Order, Concord, and Constituency 9783110876734, 9783110131048

Citation preview

Order, Concord and Constituency

Linguistic Models The publications in this series tackle crucial problems, both empirical a n d conceptual, within the context of progressive research programs. In particular Linguistic Models will address the development of f o r m a l m e t h o d s in the study of language with special reference to the interaction of grammatical components. Series Editors: Teun Hoekstra Harry van der Hulst Michael Moortgat

Other books in this series: 1

M. Moortgat, H. van der Hulst, T. Hoekstra The scope of lexical rules

2

Harry van der Hulst and Norval Smith The structure of phonological representations. Part I.

3

Harry van der Hulst and Norval Smith The structure of phonological representations. Part II.

Order, Concord and Constituency Edited by

Gerald Gazdar University of Sussex

Ewan Klein University of

Newcastle-upon-Tyne

Geoffrey K. Pullum University of California, Santa

Cruz

1983 FORIS PUBLICATIONS Dordrecht - H o l l a n d / C i n n a m i n s o n - U . S . A .

Published by: Foris Publications Holland P.O. Box 509 3300 AM Dordrecht, The Netherlands Sole distributor for the U.S.A. and Foris Publications U.S.A. P.O. Box C-50 Cinnaminson N.J. 08077 U.S.A.

Canada:

ISBN 90 70176 76 9 (Bound) ISBN 90 70176 77 7 (Paper) ® 1983 by the authors. No part of this book may be translated or reproduced in any form, by print, photoprint, or any other means, without written permission from the publisher. Printed in the Netherlands by ICG Printing, Dordrecht.

For Ed Keenan

Table of Contents

Preface Gerald Gazdar, Ewan Klein, and Geoffrey K. Pullum 1. Introduction

1

James McCloskey 2. A VP in a VSO language?

9

Robert D. Borsley 3. A Welsh agreement process and the status of VP and S

57

Susan Stucky 4. Verb phrase constituency and linear order in Makua

75

Geoffrey Horrocks 5. The order of constituents in modern Greek

95

Ronald Cann 6. Raising in Latin: a phrase structure analysis of the accusative and infinitive

113

Michael Flynn 7. A categorial theory of structure building

139

Greville Corbett 8. Resolution rules: agreement in person, number, and gender . . . .

175

References

207

Indices Index of languages Index of names Index of topics

215 216 218

Preface

The varied contributions to the present volume are drawn together by a number of common methodological strands. They are concerned with establishing language universals using a data base that is not restricted to English or a couple of the major languages of Western Europe, but draws in languages from Europe that rarely figure in general linguistic discussions (such as the Celtic languages Irish and Welsh) and from elsewhere (e.g. the Bantu language Makua and the Uto-Aztecan language Hopi). The approaches to syntax used by the contributors are 'concrete', tying the syntactic and, especially, the semantic analyses of sentences close to their surface structures rather than to abstract deep structures. Yet this attachment to a broad data base and to concrete syntax in no way diminishes the intent theoretical orientation of the contributions. In the early 1970s, the Research Centre at King's College, Cambridge, sponsored a linguistics research project, whose Senior Research Fellow for the duration of the project was Ed Keenan. In his work then and since, Ed emphasized precisely these three strands: a broad data base with no neglect of supposedly 'exotic' languages, a concrete approach to syntax, yet coupled with a high level of theoretical orientation. I was privileged to be a Junior Research Fellow on that same research project, and it is surely no accident that the three editors of the present volume also attended the seminars that Ed organized. The volume's dedication to Ed Keenan is thus a fitting tribute to his influence on the progress of linguistics. The contributions to the present volume vary considerably in scope and content. But while those looking for a new dogma to preach may be disappointed, the volume will interest all those who are concerned with syntax as a vital area of ongoing investigation. It is a pleasure to be able to present the volume to this audience. Bernard Comrie

Los Angeles, September 1982

Chapter 1

Introduction Gerald Gazdar, Ewan Klein, and Geoffrey K. Pullum

1. INTRODUCTION

The chapters of this book each deal with one or more of a family of closely related issues: verb agreement, the order of constituents, and the categorial status of verb phrases and sentences.* Although theoretically linked in this way, they focus on a typologically diverse range of languages from the Indo-European, Niger-Congo, Austronesian, and Uto-Aztecan families. Specifically, the book contains analyses of Irish in chapter 2, Welsh in chapter 3, Makua in chapter 4, Greek in chapter 5, Latin in chapter 6, Hopi, Malagasy, and English in chapter 7, and various Bantu, Romance, and Slavonic languages in chapter 8. In this introduction, we shall focus on some of the topics that unite subsets of the seven chapters that follow, and in doing so we shall attempt to outline what seem to us to be important open questions in contemporary syntactic theory. We will not attempt to provide answers to these questions, or to take sides on matters of controversy. Our role will be simply to set the scene for the contributions that make up the remainder of the book. Chapter 2, by James McCloskey, and chapter 3, by Robert D. Borsley, deal with languages from the two branches of the Celtic family: Irish (Goidelic) and Welsh (Brythonic), respectively. The importance of the Celtic languages for current syntactic theory is considerable. The VSO order of major sentence constituents that all the attested Celtic languages share sets them apart from all the other Indo-European languages typologically — as far apart as the Indo-Iranian subfamily, which is characterized by verb-final sentence structure — despite the fact that their speakers live in areas of Britain, Eire, and France that are fully accessible to speakers of non-VSO languages (e.g. English and French) outnumbering them by about a hundred to one. There has been an increasing amount of work on Celtic syntax in recent years, resulting in a useful body of linguistic literature. 1 One of the things that makes the study of VSO languages important is that it appears at first glance as if there could be no VP constituent in such languages, given the non-tangling condition on trees (see Wall 1972,

2

Introduction

148)2 and the fact that VP is assumed to contain V and 0 but not S (i.e. verb and direct object but not subject NP). Other types of language have also given rise to doubts about whether VP is a universally instantiated syntactic category. OSV languages in principle have the same property as VSO languages in that the subject intervenes between the verb and the object (though virtually nothing is known about them; see Derbyshire and Pullum 1981); SOV languages are commonly felt to offer less support for the postulation of a VP constituent than SVO languages despite having a constituent order compatible with the presence of a VP (see e.g. Hinds 1974); and languages with very free constituent order are often regarded as offering little support for constituent structure at all (cf. below for more discussion). In the 1970's, papers dealing with the status of VP's in languages of the types just referred to would have revolved around whether they could be taken to support the postulation of NP VP deep structures, and hence perhaps a universal base. But today, essentially no one believes that the correct way to deal with surface constituent order is to devise sets of transformations that can derive them from distinct underlying orders of a desired type. The authors of this book share the conviction that this is not the way to handle order and constituency phenomena. McCloskey identifies a class of constituents in Irish that he labels, pretheoretically, "progressive phrases". He develops lengthy and detailed argumentation to show that these phrases are, in fact, verb phrases. Susan Stucky, in chapter 4, also argues for the existence of VP, not in a VSO language but in a free constituent-order language. She gives indirect but nonetheless persuasive arguments pointing to the existence of VP in Makua, a Bantu language with strikingly liberal constituent order. Borsley simply assumes that Welsh has VP's rather than arguing for this position. The issue he addresses is whether VP and S are the same syntactic category in the sense of having the same bar level in terms of X-bar syntax. He argues that S and VP are in fact the same thing in Welsh, except for a feature distinction: S is [+SUBJ] and VP is [-SUBJ], the feature [SUBJ] dictating the presence or absence of an extra (subject) NP. This is a fairly radical proposal, and one which raises a nest of issues in X-bar syntax which have not been properly resolved. Thus one conclusion of McCloskey's chapter is that VP is the maximal projection of V, VP and S being distinct both in terms of bar level and in terms of major category features. Geoffrey Horrocks, in chapter 5, notes that in modern Greek, VP has the same lack of ordering constraints with respect to its sisters as do NP and PP, while S's are rigidly final in their immediately dominating category. And Ronald Cann (chapter 6) concludes his discussion of constituency in the so-called "raising to object" construction in Latin by pointing out that collapsing the Latin V introduction rules, in a manner that Borsley's proposal would

Gerald Gazdar, Ewan Klein, and Geoffrey K. Pullum

3

make very natural, has the unfortunate consequences of getting the facts about Latin verb subcategorization wrong. Cann also notes in passing that an equation of VP with S is potentially problematic from a semantic point of view. It clearly cannot be maintained given the semantic assumptions made by Michael Flynn in chapter 7, for example. There have been nearly as many different proposals concerning the status of VP and S in X-bar syntax as there have been linguists who espouse it. The following chart summarizes some of the positions that can be found in the literature. (The labels at the heads of the columns are taken to be standard informal parlance: S ' for a clause complete with complementizer, S for a clause without complementizer, and VP for the phrase that is sister to the subject NP under S. Thus the "Predicate-Phrase" of Chomsky (1965, 102) is VP in this sense. The linguists whose positions we are summarizing do not necessarily employ the informal labels that we attribute to them. For X-bar notations like " V " or " V " " , we have uniformly substituted the more easily typed and read "V2", etc.)

a. b. c. d. e. f. g. h.

Chomsky (1970) Bresnan (1976) Jackendoff (1977) Hornstein (1977) Köster (1978), van Riemsdijk (1978) Gazdar (1981) Borsley (1983) Bresnan (1982)

S' S V5 V3 SI V3 V2 V2 S2

S

s

V4 V3 S V2 V2 V2 SI

VP V2 V3 V2 V3 VI VI V2 V2

It is surprising how few of these proposals allow one to express certain generalizations in bar level terms (although of course the theories in which the proposals were embedded may have expressed the generalizations in some other way). Take "topicalization" in English, for example: VP and S' can usually topicalize, but S never can. This is shown in (2) where the topicalized constituents are bracketed.

(2)

a. They said they would achieve the quadruple somersault, and [achieve it] they did. b. That they achieved the quadruple somersault is surprising enough, but [that they did it in front of a paying audience] I can hardly believe. c. *The dangers of attempting the quadruple somersault we are well aware of, but [we will succeed] we are convinced that.

4

Introduction

Only position h allows this to be expressed directly. However, VP, S, and S ' all appear to be capable of occurring as complements in English (VP to verbs like help in Please help tidy up, S to prepositions like because in I did it because (*that) I love you, and S' to nouns like rumor in There is a rumor that digital records interfere with muscle activity). If this is the correct generalization, then only position/allows it to be expressed in bar terms. But if it is not correct, and if apparent complement occurrences of S are really always instances of S', then again only position g draws the correct distinctions. Nevertheless, there are suggestive analogies between VP and S to be found in languages other than Welsh. Stucky (1981) has argued thatMakua has the two topicalization rules shown in (3). (3)

a. S NP S/NP b. VP-* NP VP/NP

And Gunji (1981) proposed, quite independently, that Japanese required exactly this pair of rules. Under Borsley's proposal, the two rules can be collapsed rather naturally. Furthermore, consider the relation in English between verb phrases and sentences that begin with an auxiliary verb. One possible analysis could be expressed, given Borsley's proposal, as follows.3 (4)

V2[aSUBJ] -> V[AUX]

V2 [aSUBJ]

That is, either (i) a sentence or (ii) a VP may expand as an auxiliary verb followed by (i) a sentence or (ii) a VP. Borsley's main argument for identifying VP and S hinges on what he takes to be the methodological undesirability of collapsing syntactic operations via the expedient of a numerical variable ranging over bar level. It is not necessarily the case that such rule collapsing is to be eschewed, however. It is interesting that Keenan (1980) and Dowty (1982) have both proposed language-independent theories of passives which achieve their elegance and generality through the use of numerical variables. These accounts do not employ X-bar syntax; they distinguish verbal categories by the number of NP arguments they need in order to form a sentence. For instance, in Dowty's analysis, a sentence, which needs no such arguments, is analyzed as a V0; a verb phrase or simple intransitive verb needs one NP argument, so it is a V I ; transitive verbs or verb phrases ("TVP's") need two NP arguments, hence V2; and ditransitive verbs or verb phrases ("DTVP's") need three NP arguments so they count as V3. This enables Dowty to state his generalized passive rule as follows: (5)

PASSIVE(Vn) = Vn-1[+PAS]

Gerald Gazdar, Ewan Klein, and Geoffrey K. Pullum

5

This says that the passive o f some verbal constituent is a verbal constituent with one less NP argument place and one that is featurally marked for passive. The range o f values that n can take will vary from language to language. For many dialects o f English, n can only be 2, and thus the English passive maps T V P into V P [+PAS]. But for Chichewa, for example, n can be 2 or 3. This sort o f approach offers an intriguing way o f thinking about the Welsh agreement facts described in Borsley's paper, and one which may turn out to be well worth pursuing. Languages with very few apparent constraints on the order o f constituents have become a topic o f concern to linguists in recent years. There have been two methodologically quite distinct approaches to such languages. The one that has been best publicized is one that we shall call the neo-empiricist approach because it apparently derives from a feeling that hypotheses about how to handle order should spring directly out o f the way the data initially look to the investigator. The neo-empiricist approach notes that there are languages which appear to allow virtually any order o f constituents, or even o f words regardless o f their constituent membership, and proposes to account for this state o f affairs by postulating for those languages a special type o f grammar that allows all possible orders and assigns no constituency; call this a grammar o f type A. Since languages like English do not allow such anarchy in their sentence structures, they are assumed to have a quite distinct type o f grammar, o f type B, which does assign sequence and constituency. Thus the neo-empiricists end up with a gross typological dichotomy that stands in a one-to-one correspondence with the crude observation that there are languages like English with ordering restrictions and constituency and there are languages such as those in the Pama-Nyungan family in Australia with scrambled constituents and word order. The neo-empiricist approach is associated primarily with Hale and his students at MIT. Type A grammars are the "W-star" grammars o f Hale ( 1 9 8 1 ) , and the languages they are proposed for are termed "non-configurational." Surprisingly, the neo-empiricist approach is partially endorsed by both Bresnan ( 1 9 8 2 ) and Chomsky ( 1 9 8 1 ) , despite the hesitance expressed even by Hale ( 1 9 8 1 , Appendix) about the viability o f the W-star/X-bar distinction and his characterization o f his own earlier analysis o f Warlpiri as " t o o extreme". The alternative approach is argued for in Pullum ( 1 9 8 2 ) . It is developed in some detail for two very different languages by S t u c k y ( 1 9 8 1 ) andGunji ( 1 9 8 1 , 1982). This approach denies the typological bifurcation of languages entailed by neo-empiricist descriptions. Instead, it maintains that languages differ only in the particular rules they employ, not in the type o f rule. Consider, as an abstract example, a language having the following grammar.

6

Introduction (6)

i. a. b. ii. a. b.

S VP^ V < NP
- minih - a mil (*mwaaj child - dem A sa-t-oa/convince me wiira Asaapala aho

ruw - el - a

isima

that A sa/t-oa -prepare-app-t porridge 'that child, Araarima has convinced me that Asaapala prepared porridge for (him)' *b. Mwaari-ole isima Araarima aho-kaaminih-a mil wiira Asaapala aho-n-ruw-el-a

A second property of this topicalized structure in Makua is that the verb in the sentence containing the extraction site must show verb agreement with the displaced NP. Thus, in example (3b) above, the lack of agreement in the object agreement slot renders the sentence unacceptable. Note also that the embedding verb agrees not with the displaced NP, but with it's own semantic object mii ('me'). So far, we have observed that a displaced NP may appear in front of an embedding clause, potentially any number of clauses up. In addition, a displaced NP may appear in between the subject and verb of an embedding clause. This distribution is illustrated by the example in (4) below. Note that the verb agreement facts are parallel to those in example (3a). The displaced NP triggers agreement on the lower verb but not on the embedding verb.

Susan Stucky (4)

79

Arààrima mwaàn -ólé a-ho A

child - dem sa-t - oa -

wiira Asaapala aho -

- ruw - el - a

- minih - à

mii

convince-t me isima

that A sa/t-oaprepare-app-t porridge 'Araarima has convinced me of that child (as expected) that Asaapala prepared porridge for (him)' Again, two displaced NP's in the embedding clause would be unacceptable just as expected if only one gap per clause is allowed. The pattern in (4) is that which I have termed VP-Topicalization. For both the STopicalization and the VP-Topicalization it is possible to get displaced subjects as well as objects (although I have given no examples here). It would be misleading to claim that displaced NP's only appear in these two positions (immediately before the embedding verb and in between the subject and verb of the embedding clause). There is a class of verbs in Makua, intransitive verbs of perception and reporting, which govern a kind of raising. As in the topicalized patterns, an NP from a tensed lower clause is found in a higher embedding clause. Unlike topicalization, however, the displaced NP triggers object agreement in the embedding clause as well as the appropriate verb agreement (subject or object agreement) in the clause to which it belongs semantically. A displaced NP which triggers verb agreement on the embedding verb is also not subject to the distributional restrictions of topicalization, it being "freely" distributable in that clause. For this class of raising verbs, agreement on the embedding verb turns out to be optional, as illustrated by example (5) below. This optionality can be attributed to a double analysis of the structure, topicalization in one instance (in which there is no agreement allowed on the embedding verb), and raising in the other instance (in which agreement is obligatory). 6

A àhó

child - dem sa-t/oa/said-t

that

simà i*i\ U W jt ~- àa iisima

sa/t-prepare-t porridge 'Ararima said (of) the child that (he) prepared porridge'

80

Verb phrase constituency and linear order in Makua

2.2. Formalization It is the formalization of the topicalization phenomena that makes clear the value of a syntactic verb phrase for the analysis of Makua, hence the indirect nature of the argument alluded to earlier. Generalized phrase structure grammar, as implied by its name, ultimately consists simply of phrase structure rules. Thus, what would be characterized by a movement transformation in a transformational grammar is, in GPSG captured in the definitions that provide for the set of phrase structure rules. 7 Suppose for Makua, the two rules in (6) are adopted. (In these rules, the symbols N2, V2 etc. should be viewed as informal notation abbreviating bundles of features including the category specification +N, - V etc. as well as specification of bar-level. For the purposes of this analysis of Makua, assume V2 (i.e. S) to be the maximal projection of V.) (6)

a. V2 b. VI

-»•

N2 V

VI N2

The phrase structure rule in (6a) says that a partial tree in which V2 immediately and exhaustively dominates a N2 and VI (in that order) can be admitted. (6b), in analogous fashion says that a VI can be admitted just in case it immediately and exhaustively dominates a V and a N2. Now, one of the more interesting facets of GPSG is its treatment of unbounded dependencies, in particular, those of the sort found in patterns of topicalization in Makua. There are two principled additions to the formalism of phrase structure grammar in order to account for these phenomena: (i) a set of derived categories such as V2/N2 which, intuitively, corresponds to a sentence (V2) with a noun phrase (N2) gap in it someplace and (ii) a set of derived rules to admit nodes with derived categories in them. The schema which gives derived rules (Gazdar (1982)) pairs exactly one gapped category (i.e. a slashed category) with exactly one gap of the same category in the same rule. Based on the rules in (6), there will be, among others, the derived rules in (7). (7)

a. V2/N2 b. V1/N2

-> ->

N2 V

VI/N2 N2/N2

We then state the S-Topicalization and VP-Topicalization rules as in (8a) and (8b), respectively. These rules link displaced noun phrases to sentences containing noun phrase gaps.

81

Susan Stucky (8)

a. S-Topicalization b. VP-Topicalization

V2 a V2/a, where a e {N2 } VI -*• a Vl/a, where a e {N2 }

Together with other necessary rules, the rules in (7) and (8) will analyze a partial tree like that in (9) which corresponds to VP-Topicalization. V2

(9) N2

A

VI

Araarima N2

VI/N2

A

mwaanole V

N2

ahokaaminiha

mii

A A

S'/N2 [comp]

A

wiira

V2/N2 N2

A

VI/N2

Asaapala

V

N2/N2 N2 A A ahonruwela t isima

A

This notation makes explicit a syntactic binding between the displaced NP and the position in the embedded clause. The linked slashed categories constitute the projection path. Note that the addition of (8b) in particular predicts that we can find a topicalized NP between the subject and verb in any clause. This result obtains because (8a) will admit a topicalized structure at any point in a tree so long as the projection path is complete in the tree. It is this part of the formalism which accounts for the unbounded nature of topicalization in Makua. The fact that, in Makua, only one displaced NP per clause is found in an embedding clause is guaranteed by the definition of the slashed categories to admit only one syntactic gap.8 The rules for S-Topicalization and VP-Topicalization in (8) also will admit structures in which the topicalized noun phrase is "displaced" within it's own clause, as illustrated roughly by the diagram in (10). (10)

82

Verb phrase constituency

and linear order in Makua

Thus, one expects to find evidence of the hierarchical structures induced by S-Topicalization and VP-Topicalization, for example, in the orders 0 V S 10 and S DO V 10 respectively. Such evidence is argued for in the agreement analysis in section 3. More generally, it follows from the presence of the hierarchical structures induced by topicalization in GPSG that the free order of noun phrases intraclausally is not due entirely to flat structure in Makua. In fact, the argument for the topicalized structures is an argument about the surface structure of Makua sentences. The postulation of a syntactic verb phrase in Makua allows a formalization of the topicalization facts (as distinct from the raising facts) which would otherwise be difficult to account for. The evidence in the next section of this paper shows how verb agreement is different for the slashed categories than for the non-slashed categories. In this way, the analysis of verb agreement provides support for the analysis of topicalization just proposed, and, indirectly, provides reinforcement for an analysis of Makua in which there is a syntactic verb phrase. 3.1.

Description

of verb

agreement

In Makua (as in most, if not all Bantu languages), verbs display morphology which encodes agreement with the noun class prefixes of the arguments of the verb. Makua has two such agreement slots, the first such slot is after an optional negative marker and before everything else. The second agreement slot is between the tense and aspect prefixes and the verb root itself. The first slot is reserved for agreement with subjects and the second for object agreement, roughly speaking. Consider the example in (11) below. In this example the subject noun is nivaka ('spear') which is a class 5 noun. (The noun classes are traditionally given a number. Often there are singular and plural pairs of prefixes, each with distinct agreement forms.) In (11), the subject agreement prefix is ni-.9 (11)

n i - v a k a ni - ho - rith'ya - a 5 - spear sa - t - break - t 'the spear is broken'

For noun class la and for third person singular personal pronouns, no overt prefix appears. Such cases are glossed with '0' in the examples. The prefix is required regardless of whether a lexical noun actually appears in the sentence. 10 Objects, like subjects, also trigger verb agreement. Again, the appearance of the object prefix is obligatory, regardless whether an overt lexical noun is in the sentence or not. 1 1 Before defining what sort of thing counts as an object for agreement purposes in Makua, there are several morphological

Susan Stucky

83

patterns that need to be pointed out. There are object prefix morphemes only for personal pronouns and Class 1, la, and 2 nouns. There are no overt morphemes for the rest of the eleven attested classes. This morphological gap has consequences for the agreement analysis (which will become apparent as the discussion progresses). Leaving aside cases in which no overt object noun appears, the facts are that overt object nouns or personal pronouns will trigger obligatory object agreement when the verb is transitive. This is true not only of basically transitive verbs but of derived transitive verbs as well. Derived transitive verbs are those verbs exhibiting one or two verbal suffixes, the applied and the causative. The applied suffix increases the argument structure of the verb by one. The thematic role of the argument added (the applied object) is either a recipient, a beneficiary, a locative, or an instrument. The causative suffix also increases the argument structure by one, adding a causee. The example in (12a) illustrates object agreement with a basic transitive verb; (12b) illustrates object agreement with a derived transitive verb. These patterns obtain regardless of the six possible orders of the three words in each sentence. (12)

a. Aráárima a - ho - ñ - th'úm - a báásikeli A sa - 1 - oa - buy - t bicycle 'Araarima has bought a bicycle' b. Aráárima a - ho - ñ - ñth'eéh - á báásikeli A s a - t - o a - b e broken/caus-t bicycle Araarima has caused a bicycle to be broken'

One generalization to be captured (with respect to object agreement) is that an object of a transitive verb (regardless of whether it is a basically transitive verb or a derived transitive verb) will trigger object agreement. Agreement facts for the bitransitive verbs are more complicated. Consider first the causee of a derived bitransitive verb and the applied object of a derived bitransitive verb when that applied object is a recipient, a beneficiary or a locative (but not when it is an instrumental). 12 When an object of that group is of an agreeing class (i.e. of the class that has overt morphemes) then verb agreement is with that object regardless of word order. Thus, in the examples below, the subject noun is the pronoun mii ( T ) , the applied object is a beneficiary of Class 2 and the direct object is of Class 1. In this case there is an object agreement form for both the objects (e.g. -mu- for Class 1 a and -a- for Class 2). When the agreement prefix is that for Class la, only the reading in which the Class la noun is the applied object is allowed, regardless of the order of the objects postverbally. When the agreement is with the Class 2 noun, on the other hand, the readings are reversed, again, regardless of postverbal order.

84

Verb phrase constituency and linear order in Makua

In this latter case the Class 2 noun is taken to be the applied object. Compare the examples in (13) below. (13)

a. mil ki - ho - n - th'um - el a amirawo I sa - 1 - oa- buy -app - t boys 'I have bought the boys for a bicycle' b. mil ki - haa - th'um - el - a amirawo I sa - t/oa - buy - app - t boys 'I have bought a bicycle for the boys'

baasikeli bicycle baasikeli bicycle

The facts represented in (13) above also hold for recipient applied objects, locative applied objects and causees. The really tricky cases are ones in which one of the objects is of an agreeing class but the other is not. One might expect, on the basis of the evidence in (13), that overt agreement with one of the noun phrase objects would insure that the reading is always one in which the object agreed with is the beneficiary, since it is the case in examples like (13) that it is the applied object (and not the direct object) which triggers agreement. This is not the case. In the examples in (14) below, the noun baasikeli ('bicycle') is of Class la and has associated with it the object prefix form mu- (which surfaces in this example as a nasal consonant). Ntenga ('messenger'), on the other hand, is a Class 3 noun and it has no overt agreement prefix correlated with it. It turns out that in a subset of the orders, the reading with object agreement can be that of the direct object and not the applied object. Compare the examples in (14). In (14) there is agreement with the Class l a noun, baasikeli ('bicycle') and the reading is the less likely one in which the messenger was bought for the bicycle. In (14b), the word order is different, the agreement facts are the same, but, importantly, the reading is more likely to be one in which the direct object (and not the applied object) triggers agreement. (14)

a. m i i ki - ho - n - th'um - el - a ntenga baasikeli I sa - t - o a - buy - app - t messenger bicycle 'I bought a messenger for a bicycle' b. baasikeli m i i ki - ho - n - th'um - el - a ntenga bicycle I sa - t - oa - buy - app - t messenger 'A bicycle, I bought for a messenger'

Thus, under certain conditions, either an applied object or a direct object can trigger object agreement. We know from the examples in (14) that the order of the objects is relevant to determining the patterns. I will argue in the next section that this agreement pattern can be attributed to the orders induced by linking rules involving slashed categories.

Susan Stucky

85

A full analysis of verb agreement in Makua, then, must account for subject agreement in all clause types, object agreement with transitive verbs, and object agreement of various types with bitransitive verbs.

3.2.

Analysis

In this section, I will present an analysis of the verb agreement facts. It is claimed that the agreement with the direct object occurs in those orders induced by the S-Topicalization and VP-Topicalization. The formal analysis in Appendix A provides the actual notation and shows how the facts can be accounted for. Two alternative analyses, one based on discourse function and the second based on a simple linear ordering are considered. The analysis based on discourse function is shown to be, in principle, compatible with the syntactic analysis. The second, linear ordering analysis is shown to make some wrong empirical predictions. One caveat, however, is in order. The analysis of verb agreement rests on the analysis of syntactic order in Makua as a whole, not just with respect to the rules motivated in this paper. Therefore, I will discuss the clearest cases (i.e. those which can be motivated on the basis of the discussion in this paper), discussing the ramifications for other orders only generally. If it is indeed correct to attribute the appearance of agreement with direct objects (in double object constructions in which the indirect object is of a noun class with no agreement prefix) to slashed categories, then one expects that agreement should show up in the analogues to examples (14a) and (14b) in which the displaced NP is the semantic direct object belonging to the embedded clause. This is correct, as illustrated by the examples in (15) below. (15)

a. baasikely-ule Araarima a - no - ri - cuwel - a wiira bicycle - dem A sa - t - oa - know - t that Asaapala 0 - ho - ri - th'um - el - a ntenga A sa - t - oa - buy - app - t messenger 'the bicycle, Araarima knows that Asaapala bought (it) for a messenger' b. Araarima baasikely-ule a - no - ri - cuwel - a wiira A bicycle - dem sa - t - oa - know - t that Asaapala 0 - ho - ri - th'um - el - a ntenga A sa - t - oa - buy - app - t messenger 'Araarima knows of the bicycle that Asaapala bought (it) for a messenger.'

86

Verb phrase constituency and linear order in Makua

One expects also to find clause internal evidence of the agreement strategy attributed to the topicalization rules since orders like those in (10) may be analyzed as topicalized structures. The data patterns in the following way. When the order is either DO S V AO or S DO V AO, that is, when the DO is the only object preceding the verb, agreement is clearly preferred with that object. The agreement patterns in which agreement is preferred are those induced by S-Topicalization and VP-Topicalization. When the order is SO DO AO V, agreement with the DO is likewise allowed (although it is not as strongly preferred as in the cases in which only one object appears before the verb). Again, this order could be due to the VP-Topicalization. One which cannot be attributed to a VP-Topicalization is S AO DO V. In this order agreement is strongly dispreferred. Thus there is clearly a preference for agreement with the DO in those cases in which there is a possibility of attributing the order to topicalization rules. The data are, in these clause internal tests, quite fuzzy. However, they, together with the data from the unbounded dependencies, argue for an analysis which attributes the presence of agreement with a direct object in double object constructions to the slashed categories. Before discussing one class of recalcitrant cases and proposing alternative strategies, I would like to point out without an accompanying analysis (but see Stucky 1981) the same pattern can be observed in relativization. The relativization analysis has in common with topicalization the use of the slashed category notation. Thus if it is truly the slashed categories, or in slightly more theory independent terms, the presence of syntactic gaps, that is related to the agreement pattern, one would certainly expect that pattern to obtain in relative clauses. In Makua, the relative clause follows its head. There are no relative complementizers of any sort (nor any special tonal markings). And, as one expects, agreement with a relativized direct object out of a double object construction in which the indirect object does not have an overt marker is obligatory. This preference is as strong as in the cases of the displaced noun phrases in (15) above. (17)

baasikeli aa - n - th'umenle Araarima nt'eng-ole ... bicycle sa - oa - buy/app/t A messenger-dem 'The bicycle that Araarima bought for a messenger ...'

One class of cases remains to be looked at. While I haven't discussed whether or not there are any rules of the form which would allow unbounded dependencies to the right (all of the above have been to the left), in Stucky (1981) I did argue that such structures existed. Since such structures would be analyzed in GPSG by the use of slashed categories, one would expect similar options for agreement when the object noun phrases follow the verb. However, I have no attested cases of agreement

Susan Stucky

87

with a DO in the class of cases we are considering here when that DO is to the right of the verb. Note, however, that there is a gap in the data. Not only are there no instances of such patterns, there are also no cases in which the interpretation of these structures permits a DO reading of a noun phrase that has overt agreement (cf. (13) above). In general, postverbal position indicates new information of various sorts and readings on ambiguous sentences drastically differ in these cases. The lack of expected readings may be taken to be indicative of a resolution of this ambiguity rather than as an argument against the proposed analysis. Nevertheless, the fact that the cases in which the DO does trigger agreement are all preverbal positions raises the possibility of alternative analyses. One possibility is that it is primarily preverbal position that triggers this sort of agreement. This analysis falters on two accounts. Such an analysis would, presumably, have to make appeal to terminal surface strings. It would have t o pick out just DO's, which, recall, are not marked by any morphological means in Makua. Then, it would have to pick out just those DO's which semantically belong to the appropriate clause. The present analysis proposed does all of that automatically and also provides some explanation as to why the varying degrees of acceptability are found interclausally, an explanation which would not be forthcoming from a simple surface linear analysis. Such an analysis would be cumbersome at best. 1 3 A second sort of alternative analysis would be to make appeal to discourse function. After all, it has been noted that object agreement in Bantu languagues is, when optional, often tied to discourse function. And, since much of Makua word order is determined by what sort of information the speaker assumes the hearer already knows and what it is that the speaker wishes to assert, one might guess that this alternative agreement pattern is reflective of some discourse function strategy. If such an explanation is at the heart of the matter, then it is clear that no simple syntactic trick, such as an appeal to a definite marker (a medial demonstrative, in Makua) would work, since the distribution of that sort of morphology is not tied to word order in Makua. Also, there is a difference between the discourse functions associated with DO's in sentence initial position and, for instance, between the subject and verb, so that whatever these orders may have in common is not immediately clear. Certainly, there is no obvious correlation between these discourse functions and that of relativization, whatever its precise formulation will be. However, there is a syntactic characterization which these phenomenon have in common, unbounded dependencies, and that syntactic commonality is reflected in the formalism of GPSG. It may ultimately turn out that a precise formulation of discourse functions would bring out that commonality, but such a solution would, I think, fall short of predicting the syntax of the constructions. 1 4

88

Verb phrase constituency and linear order in Makua

The analysis put forth, then, claims that verb agreement is obligatory with slashed categories. The appendix to this chapter includes a precise formalization of this phenomenon. As the analysis stands, it provides additional support for the VP-Topicalization structure.

4. CONCLUSION

The answer to the question raised at the beginning of this chapter was whether or not there was syntactic evidence for a verb phrase in Makua. It was pointed out that the evidence, given the extent of order variation in the language, would be, perforce, of an indirect nature. The answer given was in the affirmative. There is evidence based on the account of displaced noun phrases as formalized in GPSG of a syntactic verb phrase. The analysis of this VP-Topicalization was supported by the verb agreement analysis for Makua. Some otherwise unruly facts seemed to follow from the analysis given. More generally, the argument about Makua structure is also surface true. That is, the analysis suggests that a surface syntactic structure of Makua should provide evidence of these hierarchical structures. 15 Therefore, any bi-level grammar which allows surface reordering would have to be prevented from violating such structures on the surface. The analysis also stands as an argument for constituent structure in a language which displays relatively free ordering, and it stands as a caution against immediately assuming that because a language has order freedom syntactically it must not have any hierarchical structure. The analysis also stands as a caution against assuming that the hierarchical structure that can be motivated for a given language must be identical for all orders. Clearly this is not so. Natural languages may provide evidence of different constituents in different orders.

APPENDIX

This appendix includes part of the verb agreement analysis of Makua. It is incomplete in the sense that the analysis I provided in my thesis accounts for more of Makua syntax than has been presented in this paper. The analysis provided here also differs in detail from that which I gave for the slashed categories and, hence, supercedes the analysis in the thesis. 16 Familiarity with GPSG is assumed for purposes of the agreement analysis. For the analysis, I am assuming a finite set of features on nouns which, for their mnemonic value, are the numbers traditionally associated with the noun classes, e.g. [la] for Class la nouns, [5] for Class 5 nouns. Such

Susan Stucky

89

features are associated with the lexical entry for nouns. In addition, a syntactic feature for verbs will be employed. This feature is a singleton set whose only member is an ordered pair. The first member of the ordered pair will represent an agreement feature from the same set of numbers used for the noun classes and it signals subject agreement. The second member of the ordered pair is also a noun class feature from the same set, but it corresponds to the object agreement prefix. Like the nouns, verbs have associated with them these syntactic features. There will be, then, features of the following sort: [ < 1 , 2 > ] (where the verb agrees with a Class 1 subject and a Class 2 object). The analysis also makes use of the Head Feature Convention (HFC) which insures that the features on a phrase level node are the same as those on the head of that phrase. Its utility will become apparent in the course of the analysis. (See Gazdar (1982) and Gazdar and Pullum (1981) for details.) 17 Finally, I assume that in the actual rules, variables which range over the permissible features may be used. Taking the features of the sort defined above for nouns and verb agreement, these are incorporated into two of the rules as in (1) below. These rules taken together with the HFC will insure that the features postulated for the N2 and VI will appear on their respective heads. (1)

a. V2 -»• N 2 M Vl[ta,0>] b. VI -»• V[] N2[0] N2[7] where a, (3, 7 e {[1st], [2nd] etc.... [1], [la], [2]...}

The rule in ( l a ) accounts for subject agreement, ( l b ) insures verb agreement with an applied object. Grammatical relations are defined derivatively along the lines of Dowty (1982) and are not, therefore, indicated in the syntactic rules shown here. As Gazdar (1982) points out, this approach eliminates the need for copying rules which involve hunting for the subject and the verb and then copying the features. Instead, the features are already there as a reflection of morphological processes in the language, and the categorial features simply insure that these features match. It is worthwhile noting that the incorporation of the features directly into the rules will make agreement obligatory (because a sentence will be well-formed only if there is agreement and only if the features match) in just the way required. For instance, the agreement schema in (1) will, together with the HFC, analyze a partial tree like that in (2a) below because the features match. They will not analyze a partial tree like that in (2b), because the features in that tree do not correlate in the manner required by the agreement schemata. Circled features indicate the non-matching features.

90

Verb phrase constituency and linear order in Makua

(2)

a.

V2 VI

N2

[]

[1] V []

N2 [2]

N2 [3]

V2

b.

VI [ ] [la]

/

V []

N2/N2 [5] [la]

N2 [6]

Thus far, the analysis makes two predictions. First, it predicts that an embedding verb will not agree with a noun phrase gap that is passed up through it. This is correct as exemplified by examples (3) and (4) in the main body of the text. As it stands, the analysis does not admit any well formed trees with slashed categories at all because no terminal symbol has been given to categories of the form a/a. The following metarule will provide such a terminal symbol and will also predict that any object gap will trigger agreement just as the analysis in section 3 of the chapter requires. The metarule is given in (6a) below and a tree corresponding to example (4) in the main body of this chapter is given in (6b). (6)

a. < VI [(a, j3)]/N2 [7] -> W N2 [0]/N2 [7] < V l [ ( a , 7 > ] / N 2 [7]

W t [] ,2, 7]

W'>

W'>

Verb phrase constituency and linear order in Makua

92

V2

b.

VI

N2 [la]

[]

\

/

V/N2

N2

[1]

[]

V

N2

[]

[1st]

[1]

S'/N2

/ /

[!] V2/N2

comp

N2 [la]

VI/N2 []

V []

t N2 [1]

[1]

N2 [9]

FOOTNOTES •Special thanks to John Wembah Rashid, who provided the data included in this chapter. I would also like to thank Chuck Kisseberth, Gerald Gazdar, Jerry Morgan, Geoff Pullum, and colleagues at the University of Illinois and the University of Massachusetts, Amherst for comments on my thesis and thus, indirectly on the analysis presented in this chapter. The primary research on Makua was supported by a grant from the University of Illinois Research Board, a University of Illinois Graduate College Dissertation Research Grant, and University of Illinois Graduate Fellowships. This chapter was written under the auspices of A.P. Sloan Foundation Postdoctoral Fellowship administered by Stanford University. 1. The data on which this analysis is based come from native speaker elicitation over a two year period at the University of Illinois. The data here are from the Imit'upi dialect of Makua, spoken in Southern Tanzania. While it appears that the other dialect that we have investigated, Ikorovere, shares many of the features that Imit'upi exhibits, the particulars of this analysis should not be generalized beyond the Imit'upi dialect. 2. I would like to stress that the use of the terms " f r e e " or " f r e e d o m " in the context of this chapter refer exclusively to syntactic phenomena. For Makua, at least, it is clear that a speaker conveys by the use of a particular order of constituents certain assumptions about the newness or oldness etc. of the information being conveyed. For some discussion of the discourse functions related to word order and tense and aspect in Makua, see Stucky (1979) and (1981). I am assuming, therefore, that there is a level of grammar at which it is appropriate to state generalizations about the syntax of a language independently of the discourse functions associated with a particular order of constituents much in the same way that one can analyze the syntax of English as separate from but related to intonational phenomena such as contrastive stress.

Susan

Stucky

93

3. Cases in which the relative clauses and/or adjectives do not appear after the noun which they modify have been elicited, however, these have the earmarks of being parenthetical phrases rather than part of the sentence they are found in. See Stucky (1981) for details. 4. The orthographic conventions are as follows. An apostrophe marks an aspirated consonant. High tone is marked with an acute accent. Low tone is left unmarked. The tone marking is not entirely phonetic. Some predictable information such as phrase final lowering of high tones and penultimate fall is unmarked. See Stucky (1981) for details. Abbreviations in the glosses include sa = subject agreement marker, t = tense/aspect affixes, oa = object agreement marker, and app = applied suffix. 5. The term S-Topicalization (and the term VP-Topicalization to be introduced) are somewhat misleading in that neither has associated with it exactly the discourse functions associated with the analogous topicalization in English. The terms are used primarily for structural reasons rather than functional ones. 6. An alternative account of the raising cases, suggested to me by Roger Higgins, should be considered. It could be that these are anticipatory or proleptic constructions. One sort of evidence for this latter analysis rather than a raising analysis would be if the same sort patterns occur with intransitive verbs in the embedded clause. Then, the downstairs NP could not have been raised. I have not been able to obtain grammatical examples of this latter sort. Even if the analysis turns out to be better characterized as a prolepsis rather than raising, the arguments in the present chapter remain intact. In elicitation subsequent to the writing of my thesis, examples with object "raising" were less acceptable than those in which subjects were involved. However, this gradation of acceptability seemed to be strongly dependent on the embedding verb. 7. In this chapter, I will be using only that part of the formalism deemed necessary to make the argument clear. The reader is particularly encouraged to consult the paper by Gazdar (1982) for details. Subsequent to the writing of my thesis, Gazdar and Pullum (1981) proposed a revised way of defining phrase structure grammars which separates out statements of linear order from linear precedence. This redefinition has important consequences for the analysis of Makua in general. But since little in this paper hinges on that redefinition, I will use phrase structure rules which define linear order and dominance relations simultaneously. One may, in fact, view these rules as those defined by the ID/LP format. The linearized versions serve to make the points in this paper more explicit. 8. The restriction to only one syntactic gap per clause is clearly not universal. Maling and Zaenen (1982) argued, for Swedish, for the necessity of multiple gaps in GPSG. Detailed discussion of the phenomenon is found in Engdahl (1980). Engdahl (1982) also outlined a class of cases of parasitic gaps in English which suggest the presence of more than one syntactic gap in English. Sag (1982b) has recast most of Engdahl's analysis in GPSG. Makua seems not to have the sort of phenomena found in Swedish. I have no idea whether or not parasitic gaps of the sort found in English can be found in Makua. 9. While the agreement prefixes are often morphologically identical to the noun class prefixes, this is not always the case. The existence of both null agreement markers and non-phonological identity precludes a simple morphological copying rule. 10. I have characterized this slot as subject agreement. This refers to the notion grammatical subject. Grammatical subjects include, for instance, passive subjects. Since they do not directly bear on the analysis in this paper, I will not discuss them here. 11. Some Bantu languages have an asymmetry between subjects and objects in this

94

Verb phrase constituency and linear order in Makua

regard. Subject agreement may be obligatory under all circumstances while object agreement is required only when there is a definite object or when there is no lexical noun. 12. The instrumental objects behave differently from the other applied objects. See Stucky (1981) for details. This difference will not be discussed in this paper. 13. It is worth noting, in addition, that no surface terminal string analysis is in the spirit of GPSG, and without substantially altering the theory, is not even a possibility. I simply raised the issue since it is an obvious first glance solution which turns out to be very difficult to explicate in any theory when it is clear what such an analysis would have to look like. 14. Certainly the commonality would not be incompatible with the syntax. Indeed, it is the syntax that expresses the commonality of these constructions rather than the commonality of discourse functions. 15. The analysis of S-Topicalization and VP-Topicalization precludes any simple concatenation rule of the sort suggested by Hale (1981) or Lapointe (1981) since their concatenation rules for free order languages necessarily concatenate categories of the same bar level. See also Stucky (1981) for further arguments against concatenation rules for Makua syntax. 16. I thank David Dowty for pointing out a technical inadequacy of the analysis in the thesis. That problem is, in the present analysis, solved. 17. The HFC insures that features are percolated down the tree. That is, features only need to be specified on V2 (S), for example; they would subsequently appear on VI and V categories by convention. In the present analysis, the verb agreement is specified at the verb phrase level, even though the verb phrase could be taken to be the head of the sentential category. For the most part, the reasons why this choice has been made have to do with the part of the analysis of Makua which is not included in this paper. 18. An additional metarule is needed for subject gaps, but since that phenomena has not been discussed in this paper, I have not included it here. Note that this metarule bears some resemblance to Sag's (1982a) slash elimination metarule for English. The motivation for the metarule in Makua is not based on coordination (as it is for English), but, rather, on insuring that slashed categories trigger agreement obligatorily.

Chapter 5

The order of constituents in modern Greek Geoffrey Horrocks

1. CONSTITUENT ORDER A N D GENERATIVE GRAMMAR*

In unconstrained versions of a theory which recognizes deep and surface structure as levels of representation it is possible in principle to use transformations to derive any surface order from any underlying order. 1 For analyses employing this apparatus to avoid charges of vacuousness, it is clearly necessary to impose principled restrictions on transformational permutation and so guarantee some limitation on pairings of deep and surface orders. With respect to languages with fairly fixed constituent order, Emonds (1980) argues that the only theoretically sanctioned and empirically motivated derivations of basic surface orders from distinct underlying orders are those of VSO from SVO and OSV from OVS by a local movement of the verb over the subject. 2 All other basic surface orders are matched therefore with identical underlying orders. With respect to languages with freer constituent order, Emonds takes the view that since a fixed order base is needed for languages like English it is not unreasonable to suppose that the grammars of all languages have fixed order base rules. Given this assumption, together with the severe restrictions on transformational permutation, it is possible to ignore the phenomenon of free phrase order in the search for a universal formulation of syntactic rules and principles. 3 The essentially superficial differences between languages with regard to the degree of constituent order freedom they permit can be accounted for by the extent to which the grammars of different languages make use of 'stylistic rules'. These apply postsyntactically to 'scramble' the fixed order output of the syntax (Emonds (1980) pp. 39-40). Within the framework of generalized phrase structure grammar (GPSG) this approach to free phrase order has been reinterpreted as involving the operation of a set of metarules on a small set of basic rules (Stucky (1981)). The transformational view obviously presupposes that it is possible to identify a 'basic' or 'underlying' order for free phrase order languages, and that it is possible to distinguish grammatically conditioned variants from purely stylistic reorderings. There is room for doubt on both counts.

96

The order of constituents

in modern Greek

Grammar- and theory-internal factors, combined with considerations such as frequency, markedness, pervasiveness, typological correlations and so on, almost always lead to indeterminate and mutually conflicting results (Stucky (1981) ch. 3). In any case the problem of arbitrary transformational permutation has simply become one of arbitrary stylistic permutation. Whichever order is selected as basic the remaining orders can in principle be derived by stylistic rules, since their potential for reordering is only very weakly constrained by Emonds' listing of their properties. 4 Furthermore these properties do not effect a clear-cut grammatical/stylistic dichotomy. For example, the 'free' inversion of subjects in several of the Romance languages is stylistic by Emonds' criteria. Yet Chomsky (1981) 240ff. treats the rule as a grammatical instance of 'move NP' since it is one of the properties of the 'pro-drop parameter' which is said to be closely connected with the theory of government and the operation of agreement rules. 5 The difficulties inherent in 'scrambling' theories can be avoided if variations in freedom of phrase order between languages are accounted for directly in terms of universally available but variable (i.e. more- or lessrestrictive) linearizing principles operating on syntactic rules. 6 The grammars of languages where these principles are less restrictive will allow more than one constituent order to count as 'basic'. Gazdar and Pullum (1981) make a proposal that has these consequences. They define a grammar as a set of immediate dominance (ID) rules and a set of linear precedence (LP) rules. The phrase structure grammar induced by such a grammar definition consists of all the phrase structure rules (which express dominance and precedence relations simultaneously) that are consistent with some ID rule and all LP rules. This 'ID/LP format' proposal presupposes that grammars have a particular property, which Gazdar and Pullum call exhaustive constant partial ordering (ECPO) — i.e. the LP rules (partially) linearizing the right hand side of ID rules expanding any one category will also linearize in the same way the right hand side of the ID rules expanding all other categories. If the grammars of natural languages universally possess the ECPO property the ID/LP format provides for a universal formulation of syntactic rules without the loss of descriptive adequacy (and empirical content) in the case of the grammars of free phrase order languages. The ID rules of the grammars of different languages turn out to be very similar; in many cases they may reasonably be expected to be identical. 7 Languages will differ, however, in the number (though not the type) of LP rules their grammars employ. The more LP rules there are in a grammar, the greater is the number of categories that are linearized with respect to each other. Consequently, relatively fixed phrase order languages will have grammars that employ a number of LP rules, and relatively free phrase order languages will have grammars employing relatively fewer LP rules.8

Geoffrey Horrocks

97

In section 2 I show h o w the rather complex facts of phrase order variation in modern Greek can be represented simply and compactly by a context-free phrase structure grammar put into ID/LP format. I assume the basic descriptive framework of GPSG as presented in Gazdar (1981, 1982) and many other works.

2. CONSTITUENT ORDER IN MODERN GREEK

I assume the ID rule (1) for sentences: 9 (1)

V2

N2[a]

VI [a]

(a = [+nominative], together with a permissible combination of person, number and gender features) and the LP rule (2): (2)

a

< HI (where a is any category, and HI represents the head of a phrase at the first bar level - see Gazdar and Pullum 1981, section 2)

I now state some ID rules for VI : 1 0 (3)

VI

(4)

VI

^

V

fevyi leaves

V, N2

aghorazi to buys

dhoro

the gift

(5)

VI

-»• V, P2

exartate apo ti yineka tu depends from the wife of-him

(6)

VI

-> V, VI [na]

prospathi na tries subj.

(7)

V1

-*• V, V2[pos]

lei pos o Sokratis ine exipnos says that the Socrates is clever

(8)

VI

(9)

VI

V, V2[na]

-»• V, N2,

sikothi get-up-3s

theli o Sokratis na yini wants the Socrates subj. become-3s filosofos philosopher

N2[GEN] dhini to dhoro tu Sokrati gives the gift the-gen. Socrates-gen.

The order of constituents in modern Greek VI

VI

VI

VI

V, N2, N2[ACC]

-+ V, N2, P2

V, N2, V I [na]

V,

N2, V2[pos]

matheni arithmitiki tin teaches aritmetic the-acc. Afrodhiti Aphrodite-acc. dhini to dhoro sto Sokrati gives the gift to-the Socrates pithi tin Afrodhiti na persuades the Aphrodite subj. aghorasi to dhoro buy-3s the gift pithi to persuades the Afrodhiti ine Aphrodite is

Sokrati pos i Socrates that the omorfi beautiful

VI -> V, P2, P2 [na]

milai sto Sokrati ya tin speaks to-the Socrates about the Afrodhiti Aphrodite

VI -* V,

iposkhete stin promises to-the na fai to subj.eat-3s the

VI

P2, VI [na]

-> V, P2, V2[pos]

Afrodhiti Aphrodite musaka moussaka

iposkhete stin Afrodhiti pos promises to-the Aphrodite that o Sokratis tha erthi the Socrates will come-3s

VI -*• V, N2, P2, V2[pos] stikhimatizi ekato dhrakhmes me bets 100 drachmas with to Sokrati pos i Afrodhiti the Socrates that the Aphrodite tha kerdhisi to vravio will win-3s the prize (N2 unspecified for case is accusative.)

Geoffrey

99

Horrocks

A few explanatory comments are in order here. Modern Greek has no infinitives; where English would have an infinitive, Greek employs a (finite) subjunctive verb marked by the proclitic particle na (which I gloss as 'subj[unctive]'). If rule (8) is compared with rules (6), (12) and (15), it will be seen that verbs may be subcategorized to take either sentential or phrasal 'subjunctive' complements. The phrasal type obviously involves control, but since the verb is finite, it must agree with its controller in person, number and gender. This could be represented by the use of agreement features in a familiar way. 11 It is important to note that only the phrasal type is freely permutable with N2 and P2 complements (see below). Sentences, whether they contain indicative or subjunctive verbs, must come last in a string of complements (at least in non-dislocated structures). Notice that this distributional difference provides one piece of evidence that Greek has a category VI distinct from V2 (see the chapters by Borsley and McCloskey in this volume for further discussion of the status of VP and S), and hence provides some evidence in favour of rule (1). It has been argued that Greek sentences with SVO order are invariably the result of topicalization operating on a basic VSO pattern, and that therefore rule (1) makes false claims about Greek sentence structure (see Warburton (1980), and cf. fn.9). Further evidence in support of rule (1) will be pointed out as we go along. It is also worth stressing that if we were to follow Chomsky (1981) pp. 24-5 et passim and insist that apparently phrasal complements are in fact sentential in character, we would have to argue that the positioning of complement sentences with respect to the other complements of V with which they co-occur depends on whether or not they have controlled PRO subjects. It is surely simpler to argue that two distinct complement types are involved, permutable VI and nonpermutable V2. Given this, we can provide a very straightforward LP rule schema for complements of a lexical head: 1 2 (18)

H


V, X>

V[a],N2[a], X>

These two metarules can be collapsed into (20): (20)

< VI [a]

Notice that, given (18), we predict that subject N2 in 'flat' sentences behave distributionally just like N2 subcategorized by V. Thus the free inversion property of pro-drop languages (Chomsky (1981) p. 240) follows automatically. As an illustration, the ID and LP rules given so far induce a set of context free phrase structure rules that will generate all the following grammatical sentences: (21)

o the tha will

Sokratis stikhimatizi dhyo lires me ton Aristoteli pos Socrates bets two pounds with the Aristotle that vrexi rain

stikhimatizi o Sokratis dhyo lires me ton Aristoteli pos bets the Socrates two pounds with the Aristotle that tha vrexi will rain stikhimatizi dhyo lires o Sokratis me ton Aristoteli pos bets two pounds the Socrates with the Aristotle that tha vrexi will rain stikhimatizi dhyo lires me ton Aristoteli o Sokratis pos bets two pounds with the Aristotle the Socrates that tha vrexi will rain o the tha will

Sokratis stikhimatizi me ton Aristoteli dhyo lires pos Socrates bets with the Aristotle two pounds that vrexi rain

Geoffrey Horrocks

101

stikhimatizi o Sokratis me ton Aristoteli dhyo lires pos bets the Socrates with the Aristotle two pounds that tha vrexi will rain stikhimatizi me ton Aristoteli o Sokratis dhyo lires pos bets with the Aristotle the Socrates two pounds that tha vrexi will rain stikhimatizi me ton Aristoteli dhyo lires o Sokratis pos bets with the Aristotle two pounds the Socrates that tha vrexi will rain 'Socrates bets Aristotle two pounds that it will rain.' They also allow for the permutation of VI but not of V2: (22)

episa ti mitera mu na ftiaxi musaka persuaded-ls the mother of-me subj. make-3s moussaka episa na ftiaxi musaka ti mitera mu persuaded-ls subj. make-3s moussaka the mother of-me 'I persuaded my mother to make moussaka.' o Sokratis theli i Afrodhiti na ton filisi the Socrates wants the Aphrodite subj. him kiss-3s *theli i Afrodhiti na ton filisi o Sokratis wants the Aphrodite subj. him kiss-3sthe Socrates 'Socrates wants Aphrodite to kiss him.' stikhimatizi o Sokratis pos tha vrexi bets the Socrates that will rain *stikhimatizi pos tha vrexi o Sokratis bets that will rain the Socrates 'Socrates bets that it will rain.'

The starred examples are ungrammatical under normal stress andintonation. They are acceptable only when the subject is identified by a marked stress and intonation pattern either as a clarificatory topic or as a highly emphatic focus (perhaps most naturally as the focus of an incredulous echo question). The rules for this kind of topicalization and focalization involve the introduction of derived categories (see below). However, before discussing such 'long-distance' topicalization (and focalization), we have first to allow for a process that might be called 'local' topicalization. It is possible in the case of direct and indirect objects

102

The order of constituents in modern Greek

(inflected in the accusative and genitive respectively) to mark these constituents explicitly as topics without recourse to 'displacement'. The device employed for this purpose is 'clitic doubling', i.e. a clitic pronoun is introduced that agrees with the object N2 in person, number and gender. In the case of subjects, even though there is no overt marking of topic status, it seems natural to assume that the features of subject N2 are copied onto V2 and interpreted as a 'null clitic pronoun'. Suppose we have a metarule like (23): (23)

X,

N2 [a] >

< V2 [a] -»• X, N2[a, +TOP] > (a = a combination of [+DEFINITE] with case, person, number, and gender) This introduces 'flat' V2 rules where V2 agrees with subject or object N2, and 'configuration^' V2 rules where V2 agrees with subject N2. The agreement features are carried down onto V by the Head Feature Convention. Subject features are realized as an inflectional suffix on the verb in the usual way, and object features are spelled out as proclitic pronouns by a rule such as (24): 13 (24)

V[a]

-»•

cl[a] + V

(a includes [+ACC] or [+GEN])

The phrase structure rules induced by the output of (23) and the LP rules already given will generate sentences like (25): (25)

ti filise tin Afrodhiti [+TOP] o Sokratis[+TOP] her kissed the Aphrodite the Socrates 'Socrates kissed Aphrodite.' (Where we already know that Socrates did something to Aphrodite.)

We are now ready to introduce rules to handle topicalization and focalization proper. Consider the following examples, where one occurrence of ti Maria must appear: (26)

(ti Maria,) su ipa pos (ti Maria,) pistevo oti the Mary to-you said-ls that the Mary believe-ls that (ti Maria,) theli (ti Maria,) na ti filisi o Yanis the Mary wants the Mary subj. her kiss-3s the John 'Mary, I told you I believe John wants to kiss'

Geoffrey Horrocks

103

{ti Maria) su ipa pos {ti Maria) pistevo oti the Mary to-you said-ls that the Mary believe-ls that (ti Maria) theli {tiMaria) na filisi o Yanis the Mary wants the Mary subj. kiss-3s the John 'It is Mary that I told you I believe John wants to kiss.' Apart from differences of stress and intonation between topicalizations and focalizations, there is one important syntactic difference. Topicalization of a N2 is always associated with clitic doubling while focalization never is. So in the first example in (26) we find the pronoun ti before filisi; this is obligatorily absent in the second example. It is also important to notice that when a constituent appears pre-verbally, as when ti Maria occurs before na filisi in (26), it must be interpreted either as a topic or as a focus. The only exception (other than clitics) is subject N2. This is made clear by an example such as that in (27), which is the very first sentence from a popular Greek novel: (27)

meta ta Khristuyena tu 1974 mnya paly a ghnosti after the Christmas of-the 1974 an old acquaintance-fern, mu me plevrise of-me me came-up-to 'After the Christmas of 1974 an old acquaintance of mine came came up to me.'

Since this is the first sentence in the book the preverbal subject cannot be either a topic or a focus; it cannot be either 'given' (notice that it is indefinite) or 'emphatic/contrastive'. It is just a subject. This special status of subjects is captured by rule (1). Subjects can, of course, be 'displaced' by long distance topicalization and focalization just like other N2: 14 (28)

o Sokratis, thelo na dhi ti dhulya mu the Socrates want-Is subj. see-3s the work of-me 'Socrates, I want him to see my work.'

To allow for the set of options illustrated in (26) we need a rule schema for (non-local) topicalization and focalization of the form: (29)

V2 a, V2/a[±TOP] (a = a variable over the set of topicalizable/focalizable constituents.)

104

The order of constituents in modern Greek

The feature [+TOP] indicates that the missing constituent in V2 has been topicalized, the feature [-TOP] that it has been focalized. It might seem on the basis of the facts presented so far that these features are redundant. For most values of a there is no syntactic difference at all between topicalization and focalization. Even when a = N2 the presence or absence of clitic pronouns follows from the treatment of local topicalization; only dependencies into locally topicalized positions (i.e. topicalizations) will involve clitic doubling. However, as we shall see below, topicalization may apply several times in succession within a single sentence while focalization may not, and there are restrictions on the way in which the two rules interact, so it is clearly necessary to draw a distinction between them. The features [±TOP] provide a simple way of doing this without at the same time losing the obvious generalization. But before pursuing these questions, let us first take a few simple examples of structures generated by rules of the type specified in (29) (together with the relevant derived rules): V2

(30)

V2/V1 na -TOP

VI [na]

V

N2

V

V2/V1 [na] I" na -TOP

na dhi ti dhulya me thelo subj see-3s the work of-me want-Is N2

o supervizor the supervisor 'I want my supervisor to see my work.''

VI/VI [na] |~na -TOP e

Geoffrey Horrocks

105

V2 N2

V2/N2 [+TOP]

ti Maria the Mary

V | pistevo belie ve-Is

V2/N2 [pos] [+TOP]

pos that

V2/N2 [+TOP] V

N2/N2 [+TOP] [+TOP] V

ti filise her kissed

e

N2

A

o Yanis the John

'Mary, I believe John kissed.' There are two things to note here. First, focalization of VI in the first example provides more evidence for VI as a category of modern Greek syntax. Secondly, there is no LP rule relevant to the rules in the set defined by (29). Consequently we get foci and topics 'displaced' to the right as well as to the left: 1 5 (31)

episa na ftiaxi ti mitera mu kulurya persuaded-ls subj. make-3s the mother of-me doughnuts 'I persuaded my mother to make doughnuts.' episa na ta ftiaxi ti mitera mu, ta kulurya persuaded-ls subj. them make-3s the mother of-me the doughnuts 'I persuaded my mother to make them, the doughnuts.'

A subject, or indeed other constituents, may in these circumstances appear to the right of a complement sentence (see (22) above and the associated discussion). There is an interesting issue that arises in connection with subject dependencies. Gazdar (1981) explains the phenomena handled by the "that-trace filter" of Chomsky and Lasnik (1977) in terms of a Generalized Left Branch Condition, which prohibits dependencies into left branches, and hence into subject position in languages like English. Where there appears to be a subject dependency, this is accounted for by the intro-

106

The order of constituents in modern Greek

duction of a tensed Vl which naturally explains the absence of a complementizer in these cases. As far as Greek is concerned, the i/wi-trace constraint is apparently inoperative. Sentences such as those in (32) are wellformed: (32)

pyos ipes pos irthe? who said-2s that came-3s 'Who did you say [that] came?' Yanis mu ipes pos ine exipnos o the John to-me said-2s that is clever.' 'It is John that you told me [that] is cleaver.'

One might argue that the GLBC holds as much in Greek as in English, and that these are examples of dependencies into post-verbal subject positions (cf. Chomsky (1981a) p. 253ff.). However, it is clear that the GLBC does not operate in Greek because of examples like: (33)

poso ine psilos o Yanis? how-much is tall the John 'How tall is John?'

There is, therefore, no reason to block dependencies into subjects on left branches. The really interesting question is whether the fairly general (genuine?) violation of the that-trace constraint in pro-drop languages can always be explained in terms of the non-operation of the GLBC, and if so, why the operation or non-operation of the GLBC should be connected with pro-drop. Discussion of such issues is beyond the scope of this paper. Returning now to the rule schema (29), it is clear that we must allow for complex derived categories in Greek because, as was pointed out above, there may be multiple application of the topicalization rule combined with application of the focalization rule:16

Geoffrey

Horrocks

107

(34) V2/N2 [+TOP] o Socratis the Socrates (V2/N2)/N2 [+TOP] [+TOP] to vivlio the book ((V2/N2)/ N2) / N2 [-TOP] [+TOP] [+TOP] tu Aristoteli to-the Aristotle V

N2/N2 [+TOP] [+TOP]

N2/N2 N2/N2 [-TOP] [+TOP] [+TOP1

to edhose it gave-3 s 'Socrates gave the book to Aristotle.' crates gave the book to someone.)

(Where we already know that So-

Notice that sentences with multiple focalization are ungrammatical: (35)

*o Yanis ti Maria mu ipes the John the Mary to-me said-2s 'You told me John saw Mary.'

pos idhe that saw-3s

as are sentences in which a topicalization is contained within a focalization: (36)

*o Sokratis, to vivlio tu Aristoteli, tu edhose the Socrates the book to-the Aristotle to-him gave-3s 'Socrates gave the book to Aristotle.' (Where we already know that Socrates gave something to Aristotle.)

It is interesting to note that w/i-questions have a number of characteristics in common with localizations. They art never associated with clitic doubling, there cannot be more than one 'extraction' of a vWi-phrase, and a wft-question cannot contain a focalization or a topicalization:

108

The order of constituents in modern Greek

(37)

*pyon ipes pos ton idhes? whom said-2s that him saw-2s 'Whom did you say that you saw [him] ?' *pyos ti mu ipes pos idhe? who what to-me said-2s that saw-3s 'Who did you tell me saw what?' *pyos ti Maria mu ipes pos idhe? who the Mary to-me said-2s that saw-3s 'Who did you tell me sawMaryV *pyos ti Maria, mu ipes pos tin idhe? who the Mary to me said-2s that her saw-3s 'Who did you tell me saw Mary?' (Where we already know that someone saw Mary.)

Suppose, then, that the rule schema for wh-questions is (38): (38)

V2 a [+wh], V2/a[-TOP] (a = any category that can be the 'focus' of a constituent question.)

This ensures that we do not get dependencies into locally topicalized N2 positions, and so guarantees the absence of clitic pronouns (cf. the first example in (37)). We can now formulate a very simple constraint that will rule out the examples in (35) and (36) together with the last three examples in (37), without blocking examples like (34): 17 (39)

*(a//3)/r [-TOP]

The effect of this is to prohibit focalizations and w/z-questions that contain other constructions involving long distance dependencies. Finally, I must point out one further piece of evidence that the N2 introduced by rule (1) are simply subjects (and not topics or whatever). While the sentences in (37) are clearly ungrammatical, the following sentence is generally felt to be grammatical but (rather) unacceptable: (40)

pyon o Yanis sinandise? whom the John met-3s 'Whom did John meet?'

If o Yanis were a topic, this should be as bad as the sentences in (37) because it would contain a derived category of the type prohibited in (39). The different status of (40) follows automatically, given rule (1).

Geoffrey Horrocks

109

3. CONCLUSION

The rules and constraints that I have given account for all the possible permutations of phrase order in declarative sentences of modern Greek. This has been done in a framework that eschews both grammatical and 'stylistic' transformations, and employs only the well-motivated apparatus (slightly generalized to allow for complex derived categories and rules to expand them) presented in Gazdar and Pullum (1981, 1982) and Gazdar (1981, 1982). Furthermore, this highly restrictive framework has permitted the facts to be described very straightforwardly in terms of simple linearization rules and a general schema for unbounded dependencies. It is also worth adding one final, theoretical, point. 18 Hale (1981) has argued that some languages have a phrase structure component in their grammars and that others do not. The former he calles 'X-bar' languages; these are supposed to exhibit NP+VP sentence structure together with (relatively) fixed phrase order, and to employ rules of the type 'move a'. The latter he calls 'W(ord)-star' languages; these are supposed to exhibit 'flat' structure under S together with very free constituent order, and not to employ rules of the type 'move a \ Against this view Lapointe (1981) has argued not only that the W* theory gives no explanation of the phenomenon of free constituent (rather than word) order, but also that the supposedly discrete properties of W* and X' languages do not in fact fall into two mutually exclusive sets. The analysis of modern Greek presented here has provided further evidence of this. The language has a NP+VP sentence rule in its grammar, but also has 'flat' sentence rules. It allows long distance dependencies of the type that transformationalists might analyze in terms of a rule of w/z-movement, but also exhibits very free phrase order. It seems, then, that the X"/W* distinction lacks secure empirical motivation. We might also conclude that it is theoretically otiose, in that languages such as modern Greek can be analyzed successfully in terms of exactly the same formal apparatus as that required for the description of fixed phrase order languages.

FOOTNOTES * I am extremely grateful to Melita Stavrou for the many hours she spent providing me with information about her native language. Thanks are also due to Bob Borsley, and to the editors of this volume, for much helpful comment and criticism. I have not always followed the advice o f f e i d , and responsibility for any errors or deficiencies that remain lies, of course, with me. 1. Ross ( 1 9 7 0 ) treats Hindi (surface SOV) as an 'underlying' SVO language; McCawley ( 1 9 7 0 ) treats English as a 'deep' VSO language; Tai ( 1 9 7 3 ) treats Mandarin Chinese (surface order usually SXVO) as underlyingly SOV; and so on. 2. One advantage of this treatment is that it permits a strictly local account of

110

The order of constituents in modern Greek

subcategorization. This locality is maintained within the framework of generalized phrase structure grammar where VSO and OSV languages can be analyzed in terms of a 'phantom' VP category, i.e. a VP whose internal structure is defined by the grammar without there being a rule that makes VP a possible daughter of S (see Gazdar and Sag (1981). 3. This theoretical decision has clearly been self-justifying: "It seems that very little has been accomplished in the elaboration of linguistic theory by reference to the properties of languages in which major constituents of a clause can be freely reordered" (Emonds (1980) p. 39). 4. These are as follows: optional, clause-bound, incapable of producing ambiguity with structures generated by non-stylistic rules, formulated without reference to particular grammatical morphemes, often associated with structures that strike the hearer as a turn of phrase used for emphasis or special effect (Emonds (1980) pp. 40-44). 5. Cf. also the controversy surrounding the status of 'Heavy NP Shift' (Fiengo (1974), Pullum (1979) pp. 142-6). 6. For a far less coherent approach to realizing the same 'leading idea' see Chomsky (1981) p. 92ff. 7. Thus even VSO and OSV languages will have an ID rule for, say, transitive verb phrases given the Gazdar/Sag approach (see note 2). 8. Gazdar and Pullum (1981) show how Makua, a very free phrase order language, employs a proper subset of the LP rules of English, a relatively fixed phrase order language. 9. This apparently uncontroversial rule is, in fact, quite controversial in the context of the grammar of modern Greek (cf. Warburton (1980). I shall take some care to justify it as the discussion proceeds. See also Borsley's and McCloskey's chapters in this volume, for a variety of interesting ideas concerning the status of VP and S. 10. This list is not exhaustive, but gives a representative sample of the types of complement structure verbs may have in modern Greek. 11. The interesting possibility of a semantics based approach to feature distribution and feature transfer is raised in Sag and Klein (1982). Since I provide no semantic rules here, this option will not be discussed further. 12. As one would hope, this makes correct predictions with respect not only to complements of V but also to complements of N and Adj. I shall not tire the reader with the relevant examples. 13. Since it is possible that projections of V will have to carry subject, direct object and indirect object agreement features simultaneously, it is clear that these will have to be structured in some way if they are not to be confused. One might follow Stucky (1981) and treat combinations of subject and object agreement features as ordered pairs (or triples) of the form < a, p, (y) >. 14. Notice that SVO sequences may be generated directly by rule (1), or may involve topicalization of focalization of the subject N2. This explains why preverbal subjects may be interpreted as topics or foci, but do not have to be. It is also worth pointing out that in the case of topicalization this may involve either the 'local' rule or the 'long distance' rule in its minimal application. Such multiple analyses are harmless. 15. Gazdar (1981) explains the apparent 'boundedness' of rightward dependencies of this sort in terms of performance considerations. Frazier (1979) and Frazier and Fodor (1978) argue for a parsing model which attaches incoming material as low as it possibly can on the parse tree. If the boundedness of rightward dependencies can be explained in this way, there is no need to impose any special constraint on the syntactic rules that admit them.

Geoffrey Horrocks

111

16. The possibility of multiple dependencies is discussed in Gazdar (1982, section 10) with reference to work by Rizzi (1978) on Italian, McCloskey (1979) on Irish, and Maling and Zaenen (1982) on the Scandinavian languages. Some might object that what I have called topicalization here is in fact left-dislocation, and that really there is no 'hole' in these cases but rather a resumptive pronoun. If this were so, it might be argued that the issue of complex derived categories does not arise, since the relationship between a left-dislocated N2 and its pronoun would be one of simple 'anaphoric' binding and there would be only one gap in sentences like (34) corresponding to the focalized N2. This approach runs into difficulties, however, when we consider Greek f/zaf-relatives, introduced by the complementizer pu. There may be an across-the-board dependency into conjoined VI where in one conjunct there is a pronoun and in the other there is a gap: aftos ine o anthropos pu ton sinandisa ke edhosa ta lefta this is the man that him met-ls and gave-Is the money 'This is the man that I met [him] and gave the money.' The main principle of the theory of coordination is that only like categories can be conjoined (Gazdar (1981)). So, unless this is to be abandoned, both conjunctsin the example above must be V1/N2. One might argue that syntactic binding was involved in both cases, but that N2/N2 was realized variously as either a gap or a pronoun. This would leave us with the problem of explaining why only relativization allows this dual realization (and then not in all cases). Thus topicalizations require a pronoun, while localizations and wh-questions require a gap. The simplest solution is to do as I have done and argue that 'syntactic' binding is involved in all cases of unbounded dependency, but that rules vary according to whether or not they permit dependencies into locally topicalized positions. If they do, the gap will be associated with clitic doubling, and if they do not, the gap will not be associated with clitic doubling. Relativization, unlike topicalization or focalization, is presumably indifferent to the status of the N2 positions involved, so that sometimes there is a pronoun and sometimes there is not. In any case, it is obvious from the (grammatical example below that the wh -island constraint does not hold absolutely in Greek, and that 'multiple dependencies' into a single constituent must be allowed for independently of the topicalization/focalization facts: aftos ine o anthropos pu anarotyeme pyon idhe this is the man that wonder-Is whom saw-3s 'This is the man that I wonder whom [he] saw.' 17. This constraint may turn out to be much more general in character, and may need to be reformulated in semantic terms as a condition on operator binding based on some notion of 'accessible scope'. For the present (39) guarantees that topicalization, vv/i-question formation and focalization interact properly. 18. The same point has been made independently by Pullum (1982) and Stucky (1981).

Chapter 6

Raising in Latin: a phrase structure analysis of the accusative and infinitive Ronald Cann

0.

INTRODUCTION*

The analysis of the accusative and infinitive (A & I, henceforth) in Latin has long been a matter of debate amongst Latinists. The problem to be confronted centres around the status of the accusative noun phrase and the possible passives of sentences containing an A & I clause. The sentence (a) can have either the two sentences ( l b ) or ( l c ) as its passive equivalent: 1 (1)

a. Omnes Belgas Caesarem necauisse all(nom.pl) Belgae(acc.pl) Caesar(acc.sg) kill(prf.inf) dicunt say(3pl.pres.act) 'Everyone is saying that the Belgae have killed Caesar' b. Belgae Caesarem necauisse ab omnibus Belgae(nom.pl) by all(abl.pl) dicuntur say(3pl.pres.pass) 'The Belgae are said to have killed Caesar by everyone' c. Belgas Caesarem necauisse ab omnibus dicitur say(3sg.pres.pass) 'It is being said by everyone that the Belgae have killed Caesar'.

How should we, then, analyse the accusative noun phrase 'Belgas' in (la)? The existence of the sentence (lb), taken on its own, would seem to indicate that it be analysed as the direct object of the main verb 'dicunt', which has been preposed to subject position in (lb), in the normal way. The structure of ( l a ) would have to be something like (2): 2

114

Raising in Latin: a phrase structure analysis

However, the existence of sentences like (lc) would seem to be evidence that the whole A & I clause is dependent on the main verb which can be preposed to subject position in passive sentences, giving (la) the structure in (3):

To account for this data, some earlier analyses have made use of both structures illustrated in (2) and (3), although giving them different emphases. The usual traditional view, summarized in Bolkestein (1976, 1979), was that the accusative noun phrase was originally a simple direct object, whilst the infinitive, being the dative/locative of a verbal noun, had an adverbial function dependent on the main verb. In time the case ending of this verbal noun became obscured, leading to the loss of recognition of its adverbial nature. This, in turn, coupled with the habitual association of the prolative infinitive with some finite verbs, led to those verbs losing their 'independence' and becoming strictly subcategorized for infinitives (to anticipate later terminology). With ordinary transitive verbs, so the argument runs, this new infinitive was felt to be a simple direct object, whereas with verbs that already had a noun phrase object, the infinitive was regarded as a second object, rather than an adverbial. The next stage in the formation of the A & I was for the infinitive to lose its dependence on the main verb and become more attached to the accusative object, thus creating a new constituent dependent on the main verb (cf. Woodcock (1959) pp. 14-17). The transformational account advanced in Pepicello (1977) is a synchronic description of the A & I, as opposed to the traditional diachronic approach sketched above. However, the synchronic analysis owes much to the historical one, retaining the dual structure approach while, in a certain

Ronald Canti

115

sense, reversing the order in which they appear. Briefly, then, Pepicello assigns the A & I sentential status in deep structure, with the surface accusative as subject of the infinitive, the whole clause being dependent on the main verb. To this structure a transformational rule applies, raising the subject of the embedded sentence into direct object position in the main clause. Further rules apply to ensure that the verb of the embedded sentence appears in its non-finite form. To this transformed structure the ordinary passive transformation may apply to give the personal passive, as found in (lb). 3 Bolkestein (1979) is an extensive criticism of both the traditional view of the A & I, and of Pepicello's transformational analysis. She argues against the accusative noun phrase being a direct object of the main verb, at any time or at any stage of a derivation, and claims instead that the construction should have the status of a non-finite sentence. Although in that paper Bolkestein offers no account of the possible passives, she does hint that perhaps some form of raising to subject (of the accusative subject of the A & I) might be involved in the personal passive. In an earlier paper (Bolkestein (1976)), she gives an analysis in terms of topic/focus distinctions. Pepicello's analysis has also been attacked from a transformational point of view. In Pillinger (1980), arguments are brought forward against any rule of Subject-to-Object raising for Latin, and consequently in favour of the non-direct object status of the accusative. Pillinger himself prefers a raising to subject analysis of the personal passive, whilst pointing to some difficulties that might be encountered in dealing with the Latin A & I by the extended standard theory or the theory of relational grammar. It is not the purpose of this chapter to review earlier discussions of the accusative and infinitive in Latin, beyond the brief mentions given above, nor to argue specifically for or against them. Rather, I shall present an analysis of the construction within the framework of generalized phrase structure grammar (GPSG). This account will adequately cover all the data found in (1) above, as well as certain other features of the non-finite verbal complement system of Latin. Although I shall not look at the Bolkestein or Pillinger papers in any more detail, I shall assume, without further argument, one of their conclusions, namely the analysis of the A & I clauses as a non-finite sentence dependent on the main verb. Thus, the structure of (la) will be assumed to be (3) rather than (2). The exact analysis of this sentence ( l a ) will be given in section 2.3., before which we must look at the notion of lexical transitivity and introduce some simple rules.

116 1.

Raising in Latin: a phrase structure analysis

PRELIMINARIES

In what follows I adopt the form of GPSG that is to be found in Gazdar (1982). Any departure from the assumptions of this paper will be noted in the text. I shall ignore for the purposes of this chapter the freedom exhibited by Latin in the ordering of constituents. Thus, although examples may show some variation in the order of subject, main verb and its dependent complements, rules will be given for one order only, that of subject-object-verb. I assume that the reordering of constituents could be given simply enough, either by using a system of metarules as done by Stucky (1981) in her analysis of Makua, or by using the ID/LP format of Gazdar and Pullum (1981), Pullum (1982) and Horrocks' chapter in this volume.

1.1. Lexical transitivity I present first a preliminary sentence expansion rule which, apart from case and agreement marking, is identical to that given for English in Gazdar (1982) p. 16: (4)

(where a is a combination of number and person features)

This rule will be amended in a later section to cover non-finite sentences. A great deal of the following analysis will depend on a notion of lexical transitivity. That is, transitivity is held to be an inherent feature of a verb that remains constant (like all lexical features) no matter what linguistic environment surrounds it. In this way, transitivity and strict subcategorization for an NP complement must be distinguished. Chomsky (1965) claims that marking a verb with the feature [Transitive] is notationally equivalent to marking it with the strict subcategorization feature +[ NP], and, therefore, only one of these features need appear in the lexical entry for a verb. However, in a discussion of ergative case marking and passive in Hindi-Urdu, Amritavalli (1979) points out that there are (at least) two problems that are encountered with this approach: (i) transitivity is treated as a feature relevant to lexical insertion rather than as a property relevant to the functioning of transformational rules, leading to undue complication in the statement of rules; (ii) each listed context of occurrence of the verb is allowed to impose a corresponding subclassification on the verb, precluding the generalization that a single subclass of verbs may appear in more than one context. (Amritavalli ( 1 9 7 9 ) p. 91).

Ronald Carni

117

Point (i), while not being directly relevant to a grammar that does not use transformational rules, nevertheless does hold for different rule-types such as are found in GPSG. We shall also see how the separation of transitivity and subcategorization allows the generalizations mentioned in point (ii) to be made. Amritavalli concludes that, although the separation of the two notions in question is desirable, it seems more than mere coincidence that there is some connection between the two. He appeals to the ever popular, but persistently elusive, notion of markedness, taking the context [ #] to be unmarked for intransitive verbs, and [ NP] to be unmarked for transitive verbs.4 I have no intention of discussing markedness in detail, but would just like to point out an inadequacy of the above statement, at least as far as Latin is concerned. Thus, whilst, in Latin, many non-transitive verbs may appear in the environment [ #], many more verbs must be marked negatively for transitivity and yet do not appear in such an environment. While it is not true that there are transitive verbs that do not appear with NP direct objects, it is the case that some non-transitive verbs may appear with a noun phrase object, albeit generally in a different case.5 I feel that all one can say at this stage is that if a verb appears with no complements (i.e. is a number of the set V[2], see below) then, unless it is associated with a special semantic rule, it is most likely to be nontransitive.6 Conversely, if a verb is transitive, then it may appear with a noun phrase object in the accusative case. Only this second point will be adopted formally here in a lexical redundancy rule to be given below. It may well be possible to formalize the first observation in a grammar that places more emphasis on the interaction of lexical semantic types and the semantic types induced by the syntactic rules, as discussed in Klein and Sag (1982). No attempt will be made along these lines in this paper but see Cann (1982) for an analysis of the Ancient Greek verb system using this method. In accordance with the above discussion, no transitivity marking will appear in the rule that introduces verbs with no complements at all: (5)

< VI [+PAS] -> W (P2[+AGT]) V; X

("P2')] >

(where & is some function over noun phrase translations, and 3* is a variable over noun phrase intensions, of type < s « e, t > t) ). The optionality of the agentive prepositional phrase (symbolized by the feature [+AGT]), may be allowed with the assumption of the optional argument convention of Gazdar (1982) p. 30. The rules in (12) which expand the prepositional phrase are again similar to those found in Gazdar (1982), with the exception that the preposition is given a full lexical entry (12c) rather than introduced syncategorematically as a terminal symbol feature. (12)

a. c. ab: P[5, +AGT,...]; XOTiT)

This is all the apparatus needed to generate simple passives of transitive verbs such as appear below: (13)

a. Liuia a Iulia amatur Livia(nom.sg) by Julia(abl.sg) love(3sg.pres.pass) 'Livia is loved by Julia'. b. malum editur apple(nom.sg) eat(3sg.pres.pass) 'The apple is being eaten'. c.*equus curritur d.*femina a senatu parcitur.12

The ungrammaticality of the last two sentences results from the nontransitivity of the verbs curro and parco.

2. THE ACCUSATIVE A N D INFINITIVE

2.1. Prolative infinitives Before going on to discuss the A & I proper, a distinction has to be made between verbs that take real A & I complements, and those that take prolative infinitives, with or without noun phrase objects.13 The distinction between the two verb-types is maintained in traditional grammars, even where the analysis of the A & I is confused. To introduce prolative

Ronald Cann

121

infinitives without any noun phrase extension we may have a rule like (14). (14)

< 6; VI -»• V1[+INF]V; X CTiT iXx[V'(*Vl'(XX(x)))(XX(x))] } >

As might be expected, members of V[6] may be either transitive or nontransitive. The transitive subset includes the verb uolo ('wish') and its derivatives nolo ('not-wish') and malo ('prefer'). Non-transitive members of V[6] include queo ('be able') and possum ('be able'). (15)

a. uolo abire wish( 1 sg.pres.act) go-away(pres.inf) 'I wish to go away'. b. Marcus mala mauult Marcus(nom.sg) apples(acc.pl) prefer(3sg.pres.act) c. queo abire be-able( 1 sg.pres.act) go-away(pres.inf) 'I can go away'. d.*Marcus mala possit.

Those verbs that take a noun phrase as well as a prolative infinitive may also be transitive or intransitive with the case of the noun phrase being dative or accusative as for rule 3. (16)

N2 [a acc, - a dat] V1[+INF] V[a TRN]; V'(*Vl' CN2'))>

All members of V[7] may appear without the infinitive but with the noun phrase alone. This is already guaranteed for transitive verbs by LRR 1, but a separate redundancy rule is needed for the non-transitive verbs. Since a significant generalization would be missed if this rule were restricted to non-transitives (i.e. that all members of V[7] are also members of V[3]), no [-TRN] specification will appear in LRR 2. For a short discussion of the new redundancy that results with this and other lexical redundancy rules, see the final section. (17)

LRR 2: V[7] C V[3].

The following examples show the grammaticality pattern that is predicted by rule 7 and LRR 2:

122

Raising in Latin: a phrase structure analysis

(18)

a. te abire cogo you(acc.sg) go-away(pres.inf) force(lsg.pres.act) 'I force you to go away'. b. te cogo 'I am forcing you'. c. abire cogeris go-away(pres.inf) force(2sg.pres.pass) 'You are being forced to go away'. d. tibi respondere permittit you(dat.sg) reply(pres.inf) allow(3sg.pres.act) 'She allows you to reply'. e. tibi permittit 'She allows you'. f.*respondere permitteris.

2.2.

The categorial status of the accusative and infinitive

We must now turn our attention to the A & I clause itself. There are two questions to be answered: what is the semantic status of the clause, and to what syntactic category should it be assigned? That is, assuming that the A & I is a simple non-finite sentence (evidence for this analysis is given in Bolkestein (1979) and Pillinger (1980)), should it also have the status of a noun phrase, either syntactically or semantically? Firstly, of course, we need a rule to expand the A & I clause. This will be a very slightly modified version of rule 1 given as (4) above. The modification involves the case of the subject. As has been seen the subject of finite sentences goes into the nominative, while the subject of non-finite sentences is in the accusative case. This information I have included in the rule below, but, as with the assignment of case to noun phrase objects, a more general theory could give the cases of sentence subjects by general principle rather than by fiat. A difference between the finite and nonfinite sentence constructions is not reflected in the rule. Non-finite verbs do not show the same agreement properties as finite verbs. Specifically, infinitives do not carry person and number features. However, we may treat this as being a result of morphology rather than syntax and allow infinitives to carry the agreement features which are simply ignored by the morphological rules. Rule 1 may, therefore, be rewritten as (19). (19)

( 1; V2[a INF]

-> N2[-a nom, - a acc, 0] VI [j3]; Vl'(~N2') >

As can be seen from the above rule, I am assuming that non-finite sentences have the same semantic type as finite sentences; i.e. they denote truth

123

Ronald Cann

values. I am, therefore, rejecting the proposal that nominalized sentences should denote sets of sets of entities, the type of noun phrases (cf. Thomason (1976) and Klein (1979) for semantic analyses of English dependent sentences). Although there may be fairly strong arguments in favour of such a type assignment, the ascription of the type of noun phrases directly to non-finite sentences would lead to greater complexity in the grammar that would preclude the statement of generalizations that should be made. However, non-finite sentences may be given noun phrase denotations directly by generating them under a noun phrase node and using a semantic function, 3\ mapping V2 denotations into N2 denotations: (20)

V2[+INF]

V[aTRN];

All the members of V[9] may also appear with the noun phrase object or the A & I alone.

128

Raising in Latin: a phrase structure analysis

(29)

a. populo nuntiauit Caesarem Rubiconem people (dat.sg) report (sg.perf.act) Caesar(acc.sg) Rubicon(acc.sg) transisse cross(pres.inf) 'She reported to the people that Caesar had crossed the Rubicon'. b. Liuiam Postumum necauisse nuntiat Livia(acc.sg) Postumus(acc.sg) kill(perf.inf) report(3sg.pres.act) 'She reports that Livia has killed Postumus' c. senatui nuntiabunt senate(dat.sg) report(3pl.fut.act) 'They will report to the senate'. d. admonuit me Marcum warn(3sg.perf.act) pronoun(lsg.acc) Marcus(acc.sg) abisse go-away(perf.inf) 'He warned me that Marcus had gone away'. e. admonuit Marcum abisse 'He warned that Marcus had gone away'. f. populum admonuerunt people(acc.sg) warn(3pl.perf.act) 'They warned the people'.

The relevant generalizations can be captured by the following lexical redundancy rules and their associated semantic rules that bind the missing argument places: 14 (30)

a. LRR 3: V[10] C V [ 3 ] b. SLR 3: If 5' is of type «s,r Np >.«s, ¿>,«s,rNp>, r»>, then 6[3]' = \iTxJGp [6' (p) (iT)(Jl)] (iT and ¿R. are variables of types (s, r N p > ) .

(31)

a. LRR 4: V[10] C V [ 8 ] b. SLR 4: If 5' is of type «s,7" Np >,«.$/>,«s,r Np >, t»>, then S[8]' = 5'(X3xX(X)).

There are, of qourse, other ways of showing the relationships between these environments. For example in rule 10, either the noun phrase or the A & I (but not both) could be made optional with the semantics being provided by argument conventions. Alternatively, in a theory like that of Klein and Sag (1982), one could exploit the clash between head type and lexical type. I shall adopt the redundancy rule approach, however, but with the reservations voiced in the final section.

Ronald Canti

129

The transitive members of V[10], may also appear (with or without the noun phrase object) with an infinitival verb phrase rather than an infinitival sentence. The same is true for members of V[8, +TRN], as might be expected. (32)

a. uolo abire 'I want to go away'. b. te admoneo abire 'I warn you to go away'. c. admoneo abire 'I warn that I am going away'. d.*credo abire.

Notice that if there is an object then it controls the infinitive, whereas if no object appears, there is subject control. Again we may postulate two redundancy rules to show this. The membership o f V [ 1 0 ] guarantees also membership of V[6], by virtue of LRR's 4 and 5. (33)

2.4.

a. LRR 5: V[8,+TRN] C V [6] b. LRR 6: V[10,+TRN] C V [ 7 ]

The personal passive

Let us now return to the possible passives of sentences containing A & I clauses. As we saw in (lb), there is a passive where the accusative subject of the A & I appears as the subject of a passive main verb. Like all passive sentences in Latin an agentive prepositional phrase may appear. (34)

a. equus abisse a me horse(nom.sg) go-away(perf.inf) by pronoun(lsg.abl) creditur believe (3 sg .pres.pass) b. 'The horse is believed by me to have gone away'. b. Romam uenisse putati sumus Rome(acc.sg) come (perf.inf) think(3pl.perf.pass) 'We were thought to have come to Rome'. c. Marcus dicitur Iulium amare say(3sg.pres.pass) Julius(acc.sg) love(pres.inf) 'Marcus is said to love Julius'.

There are two things to notice about this construction. First, only nontransitive verbs may have the personal passive construction. Thus, (35a)

130

Raising in Latin: a phrase structure analysis

is ungrammatical, and (35b) is grammatical only through the membership of the main verb of V [ 7 ] , as can be shown by the inclusion of another object (35c). (35)

a.*Caesar uultur a populo abire b. admonitus es domum uenire warn(2sg.perf.pass) home(acc.sg) come(pres.inf) 'You were warned to come home'. c.*me admonitus es domum uenire.

The second point to notice is that no second object is allowed to appear, even though this is perfectly acceptable in the active: (36)

a. mihi Marcum Iulium amare pronoun(lsg.dat) Marcus(acc.sg) Julius(acc.sg) love(pres.inf) dicit say(3sg.pres.act) 'He says to me that Marcus loves Julius'. b.*mihi Marcus Iulium amare dicitur

Thus, only the non-transitive subset of V[8] may take the personal passive. We may, therefore, state another rule and another lexical redundancy rule to give the correct distribution. Notice that since only one output is possible metarules cannot be involved (or at least their use is redundant). 15 (36)

a.

The semantics of the output ensures that the agentive prepositional phrase, if present, is interpreted as the subject of the sentence. 16 A simple morphological rule could then account for the isomorphism between the impersonal and third singular forms of the verb. Not only are rules 2 and 3 acceptable inputs to this metarule, but so are the rules that introduce non-transitive verbs with A & I complements, rules 8 and 10. The outputs will give impersonal passives of the form given in (38) and (39) without any overt subjects and without assuming any unnatural category assignments. 17 Here, then, is the tree for sentence (lc).

Ronald Canti

133

(42)

V2 -INF +IMP +PAS V2 [+INF]

Belgas

N2 [+acc]

P2 [+AGT]

f

V 8

_ V 3 +TRN

Caesarem necauisse The above sentence may also appear with an extra noun phrase argument giving evidence that it is a metarule that is at work here (i.e. there is not a single output form as with the personal passives). (43)

Belgas Caesarem necauisse mihi dicitur 'It is said to me that the Belgae have killed Caesar'.

MR 2 also predicts that the non-transitive members of V[6] and V [7] can also appear in impersonal passive constructions, as in (44): (44)

a. potestur abire 'It is possible to go away'. b. tibi respondere permittitur 'It is permitted for you to reply'. c.*respondere permitteris

Now MR 2 gives rise to passive sentences for non-transitive verbs that are identical in form with passive sentences for transitive verbs that are given different structural analyses. This is so because of the existence of rule 9 which generates neuter pronouns under an A & I node. Thus, the sentences in (45a,b) are ordinary personal passives, whilst those in (45c,d) are impersonal passives.

134 (45)

Raising in Latin: a phrase structure analysis a. aliquid auditur 'Something is heard'. b. hoc a senatu admonitum est 'This was advised by the senate'. c. aliquid mihi dictum est 'Something was said to me'. d. hoc a populo diu putabatur 'This was thought by the people for a long time'.

The PS tree for (45c) is, therefore, as in (46) and aliquid is not a syntactic subject as it is in (45a). (46)

. V2_ -INF +IMP +PAS V2 [+INF]

N2 [+dat]

N2 +acc + PRO +ntr

mihi

V "10 - TRN

]

dictum est

aliquid This aspect of the analysis is unproblematic but there is a residual issue. When the pronominal object in impersonal passives like (45c,d) is plural the main verb shows plural morphology as if the pronoun were actually subject. Thus we have (47a) rather than (47b): 18 (47)

a. talia a barbaris creduntur such(acc.pl.n) by barbarians(dat.pl) believe(3pl.pres.pass) 'Such things are believed by barbarians', b.?talia a barbaris creditur believe(3sg.pres.pass)

I suggest that the impersonal construction was readily analyzable as a personal form in these circumstances. Parsing strategies could have had an effect on this as well as the fact that for neuters the accusative and nominative forms are identical. This reanalysis of the object as the subject

Ronald Canti

135

would lead to the ascription of agreement features to the verb to bring it in line with other finite sentence forms. Although this explanation might seem a bit ad hoc, it does not seem to be implausible nor to present an insuperable objection to the analysis of the impersonal passive construction that I have presented.

3. A F I N A L PROBLEM

Thus, with a set of eleven basic rules, two metarules, a transitivity feature and a set of seven lexical redundancy rules, we can account in a fairly elegant way with the very complex data surrounding the non-finite complement system in Latin. There is, however, one area in which the grammar fails to capture some relevant generalisations. The problem involves the set of redundancy rules. A system of such rules that does not allow any more detail of environment than the statement of a rule number, imposes the restriction that environments common to different rules cannot be stated within the lexicon. Thus, if there is an environment which is common to a set of rules and conditions the appearance of a lexical item in another environment (that may or may not itself range over a set of rules), the grammar as it stands cannot express this. That there are such environments is shown by the grammar fragment presented in this paper. The lexical redundancy rules, 5 and 6, allow transitive verbs that take an A & I complement to appear with an infinitival verb phrase instead. Two rules are needed because of the additional noun phrase allowed by some verbs. That the conditioning environment is common to both rules could be shown by a lexical rule such as the following, where y indicates the position of the lexical category and the double colon connects the two environment specifications with a transitive relation that may be interpreted as "may also appear in a rule that includes the following environment". (48)

(VI X *[+TRN] VI [+INF]>

V2[+INF]>:: < VI

-» X sf[+TRN]

It might be possible to make this rule even more general to give the personal passives of non-transitive members of V[8]. This could be done by omitting the transitivity specification from the left hand side of the rule and altering the right hand side: (49)

< VI [a PAS] -> X tp [a TRN]

V1[+INF]>

136

Raising in Latin: a phrase structure analysis

with a proviso that when a is + then X may only be an agentive prepositional phrase. Similarly, LRR's 2, 3 and 4 can be reduced to a rule that allows any verb that appears with a noun phrase object and an infinitival complement, sentence or verb phrase, to appear in an environment with either, but not both, of these complements. (50)

where n = 1 or 2, and X = N or V (Notice that I am assuming that, like metarules these lexical rules have equivalent left hand and right hand sides except where specified. This means that in (50) X2 on the right hand side can only be V2 if the left hand side has V2). It is possible that there is an alternative analysis of this data. It could be argued that this is evidence for Borsley's claim (see his chapter in this volume) that verb phrases and sentences should be given the same category status except for the specification of the feature [±SUBJ], Under this view the syntactic sections of rules 6 and 8 could be collapsed as follows: (51)

( V 2 [-SUBJ]

V V2 [+INF, ±SUBJ]>

Likewise rules 7 and 10 could be collapsed, yielding a rule like (51) but with the addition of N2 into the environment. This would eliminate the need for my proposed lexical rules. However, there are problems with Borsley's approach, especially for his semantics. Leaving general considerations aside, however, these revisions fail to capture the Latin facts. This is because, although it is the case that any transitive verb that takes an A & I may also take an infinitive verb phrase, the converse is not true. The rule in (51), therefore, will not sufficiently differentiate lexical items and will consequently lead to ungrammatical sentences.

FOOTNOTES *I am grateful to Richard Coates, Gerald Gazdar, Ewan Klein, and Geoff Pullum for invaluable comments and criticism. Errors are, of course, my own. 1. Morphological information in all the following examples is included in brackets where it is considered of importance or where the form has not been met before. The abbreviations used follow customary practise and should be self evident. 2. I am using a two level X-bar syntax as used in Gazdar ( 1 9 8 2 ) and Gazdar, Pullum and Sag (1982). 3. Pepicello does not actually discuss the impersonal passive, but since he generates the embedded sentence under a noun phrase node, I presume he would want to use some form of the ordinary passive transformation. There would be a problem for him

Ronald Canti

137

if he presumed extrinsic ordering, since for ( l b ) raising to object must precede passive, whereas in ( l c ) the reverse is true. 4. For an attempt at a definition of transitivity that tries to incorporate these remarks on markedness, see Cann (1982). 5. But see section 2.2 where a rule gives rise to the possibility of non-transitive verbs appearing with accusative objects. 6. I prefer to use the term "non-transitive" for members of V J - T R N ] , rather than "intransitive" which I associate with members of V[2, - T R N ] only. 7. In this article semantic translations of lexical items will be given as primed English words that roughly translate the Latin item. 8. Actually this rule may be more general than it appears, since it may apply to any verb with a lexical type < ( s, r ^ p >, < < s, r ^ p >, t > >, wheter transitive or nontransitive. 9. It is assumed in this paper that an item may be positively specified for one morphological feature only, if the features in question belong to the same set (cf. Cann (1981)). Thus, if an item is [+acc] then it is [ - n o m , - d a t , -gen, -abl]. 10. I shall make no attempt to analyze noun phrases in any detail, either syntactically or semantically, except for proper names which receive an extensional version of Montague's (1973) translation into intensional logic. 11. I am not completely convinced that this analysis of the passive, albeit a common one, is correct. For an alternative analysis that dispenses with optional agentive phrases see Cann (1982). 12. For a note on 'intuition' in a dead language and the difficulty of marking as ungrammatical non-attested forms, see Pillir.ger (1980) pp. 55-56. 13. A prolative infinitive is simply an infinitive verb phrase for which verbs may subcategorize. 14. The numbers of the SLRs are not consecutive but refer to their associated LRR. This is a notational convenience only. 15. The [ - T R N ] specification seen in (36b) may appear either in the syntactic rule itself or in the lexical redundancy rule as here. The two approaches are equivalent. 16. It is feasible that the [ - T R N ] specification in the metarule is otiose, if there is a general principle that guaranteed that all impersonal passive verbs are non-transitive. I suspect, however, that this is not the case. 17. Comrie (1982) points out the connection between the impersonal passives. 18. I have marked (47b) as questionable rather than as ungrammatical, as I do not know enough about early Latin to say definitely that such sentences do not occur. Although examples of this sort are found in later Latin they would seem to occur under the influence of Classical Greek where neuter plural subjects regularly take singular verbs.

Chapter 7

A categorial theory of structure building Michael Flynn

0. INTRODUCTION*

In 1901 Bertrand Russell discovered the contradiction that was later to become known as Russell's paradox. It led him to formulate a theory of logical types. 1 Several years later Stanislaw Lesniewski, working in Warsaw, decided that Russell's solution to the paradox was an 'inadequate palliative' and rejected it. He turned to Husserl's theory of meaning and developed what he called a theory of 'semantical categories'. His colleague Kazimierz Ajdukiewicz then formulated an algorithm based on Lesniewski's system to determine the well-formedness of an arbitrary string in certain languages (Ajdukiewicz 1935). It is Ajdukiewicz's idea which forms a major part of the conceptual core of the theory of language proposed in this paper. To take an example from a propositional calculus, the sentence in (1) with the indicated bracketing could be specified as well formed by the formation rules in (2). 0)

McoN1^ [Spj M c o N * ! [sq] [sr]]] or in more familiar notation, (p [D] (q A r))

(2)

S -*• p,q,r, ... CON D, A S CON S S

( p , q , r , . . . are sentences in L) (D, A are connectives in L) (If a , P are sentences in L, then CON a P is a sentence in L)

Notice that the formation rules simultaneously specify the hierarchical organization (i.e. the categorial assignments) of expressions in the language and the order in which the constituents of complex (i.e. branching) categories must appear. In Ajdukiewicz's system, however, the formation rules are replaced by (sometimes fractional) categorial assignments and a method for checking order (called 'cancellation' for obvious reasons) as in (3) and (4).

140

A categoriaI theory of structure building

(3)

p,q,r,... are each members of category S S D, A are members of category—

JO

(4)

Find a combination of categories with a fractional category in the initial position, followed immediately by exactly the same categories that occur in the denominator of the fractional category. If one of these combinations is found, replace it with the category which appears in the numerator of the fractional category.

Thus, (1) represented as (5), will be found to cancel to an S, which is to say that it is a 'syntactically connected' expression of category S.

(5)

D SS

q s

p

s I ss

r S

If there is some reason to prefer a generator to a recognizer, (4) can easily be reformulated as (6), which we will call A-concatenation (A for Ajdukiewicz). (6)

W

If a is an expression of category 1

expressions of categories

— n

, and (3j ••• P n are ^

... Yfl respectively, then a

P1

Pn is of category W. A-concatenation will generate the tree in (7).

This way of looking at things has two interesting characteristics. One is that the assignment of expressions to categories suffices to determine the hierarchical organization of phrases in the language, but says nothing about ordering within constituents. 2 In our artificial example, the assignment of a connective like 'D' to the category—- says that it will combine

Michael Flynn

141

with two sentences to make a sentence. But it does not say how it will combine. A-concatenation tells us thit. Thus hierarchical organization can be made independent of left-right order. Secondly, notice that A-concatenation makes 'predictions' about the rest of the syntax of the language. For example, suppose we were to add names like John and Mary (in, say, category NP) to the language and one-place predicates like runs and walks, which combine with NPs to form sentences (category ^ p ) - A-concatenation already specifies the order. No new rules need to be added. This, of course, would not be the case for the phrase structure rules in (2), at least in the absence of a theory of phrase structure rules which could make the prediction. Though Ajdukiewicz clearly intended his system to apply to natural languages, he was well aware that, as it stood, it did not work very well. Nevertheless, it was very useful for logics and was employed by Carnap, Bar-Hillel (see the papers in Bar-Hillel 1964) and more recently, Montague. In the meantime, it became widely assumed in generative grammar that a set of context-free phrase structure rules was a major subcomponent in grammars of natural languages.3 As the notion that these rules are universal and extremely simple was discarded, it became necessary to construct a theory of phrase structure rules that incorporated sufficient constraints to permit them to be acquired by learners. The most familiar theory of this sort is the X-bar theory, in its many instantiations. But each of the instantiations that I know of either encounters severe problems, is substantially underspecified, or both. I do not think it would be fair at this point to say that phrase structure rules, or the theories that employ them, should be abandoned, but I do think it makes sense to consider a rather different alternative. 4 This chapter argues that one attractive alternative is a categorial grammar of the sort proposed by Ajdukiewicz. Phrase structure rules are discarded entirely. Categorial assignment determines hierarchical organization of phrases universally, and specification of precedence relations (for languages which have restrictions) is provided by a single, simple principle, called the word order convention, which operates simultaneously across categories and across levels. Word order conventions are very much like A-concatenation, in that by their very nature they make predictions language wide. In this chapter, I will adduce word order conventions for three languages, English, Hopi, and Malagasy, concentrating on English for the purposes of illustration. Some of the categorial assignments are adapted from Montague (1973) (hereinafter PTQ). Although a basic familiarity with Montague grammar would be helpful to the reader, I have tried to state the main ideas of the paper using a minimum of Montague's terminology. 5 Before we turn to details, let me try to articulate the approach from a broader perspective. One goal of theoretical linguistics is to shed some

142

A categorial theory of structure building

light on how a child abducts (in Peirce's sense) a rather abstract system which in part regulates linguistic behavior. From the point of view of the X-bar theory, the idea is to constrain the possible sets of phrase structure rules so that insight may be achieved into how a child adduces one set of rules over another equally compatible with the accessible data. Or to put this another way, to give some reason why, say, it so often happens that languages do not have both the phrase structure rules in (8). (This is one way of stating Greenberg's (1963) universal 4.) (8)

VP PP

-> ->

NP P

V NP

From the point of view of categorial grammar, the nature of the problem changes somewhat. The goal here is to explain why the child adduces one word order convention over another equally compatible with the accessible data. Or, how are we to construct a theory which yields the prediction that so few word order conventions specify that NP objects precede the verb, but NP objects follow their prepositions? Serious empirical proposals about universal constraints on word order conventions and the specification of a markedness theory of categories would at this point be little more than hopeful speculation (though we will see an example of the logic of the situation in the final section of this chapter). Consequently, the hope for an illuminating comparison of the categorial theory with phrase structure grammars is premature. The goal of this chapter is more modest. It is to convince the reader that the widespread confidence in phrase structure rules just might be misplaced. 1. We begin by recursively defining a set of categories. This definition is adapted from PTQ.6 (9)

Let e and t be two fixed objects. The set of categories is the smallest set CAT such that 1. e is in CAT 2. t is in CAT 3. whenever W,Y are in CAT,— is in CAT W W 4. whenever — is in CAT, — a is in CAT, where a is N, A, or V.

The categories defined by clause 3 of (9) are called fundamental categories. Those defined in clause 4 are called word class projections. N, A, and V are called word class markings. These play a role similar to that played by the multiple slash notation in PTQ. So, for example, intransitive verb phrases are assigned to -V, intransitive nominal phrases to -N, and one

Michael Flynn

143

place predicate adjectives to—A (=Siegel's (1976) t/lie). There are no t e expressions in — (the fundamental of this category) in English (though there may be in other languages). As mentioned earlier, we will let the category assignments themselves specify hierarchical organization, while a language particular word order convention will determine left-right order. It is conceptually easier to think of these two interacting specifications in two separate steps. So let us first define categories of sets of expressions: W (10)

If a is an expression of category — and 0 is an expression of

category Y, then {a,/3 } is of category W. To see an example of how this works, we will need some expressions. We will follow Montague and assign all noun phrases to the category —— (These will translate to expressions in the logic which denote sets of properties of individual concepts. For discussion, see Lewis (1970), Partee (1975), Dowty, Wall, and Peters (1981)). Notice that the argument category of this category does not have a word class marking. By a general interpretation of word class markings with respect to (10), this means that this category can apply to the fundamental category

and not to

any of the word class projections of this category (-N, - V , —A). But e e e ^ j since, as we mentioned earlier, there will be no items in English in—,— e t will never take any arguments. We will therefore, for the sake of perspicuity* abbreviate this category as 'NP\ The reader should keep in mind that NP is not, technically, a category symbol, but merely an abbreviation of a category symbol. Intransitive common nouns like man will be assigned to the nominal word class projection of—, namely,—N. Now we may regard determiners like every as being of a category which is a function from common nouns to noun phrases, so determiners must be of category — . (10) then - N

e says that the set {every, man} is of category NP. There are two things worth noting here. One is that the functor category is always uniquely determinable; there is no category which both takes and is taken by another category. This is an essential property of categorial grammar and is what is behind the attempt to use it to resolve Russell's paradox. The second is that we have two ways of determining the category of an expression. One is semantic. We followed PTQ in assigning man to a category which guarantees that it will translate to an expression which will denote a one place predicate. The other is syntactic. I doubt that anyone has clear intuitions about the type of the expression every translates to,

144

A categorial theory of structure building

but we can determine its category by noting that it combines with -N's to form NP's. The motivation for assigning NP's to—involves both methods (see Lewis 1970). I For a second example, consider intransitive verbs like walk. We will once again follow Montague and assign them to a category which will map onto a type the meaningful expressions of which will denote one place predicates (just like intransitive nouns): —V. We will also follow a Q suggestion in Bach (1980b) and treat tenses as being of a category which is a function from intransitive verb phrases to a function from noun phrases to sentences: 7 -j^-. (10) now says that {PRES, walk} is of category—.

But now we can combine this set with the one from our first example yielding (11). (11)

{{PRES, walk} , {every, man}} is of category t.

Consider a third case. We will regard transitive verbs like love as being of a category which takes noun phrases as arguments and results in one place predicates. In other words, love is assigned to —V. Hence, if Mary e

t

NP

is an NP, (10) specifies that {love, Mary} is a—V. Analogous to oursecond example, we then have (12). (12)

{PRES,

{love, Mary} } is of category j ^ .

Finally, it is clear that we may then have (13). (13)

{ PRES, { love, Mary}} , {every, man}}

is of category t.

In this way, hierarchical organization is defined universally by the principle in (10). We can now state the principle in English that determines the left-right order of constituents of phrases. First, it will be useful to introduce some notation, originally due to Lambek (1961). ^ Consider an expression a which is of category — Suppose that j3 is an expression of category Y. If a and j3 are expressions in a language, then three possibilities exist: 8 (14)

La: 1^: *a P, fi^a Lc:

cT/3, *jTa

Michael Flynn

145

*a P,*t3 a is impossible by (10). For language L a we may write the cateW gory of a as — For L^, we may write the category of a as Y\W. This means that the argument, i.e. must appear on the left of the expression a in the resulting expression. Another way of saying this is that a is 'leftwardlooking'. For L c , we may write the category of a as W/Y. Here ¡3 must appear to the right of a, or a is 'rightward-looking'. We also define the notion of major category as follows: (15)

A major category is any category whose resultant category is t.

Expressions of these categories will always translate to expressions in the logic which denote sets. Thus, so far, we have seen these major categories: (and its word class projections),-j-(i.e. NP) — ( i . e . tensed verb phrase), and t (sentence). categories.

eT NP Categories like — (TENSE), and - — are not major —N t e —V Q

With Lambek's notation and the definition of major category in (15), we can now state the word order convention for English. (16)

Word Order Convention for English W If some phrase

< every,-— > 't -N e < Mary, NP > < love, - V > e NP

t

We will follow Partee (1975, 1976b) and assume that expression are bracketed, that brackets are labelled, and that the labelled bracketing is preserved under concatenation (though we will often suppress such bracketings for perspicuity). We can now construct analysis trees as follows. Since every is of category

NP

t -gN

t

, and man is of category-N, (10), as we e

saw, specifies that {every, man} is of category NP. The word order convention for English specifies that since every does not contain a major category, it will appear to the left of the common noun it applies to. Thus we have the analysis tree in (18).

[ N P every]

man] - N e

e

Similarly, since PRES is of category

NP

—V e

is of category—V, we have (19).

(not a major category) and walk

Michael Flynn (19)

[t

147 lt

NP

NP

[t

PRES]

[t

/ \

PRES ]

7

[t

walk]] V

walk ]

-V e

NP

t

-V e

Now (19) can combine with (18). But since (19) is of a major category, it will appear to the right of (18). (20)

[t

[ N p every m a n ]

[t

PRESwalk]]

NP"

Notice that since all tensed verb phrases are of a major category, they will all appear to the right of the subject. Hence, English is subject-initial. To take the second case we considered, love is of category —V and e

NP

Mary is an NP. Since love does not contain a major category, we have the tree in (21). (21)

[t

[t

-V e

-V e

love]

[

Mary]]

"

Mary NP

Leaving verbs with multiple complements aside for the moment, it is plain that all complements of verbs will follow the verb, since no verb which takes a complement will contain a major category. Thus the word order convention specifies that English is S V Complement. Returning to our example, PRES may apply to (21) to get (22). (22)

[t NP

[t NP

t

-V e

PRESS]

[t eV

love M a r y ] ]

148

A categorial theory of structure

building

Of course it makes no difference here that the argument contains a major category, since it is only the composition of the functor category that matters. (22) may then apply to an NP like every man. Since it's a major category, it looks left. (23)

[t

[Np

every man ]

[t

PRES love Mary ] j

~NP We have, so far, shown that the generalizations in (24) are special cases of the word order convention for English. (24)

a. determiners precede the noun b. verbs precede their complements c. subjects precede the verb phrase

We noted earlier that word order conventions make predictions language wide. In the next section, we will look at how this convention for English fares in other parts of the language. Then we will examine other word order conventions and sketch a universal theory. 2. — is a rather important category in this theory. We might call it the e

'pivot' category. 11 Some categories combine with phrases to make—'s Q (or, as in English, some of its word class projections). These are phrases that, in a theory which employs a phrase structure grammar, strictly subcategorize nouns, verbs, and adjectives. Thus, the items in (25) each combine with its argument to form a —V e . (25)

< refuse, - V > e

as in refuse the offer

"NF

< decide,-V > e

12

as in decide to leave

INF

< claim, -V > S

as in claim that ontogeny recapitulates phylogeny

Since none of the categories in (25) are major categories, all the complements will appear to the right of the verb. The same is true for nouns. Suppose we adopt (mutatis mutandis) the treatment of nominalizations proposed in Chomsky (1970) and modified in Jackendoff (1975). (A detailed exposition of this is given in Flynn (1981b)). We will have the categories in (26).

Michael Flynn (26)

149

< refusal, • N >

as in refusal (of) the offer

e

NF

< decision. 4N >

as in decision to leave

INF

as in claim that ontogeny recapitulates phytogeny

e

Again, since none of the categories in (26) are major categories, the complements appear to the right of the noun. Thus we see how one kind of cross-categorial generalization is captured by the theory. The word order convention cannot tell the difference between verbs and their nominalizations and will treat their complements the same way. Jackendoff (1977) p. 61 suggests that 'semantically, restrictive modifiers map predicates into predicates of the same number.' We will say something similar: restrictive modifiers map one place predicates into one place predicates. Since the fundamental category for one place predicates is e- , t restrictive modifiers must be assigned to the fundamental category— • e

e.

Consider now prepositional phrases. Jackendoff (1977) notes that they appear as complements to nouns, verbs, and adjectives, and further, the X-bar framework must provide a mechanism to generate an indefinite number of them in the double bar level of these categories. In the categorial theory, these generalizations are captured by assigning prepositional phrases to - X , where X is a variable over N, A, or V. c t

-X

The internal structure of these phrases is transparent. Prepositions which take noun phrases into prepositional phrases are assigned to Ax c _c NP

We therefore generate the phrases in (27). Notice that since prepositions are assigned to a category which is not major, they will appear to the left of their arguments. (27)

[t -X e 1

-X e

[t -X e t

-X e

NP

in]

[

the kitchen]]

150

A categorial theory of structure building [t

[t

e

e

-X

e

with

-X

]

Maf

[Np

yn

e NP

The prepositional phrases thus generated apply to all the word class projections of - . Since these phrases contain a major category (NP), they will appear to the right of their argument. (28)

[t -N

e

[

child]

[t

-N

in the kitchen ]]

-X

e

e

t

-X

e

[t

[t

e

e

-V

-V

run]

[t

-X

with Mary]]

e

t

-X

e

We automatically get the correct relative order between subcategorizing phrases and optional restrictive modifiers because -X cannot apply to any t

- eX

t other category besides-N,A,V. We generate kiss every child in the kitchen e as follows. (29)

[t -V

e

[t -V

e

[t -V

e

kiss ] [ N p every child ] ] [{ /

in the kitchen ] ]

-X

\

e

t

-X

e

[t

kiss every child ]

[t

in the kitchen ]

-X

e

t

-X

e

[t

-V

e

NP

kiss ]

[ N p every child ]

Michael Flynn

151

Notice that in the kitchen cannot apply to [ -^v kiss]. The prepositional NP

phrase requires that we apply kiss to its NP argument before it can apply to the result. The same is true for all categories that take complements; they must apply to their arguments before any restrictive modifiers can apply to them.

We also get the structural ambiguity of kiss every child in the kitchen straightforwardly. The reading in (29) is the one that indicates where the kissing is to take place. But the other reading, where/« the kitchen indicates who is to be kissed, in generable as in (30). 13 (30)

[

kiss every child in the kitchen ] -V

e

[t

kiss ]

[ N p every child in the kithcen ]

-V

e

NP

[ N p every ] t

[t e

-N

child in the kitchen ] N

e

[t -N

e

child ]

[t

in the kitchen ]

-X

e

t

-X

e

We could, of course, add another prepositional phrase like in the living room to (30) indicating where every child in the kitchen is to be kissed. (31)

[{ -V

e

[t -V

e

kiss every child in the kitchen ] [ t

in the living room ] ]

-X

e

t

-X

e

As in the X-bar theory, the grammar will permit prepositional phrases to pile up indefinitely, though in contrast to the X-bar theory, we require no special notation to specify this. 14 Interesting questions arise about other kinds of prepositional phrases, which we put aside for now. But we are in a position to see that two aspects of the syntax of prepositional phrases are special cases of the word order convention. One is that, regardless of the complement, pre-

152

A categorial theory of structure building

positions will appear to the left, since no preposition is of a major category. Thus, English is prepositional. Secondly, prepositional phrases, since they all contain major categories, will follow the phrases they modify. 15 We can now see why —is so central. Some categories take complements to make—'s. Then other categories apply t o — t o make new—'s. Then, finally, some categories apply to — to make e the major phrases of the g sentence, 'capping', in effect, the construction of the phrase:

The position of the phrasal head and of determiners and tensing particles is fixed by the word order convention to the left of the argument. But the position of restrictive modifiers should vary depending on whether or not they contain a major category. Prepositional phrases always have major categories in them and always appear to the right (pace, note 15). But adjectives are not so uniform. Suppose we adopt, with some modifications, the analysis of adjectives in Siegel (1976). Intersective adjactives are in—A. The copula, in—V, e

e

t -A, e

applies to them and appears to the left by the word order convention. (33)

[t [

be red ] be ]

-V e

t c

-A

[{

-A e

red j

Michael Fly nn

153

Some passives and progressives may fall under this case. Now, as Siegel suggested, non-intersective adjectives like former are i n - N . These apply t -N e to common nouns, and by the word order convention, appear to the left: (34)

[t former quarterback ] -N e [ —N e t -N e

former ]

[t quarterback ] -N e

This is so because former does not contain a major category. Now, intersective adjectives appear in noun phrases as well. Suppose we posit the following category changing rule. ( A more detailed discussion of this rule is given in Flynn (1981a)). (35)

If a e - A and translatâtes as a , then a e - N and translates e e t XPXx [ P { x } V ( x ) J . ëN

This rule can be thought of as having the same effect as a transformation. It applies to anything in —A and 'preposes' it, leaving intact its intere t pretation as a one place predicate. Thus red will also be in - N , and will e t -N e appear to the left of the common noun, just like former. Notice now that any adjective which contains a major category ought to appear to the right of the common noun. And this is correct: (36)

ENP a It It department ] [ t rife with [ N p incompetents ] ] ] ] -N -N -N e e e e

And if some passives and progressives are adjectives, they behave as expected:

154

A categorial theory of structure building

(37)

[

[t

-N

e

stolen ] [

-N

book ] ] -N

e

e

e

[t

[t

book ] [ t

e

e

e

-N

-N

-N

stolen by [ N p John]] j

e

[t -N

e

[t

sleeping [ [ t

-N

child]]

-N

e

e

e

[

[{

_N - N

e

e

child ] [ t

sleeping in [ N p the living room ]]]

-N

e e

This analysis of adjectives in English, like all analyses of adjectives in English that I know of, is not without its problems. For example, it is unclear how to account for (38). (38)

a brightly shining light a light brightly shining *a shining brightly light

(Thanks to Edwin Williams for this example). And we also require a theory of category changing rules. (For some initial steps, see Dowty (1981) and Flynn (1981b)). But I think the analysis is suggestive, and it shows how the word order convention can distinguish items that behave differently but are of the same category. (For more on adjectives in the categorial framework, see Flynn (1981a) and Barss (1981)). Of course there is more to say about restrictive modification in English, but I will restrain myself here to mentioning one other case. If relative clauses are restrictive modifiers (as seems natural), they will be in -N e

t -N

Since they always contain major categories, they will appear to the right of the common nouns they modify in English. Further, if we were to analyze the complementizer position as combining with a sentence to make a restrictive modifier, we might also have an explanation for why

Michael Flynn

155

English has a leftward COMP. (We also have to assume either that COMP is empty at the relevant stage of the derivation or that its internal structure is irrelevant to the word order convention.) COMP must b e - N , which is e

t

not a major category. Consequently, we have (39). (39)

-N e

t

NP NP t

-N

-N e

(COMP=)

t -N e

t t t It actually is not necessary to assume that relative clauses are in—N. They t

NP

NP

-N e

may also be assigned t o — , giving the structure (40), s i n c e — f o r COMP _NP

t

is not a major category. (40)

NP NP NP

NP

NP NP t Of course, it doesn't matter on either analysis whether or not the phrase to the right of COMP is a t. This category plays no role in the argument, so if further analysis were to show that it should not be regarded as a sentence, the point here would not be affected. 1 6 It can also be shown, giving some

156

A categoriaI theory of structure building

natural assumptions, that the complementizer in the S complement to verbs like claim is also predicted to be leftward by the word order convention. All that needs to be assumed is that that clauses are not themselves sentences. Suppose we represent them simply as S, keeping in mind that S f t. The complementizer, then, is in—, which is not a major category and hence will appear to the left of the sentence. Before summarizing what we have said so far, it would perhaps be worthwhile to mention one of the principal constituency tests used by Jackendoff (1977): the do so test. Jackendoff calls do so a pro-V 1 and uses this to test for membership in V 1 . Strictly subcategorizing phrases may not follow do so : (41)

Ken bought a house last year, and Bob did so last week. *Ken bought a house last year, and Bob did so a car.

We can say what amounts to the same thing in the categorial theory by assigning do so to -V (though we will not elaborate here on the anaphoric mechanisms involved). Facts like those in (41) follow immediately, because buy is not a—V, but buy a house is. However, there is a difference between the two theories. There must be some mechanism to move subcategorizing complements around restrictive modifiers (example from Jackendoff):

(42)

John said

in a loud voice suddenly at 6:00

that smoking was fun.

Said subcategorizes for S, in this case, that smoking was fun. Notice that we cannot follow (42) by (43):

(43)

*but Susan did so

softly in jest that it was bad for you at 5:00

In Jackendoffs theory, (43) must be ruled out by some (unstated) 17 independent mechanism, because the S is no longer in V 1 . In the categorial theory, the ungrammatically of (43) is expected. Recall that do so is in -V. Restrictive modifiers are thus possible, but the following S is simply Q

ungenerable, because do so cannot take any arguments. Let's summarize what we've said so far. The word order convention for English, repeated here as (44), has as special cases the generalizations in (45).

Michael Fly nn (44)

157

Word Order Convention for English W If some phrase

e

e. (persuade,

~NP ADJ -V> e

~NP INF

Michael Flynn

159

f. < consider,

-V > e

~NP ADJ -V>

g. (elect,

e

NP Pred N h. < consider, I y ) e

"NP NP We return to (46i,j) in a moment. Two sorts of rules are necessary for the generation of (46a-h) in Bach's theory. First, simple concatenation is needed to combine, say, look with its particle up to the right (and similarly for the other cases). It appears that several different rules are necessary, one for each category. Notice that in our theory, these phrases behave exactly as expected. None of the categories in (48) are major categories, so these items will appear to the left of their arguments. The second kind of rule, which combines the phrases in (47) with their argument NP's, makes use of a subfunction RWRAP (Bach (1979) p. 516): (49)

RWRAP: If a is simple, then RWRAPJa,b) is a~b. If a has the form [ x p X W ], then RWRAP (a,b) is x " b W.

The rule for combining transitive verbs with their objects is then given as in (50). 19 (50)

If a e - V and (3 e NP, then RWRAP (a, j3) e -V. _e

e

NP The phrases in (46a-h) are now generable straightforwardly. (46i,j) do not involve transitive verbs at all. Promise and strike are assigned to the categories in (51) and the phrases are then generable by rules of simple concatenation. (51)

(promise, - V > e

INF NP

(strike,

-V> e

AP NP

160

A categorial theory of structure building

In a theory like the one we are developing, we cannot appeal to subfunctions like simple concatenation and RWRAP because we do not have the right kind of rules. But the effect of RWRAP is easily stateable in the framework by means of the condition on the word order convention stated in (52). (52) is to be regarded as a language particular condition. We leave open for now whether or not it should be stated more generally and also the role played by such conditions across languages. (52)

The WRAP Condition If a is a phrase of the form [ [|3] [7] ] -X e

NP _ _ then the result of applying a to its argument is [ t j3 NP y]. -X

The phrases of (46a-h) are now straightforwardly generable, as for example, (46e) in (53). (53)

[ t persuade Mary to leave ] -V e

/

[ persuade to leave ] -V e

/

~NP

/ persuade ]

Iinf

[ N P Mary]

t0

'eave 1

'-ÌV e

NP INF Turning now to (46i,j), we see that these phrases are a problem for our theory, and we have to treat them in an ad hoc way (like just about everyone else).20 To see the problem, consider the derivation of (46i). We follow Bach and assign promise to—V , therefore getting (54). e INF NP

Michael Flynn (54)

161

[t

promise Sue ]

[t

promise ]

[ N p Sue ]

e

INF NP But now promise Sue takes an INF, and the problem is clear, for promise Sue contains an expression which is assigned to a major category (i.e. Sue) and thus should appear to the right of its argument according to the word order convention. But that would be wrong. We will have to state that promise Sue is an exception to the word order convention by assigning promise to —V/INF. (Recall that the right slash '/' indicates that the argument must appear to the right of the functor.) The derivation is as in (55). (55)

[ promise Sue to leave ] -V e

[

promise Sue ]

1 V/INF e

[ IK , F to leave ]

We can summarize this section as follows. We have shown how verbs with multiple complements can be treated in our framework. The word order convention and the WRAP condition combine to give the correct order for most of the relevant cases. Promise, as usual, has to be treated as an exception in the case where it takes an infinitival complement. For some discussion of a related phenomenon, in Irish, see McCloskey's chapter in this volume. 4. In this section, we extend the framework to some other languages and take some tentative steps towards a universal theory of ordering conventions. To recapitulate the basic idea of the theory, hierarchical organization of phrases in natural languages is given by the assignment of lexical items to categories and left-right order is determined by a language particular principle which applies across categories. Thus the child learning the language must discover at least three things: the lexical items in the language with their category assignments, rules that relate categories (such as the adjective rule and the nominalization relations),

162

A categorial theory of structure building

and the word order convention. These are by no means independent of each other, since category assignments influence the word order convention and vice versa. Like every other tightly organized theory, an adjustment in one part will have effects in every other. We will see in this section that there are many places in the framework where an analysis is not rigidly specified by the theory. Indeed, there are too many. One goal of this section is to illustrate the framework's flexibility, and then to mention some ways in which this flexibility can be reduced. As a working hypothesis, we will take an orthodox stance, for the most part, on category assignments and attribute the diversity in word order restrictions that languages exhibit to the adoption of divergent word order conventions. Of course, it is not a logical necessity that this strategy is correct, since different category assignments will give different word orders under the same ordering convention. Though we will regard this as a defeasible first assumption, we will see that it does yield encouraging results. English, as we saw in the previous sections, is a'major category sensitive' language, that is, its word order convention makes crucial mention of the notion of 'major category'. Hopi is quite different. Its word order convention does not appear to require reference to major categories at all. Rather, this language relies on the fundamental category—to make the Q

relevant distinctions for determining word order. Our proposal for its convention is (56). (56)

Word Order Convention for Hopi W t For categories where Y = - X (i.e. one of the word class proY

e

t W W jections of -). y is to be interpreted as W/Y. Otherwise, y is to be interpreted as Y\W. Another way of saying (56) is this: If a category takes a pivot category as its argument, phrases of that category will appear to the left of the argument. Otherwise, they will appear to the right. The basic word order in Hopi is SOV.21 Verbs take a range of complements: (57)

a. - V : e

NP b. ! V : e

PP

ni? mit tiyo? at tiva I the boy see 'I see the boy' ni? ?itaniy ?aw yori I our mother her-to see 'I saw our mother'

Michael Fly mi

163

c. - V

ni? Hotvel-pe tiikive-ni-qa-t navoti? yta I Hotevilla-at dance FUT OBV know 'I know that there is going to be a dance at Hotevilla'

d.-V :

ni? siivat Po?k y ~yat ?aw maqa I money Po?k y aya him-to give 'I gave money to Po?k y aya'

e

e

NP PP

Since none of the complements are in —X, the word order convention e

specifies that Hopi will be verb final. (57a) illustrates another property NP

predicted by (56): determiners

—) precede nouns. (57c) shows that -N

Hopi is postpositional, and this, too, follows from (56). Turning now to the subject phrase, recall that in our analysis of English we departed from PTQ and followed Bach (1980b) and regarded tensed verb phrases as functions from noun phrases to sentences. However, in the case of Hopi it is not so clear whether this kind of analysis is appropriate. I do not have the relevant data on which to base a decision, and for that matter, such data are rather hard to come by, even for English. It works very nicely here, as far as the syntax is concerned, to allow NP's to take verb phrases into sentences, and so we will do it that way to illustrate the alternative. What we may be seeing here is another way languages can vary, though at this stage of research, this idea must be regarded as quite tentative. At any rate, Hopi verbs are not overtly marked for present or past tense or progressive aspect. They are simply entered in what 1 suppose might be called the basic form. Jeanne (1978) glosses them variously, sometimes present, sometimes past or progressive. Consider what happens when we adopt an analysis along the lines of PTQ and have NP's make sentences out of -V's: . The word order convention then states that subjects 6

-V

will appear to the left of the VP, and we've seen above that this is correct. 22 There are auxiliary elements such as ¡-nil 'FUTURE' and /-nwi/ 'NOMIC'. We can treat these as sentence operators and assign them to , which specifies that they will appear to the right of the sentence. This is correct: (58)

a. ni? pit nopna-ni I him feed-FUT 'I will feed him' b. mi? tiyo warikn w i the boy run-NOMIC 'the boy runs (habitually)'

164

A categorial theory of structure building

Though the tense marker ends up as part of the phonological word that is the predicate, Jeanne gives evidence that these suffixes must be regarded as separate from verbs at some level of representation. Verbs can be 'gapped' 23 in Hopi, leaving the tense marker behind: (59)

a. ? k n you 'you b. ?im you 'you

warikg run run and warikg run run and

pii? ni-? ti-wat warik-ni then I also run-FUT I will run also' p«-? ni? tiwat-ni then I also-FUT I will also'

Hence there appears to be some justification in regarding tenses as-j-. Notice also that the word order convention specifies that if the language has common noun modifying adjectives (-N), these will appear to the left e

t - N

of the common noun. I am uncertain about the data on this point. Jeanne (1978) p. 316 remarks that 'the class commonly called "adjective" in other languages is not to be distinguished from the verbal part of speech'. However Whorf (1946) cites the examples in (60) as cases of adjectives. (60)

a. pe-he new b. qoca white

voyo knife voyo knife

There may be a dialect difference involved here. At any rate, at least for the data Whorf gives, the word order convention makes the correct prediction. For our purposes here, we will consider one more example. Hopi makes exuberant use of topicalization or, as Jeanne calls it, the pleonastic structure, as in the following examples. (See also the discussion in Hale, Jeanne, and Platero (1977)). (61)

mi? maana, pam pakkniya the girl she cry 'the girl, she is crying'

Relative clauses also exhibit this structure: 24

Michael Flynn (62)

165

a. ni? tiyo?yat (pam) pakmimiy-qa-t hoona I boy him cry-qa-OBLIQUE sent-honie 'I sent home the boy that is crying' b. ni? tiyo?yat ?ita-ni(pit) naawakna-qa-t I boy our mother him like-qa-OBLIQUE tiwi yta know 'I know the boy that our mother likes'

Possessive phrases, postpositional phrases, and verb phrases all have a pleonastic variant: (63)

Possessive: mit tiyo?yat p o ? k o ? a t mit tiyo? yat pit po? ko? at the boy his dog 'the boy's dog' PP : mit tiyo? yat ?amim mit tiyo?yat pit ?amim the boy him him-with 'the boy, with him' VP : n i ? mit tiyo?yat tiwi?yta n i ? mit tiyo?yat pit tiwi?yta I the boy him know 'I know the boy'

It seems that what's going on here is reminiscent of the 'derived VP rule' proposed in Partee (1976b). In this case, a phrase can have some sort of NP gap, optionally marked with a pronoun, and then add a full NP which controls this position. We want the pleonastic rule in Hopi to do something like this: if you have a phrase with a pronoun in it, the rule will semantically bind the pronoun with a lambda operator, and form an expression that is 'looking for' an NP to fill the created gap. Consider a simple case like (64). (64)

mi? tiyo?yat, Taqa pit tiwi?yta that boy Taqa him know 'Taqa knows that boy'

Suppose we derive (64) as follows. We first construct the sentence (65). (65)

[ t Taqa pit tiwi?yta] translation: know (PP { x 7 } ) (t*) (t* is the translation of Taqa)

166

A categorial theory of structure building

Then we make this sentence into a phrase which is looking for an NP to make a sentence. (66)

[t

Taqa pit tiwi-?yta]

NP

translation:

X (TiT { x 7 (know'

(PP { x 7 } ) (r*))}

Now the expression in (66) applies to an NP, which according to the word order convention, appears to the left, giving (67). (67)

mi? t+yo?yat Taqa pi-tt+wi?ta translation (assuming for convenience, but probably contrary to the fact, that mil translates like the in English): V y [ A x [ boy (x)

» x = y ] a (know' (PP { y } ) (r*)) ]

In other words, there is a unique boy such that Taqa knows him. The trick now is to write the rule so that it applies to several categories. We provisionally suggest (68). (68)

The pleonastic rule in Hopi If a is a phrase with the form [w...PRO...] and translates (.. .PP { x n } ...) W than a is a phrase of category-^,, where a translates as \ n \ x

n

(.-.pp { x n }

...)}

I don't believe I have seen a rule like (68) anywhere in the literature, as it may apply to any expression which has a pronoun in it. But it appears that this is the correct generalization for Hopi. At any rate, I think this rule gives the correct syntax (and semantics, as far as this can be determined at this point) for the pleonastic construction in Hopi. 25 Let us summarize what we have noted so far in this section. The Hopi word order convention is (56), repeated here. (56)

Word Order Convention for Hopi W t W For categories —, where Y = -X, — is to be interpreted as W/Y. W Otherwise,-rr is to be intrepreted as Y\W.

Michael Fly nn

167

The following generalizations are special cases of (56). (69)

a. complements precede verbs b. subjects precede the verb phrase (i.e. Hopi is SOV) c. determiners precede nouns d. Hopi is postpositional e. adjectives (if the language has them) precede the noun f. the pleonastic noun phrase appears to the left of the phrase it is associated with.

There is much more to be said about the syntax of Hopi in the categorial theory (see Flynn (1981a) for a more complete discussion), and I don't want to suggest that this analysis is problem-free. But our principal goal here has been to illustrate the potential of the framework. We have made some initial steps towards finding whatever universal principles may be stateable within the theory. We have proposed that English is a major category sensitive language and verb phrases apply to subjects, while Hopi is a pivot sensitive language and NP's apply to verb phrases to make sentences. Do these characteristics correlate in the world's languages? Are there any other 'sensitivities' that word order conventions may have? It would be premature to attempt to answer these questions conclusively, but at least I believe we have reached the point where they can be asked. In the next few paragraphs, we will briefly survey some other languages. The categorial theory makes available languages which, in a sense, have the mirror image of Hopi syntax, that is, languages with the word order convention in (70). (70)

W t W For categories —, where Y = - X , — is to be interpreted as Y\W. Y

e

Y

W Otherwise, — is to be interpreted as W/Y. Languages with the word order convention in (70) would have the properties in (71) among others.26 (71)

a. b. c. d. e; f.

VP + Subject TVP + Object Prep + NP CNP + ADJ CNP + Relative Clause CNP + DET

168

A categorial theory of structure building

As far as I know, there is only one language with all of these characteristics (Batak, cited in Keenan (1978), though this conclusion must be regarded as tentative). There are other which are close. One is Malagasy (also discussed in Keenan (1978)). Its properties are those in (72). (72)

a. b. c. d. e. f. e.

VOS DET + CNP Prep + NP Subordinate Conjunction + Subordinate Clause CNP + Relative Clause CNP + ADJ V + ADV

To see one way the theory can accommodate such a language, let us propose a word order convention for it. First, we introduce some terminW ology from Bar-Hillel (1953). A category — is endotypic if W = Y. Otherwise, it is exotypic. A word order convention that will account for all of the data in (72) is (73). (73)

Word Order Convention for Malagasy

W W

w

For categories— if — is exotypic,—is to be interpreted as W/Y. Otherwise, it is to be interpreted as Y\W. The low-level generalizations for Malagasy in (72) are rather similar to those for English. The differences are those in (74). (74)

English:

Subject initial simple adjective + CNP CNP + complex adjective Malagasy: Subject final CNP + ADJ

If one of the choices that languages are free to make is whether the subject is a function or an argument, then a language with the Malagasy word order convention but with subjects as functions would end up subject initial like English. This language would still, however, have all adjectives following the CNP. The syntax of adjectives in English is a problem for every other theory that I know of. AP's in English appear on both sides of the head CN as in (75).

Michael Flynn (75)

169

proud man man proud of his children * Proud of his children man * man proud 27

The generalization, as we noted in section 2, seems to be that AP's with major categories appear to the right of the CN, but those without appear to the left. It was in the light of facts like these that we proposed that English is sensitive to major categories and not to endotypicality. This example shows that a small difference in word order may lead to a radically different word order convention. To conclude this section I will outline a theory of word order conventions and show one way of having word order universals follow from more general principles. Of course, one of the principal sources of such generalizations is Greenberg (1963). To begin, let us consider a VSO language with the characteristics in (76). The number of the universal which specifies each characteristic is given in parentheses. (76)

a. b. c. d. e. f. g. h.

VSO Prep + NP (Universal 3) CN + Genitive NP (Universal 2) COMP + S (Universal 12) AUX + V (Universal 16) CN +ADJ (Universal 17) CN + Relative Clause DET + N

All of the properties in (76) will follow from the word order convention for Malagasy, with the addition of a 'wrap' convention along the lines we gave for English in section 3 that specifies that the subject ends up between the verb and the object. 28 Recall our analysis of Hopi. It is an SOV language and it obeys Greenberg's universals for such languages (in particular, universals 2, 4, 5 (vacuously), 13, 16, and 24). Most of Greenberg's universals for syntax are thus reduced to two word order conventions, and our problem now is to specify a learning theory for word order conventions from which it will follow that these conventions are selected and not others. At this time an attempt to state such a theory would be premature since so few languages have been studied from this point of view. But we can list some observations about the word order conventions so far adduced, with the suggestion that some of these observations may be regarded as preliminary constraints. 29

170

A categorial theory of structure building

(77)

a. Word order conventions may be sensitive to only one of the following: pivots, major categories, endo- or exotypicality. b. VP's take subjects into S's in all languages except some pivot sensitive ones. c. Only phrases assigned to functor phrases wrap, and wrap only with a phrase on their right. (In other words, there is no operatW ion that takes some phrase a of category —, applies it to some phrase [ y 0 7] with the result [ w 0 a 7].) d. Sensitivities are defined only on functor categories. For example, there is no word order convention that says 'if the phrase assigned to the argument category contains a major category...'

The constraints in (77) have little empirical content without a markedness theory of categories. Certainly the details of such a theory are very uncertain at this point. But to indicate a chain of inferences that such a theory would make possible, consider (78), regarded as a subcase of a universal theory of categories. (78)

a. All languages have the category -V e

~NP (i.e. the category of transitive verbs)30 b. All NP's are assigned to — . (V is in parentheses because only -(V) e

languages in which NP's take VP's to sentences will have it.) Suppose the language acquisition device is equipped with (77) and (78) and further suppose that the child adduces that the language to be learned is VSO. Here is what follows immediately, with no further evidence necessary: (79)

VP's apply to NP's to make sentences (i.e. NP =

with no word e

class marking) This follows directly from (77c). From (79), (80) is deducible. (80)

The language is not a major category sensitive language. (There is no way to write a word order convention meeting (77) to yield a major category sensitive VSO language.)

Michael

Flynn

171

Hence, (81)

T h e language is either a. e x o t y p i c c a t e g o r i e s l e f t w a r d W ( i . e . — e x o t y p i c = > W / Y ) , or b. pivot initial ( i . e .Y^ , Y = -e X = >

Y\W).

If ( 8 1 a ) is true, t h e n D E T + N . If ( 8 1 b ) , t h e n N + D E T . In this w a y t w o p i e c e s o f i n f o r m a t i o n ( V S O a n d t h e order o f d e t e r m i n e r s w i t h respect t o t h e n o u n ) are s u f f i c i e n t t o u n i q u e l y d e t e r m i n e a w o r d order c o n v e n t i o n . T h e p r e c e d i n g remarks are, o f c o u r s e , q u i t e s p e c u l a t i v e , b u t I h o p e t h e m e t h o d o f our e x p l a n a t i o n o f Greenberg's g e n e r a l i z a t i o n s is clear. T h e reason w h y , say, t h e r e are n o V S O p o s t p o s i t i o n a l languages is t h a t there is n o w o r d order c o n v e n t i o n w h i c h a l l o w s this c o m b i n a t i o n .

FOOTNOTES This chapter is a condensed version of portions of Flynn (1981a). I would like * to thank Barbara Hall Partee, Emmon Bach, Edwin Williams and the editors of this volume for their suggestions and encouragement. I would also like to express gratitude to my colleagues who attended my seminar at Reed College in the spring of 1980, during which they were subjected to an early version of the theory presented here. I am also grateful to the students in m y advanced seminar at Hampshire College in the fall of 1980, and to the students and faculty at the University of Groningen, where I gave a series of lectures on some of the ideas in this paper. Everyone was patient and perceptive. They, of course, cannot be held responsible for the errors that remain. 1. See Russell (1908). Some elaborations on the remarks in these introductory paragraphs can be found under the relevant entries in the Encyclopedia of Philosophy. 2. Dowty (1981), recalling the terminology introduced in Curry (1963), refers to this distinction as that between the tectogrammatics (i.e. what we might think of the dominance relations which hold in the language) and the phenogrammatics (i.e. left-right order). As we will propose below, Dowty suggests that the tectogrammatics are essentially universal. For the deployment of this idea within a phrase structure framework, see Gazdar and Pullum (1981). 3. Bar-Hillel, Gaifman and Shamir (1960) showed that categorial grammars and context-free phrase structure grammars are weakly equivalent in generative capacity. Categorial grammars have been reintroQ'iced as tools for linguistic description from time to time ( L y o n s (1966), Lewis (1970), Geach (1972)) but most of the proposals I am aware of do not attempt to exploit the notation to achieve explanation in syntax. The work of Bartsch and Vennemann (Bartsch and Vennemann (1972), Vennemann (1973, 1975)) appears to share a similar sort of intuition about the structure of languages that I will deploy here. However, the systems are quite different.

172

A categorial theory of structure building

A detailed comparison would take us too far afield, but see Koster (1975) for remarks about the Bartsch-Vennemann theory that do not apply to the one in this paper. More recently there have been several studies which use a categorial syntax to explain syntactic phenomena. See Steedman and Ades (1981), Contreras (1981) and van der Zee (to appear). 4. For an interesting recent modification of the theory of phrase structure, see Stowell (1981). Some of the ideas presented there are quite similar in spirit to the theory in these pages, but they are deployed in a substantially different framework. A thorough-going comparison of the two approaches is beyond the scope of this chapter. 5. The theory I will explicate here departs, in a number of places and in varying degrees, from common practice in Montague grammar. I will not pause to identify each innovation. For a discussion of the Montague framework, see Dowty, Wall, and Peters (1981). 6. Categories have direct and universal semantic import. I am assuming that each category is mapped in a uniform way onto a type in an interpreted logic, along the lines specified in PTQ. For further discussion, see Flynn (1981a). 7. Actually, Bach (1980b) follows Lapointe (1980) in regarding the tensed forms of verbs as given directly by the lexicon, eliminating the need for abstract items like PRES in the syntax. I believe that our f p m e w o r k is reformulable along these lines. Walks, then, would be in -777-, loves in — . There is no effect on the points made r NP NP NP here, though we will continue to assume items like PRES for the sake of discussion. 8. This is not quite right, but we will assume it here for the sake of exposition. We will introduce a modification in section 3 that will account for discontinuous constituents. 9. We will regard the category t to fall under this definition, though I do not know of any cases where it makes a crucial difference. 10. It is unclear whether this convention is to be thought of as a rule for the construction of phrases or as an output condition. For the present purposes, the distinction will play no role and readers may have it as they wish. I believe that the word order convention may also be formulable in terms of node admissibility conditions in the sense of Gazdar (1982). Thus we may interpret (16) as an instruction to admit a node W under the conditions specified in the convention. 11. The terminology here was suggested to me by what I think is a similarly revealing metaphor in the technical vocabulary of basketball and baseball. The notion should not be confused with that of Braine (1963). 12. We beg the question of what categories INF and S abbreviate. Their exact specification, though an interesting problem, is irrelevant to the point under discussion here. 13. It is possible to formulate the principles of sentence parsing proposed in Frazier (1978) in a rather natural way within the categorial framwork. Her late closure principle can be stated as in (i) and her minimal attachment principle as in (ii). (i)

If the parser encounters a word which is ambiguous with respect lexical category, it will select a category which is a possible functor. (ii) The parser checks the next item before making a category assignment. If the next item has a category assignment that allows phrasal packaging of already encountered items, that category assignment will be selected.

Michael

Flynn

173

These principles predict that the reading in (29) is the preferred reading to that in (30) just as Frazier's principles do. For details and further discussion, see Epstein (1980). 14. In fact, any restrictive modifier that does not iterate is a problem for the theory. Ewan Klein has suggested to me that manner adverbs may be such a case. 15. Here we see one potential problem with our analysis: 'bare' prepositions as in John walked in and the people here. If in and here are assigned to the prepositional phrase category —X, the word order convention predicts *John in walked and *the e here people. I am uncertain right now what to say about this. 16. I want to make explicit the very tentative status of this treatment of COMP in relative clauses. The relevant research on unbounded dependencies in a categorial framework is only beginning, and hence the compatibility of the theory with others such as that in Gazdar (1982) and Chomsky (1981) is unclear. See Steedman and Ades (1981) and note 25 below. 17. He suggests it has something to do with the trace left by the extraposed constituent. 18. Notice that if we regard transitive verbs as being just those verbs in—V we do NP not encounter the problems noted and discussed in Gazdar (1982). We put aside the category specifications of PRT, INF, ADJ, and Pred N. Phrases such as hammer flat may be regarded as basic expressions. For discussion of this point, see Dowty (1976). The important point here is that hammer flat has an internal structure like [t lt hammer] [ A D J flat] ] _e e_ NP NP ADJ regardless of whether or not this phrase is generated by a productive syntactic rule. 19. We have reformulated this rule slightly to make Bach's notation similar to our own. The point involved is not affected. Other writers who have appealed to a rule like RWRAP include Thomason (1976) and Dowty (1978). 20. Bach's treatment is ad hoc because there is no independent motivation for the failure of phrases like promise Sue to wrap like persuade to leave. That is to say, we need an explanation for why the category ^-V combines with its argument by simple concatenation, while —V calls the subtunetion^RWRAP. _e NP 21. Parts of the analysis and all of the data (except where noted) come trom Jeanne (1978). I would like to express my thanks to Ken Hale for bringing Jeanne's work to my attention. 22. The reader can easily verify tliat Hopi would also be predicted to be subject initial if tensed VP's were assigned to ^ p as » i did for English. 23. The term here is Jeanne's. It is unclear from the examples she cites whether the rule involved is gapping or some sort of VP anaphora. 24. The relative clause marker -qa has very interesting properties which we won't go into here. For discussion, see Jeanne (1978). Flynn (1981a) suggests a treatment in categorial grammar that makes those properties special cases of the Hopi word order convention proposed here.

174

A categoriaI theory of structure building

25. No doubt the reader will notice the provocative similarity between the Hopi pleonastic rule and other rules which set up unbounded dependencies. What to make of this is not clear yet. 26. We will assume that VP's take subjects into sentences unless otherwise noted. 27. There are some instances where adjectives not containing major categories may follow the noun, but these have a rather poetic feel: (i) a melody sweet the beer refreshing the lion dying (from Shakespeare R2 5.1.29) a dozen healthy infants well formed (from Watson's famous boast about behaviorism) Cases of resistance to the word order convention show up in several places in English. The 'transportability' of adverbs ( - V and—) may be related to this as well as object c t t -V e inversion in poetry (see Austin (1977) for discussion). (ii) the lonely man's despair hunger overcame (Keats, 'Adonais') When I a fat and bean-fed horse bequile (Shakespeare, MND 2.1.45) 28. One interesting question that we will not consider here is exactly how this wrap convention is to be stated and why VSO languages are much more common than VOS languages like Malagasy. 29. Notice that it is possible that a word order convention for a language may not have an 'otherwise case'. For example, suppose we had the convention in (i). (i)

W t W For categories —, if Y = - X , then — is to be interpreted as Y\W. Y e Y

This would give us a language where all restrictive modifiers and determiners (in that order) follow the head. But the distribution of NP's would be free, since (i) does not apply to categories which take NP arguments. Makua might be such a language. (See Stucky's chapter in this volume for discussion.) There are several ways to treat languages with free or partially free order in this framework, but the pertinent research has not been attempted yet. 30. Given other assumptions that we have made, this is equivalent to the claim that all languages have a VP.

Chapter 8

Resolution rules: agreement in person, number, and gender Greville Corbett

0.

INTRODUCTION*

When noun phrases are conjoined, they may carry feature combinations which create a problem for the agreement rules as, for example, when a verb agrees with coordinated noun phrases which differ in gender. The rules which determine the form to be used are termed 'resolution rules' (Givon (1970, 1972); Vanek (1970) pp. 45-6 calls them 'feature computation rules'). The features which may require resolution are person, number and gender ( § 1). Person and number show little variation in their resolved forms; differences between languages are found mainly in the conditions under which these rules apply ( § 2). Unlike person and number resolution, gender resolution shows great diversity: some languages have rules which are basically syntactic, others rely on a semantic principle and yet others show interesting combinations of the two principles ( § 3 ) . The difference between person and number resolution on the one hand, and gender resolution on the other, stems from their differing degrees of semantic justification. As a consequence of the considerable differences in gender systems between languages, gender resolution rules are language-specific. They are, however, determined by common semantic and functional considerations (§4). The degree to which these common requirements can be met depends both on the gender structure and on the morphological possibilities of a given language.

1. FEATURES REQUIRING RESOLUTION

The features which may require resolution are person, number and gender. These will be discussed in turn. 1.1.

Person resolution

The general principle of person resolution, stated in innumerable grammars, is that the first person takes precedence over the second, and the second

176

Resolution rules: agreement in person, number, and gender

over the third. Consider the following examples from Czech, a West Slavonic language (Travnicek (1949) p. 433, Bauemoppel et al. (1968) p. 164): (1)

ja a ty zustaneme (1st pi) doma I and you will-stay at-home 'you and I will stay at home'

In (1), one of the conjuncts is first person and this takes precedence over the second person. In (2) it takes precedence over the third: (2)

bratr a ja se ucime (1st pi) hrat na klavir brother and I learn to-play on piano 'my brother and I are learning to play the piano'

In (3) there is no first person conjunct and so the presence of a second person determines the agreement form: (3)

tvuj otec a ty jste (2nd pi) si podobni your father and you are to-each-other similar 'your father and you are similar'

The resolution rules may be stated as follows: 1. if the conjuncts include a first person, first person agreement forms will be used; 2. if the conjuncts include a second person, second person agreement forms will be used. (The default condition is that third person agreement forms are used.) In Czech, these rules are ordered, the second applying only when the first fails to apply. In languages which have an inclusive/exclusive distinction in agreement forms, both rules can apply. Thus in Warlpiri, aPama-Nyungan language of Central Australia, if the conjuncts include a first person and a second person, then the first person inclusive form is used (Hale (1973) p. 319): (4)

njuntu manu qatju ka-li (present 1st dual inclusive) pula-mi you and 1 shout 'you and I are shouting'

It has been suggested that rules equivalent to those given above are universal. The suggestion appears well-founded, not only because such rules are reported frequently, but also because they match the hierarchy of reference

Greville Corbett

177

which constrains pronominal systems (see Zwicky (1977) pp. 718, 725). First person pronouns can be used to refer to 'speaker plus listener' or 'speaker plus other person'. These meanings are matched by the resolution rule which determines that a first person conjoined with a second or third person is resolved as first person. Similarly second person pronouns can be used on their own to indicate 'listener plus other person'; this is reflected in the rule which resolves second and third persons conjoined into the second person. Thus the person resolution rules have a clear semantic basis. (While it may be possible to maintain that person resolution rules always take the form given above, we shall see that they may be optional (§2.1).)

1.2. Number resolution The typical number resolution rule can be stated simply: conjoined elements require a plural. The rules are a little more complicated in the case of languages with more than two grammatical numbers. Slovene, a South Slavonic language, has singular, dual and plural. If two singulars are conjoined, then the verb stands in the dual (examples from Lencek (1972)): 1 (5)

Toncek in Igor sta (dual) prizadevna (dual) Toncek and Igor are assiduous

However, if there are more than two nouns, as in (6), or if one of the nouns is in the dual (7) or plural (8), then a plural predicate results: (6)

Toncek, Igor in Marina so prizadevni (pi) Toncek, Igor and Marina are assiduous

(7)

Marta in njegova brata (dual) bodo prisli (pi) Marta and his (Igor's) brothers will come

(8)

Igor in njegove sestre (pi) bodo prisli (pi) Igor and his sisters will come

The number resolution rules are as follows: 1. if there are two conjuncts only, both of which are in the singular, then dual agreement forms will be used; 2. in all other cases, providing there is at least one non-plural conjunct, plural agreement forms will be used.

178

Resolution rules: agreement in person, number, and gender

Of course, for languages with no dual category the first rule is not required. At first sight the restriction on the second rule appears superfluous; why should not instances where all the conjuncts are plural be covered by this rule? There is no need for a resolution rule in such instances and, as we shall see below, in some languages it is important to ensure that no resolution rule operates in such cases (§3.2). The second complication with number resolution is that it frequently does not apply. We discuss this problem in §2. When it does apply, as was the case with person agreement, it produces forms which are semantically justifiable. 1.3.

Gender resolution

While person and number resolution rules are widespread, there are many languages which do not require gender resolution rules. German and Russian both have three genders, but there are agreeing forms for these only in the singular. There is only one plural form, which serves for all three genders. The other major difference between person and number resolution rules on the one hand and gender resolution rules on the other is that the former produce forms which are semantically justifiable, while the latter often do not. Take the case of a language with two genders, masculine and feminine, in which inanimates are distributed between the two genders. If two inanimates are conjoined, one masculine and one feminine, neither resolution will be semantically justified. As might be expected, therefore, gender resolution rules show great diversity. We shall describe the possibilities in §3 below; first it is essential to distinguish genuine cases of resolution from instances where the problem is simply avoided (because agreement is with one conjunct only).

2. THE APPLICATION OF RESOLUTION RULES

Even in languages where conjunction is generally acceptable, agreement with conjoined structures may be barred under specific conditions. In Luganda, agreement with conjoined structures is possible providing all the conjuncts are human or none are human. Conjoining human and nonhuman nouns produces unnatural forms; the comitative construction is preferred (as discussed in §3.1 below; for a comparable problem involving agreement in Chippewa seeHarries-Delisle(1978)p.556).InXhosa, a Bantu language like Luganda, agreement with nouns of different genders is not acceptable, unless the agreement morphemes required by the different nouns are identical (Voeltz (1971)). In languages which have no such constraints on agreement with conjoined noun phrases, the application of the resolution rules may still not

Greville Corbett

179

be automatic. Instead, it is usually possible for agreement to occur with one conjunct only, thus avoiding the resolution rules. This raises the question of which conjunct will control the agreement (§2.1). We must also investigate the factors which favour resolution as opposed to agreement with one conjunct only (§2.2), and consider possible exceptions (§2.3). 2.1.

Agreement with one conjunct

When the resolution rules do not operate there is normally full agreement with one of the conjuncts. In Czech, person and number resolution regularly do not apply, providing the predicate precedes the subject (Travnicek (1949) p. 433): (9)

pujdu ( l s t s g ) tam ja a ty will-go there I and you 'you and I will go there'

In this example the verb agrees fully with the nearer conjunct. This situation may be represented schematically as follows:

(10) f TARGET

I NP

+

NP

Agreement with the nearer conjunct may also occur, though this is less usual, when the subject precedes the verb, as in the following Latin example (Gildersleeve & Lodge (1948) p. 184): (11)

et ego et Cicero meus flagitabit (3rd sg) and I and Cicero my will-demand 'both my Cicero and I will demand it'

This example may be represented as (12): (12) NP

+

I NP

I TARGET

Both these examples show the failure of person and number resolution to operate, and both show full agreement with the nearer conjunct. Similar examples could be given in which number and gender resolution do not operate (for instance, see Brauner (1979) p. 424 for Swahili examples).

180

Resolution rules: agreement in person, number, and gender

However, while agreement with the nearest conjunct is most frequently found in such cases, it is not the only possibility. Agreement may also be with the first conjunct which, when the subject precedes the verb, is not the nearest; this type of agreement can be found in Slovene (Toporisic (1972) p. 187): (13)

groza (fem sg) in strah (masc sg) je prevzela (fem sg) horror and fear has seized vso vas the-whole village

Here neither gender nor number resolution has operated (the resolved form would be the masculine plural) and the gender of the predicate indicates clearly that agreement is with the first conjunct, as shown in (14):

i NP

+

NP

t TARGET

Note, however, that agreement with the nearer conjunct, as in (10), is much more common. Serbo-Croat, a South Slavonic language like Slovene, also shows the possibility of agreement with a more distant conjunct: (15)

Ona stalna duboko urezana svijetla (neut pi) Those constant, deeply cut lights i sjene (fem pi) koje je naslikao and shades which has painted umjetnikov kist bila (neut pi) su jaca (neut pi) the artist's brush were stronger od realne svijetlosti. (Desnica, quoted by Megaard (1976) p. 80) than real light 'Those constant, deeply cut lights and shades which the artist's brush painted were stronger than real light.'

Here the resolved form to mark agreement with both conj uncts would be the masculine plural, as is illustrated in example (38) below. Similar examples are found in Latin (Kühner & Stegmann (1955) p. 53); while the examples quoted above involve predicate agreement, it is interesting to note that agreement with the more distant conjunct has also been found in agreement of the attribute and of the relative pronoun in Latin (Kühner & Stegmann (1955) pp. 55, 58-9). We conclude that, when the resolution rules do not apply, agreement is normally with the nearest conjunct, but that this is not the only possibility.

181

Greville Corbett 2.2. Factors favouring resolution

The factors which make resolution more likely to operate are of two types: those which involve the agreement controller (the element which governs agreement) and those which concern the agreement target (the element which marks agreement). Controllers which refer to animates, and controllers which precede their targets, are more likely to take resolved agreement forms (Moravcsik ( 1 9 7 8 ) pp. 341-2, Corbett ( 1 9 7 9 ) pp. 218-9). This can be illustrated from number resolution in the predicate. Data on Spanish (13th-15th centuries) have been derived from England (1976) pp. 813-20; statistics on German are calculated from figures in Findreng(1976)pp. 145, 165-6, 197; the Russian data are taken from modern literary texts (19301979; for details see Corbett forthcoming: Chapter 7). In each category we give the total number o f examples and the percentage in which number resolution was found. For example, in Medieval Spanish there were 288 examples of conjoined noun phrases which denoted animates and which preceded the predicate; of these 96% had a plural predicate (thus number resolution occurred in 96% of the cases). TABLE 1 inanimate

animate

/c

70

N Medieval Spanish subject-predicate

predicate-subject

plural

N

plural

288

96

243

31

German

1095

96

1702

67

Russian

115

100

67

85

Medieval Spanish

318

69

239

6

German

379

93

925

40

Russian

89

84

114

28

It is evident that both factors favour resolution. When both are present, all three languages give overwhelming preference to the resolved form. When either one is present, the resolved form is found in a significantly higher proportion of the cases than when neither is present. In Medieval Spanish and German the animacy of the subject exerts a stronger influence than its position, while in Russian the two factors are of about equal weight. (In Spanish and German there is also evidence showing that concrete subjects have plural predicates more often than abstract subjects do.) So much for controller factors; let us now consider the target. Resolution as opposed to non-resolution is a particular case of semantic versus syntactic

182

Resolution rules: agreement in person, number, and gender

agreement. It is therefore subject to the agreement hierarchy (Corbett (1979)). The agreement hierarchy consists of the following syntactic positions: attributive, predicate, relative pronoun and personal pronoun. As we move rightwards along this hierarchy, the likelihood of resolved (semantic) agreement increases monotonically. Data on number resolution in Russian are given in the following Table (Corbett (forthcoming) Chapter 9). TABLE 2 attributive N % plural

predicate N % plural

relative pronoun % plural N

34

230

10

12

70

100

personal pronoun % plural N 26

100

It can be seen that resolved forms show a monotonic increase. The fit is actually better than figures from this corpus indicate, for singular relative pronouns occur, if infrequently, and even a singular personal pronoun is possible, though exceptionally rare. It can be shown that resolution is also constrained by Comrie's predicate hierarchy (Comrie (1975), see Corbett (forthcoming) Chapter 9, for discussion).

2.3. Selective application of the resolution rules When conjoined noun phrases show features which could trigger more than one of the types of resolution rules (e.g. person and number), then the normal choice is either to apply all the appopriate resolution rules, or to apply no resolution rules and to do the agreements with one conjunct only. However, exceptions have been found in German and French. In his corpus of 20,000 pages of Modern German, Findreng found seven examples of second and third persons conjoined (all singular). Of these, four have the verb in the second person plural (as expected), while three have a third person plural form, as in the following example (Findreng (1976) p. 83): (16)

...wenn du und deine when you and your mehr bekommen werden more get will 'when you and your sister

Schwester eine ttichtige Portion sister a good portion (3rd p i ) , . . . will get a good portion more'

Greville Corbett

183

(Informant tests also revealed variation, Findreng(1976) p. 385.) In example (16) number resolution has applied (the verb is plural) but person resolution has not and the default form, the third person, is found. A possible explanation is that the ending -en (1st and 3rd pi) is a clear marker of plurality, while the ending -(e)t (2nd pi) is in some instances indistinguishable from the third person singular. In French too, examples occur of third plural verbs with subjects containing a non-third person conjunct. The following example is quoted by Grevisse (1969) p. 762: (17)

comme l'ont (3rd pi) montré Ferdinand Brunot et as it have shown Ferdinand Brunot and moi-même myself 'as Ferdinand Brunot and myself have shown'

Here the probable explanation is that in conjoined structures the normal clitic pronouns je and tu are replaced by the stressed forms moi, toi (with même in some instances); these do not regularly control verb agreement directly and so are not unambiguously marked for person as are the normal subject pronouns. As a working hypothesis we will adopt the position that the resolution rules apply or fail to apply as a set, and that exceptions to this generalization require specific explanations. However, the fact that the exceptions were found in French and German, both of which have long grammatical traditions, may indicate that exceptions are not so infrequent, but have not been picked up in languages which have yet to be subjected to such intensive study. Note that all the exceptions involve person resolution; no examples have been found of number or gender resolution failing to apply when another resolution rule has operated. In this section we have investigated the application of the resolution rules - whether they apply, the factors influencing their application and the possibility of the operation of one resolution rule independently of the others. For the rest of the paper we shall focus on those instances where the resolution rules do operate (and as a set), in order to establish what the resolved forms are and what rules are required to specify them. We have seen that in the case of person and number these are relatively straightforward. With gender resolution the situation is much more complex.

184

Resolution rules: agreement in person, number, and gender

3. GENDER RESOLUTION SYSTEMS

Gender resolution may follow two distinct principles: the semantic principle or the syntactic principle. Gender resolution by the semantic principle involves reference to the meaning of the conjoined elements and disregard for their syntactic gender. The syntactic principle refers to the syntactic gender of the conjoined items, irrespective of whether the gender is semantically justified. (Discussion of a case of gender resolution according to a morphological or phonological principle is deferred to §4.3.) in this section we shall first give examples of clear cases of semantic gender resolution (§3.1) and of syntactic gender resolution (§3.2) and then consider mixed types (§3.3). 3.1.

Semantic gender resolution

Clear examples of this type can be found in Bantu languages. Most have at least nine grammatical genders (in paired singular/plural classes). These partly correspond to semantic classifications: nouns of the 1/2 gender are human, but not all nouns referring to humans belong to the 1/2 gender (Givon (1970) pp. 250-1, (1971)). For gender resolution, the important thing is whether a noun refers to a human or to a noun-human, irrespective of the gender class. This point is illustrated in data from Luganda, presented by Givon (1970) pp. 253-4,(1971) pp. 38-9: (18)

omu-kazi (1/2), es-sajja (5/6) ne olu-ana (11/10) the-woman, the-fat-man and the-thin-child ba-a-labwa (1/2) were-seen

The resolved form for human conjoined nouns is the 1/2 form. In (18) only one of the conjuncts belongs to that class. In (19) none of the conjuncts belongs to the 1/2 class, but as all refer to humans the resolved form is again the 1 ¡2 marker: (19)

ek-kazi (5/6) aka-ana (12/14) the-fat-woman the-small-child ba-a-labwa (1/2) were-seen

ne olu-sajja (11/10) and the-tall-man

Example (19) proves that the use of the 1/2 form as the resolved form is motivated by semantic considerations. If none of the conjuncts refers to a human, then the 7/8 form is used, as in (20):

Greville Corbett (20)

en-te (9/10), the-cow, ely-ato (5/6) the-canoe

185 omu-su (3/4), eki-be (7/8) ne the-wild-cat, the-jackal and bi-a-labwa (7/8) were-seen

As was mentioned above ( § 2 ) , conjoining human and non-human forms produces an unnatural result: (21)

?omu-sajja (1/2) ne em-bwa-ye (9/10) bi-a-gwa (7/8) the-man and dog his fell 'the man and his dog fell down'

The result is unnatural providing the 7/8 (non-human) form is used; if the 1/2 form is used, an unacceptable sentence results: (22)

*omu-sajja (1/2) ne em-bwa-ye (9/10) ba-a-gwa (1/2) fell the-man and dog his

The preferred alternative is the comitative construction: (23)

omu-sajja (1/2) y-a-gwa (1/2) ne em-bwa-ye (9/10) the man fell and dog his 'the man fell down as did his dog'

Example (23) has a simple subject, with which the verb can agree fully (in the singular) and the problem of resolution is avoided. The resolution rules can be stated as follows: 1. if all the conjuncts are semantically human, then the 1/2 form is used; 2. if one or more of the conjuncts, but not all, are semantically human, then the comitative construction is preferable; 3. otherwise the 7/8 form is used. The rules as stated allow for the 7/8 form to be used for mixed conjuncts if Rule 2 is ignored. The same rules account for the Bemba data given by Givon (1972) p. 82: (24)

im-fumu (9/10) na i-shilu (5/6) ba-aliile (1/2) the-chief and the-lunatic left

(25)

ici-tabo (7/8), ubu-sanshi (14/6) na ulu-balala (11/10) the-book the-bed and the-peanut fi-li (7/8) kuno are here

186

Resolution rules: agreement in person, number, and gender

The Bantu examples show that resolution may operate according to the meaning of the conjuncts. We now turn to examples where semantic considerations are apparently irrelevant. 3.2.

Syntactic gender resolution

In French there are only two genders; if conjoined nouns are of the same gender then that gender will be used (examples from Grevisse (1969) p. 314): (26)

un livre (masc) et un cahier (masc) neufs (masc pl) a book and an exercise-book new 'a new book and exercise book'

(27)

la misère (fem) et la ruine (fem) générales (fem pl) the poverty and the ruin general 'the general poverty and ruin'

When the conjuncts include masculine and feminine nouns, then a masculine form is used (the stylists insist that it should be placed next to the masculine noun, but this requirement is not rigorously observed, as our examples show): (28)

un père (masc) et une mère (fem) excellents (masc pl) a father and a mother excellent 'an excellent father and mother'

(29)

un savoir (masc) et une adresse (fem) merveilleux (masc pl) a knowledge and a skill marvellous 'a marvellous knowledge and skill'

Here the rules apply with the same effect to animate and inanimate nouns (though the relative frequency may differ as discussed in §2.3). The rules can be stated in two different ways: A. 1. if at least one conjunct is masculine (syntactically), the masculine form is used; 2. otherwise the feminine is used. Alternatively: B. 1. if all conjuncts are feminine (syntactically), the feminine form is used; 2. otherwise the masculine is used.

Greville Corbe tt

187

For French either set of rules is adequate; indeed, the two formulations are logically equivalent just in case there are exactly two genders. However, we shall see that some languages require rules of Type A, in which one conjunct of a particular gender is sufficient to determine the agreement form, while others use Type B, in which homogeneous controllers are distinguished. An example of a language for which one rule type is clearly preferable (Type B in fact) is Slovene, which has three genders and three numbers. The predicate agreement forms are given in Table 3 (bil is the past active participle of the verb 'to be'): TABLE 3 masculine singular

bil

dual

bila

plural

bili

feminine

neuter

bila

bilo b li

bile

bila

The dual number can result from the operation of the resolution rules only if two singular nouns are conjoined (§1.2), as in the following sentences (from Lencek 1972): (30)

Toncek (masc) in Marina (fem) sta prizadevna (masc dual) Toncek and Marina are assiduous

(31)

Toncek (masc) in to dekletce (neut) sta Toncek and that little-girl are prizadevna (masc dual) assiduous

A masculine noun conjoined with a feminine, as in (30), or a neuter, as in (31) requires a masculine dual predicate. However when a feminine and a neuter are conjoined, the masculine dual is still found: (32)

ta streha (fem) in gnezdo (neut) na njej mi that roof and nest on it to-me bosta ostala (masc dual) v spominu will remain in memory 'that roof and the nest on it will remain in my memory'

188

Resolution rules: agreement in person, number, and gender

Similarly, two neuter singular nouns take a masculine dual: (33)

to that bosta will

drevo (neut) in gnezdo (neut) na njem mi tree and nest on it to-me ostala (masc dual) v spominu remain in memory

The only way in which the feminine/neuter dual form can result from the resolution rules is if two feminine nouns are conjoined: (34)

Marina (fem) in Marta (fem) sta prizadevni (fern dual) Marina and Marta are assiduous

Clearly the most economical way to write the gender resolution rules is to use the Type B formulation: 1. if all conjuncts are feminine, then the feminine form is used; 2. otherwise the masculine is used. The number resolution rules determine when the dual and when the plural form are to be used. As this is so, the rules just given will also account for gender resolution in the plural in Slovene. Thus in (35), all the conjuncts are neuter, but the masculine plural form is required: (35)

to that njem it

okno (neut), window, mi bodo to-me will

drevo (neut) in tree and ostali (masc pi) v remain in

gnezdo (neut) v nest in spominu memory

Again, the feminine is possible only if all the conjuncts are feminine: (36)

Marina (fem), Marta (fem) in Marjanca (fem) so Marina, Marta and Marjanca are prizadevne (fem pi) assiduous

Note that in the rules given there is no recourse to semantic factors — the syntactic gender is the determining factor. The fact that the rules given for gender resolution apply equally well for the dual and the plural suggests an interesting paradox. On the one hand, the resolution rules are independent of each other. Thus we have no rules which refer, say, to feminine plurals or neuter singulars. On the other hand they are interrelated in that if one type of resolution rule operates

Greville Corbett

189

then all must operate where possible (cf. §2.3). Given a subject consisting of a feminine singular and a neuter singular noun, it is not possible to apply gender resolution (to give a masculine) but at the same time to fail to apply number resolution, and so to have a masculine singular predicate. This interrelation of the resolution rules helps explain the particularly interesting situation seen in (33) and (35), where gender resolution has applied, giving a masculine predicate, even though all the nouns are of the same (neuter) gender. 2 Here number resolution is triggered by the presence of singular conjuncts; if one resolution rule operates then all must operate where possible; gender resolution does not include the possibility of assigning neuter plural endings in Slovene (we discuss why this should be so in §3.4 below), but specifies the masculine. Similar gender resolution rules are found in Serbo-Croat (though there the position is somewhat simpler as Serbo-Croat has lost the dual; at the same time, there is an added complication in that the first rule allows interesting leaks which we discuss in §4.3 below). Consider now the situation when the subject consists of neuter plurals only, as in this SerboCroat example: (37)

...ta secanja (neut pi) i razmatranja (neut pi) those memories and reflections sve su vise ustupala (neut pi) mesto novim ever have more yielded place to-new utiscima ... (Andric, Travnicka Hronika) impressions ... 'those memories and reflections increasingly gave way to new impressions'

In this sentence the resolved form (the masculine plural ustupali) would be ungrammatical. We can claim that number resolution does not apply in this case (this is the reason for the restriction on number resolution in §1.2) and so it does not trigger gender resolution. However, it would be incorrect to claim that gender resolution can be triggered only by number resolution as the following example shows: (38)

Sve njegove molbe (fem pi) i uveravanja (neut pi) All his prayers and assurances nisu pomagali (masc pi) nista. (Andric, Travnicka Hronika) did-not help notaing 'All his prayers and assurances did not help at all.'

Here we find feminine and neuter conjuncts and a masculine predicate, as required by the gender resolution rules. The correct generalization appears

190

Resolution rules: agreement in person, number, and gender

to be as follows. Gender resolution in Serbo-Croat can be triggered in two ways: either by the operation of number resolution (if one resolution rule operates, all must operate if possible), or by the presence of different genders in the conjuncts. Sentence (37) does not meet either condition, and so gender resolution cannot operate. In this sentence the neuter plural results from agreement with the nearer conjunct. The languages we have just discussed discriminate against the neuter plural, as it can arise only by agreement with the nearest conjunct and not from the operation of the resolution rules. Icelandic, however, favours it: the neuter plural is used for any mixture of genders (example (39) is from Jonsson (1927) p. 14, and (40) from Einarsson (1949) p. 133): (39)

t>au (neut pi) (drengurinn (masc) og telpan (fem)) They (the-boy and the-girl) eru freytt (neut pi) are tired

(40)

eg sa a (fem) og lamb (neut), boedi (neut pi) I saw a-ewe and a-lamb both svort (neut pi) black

The resolution rules required for Icelandic are as follows: 1. if all conjuncts are masculine, the masculine is used; 2. if all conjuncts are feminine, the feminine is used; 3. otherwise the neuter is used. In this section we have considered languages where the principle behind gender resolution appears to be purely syntactic; we shall see in §4 that, while in cases such as these the mechanism is syntactic, the motivation is semantic. Let us first consider cases where the syntactic and semantic principles are mixed. 3.3. Mixed semantic/syntactic gender resolution In this section we analyze two languages in which the semantic and the syntactic principles of gender resolution occur together. The first is Polish, a West Slavonic language; the possibilities for predicate agreement are given in the Table. Byt is the past tense of the verb bye 'to be'.

Grevìlle Corbett

191

TABLE 4 masculine personal singular plural

byl byli

feminine

neuter

byla

byto

non-personal

byly

Polish has three forms for gender agreement in the singular; in the plural there is a division into masculine personal and the remainder. The masculine personal category comprises nouns which are masculine and which refer to humans: it does not coincide completely with the semantic class of male human but it does so much more closely than do the gender classes in the singular. When nouns are conjoined and none is masculine personal, then the non-masculine personal/feminine/neuter form is used (Kulak et al. (1966) p. 249): (41)

siostry (fem) i matka (fem) czyta+y (non-masc pers) sisters and mother were-reading

If a masculine personal noun is present then the masculine personal form is used: (42)

brat (masc pers), siostry (fem) i matka (fem) brother, sisters and mother czytali (masc pers) were-reading

The resolution rules required appear to be as follows: 1. if at least one conjunct is masculine personal, then the masculine personal form is used; 2. otherwise the non-masculine form is used. These rules are of the form labelled Type A in §3.2: the first rule picks out conjoined structures which include one conjunct of a particular type (and therefore 'mixtures' will be included); in Type B rules, homogeneous structures are isolated. Unlike the French situation, we cannot rewrite these rules in the other form (1. if all nouns are non-masculine personal. . .) because there is no other motivation for labelling nouns as non-masculine personal in Polish. Rules like those given above can be found in numerous sources; they also operate in other West Slavonic languages (Corbett

192

Resolution rules: agreement in person, number, and gender

(1982)). However, Polish shows interesting exceptions. Consider the following example, given by Doroszewski (1962) p. 237: (43)

Hania (fem) i Reks (masc) bawili (masc pers) si? Hania and Rex played piik^ with-a-ball

Reks is a dog (masculine but not personal). There is no masculine personal conjunct in (43), but the predicate is masculine personal. The status of sentences like (43) has been the subject of considerable debate (outlined in Corbett (1982)). The most informative study of the question to date, that of Zieniukowa (1979), describes responses to a questionnaire by 31 pupils in liceum classes II and III (in their upper teens). For a sentence comparable to (43), only two informants used the non-masculine personal form (and one used a different construction). The masculine personal form, as in (43), is obviously the usual form. It cannot result simply from the presence of the human noun Hania, because in (41) both conjuncts referred to humans but a non-masculine personal form was used. We must check whether the presence of a masculine animate is sufficient: in the following example both conjuncts are masculine animate: (44)

pies (masc) i kot (masc) jedli (masc pers) the-dog and the-cat were-eating na podworzu in the-yard

The masculine personal form (as in the example) was the majority choice, but seven informants chose the non-masculine personal form. Thus masculine animates are less likely to produce a masculine personal form than masculine animate plus feminine human. Feminine human conjoined with masculine inanimate can also result in a masculine personal form: (45)

Mama (fem), coreczka (fem) i wozek (masc) Mother, daughter and the-pram ukazali (masc pers) si? nagle appeared suddenly

In this example informants were equally divided between the masculine personal and the non-masculine personal: ukazaiy (one informant chose neither). The rules required to cover these sentences (and other types described in Corbett (1982)) are as follows:

Greville Corbett

193

1 • if the subject includes a masculine personal conjunct, the predicate will be in the masculine personal form; 2. (optional) if the subject includes the features masculine and personal, whether these are syntactic or semantic, the predicate may be in the masculine personal form; 3. (optional) if the subject includes a masculine animate conjunct, the predicate may be in the masculine personal form; 4. otherwise the predicate will be in the non-masculine personal form. The first rule, which accounts for the form used in (42), requires no further comment. The optional Rules 2 and 3 both represent plausible weakenings of Rule 1: in Rule 2 the conditions apply to the subject as a whole rather than to a single conjunct and, more surprisingly, they allow semantic or syntactic features or a combination of these. Rule 3, on the other hand, retains the restriction to a single conjunct but reduces the requirement from personal to animate. Rule 2 accounts for the form in sentence (45) while Rule 3 has operated in (44). It is significant that when both Rule 2 and Rule 3 can apply, as in (43), then for Zieniukowa's informants the masculine personal form is almost obligatory. When none of these Rules apply, the non-masculine personal form is assigned by Rule 4, as in sentence (41). The rules refer both to syntactic gender and to semantic features. Thus Polish stands between the clearly semantic gender resolution found in Bantu and the syntactic type documented in §3.2. Latin shows a mixture of syntactic and semantic criteria of a different type. Always providing resolution occurs (in many instances it does not, see Lebreton (1901) p. 2, Kühner & Stegmann (1955) p. 44), conjuncts of the same syntactic gender take agreeing forms of that gender. However, when conjuncts are of different genders, then the resolved form to be used depends on whether the nouns refer to persons or not. For persons the masculine is used: (46)

quam pridem pater (masc) mihi et mater (fem) how long ago father to me and mother mortui (masc pi) essent dead were 'how long ago my father and mother had died7

For other conjoined elements the neuter is used:

194

Resolution rules: agreement in person, number, and gender

(47)

murus (masc) et porta (fem) de caelo tacta (neut pi) wall and gate from sky struck erant (pi) were 'the wall and the gate have been struck by lightning'

(These examples are from Kühner & Stegmann (1955) pp. 44, 52. Kühner & Stegmann state that when humans and non-humans are conjoined agreement is usually with the nearer noun, but resolution to the neuter plural is possible (1955) pp. 49, 51.) The resolution rules are as follows: 1. 2. 3. 4.

if all conjuncts are masculine, then the masculine is used; if all conjuncts are feminine, then the feminine is used; if all conjuncts refer to humans, then the masculine is used; otherwise the neuter is used.

These rules are ordered; there is no need, therefore, to stipulate that the conjuncts in Rule 3 are of mixed gender. Similarly, Rule 4 will automatically cover cases of mixed gender and those where all the conjuncts are neuterThus Latin has two resolution rules based on the syntactic principle and one on the semantic principle.

4. STRATEGIES FOR GENDER RESOLUTION

In the last section we observed that gender resolution rules may either specify that at least one conjunct be of a particular sort (Type A, as also found in person resolution), or that all the conjuncts be of a particular sort (Type B). We found no examples of the logically possible type which would refer to the majority of the conjuncts being of a certain sort. One reason is that conjunction most often involves just two conjuncts; for example, Findreng (1976) p. 196 gives separate figures for conjunction of two abstract nouns or more than two: 87% of the cases (total 2277) involved conjunction of two elements only. While Type A rules were postulated for Polish and Type B for Slovene, the effect was the same: in both languages the masculine or masculine personal form is used as the dominant resolution form. How are we to explain why different languages favour particular forms in their resolution rules? 4.1.

Markedness: an inadequate

motivation

One possibility is to appeal to markedness. There are problems in applying markedness to areas other than phonology (Bierwisch (1967) pp. 267-70),

Greville Corbett

195

but Schane (1970) attempts to extend markedness into French morphology and syntax, including gender resolution. His analysis appears attractive, but there are difficulties when we try to extend it to languages with more than two genders (though Lencek (1972) bases his explanation on markedness). The phenomena discussed by Schane are taken from morphology, syntax and semantics; in each case the unmarked form is claimed to correspond to the form used for gender resolution. We will attempt to relate these claims to the other languages we have investigated. The morphological argument runs as follows: the masculine singular form of the adjective in French consists of the stem only: the feminine singular is the stem plus feminine marker / o/ ; for the masculine plural the plural marker /z/ is added, while for the feminine plural /a/ and /z/ are both added (various deletion rules then apply). Thus the masculine plural form is unmarked for gender. In cases where gender resolution is required, the neutralized form is used, and this is equivalent to the morphologically unmarked form (1970) p. 291. Let us consider Slovene in these terms (Table 3). Here the masculine singular is signalled by the stem only, but the plurals cannot be derived from the singulars by the addition of a plural marker. If we try to substitute a more general claim, that the least complex form will be used (cf. Schane (1970) p. 292) then Polish is a counter-example, the masculine personal form, used for resolution, frequently requires a mutation of consonant, which is not found in the other forms. Schane also considers instances which are syntactic in nature. Impersonal pronouns, which are neither masculine nor feminine, take masculine agreements. It may be added that elements such as infinitives, which do not bear a gender spécification, take masculine agreements in French. However in three-gender systems, like that of Slovene, such items are assigned to the neuter gender (unless there is a special neutral form, Corbett (1980)). Schane includes discussion of interrogative and indefinite pronouns, which require masculine agreements in French. In a three-gender system like that of Slovene, these also require masculine agreement forms when they refer to humans, but neuter for inanimates. Thus the threegender system appears to have two unmarked forms — one for humans and one for other elements. Nouns and nominal adjectives referring to humans illustrate a further type of markedness, which has a semantic basis. Schane points out that les Américains (masc pi) 'the Americans' is unmarked in that it can refer to both sexes, while les Américaines (fem pi) refers only to female Americans. We might add that the phenomenon is not limited to the plural: le professeur (masc sg) 'the teacher' may be a woman. There are many such nouns, but there are few nouns of feminine gender which refer to males. Moreover, it is not difficult to find similar examples in Slovene, and

196

Resolution rules: agreement in person, number, and gender

other languages under discussion, of masculine nouns which may refer to both sexes. Leaving markedness aside, we may argue that as there are nouns in these languages which refer to both sexes and which are masculine, this means that there is independent semantic justification for the use of the masculine for gender resolution. Of the criteria discussed, the only one which applies to all the languages in question is the final point concerning semantic justification. We observed that French uses the masculine to refer to humans of both sexes and that this is the resolution form for conjuncts of different genders. Note that it is semantically justified only for nouns denoting humans; when inanimates are conjoined we cannot make the same claim. However, in such cases gender has little or no semantic justification in any case. We claim, therefore, that it is the instances where a semantically justifiable form is possible which determine the resolution form. The remainder, for which neither form is semantically justifiable, merely conform with the semantically justified cases. This appeal to a semantically justified form effectively excludes markedness from the operation of the resolution rules. We shall discuss justifiable forms for the different languages in turn. While we may view French examples like les Américains as instances of semantic neutralization, the important thing is simply that there are cases where the masculine refers to both sexes. The resolution form is established on that basis and need not refer to markedness, which thus determines the resolution form only indirectly. This view enables us to avoid a contradiction inherent in Schane's approach; he claimed the resolution form for French was the unmarked gender but the marked number. If markedness were indeed the determining factor we would expect the unmarked number to be used. Moreover he suggests tentatively that, as the first person is used when conjuncts of mixed person include a first person, then the first person should be the unmarked person ((1970) p. 292). It is generally accepted, however, that the third person is the unmarked person (it is used with items which are not specified for person). As the unmarked form and the resolution form do not coincide either for person or for number, it is not surprising that there is no direct correspondence between the unmarked gender and the gender resolution form. 4.2.

Semantic justification and clear marking of plurality

We have claimed that the forms used for gender resolution are those which have semantic justification in a given language. While this is indisputable in the case of Bantu languages, which have semantic type gender resolution, how does it apply to languages of the syntactic type? The division into syntactic and semantic types of gender resolution is accurate insofar as

Greville Corbett

197

it refers to a difference in the way the rules operate in particular languages. However, I suggest that in all instances, the motivating factor behind the choice of the form to be favoured by the resolution rules is semantic; this factor is the use of semantically justified gender forms. In instances such as the use of the masculine in French, this basic principle of semantic justification is clear enough. There is also a second motivating factor at work, namely that the resolved form should bear a clear indication of number. This factor, clear marking of number, is also based on semantics because, as we discussed above, the use of a dual or plural with conjoined elements is semantically justifiable (while it may be that none of the gender forms would have any semantic backing). A form may be favoured either because it is at least partially justified in semantic terms or because it is a clear marker of number. Thus gender may be made subservient to number (a category which more closely corresponds to the real world). The extent to which these two principles are observed depends on the morphological possibilities of a given language. We will reconsider each of the languages analyzed, bearing in mind that we must explain not only why certain forms are favoured for resolution but also why the others are also used or are excluded. In Bantu, the morphological resources are considerable. Thus of the plural predicate agreement forms in Bemba (Givon (1972) p. 17), all except one are uniquely plural. Providing this form is avoided, then any resolution rule will mark plurality clearly. Bantu can follow both principles absolutely: the gender forms used are semantically justified (based on the human/ non-human distinction) and they mark plurality unambiguously. The rules given for Bantu follow both our proposed principles and are relatively simple. None of the other languages analyzed has the same lack of ambiguity in its morphology. For this reason the two principles cannot operate with the same consistency. Let us consider how they apply to Slovene: in this language the masculine is the favoured form, the feminine is used providing all conjuncts are feminine, and the neuter is totally excluded. As mentioned above, the use of the masculine is semantically justified when humans are involved (the masculine gender is the one used for reference to both sexes). When inanimate nouns are involved, then no gender would be semantically justified and so these follow the animates where there is a semantic motivation. When the subject consists of two nouns referring to females, the use of the feminine is semantically justified, and this form is also used by inanimates. How then is the exclusion of the neuter to be explained? For this it is necessary to examine again the possible forms for predicates which show gender agreement in Slovene, as given in the Table.

198

Resolution rules: agreement in person, number, and gender

TABLE 5 masculine

feminine

neuter

singular

0

a

o/e

dual

a

plural

i

e

a

(This Table is slightly more complex than Table 3 above as we include the neuter ending -e, which is taken by certain adjectives but which did not appear in any of our examples.) Suppose the subject consists of two inanimate singular nouns; the number resolution rules specify the dual. The gender resolution rules cannot specify a form which would be semantically justified in terms of gender; we claim, therefore, that they will mark number as clearly as possible. Neither dual ending is unambiguously dual: the -a ending is found also in the singular, and the -/' ending in the plural. However, the finite verb forms end in -a for all genders; for this reason -a is a clearer marker of duality than -/ and so is favoured by the resolution rules (it is used except when there are two feminine conjuncts). Now let us consider instances where the gender resolution rules are to mark the predicate clearly as plural. Then the neuter ending -a would not be favoured because it coincides with the feminine singular and, though this is probably of less importance, with the masculine dual. Of the remaining alternatives, the feminine plural -e also occurs in the singular while the masculine -/' is found in the dual as well as in the plural. In terms of marking plurality, a case can be made for the masculine and for the feminine, but the neuter ending would be avoided. We can now assess the relative merits of the different forms. The masculine forms (dual and plural) are semantically justified in some cases (when the conjuncts refer to male persons or to persons of both sexes). In the dual, the masculine marks number more clearly than the alternative, and in the plural it marks number as clearly as or more clearly than the alternatives. As a result of these two factors, the masculine is favoured by the resolution rules. The feminine is also semantically justified in some instances (when the conjuncts refer to female persons); the feminine/neuter dual form marks number less clearly than the masculine, but in the plural, the feminine form marks number more distinctly than the neuter would. The feminine occurs as a resolution form, but is more restricted than the masculine. The neuter has no semantic backing and does not mark number clearly; it is therefore excluded from the resolution rules. In Polish there are only two agreement forms in the plural; both are used in gender resolution, the masculine personal being the favoured form

Greville Corbett

199

as it is used when the conjuncts contain a mixture of masculine personal and non-masculine personal forms. The use of the masculine personal as the semantically justifiable gender can be explained on the same semantic grounds as for French and Slovene. (Note too that the word panstwo 'Mr and Mrs, ladies and gentlemen', which refers to both sexes, takes masculine personal agreement forms.) Similarly, the use of the non-masculine personal forms is semantically justified when the conjuncts refer to female humans. The forms used, therefore, are semantically justified in some cases; the principle of clear marking of number also operates, though this is not immediately evident from the Table of agreement forms (which includes alternative forms which did not occur in our examples). TABLE 6 masculine personal

0/y/i

singular plural

i/y

feminine

neuter

a

o/e

non-personal

y/e

Table 6 shows that the available plural endings are found in the singular as well. However, the masculine personal ending is distinguished in an important way: a mutation of consonant is required in the case of many adjectives and in the past tense (byli~by4y). Thus the masculine personal form is clearly marked for plurality; again our two principles point to the same form. The West Slavonic languages are at different stages of losing gender forms in the plural; in each case, the form which is gaining ascendency is also the form favoured by the resolution rules (Corbett (1982)). It is unlikely that conjoined structures are a sufficiently frequently occurring construction to be the motivation for the change. It is more likely that gender differentiation is being lost in the plural for independent reasons; the form to survive is that which marks plurality clearly, which is for that reason the one favoured by the resolution rules. We have already discussed the semantic justification for the use of masculine forms for gender resolution in French. The principle of marking plurality also points to the masculine form, though the motivation is less strong than in Polish. In French, singular and plural agreeing forms are usually indistinguishable in speech. Some masculines are distinguishable, e.g. adjectives of the type loyal, plural loyaux. Once again the principle of using the gender form which is semantically justified (even though not in all cases) and the principle of using forms which are clearly marked as plural indicate the same form. (Recall too that the exception to the person

200

Resolution rules: agreement in person, number, and gender

resolution rules in German (§2.1) produced a form in which plurality was clearly marked.) We now turn to two examples in which the two principles do not concur. In the case of Latin we find the now familiar semantic justification for the use of the masculine (thus liberi (masc pi) can refer to children of both sexes) and for the feminine plural when all conjuncts refer to females. There is, however, semantic justification for the use of the neuter plural for inanimates: the neuter plural is used for abstract nominals, e.g. incerta 'fickle things'. Our first principle would allow the use of all three forms. To investigate the second, we must consider the actual forms available, as given in the Table. TABLE 7 masculine

feminine

neuter

singular

us/Q/is

a/0/is

um/0/e

plural

i/es

ae/es

a/ia

The masculine and feminine plural forms are unambiguously plural; in most instances the neuter plural coincides with the feminine singular, though for some agreeing elements it is clearly marked as plural. As we observed above, all three forms are used in gender resolution. While the neuter plural is not favoured by the principle of clear marking of number, it is a semantically justifiable gender in Latin; this ensures that it is used as a resolution form in Latin, unlike Slovene, where it has no semantic justification and is excluded. In Icelandic the neuter plural is the major resolution form, used for all gender clashes. This choice is semantically justified. First the neuter is used for beings of unknown sex: afkvxmi 'offspring', barn 'child', folk 'people, household', kyn 'kin, kindred, god '(heathen) god, idol', folald 'foal'. There is also the case of vi/(poetic) 'woman'. Thus the neuter does not exclude humans. The most significant examples are hjon 'man and wife' and its derivative bondahjon 'peasants (husband and wife)'. Both these nouns refer to persons of both sexes and both require neuter plural agreement forms. There is evident semantic justification for the use of the neuter plural to refer to humans of both sexes. Let us now consider the neuter plural in terms of clear marking of number.

Greville Corbett

201

TABLE 8 masculine

feminine

neuter

singular

ur/r/inn/ill

0/in/il

t/tt/iö

plural

ir/nir/lir/dir

ar/nar/lar/dar

0/in/il

The Table shows that for agreeing predicates, the neuter plural always coincides with the feminine singular (it is, however, distinct in the personal pronoun). On the other hand, the masculine and feminine are clearly marked as plural. In this instance our two principles are in direct conflict: the neuter plural is semantically justified while the masculine and feminine are clearly marked as plural. The neuter plural is the favoured form, but all three forms are used in gender resolution. When the two principles are in harmony, this leads to a restriction of the resolution forms used. In Slovene the neuter plural was excluded, while in Polish we observed that the favoured resolved form is extending its scope. More dramatically, in Bantu the majority of the classes are excluded from the output of the resolution rules. When the two principles conflict, this leads to the use of different forms, supported by one principle or the other. The dominant factor appears to be that of semantic justification: the favoured resolution form is always the semantically justified gender for conjuncts of at least one type (normally those referring to humans). When this principle is supported by that of clear marking of number, this may lead to a restriction of the forms available for gender resolution. Thus gender resolution employs semantically justifiable forms, as far as the morphology of a given language permits.

4.3. An apparent counter-example We shall now examine a language where there appears to be a morphological principle involved in gender resolution; we must demonstrate that even here there is a semantic motivation behind the resolution rules. The language involved is Serbo-Croat which, as we noted earlier, has rules similar to its South Slavonic neighbour, Slovene. It differs from Slovene in having lost the dual, so there are fewer agreement forms available. These are illustrated by the past active participle of biti 'to be'.

202

Resolution rules: agreement in person, number, and gender

TABLE 9 masculine

feminine

neuter

singular

bio

bila

bilo

plural

bili

bile

bila

The resolution rules appear to be as in Slovene, that is to say: if all the conjuncts are feminine then the feminine form is used and otherwise the masculine form is used. When there is at least one non-feminine conjunct there is no problem — the masculine form is the resolved form. In our first example feminine and neuter are conjoined: (48)

Znanje (neut sg) i intuicije (fem sg) su kod njega Knowledge and intuition have in him saradivali (masc pi) i dopunjavali (masc pi) se... worked-together and supplemented each-other (Andric, Travnicka Hronika) 'Knowledge and intuition worked together in him and supplemented each other...'

Similarly, when neuter nouns are conjoined we find a masculine plural predicate: (49)

Njegovo mesto (neut sg) u razvitku kasabe i His place in the-development of-the-town and njegovo znacenje (neut sg) u zivotu kasabalija bili his importance in the-life of-the-inhabitants were (masc pi) su onakvi kako smo ih napred ukratko opisali. such as have them before briefly described (Andric, quoted by G u d k o v ( 1 9 6 9 ) p. 91) 'His place in the development of the town and his importance in the life of the inhabitants were such as we described them earlier.'

When all the conjuncts are feminine, then we would expect feminine agreements, as we find in the following example: (50)

Opreznost (fem sg), suptilnost (fem The-discretion, subtlety till bezbrojnih poruka of-these innumerable assignments mladica ... the-young-man

sg) i pedanterija (fem sg) and pedantery zbunjivale (fem pi) su perplexed (Andric, Travnicka Hronika)

Greville Corbett

203

The feminine nouns in this example are of two different types.Pedanterija, like the majority of nouns in Serbo-Croat, ends in -a in the nominative singular. There is also a sizable group of nouns like opreznost and suptilnost which end in a consonant. These decline rather differently from nouns like pedanterija (they also decline differently from masculine nouns, the majority of which also have no inflection in the nominative singular and so end in a consonant). Though morphologically dissimilar, the two types of feminine noun generally behave in the same way for agreement purposes, which is of course why they are assigned to the same gender. However, they are involved in a most interesting complication of resolution rules, pointed out by Gudkov (1965). A masculine predicate is possible, even though all the conjuncts are feminine, providing that at least one of them ends in a consonant, as in the following example: (51)

Vredali (masc pi) su ga nebriga (fem) i Offended him the-carelessness and lakomislenost (fem) Tahir-begova. (Andric Travnicka Hronika) capriciousness of-Tahir-beg 'The carelessness and capriciousness of Tahir-beg offended him.'

Agreement of this type is not obligatory, as example (50) shows: masculine and feminine agreements are both found frequently. The gender resolution rules are similar to those required for Slovene, but we must allow for the first rule to be optional: 1. if all conjuncts are feminine, then the feminine form will be used; if at least one of the conjuncts is a feminine ending in a consonant, then this rule is optional; 2. otherwise the masculine will be used. Examples like (51), which require the condition for making Rule 1 optional, are particularly interesting because agreement rules normally refer to syntactic or semantic categories. The condition referring to a feminine ending in a consonant is of a different sort. It can be viewed as a morphological condition, referring to a particular declensional class; this has the advantage of being the 'next best thing' to a syntactic condition. Alternatively, one could consider it a phonological condition; this approach has the advantage of linking the feminines of this type to the masculines (both end in a consonant in the nominative singular), which is probably part of the explanation for the phenomenon, but it has the disadvantage of allowing phonology into syntax. There is no clear evidence to favour one view rather than the other. However, we must in either case extend the possible criteria for gender resolution proposed in §3. There we gave

204

Resolution rules: agreement in person, number, and gender

examples of gender resolution according to a semantic or a syntactic principle, as well as mixed types. In Serbo-Croat we find a morphological or phonological principle at work in the operation of the resolution rules. In §4 we claimed that whatever the conditions on the resolution rules, there was always a semantic motivation behind them. It appears that the Serbo-Croat data are particularly damaging here. It is not the use of the masculine as the favoured resolution form which is the problem: that can be explained as for Slovene since it is the semantically justified gender; and the functional argument, the fact that it marks number clearly, is stronger than in Slovene as the complicating factor of the dual is not present. The difficulty is the condition referring to feminine nouns which end in a consonant: it is not obvious how this can be justified in semantic terms. However, the declension includes a large proportion of abstract nouns, and practically no animates. When one collects examples of conjoined noun phrases it is striking that the overwhelming majority involve conjunctions of nouns of the same semantic type (all animate or all inanimate). Thus when a feminine noun ending in a consonant is one of the conjuncts then there will normally be no animates in the subject. This means, in turn, that the use of the feminine agreement form will have no semantic justification (unlike its use with animate conjuncts which refer to females). What is happening, I suggest, is that Serbo-Croat is moving from gender resolution rules operating on syntactic conditions towards a completely semantic system. This new system would take the following form: 1. if all the conjuncts refer to females, the feminine form is used; 2. otherwise the masculine form is used. The fact that the feminines ending in a consonant are affected first is to be explained, therefore, by the fact that the nouns involved are almost exclusively inanimate. The suggestion that the language is moving towards rules of the type just given is borne out by other data presented by Gudkov (1974) p. 61. Occasional examples are found in which subjects consisting entirely of feminine nouns in -a take masculine agreements: (52)

Stula (fem) i staka (fem) bili (masc pi) su sve sto je A-wooden-leg and crutch were all that tadasnja medicina mogla da mu pruzi. (Popovic) of-that-time medicine could to-him offer 'A wooden leg and a crutch were all that medical science of that time could offer him.'

Greville Corbett

205

What is particularly significant is that I have found no examples of masculine agreement with feminine nouns referring to persons. 3 The rules required for modern Serbo-Croat show that gender resolution rules may require conditions other than the more common syntactic and semantic types. However, the motivation behind the rules is still semantic. Serbo-Croat was previously similar to Slovene in that the masculine form was favoured, the feminine also used and the neuter excluded in the resolution rules; the factors at work were those of using the semantically justified gender and of marking number clearly. These same motivating factors remain in force. However, the language is moving to a position in which operation of the resolution rules will be determined by a clearly semantic principle ('if all the conjuncts refer to females . . .')• This will have the effect of making the favoured resolution form, the masculine, even more dominant than it is at present.

CONCLUSION

Agreement is a complex phenomenon, as Morgan (1972) and many others have shown (see, e.g., the chapters by Borsley and Stucky in this volume); and agreement with conjoined noun phrases is a particularly interesting part of the problem. While there are considerable differences between languages, we have shown that there are certain constraints on the choice between agreement with one conjunct or with all the conjuncts. These factors relate both to the agreement controller and to the target. When agreement is with all the conjuncts, then the rules for person and number resolution are relatively straightforward and show little variety. However, gender resolution rules are of different types and allow various conditions on their operation. Thus gender resolution rules are language-specific: the grammar of Serbo-Croat, for example, requires a gender resolution rule with a condition that we have not found necessary anywhere else. However, this does not mean that there is no pattern to gender resolution. The need for language-specific rules is a direct result of the widely divergent gender systems found in the languages of the world and of the differing morphological possibilities associated with them. While the operation of the gender resolution rules may be language-specific, the motivation behind them is constant. The principles which dictate the agreement forms to be used are semantic (use of the semantically justifiable form) and functional (clear marking of number). While the interaction of these principles with the morphological systems of particular languages may produce rules of some complexity, the principles themselves are simple and logical.

206

Resolution rules: agreement in person, number, and gender

FOOTNOTES * This chapter was written during the tenure of a Research Fellowship awarded by the Council of the University of Melbourne. I am grateful to Bernard Comrie and Alan Timberlake for comments on earlier papers which helped in the writing of this one; to Wayles Brown, Talmy Givon and especially Roland Sussex for comments on the draft, and to Maya Bradley, Jeri Jaeger, Anna Wierzbicka and Sonia Witheridge for reactions to presentations of the paper. Errors are mine. Versions of the paper were read in the Department of Linguistics (SGS), Australian National University, April 1981; at the Linguistics Symposium at the 51st Annual Congress of the Australian and New Zealand Association for the Advancement of Science, University of Brisbane, May 1981; to Edinburgh University Linguistics Circle, December 1981; and at the Spring Meeting of the Linguistics Association of Great Britain, University of Reading, March 1982. 1. Translations are not included when the meaning is evident from the glosses. Articles have been added to the glosses in the examples from Bantu and Slavonic and in the Icelandic examples. Note too that the auxiliary verb used in compound past tenses in Slavonic examples such as (13) is in fact the verb be. 2. For discussion of an additional complication, see Corbett (1982) footnote 24. 3. It might appear that deca 'children' is a counter-example; however, in Corbett (forthcoming) Chapter 5 I show that its predicate agreement forms are neuter plural. When it is conjoined with a feminine, the masculine agreements found are therefore as expected.

References

Ahlqvist, Anders (1972), Some aspects of the copula in Irish. Eigse 14. 269-274. Ajdukiewicz, Kazimierz (1935), Über die syntaktische Konnexität,StudioPhilosophica 1, 1-27. Reprinted in English under the title "Syntactic connexion" in S. McCall (ed.), Polish Logic 1920-1939. Clarendon Press, Oxford, 1967. Akmajian, Adrian, Susan Steele and Thomas Wasow (1979), The category AUX in universal grammar. Linguistic Inquiry 10, 1-64. Amritavalli, R. (1979), The representation of transitivity in the lexicon. Linguistic Analysis 5, 71-92. Anderson, Stephen R. (1981), Topicalization in Breton. Proceedings of the Seventh Annual Meeting of the Berkeley Linguistic Society, 27-39. Anderson, Stephen R. and Sandra Chung (1977), On grammatical relations and clause structure in verb-initial languages. In P. Cole and J.M. Sadock (eds.), Syntax and Semantics 8: Grammatical Relations, 1-25. New York: Academic Press. Austin, Timothy (1977), A linguistic approach to the style of the early romantic poets. University of Massachusetts Ph.D. dissertation, Amherst. Awbery, Gwenllian M. (1976), Welsh Syntax. Cambridge: Cambridge University Press. Bach, Emmon (1979), Control in Montague grammar. Linguistic Inquiry 10.4, 515531. Bach, Emmon (1980a), In defense of passive. Linguistics and Philosophy 3, 297-341. Bach, Emmon (1980b), Tenses and aspects as functions on verb phrases. In C. Rohrer (ed.), Time, tense and quantifiers, 19-37. Tubingen: Max Niemeyer. Barss, A. (1981), Adjectives, conjoined adjectives, and conjunction in categorial grammar. Unpublished ms., Hampshire College, Amherst. Bartsch, Renate and Theo Vennemann (1972), Semantic Structures. Frankfurt am Main: Athenäum Verlag. Bar-Hillel, Yehoshua (1953), A quasi-arithmetical notation for syntactic description. Language 29, 47-58. Reprinted in Bar-Hillel (1964). Bar-Hillel, Yehoshua (1964), Language and Information. Reading, Massachusetts: Wesley. Bar-Hillel, Yehoshua, C. Gaifman and E. Shamir (1960), On categorial and phrase structure grammars. The Bulletin of the Research Council of Israel 9F, 1-16. Reprinted in Bar-Hillel (1964). Bauernöppel, J., H. Fritsch and B. Bielefeld (1976), Kurze tschechische Sprachlehre, 3rd edition. Berlin: Volk und Wissen. Bierwisch, Manfred (1967), Syntactic features in morphology: general problems of so-called pronominal inflection in German. In To Honor Roman Jakobson, volume 1, 239-270. The Hague: Mouton. Bolkestein, A.M. (1976), Acl- and ut- clauses with verba dicendi in Latin. Glotta 54, 263-291. Bolkestein, A.M. (1979), Subject-to-object raising in Latin? Lingua 48, 15-34.

208

References

Borsley, Robert D. (1983), A Welsh agreement process and the status of VP and S. In this volume. Borsley, Robert D. (in preparation), The VP-complement analysis: evidence from Welsh. Braine, Martin D.S. (1963), On learning the grammatical order of words. Psychological Review 70, 323-348. Reprinted in A. Bar-Adon and W.F. Leopold (eds.) Child Language: a Book of Readings. Englewood Cliffs, New Jersey: Prentice-Hall, 1971. Brame, Michael K. (1975), On the abstractness of syntactic structure: the VP controversy. Linguistic Analysis 1, 191-204. Brame, Michael K. (1976), Conjectures and Refutations in Syntax and Semantics. Amsterdam: North-Holland. Brame, Michael K. (1980), Lexicon versus filters. In Teun Hoekstra, H.v.d. Hulst and M. Moortgat (eds.), Lexical Grammar, 73-95. Dordrecht: Foris. Brauner, S. (1979), Aktuelle Tendenzen der Entwicklung der Konkordanzbeziehungen im Swahili. Zeitschrift fur Phonetik, Sprachwissenschaft und Kommunikationsforschung 32, 422-428. Bresnan, Joan W. (1971), Sentence stress and syntactic transformations. Language 47, 257-281. Bresnan, Joan W. (1976), On the form and functioning of transformations. Linguistic Inquiry 7, 3-40. Bresnan, Joan W. (1978), A realistic transformational grammar. In Morris Halle, J.W. Bresnan and G. Miller (eds.), Linguistic Theory and Psychological Reality, 1-69. Cambridge, Massachusetts: MIT Press. Bresnan, Joan W. (1982), Control and complementation. Linguistic Inquiry 13, 343-434. Buttler, D., H. Kurkowska and H. Satkiewicz (1971), Kultura jqzyka polskiego: zagadnienia proprawnosci gramatycznej. Warszawa: PWN. Cann, Ronald (1982), Inflectional morphology and generalized phrase structure grammar. Manuscript, University of Sussex. Chomsky, Noam (1965), Aspects of the Theory of Syntax. Cambridge: MIT Press. Chomsky, Noam (1970), Remarks on nominalization. In R.A. Jacobs and P.S. Rosenbaum (eds.), Readings in English Transformational Grammar, 184-221. Waltham, Massachusetts: Ginn. Chomsky, Noam (1980), On binding. Linguistic Inquiry 11, 1-46. Chomsky, Noam (1981), Lectures on Government and Binding, Dordrecht: Foris. Ghomsky, Noam (1982), Some concepts and consequences of the theory of government and binding. Cambridge, Massachusetts: MIT Press. Chomsky, Noam and Howard Lasnik (1977), Filters and control. Linguistic Inquiry 8, 425-504. Clements, George N., Joan Maling, James McCloskey and Annie Zaenen (1981), String-vacuous rule application. Linguistic Inquiry, in press. Comrie, Bernard (1975), Polite plurals and predicate agreement. Language 51, 406418. Comrie, Bernard (1982), The theoretical significance of the Latin accusative and infinitive: a reply to Pillinger. Journal of Linguistics 17, 345-349. Contreras, Heles (1981), Discontinuous constituents in Spanish. Ditto, University of Washington. Corbett, Greville G. (1979), The agreement hierarchy. Journal of Linguistics 15, 203-224. Corbett, Greville G. (1980), Neutral agreement. Quinquereme: New Studies in Modem Languages 3, 164-170.

References

209

Corbett, Greville G. (1982), Resolution rules for predicate agreement in the Slavonic languages. Slavonic and East European Review 60, 347-375. Corbett, Greville G. (forthcoming), Hierarchies, Targets and Controllers: Agreement Patterns in Slavic. London: Croom Helm. Curry, H.B. (1963), Some logical aspects of grammatical structure. In Roman Jakobson (ed.), Structure of Language and its Mathematical Aspects: Proceedings of the Twelfth Symposium in Applied Mathematics, 56-68. Providence: American Mathematical Society. De Bhaldraithe, Tomás (1948), Varia. Eigse 6, 47-49. De Bhaldraithe, Tomás (1953), Gaeilge Chois Fhairrge - An Deilbhíocht. Baile Atha Cliath: Institiúid Ard-Léinn. Derbyshire, Desmond C. and Geoffrey K. Pullum (1981), Object initial languages. International Journal of American Linguistics 47, 192-214. Doroszewski, W. (1962), O kulturf stowa: Poradnik jqzykowy, Warszawa:PIW. Dowty, David R. (1976), Montague grammar and the lexical decomposition of causative verbs. In Partee (1976a), 201-246. Dowty, David R. (1978a), Applying Montague's views on linguistic metatheory to the structure of the lexicon. In D. Farkas et al. (eds.), Papers from the Parasession on the Lexicon, 97-137. Chicago: Chicago Linguistic Society. Dowty, David R. (1978b), Governed transformations as lexical rules in a Montague grammar. Linguistic Inquiry 9, 393-426. Dowty, David R. (1982), Grammatical relations and Montague grammar. In P. Jacobson and G.K. Pullum (eds.), The Nature of Syntactic Representation, 79-130. Dordrecht: D. Reidel. Dowty, David R., Robert E. Wall and Stanley Peters (1981), Introduction to Montague semantics. Dordrecht: Reidel. Einarsson, Stefan (1949),Icelandic: Grammar, Texts, Glossary. Baltimore: J. Hopkins. Emonds, Joseph E. (1972), A reformulation of certain syntactic transformations. In S. Peters (ed.), Goals of Linguistic Theory, 21-62. Englewood Cliffs, New Jersey: Prentice-Hall. Emonds, Joseph E. (1976), A Transformational Approach to English Syntax: Root, Structure-Preserving, and Local Transformations. New York: Academic Press. Emonds, Joseph E. (1980), Word order in generative grammar. Journal of Linguistic Research 1.1, 33-54. Engdahl, Elisabet (1980), The Syntax and Semantics of Questions in Swedish. Ph.D. dissertation, University of Massachusetts, Amherst. Engdahl, Elisabet (1982), Parasitic gaps. Linguistics and Philosophy, in press. England, J. (1976), 'Dixo Rachel e Vidas': Subject-verb agreement in Old Spanish. Modern Language Review 71, 812-826. Epstein, S.D. (1980), A formulation of late closure and minimal attachment in categorial grammar. Unpublished ms., Hampshire College. Fiengo, Robert W. (1974), Semantic Conditions on Surface Structure. Ph.D. dissertation, MIT. Findreng, A. (1976), Zur Kongruenz in Person und Numerus zwischen Subjekt und finitem Verb im modernen Deutsch. Oslo: Universitetsforlaget. Flynn, Michael (1981a), Structure Building Operations and Word Order. Ph.D. dissertation, University of Massachusetts, Amherst. Flynn, Michael (1981b), Nominalizations in categorial grammar. Paper read at the Linguistic Society of America Winter Meeting, New York. Frazier, Lynn (1978), On Comprehending Sentences: Syntactic Parsing Strategies. Ph.D. dissertation, University of Connecticut. Bloomington: Indiana University Linguistics Club, 1979.

210

References

Frazier, Lynn and Janet Dean Fodor (1978), The sausage machine; a new two-stage parsing model. Cognition 6, 291-325. Gazdar, Gerald (1981), Unbounded dependencies and coordinate structure. Linguistic Inquiry 12.2, 155-184. Gazdar, Gerald (1982), Phrase structure grammar. In G.K. Pullum and P. Jacobson (eds.), The Nature of Syntactic Representation, 131-186. Dordrecht: Reidel. Gazdar, Gerald and Geoffrey K. Pullum (1981), Subcategorization, constitutent order, and the notion 'head'. In M. Moortgat, H.v.d. Hülst and T. Hoekstra (eds.), The Scope of Lexical Rules, 107-123. Dordrecht: Foris. Gazdar, Gerald and Geoffrey K. Pullum (1982), Generalized phrase structure grammar: a theoretical synopsis. Bloomington, Indiana: Indiana University Linguistics Club. Gazdar, Gerald, Geoffrey K. Pullum and Ivan A. Sag (1982), Auxiliaries and related phenomena in a restrictive theory of grammar. Language 53, 591-638. Gazdar, Gerald and Ivan A. Sag (1981), Passive and reflexives in Phrase Structure Grammar. In J. Groenendijk, T. Janssen and M. Stokhof (eds.), Formal Methods in the Study of Language, 131-152. Amsterdam: Mathematical Centre Tracts. Geach, Peter T. (1972), A program for syntax. In D. Davidson and G. Harman (eds.), Semantics of Natural Language, 483-497. Dordrecht: Reidel. Gildersleeve, B.L. and G. Lodge (1948), Gildersleeve's Latin Grammar, 3rd edition, revised and enlarged. London: Macmillan. Givon, Talmy (1970), The resolution of gender conflicts in Bantu conjunction: when syntax and semantics clash. Papers from the Sixth Regional Meeting, Chicago Linguistic Society, 250-261. Chicago: Chicago Linguistic Society. Givon, Talmy (1972), Studies in ChiBemba and Bantu grammar. (Studies in African linguistics 3, supplement 3.) Los Angeles: Department of Linguistics and African Studies Center, UCLA. Greenberg, Joseph H. (1963), Some universals of grammar with particular reference to the order of meaningful elements. In Joseph H. Greenberg (ed.), Universals of Language, (second edition) 73-113. Cambridge, Massachusetts: MIT Press. Grevisse, M. (1979), Le bon usage. 9th edition. Gembloux: J. Duculot. Gudkov, V. (1965), Dodatak pravilima slaganja predikata sa vise subjekata. Knjizevnost ijezik 12, pt. 4, 60-61. Gudkov, V. (1969), Serboxorvatskij jazyk: grammaticeskij ocerk, literaturnye teksty s kommentarijami i slovarem. Moscow: Izdatel'stvo Moskovskogo Universiteta. Gudkov, V. (1974), Prilog o pravilima kongruencije. Knizevnost ijezik 21, pt. 1, SSÖl. Gunji, Takao (1981), A Phrase Structural Analysis of the Japanese Language, M.A. dissertation, Ohio State University, Columbus. Gunji, Takao (1982), Generalized phrase structure and Japanese reflexivization. Linguistics and Philosophy, in press. Hale, Kenneth (1973), Person marking in Walbiri. In S.R. Anderson and P. Kiparsky (eis.), Festschrift for Morris Halle, 308-344. New York: Holt Rinehart. Hale, Kenneth (1981), On the position of Walbiri in a typology of the base. Indiana University Linguistics Club. Hale, Kenneth, Laverne M. Jeanne and Paul Platero (1977), Three cases of overgeneration. In P.W. Culicover, T. Wasow and A. Akmajian (eds.), Formal Syntax, 379-416. New York: Academic Press. Harlow, Stephen (1982), The head feature convention, clause types and morphology in Welsh. Unpublished ms., University of York. Harries-Delisle, Helga (1978), Coordination reduction. In Joseph H. Greenberg, Charles A. Ferguson and Edith A. Moravcsik (eds.), Universals of Human Language: Volume 4, Syntax, 515-583. Stanford: Stanford University Press.

References

211

Hinds, John (1974), On the status of the VP node in Japanese. Bloomington, Indiana: Indiana University Linguistics Club. Horn, George M. (1974), The Noun Phrase Constraint. Ph.D. dissertation, University of Massachusetts, Amherst. Hornstein, Norbert (1977), S and X ' Convention. Linguistic Analysis 3.2, 137-176. Horrocks, Geoffrey (1983), The order of constituents in Modern Greek. In this volume. Jackendoff, Ray S. (1975), Morphological and semantic regularities in the lexicon. Language 51.3, 639-671. Jackendoff, Ray S. (1977), X syntax: a Study of Phrase Structure. Cambridge, Massachusetts: MIT Press. Jeanne, Laverne Masayesva (1978), Aspects of Hopi Grammar. Ph.D. dissertation, MIT. Jönsson, S. (1927),A Primer of Modern Icelandic. London: Oxford University Press. Kayne, Richard S. (1981), ECP extensions. Linguistic Inquiry 12, 93-133. Keenan, Edward L. (1978), The syntax of subject-final languages. In Winfred P. Lehmann (ed.), Syntactic Typology, 267-327. Austin: University of Texas Press. Keenan, Edward L. (1980), Passive is phrasal (not sentential or lexical). In T. Hockstra, H.v.d. Hülst and M. Moortgat (eds.), Lexical Grammar, 181-213. Dordrecht: Foris. Keyser, Samuel J. and Paul M. Postal (1976), Beginning English Grammar. New York: Harper and Row. Klein, Ewan (1979), On Sentences Which Report Beliefs, Desires, and Other Mental Attitudes. Ph.D. dissertation, University of Cambridge. Klein, Ewan and Ivan A. Sag (1982), Semantic type and control. In I.A. Sag and D. Flickinger (eds.), Stanford Working Papers in Grammatical Theory 2. Bloomington, Indiana: Indiana University Linguistics Club. Koster, Jan (1975), Dutch as an SOV language. Linguistic Analysis 1, 111 -136. Koster, Jan (1978), Locality Principles in Syntax. Dordrecht: Foris. Koster, Jan and Robert May (1982), On the constituency of infinitives. Language 58, 116-143. Kühner, R. and C. Stegmann (1955), Ausßhrliche Grammatik der lateinischen Sprache. Satzlehre. Erster Teil. 3rd edition. Leverkusen: Gottschalksche Verlagsbuchhandlung. Kulak, J., W. -Laciak and I. Zeleszkiewicz (1966), Jqzyk polski (Skrypt dla cudzoziemcöw). 4th edition. Warszawa: PWN. Lambek, J. (1961), On the calculus of syntactic types. In R. Jakobson (ed.),Structure of Language and its Mathematical Aspects. Providence: American Mathematical Society. Lapointe, Steven G. (1980), A lexical analysis of the English auxiliary system. In T. Hoekstra, H.v.d. Hülst and M. Moortgat (eds.), Lexical Grammar, 215-254. Dordrecht: Foris. Lapointe, Steven G. (1981), Free order phrase structure rules. In Wynn Chao and Deirdre Wheeler (eds.), University of Massachusetts Occasional Papers in Linguistics 7. Lebreton, J. (1901), Etudes sur la langue et la grtmmaire de Ciceron. Paris: Hachette. Lencek, R. (1972), O zaznamovanosti in nevtralizaciji slovnicne kategorije spola v slovenskem knjiznem jeziku. Slavisticna revija 20, 55-63. Lewis, C.T. and C. Short (1879), A Latin Dictionary. London: Oxford University Press. Lewis, David (1970), General semantics. Synthese 22, 18-67. Reprinted in Partee (1976a).

212

References

Lyons, John (1966), Towards a 'notional' theory of the 'parts of speech\ Journal of Linguistics 2, 209-236. Maling, Joan and Annie Zaenen (1982), A phrase structure account of Scandinavian extraction phenomena. In P. Jacobson and G.K. Pullum (eds.), The Nature of Syntactic Representation, 229-282. Dordrecht: D. Reidel. McCawley, James D. (1970), English as a VSO language. Language 46, 286-299. McCawley, James D. (1982), Parentheticals and discontinuous constituent structure. Linguistic Inquiry 13, 91-106. McCloskey, James (1979), Transformational Syntax and Model Theoretic Semantics: A Case Study in Modern Irish. Dordrecht: D. Reidel. McCloskey, James (1980a), Is there raising in modern Irish? Eriu 31, 59-99. McCloskey, James (1980b), A note on modern Irish verbal nouns and the VPcomplement analysis. Linguistic Analysis 6, 459-471. McCloskey, James (1981), Notes on Case and control in modern Irish. Unpublished ms., University College Dublin. McCloskey, James (1983), A VP in a VSO langauge? In this volume. Megaard, J. (1976), Predikatets kongruens i serbokroatisk i setninger med koordinete subjektsnominalfraser. Unpublished M.A. thesis, University of Oslo. Montague, Richard (1973), The proper treatment of quantification in ordinary English. In Jaako Hintikka, Julius Moravcsik and Patric Suppes (eds.), Approaches to Natural Language. Dordrecht: Reidel. Reprinted in Montague (1974), 247-270. Montague, Richard (1974), Formal Philosophy: Selected Papers ofRichard Montague. Ed. by R. Thomason. New Haven, Connecticut: Yale University Press. Moravcsik, Edith (1978), Agreement. In Joseph H. Greenberg, Charles A. Ferguson and Edith A. Moravcsik (eds.), Universals of Human Language: Volume 4, Syntax, 331-374. Stanford: Stanford University Press. Morgan, Jerry L. (1972), Verb agreement as a rule of English. In P.M. Peranteau et al. (eds.), Papers from the Eighth Regional Meeting, Chicago Linguistic Society, 278-286. Chicago: Chicago Linguistic Society. O'Brien, M.A. (1956), Etymologies and notes. Celtica 3, 168-184. O Cuiv, Brian (1944), The Irish of West Muskerry, Co. Cork. Dublin: Institute for Advanced Studies. O Murchu, Seamas (1981), Nota gramadai: an rangabhail leanunach. Teagasc na Gaeilge 2. Partee, Barbara Hall (1975), Montague grammar and transformational grammar. Linguistic Inquiry 6.2, 203-300. Partee, Barbara Hall (1976a), (ed.), Montague Grammar. New York: Academic Press. Partee, Barbara Hall (1976b), Some transformational extensions of Montague grammar. In Partee (1976a). Pepicello, W.J. (1977), Raising in Latin. Lingua 42, 209-218. Pillinger, O.S. (1980), The accusative and infinitive in Latin: a refractory complement clause. Journal of Linguistics 16, 55-83. Pullum, Geoffrey K. (1979), Rule Interaction and the Organization of a Grammar. New York: Garland. Pullum, Geoffrey K. (1982), Free word order and phrase structure rules. In James Pustejovsky and Peter Sells (eds.), Proceedings of the Twelfth Annual Meeting of the North Eastern Linguistics Society, University of Massachusetts, Amherst. Pullum, Geoffrey K. and Deirdre Wilson (1977), Autonomous syntax and the analysis of auxiliaries. Language 53, 741-788. Riemsdijk, Henk van (1978), A Case-Study in Syntactic Markedness. Dordrecht: Foris Publications.

References

213

Rizzi, Luigi (1978), Violations of the wh-island constraint in Italian and the subjacency condition. In C. Dubisson et al. (eds.), Montreal Working Papers in Linguistics 11, 155-190. Ross, John R. (1970), Gapping and the order of constituents. In M. Bierwisch and K.E. Heidolph (eds.), Progress in Linguistics, 249-259. The Hague: Mouton. Russell, Bertrand (1908), Mathematical Logic as Based on the Theory of Types. In R. Marsh (ed.), Bertrand Russell: Logic and Knowledge, 57-102. New York: Macmillan, 1956. Sag, Ivan A. (1982a), Coordination, extraction, and generalized phrase structure grammar. Linguistic Inquiry 13, 329-336. Sag, Ivan A. (1982b), On parasitic gaps. Linguistics and Philosophy, in press. Sag, Ivan A. and Ewan Klein (1982), The syntax and semantics of English expletive pronoun constructions. In I.A. Sag and D. Flickingcr (eds.), Stanford Working Papers in Grammatical Theory 2. Bloomincton. Indiana: Indiana University Linguistics Club. Schane, Sanford A. (1970), Phonological and morphological markedness. In Manfred Bierwisch and K.E. Heidolph (eds.), Progress in Linguistics, 286-294. The Hague: Mouton. Sells, Peter (1982), A phrase structure grammar for Irish relative clauses. Unpublished ms., University of Massachusetts, Amherst. Siegel, Muffy E.A. (1976), Capturing the Adjective. Ph.D. dissertation, University of Massachusetts, Amherst. Sommerfelt, Alfred (1965), Sentence Patterns in the Dialect of Torr. In Lochlann, A Review of Celtic Studies, vol. 3, 256-277. Oslo: Universitetsforlaget. Steedman, Mark J. and Anthony E. Ades (1981), On the order of words. Unpublished paper, University of Warwick. Stenson, Nancy (1976), Topics in Irish Syntax and Semantics. Ph.D. dissertation, University of California, San Diego. Stowell, Timothy (1981), Origins of Phrase Structure. Ph.D. dissertation, MIT. Stucky, Susan (1979), Focus of contrast aspects in Makua: syntactic and semantic evidence. Proceedings of the 5th Annual Meeting of the Berkeley Linguistics Society, 362-372. Stucky, Susan (1981a), Word Order Variation in Makua: a Phrase Structure Grammar Analysis. Ph.D. dissertation, University of Illinois, Urbana-Champaign. Stucky, Susan (1981b), Linear order and case marking. Presented to the 56th annual meeting of the Linguistic Society of America, New York, December 1981. Stucky, Susan (1983), Verb phrase constituency and linear order in Makua. In this volume. Tai, James H.-Y. (1973), Chinese as an SOV language. In C. Corum etal. (eds.),Papers from the Ninth Regional Meeting, Chicago Linguistic Society, 659-671. Chicago: Chicago Linguistic Society. Thomason, Richmond (1976a), On the semantic interpretation of the Thomason (1972) fragment. Bloomington, Indiana: Indiana University Linguistics Club. Thomason, Richmond (1976b), Some extensions of Montague grammar. In Partee (1976a), 77-118. Toporisic, J. (1972), Slovenski knjizni jezik, v >lume 3. Maribor: Zalozba Obzorja. Travnicek, F. (1949), Mluvnice spisovne cestiny, volume 2, Skladba. Prague: Melantrich. Vanek, Anthony L. (1970), Aspects of Subject- Verb Agreement. Edmonton: Department of Slavic Languages, University of Alberta. Reprinted in Current Inquiry into Language and Linguistics 23, Edmonton: Linguistic Research.

214

References

Vennemann, Theo (1973), Explanation in syntax. In J.P. Kimball (ed.), Syntax and Semantics 11, 1-50. New York: Seminar Press. Vennemann, Theo (1975), An explanation of drift. In C.N. Li (ed.), Word Order and Word Order Change, 269-305. Austin: University of Texas Press. Voeltz, Erhard (1971), Surface constraints and agreement resolution: some evidence from Xhosa. Studies in African Linguistics 2, 37-60. Wall, Robert E. (1972) Introduction to Mathematical Linguistics. Englewood Cliffs, New Jersey: Prentice-Hall. Ward, A. (1974), The Grammatical Structure of Munster Irish. Ph.D. dissertation, Trinity College Dublin. Warburton, Irene P. (1980), Provlimata skhetika me ti sira ton oron stis Elinikes protasis. [Problems to do with the order of constituents in Greek sentences]. Unpublished ms., University of Reading. Whorf, Benjamin Lee (1946), The Hopi language, Toreva dialect. In H. Hoijer (eH ), Linguistic Structures of Native America, 158-183. New York: Viking Press. Woodcock, E.C. (1959),/! New Latin Syntax. London: Methuen. Zee, N.v.d. (to appear), Samentrekking, een kategoriaal perspectief, GLOT 5.2. Zieniukowa, J. (1979), Sk-ladnia zgody w zdaniach z podmiotem szeregowym we wspó+czesnej polszczyznie. Slavia Occidentalis 36, 117-129. Zwicky, Arnold M. (1977), Hierarchies of person. In W.A. Beach et al. (eds.), Papers from the Thirteenth Regional Meeting, Chicago Linguistic Society, 714-733. Chicago: Chicago Linguistic Society.

Indices

INDEX OF LANGUAGES Bantu 2, 75, 87, 93, 178,184-186. 193, 196, 197, 201, 206 Batak 168 Bemba 185, 197 Breton 9, 52 Chichewa 5 Chippewa 178 Czech 176, 179 English 1, 3 , 4 , 5, 1 1 , 4 3 , 4 8 , 69-71,72, 7 5 , 9 4 , 105, 110, 117, 199, 141,144162, 167, 168-169, 173 English, Hiberno- 43-44 French 182, 183,186-187,191,195-196, 197, 199 Gaelic, Scots 53 German 178, 181-183, 200 Greek 7, 95-112 Greek, Ancient 117, 137 Hindi 110, 116-117 Hopi 141, 162-167, 169, 173 Icelandic 190, 200-201, 206 Ikorovere 92 Imit'upi 92 Irish 1, 2, 8, 9-55, 111, 161 Irish, Early Modern 27 Irish, Old 27 Italian 111 Japanese 4

Latin 2, 3, 113-137, 179, 180,193-194,

200

Luganda 178, 184-185

Makua 2, 4, 7, 75-94, 110, 1 1 6 , 1 7 4 Malagasy 141, 168-169, 174 Mandarin 110 Munster 27 Pama-Nyungan 5, 176 Polish 190-193, 194, 195, 198-199, 201 Romance 96 Russian 178, 181, 182 Scandinavian 111 Serbo-Croat 180, 189-190, 201-205 Slavonic, South 177, 180 Slavonic, West 176, 190-191, 199 Slovene 177, 180, 187-189, 194, 195, 197-198, 199, 200, 201, 203, 205 Spanish, Medieval 181 Swahili 179 Swedish 93 Ulster 30, 51, 54 Warlpiri 176 Welsh 1, 2, 4, 5, 8, 57-74 Xhosa 178

216

Indices

INDEX OF NAMES Ades 172, 173 Ahlqvist 53 Ajdukiewicz 139-41 Akmajian 74 Amritavalli 116-117 Anderson 9, 52 Austin 174 Awbery 73

Fiengo 110 Findreng 181, 182-183, 194 Flynn 3, 6, 148, 153, 154, 167, 171, 172, 173 Fodor111 Frazier 111, 172-173 Gaifman 171

Bach 144, 157-161, 163, 171, 172, 173 Bar-Hillel 141, 168, 171 Barss 154 Bartsch 171-172 Bauernöppel 176 Bierwisch 194 Bolkestein 114-115, 122 Borsley 1, 2, 3 , 4 , 5, 7, 73, 99, 110, 136, 205 Bradley 206 Braine 172 Brame 64, 123 Brauner 179 Bresnan 3, 5, 10, 20, 45, 51, 64, 117 Brown 206 C a n n 2 , 117, 137 Carnap 141 Chomsky 3, 5, 9, 10, 45, 70, 71, 72, 73, 96, 99, 100, 105, 110, 116, 118,123, 148, 173 Chung 9 Clements 30, 53 Coates 136 Comrie 137, 182, 206 Contreras 172 Corbett 7, 181, 182, 191-192, 195, 199, 206 Curry 171 De Bhaldraithe 27, 2 9 , 5 2 , 5 4 Derbyshire 2 Doroszewski 192 Dowty 4, 9, 89, 94, 143, 154, 158, 171, 172, 173 Einarsson 190 Emonds 3, 9, 53, 74, 95, 96, 110, 125 Engdahl 93 England 181 Epstein 173

Gazdar 3, 6, 7, 9, 45, 57, 63, 64, 66, 68, 69, 71, 72, 73, 74, 76, 80, 89, 9 0 , 9 2 , 93, 96, 97, 105, 109, 110, 111, 116, 119, 120, 123, 136, 171, 172, 173 Geach 171 Gildersleeve 179 Givon 175, 184, 185, 197, 206 Greenberg 12, 52, 142, 169, 171 Grevisse 183, 186 Gudkov 203, 204 Gunji 4,5 Haie 5, 94, 109, 164, 173, 176 Harlow 8, 73 Harries-Delisle 178 Higgins 93 Hinds 2 Horn 72 Hornstein 3, 10, 45, 74 Horrocks 2 , 6 , 7 , 7 6 , 116 Husserl 139 Jackendoff 3, 10, 45, 70, 74, 148, 149, 156 Jaeger 206 Jeanne 163-164, 173 Jonsso n 190 Kayne 45 Keenan 50, 168 Keyser 69 Kisseberth 92 Klein 7, 71, 73, 74, 110, 117, 123, 128, 136,173 Koster 3, 66, 70, 172 Kuhner 180, 193-194 Kulak 191 Lambek 144, 145 Lapointe 7, 94, 109, 172 Lasnik 105, 123 Lebreton 193

217

Indices Lencek 177, 187, 195 Lesniewski 139 Lewis 143, 144, 171 Lodge 179 Lyons 171 Maling 93, 111 May 70 McCawley 8, 69, 110 McCloskey 1, 2, 12, 13, 20, 27, 30, 35, 36, 38, 39, 44, 46, 47, 52, 54, 7 5 , 9 9 , 110, 111, 161 Montague 137, 141, 143, 144, 158, 172 Moravcsik 181 Morgan 92, 205 O Cuiv 53 OMurchu 2 7 , 5 3 , 5 5 O'Brien 27-28 Partee 143, 146, 158, 165 Peirce 142 Pepicello 114-115, 123, 126, 136 Peters 143, 172 Pillinger 115, 122, 123, 125, 126, 131, 137 Platero 164 Postal 69 Prince 8 Pullum 2, 5, 6, 7, 45, 57, 63, 64, 68, 69, 71, 72, 73, 74, 76, 89, 92, 93, 96, 109, 110, 112, 116, 136, 171 Rashid 92 Riemsdijk 3, 37, 66 Rizzi 111 Ross 110 Russell 139, 143, 171 Sag 7, 9, 45, 57, 63, 64, 68, 69, 71, 72, 73, 74, 76, 93, 94, 110, 117, 128, 136

Schane 195-196 Sellis 8 Shamir 171 Siegel 152-153 Sommerfelt 12 Stavrou 110 Steedman 172, 173 Steele 74 Stegmann 180, 193-194 Stenson 12, 13, 35, 38, 47, 53, 55 Stowell 45, 172 Stucky 2, 4, 5, 6, 7, 78, 86, 92, 93, 94, 9 5 , 9 6 , 110, 112, 116, 174, 205 Sussex 206 Tai 110 Thomason 123, 158, 173 Timberlake 206 Toporisis 180 Trávnícek 176, 179 Vanek 175 Vennemann 7, 171-172 Voeltz 178 Wall 1, 143, 172 Wallace 8 Warburton 99, 110 Ward 12 Wasow 74 Whorf 164 Wierzbicka 206 Williams 154, 171 Wilson 69 Witheridge 206 Woodcock 114 Zaenen 93, 111 Zee 172 Zieniukowa 192-193 Zwicky 177

218

Indices

I N D E X O F TOPICS adverbials 10, 16, 22, 23, 43, 53, 55 agreement 5, 7, 57-74, 76, 78-79, 82-92, 93, 94, 96, 97, 100, 102, 110, 116, 119,135, 175-206 agreement hierarchy 182 auxiliaries 52

lexical redundancy rules 117-118, 121, 128-130,135-136

case marking 35, 52, 77, 102, 114, 118, 119, 122, 137 categorial grammar 139-174 clefts 13-14, 17, 2 4 - 2 5 , 4 5 , 4 9 - 5 0 , 5 2 , 5 3 clitic doubling 102, 103-104, 107, 111 clitics 58-61, 102, 104, 183 concord 5, 7, 57-74, 76, 78-79, 82-92, 93, 94, 96, 97, 100, 102, 110, 116, 119, 135, 175-206 constituent order 6, 7, 75-94, 95-112, 116, 119, 140-142, 144-171 control 46 coordination 7, 31-32, 111, 175-206 copula 14, 53, 152

negation 47-48 nominalization 148-149 noun classes 82-85, 88-89, 184-186 noun phrase constraint 72 number 7, 64, 116, 122, 131, 134, 175, 177-178, 179, 180, 181, 182, 183, 187, 189, 196, 197, 199, 200, 201, 202, 205

demonstratives 76-77 discourse 87, 92, 93 duals 177-178, 187-188, 198 equi 55, 58 focus 14, 17-18, 20-25, 50-51, 53, 101109, 111, 112 free word order 2, 5, 75, 82, 92, 95-96, 110, 174 gender 7, 131, 175, 178, 179, 180, 183, 184-205 generalized left branch constraint 105-6 generalized phrase structure grammar 57, 66, 70, 71, 72, 76, 80, 82, 86, 87,88, 93, 94, 95, 97, 115-116, 123-124 grammatical relations 89

markedness 117-118, 131,137,142,170, 194-196 mutation 29-30, 59, 73, 195

object agreement 7, 79, 82-92, 110 OSV word order 2, 7 5 , 9 5 , 1 1 0 OVS word order 75, 95 parasitic gaps 93 parsing 111, 134, 172 passives 4, 5, 30-31, 50, 54, 113, 115, 119-120, 125, 127, 129-135, 136, 153, 158 person 7, 64, 116, 122, 175-177, 179, 182, 183, 205 phantom categories 74, 110 progressive phrases 10, 12-55, 153 projection path 81 pronouns 35, 41, 58-59, 61, 65, 74, 8283, 111, 125, 133, 134, 182, 183, 195 questions 48-49, 70-71, 107-108, 111, 112

head feature convention 71, 73, 89, 94, 102 heavy noun phrase shift 11, 110

raising 58, 79, 93, 113-115, 120-137 reflexives 35 relational grammar 115 relative clauses 86, 87,154-156,164,173 resolution rules 175-206 right wrap 159-161, 170, 173 rightward dependencies 111

ID/LP rule format 6, 93, 96-97, 116, 119 infinitives 38-41, 48-49, 64, 77, 99, 113115, 120-137, 161, 172 islands 2 9 , 7 6 , 111

sentences and sentential complements 2, 3, 4, 45, 51, 55, 57-74, 77, 99, 105, 156 SOV word order 75, 110, 169

219

Indices subcategorization 3, 20, 110, 114, 116, 118, 123, 150, 156 subjects 4 6 - 4 7 , 48, 52, 58, 62, 74, 75, 102, 103, 105-106, 108, 111, 148, 170, 173 subjunctive 99 SVO word order 12, 75, 95, 9 9 , 1 1 0 , 1 1 1 that-trace facts 105-106 tone 93 topicalization 3, 4, 6 9 , 77-82, 86, 9 0 , 9 4 , 101-109, 111, 112, 164-166 transitivity 83-85, 114, 116-121, 126136, 137, 144, 159, 173 u n b o u n d e d dependencies 30, 70-71, 78, 80, 8 6 , 8 7 , 1 0 3 , 1 0 9 , 111, 173 verb phrase c o m p l e m e n t s 4, 69-70, 74, 77, 99

verb phrase ellipsis 75 verb phrase fronting 3, 69, 71, 74, 75 verb phrase topicalization 76, 77-82, 86, 88, 90, 94 verb phrases 1, 2, 3, 4, 7, 9-55, 57-74, 75-94, 99, 105, 110, 136, 147, 170, 173, 174 verbal nouns 13, 28, 29, 38-39, 45, 54, 5 5 , 6 0 , 6 2 - 6 3 , 114 VOS word order 75, 168, 174 VSO word order 1, 9-55, 75, 9 5 , 9 9 , 1 1 0 , 169, 170-171, 174 word f o r m a t i o n 39-40, 4 2 X-bar syntax 2, 3, 10, 1 7 - 5 1 , 5 3 , 6 3 - 7 4 , 109, 136, 141, 142, 149, 151