passive auxiliary be and of the element to. However, I assume that both these are Vs24 having essentially the properties of tend, so that what is left out is just two repetitions of the structure defining the upper clause in [19].
< previous page
page_17 If you like this book, buy it!
next page >
< previous page
page_18
next page > Page 18
Let us concentrate on the lower clause, defined by node 200. This manifests the structures defining both passives and what is called in RG terms 3 TO 2 ADVANCEMENT [32A]. For previous detailed discussion of this phenomenon in many different NLs, see for example Aissen. (1979, 1983, to appear), Chung (1976a), Crain (1979), Fauconnier (1983), Franz (1980), Gibson (1980), Gibson and Raposo (to appear), Hermon (1981), Johnson (1977, 1979), Joseph (1982a), Perlmutter and Postal (1983a), Postal (1982) and many other references in Dubinsky and Rosen (1983). The structure of the lower clause is characterized by the fact that I, a 3 arc, sponsors the 2 arc H. I and H are distinct, but share the same head node. Arcs sharing the same head node are said to overlap. Distinct overlapping arcs represent the fact that a single element bears more than one distinct grammatical relation. The relation between H and I is an instance of a fundamental APG notion particularly relevant to what follows: the SUCCESSOR RELATION. A is a successor of B if and only if B is a distinct overlapping sponsor of A. The successor relation is inter alia the APG reconstruction of that lingustic reality which transformational grammar has attempted to capture via the notion movement transformation. Successors can be either of two types, depending on whether the pair share the same tail node or not. Arcs which share tail nodes are said to be NEIGHBOURS. Distinct neighboring arcs represent instances where distinct elements bear relations to the same linguistic elements. If a successor/predecessor pair are neighbors, one speaks of LOCAL Successors, if not, of FOREIGN successors. More generally, local appended to any relation between arcs indicates that those arcs are neighbours, while Foreign indicates that they are nonneighbors. Hence in [19] H is a local successor of I. B, on the other hand, is a foreign successor of G. So-called 'raising' is, evidently, a special case of a foreign successor structure. The existence of arc I in [19] indicates that Melvin bears the same initial relation in the complement clause as it does in [21a] and the existence of H indicates that it bears the same noninitial relation in the complement as it does in [21b]:
It is a 2 = 3 local successor relation of the sort holding between H and I which characterizes the similarity between pairs like [21a,b], that is, which represents the fact that although Melvin is a superficial 2 in [21b], at another level it bears the same [3] relation that it bears nonsuperficially in [21a]. Similarly, the 8 = 2 local successor pair (K,J) represents the fact that, although a superficial non-2 (in fact, 8) in [21b], gifts initially bears the same relation, namely 2, as in [20a]. For background on this multilevel view of grammatical structure, see for example Perlmutter (1981, 1982, 1983b, 1984a, 1984b) and Postal (1982).
< previous page
page_18 If you like this book, buy it!
next page >
< previous page
page_19
next page > Page 19
A systematic relation holds between the coordinates of a local successor and those of its predecessor. Namely, the first coordinate index of the local successor is always + 1 of the last coordinate index of the predecessor. This means that the local successor relation is such that an element having a local successor grammatical relation at level i does not bear at i the relation it bore at level i-1. In this way, the local successor relation reconstructs aspects of earlier accounts in which one said that e.g., 'a 2 becomes a 1', meaning that some nominal was a 2 at one level, a 1 at the subsequent level, and in addition not a 2 at that subsequent level. Informally, and following RG usage, I will say in cases of local successor pairs that the element corresponding to the head node has either ADVANCED or DEMOTED from R1 to R2, where R1 is the R-sign of the local predecessor, R2 the R-sign of the local successor. The not very significant contrast between advance and demote has to do with the relative postion of the R-signs of local predecessor and successor on the hierarchy: (1 > 2 > 3 > Obliques and 8) where 1 is highest. Thus, for example, a nominal advances from 2 or 3 to 1 but demotes from 2 to 8 or from 1 to 3. It will be noticed that some arcs, all with the coordinate c1 have no sponsors indicated in [9]. This is a purely notational device which takes advantage of the claimed regularity that all and only arcs with coordinate c1 sponsor themselves. Every arc has a sponsor, some have two, but no self-sponsoring arc has a DISTINCT sponsor. The collection of selfsponsoring arcs is extensionally equivalent to the set of arcs with coordinate c1. These are also referred to as INITIAL arcs. The collection of all initial arcs in a PN forms a graph-theoretic object called an L(OGICAL)-GRAPH. JP (1980) claims that the meanings of sentences are represented by L-graphs. The erase relation serves to pick out another, distinct subcollection of all of the arcs in the graph ('RN' in RG) defined by a PN, namely, that subcollection none of whose members are erased. These are SURFACE ARCS, and the maximal collection of such in a PN defines the S(URFACE)-GRAPH of the PN. This represents structure at a level of abstraction roughly equivalent to the surface structures of transformational linguistics. Of the eighteen explicit arcs in [19], only A, C, E, L, P, Q, R, S and T are surface arcs (unerased). PNs can represent the fact that a single element (node) can bear more than one relation (in fact, it can bear an unbounded number of them). Formally, a single element can head any number of arcs. In [19] all such cases involve successor relations. But at least one other type of multiple relation pattern exists, that in which a node bears more than one initial relation. This is claimed to be the right reconstruction of the phenomenon usually discussed in terms of what I claim is the nonexistent relation 'coreference'. Space precludes a discussion; see JP (1980: Chapter 11), Postal (1982), Perlmutter and Postal (1984b), Perlmutter (1983), Perlmutter (to appear), Rosen (1981a, 1982). In [19], the nominal Melvin is represented as
< previous page
page_19 If you like this book, buy it!
next page >
< previous page
page_20
next page > Page 20
bearing five distinct relations: it is a 3, 2 and 1 of the complement cause and a 2 and a 1 of the main clause. Of these, all but the first are successor relations and only the last is a surface relation, represented by an unerased arc, namely A. All the other arcs headed by node 50 are erased. Thus, from the point of view of such superficial facts as word order, Melvin is a main clause 1 and only that. I said earlier that the complement clause in [19] was passive. This follows from the RG analysis in [4] above, based on the triple of arcs G, H and N. One can now reformulate the RG account, translating it into APG terms. First, a passive clause contains a 1 arc local successor of a 2 arc. But this does not suffice, as already indicated. For instance, the main clause in [19] meets this condition, but one does not want it to turn out to be passive. The sufficiency condition involves the existence of an earlier 1 arc, instantiated by N in [19]. In APG terms, this can be made a little stronger by saying that the 1 arc local successor must OVERRUN some arc. A overruns B if and only if (i) they are neighbors, (ii) have the same R-sign and (iii) A's first coordinate index 1 greater than B's last coordinate index. Hence in [19], G overruns N. A passive clause is then one such that [22] holds (but see Chapter 7 for a generalization):
It correctly follows that the clause defined by 200 in [19] is passive but that defined by node 100 is not, since it fails the second conjunct. Such clauses are unaccusative advancement structures. I remarked earlier that although it heads five distinct arcs, three of them complement arcs, Melvin in [19] functioned as a main clause 1 from the point of view of word order. This raises the important issue of how constituent order is represented in PNs. Several different techniques are considered in JP (1980: Chapter 12), including the recognition of arcs representing a primitive precedence relation or a third primitive relation between arcs. But these ideas were mistaken, since constituent order is representable without additional apparatus in terms of independently existing features of PNs. One can represent order without new primitives by taking advantage of the arithmetical properties of nonterminal nodes like 50, 60, etc. It was originally assumed that the choice of numbers is arbitrary. But, by making it partially nonarbitrary, and by focusing on unerased arcs, one can take the arithmetical order to represent constituent order. 25 Let us pick some nonterminal node, like 200 in [19], which is the tail of at least two unerased arcs. The order of substrings corresponding to the heads of such arcs is given by the arithmetical order of their head nodes. For instance, the English rule saying is 1s precede Ps can now be interpreted as specifying that if A and B are neighbouring surface arcs and A is a 1 arc and
< previous page
page_20 If you like this book, buy it!
next page >
< previous page
page_21
next page > Page 21
B is a P arc, then A's head node must be less than B's head node. The linguistic notion 'precede' thus reduces to the arithmetical relation 'less than'
< previous page
page_22
next page > Page 22
1.5 A Theoretical Note Certain comments on earlier versions of this study suggest the need for the following remarks about theoretical complexity. One feature of what follows is the introduction of a number of new derivative concepts, each precisely defined in terms of the primitives and defined concepts of JP (1980). These notions are then appealed to in various grammatical rules, whose statement they greatly simplify. From one point of view, stressed by critics, these definitions represent a complication of the framework. Obviously, if one measures theoretical complexity even partly by counting definitions, each definition worsens a theory. But I see no sense to such a measure. Exactly what one wants from a primitive theoretical basis is that it permit the definition of a rich set of defined concepts suitable for elegant formulations of the principles governing the subject matter. From this point of view, theoretical definitions introduce no new complexity. They simply make terminologically explicit the notions already present in the underlying system of primitives. Any defined concept is in fact eliminable at the price of complicating theoretical discourse. If one eliminated any subset of theoretical definitions while continuing to appeal to the defined concepts, each previously defined construct could be taken as a new primitive. But this would save nothing, since the earlier definitions would have to show up as new laws governing the relations between items in the enriched primitive vocabulary. For example, [2.26a] below defines the notion free in terms of the earlier APG concept Replace. If, however, free is taken as primitive, then the content of the definition must appear as a law saying that free arcs are all and only those which are not replaced. Nothing is gained. A related point made by certain critics is that the richness of theoretical concepts suggests that anything whatever is describable in current terms. But since, as already indicated, defined terms are eliminable and only highlight notions implicit in the primitive base, there is no clear way in which defined terms extend the ability of a framework to accommodate linguistic diversity. Descriptive flexibility is limited in APG terms by a rich system of laws, which greatly constrain what is describable. One hundred and sixteen such principles are posited in JP (1980). They are all obviously to be regarded as tentative hypotheses about NL sentence structure. Many will have to be eliminated, replaced, revised or subsumed under deeper generalizations. But the general methodology of postulating such principles should be clear and unobjectionable. Perhaps some critics are assuming that only certain terms are available for the construction of APG rules, and that by increasing the class of defined concepts one is increasing the class of allowed rules. From the standpoint of the notation provided, this is true, but irrelevant. Viewed extensionally, definitions, by their very nature, do not alter the descriptive capacity of a framework. Defined terms merely permit one to describe various phenomena in a relatively more compact fashion. So, the term free mentioned earlier
< previous page
page_22 If you like this book, buy it!
next page >
< previous page
page_23
next page > Page 23
permits statement of rules which otherwise would have required more complicated statements appealing to Replace or to its defining components. Tradeoffs like these seem to be independent of the worth of a theoretical framework. Moreover, and quite crucially, for reasons considered in JP (1980: Chapter 14), the need to constrain grammatical rules felt in other frameworks does not really arise in the nonconstructive APG framework. In the latter, factual constraints are imposed directly on sentences and collections of sentences without the mediation of constraints on rules. Informally stated, it is not necessary to specify the nonexistence of a sentential phenomenon X via statements about grammars or rules if the laws of the theory simply allow no NL sentence to manifest X. My conclusion is that theoretical definitions of the sort multiply offered below should not, via a false analogy, be compared, for example, to adding new primitives or to adding new rule types or types of operations to a system which previously lacked them. Nothing in the present study represents an elaboration of the latter type, except for the occasional peripheral recognition of several primitive grammatical relations not previously noted. This recognition is theoretically parallel to the positing in other frameworks of new grammatical categories. The expansion of grammatical relations hinted at here, which should, I believe, be greatly extended (see Postal (to appear c)), pales to near insignificance when compared to the increases in recognized grammatical categories seen in other frameworks in recent years. If we assume that they are all methodologically proper and mutually consistent, the definitions offered here can be validly criticized in principle, e.g. as introducing terms that are ad hoc or without redeeming utility or as introducing terms subsumed under other terms. But the fact that there are lots of them seems to have no relevance to any theoretical or factual issue. Nor can they reasonably be claimed to increase the theoretical complexity of the underlying framework. Notes: Chapter 1 1. In recent works, Chomsky has begun to suggest that there are no such things as NLs, a position I find incoherent; see Langendoen and Postal (1984: Chapter 6) for discussion. 2. In Platonist terms, it would, of course, be unproblematic if it turned out that so-called English-speakers only knew part, not all of English. 3. More generally, 'attestable' should denote a relation between the collection of NLs and fixed types of creatures. It is, for instance, logically possible that there are NLs that are attestable for men and not for women, etc. 4. In the framework of JP (1980), each NL has an infinite collection of morphemes. The finite subcollection normally spoken of in this connection is the collection of constrained morphemes, those limited by the grammar to some specific semantic and/or syntactic properties. This finiteness is a trivial nonlinguistic feature of attestable NLs.
< previous page
page_23 If you like this book, buy it!
next page >
< previous page
page_24
next page > Page 24
5. The indefinite article is deliberate. There is no guarantee that there is a unique correct grammatical theory. 6. I say 'generally' because in the complements of the so-called causative construction, no auxiliary (or past participle) is possible in what are arguably passive (complement) clauses:
7. Chomsky (1977: 159) expresses considerable doubt about the validity of a cross-linguistic notion: 'But ''passive" does not seem to me a unitary phenomenon either in a single language or across languages.' Similar negative remarks appear in Chomsky (1981: 120-121). 8. Zribi-Hertz (1980: 5) finds Grevisse's (1961: 561) account of passives '... empty of any interpretation of the facts ...' because he subsequently notes exceptions to his generalization. I find this rather unjust. 9. Later in the same work (Chomsky (1981: 128-135)), the author tries to develop an account of Japanese passives which is abstractly parallel to that for English passives. This is attempted despite his assumption that Japanese is a 'nonconfigurational' NL and, in particular, has base rules which introduce no VP node. Chomsky's common account of English/Japanese passives depends on talking about the same grammatical relations in both NLs. This renders the account incoherent as far as I can see. For Chomsky assumes such relations to be defined; however, under his assumptions, Japanese lacks the structures to properly invoke the universal definitions of the relations he assumes for it. 10. The last two relations, introduced in JP (1980), are relevant for describing prepositional phrase structures; see Chapter 6. 11. In RG work, the original 1 is always taken to be an 8. This was, for example, imposed by a law variously named 'The Relational Annihilation Law' and 'The Chômeur Law'. In the framework of JP (1980), this law is not imposed, and the original 1 is sometimes a later 8, sometimes a later 3, sometimes bears a different later relation and sometimes (in so-called 'short' passives) bears no later relation at all. 12. In my view, while English lacks superficial clauses which are impersonal passives, it does have nominals which are nominalized impersonal passive clauses, e.g.:
Such an analysis requires serious support, of course, but I believe that this is possible. A major argument is that the class of possible nominals like (ia) is determined by the same principles as determine the class of impersonal passives; in particular, unaccusative predicates are no better in English nominals like (ia) then they are in, e.g., French impersonal passive clauses:
< previous page
page_24 If you like this book, buy it!
next page >
< previous page
page_25
next page > Page 25
13. Nerbonne (1982) argues that the form es in German clauses like (7a) is not a nominal. While such forms do have unique properties, I am unconvinced of his conclusion. In any event, the claim in the text is not in jeopardy. It can, for example, be illustrated by alternations in French subjunctive clauses:
In these complements, the dummy nominal which normally manifests as the clitic il can optionally not appear. Such cases are discussed in Chapter 4. 14. A claim that IPs are not possible with transitive predicates might be buttressed through an appeal to the distinction between true IPs and impersonal PPs drawn in Chapter 4. 15. The literature sometimes gives the impression that French lacks IPs. This quite erroneous idea may be based on the fact that such clauses belong only to a rather formal and certainly noncolloquial style. 16. The claim that French does not allow IPs of transitives with invisible, unspecified 2s is true for the consultants I have worked with. It is not, however, true of the French of Zribi-Hertz (1981), whose dialect is, along this parameter, like German. That is, in this variant of French, [9c] has an IP reading. 17. Well-formed passives like (i) do not falsify this claim (data from Smith (1982)):
For such can be regarded as 'extraposed' versions of PPs, the 'extraposition' being the same phenomenon which relates the nonpassives [iia and b]:
18. Avoidance of the term 'generate' here is deliberate since, as demonstrated in Langendoen and Postal (1984), NLs have no generative grammars and cannot be generatedi.e., recursively enumerated. 19. I use 'collection' instead of 'set' for essentially the same reasons cited in the previous note. The argument shows that NLs have too many sentences for their sentences to comprise sets. 20. This term, not used in JP (1980), is introduced in Postal (1982). 21. For discussion of metagrammars and their role in generating infinite grammars, see Langendoen (1976, 1979), Gazdar (1981a, 1982), Gazdar and Pullum (1981) and Pullum (1982a). 22. Integers smaller than 20 are reserved for other roles, e.g., representing certain R-signs. 23. Suppose one introduces the following notion:
< previous page
page_25 If you like this book, buy it!
next page >
< previous page
page_26
next page > Page 26
Then the constraint referred to in the text can, I think, be essentially represented as:
24. For arguments for the verbal character of infinitival to, see Pullum (1982b). 25. Taking nodes as integers works only if all sentences are finite. To make the present approach to constituent order consistent with the conclusion of Langendoen and Postal (1984) that there are transfinite sentences of every cardinality, the colleciton of nodes must be expanded to include transfinite cardinals, over which '
< previous page
page_42
next page > Page 42
This data differs from that for my informants in two crucial respects. First, the speakers Kayne describes allow plain passives with (dés)obéir, (Ae and f), but not reflexive passives, that is, not (Ah and i); see Chapter 4. Second, speakers referred to by Kayne do not permit this V in the 'object raising' construction illustrated by (Ag). My informants always accept the latter and have a strong tendency to accept the reflexive passive cases. Evidently then, two different grammars are involved which must differ in particular with respect to rule [23]. While this is adequate for my informants, it allows too many sentences with (dés)obéir for the speakers underlying Kayne's account. I suggest that the dialect described by Kayne can be characterized by replacing rule [23] with:
The notion Copies can be defined in terms of the concept 'Copy arc' of JP (1980: Chapter 11); see Chapter 4 for discussion of this idea.
The point is that in reflexive passives like (Ah and i), the members of the 1 = 2 local successor pair defining the clause as passive jointly sponsor a 2 arc replacer of the 2 arc predecessor. Since the dialect described by Kayne precludes such clauses and only allows nonreflexive passives with (dés)obéir, the final conjunct of the consequent of (B) is necessary. In the dialect Kayne describes, clauses with 32A and (dés)obéir are 'saved' only in one way: by advancement of the 2 to 1 in a nonreflexive passive. In the dialect described earlier, several other structures can 'save' such advancement cases.
< previous page
page_42 If you like this book, buy it!
next page >
< previous page
page_43
next page > Page 43
The more restricted French attested by Kayne is equally describable in the current framework. But the extra restrictions make it less interesting, since it does not allow the sentences which strongly justify, for the dialect described in section 2.3.2, the claim that the condition involved requires reference to the notion '1 arc R-successor', that is, an ancestral relation whose extension covers both 1 arc local and foreign successors and R-successors. In the dialect Kayne describes, the relevant successors must be local successors, and the phenomenon shows far less. But, the fact that the latter dialect only supports a less interesting argument for the present framework does not, of course, indicate that it is in any way problematic for this framework. 2.3.3 A Restriction in Ancient Greek Consideration of the French rule proposed in [23] offers some potential insight into certain facts of Ancient Greek taken by Feldman (1978) to cast doubt on the RG reconstruction of 'passive clause'. While Ancient Greek allows PPs of 2s, it does not permit clear instances of 32A, since, according to Feldman, underlying 3s cannot appear as superficial 2s. Nonetheless, there are apparent PPs where the underlying 3 appears as 1 of the passive. To quote (Feldman (1978: 500)): Greek can also passivize on a Dative IO in a ditransitive clause using exactly the same strategy. In this case, the IO becomes Nominative and the DO remains Accusative.
Given these facts, Feldman (1978: 500) concludes: The process exemplified in [6] must therefore involve the advancement of an IO to SU without any intermediate step in the derivation. Note that the DO in [6] is entirely unaffected by the transformationthe passive sentence is transitive, contrary to P&P's claim cited in [3] above. The process thus lacks two of the three characteristics of the Universal Passive, and [6b] must not be a passive form at all. Yet to all appearances it is Passive: the Agent has been deleted, another argument is in the Nominative case and triggers verb agreement, and the verb has 'passive' morphology.
< previous page
page_43 If you like this book, buy it!
next page >
< previous page
page_44
next page > Page 44
However, Feldman's conclusions are not imposed by the factual situation he describes. Suppose Ancient Greek did permit 32A, so that the passivized original 3 in Feldman's [6b] is, contrary to his claim, a 2 at an intermediate level (in APG terms, there is a 2 arc local successor of the 3 arc, the 2 arc having the 1 arc local successor defining the clause as passive). But assume that 32A in Ancient Greek is masked, just as 32A is in French with Vs like (dés)obéir, via the rule [27]:
This would account for the possibility of passives like Feldman's [6b] while not allowing his *[7]. The claim would be that the earlier 2 in cases like his [6b] is a final 8, even though marked accusative. This kind of' accusative case marking of 8s which realize earlier 2s in NLs where 2s are marked accusative is independently attested, even in cases where there are good arguments that the accusative marked 8 is not a final 2 (see Perlmutter (1982), Perlmutter and Postal (1984b)). That is, the only evidence that Feldman offers that állo ti méìdzon is a final 2 in his [6b] is that it is marked accusative. But such case marking facts have proved to be an unreliable guide even to superficial grammatical relations. 15 Of course, the analysis I am proposing for Ancient Greek could be wrongthere could be some evidence against an intermediate 2 stage, etc. But given the possibility of rules like [27], the facts cited by Feldman alone cause no inherent problems. This possibility is supported by the circumstances in French strongly motivating a rule like [23]. One issue raised by the discussion of Ancient Greek is whether it is ever possible for a 3 arc to have a 1 arc local successor. If a law blocking this could be maintained, then cases like that cited by Feldman and the Japanese situation mentioned in note 15 of this chapter would clearly require the kind of prior advancement to 2 1 have suggested. But such a law seems impossible to maintain. First, various facts in certain Bantu NLs seem to undermine it. There are cases where there is evidence distinguishing between final 2 status and final 8 status for the earlier 2 in a passive-like clause in which an earlier 3 is final 1. In particular, earlier 2s in Kinyarwanda seem to behave like final 2s in such cases; see the discussion in Perlmutter and Postal (1983a). Further, more recently, Jake (1983) has argued convincingly that Quechua has a class of clauses in which 3s directly advance to 1. The evidence consists of showing that earlier 2s in such clauses are also final 2s, which would be incompatible with advancement to 1 via an intermediate 2 stage (see Chapter 7). One should, I think, develop the following perspective. Since, for example, the cases considered by Jake involve the same 'passive' morphology as standard 21A passives, there is no a priori reason why the cases considered by Feldman could not be true 31A structures. But even if they were, this would not really show that the approach of Perlmutter and Postal
< previous page
page_44 If you like this book, buy it!
next page >
< previous page
page_45
next page > Page 45
(1977) or that of the present work is seriously in error. The right conclusion is, I think, simply that what have so far been here called passive clauses, that is, those with 21A, are simply a special case of a more general class of clauses in which the R-sign of the predecessor varies over 2, 3, etc. That is, one can define the notion GENERALIZED PASSIVE CLAUSE, just as we have 'Passive Clause', redefining the latter as that special case of the former where the predecessor arc is a 2 arc. It would then have to be said that in such and such NLs the only generalized passive clauses are those where the predecessor is a 2 arc. All the basic notions carry over. This sort of generalization is motivated further and carried out in Chapter 7. Hence while I have argued that Feldman has not strongly supported the case for 31A in Ancient Greek, even if he had, not much would follow. 2.3.4 A Restriction in Modern Greek The discussion of 31A at the end of the previous subsection connects naturally to a recent discussion of Modern Greek (MGreek hereafter) in Joseph (1982a), which argues that MGreek has instances of 31A. I argue here that this conclusion is not necessarily imposed by the material Joseph presents. This position is supported through an alternative generalization which captures the principles Joseph has uncovered. Joseph considers paradigms based on the single V didásko 'to teach'. This occurs in three different active-voice patterns:
Joseph takes the distinction between [28a and b] to involve alternate morphological realizations for the marking of what he takes to be the initial and final 3, Yani. My guess would be that [28b] involves demotion of the initial 3 to a relation with no traditional name. This is called QUASI-OBJECT in my recent work on French (see Postal (to appear a,c)), where its R-sign is taken as 6. 16 If so, [28b] involves 3 to 6 demotion [36D]. This is irrelevant for what follows. Perhaps the same analysis is right for the English Vs mentioned in note 18 of Chapter 6. However, naturally enough, Joseph takes [28c] to involve 32A, with the initial 3 ending up as final 2, accounting initially for its accusative case marking. More tellingly, there is, according to Joseph, a mechanism in MGreek which permits 'cross-indexing' nominals with an accusative clitic
< previous page
page_45 If you like this book, buy it!
next page >
< previous page
page_46
next page > Page 46
pronoun, a possibility which is, he states, limited to final 2s. These facts are illustrated in [29]:
The point is that just those nominals permit clitics which would be final 2s, in particular, ton Yáni in the putative 32A structure [29a]. Consider now the passives associable with [28]:
As can be seen, these three expressions (two are compressed in [30a]) correspond exactly to the expected PPs from the three cases of [28], if [28c] involves 32A. However, problems arise. First, while according to Joseph, only some speakers permit the putative 32A case [28c], all speakers allow [30b]. This situation is reminiscent of the restriction described by Feldman for Ancient Greek, discussed in the previous subsection. From this, Joseph suggests that the relevant V permits 31A advancement for all speakers, 32A only for some. Moreover, there appears to be an even stronger argument for a 31A analysis of the MGreek V for 'teach'. As already indicated, accusative clitics seem only to be possible with finals 2s. But, as Joseph observes, an accusative clitic is possible in an analog of [30b]:
< previous page
page_46 If you like this book, buy it!
next page >
< previous page
page_47
next page > Page 47
Here the clitic corresponds to the initial 2 of the clause. If [30b] and [31] involve 32A with subsequent passivization, then that nominal would be a final 8, and the clitic in [31] would apparently be anomalous. If, however, [30b] and [31] simply involve 31A, the initial 2 is a final 2 even in the passive and the clitic possibility follows from the general rule for accusative clitics Joseph states in terms of final 2hood. These facts appear to show, and Joseph takes them to show, that MGreek permits both passivization and 31A. While this may be correct, I want to point out an alternative. Consider a hypothetical NL that is not like what Joseph takes MGreek to be, that is, an NL which permits no 31A at all. Then, in APG terms, it needs as a rule or theorem of some rules an analog of the principle [25] above, repeated: Since, as previously remarked, Jake (1983) shows Quechua to have 31A (see Chapter 7), the grammar of Quechua will differ from that of English inter alia in containing neither [32] nor anything which entails it. Now I judge from Joseph's discussion that the postulated 31A in MGreek occurs at best only with didasko *; in general, the 3 of an arbitrary V has no possibility of advancing to 1.31A is thus impossible for the 3s of intransitives. Hence the 31A analysis requires complicating [32], to a rule which allows 31A only with this single predicate. This conclusion is drawn to indicate that Joseph's proposal is not free in contrast to an analysis which rejects 31A and allows only passive and 32A in MGreek. The key issue is the handling of accusative cliticization. Obviously, if Joseph's hypothesis that this is limited to determination by final 2s stands, the analysis I am suggesting, namely, that [30b] and [31] involve 32A cannot. Hence I consider an alternative to Joseph's assumption, and hypothesize that the cliticization in question is determined by a special case of a notion defined in previous RG work, so-called ACTING 2S; see Perlmutter and Postal (1984b), Perlmutter (1982). Roughly, an acting 2 is a nominal which is a 2 at an earlier level and not a distinct TERM at a later level. I will define a closely related notion in APG terms as a property of arcs:
< previous page
page_47 If you like this book, buy it!
next page >
< previous page
page_48
next page > Page 48
Since 'Term arc' here varies over 1, 2 and 3 arcs, [33] is a compressed definition of three distinct notions. Basically, [33] picks out those term arcs whose realizations in a clause do not head later stratum distinct term arcs. We are interested here in the case where Term = 2. In these terms, acting 2 arcs include all 2 arcs with 8 arc local successors. Hence, in a standard 32A structure, both the 2 arc local successor of the 3 arc and the earlier 2 arc are acting 2 arcs. However, in any passive, the prepassive arc, although a 2 arc is not an acting 2 arc, since it has a 1 arc, hence term arc, local successor. It is clear that MGreek accusative clitics are not determined by acting 2s as such, where by acting 2s one means nominals corresponding to the heads of acting 2 arcs. For this would mean that in 32A structures like *[29b and c], either the initial 2 or initial 3 should determine a clitic. But *[29c] shows that only the initial 3 (final 2) can. Informally, my suggestion is that such clitics are determined by the last acting 2 in a clause. To make this precise, one can introduce a general notion LAST, holding of a certain set of properties of arcs. The notion is introduced in this general way, because it proves to be true that various properties of arcs are subcategorized by the notion Last in question. That is, Last is useful in defining other notions besides Last Acting Term.
The concept Outlast of the consequent of [34] is defined in Postal (1982: 114). 17 In these terms then, my suggestion about Mgreek accusative clitics is:
It is evident immediately that [35] does not allow the 8 determined by simple 32A with no passivization to determine a clitic. Because, although that nominal heads an acting 2 arc, it is not the last acting 2 arc:, since the 2 arc local successor of the 3 arc is in a later stratum. Significantly, [35] predicts that the very 8 which cannot determine a clitic in a 32A active can determine one in the corresponding passive. Because in a passive, what was the last acting 2 arc in the active is not an acting 2 arc at all. So the initial 2 arc in such a passive is the last acting 2 arc. The interaction of [35] with 32A structures and passivization is shown in the following simplified diagrams of structures containing an initial 2 and 3:
< previous page
page_48 If you like this book, buy it!
next page >
< previous page
page_49
next page > Page 49
Consideration of the 2 arcs in [36] shows that only A in [36a], A in [36b] and E in [36d] are last acting 2 arcs. This corresponds precisely to the fact that accusative clitics are possible only with the initial 3 (final 2) in [29a] = [36b], with the initial 2 (final 2) in [29e]=[36a] and with the initial 2 (final 8) in [31] = [36b], but with no nominal at all in cases like [30a] = [36c], which contains no acting 2 arc at all. It turns out that [35] differs hardly at all in complexity from the analog to [35] needed under Joseph's 3 IA account; the difference is just that between a single reference to 'final 2 arc' versus 'last acting 2 arc.' Moreover, [35] permits the simple [32] as opposed to the variant complicated to permit MGreek 31A just with the single V at issue. The remaining problem then is the failure of some MGreek speakers to allow what I am taking to be the 'intermediate' 32A structures like [28c] and
< previous page
page_49 If you like this book, buy it!
next page >
< previous page
page_50
next page > Page 50
[29a]. But this also, I believe, does not favor a 31A analysis. In the later terms, such speakers will have a rule blocking 32A like [37]:
In the terms I am suggesting, the constraint is [38]:
This does not seem appreciably more complex than [37] and has the intriguing property of being parallel to the 32A masking conditions suggested for both French (dés)obéir and the Ancient Greek facts of the previous subsection. While I have in no way shown Joseph's 31A analysis of MGreek to be incorrect, I have indicated that there is an alternative which appeals only to 32A. Of course, as discussed at the end of the previous section, since there seems to be no doubt that 31A exists in some NLs, 19 and, moreover, that it can be associated with the same verbal morphology as 21A, nothing very much is at stake. No matter which analysis of MGreek is right, there will be an APG treatment expressing the generalizations based on the notion generalized passive clause, which is one in which a 1 arc local successor overruns some arc. 2.3.5 A Restriction in Korean A case relevant to the discussion of the constraints in French, Ancient Greek and Modern Greek is provided by Shibatani's (1977: 804-805) brief account of Korean. Shibatani tries to argue that the distribution of floating quantifers in Korean cannot be stated, as generally assumed in early RG work, in terms of grammatical relations. He describes a construction which involves in RG/APG terms 32A, and 32A with essentially the kind of case marking found in Ancient Greek. Thus he initially considers the examples:
Although Korean has a passive construction, Shibatani points out that in general 3s are not allowed to be the ls of corresponding PPs. Thus he claims that passivization '...provides a syntactic test for distinguishing ...' 3s from
< previous page
page_50 If you like this book, buy it!
next page >
< previous page
page_51
next page > Page 51
2s. Consider [39b]. The natural RG analysis would take this as a case of 32A, with accusative marking of the 8. But Shibatani observes that 'the child' cannot be passivized from [39b]:
Shibatani concludes that the difference between [39a] and [b] is not, in our terms, one describable by 32A, claiming in effect that the 'accusativized' 3 in [39b] is a superficial 3, not a superficial 2. Since it behaves like a 2 with respect to quantifier floating, for Shibatani, the latter is to be stated in terms of case marking. But another analysis would keep these Korean facts consistent with the relational claim that quantifiers float only off 1s and 2s and not off 3s. This would be a constraint on 32A parallel to those already suggested for French and Ancient Greek, only involving a negative. Suppose [39b] does involve 32A but that its lack of a corresponding PP is due to rule [41]:
This allows [39b] as a case of 32A, with 'the child' as superficial 2, allows PPs like [40a], but correctly blocks *[40b]. Moreover, [41] helps account for the fact, which Shibatani explicitly notes in his footnote 19 is unaccounted for in his terms, that the variant of [40a] in which 'child' is marked accusative is impossible. In present terms, this marking is possible only when the 3 advances to 2. But in a passive like [40a], the 3 has not advanced to 2. 20 The possibility of rules like [41] brings out a logical problem with Shibatani's claim that 'passivization is a test for 2hood'. In current terms, 2hood is, even if we ignore the issue of generalized passive clauses, only a necessary not a sufficient condition for a passive of any type. If an NL forms passives only with earlier 2s, then it is truly necessary that the superficial 1 of a passive also head a 2 arc, thus be the 2 of the clause at some earlier level. But there is no necessity that the superficial 2 of an active correspond to the 1 of some passive. This is only the case if no rules blocking passives intervene. It is a logical error to conclude that an unpassivizable nominal is necessarily not a 2. Passivization failure is consistent with both 2hood and non-2hood. These brief remarks about Korean illustrate how constraints on successors and on R-successors, easily stated in APG terms, can deal with what
< previous page
page_51 If you like this book, buy it!
next page >
< previous page
page_52
next page > Page 52
otherwise might seem problematic restrictions. Further, Korean supports the comment made in the discussion of Ancient Greek that case marking is far from an infallible guide to superficial grammatical relations. In particular, Korean illustrates the assumption made for Ancient Greek: that 8s due to 32A can be marked accusative, just like superficial 2s. 2.3.5 Postscript After I completed this section, D. Gerdts of the State University of New York at Buffalo, pointed out to me that rule [41] probably need not be stated since it follows from a broader generalization about Korean passives. Gerdts, Choi, Chun and Youn (1982) argue that only initial 2s can form passives of the sort under discussion; see also Gerdts (to appear). If this is true, [41] is a trivial theorem, since local successors and initial arcs are mutually exclusive.' 2.3.6 Quiché Instrumental Advancement Next I consider data from Quiche, as described in Norman (1978). Here also the major restriction does not directly involve passives. But passives interact with that restriction so as to justify an R-predecessor rule analogous to French [23]. This supports the view that Quiché passives involve 1 arc local successors, as determined by the APG universal account of passive clauses. The facts involve, in RG/APG terms, the advancement of instrumental nominals to 2 (Inst2A), a common enough feature in various NLs. This initial conclusion is motivated in Quiché as follows. 2s occur without case marking or prepositions, determine absolutive agreement on the V, permit corresponding PPs and can be 'extracted' for focusing and questioning without special marking of the V. Instrumental nominals, in contrast, do require case marking, do not determine V agreement, cannot appear as 1 s of corresponding PPs and, when instrumentals are 'extracted', a particle wih/wi must appear on the V. To get an initial sense of Quicheé clause structure, consider the following clauses:
< previous page
page_52 If you like this book, buy it!
next page >
< previous page
page_53
next page > Page 53
Since Vs agree with final 1s ergatively and absolutively and with final 2s absolutively, [42] and [43] share the same absolutive agreement marker. In [42] it is determined by the final 2, in [43] by the final 1. Second, [43] lacks an ergative agreement marker because it has no final 2; thus its final 1 is absolutive. Third, Quiché is an NL of the sort cited earlier in connection with the inadequacy of attempts to characterize passive clauses in terms of the devices of transformational grammar, where the word order in active and PP is essentially the same. This follows, apparently, because, although superficial 2s precede superficial 1s in the same clause, superficial 1s precede superficial 8s. The V is, however, in all cases systematically initial. Key structures for current purposes are those where the V has as suffix the voice marker -b'e. In this case, as Norman (1978: 461) puts it: '... the underlying instrument has all those properties which characterize the direct object while the underlying object has none.' So compare:
In [44b], the V absolutively agrees with the instrumental, not with 'you'; the instrumental is not marked with a preposition and, in the PP corresponding to [44b], it is the instrumental which is the final 1:
I comment below on the preverbal order of the form for 'machete' in [44b] and [45]. Finally, in sentences with the voice marker b'e-, the instrumental is focused like a 2, that is, without the verbal marker wih. One could account for these facts roughly as follows. Assume Inst is the R-sign for the grammatical relation Instrumental, initially borne by nominals like ch-ee jun ch'iich' 'a machete' in [44a] and by the corresponding nominal in the English translation. Inst arcs can, in Quiché, although not in general in English, have 2 arc local successors; concomitantly, the cooccurring earlier 2 arcs have 8 arc local successors. Thus, unlike English, which
< previous page
page_53 If you like this book, buy it!
next page >
< previous page
page_54
next page > Page 54
does not allow Inst2A, Quiché lacks a rule blocking 2 arc local successors of Inst arcs. When a Quiché clause does contain a 2 arc local successor of an Inst arc, its V is marked with the voice marker b'e. Vs agree (absolutively) with final 2s, hence with earlier instrumentals when b'e- appears, and passive clauses involve, as always, 1 arc local successors of 2 arcs. Hence instrumentals can be passivized only in clauses with b'e-, because only in these do they head 2 arcs. In these terms, the structure of [45] would include the elements in [46]:
So far Quiché instrumental clauses involve nothing whose analog has not been described in RG terms many times. But Quiché instrumentals go beyond more typical cases in that, as not made clear so far, Inst2A clauses are always illformed unless the resulting 2 is 'extracted', that is, focused, questioned, etc. (Instrumentals are only questionable or relativizable if they do advance to 2, although they are directly focusable.) As Norman (1978: 463) puts it informally: Besides this functional link between the advancement rule and the extraction rule, a formal link also exists: any instrument which has advanced to object is obligatorily extracted, i.e. it must be focused, questioned or relativized. Sentences where an instrument advances to direct object without being extracted are ungrammatical, as shown by [15].
< previous page
page_54 If you like this book, buy it!
next page >
< previous page
page_55
next page > Page 55
The question arises then how to state this somewhat unusual restriction (unusual in general, though similar restrictions are attested elsewhere in Mayan; see Craig (1977)). In APG terms, 'extractions' of the sort in question involve (foreign) successor arcs whose R-signs are members of an Rsign class called OVERLAY R-SIGNS; these include Top, Wh, etc. This set of R-signs defines the class of Overlay arcs. Consider the following alternate hypotheses as to the relevant restriction in Quiché:
[48a] says that the 2 arc local successor of an Inst arc must have an Overlay arc successor. [48b] makes the weaker claim that such a local successor must have an Overlay arc R-successor. Only [48b] has a chance of being correct since [48a] is already falsified by Norman's [11] = [45]. This is a PP of an Inst2A clause, so the 2 arc local successor of the Inst arc has a I arc local successor. No arc can have more than one successor, 21 and hence the 2 arc has no Overlay arc successor. But [45] is consistent with [48b]. For, as not so far stressed about PPs like [45], and not represented in [46], the final 1 of the PP has been focused; unfocused l s are postverbal. And, crucially, as Norman (1968: 463) emphasizes: The condition also holds even when an instrument advances to direct object and then to subject by Passive. Sentences like [16] are ungrammatical, since the underlying instrument has advanced to direct object and then to subject without being either focused, questioned or relativized.
Exactly this result would follow from [48b], since this requires the 2 arc local successor of an Inst arc to have an Overlay arc R-successor. Thus, in *[49], the 2 arc in question has a 1 arc local successor, and [48b] can be satisfied only if that 1 arc (or some successor of it, or some successor of some successor of it, etc.) has an Overlay arc successor. This is not the case in the extraction-free structure *[49], where the I arc of the passive has no successor and is a surface arc. The condition is met, however, in [45], where the final 1 arc of the PP has what we assume to be a Top arc successor. The key point is that rule [48b] captures simultaneously the generalization
< previous page
page_55 If you like this book, buy it!
next page >
< previous page
page_56
next page > Page 56
underlying the ill-formedness of both Quiché actives like *[47] and passives like *[49]. The implication of this discussion is this. The restriction on Quiché Inst2A, which Norman states informally very clearly but does not embed in a precise framework, is easily represented in APG terms by rule [48b]. This statement makes reference to R-PREDECESSOR, the ancestral of the predecessor relation, just as the French rule in [23] does. Crucially then, [48b] can only properly block Quiché passives like *[49] while failing to block well-formed 'extracted' instrumental PPs like [44b] if these clauses can be analyzed so as to involve an element instantiating the relation 'Rsuccessor of a 2 arc local successor of an Inst arc'. But the superficial 1 arcs of the relevant PPs instantiate exactly that relation under the APG version of the universal view of passive clauses, defined by the presence of a 1 arc local successor of a 2 arc. I conclude that evidence for the constraint [48b] is evidence favoring the universal account of passives I have been advocating. It would be instructive to see how the French constraint discussed in section 2.3.2 and the Quiché constraint discussed by Norman could be stated in frameworks distinct from that advocated here. Such facts provide a relevant challenge to alternative viewpoints. 2.3.7 Agreement and Passives in Chichews This section and the next consider some intriguing and complicated facts from two Bantu languages, facts involving verbal agreement and passivization. The goal is to indicate how the APG framework permits an account of the generalizations governing these phenomenona, generalizations which have, to my knowledge, never really been statable before. Chichewa is a Bantu language of Malawi and eastern Zambia. All the data here are taken from Trithart (1976). As is typical for Bantu languages, Chichewa has noun classes and obligatory agreement in class of a V with its final 1. This 1 agreement is manifested by a verbal prefix. Interest here centers on the fact that Chichewa also manifests optional agreement with various non-1s; this agreement is manifested as an infix between the tense marker and the stem. Only one infix agreement marker is allowed per V. A typical example is [50] (the V is bracketed in all examples):
Here the V can contain the infix zi- in agreement with the noun n-thochi. From examples like [50], one could initially conclude that optional infix agreement (henceforth IA) is determined by final 2s. But, although there is a sense, clarified below, in which IA is 2 agreement, IA determination is in fact
< previous page
page_56 If you like this book, buy it!
next page >
< previous page
page_57
next page > Page 57
enormously more complicated. In particular, it is also possible in PPs corresponding to actives like [50]:
The V in [51] contains the obligatory 1 agreement prefix and can optionally contain the IA marker as well. In the latter case, the final 1 nominal determines IA. Hence IA certainly cannot be characterized in terms of final 2s, since [51] contains no final 2, and both the prefix and infix agreement markers are determined by the nominal which is final 1. The facts are even more complicated in clauses containing what would be 3s in RG/APG terms:
[52a] shows that when the earlier 3 manifests as a superficial 3, flagged with the preposition kwá, it cannot determine IA. But in a structure interpretable as involving 32A in RG/APG terms, either the earlier 3 or earlier 2 can determine IA. This is consistent with the view that IA is in some sense 2 agreement and begins to suggest a generalization about IA like:
This accounts for all of the data in [50]-[52] if, as is claimed by our underlying view, passive clauses like [51] are such that the final 1 is an earlier 2 and if, in actives like [52b], both of the postverbal nominals head 2 arcs at different levels, a consequence of a 32A analysis of clauses like [41b]. While [53] is an approximation in the right direction, the actual conditions on IA are still more complicated, as shown by the two distinct PPs corresponding to [52b]:
< previous page
page_57 If you like this book, buy it!
next page >
< previous page
page_58
next page > Page 58
Regardless of whether the prepassive arc is headed by the initial 2 22 as in [54a], or by the initial 3 as in [54b], IA can, unlike in the active [52b], only be determined by the initial 2. While consistent with [53], these extra restrictions do not follow from that statement, which would wrongly not block the ill-formed IA structures in both [54a,b]. This is true under the assumption, which I believe correct, that [54a,b] both involve 32A, and are entirely parallel to the English pair [55]:
I take [54a,b] to have, respectively, structures including the following elements:
< previous page
page_58 If you like this book, buy it!
next page >
< previous page
page_59
next page > Page 59
In these terms, [54a], like [55a], illustrates a relatively rare but nonetheless existing structure, in which a (nonreflexive) passive clause is transitive in its final stratum, due to advancement of another nominal to 2 subsequent to the level defining the clause as passive. 23 My interpretation of the agreement facts in [54] is that [53] is correct but represents only part of the conditions on IA. The contrast between the good and bad IA cases in [54] is based on a distinction between those 2 arcs which are local successors of 3 arcs and other 2 arcs, in particular, the initial 2 arc. Roughly, the initial 2 in a clause can determine IA more or less independently of other considerations.24 But a 2 which is an earlier 3 can only determine IA subject to further restrictions. To move closer to a specification of Chichewa IA, it is necessary to represent these conditions. First, though, a few words are necessary about agreement per se. It is not possible here to consider seriously how IA in particular or agreement in general is to be represented in terms of PNs. It suffices to assume that agreement involves the sponsoring of certain arcs by certain others. For concreteness, assume that agreement involves inter alia a specified class of arcs, say with the R-sign Ag. The informal statement that a certain nominal N determines agreement is then more precisely that N
< previous page
page_59 If you like this book, buy it!
next page >
< previous page
page_60
next page > Page 60
corresponds to the head of some arc which sponsors an Ag arc. In the case of Chichewa IA, if we continue to assume the correctness of illustration [4] in Chapter 1, agreement would be represented by structures of the form:
Certain principles, probably statable as laws, would then have to guarantee that the head of an arc like B is assigned to a certain set of categories identical to a subset of categories to which the head of A is assigned. I will not consider this further. Certain other principles will both have to specify the 'location' of arc B, that is, indicate that it is a branch of a P arc neighbor of A, and will also have to determine the relative ordering of the IA marker. But here I want to concentrate exclusively on the principles which pick out the class of arcs like A, arcs one can refer to informally as IA CREATORS. 25 Consider the following rule governing IA:
This rule now allows IA freely with, e.g., the initial 2 of a clause, but allows it with a 2 advanced from 3 only if the 2 is final. Thus [58] correctly blocks IA in [54b] determined by the initial 3 (later 2) because the 2 arc is not a final 2 arc, since it has a 1 arc local successor, which erases it. However, [58] is still not articulated enough, because it fails, under a 32A analysis of the sort
< previous page
page_60 If you like this book, buy it!
next page >
< previous page
page_61
next page > Page 61
advocated here, to block the IA in [54a] determined by the initial 3/final 2. In fact, a minimal pair is formed by the active 32A structure [52b], in which the earlier 3 can determine IA, and the 32A passive [54a], in which it cannot. While requiring an IA-determining 2 which is an earlier 3 to be final 2 may be correct, it is evidently not sufficient. A sufficiency condition can, I suggest, be found by noting that in the active [52], where the initial 3 can determine IA, the initial 2 ends up as an 8, while in the PP [54a], where the initial 3 cannot determine IA, the initial 2 ends up as a final 1. Put slightly differently, the 2 heading the arc overrun by the 2 arc local successor of the 3 arc heads an 8 arc in the good cases. Suppose then rule [58] is expanded to [59]: 26
This requires that a IA creator 2 arc local successor of a 3 arc be a final arc and, in addition, that the 2 arc which it overruns have an 8 arc local successor which is a surface arc, that is, which is not erased. We have so far motivated the requirement that the overrun 2 arc have an 8 arc R-local successor, for it is this which distinguishes the well-formed initial 3-determined IA in [52b] from the ill-formed initial 3-determined IA in [54a]. But the 'surface' condition is not yet motivated. I return to this presently. First though, I consider additional support for the other conditions, derivable from benefactive structures. Chichewa has a class of nominals interpreted as benefactives which occur in clauses with the properties of 2s. There is no way to present a benefactive nominal obliquely or with a preposition. In particular, a benefactive cannot be flagged with the preposition kwá, used with 3s. One finds:
Since the benefactive which ends up as final 1 in the PP can determine IA, the facts would follow trivially from the assumption that such benefactives are simply initial 2s, the properties of [58] then predictably matching those of the pair [50] and [51]. However, I reject this view and claim that the benefactive nominals head initial oblique arcs with the R-sign Ben. Cases like [60] then
< previous page
page_61 If you like this book, buy it!
next page >
< previous page
page_62
next page > Page 62
result from advancement of benefactives to 2 (Ben2A), which is obligatory. This analysis is motivated, beyond general universalist assumptions which preclude ever taking benefactives as initial 2s, by the behavior of benefactives with respect to IA in transitive clauses. For [60] involves a'structure which is, the benefactive aside, intransitive. The key examples would then involve the combination of benefactive nominals with independently transitive clauses. These exist:
In the active [61a], either postverbal nominal can determine IA, that is, either the nominal that would be the initial benefactive or the one which would be the initial 2, under a Ben2A analysis. But in either of the passives, only the initial 2 can determine IA, regardless of which nominal is final 1. In other words, with respect to IA, benefactives behave exactly like those 3s which have advanced to 2. There are at least two different ways to represent this in a grammar. First, the facts would follow without alteration of rule [48] if it is assumed that benefactives advance to 2 via 3. Second, the facts would follow without this assumption if subcondition [626] in rule [59] were replaced by the disjunction [62b]:
The former approach would require a rule for obligatory advancement of benefactives like [63a], while the latter would require [63b] below; for the notion Free, see [26a] above:
< previous page
page_62 If you like this book, buy it!
next page >
< previous page
page_63
next page > Page 63
Thus the view that benefactives advance to 3 first simplifies the agreement rule but complicates the advancement of benefactive rule, while the view that benefactives advance directly to 2 has the reverse consequences. 27 I see no way to choose between them at the moment and rather arbitrarily choose the solution that leaves rule [59] as is and picks [63a]. Of course, this choice would be mandated if there were a law that benefactives could only advance directly to 3, so that all apparent advancements to 2 involve an intermediate 3 stage. But such a law is implausible.28 Given these assumptions, both the facts in [61] and the contrasting situation in the intransitive benefactive clauses in [60] follow from rule [59] without alteration. IA in [60b], determined by the initial benefactive, advanced successively to 3, 2 and 1, is possible because the 2 arc local successor, which is the prepassive arc, overruns no arc, and thus does not invoke the internal conditional antecedent in rule [59]. In clauses like [61], which have 2s independently of the advanced benefactive, the benefactive can head a 2 arc local successor only by overrunning the initial 2 arc.29 Since the 2 arc local successor is, under current assumptions, the local successor of a 3 arc, IA in such cases is subject to the second complex condition in rule [59]. This thus prevents benefactives from determining IA in passives, regardless of whether the benefactive or the initial 2 is the final 1. I take the behavior of benefactives with respect to IA determination thus to support both rule [59] and the view that Chichewa benefactives involve 2 arc local successors (of 3 arcs) for earlier Ben arcs. In particular, the fact that benefactive in PPs like [6lb,c] do not behave like initial 2s supports an advancement view of Chichewa benefactives, which in turn supports rule [59]. More generally, of course, these facts support the existence of obligatory and hence 'masked' advancements. Although rule [59] as is is reasonably simple, considering the complexity of the facts it covers, a more elegant statement is possible using the concept INDUCE, itself defined in terms of THREATEN:
Hence A induces B if A is a successor which overruns B's local predecessor. Looking at the pair of structures [56], one can see that in [56a], D induces A
< previous page
page_63 If you like this book, buy it!
next page >
< previous page
page_64
next page > Page 64
and A induces H, while in [56b] B induces E and A induces I. Given this notion, rule [59] can be refined to:
This has the virtue of suppressing the existential quantifier and its associated variable needed in [59], as well as eliminating any need for the rule to mention an ancestral. Possibly, all ancestrals can be eliminated from rules in favor of a restricted set of concepts like Induce. Finally, let us consider the reason for the specification in rule [66] that the 8 arc R-local successor of the overrun 2 arc must be a surface arc. This is motivated by the following general fact, which Trithart documents at length. IA determination by a 2 which is an earlier 3 or benefactive is blocked even when that 2 is final and even when the nominal heading the overrun 2 arc is a final 8, if that 8 is not visible in the clause as a result of what would be, (in transformational terms), extraction or deletion involved in question, relative clause, cleft or raising structures. In other words, when the earlier 2 of a clause which independently permits two distinct IA determinations because of the advancement of a 3 (benefactive) to 2 is extracted or deleted, only a single IA determination is possible, that by the initial 2that is, that by the nominal not visible in the clause. Compare [52b] above with the corresponding relativization structure:
Assuming, as is necessary in APG terms, that the embedded clause here involves an erased 8 arc, the failure of IA determined by the earlier 3 follows from the surface arc condition in rule [66]. It is striking that parallel invisibility of the secondary 2 itself has no consequences for IA determination. Thus, in contrast to [67], the relative clause in [68] has the same dual possibilities for IA determination as a simple clause:
< previous page
page_64 If you like this book, buy it!
next page >
< previous page
page_65
next page > Page 65
This follows since in such cases the head of the overrun 2 arc does end up as a surface 8, and thus all conditions for IA determination by a nonfirst 2 are satisfied. Many other properties of Chichewa IA deserve comment. For instance, instrumental and locative nominals which advance to 2 determine restrictions partially distinct from those discussed here involving 3s and benefactives. These restrictions do not bear on the correctness of rule [66] but simply show the need for further conjuncts to yield a complete description of the conditions on Chichewa IA. To summarize, the support for the general APG view of passivization derivable from Chichewa IA is this. Rule [66], which expresses the generalizations underlying the data we have considered, takes IA to be an instance of 2 agreement. Only nominals which head 2 arcs can determine IA by this account. However, the nominals determining IA are not restricted to being final 2s; they can be final 2s, final ls, or final 8s. Only a view of grammar which recognizes multiple levels seems to have a chance of capturing the generalizations existing here. In particular, only the view of passive clauses which takes their final 1 arcs to be local successors of 2 arcs allows a relatively simple generalization like [66] to subsume the fact that only some ls, (namely, a subset of those which are final ls of passive clauses), can determine IA, (which is agreement with 2s), because only this view of passive clauses determines that the final ls of passives are earlier 2s. 2.3.7 Postscript After completing section 2.3.7, I have become aware, through diverse sources, that there are serious problems with some of the Chichewa data described. Several linguists, including G. K. Pullum and W. Davies, have reported to me a failure to reproduce the acceptability of passive clause IA with Chichewa speakers consulted here in the United States. Further, D. Perlmutter has informed me of a correspondence with L. Trithart, who gathered the data in question, in which she herself expresses a lack of confidence that IA is truly possible in passive clauses. An anonymous reader of this study, apparently familiar with these problems, strongly recommended that section 2.3.7 be eliminated, suggesting that inclusion of that material would lead to a 'pollution' of the literature with nonexistent data. After considerable reflection, I have decided to leave the section unaltered, for several reasons. First, granting the worstthat is, that IA agreement is not possible in passive clauses for the Chichewa speakers studied by Trithart, it is at least remotely possible that this phenomenon exists either in other varieties of Chichewa, in other forms of Bantu or in other non-Bantu NLs. I know of no grammatical theory which presently has otherwise true principles provably strong enough to preclude such an agreement situation. If such agreement does exist in NLs, then the account in
< previous page
page_65 If you like this book, buy it!
next page >
< previous page
page_66
next page > Page 66
section 2.3.7 can, I claim, provide insight into 'the correct way to describe it. If, on the contrary, it is truly impossible, it is important to find ways of blocking it in principle. The simplest APG proposal, namely, to limit Ag arc sponsors to being final arcs, is impossible, given the existence of documented agreement determined by initial arcs, some of which are not final arcs. Such a case from Achenese is discussed in Perlmutter (1981, 1982), where what is involved is agreement determined by initial 1 arcs, including those which, in passives, have 8 arc local successors, and thus are not final ares. Thus even if there is no agreement in NLs of the sort claimed by Trithart (1976), the discussion in the present study can at least focus attention on the key question of what grammatical laws preclude this. But if there is no IA agreement in Chichewa passives, then the support derived from it for the RG/APG multilevel approach to passives of course does not stand. Second, there is much of interest to Trithart's data which does not seem to be in any doubt, in particular, the existence of pairs of passives like [54a and b], especially those like [54b]. These are referred to as TERTIARY PASSIVES in the following section, which compares Chichewa and Chi-Mwi:ni (Bantu) in this regard, noting the nonexistence of tertiary passives in the latter. Thus the contrast between NLs permitting and those not permitting tertiary passives, whose explicit recognition and description is an important feature of this study, would have been lost if section 2.3.7 had been eliminated. Finally, the preceding discussion of Chichewa can hardly be accused of bringing about any 'pollution' of the literature. The Chichewa facts presented by Trithart are already discussed in at least four separate published articles, Trithart (1975, 1979), and Dowty (1982a, 1982b). Moreover, it is claimed in several places in these works, as partly indicated in notes 23 and 27 of this Chapter, that Chichewa data (in a manner unrelated to the dubious agreement facts), cause problems for the RG framework. It is thus not irrelevant to show how the current relational framework: encounters no difficulty at all with either the challenged or unchallenged material in Trithart (1976). For all these reasons, it has seemed reasonable to me to leave section 2.3.7 unaltered, since it has a certain value even if IA agreement in Chichewa passives is chimerical. 2.3.8 Agreement and Passives in Chi-Mwi:ni Chi-Mwi:ni is another Bantu language, apparently a dialect of Swahili. All the data here are taken from the factually rich description in Kisseberth and Abasheikh (1977) (both this work and these authors are referred to as KA hereafter). Our concern will be phenomena related to those discussed with respect to Chichewa. Although the conditions in ChiMwi:ni are reminiscent of those in Chichewa, they differ in striking ways, and their statement must therefore be partially the same and partially different.
< previous page
page_66 If you like this book, buy it!
next page >
< previous page
page_67
next page > Page 67
First, and quite parallel to Chichewa, I will be concerned with the conditions governing the determination of the object agreement infix. I will utilize the same terminology as in the preceding section, in particular, 'IA', and 'IA creator'. KA refer to the determination of an OP (object prefix). Like Chichewa, Chi-Mwi:ni has the typical Bantu noun classes, with obligatory agreement of Vs with the final 1 in noun class. Also, IA is again in one sense optional in Chi-Mwi:ni, in that the absence of the IA marker normally does not determine ill-formedness, but rather only the indefinite-ness of the nominal that could determine IA does. Second, I will be concerned with the possibilities of passivization. Consider the following examples (as in the previous section, Vs in all examples will be bracketed):
[69a] is a standard active transitive clause, showing agreement with both the final 1 and the final 2. This initially suggests that IA is determined by the final 2. Moreover, unlike the situation claimed by the original source for Chichewa (Trithart, 1976), such an initial impression is maintained when the corresponding passive is considered, for not only is no IA present in [69b], none is possible. Apparently, no Chi-Mwi:ni passive clause can have a V with an IA marker. This is consistent with the view that Chi-Mwi:ni IA is determined by final 2s, given that in RG/APG terms, a PP like [69b] has no final 2. Hence an initial conclusion is that the IA rule for Chi-Mwi:ni might include [70]:
Statement [70] will, however, have to be radically revised. Consider some clauses involving Vs which take what would be 3s in RG/APG terms:
< previous page
page_67 If you like this book, buy it!
next page >
< previous page
page_68
next page > Page 68
These examples suggest that the stem 'write' permits both 2s and 3s, and that superficial 3s are flagged with the preposition ka. Passives of Vs with superficial 3s are standard; the initial 2 of the active corresponds to the final 1 of the passive. However, like Chichewa, Chi:Mwi:ni also permits (and in many cases apparently requires) structures interpretable in APG terms as involving 32A. Consider:
Notice the presence of the applied suffix here, not present in [72]. This seems to be a marker indicating that some nominal has advanced to 2, although there seem to be certain cases of such advancement in which this suffix is not present. Independent of agreement contrasts between Chichewa and Chi-Mwi:ni, here one finds a passivization contrast. Comparing [73] with [54] above, it is clear that while Chichewa permits two passives corresponding to a 32A active, Chi-Mwi:ni does not. Only what would be the earlier 3 in such an active can correspond to the final 1 of a Chi-Mwi:ni passive. An initial formulation of this restriction could be the perfectly general statement in [74], which has no analog in the grammar of Chichewa:
Statement [74] makes key use of the notion Induce, introduced in the previous subsection. It says that a 2 arc local succressor can only induce an 8 arc, and hence, in particular, not a 1 arc. The structure of the ill-formed *[73c] would be, in current terms, [75]:
< previous page
page_68 If you like this book, buy it!
next page >
< previous page
page_69
next page > Page 69
Here the 2 arc local successor D induces the 1 arc A, violating [74]. It will be seen below, however, that [74] is slightly too general. Example [73a] reveals another contrast between Chichewa and Chi-Mwi:ni, this one with respect to IA. Whereas in 32A actives like [52b], Chichewa permitted IA determination by either the earlier 3 or earlier 2, Chi-Mwi:ni permits it only by the earlier 3. This already follows from the statement in [70], since, in a 32A active, only the earlier 3 can be a final 2, given the Stratal Uniqueness Law. 30 The patterns and generalizations noted so far are extremely general for Chi-Mwi:ni and hold, apparently, for all advancement to 2 structures not involving advanced instrumentals. Thus they hold for 3s advanced to 2, both with Vs which require the applied suffix, like that in [73], and with Vs which do not take this suffix. Similarly, they hold for benefactives advanced to 2this occurs only with the applied suffix. A similar question arises for Chi-Mwi:ni as Chichewa, namely, do advanced benefactives involve 3 arc local successors as well as 2 arc local successors. A possible argument for direct Ben2A is discussed at the end of this section. However, in the case of instrumentals advanced to 2, the facts are much more complicated, initially bewilderingly so, with respect to both IA and passivization. Advanced instrumentals require the applied suffix, and can be illustrated by [76]:
< previous page
page_69 If you like this book, buy it!
next page >
< previous page
page_70
next page > Page 70
[76a] shows that instrumentals can occur obliquely, that is, heading surface oblique arcs (more precisely, those with the R-sign Inst), in which case, the preposition ka: is required. But an instrumental can also occur unflagged in actives, as in [76b], in which case, the V requires the applied suffix. This possibility is limited by considerations which need not directly concern us such as 'topicalization'. This is shown by the fact that a simple clause in which an instrumental occurs (postverbally) without a flag and with the applied suffix is ill-formed. The nature of these restrictions, discussed by KA (1977: 196-201), is unclear to me. They are, however, not completely parallel to those found in Quiché in section 3.6, since the unflagged instrumental can occur untopicalized if the earlier 2 is questioned:
Evidently, it is natural to assume in APG terms that unflagged instrumentals like those in [76b] and [77] have advanced to 2. The 'topicalization' requirements, whatever they are, are then restrictions on Top arc successors of arcs with the property 2 = + = InstLocal Successor(A). The claim that cases like [76b] involve Inst2A with concomitant demotion of 2s to 8 is initially supported by the fact that in these cases, in contrast to those where an instrumental is flagged, the initial 2 can neither determine IA nor have a corresponding passive:
In [78], where no Inst2A is involved (note the lack of the applied suffix and presence of the preposition ka;), the initial (and also final) 2 determines IA, and there is a passive, [78b], where the correspondent of the initial 2 is final 1. But this is impossible for the analogs of [76b]:
< previous page
page_70 If you like this book, buy it!
next page >
< previous page
page_71
next page > Page 71
The failure of the earlier 2 to determine IA in cases of Inst2A follows directly from the initial IA condition in [70]. Similarly, the absence of a passive in which the earlier 2 is final 2 follows from the induction condition in [74]. So far then, Inst2A cases are no different from 32A or Ben2A structures, and require no rule revisions. However, in cases like [76b], the putatively advanced instrumental, which, in current terms, would be a final 2, can also not determine IA:
Here -sh- is the agreement form for chisu. Note that the inability of the advanced instrumental to determine IA cannot be attributed to the fact that in [80] this nominal is topicalized. There is no general ban in Chi-Mwi:ni on topics determining IA, as shown, for example, by KAs [61a and b] and [62a and b], in which topicalized noninstrumentals determine IA. In these examples, the topics are initial 2s, 3s or benefactives. Although the failure of the instrumentally interpreted nominal to determine IA might seem to cast some doubt on the claim that the instrumental has advanced to 2, the latter view is supported by the fact that the instrumental in such cases can end up as 1 of a passive:
Consider also:
< previous page
page_71 If you like this book, buy it!
next page >
< previous page
page_72
next page > Page 72
Hence the passive facts in [81] and [82] follow from the Inst2A view together with constraint [74]; the failure of IA thus suggests an additional constraint on IA determination restricted to Inst2A structures. Before attempting any modification of earlier rules, however, it is necessary to consider further complications. The pattern of IA and passivization in what I am taking to be Inst2A structures illustrated in [76]-[82] is not fully representative and depends in part on the fact that the earlier 2 in these examples is inanimate. A paradigm where the corresponding nominal is animate is as follows:
[83b] shows that the initial (animate) 2 can determine IA, although in our terms it is a final 8, not a final 2. But *[83c] shows that the advanced instrumental, which is a final 2 in our terms, cannot in this case determine IA. The passive facts for earlier animate nominals in instrumental advancement cases are given in [84]:
[84a] shows that the advanced instrumental can passivize. [84b and c] show, however, contrary to all earlier examples, and in violation of [74], that the earlier 2 can also passivize, at least if it ends up in the 'nontopical' postverbal position. 31.
< previous page
page_72 If you like this book, buy it!
next page >
< previous page
page_73
next page > Page 73
Let us summarize the IA facts for advanced instrumentals first. Although in general in Chi-Mwi:ni, IA is determinable only by final 2s, in Inst2A structures, the following situation obtains. If an advanced instrumental displaces 32 an inanimate earlier 2, no IA is possible, while if an advanced instrumental displaces an animate earlier 2, IA is possible only with the earlier 2. One can revise rule [70] to take account of these observations as follows, using Animate as a schematic representation of a predicate indicating that the head of an arc is an animate nominal:
Recall from [26c] that facsimiles are neighbors with the same R-sign. Comparing statement [85.] to the initial generalization [70-], one observes the following. [85] maintains the assumption that an IA creator is a 2 arc, but abandons the claim that it is always a final arc, in the face of cases like [83b], in which the only final arc headed by the nominal determining IA is evidently an 8 arc. The Facsimile requirement in the antecedent of [85] does not require that an IA creator have a distinct neighboring 2 arc, since Facsimile is not irreflexive. The overall consequent of [85] has four conjoined conditions. The first specifies that an IA creator is a 2 arc. The second states that no same R-sign neighbor of an IA determiner can have a 1 arc local successor. This blocks IA in all passive clauses, regardless of whether the prepassive arc is the IA creator.33 Such a condition was not evidently necessary given the simple formulation in [70], which assumed falsely that only final arcs could be IA creators. The finality condition would thus block all cases in which a prepassive arc was an IA creator. But, since the finality condition cannot be maintained, something else is needed for this task, which the first condition in the consequent supplies. Moreover, the statement in [70] does not block IA determination by the final 2 arc in a passive in which Inst2A induces a 1 arc. But this is properly blocked by the second consequent condition in [85], since in such a case a facsimile of the IA creator, namely, the arc it overruns, would be the local predecessor of a 1 arc. Hence the second element of the consequent correctly allows IA only in Chi-Mwi:ni active clauses. The notion Outlast, as noted earlier, is defined in Postal (1982). Recall that one arc, A, outlasts another, B, if they are neighbors and A's last coordinate index is greater than B's last coordinate index. In other words, A is in some later stratum which B is not in. Obviously then, no arc outlasted
< previous page
page_73 If you like this book, buy it!
next page >
< previous page
page_74
next page > Page 74
by another is ever a final arc. The third conjunct of the consequent thus embodies the abandonment of the earlier incorrect claim that all Chi-Mwi:ni IA creators are final arcs. The third conjunct of the consequent of [85] amounts to a limitation on the class of cases where an IA creator can be nonfinal, restricting this to animate IA creators which are overrun. The animacy claim distinguishes paradigms like [82], where the inanimate nominal displaced by an advanced instrumental cannot determine IA, from paradigms like [83], where the animate nominal displaced by an advanced instrumental can determine IA. Finally, the fourth (biconditional) conjunct of the consequent of [85] makes several claims. Read from left to right, it augments the requirement imposed by the third conjunct of the consequent, limiting clauses in which an IA creator is not the final 2 arc to those containing a 2 arc local successor of an Inst arc. Moreover, it specifies that in this case, the nonfinal IA creator must be overrun by the local successor of the Inst arc. Read from right to left, the fourth conjunct of the consequent states that an IA creator with a facsimile which is a local successor of an Inst arc must be overrun by that local successor. Since Overrun is an irreflexive relation, this states that the 2 arc local successor of an Inst arc can never itself be an IA creator. In other words, even a final 2 arc is not a possible IA creator in Chi-Mwi:ni if it is the local successor of an Inst arc. Thus, overall, [85] says that a Chi-Mwi:ni IA creator must be a 2 arc and must be final except in cases of Inst2A, in which case, it must be animate. Moreover, an instrumental advanced to 2 can never (even when final) be an IA creator, 34 and no arc in a passive clause is an IA creator. Viewed in isolation, rule [85] is hardly elegant. But viewed in context with the facts of Chi-Mwi:ni IA determination, and, in particular, with the highly ad hoc conditions governing this in Inst2A clauses, [85] does, I believe, express the generalizations governing this domain, predicting correctly facts not yet considered. All the examples cited so far involve at most a single advancement to 2 per clause, that is, a single 2 arc local successor. However, there are ChiMwi:ni clauses which must be analyzed as involving multiple advancements, in particular, where both an instrumental and another nominal advance to 2. Given the Stratal Uniqueness Law, this' advancement must be 'successive.' Consider [86]:
This sentence contains an advanced benefactive, which cant determine IA, and an unadvanced instrumental, which of course cannot. Likewise, the
< previous page
page_74 If you like this book, buy it!
next page >
< previous page
page_75
next page > Page 75
initial 2, 'clothes' cannot determine IA. These facts follow directly from rule [85], since 'children' is clearly the final 2 in [86], as indicated further by the fact that there is a passive corresponding to [86] in which 'children' is final 1, although there is none where 'clothes' is final 1:
It has not previously been observed that there is no way in Chi-Mwi:ni, as in Chichewa, to form a clause with a benefactive nominal unless it is advanced to 2 (determining the presence of the applied suffix). Hence one of the following conditions must be part of Chi-Mwi:ni grammar:
The choice depends again on whether the benefactive first advances to 3. Consider then the examples in [89], which indicate that in a clause with an (advanced) benefactive, an instrumental can also advance to 2:
The lack of flagging of the instrumental in [89a] indicates that it has advanced to 2, as does the fact that in [89b], it advances to 1 in the passive. 35 Note though that in [89a], IA is determined by the same nominal as in [86], namely, the benefactive. It is clear from KAs discussion that in these cases only the benefactive can determine IA. This follows from rule [85], if the 2 arc local successor of the Inst arc overruns the 2 arc local (R-) successor of the Ben arc, in other words, if structures with both advanced benefactives and advanced instrumentals involve structures like [90];36
< previous page
page_75 If you like this book, buy it!
next page >
< previous page
page_76
next page > Page 76
In such structures, the only possible IA creator is D. Let us see why. First, since B, D, G are all and only the 2 arcs, only these are the potential IA creators in [90]. Consider B. It is outlasted by D, and thus must be overrun by D, and be animate. It is so overrun. If it is not animate, [85] is still not satisfied. Suppose it is animate. Then [85] is still not satisfied because its fourth conjunct requires that B be overrun by the local successor of an Inst arc, while B is overrun instead by the local successor of a Ben arc. Next consider D. It is outlasted by G, hence must be overrun by it and be animate. D meets these two requirements. Moreover, D is overrun by the local successor of an Inst arc, namely, G, and hence meets the fourth conjunct of [85] as well. Hence D is a possible IA creator under [85]. Finally, consider G. It is a local successor of an Inst arc, hence must, according to [85], overrun any IA creator, in particular, itself. Since Overrun is irreflexive, this is not possible. Hence, G is not a possible IA creator and, correctly, only D in [90] has this property. Consider though structures like [91]:
< previous page
page_76 If you like this book, buy it!
next page >
< previous page
page_77
next page > Page 77
This differs from [90] in that Inst2A is earlier than Ben2A. Such structures would, first, not interact correctly with rule [85]. For the latter would permit no arc at all in [91] to be an IA creator, making the false claim that there are active Chi-Mwi:ni clauses with both advanced benefactives and instrumentals in which no nominal can determine IA, when in fact, the benefactives can. [85] would block all IA in [91] for the following reason. The rule says that any IA creator must be overrun by any facsimile which outlasts it. This excludes B as an IA creator in [91], since both C and G outlast B, but only G overruns it. Consider C. This is the final 2 arc, and hence if an IA creator, is neither subject to the third conjunct of [85], nor to the left-to-right part of the fourth conjunct. But it is subject to the right-to-left part of the biconditional, which it cannot satisfy. For, in the case of [91], this requires that C overrun itselfan impossibilitysince Overrun is irreflexive. Finally, consider G. [85] makes it impossible for G to be an IA creator since it requires any 2 arc local successor of an Inst arc to outlast a neighboring IA creator, and no arc can outlast itself. Consequently, if [85] is both correct and sufficient, something must exclude structures like [91] from the class of Chi-Mwi:ni clauses.
< previous page
page_77 If you like this book, buy it!
next page >
< previous page
page_78
next page > Page 78
Moreover, the same conclusion is imposed by consideration of passives. For if [91] is a legal Chi-Mwi:ni structure, nothing would preclude the existence of a related clause in which the analog of C was a prepassive arc. This would incorrectly allow a passivized benefactive in a clause with an advanced instrumental (subject to the 'topicalization' requirements on Inst2A, which could be met). Something must then guarantee that Ben2A 'precedes' Inst2A, more precisely, that the 2 arc local successor of an Inst arc outlast the 2 arc local successor of a neighboring Ben arc. The easiest representation of this in Chi-Mwi:ni grammar would apparently specify that the 2 arc local successor of an Inst are cannot be overrun, width no analogous constraint holding for 2 arc local (R-)successors of Ben arcs:
This properly blocks all structures like [91], which contains an overrun 2 arc local successor of an Inst arc, namely, G (illegally overrun by c). Rule [92] determines that in a clause with multiple advancements including an instrumental advancement, the local successor of the Inst arc will always outlast the other 2 arc local successors. Chi-Mwi:ni rule [85] claims, via the second conjunct of its consequent, that, in contrast to the Chichewa rule in [55], there can be no IA determination in passives. What is in effect a condition on Chi-Mwi:ni passives was formulated in [74]. But the discussion of Inst2A clauses indicates that [74] is too general; cases like [84b,c] show, in current terms, that some 2 arc local successors can induce 1 arcs, namely, some of those which are local successors of Inst arcs. For in (certain) cases of advanced instrumentals, either the advanced instrumental or the earlier 2 can passivize. The latter is a unique feature of Inst2A in Chi-Mwi:ni; it was already stressed that the analog never holds for 3 or benefactive advancement. However, KA observe that this possibility of 'extra' passivization in Inst2A structures is not completely general. Recall that it was impossible in cases like *[82e], where the nominal displaced by the advanced instrumental is inanimate. Moreover, KA show that the relevant property here is not animacy, as in the case of IA determination, but rather membership in certain noun classes. Only nominals which are members of what KA call noun classes 1 and 2 can be heads of 1 arcs in passives in which an instrumental advances to 2. While all members of these classes are animate, not all animates are members of these classes. My assumption is that the 'extra' passives possible in certain Inst2A structures in .Chi-Mwi:ni are entirely parallel to the passives in Chichewa constructions like [43a] and [50c] or in English [44]. Thus I take Chi-Mwi:ni [84b,c] to have structures including [93]: 37
< previous page
page_78 If you like this book, buy it!
next page >
< previous page
page_79
next page > Page 79
This is a case where the 2 arc local successor (of an Inst arc), D, induces a 1 arc, A, rather than an 8 arc. To limit this phenomenon to (i) advanced instrumentals and to (ii) displaced class 1 and class 2 nominals, the following rule replacing the too general [74] is apparently adequate 38:
The class assignment statement is of course entirely schematic. Ignoring this, the rule limits 1 arc-inducing 2 arc local successors to those which (i) are local successors of Inst arcs and which (ii) overrun arcs headed by class 1 or class 2 nominals. Let us summarize the discussion and indicate how the facts of Chi-Mwi:ni bear on the central issue of this chapter: how the universal view of passive clauses advocated here permits desirable statements of NL-particular restrictions. Investigation has led to the view that Chi-Mwi:ni grammar includes the three rules [85], [92] and [94]. [85] is related to questions of passivization only indirectly, but slightly more directly than it might at first seem. This is brought out by a comparison of Chi-Mwi:ni with Chichewa. It might have been thought a priori that 'object agreement' was not possible in passive clauses. But Trithart's Chichewa data, if accurate,
< previous page
page_79 If you like this book, buy it!
next page >
< previous page
page_80
next page > Page 80
indicates that this is incorrect, even internal to Bantu. Thus the fact that Chi-Mwi:ni has 'object agreement' (IA) and yet does not permit it in passives indicates that some Chi-Mwi:ni rule must guarantee this result. [85] does this via the second conjunct of its consequent. This statement, which references 1 arc local successors of 2 arcs, can easily be integrated into a rule which must, on entirely distinct grounds, refer to 2 arc IA creators and to their facsimiles. Of course, such a formulation is only possible under the: view that Chi-Mwi:ni passives involve 1 arc local successors of 2 arcs. But just this follows from the general view I am seeking to support, namely, that all passives involve 1 arc local successors of (in particular) 2 arcs. Moreover, rule [85] is indirectly related to support for the present universal conception of passive clauses in another way. This statement is designed to interact properly with an analysis of clauses with unflagged instrumentals which treats these as involving Inst2A. But, as has been shown, just such clauses have unique (for Chi-Mwi:ni) passivization possibilities. Only these among Chi-Mwi:ni advancement to 2 clauses allow an earlier non-2 to appear as final 2 of a passive clause. This peculiarity is easily represented in APG terms via rule [94], or possibly as rule (i) of note 38. But such formulations depend critically on the view that passives involve 1 arc local successors of 2 arcs. Finally, rules [85] and [94] integrate properly with a description of cases of multiple advancements to 2, in particular, benefactives and instrumentals. The key fact here is that the advancement of instrumentals must involve strata subsequent to those of other advancements to 2, a fact imposed by rule [92]. This interacts with rule [85] to correctly predict that in such multiple advancement cases only the benefactive, and not the instrumental or the initial 2, can determine IA. And it interacts with rule [94] to properly indicate that in such cases, either the benefactive or the instrumental can be the 1 of a corresponding passive. Overall then, the view of passives as involving 1 arc local successors of 2 arcs facilitates a relatively compact description of an initially perplexing 39 set of facts in Chi-Mwi:ni. A key general point about passive clauses to emerge from the discussions of Chichewa and Chi-Mwi:ni involves what can be called informally 'tertiary passives'.40 I intend this notion to cover Chichewa passives like [43a], [50c], ChiMwi:ni passives like [84b,c] and English passives like [44a]. The latter is repeated here as [95e]:
< previous page
page_80 If you like this book, buy it!
next page >
< previous page
page_81
next page > Page 81
The facts discussed show that in some NLs there are three different types of passives structure ultimately corresponding to a transitive clause with one other constituent nominal. That is, there are three distinct types of PP independent of any questions of impersonal constructions. I will refer to passives like [95c] as PRIMARY passives, to passives like [95d] as SECONDARY passives. Primary passives are those in which the earliest (in [95c], the initial) 2 arc is the prepassive arc and in which no advancement to 2 is present. Secondary passives are those in which the prepassive arc is itself a 2 arc local successor. Tertiary passives are thus those in which the earliest 2 arc is prepassive but where there is an advancement to 2. Thus tertiary passives are passive clauses in which a 2 arc local successor is a final arc and hence not prepassive. Put differently, tertiary passives are passives which are final stratum transitive clauses. They thus show that while in general passivization is detransitivizing, this is neither a logical necessity nor an invariant property of existing passives in attested NLs. 41 The structural similarities and differences between primary, secondary and tertiary passives are shown in the following illustrations, which represent a case (say that in [95]) where the secondary and tertiary passives involve 2 arc local successors of 3 arcs (though the notion of secondary and tertiary passives generalizes over the R-signs of these local predecessors):
< previous page
page_81 If you like this book, buy it!
next page >
< previous page
page_82
next page > Page 82
< previous page
page_82 If you like this book, buy it!
next page >
< previous page
page_83
next page > Page 83
In the terms introduced here, tertiary passives can thus be characterized slightly differently as clauses in which the 1 arc local successor of a prepassive arc is induced by a 2 arc local successor. This view of alternative passives like those in [95] is permitted only by the APG conception of 'passive clause'. This has the virtue of yielding a simple statement of restrictions limiting the possibilities of tertiary passives (via rules like [94]). For while tertiary passives may be quite freely allowed in Chichewa, it has been seen that this is not the case in Chi-Mwi:ni. They are also quite restricted in English. Compare [95] with
Hence certain individual NLs will need rules analogous to Chi-Mwi:ni [94], limiting cases of 1-inducing 2 arc local successors to certain types of structure. There are NLs with passive clauses where, apparently, no cases of such 1 arc-inducing are allowed, that is, which completely lack tertiary passives. Tzotzil is apparently like this; see Aissen (1983). Such NLs perhaps have rules as simple as [74] above, or, possibly, like:
The important point is that within the APG framework, which recognizes passive clauses as involving 1 arc local successors, the existence of tertiary passives as in many Bantu NLs and English is easily accounted for, as is their absence in NLs like Tzotzil. Moreover, NL-particular limitations on such passives, like that for Chi-Mwi:ni represented by [94], are also not difficult to construct. Chi-Mwi:ni permits both a benefactive and an independent 3 to occur in a single clause. And, as has been illustrated, Chi-Mwi:ni requires benefactives to advance to 2 and also allows 3s to do so. The cases where benefactive and earlier 3 cooccur offer an argument that benefactive advancement in Chi-Mwi:ni is direct advancement to 2. For, as KA show, although in general 3s can advance to 2, this is impossible with a benefactive present; see KAs [128]-[132]. In such clauses, the earlier 3 must show up flagged, and with the same preposition found with 3s when no benefactive is present. This suggests that such earlier 3s are superficial 3s, which would not, given the Stratal Uniqueness Law, be possible if benefactives advanced to 3, for this would yield overrunning of the earlier 3 arc. If this conclusion is correct, then the blockage on 32A in the presence of a benefactive can be simply
< previous page
page_83 If you like this book, buy it!
next page >
< previous page
page_84
next page > Page 84
represented as in [99] (while Chi-Mwi:ni grammar can contain the simpler of the alternatives in [88]):
Obviously, if an analysis like this is correct, a law like that mentioned in the discussion of Chichewa, which would only allow benefactives to advance to 3, cannot exist. 2.3.9 Halkomelem Passives and Raising Halkomelem is a Salish language of southwestern British Columbia. All the following data are from Gerdts (1981). At first glance, Halkomelem seems to have relatively straightforward PPs for a VSO type NL:
The passive in [100b] differs from the active [100a] in that while the former has transitive verbal marking, the latter has this followed by an intransitive suffix, here . Second, while the V in [100a] has third person agreement with 'John', there is no agreement in the passive. Finally, Halkomelem has two nominal cases Gerdts refers to as STRAIGHT and OBLIQUE. In [100a], the 1 is in the straight case, while its correspondent in [100b] is in the oblique case. These properties largely support, or are consistent with, a PP analysis of cases like [100b], as Gerdts considers in detail. In particular, she shows that the nominal which is a final 1 in cases like [100a] is not a final 1 in those like [100b]. The apparent PPs in Halkomelem are subject to certain unusual constraints. As Gerdts discusses, first and second person pronominals cannot be 8s in passives. Such a constraint can be stated in current terms as [101 a], with 'I/II' a schematic predicate indicating that the head of an arc is a first or second person pronoun or, alternatively, as [101b] with 'III' indicating third person:
< previous page
page_84 If you like this book, buy it!
next page >
< previous page
page_85
next page > Page 85
'Passive-Destroyed' is formally defined at the beginning of Chapter 4. It designates the earlier 1 arc of a passive clause, that overrun by the defining 1 arc local successor. Although I have assumed Halkomelem passives like [100b] are PPs, Gerdts does not merely assume this, but supports it. In particular, she argues against an analysis suggested by Hukari (1980) which would take such clauses to involve 'unmotivated chômage', 43 that is, demotion of the 1 to 8, but no advancement of the earlier 2 to 1. Any such analysis would violate two RG/APG laws, the Final 1 Law and the Motivated Chômage Law.44 Gerdts shows such an analysis is incorrect and that the third person passives like [100b] are PPs, via appeal to a raising construction. Because the raising construction in Halkomelem is 'fed' by a somewhat broader class of complement nominals than its English analog, the argument in Halkomelem is a significant one in the overall attempt to support the current APG view of passive clauses. The Halkomelem raising construction is apparently possible only with main Vs based on a few stems, including xec 'figure out/check out/wonder'. An example of paired complex structures, the second of which involves raising, is provided by [102]:
In each example here and in what follows, the subordinate clause is bracketed. In the case taken to involve raising, [102b], the nominal corresponding to the complement 1, 'the white man', is unambiguously in the main clause. This is determined merely by its position in a VSO NL like Halkomelem. For if that nominal were in the subordinate clause, it would have to follow the complement V, as its analog does in [102a]. Gerdts shows that while a construction like [102b] is possible when the main clause 2 corresponds to the complement I, this is not possible when the main clause 2 corresponds to the complement 2:
< previous page
page_85 If you like this book, buy it!
next page >
< previous page
page_86
next page > Page 86
Hence 2s like 'the white man' in [103b] cannot raise. Consider then the case where the complement is a clause of the sort in [100b], i.e., one taken to be a PP. Then its earlier 2/final 1 can raise, as illustrated by Gerdts (1981: 209). This property of putative PPs supports their PP status, as Gerdts notes, since it would be compatible with the view that a nominal which raises to 2 in Halkomelem is, as in English, a final 1 and only that. However, the Halkomelem situation differs, since when the complement is passive, surprisingly, the earlier 1 can also raise:
These three examples are all essential paraphrases and all have a PP subordinate clause. There is no raising in [104a], raising of the final 1 in [104b] and raising of the final 8 in [104c]. As Gerdts notes, the generalization governing the class of nominals which can be raised in Halkomelem is, under a PP analysis of the relevant passive clauses, the class of acting 1s. Recalling the notion Acting Term arc from section 3.4 [33], one can easily state this in current terms by first defining:
< previous page
page_86 If you like this book, buy it!
next page >
< previous page
page_87
next page > Page 87
Thus an imitation 1 arc is a remote local successor of an acting 1 arc. This definition permits one to say, for example, that both a final 2 arc and an 8 arc local successor of a 2 arc are imitation 2 arcs. Terminology has now been provided for picking out both an acting Term arc and its R-local successors. Let us consider limiting the class of nominals which can 'feed' a raising to 2 construction. In current terms, this is simply a matter of controlling the class of predecessors of certain 2 arc foreign successors. For the English construction studied in Postal (1974), where only final 1s raise, the relevant constraint would be along the lines of [106]: But Gerdts' different generalization about Halkomelem involving acting 1 s is now statable as [107]: 45
Much of the earlier discussion of passive clauses, in particular, that involving Chichewa in section 2.3.7, was designed to argue for the RG/APG view that the final I of a passive clause is an earlier 2. This discussion of Halkomelem supports the different view inherent to the overall RG/APG conception of passive clauses, namely, that the final 8 of a passive is a 1 at an earlier level.46 For only this assumption, made of course entirely independently of any knowledge of Halkomelem, provides a basis for the generalization which picks out the class of nominals which can raise in Halkomelem. In a distinct view of passives, the fact that in Halkomelem the final 1s of both actives and passives and the passive 8 of passives but not, e.g., the final 2 of actives or other functional types, raise would, apparently, be an incomprehensible anomaly. Generalizations like [106] seem beyond many views of passives now current which recognize no grammatical structure beyond that directly manifested in surface forms. Again then, the relevant facts provide a challenge to such views. 2.3.10 Summary In section 2.3, I have tried briefly to illustrate substantively two of the three potential means of validation for a universal approach to passive clauses in general, by showing how such an account yields lawful predictions about various passives in arbitrary NLs and how it permits insightful and relatively
< previous page
page_87 If you like this book, buy it!
next page >
< previous page
page_88
next page > Page 88
straightforward representations of apparently bizarre and complicated NL-particular restrictions, some of which otherwise seem unstatable other than by the equivalent of disjunctions (lists). The discussion has involved support for a universal approach to both PPs and IPs. The third type of possible justification for a universal approach is illustrated in Chapter 4. Notes: Chapter 2 1. The description of pseudo-passives has been a vexed issue and there seems to be nothing approaching a full account of them in any framework so far. A description in APG terms is provided in Chapter 6, along with a criticism of a certain common assumption about these structures. 2. Wachtel (1979) argues that the conclusion can be avoided since there is a factually equivalent alternative to the 1AEX which makes the desired predictions in other cases with no postulation of dummy nominals. His proposal is criticized in Perlmutter and Zaenen (1984). 3. This claim is clarified in Chapter 5. 4. However, even this weaker claim is potentially incompatible with certain data, like those from Achenese cited in Perlmutter (1982: 297). 5. For discussion of inversion clauses, see Dryer (1982), Harris (1981, 1984a, 1984b), Davies (1981a, 1981b, 1981c), Hubbard (1980, 1982), Perlmutter (1978b, 1984b), Perlmutter and Postal (1984a), Sridhar (1976a, 1976b), and Jackson (1982). 6. No such advancement to 1 takes place in structures involving what has been called 'impersonal inversion' in the RG literature; see Perlmutter (1983b), Jackson (1982). Examples are provided by clauses with the French V falloir; see (4.8a) and (4.9) below. 7. In one sense, passive arid inversion can combine in a single clause in a manner compatible with the IAEX; for discussion, see Perlmutter and Postal (1984a: appendix). 8. In some cases, there are independent arguments for the inversion character of the relevant predicates. Chapter 4 provides such an argument with respect to French falloir. 9. There are others, varying across speakers, including pardonner 'to pardon', consentir 'to agree to', etc. 10. Rule [23] combines with the view of the Extraposition of Indefinite (EXI) construction discussed below in Chapters 3, 4 and 5 to predict the apparent anomaly noted in Kayne (1975) and discussed in detail in Postal (1982) that (dés)obeir passives have no EXI variants:
< previous page
page_88 If you like this book, buy it!
next page >
< previous page
page_89
next page > Page 89
The contrast between (aii) and *(bii) is a result of rule [23], given the view of the EXI construction discussed below. Under this, it involves the sponsoring of a dummy 2 arc, B, by a 2 arc, A, which must have an 8 arc local successor. This means that e.g. beaucoup d'ordres in (aii) and (bii) is a final 8. But, of course, given [23] it is impossible for the only nondummy 2 arc in (bii), namely, the 2 arc local successor of the initial 3 arc, to have an 8 arc local successor, since [23] requires in effect that it have a 1 arc local successor. This account is slightly different from that in Postal (1982), since it assumes a different view of the EXI construction. 11. As Y.-C. Morin of the University of Montreal has kindly reminded me, there is at least one other apparent exception to this generalization, namely, the intransitive V vivre (avec) 'to live (with)'. This permits no 2 in actives, permits no passives, but nontheless occurs unproblemati-cally in the 'object raising' construction (i):
My guess is that these facts should be described by claiming that the relational type occurring with vivre in simple clauses like (ia), call it a comitative, whose associated R-sign is Com, is allowed to advance to 2, but only if the resulting 2 arc local successor has a 2 arc foreign successor defining the 'object raising' construction. The relevant rule might be along the lines of (ii):
Under these assumptions, the true irregularity in (id) involves exceptional advancement to 2, rather than exceptional object raising of a non-2. These facts have, I believe, no direct bearing on the correct account of the apparently anomalous behavior of (dés)obéir in the 'object raising' construction. However, if (ii) is essentially correct, then the account of the anomalies in (id) and [24f] in the text are the same. Each involves exceptional advancement to 2, masked by some requirement that the resulting 2 arc local successor have a successor. 12. This analysis assumes, nonessentially, that 'object raising' involves raising of the complement 2 to be a main clause 2, with subsequent advancement to its final status as main clause 1. For discussion of French 'object raising' in transformational terms, see Fauconnier (1974), Ruwet (1982: Chapter 2). 13. Replace is used inter alia in the description of anaphoric relations and the description of prepositional flagging, for the latter, see Chapter 6. Many rules like [26e] requiring (local) successors for arcs, e.g. rules imposing obligatory advancement or demotion, need to pick out the relevant predecessors in terms of Free. The reason is that without this limitation it would, given the APG representation of 'coreference' via overlapping arcs, often wrongly be required that an arc which is replaced by a 'coreferential' arc also have a local successor. For instance, a typical representation of a clause in which the initial 1 and 2 are 'coreferential' is, in APG terms, in relevant part (i):
< previous page
page_89 If you like this book, buy it!
next page >
< previous page
page_90
next page > Page 90
In this case, A and B cosponsor C, which replaces B, which is hence not free. As it stands, a rule like [26e] properly could at best require that C have a term arc local successor. But a rule analogous to [26e] without a specification analogous to the Free condition could wrongly require this of B as well, blocking many perfectly grammatical sentences. For another example of the use of Free in rules, see [63] in the text. 14. Marantz's comments are independently odd in that Japanese permits passives in which the 3 of a ditransitive clause ends up as the surface 1; see Kuno (1973: 348-349). Hence this property appears to be a poor ground for assigning the status 2. 15. The facts cited for Ancient Greek appear quite parallel to those made reference to in the previous note for Japanese. That is, the 3 of a Japanese distransitive clause can be the 1 of a passive clause, but there are no clauses in which the 3 of a ditransitive appears as the superficial 2 of an active clause. Japanese might then also have a rule like [27]. 16. See Postal (to appear a, to appear c); in the former, it is argued that the relation 6 underlies the French clitic en. 17. Outlast is a predicate which says of two neighboring arcs that the last coordinate index of the one instantiating the first argument is greater than the last coordinate index of that instantiating the second argument. Hence if A outlasts B, it is in a later stratum than B. 18. An equivalent formulation would have the consequent: Prepassive(A), where the notion Prepassive is that introduced in Chapter 4, section 1; see note 22. 19. Many Phillipine languages, e.g. Cebuano (see Bell (1976, 1983)) have been analyzed as allowing quite free advancement to 1, including 31A. But, in so far unpublished work, Donna Gerdts of the State University of New York at Buffalo argues convincingly that these are instead instances of advancement to 2. Hence these cases, which at first glance seem to preclude the possibility of banning 31A, do not do so. 20. Technically, this can only be guaranteed by specifying that Korean lacks what are called 'tertiary passives' in section 2.3.8 below.
< previous page
page_90 If you like this book, buy it!
next page >
< previous page
page_91
next page > Page 91
21. At least in the system of JP (1980); see Theorem 19. I can imagine grounds for weakening this position, and allowing multiple successors in certain cases. But any such formulation would still preclude multiple successors in cases like those in the text. 22. The term 'prepassive' referring to a property of arcs is precisely defined in Chapter 4, section 1. Roughly, an arc is prepassive if it is the 2 arc local predecessor of a 1 arc in a passive clause. 23. Early RG work had a principle, called the Relational Annihilation Law, which, inter alia, precluded the existence of structures like (56b). Roughly, this principle said, in current terms, that an arc overrun by a term arc successor had to have an 8 arc local successor. This principle, later referred to as the Chômeur Law, has been abandoned in both RG work (see Perlmutter and Postal (1983a: section 8)) and in JP (1980: Chapter 8). Dowty (1982a: 121-123) cites Chichewa passives like (54a) as posing a problem for RG. But the only problem known is eliminated with the elimination of the Chômeur Law, a rejection demanded by many other facts as well. This rejection does not, contrary to the implications of Dowty's remarks, require any special assumptions about the structure of cases like (54a), which I stress is also found in English examples like (55a). Further discussion of the phenomenon involved is found in the treatment of Chi-Mwi:ni below. 24. This remark holds only for structures with advanced 3s; when advanced instrumentals and locatives are considered, the situation is more complex. 25. One can think of IA creators as simply a subset of arcs with the property + =AgSponsor(A), where the Ag arc is a branch of the P arc neighbor of A. The grammar then must segregate this class of arcs into two subsets, one relevant for prefix agreement, where the Sponor is a final 1 arc, and the other relevant for IA. 26. A formulation simpler than [59] is possible making use of the notion Induce introduced below; see [66]. 27. Dowty (1982a: 121) implies that the fact that Chichewa benefactives only appear in nonpassives as 2s presents a problem for RG. But he fails to indicate what the problem is. The restriction involved is easily represented in APG terms along the lines in [63]. In other cases in other NLs, good evidence has been presented for such 'obligatory' advancement. Postal (1982) presents such for the advancement of 3s to 2 with certain Vs in French; Gerdts (1981) presents evidence for the necessary advancement of 3s and benefactives to 2 in Halkomelem; Aissen (1983) presents relevant evidence for Tzotzil. 28. There are reasonably clear cases where benefactives advance to 3, but do not then advance to 2, for instance, in French examples like (i):
In (ib,c), the benefactive nominal is a 3, although in (ia), it is not. For further discussion of benefactive advancement, see the end of section 2.3.8. 29. To guarantee the truth of this statement for structures like the passive in (61c) involves a technical issue in APG theory, namely, the necessity to block a priori substructures like (i):
< previous page
page_91 If you like this book, buy it!
next page >
< previous page
page_92
next page > Page 92
For in (i) C does not overrun B, since its first coordinate index is not + 1 of Bs last coordinate index. We want a principle which only allows structures like (ii):
The matter is discussed in JP (1980: Chapter 8); unfortunately, no fully satisfactory principle guaranteeing the desired result without blocking other apparently allowable structures has been found. The discussion in the text assumes though that some sentence law determines that (i) is not a possible substructure in any NL. 30. The Stratal Uniqueness Law (see Perlmutter and Postal (1983a), JP (1980)) says that distinct facsimile term (1, 2 or 3) arcs cannot share a coordinate (hence stratum).
< previous page
page_92 If you like this book, buy it!
next page >
< previous page
page_93
next page > Page 93
31. It is possible and I think likely that the postverbal word order illustrated for a nominal like mwi:zi in [84c] is not a case of postverbal positioning of a surface 1, but rather of an 8 (earlier 1), determined by an invisible dummy final 1. For discussion of a similar idea with respect to many cases in Italian and discussion of the general issue, see Perlmutter (1983b). 32. 'Displace' is a convenient term used by KA, but one not precisely defined. Informally, one can say that one nominal, Na, displacs another, Nb, if Na corresponds to the head of a successor arc which overruns an arc headed by Nb. 33. Note that [85] also blocks IA in all clauses involving unaccusative advancement. But no data in KA are relevant to testing this claim. 34. Rule [85] thus claims that there could be no IA in a previously intransitive clause in which an instrumental advances to 2, that is, one in which the advanced instrumental has displaced no other nominal. This would include clauses like The man farmed the hoe meaning 'The man farmed with the hoe', etc. KA give no data relevant to this prediction, but it would not be very surprising if it was incorrect. In this case, [85] would have to be somewhat modified to allow the 2 arc local successor of an Inst arc to be an IA creator if it overruns no arc. 35. The fact that the Vs in these examples have the applied suffix does not constitute an argument that the instrumental has advanced, because the advancement of the benefactive suffices to determine the presence of that suffix in ChiMwi:ni. 36. For simplicity, the immediately following discussion ignores the possibility that benefactives advance to 2 via 3. This does not materially affect the discussion. 37. [93] does not include a representation of the fact that Luti is a topic, or of the postverbal ordering property of mwi:zi in [84c]. 38. It is probably possible to simplify [94] to something like:
This would be possible under the surely correct assumption that no foreign successor of an Inst arc ever overruns anything and that no foreign successors can induce 1 arcs. I know of no cases of the latter sort in any NL, so the relevant simplifications seem plausible. Moreover, no doubt nothing in Chi-Mwi:ni requires that instantiations of the variable A be characterized as 2 arcs since I assume that all local successors of Inst arcs are 2 arcs (in Chi-Mwi:ni, and possibly in all NLs). Hence the 2 arc specification in (i) is probably also not necessary. 39. Compare the description offered here, made possible by the APG framework, with the conclusions of KA themselves (1977: 217-218), who could offer no grammatical rules for the phenomena they discussed and who were driven to appeal to the unclear, undeveloped, and, I suspect, incoherent notion that the object relation in Chi-Mwi:ni is nondiscrete: Under this view, instruments are not as fully principal objects as, say, beneficiaries (since instruments cannot control an OP whereas beneficiaries do). On the other hand, instrumentssince they can passivizedo participate in the principal object relationship in part, unlike the NPs displaced by a beneficiary or indirect NP ...' 40. English tertiary passives are discussed in greater detail in Postal (to appear b), which justifies such an analysis for the clauses in question, as opposed, for example, to one similar to that which has sometimes been suggested in transformational terms, which would see the postverbal nominal as a superficial 3 with a deleted preposition. 41. In general, reflexive passives, discussed in the text in Chapter 4, section 6, are necessarily transitive in the first stratum containing the 'new' 1 arc defining a passive.
< previous page
page_93 If you like this book, buy it!
next page >
< previous page
page_94
next page > Page 94
42. [97] illustrates constraints on tertiary passives having to do with the choice of V. There are also constraints as to the nature of the nominal heading the 3 arc; in my variant of English, it may be that this nominal must be a single word:
Or, perhaps, as E. Tauss has observed, even a monosyllabic word: 43. For discussion of this notation and its claimed lack of reference in NLs, see Perimutter and Postal (1984b). 44. The former law says that every basic clause final stratum contains a 1 arc; the latter claims that an arc with an 8 arc local successor always is overrun. For discussion, see Perlmutter and Postal (1984a, 1984b), JP (1980). 45. There is no need in rules like [105] and [106] to specify that arcs instantiating the variable A be final. As discussed in JP (1980: Chapter 14), this is made unnecessary by the Unique Eraser Law, which limits each arc to a maximum of one eraser. For a nonfinal arc always has a local eraser and hence, if it has a foreign successor, it will have two erasers, certain special (but irrelevant) cases aside. 46. This idea is further supported by the discussion of French 'control' facts in Chapter 3, section 3.
< previous page
page_94 If you like this book, buy it!
next page >
< previous page
page_95
next page > Page 95
Chapter 3 The Inadequacy of the Relational Grammar Account of Impersonal Passives 3.1 The Relational Grammar View Previous discussion has largely taken for granted for RG view of IPs; see Perlmutter (1978a, 1982), Perimutter and Postal (1977, 1984a, 1984b). In these terms, IPs have been characterized as passive clauses in which the nominal heading a 2 arc at one level and a 1 arc at the next is a 'dummy'. One problem with this view is that, within RG, the notion 'dummy' is not characterized and remains informal. More precisely, no way has been given to mechanically scan an RG RN and determine which arcs are headed by 'dummies'. Let us assume this lacunae to be filled by the APG account in JP (1980: Chapter 10), where the notion 'dummy nominal' is precisely formulated, as briefly discussed in Chapter 4, section 2. However, even supplemented by a precise account of 'dummy nominal', the extant RG view remains inadequate; its account of IPs provides only a necessary condition for IPhood. This is noted in Perlmutter and Postal (1984a; 1984b). But no proposal was made to deal with the problem that certain passive clauses in which the nominal occurring as both 2 and 1 is a dummy are nonetheless not IPs but rather PPs.
< previous page
page_95 If you like this book, buy it!
next page >
< previous page
page_96
next page > Page 96
3.2 English Personal Passives with Dummies Dvamced From 2 to 1 Consider [1]: This is uncontroversially a passive with a dummy as its superficial 1. English does not in general permit IPs, as shown by [2]: 1
Clearly, the dummy it in [1] is linked to the extraposition construction found also in nonpassives like [3] and thus is independent of passivization:
Analyzing [1] as an IP would thus have the following correlated drawbacks. Given *[2a-d], such a step renders the possibility of cases like [1] exceptional and ignores as an accident that the independently existing English extraposition construction provides a (non-IP) analysis of [1]. Thus taking [1] as an IP pointlessly recognizes an entire exceptional construction in English, when the only sentences at issue fall out from the analysis of an independent construction. Evidently then, a correct account of IPs should determine that [1] is not an IP.2 Since [1] is unchallengeably a passive, the only way to guarantee the result is to analyze it as a PP. To achieve this under the RG view of IPs, the dummy in [1] must not be analyzed as an earlier 2. This is accomplished by taking [1] as the 'extraposed' version of [4]: Under this view, the nominal in [1] which is both a 2 and a later 1, defining the clause as a PP, is the that complement, not the dummy. While initially plausible and possibly viable for certain cases, including [1] itself, this approach to guaranteeing that all cases like [1] are not IPs in RG terms cannot really work because of Vs like those in [5], cited in Chomsky (1981: 122) and Williams (1981):
Since Vs like hold and feel do not permit clausal 1s in passives, it is initially unclear whether anything motivates saying that complements of these Vs can
< previous page
page_96 If you like this book, buy it!
next page >
< previous page
page_97
next page > Page 97
ever be 1s. It might appear at first that *[5b] is not relevant to this issue, since the necessary constraint on these Vs might seem to be stated just as economically under the view that [5a] involves passivization of the complement and 'insertion' of it only as a 1. That is, the constraint could just be, informally, [6]: This would block cases like *[5b] while allowing [5a], even under an analysis of the latter where the complement is a 1 at one level. Other Vs which pattern like hold with respect to the constraint in [6] include conclude, find, recall and say. However, there is good reason to reject a constraint like [6] and to conclude that the proper restriction on these Vs is incompatible with the claim that the complement in cases like [5a] is a 1 at any level. My overall view is that 'extraposition' is in general free for that clause 2s just as it is for 1s. This is disguised by the fact that, with few exceptions, the dummy it associated with 'extraposition' must be invisible as a 2, though not as a 1. Against this background, Vs of the class covered by [6] differ from freer Vs like believe, for which 'extraposition' is not required, in that 'extraposition' is required for their complement clause 2s. 'Extraposition' is to be understood here as meaning the existence of a pair Sponsor(A, B), where A is the 2 arc whose head corresponds to the complement clause and B is a neighboring GHOST 2 ARC 3 whose head would be a dummy nominal. Associated with such a sponsor pair is a local successor sponsor pair including the 2 arc whose head is the complement clause. In this case, the sponsored arc is an 8 arc, indicating that the complement ends up as an 8. In these terms, 'extraposed' 1s and 2s correspond to the respective APG substructures [7]:
< previous page
page_97 If you like this book, buy it!
next page >
< previous page
page_98
next page > Page 98
In contrast to dummies associated with 'extraposed' 1s, those associated with 'extraposed' 2s are rarely VISIBLE in English because of a rule requiring, with few exceptions, that 'extraposition' dummy 2 arcs self-erase (if they do not have (1 arc) local successors). 4 Normally in English, the substructure in [7b] must actually be a subpart of one of the more articulated structures [8]:
[8a] corresponds to a passive, [8b] to an active in which the 8 status of the complement is disguised. But the dummy is actually visible as a superficial 2 in cases like [9]:
The V take of [9a] precludes invisibility of the extraposition dummy, while most Vs require it. The V mention, for many speakers at least, neither requires nor precludes such invisibility: The analysis I am suggesting claims that whereas with a V like believe, which is not subject to obligatory extraposition of its complement,5 an active clause can contain a surface complement as 2, this is impossible for Vs like hold, feel, etc., whose surface complements must be 8s. This predicted difference shows up in contrasts for topicalization and 'object raising'. These are both impossible for straightforward 'extraposition' 8s, as shown by [11]-[14]:
< previous page
page_98 If you like this book, buy it!
next page >
< previous page
page_99
next page > Page 99
Strikingly though, the complements that go with Vs like conclude, feel, hold, etc., behave like 'extraposed' complements in contrast to those with Vs like believe:
These facts support the view that Vs like hold do have 'extraposed' complements, even in active clauses. 6,7 It is then not true, as a constraint like [6] suggests, that the special characteristics of Vs like hold are limited to an impossibility of complement 1s. Something more general must be said. Informally, I suggest that the relevant constraint is [16]:
This constraint, unlike that in [6], determines both the restriction leading to *[5b] as well as those underlying the starred versions of [15]. But this statement of the constrained precludes a complement with these Vs from ever being a 1.9 If [16] is correct, clauses like [5a] can, in pure RG terms, be properly characterized as passives only if the dummy is both a 2 and a later 1. But while yielding the correct result that [5a] is a passive clause, this has the unwanted, surely incorrect consequence that [5a] is an IP. This is a first indication that the RG approach to IPs yields only a necessary not a sufficient condition for IPhood. 3.3 French Personal Passives with Dummies Advanced From 2 to 1 A case similar to that in section 3.2 can be built on parallel phenomena in French. The case is stronger because French does permit clausal IPs, and it is thus possible to show that some clauses which would apparently be IPs by the RG account are not only intuitively non-IPs but also are in contrast with genuine French IPs; see Chapter 4, section 4. Consider [17]:
< previous page
page_99 If you like this book, buy it!
next page >
< previous page
page_100
next page > Page 100
Both [17a and b] are uncontroversially passives, having standard features of French passives independent of cases with dummy 1s. These include a future form sera of the passive auxiliary être, in agreement with the final 1 of that auxiliary, il, 10 the V in past participle form (mis), and an optional nominal (Pierre) in a par prepositional phrase understood as the initial 1. While both [17a and b] are passives, and impersonal in the sense of having a dummy superficial 1, only [17a] is a genuine IP. One wants [17b] to turn out to be a PP, corresponding to the clear PP in [18]:
More specifically, [17b] should be a special case of the 'extraposition of indefinite' (EXI hereafter) construction which exists in French independently of passivization; see Postal (1982). The relation between [17b] and [18] is then just an instance of that holding between the nonpassive pair [19]:
For an extensive RG discussion of structures parallel to [19b], see Perlmutter [1983b]. In particular, the constraints on the class of postverbal nominals which can occur in clauses like [19b], constraints inadequately subsumed under rubric 'indefinite', hold identically for the parallel nominals in clauses like [17b]. As with English [1], [17b] does not initially seem to cause too much problem for the RG view of IPs. It can be claimed that the nominal in [17b] which is both a 2 and a 1 is certaines bouteilles and that the dummy is only a 1, just as the dummy in [19b] is only a 1. This is not only initially plausible but is consistent with much past description of the EXI construction, which has treated it as a phenomenon in which Is take on some special status, with the dummy il acceding to the status of 1. In particular, this approach is consistent with the account of EXI in Postal (1982). However, I claim that this traditional view is mistaken and that there is no real evidence that a postverbal nominal like certaines bouteilles in [17b] is a 1 at any level.11 On the contrary, there is evidence, considered presently,
< previous page
page_100 If you like this book, buy it!
next page >
< previous page
page_101
next page > Page 101
that it is not a 1 at any level. If so, an RG characterization of cases like [17b] as passives can only be based on the fact that the dummy is an earlier 2 advanced to be a later 1. But then the RG account of IPs as is would wrongly specify [17b] as an IP. This is an unacceptable conclusion intuitively and one which offers no basis for an important contrast between [17a and b], considered in Chapter 4, section 4. Let us turn to the question of whether nominals like certaines bouteilles in [17b] are 1s at any level. For convenience, let us refer to these as PIVOT (nominals). Consider the following partial 'control' paradigm due originally to Ruwet (1972: 120):
Although these examples contain an untensed gerundive expression en hurlant with no expressed 1, the unexpressed 1 of the gerundive is understood as identical to one of the nominals which precede it. Strikingly, as Ruwet observed, in the active [20a], the invisible 1 can only be understood to be the clausal 1, les policiers. But in the PP [20b], the invisible 1 is ambiguously understandable to be either the superficial clausal 1, Beaucoup d'étudiants or the nominal in the par phrase. Further relevant data include [21]:
In [21a,b and c], the invisible 1 of en hurlant is unambiguously identified with Pierre. It is impossible to understand it to be the nominal after the prepositions sur, à, or de. The generalization seems clear. The invisible 1 of an adverbial phrase like en hurlant can always be understood to be the final 1 of the clause containing it 12 but in general no other nominals except that in
< previous page
page_101 If you like this book, buy it!
next page >
< previous page
page_102
next page > Page 102
the par phrase of a passive. 13 Similar constraints govern expressions like après avoir bu 'after having drunk'. Why this peculiar disjunction? Why does the nominal in passive par phrases have a feature not associated with most other nominalsand specifically not with most others in prepositional phrasesthe property of behaving like a superficial 1? Under the RG/APG account of passive clauses underlying the present study, an answer is immediate. The disjunction is just an artifact which comes from the special structure of passive clauses. Informally, the real principle governing the relation between the invisible 1 of gerundives like that in [20] and the nominals of the main clause is the disjunction-free statement [22]:14
Thus in the active, [20a], where les policiers is both initial and final 1, only it is equatable with the invisible 1. But in current terms, both nominals in a PP like [20b] meet [22], les policiers because, although a final 8, it is the initial 1 and beaucoup d'étudiants because, although an initial 2, it is a later and final 1. For cases like [21], only the superficial 1 is a possible 'controller' because only it is a 1 at any stage. Thus, in current terms, the apparently exceptional behaviour of the par phrase nominal in passives reduces to a regularity because among the nonsuperficial 1s considered, only this nominal is a 1 at an earlier level. Facts like these provide one of many similar arguments for the kind of multilevel approach to passivization characteristic of RG and APG. More precisely, such facts support the claim that nominals referred to as passive 8s in RG/APG terms are 1s at some (in fact, earlier) level.15 For previous RG work supporting this kind of multilevel analysis, see Perlmutter (1981, 1982, 1983b, 1984a, 1984b). Hubbard (1982), Jackson (1982), Davies (1981a, 1981b, 1981c, 1982, 1984), Rosen (1981a, 1982), Timberlake (1982) and references therein. Directly pertinent to present concerns, [22] provides a way to determine whether EXI pivots like that in [17b] are 1s at any level. If they are, [22] will not preclude passives like [17b] from having the ambiguous 'control' properties of ordinary PPs like [20b]. If, however, pivots are not 1s at any level, [22] determines that passives parallel to [17b] are unambiguous with respect to 'control', which is in fact correct; see Ruwet (1982: 288)
Here the invisible 1 of en hurlant is understandable only as being les policiers.16 This conclusion is fortified by 'short' ordinary PPs like [24]:
< previous page
page_102 If you like this book, buy it!
next page >
< previous page
page_103
next page > Page 103
These clauses, understood to have unspecified initial 1s, are unambiguous with respect to the en hurlant phrase, which is only understood as having its 1 identical to the specified nominal beaucoup d'étudiants. That is, apparently the 'controller' of a phrase like en hurlant cannot be an invisible, unspecified nominal. Predictably then, the pivotcontaining variant of [24] is ill-formed since there is then no nominal at all available to be understood as the I of the adverbial phrase: 17
The same considerations predict the following facts:
*[26c] is ill-formed because the only available 'controller' is beaucoup de murs, an inanimate nominal incapable of being the 1 of a predicate 'scream'. The 'control' generalization in [22] provides a strong argument that EXI pivots are not at any level Is of the passive clauses containing them. For, if this is the case, the independently needed [22] suffices to predict the impossibility for pivots to serve as 'controllers'. If, on the contrary, pivots were 1s at one level, only an additional, apparently ad hoc, constraint could yield this result. Since examples like [17b] are uncontroversially passives, the conclusion that pivots do not head 1 arcs means that an RG approach must take the dummy superficial 1 to be an earlier 2. As in the earlier discussion of English examples like [5a], while correctly specifying the relevant clauses as passives, it yields the unwanted result that cases like [17b] are IPs. I take this as a further indication that the RG account of IPs as is is too limited, that is, provides only a necessary condition for IPhood.
< previous page
page_103 If you like this book, buy it!
next page >
< previous page
page_104
next page > Page 104
3.4 More English Personal Passives Involving Advanced Dummies Sections 2 and 3 provided somewhat involved and indirect arguments for the following view:
To complete the present support for [27], it suffices to observe English examples like [28]:
Each is a passive clause with a dummy superficial 1, and each appears to be simply the respective passive of the corresponding active in [29]:
There is then a good prima facie case that the dummies in [28] are 2s at an earlier level, just as they are 2s in [29]. Thus, an unadorned version of the RG characterization of IPs makes [28a,b and c] IPs, contrary to intention and contrary to the otherwise unexceptional generalization that English does not allow clausal IPs. Evidently, a correct account should determine that [28a,b and c] are merely the PP correspondents of [29a,b and el. 3.5 Summary The conclusion so far is then that while all IPs are passive clauses in which a dummy nominal is both a 2 and later level 1, this is only a necessary condition. To provide a sufficiency condition requires a more articulated account of the notion IP than is possible in RG terms alone. However, the basic concepts of APG easily permit such an articulation. Notes: Chapter 3 1. In transformational work, it was sometimes assumed that clauses like (i): are related directly to those like (ii): This approaches calling (i) an IP. However, I think the parallelism is misleading and that cases like (i) involve not a main clause predicate bombed but a nominal a building bombed.
< previous page
page_104 If you like this book, buy it!
next page >
< previous page
page_105
next page > Page 105
Put differently, I take (i) to be parallel to French (iii) not to (iv):
2. This is contrary to the discussion in Keenan (1975: 348), who takes cases like [1] to be IPs. 3. The notion Ghost arc is introduced in JP (1980). Roughly, such an arc has a single nonoverlapping sponsor and an Rsign which is a member of the class of nominal R-signs, e.g., 1, 2, etc. See Chapter 4 for further discussion. 4. The restriction in question is statable in terms of the APG notion Output arc, that is, one with no local eraser. As background, I assume a general restriction limiting invisible pronominal 2 arcs in English to those which are ghosts:
The invisibility of relevance is taken to be due to self-erasure. The restriction is then:
The logic of the description is this. Rule (i) allows ghost 2 arcs to erase themselves in general in English. Selferasing arcs are not output arcs. Therefore, rule (ii) says that ghost 2 arcs which have not erased themselves must cooccur with the limited class of Vs. Note that previous rules do not preclude ghost self-erasure with take: This requires the ad hoc statement:
The reference to 'output arc' in rule (ii) correctly does not preclude the extraposition ghost 2 arcs from having I arc local successors to yield passives like [5a]. For arcs with local successors are never output arcs. Note that rule (iv) correctly blocks passives when take has a complement: Since (iv) requires the ghost to be an output arc, it cannot be the local predecessor of the I arc needed to determine the passive; local predecessors are always erased and hence never output arcs. 5. This is a discursive oversimplification. The actual constraint for the limited class of Vs requires, I think, that the initial 2 arc in such cases have an 8 arc local successor. This is satisfied both in extraposition constructions and raising constructions like (i): Under a raising analysis, the clausal complement is a final 8. 6. Williams (1981: 95-96) suggests in effect that the contrast between pairs like (i) and (ii):
< previous page
page_105 If you like this book, buy it!
next page >
< previous page
page_106
next page > Page 106
lies in the fact that Vs which pattern like that in (i) occur with NP objects; while those which pattern like that in (ii) do not. Hence he offers such contrasts as:
Interpreted in the strongest way, Williams' claim is a biconditional which says, schematically:
While there may be some tendency in the direction of this claim, it is not actually true. For instance, say with the habitual meaning of French on dit patterns like (ii) but occurs with NP objects in Williams' terms.
Moreover, note that in this case, say patterns like [11]-[13] with respect to complement topicalization: Thus, contrary to what Williams claims, the constraint blocking clausal 1s in cases like (iia) cannot be attributed to the failure of the relevant Vs to take NP objects; (iva) is not literally true. On the other hand, I know of no exceptions to (ivb). 7. Relevant to the remarks of the preceding note is the behavior of insist:
The paradigm of insist is not like that of either Vs like believe or those like hold; for me, insist permits no passives with complements (but cf. Rosenbaum (1967: 83)), although it does otherwise: Hence this V is also inconsistent with [4a] of the previous note. 8. As indicated in note 5, the constraint is more general. 9. This follows since, in the system of JP (1980), no arc can have two successors. Hence, if the initial 2 arc is required to have an 8 arc local successor, it cannot have the I arc local successor needed to define a passive clause in which the complement is final 1. 10. This is a slight simplification in that il is arguably not the final I but a clitic heading an arc sponsored by the final I arc. 11. Burzio (1981) takes the postverbal nominal in cases like [17b] in effect to be both an initial and final 2 whose contrast with other final 2s is due to the fact that its abstract case-marking is nominative rather than the expected accusative. While I regard this as incorrect, the differences between that view and the current one, in which the nominal is an initial 2 and then final 8, are rather subtle and it is not easy to find cross-framework evidence to distinguish them.
< previous page
page_106 If you like this book, buy it!
next page >
< previous page
page_107
next page > Page 107
12. Assuming that nominal has some semantic content and is not a dummy like that underlying il. 13. One other case, that of inversion 3s, is discussed below. 14. [22] is, of course, only a necessary condition. 15. As argued in different terms in Chapter 2, section 9. 16. Word order is irrelevant. The same relations hold in (i) (although, to be sure, it is stylistically very heavy): 17. As expected, if the adverbial is removed from *[25], the result is well-formed.
< previous page
page_107 If you like this book, buy it!
next page >
< previous page
page_108
next page > Page 108
Chapter 4 An Arc Pair Grammar Definition of Impersonal Passives 4.1 The Arc Pair Grammar View of Clause Structure and the Problem of Impersonal Passives As mentioned in Chapter 1, the essential APG idea with respect to sentence structure is that sentences consist of two basic binary relations between arcs, Sponsor and Erase. Put differently, an NL sentence is a set of two collections of ordered pairs of arcs, where the two collections correspond to these two relations. Every arc has at least one sponsor, some have two. If an arc is in the range of Erase, it is not a 'superficial' arc. I now apply the APG view of sentence structure to the problem of IPs, which, at this point, has in effect been argued to have the following character:
< previous page
page_108 If you like this book, buy it!
next page >
< previous page
page_109
next page > Page 109
Before turning directly to [1b,c and d], recall the APG 'translation' of the RG notion of passive clause, in terms of which it is possible to introduce some new and useful terminology. Within the APG framework, a passive clause involves the following configuration:
Thus a passive contains three arcs A, B, C such that 1 = 2Local Successor(A, B) and Overrun(A, C). One can then introduce [3]:
These definitions in effect convert the structural configuration in [2] into properties of the relevant arcs. 1 4.2 Ghost Arcs and Dummy Nominals The problem of IPs is now interpretable as that of picking out a certain class of arc-passive arcs, or, equivalently, of prepassive arcs. Previous discussion
< previous page
page_109 If you like this book, buy it!
next page >
< previous page
page_110
next page > Page 110
indicates that an arc-passive/prepassive pair in an IP will 'be headed by a dummy nominal. It is thus necessary to inquire as to what APG says about these, since I have adopted the RG view that this notion is fundamental to an account of IPs. In JP (1980) dummy nominals are those corresponding to the heads of arcs called GHOST arcs. Roughly, these are arcs having one NONOVERLAPPING sponsor whose R-sign is a member of a class associated with arcs whose heads are nominals. This set includes 1, 2, 3, 8, etc. However, that work assumes, and nothing so far motivates weakening this claim, that all ghost arcs and all ghost arc sponsors are either 1 or 2 arcs. There are then at most the four possible local 2 ghost sponsor situations in [4]:
< previous page
page_110 If you like this book, buy it!
next page >
< previous page
page_111
next page > Page 111
Of these, only the first three seem to be attested. The first new claim that can be made about IPs is this: Principle [5] immediately reveals why English passives involving dummies, like [3.28b] above (repeated here as [6a] together with parts of its structure) are not IPs. The representation in [6b] ignores the upstairs passive auxiliary and the infinitival marker to:
[6b] is a raising structure. So the prepassive main clause arc B, a FOREIGN SUCCESSOR of the final 1 arc, C, of the complement, has an overlapping sponsor, and hence is not a ghost. The ghost arc in the structure of [6], D, which defines there as a dummy, is a complement clause arc, thus irrelevant to the nature of the main clause. Consequently, the latter is properly characterized as a non-IP by [5].
< previous page
page_111 If you like this book, buy it!
next page >
< previous page
page_112
next page > Page 112
4.3 Ghost Sponsors and the Characterization of Impersonal Passives While [5] represents an evident improvement on the RG account of IP, it remains insufficient. In particular, [5] fails to specify that English clauses like [3.5a] and French clauses like [3.17b], both argued earlier to involve dummies introduced at 2s, and thus, in APG terms, ghost 2 arcs, are not IPs. However, the APG assumption that every arc, hence every ghost 2 arc, has a sponsor suggests that the sufficiency condition for IPs might be statable in terms of the nature of the sponsor of the 2 arc ghost required by [5]. This approach was in effect developed in JP (1980: Chapter 10). Adopting that view, an IP is precisely defined as a passive whose prepassive arc, A, is not only a ghost but one sponsored by the 1 arc which the arc-passive arc (the local successor of A) overruns. In other words, an IP is a passive whose prepassive arc is a ghost sponsored by the passive-destroyed arc; all IPs therefore model the ghost sponsor pattern in [4c]. JP (1980) distinguishes two types of ghosts, depending on whether their sponsors are facsimiles (same R-sign neighbors) or not. A ghost having a nonfacsimile sponsor is said to be UNSTABLE, otherwise STABLE. The current claim is then that IPs are all and only those clauses containing unstable prepassive arcs. Let us introduce the notion [7]:
Since arc-passive arcs are 1 arc local successors of 2 arcs, [7] says that an arc-passive arc is IP-arc-passive if and only if its (prepassive) local predecessor is unstable. This is equivalent to saying that the prepassive arc is sponsored by the arc overrun by the arc-passive arc. In these terms, an IP is simply a clause having an IP-arc-passive arc, a PP a passive clause which is not an IP. To illustrate how the extended APG account of IPs just introduced provides a solution to the problem of characterizing IPs, let us return to the two French clauses in [3.17] above and add a third, so that together these illustrate, I claim, all three attested ghost arc patterns in [4c]:
< previous page
page_112 If you like this book, buy it!
next page >
< previous page
page_113
next page > Page 113
What should be shown in particular is how a definition of IP in terms of the notion provided in [7] 3 correctly distinguishes [8b and c], specifying only the latter as an IP in a way in which neither the RG account nor an APG account limited to [5] could. While each of [8a,b and c] has a final dummy 1, and thus, in APG terms, some kind of ghost arc, no two have the same structure. [8a] manifests none of the properties of French passive clauses. I claim that it involves INVERSION (that is, 1 to 3 demotion [13D]), mentioned earlier, and that the ghost arc is a 1 arc. The relevant structure of [Sa] is [9]:
The structure of [8a] thus instantiates ghost pattern [4a], with A a stable 1 arc ghost. Since no arc in [9] is a 1 arc local successor of a 2 arc, none is arc-passive, and hence none is IP-arc-passive. The assumption that [8a] is a genuine 13D structure, that is, the view that its uncontroversial superficial 3 arc is actually a local successor of a 1 arc, can be supported via the argument (cited for French earlier) sustaining constraint [3.22]. This states that the unexpressed 1 of various 'adverbial' expressions can only be understood to be a nominal which heads some 1 arc in the main clause. For the superficial 3 of falloir clauses like [8a] is a possible controller for such expressions, as claimed by this constraint only given the 13D analysis, under which the 3 arc of these clauses is the local successor of an earlier 1 arc:
< previous page
page_113 If you like this book, buy it!
next page >
< previous page
page_114
next page > Page 114
Strikingly, the hypothesized 13D 3 in [10a] is a possible 'controller' of the invisible 1, but the non-13D (in fact, initial) 3 in [10b] is predictably not. Even more strikingly, the non-13D (in fact, initial) 3 in the EXI PP sentence [10c] is also not a possible 'controller'. This shows that no account of cases like [10a] which appeals merely to the presence of a superficial dummy can properly characterize the class of relevant 'controllers'. Structure [9] assumes that the dummy nominal in [8a] underlying il heads a ghost 1 arc, and no other arc. The possibility that this dummy heads a ghost 2 arc is excluded inter alia by the fact that falloir occurs with final 2s and that, e.g., certaines bouteilles in [Sa] is a final 2. This is supported by the fact, illustrated in [10d], that this type of nominal can determine an accusative clitic like les, which correlate strictly with final 2s. If the dummy nominal in such cases headed a 2 arc, that would overrun the initial 2 arc, which could then not, given the Stratal Uniqueness Law, be a final 2 arc. Moreover, if the dummy headed a 2 arc, there is no reason why falloir clauses would not be characterized as passives, although they manifest no characteristic properties of French (nonreflexive) passives. These conclusions are supported by the contrast between the falloir paradigm, which I claim involves final 2s, and those of certain relatively unique EXI Vs. Example (10ei) is ill-formed because the final pronominal 2 has not determined an (accusative) clitic, as required. 4 This behavior contrasts with that of the EXI V rester, illustrated in [10f]. This differs from other EXI Vs in permitting definite pivot nominals very freely, including definite pronouns. But since I
< previous page
page_114 If you like this book, buy it!
next page >
< previous page
page_115
next page > Page 115
claim that the pivot of an EXI construction is a final 8, the postverbal nominal in, e.g., [10fii] is a final 8. Hence both its failure to determine an accusative clitic and the occurrence of well-formed full pronouns is predicted. Thus the falloir and rester paradigms illustrate two quite different structures corresponding to the French sequence . In the former, the dummy is a 1 and only a 1 and the postverbal nominal is a 2 and only a 2. In the latter, the dummy is an earlier 2 advanced to 1, and the postverbal nominal is an earlier 2 demoted to 8. Examples [8b and c] both differ from [8a] in that each contains a 2 arc ghost, and both these are prepassive. The structure of [8b] is as follows, artificially ignoring the auxiliary and the locative element:
The key here is that the ghost arc, B, is STABLE, since it is sponsored by the 2 are D. This correlates with the fact that certaines bouteilles in [11] is not a 1 at any level. This claim was already supported by the 'control' argument of Chapter 3, section 3. The stable 2 arc analysis in [11] amounts, of course, to the claim that the entire EXI construction in French is characterized by 2 arc sponsors of ghost 2 arcs. 5 In any event, while A in [11], unlike the final 1 arc A of [9], is arcpassive, it is not IP-arc-passive.
< previous page
page_115 If you like this book, buy it!
next page >
< previous page
page_116
next page > Page 116
A useful notion from JP (1980) is that of DUMMY ARC. Dummy arcs are all and only the R-successors of ghosts. Hence all ghosts are dummy arcs, but not conversely. In particular, all arc-passive successors of ghosts are dummy arcs, but not ghosts. Similarly, the foreign successor 2 arc B in the structure [6b] is a dummy arc but not a ghost. Let us introduce a further term, distinguishing those dummy arcs which are R-successors of stable ghosts from R-successors of unstable ghosts:
In these terms, an IP could be equivalently redefined as a clause whose arc-passive arc is an instability dummy arc. Having distinguished [8a and b], it remains for us to consider the much more subtle contrast between [8b and c], revealed by the structure of the latter: 6
< previous page
page_116 If you like this book, buy it!
next page >
< previous page
page_117
next page > Page 117
Ignoring irrelevancies concerning the unclear status of the complex predicate, the relevant point is that the ghost arc B is unstable and, in fact, sponsored by the initial 1 arc, I. Hence, according to the definitions, A, the arc-passive arc, is IP-arc passive, and also an instability dummy arc. 4.4 An Argument Distinguishing French Impersonal Passives From Certain ''Impersonal' Personal Passives I stated earlier that facts internal to French motivated the fundamental distinction drawn here between clauses like [8b and c]. This claims that although both [8a and c] are passives involving dummies corresponding to the heads of 1 arc local successors of 2 arcs, only [8c] is a true IP, while [8b] is a PP of the EXI sort. The distinction between these two types, now rendered precise via the notions Arc-passive, IP-arc-passive, Prepassive, Stable, etc., can be supported via a contrast noticed by Kayne and Pollock (1978). When true French IPs are embedded in either certain subjunctive or 'WH-movement' contexts, the dummy 1 can be invisible, not otherwise in general allowed in French. For discursive simplicity, restrict attention to the subjunctive cases:
In the subjunctive complements of Vs like Vouloir 'to want', exiger 'to require', etc., the final 1 of IPs can evidently be invisible. Kayne and Pollock (1978) indicate that the only type of dummy nominal capable of invisibility in these subjunctive embeddings is that of IPs. Other dummies, including that of weather expressions must be visible here as elsewhere (but see section 4.7 Postscript below):
Crucially, Kayne and Pollock (1978) point out that the dummy associated with EXI in nonpassives also must remain visible in these contexts:
< previous page
page_117 If you like this book, buy it!
next page >
< previous page
page_118
next page > Page 118
The present view that [8b and c] differ fundamentally in structure, with only the former being an IP, then receives support from the hitherto unnoticed fact that such structures contrast in subjunctive environments:
Only examples of the type claimed here to be true IPs permit invisible dummies in the subjunctive environments. This provides an argument for the claim that pairs like [8b and c] differ in structure. Put differently, it argues that the EXI impersonal passives are non-IPs, as claimed by the current reconstruction of the notion 'IP' in terms of IP-arc-passive arcs (equivalently: unstable prepassive arcs). Rather, they are 'accidentally' impersonal PPs. The conclusion follows not only from the fact that pairs like [8b,c] contrast in the subjunctive environments but that the dummy Is of the [8b] type behave like dummy Is of nonpassives. The right dummy Isthat is, those which can be invisible in these environments, are picked out by the notion Instability dummy arc, under the contrasting analyses of the two cases provided in the present framework. 4.5 Formalizing the Subjunctive Restriction The contrast noted by Kayne and Pollock is representable in French grammar using devices already introduced. I assume that the invisible dummies of cases like [17b] involve self-erased arcs, so that the problem is simply to control which French dummy 1 arcs self-erase. I will make no attempt to represent seriously either the subjunctive contexts which have already been shown to permit invisible IP dummies or the other class of contexts shown by Kayne and Pollock (1978) to allow this, namely, so-called stylistic inversion/'WH-movement' environments. Assuming the contexts cited by Kayne and Pollock to be, with certain lexical exceptions, the only ones in French where arc-passive dummy arcs self-erase, one can specify the rule roughly as in [18]:
< previous page
page_118 If you like this book, buy it!
next page >
< previous page
page_119
next page > Page 119
Of course, the inscriptions 'FALLOIR', 'SUBJUNC' and 'STYLE-INV' in this rule are entirely schematic, since the way to represent these contexts is not relevant here. The former inscription refers to the fact that, exceptionally, a few predicates, most notably falloir, permit a nonpassive dummy to self-erase:
The schematic inscription is thus intended to keep rule [18] consistent with the short form of [19]. The rule then claims that otherwise, a French dummy 1 arc can self-erase only if it is both IP-arc-passive and in a subjunctive or stylistic inversion environment. 9 Notice that [18] properly does not require the relevant R-successors of IP-arc-passive arcs to self-erasethe invisibility of the relevant dummies is 'optional'. Consider the following cases of dummy 1s:
< previous page
page_119 If you like this book, buy it!
next page >
< previous page
page_120
next page > Page 120
In all these cases, invisibility of the dummy is not allowed, which follows from rule [18]. In [20] there are extraposition dummies, in one case with an infinitive, in the other with a que clause. Neither contains an arc-passive arc, hence no IParc-passive arc. In [21] the complement has the dummy associated with the impersonal inversion structure of falloir, as represented in [9] above. Again there is no arc-passive arc, hence no IP-arc-passive arc and [18] correctly blocks selferasure. In [22] the dummy is a.n EXI element, combined with a 'cataphoric' structure; there is no arc-passive arc and thus no IP-arc-passive arc, precluding satisfaction of [18]. Example [23] contains a weather dummy, and again there is no arc-passive arc capable of satisfying rule [18]. The most interesting case is [24], which does contain an arc-passive arc. But self-erasure of the dummy 1 arc is correctly predicted to be impossible by rule [18], since there is no IP-arcpassive arc. The prepassive arc is headed by an extraposition dummy 2 arc, which is stable. All the cases of 1 arc selferasure in [20]-[24] are blocked, even though the dummies do occur in subjunctive environments. Such blockings give initial support to rule [18]. Further predictions are made by [18]. Consider the case where true IPs are embedded below Vs which permit 'subjectraising', V's like sembler 'to seem', commencer 'to begin', risquer 'to be likely', etc. Such embedding is entirely possible, but [18] claims that the dummy which appears raised will not be able to self-erase. This is apparently correct:
< previous page
page_120 If you like this book, buy it!
next page >
< previous page
page_121
next page > Page 121
The incompatibility of self-erasure with these raising cases follows from [18], for the reason given below. Rule [18] demands that the self-erasing 1 arc be IP-arc-passive. This means that it must be the local successor of an unstable prepassive arc, and thus only the final 1 arc of the IP clause itself can be invisible according to [18]. But in all the bad cases of [25] and [26], the self-erased arc is not in the passive clause itself, but in a higher clause, as the result of raising. Of course, this conclusion brings out an obvious problem with the present formulation. Given that I would adopt a biclausal analysis of structures occurring with passive auxiliaries like the French être, with tense auxiliaries, etc., as it stands, [18] is in fact wrongly incompatible with self-erasure even in the standard good cases like [17b]. For these then involve self-erasures of arcs in higher clauses than the actual IP clause, namely, in the auxiliary clause, or in a tense clause higher than the auxiliary, etc. Evidently, a proper refinement of [18] will have to distinguish the sorts of raising associated with auxiliaries from that associated with Vs like sembler, risquer, etc. This is independently necesary on other grounds. For example, as is well-known, pronominal clitics associated with subordinate elements 'raise' up to passive and tense auxiliaries, but never up to Vs like sembler when these have verbal complements:
< previous page
page_121 If you like this book, buy it!
next page >
< previous page
page_122
next page > Page 122
In [28a], the clitic associated with the 3 of the participle ends up on the highest auxiliary, while in [28b] the clitic determined by the 3 of the complement V must end up on that V. Thus a proper refinement of [18] must distinguish two sorts of raising domains and permit the self-erasure of certain R-successors of IP-arc-passive arcs only when these occur with the auxiliary class of cases. I will not pursue this issue here, since it is not directly relevant to the central problem; see Postal (to appear c). The crucial property of [18] is that, aside from the problem just discussed, dummy self-erasure is only allowed for those arcs which are local successors of unstable ghost arcs. Hence the respective analyses of [8b and c] in [9] and [11] correctly predict the distinction between [17b and d], since only the structure of the former contains an IP-arc-passive arc. Thus the contrast between [17b and d] provides some initial justification for the subtle distinction drawn here between two types of 'impersonal' passives, one containing an IP-arc-passive arc, the other not. Surprisingly, this distinction manifests itself in French in determining a contrast in dummy invisibility. Under a standard and informal view in which both [17b and d] are just 'impersonal' passives, there is, however, no reason to expect that in certain environments the dummy of one but not the other could be invisible. 4.6 Reflexive Passives The argument in sections 4 and 5 for a structural distinction between French passives like [8b and c] can perhaps be supported somewhat in a way which, in addition, favors both the general goal of constructing a universal reconstruction of passive clauses and the RG/APG view that genuine REFLEXIVE PASSIVES, so far largely ignored, are a subtype of passive clause. I say 'perhaps' because the facts involved are rather marginal and involve subtle distinctions of acceptability. I think the matter is worth considering in any event because it offers an opportunity to briefly examine the treatment of the general class of reflexive passive clauses. The distinction between reflexive and nonreflexive (plain) passives, (henceforth RF-PASSIVES vs. PL-PASSIVES) is entirely orthoganal to the distinction between IPs and PPs. This yields in principle and in fact (as discussed in Perlmutter and Postal (1984a, 1984b), JP (1980: Chapter 11)) four distinct major types of passive clause, representable as [29]:
< previous page
page_122 If you like this book, buy it!
next page >
< previous page
page_123
next page > Page 123
Plain Reflexive
PP PL-PP RF-PP
IP PL-IP RF-IP
I have already given many French examples of the two PL-passive types. And, as is well-known, French also possesses the RF-PP type; this is the so-called se-moyen construction, discussed, for example, in detail in Ruwet (1972) and Zribi-Hertz (1982). Compare [30]:
Here [30b] is a RF-PP contrasting with the PL-PP in [30c]. French RF-PPs lack the past participle/auxiliary passive morphology of PL-passives, but share the se reflexive clitic morphology of 'true' reflexive clauses like [31]: 10
RF-PPs differ from PL-IPs in that no visible 811 corresponding to an earlier 1that is, no par phrase or any visible analogis possible. French RF-PPs are systematically understood to have an 'unspecified' initial 1 referencing sentient beings, corresponding to active clauses with the clitic on, on one reading of the latter.12,13 As discussed in JP (1980: Chapter 11: section 8), an account results directly from the APG approach to passives and anaphora for the recurrent use of 'reflexive' morpho-syntax in what are functionally passive clauses. Space prevents a consideration of this here. In APG terms, an RF-PP like [30b] would have a structure like [32]:
< previous page
page_123 If you like this book, buy it!
next page >
< previous page
page_124
next page > Page 124
Such representations evidently meet the conditions for being passive clauses and have arc-passive arcs as well as prepassive arcs which are not unstable, etc. The APG claim is that RF-passives are defined by the existence of COPY ARCS, like D in [32]. Such arcs are defined precisely and discussed in detail in JP (1980). Copy arcs, which are special cases of the more general fundamental APG category of REPLACERS, have two sponsors, which are a predecessor/successor pair. The replacer by law must erase one of its sponsors, which is a neighbor and has the same R-sign. Taking for granted the notion Copy Arc from JP (1980), one can formally characterize RF-passive and PL-passive clauses analogously to previous characterizations by specifying the relevant 1 and 2 arcs:
< previous page
page_124 If you like this book, buy it!
next page >
< previous page
page_125
next page > Page 125
[33c and d] are compressed definitions of two new terms each, RF-ARC-PASSIVE and RF-IP-ARC-PASSIVE, PLARC-PASSIVE and PL-IP-ARC-PASSIVE. Given these concepts, an RF-passive clause is simply one whose arcpassive arc is RF-arc-passive, equivalently, one whose prepassive arc is RF-prepassive. I have introduced the four terms in [33c-f], which are particular to passive constructions, in terms of the more general ideas defined in [33a and b] because RF-passive constructions are only a special case of a variety of copy constructions existing in NLs as alternatives of noncopy cases. Thus, ultimately, the terms in [33a and b] can be useful in describing a broader range of phenomena. 4.7 Reflexive Impersonal Passives and The Subjunctive Restriction Although discussion of RF-passives in French is often limited to RF-PPs, Gross (1975: 102) cites clauses which appear to be RF-IPs. Some French speakers accept such forms as [34c]:
< previous page
page_125 If you like this book, buy it!
next page >
< previous page
page_126
next page > Page 126
I want to argue that rather marginal French structures like [34c] are indeed RF-IPs, having a structure including the elements [35]: 14
Such a conclusion is suggested initially by the following factors. An RF-IP analysis (i) provides an account of the presence of the reflexive clitic; (ii) offers an account of the dummy final 1; (iii) yields a basis for the nonvisibility of the understood initial 1 via the same mechanisms appealed to for PL-passives of both types and for RF-PPs; (iv) permits subsuming the restriction prohibiting a par phrase under the same restriction needed for RF-PPs (see [54] below); (v) grounds in terms of the 1AEX an account of the fact that none of the predicates permitting clauses like [34c] is or a 13D predicate. Together, these considerations make an RF-IP analysis of clauses like [34c] more than plausible.
< previous page
page_126 If you like this book, buy it!
next page >
< previous page
page_127
next page > Page 127
As noted in Perlmutter and Postal (1984b) and in JP (1980: Chapter 11), the existence of RF-IPs in a variety of NLs provides a strong argument for a dummy advancement analysis of IPs in general. For only under an advancement analysis will the presence of reflexive morpho-syntax in such structures follow from the principles determining such morpho-syntax in nonpassives. Moreover, internal to French, there is the basis of an argument justifying the following: (i), in particular, the view that cases like [34c] are RF-IPs; (ii), more generally, the view that RF-IPs and PL-IPs are instances of a common category of IPs; (iii), still more generally, that there is a genuine notion 'passive clause'. This argument is not, hopefully, undermined by the marginality of the construction in [34c]. Although this matter is not discussed by Kayne and Pollock (1978), for at least some French speakers, RF-IPs seem to have the possibility of occurring with an invisible final 1 in subjunctive environments:
The situation seems to be as follows. For some reason, the embedding of RF-IPs in subjunctive contexts yields sentences like [36a]. These are strained at best, possibly even ill-formed. The same properties appear in [36b]. But, significantly, regardless of the status of [36a], the invisibility of the dummy in [36b] apparently yields no additional decay in acceptability. I conclude that embedded RF-IPs are not incompatible with the invisibility of the final dummy 1. But just this is predicted by rule [18] without alteration 15 under the view that clauses like [34c] are RF-IPs. The fact that the constraint in [18] groups together16 such morphologically different clauses as [34b and c] is an interesting argument for the reality of the notion 'passive clause' underlying this description. The possibility of invisible unstable dummies like those in [36] is especially notable since, within the class of RFpassives, the subjunctive contexts again distinguish true IPs from EXI versions of PPs (here RF-IPs), from EXI variants of RF-PPs. Thus [36] should be compared with [37]:
< previous page
page_127 If you like this book, buy it!
next page >
< previous page
page_128
next page > Page 128
The properties of [37a] and [37bi] are quite parallel to those of [36a], that is, they are at best extremely strained. But, notably, [37bii] seems to have a much worse status. Given that RF-PP cases like [37bi], like EXI versions of PL-PPs, involve stable prepassive ghosts, the ill-formedness of cases like *[37bii] follows from the failure to satisfy the IP-arcpassive condition of rule [18]. This condition is sufficiently sharp that even though the embedding of either RF-PP or RF-IPs in subjunctive contexts like [36] and [37] is strained or far-fetched, there appears to be a perceptible contrast between the dummy invisibility in [36b], which has no decay effect, and that in [37bii], which yields complete rejection, carrying the sentence with visible dummy to a structure which is completely un-French. Given the subtlety of the facts and judgements involved, the reflexive cases obviously demand much further study. But one can at the least take them as hinting at the possibility that French RF-passives obey the same constraint as PLpassives. If so, this yields an unusual justification for the RG/APG claim that both these types are instances of the single universal category of clauses having arc-passive arcs; the existence of this category is independent of whether or not arc-passive arcs sponsor copy arcs and thus define RF-passives. If the characterization of the facts offered here proves viable, the linkage of French PL-IPs and RF-IPs under rule [18] provides a valuable case internal to a single NL where there is a relevant notion of passive clause independent of notions of 'passive morphology'. Such linking is nonproblematic in the current framework, where the morphologically distinct clauses share the key property of having arc-passive arcs. 4.7 Postscript After completing the preceding sections of this chapter, I became aware of certain facts which are problematic independently of any matters previously discussed. It appears that my principal consultant will accept cases of dummy invisibility like (A):
The grammaticality of (Ab) is unexpected given rule [18]. The reason is that (Aa) seems to be a simple passivized clausal extraposition structure. This is
< previous page
page_128 If you like this book, buy it!
next page >
< previous page
page_129
next page > Page 129
one most plausibly analyzed so that the initial 2 arc corresponding to the complement sponsors a stable 2 arc ghost, which is the prepassive arc for what is then a non-IP. Alternatively, (Aa) might involve a stable 1 arc ghost, with the prepassive arc being the initial complement 2 arc. Again there would be no IP-arc-passive arc and [18] would not be satisfied. Hence, apparently (Ab) is inconsistent with rule [18] and, more generally, with the claim that subjunctive context dummy invisibility is limited to cases of true IPs. Several remarks are in order with respect to integrating (A) into the discussion of previous sections. First, while my principal consultant has, over a period of many months consistently accepted (Ab), other French speakers I have asked reject it. Conceivably then, there are variants of French in which the discussion of earlier sections goes through unaltered. This would leave open just how to characterize the NL in which (Ab) is well-formed. The following is at least a possibility. Footnote 16 to Chapter 1 observed that Zribi-Hertz (1981) characterizes a type of French in which, contrary to that of my consultant, true IPs are allowed with transitive structures containing invisible unspecified 2s. Suppose, analogously, that some speakers permit true IPs with transitive structures in which the 2 is a complement, and that (A) is an instance of this structure. This would mean that example (Aa) has a dual analysis as both an extraposition structure and an IP, although (Ab) is exclusively an IP. Such a treatment would keep (Ab) consistent with rule [18], but only by locating the irregularity elsewhere in the grammar of the relevant form of French. While I do not know whether this account of (Ab) is viable, the account of unspecified transitive IPs in Zribi-Hertz (1981) at least suggests that it is not altogether far-fetched. For the variation I am proposing with respect to complements is entirely analogous to that represented by the difference in IP distribution between the French of my consultant and that of Zribi-Hertz (1981). Third, independent of the suggestion just made, there is a fact which is extremely relevant to the earlier claim that the phenomenon of subjunctive environment dummy invisibility supports the RG/APG grouping together of RF-passives and PL-passives in a single category. This is that (A), containing a PL-passive, is exactly matched for my consultant by the acceptability of(B), containing an RF-passive:
Consequently, while facts like (A) and (B) in one way raise a real problem for the conclusions of the earlier discussion, in another they support certain aspects of the argument. Namely, they favor the claim that morphologically
< previous page
page_129 If you like this book, buy it!
next page >
< previous page
page_130
next page > Page 130
distinct plain and reflexive clauses are instances of a single universal clause typepassives. 4.8 The Ghost Arc Law and a Fundamental Property of Impersonal Passives The framework of passive description developed in APG terms determines without NL-particular specification one obvious property of both PL-IPs and RF-IPs in all NLs known to have them. The dummy nominal characteristic of them, whether visible or not, never has an analog in the closest active corresponding to any given IP. This means inter alia that there are contrasts like the following in French:
The question is why the dummy characteristic of IPs cannot appear in actives. What 'forces' passivization in such cases? This need has in effect been taken as an objection to any analysis of IPs which posits 21A based on the existence of a dummy. But such a criterion loses its force if, as is the case, there is a law characterizing all ghost arcs which predicts the asymmetrical distribution of dummies corresponding to unstable ghost arc heads. In APG terms, the law is [39] from JP (1980: Chapter 10):
This principle requires that every ghost arc, A, have an R-successor which is a neighbor of, and which has the same Rsign as, A's sponsor. The Ghost Arc Law is trivially satisfied by all stable ghost arcs, whose sponsors are by definition facsimiles. But the Ghost Arc Law is not necessarily satisfied in the case of unstable ghost arcs. It then requires that each of these have some successor, and an
< previous page
page_130 If you like this book, buy it!
next page >
< previous page
page_131
next page > Page 131
R-successor which is a facsimile of the sponsor. For IPs, which involve the ghost sponsor pattern in [4c], [39] is only satisfied if the 2 arc has a 1 arc R-successor. In effect, this means it is only satisfied if the 2 arc has a 1 arc local successor. Thus representations like [4c] all alone, which would be the structure of *[38c], violate the Ghost Arc Law, not only in French but in any NL having IPs. Discussion of the role of the Ghost Arc Law in explaining the universal lack of dummies in active corresponding to IPs raises the question of a parallel absence in actives corresponding to what have here been taken to be PPs, like French [8b]. Compare [40a and b]:
If, as has been claimed, the ghost arc underlying il in [40a] is stable, the Ghost Arc Law provides no basis for the impossibility of such ghosts in actives. There is no reason, however, to think that this raises any real problem, and still less reason to think that it casts any doubt on a stable ghost analysis of [40a]. Instead, the ill-formedness of the dummy-manifesting version of [40b] could be due to exactly the sort of constraint (discussed in Chapter 3, section 2) which renders most English extraposition dummies invisible if they head stable 2 arcs lacking 1 arc local successors. A parallel analysis is even more highly motivated in French, since there are no cases whatever of visible French dummy 2s. Yet there exist well-known contexts where these would be expected because they are contexts in which complement 1 s head main clause foreign successor 2 arcs:
< previous page
page_131 If you like this book, buy it!
next page >
< previous page
page_132
next page > Page 132
Cases like [41b and d] suggest, quite independently of those like [40], that French has the rule [42]: 18
Since arcs with local successors are never final (i.e., members, of the last level of their tail node), [42] in effect forces self-erasure by dummy 2 arcs lacking l arc local successors. Thus the stable ghost arcs that are taken to underlie passives like [40a] cannot occur in actives. Rule [42], motivated independently, properly blocks the starred version of [40b]. This analysis would mean that active examples like [40b] without the impossible visible dummy would have two distinct S-graphs, one in which certaines bouteilles heads a 2 arc, one in which it heads an 8 arc. But this result, analogous to that suggested for complement clauses with English Vs like believe in Chapter 3, section 2, has no known deleterious consequences.19 Since the antecedent of Rule [42] would not hold in [41e and f], which contain no ghost arcs, hence no dummy nominals, they lack the possibility of invisible final 2s. 4.9 Rules Involving Passive Clauses The concepts introduced in previous subsections provide simple and uniform ways of incorporating into specific grammars various recurrent major restrictions on, and properties of, passive clauses in NLs. This provides, I suggest, support of the third type mentioned in the preface for constructing a universal account of passive clauses. For instance, French rule [18] is perhaps only a somewhat more elaborate and restricted version of a rule type manifested by many NLs having IPs with invisible superficial Is. For NLs like Welsh, where all IPs have such 1s (see Awbery(1976)), the rule might simply be [43]:
EGRESSOR and ENTRANT are logically parallel concepts from JP (1980). The former refers to arcs with no successors, the latter to arcs with no predecessors. Rule [43] forces all 1-arc R-successors of IP-arc-passive arcs to selferase unless they have successors. The latter proviso is designed inter alia to allow the common situation where dummy 1s of IPs can, as it is commonly put informally, 'raise', in which case it is the resultant 1 arc R-successor in the higher clause which would self-erase, unless that in turn
< previous page
page_132 If you like this book, buy it!
next page >
< previous page
page_133
next page > Page 133
has a successor, etc. This is just the sort of interaction between raising and self-erasure of dummy arcs seen to be precluded by French rule [18], except in auxiliary cases. Common major constraints relating to passive clauses include [44]:
The constraint in [44a] can be represented equivalently as either [45a or b]:
For NLs like French and English, which lack constraints of the generality of [44a], there can nonetheless be analogs of [45a] for particular items which resist occurrence in passive clauses, e.g., have, which almost completely precludes passivization. 20 Such restrictions are also easily stated:21,22
The constraints in [44b,c] are represented as respectively [47a and b]:
Occam's razor would permit a grammar to contain one of these rules only if it lacks the more general [45a], which would render the antecedents of either [47a and b] uniformly false, thus making [47a and b] vacuously true in all cases. I take this account to solve certain problems which have not been extensively discussed before and which even lack obvious ad hoc solutions in other frameworks, still less a uniform cross-linguistic treatment. For instance, Williams (1981: 95) presents an approach to passivization quite foreign to the current framework, citing the German IP in [48a] and its English translation in [48b]:
< previous page
page_133 If you like this book, buy it!
next page >
< previous page
page_134
next page > Page 134
He then remarks: The [30a] case shows that English and German are decidedly different: the English translation is ungrammatical, and we have no explanation for this difference. In current terms, the problem faced by Williams is precisely dealt with. 23 The grammar of an NL like English, which has PPs, but is assumed to lack IPs,24 would contain [47a]. This blocks cases like *[3.2] above, while allowing e.g. [3.5a], since these turn out to be PPs under the current analysis. German grammar would contain neither [47a and b], and thus German permits both IPs like [48a] and PPs. The concepts introduced also offer a simple, cross-linguistically uniform way of specifying the recurrent restriction [49]:
This is representable as [50]:
On the other hand, for NLs like German or the French of Zribi-Hertz (1981), which do allow IPs of unspecified transitive structures but not those with visibly expressed earlier 2s, the relevant rule would presumably be along the lines of [51]:
Given the laws in JP (1980: Chapter 9), such erasure is possible only for arcs headed by unspecified nominals;25 hence [51] precludes IPs with earlier 2 arcs headed by full nominals, as desired. The laws just mentioned also play a role in the functioning of the rule which accounts for another extremely common recurrent restriction on passives:
Thus, in some NLs, e.g., Arabic (see (Saad (1975)), all passives are analogous to English [53a], those parallel to [53b] being impossible:
This restriction is statable in current terms as [54a] or as the even simpler [54c] if the concept defined in [54b] is appealed to:
< previous page
page_134 If you like this book, buy it!
next page >
< previous page
page_135
next page > Page 135
The laws mentioned earlier guarantee that only unspecified nominals can head such overrun arcs. 26 Rules like [45a], [47a and b], etc., are schematic in a way not pointed out, since they ignore the PL-passive/RF-passive distinction. For instance, an NL could be like English in allowing PL-PPs but not RF-PPs,27 or like Lardil (see Klokeid (1976)) in allowing RF-PPs but not PL-PPs. Similarly, an NL can be like possibly Maasai and allow PL-IPs but not RF-IPs. These more articulated restrictions can be imposed by refining earlier rules which use concepts RF-(IP)-arcpassive. For instance, an NL allowing no IPs and only PL-PPs would have the rule [55]:
An NL allowing no PPs and only RF-IPs would have the rule [56]:
An NL like French allowing either 'short' or 'long' PL-passives but only 'short' RF-passives,28,29 would have a rule like [57]:
Note that [57] generalizes over both RF-PPs and RF-IPs, which is correct, as shown by the starred version of [34c] above. Formulation of other refinements of passive-limiting rules to take account of the RF/PL dimension is evidently straightforward. The uniformity and elegance with which the various cited constraints can be stated in the current framework provides an argument of the third sort which I indicated in the preface as potentially supporting the validity of the goal of constructing a universal approach to all passive clauses in all NLs. Moreover, of course, these desirable properties furnish specific support to the substantive ideas of APG, which permit realization of this goal in the ways indicated.
< previous page
page_135 If you like this book, buy it!
next page >
< previous page
page_136
next page > Page 136
Notes: Chapter 4 1. One can define a further concept that picks out the local successor of the 1 arc overrun in a passive clause:
This can, I believe, be usefully employed for contrasts like that in French:
That is, some Vs require that the earlier I of a passive be marked with de, not par. I believe that this is a contrast in grammatical relations and that while the earlier I in (iia) has an 8 arc local successor, that in (iia) has a local successor representing the relation Quasi-object posited in Postal (to appear a, to appear c), whose R-sign is 6. This is the relation associated independently of passives with the pronominal clitic en. This accounts for why the earlier 1 of a passive like (iib) but not (iia) can determine such a clitic:
Under these assumptions, one can use the concept defined in (i) to state the rule relevant for the contrast in (ii) along the lines of (iv):
2. JP (1980) recognizes the possibility of foreign sponsors for certain ghost arcs. This is not relevant here. 3. Technically, I have not really formally defined 'passive clause'. This could be done by associating the clause with the nonterminal node defining it. Such a node is now passive if and only if it is the tail of an arc-passive arc. 4. Irrelevant to the present argument, the ill-formedness of (10eii) is unaccounted for by anything said so far. This is actually the expected form given regular paradigms like:
The fact is that falloir is some kind of exception to the regular requirement that the 'expected' (ia) show up as (ib). For discussion of these matters, see Postal (to appear c). 5. This analysis is partially incompatible with that of Postal (1982), which took the relevant construction to involve 1 arc ghosts. However, the view that the EXI construction involves exclusively 2 arc ghosts is consistent with the view developed and defended in RG terms in Olie (1983); this study claims that the 'extraposed' nominal in an EXI construction is a 2 demoted to 8 in all cases.
< previous page
page_136 If you like this book, buy it!
next page >
< previous page
page_137
next page > Page 137
6. Although it is not relevant here, I would assume arc G involves the R-sign 5, the name of a traditionally unrecognized grammatical relation I call Semi-object; see Postal (to appear c). This relation, arguably basic to the description of much of French clause structure, is associated with the pronominal clitic y, which is the manifestation of pronominal 5s:
7. Example *[16b] must not be confused with the grammatical (i):
The two are homophonous; see Kayne and Pollock (1978). 8. Exactly parallel to the remark in the preceding note, *[17d] should not be confused with the grammatical (i):
Note that in this sort of structure, the V is plural in agreement with certaines bouteilles. But this agreement is only visible when the relevant nominal is plural. 9. Rule [18] of course says nothing about the self-erasure of non-dummy I arcs, even pronominal I arcs, which often self-erase in other NLs. This is in general not permitted in French, a fact which must be expressed in other rules. 10. Actually, examples like [31] are often ambiguous, having both true reflexive/reciprocal meanings but also RF-PP interpretations. Thus [31a] can also mean 'One understands that man'. 11. See note 11 of Chapter 1, 12. This is the reading of on where it refers to an unspecified mind-possessing entity or group of entities. 13. Ruwet (1972: 87-125) provides what are in effect several arguments for the existence of an invisible initial 1 in these clauses. See in particular his discussion of the interpretation of adverbs like à regret. The conditions governing this differ from those relevant for e.g. après avoir bu, etc. 14. I ignore the structure of the abstract locative en cet institut. 15. This remark holds, of course, only for speakers who judge [36b] like, e.g., [17b]. For speakers who accept [36a] but reject [36b], further specification would be required. 16. In particular, the grouping in question argues against any attempt to base a reconstruction of the notive Passive on some particular morphological manifestation. 17. More precisely, [38c] contains an accusative clitic corresponding to a dummy 2. 18. Of course, this rule alone would block the clitic in cases like [38c] since the dummy arc would have to self-erase, precluding it from sponsoring a clitic arc if, as I assume, clitic arcs in French regularly erase their sponsors. For no arc can have more than one eraser. 19. Moreover, the result would be avoided if, as is not implausible, French requires that all 2 arc ghosts have I arc local successors, e.g., if the grammar entails (i):
Under these conditions, the ghost arc in [40b] would never get a chance to self-erase, always having a local successor and being erased by a distinct arc.
< previous page
page_137 If you like this book, buy it!
next page >
< previous page
page_138
next page > Page 138
20. 'Almost' because there are a few exceptions, e.g., the one linked to the meaning 'harmfully deceived': Rule [46] artificially ignores these cases. 21. Bach (1980) suggests that the constraint blocking passives with have be attributed to the fact that it does not, in his terms, occur with a transitive verb phrase. Any such account fails to generalize to French, where the analogous avoir manifests the same sort of passive limitations. The problem for Bach's account is that in other ways avoir behaves like a transitive V, e.g., in taking accusative clitics, in its behaviour as a clause union construction complement, etc. This strongly indicates that French has an analog of [46] which does not interfere with these other transitive properties. Moreover, such an analog correctly predicts the failure: of RPs with avoir as well as that of regular PPs:
22. Rule [46] illustrates one class of constraints on passivization, namely, where it is blocked for particular Vs. Consider the possibility of the 'opposite' case, where it is obligatory for certain Vs. Examples seem rarer, but one such is English want on the meaning 'seek to arrest':
Another is English rumor:
That the latter does not involve an adjective is shown by such contrasts as: since this environment allows adjectives but not passive participles. To state such constraints, it is useful to introduce the notion Secure.
Recall from [2.26a] that a free arc is one which is not replaced. Given this, constraints requiring passivization in cases like (i) and (ii) can be stated along the lines of (v):
23. Williams' use of 'explanation' seems to me without basis. He gives no reason to thinkand I believe there is nonethat the fact that German has clausal IPs and English does not has a grammatical explanation, although, of course, it requires a description. 24. See note 1 of Chapter 3. 25. The point is that examples like German [1.10c] and [1.9c] in some forms of French are only understandable as having an unspecified initial 2; the laws in question were :intended to make this and similar restrictions an invariant of NLs.
< previous page
page_138 If you like this book, buy it!
next page >
< previous page
page_139
next page > Page 139
26. Such laws may encounter problems with facts like those cited for Algonkian by Jolley (1982: 20-21). The difficulty is that certain clauses seem likely to be analyzable as 'short' passives, that is, with no visible nominal representing the initial 1, and yet they are interpreted not as having an 'unspecified' initial 1 but rather one referring explicitly to 'higher powers'. See also note 20 of Chapter 6. 27. This ignores the possibility that so-called 'middle' sentences like (i) might be RF-PPs with an invisible copy pronoun: 28. The French of an earlier period also allowed 'long' RF-PPs; see Ruwet (1972: 110). Occasionally, grammatical works cite certain 'long' RF-PPs as well-formed in modem French, so perhaps this is marginally possible with some Vs for some speakers; see Roberts (1980: 79). 29. Eckman (1974) argues that there is a regularity yielding an asymmetrical distribution of 'short' and 'long' passives. The claim is that every NL having 'long' passives has 'short' ones but not conversely. Eckman tries to explain this situation via an unusual 'raising' account of passives, which does not seem interpretable in the current framework. As noted in Keenan (1980: 212), it is unclear that the antecedent of Eckman's conditional is true, since the Austronesian NL Achenese seems to have 'long' but not 'short' passives; see Lawlet (1977). Moreover, Ostman (1981) indicates the existence of a (little used) Finnish PP construction in which the representative of the earlier 1 is obligatory. Hence it is doubtful that there is any true property of NLs to be explained here, although a clearly dominant tendency characterizes attested NLs.
< previous page
page_139 If you like this book, buy it!
next page >
< previous page
page_140
next page > Page 140
Chapter 5 Apparent Impersonal Passives of Unaccusative Structures and Possible Violations of The 1AEX 5.1 Remarks As indicated in Chapter 2, section 2.2, one consequence of the 1AEX of Perlmutter and Postal (1984a) (as represented in [2.2a]) in combination with the current view of passives is that no passive clause of any type in any NL can be based on an unaccusative predicate. If the class of unaccusative predicates is universally determined by their meanings, as specifically considered in Perlmutter (1978a) and as follows from what is called the Principle of Initial Determination (PID) in Postal (1982) and from the Universal Alignment Hypothesis in Perlmutter and Postal (1984a), Rosen (1981b), then for every NL, the class of semantic predicates found in IPs must be disjoint from the class of unaccusative predicates. A certain amount of potential counterevidence to this view has recently been brought forward. This might be taken to indicate that, for instance, the class of unaccusative predicates cannot be universally specified. This conclusion is explored and defended at length in Rosen (1981b) and is also accepted in Perlmutter (1982). Or, it might be taken to indicate that other putatively universal relational principles like the 1AEX cannot be maintained as grammatical laws. While such theoretically unhappy results might, of course, have to be accepted, it will be argued by dealing with several cases, that the justification required for such negative conclusions goes considerably beyond that
< previous page
page_140 If you like this book, buy it!
next page >
< previous page
page_141
next page > Page 141
typically advanced. What follows is by no means a defense of the Universal Alignment Hypothesis with respect to the assignment of intransitive predicates to the categories unaccusative and unergative. I do not know whether such a defense is possible, although I am not convinced of its impossibility. But it certainly is a defense of the 1AEX. Moreover, I take the following discussion to suggest minimally that even the issues relevant to the Universal Alignment Hypothesis need to be more deeply studied before the sort of negative conclusions now sometimes being drawn are accepted as inevitable. 5.2. Turkish Perlmutter (1978a) described the unaccusative hypothesis in some detail and developed its consequences for IPs given the 1AEX. He illustrated the claims with data from Dutch and Turkish. But Özkaragöz (1980), in a work designed to support the general outlines of the unaccusative hypothesis with data from Turkish, observes apparent counterexamples to the claims that have been made. The counterexamples are what appear to be Turkish IPs based on predicates whose semantics should make them universal unaccusatives, e.g. [1]:
Such predicates do, as expected, preclude IPs in other IP-possessing NLs. Apparently then, the lawful consequence of the universal account of passives, the 1AEX and a universal approach to unaccusative assignment can not be jointly maintained, just because of the relevant Turkish impersonal clauses. Before accepting this conclusion, however, one should note that Özkara-göz develops certain Turkish-internal criteria for unaccusativity independent of IPs. And, by these criteria, the predicates in [1] and similar clauses behave as unaccusatives. Thus there is no solution to the problem of [1] based on claiming that Turkish, exceptionally, treats these predicates as unergatives. I take this to suggest that, even from the parochial viewpoint of Turkish grammar itself, it is worthwhile to try and maintain all of the relevant universal claims about NLs even in the face of examples like [1]. There is only one apparent way in which this could be done: to deny that the cited examples are IPs. Let us explore what this means. According to the unaccusative hypothesis, the unspecified nominal in examples [la and b] call it ON for convenienceis an initial 2. For [la and 1b] to be counterinstance IPs, this unspecified 2 must also head a local successor 1 arc, which is
< previous page
page_141 If you like this book, buy it!
next page >
< previous page
page_142
next page > Page 142
overrun by the IP-arc-passive arc which is the local successor of the prepassive unstable ghost defining an IP. In other words, to be true counterexamples, examples like (la and b) must involve at least the structure [2]:
But what grounds motivate a structure like [2]? In particular, what motivates the key feature of [2] as a counterexemplification of universal unaccusativitythe claim that ON heads a 1 arc at one level in examples like [1]? The key motivation for this assumption is evidently the occurrence of the passive marker on the Vs of these apparent IPs. The justification for calling this affix 'passive' is, of course, that it is the suffix which marks the Vs of uncontroversial PPs in Turkish, and also the Vs of those apparent Turkish IPs which are not problematic because they are based on unergatives. However, while this is real evidence for the unwelcome conclusion, it is quite weak. For just like accusative case marking, mentioned earlier to be an imperfect guide to superficial grammatical relations, voice markers, etc., passive markers are an equally imperfect guide to clausal grammatical structure. Nonetheless, no other evidence is brought forward to support the view that the sentences in [1] are truly IPs. This is especially important since there are evidently alternative analyses of cases like [1] which are consistent with all the universal constraints I wish to maintain. One such :analysis would agree with the IP analysis in taking Turkish examples like [1] to involve a dummy 2 which advances to 1. But it would differ in that the initial 2 would never advance to 1. That is, [la and b] would involve unaccusative advancement to 1 and thus the relevant clauses would manifest only a single advancement to 1, consistent with the 1AEX. For extensive discussion of such structures in RG terms, see Perlmutter (1983b). A conceivable APG structure for these examples would then include [3]: 1
< previous page
page_142 If you like this book, buy it!
next page >
< previous page
page_143
next page > Page 143
This structure is not passive and since its only 1 arc local successor, A, overruns no arc, is not arc-passive, or even potentially IP-arc-passive. Further, the ghost arc B is stable. Consequently, the advancement to 1 found in [1], and presumably required, is not forced by the Ghost Arc Law. This is no problem because some advancement to 1 is forced by the Final 1 Arc Law, which requires there to be a 1 arc at the final level of every basic clause. 2 Evidence which could show that [3] is overwhelmingly inferior to an IP analysis like [2] for the relevant Turkish cases has not been developed. Such evidence could take the form of, e.g., 'control' facts like those used in connection with French EXI passives such as [4.8b] to argue against the view that pivot nominals were 1 s at any level. Or, it could take the form of some kind of marking facts relevant to the initial 2. If, for example, this was marked with a case or preposition uniquely characterizing passive 8s, an IP analysis would be supported. But for putative Turkish IPs like [1], no such evidence has been brought forward.3 While I do not, of course, claim to have shown the theoretically preferable analysis in [3] to be internally superior to an IP analysis for Turkish, it has been shown that evidence favoring the latter is at best extremely slim, too slim to justify rejecting the universal claims which the cases might counterexmplify. 5.2 Postscript Many months after completing section 5.2, I became aware of Özkaragöz (1982a), a study of certain Turkish 'double passives'. These clauses are so called because they contain two instances of the passive suffix. This material
< previous page
page_143 If you like this book, buy it!
next page >
< previous page
page_144
next page > Page 144
suggests that the analysis of the apparent IPs of unaccusatives proposed via [3] is not correct. Examples of 'double passives' include (A):
The property illustrated by (Aa and b), that both the understood 1 and 2 must be the generic unspecified nominal, is systematic. Such sentences are apparently only possible with transitive V-stems and in the aorist tense. It would be possible to argue, and is indeed argued by Özkaragöz (1982a), that 'double passives' like (A) are IPs of PPs. If so, such structures are directly inconsistent with the 1AEX, and the relevant sentences apparently cause even greater problems for the conceptual framework of this study than the apparent IPs of unaccusative predicates. However, it is possible to propose an analysis which (i) serves as an alternative to the earlier analysis (as in [3]) of the apparent IPs of Turkish unaccusative predicates, (ii) provides a proper structure for the 'double passives', (iii) is consistent with all cited Turkish-internal generalizations about relevant sentences and (iv) is consistent both with the 1AEX and even with the universal assignment of unaccusative status. The basic idea of the structure in [3] was that while the relevant clauses involved a dummy 2, as in true IPs, and a dummy 2 which advances to 1, the clauses were not passives, hence not IPs, because there was no overrun 1 arc. While this approach might be applied to the 'double passives' in (A), I will here develop an alternative approach to all the sentences. Therefore, let us propose that the relevant structure of Turkish [1] is not [3] but (B):
< previous page
page_144 If you like this book, buy it!
next page >
< previous page
page_145
next page > Page 145
(B) differs from [3] in that the stable ghost arc, A, is a 1 arc, not a 2 arc, and the nominal which advances from 2 to 1 is the initial 2, the unspecified mind-possessing nominal represented as ON. Although impersonal, like [1], the structure is not passive, and no universal constraints are violated since there is only a single advancement to 1. Among the issues left open is why such clauses, if not passives, have the 'passive' verbal marker, just as PPs do and just as those apparent IPs based on unergative predicates do. I return to this below. Postulation of structures like (B) amounts to the claim that Turkish has a construction in which a 1 arc whose head is ON can determine a neighboring 1 arc ghost, which erases both its sponsor and itself. As remarked in note 1, for the case of [1], the analysis in (B) also requires elimination of PN Law 85 of JP (1980). Somehow, the ghost arc in (B) must be associated with the 'passive' verbal marker. One can then utilize the same notions to provide an analysis for the 'double passives'. This takes them to be PPs in which the initial 2 arc is also headed by ON and in which the arc-passive local successor also sponsors a ghost 1 arc which erases the arc-passive arc. Hence, parallel to (B), the structure of (A) would be (C):
Structure (C) is compatible with the 1AEX, since there is only one advancement to 1. Further, (C) is consistent with the basic arguments of Özkaragöz (1982a), since these explicitly show only that the final 1 of 'double passives' is a dummy. This is the case in (C) since B is a ghost arc, so its head is by definition a dummy; see JP (1980: Chapter 10).
< previous page
page_145 If you like this book, buy it!
next page >
< previous page
page_146
next page > Page 146
Key issues then are how these analyses deal with the passive morphology and how the distribution of the postulated dummy 1 construction can be controlled. The two issues are related, as shown below. Suppose it is claimed that the apparent IPs of Turkish unergatives are, as previously assumed, true IPs. Then the stable 1 arc ghosts postulated in diagrams (B) and (C) apparently occur only in clauses containing a 1 arc local successor of a 2 arc whose head is ON. There might then be a Turkish rule like: (D):
Here the inscription 'ON(A)' is just schematic. However, suppose the apparent IPs of unergatives are not true IPs but just instances of the same stable 1 arc ghost construction illustrated in (B) for unaccusative predicates. Then the second conjunct of (D) is too strong, as there will be clauses with the structure in (E):
However, one cannot just eliminate the final consequent conjunct of (D). For something must prevent the dummy structure in the case of unpassivized transitives, since there are apparently no sentences analyzable as in (F):
< previous page
page_146 If you like this book, buy it!
next page >
< previous page
page_147
next page > Page 147
The proper weakening of (D) might well then be to require the ghost sponsor to be an ABSOLUTIVE 1 arc (see JP (1980: Chapter 7)), that is, one not sharing its final coordinate with a neighboring 2 arc, yielding (G):
This allows such ghosts to be sponsored by initial unergative 1 arcs, by 1 arc local successors of prepassive 2 arcs, etc., but not by the initial 1 arc of an unpassivized transitive clause, blocking cases like (F). No matter which assumption about the apparent IPs of unergatives is made, there is, therefore, an effective generalization controlling the posited ghost structure. There is at least one reason to suspect that (G) is correct and thus that Turkish has no true IPs at all. Although in PPs, the initial 1 can in some cases show up in a postpositional phrase in tarafindan, this is never possible with the apparent IPs of intransitives regardless of whether these are based on unergative or on unaccusative predicates; see Kornfilt (1976: 438). The single central nominal argument of such clauses is always interpreted as ON. If one takes the apparent IPs of unergatives also to be instances of the stable 1 arc ghost structure rather than true IPs, then this fact can be built into the rule sketched in (G), essentially by extending it to (H):
< previous page
page_147 If you like this book, buy it!
next page >
< previous page
page_148
next page > Page 148
Given the laws in JP (1980: Chapter 9), the conjunct 'Erase(B, A)' of (H) determines that As head must be an unspecified nominal. Crucially, the extension of (G) to (H) is required no matter how apparent IPs of unergatives are analyzed, since in no case does the posited. ghost structure determine a visible 8 realization for the earlier 1. This is impossible in both the 'double passives' and the apparent IPs of unaccusatives. However, if the apparent IPs of unergatives are true IPs, then they are not characterized by rule (H) (that is, they do not satisfy its antecedent) and the impossibility of a visible 8 marked with tarafindan in the impersonal unergative clauses would require independent specification. I conclude that (H) is the preferred statement on the basis of current information and thus that Turkish has, in the technical sense developed in JP (1980) and in this study, no true IPs. That is, none of the Turkish impersonal clauses discussed instantiates the ghost sponsor pattern in [4c] of Chapter 4. This conclusion raises the following issue, properly stressed to me by I. Özkaragöz (personal communication 10/18/83). It is well-known (see Aissen (1974a, 1974b); Zimmer (1976); Aissen and Hankamer (1980); Gibson and Özkaragöz (1981)) that Turkish PPs can in general not be embedded as the complements in the Turkish causative (clause union) construction. But neither can the apparent IPs of intransitives, which it has now been claimed are not true IPs, moreover, not passives of any kind. Doesn't this lead to a loss of generalization? The answer is not necessarily. For, although they are not passives, structures like (B) and (E) share certain structural properties with PPs. In particular, they share the property that their first (free) 1 arc is not the final 1 arc. Hence in (B), the first free 1 arc is B, but the final 1 arc is A. But of course in any PP the first free 1 arc will be either the initial 1 arc or some anaphoric replacement of it, while the final 1 arc will be related to the arc-passive arc defining the clause as a PP. There is thus no possibility of these two arc categories coinciding in the same arc. The constraint on Turkish causative complements could then be simply something like (I):
This is no more complicated than a statement precluding passives in such complements; so, unless (I) can be overthrown, the causative,' constraint does not argue against the current analysis of apparent IPs. Note 22 of Chapter 4 introduced the notions SECURE, designating free arcs (those lacking replacers) which are not overrun. Consider the complement of this class among the class of free arcs, the class of INSECURE arcs, defined as in (J)
< previous page
page_148 If you like this book, buy it!
next page >
< previous page
page_149
next page > Page 149
Given this, it might be proper to reformulate (I) even more simply as (K):
Previous discussion has left open the question of the passive morphology. Under an analysis like that of Özkaragöz (1982a), the generalization would appear to be that the V of a clause K contains one instance of the passive suffix for each arc-passive arc in K, hence two in cases like (A), analyzed by Özkaragöz to involve both personal and impersonal passivization, violating the 1AEX. But in current terms, making use of the same property appealed to for the causative complement constraint, one can say that the V of a Turkish clause K contains one instance of the passive suffix for each insecure 1 arc in K. The current suggestion is then that Turkish passive morphology is actually a morphology of clauses with insecure 1 arcs. The competing proposals about Turkish passive marking reduce to (L):
(Lb) is no less general than (La). It applies properly to sentences like (A) analyzed as in (C) to determine two passive suffixes, but is, nonetheless, entirely consistent with the 1AEX. Of course, this account of the passive suffix is informal. It has not been indicated how to represent such elements formally in APG terms, what kind of arcs they head, what sponsors them, etc. But this is unnecessary here. For there is no more problem in formalizing (Lb) than (La), which appeals to multiple passivation and to violations of the 1AEX. Any formal development capable of sustaining L(a) would likewise sustain L(b). Consequently, nothing about the facts of Turkish considered here or in the cited works have been shown to undermine either the 1AEX or the universal principles assigning unaccusative status. The account of passive morphology in (Lb) fails to predict the passive marker on the embedded V in the 'infinitival double passives' studied in George and Kornfilt (1977). These are sentences of roughly the form The authors were wanted to be applauded by the audience with the meaning 'The audience wants to applaud the authors'. Although both the main and subordinate Vs in such structures have the passive marker, George and Kornfilt argue convincingly, in transformational terms, that only the main clause has undergone the passive rule. In the current framework, this means that only the main clause is passive and that the subordinate clause contains no prepassive arc. I would assume, essentially in keeping with the views of George and Kornfilt, that these clauses involve a variety of clause union, permitting the subordinate 2 to be a main clause 2, and thus providing the prepassive arc required for main clause passivization of an original subordi-
< previous page
page_149 If you like this book, buy it!
next page >
< previous page
page_150
next page > Page 150
nate clause constituent. Clause union means, of course, that the subordinate clause P arc heads a U arc foreign successor :neighbor of the main clause P arc. Some principle must then extend Turkish passive morphology to the V which corresponds to the head of such a U arc. It does not seem difficult to construct such an extension. But exactly how to do this remains partially obscure because I do not know how the 'infinitival double passives' discussed by George and Kornfilt (1977) interact with the 'double passives' discussed by Özkaragöz (1982b). Specifically, the question arises whether there are impersonal variants of 'infinitival double passives' in which the main clause V has two instances of PASS. This question aside, the passive marking of the originally subordinate V in an 'infinitival double passive' can be accounted for by modifying informal principle (Lb) to take account of the presumed structure (Mb) for an 'infinitival double passive' like (Ma). Note that (Mb) ignores the subordinate clause initial 1, which is erased as a function of the 'equi' structure:
The key point about any clause union analysis like (Mb) is that the main clause has, in one sense, two Vs, although these head distinct types of arcs
< previous page
page_150 If you like this book, buy it!
next page >
< previous page
page_151
next page > Page 151
(P vs. U). Suppose though that these are subsumed under a category VERBAL ARC, via (N):
Then it would be rather straightforward to generalize and modify principle (Lb) to something like (O):
Because of its reference to verbal arcs, voice marking principle (O) requires each of arcs H and I in (Mb) to have a branch ending up as the PASS marker, because each is a neighbor of the insecure 1 arc A. While (O) remains far from fully precise, it should indicate how one could, without great difficulty, integrate the phenomenon discussed by George and Kornfilt into the framework developed here. This phenomenon appears to offer no support for (La) against (Lb). Consequently, the facts discussed by George and Kornfilt offer no support for analyses of Turkish inconsistent with the 1AEX or with any other current relational assumptions. As stressed to me by I. Özkaragöz (personal communication: 12/12/83), there is a further problem for an analysis of Turkish impersonal clauses like that sketched in this postscript. This involves clauses in which, apparently, indefinite ls can be incorporated in the V. Hence alongside the ordinary SOV clause in (Pa), there is the structure in (Pb):
According to Özkaragöz, with respect to a variety of phenomena including word order, scrambling, relativization and raising, the realization of the initial 1 in (Pb), contrary to that in (Pa), does not behave like a 1. Following a suggestion of Hankamer and Knecht (1976), Knecht (1976), Özkaragöz considers the idea that this nominal has been demoted. This would mean in RG/APG terms that it corresponds to an 8 arc local successor of a I arc. And, given the Motivated Chômage Law and the Final 1 Arc Law, some later 1 arc would have to be present. This could apparently only be one headed by an invisible dummy. The problem is that this interpretation of the analysis suggested by Hankamer and Knecht assimilates the incorporation structure in (Pb) to the stable 1 arc ghost structure posited here for the impersonal clauses and
< previous page
page_151 If you like this book, buy it!
next page >
< previous page
page_152
next page > Page 152
'double passives'. Yet, crucially, the Vs of the incorporation sentences do not manifest the passive morphology. Thus the properties of the incorporation clauses might appear to threaten principle (Lb). So far though, this conclusion is unwarranted. That the data show the indefinite element in clauses like (Pb) not to behave like a 1 does not necessarily mean that this element fails to correspond to a final I arc. Rather, it can be assumed that the data show only that the relevant arc, though a final 1 arc, is not a surface 1 arc, as.a result of the existence of the structure which defines the fact that the relevant element is 'incorporated'. Such sentences inevitably raise the issue of how 'incorporation' is to be characterized, a matter about which essentially nothing has been said in either RG or APG terms, or indeed, in the terms of almost any modern framework. Consider the following APG proposal. When it is said that 'a 1 incorporates', this means that the relevant 1 arc sponsors an arc representing a certain relation, call it Incorporated (R-sign = Inc). Inc arcs have P arc heads as their tails and are neighbors of Stem arcs. For structures with 'doubling' incorporation, as in Mohawk and other Iroquoian languages (see Postal (1979)), where the incorporated noun stem can occur on the V in addition to a matching external noun, there is no erasure of the Inc arc's sponsor. But for Turkish clauses like (Pb), where 'doubling' is impossible, a rule will say that the Inc arc erases its sponsor. This yields a structure for (Pb) something like (Q):
< previous page
page_152 If you like this book, buy it!
next page >
< previous page
page_153
next page > Page 153
Under this analysis, regardless of any other properties of 1 arc B, it is provably a final stratum 1 arc, since it has a FOREIGN ERASER, D. And Theorem 44 of JP (1980: 182) says that any foreign erased arc is final. Consequently, under a structure like (Q), the absence of passive morphology is predicted by principle (Lb) or (O), for (Q) contains no insecure 1 arc. Note that in (Q), both the 1 arc and the Inc arc have as R-branches L arcs (representing the Labels relation) with the same head node. Since this is evidently a defining rather than an accidental feature of incorporation structures, grammatical laws presumably guarantee this situation. But such laws need not be considered here. The key issue is whether the analysis proposed in (Q) for incorporation can handle the facts previously taken to involve demotion of the 1. I suspect that it can. It appears that those phenomena with respect to which the incorporated element 'does not behave like a 1' can be described in terms of surface lhood or related notions. Observe that 1 arc B in (Q), although a final 1 arc, is not a surface 1 arc, since it is erased (by D). Surely, word order constraints never reference final lhood, and it seems that most, and possibly even all, word order restrictions involve surface relations; see Chapter 1, section 4.4. 4 The fact that elements like ari in (Pb) cannot 'raise' involves foreign successors, which determine the erasure of the predecessor. But arc B in (Q) already has an eraser. Hence if it had a successor, it would have two erasers, violating one of the most basic APG laws limiting each arc to a maximum of one eraser.5 Similarly, relativization involves the erasure of the 1 arc, which, in the case of (Q), would also violate the Unique Eraser Law. Other parochial facts about Turkish relativization which Hankamer and Knecht (1976) state in terms of the presence or absence of a 1 also do not seem incompatible with structures like (Q). Hankamer and Knecht (1976: 218) claim: In these examples, ..., we propose that the subject has not only moved but has been demoted from subjecthood, and that the sentences are subjectless at the time of RC formation. But their notion 'subjectless' can, as far as I can see, be taken to involve the presence of an erased 1 arc. Since B in (Q) is erased, ari in (Pb) will be treated differently from the head of the unerased 1 arc in (Pa), as is descriptively required. According to Hankamer and Knecht (1976), whenever a clause is 'subjectless', if any constituent of it is relativized, the result must be formed with the so-called 'subject participle' (SP) rather than the so-called 'object participle' (OP). In particular, by this criterion, the apparent impersonal passives of Turkish are 'subjectless'. But this is consistent with the present account, since, under the non-IP analysis of such clauses, as in (B), there is no surface 1 arc. The final 1 arc is the stable ghost, which self-erases. Overall, it seems to me that all of Hankamer and Knecht's observations about the
< previous page
page_153 If you like this book, buy it!
next page >
< previous page
page_154
next page > Page 154
choice of the OP versus the SP can be given an APG account as in (Rb) based on the generalizations they discovered, assuming that a relative clause is represented as in (Ra):
The informal notion 'relativized' arc refers to an arc like G in (Ra), that is, one erased by the arc corresponding to the relative clause head. The claim is then that the OP is used only when relativizing some subconstituent of any constituent of the relative clause which cooccurs with a surface 1 arc. The key point is that C in (Ra) must be a surface arc for the relative to have the OP. Were C erased by Bas it would be in relative clauses of the sort occurring in phrases meaning 'the man who ate the apple'the prediction is made by (Rbii) that the SP is required, which is of course correct. 6
< previous page
page_154 If you like this book, buy it!
next page >
< previous page
page_155
next page > Page 155
The account of 1 incorporation in examples like (Mb) should be related to other instances of Turkish incorporation. Knecht (1976: 318-321) discusses cases in which 'generic' 2s incorporate. Thus (Sb) exists alongside the expected (Sa):
It turns out that the non-case-marked form in (Sb) does not behave like an ordinary 2 with respect to various factors, e.g., freedom of word order. It is thus proposed that 'generic' 2s incorporate. Knecht reports J. Hankamer's observation that a contrast similar to that between (Sa and b) shows up in various ways in passives:
(T) is ambiguous, corresponding either to (Sa) and meaning 'The squash were eaten' or to (Sb) and meaning 'Squash were eaten'. But Hankamer observed that if kabak is separated from the V in (T), e.g., by an adverb, only the 'nongeneric' meaning of (Sa) is possible. Hankamer also noted other ways in which the 'generic' nominal of one reading of (T) behaves differently from a true passive 1, e.g., in not being subject to raising to 2 with Vs like 'believe'. Knecht provides further arguments that in the 'generic' variant of (T) the nominal meaning 'squash' does not behave like an ordinary 1. Hence it seems that the 'generic' form is also incorporated in passives like one reading of (T). As Knecht notes in transformational terms, this raises an issue about the interaction of passivization and incorporation. One analysis would have generic object incorporation precede passive. Personal passivization would thus be bled when incorporation actually takes place; hence the clause manifesting the 'generic' reading of (T) could not be a PP. It would then be interpretable as an impersonal clause, in which, of course, kabak is not a 1. However, Knecht (1976: 355) also observes the existence of an analysis, suggested by Hankamer, in which generic 2s can undergo passive and then are subsequently incorporated as 1s. Knecht was aware of no evidence for or against either position. Since it has already been shown necessary for the sentences advanced by Özkaragöz to recognize incorporation of ls, the natural analysis of clauses
< previous page
page_155 If you like this book, buy it!
next page >
< previous page
page_156
next page > Page 156
like (T) in current terms might seem to be to take them to be PPs with incorporated final ls. For the same reasons as the .ls discussed in Özkara-göz's examples, these incorporated ls would then fail to 'behave like ls' with respect to raising, word order, and other phenomena which lead to their being considered incorporated. That is, the assumption that the relevant constraints refer to surface lhood, internal survivor lhood, etc., would also predict these facts. For the PPs in question would have final 1 s but neither surface nor internal survivor 1s. In such terms, examples like (T) would be PPs on both readings, with the arc-passive arc erased by the Inc arc it sponsors in the 'generic' case. However, there are grounds to reject the view that the 'generic' reading of (T) is a PP (whose final 1 arc determines incorporation). The PP analysis of the 'generic' reading of (T) makes an easily tested prediction which conflicts with that made by the view that the sentence is some form of impersonal. As already noted, Turkish PPs can on occasion occur with an explicit 1 chômeur marked with the postposition tarafindan. This is never possible with the impersonal clauses. And, while the published sources consulted here for Turkish provide no information, I. Özkaragöz (personal communication: 2/3/84) and J. Kornfilt (personal communication: 2/12/84) have both kindly informed me that the 'generic' reading of (T) is lost when a tarafindan phrase is added. This argues strongly against a PP analysis of the 'generic' reading of (T) and suggests instead that this variant of (T) should be treated analogously to our treatment of the so-called impersonal passives like (la and b) and those with unergative predicates. This would yield a structure for the 'generic' reading of (T) like (U):
< previous page
page_156 If you like this book, buy it!
next page >
< previous page
page_157
next page > Page 157
This structure is evidently not passive at all, but an impersonal structure of the sort already discussed. The 'passive' morphology is motivated by principle (Lb). However, (U) raises an important problem for the analysis of Turkish impersonal structures which I have suggested and tried to defend. Note that the 2 are C in (u) has a foreign eraser, E, and is thus lawfully a final stratum arc. Therefore, the absolutive condition in principle (G), taken earlier to control the 1 arc ghost structure assumed to underlie Turkish impersonals, is not met, and (G) should forbid structures like (U). Clearly then, the absolutive condition in (G) would have to be weakened in some way so that the incorporation of a 2 leads to a clause being treated like an intransitive clause even though it has a final 2. One possibility for formulating such a weakening is to posit that an impersonal clause, that is, one containing the impersonal 1 arc ghost structure, not contain any internal survivor 2 arc. For this notion, see note 6. The 2 arc C in (u) is not an internal survivor of the clause defined by node 811 since it is erased by E, which is internal to the constituent defined by 811. Of course, all 2 arcs with distinct local erasers fail to be internal survivors, so this condition would cover the cases covered by the absolutivity condition of (G). Nonetheless, I do not know how viable this suggested revision is. If it runs into problems, this would suggest some error in the deeper assumptions made so far. In conclusion, the extremely interesting facts about Turkish incorporation clauses do not so far seem to cause insuperable problems for the approach to Turkish impersonal and double passives proposed here, even given the difficulty with (U) just discussed. They would only be troublesome under the view that the incorporated 1 has been demoted by the presence of a stable 1 arc ghost. But this assumption has not really been justified against possible alternatives, like (Q), in which the 1 arc whose head incorporates is a final 1 arc but not a surface I arc. The analysis proposed in this postscript for Turkish impersonal and double passives is, interestingly, largely identical to that proposed below in section 5.5.1 for certain Lithuanian structures which have also been claimed to raise problems for the relevant collection of claimed universal relational principles. 5.3 Extensions of Passive Morphology Section 2 suggested in effect that Turkish 'passive' morphology is extended to a limited class of nonpassives, including those involving unaccusative predicates with the initial ON-headed 2 arc sponsoring a stable ghost. It is relevant that certain facts from English and Welsh suggest that such 'extensions' of passive morphology are independently possible in NLs. The facts involve Vs meaning 'to be born', surely unaccusative by universal
< previous page
page_157 If you like this book, buy it!
next page >
< previous page
page_158
next page > Page 158
criteria. This correctly predicts, for example, that in French, there is no IP of the V naître:
Unlike its French equivalent, English born, does not occur in clauses with ordinary active morphology. But it does occur in clauses which have the look of passives:
However, I claim that [5c] is not a passive but an unaccusative clause with the baby as initial 2, advanced to 1. The V in question thus irregularly determines 'passive' morphology. 7 The denial of passive status is supported in spite of the verbal morphology since the relevant clauses do not involve the standard passive 8 marking with by. That is, a true IP of an unaccusative predicate like born would be expected to show up as the impossible: I take these facts to indicate the possibility for specific (in this case lexical) determination of passive morphology for nonpassive clauses, that is, marking of the kind needed to maintain a non-IP analysis of Turkish cases like [1] proposed in section 2. The possibility of such marking is supported by similar anomalous marking found in Welsh with the V meaning 'born'. The latter facts are even more striking because Welsh has both PPs and IPs, but the two have utterly different verbal morphologies; see Perimutter and Postal (1984b). PPs involve a form of the V cael 'get' as auxiliary, and the 'verbal noun' form of the main V, with a possessive form of pronoun as a (raising) copy of the passive 1.8 IPs involve no auxiliary but involve instead only a specific impersonal form of the V. Both types of passives share, however, the use of the preposition gan as a passive 8 marker. Consider then [7]:
< previous page
page_158 If you like this book, buy it!
next page >
< previous page
page_159
next page > Page 159
These illustrate that geni 'to be born' occurs in clauses with both types of passive morphology. And yet there is evidence supporting the claim required by the 1AEX and a universal assignment of 'born' to unaccusative status that these clauses are not passives. For in neither example does the nominal Wyn occur with the standard passive 8 marker gan. I assume that [7a] involves unaccusative advancement of the initial 2 to 1 so that the structure is parallel to that of English [5c] (ignoring the auxiliary structure of the latter). [7b] involves, however, essentially the structure in [3], except that Wyn replaces ON. Thus it would appear that Welsh has ad hoc passive morphology marking for unaccusative clauses based on geni, requiring PP morphology if the initial 2 advances to 1, IP morphology if the nominal which advances is a dummy. 9 The latter statement is essentially what would be needed for the analysis of the Turkish cases in [1] presented in section 2, since those Turkish Vs do not permit the passive marker when the initial 2 simply advances to 1. Together then, the English and Welsh facts seem to indicate that an NL can utilize passive morphology in restricted ways for nonpassives. The several instances of this sort so far examined are in a sense nonrandom in that the passive morphology is, in RG/APG terms, extended to nonpassive clauses which do involve 21A. But just this is needed to permit the analysis which defends the universal view of unaccusativity from the Turkish clauses in [1], which, it was suggested, are not IPs at all, but sentences with the same general nonpassive structure as Welsh [7b] (but see the postscript to 5.2). In other words, it could be claimed that Turkish merely has a more general version of what Welsh has for the single V geni: extension of the passive verbal marker to a generalized 21A marker. Turkish would then use the passive suffix in just those nonpassive 21A clauses where the 2 arc is a stable ghost. One might then expect to find NLs in which this situation is maximally generalized, that is, where some single morphology marks all and (ideally only) passive and unaccusative advancement clauses. Just this situation has been described for Albanian by Hubbard (1979, 1980). In one sense, the view that passive morphology can be extended to nonpassive clauses is traditional. For just this is traditionally said about Latin deponent Vs; see, e.g., Allen (1874: 22). In this case though, the extension seems to go beyond unaccusative clauses and is found with Vs whose meanings would make them universal unergatives or transitives (e.g., hortari, exhort). While I am not familiar with the details of Latin, this is a prima facie case again that NLs permit 'passive' verbal morphology in non passive clauses. 5.4 Sanskrit On the basis of Sanskrit data, Ostler (1979: Chapter v) makes in some detail the same sort of objection just considered for Turkish to the claimed
< previous page
page_159 If you like this book, buy it!
next page >
< previous page
page_160
next page > Page 160
universal incompatibility of IPs and unaccusative predicates. Marantz (1981) takes Ostler's account as showing the untenability of the relevant RG universals. Consider some background facts. Sanskrit, an NL with a rich system of superficial case marking on nominals, has uncontroversial PPs, illustrated by the following active/passive pair:
In the active, the final 1 is marked nominative, the final 2 accusative; in the PP, the final 1/earlier 2 is also nominative, while the initial 1/final 8 (I assume) is marked with the instrumental. The alternative verbal forms in [8b] illustrate three distinct passive morphologies, one involving insertion immediately after the root of the morpheme -ya-, the second involving the past participle passive marker -ta-/-na-, and the third involving an ending -i. These have different agreement and tense combination properties, which I need not consider. Sanskrit also allows, according to Ostler, PPs based on intransitive predicates like sit, where oblique elements show up as final ls. These are partially similar to English pseudo-passives. Compare [9a and b]:
So far there are no particular problems. But there also exist other sentences, based exclusively on intransitive predicates, which Ostler (1979: 367) calls 'impersonal passives'. An example is [10]:
< previous page
page_160 If you like this book, buy it!
next page >
< previous page
page_161
next page > Page 161
The key visible difference between [9b] and [10] is that the latter has no visible superficial 1 (no nominal in the nominative). Taking cases like [10] to be genuine IPs, Ostler offers evidence against a theory which maintains all of the following: (i) an advancement view of passives, (ii) the 1AEX and (iii) the universal assignment of unaccusative status. He does this by noting that Vs which would, he assumes, be universal unaccusatives, occur in the apparent IPs:
[11b] would appear to be the IP of an auxiliary V, of the sort taken by Perlmutter and Postal (1984b) to always be unaccusative. I agree with Ostler that these are apparent counterexamples in that they do involve what would be universally unaccusative predicates. However, these forms are only real counterexamples if they are genuine passives. As for Turkish, one must then examine the nature of the evidence that the apparently counterexemplifying clauses are truly passive. The grounds seem to be twofold. First, in these impersonal constructions, the nominal 'expected' to be superficial 1 does not occur as final 1 in the nominative but rather occurs marked with instrumental case, just like the earlier 1/final 8 of an uncontroversial PP. Second, the verbal morphology is of one of the sorts typical of PPs. Thus, the case seems to be stronger than that in Turkish, where the chief evidence for passivehood of the clauses involved the verbal morphology; in Sanskrit, it appears in addition that the nominals are marked just as passive 8s are marked in PPs. Despite this, the data Ostler gives do not support the negative conclusions he draws. For nothing supports an IP analysis of clauses like [11] against an analog of the analyses proposed above for Welsh [7] and Turkish [1]an analog not considered by Ostler. Suppose, that is, that one takes the structure of [11a] to include [12]:
< previous page
page_161 If you like this book, buy it!
next page >
< previous page
page_162
next page > Page 162
In this analysis, 'trees' also ends up as an 8, but one originating as a 2 rather than a 1. But this would apparently motivate instrumental case marking as well as an IP analysis under which 'trees' would head an 8 arc local successor of a 1 arc. For Ostler (1979) himself points out cases involving instrumental 8s in structures where there is no question of their being 1s at any level:
I take these examples to indicate that the ditransitive Sanskrit V 'apportion' occurs with 3s, which are marked dative if they, surface, as in [12b], but that this V also permits 32A. When the original 3 advances to 2, it is regularly marked accusative, while the original 2 demotes to 8 and, crucially, is marked instrumental. Thus the instrumental is independently a 2-chômeur, as well as a 1-chômeur, case, although apparently not all 8s in Sanskrit are
< previous page
page_162 If you like this book, buy it!
next page >
< previous page
page_163
next page > Page 163
marked instrumental. This is not unusual since 8 marking is rarely as fully systematic as term marking, itself not invariable. The instrumental marking of the nominal in putative IPs like [11] thus fails to support an IP analysis against a structure like [12]. There remains the fact that clauses like [11] involve the various passive verbal morphologies, those found on Sanskrit PPs. But this also is probably of limited relevance since, apparently, at least some of these are more widely used, and used in fact to mark what would be unaccusative advancement clauses in RG/APG terms. Thus, for the passive marker ya-, Rosen (1981 a: 39) cites the following paradigm:
Given that [14d] is the kind of clause in which 'rice' would, under the unaccusative hypothesis, be an initial 2, as in [14a], the presence of -ya- in [14d] suggests strongly that this morpheme can (must?) be the marker of 21A in nonpassives as well as passives. But this motivates the passive marking of the V of the sort proposed in [12] even under a non-IP analysis of Sanskrit cases like [11]. Obviously it has not been shown that [12] is the correct analysis for the putative Sanskrit IPs. But clearly Ostler has not made a real case that what he calls 'impersonal passives' in Sanskrit are actual IPs in a sense where they would be a problem for universal unaccusativity, rather than instances of an apparently independently existing non-IP, nonpassive unaccusative advancement structure like [12]. Ostler (1979) also claims there is counterevidence to the claimed universal incompatibility between IPs and unaccusative predicates based on what he calls 'reflexive passives'. Thus alongside (8a, b) above one finds [15]:
However, these clauses are passives only in the sense of having the three possible passive verbal morphologies. They also contain the optional form
< previous page
page_163 If you like this book, buy it!
next page >
< previous page
page_164
next page > Page 164
translated reflexively. It is unclear to me what determines. this translation, since Ostler (1979: 383) states that true reflexive clauses have a different form. In any event, a natural approach to clauses like these, given what has been seen of the nature of -ya-, would deny that they are passive clauses and take them to involve intransitive, unaccusative structures. This is consistent with their meanings as given, which involve no element corresponding to what would be an initial 1 in RG/APG terms. The relation between cases like [8a] and [15] seems to be essentially that between French pairs like [16a and b]:
Despite the fact that casser is a typically transitive V, I take structures like [16b] to be initially intransitive. The occurrence of a paired morphologically related intransitive incohative item for a causative transitive item is a common feature of attested NLs. Moreover, it is also common that the incohative item be 'reflexive'. Although French does have true reflexive passives, the se-moyen construction discussed earlier, Ruwet (1972) showed that these contrast with structures like [16b], which he called 'neuters'. In current terms, [16b] involves a single initial nominal, a 2, which advances to 1 leaving a copy. 10 I see no objection to assuming that Sanksrit examples like [15] have a similar structure. The form -ya- or other passive markers would be motivated in these nonpassives for the same reason as in Rosen's example [14d]. The key problem for RG claims raised by the so-called reflexive passives is that they permit what Ostler calls 'impersonal reflexive passives'. Thus corresponding to [15] one finds [17]:
The differences between [15] and [17] are that the nominal is in the instrumental in [17] and that the verbal forms are 3sg. neuter, instead of masculine in agreement with the final (nominative) 1, as in [15]. But again there is no genuine evidence that 'grain-holder' is a 1 at any level, and so there is a plausible analysis of such clauses parallel to that suggested for [11a], one which avoids taking it as any kind of passive. It would be a variant of [15] in which the initial 2 arc sponsors a stable ghost arc, the latter having a 1 arc local successor, while the initial 2 arc has an 8 arc local successor. The
< previous page
page_164 If you like this book, buy it!
next page >
< previous page
page_165
next page > Page 165
latter would then determine instrumental marking for its head just as in earlier cases of 8s originating directly as 2s. The clause would involve unaccusative advancement, motivating the passive morphology. It would be the invisible dummy which is final 1; quite possibly this would regularly determine the neuter agreement of the verbal form. My suggestion is thus that the impersonal Sanskrit clause in [17] has a structure essentially like the EXI variant of the French clause in [16], which is [18]:
Again, although Ostler has not succeeded in providing real evidence for the view that the instrumental marked nominal in [17] is a 1 at any level, this would be a necessary (though insufficient) condition for a counterargument to the joint maintenance of the 1AEX and the universal assignment of unaccusative status. 5.5 Nerbonne's Critique One of the sharpest and most extensive criticisms so far of the claim that passive clauses based on unaccusative predicates are universally impossible is that in Nerbonne (1982). Sections III and IV of this study present data from Lithuanian, Irish and Estonian, all claimed to show that predicates which would be unaccusatives by the universal criteria suggested in Perlmutter (1978a) nonetheless allow 'impersonal passives'. Nerbonne's tack in all the cases is essentially the same. He isolates a certain verbal morphology, call it VM, which he takes to be a passive morphology. In some, but not all cases, VM is found with clear and for the most part uncontroversial PPs and in all cases VM is in contrast with the verbal morphology of uncontroversial active clauses. He refers to VM, presumably in general accord with traditional descriptions of these NLs, as respectively 'passive' (Lithuanian), 'autonomous' (Irish) and 'impersonal' (Estonian). Let us refer to the claim that unaccusative predicates do not permit true IPs as NEG-IP. The logic of what Nerbonne considers, and repeatedly refers to as, a refutation of NEG-IP 11 then reduces chiefly, but not exclusively, to the citation from these NLs of predicates which are both unaccusative by the hypothesized universal criteria and which yet occur with VM. There is no doubt that Nerbonne has in fact accomplished the presentation of such an array of VM-marked unaccusative predicates. Were the discovery and display of such VM-marked universal unaccusative predicates sufficient grounds for rejecting NEG-IP, Nerbonne's discussion would indeed have refuted it. But of course, the just stated conditional is false. Hence, while it might be possible to build a refutation on the kind of data
< previous page
page_165 If you like this book, buy it!
next page >
< previous page
page_166
next page > Page 166
Nerbonne cites, he has at best only provided one step of the required argument. In particular, he hardly considers alternative syntactic analyses of the clauses he considers to be IPs of unaccusative predicates, analyses like those suggested above for Turkish and Sanskrit cases which have been taken to counterexemplify NEG-IP on much the same grounds Nerbonne appeals to. 5.5.1 Lithuanian Nerbonne's most extensive case is drawn from Lithuanian, which has a rich nominal case system. Moreover, predicates and adjectives agree with their final 1, apparently in case, gender and number. He states that Lithuanian has what is, in APG terms, a (nonreflexive) PP, whose verbal structure involves combining the present passive participle in -m- or 'the past participle in -t- ith the auxiliary V buti 'to be', in any tense. The final 1 of such PPs is in the nominative, which then determines this case for the participle, via agreement. An example of such a PP is [19]:
Although Nerbonne gives no examples of clear 'long' PPs in which the initial 1 manifests as a superficial nominal, he states (p. 74), that under this condition, such a nominal appears in the genitive. Nerbonne then indicates that Lithuanian allows true IPs, by citing [20]:
Nerbonne takes [20] to be an IP, so that Lithuanian apparently uncontroversially allows IPs; [20] then seemingly illustrates that Lithuanian marks the initial 1 of a passive clause with the genitive case when this nominal visibly manifests in such clauses as a non-1. It also indicates that in impersonal cases, where there is no visible final I, the participle which normally agrees in gender with the final 1 takes the neuter form. One further fact is cited by Nerbonne to indicate a priori that Lithuanian has IPs. This is that there are apparent IPs of transitive structures. Hence, alongside [19] above, one also finds [21]:
< previous page
page_166 If you like this book, buy it!
next page >
< previous page
page_167
next page > Page 167
This is superficially different from the PP in [19] in two ways. The original 2 is here in the genitive, as is the corresponding nominal in the corresponding active (evidently irregularly), and the participle is neuter, indicating that this nominal is not the final 1. But [19] and [21] share VM. From this Nerbonne (1982: 74) concludes: 'It is clear that we are dealing with impersonal passives.' Having established this much about Lithuanian, Nerbonne constructs an alleged refutation of NEG-IP essentially by running through Perlmutter's (1978a) list of semantic categories determining initial unaccusatives and finding a Lithuanian intransitive V of each category which occurs in a clause which manifests (i) VM and (ii) the neuter form of this. Typical examples are [22a and b]:
For Nerbonne then, the conclusion is relatively straightforward. On Lithuanian-internal grounds, [22a, b] are IPs, but their predicates are universal unaccusatives; hence NEG-IP is refuted. There are two ways in which one might try to resist Nerbonne's argument against NEG-IP. First, one might haggle about the universal unaccusative status of the predicates. Nerbonne really considers only this approach and rightly rejects it. It is clear that if there is a universal unaccusative category, most of the cases he cites must be members of it. But a second way to avoid the conclusion would appeal to the possibility that sentences like [22a and b] are not passives. Nerbonne does not even contemplate this view. But, despite the morphological facts cited so far, I suggest that not only is this (obviously) a logical possibility, it is plausible and is precluded by no material in Nerbonne (1982). If, however, the clauses in question are not passives, they are not IPs, and hence their existence can refute no claim about passives, and in particular, not NEG-IP. Thus a natural approach is to reject the view, quoted above, that the VM cases treated by Nerbonne are clearly IPs. More precisely, it will be argued that the clauses of relevance, except for [21], are not passives of any sort. Since Nerbonne (1982) is a secondary source about Lithuanian and since I know nothing about this NL independently, I will make an assumption not fully supported by data he cites. However, this assumption strengthens Nerbonne's case, so it can hardly be considered unfair. I will simply assume
< previous page
page_167 If you like this book, buy it!
next page >
< previous page
page_168
next page > Page 168
that unaccusative predicates like those in [22a and b] can occur in active clauses, that is, with the non-VM morphology and with the single argument in the nominative. This assumption is necessitated by the fact that Nerbonne does not in general cite: active counterparts of the clauses he takes to be impersonal passives of intransitives. However, he does speak [76] of active correspondents of certain of these, so I am simply making the maximal generalization. The assumption seems uncontroversial; were things otherwise, the situation would be unusual enough to require comment in the sources. Given this, the way to defend NEG-IP from Nerbonne's data is to pursue the view, as in the parallel cases treated earlier from Turkish and Sanskrit, that examples like [22a and b] are not IPs. If not, then despite the presence of the VM morphology on uncontroversial PPs like [19], this morphology should not be referred to uncritically as 'passive'. This amounts, in APG terms, to the claim that although VM is, at least over the domain discussed by Nerbonne, systematically determined by clause structure, it is not defined by clauses having arc-passive arcs, nor is it defined by the mere property of 21A, as suggested in certain earlier cases. For given my assumption of the previous paragraph, some Lithuanian clauses with unaccusative advancement have active morphology. The two key issues are the following: (i) what is the essential structure of both what appear to be IPs of unergative predicates, like [20], and of unaccusative predicates, like [22a and b], and (ii) what is the general clausal feature which determines VM morphology? It will be argued that (i) there is a standard APG analysis for the intransitive clauses taken by Nerbonne to be impersonal passives under which they are not passives at all; (ii) this analysis yields a general, nondisjunctive prediction of VM morphology for all the Lithuanian clauses cited in Nerbonne (1982) and (iii) this analysis also yields a general, nondisjunctive analysis of the occurrence of genitive case as a marker of 8s. The basic idea is that the Lithuanian clauses taken by Nerbonne to be IPs actually have the distinct impersonal structure in [4a] of Chapter 4, section 1. They are structures where a 1 arc locally sponsors a ghost 1 arc. In this case, if the earlier 1 arc is headed by a specified nominal, it has an 8 arc local successor. Thus I am suggesting that the clauses taken by Nerbonne to be IPs actually have a structure similar to English clauses like [23b]:
In such cases, a dummy 1 'displaces' an earlier 1, which ends up as an 8. In these terms, the structure of the long form of unergative example [20] would be, concentrating on essentials, [24]:
< previous page
page_168 If you like this book, buy it!
next page >
< previous page
page_169
next page > Page 169
The intransitive clause is represented as having an initial 1 arc, B, and hence is unergative. The initial 1 is, however, a final 8, since the initial 1 arc sponsors A, a 1 arc ghost (hence a stable ghost). This self-erases, an unsurprising fact, since even nondummy final 1 pronouns seem, from Nerbonne's data, to be (at least optionally) invisible in Lithuanian. The self-erasure accounts for the absence of a visible final 1 in [20]. The key thing about [24] is that it is a standard dummy structure, one taken in Postal (1982) 12 to underly such clear nonpassives as French [25b] from Boons, Guillet and Leclère (1973: 189):
Perimutter and Zaenen (1984) argue that this pattern underlies many Dutch clauses. Perlmutter (1983b) considers these structures as one of the total class of impersonal structures possible in NLs. The existence of the kind of structure in [24] is thus supportable independently of what is said about any Lithuanian clauses. A relational grammatical theory is going to countenance such structures in any event. A question immediately arises as to why the 8 determined in [24] is marked with the genitive case, the same case which marks what would be the final 8 (initial 1) in a Lithuanian PP. The answer can be taken to be generalization [26], which covers every case in Nerbonne (1982):
< previous page
page_169 If you like this book, buy it!
next page >
< previous page
page_170
next page > Page 170
Formulation [26] is a maximally general and simple statement, given the presence of genitive marking on the 8s of true PPs. Consider now the apparent IPs which Nerbonne takes to refute NEG-IP, for example [22b], which would, under the analysis I am proposing, have the relevant structure in [27]:
Since 'disappear' is a putatively universal unaccusative, the clause involves an initial 2 arc, D. But since [22b] is taken to be an active and not a passive clause, [26] represents it as involving unaccusative advancement. It then differs from a simple active clause in having the same stable 1 arc ghost structure as [24], yielding an invisible final 1 heading a selferased ghost arc. Thus the general idea is that Lithuanian ls have the possibility of being 'displaced' by invisible 1 dummies with some freedom. Significantly, [27] is not a passive clause according to the universal account, but an unaccusative advancement clause consistent with the 1AEX, since it involves only a single 1 arc local successor. While both A and C are neighboring 1 arcs, ony C is a local successor, since A is a ghost, and thus not a successor at all. If [27] is essentially the right structure for [22b], Nerbonne's claim that such examples refute NEG-IP is incorrect. But one must face the most obvious criticism of [27], the objection that it provides no conjunctive generalization which determines in a non-ad hoc way that clauses like [22b] have the VM morphology. This objection might be bolstered by the observation that French EXI impersonal sentences like [28b] notably have an active verbal morphology like the personal [28a] and distinct from French nonreflexive passive morphology, found only in the ill-formed *[28c]:
< previous page
page_170 If you like this book, buy it!
next page >
< previous page
page_171
next page > Page 171
*[28c] shows French passive morphology, the passive auxiliary, etc., but is ill-formed. If then, it is claimed that Lithuanian structures like [22b] are not passives, why do they have the VM morphology typical of undisputed passives like [19]? A natural answer is that while the morphology called 'passive' in French is indeed a passive morphology in the sense of occurring systematically 15 in no nonpassive clause,16 the VM morphology of Lithuanian can be claimed to be based on a distinct structural property. Moreover, this structural property is distinct not only from passive, but also from mere 21A, since it is not systematically present in Lithuanian unaccusative clauses, but only in those here taken to have dummy final Is, that is, no visible final 1. The question then is what systematic structural property [24] and [27] share with a standard passive clause which yet distinguishes them from both simple unergative and unaccusative Lithuanian clauses. (These, it is here assumed, always involve the active verbal morphology, not VM.) That is, what property does the stable 1 arc ghost structure share with a passive which it fails to share with either simple unergative clauses, simple transitive clauses, or simple unaccusative advancement clauses? To answer, consider simplified examples reduced to essentials of the three basic clause types in Lithuanian assumed to require active morphology;
< previous page
page_171 If you like this book, buy it!
next page >
< previous page
page_172
next page > Page 172
< previous page
page_172 If you like this book, buy it!
next page >
< previous page
page_173
next page > Page 173
Each of [29a-c] shares the property of not containing a free 1 arc which is overrun. In terms of the notion Insecure defined in the Postscript to Section 5.2, none contains an insecure 1 arc. On the other hand, in any passive clause, the first (free) 1 arc is always overrun, and, since not replaced, is insecure. Moreover, in [24] and [27], which share the VM morphology with Lithuanian PPs, the first 1 arcs, C in [24] and C in [27], are both free and overrun (by A and A, respectively). For the available data then, the relevant generalization appears to be [30a], which is no less general than the account implicit in Nerbonne's view that VM is the morphology is of passive clauses, which is essentially [30b]:
The analysis suggested here combines with principles [26] and [30a] to offer a suitable analysis for examples like [21], namely [31]:
< previous page
page_173 If you like this book, buy it!
next page >
< previous page
page_174
next page > Page 174
That is, rather than being taken to be IPs, such examples are analyzed as involving both the structure of PPs, existing independently in Lithuanian, and the stable 1 arc ghost structure proposed for the intransitive cases. The initial 2 in such clauses is thus a later 1, and a still later 8, hence a l-chômeur, being predictably genitive-marked via [26]. The VM morphology of the V is also predicted, in fact doubly so, since both the initial 1 arc, A, and the arc-passive arc, D, headed by the initial 2 are insecure. As observed by D. E. Johnson (personal communication), an alternative to [30a], namely [32], appears to cover the facts in Nerbonne (1982), under analyses like [24] and [27]:
This says that VM is the morphology of clauses containing 1-chômeurs. The problem is that [32] is incompatible with the APG account of 'short' passives, which, as previously seen, involves no (invisible) 8 arcs but instead involves erasure (CLEARING in the sense of [4.54b]) of the initial 1 arc by the arc which overruns it; see JP (1980: Chapters 8 and 9). That is, in this account, unlike normal RG practice, not all insecure arcs have local successors. Lithuanian of course has 'short' passives with VM morphology. Consequently, [32] does not seem to be a viable alternative to [30a], given that there seems to be no motivation to abandon the APG analysis of 'short' passives just to adopt [32], which has no advantage over [30a]. I believe that conditions (A, B and C) set out above for the description of the Lithuanian data in Nerbonne (1982) have been met. It has been shown that the putative IPs of Lithuanian claimed to refute NEG-IP have analyses consistent with NEG-IP which provide general accounts of both the distribution of genitive marking on 1-chômeurs and of the distribution of VM. There is no evident way in which this analysis is inferior to one which takes e.g., [22a and b] to be IPs. Therefore, there is so far no argument that Lithuanian VM clauses involving universal unaccusatives have any negative bearing on NEG-IP. This could only be shown by providing some argument against the sort of syntactic analysis sketched here. One objection which is to be anticipated in the defense of NEG-IP from the Lithuanian facts (brought forward by Nerbonne) like the one I have proposed is a claim that it renders the prediction of a universal incompatibility between IPs and (universally determined) unaccusative predicates empty or factually vacuous. It could be maintained that no pattern of facts could fail to receive some treatment like that just sketched. But I will now argue that this is not the case. At issue is the need to show that data from particular NLs could potentially distinguish the sort of clause structures I
< previous page
page_174 If you like this book, buy it!
next page >
< previous page
page_175
next page > Page 175
have posited for Nerbonne's putative IPs based on unaccusative predicates from true IPs of these. Compare then [27], the structure posited here for [22b], with the structure that [22b] would have if it were, in current terms, a true IP of an unaccusative predicate:
While [27] and [33] have many similarities, they also differ, and this offers the possibility of facts which disconfirm an analysis like [27]. The latter structure involves a stable 1 arc ghost and thus a dummy 1 arc, A, while [33] involves an unstable 2 arc ghost, B, with A an instability dummy 1 arc. But it was seen in the discussion of French in Chapter 4, sections 4-8 that stability and instability dummy 1 arcs can contrast in behavior. Hence, if one could independently show that Lithuanian stability dummy I arcs had some property lacking in A in [27], this would count against [27] and, indirectly, favor [33]. Further, the analysis proposed here groups the putative IPs with clauses parallel to English transitive clause cases like [34b], in which, evidently, a dummy 1 arc 'displaces' an earlier (clausal) 1 arc sponsor:
< previous page
page_175 If you like this book, buy it!
next page >
< previous page
page_176
next page > Page 176
If then, the determination of VM morphology by clauses containing insecure 1 arcs is maximally general, the present hypothesis predicts, contrary to an IP account, that the analog of [34b] in Lithuanian would require VM morphology. The most important: way in which analyses like that suggested here for Lithuanian impersonals can in principle be factually differentiated from true IPs lies in appeal to the grammatical laws which limit each type of structure. So far, the burden in this area has been borne exclusively by the 1AEX. But other potential laws could in principle show that a given impersonal structure was properly represented along the lines of [27] and not [33], or conversely. There are some clear Lithuanian data as yet undiscussed which suggest the relevance of such appeal to lawful properties of contrasting impersonal clause types. An important feature of the Lithuanian impersonal construction is that it is possible with predicate nominals, as illustrated by [35]:
These sentences have an important feature not really dealt with by Nerbonne. The predicate nominal agrees in case with the genitive nominal, which, however, determines no agreement at all on the main V. One can account for these facts under the current assumptions as follows. As is motivated in the postscript to this section below, there is a constraint which limits the Lithuanian stable 1 arc ghost structure posited here to clauses whose ls do not undergo the raising associated with auxiliaries and Vs like pasiródyta 'turn out'. Assume then that in clauses like [35], the predicate nominal constituent defines a separate clause embedded below the unaccusative main V. Since the 1 of the subordinate clause ends up as a main clause constituent, the main V would determine raising of the complement 1. It follows that the invisible dummy in [35a, b] is only a constituent of the highest clause. The relevant structure of [35b] would be [36]:
< previous page
page_176 If you like this book, buy it!
next page >
< previous page
page_177
next page > Page 177
The key point is that the genitive nominal, jõ, is the final 1 of the lower clause. But it is entirely regular that Lithuanian adjectives and predicate nominals agree (inter alia) in case with their final 1. The fact that this nominal only receives its (genitive) case as a function of its final 8 status in the main clause is irrelevant. The final 1 of the main clause is the invisible dummy corresponding to ghost arc B, which can be taken to determine the form of the main clause V, also by regular agreement with the final 1. The subordinate agreement would not work out correctly if the ghost arc structure were an aspect of the lower clause. Notably then, the impersonal Lithuanian structure whose status is at issue combines freely (i) with auxiliary Vs, (ii) in such a way that a dummy arc is only present in the main clause, and (iii) with predicate nominal structures. These properties may well offer grounds for factually differentiating the kind of impersonal analysis advocated here from a true IP analysis of Lithuanian impersonal clauses. First, as noted in Perlmutter and Postal (1984b), true IPs are in a clear sense incompatible with auxiliaries and related forms. The basic observation is that a predicate incompatible with IPs still permits no IP when embedded below an auxiliary or other intransitive raising predicate. Thus, if an auxiliary has as its final 1 the
< previous page
page_177 If you like this book, buy it!
next page >
< previous page
page_178
next page > Page 178
dummy of a true IP, it is only because the complement of the auxiliary is itself independently an IP, and the dummy is a complement constituent. This follows from the 1AEX under the view that such auxiliaries and, more generally, all intransitive predicates taking complements are unaccusative. Hence in NLs like French, German, etc., where fixed unaccusative predicates preclude IPs, embedding these predicates below auxiliaries or other raising predicates; does not help. Similarly, the auxiliary of, e.g., an adjective is not subject to IP formation. These constraints are illustrated for French in [37]:
In current terms, the 1AEX explains *[37c-e and g] to the same extent as it does *[37b]. But Lithuanian allows impersonal versions of sentences with auxiliaries. The following forms an essential minimal pair with *[37g]:
This contrast between the apparent IPs of Lithuanian and the true IPs of French needs an account in any framework. In current terms, it follows from the view that French actually has IPs, which are predictably incompatible under the 1AEX with all unaccusatives, including intransitive raising forms, while Lithuanian has a stable 1 arc ghost structure, about which the 1AEX says nothing.
< previous page
page_178 If you like this book, buy it!
next page >
< previous page
page_179
next page > Page 179
Second, [35a, b] illustrate the direct compatibility of predicate nominals with the Lithuanian impersonal forms. On the contrary, in the NLs known to me where the 1AEX and the universal assignment of unaccusative status relatively unproblematically predict the incompatibility of IPs with unaccusatives, there are no IPs of predicate nominal constructions. Consider the semantically equivalent French/German pairs in [39]:
True IP constructions of French and German are incompatible with predicate nominals. 17 With respect to French, one might claim that *[39b and d] are blocked by the same constraint (whatever it is) that precludes French IPs of transitive clauses; see the discussion of [1.9]. It might be suggested that French IPs are simply incompatible with a post-verbal nominal. Strikingly though, French IPs are also incompatible with those French predicate nominals that are marked with prepositions:
< previous page
page_179 If you like this book, buy it!
next page >
< previous page
page_180
next page > Page 180
Any attempt to reduce *[39b] to the same constraint blocking *[40b] 18 (but allowing [40d]) runs into the impossibility of *[40f]. This does not involve a postverbal nominal but does involve the IP of a predicate nominal clause. The question arises whether the correlation between the possibility of impersonal forms with unaccusatives and the possibility of impersonal forms of predicate nominal clauses is an accident. It is possible that grammatical laws preclude IPs of predicate nominal clauses. If so, the existence of impersonal predicate nominal sentences in Lithuanian would be a clear indication that these are not IPs and the behavior of predicate nominal clauses with respect to impersonal forms would yield a way of distinguishing the factual claims made by true IP analyses and nonpassive impersonal structures like that proposed here for Lithuanian. JP (1980) presents a novel account of predicate nominal clauses in which the (nonauxiliary) clause is characterized by a noninitial P arc more precisely, a P arc local successor of a 2 arc. Thus the nonauxiliary active clause in [39a] would include the structure in [41]:
Although it may at first seem bizarre to analyze a predicate nominal as an earlier-level 2, this idea is motivated in general terms in JP (1980: Chapter 7). Moreover, rather specific evidence for it can, on occasion, be found. It is a well-known and much-discussed fact (see Kayne (1975: 382-383), (1978: 712), (1980: 86-92), (1981: 127); Gross (1968: 25); Ruwet (1972: 49); Milner (1978: 49); Couquaux (1981: 43-44)); Burzio (1981: 210-211)) that
< previous page
page_180 If you like this book, buy it!
next page >
< previous page
page_181
next page > Page 181
the French pronominal clitic enmore precisely, those instances of en associated with quantificational forms like beaucoup, 'many', certains 'certain', with cardinal numbers, etc., is only possible when the nominal to which they relate has a certain status in a clause. Compare [42a-d]:
It is normally agreed that these instances of en must be associated with 2s. But a neglected fact 19 is that such en are grammatical with predicate nominals:
[43a and b] are perfect, while [43c] seems only as far-fetched as its English translation. By taking predicate nominals to head 2 arcs, the possibility arises of capturing the generalization linking [43] to [42a]. This could be done via appeal to the notion acting 2 arc, as discussed in Chapter 2. Recall that an acting 2 arc is a 2 arc whose local successor, if any, is not a term arc. Clearly, a 2 arc with a P arc local successor is an acting 2 arc and, under the analysis of JP (1980), predicate nominals are thus acting 2s, just as are nominals corresponding to 8 arc local successors of 2 arcs.20 Although the actual formulation of the generalization requires considerable further work, it is clear already that an analysis which approaches the structure of predicate nominals to that of 2s is supported by the possibility of en in cases like [43]. Incidentally, one can not seriously claim that the 'source' of such en can simply be any postverbal nominal. This is by no means true, as shown by [44] and contrasts like [45] from Pollock (1981):
< previous page
page_181 If you like this book, buy it!
next page >
< previous page
page_182
next page > Page 182
In present terms, en is impossible in [44b] and [45d] because the 'source' nominal is an adverb in these cases and heads no 2 arc at all. On the contrary, in [45a], the nominal does head a 2 arc as shown, inter alia, by the fact that, in contrast to that in [45b], it permits a corresponding PP:
Other contrasts between passer and rester consistent with the characterization given here are cited by Pollock (1981). Further evidence of the adverbial character of the postverbal nominal in [45b] is provided by such data as [47]:
Overall then, support for the claim that the type of en cliticization at issue depends on the source nominal heading a 2 arc is far from insubstantial, which supports the suggestion of JP (1980) concerning: the 2=PLocal Successors(A, B) analysis of predicate nominal structures. Consider again then the failure of predicate nominals to combine with true IPs in French, German, etc. None of the principles of JP (1980) seem to preclude the combination of the structure of [41] with the unstable ghost arc structure taken here to define true IPs. But various potential laws would have this effect. Wanted are principles which would entail, e.g., [48]:
< previous page
page_182 If you like this book, buy it!
next page >
< previous page
page_183
next page > Page 183
At worst, one could stipulate [48] as a law. But it could well follow from something more general. Conceivably, for instance, a 2 arc local predecessor of a P arc is incompatible with any other neighboring 2 arc whatever. That is, there could be the grammatical principle [49]:
Such a law would entail inter alia that the kind of detransitivization imposed by the structure in [41] is not subject to later stratum retransitivization. If [49] can be maintained as a law, the fact that French and German predicate nominal constructions are incompatible with true IPs is subject to an explanation. Furthermore, the compatibility of Lithuanian impersonals with predicate nominals would then indicate that these impersonals are not IPs. Some support independent of IPs for a principle like [49] is probably derivable from French. 21 The EXI construction touched on in several places earlier (see [3.17]) ff. and especially [4.11]) is also incompatible with predicate nominals:
The incompatibility illustrated by all the variants of *[50d] also follows from principle [49] given the analysis of the EXI construction suggested in Chapters 3 and 4. Under that analysis, this construction is defined by the sponsoring of a stable 2 arc ghost by a 2 arc, which has an 8 arc local successor. In these terms, the postverbal nominal of [50b] is thus a final 8. But it can be a source for the clitic en since it heads an acting 2 arc. In some cases, for some speakers, certain nominals can head the sort of 2 arcs which can sponsor the 2 arc ghosts defining EXI structures only via the possibility of 2 arc local successors for certain 1 arcs. Thus some, but only some,22 speakers accept EXI with presumably unergative intransitives, as in [25]. But
< previous page
page_183 If you like this book, buy it!
next page >
< previous page
page_184
next page > Page 184
this possibility is never known to be extended to cases like *[50d]. Principle [49] combines with the view that predicate nominals involve P arc local successors of 2 arcs to offer an explanation. For the ghost 2 arc defining the EXI construction in *[50d] is necessarily distinct from the 2 arc local predecessor of the P arc:, contravening principle [49]. Examples like *[40f] were taken above to argue the inadequacy of forbidding IPs with (bare) postverbal nominals, since *[40f] contains a predicate nominal of the sort marked with a preposition. This supported an appeal to a principle like [49] for the IP/predicate nominal incompatibility. Significantly, the same structures also undermine an attempt to reduce the ill-formedness of cases like *[50d] to the well-known fact that the French EXI construction is incompatible with bare postverbal nominals distinct from the EXI nominal itself. 23 For again EXI is also incompatible with predicate nominals marked with prepositions:
Therefore, the failure of EXI in cases like [50d] remains apparently unaccounted for without appeal to a principle like [491, which is again supported, under the view that French preposition-marked predicate nominals, like bare ones, involve 2 arc local predecessors of P arcs.24 The implication of this discussion is that there is nontrivial support for an APG law which would explain the incompatibility of true IPs and predicate nominals.25 In any event, the mere possibility of such a law shows that the sort of impersonal analysis suggested here for Lithuanian is in principle distinguishable from a true IP structure. As the class of grammatical laws is elaborated further, new ways of distinguishing even such subtly differing structures as [27] and [33] may become apparent. 5.5.1 Postscript After having essentially completed section 5.5.1, I became aware of Timber-lake (1982). Appealing to Lithuanian data related to that cited in Nerbonne (1982), Timberlake also produces a critique of the account which predicts that there are no true IPs of unaccusative predicates in any NL (NEG-IP). Although Timberlake's argument is partially similar to Nerbonne's, it is also partially different: it is elaborated in greater detail, and presents certain distinct data of interest. Thus it deserves comment. But I stress that the analysis already proposed in section 5.5.1 suffices, as far as I can see, to defend NEG-IP from all of Timberlake's material, just as it does from all of Nerbonne's. Timberlake provides, whereas, as noted, Nerbonne did not, good examples of related active/'long' passive pairs like (A):
< previous page
page_184 If you like this book, buy it!
next page >
< previous page
page_185
next page > Page 185
The standard PP in (Ab) shows the initial 1/final 8 marked genitive, as was only assumed in our earlier discussion. Timberlake provides various arguments which in effect support the general RG/APG view of nominals like that marked genitive in (Ab), namely, that this is a final non-1 which is an earlier stratum 1. Timberlake supports the least obvious of these claims, the assignment of earlier 1 status, by appeal to data involving the control of reflexive elements. The relevant principle is that only Is (at some level) can antecede reflexives. The argument is quite close to that given for Russian in Perlmutter (1978b, 1982, 1984b). Genitive nominals like that in (Ab) are then shown by Timberlake to be possible antecedents of reflexives. This is true both of the earlier 1s of PPs like (Ab) and of the earlier 1s of the Lithuanian impersonal construction. Note that although the analysis advocated here denies that the latter are passives of any type, it nonetheless specifies that the genitive-marked nominal in an impersonal construction is an earlier 1. Significantly, Timberlake provides examples similar to [21] above but in which both the initial I and initial 2 of a transitive show up in a VM-marked clause as genitive marked nominals. Such clauses are not dealt with by Nerbonne, but Timberlake explicitly takes them to be IPs of PPs:
While for Timberlake (Bc) is an IP of a PP, here it is naturally analyzed like [21].
< previous page
page_185 If you like this book, buy it!
next page >
< previous page
page_186
next page > Page 186
In (Bc), the participle, which agrees with the initial 2, 'leaf, is genitive, in contrast to the participle of the simple PP in (Bb), which is nominative. This can be accounted for by a constraint on the position in a chain of embedded forms in which the 1 arc ghost structure posited in section 5.5.1 can occur. If Lithuanian VM participles always required an immediately higher auxiliary 'be', one might be able to state this as limiting the ghost structure to forms of 'be', or, possibly, to finite clauses. But as examples like (Bb), [22b], etc., show, many VM participles do not require a form of the auxiliary. The principles determining the presence of the latter are unclear' to me. Assuming though that the auxiliary, when present, determines that the final 1 of its complement raises to be a 2, and then a subsequent 1, one can suggest that the constraint on the ghost structure is that it is only possible in the highest of a chain of such raising 'triggers'. This is equivalent to claiming that the ghost 1 arc can never be the predecessor for the sort of raising associated with auxiliaries. Hence, when the auxiliary is present, the ghost can only exist in the auxiliary clause, etc. These assumptions determine that in (Bc) the ghost can only exist in the main clause, whose predicate is buta 'been', yielding a structure like (C) for (Bc):
< previous page
page_186 If you like this book, buy it!
next page >
< previous page
page_187
next page > Page 187
This permits an account of the agreement facts without giving up the view that all predicate agreement in Lithuanian is with final 1s. For in (C), the final 1 arc of the lowest predicate, that of the participle, is I, the arc-passive arc defining the complement as a PP. This arc has a foreign successor, under the raising determined by the passive auxiliary, and ultimately the final 1 of the complement is a 1 of the main clause. But it is not the final 1 of the main clause since it must be demoted to 8, given the ghost 1 arc structure. So the final 1 of the main clause is a dummy corresponding to arc A. Under this analysis, only the unaccusative main auxiliary clause manifests the ghost structure. One might have expected Timberlake to take (Bc) to disprove the 1AEX. But he does not. On the contrary, he claims to actually provide evidence from Lithuanian supporting the 1AEX and concludes, instead, apparently even more radically, that (Bc) disproves the idea that passives involve advancement (of a dummy). He claims (1982: 518): The formation of impersonal passives from personal passives, which is fully grammatical in Lithuanian, clearly violates the advancement analysis of passives in RG. Timberlake comes to the view that it is the Motivated Chômage Law which should be abandoned, so that cases like (Bc) can involve unmotivated chômage of the 1 determined by personal passivization. But, given structures like (C), this is not required at all. Underlying Timberlake's discussion, like Nerbonne's, is the unargued-for assumption that the impersonal cases are IPs. If this view is given up, neither any universals nor any cited Lithuanian-internal generalizations need be abandoned. In fact, Timberlake (1982: 522-523) himself comes very close to such an analysis, saying: One can anticipate two possible modifications of RG that might allow it to maintain a Motivated Chomage and/or advancement analysis of Passive. First, one might suggest that impersonal passives of all intransitives are derived by insertion of a dummy nominal directly as subject (italics mine): which would make the subject a chomeur. Demotion by dummy insertion is conceivable, given that it is already used for direct objects. On the assumption that demotion by dummy insertion is motivated, this would superficially allow the principle of Motivated Chomage to be maintained, at the expense of giving up the advancement analysis of Passive. First, the word 'superficially' here is superfluous and misleading. Second, it is incorrect to say that the analysis Timberlake considers, one version of which has been proposed and argued for here, has as an expense the necessity to give up ''the advancement analysis of Passive". All that is required is giving up the idea that certain specific Lithuanian sentences, like certain sentences
< previous page
page_187 If you like this book, buy it!
next page >
< previous page
page_188
next page > Page 188
from Turkish considered earlier, involve passivization. But this idea has never been truly motivated in a nonsuperficial way by either Nerbonne or Timberlake. Nor is there any known reason to consider giving up this assumption an expense. Since all that is involved are descriptive decisions about Lithuanian among alternative structures always allowed in RG as well as in APG terms, it is also mistaken to suggest that the Lithuanian facts demand some theoretical modifications. So far, nothing in Lithuanian has been shown to require this. Timberlake's viewcontra Nerbonnethat the 1AEX is not threatened by the Lithuanian passive facts is based on certain extremely interesting data whose analyses are, however, less obvious than Timberlake assumes. Timberlake takes these data to positively support the 1AEX; hence his conclusion that what he regards as IPs of PPs, etc., must falsify some other RG assumptions, in particular, the Motivated Chômage Law. The relevant facts involve what seems to be passivization out of the infinitival complements of Vs like intend. In the following examples, the capital H stands for the Lithuanian phrase i honoraro 'from honorarium':
(Dc) shows that the PP in which the 2 of the infinitive ends up as final 1 of the main clause is subject to the impersonal structure. Under the current assumptions that the latter is possible only in a clause whose 1 need not raise, the middle verbal form numatyto agrees in case with its final 1, which is the nominal corresponding to the arc-passive arc of a PP. This case is genitive as a result of the raising of this nominal into the main clause of the passive auxiliary, where it is demoted to 8 by the stable ghost arc structure. What Timberlake considers to be evidence favoring the 1AEX comes from a constraint on such infinitival passives, illustrated by the related forms in (E):
< previous page
page_188 If you like this book, buy it!
next page >
< previous page
page_189
next page > Page 189
According to Timberlake, (Ec) shows that when the main clause 2 is not visible (which can mean, apparently, that it is either unspecified or 'discourse-deleted'), the infinitival 2 can be the 1 of the main clause. The PP in (Eb) has no visible surface 2, since the main clause 2 has itself been passivized. Timberlake then reasons that the internal constraints on Lithuanian passivization themselves would all be satisfied if (Eb) were 'repassivized', with the infinitival 2 then the 1 of a new main clause passive. But *(F) shows such multiple passivization is utterly impossible:
Timberlake infers that it is the 1AEX which prevents a 'second passivization', so that these biclausal passives support the 1AEX. But he notes that in contrast to *(F), the impersonal form of (Eb) is well-formed:
Timberlake concludes from the *(F)/(G) contrast that the 1AEX holds in Lithuanian and thus that the impersonal forms of PPs like (G) are demonstrably not subject to it. This is the case according to Timberlake because the impersonal forms, though passives, are not formed by advancement. In present terms of course, the forms also do not involve advancement,
< previous page
page_189 If you like this book, buy it!
next page >
< previous page
page_190
next page > Page 190
and there is no reason to call them passive or to think they cause problems for the relational assumptions of interest. Timberlake's assumption that *(F) supports the 1AEX is far from obvious at this point. He has not considered the question of how passives like (Ec) can be made consistent with the RG/APG view of passives at all. For, on the face of it, the nominal which ends up as final 1 of these PPs is not a main clause constituent, and hence would not head a 2 arc capable of being a 2 arc local predecessor of an arc-passive arc. Unless this issue is resolved, the biclausal passives far from supporting the 1AEX, constitute a problem for the whole relational framework of passivization. However, there is, of course, a standard way of solving the problem, namely, via appeal to the 'clause reduction' or 'equi clause union' phenomenon discussed for Spanish in Aissen and Perlmutter (1976, 1983), JP (1980: Chapter 8) and for Italian in many other works in different frameworks; see e.g. Rizzi (1982): Chapter 1), Burzio (1981, 1983). I proposed an APG version of such an analysis for the type of Turkish 'infinitival double passives' discussed by George and Kornfilt (1977); see (M) of the Postscript to section 2 of this chapter. Under this view, the P arc of the infinitival subordinate clause has a U arc successor in the main clause, and all of the internally unerased neighbors of that P arc also have main clause successors whose R-signs are determined by fixed principles. In particular, lawfully, the 2 arc of the complement will have a 2 arc foreign successor in the main clause. This successor 2 arc could then serve as the prepassive arc underlying PPs like (Eb) in full consistency with the current view of passives. If, however, this view is adopted, the question of the constraints governing this kind of clause union in Lithuanian arises. Since Timberlake provides no information about this, the matter cannot currently be pursued. But it is evidently possible that constraints on equi clause union would themselves preclude *(F) with no appeal to the 1AEX. There are, to my knowledge, no known cases where equi clause union combines with a main clause passive in a way relevant to Timberlake's discussion, so this possibility is perhaps more than just logical. But if there are constraints of this sort, either universal or Lithuanian-particular, *(F) fails to support the 1AEX. In summary, while Timberlake (1982) provides many interesting and important facts about Lithuanian which suggest various lines of research that deserve to be pursued, the material cited by Timberlake seems to determine no real problems for the analysis already suggested to deal with the facts brought forward in Nerbonne (1982). The claims made by these critics that Lithuanian impersonal forms require that some relational law or assumptions be abandoned have in neither case been supported. 5.5.2 Irish Nerbonne's supposed refutation of NEG-IP from Irish is briefer than that from Lithuanian. It consists of (i) taking the Irish autonomous form to
< previous page
page_190 If you like this book, buy it!
next page >
< previous page
page_191
next page > Page 191
constitute passive morphology, and then (ii) presenting predicates that Nerbonne considers clear universal unaccusatives in the autonomous form. The case is, however, much weaker than that dealt with in Lithuanian. For there is no a priori way to associate the autonomous form with the morphology of PPs, since the morphology of Irish PPs is distinct from the autonomous form. Moreover, the predicates cited are not so clearly universal unaccusatives. I need not pursue this line of discussion, however, because even if they are, nothing brought up by Nerbonne precludes a syntactic analysis for Irish parallel to that just argued for for Lithuanian. There would simply be a different rule determining verbal morphology, one yielding the autonomous form (presumably) only in cases of a stable dummy arc. Nerbonne's remarks about Irish have a feature which is absent from his discussion of Lithuanian, although it is present in Timberlake's remarks, via cases like (Bc) of the postscript to section 5.1 of this chapter. Nerbonne tries to show that Irish counterexemplifies not only NEG-IP but directly refutes the 1AEX independently of unaccusative predicates. The attempted demonstration has the following form. Nerbonne indicates that Irish has, besides the autonomous form, what he calls a genuine passive:
Nerbonne points out that the underlying 2 is clearly marked as final 1 of the passive both by verbal agreement and by its position. Let us acceptas I see no reason not to 26 that [52b], more precisely, the complement of the auxiliary of [52b], is a genuine PP in current terms. Nerbonne's putative refutation of the 1AEX then consists of noting that such PP constructions have their own autonomous forms:
He then concludes (p. 85): The construction is straightforward and regular, involving two advancements to I in a single clause in the analysis in which autonomous forms are derived by 2 1 advancement of a dummy. To back this up, Nerbonne cites (p. 90) a remark by a traditional Irish grammarian a propos of the autonomous form: 'No, it is not passive, for it
< previous page
page_191 If you like this book, buy it!
next page >
< previous page
page_192
next page > Page 192
has a passive of its own.' Of course, more precisely, it is the passive auxiliary in [53] which manifests the autonomous form. Example [53] notwithstanding, nothing indicated by Nerbonne provides any problem for the 1AEX. For, as in earlier cases, he has nowhere argued for or supported the key premise of the argument: that Irish autonomous forms are associated independently with some 21A. The fact that this was assumed in the version of Perlmutter and Postal (1984a), to which Nerbonne had access when writing Nerbonne (1982), is not relevant. The reader will note that the facts brought forward here by Nerbonne with respect to impersonal forms of Irish PPs are quite parallel to the existence of the 'double passives' considered in the postscript to section 5.2 and to the impersonal forms of Lithuanian PPs considered in section 5.5.1 and its postscript. It would appear that the best analysis on the basis of the tiny amount of data considered in the discussion would simply reject the view that Irish autonomous forms involve an advancement and would instead treat them along the lines of our treatment of the non-PP Lithuanian VM forms. That is, they would be taken to involve a stable I arc ghost structure. Without developing some positive argument for associating these forms with 21A or a negative argument against the alternatives, Nerbonne cannot justifiably claim to have provided an argument from Irish against NEG-IP or against the 1AEX or against their conjunction. 5.5.3 Estonian Nerbonne's data from Estonian consists of citing, apparently from questionnaire data, presumed universal unaccusatives in a verbal form called 'impersonal'. No argument is given to associate this form with PPsand whether these exist is not even discussednor is any syntactic reason whatever given for claiming that the impersonals are passives. Such material is too far from offering the basis for a refutation of NEG-IP to require discussion. 5.5.4 Lithuanian and German Medio-passives Nerbonne offers two other bodies of data that supposedly counterexemplify the IAEX; this time, they are examples of what he calls 'medio-passives'. This discussion was evidently motivated by the RG and APG recognition of reflexive passive clauses. But the putative refutation does not really get off the ground. The data in question, which are from Lithuanian and German, are examined below. Turning to the Lithuanian facts first, consider:
< previous page
page_192 If you like this book, buy it!
next page >
< previous page
page_193
next page > Page 193
This is apparently a PP, given the nominative agreement and the initial 2 in the nominative. Nerbonne considers it a counterexample to the 1AEX because he has decided, although without argument or justification, that the derived verbal form prisiminti, which contains an affix -si-, and which is based on the more elemental form priminti 'to remind', is somehow related to an independent 21A. Impersonal forms with the VM structure can also be formed on Vs with this derivational affix. So Nerbonne also takes these to refute the combination of the 1AEX and universal assignment of unaccusative and unergative status. But even if we ignore the earlier discussion, which showed the absence of grounds for taking VM morphology to indicate 21A with intransitives, Nerbonne does not even attempt an argument that what he calls the reflexive affix in Lithuanian is associated (ever, still less systematically) with any sort of advancement to 1. Lacking this though, the discussion of Lithuanian reflexive medio-passives has no bearing on the relevant issues. Consider next the German data. As observed in Perlmutter and Postal (1984a), JP (1980), and elsewhere, German has what appear to be clear RF-PPs and RF-IPs, e.g.:
Nerbonne (1982: 81) refers to [55a] as a 'medio-passive', and the supposed further refutation of NEG-IP is the claim that medio-passives interact in a single clause with non-RF-IPs. The total body of data offered in support of this claim is that found in [56]:
< previous page
page_193 If you like this book, buy it!
next page >
< previous page
page_194
next page > Page 194
Nerbonne's discussion of these examples is very brief and strangely and irrelevantly phrased in part in terms of a Montague grammar rule. It is hence hard to follow the argument he believes they support. One interpretation of Nerbonne's argument is as follows. He refers to genuine RF-PPs like [55a] as 'reflexive mediopassives' and uses the same terminology for [56c,f and g]. Suppose then Nerbonne's argument assumes (A) that the latter examples are genuine RF-PPs in the sense of RG and APG, and thus that they involve an inherent 21A like [55a], with a nondummy nominal advanced. And it is supposed (B), that [56c,f and g] are nonreflexive IPs, also independently associated with a 21A, this time with a dummy advanced. Hence [56c,f and g] are, under this interpretation, apparently directly inconsistent with the 1AEX. If both (A) and (B) hold, the 1AEX is indeed falsified by the well-formedness of [56c,f and g]. I certainly accept assumption (B). These examples seem to have the properties of German nonreflexive IPs. However, I see no reason to accept (A), for which Nerbonne provides no argument at all. One might think that the analysis is supported by the meaning of [56b and e]. But this is anything but clear. Consider [56b]. If this is an RF-PP, then it should have a transitive meaning, roughly 'They are gathered by people', just as [55a] means 'Such things aren't said. by people'. Nerbonne, however, gives as a translation something quite different, namely, 'They gather', which then he annotates by saying: '... not "They gather themselves" but "Something or someone gathers them".' I see no reason to accept this. 27 Moreover, it is not really relevant since Nerbonne correctly translates the corresponding IP with an exclusively intransitive meaning. If, as the argument assumes, [56c] were the IP of an RF-PP, it would mean 'People should now be gathered by UNSPECIFIED'. The other examples are open to the same objection. It follows that this interpretation of Nerbonne's argument yields no reason for taking the so-called German medio-passives to be RF-
< previous page
page_194 If you like this book, buy it!
next page >
< previous page
page_195
next page > Page 195
PPs, or indeed passives of any sort. Hence there is no argument taking such examples as [56c] to involve double passivization which attacks the 1AEX. In connection with the possibility that the German medio-passives are RF-PPs, Nerbonne has not considered a distinction for German of the sort argued in detail for probably parallel cases in French by Ruwet (1972: Chapter 3). Ruwet distinguished in many ways two sets of reflexive constructions, one which he called 'neuters', the other 'middles'. Only the latter are passives (RF-PPs). My guess is that Nerbonne's medio-passives are actually the same sort of clause as Ruwet's neuters. It is, evidently, an important question how to analyze such clauses, but not one which has been shown to this point to have a clear bearing on the 1AEX. There is another interpretation of Nerbonne's argument from the German medio-passives, and a personal communication (12/5/1983) makes clear that it is this interpretation that he actually had in mind. Under this second interpretation, the medio-passives like [56b,e] are not interpreted as RF-PPs but as unaccusative advancement structures. This is consistent with my suggestion of a parallelism between German medio-passives and French neuters, since the latter are demonstrably not RF-passives (most are unaccusative structures). Since the unaccusative analysis also takes German medio-passives to involve an instance of 21A, the argument from the nonreflexive IPs goes through. For these would then involve two advancements to 1 in the same clause, one the unaccusative advancement of the initial 2, the other that of the dummy 2 determining the nonreflexive IP structure. The trouble with this attack on the 1AEX is that no argument has been given to justify the claim that the relevant nonIP clauses involve any instance of 21A. Consider [56g], which corresponds to an active medio-passive something like the semantically rather strange [57]:
The view that [57] is an unaccusative advancement structure apparently is justifiable as follows:
But I do not think one can conclude on the basis of data so far studied that [58c] is required by [58a and b], even given a universal assignment of
< previous page
page_195 If you like this book, buy it!
next page >
< previous page
page_196
next page > Page 196
nominals to initial 1 or 2 based on such notions as 'annoyee'. [58c] is only required if the 2s in the clauses of the [58b] type are initial 2s. This is suspicious. For in general it is initial 1s that are required by fixed Vs to refer to 'mindpossessing' entities. But this property is associated with the 2 in [58b]. An available alternative then is that the transitives of relevance to [58b] are not basic but involve causatives of an embedded intransitive annoy in which the required mind-possessor is referred to by an initial 1 not an initial 2. This amounts to saying that the proper analysis of triples like [59] takes [59b] to be most basic and takes the 'annoyee' to be an initial 1 in all three, although the initial 1 of a nonsurface clause in [59a]:
Unless an analysis of this sort can be countered, which, notably, Nerbonne has not attempted, there is no justification for the view that German active clauses like [57] involve any instance of 21A. It is, I believe, significant that all the German medio-passives cited by Nerbonne involve human final l s, and, moreover, the sort of final 1 s capable of being interpreted as initial (unergative) 1s, since their cognates in other NLs are such. Relevant then would be medio-passives corresponding to actives whose final 1 is inanimate, where the predicates correspond to uncontroversial unaccusatives in other NLs. The prediction would then be made that, contrary to the cases in [56], the corresponding (nonreflexive) IPs are impossible. Consider then the following facts kindly provided by my colleague E. Mueckstein:
The key point is that [60b] is only interpretable as referring to people, and cannot refer to the gathering of water. No IP corresponds to [60a]. Hence when a medio-passive is rather clearly based on an unaccusative, far from being refuted, the prediction of the 1AEX and the universal assignment of unaccusative status that no IP is possible is supported. The same point holds
< previous page
page_196 If you like this book, buy it!
next page >
< previous page
page_197
next page > Page 197
for [60c]. Here the corresponding IP is apparently ill-formed, except, possibly on a metaphorical meaning where it refers to appointments. In any event, [60d] is unrelated to [60c]. Hence in the few cases I have been able to look into where medio-passives are based on uncontroversial unaccusative predicates, they are not, as the 1AEX predicts, subject to formation of corresponding IPs, given the RG/APG view of IPs as involving 21A. Curme (1952: 338), cited in Perlmutter and Postal (1984a), notes the existence for some speakers of IPs of mediopassives on the same page where he claims that IPs can '... only be used with intransitives which express an activity or condition that stands in relation to a free moral agent'. Curme's terminology would seem equivalent to a modern usage which permits IPs only with Vs taking experiencer or agent nominals. But these are all members of the class of universal unergatives. It would seem then that if Curme was right, one will find no German IPs of medio-passives based on unaccusatives, supporting the 1AEX. Clearly though, there is a need for detailed study of the range of medio-passives (and other clauses) which permit corresponding IPs in German. I have considered two interpretations of the claim that the IPs of certain German medio-passives falsify the 1AEX. Under one interpretation, the medio-passives are RF-PPs. This analysis seems incompatible with their meanings and has no independent justification. Under the other analysis, the medio-passives involve unaccusative 21A. But this analysis has not been motivated in the case of those medio-passives permitting corresponding IPs for some German speakers. 28 For other medio-passives, specifically those involving inanimate final 1s, a claim of unaccusative status and 21A independent of that of the IPs seems justified. But in this case, according to the little data so far available and according to Curme's claim, a corresponding IP is impossible even for a speaker who accepts the IPs of the mediopassives with human 1s. Consequently, for the former medio-passives, there is no argument against the 1AEX and for the latter, the available evidence actually supports it. Overall then, Nerbonne has not grounded any argument from German medio-passives against the 1AEX or against any relational assumption underlying the current universal view of passive clauses in NLs, or against the currently maintained conjunction of these principles. To conclude, it seems that none of Nerbonne's data, neither that from Lithuanian, nor that from Irish, nor that from Estonian nor that from German, supports a refutation of either NEGIP or the 1AEX, contrary to his conclusions. 5.6 Summary Although I have been critical of the discussions of Özkaragöz, Ostler, Nerbonne and Timberlake, such attempted refutations of the 1AEX, of
< previous page
page_197 If you like this book, buy it!
next page >
< previous page
page_198
next page > Page 198
universal assignment of unaccusative status, etc. should be encouraged. It is only by confronting universal claims about passive structures with the most difficult and problematic cases that one can gain confidence in their validity or, to the contrary, show they are false. The fact, since I believe it has been shown to be a fact, that no data have yet been brought forward from Turkish, Sanskrit, Lithuanian, Irish or Estonian which require abandoning any of the claims in Chapter 2, section 2 is encouraging, though hardly definitive, especially in view of problems with the universal assignment of unaccusative status of the sort considered in Rosen (1981b), Perlmutter (1982), etc. Really needed are extensive studies which are concerned with the scope of genuine IPs in as many diverse NLs as possible. Consider, for example, the claims made in section 5.1 concerning the incompatibility between true IPs and predicate nominals. In how many NLs would it currently be possible to test practically such a claim on the basis of the available literature? Unfortunately, in hardly any. But, as the discussion of this chapter should make clear, evaluation of abstract principles like the 1AEX and the universal assignment of unaccusative status cannot adequately be based on facile or superficial analyses. Without detailed attempts to propose and justify syntactic analyses, it will be impossible to really undermine (or, of course, support) such principles. 5.7 A Note on Italian NEG-IP is, as a truly substantive principle, based on the possibility of being able to identify unaccusative predicates independently of their susceptibility to IP formation. The most general way to determine this identification is via the universal assignment of intransitive predicates to the disjoint categories of unaccusative and unergative. I have already mentioned recent work which casts doubt on the feasibility of a fully general and unconditional universal assignment. Perhaps the strongest argument against this is found in recent work on Italian by Perlmutter (to appear a) and Rosen (1981 a, 1982), which offers a novel type of argument against such assignment. Roughly, the argument has this form. The choice of past tense auxiliary can be shown to be based on a condition sensitive to the presence of 2 arcs. Clauses with unaccusative advancement in particular regularly determine the auxiliary essere. However, the Italian-internal assignment of predicates needed to be consistent with the general auxiliary-choice rule conflicts in both directions with universal assignment of predicates to the ergative and unaccusative categories. Some of what would be universal unaccusatives behave like unergatives with respect to auxiliary choice, while some of what would be universal unergatives behave like unaccusatives with respect to auxiliary choice. I think it clear then that the following has been shown. If the proper rule for auxiliary choice is that sketched in Perlmutter (to appear a), the universal
< previous page
page_198 If you like this book, buy it!
next page >
< previous page
page_199
next page > Page 199
assignment of intransitive predicates to the categories unaccusative and unergative cannot be maintained. However, to turn this conditional into a true refutation of universal assignment, one must accept the truth of the offered auxiliary choice rule. While I believe this rule incorporates important insights, I am unconvinced that it is literally correct. And I note that its acceptance as is entails certain extremely undesirable conclusions, e.g., the necessity to recognize two distinct formal representations for 'coreference', one the overlapping arc approach of JP (1980), the other some sort of indexing, as in more standard approaches. This alone is, I believe, reason enough to seek for an alternative to the rule proposed in Perlmutter (to appear a). Moreover, the rule in Perlmutter (to appear a) has not been shown to correctly deal with auxiliary choice in sentences involving so-called equi clause union. In any event, the issue is conceptually relatively clear. Anyone wishing to maintain a reasonably strong form of the universal assignment hypothesis is committed inter alia to finding an alternative rule of auxiliary choice for Italian. Notes: Chapter 5 1. This analysis requires abandoning at least one sentence law from the system of JP (1980), namely, PN Law 85. In current terms, this requires that the arc overrun by a stable ghost, that is, its sponsor, have an 8 arc local successor. Fortunately, this law was only weakly motivated, and stood because of a lack of knowledge of attested cases properly analyzable as counter-examples. It played no role in explaining any attested facts. 2. The notion Basic Clause is defined precisely in JP (1980). Roughly, it is a clause containing a P arc. 3. Such evidence is findable at least in principle. The postposition tarafindan seems to uniquely mark the 8s of passive clauses. See Kornfilt (1976), George and Kornfilt (1977). 4. Depending on how this claim is made precise, it would in all probability be incompatible with the claims of McCawley (1982) that NLs contain surface representations with discontinuous constituents. To the extent that McCawley's claim is true, an approach to word order along the lines sketched in Chapter 1, section 4.4 could not be fully successful. Such an approach depends on ordering exclusively surface sisters. 5. JP (1980: Chapter 14, section 5) presents a detailed discussion of how the Unique Eraser Law blocks a variety of cases, doing much work otherwise attributed to rule ordering, cycles, etc. 6. There is at least one problem with (Rbii), revealed by the following fact, already noted in Knecht (1974: 25) and helpfully verified by I. Özkaragöz (personal communication 2/3/84). Turkish pronominal 1 s can be invisible. But regardless of whether the pronominal 1 of a relative clause is visible or not, the relative requires the OP form. Thus Knecht (1974: 25) gives the following examples:
< previous page
page_199 If you like this book, buy it!
next page >
< previous page
page_200
next page > Page 200
*(ic) is ill-formed because it wrongly contains the SP form. The correct inflection of such relatives is indicated in (ii), from Özkaragöz (personal communication 2/3/84):
This is not predicted by (Rbii) under the standard APG analysis of invisible pronominal cases of this sort, which takes them to involve self-erased arcs. Hence in these terms, the relative clause of (i) would have a self-erased 1 arc, and the 'relativized' constituent would not be an R-branch of a neighbor of a surface I arc. One can, I suspect, deal with this situation by slightly weakening (Rbii) to replace the reference to 'surface' with that to the notion Internal Survivor of JP (1980: 526). Certain currently irrelevant refinements aside, Internal Survivor is a relation between an arc A and some constituent (set of arcs meeting fixed conditions) X, where A has no assassin (distinct eraser) internal to X. Crucially then, self-erased arcs fall together with unerased arcs as internal survivors of constituents since, given the Unique Eraser Law, no self-erased arc has an assassin. Principle (Rbi) could then be reformulated to keep all of its previously discussed consequences and yet still maintain consistency with facts like (i) and (ii) if it required the 1 arc not to be a surface arc but only an internal survivor of the relative clause structure. 7. Presumably, the necessity of the auxiliary be in the cases is not an independent fact but, as with true passives, can be taken as predictable from the morphology of the subordinate V. Note that just like true passive participles, born can occur in certain environments distinct from the complement of be.
8. As in the analogous English cases, I assume that the auxiliary V obligatorily determines raising of the final 1 of its complement (the passive clause per se), this raising being of the copy sort. Some further features determine that this copy ends up genitive; see the discussion in Perlmutter and Postal (1984b). 9. There is no doubt a generalization here which links the IP morphology to the advancement of a dummy 2 to 1; but it is unclear how to state it precisely. 10. More precisely in APG terms, the initial 2 arc and its I arc local successor cosponsor a copy 2 arc, a replacer of, and hence eraser of, the initial 2 arc; see Chapter 4, section 6. French grammar must distinguish among unaccusative Vs those which require such copy arcs, those which allow but do not require them and those which preclude them altogether. 11. This conclusion is apparently taken as demonstrated by Dowty (1982b). 12. I now think wrongly so, since, as indicated in e.g. [4.11], I assume that the ghost arc defining the EXI construction is a 2 arc sponsored by a 2 arc. 13. Not all French speakers accept the extraposed version of [25]. My current consultant very emphatically does not. 14. The star on [28c] refers to the intransitive use of passer. On a transitive use, [28c] has a well-formed passive meaning 'Two trains were passed to unspecified by unspecified'. In this case, [28c] is an EXI structure like [4.11].
< previous page
page_200 If you like this book, buy it!
next page >
< previous page
page_201
next page > Page 201
15. This claim is not falsified by the fact, indicated in note 6 of Chapter 1, that some embedded French passive clauses do not permit this morphology. 16. This ignores the uses of the past participle morphology for the distinct function of marking past tense with certain classes of verbal elements. 17. I am indebted to my colleague E. Mueckstein for providing the German data in [39]. 18. Of course, I am interpreting *[40b] as a true IPthat is, with a structure like [4.13]and not as an errant EXI structure like [4.11]. On the latter reading, *[40b] would be ill-formed simply because the EXI nominal is illegally definite. 19. Most notably by Burzio (1981: 210-211), who states: If we assume the syntax of French en essentially analogous to that of Italian ne as would seem independently plausible, namely if we assume that, like ne, en cliticizes only from direct object positions, then the i-subjects in [46] here below must be in such a position . . . .
Thus, contrary to our analysis, in which plusieurs heads a final 8 arc, Burzio considers it a surface 2. Crucially though, Burzio's claim that such en require sources which are surface 2s offers no basis for text cases like [43]. Perhaps Burzio has been misled here by the analogy with Italian ne, which is quite possibly only partial. For in briefly attempting to check analogs of examples like [43] with one Italian informant, I was unable to find any wellformed structures like [43]. 20. Thus the nominal certaines bouteilles of [4.8b] represented in [4.11] is a final 8 but heads an acting 2 arc. Such nominals can then predictably be the 'source' for en. 21. An intriguing possibility is that such a principle might explain in APG terms the impossibility of a 'coreferential' pronoun as predicate nominal:
In APG terms (see JP (1980: Chapter 11) and note 13 of Chapter 2), 'coreference' involves pairs of initial overlapping arcs. This idea combines with the analysis of predicate nominals as involving P arc local successors of 2 arcs to determine that ill-formed structures like (i) and (ii) would involve a distinct 2 arc neighbor of the P arc local successor. For that 2 arc would be the pronominal arc replacer of the initial 2 arc representing the 'coreference'. 22. See note 13. 23. That is (see Postal (1982)):
Certain lexically limited marginal apparent counterexamples to the impossibility of cases like *(ib) and *(id) are discussed in Kayne (1978), Pollock (1981). 24. There is, however, a problem for our account with respect to paradigms like [50]. Namely, preposition-marked predicate nominals are sharply incompatible with en:
< previous page
page_201 If you like this book, buy it!
next page >
< previous page
page_202
next page > Page 202
This follows from nothing indicated so far. It is possible that the incompatibility of English pseudo-passives (see Chapter 6) with predicate nominals also supports principle [49]:
Compare (iv), which involves a non-predicate nominal with the same preposition:
*(iiib) thus follows from no standard assumptions. But under the analysis of pseudo-passives in Chapter 6, *(iiib) does follow from principle [49], since the 2 arc predecessor of the P arc defining the predicate nominal would cooccur with the distinct prepassive 2 arc defining the clause as a passive. 25. A potential problem for principle [49] is raised by examples like (i) cited in Bach (1980: 302): It seems that us is a final 2 and it might well be argued that it is an earlier benefactive. Since a good chairperson is a predicate nominal, it might be claimed that (i) involves a later 2 arc than that underlying the predicate nominal. A possible way out is to regard make, like the auxiliary be, as defining a higher clause than that of the predicate nominal proper. The 2 arc local successor is then an element of that clause, consistent with [49]. 26. McCloskey (1979: 141) claims that Irish does not have a transformational passive, a conclusion with no bearing on current assumptions. 27. And I am indebted to my colleague E. Mueckstein for verifying that structures like [56b] have only an intransitive meaning. 28. E. Mueckstein does accept them, but observes as does Nerbonne, that many speakers do not.
< previous page
page_202 If you like this book, buy it!
next page >
< previous page
page_203
next page > Page 203
Chapter 6 English Pseudopassives 6.1 Background The English structures referred to as prepositional passives or pseudo-passives (Ps-Ps hereafter) are cases where it appears that the object of some preposition ends up as the 1 of a corresponding passive (but see below). Hence, corresponding to actives [la-g] are the Ps-Ps [2a-g]:
< previous page
page_203 If you like this book, buy it!
next page >
< previous page
page_204
next page > Page 204
The grounds on which the examples of [2a-g] are passives, that is, clauses properly classed with the uncontroversial PPs of [2h and i], are straightforward. The Ps-Ps share the past participle/auxiliary be morphology of [2h and i] and utilize the same prepositional phrase structure in by for marking the nominal understood as the initial 1. Moreover, the Ps-Ps have the same interpretation properties of PPs. The pre-by nomiual is understood as the initial 1 of the corresponding active while the final 1 of a Ps-P is understood as the object of the stranded preposition of the corresponding active. In fact, other than their relation to prepositional structures, there is no reason whatever to doubt that Ps-Ps are passives. Viewed from many points of view, Ps-Ps seem odd or anomalous. The reason is that in general the final I of an English PP, as is also obviously the case in many other NLs, correlates with the 2 of a corresponding active clause, that is, with the immediately post-V nominal not marked with a preposition. But as just noted, this is not the case with Ps-Ps. In terms of transformational grammar, the problem of Ps-P description is how to allow the NP movement rule that is assumed in transformational terms to apply in passives 1 to apply to prepositional objects specifically like those in [1], but not to any random prepositional objects. A parallel issue arises in the nontransformational lexical/functional framework of Bresnan (1982), where the issue is discussed in some detail; see section 3 below. A certain approach has emerged as an apparently widely agreed-on part of the 'solution' to Ps-P description in both common transformational and nontransformational frameworks; see Chomsky (1975: 242, fn 43; 1981: 123, 292-300), van Riemsdyk (1978: 215ff), Hornstein and Weinberg (1981), Williams (1980: 204), Bresnan (1982a), Johnson (1979). This is the idea that Ps-Ps are possible because some sort of grammatical mechanism can turn English V + Prepositional Phrase structures into those in which the V and preposition form a complex V with the original prepositional object as, in effect, the 2 of that V. This mechanism could then be claimed to be a peculiarity of English but, notably, to be missing in e.g. German or French, which lack analogs of Ps-Ps. Compare, for example the English and French cases in [3]:
< previous page
page_204 If you like this book, buy it!
next page >
< previous page
page_205
next page > Page 205
However, the idea that Ps-Ps are possible because certain Vs and prepositions can be reanalyzed as single complex Vs essentially misses the point. This is so even if, as is likely, some Ps-Ps do, or at least can, involve, the incorporation of the stranded preposition in the V complex. I argue that even if such incorporation is a real phenomenon, it is only marginallyand in a sense accidentallyconnected to the existence of Ps-Ps. Suppose, following such works as Johnson (1979), Hornstein and Weinberg (1981) and others, that Ps-Ps are possible because English has a 'reanalysis' rule which in essence converts a V plus preposition into a 'complex V'. The 'reanalysis' rule is intended to have the effect in [4]:
Structure [4b] has the putative form of an ordinary transitive structure. Under such assumptions, Ps-Ps can then presumably be characterized by the standard mechanisms operating for typical transitive structures, however these are conceived. An implication of this approach, if we assume that 'reanalysis' is optional, is that an example like [5a], which has a Ps-P correspondent [5b], has both the structures in [5c,d], while an example like [5e], which has no Ps-P correspondent, would lack the structure corresponding to [5d]:
This consequence is made quite explicit in Bresnan (1982a: 52). Moreover, it is hard to see how any proponent of 'reanalysis' could take it to be anything but optional since, obviously, in many ways the prepositional phrase in cases like [5a] behaves as a unit:
There are, of course, many questions about the way in which 'reanalysis' is to be formulated in various frameworks. However, it is unnecessary to consider these here. Rather, one can concentrate on the core 'reanalysis' view that the existence of Ps-Ps depends on the existence of active clause
< previous page
page_205 If you like this book, buy it!
next page >
< previous page
page_206
next page > Page 206
structures like [5d]. This makes a basic and quite general claim: any grammatical constraints which treat postverbal prepositional phrases differently from postverbal 2s. should be able to treat the: objects of the apparent prepositional phrases of actives like [5a] as postverbal 2 nominals. But in at least five distinct domains this prediction is directly contradicted by the evidence. First, an optional 'reanalysis' rule makes an easily testable prediction for varieties of English, like my own, in which generalization [7] holds:
Principle [7] accounts for such facts as [8]:
It follows that the existence of an optional 'reanalysis' rule predicts, mutatis mutandis, that in just those actives corresponding to well-formed Ps-Ps, the negative should be able to follow the preposition. But this is uniformly impossible, and in each case the negative can only be positioned as with other ordinary prepositional phrases:
Second, consider the distribution of only in cases like [10]:
< previous page
page_206 If you like this book, buy it!
next page >
< previous page
page_207
next page > Page 207
It seems that with a proper nominal, postverbal only can precede a nominal, or a prepositional phrase, but not a nominal inside of a prepositional phrase. 3 But this is just as true in the structures which would have to undergo 'reanalysis' to account for Ps-Ps:
Again the constituent structure posited by 'reanalysis' is not only not supported but disconfirmed. Third, Ross (1967: 226-227) observed that, in transformational terms, the object of a preposition is not subject to rightward movement, e.g. under Heavy NP Shift. But of course 2s are. A 'reanalysis' approach then predicts that those verbal forms which permit Ps-Ps should allow Heavy NP Shift of the same nominal which can passivize. But consider [12]:
Again, the constituent structure posited by 'reanalysis' fails to manifest and the apparent prepositional phrases behave only like prepositional phrases. Fourth, there is an argument to the same effect, based on the gapping phenomenon and which is due essentially to Ross (1981). The general case is that gapping can lead to the absence of a V in a coordinate clause to the right of an antecedent clause, but it cannot in general determine the disappearance of the preposition of a prepositional phrase:
< previous page
page_207 If you like this book, buy it!
next page >
< previous page
page_208
next page > Page 208
However, since 'reanalysis' claims that the strandable prepositions of Ps-Ps are incorporated in Vs even in actives, no non-ad hoc constraint blocks the absence of just these prepositions in gapping structures. Butt, as Ross (1981) observed, such absence is impossible for active clause prepositions corresponding to the strandable prepositions of Ps-Ps:
Fifth, an argument similar to the preceding one can be based on the following slightly marginal 'pseudo-gapping' phenomenon discussed in Levin (1978).
[15a] is not a case of gapping, which, as is well-known, cannot 'strand' an auxiliary like will. As [15b] shows, 'pseudogapping' cannot delete the preposition of a prepositional phrase, though it does, of course, delete Vs. A 'reanalysis' approach to Ps-Ps then should predict that 'pseudo-gapping' should be able to delete just those active clause prepositions which correspond to the strandable prepositions of Ps-Ps. But this is not possible:
The arguments given against 'reanalysis' reveal a single general fact. The preposition of active clauses that corresponds to the stranded preposition of Ps-Ps behaves like the head of a prepositional phrase that is not like part of a V. Thus the claim that Ps-P existence depends on the sort of constituent structure brought about in some way by 'reanalyzing' V + preposition structures as compound Vs is in conflict with the syntactic behavior of the relevant phrases in at least five distinct respects. This evidence against 'reanalysis' obviously has to be weighed against arguments that have been given in favor of it. I consider below three basic arguments offered by Bresnan (1982a). But independently of such comparisons, it has already been
< previous page
page_208 If you like this book, buy it!
next page >
< previous page
page_209
next page > Page 209
shown that it is, to say the least, reasonable to attempt to develop an account of Ps-Ps which is independent of 'reanalysis' or any analogous assumptions. Such an account in APG terms is sketched in the following section. 6.2 An Arc Pair Grammar Analysis of Pseudopassives From the APG point of view, the key issue about Ps-Ps is to specify where the requisite prepassive arc that is claimed to underly all true passive clauses is to be found. In what sense does a prepassive arc exist in a Ps-P? This question can be answered by taking Ps-Ps to involve advancement to 2 of various nominals which head 3 arcs, oblique arcs, etc.; this advancement has, however, two special, though far from unique, properties. The most obvious special property is that there is no instantiation of the putative advancement structure in which the advanced nominal behaves like a surface 2. Consider [17]:
Without further constraints, positing advancement to 2 would predict the existence of forms like *[17b], in which the nominal supposedly advanced to 2 manifests itself as a superficial 2. Suppose, contrary to fact, that 'intermediate' examples like *[17b] were systematically well-formed. Then there would be no difficulty in accounting for Ps-Ps like [17c]. These would just be the quite regular PPs corresponding to [17b]. The problem of Ps-Ps would then be seen not to strictly concern passives at all, but rather to involve essentially the description of structures like *[17b] and their relation to those like [17a]. In an abstract way, the proposal to relate [17c] to a hypothetical structure like [17b] is partially similar to the appeal to 'reanalysis'. But it also differs sharply. Specifically, it does not claim that [17a] is structurally ambiguous and does not require (but also not preclude; see below) that cases like [17c] involve a compound V or an incorporated preposition. One can, and, I claim should, look upon the issue of Ps-Ps in just the way depicted in [17]. For the ill-formedness of the intermediate structures can easily be accounted for, by obligatory linkage of the posited advancement to 2 to a subsequent advancement to 1, thereby determining a passive (see below). Adopting this approach, the ill-formedness of the 'intermediate' active forms like [17b] is no more a problem for the advancement to 2 analysis of Ps-Ps than the illformedness of the 'intermediate' form *[18b] is a problem for a successive cyclic transformational treatment of forms like [18c] in which nominals move only to the comps of their own clause:
< previous page
page_209 If you like this book, buy it!
next page >
< previous page
page_210
next page > Page 210
I conclude that the proper way to think of Ps-Ps is that they involve advancement of various nominals to 2, thereby permitting passives in the standard way. This advancement, however, is masked by being linked to a necessary subsequent advancement to 1. Such masking is not novel; Chapter 2, section 3.2 dealt with a partially parallel situation from French where 32A with the Vs (dés)obéir is masked by a requirement of the existence of a 1 arc R-successor for the resulting local successor 2 arc. One can then concentrate on the analysis of the structures of clauses like *[17b], which, by hypothesis, exist in effect as aspects of Ps-P clauses. These manifest unusual features not accounted for by the advancement to 1 which masks the prior advancement to 2. The special nature of the advancement characteristic of Ps-Ps can best be seen by comparing the 32A typical of a Ps-P structure like [17c] with the more ordinary 32A found in many English cases like [19b]:
These have the sort of structures already considered briefly in Chapter 2; see [2.18], [2.19], [2.20], [2.56], [2.75] and [2.96b]. What is unusual about the posited 32A in a Ps-P like [17c] is the presence of the stranded preposition, which is not only not characteristic of 32A cases like [19b] but always impossible:
The present hypothesis about Ps-Ps thus requires recognition of two contrasting types of 32A. But this is no surprise, since the APG framework allows, in effect, two types of advancement, as already touched on in the discussion of reflexive passives in Chapter 4, section 6. One is the plain kind found with Vs like forward, the other the copy kind found by hypothesis in Ps-Ps. The contrast involves whether or not the local successor/predecessor pair defining the advancement cosponsor a copy arc or not. As indicated in Chapter 4, section 6, a copy arc (see JP (1980: Chapter 11)) is a replacer, an arc with two sponsors which is a facsimile of one of its sponsors (the arc it replaces), which it lawfully erases. Hence 32A involving a copy arc is a structure like [21]:
< previous page
page_210 If you like this book, buy it!
next page >
< previous page
page_211
next page > Page 211
In these terms then, the first special property of Ps-Ps is that they involve advancement to 2 masked by a linkage to advancement to 1; the second special property is that the advancement to 2 in question is of the copy variety. Even this limited level of analysis permits a simple characterization of the difference between English and NLs like French and German, which lack both Ps-Ps, and actives like *[17b]. One simply specifies that the grammars of the latter NLs block copy advancement to 2. That is, these grammars contain either a rule like [22] or still more general principles which entail it: 4,5
While English does not have a rule with the generality of [22], it does, of course, on present assumptions need to control copy advancement to 2. It can do so by minimally imposing the linkage to 1 advancement to allow [17c] while not allowing *[17b]. The relevant constraint can be initially formulated as [23]:
< previous page
page_211 If you like this book, buy it!
next page >
< previous page
page_212
next page > Page 212
The second conjunct of [23] restricts the advancement to 1 masking condition to copy cases, and thus correctly does not block noncopy advancement to 2 active clauses like [19b]. Note that [23] does not specify the R-sign of B, correctly claiming that the properties specified hold for all Ps-Ps regardless of the R-sign of the copied predecessor arc. The hypothesis that Ps-Ps involve copy advancement to 2 does not account for all the properties of posited structures like [17b]. The key interrelated questions are why there are prepositions in such structures at all, and why these are stranded. That these matters are independent of the condition which masks the advancement to 2 is shown by comparison with the French (dés)obéir constructions treated in Chapter 2, section 3.2. For although those were taken to involve 32A masked by a requirement of 1 arc-R-successors, no prepositions occur in the resulting good sentences when the relevant R-successor is an arc-passive arc. Naturally, one will account for the contrast between e.g. English Ps-Ps and French (dés)obéir passives, that is, the contrast between [24b] and [24d], by invoking the assumption that the 32A in [24d] is the copy variety, while that in [24b] is not:
This requires a slight extension of the French rule given earlier to specify that French 32A is of the noncopy variety. But even the recognition of copy advancement to 2 in English Ps-P cases like [17c] still leaves partially unanswered the question of why copy advancement yields prepositions. The reason is that arcs of the sort which allow copy advancement to 2that is, copied local predecessors of 2 arcsare members of an English arc class which must, quite independently of what one says about Ps-Ps, define prepositional phrases under specified conditions. For example, under conditions X, a 3 are requires branches defining a prepositional phrase, as illustrated by ordinary active structures like [25]:
< previous page
page_212 If you like this book, buy it!
next page >
< previous page
page_213
next page > Page 213
If the copy arcs involved in the advancement to 2 underlying Ps-Ps are 3 arcs, and if they satisfy conditions X, then they must define prepositional phrases for the same reason as the 3 arcs in [25]; similarly for other arc types serving as local predecessors of copy advancement to 2. What are the conditions X? The simplest suggestion, and the one tentatively adopted here, is that they involve the property of being a surface arc, that is, of having no eraser. The constraint for English is, I believe, that a broad class of surface arcs including 3 arcs and most oblique arcs require branches defining prepositional phrases. 6 One can state the needed constraint more precisely by appealing to the basic APG account of the structure of prepositional phrases.7 This can be sketched via a representation of e.g. [25b], which would be [26]:
The basic ideas are these. Certain arcs, including 3 arcs, are members of a class which, if surface arcs, must be MARQ CLOSURES. The latter concept designates arcs like D in [26], which are replacers having Marq arc branches as one of their two sponsors. Thus a structure essentially identical to [26] except that everything to the right of arc C is eliminated would be ill-formed in English since C would then be a surface 3 arc which is not a marq closure. Prepositional phrases in these terms thus involve the relations Marquee (R-sign = Marq) and Flag (R-sign = F) representing a structure like that in [26]. Specifically, there is some arc like C which has a Marq arc foreign successor and which also sponsors the F arc neighbor of that successor. As a result of the existence of the replacer 3 arc, the earlier 3 arc is lawfully erased. Cs head, which is initially a clausal constituent, thus ends up as a surface prepositional phrase constituent, since the Marq arc is not erased but the
< previous page
page_213 If you like this book, buy it!
next page >
< previous page
page_214
next page > Page 214
clausal arc, C, is. Correspondingly, the whole prepositional phrase structure is a surface clausal constituent, since arc D is not erased. In these terms, the relevant general constraint requiring prepositions in English might be something like [27]:
Left open is the range of arcs over which R varies, a question which need not concern us here, except for one proviso. All local predecessors defining the copy advancement to 2 of Ps-Ps are subsumed by R. The restriction to surface arc determines inter alia that self-erasing arcs and. arcs with (non-marq closure) successors will not determine prepositional phrase structures. Although this is not strictly relevant, it is possible that the: converse of [27] without the specification 'R = ' is a grammatical law. This 'would mean that there are no invisible prepositional phrases. By the definitions of the two concepts, copy arcs are not marq closures. Therefore, given [27], the copy R arc defining the Ps-P structure cannot be a surface arc. This can be systematically avoided if it is the predecessor of a Marq arc, yielding a structure which satisfies both rules [23] and [27]. The structure so far determined for a typical Ps-P like [24d] is thus like [28], with the auxiliary structure and by phrase artificially suppressed:
< previous page
page_214 If you like this book, buy it!
next page >
< previous page
page_215
next page > Page 215
Here C is the initial 3 arc, B its 2 arc local successor, E the copy arc, G the marq closure replacer of the copy arc. Both rules [23] and [27] are satisfied. Ps-Ps were initially described here, as is in effect both traditional and standard, as involving structures in which the object of a preposition apparently ends up as the 1 of a passive clause. Clearly though, internal to the analysis developed here, this informal account is not correct. The nominal which is the final 1 of a Ps-P is, in current terms, not at any level of structure the object of a prepositional phrase. More precisely, it is not the head of a Marq arc. Only the invisible copy pronoun has this property. This is clear in [28], where Peter heads no Marq arc and is entirely independent of the prepositional phrase structure, which is built around the copy pronoun. Representation [28] specifies the structural property that the Marq arc self-erases. This is intended to account for the invisibility of the otherwise expected copy pronoun, which is, of course, uniformly required. But while [28] contains this, no grammatical statement so far ensures it. This lacunae can be filled by replacing rule [27] with the more articulated [29]:
The new specifications thus allow [30a], but not [30b]:
While I have not provided anything like a complete description of Ps-Ps, I have, with one exception, sketched the basic structure of this class of clauses in sufficient detail for present purposes. A notable fact about Ps-Ps, illustrated by the case of 32A, is that they involve the kind of advancement to 2 found in [3lb] and never that found in [31c], as indicated by [31d and e]; see Chomsky (1965: 104):
That is, no visible nominal corresponding to the head of an earlier 2 arc is possible in Ps-Ps. 8,9 Stated informally, PsPs cannot correspond to active transitive structures. In APG terms though, [31d] involves an invisible unspecified nominal. The difference between the two advancement types in current terms is that the later 2 arc in cases like [31d], whose head is Melvin, erases the earlier 2 arc it overruns, while in cases like *[31e], it does not. Rather, that 2 arc has an 8 arc local successor (which erases it). The former
< previous page
page_215 If you like this book, buy it!
next page >
< previous page
page_216
next page > Page 216
type of erasure is claimed (see JP (1980: Chapter 9)) to only be possible when the overrun arc in question is headed by an unspecified nominal. One can then build the requisite restriction into the grammar by expanding rule [29] to [33], via the term defined in [32]: 10
I refine rule [33] further below and then consider ways to simplify it and to reduce some of its features to independent principles of English grammar. 6.3 Arguments for 'Reanalysis' Having sketched the outlines of an APG analysis of Ps-Ps, I return to the question of the viability of a 'reanalysis' approach to such structures, and, in particular, consider Bresnan's (1982a) arguments for a version of 'reanalysis' formulated in terms of her lexical/functional framework. Bresnan (1982a: 51) proposes that there is a rule of V-P Incorporation underlying Ps-Ps, a lexical rule which incorporates a preposition into a V, yielding a single complex V that governs a direct object identical to the original object of the incorporated preposition. The rule is [34]:
Bresnan (1982a: 51) cites three arguments favoring the view that [34] is the correct account of Ps-Ps. Let us consider the third argument first. It is based on the claim that this approach helps explain the contrast between the pairs in [35]:
Without considering the internals of the argument, it cannot show what is assumed, since the supposed contrast is dialect-particular. For me, [35bii] is also ill-formed, although my dialect of course has a full range of Ps-Ps. Hence, if true that [34] helps predict that [35bii] is well-formed, this is not a virtue. Bresnan's second argument is that [34] correctly predicts the fact that adverbial and parenthetical material cannot be interpolated between the stranded preposition of a Ps-P and the preceding verbal stem:
< previous page
page_216 If you like this book, buy it!
next page >
< previous page
page_217
next page > Page 217
Although this argument seems basically correct, there are certainly certain troublesome cases; see note 14. The argument seems even stronger when it is observed that there is, of course, no general restriction requiring a stranded English preposition to be contiguous to a V:
Bresnan's first, and surely most striking, argument for [34] is that it predicts that the complex Vs it postulates can undergo a rule limited to Vs, one that Bresnan calls 'Participle Adjective Conversion'. This converts a basic participle to one which is in addition a member of the category Adjective. Evidence for this is provided by the possibility of V + preposition combinations in such nominals as those in [38]:
The point is that in general English predicational forms cannot occur in such positions unless they are adjectives, and even then, only when the adjective is phrase final. While apparently a striking argument for the central role of a rule like [34] in Ps-P description, the phenomenon in [38] actually provides a basis for rejecting this conclusion. For Bresnan failed to observe that only a proper subset of Ps-P participles can behave like adjectives. Consider, for example, the well-formed Ps-Ps in [39] with their ill-formed corresponding adjectival uses:
< previous page
page_217 If you like this book, buy it!
next page >
< previous page
page_218
next page > Page 218
Since only a proper subset of the passive participles found in Ps-Ps have adjectival uses, the question is how, in terms like Bresnan's, this fact can be made compatible with the claim that all the passive participles in Ps-Ps involve incorporated Ps as parts of complex Vs. There seem to be two possibilities. First, some ad hoc constraints could be invoked. If this path is taken, the argument that the adjectival use of some Ps-P participles supports the rule of V-P Incorporation largely disappears. Second, it could be claimed that the class of allowable adjectival uses is determined by independently needed constraints. Here the only candidate in the material Bresnan actually presents is the constraint in her rule Participle-Adjective Conversion:
Bresnan claims that [40] is applicable only to Vs whose associated predicate's SUBJ is its theme. The theme restriction is intended to predict, inter alia, that intransitive (nonpassive) participles undergo the rule only if their SUBJ is a theme, accounting for such contrasts as [41]:
The idea is that fall takes a theme subject, while jump does not. 11 The trouble with this approach to the problem in [39], indeed to any problem, is that notions like theme are quite vague and unclear, a defect Bresnan (1982a: 24) is not unaware of:12 While there are obvious difficulties in providing a consistent thematic analysis of nonmotional and nonlocational verbs, nevertheless, some appropriate semantic restriction on the subject argument of [36] is required,... Bresnan claims that the theme argument in a predicate is that which undergoes the motion or change in state denoted by the predicate. Even putting vagueness aside, this notion seems to have no application to the issue of which Ps-Ps have corresponding adjectival uses. Consider, for instance, [42]:
If it is the condition in [40] that allows this use of hear to undergo Participle-Adjective Conversion but not the predicates in [39], then that island must be
< previous page
page_218 If you like this book, buy it!
next page >
< previous page
page_219
next page > Page 219
the theme of the predicate in [42b]. But there is no motion or change of state in [42], so the definition should not apply. On the other hand, in [39a,b] there is apparently change of state, but the adjectival use is impossible. Consider also [43]-[45]:
It seems that any claim that those cases which permit adjectival uses involve themes while those which do not can only, at this point, be maintained via a notion of 'theme' that is empty of all independent content. Thus any appeal to the condition in [40] really reduces the second solution to the first, thereby eliminating the argument for rule [35]. I conclude that nothing in Bresnan's account so far yields insight into which class of Ps-Ps have corresponding adjectival forms. Therefore this phenomenon offers no current support for the view that Ps-P clauses inherently involve incorporation of prepositions into complex Vs. There is, rather, an argument to the opposite effect. Suppose that the adjectival use of a V preposition does indeed depend on the incorporation of the preposition in the V; hence adjectival V + preposition combinations involve complex Vs. There is then a way to account for the subsetting facts about adjectival use of Ps-P participles as follows. Some Psparticiples, but only some, permit incorporation of the stranded preposition, this incorporation being a secondary, nonessential and optional feature of (only some) Ps-P clauses. 13 Of course, this distinction is at present ad hoc, but no one has shown any principled basis for it in any terms. If some principle is involved, e.g. a semantic one, it could be interpreted as predicting which participles permit preposition incorporation. This approach to the adjectival use problem remains entirely consistent with the five arguments of section 1 against an incorporation view, those involving: (i) negatives, (ii) only, (iii) the Heavy NP Shift phenomenon, (iv) gapping and (v) 'pseudo-gapping'. Recall that those arguments showed that Ps-Ps do not correspond to active structures in which a < V Preposition Nominal X > sequence is bracketed < < V P> Nominal X >. But in what was just proposed here, preposition incorporation would only be found in some passive structures, never in actives.
< previous page
page_219 If you like this book, buy it!
next page >
< previous page
page_220
next page > Page 220
Bresnan (1982a) gives three arguments in favor of 'reanalysis'. One was seen to have no force, the second to be largely sound but not without serious difficulties and the third actually to count against 'reanalysis'. When these conclusions are combined with the five arguments against 'reanalysis' given in section 1, the conclusion seems to impose itself that 'reanalysis' cannot be the basis for the existence of English Ps-Ps. But of course it is not in the approach sketched in section 2. 6.4 The Pseudo-Passive Control Rule If the conclusions reached so far are accurate, it is incorrect to attribute to preposition incorporation, which is an only optional property of only some Ps-Ps, the fact, discussed by Bresnan (1982a) and systematically true of all Ps-Ps, that the stranded preposition of a Ps-P is contiguous to the V. This will thus have to be built into rule [28] as an ad hoc constraint, expanding it to [46]: 14
The concept Immed(iate) Arc Precede of this rule can be trivially defined in terms of Arc Precede, defined in section 4.4 of Chapter 1:
Even if it covers the facts properly, rule [46] is sufficiently complicated as to be rather suspicious. Fortunately, various ways of simplifying it exist. A certain amount of the complexity of [46] is due to the need to specify the self-erasure of the relevant Marq arcs, to yield stranded prepositions in Ps-Ps. Of course, there are many other stranded prepositions in English, including those in [48]:
< previous page
page_220 If you like this book, buy it!
next page >
< previous page
page_221
next page > Page 221
Suppose that it is assumed, as is plausible in APG terms, that these cases also involve copy pronouns determined by successor/predecessor pairs in which the successor is an Overlay arc; see JP (1980: Chapter 7). Then it would also be necessary to require the self-erasure of the relevant Marq arcs in these cases:
This suggests that the rule demanding self-erasure of the Marq arcs posited under the present analysis of Ps-Ps has nothing specifically to do with Ps-Ps. Rather, the relevant erasure could be attributed to a far more general English rule, along the lines of [50]:
If this rule exists, then [46] can be significantly simplified. Before showing this, let us consider other ways of obtaining such simplification. Part of the complexity of [46] is due to the need to isolate the marq closure existing in Ps-Ps in order to specify the contiguity of the passive V and the stranded preposition. One can, however, do this more elegantly by defining a relation existing between the predecessor of the prepassive 2 arc and the Marq arc whose support is picked out in [46]. This relation can be grasped by noting the following. The predecessor A is related to a copy arc B, which is a replacer of B. The copy arc is related to a marq closure C, which is a replacer of B. It is this marq closure which must be in the immediate arc precede relationship with the P arc of a Ps-P. It follows that the relation between A and C is the ancestral of Replace, R-Replace. Thus C can be uniquely picked out with respect to A as the last in a chain of arcs related by RReplace, where 'last' makes reference to the fact that C has no replacer. This permits introduction of the defined notion [51]:
< previous page
page_221 If you like this book, buy it!
next page >
< previous page
page_222
next page > Page 222
Given [51], the marq closure which occurs in a Ps-P is a residue of the predecessor of the prepassive arc. This will permit considerable simplification of [46]. A final observation is pertinent to a more elegant reformulation of [46]. Chapter 4 introduced a series of notions which distinguished successor relations according to whether the successor/predecessor pairs sponsored a copy arc or not. This led to such defined notions as Copy*Successor, Copy*Local Successor, etc. But no use is made of these in [46]. In such terms, the prepassive arc defining a Ps-P is an instance of the relation 2 = Copy*Local Successor, that is, is a 2 arc local successor which cosponsors with its predecessor a copy arc. This permits reformulation of [46] as [52]:
Given the complexity and uniqueness of the facts surrounding Ps-Ps, formulation [52] does not seem unacceptable from the point of view of simplicity. 6.5 An Alternative Arc Pair Grammar View of Pseudo-Passives 6.5.1 The Issue Sections 1-4 have ignored the possibility of an alternative APG approach to Ps-Ps, one permitted by the conclusions of Chapter 7 below concerning the existence of passive clauses in which the prepassive arc is not a 2 arc. The alternative analysis would take Ps-Ps to be generalized passive clauses in the sense of Chapter 7, with the prepassive arc thus some non-2 arc. Ps-Ps would then involve direct advancement to 1 of non-2s. Since the identity of the non-2 R-signs is largely irrelevant to answering the questions that follow, one can, for simplicity, concentrate on cases involving: a single R-sign. I have chosen an alternative analysis which would involve direct advancement to l of a 3. Consequently, the discussion can focus on Ps-Ps such as [53]: Under the analysis advocated so far, this involves copy advancement of the initial 3 to 2 with subsequent advancement of the 2 to 1. The alternative analysis would involve copy advancement of the 3 to 1, determining a copy 3 arc, and yielding a structure along the lines of [54] for the short form of [53]:
< previous page
page_222 If you like this book, buy it!
next page >
< previous page
page_223
next page > Page 223
Given structures like [54], the proper rule controlling Ps-Ps could, of course, not be [52], which presumes advancement to 2. The analog assuming direct advancement to 1 might be initially [55]:
Evidently, since the overall approach to passive clauses of this volume, generalized as in Chapter 7, allows both the original analysis with intermediate advancement to 2 and the analysis with direct advancement to 1 for English Ps-Ps, one must try to find grounds for choosing between them. The following remarks attempt to bring to bear some facts which seem to favor the advancement to 2 analysis. It must be admitted, however, that the arguments are far from decisive and involve a number of less than fully secure assumptions. But it is worth developing the issue, at least to clarify the problem. As is, rule [55] is simpler than [52], which seems to offer an initial advantage to the direct advancement view. However, an immediate comparison of these two is meaningless. As things stand, [55] and [52] are incomparable for several reasons. First, it might be necessary to expand [55] to indicate that direct copy advancement to 1 is not possible for 2s. Second, and more important, the latter rule has not built in the required restriction embodied in that part of [52] which makes reference to the property 'Pure'. It is not clear how to build the necessary constraint into [55]. Clearly, Pure
< previous page
page_223 If you like this book, buy it!
next page >
< previous page
page_224
next page > Page 224
cannot be used, since, under the assumption of direct advancement to 1, no 2 arc is overrun, and Pure would never be relevant. 6.5.2 Blocking Pseudo-Passives with Any 2 Arc One might suppose that the correct statement simply blocks copy advancement to 1 in any clause containing any 2 arc at all. This would mean adding to the consequent of [55] the condition [56]:
However, internal to usual APG assumptions, [56] is not possible. For standard APG analyses of the clauses in [57a,b] take them to have invisible 2s:
In [57a], even when the pronoun is not visible, the clause contains a 2 arc corresponding to it, one which self-erases. [57b] contains an initial 2 arc whose head is an unspecified nominal, a 2 arc which also self-erases; see JP (1980). But clauses like [57a and b] are subject to Ps-P formation, as [57c and d] show. However, the variants of [57a] where the pronoun is visible cannot have corresponding Ps-Ps:
Given the assumption that all the cases of [57a and b] involve 2 arcs, it can be claimed that the facts follow, under the advancement to 2 approach, from the Pure condition in [52]. As discussed in JP (1980: Chapter 9), when the unspecified arc in [57b] is overrun in a passive like [57d], the overrunner arc erases it, precluding its otherwise necessary self-erasure. This makes the overrunner pure. An analogous treatment can be imagined for [57a], although this raises problems for certain principles of JP (1980). 15 It follows that the treatment of cases like [57c,d] suggests that Ps-Ps are possible correspondents of actives containing certain 2 arcs, and thus that [56] is not a possible way to build the required restriction into the direct advancement approach. 6.5.3 Some Earlier Observations To the extent that examples like [57c and d] or other factors, known (see below) or unknown, show the untenability of [56], there is an argument for
< previous page
page_224 If you like this book, buy it!
next page >
< previous page
page_225
next page > Page 225
the advancement to 2 view of Ps-Ps related to some data discussed in early transformational work. The basic facts seem to have been first noted in Alexander and Kunz (1964: 26), then independently by Higgins (1973: 184-185), whose remarks are discussed by Emonds (1976: 124, fn.). Dozens of English Vs take a 3 and a following (that) complement, as in [59]:
The key fact is that, although 3s are, as has been seen, subject to formation of Ps-Ps, none of Vs of the class illustrated in [59] permits a Ps-P corresponding to structures like [59], as noted by Alexander and Kunz (1964: 26), 16 and, for specific cases, by Higgins (1973), Emonds (1976: 124) and Bach (1980: 326):
Given the plethora of particular lexical, semantic, and pragmatic constraints on Ps-P formation,17 which have here largely been ignored, it might be claimed that it is just a grammatical accident that clauses of the sort in [59] do not permit Ps-Ps. This is suspicious given the regularity of the blockage for the dozens of Vs permitting both 3s and complements.18 Moreover, it is worse than suspicious, since, as may not have been observed previously, members of the V class exemplified in [59] permit Ps-Ps based on the 3 if they do not occur with a complement:
Evidently then, *[60] represents not a lexical incompatibility between certain Vs and Ps-Ps but a syntactic constraint somehow related to the presence of the complement clause. A precise APG account of this constraint can be given which is, theoretical differences aside, essentially consistent with the view proposed by Higgins (1973) and accepted by Emonds (1976: 124). The relevant restriction reduces to a combination of certain APG laws and the Pure condition of the consequent of [52]. The latter is, as already seen, motivated independently of questions of cases like [60] and [61] by such impossible Ps-Ps as those in [62]:
< previous page
page_225 If you like this book, buy it!
next page >
< previous page
page_226
next page > Page 226
To explain the starred examples of [61] via the constraint operative in [62], observe that examples like [59] involve extraposed clauses, moreover, extraposed clausal 2s, of the kind discussed in Chapter 3, section 2. So there is a dummy 2 arc in all such cases and the complement heads an initial 2 arc but a surface 8 arc. This view accounts for the difference, in word order between e.g. [63a and b] in terms of the independently needed principle that surface 3s follow surface 2s but precede surfaces 8s. The difference between the two variants of [63c] is then only a matter of whether the dummy 2 arc self-erases or not:
One then obviously wants it to follow from this view that the Ps-Ps in [64] are all ill-formed for the same reason:
This should be the Pure condition in rule [52]. Under the assumption that all Ps-Ps involve advancement to 2, the 2 arc local successors of the 3 arcs in [64], which overrun the ghost arcs defining extraposition, must erase those ghost arcs in order to be pure. But, given the principles of JP (1980), this is impossible. As considered in JP (1980: Chapter 9), and, as already touched on, such erasure is only allowed when the erased arc is unspecified. But dummy arcs are not unspecified in this sense. 19 A problem was noted for this view with respect to grammatical Ps-Ps like the reflexive-containing one in [57c]; see note 15. This perhaps suggests no more than that the relevant principles have to be fixed up to allow an arc overrunning an insecure arc to erase that insecure arc if it is a reflexive pronominal arc as well as an unspecified one. While this would be quite inelegant lacking some discovery of a formal similarity between unspecified nominals and (just) reflexive pronominals, it would still exclude the relevant sort of erasure of most arcs, including dummy arcs.20 I have been trying to argue that the direct advancement to 1 view of Ps-Ps is inferior to that involving intermediate advancement to 2. Given that the former cannot preclude all direct advancement to 1 in clauses containing
< previous page
page_226 If you like this book, buy it!
next page >
< previous page
page_227
next page > Page 227
2 arcs, the facts in *[60] and *[61] have been shown to indicate an advantage for the advancement to 2 view. Given certain APG laws, the long noted impossibility of Ps-Ps follows from the Pure condition of rule [52], which has so far no proper analog in the direct advancement to I approach. However, the relevance so far of *[61] and *[62] to the choice between the two APG views of Ps-Ps is conditional on the existence of other facts which preclude incorporating [56] into rule [55]. For [56] would suffice to block *[61] and *[62] on the extraposition analysis of clauses like [59] assumed here. Note though that *[61] and *[62] do unconditionally support the advancement to 2 hypothesis against the direct advancement view in which rule [55] is extended not by condition [56] but by [65]:
On the direct advancement view, clauses like the starred forms in *[62] contain surface 2 arcs, and would be directly blocked by [65]. However, since cases like *[60] contain no surface 2 arcs, to the extent that all the examples of *[64] are blocked for the same reason, [65] is not a possible alternative to the Pure condition of rule [52], which is possible under the advancement to 2 view. 6.5.4 Some Demotion Paradigms Given the conditionality of *[61], etc., with respect to [56], it is important to seek other objections to an account incorporating [56] into rule [55]. I believe such are furnished by several partially unique paradigms of Vs taking complements. Let us begin with that of the peculiar V see, on the reading illustrated in (66).
I interpret these facts as showing the following. Being unpassivizable, untopicalizable, and unobject-raisable, the complements in both [66a,b] have the properties of extraposed clauses, as discussed in Chapter 3, section 2. But, in the absence of counterevidence, clauses which share the properties of extraposed clauses are extraposed clauses. 21 Hence, in this sense, see is partially like such complement-taking Vs as hold, take, and feel which require extraposition of their complement and thus, in contrast with those
< previous page
page_227 If you like this book, buy it!
next page >
< previous page
page_228
next page > Page 228
like e.g., believe, never permit the complement to be topicalized or to appear as the 1 of a passive. Thus the relevant sense of see is one of the Vs governed by the constraint in [3.16]. Moreover, the V of [66] is like take, but unlike most others, in permitting the extraposition dummy to show up as a non-1. But differences also manifest themselves since, obviously, the grammar must account for the preposition contrast:
Moreover, there is a deeper issue. I have spoken of [66a and b] as involving extraposed clauses. How can such structures, especially [66b], be integrated into a general account of English extraposition? Specifically, in current terms, how can [66a and b] be subsumed under the alternative types of representation allowed in [3.7], which permits only extraposed 1s and 2s? Since the complements in [66] are clearly not extraposed 1s, the alternative is that they are extraposed 2s. If e.g. [66b] involves an extraposed 2, the dummy should be a 2. But English surface 2s are not in general, indeed I think not ever, marked with any preposition. Moreover, the presence of the preposition is independent of the presence of the complement, since it occurs in [66g,h and i] as well. On the face of it then, the paradigm in [66] seems to require a unique extension of the notion 'extraposition' to permit both (i) a linkage between an extraposed clause and a dummy inside of a prepositional phrase and (ii) an extraposed clause with no dummy in a paradigm which appears not to permit 2s at all. Suppose though that the sense of English see in (66) takes an initial 2, which can be a complement or an abstract nominal But assume 2s of this V are subject to obligatory demotion. It is unclear what relation the 2 demotes to, but it is not crucial at this point to specify this, except in one respect. Given the Chômeur Advancement Ban of Perlmutter and Postal (1983a) and the related, but more general, Cho Arc No Local Successor Law of JP (1980: 310), which entails the former, the relevant relation cannot be 8. For the present analysis of Ps-Ps like [66i] will, of course, involve advancement (to 1) subsequent to the demotion at issue. The APG law in question precludes all local successors for 8 arcs; hence, if the demotion relation were 8, it would preclude passives like [66i]. Moreover, if the relation were 8, it is not clear how the Motivated Chômage Law could be satisfied, since there is not in general any later 2 arc to motivate demotion. Let us guess that the relevant demotion is to the relation SEMI-OBJECT (R-sign = 5) posited in Postal (to appear c) and mentioned with respect to French in Chapter 4, note 6. One of the prepositions marking English surface 5s would then be to. Later remarks suggest that at and possibly for also have this function. Under the assumption being considered, in [66g], the initial 2 demotes to 5, while in [66b], it is the noninitial dummy that does. In each of these cases, it is the last 2 which demotes to 5. However, this cannot be the right general
< previous page
page_228 If you like this book, buy it!
next page >
< previous page
page_229
next page > Page 229
constraint given [66d and i] under current assumptions, since in these cases the last 2 is a later 1. The demotion thus seems to pertain to last 2s which are not Is. Thus it might seem that the demotion for a V like see is demotion of the last acting 2 arc. This can also be stated in terms of the notion Secure of Chapter 4, note 22. For the initial 2 arc in cases like [66b and d] would be insecure since free and overrun by the stable dummy arc it sponsors. But the resulting dummy 2 arc is secure. Since the prepassive 2 arc taken to underly Ps-Ps would also be secure, the constraint would then pertain to the first secure 2 arc, 'first' being understood analogously to the 'last' of Chapter 2, section 2.3.4, as made precise in [68a]. Hence the demotion requirement can almost be [68b], with a purely schematic V identification statement:
In these terms, the structure of [66i] includes the elements in [69]:
< previous page
page_229 If you like this book, buy it!
next page >
< previous page
page_230
next page > Page 230
I said the constraint on see could 'almost' be [68b]. Rule [68b] is incorrect as is because it wrongly precludes the wellformed [66a]. One obviously wants to analyze this as involving the same structure as [66b] except that the dummy 2 arc defining extraposition self-erases, as do those in most other extraposition 2 cases, that is, those not involving 5 demotion. In this respect, [66a] should be treated like the complement cases of Vs such as hold, feel, etc. Recall from Chapter 3 that, with these Vs, extraposition was taken as obligatory but, as in general, the dummy 2 arc defining this had to self-erase if it had no I arc local successor. Thus one wants the dummy in [66a] to be invisible for the same reason, via the same rule, as those in e.g., [70a], as compared to the marginal possibility in [70b-d]:
The alternative would be to posit entirely ad hoc erasures for dummy 5 arcs. Thus relevant part of the structure of [66a] should be [71]:
Consequently, one must protect self-erasing 2 arcs like C in [71] from the necessity of demotion, even when they are the first secure 2 arc. For even with Vs whose 2s are otherwise subject to demotion, the relevant nominals
< previous page
page_230 If you like this book, buy it!
next page >
< previous page
page_231
next page > Page 231
are freed from this obligation if they head self-erasing 2 arcs. One can account for this via the concept in [72]:
Rule [68b] can then be properly weakened slightly by replacing 'First 2 = Secure(A)' by 'First 2 = Full(A)', so that if the first secure 2 arc self-erases, the requirement of demotion does not exist. The actual reformulation of [68b] is delayed until after it is argued that the rule can be generalized to cover the behavior of other Vs. Let us summarize what a demotion analysis embodying the improved version of rule [68b] does. By permitting one to take the extraposed complements in [66a and b] to be initial 2s, the demotion hypothesis permits the retention of a restrictive view of English clausal extraposition under which this is limited to 1s and 2s. It eliminates any necessity for special devices to account for the relation between an extraposed complement and a dummy inside a prepositional phrase in sentences like [66b]. Such sentences turn out to illustrate the case where the dummy 2 uniformly associated with extraposition of 2s itself demotes to 5. 22 Further, the demotion hypothesis permits one to relate exact paraphrase pairs like [66a and b] via appeal to an independently motivated rule which forces self-erasure of extraposition dummy 2 arcs which lack 1 arc local successors. Under a demotion analysis, [66a and b] are related by the same rule which distinguishes [70a] from [70b,c and d]. The demotion hypothesis accomplishes the previous tasks while preserving the generalization that English surface 2s are not marked with prepositions, despite the assumption that the relevant V takes initial 2s.23 Finally, postulation of demotion permits preservation of the restrictive APG principle that all ghost arcs and all ghost arc sponsors are nuclear term (1 or 2) arcs, even given the occurrence of dummy nominals as non-1s or 2s in cases like [66b]. These seem like sufficiently useful consequences to initially justify the price of positing demotion, which is a rule like [68b]. Let us return to the central goal of this discussion, which is to find grounds for distinguishing the advancement to 2 view from the direct advancement view of English Ps-Ps. An obvious consequence of the demotion analysis of paradigms like [66] is that English Ps-Ps can be formed from clauses containing 2 arcs (distinct from the prepassive 2 arc posited only under the advancement to 2 analysis of Ps-Ps), For example, the Ps-P in [69] contains not only the prepassive 2 arc, B, but also the initial 2 arc D. However, the presence of such 2 arcs is consistent with the Pure condition of [52]. For the prepassive arc B of [69] does not, and could not, overrun the initial 2 arc. Hence B is pure, because it satisfies vacuously the Pure condition of [52]. However, the situation in question is not, of course, consistent with the direct advancement to 1 analysis based on the combina-
< previous page
page_231 If you like this book, buy it!
next page >
< previous page
page_232
next page > Page 232
tion of rule [55] with condition [56]. For that wrongly precludes any 2 arcs in Ps-Ps, inconsistent with the 2 to 5 demotion analysis of paradigms like [66]. This is a reason for thinking the paradigm in [66] supports the advancement to 2 analysis. A final remark about structures like [69]. The passivized phrase in [69] heads a 2 arc, D, at one level, a local successor of that, C, and then the prepassive 2 arc, B, local successor of that, as well as the ultimate 1 arc local successor defining the clause as passive. This 'back and forth' structure may seem odd, but it violates no APG laws or other wellestablished principles and I see no reason to reject it a priori. Formally, there is a parallelism with the treatment of ANTIPASSIVE constructions proposed in Postal (1977), discussed in JP (1980) and Postal (1982), and supported in Davies (1981a, 1984). In this analysis, a single phrase heads a 1 arc, a 2 arc local successor and then a 1 arc local successor of that; hence there is a representation having the same 'back and forth' properties. Some, but surely only some, of the scepticism as to the validity of a demotion (to 5) analysis of paradigms like [66] will arise from the fact that the demotion analysis so far is ad hoc to a single V. Significantly though, the same demotion approach to other Vs can quite possibly provide an adequate solution to a problem dealt with, but inadequately, in Rosenbaum (1967). Rosenbaum (1967: 83) noted that certain complement-taking Vs, e.g., marvel and rejoice, cannot occur with postverbal. nominals but do take prepositional phrases. However, these Vs do occur with complements and, moreover, in contrast to e.g., see, permit the complement to passivize. Rosenbaum listed such paradigms as [73a-c], to which I have added [73d-f]:
The starred versions of [73e and f], as well as Rosenbaum's *[73c], argue that the surface complement which occurs postverbally with such Vs is an extraposed clause. The paradigm of such Vs is thus partially similar to, but also partially different from, that of see. One difference from see is that extraposition cannot be said to be required. For that would wrongly preclude [73b] and the good versions of [73e and f]. A key question for Rosenbaum about paradigms like [73] was how to account for examples like [73a], which contain no preposition. He argued against the idea of proposing two different base structures, one with a prepositional phrase and the other with bare complements. Such a proposal offers no obvious source for, e.g., passives like [73b], topicalized structures like [73d] or object-raised structures like [73f].
< previous page
page_232 If you like this book, buy it!
next page >
< previous page
page_233
next page > Page 233
Rather, assuming that there was an obligatory rule deleting prepositions in front of, in particular, that complements, Rosenbaum proposed to generate with Vs like marvel prepositional phrases whose objects were that complements, although these prepositional phrases are, of course, never found in English surface structures. This was not unnatural since, for Rosenbaum, that complements were members of the category NP, contrary to much later work, which was stimulated by Emonds (1969). Rosenbaum (1967: 84-85) thought he had found an important argument for the existence of such a preposition deletion rule beyond that of its role in regularizing paradigms like [73]. Namely, this was supposed to predict the existence of 'long' passives, which lack the characteristic and normally obligatory by, as in [74c] and [75c]:
However, the line of reasoning attempting to relate putative passive structures like [74c] and [75c] to facts like the possibility of [73a] was clearly mistaken. 24 For [74c] and [75c] are not passives but adjectival clauses. This is shown, first by their compatibility with adjectival modifiers like very and contexts like Nora seem ..., which do not combine with passive participles, and, second, by their incompatibility with environments like Nom see Nom ... and Nom had Nom ..., which take passives but not adjectives:
In addition, if one picks cases analogous to [74c] and [75c] which are not subject to a possible interpretation as adjectival clauses, the result of removing the preposition is totally ill-formed:
< previous page
page_233 If you like this book, buy it!
next page >
< previous page
page_234
next page > Page 234
Thus, while Rosenbaum's. joint assumptions that (i) that clauses are NPs and (ii) there is a rule deleting prepositions directly before such clauses at first seem to regularize paradigms like [73] and to be supported. by clauses like [74c] and [75c], in fact, the passive distributions suggest that these ideas do not offer a correct account. A different approach to paradigms like [73] which does not have the false consequences for passives like *[78b and d] must thus be found. Rosenbaum's proposal is objectionable in another way. lit offers no basis for the fact that the complement in [73a] behaves like an extraposed clause. For Rosenbaum, this complement is not extraposed; it is simply an 'oblique' complement whose preposition has been deleted. Significantly then, the approach developed for see offers an account which makes no false predictions about passives like *[78b and d] and which determines that the complement in [73a] is an extraposed clause. One can assume that Vs like marvel take initial 2s subject to demotion. Since it is again unclear what the demotion relation would be, let us once more just guess that it is 5. This permits transforming the improved version of rule [68b] motivated earlier for one sense of see from an ad hoc statement about that V into a generalization about a class including see, marvel, rejoice, and no doubt others. 25 [68b] can now be reformulated something like [79]:
Since the set described in [79] includes the Vs cited by Rosenbaum, most of the unique properties of these reduce to the same principle underlying the see paradigm. However, there is a distinction other than choice of preposition (to vs. at). Contrary to the situation with see, extraposition cannot be said to be required. For that would wrongly preclude [73b] and the good versions of [73e and f]. One might try to say that unlike see, a V like marvel, is a member of the majority class of Vs taking transitive 2 complements which extrapose. As discussed in Chapter 3, section 2, and in the previous subsections, these require that the extraposition dummy 2 arc self-erase if it has no I arc local successor; see (ii) of note 4 of Chapter 3. This would account for *[80a] and its contrast with [73b]:
However, this proposal does not suffice to account for the ill-formed variant of *[80b] in which the preposition is present. The ill-formedness of *[80b] separates such Vs from the standard extraposition case illustrated by [80c].
< previous page
page_234 If you like this book, buy it!
next page >
< previous page
page_235
next page > Page 235
Some other constraint must be operative here which differentiates Vs like marvel both from see and from those like believe. Given the possibility of [73b], it cannot be a constraint blocking all passives with the former. An obvious suggestion is that marvel is a member of a V class subject to a stronger constraint on the extraposition dummy 2 arc than that it self-erase if it has no 1 arc local successor. These Vs can be said to require unconditionally that this arc selferase:
Rule [81] blocks both *[80a and b], as required. *[80a] is blocked for obvious reasons; *[80b] is bad because, given [81], it would have to involve the impossible marking with a preposition of an extraposed clause, something never found in English. Other Vs which should be mentioned in [81] probably include, at least for me, insist (see note 7 of Chapter 3), and maybe rejoice. pray, hope and wish (see note 25). Thus, via rule [81], the Vs subject to 2 demotion to 5 other than see require self-erasure of the dummy arc defining extraposition. The key difference between, e.g., see and marvel is then the same as that accounting for the contrast between, e.g., resent and hold in [70]. 26 It might be argued that rule [81] is an ad hoc statement required by the demotion approach to Vs like marvel. But this is untrue in two senses. First, any approach to English grammar needs some statement to preclude structures like [82]:
In current terms, [81] correctly blocks all such sentences. Second, and quite significant from the point of view of justifying a demotion analysis of Vs like marvel, there is evidence that a rule like [81] is needed for Vs which uncontroversially take 2 complements, in particular, the sense of say on which it refers to actual utterances and takes a 3. For me, this V has the asymmetrical that clause paradigm [83]:
< previous page
page_235 If you like this book, buy it!
next page >
< previous page
page_236
next page > Page 236
[83a,e and f] show that the surface complement occuring with this sense of say is an extraposed clause. [83h] shows that the ill-formed passive *[83d] 27 cannot be attributed to any general blockage on passives with this V. But the paradigm is accounted for under two assumptions: (i) that this V is subject to obligatory extraposition, as with others encountered earlier, that is, that this sense of say is another V governed by the constraint in [3.16]; (ii) that the V is also one of those listed in [81], hence acquiring its ghost 2 arc if any to self-erase. Another V having a paradigm like [83] for me is write:
As with the see paradigm discussed earlier, the demotion analysis of the Vs treated by Rosenbaum subsumes them under independently needed assumptions about English extraposition. Under the demotion analysis, extraposed clauses like that in [73a] are reduced to the paradigm of extraposed 2s, as shown inter alia by their sharing the properties specified in rule [81]. This rule is, as seen from [83] and [84], motivated for nonprepositional cases; but under the demotion analysis, it also properly governs prepositional examples like those of the marvel paradigm. Moreover, unlike Rosenbaum's account involving a preposition deletion rule, the demotion approach to the marvel paradigms yields no false predictions about passives like *[78b and d]. The absence of preposition in [73a] is here attributable to the regularity that English extraposed complements never take prepositions. The last remark is oversimplified in one respect. It leaves open the question of why there are not instances of what I have taken to be demotion paradigms like [82] on an analysis where the complement is not an extraposed clause (that is, final 8) but simply a clausal 5 due to demotion:
Similarly, the account has left open the question of the absence of topicalized structures like: [85] and [86] are both consistent with the rules given so far, in that they would simply involve demotion to 5 of the initial 2 complement, required by rule [81] in the absence of the dummy 2 arc, which only exists given
< previous page
page_236 If you like this book, buy it!
next page >
< previous page
page_237
next page > Page 237
extraposition. And, while obligatory for see, extraposition is, as already stressed, optional for marvel. Rosenbaum's preposition deletion account was designed, inter alia, to account for the facts of *[85b], *[86]. The fact is that no English surface structure contains a that complement embedded in a prepositional phrase. It may be then that a rule banning such complements from being the heads of Marq arcs is necessary. Such a rule would not be inconsistent with the existence of Ps-Ps in which the final 1 is a that complement. For under the present treatment, the final 1 of a Ps-P never heads a Marq arc, as already stressed. Only the copy heads such an arc. Similarly, topicalized structures like *[86] would also involve copies and not complements inside of prepositional phrases. I will not formulate a rule banning that complements as heads of Marq arcs since I have some doubts that this obvious approach is correct. Note though that without the demotion approach I have sketched, such a ban would not suffice to regularize the marvel paradigm or related paradigms. It still requires an account that in, e.g., [73a], the complement behaves like an extraposed clause; it still requires an account that passives with dummy 1s like *[82b] are impossible. Blocking that clauses inside of prepositional phrases does not impose these restrictions. The chief thrust of the discussion of the previous two subsections is this. Via appeals to a demotion analysis, etc., grounds have been presented for taking seriously the view that there are Ps-Ps, e.g., with see and marvel, in which there is more than one, in fact, sometimes two, sometimes three distinct 2 arcs. These are the initial complement 2 arc, sometimes a dummy 2 arc and in all cases a prepassive 2 arc local successor of the demotion arc itself, headed by the complement in the absence of extraposition, by the extraposition dummy otherwise. If correct, these conclusions show that rule [55] cannot incorporate a simple condition like [56]. They thus leave it obscure how one would impose on [55] the needed analog of the Pure condition of rule [52], possible under the advancement to 2 analysis. The demotion paradigms thus provide, I conclude, another reason for believing, that the advancement to 2 approach is the correct analysis of English Ps-Ps. 6.6 Open Questions There are a number of issues related to the descriptions of English Ps-Ps which I have not dealt with and which space precludes any serious treatment of here. I will simply mention several such problems. 6.6.1 Other Constraints Limiting the Class of Pseudo-Passives First, I have not tried to specify the constraints limiting the class of local predecessors for the prepassive arcs taken to define Ps-Ps, constraints which
< previous page
page_237 If you like this book, buy it!
next page >
< previous page
page_238
next page > Page 238
restrict just which prepositions with just which Vs permit Ps-Ps. That is, I have not discussed what blocks such impossible Ps-Ps as those in [87]:
A vast body of data relevant to any study of this question is found in Couper-Kuhlen (1979). However, many such cases of impossible Ps-Ps follow from the 1AEX and the unaccusative hypothesis, that is, the view that the relevant intransitive Vs take initial 2s. Examples include [88]:
If, as claimed here, the Vs in [88] take only initial 2s, the impossibility of the passives follows from the 1AEX. This is true regardless of whether one adopts the advancement to 2 approach to Ps-Ps, argued for here, or the direct advancement to 1 approach. The same ideas help account for contrasts in passivizability which have been noted, e.g., [89], due to Svartik, and discussed in Riddle and Sheintuch (1983):
The V of *[89b] is an ordinary unaccusative; any passive based on it is thus blocked by the 1AEX. [89a] however involves a semantically distinct V which takes an initial 1 and either an initial 2 which demotes, possibly to 5, or an initial oblique. Nothing blocks advancement of the non-1. Similarly, as touched on in Chapter 2, the 1AEX blocks all Ps-Ps related to the 3 which is the earlier 1 of an Inversion construction, hence inter alia cases like [90]:
< previous page
page_238 If you like this book, buy it!
next page >
< previous page
page_239
next page > Page 239
6.6.2 Additional Complex Passives A second issue related to Ps-Ps involves much-discussed cases of passives in which the predicate seems to be complex, independent of the stranded preposition of Ps-Ps. Such cases, discussed for example in Bach (1980: 323-324) and in Bresnan (1982a: 57-62), include [91] and [92]:
[91] and [92] illustrate two different subtypes. In the former, there is only an 'outer' passive, that is, a Ps-P, while in the latter, there is a Ps-P and an 'inner' passive, at least subject to certain conditions. Such cases raise problems for the account of Ps-Ps developed here for several reasons. First, the possibility of [92c] suggests that the advantage phrase in [92b] is a 2, which would violate the Pure condition of rule [52]. Moreover, if this phrase is a surface 2 in [92b], then the Immediate Arc Precede condition would not hold. Indeed, the latter cannot be met unless advantage in [92b] corresponds to an R-branch of the P arc. Bresnan (1982a) suggests a 'reanalysis' approach to such phrases. While that might be attractive in certain ways, integrating such a view with previous assumptions is far from straightforward. Moreover, one can suspect that 'reanalysis' in some sense misses a key feature of structures like those in [91] and [92]. Intuitively, elements like advantage, fun and sight in [91] and [92] are predicational. In general, they are equivalent to individual Vs. Hence [91a] is paraphrased by [93]: This suggests to me that it would be worthwhile considering a radical analysis of all such structures in which the elements just cited head initial Stem arc branches of P arcs. In APG terms, this suggestion could be initially realized by an analysis for, e.g., [91a] something like [94]. This structure ignores those aspects of [91a] involving the specifics of the prepositional phrase:
< previous page
page_239 If you like this book, buy it!
next page >
< previous page
page_240
next page > Page 240
This representation thus hypothesizes that Jerome in [91a] heads an initial 2 arc, just as it does in [93]. The hypotheses in [9,4] play a role in giving semantically identical structures similar initial representations, in accord with basic APG ideas. However, in [94], the nominal Jerome is demoted to 6 when the initial predicate stem ascends to be a clausal 2. Thus such structures claim that NLs involve a hitherto undiscussed type of raising: ascension of the stem of a P arc to head a 2 arc neighbor of that P arc. One virtue of this view is that it offers an approach to the relation between pairs like make/fun, etc., in terms of the syntactic notion of pronominalization. In [94], G is in fact a copy arc sponsored by H and C. Hence Copies(G, H). Therefore, G meets the conditions for being a pronominal arc in the sense of JP (1980: Chapter 11). The idea that empty Vs like make in [91a], catch in [91b] and take in [92] are pronominal replacements for the true meaning-bearing elements seems attractive. Because elements like fun in [94] head 2 arcs, the possibility of simple PPs like [92c] is in principle allowed for. Of course, heavy constraints exist on this possibility, as *[91e,f] show. That is, most 2 arc foreign successors of Stem arcs would have to be prevented from being prepassive arcs. The possibility of Ps-Ps based on elements like Jerome in [91a] is allowed for because these head arcs of the sort which are independentlly subject to copy advancement to 2. Nonetheless, the view of Ps-Ps like [91b and d] and [92b] just sketched is incompatible with at least two aspects of rule [52]. Under assumptions like those in [94], the Immediate Arc Precede condition would not hold, and the
< previous page
page_240 If you like this book, buy it!
next page >
< previous page
page_241
next page > Page 241
Pure condition would be violated. With respect to the latter, one would apparently have to say that when 2 arcs like C in [94] are overrun in the formation of Ps-Ps, they have (presumably) 8 arc local successors. Rule [52] would then have to be modified so as not to require that the advancee to 2 be Pure in the case where it overruns a successor of a Stem arc. This is not itself too troubling. But then one observes that under just these conditions, the Immediate Arc Precede condition also does not hold, because the hypothesized final 8 (e.g. fun in [91b]) intervenes between the predicate and the stranded preposition structure. Conceivably, one could eliminate part of this suspicious correlation of violations by positing an incorporation of such 8 arcs into the predicate, that is, by allowing such 8 arcs to have foreign successors which are branches of the P arc. Thus there would be a curious 'back and forth' structure in which, e.g., fun would head an initial branch of a P arc, then a clausal 2 arc foreign successor, then an 8 arc local successor of that, and then some kind of branch of the P arc which is a foreign successor of that 8 arc. While complicated, these assumptions would offer a way to block a wide variety of Ps-Ps in cases like [95]:
These cases could be analyzed as in [94], but it could be claimed that, contrary to expressions like fun, those like bet, etc., do not permit incorporation of the relevant 8s. Hence Ps-Ps like *[95b and d] would violate the Immediate Arc Precede condition. Alternatively, it could be claimed that cases like [95] do not fall into the set of cases permitted to be free of the Pure condition. Obviously then, a serious APG account of these marginal and variable cases remains to be worked out. Lacking that, they may serve as a wedge for a criticism of the general approach to Ps-Ps advocated here. But I think that the speculative ideas just presented offer some hope of permitting as good an analysis of these irregular and highly exceptional structures as is possible. 28 Notes: Chapter 6 1. Chomsky (1981: 122) suggests that certain English passives are, in his transformational terms, not formed by NP movement. This novel idea is motivated, as far as I can see, by the class of structures treated in Chapter 3, section 2 in which a transitive complement cannot be the final 1 of a corresponding passive. In current terms, of course, such cases in no way undermine the uniform characterization of passive clauses. 2. The form dessus is the regular variant of the preposition sur when the prepositional phrase has no surface object.
< previous page
page_241 If you like this book, buy it!
next page >
< previous page
page_242
next page > Page 242
3. The restriction to names seems required because of cases like:
4. The property + + Copies is obtained from the relation Copies(A, B) defined in (C) of the Postscripts to Chapter 2, section 3.2 via our standard '+ +' notation. This converts a binary relation between arcs into a property of the second term of the relation. Hence + + Copies(A) designates an arc which, with its successor, cosponsors a copy am replacer (of A). 5. Something like [22] would not need to be specified if there were more general blocks on advancement to 2 covering also the noncopy sort. But Postal (1982) and Fauconnier (1983) argue that French has (noncopy) advancement of 3s to 2s. 6. This view raises questions, though not insuperable problems, concerning the treatment of cases like (i): Does the relevant prepositional phrase structure involve (a) a surface Top arc which is a marq closure, or (b) a Top arc foreign successor of a 3 am which is a marq closure? In case (b), the conditions in the text would not hold. I conclude tentatively that alternative (a) is correct. Hence cases like (i) involve Top arc successors for 3 arcs which do not have prepositional branches, with these Top arc successors then having Marq arc foreign successors defining the prepositional phrase. An open question is what general principle forces the existence of the preposition in (i) as well as those in [25a and b]. It is important to observe that only alternative (a) above is consistent with a possible law suggested in the text below. This would require in effect that all marq closures be surface arcs. 7. Actually, the account given here involves some minor modifications of that in JP (1980). These seem to yield simpler structures with no loss. 8. For my English, thins is true despite works like Bolinger (1975), cited in Ziv and Sheintuch (1981). According to the latter, Bolinger rejects the constraint suggested in the text and claims that examples like (i) are well-formed:
Ziv and Sheintuch (1981) suggest that sentences such as (ii) are also marginally acceptable:
But for me, all of (i) and (ii) are at best jokes and must be considered ungrammatical. This is not to deny the existence of other dialects in which they are grammatical. Such dialects would be of interest with respect to the issue of 'reanalysis' because the word order in (i) and (ii) renders it impossible to regard the preposition as incorporated in the V unless one claims the same about the preceding nominal. 9. The informal remark: in the text ignores cases like [66d and i] below. The discussion of section 5.4 shows how a certain class of earlier 2 arcs like those in [66d and i] are permitted in Ps-Ps consistently with the condition introduced in rule [33]. Further problems related to the claim in the text are raised in section 6.3. 10. There is a close relation between the notion Pure and the concept Clears of [4.54b]. Recall that Clears(A, B) if and only if A both overruns and erases B. Pure(A) is then more general than the property + Clears(A) in holding of A if A fails to overrun any other arc as well as if + Clears(A).
< previous page
page_242 If you like this book, buy it!
next page >
< previous page
page_243
next page > Page 243
11. Naturally, I would argue that the real systematic difference between pairs like [41a and b] depends on the fact that former involves an unaccusative V and the latter an unergative V. The former type falls together with passives, as in: because both instantiate clauses including the substructure (ii): This is necessary but insufficient for adjectival use of a past participle. 12. In my view, 'thematic roles' play no part in a proper grammatical theory. One virtue of a class of primitive grammatical relations stratified into levels is that it offers an alternative mode of description of many facts for which thematic relations are otherwise appealed to, as in the previous note. 13. It is not possible here, nor directly relevant, to offer any account of the APG structure of whatever real preposition incorporation exists. 14. Although this has hardly been discussed in the literature up to this point, any complete analysis of Ps-Ps must take into account coordinate cases such as:
Depending on specific assumptions about coordination, these expressions could raise problems not so far treated in the literature, including the present discussion. For instance, a 'reanalysis' approach must take the quantifier both in (i) to be incorporated if the coordinate preposition is. Similarly, 'reanalysis' seems faced with two unpleasant choices for (iii). It could be claimed that (iii) is a simple structure in which all of over and then under is incorporated. Or, it could be argued that only the first preposition is incorporated and that the rest represents a second clause reduced by ellipsis. But then, since that clause would have to be taken as a Ps-P, some type of ellipsis would have to be capable of deleting only parts of Vs, that is, everything but the supposedly incorporated preposition. All these problems are worsened in (iv): 15. This problem is already noted and discussed in the postscript to Chapter 9 of JP (1980). 16. These authors had some doubts about the ill-formedness of cases like [60], which I do not share. Others I have asked about these cases likewise do not share such doubts. 17. See Riddle and Sheintuch (1983) and Couper-Kuhlen (1979) for many relevant examples and discussion. 18. The phenomenon is even more general and involves combinations of complements with nominals that we have no reason to consider to be (surface) 3s, as in:
The Vs in question include beg, ask, request, demand, and require. In some cases, the (id) passive is truly bad but the general pattern otherwise shows an extraposed clause, as in the cases with 3s. The same principles appealed to below for the latter can be seen to hold for those in (i). 19. Actually, at least one new law is, I believe, required to guarantee this under the assumptions of JP (1980). In that work, 'unspecified' in the relevant sense was taken to mean having a certain
< previous page
page_243 If you like this book, buy it!
next page >
< previous page
page_244
next page > Page 244
sort of arc as a branch. Something must then prevent ghost arcs from having such branches for it to be formally guaranteed that the laws of JP (1980) preclude treating ghost arcs as unspecified. To accomplish this, I tentatively propose to supplement the principles of JP (1980: Chapter 9) with the constraint in (i). To understand (i), it suffices to indicate that the class of Inexplicit nodes is the class corresponding to heads of unspecified arcs; see JP (1980: Chapter 9).
Details aside, (i) says in effect that any arc having a branch of the sort defining it as unspecified must have the same head as some self-sponsoring arc. It is shown in JP (1980) that no ghost are can have this property, since ghost arcs are grafts. Anti PN Law 13, The Graft Overlap Law, blocks grafts and self-sponsoring arcs from overlapping. 20. Problems for the relevant laws are, however, raised by Jake (1983) on the basis of data in Quechua and English. The English data involves cases like: The problem is that passives like (i) correspond to actives in which the only possible 1 is a nominal hitherto taken to be a dummy: The laws in JP (1980) should then prevent 'short' passives like (i) in which, apparently, an overrun dummy arc is erased by the are which overruns it (that is, where the dummy arc is cleared). I have no happy solution. Perhaps this and other facts which show that weather 'dummies' behave in special ways should be taken to suggest that these elements are not dummies in the sense of the relevant laws. But a precise account yielding this distinction remains to be developed. The facts from Quechua also seem serious, but space precludes a discussion here. 21. The relevant constraint has not so far been formally stated but, informally, has been taken to preclude an 8 arc determined by an extraposition ghost from having a successor. A first approximation might be (i):
22. This causes some problems for previous RG/APG laws about dummies beyond those mentioned in note 20; see the Nuclear Dummy Law of Perlmutter and Postal (1983a) and PN Law 87, the Dummy Arc Law, of JP (1980). These laws have claimed that the only central arcs dummies can head are nuclear term arcs. But I see no viable means of keeping this claim consistent with examples like [67a]. It thus seems that the relevant laws must be weakened. Perhaps, depending on resolution of the issues in note 20, it is correct to allow dummies to head 1, 2, 5, and 6 arcs, but still not 8 arcs. 23. Various universal assumptions might ultimately mandate an analysis like this, e.g., assumptions about what sort of initial arcs a full sentence-like complement can head, or assumptions about what kind of initial arcs a V with the relevant meanings can take. 24. That the analysis proposed by Rosenbaum is tempting is shown by the fact that a quite similar proposal is made for French by Gross (1968: 108-111). Gross's proposal differs in that the preposition deletion rule is optional. I believe that it can be shown to be inadequate for reasons parallel to those given in the text for English.
< previous page
page_244 If you like this book, buy it!
next page >
< previous page
page_245
next page > Page 245
25. Vs which are candidates for being listed in [79] include pray (for), hope (for), wish (for) and brag (about). 26. Compare the present treatment of the see paradigm with that of Williams (1980: 222), who remarks: It, the inert kind, appears sporadically in positions other than subject position, as in the following sentences:
Our theory is incapable of describing these cases: only subject its are predicted to occur under the predication theory, because it is only when an S is interpreted as a subject that an NP is in need of being made inert by it. I do not see this as a deficit in the predication theory, because it is only in subject position that the it can occur with any generality, and there it occurs completely generally. In other positions, it occurs only sporadically:
As for the cases where it does occur in object position, such as regret it and see to it, I offer no general solution, nor, in my opinion, should one be sought. Perhaps regret it and see to it can optionally be reanalyzed as complex verbs taking S complements. One virtue of the present account is that it integrates the unique structures noted by Williams into an overall account of extraposition and dummy it. This is an alternative to Williams' defensive view that the descriptive inadequacy of his account is somehow a virtue. 27. *[83d] should not be confused with examples like: These involve a different, 'generic', V, one which can never be punctual and which takes no explicit 3. Observe that the 'generic' V permits raising structures, while that in the text doesn't:
28. There is an alternative APG analysis of cases like [91b] which deserves study. This would posit not raising of the stem to 2 as in [6.94] but rather advancement of the predicate to 2, an advancement of the copy type. Under this analysis, made would be the head of a copy P arc which erases the initial P arc, and there would be only a single Stem am branch of the latter. This analysis is something like the reverse of that introduced in JP (1980) for predicate nominals and discussed briefly in Chapter 5, section 5.1. In the latter, 2 arcs have P arc local successors, while in what was just suggested, P arcs have 2 arc local successors.
< previous page
page_245 If you like this book, buy it!
next page >
< previous page
page_246
next page > Page 246
Chapter 7 Generalized Passive Clauses 7.1 Remarks From the beginning of work in RG, with a few notable exceptions such as Johnson (1979), the RG analysis of passive clause has been closely bound up with the 2 relation. I have closely followed this tradition in the preceding parts of this work. Specifically, the characterization in [4.2] and the definitions in [4.3] all assume unequivocally that a passive clause is one in which a 1 arc local successor of a 2 arc overruns some arc. Logically, though, there is no particular reason for this limitation. One: can consider the notion GENERALIZED PASSIVE CLAUSE, (G-passive clause), by which would be meant any clause containing a 1 arc local successor which is. an overrunner. Particular subtypes. of passive would then be defined by the R-sign of the local predecessor. What have hitherto been called passive clauses would then be a (to be sure dominant) subtype where this predecessor is a 2 arc. A certain amount of the material in Chapter 2, e.g., in sections 2.3.3 and 2.3.4, was based on resistance to the idea of recognizing passive clauses whose prepassive arcs were not 2 arcs. That is, I was originally motivated by the vague hope that no such clauses existed. It now seems that this hope is completely vain and that an adequately flexible grammatical theory must take passive clauses in the sense of previous sections to be only one subtype of a broader category. Fortunately, this requires only trivial changes in the concepts introduced.
< previous page
page_246 If you like this book, buy it!
next page >
< previous page
page_247
next page > Page 247
7.2 Theoretical Modification Basically, all that has to be done beyond introduction of the notion generalized passive clause is to redefine the concept Arc-Passive(A) of [4.3]. The generalization of this notion will automatically spread to other derivative concepts like Prepassive, Passive-destroyed, etc., which are defined in terms of Arc-passive. Suppose then that [4.3a] is replaced by [1]:
[1] differs from [4.3a] only in dropping the specification that A be the local successor of a 2 arc. Given [1], the earlier notion Arc-Passive is now equivalent to [2]:
Suppose that all the passive clauses of some NL have prepassive arcs which are 2 arcs. Then, given [2], it could have the rule [3]:
I assume, as made clear in Chapter 6, that [3] is correct for English and, as is implicit in Chapter 2.3.2, that it is also right for French. 7.3 Weaker Constraints on Prepassive Arcs: Imbabura Quechua So far I have shown how to generalize the current approach to allow passive clauses in which the prepassive arc is not a 2 arc, but no real reasons for doing so have been given. These grounds will consist of cases in which (i) an NL has passive clauses; (ii) some of these are clauses where the final 1 is an earlier non-2; (iii) the possibility that the earlier non-2 advanced to 2 and only subsequently to 1 is factually excluded. A rather clear case of this sort is provided by Imbabura Quechua (IQ hereafter) as described by Jake (1983). IQ is an agglutinative NL with a complex system of suffixes. Nominals are marked with case endings, including topical (TOP) for topics, nominative (NOM), for 1s, accusative (ACC) for 2s, dative (DAT) for 3s, and instrumental (INST) for instruments. Vs seem to agree with their final 1s in person and number. The passive construction of relevance is composed of a past participle form of a V plus a form of the auxiliary ca- 'to be', inflected for person, number and tense. A relatively typical IQ PP is then [4]: 1
< previous page
page_247 If you like this book, buy it!
next page >
< previous page
page_248
next page > Page 248
Besides the irrelevant nonobligatory tendency of final (passive) 1s to topicalize, the most notable feature in [4b] is that the apparent 8 of a passive is marked, like final 1s, with the phonologically null nominative case. Another example is [5]:
Jake argues at length that the nominals translated as final 1s in English are also final 1s in IQ. So far then, IQ PPs present nothing that goes beyond the narrow traditional framework of passive description. The key fact is that both 3s and instrumentals can appear as the final 1s of IQ passives, as illustrated in [6] and [7]:
The morphological and syntactic properties of [6b] and [7b] leave little doubt that they are, internal to IQ, PPs in the same sense as, e.g., [4b] and [5]. Both manifest the same pattern of case marking as earlier PPs, both use the same passive verbal morphology and the same auxiliary. Internal to IQ, all the examples clearly fall together as passives. Therefore, one can argue that [6b] and [7b] are PPs in which the prepassive arc is not a 2 arc, if one can find some argument that the initial 3 arc and instrumental arcs in such passives do not have 2 arc local successors.
< previous page
page_248 If you like this book, buy it!
next page >
< previous page
page_249
next page > Page 249
A first observation is that although 3s can advance to 2 in active clauses, that is, Q has a relatively standard 32A construction (see below), there is no analog for instruments. But of course this is by no means decisive, since there could be a rule of a type appealed to several times already, linking Inst2A to a subsequent advancement to 1, e.g., a rule like [8]:
Given a rule like [8], it would be possible to claim that passives like [7] involve prepassive 2 arcs. But the idea that all IQ passives involve prepassive 2 arcs is strongly countered by evidence developed by Jake. In particular, by considering the interaction of passivization with a productive raising to 1 construction, Jake is able to show that at least some IQ passives in which an earlier 3 heads the arc-passive arc do not involve an intermediate prepassive 2 arc. Hence such passives involve prepassive 3 arcs, justifying the theoretical recognition of G-passive clauses. The relevant argument that not all passivized 3s in IQ can be taken to head intermediate 2 arcs is found in Jake (1983: Chapter 5). Consider the IQ 32A construction, illustrated by [9]:
Significantly, the order of the two accusative nominals in [9b] is free. Jake argues that [9b] is a 32A construction, in which jari-ta is an initial 3, final 2, and aswa-ta an initial 2, final 8. It turns out to be a general fact that IQ 8s are case marked like the terms they were at an earlier stratum. Recall that the earlier 1/final 8 of a passive is nominative. A key argument for the 32A analysis is that despite the identity of case marking and word order of the two accusative nominals in [9b], only that which is an earlier 3/final 2 under a 32A analysis is passivizable. Given the existence of 32A structures like [9b], it is then natural to suspect that all passivized 3s can be taken to be 2s at an intermediate stage. But Jake is able to invalidate this hypothesis by considering the interaction of passivized 3s with raising to 1. The logic of the argument is this. If, in a structure containing a 2, a 3 is directly advanced to 1, the 2 can, and mutatis mutandis will, remain a 2 at subsequent levels, and hence be capable of behaving like a final 2. On the contrary, if, as in [9b], a 3 advances to 2 at
< previous page
page_249 If you like this book, buy it!
next page >
< previous page
page_250
next page > Page 250
level i, an earlier 2 must take on some other relation at level i, and could be expected to behave as a final non-2, in particular, as a final 8. Significantly then, in the relevant raising construction, final 2s are permitted to raise, as are final 1s, 3s, benefactives and instruments. But, crucially, final 8s are not. Jake illustrates these facts at length, showing in particular, that final 8s which are earlier 1 s of passives and final 8s which are earlier 2s of active 32A structures are not raisable to 1. The question then is to consider the raisability of the earlier 2 in a passivized ditransitive clause in which the 3 is final 1. The IQ raising to 1 construction is based on adjectives like sinchi 'tough', ishta 'strange', and aspectual Vs like chaya 'begin' and tucu 'happen, become'. According to Jake, these predicates take sentential 1s, and the raising is thus out of sentential 1s. The fact that it is raising to 1 thus conforms to the RG law. The Relational Succession Law; see Perlmutter and Postal (1983b). 2 A typical pair showing unraised and raised examples is [10], in which the raised nominal is the final 1 of the complement.3 Following Jakes' usage, complements are bracketed in both raised and unraised structures:
Most notably, whereas the main V in [10a] is third person in agreement with its final 1, which is the sentential complement, that in [10b] is second person in agreement with the raised nominal. Moreover, the second person nominal in [10b] is separated from the remanant of the clause of which it is understood as the final 1. Considerable other evidence based on validators, reference morphology and cliticization leaves little doubt that what is called the raised nominal is a surface main clause constituent. So far, the IQ raising to 1 construction does not seem that different from e.g. the English raising to 1 construction with seem, likely, etc. That is, it is not much different than the minimal distinctions imposed by independently existing differences between IQ and English in word order, case marking and other morphological properties. However, the IQ raising to 1 construction does differ radically from that in English in one fundamental and critical respect. As Jake shows in detail, the class of nominals which can raise to 1 in this construction include not only final 1s as in English but also (and only) final 2s, 3s, instruments and benefactives. Some examples include [11]-[13]:
< previous page
page_250 If you like this book, buy it!
next page >
< previous page
page_251
next page > Page 251
In [11], a final 2 has raised, in [12] a final 3 and in [13] a final instrument. The behaviour of these raised nominals in the main clause is the same as that of raised 1s. As Jake shows, these nominals are also final 1s of the main clause. The fact that all raising to 1 is a unified phenomenon is shown additionally by the fact that the Vs which allow one type of nominal to raise are just those which allow the others. A V which allows 1s to raise, allows 2s, 3s, instruments and benefactives to do so as well. Further, if a V does not allow one of these latter four types to raise, it does not allow 1s to raise either. Moreover, regardless of what type of element raises, it can only raise into the immediately higher clause, as in the analogous English construction with seem. Finally, in all cases, the morphological structure of the complement permitting raising is the same. The existence of a unified raising to 1 construction not limited to final l s is therefore hardly in doubt. The most important point for current purposes is that despite the range of final complement relations manifested by raised nominals, not just any nominals can raise. In fact, this is only possible for the five final relations just mentioned. Jake shows this in detail. Hence, for example, locatives, ablatives, and comitatives cannot raise. Most crucially, it is then also predictably impossible for final 8s to raise. This is illustrated by [14] for the final 8/earlier I of a passive and by [15] for the final 8/earlier 2 of an active 32A structure:
< previous page
page_251 If you like this book, buy it!
next page >
< previous page
page_252
next page > Page 252
Given then the existence of a raising to 1 construction in which final 2s can raise but final 8s cannot, the crucial question is to determine the raisability of the initial 2 of a passivized ditransitive clause in which the 3 is the final 1. If this 3 is an intermediate 2, the earlier 2 should be a final 8, hence unraisable. But if the 3 advances directly to 1, the earlier 2 can be a final 2 and hence raisable. An example of such a passive is [16a] and, as [16b] shows, it is raisable:
[16b] thus shows that a generalization would be lost if one did not recognize direct 3 to 1 advancement in IQ. I conclude that [16a] should be analyzed as such a case and thus that grammatical theory must recognize G-passives in which the prepassive arc is not a 2 arc. This suggests that in contrast to NLs like English which have a rule like [3], IQ specifies more liberally [17]:
Although there is no argument for instruments parallel to that for 3s in [16], it seems plausible that instruments also advance directly to 1. This permits elimination of a rule like [8] at the mere cost of the final conjunct of the consequent of [17]. It should be observed that Jake (1983: 101-102) draws a partially unfounded conclusion about facts like [16]. She infers that examples like [16a] are structurally unambiguous and have only the direct 31A analysis. This is concluded, however, solely from facts like: [16b]. But [16b] fails to determine this consequence. All that it shows is that on at least one analysis,
< previous page
page_252 If you like this book, buy it!
next page >
< previous page
page_253
next page > Page 253
the passive clause has a 31A structure. Nothing precludes it having also a 32A structure in which the 2 arc local successor is a prepassive arc. In fact, barring other arguments, the simplest grammar will indeed predict this dual structure for any V which independently permits 32A. To conclude the discussion, IQ seems to provide solid grounds for the conclusion that not all prepassive arcs are 2 arcs. Most interesting about the argument that 3s can advance directly to 1 is the fact that this conclusion can be supported even though IQ permits 32A in actives and thus apparently has no 'need' to allow 31A. That is, apparently all the simple sentences permitted by direct 31A are also permitted by 32A and passive with a prepassive 2 arc. Nonetheless, the raising facts indicate that such sentences also are characterized by structures with 31A. 7.4 The Scope of Non-2 Prepassive Arcs Although it has been argued that NLs permit non-2 prepassive arcs, it remains far from clear exactly what the scope of this phenomenon is. The IQ cases involve PL-PPs. Are RF-PPs possible in which the prepassive arc is not a 2 arc? Only further research can answer this question. Consider though the interactions between non-2 prepassive arcs and IPs. If the RG Nuclear Dummy Law, which precludes non-nuclear term dummy arcs, were correct, then there would be no IPs with non-2 prepassive arcs. For in terms of Chapter 4, each IP involves an unstable ghost as a prepassive arc, and all ghost arcs are dummy arcs. However, it may be, as suggested in Chapter 6, note 22, that the Nuclear Dummy Law and its APG analog have to be weakened. It could still be correct though, as suggested in the previous Chapter, that a principle of the form [18] is a truth of grammatical theory:
This constraint is consistent even with the cases of Chapter 6 taken to suggest that not all central dummy arcs are nuclear term arcs. For those cases involved 5 arc local successors of ghost 2 arcs, not 5 arc ghosts. If something like [18] is a viable law or theorem of other principles, as appears consistent with currently available evidence, then the analysis of IPs in Chapter 4 entails that all IPs have 2 arc ghosts; the domain of passives with non-2 prepassive arcs is then exclusively some subdomain of PPs. Notes: Chapter 7 1. The suffix glossed 'wit' (witness), is one of the so-called 'validators'. These indicate certain information about the speaker's beliefs as to the truth of the sentence, as well as focus or new information. The 'wit' suffix apparently indicates that the speaker has first hand knowledge of the assertion.
< previous page
page_253 If you like this book, buy it!
next page >
< previous page
page_254
next page > Page 254
2. My own guess would be, however, that the relevant predicates take initial (sometimes unaccusative) 2s, and that the raising is to 2, with subsequent unaccusative advancement to 1. If no raising takes place, then the initial complement 2 advances to 1; in each case, the advancement is forced by the requirement that a basic clause have a final I are. But the possibility suggested here has no bearing on the argument in the text. 3. The verbal suffixes ngapaj and chun are reference markers. The former indicates roughly that the complement and main clauses have final Is with the same reference; the latter that these final 1s differ in reference. Naturally then, in a raising structure, only the former is possible on the subordinate clause. For reasons discussed by Jake, however, either marker is, curiously, possible on the subordinate clause if no raising occurs.
< previous page
page_254 If you like this book, buy it!
next page >
< previous page
page_255
next page > Page 255
References Aissen, J. (1974a) The Syntax of Causative Constructions Harvard University Doctoral Dissertation, Cambridge, Massachusetts. Aissen, J. (1974b) 'Verb Raising', Linguistic Inquiry 5: 325-366. Aissen, J. (1979) 'Possessor Ascension in Tzotzil', in L. Martin (ed.), Papers in Mayan Linguistics, Lucas Brothers Publishers, Columbia, Missouri. Aissen, J. (to appear) 'Theme and Absolutives: Some Semantic Rules in Tzotzil', in Cook and Gerdts (to appear). Aissen, J. (1983) 'Indirect Object Advancement in Tzotzil', in Perimutter (1983a). Aissen, J. and J. Hankamer (1980) 'Lexical Extension and Grammatical Transformations', in Proceedings of the Sixth Annual Meeting Berkeley Linguistics Society, Berkeley, California. Aissen, J. and D. M. Perlmutter (1976) 'Clause Reduction in Spanish', in Proceedings of the Second Annual Meeting Berkeley Linguistics Society, Berkeley, California. Aissen, J. and D. M. Perlmutter (1983) 'Clause Reduction in Spanish', in Perlmutter (1983a). Alexander, D. and W. J. Kunz (1964) 'Some Classes of Verbs in English', Unpublished report, Linguistics Research Project, Indiana University, Bloomington, Indiana. Allen, J. B. (1874) An Elementary Latin Grammar, Clarendon Press, Oxford, England. Allen, S. (ed.) (1982) Proceedings of the Nobel Symposium on Text Processing, Almqvist and Wiksell, Stockholm, Sweden.
< previous page
page_255 If you like this book, buy it!
next page >
< previous page
page_256
next page > Page 256
Awbery, G. M. (1976) The Syntax of Welsh, Cambridge University Press, Cambridge, England. Bach, E. (1980) 'In Defense of Passive', Linguistics and Philosophy 3: 297-342. Bell, S. (1976) Cebuano Subjects in Two Frameworks, MIT Doctoral Dissertation, Cambridge, Massachusetts. Bell, S. (1983) 'Advancements and Ascensions in Cebuano', in Perlmutter (1983a). Bever, T., J. Carroll and L. Miller (eds.) (1984) Talking Minds: The Study of Language in Cognitive Science, The MIT Press, Cambridge, Massachusetts. Bolinger, D. (1975) 'On the Passive in English', in A. Makkai and V. Makkai, eds., The First LACUS Forum, Hornbeam Press, Columbia, South Carolina. Boons, J.-P., A. Guillet and C. Leclère (1973) La Structure des phrase simples en français, Laboratoire d'Automatique Documentaire et Linguistique, University of Paris VIII, Paris. Bresnan, J. (1982a) 'The Passive in Lexical Theory', in Bresnan (1982b). Bresnan, J. (1982b) 'The Mental Representation of Grammatical Relations, The MIT Press, Cambridge, Massachusetts. Burzio, L. (1981) Intransitive Verbs and Italian Auxiliaries, MIT Doctoral Dissertation, Cambridge, Massachusetts. Burzio, L. (1983) 'Conditions on Representation in Romance Syntax', Linguistic Inquiry 14: 193-221. Chomsky, N. (1965) Aspects of the Theory of Syntax, The MIT Press, Cambridge, Massachusetts. Chomsky, N. (1975) Reflections on Language, Pantheon Books, New York. Chomsky, N. (1977) Language and Responsibility, Pantheon Books, New York. Chomsky (198 l) Lectures on Government and Binding, Foris Publications, Dordrecht, Holland. Chung, S. (1976a) 'An Object-Creating Rule in Bahasa Indonesia', Linguistic Inquiry 7: 41-88; also in Perlmutter (1983a). Chung, S. (1976b) 'The Subject of Two Passives in Indonesian', in Li (1976). Cole, P. and J. M. Sadock (1977) Syntax and Semantics 8: Grammatical Relations, Academic Press, New York. Couper-Kuhlen, E. (1979) The Prepositional Passive in English, Max Niemeyer Verlag, Tubingen, West Germany. Couquaux, D. (1981) French Predication and Linguistic Theory', in May and Koster (1981). Comrie, B. (1977) 'In Defense of Spontaneous Demotion: The Impersonal Passive', in Cole and Sadock (1977). Cook, E. D. and D. Gerdts (eds.) (to appear) Syntax and Semantics 16: The Syntax of Native American Languages, Academic Press, New York. Craig, C. (1977) The Structure of Jacaltec, University of Texas Press, Austin, Texas. Crain, C. (1979) 'Advancement and Ascension to Direct Object in Chainotto', in Linguistic Notes from La Jolla 6, University of California at San Diego. Curme, G. (1952) A Grammar of the German Language, Frederick Ungar Publishing Company, New 'York. Davies, W. D. (1981a) Choctaw Clause Stracture, University of California at San Diego Doctoral Dissertation, La Jolla, California. Davies, W. D. (1981b) 'Choctaw Subjects and Multiple Levels of Syntax', in T. Hoekstra et. al. (1981).
< previous page
page_256 If you like this book, buy it!
next page >
< previous page
page_257
next page > Page 257
Davies, W. D. (1981c) 'Choctaw Switch Reference and Levels of Syntactic Representation', in Cook and Gerdts (to appear). Davies, W. D. (1982) '2-3 Retreat, The Notion 'Absolutive', and Levels of Grammatical Relations', in Flickinger et. al. (1982). Davies, W. D. (1984) 'Antipassive: Choctaw Evidence for a Universal Characterization', in Perlmutter and Rosen (1984). Dowty, D. (1982a) 'Grammatical Relations and Montague Grammar', in Jacobson and Pullum (1982). Dowry, D. (1982b) 'More on the Categorical Analysis of Grammatical Relations', in Zaenen (1982); also in Joseph (1982b). Dryer, M. (1982) 'Passive and Inversion in Kannada', Proceedings of the Eighth Annual Meeting Berkeley Linguistics Society, Berkeley, California. Dubinsky, S. and C. Rosen (1983) 'A Bibliography on Relational Grammar Through April, 1983 with Selected Titles on Lexical-Functional Grammar', Distributed by Indiana University Linguistics Club, Bloomington, Indiana. Eckman, F. (1974) 'Agentive and Agentless Passives', Working Papers on Language Universals Number 14, Stanford University, Stanford, California. Emonds, J. (1969) Root and Structure-Preserving Transformations, MIT Doctoral Dissertation, Cambridge, Massachusetts. Emonds, J. (1976) A Transformational Approach to English Syntax, Academic Press, New York. Engdahl, E. and M. J. Stein (eds.) (1979) Studies Presented to Emmon Bach by His Students, The University of Massachusetts, Amherst, Massachusetts. Fauconnier, G. (1974) La coréférence: syntaxe ou sémantique?, Editions du Seuil, Paris. Fauconnier, G. (1983) 'Generalized Union', Communication and Cognition 16: 195-229. Feldman, H. (1978) 'Passivizing on Datives in Greek', Linguistic Inquiry 9: 499-501. Flickinger, D. P. et. al. (eds.) Proceedings of the First West Coast Conference on Formal Linguistics, Stanford University, Stanford, California. Gazdar, G. (1981) 'Unbounded Dependencies and Coordinate Structure', Linguistic Inquiry 12: 155-184. Gazdar, G. (1982) 'Phrase Structure Grammar', in Jacobson and Pullurn (1982). Gazdar, G. and G. K. Pullum (1981) 'Subcategorization, Constituent Order, and the Notion ''Head"', in Moortgat, Hulst and Hoekstra (1981). George, L. and J. Kornfilt (1977) 'Infinitival Double Passives in Turkish', Proceedings of the Seventh Annual Meeting of the North Eastern Linguistic Society, Cambridge, Massachusetts. Gerdts, D. (1981) Object and Absolutive in Halkomelem Salish, University of California at San Diego Doctoral Dissertation, La Jolla, California. Gerdts, D. (to appear) Korean Relational Grammar. Gerdts, D., S. Choi, S. Chun and C. Youn (1982) 'A Constraint on ci-Passives in Korean', unpublished paper, State University of New York at Buffalo, Buffalo, New York. Gibson, J. (1980) Clause Union in Chamorro and in Universal Grammar, University of California at San Diego Doctoral Dissertation, La Jolla, California. Gibson, J. (1982) 'The Chômeur Relation and Multiple Levels', in Flickinger et. al. (1982).
< previous page
page_257 If you like this book, buy it!
next page >
< previous page
page_258
next page > Page 258
Gibson, J. and I. Özkaragöz (1981) 'The Syntactic Nature of the Turkish Causative Construction', in Papers from the Seventeenth Regional Meeting Chicago Linguistic Society, Chicago, Illinois. Gibson, J. and E. Raposo (to appear) 'Clause Union, the Stratal Uniqueness Law and the Chomeur Relation'. Grevisse, M. (1969) Le bon usage, Editions J. Duculot, Gembloux, Belgium. Gross, M. (1968) Grammaire transformationnelle du français, syntaxe du verbe, Librairie Larousse, Paris. Gross, M. (1975) Méthodes en syntaxe, Hermann, Paris. Hankamer, J. and J. Aissen (1976) eds., Harvard Studies in Syntax and Semantics Volume 2, Harvard University, Cambridge, Massachusetts. Hankamer, J. and L. Knecht (1976) 'The Role of the Subject/Nonsubject Distinction in Determining the Choice of Relative Clause Participle in Turkish', in Hankamer and Aissen (1976). Also in Papers from the Sixth Meeting North Eastern Linguistic Society, Montreal Working Papers in Linguistics, University of Montreal, Montreal, Canada. Harris, A. (1981) Georgian Syntax: A Study in Relational Grammar, Cambridge University Press, Cambridge, England. Harris, A. (1984a) 'Case Marking, Verb Agreement and Inversion in Udi', in Perlmutter and Rosen (1984). Harris, A. (1984b) 'Inversion as a Rule of Universal Grammar: Georgian Evidence', in Perlmutter and Rosen (1984). Hermon, G. (1981) 'The Relationship of Meaning and Underlying Grammatical Relations' in Proceedings of the Seventh Annual Meeting Berkeley Linguistics Society, Berkeley, California. Higgins, R. (1973) 'On J. Emonds' Analysis of Extraposition', in J. Kimball, ed., Syntax and Semantics Volume 2, Academic Press, New York. Hoekstra, T., H. van der Hulst and M. Moortgat (eds.) (1980) Lexical Grammar, Foris Publications, Dordrecht, Holland. Hoekstra, T. H., H. van der Hulst and M. Moortgat (eds.) (1981) Perspectives on Functional Grammar, Foris Publications, Dordrecht, Holland. Hornstein, R. and A. Weinberg (1981) 'Case Theory and Preposition Stranding', Linguistic Inquiry 12: 55-91. Hubbard, P. (1979) 'Albanian Neapolitan Morphology: Passive, Multi-Attachment and the Unaccusative Hypothesis', in Linguistic Notes from La Jolla, Number 6, University of California at San Diego. Hubbard, P. (1980) The Syntax of the Albanian Verb Complex, University of California at San Diego Doctoral Dissertation, La Jolla, California. Hubbard, P. (1982) 'Dative Clitics in Albanian: Evidence for Syntactic Levels', in Proceedings of Eighth Annual Meeting Berkeley Linguistics Society, Berkeley, California. Also in Working Papers in Relational Grammar, Department of Linguistics, University of California at San Diego, La Jolla, California. Hukari, T. E. (1980) 'Subjects and Objects in Cowichan', paper presented at the 15th ICSL, Vancouver, B. C. Jackson, C. H. (1982) 'Multi-Level Syntactic Description: Evidence from Tamil', in Working Papers in Relational Grammar, Department of Linguistics, University of California at San Diego, La Jolla, California.
< previous page
page_258 If you like this book, buy it!
next page >
< previous page
page_259
next page > Page 259
Jacobson, P. and G. K. Pullum (eds.) (1982) The Nature of Syntactic Representation, D. Reidel Publishing Company, Dordrecht, Holland. Jain, J. (1977) 'The Hindi Passive in Universal Grammar', unpublished paper, San Francisco State University, San Francisco, California. Jake, J. (1983) Grammatical Relations in Imbabura Quechua, University of Illinois Doctoral Dissertation, Urbana, Illinois. Jespersen, O. (1924) The Philosophy of Grammar, George Allen and Unwin, London. Johnson, D. E. (1977) 'On Relational Constraints on Grammars', in Cole and Sadock (1977). Johnson, D. E. (1979) Toward a Theory of Relationally Based Grammar, Garland Publishing Company, New York. Johnson, D. E. and P. M. Postal (1980) Arc Pair Grammar, Princeton University Press, Princeton, New Jersey. Jolley, C. A. (1982) 'On the Plains Cree Passive: An Analysis of Syntactic and Lexical Rules', in Joseph (1982b). Joseph, B. D. (1982a) 'On Some Advancements to Subject in Greek', in Joseph (1982b). Joseph, B. D. (ed.) (1982b) Grammatical Relations and Relational Grammar, Working Papers in Linguistics Number 26, The Ohio State University, Columbus, Ohio. Katz, J. J. (1981) Language and Other Abstract Objects, Rowman and Littlefield, Totowa, New Jersey. Katz, J. J. (1984) 'An Outline of Platonist Grammar', in Bever, Carroll and Miller (1984). Kayne, R. (1975) French Syntax, The MIT Press, Cambridge, Massachusetts. Kayne, R. (1979) 'Rightward NP Movement in French and English', Linguistic Inquiry 10: 710-719. Kayne, R. (1980) 'Extensions of Binding and Case-Marking', Linguistic Inquiry 11: 75-96. Kayne, R. (1981) 'ECP Extensions', Linguistic Inquiry 12: 93-133. Kayne, R. (1983) 'Le datif en français et en anglais', in Analyses grammaticales du français, Revue Romane Special Number 24, University of Copenhagen, Copenhagen, Denmark. Kayne, R. and J.-Y. Pollock (1978) 'Stylistic Inversion, Successive Cyclicity, and Move NP in French', Linguistic Inquiry 9: 595-622. Keenan, E. (1975) 'Some Universals of Passive in Relational Grammar', in Papers from the Eleventh Regional Meeting Chicago Linguistic Society, Chicago, Illinois. Keenan, E. (1976) 'Remarkable Subjects in Malagasy', in Li (1976). Keenan, E. (1980) 'Passive is Phrasal (Not Sentential or Lexical)', in Hoekstra, van der Hulst and Moortgat (1980). Kirsner, R. S. (1976) 'On the Subjectless "Pseudo-Passive" in Standard Dutch and the Semantics of Background Agents', in Li (1976). Kisseberth, C. W. and M. I. Abasheikh (1977) 'The Object Relationship in ChiMwi:ni: A Bantu Language', in Cole and Sadock (1977). Klokeid, T. (1976) Topics in Lardil Grammar, MIT Doctoral Disssertation, Cambridge, Massachusetts. Knecht, L. (1974) 'Turkish Relative Clause Participles: A Rejoinder and A New Proposal', unpublished paper, MIT, Cambridge, Massachusetts. Knecht, L. (1976) 'Turkish Comparatives', in Hankamer and Aissen (1976).
< previous page
page_259 If you like this book, buy it!
next page >
< previous page
page_260
next page > Page 260
Kornfilt, J. (1976) 'The Cycle Against Free Rule Application', in Hankamer and Aissen (1976). Kuno, S. (1973) The Structure of Japanese, The MIT Press, Cambridge, Massachusetts. Kurylowicz, J. (1946) 'Ergativeness and the Stadial Theory of Linguistic Development', translated from the Russian by P. Cullicover. The Study of Man 2: 1-21 (1973). Langendoen, D. T. (1976) 'On the Weak Generative Capacity of Infinite Grammars', CUNY Forum 1: 13-24. Langendoen, D. T. (1979) 'The Grammatical Analysis of Texts' in Allen (1982). Langendoen, D. T. and P. M. Postal (1984) The Vastness of Natural Languages, Basil Blackwell, Oxford, England. Lawler, J. (1977) 'A Agrees with B in Achenese', in Cole and Sadock (1977). Levin, N. S. (1978) 'Some Identity-of-Sense Deletions Puzzle Me. Do they you?', in Papers from the Fourteenth Regional Meeting Chicago Linguistic Society, Chicago, Illinois. Li, C. N. (ed.) (1976) Subject and Topic, Academic Press, New York. Marantz, A. (1981) On the Nature of Grammatical Relations, MIT Doctoral Dissertation, Cambridge, Massachusetts. May, R. and J. Koster (1981) eds. Levels of Syntactic Representation, Foris Publications, Dordrecht, Holland. McCawley, J. D. (1982) 'Parentheticals and Discontinuous Constituent Structure', Linguistic Inquiry 13: 91-106. McCloskey, J. (1979) Transformational Syntax and Model Theoretic Semantics, D. Reidel Publishing Company, Dordrecht, Holland. Milner, J.-C. (1978) De la syntaxe à l'interprétation, Editions du Seuil, Paris. Moortgat, M., H. Hulst and T. Hoekstra (eds.) (1981) The Scope of Lexical Rules, Foris Publications, Dordrecht, Holland. Nerbonne, J. A. (1982) 'Some Passives Not Characterized by Universal Rules: Subjectless Impersonals', in Joseph (1982b). Norman, W. M. (1978) 'Advancement Rules and Syntactic Change: the Loss of Instrumental Voice in Mayan', in Proceedings of the Fourth Annual Meeting Berkeley Linguistics Society, Berkeley, California. Olie, A. (1983) 'Overt and Silent Dummies in French: A Relational Approach to Impersonal Constructions', unpublished paper, University of California at San Diego, La Jolla, California. Ostler, N. D. M. (1979) Case-Linking: A Theory of Case and Verb Diathesis Applied to Classical Sanskrit, MIT Doctoral Dissertation, Cambridge, Massachusetts. Ostman, J.-O. (1981) 'The Finnish "Passive" and Relational Grammar', in Papers from the Seventeenth Regional Meeting Chicago Linguistic Society, Chicago, Illinois. Özkaragöz, I. (1980) 'Evidence from Turkish for the Unaccusative Hypothesis', Proceedings of the Sixth Annual Meeting Berkeley Linguistics Society, Berkeley, California. Özkaragöz, I. (1982a) 'Monoclausal Double Passives in Turkish', to appear in Proceedings of the Conference on the Turkish Language and Linguistics in Ataturk's Turkey.
< previous page
page_260 If you like this book, buy it!
next page >
< previous page
page_261
next page > Page 261
Özkaragöz, I. (1982b) 'Transitivity and the Syntax of Middle Clauses in Turkish', in Working Papers in Relational Grammar, University of California at San Diego, La Jolla, California. Perlmutter, D. M. (1978a) 'Impersonal Passives and the Unaccusative Hypothesis', Proceedings of the Fourth Annual Meeting Berkeley Linguistics Society, Berkeley, California. Perlmutter, D. M. (1978b) 'Evidence for Inversion in Russian, Japanese and Kannada', unpublished paper, University of California at San Diego. Perlmutter, D. M. (1981) 'Functional Grammar and Relational Grammar: Points of Convergence and Divergence', in Hoekstra et. al (1981). Perlmutter, D. M. (1982) 'Syntactic Representation, Syntactic Levels, and the Notion of Subject', in Jacobson and Pullum (1982). Perlmutter. D.M. (1983a) Studies in Relational Grammar 1, The University of Chicago Press, Chicago, Illinois. Perlmutter. D.M. (1983b) 'Personal Versus Impersonal Constructions', Natural Language and Linguistic Theory 1: 141200. Perlmutter. D.M. (1984a) 'Working 1s and Inversion in Italian, Japanese, and Quechua', in Perlmutter and Rosen (1984). Perlmutter. D.M. (1984b) 'The Inadequacy of Some Monostratal Theories of Passive', in Perlmutter and Rosen (1984). Perlmutter D. M. (to appear) 'Multiattachement and the Unaccusative Hypothesis: the Perfect Auxiliary in Italian'. Perlmutter D. M. and P.M. Postal (1977) 'Toward a Universal Characterization of Passivization', Proceedings of the Third Annual Meeting Berkeley Linguistics Society, Berkeley, California. Perlmutter, D. M. and P.M. Postal (1983a) 'Some Proposed Laws of Basic Clause Structure', in Perlmutter (1983a). Perlmutter, D. M. and P.M. Postal (1983b) 'The Relational Succession Law', in Perlmutter (1983a). Perlmutter, D.M. and P.M. Postal (1984a) 'The l-Advancement Exclusiveness Law', in Perlmutter and Rosen (1984). Perlmutter, D. M. and P.M. Postal (1984b) 'Impersonal Passives and Some Relational Laws', in Perlmutter and Rosen (1984). Perlmutter, D. M. and C. Rosen (eds.) (1984) Studies in Relational Grammar 2, The University of Chicago Press, Chicago, Illinois. Perlmutter, D. M. and A. Zaenen (1984) 'The Indefinite Extraposition Construction in Dutch and German', in Perlmutter and Rosen (1984). Pollock, J.-Y. (1981) 'On Case and Impersonal Constructions', in May and Koster (1981). Postal, P.M. (1974) On Raising, The MIT Press, Cambridge, Massachusetts. Postal, P.M. (1977) 'Antipassive in French', Lingvisticae Investigationes 1: 333-74. Postal, P.M. (1979) Some Syntactic Rules in Mohawk, Garland Publishing Company, New York. Postal, P.M. (1982) 'Some Arc Pair Grammar Descriptions', in Jacobson and Pullum (1982). Postal, P.M. (to appear a) 'A French Construction Which Isn't What it Seems'. Postal, P.M. (to appear b) 'Masked Advancement'. Postal, P.M. (to appear c) French Indirect Objects and Semi-Objects.
< previous page
page_261 If you like this book, buy it!
next page >
< previous page
page_262
next page > Page 262
Pullum, G. K. (1982) 'Free Word Order and Phrase Structure Rules', in Proceedings of the Twelfth Meeting: North Eastern Linguistic Society. Pullum, G.K. (1982) 'Syncategorematicity and English Infinitival To" Glossa 16: 181-215. Riddle, E. and G. Sheintuch (1983) 'A Functional Analysis of' Pseudo-Passives', Linguistics and Philosophy 6: 527563. Riemsdijk, H. van (19'78) A Case Study in Syntactic Markedness: The Binding Nature of Prepositional Phrases, Foris Publications, Dordrecht, Holland. Rizzi, L. (1982) Issues in Italian Syntax, Foris Publications, Dordrecht, Holland. Roberts, J. S. (1980) French Causatives in Generative Syntax, Georgetown University Doctoral Dissertation, Washington, D.C. Rosen, C. (1981a) The Relational Structure of Reflexive Clauses: Evidence from Italian, Harvard University Doctoral Dissertation, Cambridge, Massachusetts. Rosen, C. (1981b) 'The Interface Between Semantic Roles and Initial Grammatical Relations', in Zaenen (1981) and in Perlmutter and Rosen (1984). Rosen, C. (1982) 'The: Unaccusative Hypothesis and the "Inherent Clitic" Phenomenon in Italian', in Papers from the Eighteenth Regional Meeting Chicago Linguistic Society, Chicago, Illinois. Rosenbaum, P. (1967) The Grammar of English Predicate Complement Constructions, The MIT Press, Cambridge, Massachusetts. Ross, J. R. (1967) Constraints on Variables in Syntax, Doctoral Dissertation, MIT, Cambridge, Massachusetts. Ross, K. (1981) 'Reanalysis Reconsidered', unpublished paper, Texas Instruments Corporation, Dallas, Texas. Ruwet, N. (1972) Théerie syntaxique et syntaxe du français, Editions du Seuil, Paris, France. Ruwet, N. (1982) Grammaire des insultes et autres études, Editions du Seuil, Paris, France. Saad, G. N. (1979) 'Passivization', in Engdahl and Stein (1979). Shibatani, M. (1977) 'Grammatical Relations and Surface Cases', Language 53: 789-809. Smith, M. (1982) 'Subjecthood and Ephemeral Dummies in German', unpublished paper, University of California at San Diego, La Jolla, California. Sridhar, S. (1976a) 'Dative Subjects', Papers from the Twelfth Regional Meeting Chicago Linguistic Society, Chicago, Illinois. Sridhar, S. (1976b) 'Dative Subjects, Rule Government and Relational Grammar', Studies in the Linguistic Sciences 6:130-151. Sweet, H. (1898) A New English Grammar, The Clarendon Press, Oxford, England. Timberlake, A. (1982), 'The Impersonal Passive in Lithuanian', in Proceedings of the Eighth Annual Meeting Berkeley Linguistics Society, Berkeley, California. Trithart, L. (1975) 'Relational Grammar and Chichewa Subjectivization Rules', in Papers from the Eleventh Regional Meeting Chicago Linguistic Society, Chicago, Illinois. Trithart, L. (1976) Relational Grammar and Chichewa Subjectivization Rules, University of California at Los Angeles Masters Dissertation, Los Angeles, California. Trithart, L. (1979) 'Topicality: An Alternative to the Relational View of Bantu Passive', Studies in African Linguistics 10: 1-30.
< previous page
page_262 If you like this book, buy it!
next page >
< previous page
page_263
next page > Page 263
Wachtel, T. (1979) 'The Demotion Analysis of Initially Unaccusative Impersonal Passives', in Papers from the Fifteenth Regional Meeting Chicago Linguistic Society, Chicago, Illinois. Williams, E. (1980)'Predication', Linguistic Inquiry 11:203-238. Williams, E. (1981) 'Argument Structure and Morphology', The Linguistic Review 1: 81-114. Zaenen, A. (1981) Subjects and Other Subjects, Indiana University Linguistics Club, Bloomington, Indiana. Zimmer, K. (1976) 'Some Constraints on Turkish Causativization', in M. Shibatani, ed., Syntax and Semantics Volume 6, Academic Press, New York. Ziv, Y. and G. Sheintuch (1981) 'Passives of Obliques Over Direct Objects', Lingua 54: 1-17. Zribi-Hertz, A. (1980) 'Presentation', Langue Française 46: 5-7. Zribi-Hertz, A. (1981) 'Towards a Transformationally-Expressed Explanation of Passive Verbal Morphology in French and English', Indiana University Linguistics Club, Bloomington, Indiana. Zribi-Hertz, A. (1982) 'La construction 'se Moyen' du français et son statut dans le triangle moyen-passif-réfléchi', Lingvisticae Investigationes 6: 245-401.
< previous page
page_263 If you like this book, buy it!
next page >
< previous page
page_264
next page > Page 264
Index A ABASHEIKH, M. 66 ABSOLUTIVE (ARC) 147, 157 ACHENESE 66, 88, 139 ACTING (TERM ARC) 47, 48, 86, 87, 229 ADVANCE(MENT) 18, 19, 20 AINU 9 AISSEN, J. 5, 18, 83, 91, 148 ALBANIAN 159 ALEXANDER, D. 225 ALLEN, J. 159 ANCESTRAL 34, 43, 64 ANCIENT GREEK 43-45, 46, 50, 51, 52, 90 ANDEAN xi ARABIC 9, 134 ARC 7, 8, 15-23 AWBERY, G. 11, 132 B BACH, E. 138, 202, 225, 239 BASIC CLAUSE 199 BELL, S. 90 BENEFACTIVE TO 2 ADVANCEMENT (= BEN2A) 62, 63, 69, 71, 77, 78 BOLINGER, D. 242 BOONS, J.-P. 169 BRESNAN, J. 10, 204, 205, 208, 216-220, 239 BURMESE 133
< previous page
page_264 If you like this book, buy it!
next page >
< previous page
page_265
next page > Page 265
BURZIO, L. 106, 180, 201 C CEBUANO 90 CHI-MWI:NI 34, 66-84, 91, 93 CHICHEWA 34, 56-66, 67, 68, 69, 75, 78, 79, 80, 83, 84, 87, 91 CHO ARC NO LOCAL SUCCESSOR LAW, THE 228 CHOCTAW 133 CHOI, S. 52 CHOMEUR ADVANCEMENT BAN, THE 228 CHOMEUR LAW, THE 24, 91 CHOMSKY, N. l, 3, 7, 10, 96, 204, 215, 241 CHUN, S. 52 CHUNG, S. 18 CLAUSE UNION 148, 149, 150 CLEAR 135, 242, 244 COMRIE, B. 9, 10 COORDINATES (OF ARCS) 8, 15, 19, 20 COPIES 211, 215, 216, 220, 221, 240, 242 COPY ADVANCEMENT 210, 211, 212, 213, 214, 222, 223, 240 COPY ARC 124, 210, 213, 214, 215, 221, 222, 240, 242 COUPER-KUHLEN, E. 238, 243 COUQUAUX, D. 180 CRAIG, C. 55 CRAIN, C. 18 CURME, G. 197 CZECH 9 D DAVIES, W. 65, 88, 102, 232 DEMOTE 19 DISPLACE 78, 79, 93 DOWTY, D. 66, 91, 200 DRYER, M. 88 DUBINSKY, S. 8, 18 DUMMY ARC 89, 98, 116, 117, 118, 119, 120, 132, 133, 137, 175, 177, 191, 226, 229, 235, 236, 253 DUMMY (NOMINAL) 10, 93, 95, 96, 97, 98, 99, 100, 101, 103, 104, 107, 108, 109, 110, 111, 112, 114, 117, 118, 119, 120, 122, 127, 128, 129, 130, 131, 132, 137, 142, 144, 146, 151, 159, 165, 168, 169, 170, 171, 176, 178, 187, 191, 194, 195, 200, 228, 237, 244, 245 DUTCH 9, 141 E ECKMAN, F. 139 EGRESSOR (ARC) 132, 244 EMONDS, J. 225, 233 ENGLISH x, xi, 13, 17, 20, 21, 23, 24, 45, 47, 53, 78, 80, 83, 85, 86, 91, 93, 94, 96-99, 100, 103, 104, 105, 131, 132, 133, 134, 135, 138, 158, 159, 160, 168, 175, 181, 200, 202, 203-245, 247, 248, 250, 251, 252
ENTRANT (ARC) 132 ERASE 16, 17, 19, 20
< previous page
page_265 If you like this book, buy it!
next page >
< previous page
page_266
next page > Page 266
ESTONIAN 165, 198 EXTRAPOSITION OF INDEFINITE (FRENCH) (= EXI) 88, 100, 102, 103, 114, 115, 117, 118, 120, 127, 128, 136, 143, 165, 170, 183, 184 EXTRAPOSED/EXTRAPOSITION (ENGLISH) 96, 105, 226-237, 243, 244, 245 F FACSIMILE (ARC) 40, 48, 210 FAUCONNIER, G. 18, 89, 242 FELDMAN, H. 43, 44, 45, 46 FINAL ! ARC LAW, THE 143, 151 FINNISH 9, 139 FIRST 229, 231, 234 FOREIGN (SUCCESSOR) (ARC) 18 FRANZ, D. 18 FREE (ARC) 22, 40, 75, 90, 138, 148, 173, 229 FRENCH 4, 9, 10, 11, 24, 25, 31, 32, 33, 35-43, 44, 45, 50, 51, 52, 56, 88, 89, 90, 91, 94, 99-103, 105, 106, 112-123, 125-131, 132, 133, 134, 135, 136, 137, 138, 139, 158, 164, 165, 169, 170, 171, 175, 178-184, 195, 200, 201, 204, 210, 211, 212, 228, 242, 244 FULL (ARC) 231, 234 G GAZDAR, G. 25 GENERALIZED PASSIVE CLAUSE (= G-PASSIVE CLAUSE) 45, 50, 51, 246, 247 GEORGE, L. 149, 150, 151, 199 GERDTS, D. 52, 84, 85, 86, 90, 91 GERMAN 9, 11, 25, 133, 134, 138, 178, 179, 182, 192-197, 201, 204, 211 GHOST (ARC) 97,, 99, 105, 109, 110, 111, 112, 113, 114, 115, 116, 117, 122, 128, 129, 130-132, 136, 137, 142, 143, 145, 146, 147, 148, 151, 153, 157, 159, 164, 168, 169, 170, 171, 174, 175, 176, 177, 178, 182, 183, 186, 187, 236, 244, 253 GIBSON, J. 18, 148 GHOST ARC LAW, THE 130, 131, 143 GRAFT OVERLAP LAW, THE 244 GREVISSE, M. 6, :24 GROSS, M. 125, 180, 244 GUILLET, A. 169 H HALKOMELEM 34, 84-87, 91 HANKAMER, J. 148, 151, 153, 155 HARRIS, A. 88 HERMON, G. 18 HIGGINS, R. 225 HORNSTEIN, N. 204, 205 HUBBARD, P. 88, 102, 159 HUKARI, T. 85 I IMITATION (TERM ARC) 87 IMMED(IATE) ARC PRECEDE 220, 222
IMPERSONAL. INVERSION 88 IMPERSONAL PASSIVE (= IP) 8-11, 24, 25, 95-104, 108-132, 140-198, 253
< previous page
page_266 If you like this book, buy it!
next page >
< previous page
page_267
next page > Page 267
INCORPORATE/INCORPORATION 151, 152, 153, 155, 156, 157 INDONESIAN 4 INDUCE 63, 68, 69, 73, 78, 79, 83, 91, 93 INEXPLICIT NONGRAFT LAW, THE 244 INSECURE (ARC) 148, 149, 151, 153, 173, 174, 176, 226 INSTABILITY DUMMY ARC 116, 117, 118, 175 INSTRUMENTAL TO 2 ADVANCEMENT (= INST2A) 52, 54, 55, 56, 249 INTERNAL SURVIVOR (ARC) 200 INVERSION (SEE I TO 3 DEMOTION) 88, 107, 113, 120 IRISH 165, 190-192, 198, 202 ITALIAN 198, 199, 201 J JACKSON, C. 88, 102 JAIN, J. 10 JAKE, J. 44, 47, 244, 247, 248, 249, 250, 251, 252, 254 JAPANESE 44, 90, 133 JESPERSEN, O. 6 JOHNSON, D. 18, 174, 204, 205, 246 JOLLEY, C. 139 JOSEPH B. 18, 45, 46, 47, 50 K KATZ, J. 1 KAYNE, R. 117, 118, 127, 137, 180, 201 KEENAN, E. 5, 10, 105, 139 KINYARWANDA 44 KIRSNER, R. 9 KISSEBERTH, C. 66 KLOKEID, T. 135 KNECHT, L. 151, 153, 155, 199 KOREAN 50, 51, 52, 90 KORNFILT, J. 147, 149, 150, 151, 156, 199 KUNO, S. 90 KUNZ, W. 225 KURYLOWICZ, J. 6 L LANGENDOEN, D. T. 1, 2, 12, 13, 23, 25, 26 LARDIL 135 LAST 48, 49 LATIN 3,9, 159 LECLERE, C. 169 LEVIN, N. 208 LIFTER (ARC) 25, 26
LITHUANIAN 6, 157, 165, 166-190, 191, 192, 193, 198 LOCAL (RELATIONS BETWEEN ARCS) 14, 15, 18, 19, 20 LOGICAL GRAPH (= L-Graph) 19 LONG PASSIVE 135, 139, 184 LONG PP 166 M MAASAI 9, 133, 135 MALAGASY 5
< previous page
page_267 If you like this book, buy it!
next page >
< previous page
page_268
next page > Page 268
MANDARIN 133 MARANTZ, A. 10, 90, 160 MARQ CLOSURE 213, 214, 215, 221, 222, 242 MCCAWLEY, J. 199 MCCLOSKEY, J. 202 MILNER, J.-C. 180 MODERN GREEK (= MGREEK) 45-50 MOHAWK 152 MORIN, Y.-C. 89 MOTIVATED CHOMAGE LAW, THE 228 MUECKSTEIN, E. 196, 201, 202 N NEIGHBOR 18, 20 NERBONNE, J. 165, 166, 167, 168, 169, 170, 174, 175, 176, 184, 187, 188, 190, 191, 192, 193, 194, 195, 197, 202 NODE 7, 8, 15, 16, 17, 18, 19, 20, 21 NONTERMINAL NODES 15 NORMAN, W. 52, 53, 54, 55, 56 NORTH RUSSIAN 9 NUCLEAR DUMMY LAW, THE 244, 253 NUCLEAR TERM ARCS 14 O OLIE, A. 136 OSTLER, N. 159, 160, 161, 162, 163, 164, 165 OSTMAN, J.-O. 139 OUTLAST 48, 73, 76, 77, 78, 90 OUTPUT (ARC) 105 OVERLAP 18, 244 OVERLAY ARC 55 OVERRUN 20, 109, 112, 114, 134, 135, 136, 138, 142, 143, 144, 148, 173, 174, 216, 224, 226, 229, 231, 241, 242, 244, 246, 247 OZKARAGOZ, I. 141, 143, 144, 145, 148, 149, 150, 151, 155, 156, 197, 199 P PAIR NETWORK (= PN) 12, 15, 16, 17, 19, 20, 21 PASSIVE-DESTROYED (ARC) 109, 112, 136, 247 PASSIVE-INDUCED (ARC) 136 PERLMUTTER, D. 6, 10, 18, 19, 44, 47, 88, 91, 92, 93, 94, 95, 100, 102, 122, 127, 133, 140, 141, 142, 165, 167, 169, 177, 185, 193, 197, 198, 199, 200, 228, 244, 250 PERSONAL PASSIVES (= PP) 9, 10, 11, 25, 43, 46, 50, 51, 52, 53, 55, 56, 57, 81, 84, 85, 86, 88, 95, 96, 100, 101, 102, 104, 108, 112, 114, 117, 118, 122, 123, 125, 126, 127, 128, 131, 133, 134, 135, 137, 138, 139, 142, 144, 145, 147, 148, 155, 156, 158, 159, 160, 161, 163, 165, 166, 167, 168, 169, 170, 173, 174, 182, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 197, 204, 209, 247, 248, 253 PL-ARC-PASSIVE (ARC) 125 PL-IP 123, 125, 127, 128, 130, 135 PL-PASSIVE 122, 123, 124, 126, 128, 129, 135 PL-PP 123, 253
PL-PREPASSIVE (ARC) 125 POLLOCK, J.-Y. 117, 118, 127, 137, 181, 182, 201 POSTAL, P. 6, 10, 12, 13, 15, 17, 18, 19, 23, 25, 26,
< previous page
page_268 If you like this book, buy it!
next page >
< previous page
page_269
next page > Page 269
44, 45, 47, 48, 87, 88, 90, 91, 92, 93, 94, 95, 100, 122, 127, 133, 140, 152, 169, 177, 193, 197, 200, 201, 228, 232, 242, 244, 250 PREDECESSOR (ARC) 15, 18, 19 PREPASSIVE (ARC) 78, 81, 83, 90, 91, 109, 110, 111, 112, 115, 117, 118, 120, 121, 124, 125, 128, 129, 133, 134, 135, 138, 142, 147, 149, 222, 229, 231, 246, 247, 248, 252, 253 PRIMARY PASSIVE 81 PRINCIPLE OF INITIAL DETERMINATION (= PID) 140 PSEUDO-PASSIVE (= PS-P) 10, 88, 160, 202, 203-241, 242, 243 PULLUM, G. 25 PURE (ARC) 216, 220, 222, 224, 225, 226, 227, 231, 237, 239, 241, 242 Q QUASI-OBJECT 45, 136 QUECHUA (= IQ) 44, 47, 244, 247-253 QUICHÉ 34, 52-56, 70 R R-PREDECESSOR (ARC) 55, 56 R-SIGN (OF ARCS) 7, 8, 14, 15, 19, 20, 105, 110, 112, 124, 130, 136, 152, 212, 213, 214, 228, 246 R-SUCCESSOR (ARC) 34, 38, 39, 43, 51, 55, 56 RAISING 17, 18, 120-122 RAPOSO, E. 18 REFLEXIVE PASSIVE 93, 122-128, 163, 164 RELATIONAL ANNIHILATION LAW, THE 24, 91 RELATIONAL GRAMMAR (= RG) ix, 6, 7, 8, 9, 10, 11, 15, 16, 18, 19, 20, 24, 43, 47, 50, 52, 85, 87, 88, 91, 95, 96, 99, 100, 101, 102, 103, 104, 108, 109, 110, 112, 113, 122, 142, 151, 152, 159, 160, 164, 174, 185, 187, 188, 190, 192, 246, 250, 253 RELATIONAL NETWORKS (= RN) 7, 15, 16, 19, 95 RELATIONAL SUCCESSION LAW, THE 250 REPLACE 22, 23, 89, 90, 124, 200, 201, 210, 213, 215, 221 RESIDUE 221, 222, 223 RF-ARC-PASSIVE(ARC) 125, 135 RF-IP 125, 126, 127, 128, 130, 135 RF-IP-ARC-PASSIVE (ARC) 125, 135 RF-PASSIVE 122, 124, 125, 127, 128, 129, 135, 195 RF-PP 123, 125, 126, 127, 128, 135, 194, 195, 197, 253 RF-PREPASSIVE (ARC) 125 RIDDLE 238, 243 RIEMSDYK, H. van 204 RIZZI, L. 190 ROBERTS, J. 139 ROSEN, C. 8, 18, 19, 102, 140, 198 ROSENBAUM, P. 232, 233, 234, 236, 237, 244 ROSS, J. 207 ROSS, K. 207, 208 RUSSIAN 185
RUWET, N.89, 101, 102, 123, 137, 139, 164, 180, 195 S SAAD, G. 9, 134 SALISH xi
< previous page
page_269 If you like this book, buy it!
next page >
< previous page
page_270
next page > Page 270
SANSKRIT 6, 159-165, 198 SECONDARY PASSIVE 81, 82 SECURE (ARC) 138, 148, 229, 230, 231 SELF-ERASE/SELF-ERASURE 98, 105, 118, 119, 120, 121, 122, 132, 133, 146, 147, 153, 169, 170, 200, 215, 216, 220, 221, 226, 230, 231, 234, 235, 236 SELF-SPONSOR 19 SEMI-OBJECT 137, 228 SHEINTUCH, G. 238, 242, 243 SHIBATANI, M. 50, 51 SHORT PASSIVE 24, 134, 135, 139, 174 SMITH, M. 25 SPONSOR 16, 17, 18, 19 SRIDHAR, S. 88 STABLE 112, 113, 115, 116, 117, 120, 157, 169, 170, 171, 174, 175, 176, 178, 183 STRATAL UNIQUENESSS LAW, THE 69, 74, 83, 92, 114 SUCCESSOR (ARC) 14, 15, 18, 19, 20, 25 SURFACE ARC 19, 20, 21 SURFACE GRAPH (= S-GRAPH) 19, 132 SURFACE MATE 21 SVARTIK 238 SWAHILI 66 SWEET, H. 6 T TERM ARC 14, 47, 48 TERMINAL NODES 15 TERTIARY PASSIVE 66, 80, 81, 82, 83, 90, 93, 94 THREATEN 63 TIMBERLAKE, A. 102, 184, 185, 187, 188, 189, 190, 191, 197 TRITHART, L. 56, 57, 58, 62, 64, 65, 66, 79 TURKISH 6, 9, 141-157, 158, 159, 161, 198, 199 TZOTZIL 5, 83, 91, 133 U UNACCUSATIVE 8, 20, 24, 140-199, 200, 238, 243, 254 UNERGATIVE 141, 142, 145, 146, 147, 148, 156, 159, 168, 171, 183, 193, 196, 197, 198, 199, 243 UNIQUE ERASER LAW, THE 94, 153, 199, 200 UNIVERSAL ALIGNMENT HYPOTHESIS, THE 140, 141, 195 UNIVERSAL SENTENCE (= US) 12, 13 UNSTABLE 112, 116, 117, 118, 121, 122, 124, 130, 134, 142 V VERBAL ARC 151 W WACHTEL, T. 88
WALBIRI 133 WEINBERG, A. 204, 205 WELSH 9, 10, 11, 132, 158, 159, 161 WILLIAMS, E. 96, 105, 106, 204, 245 Y YOUN, C. 52
< previous page
page_270 If you like this book, buy it!
next page >
< previous page
page_271 Page 271
Z ZAENEN, A. 88, 169 ZIMMER, K. 148 ZIV, Y. 242 ZRIBI-HERTZ, A. 24, 25, 123, 129, 134 1 TO 3 DEMOTION (= 13D) (SEE INVERSION) 113, 114, 126 1 ADVANCEMENT EXCLUSIVENESS LAW (= 1AEX) 27, 28, 29, 30, 31, 32, 33, 88, 126, 140, 141, 142, 144, 145, 149, 151, 159, 161, 165, 170, 176, 178, 179, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 238 2 TO I ADVANCEMENT (= 21A) 8, 44, 45, 130, 193, 194, 195, 196, 197 2 TO 3 DEMOTION (= 23D) 40, 41 3 TO I ADVANCEMENT (= 31A) 44, 45, 46, 47, 49, 50, 252, 253 3 TO 6 DEMOTION (= 36D) 45 3 TO 2 ADVANCEMENT (= 32A) 18, 35, 36, 37, 38, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 57, 61, 68, 69, 71, 162, 210, 212, 215, 249, 250, 251, 253
< previous page
page_271 If you like this book, buy it!