Perspectives on Functional Grammar [Reprint 2020 ed.] 9783112329603, 9783112329597

Citation preview

PERSPECTIVES ON FUNCTIONAL GRAMMAR

PERSPECTIVES ON FUNCTIONAL GRAMMAR

Editors: Teun Hoekstra Department of Dutch State University Leyden Harry van der Hülst INL, Dutch Lexicological Institute Leyden Michael Moortgat INL, Dutch Lexicological Institute

Leyden

¥ 1981

FORIS PUBLICATIONS Dordrecht - Holland/Cinnaminson - U.S.A.

This volume also appeared as no. 3(1980)3/4 (Leiden Linguistic Journal)

o/GLOT

© 1980 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 authors. ISBN 90 70176 27 0 Printed in the Netherlands by Intercontinental Graphics.

Table of Contents

Introduction Teun Hoekstra An Outline of Functional Grammar

3

Studies in Functional Grammar Simon Dik & GvozdanoviC Jadranka Subject and Object in Functional Grammar

21

Simon Dik et al. On the Typology of Focus Phenomena

41

Casper de Groot On Theme in Functional Grammar. An Application to Some Constructions in Spoken Hungarian

75

Jan de Jong On the Treatment of Focus Phenomena in Functional Grammar

89

Judith Junger Copula Constructions in Modern Hebrew

117

Sita Kishna

135

The Recipient State Construction in Sarnami

Josien Lalleman The Functional Grammar View on Dutch Word Order: Data from Spoken Dutch of Foreign Workers Kwee Tjoe Liong In Search of an Appropriate Relative Clause

157 175

Jack Prentice "Beneficiary" and Other Semantic Functions in Murut

191

Workgroup on Functional Grammar On the Functional Grammar of Teaching Verbs Comparison

with

other

203

theories

William D. Davies Choctaw Subjects and Multiple Levels of Syntax

235

Theo M.V. Janssen Montague Grammar and Functional Grammar

273

J. Lachlan Mackenzie Functions and Cases

299

David M. Perlmutter Functional Grammar and Relational Grammar: Points of Convergence and Divergence

319

Preface and Acknowledgements

The goal of this volume is to inform the reader about a linguistic theory developed in Dik (1978):Fnnetional Grammar. To achieve this goal we invited not only a number of linguists working within the context of this theory, but also proponants of other theoretical frameworks to discuss FG against the background of their own views. The appearance of the collection owes much to the encouraging reactions of Simon Dik. We thank all contributors for their cooperation ( some of them especially for typing out their own manuscript). Thanks for various kind of help go to Mans de Nijs and Marcel Thelen.

Introduction

An Outline of Functional Grammar

Teun Hoekstra State University Leyden 1. Preliminaries 1 Functional Grammar (FG) is one of the various theories of language that have been developed as alternatives to transformational grammar during the last decade. The theory was presented by Dik in Functional Grammar (Dik (1978a)), although many of the basic ideas behind it can be traced back to earlier work of Dik (e.g. Dik (1968), (1975), (1977)). Since its presentation in 1978, FG has met with many positive reactions as will be clear from the numerous references in the literature as well as from the increasing number of linguists working within this framework. Especially in the Netherlands, Dik has succeeded in inspiring many linguists (see e.g. the proceedings of the annual meeting of the Dutch Linguistic Society (Zonneveld & Weerman (1980)). Therefore, the journal GLOT decided to devote a theme-issue to FG. It collects a number of studies within the framework of FG and it provides a platform for critical discussion by representatives of other frameworks. FG is an eclectic model. With Relational Grammar it shares the conviction that generalizations within and accross languages can be captured only in terms of primitive relations like subjefct and object, whereas these are taken as derivative in generative grammar. In another fundamental respect, it is related to Fillmore's case theory, in that it is assumed that semantic functions are needed as independent notions in order to describe the syntactic structure and the ultimate expressions. Within FG, then, syntax is not regarded as an autonomous system, but as a system based on semantics. This view of grammar contrasts with the autonomy thesis defended within generative grammar. Other features of FG can be traced back to the Prague School of linguistics and the pragmatic theory of the Oxford School in

4

HOEKSTRA

Philosophy (Austin and Searle etc.). We won't discuss these commonalities with other frameworks in any detail. The present introduction is meant to provide the necessary background for the papers in this volume. Section 2 gives the basic outline of the theory. The papers collected here can be divided into two groups. First, there is a number of papers written within the framework of FG. They deal with some theoretical issue 2 or tackle some descriptive problem . A second group of papers compares FG with other related frameworks. Within this category, Perlmutter's and Davies' papers represent Relational Grammar and MacKenzie1s paper discusses the justification of the semantic functions recognized by FG in relation to the cases, i.e. semantic roles, in case grammar. 2. The structure of Functional Grammar 2.1. Predications Within the formation of linguistic expressions according to FG, two steps can be distinguished. First, a predication is formed, which is then mapped onto its actual form by means of expression rules. Before the application of expression rules, two levels of function assignment are relevant: the assignment of syntactic functions and the assignment of pragmatic functions. predications

syntactic function

pragmatic function

expression

assignment

assignment

rules

The basic element of a predication is a predicate. Predicates are represented in the lexicon in the form of predicate frames which contain in a sense the mini-grammar of the sentences in which they occur. Predicate frames are specified for the number of arguments a predicate takes. The argument positions are labeled for semantic functions like Agent, Goal, Recipient etc. Furthermore, selectional restrictions with respect to terms that fill the argument positions, are given in the lexical representation of predicate frames as well. (1) illustrates the way in which the predicate give is represented in the lexicon: (1)

give (x.:animate(x,)) V I 1 Ag

2 Go

J

:animate(x_) ) j Ree

(1) provides the information that give is a predicate of the category V(erb) which takes three arguments, an Agent, a Goal and a Recipient, the first and last of which are specified with a selectional predicate animate.

OUTLINE OF FG

5

From predicate frames, predications are formed by inserting appropriate terms into the argument positions. Objects thus constructed are referred to as nuclear predications. Nuclear predications are divided into four types according to the parameters 'control' and 'dynamism' in the following way: dynamism

control

(2) Actions

+

(3) Processes

+

-

(4) States

-

-

(5) Positions

-

+

+

The following examples illustrate these four types of states of affairs: (2') (3') (4') (5')

John hit the ball The ball went over the fence The ball lies in the middle of the field John stood in the middle of the field

(Action) (Process) (State) (Position)

Dynamic states of affairs are characterized as involving some kind of change, whereas controlled states of affairs involve some participant controlling the state of affairs. In Dik (1978a) the number of semantic functions available in FG is left open. We shall not dwell on the subject here, but it is important to note that the semantic functions are thought to correlate with these basic states of affairs. E.g. all and only action predications have Agents; Goal function is assigned to those entities which are affected (or effected) by the operation of some controller (Agent or Positioner) or Force; Force is the function assigned to entities presented as non-controlling instigators of processes, e.g. the wind in 'The wind opened the door'; etc. The kind of state of affairs that is expressed by a nuclear predication is relevant for the formation of full predications from nuclear predications. Full predications are constructed from nuclear predication by extending the latter with satellites. What satellites a predication can be extended with is predictable on the basis of the type of state of affairs: Action predications can always be extended with a manner satellite, whereas State predications never can, etc. Functions like Instrument, Beneficiary, Temporals and Locatives are normally expressed by satellites^. An example of a full predication in which appropriate terms have been inserted is: (6) (dancev O

V

'

^

W

Action

(y

l : b e a u t i f u l ( y l " Manner

6

HOEKSTRA 2.2. The structure and function of terms Full predications are arrived at through the operation of term insertion inserting terms into the argument and satellite slots. The inherent features of the terms must match the selectional restrictions specified for the argument positions. If not, some special interpretive strategy is called for. Terms are expressions used to refer to entities in some world. Reference itself is regarded as a pragmatic notion: in order to make sure that the addressee

will pick out the intended referent of the term, the speaker narrows

the set of potential referents of the term down by using Restrictive predicates (which are of the same formal type as the ordinary predicates discussed in the previous section), such that he can be confident that the amount of descriptive information will suffice. A small set of terms is taken to be basic. They are stored as such in the lexicon. Examples of these basic terms are proper names and pronouns. Most terms, however, are constructed by means of productive term formation rules according to the following schema: (7) ( ]: (11)

assassinate,, (x, -.human (x,)) _ (x_ :human (x„)) „ V I 1 Ag 2 2 Go ~df murder^x,) (x„:politician(x„):important(x.))_ 1 Ag 2 ^ 2 2 Go

(12)

murder,, (x. : human (x.)) _ (x_ .-human (x_))_ V I 1 Ag i ¿. Go =

df

tklll

V

'VAg^WAction.

(y

l:

Untend

V

«Vag»

0

"

0

»!^

1

>Clrc

i.e. assassinate is defined as 'murder an important politician' and murder is to kill someone with the circumstance that one intends to perform that Action. Kill itself can also be defined, in terms of cause to die. Unlike the procedure of generative semantics, this theory of lexical decomposition does not allow the defining predicates to be within the scope of syntactic rules. Hence, meaning definitions can only be used in interpreting predicates, but not in forming them. Because of the rejection of any kind of meta-language, the analysis of an object-language is strongly tied to the language itself, which is regarded as a point in favor (1978a:47)).

(Dik

OUTLINE OF FG

9

2.4. Syntactic function

assignment

Before full predications are mapped onto actual linguistic expressions, two kinds of functions are assigned to elements of the predication, syntactic and pragmatic functions. FG only recognizes two syntactic functions, subject and object."'

Although the name suggests something else, the syntactic func-

tions are interpreted in a semantic way. Predications are said to describe a particular state of affairs. The syntactic functions have the effect of giving a specific point of view on this state of affairs. Subject is assigned to the argument that refers to the most salient participant in the state of affairs described, whereas the constituent with object function gives the secondary vantage point. Alternative subject and object assignments may result in different expressions, but, since subject and object assignment are defined as specifying a certain perspective on some state of affairs description, a condition for two sentences to be regarded as the result of different syntactic function assignments is that the sentences describe the same state of affairs or, put differently, that they are truth conditionally synonymous. If this condition is not satisfied, we are dealing with two distinct predicates, related to each other by means of a predicate formation rule or both listed in the lexicon.

Morphological identity is no barrier to

invoke a predicate formation rule. The alternation in the spray and load etc. sentences illustrated in (13) is accounted for by means of a predicate formation rule which relates two predicates spray etc. to each other:

(13)

a.

They sprayed paint on the wall

b.

They sprayed the wall with paint

(13a) and (13b) cannot be regarded as the result of alternative syntactic function assignment, since they can refer to different states of affairs (see Dik 1980, ch. 2). 6 There is variation among languages with respect to the possibilities of syntactic function assignment. The variation is limited by means of the continuity hypothesis. Among the semantic functions, a hierarchy is assumed, determined by the degree to which the semantic functions are central to the predication. This hierarchy is comparable to the Keenan & Comrie (1977) Hierarchy of Accessibility. It differs from the latter with respect to the nature of the positions on the hierarchy: semantic functions in FG, grammatical relations in the Keenan & Comrie Hierarchy. Since no definite inventory of semantic functions is available, the Semantic Function Hierarchy

HOEKSTRA

10 gives the following partial ordering:

(14)

Semantic Function Hierarchy Agent > Goal> Recipient > Beneficiary > Instru > Loc > Temp

In general, subject and object assignment become more difficult and more marked as we descend down the hierarchy. The continuity hypothesis now states that for each function assignment a continuous stretch of the hierarchy is available: languages vary with respect to the

cut-off points they

select for each function assignment. So, if subject assignment is possible up to the Recipient function, as in English, the continuity hypothesis predicts that subject can also be assigned to Goals and Agents. The cut-off point for object assignment in English is assumed to be Beneficiary. The following examples illustrate the subject and object assignment possibilities in English:

(15)

(16)

a.

John A g

b

"

J

C

"

MarY

°

h

%

gave the b o o k ^

Subj

gaVe Mary

Subj

RecSubj

Was

to Mary R e c

obj

Rec Obj

g l v e n t h e book

book

the

Go

by

d.

The b o o k Q o S u b j was given Mary R e c

a.

John A g

b.

John

c.

*Mary

bought a book G o

Subj

Ag Sub]

Qbj

Go

J o h Qbj

%

by John

for Mary

bought Mary„ „ . a book, J " Ben Ob] Go

Ben Sub]

was bought a book by John. „ .. 3 Go -1 Ag Subj

The markedness of assignments around the cut-off point is related to the following observations: there is individual variation regarding the acceptability judgments of sentences involving assignments of any of the functions 7 around the cut-off point,

language change with respect to assignment possi-

bilities always takes place around the cut-off points and there may be restrictions on the possibility of assigning a syntactic function to the cutoff point. Assignment of a syntactic function to a term will result in giving the constituent a more fixed and prominent position in the sentence and in suppressing the formal marking of the semantic function (e.g. by for Agents, to for Recipients in English etc.). Given the semantic definition of the syntactic functions, one expects the constituent with subject function

OUTLINE OF PG

H

to precede the constituent with object function. We shall return to this in section 2.7. There may be languages for which assignment of subject and/or object is not relevant. Criteria to determine the relevance of object assignment in a particular language involve the existence of the so-called dative alternation as illustrated in English in (15a) and (15b) and of the so-called raising to object constructions as in "John believed Bill to be a fool" (see Dik (1979a)) . If a language does not display any of these possibilities, one can conclude that object assignment is irrelevant. The presence of subject assignment in the grammer of a particular language can be argued for on the basis of passives. For Serbo-Croatian (Dik & Gvozdannvic (this volume)) and Hungarian

(De Groot (this volume)), it is argued that neither subject

nor object assignment is relevant. This tallies with the relatively free word order of those languages (see also Dik (1980, ch. 6)).

2.5. Pragmatic function assignment After the assignment of syntactic functions, pragmatic functions are assigned. These are functions which specify the informational status of the constituents involved within the wider setting in which they occur. What this setting is, depends upon extralinguistic knowledge of the speaker (his beliefs, knowledge of the particular situation etc). It should be noticed, however, that the pragmatic functions themselves are relevant for determining the actual form of linguistic expressions. Of the four pragmatic functions available in FG, two are regarded as belonging to constituents outside of the predication proper. Theme is assigned to constituents preceding the predication, tail to constituents following it. Thus, the general schema assumed is: (17)

Theme, Predication, Tail

Theme and Tail constituents can therefore be compared with constituents in left and right dislocation position resp. A constituent with Theme function is interpreted as presenting the domain or universe of discourse within which it is relevant to express the predication. Just as in current versions of generative grammer, the Theme constituent is not assumed to be extracted from the main predication. This would be impossible in cases like:

12

HOEKSTRA (18)

zoo wa hana ga nagai elefant Theme nose SUBJ long "As for elefants, their noses are long"

(example from Japanese, taken from Li & Thompson (1976)) Dik (1978a) distinguishes between two types of relations that can hold between the Theme and the predication. In the first case, the relation is specified in terms of pragmatic relevance only, as in sentences like (18), in the second case, this pragmatic relation is supplemented by a structural relation: for a given Theme x^, the predication following it may be an open predication on x^, i.e. some term within the predication is coreferential with the Theme» as in: (19)

That man, I hate him

Languages vary in the way they express the variable in the predication: they may either insert a pronominal element (as in English) or leave the variable unexpresssed. In the latter case, the structure should not be confused with Topic initial structures like: (20)

That man I like

In the latter cases, the initial constituent is not set off by a comma intonation, which is characteristic for Theme-predication constructions. Tail is interpreted as modifying or clarifying (some constituent of) the predication, i.e. as an afterthought. Therefore, the relation between the Tail and the predication will be more specific than in the case of Themes. Languages may vary with respect to the ease with which they make use of the Tail construction. In some languages it will hardly ever be used, whereas it may even be grammaticalized in others (Dik (1978a: 156)) . If the Tail construction is grammaticalized, the pronominal element corresponding to the Tail will wear off until it becomes a crossreferential element. This would be an instance of "markedness shift" (see below). In Dik (1980, ch. 7), this idea is worked out to explain the existence of subject final languages. Within the predication proper. Topic and Focus are distinguished. Topic is defined as the function assigned to constituents presenting the entity about which the predication predicates something in the given setting. The constituent with Focus function presents the relatively most important information with respect to the message conveyed to the addressee. Notice that this definition of these pragmatic functions does not have a dichotomizing effect

OUTLINE OF FG

13

such that the first part of a sentence presents the old and the other part the new information, nor could such a situation be accommodated in FG (i.e. VP-focus is impossible). For a discussion of focus, see Dik et al. (this volume) and De Jong (this volume). Constituents in sentence initial position cannot a priori be identified as bearing the Topic function. Nevertheless, expression rules may be sensitive to pragmatic functions and, in general, Topics will be found in a prog minent position in the sentence.

The same applies to focus constituents:

generally, constituents with focus function will bear nuclear stress, but this is not necessary. 2.6. Expression

rules

Through the application of the procedures sketched in the previous sections, we arrive at fully specified predications in which the constituents are labeled for category and for semantic, syntactic and pragmatic functions. The expression rules determine the way in which these fully specified predications are actually expressed in sentences. Expression rules account for the following devices: a. the form in which terms are realized, in particular by cases and/or adpositions b. the form in which the predicate is realized, in particular the voice, auxiliary elements and agreement phenomena c. the order of elements d. stress and intonation Generally, but not exclusively, semantic functions will be expressed either by case marking or adpositions. Their position is usually relatively free. Syntactic functions are also expressed by means of case marking or adpositions , but typically, the cases used for subject and object are the most •unmarked1 cases of the language. As stated before, assigning a syntactic function to a constituent will generally result in masking the marking of its semantic function. Very often, constituents with syntactic functions will have a zero marking. Agreement ana crossreference phenomena will more often be triggered by constituents with a syntactic function than by constituents that have only a semantic function. Moreover, unlike constituents with only a semantic function, subjects and objects will generally have a specific position (cf. the remarks above about free word order languages). As for case systems, FG distinguishes between three different systems. First there is the nominative-accusative system employed in many of the

14

HOEKSTRA

Indo-European languages. Secondly, there is an ergative case marking system and finally, the active case marking system. Dik's (1978) discussion of expression rules is for the most part concentrated on these different case systems. This includes a very interesting discussion of so-called ergative switching which it would take us too far to summarize here. Pragmatic functions can be expressed by means of special Topic and Focus markers. These can have the same masking effect on the expression of semantic and syntactic functions as have the syntactic functions on the expression of semantic functions. This is illustrated by the following Japanese examples: (21)

Agent

= ni

Agent

Subj

Agent

Subj Topic = wa

= ga

ni is used for constituents with only agent function, ga is used for all subjects, including Agent subjects, and wa is used for all Topics, including subjects. If, however, no syntactic function is assigned, the original semantic function expression is retained. Thus, an Agent Topic is marked with ni wa.

As will be clear from the above discussion, expression rules have access 9 to what would be called global information in generative grammar : they can be sensitive to semantic, syntactic and pragmatic functions and even to inherent semantic features of terms. For example, some semantic function could be expressed by means of a certain adposition only when it is a [+human] term. The fact that it is necessary for expression rules to have access to this amount of information, argues for the necessary character of this global information. The expression rules for word order are discussed in the following section. 2.7. Word order in FG

Predications are regarded as unordered sets of constituents. A linear order is imposed after the assignment of functions at both the syntactic and the pragmatic level. The rules assigning the order belong to the expression component of the grammar. According to FG, the ordering patterns of each language result from a number of different principles which partly counteract each other. The

OUTLINE OF FG

15

following tendencies are relevant for the determination of a specific word order. 1. the tendency to place constituents with the same functional specification in the same position 2. the tendency to place special categories in special positions, more specifically certain designated categories like complementizers and constituents with a pragmatic function 3. the tendency to order elements from left to right according to their categorial complexity. The third tendency is discussed in section 2.7.2. 2.7.1. Functional patterns of word order The first tendency mentioned above is accounted for by assuming that all languages have one or more functional patterns which define a number of relevant positions for specific categories of constituents. The variation between the possibilities is restricted by the following schema which is assumed to be universal: (22)

Theme, Pi (V) S (V) O (V), Tail

As stated before, Theme and Tail are so defined as to precede and follow the predication respectively. S in this schema stands for the subject which should, according to the semantic definition of subject, precede O(bject). The verb can either precede them both, stand in between or follow them. Notice that subject final languages would deviate from this language independent pattern.

Such languages can be accommodated either by assuming that

subject and object assignment are irrelevant or by assuming that the predication-Tail construction is grammaticalized. These hypotheses are put forward and substantiated in Dik (1980, ch. 6). In accordance with (22), the functional pattern for English would be Pi

S

V

O

Pi stands for a specific sentence initial position that can be used for elements mentioned in 2. above, i.e. complementizers ir. subordinate clauses, interrogative and relative pronouns and for Topic and Focus constituents. Example (20) above has the constituent that man in Pi, preceding the subject. Pi is therefore comparable to the COMP category in generative grammar. Languages may vafy as to the obligatoriness of filling Pi. The functional pattern assumed for Dutch main clauses is:*1"1

lb

HOEKSTRA

PI V . S O V. f

X

V f stands for the finite verb form, V^ for all other verbs. The PI position must obligatorily be filled, unlike PI in English. However, in many cases the Pi position will be filled by the subject since it is often the Topic as well. This may result

in markedness shift: PI as a marker of Topic func-

tion gets reinterpreted as the position marking subjects. This, in turn, may result in creating a new PI position, giving rise to a new functional pattern PI S VO. A diagnostic criterion for the difference between PI VSO and PI SVO languages is that in the latter the placement of constituents other than subjects in PI will result in having two constituents before the verb, whereas in the former, the subject will follow the verb in such cases. Compare (20) with Dutch (23): (23)

Die man mag ik niet That man like I not = (20)

2.7.2. LIPOC Although languages prefer to have constituents with the same functional specification in the same position, another tendency counteracts this tendency. This is the tendency refered to as LIPOC (language independent preferred order of constituents) according to which constituents are ordered according to their categorial complexity. Relative complexity of constituents is defined by the following hierarchy:

(24)

clitics

pronouns > NP' s > NP+postposition > V > PP ' s > embedded clauses

The effects of LIPOC are accounted for by means of such rules as clitic movement and extraposition in transformational theories. They can be observed in many languages: in French, nominal objects follow the verb, whereas pronominal objects precede it; in Dutch, nominal objects precede the verb, whereas prepositional objects can either precede or follow it and sentential objects invariably follow the verb. If LIPOC prefers an ordering of constituents at variance with the order defined by the functional patterns, this pattern will be under tension and it is likely to be adapted. LIPOC, then, gives a kind of functional explanation for the phenomena described by means of the transformations mentioned above. Moreover, it explains many changes in word order found across different languages. It remains necessary, however, to specify for each language

OUTLINE OF FG

17

what positions a constituent with a certain functional specification and a certain categorial complexity can take relative to other elements, in case there is variation. Therefore, LIPOC belongs to meta-theory, rather than to grammars. This concludes our exposition of the structure of FG. This sketch of the structure of FG will suffice to provide the necessary background for the papers in the present volume. The reader is refered to Dik's own work (especially Dik 1978a), for a fully detailed account. NOTES *This introduction is intended only to present necessary information on FG. No critical comments are given, nor are the relations with other frameworks discussed in any detail. On some occasions reference is made to concepts of generative grammer as far as general familiarity with these can be presupposed. For critical discussion and reviews of FG, see Hoekstra (1978), Comrie (1979), Hoff (1979) and Watters (1980). 2 No further exposition is given here of the papers written within the framework of FG. In the remainder of the text of this introduction reference will be made to the papers where the relevant concept of FG is discussed. ^There is no non-intuitive way of deciding between nuclear arguments and satellites. This problem is not unique for FG, however. The same problem is encountered by generative grammar, i.e. the distinction between vl complements and V complements in X1 theory. 4 The terms extensional and intensional are used here in the broad sense of reference and sense respectively. FG as developed so far is in fact an uninterpreted syntax. See Janssen (this volume) on this matter. ^See Perlmutter (this volume) who argues that at least a third syntactic function, indirect object, is needed as well. ^Since predications are regarded as state of affairs descriptions and predications are semantically defined, we expect to find generalizations across languages with respect to predicates describing the same state of affairs, especially with respect to the semantic functions of the arguments. The paper of the Workgroup "Teaching Verbs" (this volume) examines the class of teaching verbs crosslinguistically. Both the expectation of recurrent types of predicates and the claim that different predications involve different state of affairs descriptions are substantiated. ^For many speakers of English, (15d) is of doubtful grammaticality. The reason for that is unclear. Likewise, (16c) is acceptable for some speakers. This illustrates the variation in judgments found in cases of assignments around the cut-off points. g

Kishna (this volume) argues that SarnSmi, a language spoken in Sarnami is a Topic initial language. Moreover, she claims that Topic function may be relevant for the determination of controllers of reflexives etc. This supports here analysis of the "Dative subject construction" and undermines

18

HOEKSTRA

much of the arguments Relational Grammar puts forward for the analysis of these constructions, refered to as inversion construction (cf. Perlmutter (this volume) and references cited there). 9 Since information concerning semantic function and syntactic function is retained, this global information is locally available. For a detailed illustration of expression rules, see Kwee (1979, this volume). ^Lalleman (this volume) discusses the functional patterns assumed to handle word order in Dutch, with particular reference to the acquisition of the correct orders by foreign workers.

References Comrie, Bernard 1979

Review of Dik (1978a). Studies in language 3. 267-76.

Dik, Simon C. 1968 Coordination. Its implications for the theory of general linguistics. Amsterdam: North-Holland. 1975 The semantic representation of manner adverbials. In: A.Kraak (ed.) Linguistics in the Netherlands 1972-73. Assen: Van Gorcum. 1977 Vraagzinnen in een funktionele grammatika. Spectator 6. 407-12. 1978a Functional Grammar. Amsterdam: North-Hoiland. 1978b Stepwise lexical decomposition. Lisse: The Peter de Ridder Press. 1979a Raising in a functional grammar. Lingua 47. 119-40. 1979b Funktionele morfologie. In: T. Hoekstra & H. van der Hülst (eds.). Morfologie in Nederland. Leiden. 1980 Studies in Functional Grammar. London/New York: Academic Press. Hoekstra, Teun 1978

Funktionele Grammatika. Forum der Letteren 19. 293-312.

Hoff, B.J. 1979

Review of Dik (1978a). GLOT 2. 159-163.

Jackendoff, R.S. 1975 Morphological and semantic regularities in the lexicon. Language 51. 639-71. Keenan, Edward L. & B. Comrie 1977 Noun phrase accessibility and universal grammar. Linguistic Inquiry 8. 63-99. Li, Charles N. & Sandra A. Thompson 1976 Subject and topic; a new typology of language. In: C.N. Li (ed.). Subject and Topic. New York: Academic Press. Waiters, John R. 1980

Review of Dik (1978a). Lingua 50. 155-171.

Zonneveld, Wim & Fred (eds.) 1980 Linguistics inWeerman the Netherlands 1977-1979. Dordrecht: Foris.

Subject and Object in Functional Grammar

Simon Dik & Gvozdanovid Jadranka University of Amsterdam

0. Introduction This paper discusses the status of the syntactic functions Subject and Object in the grammar of Serbo-Croatian. It will be argued that, given the framework of Functional Grammar (FG, cf. Dik (1978)), Object assignment has no relevance for Serbo-Croatian. Subject assignment might be used in order to account for the 'pseudo-reflexive' construction in which Goal terms appear to take Subject function. It will be shown, however, that such a treatment cannot account in a natural way for certain semantic properties of this construction, and that it presupposes a rather sharp distinction between this particular use of the reflexive element, and other constructions in which this element occurs. For that reason, another analysis of the pseudo-reflexive construction is considered, in which this construction is regarded as due to a rule of predicate formation. To the extent that such an analysis can be maintained, it can be argued that Subject assignment has just as little relevance in the grammar of Serbo-Croatian as Object assignment. Some consequences of this view are discussed at the end of this paper.

1. Subject and Object assignment in FG The theory of Subj and Obj assignment developed in FG (Dik (1978: ch. 5, cf. 1980: ch. 6)) can be summarized by means of the following diagram: (1)

Ag Subj Obj

Go

+ > + > +>

Rec

Instr

Loc

Time

+

>

Ben +

>

+

> + >

+

+

>

+

>

+

> + >

+

DIK & GVOZDANOVIC This diagram expresses that Subj and Obj assignment are claimed to be sensitive to a hierarchy of semantic functions, in this sense that Agent terms are the first candidates for Subj assignment, then Goal terms, and so on through the hierarchy. Similarly, Goal terms are judged to be the first candidates for Obj assignment, then Recipient terms, and so on through the hierarchy. Each language is hypothesized to have a continuous sequence of assignment possibilities both for Subj and for Obj assignment, up to the 'cut-off point' for those assignments in the given language. As we shall see in this paper, this theory is compatible with the possibility that a given language makes no use of Subj and Obj assignment possibilities at all, i.e. does not enter the hierarchy in the first place. 2. Object assignment in Serbo-Croatian Object assignment is used in FG in order to differentiate between such pairs as: (2)

a.

Ivan gave the book (GoObj) to his brother (Rec)

b.

Ivan gave his brother (RecObj) the book (Go)

These two constructions differ both in the order (the book - to his brother) and in the formal expression of the Recipient term (to his brother - his brother). This is accounted for by the assumption that in (2a) the Goal term, but in (2b) the Recipient term has received Obj function. The Serbo-Croatian equivalents to (2a-b) can likewise have different orders for the Goal and the Recipient term: (3)

a.

Ivan

je

dao

knjigu

svom

bratu

Ivan (nom) aux given book (acc) his brother (dat) ' Ivan gave the book to his brother' b.

Ivan

je

dao

svom

bratu

knjigu

'Ivan gave to his brother the book' In fact, quite a few other orders are possible for the constituents of (3a-b)

SUBJ & OBJ IN SERBO-CROATIAN (4)

23

a.

Dao

je

Ivan

knjigu

b.

Dao

je

Ivan

svom

c.

Knjigu

d.

Knj igu

svom

bratu

bratu

je

dao

Ivan

svom

je

dao

svom

bratu

e.

Svom

bratu

je

dao

Ivan

f.

Svom

bratu

je

dao

kn j igu

knjigu bratu Ivan knj igu Ivan

As can be seen from these examples, however, these alternative orderings do not involve any differences in the case marking on the Agent, Goal, and Recipient terms: the Agent is always expressed in the nominative, the Goal in the accusative, and the Recipient in the dative. Therefore, there is no reason to assume that differential assignments of Object functions are involved in these cases: the alternative orderings can be accounted for in terms of constituent ordering rules which are sensitive to the pragmatic functions assigned to the constituents. Nor does Serbo-Croatian appear to have other construction types for which alternative Object assignment would be required. In Dik (1979) it was argued that English pairs such as: (5)

a.

John believed that Peter played a sonata

b.

John believed Peter to play a sonata

could be handled by assuming that in (5b) the Object function relevant to the believe-level has been assigned to the Subject of the embedded predication (Peter). This would explain both the position and the form (cf. pronominal him) of this constituent in (5b). Compare now the following pair in Serbo-Croatian: (6)

a.

Ivan Ivan

misli

da

Petar

svira

sonatu

thinks that

Peter

(nom) plays sonata (acc)

'Ivan thinks that Peter plays a sonata' b.

Ivan

misli

Ivan

thinks

Petar

da

svira

sonatu

Peter (nom) that plays sonata (acc)

'Ivan thinks that Peter plays a sonata' Again, the difference between (6a-b) is only a matter of constituent ordering: Petar is in the nominative in both cases, and for the rest the constructions are identical. There is no need to assume that alternative

24

DIK & GVOZDANOVIC

Object assignment is involved here. We do find pairs where there is a nominative in the one, and an accusative in the other construction: (7)

a.

Ivan

vidi

kako

Petar

Ivan

sees

how

Peter (nom) plays a sonata (acc)

svira

sonatu

"Ivan sees how Peter is playing a sonata' b.

Ivan

vidi

Petra

Ivan

sees

Peter (acc) how plays a sonata (acc)

kako

svira

sonatu

'Ivan sees Peter, how he is playing a sonata' In this case, however, it can be argued that Petra in (7b) represents a real Goal argument to vidi rather than an argument of the embedded verb which has been 'raised' out of the embedded predication through Object assignment (for further discussion of such constructions as these, cf. Gvozdanovic (1981), and Bolkestein et al. (1981)). We may conclude that there is no clear indication in Serbo-Croatian that alternative Object assignment is of any relevance. Some implications of this will be discussed at the end of this paper. 3. Subject assignment in Serbo-Croatian Subject assignment is used in FG to account for such differences as between: (8)

a.

John (AgSubj) gave the book to Peter

b.

The book (GoSubj) was given to Peter by John

c.

Peter (RecSubj) was given the book by John

Again, Serbo-Croatian has no direct equivalents to the passive constructions (8b) and (8c). There is, in fact, no productive passive construction at all. The only construction type for which Subject assignment to Goal terms might be relevant is the pseudo-reflexive construction exemplified by the following cases: (9)

Jabuka

se

jede

apple (nom) refl eats 'An apple is being eaten'

SUBJ & OBJ IN SERBO-CROATIAN (10)

Kuca

se

25

gradi

house (nom) refi builds 'A house is being built1 Compare these constructions to their active counterparts: (11)

Ivanka

jede

jabuku

Ivanka

eats

apple (acc)

'Ivanka is eating an apple* (12)

Ivan

gradi

kucu

Ivan

builds

house (acc)

'Ivan is building a house1 On the assumption that Subject assignment is involved here, one might say that the Goal terms marked by the accusative in (11)-(12) appear as Subjects, marked by the nominative, in (9)-(10). In order to arrive at (9), for instance, one would then start with the predicate-frame for jesti 'to eat' : (13)

je sti v

(Xl)Ag (X2>Go

Term insertion and Subject assignment to the Goal position would result in:2 (14)

jestiv (xl)ftg

(x.: jabukaN

(x.))^^

Expression rules would then have to map this structure onto the pseudoreflexive construction given in (9). The active construction (11), on the other hand, would get the following underlying representation: (15)

jestiv (x.: IvankaN ( x ^ ) ^ ^

(x. : jabukaN (x.))^

If this line of analysis is followed, however, it must immediately be added that Subject assignment to Goal terms is a heavily constrained operation. In the first place, there can be no expressed Agent in constructions of the form (9)-(10). Thus, given a predicate-frame such as (13), it must be stipulated that Subject can be assigned to Goal only when the Agent position has been left empty. In the second place, the pseudo-reflexive construction is possible only

26

DIK & GVOZDANOVIC

when the Goal position is filled with an inanimate term. Thus, by the side of an active construction such as (16), there is no pseudo-reflexive counterpart of the form (17): such a construction can only be interpreted as a real reflexive, or as a kind of reciprocal construction: (16)

Ivan

tu£e

Ivan beats

brata brother (acc)

'Ivan is beating his brother' (17)

Brat

se

tuSe

brother (nom) refl beats = 1(My) brother is beating himself1 = '(My) brother is fighting' / '(My) brother is being beaten' Should one want to express the fact that my brother is being beaten without specifying the Agent, then one must use an active construction with a 3rd person plural verb, as in: (18)

Brata

tuku

brother (acc) they-beat 'They are beating my brother' The expression rules required for mapping constructions of type (14) onto expressions of type (9) could be formulated as follows: (19)

When Subj is assigned to Goal, then (i) the GoSubj is expressed in the nominative, (ii) the Verb gets a reflexive form, (iii) the Verb agrees in number and gender with the Subj (iv) the GoSubj takes those positions which are characteristic for Subjects in general (i.e. those positions where we also find AgSubj terms).

According to this analysis, then, Serbo-Croatian does indeed have Subject assignment to Goal terms, although of a very restrictive type.

SUBJ & OBJ IN SERBO-CROATIAN

27

4. Problems for the analysis The analysis of pseudo-reflexive constructions in terms of Subject assignment to Goals meets a number of problems which cast doubt on the correctness of this analysis. 4.1. Semantic differences. If the pseudo-reflexive construction comes about through Subject assignment to Goal terms, then one would not expect semantic differences beyond the differences in 'perspective* which, according to FG, are associated with alternative assignments of Subject and Object function. Constructions due to alternative Subject and Object assignment should be semantically equivalent to the extent that they should be capable of describing the same set of states of affairs. In the case of the Serbo-Croatian pseudo-reflexive construction, however, there seem to be semantic differences involved which go beyond differences in perspective. Thus, the pseudo-reflexive tends to take on a general or habitual meaning not present in the corresponding active construction . Compare: (20)

Ivan

piSe

pismo

perom

Ivan (nom) writes letter (acc) pen (instr) 'Ivan is writing a letter with a pen' (21)

Pismo

se

pi§e

perom

letter (nom) refl writes pen (instr) = 'A letter is being written with a pen' = 'A letter is (normally) written with a pen' Thus, there is a tendency to interpret (21) as if 'being written with a pen' is assigned as a property to letters. On the other hand, (20) describes an actual event rather than a general property of Ivan's. If (20) and (21) are derived from the same underlying predicate-frame through alternative Subject assignment, it is not clear how this semantic difference is to be accounted for. There are even cases in which only a generic meaning can be assigned to the pseudo-reflexive construction. Compare the following cases:

28

DIK & GVOZDANOVIC (22)

a.

Ivan

dr2i

no£

Ivan (nom) holds knife (acc) 'Ivan is holding a knife1 b.

*No2

se

dr£i

knife (nom) refl holds 'A knife is being held' c.

No2

se

dr£i

u

ladici

knife (nom) refl holds in drawer 'Knives are kept in the drawer' Again, it would be difficult to see how these differences can be accounted for if (22a) and (22c) are to be derived from the same underlying predicateframe through alternative Subject assignment. For another example compare the following:

(23)

a.

Ivan

predaje

Ivanki

Ivan

teaches

Ivanka (dat) mathematics (acc)

matematiku

'Ivan teaches mathematics to Ivanka' b.

*Matematika

se

predaje

Ivanki

mathematics (nom) refl teaches Ivanka (dat) 'Mathematics is being taught to Ivanka' (24)

Sveuiilistu

u

mathematics (nom) refl teaches at university

Matematika

se

predaje

na

in

Zagrebu Zagreb

'Mathematics is taught at the university of Zagreb'

(23a) describes an actual event; there is no corresponding pseudo-reflexive construction as indicated in (23b). (24), however, is acceptable, since it describes a general fact rather than an actual event.

4.2. Other uses of the reflexive element. The reflexive element se has a number of other uses in Serbo-Croatian. If the pseudo-reflexive construction is simply seen as an expression device for constructions in which Subject has been assigned to Goal, then it is not easy to connect the pseudo-reflexive to these other uses of the reflexive pronoun. It would then seem to be a mere accident that this rather than some other device is used. Let us briefly review the main uses of the reflexive pronoun se. Notice that this pronoun is insensitive to Person and Number, and only inflects

SUBJ & OBJ IN SERBO-CROATIAN

29

for Case. (A)

lexical reflexive

Certain verbs require the reflexive in all their occurrences. Examples are: (25)

a.

bojati se

'to be afraid of'

sjecati se

'to remember'

smijesiti se

'to smile'

In these cases, clearly, the choice of the reflexive is an automatic corollary of the choice of some verbal predicate. The reflexive element must thus be part of the predicate-frames of these verbs, as they occur in the lexicon. (B)

'true' reflexive

In the second place the reflexive pronoun is used to mark identity of reference between some term and another, as in: (26)

On

se

je

ubio

he refl aux killed •He killed himself' In this case, se indicates that the Goal of ubiti 'to kill' is referentially identical to the Agent. (C)

reciprocal

Constructions with se can also get a reciprocal interpretation, as in: (27)

Djeca

su

se

vidjela

children aux refl seen =

(i) 'The children saw themselves'

= (ii) 'The children saw each other' This interpretation can only be assigned when the first argument is nonsingular. There would seem to be a clear semantic difference between the reflexive and the reciprocal meaning, but this is not paralleled by a corresponding difference in form.

30

DIK & GVOZDANOVIC (D)

impersonal

The reflexive element can be used to form impersonal constructions. Compare: (28)

Ivan

spava

na

podu

Ivan

sleeps

on

ground

'Ivan is sleeping on the ground' (29)

Spava

se

sleeps refl

na

podu

on

ground

'It sleeps itself on the ground' = 'One sleeps on the ground' This impersonal construction

seems closely related to the pseudo-reflexive

one discussed in section 3 above. The impersonal construction, too, has a tendency to signal a generic or habitual meaning. Thus, there are both semantic and formal parallels between the impersonal and the pseudo-reflexive construction (cf. Mizirk. (1969)). However, if the latter is analyzed in terms of Subject assignment to Goal, then the former can obviously not be described in the same manner. For the whole point of the impersonal construction is that there is no specified argument to which Subject could be assigned, and certainly no Goal argument, since we are dealing here either with intransitive verbs or with transitive verbs for which the Goal is not specified. Thus, it would seem that the close parallelism between such constructions as the following would be insufficiently captured in the description:

(30)

Pismo

se

piSe

perom

letter (nom) refl writes pen (instr) 'A letter is written with a pen' (31)

Ovdje

se

here refl

pi£e writes

perom pen (instr)

'Here one writes with a pen' Given the similarity both in form and in meaning between these two construction types, one would wish to have a way of more directly relating them in the description.

SUBJ & OBJ IN SERBO-CROATIAN

31

4.3. The so-called 'modal'construction. Both the pseudo-reflexive and the impersonal construction with se can be extended with a nominal term in the dative. This results in a construction usually called 'modal', which is interpreted along the lines of 'X feels like Y 1 , where X is the dative constituent, and Y is the event expressed in the pseudo-reflexive/impersonal construction. Compare the following cases:

(32)

Meni

se

jede

jabuka

me (dat) refl eats apple (nom) 'I feel like eating an apple' (33)

Ivanu

se

ne

gradi

kuia

Ivan (dat) refl neg builds house (nom) 'Ivan does not feel like building a house' (34)

Njemu

se

spava

ovdje

he (dat) refl sleeps here 'He feels like sleeping here' (35)

Meni

se

kuha

me (dat) refl cooks 'I feel like cooking' Again, this construction points to a close parallelism between the pseudoreflexive and the impersonal construction. It is not at all clear how the modal construction should be described. Suppose, first, that we should want to directly relate such constructions as the following: (36)

Ivanka

jede

jabuku

Ivanka (nom) eats apple (acc) 'Ivanka is eating an apple' (37)

Ivanki

se

jede

jabuka

Ivanka (dat) refl eats apple (nom) 'Ivanka feels like eating an apple1 We could then try to derive (37) on the basis of an underlying predication of the form: (38)

jestiv (x.: IvankaN (x.))Ag

(x. : jabukaN

(x.,))^^

DIK & GVOZDANOVIC However, in terms of this underlying structure it would be difficult to explain why the meaning of the resulting construction is not 'An apple is being eaten by Ivanka', but: 'Ivanka feels like eating an apple'. Another problem with an underlying representation such as (38) is that constructions of the form (37) do not so much describe an event in which a person is actively involved, but rather a feeling which irresistibly imposes itself on a person, a process over which that person has no control. For these various reasons we would rather want to analyze the modal construction according to the following bracketing: (39)

(Ivanki)

(se jede jabuka)

in which Ivanka is added as a Recipient to that which is expressed in the pseudo-reflexive or impersonal construction. This, however, would not be compatible with an analysis of the pseudo-reflexive such that jabuka se jede is analyzed as 'an apple is being eaten by someone'. For a literal paraphrase of the form 'To Ivanka, an apple is being eaten by someone' would not naturally bring us to what (37) actually means. 5. An alternative: reflexive predicate

formation

For the various reasons mentioned in section 4 it seems reasonable to try and find some other analysis of constructions with se such that (i) the various uses of se can be more naturally related to each other, and (ii) the semantic properties of these constructions can be more naturally accounted for. Such an analysis can be found if we regard constructions with se as due to rules of predicate formation. Predicate formation rules are rules which take a predicate-frame as input and deliver another, derived predicate-frame

as output (for some discussion of such rules cf. Dik

(1980: ch. 2 and 3)). In the particular case of Serbo-Croatian constructions with se we shall speak of reflexive predicate formation. Notice first that, in order to capture the inherently reflexive predicates mentioned in section 4.2. under (A), we must assume that the lexicon of Serbo-Croatian contains basic (non-derived) predicate-frames containing se. In the case of these lexical reflexives, there is no reason to assume that se represents an independent argument position. Thus, a one-place lexical reflexive such as smijeSiti se 'to smile1 can be represented as:

SUBJ & OBJ IN SERBO-CROATIAN (40)

33

smijeSiti se (x. ) 1 Ag V

Other lexical reflexives are two-place, where the second argument may be realized in the genitive, the dative, or the instrumental case. Although these case assignments appear to be lexicalized to a certain degree, they do seem to roughly correspond to differences in semantic function of the second argument. Thus, we could assume inherently reflexive predicateframes such as the following: (41) (42) (43)

bojati se (x.)„ (x„)„ V 10 2 Source dopadati., se (x. )„ (x„) 10 2 Rec V baviti,, se (x, ), (x-,),. V 1 Ag 2 Instr

'to be afraid of' 'to be pleasing to' 'to be occupied with' c

where Source will be mapped onto the genitive, Recipient onto the dative, and Instrument onto the ablative case. We shall now assume that other constructions with se come about through the application of a productive predicate formation rule which results in predicate-frames similar to those given in (40)-(43). This rule of reflexive predicate formation has the general effect of reducing the number of argument positions of the input predicate-frame by one, and of effecting a semantic modification which can be formulated in general terms. Let us first give some specific examples, and then try to formulate the predicate formation rule in general terms. In the case of jesti 'to eat', reflexive predicate-formation would have the following effect: (44)

INPUT:

jestiv

OUTPUT:

jestiy se

(^Go

MEANING: 'a jesti-event applies to x^1 In the case of a one-place input predicate such as spavati 'to sleep', the effect of the predicate formation rule would be as follows: (45)

INPUT:

spavativ (Xj)^

OUTPUT:

spavatiy se (0)

MEANING: 'a spavati"event applies' Thus, in the case of a two-place predicate the output is a one-place pred-

34

DIK & GVOZDANOVIC

icate, and in the case of a one-place predicate the output is a zero-place predicate (a predicate without argument positions). A zero-place predicate can only be realized in the form of an impersonal construction with a 3rd person singular verb form. The semantic modification in both cases leads to a meaning which can be formulated as 'a V-event applies (to x^)', where V is the input predicate, and Xj is the remaining first argument, if there is any. On the basis of these examples, we can now give the following general formulation of the rule of reflexive predicate formation: (46)

REFLEXIVE PREDICATE FORMATION INPUT:

predicatev (x^) (x2) ... (x^)

OUTPUT:

predicatev se (x^) (x^)

(x^

)

MEANING: 'a predicatev~event applies to x^' As we saw above, transitive verbs can also lie at the basis of impersonal constructions, as in: (47)

Ovdje

se

here refl

pi£e writes

perom pen (instr)

'One writes with a pen here' (48)

Ovdje

se

here refl

ne

umiva

neg

washes

'No washing is done here1 The question is: how are we to account for this type of impersonal construction in terms of the theory developed here? Two possibilities seem to present themselves. The first is to assume that a transitive verb such as pisati 'to write' is reduced by one argument according to schema (46) in the following way: (49)

INPUT:

pisativ ( x ^

OUTPUT:

pisativ se ( x ^

( x ^

(X3)

^

(x 2 > I n s t r

MEANING: 'a pisati-event applies to x^' and to assume that the (x^) position in the derived predicate-frame can either be filled by a term such as pismo 'letter', to result in constructions of the form:

SUBJ & OBJ IN SERBO-CROATIAN (50)

Pismo

se

pise

letter refl writes

35 perom pen (instr)

'A letter is written with a pen' or can be left unfilled, and will then lead to impersonal constructions of the form of (47). Another possibility would be to assume that the reflexive predicateframe given as OUTPUT in (49) can be re-entered as INPUT into the reflexive predicate formation rule (46), so that the number of arguments is again reduced by one. Strictly speaking, this would have to lead to a doubly-derived predicate with double se. In order to avoid this, we would have to assume that se is not added to a predicate which already contains se. The result would then be something like:

(51)

INPUT:

pisatiy se ( x ^

(x 2 > I n s t r

OUTPUT:

pisati se (0) (x,) . V 1 Instr MEANING: 'a pisati-event applies, with the instrument x^' We shall not try to answer the question as to which of these two possibilities is to be preferred in treating impersonal constructions based on underlying transitive predicates^. 6. An extended

example

Let us now illustrate the theory developed in section 5 with an extended example of those constructions which can be formed on the basis of a transitive verb

such as prati 'to wash1. This verb takes an Agent and a Goal

argument, where the Goal position can be filled either with an animate or with an inanimate term. Now consider the following paradigm of possible constructions:

(52)

Ivanka

pere

odijela

Ivanka (nom) washes clothes (acc) 'Ivanka is washing clothes' (53)

Ivanka

pere

djevojke

Ivanka (nom) washes girls (acc) 'Ivanka is washing girls'

36

DIK & GVOZDANOVIC (54)

Ivanka

se

pere

Ivanka (nom) refl washes 'Ivanka washes herself' (55)

Djevojke

se

peru

girls (nom) refl wash (3rd person plural) = a. 'The girls are washing themselves' b. 'The girls are washing each other' (56)

Ovdje here

se

peru

odijela

refl

wash

clothes (nom)

'Clothes are being washed here* (57)

Ovdje here 1

(58)

se

ne

pere

refl

neg

wash

No washing is being done here'

Meni

se

ne

pere

me (dat) refl neg wash 'I don't feel like washing' (59)

Meni

se

ne

peru

me (dat) refl neg wash

odijela clothes (nom)

'I don't feel like washing clothes' Sentences (52) and (53) exemplify the transitive use of the verb prati, with an inanimate and an animate Goal, respectively. In (54) we find the derived reflexive predicate prati se with the meaning 'a washing-event applies', with an animate argument, thus leading to 'a washing-event applies to Ivanka'. The most natural interpretation of this is that Ivanka is washing herself. In (55) we find: 'a washing-event applies to the girls'. How can a washing-event apply to a group of girls? Well, either they are washing themselves, or they are washing each other. Again, the interpretation in which the girls are actively involved in the event is the most natural one, but the construction is neutral as to whether the washing-relation obtains reflexively or reciprocally within the group of girls. Thus, the difference between reflexive and reciprocal is probably irrelevant to the analysis of this construction. In (56) it is expressed that a washing-event applies to clothes. Since clothes do not normally wash themselves, the most natural interpretation is that they are washed by someone. Thus, when the argument position of the reflexive predicate is filled with an animate term, then the interpretation

37

SUBJ & OBJ IN SERBO-CROATIAN in which the referent of that term is actively involved in the event is

preferred, and when it is filled with an inanimate term, then the preferred interpretation is that the referent is passively involved in the event. Although these principles of interpretation constitute strong preferences, they are not absolute: they can be broken in exceptional cases. Thus, if one wants to speak of self-washing windows, one can say:

(60)

Ovi

prozori

these windows

se

peru

sami

refl

wash

selves

'These windows wash themselves' In this case, although the argument is inanimate, it is nevertheless interpreted as being actively involved in the washing-event. Conversely, a construction such as:

(61)

Ovdje here

se

peru

djevojke

refl

wash

girls

could be exceptionally interpreted as 'Girls are washed here". This strongly suggests a place where one can bring one's girls to have them washed, and thus to a certain extent dehumanizes the girls. This also explains the somewhat humorous effect of a construction such as: (62)

Ovdje here

se

peru

odijela

i

refl

wash

clothes

and

djevojke girls

'Clothes and girls are washed here' In (57) we find the impersonal use of the reflexive predicate, to be interpreted as 'No washing-event applies here'. Following the same line of analysis, we can now interpret (58) and (59) as based on the derived reflexive predicate-frame, extended with an extra term specifying a human Recipient. These sentences can then be semantically interpreted as 'No washing-event applies, with respect to me', and 'No washing-event applies to clothes, with respect to me". On the basis of such paraphrases, it is not too difficult to arrive at the semantic interpretation conventionally associated with this construction type. Notice that this analysis correctly predicts that the Recipient is not actively involved in the event: the whole construction rather specifies a negative feeling of the person in-

38

DTK & GVOZDANOVIC

volved, with respect to the washing-event or the clothes-washing-event specified by the reflexive construction. It would seem, then, that the analysis in terms of reflexive predicate formation with its argument-reducing and meaning-modifying properties allows for a more unified treatment of the various uses of the se-construction. At the same time, this analysis yields a more natural analysis of the various semantic properties of the constructions in question.

7. Some

implications

If the analysis developed in sections 5 and 6 is accepted, then we may conclude that Subject assignment is not distinctively used in Serbo-Croatian. As to case marking and agreement, we can simply say that the first argument of any predicate-frame (basic or derived) is to be expressed in the nominative, and that the main verb agrees with this first argument in number, person, and gender. This would mean that neither Subject nor Object function would be relevant to the grammar of Serbo-Croatian. This has certain consequences for the treatment of constituent order in Serbo-Croatian. When a language makes use of Subject and Object function, it is usual for constituents which have acquired these functions to go to specific positions in the clause (cf. Dik (1978: ch. 8)). When a language does not use the functions Subject and Object, it can thus be predicted that its constituent order possibilities will have greater freedom. This is certainly correct for Serbo-Croatian: cf. (3) and (4) above, and the discussion in Gvozdanovifi (1981). At the same time, we may expect constituent order variants to be mainly sensitive to pragmatic factors. This, too, seems to be correct in the case of SerboCroatian.

NOTES 1 (3) and (4) do not exhaust the possibilities of ordering these constituents. See Gvozdanovic (1981) for a general discussion of constituent ordering possibilities in Serbo-Croatian, and of the different pragmatic functions associated with these orderings. 2 Term operators are left unspecified here, because they involve problems irrelevant to the present discussion. 3 However, the following facts could be adduced in support of the second solution mentioned, involving repeated application of the reflexive predicate formation rule, and a blocking of double se expression. Compare:

SUBJ & OBJ IN SERBO-CROATIAN (i)

a. b.

(ii)

a. b.

39

Ivan spava ovdje Ivan sleeps here Ovdje se spava Here refl sleeps 'One sleeps here' Ivan se smije£i Ivan refl smile 'Ivan smiles' Ovdje se smijeSi Here refl smile 'One smiles here'

The impersonal construction (ib) presumably originates through application of reflexive predicate formation on the one-place predicate spavati 'to sleep'. Notice, now, that a lexically reflexive predicate such as smijesiti se 'to smile' can also be used impersonally, as in (iib). Presumably, the same rule must be used to derive (iib) from the predicate-frame of smijeSiti se, as is used to derive (ib) from the predicate-frame of spavati. But this implies that we need a 'double se constraint' anyway, in order to avoid constructions of the form *Ovdje se se smijeSi. REFERENCES BOLKESTEIN, A. Machtelt et al. 1981

Predication and expression in Functional Grammar. To appear.

DIK, S.C. 1978 Functional grammar. North-Holland Linguistic Series 37. Amsterdam: North-Holland. 1979 Raising in a Functional Grammar. Lingua 47, 119-141. 1980 Studies in Functional Grammar. London & New York: Academic Press. GVOZDANOVIC, J. 1981 Word order and displacement in Serbo-Croatian. In: Bolkestein et al. (1981). M0RK, H. 1969 Impersonal and passive sentences in Serbo-Croatian (a generative study). Scando-Slavica 15, 247-262.

On the typology of Focus Phenomena

Simon Dik et al. University of Amsterdam

0. Introduction This paper discusses some aspects of the typology of the pragmatic function Focus within the framework of Functional Grammar. We will mainly concentrate on the following two questions: (a) what sorts or subtypes of Focus have to be distinguished if we want to adequately account for the variety of Focus-related constructions to be found in different languages ? (b) under what conditions do have have to assume the presence of more than one Focus within one construction ? The paper is organized as follows. Section 1 briefly sketches the status of the Focus function within FG. Section 2 more clearly defines the problems involved in questions (a) and (b). Section 3 summarizes the theory of Focus types developed by Watters (1979) in his description of Aghem. The influence of Watters' work will be evident throughout this paper. Section 4 discusses scope differences connected with the Focus function. Section 5 is concerned with the relations between Focus and Contrast: must these be seen as distinct phenomena, or can they be interpreted as subtypes of one focussing mechanism ? Section 6 presents a typology of the Focus function in terms of a number of relevant parameters. Section 7, finally, illustrates how more complex Focus constructions could be actually described according to the principles of FG. The examples from different languages used in this paper were determined by the knowledge and interests which happened to be represented in our group. No attempt has been made to arrive at anything like a representative sample of languages. Obviously, our contribution is anything but definitive, as may be expected in this relatively unexplored area of linguistic organization.

DIX ET AL.

42 1. The status of Focus in FG. Consider the following question-answer pair: (1)

A:

What did John buy ?

B:

John bought an umbrella.

According to the principles of FG, (IB) would in this context get the following underlying representation: (2)

Past buy,, (dix.: John(x. ))„ _ . ,m J V i i AgSubjTop (ilx.: umbrella (x.)) _ ._ 3 j GoObjFoc

The functional information contained in this representation tells us that John is the Agent, the Subject, and the Topic, and that an umbrella is the Goal, the Object, and the Focus of this construction. The pragmatic functions Topic and Focus were defined as follows (Dik 1978: 19): Topic: the Topic presents the entity 'about' which the predication predicates something in the given setting.1 Focus: the Focus represents what is relatively the most important or salient information in the given setting. Focus has been assigned to the Goal in (2), because that constituent contains the crucial information requested in (1A). A constituent with Focus function presents information 'bearing upon the difference in pragmatic information between Speaker and Addressee, as estimated by the Speaker.' (Dik 1978: 149). By uttering (1A) the speaker indicates that there is a difference between him and the addressee concerning the identity of the thing that John bought. The answer (IB) levels out this difference. It seems safe to assume that the Focus function, in the general sense outlined above, is relevant to the organization of all natural languages. Languages differ, however, in the expressive devices which they use for signalling Focus in the actual form of linguistic expressions. These devices come in the following sorts: (i) (ii)

intonational prominence: extra stress, higher tone; special constituent order: special positions for Focus constituents in the linear order of the clause;

43

TYPOLOGY OF FOCUS (iii)

special Focus markers: particles marking off the Focus from the rest of the clause;

(iv)

special Focus constructions: constructions which intrinsically define a certain distribution of Topic and Focus over the structure of the clause, such as cleft- en pseudocleft constructions. 2

Different languages use some or all of these devices in different combinations .

The above account of Focus, though probably not incorrect, is certainly too vague and "unspecified to arrive at a correct understanding of the variety of Focus phenomena found in natural languages. The aim of the present paper is to arrive at a more detailed and differentiated picture of the typology of Focus phenomena.

2. One Focus or more ? So far we have acted on the assumption that only one Focus is assigned per predication, and that there is only one, undifferentiated type of Focus to be assigned. Both of these assumptions will turn out to be incorrect simplifications of the facts to be accounted for.

2.1. More than one Focus per

predication.

It is easy to see that it must be possible for more than one Focus function to be assigned to the constituents of a single predication: questioned terms necessarily have Focus, and many languages allow more than one term to be questioned within a single predication, as in:

(3)

Who ate what in the restaurant ?

Similarly, in an answer to a question such as (3), each term corresponding to a questioned term in (3) must be assigned Focus:

(4)

JOHN_ ate CHILI (in the restaurant) Foe Foe

From these examples it is clear that Focus must be assignable to more than one constituent within a single predication. In this respect Focus assignment differs from the assignment of semantic functions and syntactic

44

DIK ET AL.

functions: functions such as Agent and Subject cannot be assigned to more than one constituent per predication. Another difference between Focus and the other functions is, that Focus can be assigned to the predicate or to the predication as a whole, whereas the other functions are restricted to terms. 2.2. Parameters determining different types of Focus. Focus was generally defined above as characterizing that part of a linguistic expression that contains what is relatively the most important or salient information within the given setting of use. Focus will usually relate to differences in the pragmatic information of speaker and addressee, as estimated by the speaker. There are different ways, however, in which a piece of information may be important or salient. Both informationally and syntactically, different sorts of Focus constructions may come out according to the different values that the Focus function may take along a number of parameters. These parameters, briefly outlined here, will return for further discussion in the course of this paper. (a)

Scope of the Focus: is Focus assigned to the predication as a whole (or its truth value), or only to some constituent of the predication ?

Consider the following examples: (5)

John went to the market

(6)

John DID GO to the market

(7)

John went to the MARKET

(5) can be taken as a Focus-neutral assertion of a certain fact. In (6), the Focus is on the truth value of the predication as a whole. This construction could be used in a context in which the issue was whether or not it was the case that John went to the market, to assert emphatically that this was indeed the case. In (7), Focus is only on the Direction constituent. This construction can be used when the issue is not whether or not John went somewhere (that he went somewhere is presupposed in this case), but when the issue is where precisely he went. Thus, scope differences of the Focus function lead to quite different types of linguistic expression. Comparable differences are found in other languages. Compare the equivalents of (5) - (7) in Bahasa Indonesia:

45

TYPOLOGY OF FOCUS (8)

(9)

(10)

Ali

pergi

ke

pasar

Ali

go

to

market

'Ali

went to the market' PERGI

(Focus-neutral)

ke

pasar

to

market

Ali

ADA

Ali

exist go

'Ali

DID GO to the market'

(Focus on truth value)

Ali

pergi

ke

PASAR

Ali

go

to

market

'Ali

went to the MARKET' (Focus on Direction)

Notice that Bahasa Indonesia, too, uses an emphatic auxiliary verb (with a literal meaning of 'to exist') in order to express Focus on the truth value of the predication as a whole.

(b)

Emphasis or Contrast: does the Focus merely emphasize the importance of a given constituent, or does it contrast the information of that constituent with the information given in some other constituent ? In the latter case, is the constituent with which contrast is established contained in the same linguistic expression, or in some other linguistic expression, or is it presupposed ?

Consider the following examples:

(11)

John bought a TOYOTA .' (you wouldn't believe it)

(12)

John bought a TOYOTA, not a VOLKSWAGEN !

(13)

John bought a TOYOTA : (not a Volkswagen, as you seem to assume)

In (11), there is not necessarily a contrast with another specific type of car: the speaker can use (11) merely to emphasize the unexpectedness of the event described. In (12), there is an explicit contrast between two constituents contained in the same linguistic expression. And in (13), there is a contrast with respect to the presumed presupposition of the addressee. In the particular case of (11) and (13), English has no formal difference corresponding with emphasis versus contrast. As we shall see below, other languages do have such formal differences.

(c)

Relation to pragmatic information of addressee: is the Focus constituent meant to fill in, to expand, to restrict, or to replace

46

DIK ET AL.

a given piece of information contained in the pragmatic information of the addressee ? Consider the following examples: (14)

John went to the MARKET ("filling in')

(15)

John not only went to the MARKET, but also to the STORE ( ' expanding' )

(16)

John didn't go to the MARKET, he only went to the STORE ( ' restricting')

(17)

John didn't go to the MARKET, he went to the STORE ('replacing')

The relevance of these distinctions will be discussed in section 6 of this paper. (d)

New or Given-, does the Focus present information new to the addressee, or does it select a salient item from among a •given' set of possible items ?

It is often assumed that Focus can only characterize information new to the addressee. There certainly is a strong correlation between Focus and 'new information'. Indeed, how could information (assumed to be) 'given1 to the addressee constitute the most important or salient information contained in a linguistic expression ? However, consider the following exchange: (18)

A:

What did John and Bill finally decide on ?

B:

JOHN bought a TOYOTA and BILL a VOLKSWAGEN

In (18B), both John and Bill have Focus, although these constituents in themselves do not contain new information. What one could say in such a case, of course, is that the relations John - Toyota and Bill - Volkswagen are (presumed to be) 'new' to the addressee, and that this explains the Focus distribution in this case. We return to this matter below, in section 5. (e)

Exhaustive or not: is the information transmitted meant to be exclusively true of the focussed item, or does it leave open the possibility that it may be true of other items as well ?

TYPOLOGY OF FOCUS

47

Consider the following examples: (19)

JOHNSON voted against the bill

(20)

It was JOHNSON who voted against the bill

In (19) it is emphatically asserted that Johnson voted against the bill. This assertion is compatible with the possibility that others may also have voted against the bill. (20) , however, will normally be -used to indicate that Johnson was the only person who voted against the bill. This latter sort of Focus is what Kuno (1972) has called 'exhaustive listing' Focus. 2.3. Multiple Focus and different Focus types. It is evident from the above discussion that the questions discussed in 2.1. and 2.2. are interrelated in several ways. When there are different Focus constituents in a single predication, these may be of different types, as in: (21)

John didn't buy a TOYOTA, he bought a VOLKSWAGEN

Here, there is Focus on both Toyota and Volkswagen, but these Focus constituents function contextually in different ways: the former is meant to remove information from the pragmatic information of the addressee, the latter is used to insert new information into that pragmatic information. Conversely, a given type of Focus may lead to multiple Focus constituents within one predication, as is the case in 'parallel' constructions such as: (22)

JOHN bought a TOYOTA and BILL a VOLKSWAGEN

An adequate theory of Focus will have to account for these interrelationships . 3. Watters 1 analysis of Aghem. 3.1. Summary of the analysis. In his analysis of Focus phenomena in Aghem, a Grassfields Bantu language of Cameroon, Watters (1979) demonstrates that the following Focus types must be distinguished in order to account for the syntactic properties of Focus constructions in this language.

48

DIK ET AL.

Types of Focus in Aghem: a. Unmarked Focus: occurs when the focus or foci are not formally marked on the surface, the sentence having the basic word order. Example: Inah gave fufu to his friends. b. Assertive focus: that information which the speaker believes, assumes or knows the hearer does not share with him or her. Example: Inah gave FUFU to his friends. c. Counter-assertive focus-, that information which the speaker substitutes for information which the hearer asserted in a previous utterance. Example: Inah gave FUFU (not yams) to his friends. d. Exhaustive listing focus: that information which the speaker asserts is unique in the sense that the rest of the sentence is true only with respect to it and false with respect to all other units of information which could be appropriately substituted for it in the sentence. Example: Inah gave FUFU ONLY (and nothing else) to his friends. e. Polar focus: the truth value 'true' or 'false' which the speaker asserts concerning a proposition. Example: It is TRUE/THE CASE that Inah gave fufu to his friends = Inah DID give fufu to his friends. f. Counter-assertive polar focus: the truth value 'true' or 'false' which the speaker asserts, contradicting the hearer's previous utterance concerning the truth value of the sentence. Example: It is TOO the case/true (contrary to your denial) that Inah gave fufu to his friends = Inah DID TOO give fufu to his friends. (this is from Watters 1979: 177). Hie Focus mechanisms of Aghem are rather complicated in detail, but the following summary gives the basic principles. Aghem uses the following devices for expressing Focus types: (a) constituent order, (b) a special Focus-bound completive Past tense marker, (c) a special Focus marker, (d) the cleft construction. The basic functional pattern of the clause can be represented as: (23)

S

Aux

Pb

V

Pa

O X

Especially important for Focus are the

1

special positions1 Pb (immediate-

ly before the verb) and Pa (immediately after the verb). Of these two. Pa is the main position for Focus terms. In this respect, Aghem deviates in an interesting way from the more common pattern in which the clauseinitial position PI is used for these purposes.

TYPOLOGY OF FOCUS

49

The main rules for the expression of Focus are as follows: (i)

any Focus term, whether questioned or non-questioned, goes to Pa.

(ii)

when the Focus term in Pa is not the Subject, then up to two non-Focus, non-Subject terms may go to Pb; these are then necessarily part of the presupposition (or Topics), and the Focus term in that case is counter-asserted rather than simply asserted.

(iii)

when the Focus term in Pa is the Subject, then there will be a dummy Subject in the S position, and there may be other Focus terms in the area after Pa; up to two of these may alternatively be placed in Pb. In both cases, the result is a construction with multiple Foci. Notice that multiple term Focus can only be expressed when at least the Subject is in Focus and thus in Pa.

(iv)

there is one neutral and one Focus-signalling completive Past tense marker. When the latter is placed in Aux, Focus is on the value of the predication as a whole: Polar focus,

(v)

when in the same condition all verbal complements are placed in Pb so that the verb is in final position, the result is Counter-asserted polar focus,

(vi)

when the special Focus marker no is placed after a constituent to the right of and including the verb, it adds the value 'Counter-asserted' to that constituent; if the constituent in question already has this value on other grounds, it adds the value 'Exhaustive listing',

(vii)

the cleft construction is used for only exhaustive listing Focus.

We give one example for each of the Focus types cited under a-f above, to show how these principles work: a. Unmarked (24)

focus

ffl

(Focus-neutral): a

mo

zf

ki-b£ an 'son

friends SM Past eat fufu in farm 'The friends ate fufu in the farm' Comment: neutral word order, both Pa and Pb empty. SM = Subj marker.

50

DIK ET AL.

b. Assertive focus: (25)

a

mo

zi

DS Past eat

enao

b£-ko

an

'som

INAH

fufu

in

farm

'INAH ate fufu in the farm' Comment: Subj-Focus in Pa, dummy Subj in S (= DS): assertive Focus on Subj. c. Counter-assertive focus: (26)

enao

mb

Inah

Past in

an

•(No), Inah ate

'som

zf

a-lfm

farm

eat

YAMS

YAMS in the farm'

Comment: Object in Pa, Locative term in Pb: Counter-assertive Focus on Object. d. Exhaustive listing focus-. (27)

a

mb

15 baotom

wfl 'a 6

DS Past be CHIEF 1

mb

buo

this Rel he Past come

It was the CHIEF who came'

Comment: cleft construction: Exhaustive listing focus on 'the chief1. e. Polar focus: (28)

ena

ma'a

Inah

Past-Foe give fufu

fuo

b£-ko

a

ffn-gho

to friends

'Inah DID give fufu to his friends' Comment: Focus Past marker rather than simple Past marker: Polar focus on predication. f- Counter-assertive polar focus:

(29)

fil

a

maa

i bc-k5

friends SM Past-Foe fufu

zf eat

'The friends DID eat fufu' Comment: Focus Past marker, Obj in Pb and verb in final position: counter-assertion of truth of predication.

TYPOLOGY OF FOCUS

51

3.2. Some comments on Watters' analysis. As will become clear in the further course of this paper, we believe that Watters' distinctions are essential to the development of a general typology of Focus phenomena. Some points in his analysis, however, would seem to require some modification. We mention the following points: (i)

We interpret 'unmarked Focus' as being equivalent to 'absence of Focus'. That is, we assume that a linguistic expression which does not relate in any way to a presupposition of the addressee, and has no internal contrast either, is simply Focus-neutral: no Focus function is assigned either to the expression as a whole, nor to any part of the expression.

(ii)

We find the term 'Counter-assertive1 less fortunate, because the type of construction falling under this label can be produced in precisely the same form in relation to (a) a specific assertion which the other participant has just made, or (b) a specific presupposition which the speaker assumes the addressee possesses at the moment of speech. The essential factor would thus seem to be, not what the other has just said, but what the speaker assumes about the pragmatic information of the other at the moment of speaking. The constructions involved are thus 'Counter-presuppositional' rather than 'Counter-asserted'.

(iii)

Some distinctions made by Watters must be further refined, because some of his categories may result in different construction types according to (a) the precise presuppositional conditions involved, and (b) that which the speaker wishes to achieve by assigning Focus to a given item in a linguistic expression.

These points will be clarified in the further course of this paper. 4. Scope differences. In 2.2. above we noted sub (a) that one of the parameters differentiating Focus types concerns the scope of the Focus function: i.e., the question as to what part of the predication is operated upon by the Focus function. A predication consists of a predicate, applied to one or more argument terms, possibly extended with one or more satellite terms. Thus, we get the

52

DIK ET AL.

following possible scope distinctions: The scope of Focus is: (a)

the predication as a whole;

(b)

some constituent(s) of the predication; (bl) the predicate; (b2) some term(s) .

In respect to these scope distinctions, there are close connections between Focus and Negation, and this is no coincidence: Negation typically applies to some piece of information concerning which there is some (real or presumed) difference between the pragmatic information of the speaker and that of the addressee. Thus, what is negated is usually in Focus. We might thus distinguish between positive Focus and negative Focus. The main difference between these is, that negative Focus is always marked by some explicit segmental element in the clause, whereas positive Focus may lack such a segmental marker and may be solely expressed through constituent order and/or suprasegmental distinctions. The scope differences distinguished above may thus be compared to such distinctions as between 'sentence negation' and 'constituent negation': in the former case, the whole sentence (or predication) is in the scope of Negation, in the latter case only some constituent of the predication (either the predicate or some term) is in the scope of Negation. 4.1. Focus on the predication

as a

whole.

If the Focus is on the predication as a whole, it naturally concerns the illocutionary 'point' of the predication: if the predication is presented as an assertion, then the Focus will fall on the truth value of the assertion, as in: (30)

John DID

go to the market

(31)

John did NOT go to the market

In (30) it is emphatically asserted that John went to the market, in (31) it is emphatically denied that he did so. This is a case of Polar Focus in Waiters' terminology. If the predication is presented as an invitation or an advice (in which

TYPOLOGY OF FOCUS

53

case truth value is not at stake), the Focus will concern the force with which such speech acts are presented to the Addressee: (32)

DO come over for dinner :

(33)

DON'T give up !

Since Focus, when it has the whole predication in its scope, may have different values according to the type of speech act in which the predication occurs, we shall use the neutral term Predication Focus for this type. As Watters has shown for Aghem, Predication Focus may differ according to whether the predication is emphatically asserted or counter-asserted. In Dutch, positive asserted Polar Focus has no specific marker (except emphatic intonation), but positive counter-asserted Polar Focus is marked by the emphatic particle wel, which quite clearly brings out the parallelism with negative Polar Focus, as in the following exchange: (34)

A:

Jan heeft zijn huis verkocht 1 John has

his

house sold

'John has sold his house 1' B:

Nee, hij heeft zijn huis NIET verkocht. No , he

has

his

house NOT sold

'No, he has NOT sold his house' A:

Ja, hij heeft zijn huis WEL verkocht ! Yes, he has

his

house

sold

'Yes, he HAS (indeed) sold his house' Thus, Polar Focus is a specific type of Predication Focus; Predication Focus is the positive counterpart of what is usually called sentence negation; and the difference between assertion and counter-assertion is relevant to both the status and the expression of Predication Focus. 4.2. Focus on some constituent of the predication. When Focus is on some constituent of the predication, it can either have the predicate or some term in its scope. In both cases it correlates with predicate or term negation. Thus, Focus is on the Recipient term in all of the following examples:

54

DIK ET AL. (35)

John gave the book to PETER

(36)

John gave the book to PETER, not to CHARLES

(37)

John did not give the book to CHARLES

And in the following examples Focus is on the predicate: (38)

John GAVE the book to Peter

(39)

John GAVE the book to Peter, he didn't SELL it to him

(40)

John did not GIVE the book to Peter

A characteristic of Constituent Focus is that it corresponds to a specific presupposition, which can be reconstructed by taking out the Focus constituent and replacing it by a variable (cf. Chomsky 1972, Jackendoff 1972). Thus, (35) connects with the presupposition 'John gave the book to someone': the only issue between Speaker and Addressee is the identity of the 'someone'. And (38) connects with the presupposition 'John did something with the book with respect

to Peter'; the only issue is what

precisely he did with the book. Sentences with term Focus can typically be given as answers to term questions. Thus, (35) would be a correct answer to the question: (41)

To whom did John give the book ?

In a similar way, we would expect constructions with predicate Focus, as in (38)-(40), to be correct answers to constructions in which the identity of the predicate is questioned. For some reason, however, most languages have no straightforward possibility of questioning the identity of predicates. E.g., if there is a presupposition of the form: (42)

John Verb-ed the book to Peter

then there is, in English and many other languages, no straightforward construction of the form: (43)

*John what-ed the book to Peter ?

Instead, use has to made of circumlocutions such as:

TYPOLOGY OF FOCUS (44)

What did John do with the book ? Did he GIVE it to Peter ?

Thus, terms and predicates behave differently Tjnder Question Focus in these languages. The non-questionability of predicates, however, is not a -universal property of languages. Thus, Indonesian has three verbs derived from the question word apa 'what' which can serve to question the identity of the V: mengapa do what', mengapa(apa)i

'to do what to somebody', and mengapakan

'to 'to do

what with something or somebody', as in the following examples (where PASS stands for Passive) : (45)

A:

Mengapa Ali di kamar nya ? do what Ali at room

his

'What is Ali doing in his room ?' B:

Menulis surat. write

letter

'Writing letters' (46)

A:

Kau-apai

anak

ini, hingga

ia menjerit-jerit ?

you do what to PASS child this so that he scream (This child is done what to by you so that he is screaming^) 'What did you do to this child to make him scream ?' B:

Ku-pukuli

dia

I hit (repeatedly) PASS he (He was hit (repeatedly) by me =) 'I gave him a beating' (47)

A:

Hendak kau-apakan want

buku ini ?

you do what with PASS book this

(What is going to be done with this book by you =) 'What are you going to do with this book ?' B:

Hendak ku-jual want

pada Ali

I sell PASS to

Ali

((It) is going to be sold by me to Ali =) 'I'm going to sell it to Ali'

DIK ET AL.

56 4.3. The distinction

between

term scope and predicate

scope.

Although predicates in Indonesian can be questioned just like terms, it possesses other construction types which discriminate sharply between term scope and predicate scope. One of these is question formation by means of the particle kah. Indonesian Yes-no questions can be formed by means of the question particle kah. Compare: (48)

Ali

membeli

buku

Ali

buy

book

'Ali bought a book (or: books)' (49)

Ali

membeli

buku

kah

?

'Did Ali buy a book ?' The particle kah must have the predicate in its scope. It cannot have term-scope. This can be seen in the following way. In a construction type alternative to (48)-(49), the Subject and the predicate can appear in inverted order. This order gives Focus to the predicate: (50)

Membeli buku, Ali "Ali did buy a book1

(51)

Membeli buku kah, Ali ? 'DID Ali buy a book ?'

It is impossible, however, to get terms in front position with kah: (52)

*Buku

kah, Ali

membeli ?

(53)

*Ali kah, membeli buku ?

A nominal can only be questioned by kah if it functions as a predicate, not as a term: (54)

Ali

guru

Ali

teacher

'Ali is a teacher1 (55)

Ali

guru

kah

?

'Is Ali a teacher ?'

57

TYPOLOGY OF FOODS (56)

Guru, All I 'Ali is a TEACHER !'

(57)

Guru kah, Ali ? 'Is Ali a TEACHER ?'

Fronting with kah can thus be used as a criterion for determining whether a given constituent is a term or a predicate. This is interesting in relation to the analysis of the pseudo-cleft construction. Consider: (58)

Ali

(lah), yang

membeli

buku

itu

Ali

Foe

buy

book

the

Rel

'ALI (was the one) that bought the book' In this constructions Ali is the focussed constituent (optionally marked with the Focus marker lah), and the rest has the properties of a relative clause. Consider now: (59)

Ali

kah, yang membeli

buku itu

?

'Was it ALI who bought the book ?' (60)

Siapa (kah) , yang membeli

buku itu ?

'Who was it that bought the book ?' From these examples it is clear that the Focus constituent of the pseudocleft construction can be questioned by means of kah. The conclusion must be that this constituent functions as a predicate rather than as a nominal term. 3 5. Focus and Contrast. A recurrent problem in connection with Focus phenomena is the precise relationship between the notions of 'Focus' and 'Contrast'. The main question in this respect is: do we need a distinct device for Contrast by the side of the pragmatic Focus function, or can Contrast be handled in terms of (specific instances of) Focus assignment ? In part, the answer that one wants to give to this question is a matter of terminology and definition. We have defined Focus as marking that part of a predication which contains its relatively most important or salient information in the given setting. This implies that there will always be

58

DIK ET AL.

a certain contrast between the Focus part of the predication, which is 1

foregrounded' in one way or another with respect to the 'backgrounded'

rest of the predication. If the notion of contrast is used in this broad sense, then any type of Focus assignment will necessarily imply Contrast. But this is then a contrast between that which is focussed upon, as against that which is not focussed upon.11 'Contrast' is usually restricted, however, to the more specific case in which one piece of information, say X, is explicitly or implicitly opposed to some other piece of information, say Y, which stands in some specific relation

of opposition to X in the given setting. This sort of

Contrast is found in such patterns as the following: (61)

A:

It is the case that Y

B:

No, it is the case that X

(62)

It is the case that X, not Y

(63)

It is not only the case that Y, but also that X

(64)

X did V, but Y did W

From these examples it is clear that the contrast between X and Y may have different sources: B may counterassert X to an assertion Y of A; or the speaker may oppose X to a presumed presupposition Y. In these cases X is meant to correct the conversationally 'given' Y. But Contrast may also be determined by factors internal to the linguistic expression as such, as in (64), where the speaker asserts two different things, V and W, about two different entities, X and Y. If Contrast is used in the restricted sense with respect to an opposition between X and Y, then it follows that Focus does not necessarily imply Contrast: speakers can emphatically assert X without opposing it to an explicitly given or presupposed Y. On the other hand, Contrast will necessarily imply Focus on X, and also on Y if Y is explicitly given in the expression. The opposition between X and Y is obviously sufficient to give both X and Y the status of 'most important or salient information'. In this context it is useful to also consider the relations between Focus and Contrast on the one hand, and the notions 'New' and 'Given' information on the other.

TYPOLOGY OF FOCUS

59

As we saw above, a constituent with Focus function is assumed to present information bearing upon the difference in pragmatic information between speaker and addressee, as estimated by the speaker. It is not immediately clear, however, that this information should necessarily be 'new' to the addressee in the given setting. It may certainly be, but the speaker may also Focus on information which he knows is not new to the addressee, either to stress its importance in the given setting, or to reactivate that information in the addressee's memory. Thus Focus does not necessarily imply that the information focussed upon is new to the addressee. The same applies to Contrast. Although certainly a piece of 'new' information may be contrasted with a piece of 'given' information, the contrast may also be between two pieces of 'given' information, or between two pieces of 'new' information. From this it follows that there is no simple one-to-one relation between Focus or Contrast on the one hand, and 'new' information on the other. 6. A typology of the Focus

function.

Against the background of the various factors discussed in the preceding paragraphs, we shall now propose and discuss a subdivision of the Focus function, based on the various uses to which this function can be put. This subdivision may be seen as an attempt to get a picture of the 'etics' of Focus, in the sense that it distinguishes the various ways in which Focus can manifest itself, without necessarily implying that these different ways are also coded in different grammatical constructions in natural languages. In order to arrive at a clearer picture of the 'emics1 of Focus (i.e. a picture of which of these distinctions are actually coded in grammatical oppositions) , more research on the Focus systems of individual languages will be required. The distinctions that we think are important for an understanding of Focus phenomena can be represented as in Diagram 1. It is to be understood that the different terminal Focus types (italicized in the schema) are yet to be cross-classified for the scope differences discussed in section 4 above. Most Focus types distinguished occur either with Focus on the predication as a whole, or on some constituent of the predication - and in the latter case, either with Focus on the predicate or on a term. These scope differences will be illustrated in our discussion of the different types distinguished.

DIK ET AL.

60 Focus +Contrast

-Contrast

+Specific Presupposition -Corrective

-Specific Presupposition

+Corrective

Completive Selective Expanding Restricting Replacing Parallel Diagram 1. A subdivision of Focus types. As we saw in section 5, Focus can, but need not involve Contrast. If no Contrast is involved, then the Focus information can be seen as 'filling in' a gap in the pragmatic information of the addressee (Completive Focus) . If there is Contrast, the Contrast established can either hold between the Focus information and a specific presupposition, or it can be determined by construction-internal factors, without any specific presupposition being involved. If a specific presupposition is involved, the Focus information can either be selected from an antecedently presupposed set (-Corrective) , or it can be meant to bring about some modification in the pragmatic information of the addressee (+Corrective). These and the other distinctions made in the diagram will be clarified and illustrated below. 6.1. Completive Focus We speak of Completive Focus when the Focus information is meant to fill in a gap in the pragmatic information of the addressee. The clearest case of this is answers to Q-word questions. Thus, when someone asks: (65)

What did John buy ?

his partner can infer that he has the following presuppositional structure in his pragmatic information: (66)

John bought x; x =

.

TYPOLOGY OF FOCUS

61

Question (65) will be interpreted as a request to fill in the blank within this presuppositional structure. When this question is answered by:

(67)

John bought COFFEE

the Focus on COFFEE signals that this particular bit of information is meant to fill in the blank in (66), and is by virtue of that fact the most important information in (67).

Completive Focus does not involve any specific contrast. It relates to a presupposition, but not to a specific presupposition concerning the identity of the unknown entity x in (66). (67) is an example of Completive Focus on a term. Completive Focus on the predicate is found in the following exchange: (68)

A:

What did John do with the book ?

B:

He SOLD it.

It is not immediately clear whether there is any application for the notion Completive Focus with respect to the whole predication. Consider an exchange such as: (69)

A:

Did John sell the book ?

B:

He did.

From one point of view, (69A) could be analyzed as involving an informational gap with respect to the truth value of the predication 'John sold the book': (70)

It is x that John sold the book; x =

.

The answer could then be seen as filling in the blank with the value TRUE. It is clear, however, that there is a difference between Q-word questions and Yes-no questions in this regard: in the latter case, the blank can only be filled in with one of two values, so that the presuppositional structure associated with the question could just as well be given as:

62

DIK ET AL. (71)

It is x. that John sold the book; x is TRUE or x is FALSE.

In that case the answer (69B) would select the value TRUE from among the presupposed possibilities, and the Focus would be of the Selective rather than of the Completive type. 6.2. Salective Focus. We speak of Selective Focus when the Focus information selects one item from among a presupposed set of possible values. Consider the following example: (72)

A:

Did John buy coffee or rice ?

presupposition: John bought x; x = coffee or x = rice B:

He bought COFFEE

The Focussed item in (72B) indicates that coffee is the correct value for x. Selective Focus involves Contrast in this sense that it explicitly or implicitly excludes the other presupposed value(s) as the correct value for x. If this is made explicit, we get constructions such as: (73)

He bought COFFEE, not RICE

Thus, Selective Focus may involve two constituents with Focus, one selecting the correct value, and the other rejecting the incorrect value. But Selective Focus is not 'corrective' in the sense to be clarified below, because A's presupposition in (72) is not incorrect: it has to be specified rather than corrected. Selective Focus on the predicate is found in a case such as: (74)

A:

Did John hire or buy that car ?

B:

He BOUGHT it.

As for Selective Focus with respect to predication scope, compare the preceding section. In a case such as the following, we clearly have Selective Focus: (75)

A:

Did John buy the car or didn't he ?

B:

He DID.

TYPOLOGY OF FOCUS

63

ïhus, explicitly disjunctive yes-no questions lead to Selective Focus. The question of whether non-disjunctive yes-no questions have to be regarded in the same way is thus related to the question of whether such questions must or must not be analyzed as implicitly disjunctive. 6.3. Replacing Focus. We speak of Replacing Focus in cases in which a specific item in the pragmatic information of the addressee is removed and replaced by another, correct item. Consider the following examples: (76)

a.

A:

John went to London.

b.

B:

No, he didn't go to LONDON, he went to NEW YORK

c.

No, he went to NEW YORK, not to LONDON

d.

No, he didn't go to LONDON

e.

No, he went to NEW YORK

It is clear that the full operation in this case involves two distinct steps: removing the incorrect information (to London) and substituting the correct information (to New YorkJ. These two steps we will call

the

Rejection and the Correction, respectively. We can then say that in (76b-e) we have the following patterns: (77)

b.

No, Rejection, Correction

c.

No, Correction, Rejection

d.

No, Rejection

e.

No, Correction

In all these patterns, no indicates that the speaker does not agree with the addressee, and believes that something is to be corrected in his pragmatic information. (76b) would seem to be the most 'logical' form for replacing information. In (76c) the speaker first gives the Correction and then, as an afterthought, the Rejection. (76d) is not complete, in a sense: the addressee is told that something is wrong with his pragmatic information, but he is not told what he is supposed to substitute for it. A typical reaction of A's in such a situation could be: (78)

Well, where DID he go then ?

64

DIK ET AL.

On the other hand, (76e), though only giying the Correction, is complete in that the Correction presupposes the Rejection, which can be reconstructed on the basis of the Correction. When a construction with Replacing consists of a Rejection

followed by

a Correction, both the rejecting and the correcting constituent carry Focus function, but the functionality of these two Pocus constituents is quite different: (79)

Rejection

Correction

rejects information presumed to be presupposed

inserts information presumed to be correct

1 announces1 Correction

fulfills announcement

is incomplete without Correction

presupposes Rejection and is thus complete without explicit Rejection

The 'open1 and 'announcing' character of the Rejection comes out quite clearly in a construction such as the following (Latin, Cicero): (80)

... cum ille dixisset non eum sua, when he patriae

sed

had-said not him by-his-own, but gloria

splendorem

of-fatherland by-glory splendid reputation

adsecutum having-achieved

'... when he had said that he had achieved his splendid reputation not through his own glory, but through the glory of his fatherland.' In this construction the Rejection non e m sua must be interpreted along the lines of non eum sua gloria splendorem adsecutum "that he had not through his own glory achieved his splendid reputation' . In the shortened form in which the Rejection is given in (80), it is not only pragmatically 'incomplete' in the sense that it announces a Correction, but it is also grammatically incomplete in the sense that non eum sua would have been uninterpretable without the following Correction. Although the Rejection is in principle incomplete without the following Correction, there are circumstances in which the content of the Correction can be fully or partially predicted on the basis of the Rejection. Consider :

TYPOLOGY OF FOCUS

(81)

The number I have in mind is not EVEN.

(82)

I have not painted my door YELLOW.

65

In (81) the addressee may rightly conclude that the correct property is 'odd', even when this is not made explicit in the Correction. This is of course due to the fact that even-odd constitutes a binary opposition. In the case of (82) the addressee may rightfully conclude that the speaker has painted his door some other colcrur. From this it follows that in constructions of the form not X but Y there must be a compatibility between X and Y in the sense that these must be drawn from the same oppositional dimension. If that dimension is a binary one, the Correction can be predicted on the basis of the Rejection, and the Rejection is thus sufficient in itself. But even if the dimension involved is not a binary one, the type of predicate to be expected in the Correction can be predicted on the basis of the Rejection. 6.4. Expanding

Focus.

By Expanding Focus we understand cases in which the Focus information is meant to be added to the antecedently given presupposed information. Consider the following example: (83)

presupposition of A: John bought x; x = coffee B:

John not only bought COFFEE, he also bought RICE

B:

Yes, but he also bought RICE

Expanding Focus contains an element of correction in this sense that B indicates that he believes A's information is incomplete with respect to the value for x. A's presupposition, however, is not incorrect in the sense that it must be (partially or wholly) replaced, as in the other types of Corrective Focus. Therefore, the label +Correctlve is less clearly applicable to Expanding Focus than to the other cases to be discussed below. The following example gives Expanding Focus with the predicate in its scope: (84)

John not only BOUGHT the house, he's also going to LIVE in it.

66

DIK ET AL.

We see no clear application of Expanding Focus to cases where the whole predication is in scope. 6.5. Restricting Focus. Restricting Focus is a type of Focus by which an antecedently given presupposed set is restricted to one or more correct values. Consider the following example: (85)

presupposition of A: John bought x; x = coffee and rice B: No, he didn't buy RICE, he only bought COFFEE B: No, he only bought COFFEE

Restricting Focus clearly corrects the presupposed information of A, in that at least one value for x is explicitly or implicitly rejected as incorrect. Restricting Focus with the predicate in scope is found in: (86)

No, John only BOUGHT the house, he is not actually going to LIVE in it.

Again, we see no application for Restricting Focus with the whole predication in scope. 6 . 6 . Parallel Focus. We speak of Parallel Focus in cases in which a speaker contrasts two pieces of information within one linguistic expression, as in: (87)

JOHN bought a BIKE, but PETER a CAR

In such a case each of the capitalized items has Focus function, but this is occasioned by the internal relations between the contrasted pairs (JOHN, BIKE) and (PETER, CAR) rather than by any specific relationship with the presumed pragmatic information of the addressee. The Focussed items in this case do not necessarily reject or correct anything in the pragmatic information of the addressee, although the Parallel Focus construction may be used in a presuppositional context, as in:

67

TYPOLOGY OF FOCUS (88)

A:

I know that John and Peter bought a Volkswagen and

B:

JOHN bought a TOYOTA, and PETER a VOLKSWAGEN

a Toyota. But who bought what ?

The point is, however, that Focus function in such a case is not occasioned by relations between the speaker's assertion and the addressee's presuppositions, but by the relationships of contrast internal to the predication proper. (88) is thus also an example of a case in which Focus is assigned to constituents which in themselves do not provide 'new' information. Parallel Focus involves strong parallelisms between two or mare pairs of constituents which pairwise belong to the same semantic category, usually have the same functional status, and often (except in the case of chiasmus, see below) take the same ordering in their respective domains. Consider the following examples from New Testament Greek: (89)

ton

Jakob

egapesa,

ton

the

Jacob

I-loved,

the

de

Esau

emisesa

Esau

I-hated

'Jacob I loved, Esau I hated' (90)

to

men

the

soma

nekron

dia

body

dead

through sin

dia

hamartian,

pneuma

zoe

spirit

live through righteousness

to

de

the

dikaiosunen

'The body is dead through sin, but the spirit is alive through righteousness' Notice that in (90) there are three pairs of parallel constituents. The opposition between the two members of the parallel construction is signaled by de in (89), and by men ... de ('on the one hand ... on the other hand') in (90). Notice that in all examples given here and below, use is made of the initial PI position for placing one pair of parallel constituents. This shows, again, that PI is a preferred position for Focus constituents. Compare in this respect the difference between: (91)

I met JOHN in PARIS, and PETER in MADRID

(92)

JOHN I met in PARIS, and PETER in MADRID

68

DIK ET AL.

Although (91) is certainly grammatical in English, (92) impresses us as more idiomatic in the case of the parallel construction. As far as the expression of constructions with Parallel Focus is concerned, asyndeton (absence of overt coordinators) seems to be used rather frequently for this purpose in quite different languages. Compare the following examples: (93)

Speiretai en astheneiai, egeiretai en dunamei (NT Greek) it-is-sown in weakness,

(94)

Tu

fortunatus,

it-is-awakened in strength

ego miser

you fortunate , I

(Latin, Plautus)

miserable

'You are fortunate, 1 am miserable' (95)

Bapak

orang besar,

father man

big

saya

orang

kecil (Indonesian)

I

man

small

'You sir, are an important person, I am but an ordinary man' A device with a strong stylistic effect, greatly appreciated in classical antiquity, was the so-called figure of chiasmus in which the members of the parallel pairs were arranged crosswise in A - B, B - A fashion. Consider the following examples from Cicero and from Varro: (96)

(97)

assequor

omnia,

I-reach

everything, if I-hasten; if I-linger, I-lose

si

propero,

lucifer

interdiu,

noctu

morning star by-day, by-night

si

cunctor, omitto

hesperus evening star

7. Formal treatment within FG. In this final section we discuss some aspects of the question how the different types of Focus configuration distinguished earlier in this paper could be formally handled within a Functional Grammar.5 Since this question is naturally quite complicated, we shall restrict ourselves to one construction in Dutch, which, however, illustrates most of the problems involved. Consider the following cases:

69

TYPOLOGY OF FOCUS (98)

Jan heeft Piet niet het BQEK gegeven, maar de PLAAT. John has

Pete not

the BOOK given

, but

the RECORD.

'John has not given Pete the BOOK, but the RECORD' (99)

Niet het BOEK heeft Jan Piet gegeven, maar de PLAAT. Not

(100)

the BOOK has

the RECORD.

Jan heeft Piet niet het BOEK, maar de PLAAT gegeven. John has

(101)

John Pete given , but

Pete not

the BOOK, but

the RECORD given.

Niet het BOEK, maar de PLAAT heeft Jan Piet gegeven Not

the BOOK, but

the RECORD has John Pete given

These are four variants of a construction which, in terms of distinctions made earlier in this paper, we could characterize as follows: this construction involves two Focus constituents (het BOEK, de PLAAT) , which are in Contrast with each other. The Contrast involves the following specific presupposition: (102)

Jan heeft Piet het boek gegeven 'John has given Pete the book1

The construction is Corrective: the speaker wishes to effect some change in the pragmatic information of the addressee; more specifically, he wishes to replace the information het boek in presupposition (102), which he presumes is contained in the addressee's pragmatic information, by the information de plaat. Construction (98) is thus an example of Replacing Corrective Focus. Replacing Focus involves two distinct acts of the speaker: the Rejection (represented by niet het BOEK} and the Correction (represented by maar de PLAAT) . Given the principles of FG, Focus function must be assigned both to het boek and to de plaat. This is necessary (i) in order to specify the special pragmatic status of the construction, (ii) in order to account for the intonational prominence of these constituents (contrastive stress), (iii) in order to be able to specify the special positional properties of these constituents. Further, because of points (i) and (iii), we also have to be able to distinguish the Focus functions as assigned to these two constituents: we saw that, although they can both be said to represent 'the most important or salient information in the predication' , they do so on different grounds:

70 het

DIK ET AL. boek

directs the addressee towards that piece of information which

is going to be replaced (we can here speak of Replacing Focus); de plaat gives the information which has to be substituted for het boek (this we can call Correcting Focus). And as we can see from (98)-(101), the constituents carrying the Replacing Focus (RF) and the Correcting Focus (CF) have different privileges of occurrence within the clause. The relevant rules can be given as follows: (R) (a)

place RF + CF in the appropriate pattern position (100);

(b)

place RF + CF in PI position (101);

(c)

place RF in pattern position, CF at the end of the clause

(d)

place RF in PI position, CF at the end of the clause (99).

(98) ;

As long as RF and CF stay together, as in (a) and (b), placement rules could globally refer to niet het boek, maar de plaat as a unit; but when they are separated in the constituent order of the clause (as in (c) and (d)), placement rules must be able to pick out the correct subpart and bring it to its place in the clause. In principle, correct results could be achieved even if only one undifferentiated Focus function is used. We would then get representations of the form: (103)

niet (het boek) maar (de plaat) Foe Poc

and placement rules would have to be formulated in such a way that they can contextually identify the correct Focus constituent within configurations such as (103) for proper treatment. However, this would require rather complex context-dependent statements for the rules involved. A simpler solution, then, is to accept RF and CF as sub-functions of the general Focus function, and to have these subfunctions properly assigned to their respective constituents. Later rules can then directly refer to these subfunctions without taking the context into account (for discussion of this question, see Watters 1979). A last question is: how do we generate configurations such as (103) ? Within a transformational framework, it would of course be attractive to derive such constructions as (98)-(101) from some underlying representation of the form:

71

TYPOLOGY OF FOCUS

(104)

Jan heeft Piet niet het boek gegeven, maar (Jan heeft Piet) de plaat (gegeven)

through some form of Conjunction Reduction. This course is excluded in FG, however, because this model does not allow one to formulate the structure-changing operations required in this treatment. Dik (1980), chapter 9, argues that constructions with term coordinations as well as so-called 'Gapping' constructions should be handled in terms of expansion rather than of reduction. The expansion analysis is formulated in terms of a very general coordination schema: (105)

a

- 2)

which, operating on some element a, expands this element into an n-ary series of coordinated elements of the same type. This coordination schema will have to be specified (a) with respect to the possible values that a can take, and (b) with respect to what sorts of overt coordinating elements may or must appear in coordinated series of different types. Adapting this approach to the present construction type, we would arrive at a description in which a given term position, representing a, could be immediately expanded into a configuration of the form niet a 1 maar a 2 . In the specific case of our examples, we would need an expansion as represented in: (106)

geven

nlet

- ( x 2»Go

^ " ^ ' ' g o

This effect can be reached by formulating (107) as a subcase of the general coordination schema: (107)

a

•* niet- a 1 maar - a 2

and by allowing a to take any term position (including its semantic function) as its possible value. Notice that this approach has the added advantage that it automatically excludes constructions with terms with nonmatching functions, such as:

72

DIK ET AL. (108)

John did not read the book with pleasure, but

with

his girl friend. John did not wait on the platform, but for three hours

(109)

(see Dik, ibid., for further discussion of these points).

The final question now is: how to assign the pragmatic functions RF and CF in the proper way to the relevant subparts of the configuartion introduced by rule (107). One possibility is to formulate pragmatic function assignment rules operating on the output of this rule, contextually specified to achieve the correct result. However, since rule

(107) is asso-

ciated with a unique distribution of these pragmatic functions

(i.e.,

W

will necessarily attach to the first part, and CF to the last part of its output), a simpler solution would be to have the rule itself attach these pragmatic functions to its output constituents. This can be done by reformulating

(110)

(107) as follows:

a

-»• niet- a 1 ^ , maar- a 2 c p

In this way, then, we define the following derivation for the constructions (98)- (101) : Start with the predicate-frame for geven

(111)

geven y

Apply schema

(112)

( x ^

'to give':

(x^^

(1.10) to the Goal argument position in (111), to produce:

gever^ ( x ^

niet- (x2> ^

maar-

, ) ^

Apply term insertion and syntactic function assignment 6 , so as to arrive

(113)

geven v

< J a n ) A g S u b j niet-(het b o e k ) ^ (Piet)

Sentences

maar-(de

plaat)^

RecObj

(98)-(101) can now be derived from this underlying predication

by applying expression rules which will properly specify both the form and the order of the constituents. RF and CF will lead to contrastive stress on boek and plaat

in any realization of (113). And placement rules corres-

TYPOLOGY OF FOCUS

73

ponding to (R) (a)-(d) formulated aboye will properly carry the constituents of (113) to their possible positions within the clause. Although this approach would have to be refined and specified in several ways in order to accomodate the different types of Focus construction distinguished in this paper, we trust that the above sketch gives the basic elements for answering the question of how such constructions could be formally handled within a Functional Grammar. NOTES 1 For a recent attempt to arrive at a more precise definition of the notion Topic, see Reinhart (1980). Topic will not be further considered in the present paper. 2 For a discussion of these constructions within the framework of FG, cf. Dik (1980), ch. 10. 3

Compare Dik (1980), ch. 10.

"Quirk et al. (1972: 939) use the term 'Contrastive Focus' both for the case where the Focus constituent 'fills in' a gap in the information of the addressee, as in: (i)

(Who was born in Swansea ?) Dylan THOMAS was (born in Swansea)

and for the case in which the Focus constituent is meant to replace some piece of information, as in: (ii)

DYLAN Thomas was born in Swansea (not EDWARD Thomas)

As is clear from our discussion we believe that these two cases must be more carefully distinguished, and that in (i) there is no contrast in the sense in which contrast is relevant to (ii). 5

For a more extensive discussion of this problem, see De Vries (1980).

6

Syntactic function assignment must be properly constrained in such a way that parallel term positions introduced by (110) are treated in parallel fashion with respect to the syntactic functions Subject and Object. REFERENCES Chomsky, N. 1972

Studies on semantics in generative grammar. The Hague: Mouton.

Dik, 1978 S.C.Functional Grammar. Amsterdam: North-Holland. 1980

Studies in Functional Grammar. London & New York: Academic Press.

Jackendoff, R.S. 1972 Semantic interpretation in generative grammar. Cambridge, Mass. : MIT Press. Kuno, S. 1972 Functional sentence perspective: a case study from Japanese and English. Linguistic Inquiry 3, 269-320.

74

DIK ET

Reinhart, T. 1980 Pragmatics and linguistics: an analysis of sentence topics. Paper, Tel-Aviv University. To appear in Philosophica. Vries, L. de 1980 Focus en Kontrast: aspekten van de verantwoording van kontrast in een Fmiktionele Grammatika. M. A. thesis, Amsterdam: Free University. Watters, J.R. 1979 Focus in Aghem: a study of its formal correlates and typology. in: Larry M. Hyman ed., Aghem grammatical structure. SCOPIL 7, 137-197. Los Angeles: UCLA.

On Theme in Functional Grammar

An Application to some Constructions in Spoken Hungarian

Casper de Groot University of Amsterdam

0. Introduction In Functional Grammar one of the three types of functions are the pragmatic functions. These functions mark the informational status of constituents within the wider communicative setting in which they are used. Four pragmatic functions are distinguished, two external (Theme and Tail) and two internal (Topic and Focus) to the predication proper. Theme, Topic and Focus are defined as follows: Theme: The Theme specifies the universe of discourse with respect to which the subsequent predication is presented as relevant. Topic: The Topic presents the entity "about" which the predication predicates something in a given setting. Focus: The Focus presents what is relatively the most important or salient information in the given setting. I illustrate here the functions Theme and Topic: (1)

That trunk, put it in the car

(2)

As for Paris, the Eiffel Tower is really spectacular

Themes are that trunk and as for Paris; Topic constituents are it and the Eiffel Tower. The Theme constituents present the entity that the subsequent predications are going to bear upon, while the Topic constituents present the entity about which the predication predicates. For certain sentences beginning with a Theme one can entertain the idea that the Theme is extracted from the predication following it. The original

76

DE GROOT

position of the constituent in the predication will be marked by a pronoun, for instance sentence (1). This analysis does not account for the type of Theme-predication shown in (2). In that case the Theme cannot be thought to originate in the antecedent predication, thus it cannot be extracted from that predication. That is why Dik (1978, chapter 6) proposes the following general schema:

(3)

( 1*Theme . •

Predication

The Theme is independently generated, the predication is chosen in such a way that it does not conflict with the Theme. There is a pragmatic relation of relevance. When there is no such relation, the sentence does not make sense. For instance: (4)

*As for Paris, the Empire State Building is really spectacular

One of the arguments Dik adds in favour of this interpretation is based on the consideration that one would -xpect that a Theme, when it is extracted, will have the semantic and syntactic functions, and the corresponding markers, of its source constituent within the predication. In several languages a left-dislocated Theme constituent occurs in a sort of "absolute form", characterised by the most unmarked case (typically the nominative). There are languages, however, in which the Theme can occur in a more specific case. This might be taken as an argument in favour of an extraction analysis. Since Hungarian has both constructions with a Theme in the absolute form1 and in a specific case, an investigation of Theme in Hungarian might give information relevant to the descriptive model of Theme-predication organisation. I shall concentrate on the type of construction with a Theme as given in sentence (1), because that construction can be problematic for the general schema. Sentences which might contain a Theme I have had judged by ten informants. In many cases it proved to be difficult to say whether they were grammatically correct or not. "It is used often" or "it sounds acceptable" were common reactions. If I got six or more positive reactions I took the sentence to be correct, four or five earned a question-mark and three or less I considered to be bad, ungrammatical. For a better understanding of the notion Theme in discussing the Hunga-

THEME IN HUNGARIAN

77

rian language it is necessary to summarize briefly the status of Topic in that language. 1. Topic in Hungarian Hungarian is a so-called free word order language which cannot be easily described in terms of Subject, Object and Verb. Although Hungarian is reputed to be a SOV language the idea that it should be considered as a TVX or rather a TFVX language is gaining more and more ground. That is, if there is a Focus it immediately precedes the verb; the constituent(s) preceding the Focus is (are) Topic. According to £. Kiss (1980) the word order within the Topic is free. Szabolcsi (1980) adds that there are obligatory Topics in the following cases:

(5)

Uulatni Peter mulatott To-enjoy Peter enjoyed 'Enjoy himself, Peter did'2

(6)

Versnek verset irt Poem, poem wrote-he dat acc 'A poem, he wrote one'

The constituents in italics are the obligatory Topics. Thus if they occur in a sentence, they must be in Topic position. Because of its intonation contour Szabolcsi speaks about emphatic Topic. The functional pattern for Hungarian is: (7)

P2

,

Pt

PQ

V

X

,

P3

The schema does not contain positions for Subj and Obj because these functions are not assigned in Hungarian (See De Groot (1981)). The special positions correspond to the four pragmatic functions. P^ is the position for Theme, Pj for Topic, P^ for Focus and P^ for Tail. The special property of Pj is that it can accommodate more than one constituent. When Topic has been assigned to an embedded predication a dummy element representing the embedded predication will appear in P^ of the main clause. For instance:

78

DE GROOT (8)

nem_ hisz, (£n)„ (ismer., (Mari)„m (Peter)„ )_ „ roc V Po V 0Top Go GoTop not

believe I

know

Mary

Peter

This predication can be expressed in (9): (9)

ftzt nem„ hiszem, hogy Mari ismeri Petert Top Foe Top P V P P 1 1 That not0 believe-I that Mary knows Peter 'I do not believe, that Mary knows Peter'

Constituents with Topic function within the embedded predication can optionaly be displaced to the Pj of the higher predication when the embedded predication has Topic function as well. As we can see Mari in (8) has Topic function and it is within an embedded predication with Topic function. Thus Mari can be displaced: (10)

Mari nem hiszem, hogy ismeri Petert p p v i o 'I do not believe, that Mary knows Peter1

Notice that Mari replaces the dummy representation azt. (See for displacement resp. raising in Hungarian De Groot (1981) and fi. Kiss (1980)). The word order within the Topic seems to be free. I would like to make the following remark. Let us consider the sentences (11) and (12) which should have the same meaning since there is only a different word order within the Topic. (11)

I IkônyvetI„ adott 1 Péter MarinakI„ 'Top 1 1 'Foe Peter Mary-to book gave ' Peter gave Mary a book *

(12)

|Marinak Péter|Top |kônyvet|poc adott

If the intonation of (11) and (12) is unmarked, the first constituent has middle pitch, the second constituent has a more neutral intonation while the third constituent, the Focus, receives the main stress. The interpretation of the unmarked sentences is for (11): Some people gave Mary something; Peter gave a book. For (12): Peter gave some people something; to Mary he gave a

THEME IN HUNGARIAN

79

a book. With these interpretations the constituents Pétez and Marinak fulfil exactly the definition of Topic. It becomes even clearer when we want to express that Peter gave a book to Mary and an apple to Judy. Only one word order within the Topic is allowed3: (13)

|Péter Marinak I IkônyvetL adott, 1IjutkânakL I aimâtL 'Top ' 'Foe 'Top ' 'Foe Peter Mary-to

book

gave

Judy-to

apple

'Peter gave Mary a book and Judy an apple' The sentences (11)-(13) show us that the word order in the part of the sentence preceding the Focus is not always free and they strongly indicate that Hungarian prefers one Topic constituent, in initial position of the sentence. The sentences (11) and (12) have the same interpretation when they are pronounced with the same stress on Péter resp. on Marinak. When the second constituent, however, is pronounced with more stress than the first one, most informants consider the sentence as ungrammatical. The sentence intonation does not necessarily have to be the intonation of (11). In contrasts, middle pitch can be given to any constituent or any combination of constituents. Even all constituents in Topic position can have the same stress. This may for instance be the case when there is no Focus in the sentence.11 2. Theme in Hunqarian 2.1. Theme and main clause Constructions in spoken Hungarian which can be regarded as being constructions with a Theme can be devided into three types. Type 1: Theme in the absolute form (14)

A pék -- neki nem adtam semmit The baker , _ him, . not gave-I nothing absf dat acc 'The baker, I did not give him anything'

In (14) there is a clear pause between pék and neki. I matk this pause with the diacritic sign — . The personal pronoun is preferred to the demonstrative pronoun. The latter, however, can be used in the case when a speaker wants to express that there is a great distance (social) between him and the baker:

80

DE GROOT (15)

A p£k —

annak nem adtam semmit

The baker . _ that. , not gave-I nothing absf dat acc Sentence (15) is very exceptional. Type 2: Theme in a case a. constructions without a pause (16)

A pSknek annak nem adtam semmit The baker. that. . not gave-I nothing dat dat acc

In sentence (16) the use of the demonstrative pronoun seems to be obligatory. This sentence with a personal pronoun is ungrammatical. Between the Theme peknek 'baker^^' and the Topic annak 'that^^' there is no pause longer than normally would be the case between two constituents.5 b. constructions with a pause (17)

A peknek —

neki nem adtam semmit

The baker. him^ . not gave-I nothing dat dat acc Notice that the personal pronoun is used here. Sentence (17), however, will be used as an answer to (18): (18)

Mit adtal a peknek? What gave-you the baker. ^ acc dat 'What did you give the baker'

When (17) is an answer to (18), the Theme in (17) does not anticipate the predication that follows it, but repeats a part of the question. That is why a Theme can even have a case which does not correspond to the function of its coreferential element in the predication following it. The next dialoque shows this:

THEME IN HUNGARIAN (19)

A:

81

Hovâ sietsz? 1

Where are you running to? '

B:

Szeretnék mondani valamit Péternek.

A:

Péternek? Azt semmi sem érdekli.

'I would like to say something to Peter.' Peter, . that nothing not interests dat acc nom 'To Peter? Nothing interests him.' Notice that the form azt "that ' can be used here. Azt to express 'him' acc only occurs in constructions with a Theme. The personal pronoun St 'him^^' can be used as well. Type 3: Infinitive and dative-construction as Theme (5)

Mulatni Péter mulatott To-enjoy Peter enjoyed 'Enjoy himself, Peter did'

The infinitive behaves like a Topic in this respect that it prefers the initial position in the sentence. In other positions preceding the Focus it can occur as well: (20)

Péter mulatni mulatott

With respect to the informational status, however, the infinitive behaves like a Theme. Szabolcsi even compares it with: (21)

Marit azt nem szeretem Mary

that not like-I acc acc "Mary, I do not like her'

She also compares it with another construction, which can be considered as having a Theme: (22)

a.

Elmenni Péter ment el To-go away Peter went away 'Go away, Peter did'

b.

El —

Péter ment

Away

Peter went

82

DE GROOT

(22)a and (22)b have the same meaning. Parallel to the construction with an infinitive (for verbs), there is a construction with a dative, which is used in relation with a substantive or an adjective: (23)

Versnek verset irt, de jonak nem volt j6 Poemn poem wrote he but good. not was good dat acc dat nom 'A poem, he wrote one, but good, it was not'

2.2. Theme and subordinate clause In subordinate clauses there are no other constructions which can be regarded as constructions having a Theme than those mentioned in section 2.1. above. I summarize here the forms in which they appear. Type 1 and 2b: The Theme only occurs in P^ of the main clause: Type 2a and 3: The Theme constituents have within the subordinate clause the same possibilities as those of the main clause. They can also occur in the initial position of the main clause. The pronoun of Type 2a always follows the Theme immediately. Thus a Theme of a subordinate clause is either in the initial position of the main clause or after the conjunction; a Theme in the subordinate clause is never followed by a pause. Some examples: (24)

Janos —

azt nem akarom, hogy neki puszit adjal

John

that not want-I that him. kiss give-you absf acc dat acc 'John, I do not want, that you give him a kiss'

(25)

Azt nem akarom, hogy Janosnak annak puszit adjSl That

not want-I that John. . that. ^ kiss give-you acc dat dat acc 'I do not want, that John you give a kiss'

(26)

Janosnak annak nem akarom, hogy puszit adjal John. . that. . not want-I that kiss give-you dat dat acc 'John, I do not want, that you give a kiss'

(27)

Azt nem hiszem, hogy mulatni Peter mulatott That

not believe-I that to-enjoy Peter enjoyed acc nom 'I do not believe, that enjoy himself, Peter did'

(28)

Mulatni azt nem hiszem, hogy Peter mulatott enjoyed not believe-I that Peter acc nom 'Enjoy himself, I do not believe, that Peter did1

To-enjoy that

THEME IN HUNGARIAN

83

In sentence (25) the subordinate clause is represented by the dummy azt (see section 1 above). In sentence (26) the dummy has disappeared. Does it follow that annak in (26) replaced the dummy representation? Inclusion of azt in this sentence, either in Topic position or neutral postverbal position makes this sentence ungrammatical. A sentence, however, with azt in Topic position without a pronoun referring to the Theme is grammatical: (29)

Jánosnak azt nem akarom, hogy puszit adjál

These facts can be interpreted as follows. In (29) Topic has been assigned to the embedded predication and is therefore represented by the dummy in Topic position. Optionally there can be a pause after Jánosnak. If Focus would have been assigned to the embedded predication (30) would be the output: (30)

Jánosnak annak azt akarom, hogy puszit adjál John. ^ that, that want-I that kiss give-you dat dat acc acc 1 'To John, I want, that you give a kiss

We can conclude that if the Theme Jánosnak in (26) would be extracted from the subordinate clause, it is not by way of the dummy representation of that clause. It is not displaced in the way Mari is displaced in (10), where Mari replaces the dummy: (10)

Mari nem hiszem, hogy ismeri Pétert Mary

nom

not believe-I that knows Peter

acc

'I do not believe that Mary knows Peter' An example of a sentence with a Theme and a displaced Focus: (31)

Pista 1Ima este|„ szeretném, ha felhivna 'Foe Steve to-night I-would-like if calls-he nom 'Steve, I would like him to call to-night'

One can doubt whether the first constituent in (31) is a Theme or rather a Topic. First É. Kiss and then De Groot have argued that displacement of both a Topic and a Focus is not possible in Hungarian. If Pista is a Topic, sentence (31) is a counter example. There are, however, two arguments in

«4

DE

GROOT

favour of the idea that Pista is a Theme. A pause can be added after this constituent and Pista is expressed in the nominative case according to the function it has in the subordinate clause. If it would have been displaced as a Topic it would have come into the domain of the verb of the higher predication and therefore expressed by the accusative case. For instance: (32)

Pistât szeretném, ha ma este felhivna Steve I-would-like if to-night calls-he acc 'I would like Steve to call to-night'

Morphological readjustment of Pista in (31) leads to ungrammaticality: (33)

"Pistât ma este szeretném, ha felhivna Steve to-night I-would-like if calls-he acc

3. Discussion Let us see what information the examples give us in relation to the two approaches for the description of a Theme-predication organisation. Approach I: the Theme is generated independently of the predication; Approach II: the Theme is extracted out of the predication following it. The Theme constructions of type 1 and 2b can without doubt be treated by Approach I. The types 2a and 3 are problematic. I start with type 2a in the subordinate clause. The examples (26) and (30) show that the Theme is not displaced out of the subordinate clause by way of the dummy representation of that clause. The sentences (31)—(33) illustrate that the Theme of the subordinate clause does not come into the domain of the higher predication. These two facts could be interpreted in favour of Approach I if not a third fact, illustrated by the examples (29) and (31), showed that the Theme remains in the domain of the embedded predication. The Recipient function in (29) is only expressed by means of the dative case of the Theme. There is no pronominal element in the subordinate clause referring to the Theme. As for type 3, the Theme constituents are free within the part of the clause preceding the Focus, thus they are not bound to P^. This leads us to the conclusion that constructions with a Theme of type 2a and 3 cannot be treated by Approach I. Approach II, however, does not satisfy either. This approach wants to extract, but there is no reason to assume that Jânosnak in (29) is taken out of the predication. On the contrary,

THEME IN HUNGARIAN

g5

it bears the marker of its semantic function and there is no pronoun that marks its "empty" place in the predication. Sentence (29) is even ungrammatical with a pronoun: (34)

"Janosnak azt nem akarom, hogy neki/annak puszit adjSl John, . that not want-I that him/that. ^ kiss give-you J dat acc dat acc 3

I cannot but conclude that the Theme constituents mentioned are within the embedded predication and can be displaced in one swoop to the initial position of the clause. However, this conclusion contradicts

our original definition, because

Theme is a pragmatic function external to the predication proper (in our case it is not external). Secondly, the initial position of the clause, P^, is reserved for the Topic (a Theme in P^?). This is a reason to assume that we are not dealing with Theme but with Topic. On the other hand their informational status agrees with a Theme and not with a Topic and they have special markers (pronoun, pause, infinitive, dative). A pause in (29) can, since it is caused by the lack of a pronoun, be considered as a marker to indicate that the first constituent does not belong to the main clause. What about type 2a in the main clause? Let us consider (16). Theme has a case and there is no pause between the Theme and the Topic. It is difficult to ascertain whether PSknek is within or outside the predication. The case supports the idea that the Theme originates in the antecedent predication; the absence of a pause can be interpreted in favour of the idea that the Theme constituent is within the predication. The situation at the moment is that Approach I is satisfactory only for type 1 and 2b. Approach II does not satisfy at all. The status of the Theme of type 2a and 3 seems to be in between Theme and Topic. In other words, the types of Theme

do not fit the definition of the pragmatic function. We can

search for other approaches, but I think it is better to reconsider the definition of the pragmatic function Theme. With respect to the informational status of the Theme constituents of type 2a and 3 there is no reason to doubt the correctness of the definition. The problem is that the pragmatic function Theme is said to be external to the predication. By no means it can be proved that to these types of Theme subsequent predications are presented as relevant.

86

DE GROOT Let us see what the consequences are if we leave the notion "external",

which is in my opinion rather a structural than a pragmatic notion. It seems to me that two questions will arise: first, do we have to distinguish between Themes within and outside a predication and secondly, what is the relation between the "within" Theme and the Topic? I have already mentioned that the "within" Theme differs from a Topic (according to its definition) by a difference in pragmatics. The Theme differs from other constituents in the part of the sentence preceding the Focus in that it is marked by a pronoun, dative etc. and because it always has the strongest stress within that part of the sentence. If we assume that Topic is the only constituent with middle pitch (see section 1 above) the "within" Theme fills this part. It can be considered as a special Topic, as is done by Szabolcsi. When we consider all constituents

within the initial part of the sentence as Topic constituents, the

"within" Theme excludes all Topic constituents with middle pitch. There is, however, still another difference between the "within" Theme and a Topic, or rather between sentences with or without any kind of Theme: a sentence with a Theme must have a Focus.6 All examples with a Theme given in this paper are ungrammatical without a Focus. This fact points in my opinion to the solution of the problem. It becomes clear why a construction with a Theme can be used: its purpose is to emphasize the Focus. Now the difference between the Theme of type 1 and 2b on the one hand and the Theme of the type 2a and 3 on the other hand is that 2a and 3 use this special property of Theme on purpose. A Theme of type 1 and 2b is followed by a pause, a hesitation, the speaker mentions a subject but he does not necessarily have to know how he shall continue. The Themes of type 2a and 3 are used in another way. The speaker knows exactly what he is going to say, there is no hesitation. He knows which semantic function the Theme will have in the predication and adds the appropriate case. Maybe the recurrent use of expressions with a Theme-predication organisation have led to partial grammaticalization. Thus, some Themes were incorporated in the predication. This may also explain, that the "within" Theme is not always bound to the first position of the clause; it cannot occur after the Focus of course. 4. Conclusion The application of the pragmatic function Theme to Hungarian is not without problems. The general schema as a descriptive model for constructions with a Theme,

THEME IN HUNGARIAN (3)

(x,)„. 1 Theme

87 , Predication

is only useful for those cases in which the Theme does not have a case as an expression of a semantic function predicted by the predication following it. For other constructions with a marked Theme there are no reasons to assume that the Theme constituents are outside the predication. They can be considered as special Themes or Topics with the function to emphasize the Focus. An alternative extraction approach for a description of Theme is therefore rejected. A description of Hungarian necessitates a destinction between the functions Topic and Theme. The definition of the latter, however, should be reformulated with respect to the notion "external" and it is advisable to take into consideration the relation between Theme and Focus, because, as it seems to me, the observations concerning Hungarian are equally valid for other languages. ACKNOWLEDGEMENTS This study has been made possible by the support of the Institute of Cultural Relations (Kulturâlis Kapcsolatok Intézete) in Budapest. I would like to thank the Institute of Linguistics of the Hungarian Academy of Sciences for the hospitality it gave me during the period I was working on this project. I am especially endebted to my informants for their patience and constant readiness to give thoughtful answers to my questions. NOTES I I use the term absolute form (absf) and not nominative or absolutive. The nominative I consider to be the case expressing e.g. the Agent in Hungarian. Using the absolutive might be confusing since it normally indicates a case in ergative languages. 2NO effort has been made to translate the Hungarian examples into English, because it is impossible to render these constructions in that language. I have merely tried to make their meaning understandable. 3 In the example I give here the constituents Marinak 'to-Mary' and Jutkânak 'to-Judy' are not to be considered as contrasts. The constituents meant have a neutral or reduced Topic intonation. Sentence (13) can of course have an interpretation with a contrast. The contrast property can be considered as Focus, which might explain the position of the contrast constituents here: they immediately precede the Focus.

''For instance the answer to the question: What happened this afternoon? (i)

Péter Marinak kônyvet adott Peter to-Mary book gave

88

DE GROOT

It can be argued that Focus has not been assigned to konyv 'book', or konyv can be considered as an "unmarked Focus". The answer to: What did Peter give to Mary this afternoon? contains a (marked) Focus: (ii)

Peter Marinak kSnyvet adott Foe

I do not want to discuss Focus here. For discussion of this pragmatic function see the articles on Focus in this volume by Jan R. de Jong and a Workgroup Functional Grammar. s

Annak •that

• cannot be used attributively. The Hungarian egression

for 'that baker. ^ 1 is annak a p6knek 'that. ^ the baker. ^ 1 . dat dat dat 6

See note 4. Like for instance the Focus in (ii).

REFERENCES DIK, S.C. 1978 Functional Grammar. North-Holland Linguistic Series 37. Amsterdam: North-Holland. GROOT, C. de 1981 Sentence-intertwining in Hungarian. To appear in Bolkestein, A. Machtelt, et alii, Predication and Expression in Functional Grammar. KISS, K. 1980 Structural Relations in Hungarian, a "free" word order language. To appear in Linguistic Inquiry. SZABOLCSI, A. 1980 Az aktucilis mondattagolSs szemantikajShoz (On the semantics of functional sentence-prespective) . To appear in NyelvtudomSnyi KBzlemSnyek. SZALAMIN, E. 1978 A mai magyar beszSlt nyelvraondattan^nakk£rdesehez (Questions concerning the grammar of modern spoken Hungarian). Magyar Nyelv 74, 293-303. ZOLNAI, G. 1926 MondatStszovodes (Sentence-intertwining). irtekezesek a Nyelv- £s SzSptudomanyok Korebol 23/8

On the Treatment of Focus Phenomena in Functional Grammar Jan de Jong University of Amsterdam

0. Introduction and summary 1 This paper is a report of an investigation into focus phenomena. Special attention has been paid to the expression of different focus types in some African languages. I did not collect any new data on these languages but simply made use of existing publications. In the case of Efik Cook's (1976) rendering of the Efik focus system has been supplemented by some additional information from the author. It is well known that different sentences used in different situations may vary considerably as to the type of focus they contain (a number of these types have been distinguished in the paper by Dik et al. (this volume)). The intention of this paper is to show that at least in the languages under consideration there are two important dimensions of variation with regard to focus type. These dimensions are scope, i.e. whether a greater or smaller part of a sentence (or its underlying predication) is in focus and contrast, i.e. whether the sentence contradicts some previously uttered sentence or some presupposition. Evidence for the relevance of the dimension scope is brought forward by Oomen (1978) on Rendille and Givon (1975) on some Bantu languages. In these languages morphological means are used to indicate whether the 'salient' or 'new' information is contained within an NP or within the VP (or the sentence as a whole). To put it another way: whether or not some term has been singled out to receive focus. Watters (1979), in a study of focus phenomena in Aghem, also provides some evidence on the relevance of the scope dimension. Moreover, in this language there exists a formal distinction between non-contrastive sentences and contrastive sentences. It is in the Efik language (dealt with in Cook (1976)), however, that the dimensions of scope and contrast combine most clearly to form a 'focus system' . In this language different types of focus are marked by means of

90

DE JONG

verbal morphology. These types can be shown to differ among themselves along the dimensions of scope and contrast, not only with regard to their function, but also each dimension appears to have a separate formal correlate: presence or absence of a verbal morpheme. Before discussing the details of focus marking in Efik, Aghem and Rendille (in sections 2, 3 and 4 respectively) in section 1 I will explain one possible way to deal with the notions of scope and constrast in Functional Grammar. 1. Theoretical

preliminaries

In Dik (1978) focus was defined as the function that presents "the relatively most important or salient information". Above I suggested that this most salient information may be presented in different ways: it may be of greater or smaller scope and it may, or may not be contrastive. In order to see how the concepts of scope and contrast fit into Functional Grammar we will first have to analyse the notion of focus itself. I will try to answer the following questions: what kind of linguistic units may be said to carry information ? And: what is it that makes this information salient ? 1.1. The notion

of

'information'

If we apply the first question: "what kind of structures may be said to carry information" to logical predications, such as e.g. P(a,b), I think one would agree with the thesis that it is the predication as a whole, rather than any of its constituent parts (the arguments a or b or the predicate P itself) that has to be considered as a unit of information. In other words, a set of unconnected predicates and terms could not count as "information". However, if we admit that there is a basic analogy between predications and sentences in natural languages as well as between terms in predicate logic and noun phrases in natural languages, how are we to interpret

state-

ments made by linguists such as "in sentence p the object NP represents the new information"? Could an NP indeed by itself contain information ? For the sake of illustration let us imagine that someone would make out of the blue the following utterance: (1)

John

We would not expect the knowledge of the Addressee to be augmented by such a remark, i.e. it would not convey any information. But if the same expression would state the answer to the question

TREATMENT OF FOCUS (2)

91

Whom did Mary visit?

the knowledge of the Addressee would have increased, as a result of the conjunction of one meaningful element, the NP John, with another, the predicate visit. (That there is a relation between the two is in some languages made explicit by case-marking on the noun). The same facts could also be stated thus: (1) as an answer to (2) conveys the following piece of (new) information: "John participates (as a Goal) in the state of affairs visit (x,) (x.) ". i Ag j Go In Functional Grammar we might represent the structures underlying (2) and (1) (as an answer to (2)) as follows (stated here, as in the rest of this paper, in a simplified way, omitting term and predicate operators): (3)

visit

(4)

(x. : John (x .)) , . ] ] GoOb]

V

(x. : Mary(x.)) (Q x.)_ 1 i AgSub] ] GoOb]

The piece of information conveyed by (1) in the context of (2) can be paraphrased as "the Goal in the state of affairs visit (x.)_ (x.)„ „ a. Ag 3 Go is John. If we want to find some structure that contains just this piece of information we will have to find a predication with John as a (nominal) predicate and with an argument expression that refers to the Goal in the state of affairs visit(x.) i Ag (x.) j Go. The best candidate seems to be the predication John (x.) within term (4). It contains the nominal predicate John and the variable x^ links it to the Goal of the predicate visit in (3). Note that (1) uttered in isolation does not convey information because the variable in John (x) is in that case not 'bound' by a higher predicate. It should be noticed that there are cases where the identification of new information with predications is much easier. If a sentence such as "John visits Mary" is an answer to the question "What happened?" we could simply say that the whole of the predication underlying the answer consists of new information.

92

DE JONG I shall assume therefore that for every sentence the new information

that it conveys can be identified with predications in underlying structure and not, as it is often claimed to be the case, with constituents in surface structure. In this paper I will investigate the consequences of this assumption, both theoretical and with regard to the description of a number of languages, although I am aware that there are other possibilities to describe the relationship between the notion of information and sentence structure. First let us apply the view on the notion of information outlined above to the following exchanges. The answers are often said to contain in each instance differently distributed 'old' and 'new information', though the 'prepositional content' in each case is the same. (5)

Q: What happened ? A: The cat broke the vase.

(6)

Q: What did the cat do ? A: The cat broke the vase.

(7)

Q: What did the cat break ? A: The cat broke the vase.

Those who wish to identify information with constituents would say that in (5) the entire sentence contains new information, in (6) only the VP and in (7) only the object NP. In view of what was assumed above our representation of the facts would be as follows. Given the underlying predication (8) for each of the sentences (5)-(7): (8) break^ Cx± : cat(x. ) ) A g S u b j (which has to contain the syntactic variable x ) bu expanding x n n T-RCn:

to a suitable relative pronoun.

\u [a1(u ) A A' ] n n

S-TI^: Term Insertion 3: Insert term a in position 3 of a verbal predicate frame in which positions 1,2,3 are empty. T-TI3:

XZj Az2 [oKPtej) (a2))]

5. Discussion In this section I will make some general remarks about the relation between functional grammar and Montague grammar. The first remark I want to make, concerns the difference between the version of functional grammar as it figures in the previous sections and its more standard form. I did not make use of all the information present in a predicate frame. Semantic functions (e.g. Goal), features (e.g. human), and variables (the x^'s in frames) were not taken into consideration. Semantic functions express semantical details, for example that a certain position of a frame will contain the goal, and another one the locative satellite. The kind of semantics currently used in Montague grammar is not yet that refined enabling to take these differences into account. In order to do so the logic and the model probably would have to be enriched. We had a limited aim (see section 1), for which it was not necessary to bother about these details. For larger fragments of English even reaching this limited aim is even difficult enough. The features provide the information what kind of terms can be put into which positions. Certainly, Montague grammar will have to deal with such selection restrictions as well, and the information contained in these features has to be incorporated somehow. It has both been proposed to do so in the semantics by means of partially defined functions (Cooper (1975), Waldo (1979)), and to to so in syntax by means of subcategorization (Janssen (1980b)). Either one of these methods could be followed here. The role of the variables in functional grammar is taken over, in a new

292

JANSSEN

way, by the quantification rules. Since functional grammar was not designed together with an explicit formal semantics, it is not too surprising that some details had to be changed. A parallel between functional grammar and Montague grammar is the tendency to eliminate transformations. In transformational grammar it was often their semantic relation that motivated a transformational relation between sentences. The basic difference between Montague grammar and transformational grammar is that Montague grammar has an explicit semantic component. Therefore Montague grammar has the possibility to formalize such relations among sentences at the level of semantics. Sentences may be produced independently syntactically, while their semantic relations can still be dealt with, but now in the semantic component. Thomason (1976) treats passive sentences as produced independently of the active sentences, and accounts for the relation between them in semantics. Gazdar (1980) propagates a context-free - i.e. a non-transformational - approach to syntax. Many of his rules are possible because his grammar has an explicit semantic component analogous to that of Montague grammar. Several governed transformations are treated non-transformationally in Dowty (1979), and Bartsch (1979), again by having the semantics do the job. So there is a general tendency among Montague grammarians to avoid transformations. Not using transformations is one of the aims of functional grammar. It is to be expected that many of the results in the field of Montague grammar can be applied in functional grammar as well. The following is an example. How could one allow for the possibility that both (62) and (63) are produced? (62)

John serves the cake to Mazy.

(63)

John serves Mary.

In transformational grammars (63) usually is obtained from a source like (62) by means of a transformation called 'object deletion1. Dowty (1979) does not use a deletion rule. Instead he proposes a rule which makes out of the three place verb serve, a two place verb. This rule has no visible syntactic effect: the verb obtains the status of a two-place verb without changing its form. Semantically, this rule has the effect of introducing some existential quantifier. The same idea can be applied in functional grammar: deleting nothing, just change the status of the verbal frame (e.g.

MCKTTAOUE GRAMMAR AND FG

293

by inserting a dummy term or removing the third position). An important difference between functional grammar and Montague grammars as they are usually described in the literature is that, whereas a Montague grammar produces rather natural phrases, a functional grammar produces abstract structures in the non-final stages. This has as consequence that in a Montague grammar two only superficially different phrases such as (64) and (65) are to be considered as different syntactic objects. (64)

give Mary a book

(65)

give a book to Mary

In the more abstract approach of FG they correspond to the same predication, leaving it to the expression rules to make the difference. Another example of this phenomenon is the active/passive distinction. A functional grammar can have somevhat simpler rules since superficial differences have not to be accounted for in the stage where the meaning is determined. An advantage of the abstract structures becomes apparent if we consider syntactic variables. A Montague grammar produces strings of words, so a variable has to be represented as a word. They are represented as a male pronoun to which an index is attached: by he^ and by hiro^. This representation has the disadvantage that an arbitrary choice is made: the variables look 'male' and 'singular'. By later rules these

decisions often have to

be withdrawn. Furthermore one is tempted to consider the variables syntactically as pronouns, which they are not (Janssen (1980a)). Functional grammar has the advantage that syntactic variables can be represented as what they really are: abstract elements. No premature decisions have to be taken. These syntactic advantages have their price. The frames of a functional grammar are rather distinct from real sentences. One might ask whether the available abstract information is the kind of information one needs for producing correct sentences. In any case, this is not the kind of information which traditionally is considered to be of the required kind: a constituent structure. Kwee Tjoe Liong has developed a computer program implementing Dik (1981). Consequently he had to be very explicit, in particular in the formulation of the expression rules. There turned out (Kwee (1979) and Kwee (this volume)), to be a lot of unsolved problems in formulating rules producing sentences out of frames. The fact that the available structural information is rather abstract, brought me into

294

JANSSEN

problems when formulating of S-Qn. I will turn to these problems now. Pronominal reference is always possible from right to left: a suitable pronoun can always refer to a term occurring earlier in the sentence. In certain circumstances pronominal reference also is possible in the other direction. But it is rather difficult to characterize these situations. So in case there are two positions in a sentence, one to be filled with a term, and the other with a pronoun referring to that term, then a safe strategy is to put the term in the leftmost position. This safe strategy is followed in PTQ in the formulation of the rule corresponding to S-Qn. When formulating S-Qn in FG this strategy cannot be applied since it is not possible to say in advance in which linear sequence the terms will occur in the sentence which is finally obtained. The assignment of syntactic functions may have as consequence that anaphoric variables are raised from deep embeddings and the expression rules can make that the corresponding anaphoric pronouns occur earlier in the sentence than the terms they refer to. I do not know how to formulate S-Qn in such a way that only correct reference patterns result. It is not attractive to try to solve this problem by allowing 'unrestricted1 expansions in S-Qn, and to put restrictions on the rules assigning syntactic functions and the expression rules, which would require that pronouns should occur to the right of the term they refer to. For this would introduce a filter into the grammar since there is no guarantee that from a given frame with several pronouns a sentence can be produced which obeys the requirements. Such a filter is undesirable from the point of view of functional grammar itself. (See section 2.1 in Dik (1981)). The safe strategy followed in Montague grammar does not guarantee that all the sentences which are produced are correct. Although a pronoun may always refer backwards, sometimes it has to be a personal pronoun, and sometimes a Reflexive pronoun. Since the rules of PTQ do not treat reflexives, the sentence John sees him is produced, with the meaning that John sees himself. It is difficult to characterize the positions in which a reflexive is required. It is even more difficult to characterize the positions where forward referring is allowed. In transformational grammars there has been some progress in characterizing these two kinds of positions. These characterisations use structural properties (Reinhart (1979)), such as the notion C-command. In a PTQ-style grammar plain strings are produced without any structure; here a correct treatment of pronouns seems impossible. But one might enrich these strings with markers indicating the relevant structural

MONTAGUE GPAMMAR AND FG

295

information (Bennett (1976)). It would be more straightforward to have the grammar produce trees (or equivalently labelled bracketings) instead of plain strings. If these trees are of the same as the trees used in transformational grammars, then all insights from that field can be used in Montague grammar as well. The idea of such a grammar is due to Partee (1973), and has been worked out in Partee (1979a,b) and Bach (1979). In a functional grammar structures are produced of a completely different kind, so the insights from transformational grammars cannot easily be adopted. This leaves the challenge for the functional grammarians to characterize in the terminology of functional grammar the configurations where reflexives or forward referring pronouns may occur. This seems to me a very difficult task, since a frame does not give much information about the surface forms it can take. Let me summarize my views on the relation between functional grammar and Montague grammar. There is no problem in considering functional grammar to be an instance of the framework of Universal Grammar (Montague (1970)). This has the consequence that meanings are not considered as to be determined by frames as such, but only given by their derivational histories. Derivational histories as they are designed in Montague grammar, can be imitated in functional grammar. Sometimes a simplification is possible because of the rather abstract structures produced in a functional grammar. This same fact, however, leaves certain syntactic problems to be solved. Of course, such an imitation may have the effect that certain production processes of functional grammar have to be changed; an example is the relative clause construction given in section 4 (for a discussion see Partee (1973) and Janssen (1980a)). In any case the program for dealing with semantics in a functional grammar is established. One has to provide for formulas for the basic frames, and for each rule a (polynomially defined) operation on formulas.

Acknowledgements I wish to thank Simon Dik, Kwee Tjoe Liong, and especially Jeroen Groenendijk for their comments on a previous version of this paper.

296

JANSSEN

REFERENCES BACH, E. 1979 Montague grammar and classical transformational grammar. In Davis & Mithun 1979, pp.3-49. BARTSCH, R., 1973 The semantics and syntax of number and numbers. In J.P. Kimball (ed.) Syntax and semantics. Volume 2. Academic Press, New York, 1973, pp. 51-93. 1978

Infinitives and the control problem in categorial grammar. In Theoretical Linguistics 5, 217-250.

BENNETT, M. 1976 A variation and extension of a Montague fragment of English. In Partee 1976, pp. 119-163. COOPER, R.H. 1975 Montague's semantic theory and transformational syntax. Dissertation, Univ. of Massachusetts, Amherst. Published by: Xerox University Microfilms. DAVIS, S. & M. MITHUN (eds.) 1979 Linguistics, philosophy and Montague grammar. (Proc. conf. Albany 1977). Univ. of Texas Press, Austin, Texas. DIK, S.C. 1981 Seventeen Sentences: basic principles and application of functional grammar. In Proc. Conf. on current approaches to syntax, Milwaukee (Wise.) 1979, Syntax and Semantics 13, Academic Press, to appear. DOWTY, D. 1978 Governed transformations as lexical rules in a Montague grammar. In Linguistic Inquiry 9, 393-426. 1979 GAZDAR,

1980

Dative movement and Thomason's extensions of Montague grammar. In Davis & Mithun 1979, pp. 153-222. G.

Phrase structure grammar. In G.K. Pullum & P. Jacobson (eds.) On the nature of syntactic representation, to appear.

GROENENDIJK, J. & M. STOKHOF 1978 Proc. of the second Amsterdam coll. on Montague grammar and related topics. Amsterdam papers in formal grammar II, Centrale Interfaculteit, Univ. of Amsterdam. 1980

The semantics of wh-complements. In Groenendijk, Janssen, Stokhof 1980.

jROENENDIJK G., T. JANSSEN & M. STOKHOF (eds.) 1980 Formal methods in the study of language. Proceedings of the third Amsterdam colloquium. Mathematical Centre Tract, Mathematisch Centrum, Amsterdam. GUENTHER, F. & J.S. SCHMIDT (eds.) 1979 Formal semantics and pragmatics for natural languages. Synthese language library 4, Reidel, Dordrecht. HENY, F. & H.S. SCHNELLE (eds.) 1979 Selections from the third Groningen round table, Syntax and Semantics 10, Academic Press, New York.

MONTAGUS GRAMMAR AMD FG

297

JANSSEN, T.M.V. 1978 Compositionality and the form of the rules in Montague grammar. In Groenendijk & Stokhof 1978, pp. 101-124. 1980a

Compositional semantics and relative clause formation in Montague grammar. In Groenendijk, Janssen, Stokhof 1980.

1980b

On problems concerning the quantification rules in Montague grammar. In Rohrer 1980, pp. 113-134.

KWEE 1979

Tjoe Liong, A68-FG(3). Simon Dik's funktionele grammatika geschxeven in ALGOL68, Versie nr. 3 Publicatie 22, Instituut voor algemene Taalwetenschap, University of Amsterdam.

(this volume) In search of an appropriate relative clause. LEWIS, D. 1972 General semantics. In D. Davidson & G. Harman (eds.) Semantics of natural language, Synthese library 40, Reidei, Dordrecht, 1972, pp. 169-248. Reprinted in: Partee 1976, pp MONTAGUE, R. 1970 Universal grammar, Theoria 36, 373-398. Reprinted in Thomason 1974, pp. 222-246. 1973

The proper treatment of quantification in ordinary English. In K.J.J. Hintikka, J.M.E. Moravcsik & P. Suppes (eds.) Approaches to natural language. Synthese Library 49, Reidel, Dordrecht, 1973, pp. 221-242.

PARTEE, B.H. 1973 Some transformational extensions of Montague grammar. Journal of Philosophical Logic 2, 509-534. Reprinted in: Partee, 1976, pp.51-76. 1976

Montague grammar. Academic Press, New York.

1979a

Constraining transformational Montague grammar: A framework and a fragment. In Davis & Mithun 1979, pp. 51-102.

1979b

Montague grammar and the well-formedness constraint. In Heny & Schnelle 1979, pp. 275-314.

REINHART, T. 1979 Syntactic domains for semantic rules. In F. Guenther & J.S. Schmidt 1979, pp. 107-130. ROÖRER, C. 1980 Time, tense and quantifiers. (Proc. conf. Stuttgart 1979). Linguistische Arbeiten 83, Max Niemeyer Verlag, Tuebingen. THOMASON, R.H. 1974 Formal philosophy. Selected papers of Richard Montague. Yale University Press, New Haven. 1976

Some extensions of Montague grammar. In Partee 1976, pp. 77-117.

WALDO, J. 1979 A PTQ-semantics for sortal incorrectness. In Davis & Mithun 1979, pp. 311-331.

Functions and Cases

J. Lachlan Mackenzie Free University, Amsterdam

1. Introduction An examination of functional grammar (FG) from a case grammar (CG) viewpoint is a potentially fruitful undertaking, if for no other reason than that FG owes an obvious intellectual debt to CG (Dik, 1978: v, 31), yet purports to offer an analysis truer to observed formal distinctions in language (Dik, 1978: 39 ff., 99-100).

A direct comparison of FG and

CG is however impossible, given that CG is transformational-generative and thereby falls under the "formal paradigm" which Dik (1978: 1-6) regards as being diametrically opposed to the "functional paradigm" within which FG is conceived.

I shall therefore not seek to argue that

FG and CG (or subcomponents thereof) are notational variants of each other (cf. Hoekstra, 1978; Dik, 1979) in the way that Anderson (1980) has claimed that relational grammar (also transformational-generative) is case grammar.

Rather I shall endeavour to show that FG could benefit

from an even greater incorporation of CG insights into linguistic structure, with special regard to the areas in which CG and FG are the most similar, namely the analysis of semantic and syntactic functions. That CG is indeed transformational-generative requires little support. Its origins are to be traced to Fillmore (1966), which attacks the confusion of categorial and functional information in Chomsky's "illustrative fragment of the base component" (cf. also Meisel, 1973; Anderson, 1977) and proposes that all deep-structure NPs and PPs be dominated by a functional label; furthermore, Fillmore points to the necessity for certain new transformations, Subjectivalization and Copying.

Later

developments of CG have not changed the fundamental methodological

300

MacKENZIE

commitment of its practitioners: Fillmore (1977: 70) recognizes three levels of representation, surface structure, deep structure (in much the EST sense, and mapped into surface structure in the familiar way), and "a level of representation which recognizes case functions", the relation of which to deep structure is less clear; Anderson (1977), on the other hand, argues for a model reminiscent of generative semantics, with an unordered semantic base structure, cyclic and post-cyclic transformations, and cyclic lexical insertion. That CG is a grammar (in the broad sense) is a less defensible proposition.

Fillmore (1977: 62) regrets that he has been interpreted

as wishing to present a general model of linguistic structure, e.g. by those who have written to ask how CG handles intonation.

A very wide

range of syntactic and semantic issues has, however, been tackled by Anderson in a profusion of publications since 1968 (for a select bibliography, see Anderson, 1977); Anderson and Jones (1977) have moreover put forward a theory of phonological structure that is empirically and methodologically consistent with Anderson's work on syntax and semantics.

The entirety of this output may be regarded as

representing a coherent effort towards a grammar.

On these grounds, it

will be Anderson's work that I shall take as representative of CG in the following pages; it behooves me therefore to offer a brief sketch of his proposals. Andersonian case grammar (ACG) is characterized by dependency representations such that the highest node is a predicate immediately governing nodes associated with a case-relation label; each of these nodes requires a dependent N(oun).

ACG recognizes no more than four

(universal) case-relations, which are subject to componential analysis in terms of the atomic spatial notions of "place" and "source": (1)

absolutive: locative: ergative: ablative:

-place, +place, -place, +place,

-source -source +source +source

This is an expression of the "localist hypothesis", according to which "spatial expressions are more basic, semantically and grammatically, than various other kinds of non-spatial expressions" (Lyons, 1977: 71B). Anderson specifically claims that "all of the case-relations apparently

FUNCTIONS AND CASES

3Q1

manifest varying degrees of abstractness, depending crucially on the character of the predicate on which they depend and the arguments subordinate to them.

In fact, abstractness/concreteness is irrelevant

to the characterisation of the case-relations as such" (1977: 116). ACG differs from the proposal made by Fillmore (1968) (the insufficiency of which has been demonstrated by Huddleston, 1970) to have each NP associated with only one case-label in permitting certain combinations of case-relations, so that in (2): (2)

His acting afforded me much pleasure.

(Cf. Anderson, 1977: 109)

his acting is both ergative and ablative, me is both absolutive and locative, and much pleasure is simply absolutive.

As is also clear from

(2), Fillmore's requirement (similarly disputed by Huddleston, 1970) that only one instance of each case occur per proposition is relaxed in ACG, but only for absolutive, thereby meeting Huddleston*s strictures. Further details of ACG will emerge in what follows; for a full account and justification of ACG, the reader must turn to Anderson (1977). 2. Arguments and Satellites The most salient parallelism between CG and FG is that both, in contradistinction to EST and relational grammar (RG), take the semantic relations between predicate and argument (Dik: semantic function; Anderson: case-relation) to be fundamental, in the sense that they are alone present in initial structure (Dik: predicate frame; Anderson: base structure).

In both approaches, these initial structures are linearly

unordered (Dik, 1978: 70; Anderson, 1977: 74), and in both, all "contentives" are treated as predicates (Dik, 1978: 15; Anderson, 1977: 99). Both, too, distinguish between arguments and satellites (Anderson: participants and circumstantials), and share an analysis of instrumentals, treated by Fillmore (1968) as arguments, as satellites (Dik, 1978: 27, 39-41; Anderson, 1977: 128) (pace Dik's, 1978: 27-28, unconvincing analysis of the object of use as an instrumental argument).

Both agree

that the number of arguments per predication is limited: Anderson's subcategorization and dependency rules allow no more than three per predication (1977: 109), while Dik states that "nuclear predications

302

MacKENZIE

in natural language hardly ever have more than four different argument positions.

Even predicates with four nuclear arguments are rare in any

language." (Dik, 1978: 30) Dik's criteria for distinguishing between arguments and satellites do not emerge at all clearly from his writings, despite the wide recognition of the subtle nature of this distinction. (a) (b) (c) (d)

He does state (1978: 25) that

satellites "do not function in the definition of the state of affairs as such" satellites "give further information pertaining to the state of affairs as a whole" satellites indicate the time, location, reason, cause, ... of the state of affairs satellites are "optional", whereas arguments are "necessary"

With (a), however, he begs an obvious, but unanswered question and opens himself to accusations of circularity if it is a nuclear predication, i.e. one without any satellites, that defines a set of states of affairs (cf. Dik, 1978: 69).

No serviceable criterion can be derived from (b),

and, as to (c), Dik admits that arguments can be characterized by many of the same semantic functions as satellites.

And (d) also fails as the

basis of a possible criterion given that, as Dik (1978: 25) points out, arguments "may be left unspecified under certain contextual or situational circumstances". The distinction between arguments and satellites has also been troublesome for CG.

Fillmore (1968) handles the opposition between

locative arguments and satellites in terms of "highly restricting" and "weakly restricting" locatives respectively.

He offers no definitive

answer to the question whether the latter belong to the "propositional constituent" ("a tenseless set of relationships involving verbs and nouns" - Fillmore, 1968: 23) or to the "modality constituent" (which "will include such modalities on the sentence-as-a-whole as negation, tense, mood and aspect" - Fillmore, 1968: 23).

If weakly restricting

locatives are assigned to the propositional constituent of CG*, this imposes an analysis of certain sentences that infringes the constraint alluded to in §1 above according to which "no case category appears more than once" (Fillmore, 1968: 24). sentence such as (3):

Consider, in this regard, a

FUNCTIONS AND CASES (3)

303

In the dining-room he stood on the table.

The constraint, which is otherwise well-motivated (with the proviso that Anderson's absolutive is exempt), could be saved by analysing highly restricting and weakly restricting locatives as two different cases, but, in as much as cases are distinguished semantically in the defining metalanguage, it is surely inadmissible to set up two (or more) equivalent but syntactically differentiated case-relations. The alternative is to assign weakly restricting locatives to the modality constituent (M).

Pak's (1974: 34) criticism of this proposal,

namely that "it follows ... that all cases, adverbial or not, may be modalized [since] there is no way to determine which should be case and which should be modality" is threatening if no criteria for distinguishing weak and strong restriction are available. for cases to

Fillmore does in fact allow

figure in M: "It is likely ... that certain 'cases' will

be directly related to the modality constituent as others are related to the proposition itself, as for example certain temporal adverbs" (1968: 32).

If, with "certain temporal adverbs", Fillmore is referring to those

that are "weakly restricting", it would appear justified to situate weakly restricting locatives in M also.

This apparent solution is however

based on an equivocal use of the term "case".

Although the internal

structure of M is never made very clear in Fillmore's writings, it is unquestionable that it does not contain the predicate, which is anchored in the proposition.

The function of case-relations is however to associ-

ate NPs with the predicate.

It is therefore very hard to see how NPs

dominated by a case-node could be present in M, since there is no direct syntactic relation between these NPs and the predicate. In a footnote, Fillmore (1968: 23, n. 29) suggests that such adverbs as' unfortunately,

willingly,

superordinate sentences.

easily

and carefully

should be d e r i v e d

from

This proposal is extended in Fillmore (1971) to

weakly restricting locatives and temporals: these are treated as being situated in underlying structure in higher Ss containing as their predicate occur or happen.

Carried to its logical conclusion, this

proposal obviates the difficulties inherent in assigning case-nodes to M.

Fillmore (1972) analogously suggests analysing adverbials as

"disguised" embedding verbs and specifically endorses Lakoff's (1968) proposal that instrumental with-phrases be derived from underlying

304

MacKENZIE

structures containing the verb use (thereby distancing himself from his 1968 analysis of such phrases as strongly restricting).

This incor-

poration of generative-semantics proposals into CG is also favoured by Anderson (1972), who treats locative, temporal, reason and "some manner" adverbials as deriving from higher "quasi-predications".

Indeed ACG has

no modality constituent as such: in general, all the categories which Fillmore has appear in M are assigned by Anderson to such superordinate quasi-predications.

This solution has several advantages: it avoids

Pak's objection that M is little more than a "semantic dump" for phenomena that cannot easily be handled within the propositional constituent (in Pak's pregnant formula, M = S - P); it provides a natural representation for the variable scope of negators, quantifiers and adverbials in terms of their relative position in the dependency tree; it replaces the unmotivated catenation of M and P (S •) M + P) with a unified representation in which the position of the various elements, propositional and extra-propositional, is motivated on semantic grounds. Both FG and CG thus stand in need of a reliable, cross-linguistic characterization of argumenthood.

Carvell and Svartvik (1969: 12) have

noted that the distinction between argument and satellite is "often based on the feeling that in the first ... the verb and the preposition form a close unit, in the second ... they do not.

However, although it is

indispensable to the analyst, such linguistic feel should not be the only basis for ordering the material".

Yet the tests that have been

proposed either fail to satisfy the analyst's intuitions and/or are restricted in their applicability to one language.

Thus Chomsky's

(1965: 101) claim that only satellites can be "optionally preposed to the sentence" (cf. also Dik, 1978: 53) is empirically dubious: Carvell and Svartvik point out in this connection that "emphatic front position is almost always possible with emphatic intonation and stress", and the test is inapplicable to, say, German.

Chomsky (1965: 105) further

asserts that NPs in satellites are not subject to "pseudopassivization", but this is of course equally true of many NPs in arguments (cf.*rhree hours were lasted for), so that no criterion results.

Anderson suggests

(1977: 135-136) that a satisfactory test for English is derivable from Browne's (1972) observation that only satellite question-words can be conjoined:

FUNCTIONS AND CASES (4) (5)

305

Where and when did John buy his car? *How long and where did you keep the car?

But (6), with what are intuitively {and according to the do so test discussed below) two conjoined arguments, is acceptable: (6)

Where and with whom does John live?

so that this test must also be discarded.

A more reliable test for

English is Lakoff and Ross' (1966) do so test which is applicable to all but copular sentences.

This test involves conjoining to S^, the

sentence being tested, and X do so Y too (in the appropriate tense), where X is a subject referentially non-identical to the subject of S^ and Y is the (repeated) element whose argumenthood is in question.

Only

if Y is a satellite will the resultant conjunction be grammatical: (7) (8)

John bought a car in Amsterdam and Bill did so in Amsterdam too. 'John lives in Amsterdam and Mary does so in Amsterdam too.

Esau (1973: 66) attempts to use a similar test for German, employing dasselbe tun as an equivalent of do so too, but the resultant sentences are so unusual that it is difficult to assess their grammaticality.

In

any case, Lakoff and Ross* test, which makes crucial reference to the syntactic constituent V' in English (cf. Jackendoff, 1977: 58), will presumably be inapplicable to those languages which lack a verb phrase. Generally, it remains to be shown that the argument-satellite opposition corresponds to a universal semantic distinction, for Dik presents it as such, and that it is not a grammatical artefact stemming from a generalization of intuitions influenced by the syntax of English and directly comparable languages.

In this connection, it is interesting

to observe that the semantic function hierarchy, to which I shall return in §4 below, cuts across the argument-satellite distinction.

English

Object assignment, for example, has its cut-off point after Ben(eficiary) (Dik, 1978: 78), which is introduced as a satellite (Dik, 1978: 52). Since the subject function can be assigned, given the appropriate predicate, to wellnigh any semantic function, whether argument or satellite, as is shown for English by Couper-Kuhlen (1979), the need for

MacKENZIE

306

such a distinction in the semantic predicate-frame falls away.

To the

extent that ACG semantic representations indicate, through the use of hierarchical dependency structure, the scope-relations obtaining between sentence elements, independent of their surface-syntactic status, they avoid the criticism here levelled at FG predicate frames. Semantic functions Dik (1978: 31) observes that the "semantic functions used in FG ... are somewhat reminiscent of those distinguished in Fillmore's (1968) Case Grammar".

Indeed, Dik's definition of semantic functions (1978:

32) is entirely consistent with the interpretation ascribed to caserelations by CG (cf. Anderson, 1977: 9: "labels for the semantic role that a particular NP fulfils in the predication").

There are also many

correspondences of detail: Dik (1978: 42) shares with Fillmore (1977: 78) and Anderson (1977: 45) the conviction that animacy should not be used to distinguish semantic functions; he agrees with Anderson (1977: 121) that "Experiencer" is not a separate case-relation.

However,

whereas the rules in Anderson (1977) allow for four simple and five multiple case-tables: (9)

abs loc erg abl

abs,loc loc,erg erg,abl

abs,erg

abs,abl

(plus, possibly, loc,abl for "path") Dik (1978) proposes no fewer than twenty-four semantic functions, of which eleven may be "nuclear", i.e. may characterize arguments.

The

nuclear semantic functions differ from CG case-relations in being relativized to a- typology of states of affairs, classified according to the presence or absence of control and dynamism (cf. Dik, 1978: 32-39).

From

the viewpoint of CG, this introduces needless complexity into the specification of the semantic functions, losing, for example, the generalization that "all verbs take an abs" (Anderson, 1977: 82), that is, one semantic function (absolutive) is common to all states of affairs. More seriously, perhaps, Dik's taxonomy of nuclear semantic functions can be shown to be inconsistent, or at least unclear, and to undergeneralize.

FUNCTIONS ÄND CASES

307

As a result of the linking of nuclear semantic functions to types of states of affairs, not all entities that play the same semantic role are assigned the same semantic function in FG.

Thus, the "controller" of an

Action, a state of affairs that is [+controlled, -adynamic] , is termed Agent, while the controller of a Position [(-controlled, -dynamic} is 2 termed Positioner.

Similarly, the entity to which an Action is applied

and the entity which is affected by a Process [-controlled, +dynamicj are termed Goal and Processed respectively, despite the parallel roles they play in, for example, John opened the door and The door opened. Yet it is not the case that a role which occurs in more than one kind of state of affairs always receives a different appellation in each: Recipient, Direction and Source occur in Actions and Processes; Location occurs in Positions and States [^-controlled, -dynamic} ; and Goal occurs in Actions, Positions and Processes.

The criteria for the demarcation of semantic

functions are in this respect far from clear. Let us consider Goal further.

It is defined as "the entity to which

the Action is applied by the Agent" (Dik, 1978: 37), a definition which should exclude it from occurrence in other types of state of affairs, although a later definition, p. 41, assigns Goal to "those entities which are affected (or effected) by the operation of some Controller ( ... ) or a Force".

Unlike the abs of ACG, which Anderson (1979: 2) defines as

"the argument most intimately associated with its predicate, with which it enters into the potentially most specific selectional restrictions: with location or movement predicates it introduces the located or moving entity; with action or experience it is that which is acted upon or experienced; with processes or descriptions it is undergone or described". Goal is not generalized to the Processed function or to the 0 function attributed to "entities in a certain state" (Dik, 1978: 41).

The moti-

vation for the semantic function Goal, as against the more general abs, is however methodologically unsatisfactory.

Dik (1978: 41) claims without

further substantiation that the sort of role indicated by Goal receives a better characterization and, secondly, "that the Goal function as we have introduced it can be used in formulating the rules for Subject and Object assignment".

In other words, the motivation for the semantic function

Goal is that it facilitates the specification of syntactic regularities, an assertion which undermines the autonomy of the levels of semantic functions and syntactic functions in an undesirable manner.

308

MacKENZIE

One function. Positioner, is assigned to the argument corresponding with the subject of each of the following sentences: (10a)

John stayed motionless.

(10b)

John was sitting in his chair.

(10c)

John kept his money in an old sock.

(lOd)

John kept the present.

(Dik, 1978: 38)

In each case, "John" is said to control a non-dynamic state of affairs: hence the assignment of Positioner. one semantic role is involved.

It is however not clear that only

In (10a), (10c), and (lOd), John is, in

the desired interpretation, the "causer" of the state of affairs; in (10b), although in control of the situation, John cannot be said to be causing it. (11)

A clear example of the difference is (11):

John was lying on the ground.

Dik would predict that such a sentence was ambiguous according as John is Positioner or 0, i.e. in control of the situation or not.

Lakoff's

ambiguity test (Lakoff, 1970), however, shows that (11) is merely "vague", since (12) has four, and not two understandings: (12)

John was lying on the ground and Bill was lying on the ground.

I conclude that there are no grounds for distinguishing Positioner in sentences like (10b) and (11) from 0; indeed both would be simply abs in ACG.

As for the Positioners in (10a), (10c), and (lOd), they seem to be

indistinguishable from the Agents that are arguments of such predicates as give, send, take, with the distinction that, rather than causing a change, they cause no change to take place.

A further advantage of

collapsing the Positioners of (10a), (10c), and (lOd) with Agent is that the arguments corresponding to the Subjects of John remained in Africa and John did not leave Africa would have, as seems intuitively correct, the same semantic function.

In ACG, all Agents and causative Positioners

have the case-label erg, with or without further case-specifications. In FG, Location occurs only in Situations (non-dynamic states of affairs), and Recipient, Direction and Source only in Events (dynamic

FUNCTIONS AND CASES states of affairs).

309 Moreover, Recipient and Direction do not co-occur,

although either may co-occur with Source.

That is, Recipient and

Direction are in complementary distribution, and they together are in complementary distribution with Location.

What is then to prevent FG

from following ACG and recognizing only one function. Location, which designates "the entity to which something is transferred" (= Recipient) or "towards which something is moved" (= Direction) if it occurs in a dynamic predication, and which designates a static location if the predication is non-dynamic?

This question is not addressed by Dik (1978)

but one suspects that the Semantic Function Hierarchy (SFH) which determines syntactic Subject and Object assignment, and in which Rec and Loc appear separately and discontinuously, Dir not appearing at all, may have motivated distinguishing Location, Recipient and Direction as semantic functions. The distinction between Recipient and Direction resides, according to the definitions given above, in the difference between "transfer to" and "move towards" (cf. Dik, 1978: 37); "move to" might even be a better characterization of Direction, to judge by Dik's examples The wind blew the leaves into the kitchen and John sent a book to London. 1 take it that the use of "transfer" in the definition of Recipient indicates movement into the Recipient's "possession"; note, moreover, that the Recipient is "typically" animate (Dik, 1978: 78).

Could it then not be

that, given their complementary distribution, Recipient and Direction are but one semantic function, the Recipient being inherently characterized as "possessing" (in a broad sense) and preferably animate? Again, the distinction between two semantic functions seems to have been motivated by syntactic considerations, namely that Recipients are in some languages susceptible to Subject and Object assignment, whereas Directions, at least according to the SFH, never are. Finally, the distinction in FG between the semantic functions Location and Time (Temp), also embodied in the SFH, could, in my view, with benefit be dropped, since the distinction resides more in the nature of the argument itself than in the semantic role it plays in the predication (cf. Lyons, 1977: 719, who distinguishes between "first-order entities" in spatial location and "second-order entities" in temporal location). A collapsing of Loc and Temp would of course also lead to a simplifi-

310

MacKENZIE

cation of the expression rules, given that spatial and temporal expressions are in many languages closely related. At this point, it may be useful to present a (partially summarizing) table of correspondences between FG nuclear semantic functions and ACG case-relations:

FG nuclear semantic functions

ACG (underlying) case-relations

Agent, "causative" Positioner

erg r ~i Itcase^j

Goal, Processed, 0, "noncausative" Positioner

abs

Source

abl

Recipient

loc (in embedded predication)

Location, Direction, Time Force

loc

ea

4. Syntactic Functions Both ACG and FG incorporate the insight that a grammar which makes reference to syntactic functions alone (e.g. RG) will be forced to have recourse to cumbersome and unnatural formulations of regularities that are elegantly and naturally attributable to semantic functions.

Thus

Dik states that only [^Controlled} states of affairs allow predications "occurring under the scope of the imperative mood" (Dik, 1978: 35) and Anderson (1979: 3-4) similarly argues that "imperative deletion" should be restricted to the agent in action clauses.

Dik's formalism allows

the full range of functional information to be available to all rules; Anderson (1979: 32, n.l) considers what is presumably the transformationalgenerative equivalent of such a procedure, namely one "whereby caserelations and grammatical relations are equally available throughout a derivation", and concludes that it constitutes a weaker hypothesis than his own (cf. also Anderson, 1980: 23). ACG's hypothesis is that:

311

FUNCTIONS AND CASES (a)

in the cyclic clause, all rules makes reference only to caserelations;

(b)

rules that relate to subordinate clauses (e.g. Equi) make reference only to grammatical relations in those clauses (if the language has grammatical relations);

(c)

subject-formation is a cycle-terminating rule.

This hypothesis thus predicts that one identifiable class of syntactic operations will be sensitive to case-relations and another will be sensitive to grammatical relations; Dik's position is, in this respect, weaker in that either semantic functions (case-relations) or syntactic functions (grammatical relations) may in principle be referred to by any rule.

For further comments on this matter, see Hoekstra (1978).

The grammatical relations of ACG and the syntactic functions of FG are not directly comparable: what Dik terms a "subject" in English is also "subject" to Anderson, but what Dik calls "subject" in Tagalog and Kapampangan is to Anderson a "prime" (1979: 24).

Whereas Dik's defi-

nition of subject is notional ("that constituent which refers to the entity which is taken as a point of departure for the presentation of the state of affairs in which it participates"), Anderson's is morphosyntactic ("A language L has subjects if the agent in (actional) intransitive clauses in L shares morpho-syntactic properties with the argument in intransitive clauses in L which neither shares with the patient in transitives: NPs sharing these properties are subjects" 1980, 11).

Thus Dik recognizes subjects in Dyirbal (1978: 167-168),

tut Anderson recognizes subjects only for those pronominal constructions in Dyirbal which manifest "nominative" morphology (in the sense of Dik, 1978: 164). Whereas FG treats Subject assignment as the addition of a different type of function to an argument already marked for a semantic function, ACG (Anderson, 1977: 162) hypothesizes that subject-formation is the substitution of an erg (i.e. a case-relation!) for the case-relation highest on the "subject selection hierarchy" (equivalent to the SFH), unless it is already erg, in which circumstance subject-formation applies vacuously.

The choice of erg, "the quintessential human role"

(Anderson, 1977: 119), is partly motivated by the tendency for speakers and hearers anthropocentrically to "empathize" with subjects, and presumably also to personify subjects more readily than other grammatical relations.

In ACG, an object is defined as "the hierarchically

MacKENZIE

312

highest absolutive argument in a predication which prior to subjectformation also contains a separate ergative argument" (Anderson, 1977: 275); thus, as in FG, the presence of an object is dependent on that of a subject, but the definition is again in terms of case-relations rather than notions of "perspective". The idea of perspective so central to FG's definitions of syntactic functions is intuitively appealing, especially in the case of subject (although it is admittedly difficult to see how these definitions or their applicability to individual cases could ever be falsified).

Dik

stresses that Subject and Topic are not to be confused, but the fact that both are treated as "starting-points". Subject for the presentation of the state of affairs and Topic for the predication (in as much as the predication says something about the Topic) brings out the intimate relation between subjects and topics that has long been recognized. Anderson (1979: 28) reviews the evidence for a historical derivation of subject-forming languages from "topic-prominent" languages and concludes that subjects may have arisen from the institutionalization of a tendency to "empathize" with topicalized erg NPs.

FG's notion of

perspective is almost certainly related to "empathy" in this sense; cf. also Fillmore's related discussion of perspective (1977). The difference between the FG and ACG approaches to subjecthood emerge most clearly in relation to languages like Tagalog, in which, according to Dik, Subject may be assigned (and in most sentences is) to any of Ag, Go, Rec, Ben, Inst or Loc.

Anderson (1979: 21-25),

considering the same data, concludes that the phrases in question simply fail to satisfy the morpho-syntactic definition of "subject" quoted above.

He further observes, with Dik, whom he cites as

presenting a more careful discussion, that the phrases are not "topics" in any generally accepted sense, and proposes a new label for such phenomena, namely "prime", which differs from subject in, for instance, not having preference for any case-relation(s).

Anderson's position (1979,

1980) is that every clause has a "principal relation", which is, inter alia, the most accessible to relativization.

The abs, the only obli-

gatory relation in every clause is this principal relation, unless it is usurped by a "derived principal relation", which may be subject (as in English), prime (as in Tagalog), or topic or focus (as in Hungarian - see Anderson, 1979: 26-27).

FUNCTIONS AND CASES

313

Let us also consider the contrast between the FG and ACG approaches to the morpho-syntax of non-pronominal constructions in Dyirbal (Dik, 1978: 167-170; Anderson, 1977: 253-261; 1979: 5-11). As regards such constructions, Dyirbal is an "ergative language", and relativization, conjunction-reduction, equi and inflexional morphology are consequently sensitive to the principal relation abs.

The problems noted by both Dik

and Anderson is that, in the much-discussed "anti-passive" construction, the Agent may occur with absolutive morphology, the verb is marked with the suffix or dative case.

and the Goal (= underlying abs) appears in the ergative Dik shows that this cannot be treated merely as a

question of alternative Subject assignment, proposing that, since the V-gai form acts as an intransitive verb, the ergative phrase of the "anti-passive" is an instrumental phrase (morphologically identical to an ergative phrase in Dyirbal) included in the predicate.

There are two

problems with this analysis: firstly, it is not clear how the formalism of FG can permit a term to change semantic function, i.e. from Goal to Instrument (except by an ad hoc expression rule); secondly, Dik cannot account for the occurrence of dative morphology as an alternative to ergative in the "anti-passive" construction, dative morphology which is obligatory if the Goal is expressed by a pronominal.

I wish to present

an alternative account, in the spirit of ACG, which is exempt from these shortcomings. If the "anti-passive" is treated as a version of passive, the fact that V-Jpai functions as an intransitive verb follows naturally.

Anderson

(1977: 205) proposes and justifies (13) as the underlying structure for passives, claiming that it embodies the traditional insight that, in the passive, "the subject receives the action of the verb" (hence JjLoc,ergJ ) , and that it offers a natural explanation for the frequent use of ablative morphology for agent-markers (cf. German von, Old English of, etc.).

MacKENZIE

314

The derivation of passives involves double application of equi, verb raising and subjunction (cf. Anderson, 1977: 137) of the empty absphrase to V^.

The derivation of anti-passive, I would propose, involves

one operation of equi and one of raising, whereby the erg N^ under V^ is raised into the lexically empty

If equi is a form of raising (into

a co-referential N, cf. Anderson, 1977: 169, 209), ordering problems will not arise.

And the abl-node can be deleted to give an agentless

passive by a rule as will presumably also be required in such languages as Latvian, Classical Arabic and Maninka which Keenan (1980: 212) quotes as possessing necessarily agentless passives.

In the resulting

structure, (14): (14)

V

N , Dik's Agent, is marked as abs, and N , Dik's Goal

complex case-label that is naturally realized by either dative (= loc) or ergative.

The advantages of such an account are that no reference

is made to subjecthood (which Anderson claims is lacking in Dyirbal nonpronominal constructions), the intransitivity of the verb-form is naturally accounted for and the right predictions about case-marking are made. Object assignment, "a way of defining a perspective on the terms remaining after Subj assignment has taken place", can, as the definition suggests, occur only once per predication. only one object in John gave my brother

Thus FG, like RG, recognizes

the books,

the books being merely

Goal in FG 'and a chômeur in RG; CG however analyses both my brother and the books as objects, since both fulfil the definition quoted above. Anderson (1977: 184-185) points out that RG cannot account for the generalization that, in English, "only an object can intervene between an object and its verb", cf.: (15a)

Jim gave necklaces very frequently to the girls.

(15b)

*Jim gave very frequently necklaces to the girls.

FUNCTIONS AND CASES

315

(15c)

*Jim gave very frequently the girls necklaces.

(15d)

*Jim gave the girls very frequently necklaces.

FG could presumably handle these data in terms of a requirement that such adverbials as very frequently may not occur before Goal, given the assignments: (15a) = GoObj Adv Rec (15b) = Adv GoObj Rec (15c) = Adv RecObj Go (15d) = RecObj Adv Go but such a constraint would go against what I perceive as the spirit of FG, namely that only syntactic and pragmatic functions are referred to by constraints on constituent ordering (cf. Dik, 1978: 74, " ... rules determining constituent ordering will not usually refer to specified semantic functions").

Such matters as the placement of adverbials are

not discussed in Dik (1978), so I shall press the point no further. My final point relates to the SFH and the notion of "cut-off point", defined (Dik, 1978: 9) as "the limit of applicability" of the hierarchy. This formulation involves a fundamental equivocation which has, to my knowledge, never been elucidated in print.

Consider a language such as

English, in which the cut-off point for Subject assignment is said to be "around the Beneficiary".

Is this to be interpreted as a claim that all

Agents, Goals and Recipients, some Beneficiaries and no Instruments, Locations and Times are susceptible to Subject assignment? claim is readily falsifiable.

If so, the

As is well known, many recipients cannot

be assigned Subject function, whereas certain Instruments, Locations and Times can:

(16a)

The church was given/»donated ?10,000.

(16b)

John was offered/*returned the money.

(16c)

I was told/*muttered the secret.

(17a)

Twenty dollars will buy you a square meal.

(17b)

The tank has sprung a leak.

(17c)

The last few weeks have been cold.

316

MacKENZIE

If the cut-off point is to be interpreted as the point after which further Subject assignment is prohibited, then the examples in (17) still furnish counter-evidence, and the status of Beneficiary as being beyond the cut-off point (cf. Dik, 1978: 80) cannot be justified.

In

general, one can assent to the notion of a hierarchy, with the associated hypotheses of continuity, increasing markedness, and a progression from clearly grammatical through doubtfully grammatical to clearly ungrammatical (cf. Dik, 1978: 76-77) without requiring the additional hypothesis of a cut-off point. 5. Conclusion This discussion of FG from the viewpoint of CG has of necessity been selective. Many matters have been disregarded: the interested reader may wish to compare FG's approach to the function and structure of terms with that of Thrane (1980), who adopts a position compatible with and in many respects identical to CG; linearization within a dependency CG is dealt with by Anderson (1976); the equivalent of expression rules in CG are discussed in preliminary fashion by Mackenzie (1980).

The lack of

methodological correspondence between FG and CG, and the many aspects of linguistic structure that have not yet been dealt with by FG, or indeed CG, have made a point-by-point comparison impossible.

I do hope to have

shown, however, that, at least as regards semantic and syntactic functions, FG would benefit from a re-direction towards current work in CG. NOTES 'This appears to be the solution favoured by Fillmore. He claims (1968: 26, n. 34) that the appearance of "outer (= weakly restricting) L" and of B(enefactive) is dependent on that of A(gentive). The only way of stating this dependency relation between cases economically would be for outer L, as B does for Fillmore, to figure in the proposition. Unfortunately, this argument is vitiated by the fact, pointed out by Anderson (1977: 28), that outer L does not require the co-presence of A: (i) (ii)

Fred died in his bed. The butter in this dish has melted in that cupboard.

2 In ACG, Agent and Positioner would both receive the same case-label erg (with or without additional case-specifications). Erg is defined by Anderson (1979: 14) as the "'potential controller', requiring an entity high on the humanness hierarchy".

317

FUNCTIONS AND CASES REFERENCES

ANDERSON, J.M. 1972 Remarks on the hierarchy of quasi-predications. RRLing 17, 23-44, 121-140, 193-202, 319-335. 1976 On Serialization in English Syntax. Ludwigsburg Studies in Language and Linguistics, 1. 1977 On Case Grammar:Prolegomena to a Theory of Grammatical Relations. Croom Helm, London. 1979 On Being without a Subject. Indiana University Linguistics Club. 1980 Anti-Unaccusative, or Relational Grammar is Case Grammar. Linguistic Agency, University of Trier. ANDERSON, J.M. and C. JONES 1977 Phonological Structure and the History of English. North-Holland, Amsterdam. BROWNE, E.W. 1972 Conjoined question words and a limitation on English surface structures. LI 3, 223-226. CARVELL, H.T. and J. SVARTVIK 1969 Computational Experiments in Grammatical Classification. Mouton, The Hague. CHOMSKY, N. 1965 Aspects of the Theory of Syntax. MIT, Cambridge, Mass. COUPER-KUHLEN, E. 1979

The Prepositional Passive in English. Niemeyer, Tübingen.

DIK, 1978 S.C. Functional Grammar. North-Holland, Amsterdam. 1979 Over Funktionele Grammatika. Forum der Letteren 20, 42-51. ESAU, H. 1973 Nominalization and Complementation in Modern German. NorthHolland, Amsterdam. FILLMORE, Ch.J. 1966 A proposal concerning English prepositions. MSLL 19, 19-33. 1968 The case for case. In E. Bach and R.T. Harms (eds.) Universals in Linguistic Theory. Holt, Rinehart & Winston, New York. 1971 Some problems for case grammar. MSLL 23, 35-56. 1972 Subjects, speakers and roles. In D. Davidson and G. Harman (eds.) Semantics of Natural Language. Reidel, Dordrecht. 1977 The case for case reopened. In P. Cole and J.M. Sadock (eds.) Syntax and Semantics VIII: Grammatical Relations. Academic Press, New York. HOEKSTRA, T. 1978 Funktionele Grammatika: Naar aanleiding van S.C. Dik, Functional Grammar. Forum der Letteren 19, 293-312. HUDDLESTON, R.D. 1970

Some remarks on case grammar.

LI 1, 501-511.

JACKENDOFF, R.S. 1977 X' Syntax: A Study of Phrase Structure. MIT, Cambridge, Mass.

MacKENZIE

318

KEENAN, E. 1980 Passive is Phrasal not (Sentential or Lexical). GLOT 2, 395-427. Also in Hoekstra, T., v.d. Hulst, H. and M. Moortgat (eds.) Lexical Grammar. Foris, Dordrecht. LAKOFF, G. 1968 Instrumental adverbs and the concept of deep structure. 4-29. 1970

A note on vagueness and ambiguity.

U

1, 357-359.

LAKOFF, G. and J.R. ROSS 1966 A criterion for verb-phrase constituency. tation Laboratory, Harvard University. LYONS, J. 1977

Semantics.

FL 4 ,

Report NSF-17, Compu-

Cambridge University Press, Cambridge.

MACKENZIE, J.L. 1980 The morphology of case grammar: a reconnaissance. In Fry, A.J., Mackenzie, J.L. and R. Todd (eds.) Free University Studies in English I. Vrije Universiteit, Amsterdam. MEISEL, J.M. 1973 Prepositions in object and adverbial constructions. Lingua 31, 213-236. PAK, T. 1974

Absurdities in Fillmore's case grammar.

SL 28, 19-50.

THRANE, T. 1980. Referential-semantic Analysis: Aspects of a Theory of Linguistic Reference. Cambridge University Press, Cambridge.

Functional Grammar and Relational Grammar

Points of Convergence and Divergence

David M. Perlmutter University of California, San Diego

0. Introduction* In this paper I point out some of the ways that functional grammar (FG) and relational grammar (RG) are alike, and concentrate on the single most important difference between them -the question of grammatical relations at more than one level. This paper is limited to issues of syntax narrowly conceived, i.e. areas where both FG and RG have claims to make. Issues such as pragmatics, the consequences of viewing language as an instrument of social interaction, etc. are ignored here. 1. Some Points of Convergence As Hoekstra (1978) has pointed out, FG is somewhat eclectic, having adopted elements of various models developed in the decade 1967-1977. Some of the elements adopted from RG make the two frameworks alike in a number of important ways. The claims the two frameworks share include the following: (1)

a.

Linguistic theory must posit grammatical relations that are independent of semantic notions such as Agent, Recipient, Benefactive, etc.

b.

These grammatical relations play a role in the syntax of languages that differ in word order, case patterns, etc,

c.

Grammatical constructions in individual languages are to be analyzed (at least in part) in terms of these grammatical relations.

d.

The grammatical relations include the subject ('1' in RG) and direct object ('2' in RG) relations.

320

PERLMUTTER e.

Phenomena such as word order, case marking, and verb agreement are accounted for by rules sensitive to grammatical relations.*

At a time when different syntactic frameworks diverge on a large number of issues, these points in common make FG and RG relatively close. Both accord a great deal of importance to grammatical relations and seek to account for data in a wide variety of languages in terms of them. As a result, even the particular analyses proposed in one framework are easily translated into the other. For example, Dik (1978) translates a significant portion of Chung's (1976) analysis of Indonesian into FG. He accepts her arguments that the notional indirect object is, in many sentences of Indonesian, a direct object. Similar analyses are proposed for other languages. The fact that FG and RG have so much in common makes meaningful comparison between the two frameworks possible. This is not the case with frameworks whose basic assumptions are so different that there is little or nothing that can be held constant while attention is focussed.on points of difference.

2. A Major Difference between FG and RG The major area where FG and RG make distinct and incompatible claims concerns multiple

levels

in syntax. RG claims that a given nominal can

bear distinct grammatical relations to a clause at distinct syntactic levels, whereas FG recognizes grammatical relations at only one level. For example, consider the sentences (2)

a.

Marie gave the watch to Tom.

b.

Marie gave Tom the watch,

c-.

Tom was given the watch by Marie.

In RG, these sentences would have the following representations in terms of stratal diagrams:

2

RELATIONAL GRAMMAR AND FG (3)

b.

a.

give

Marie

the watch

Tom

Marie

the watch

Tom

9ive

Marie

the watch

Tom

c.

give

(3a) represents the fact that Marie bears the 1-relation, the watch the 2-relation, Tom the 3-relation (indirect object), and give the P-relation (Predicate). In this example, the elements of the clause each bear only one grammatical relation to it. In (2b/3b) , however, Tom bears the 3-relation in the initial stratum (level) and the 2-relation in the final one; the watch is an initial 2 and final chomeur. In (2c/3c), Tom has advanced from 3 to 2 to 1; this nominal bears three distinct grammatical relations in three distinct strata (levels). In FG, on the other hand, granmatical relations are recognized at only one level, so a nominal can bear at most one grammatical relation. Adopting RG's notation of stratal diagrams to represent the semantic roles and grammatical relations FG posits for the sentences in (2), we have: b.

(4)

Marie

the watch

Tom

the watch

Tom

PERLMUTTER

322 c.

The semantic roles in the three sentences are the same; they differ in the assignment of grammatical relations. In (2a/4a), Marie is the 1 and the watch the 2, while in (2b/4b), Marie is the 1 and Tom the 2. In (2c/4c), 4 on the other hand, Tom is the 1. The key point is that in FG a given nominal bears at most one grammatical relation. The prominence of semantic functions in FG brings out the fact that work in RG has not been explicit about their place in the theory. They have generally been assumed to be an aspect of semantic representation, and have therefore not been included in syntactic representations. If it were desired to represent

semantic roles in syntactic structures, (2c) could be given

the representation in (5): (5)

This would require some modifications in the class of R-signs, the assignment of coordination, the definitions of notions such as 'initial stratum', etc. It therefore seems preferable not to adopt such representations unless there is clear evidence that syntactic rules need to refer to semantic roles. Comparison

of (5) with (4c) makes the real difference between RG and FG

stand out clearly: RG posits grammatical relations (1, 2, 3, etc) at more than one level; FG does not. In (5) and (3c), initial and final 1 are distinct. This is not the case in (4c). As Dik (1978:87) remarks: "It should further be realised that FG

RELATIONAL GRAMMAR AND FG

3^3

recognises only one sort of Subject function, and does not distinguish (indeed, cannot distinguish) between 'deep* subjects and 'derived1 ones." Where FG posits grammatical relations, they generally correspond to the final grammatical relations in RG. In no case does FG posit grammatical relations at other levels. The issue is clear: is it necessary to posit grammatical relations at more than one level? It is an empirical question whether there are generalizations in grammars whose statement multiple syntactic levels. If there are not,

requires

a single-level theory such

as FG is preferable to RG. If there are such generalizations, FG must be rejected in favor of RG. i 3 is devoted to arguments that multiple levels are needed to state generalizations in the grammars of natural languages. Of the various syntactic levels, two will be important here. The first stratum of a given node is its initial stratum. Where there is a k*"*1 stratum of a node b and no k+l s t stratum of b, the k^ 1 stratum is the final stratum of b. A nominal that is a 1 in the initial stratum of b is the initial 1 of b, and a 1 in the final stratum of b is the final 1 of b. This terminology extends to the other grammatical relations.

3. Some Inadequacies of FG as a Single-Level Syntactic Theory 3.1. Generalizations in terms of Initial Grammatical Relations Verb Agreement in Achenese, an Austronesian language studied by Lawler (1977), provides an argument for recognizing granmatical relations other than final ones. Lawler shows that in Achenese, verb agreement is 4 the same in an active sentence and its corresponding passive:

(6)

(7)

a.

Gopnyan ka g±-com Ion. 3h PERF 3h-kiss me 'She (already) kissed me.1

b.

Lon ka g±-com le-gopnyan. I PERF 3h-kiss by-3h 'I have (already) been kissed by her.

a.

Brsn ni-pajoh boh-mamplam. you 2h-eat fruit-mango 'You eat the mango.'

b.

Eoh-mamplam n±-pajoh le-dr^ji.. fruit-mango 2h-eat by-you 'The mango is eaten by you.'

PERLMUTTER

324

In (6a-b), g±- is the verbal prefix agreeing with a third person nominal referring to someone higher in status (or older) than the speaker. In (7a-b), ni- is the prefix agreeing with a second-person nominal. To determine the structures of individual sentences in RG, as in any other aspect of grammar construction, it is necessary to construct alternative grammars and grammar fragments and find arguments for adopting one over others. In some cases, certain analyses will be forced or excluded by proposed universals of grammar. For example, the universal characterization of Passive clauses proposed in Perlmutter and Postal (1977) imposes an analysis on examples like (6-7). Of interest here is the fact that for all the Achenese data presently available, this analysis also makes possible an adequate account of verb agreement, as will be seen below. In RG, the structures of (7a-b) can be given in the following simplified stratal diagrams:

(8)

b.

a.

pajuil

dron

pajuil

boh-mamplam

dron

boh-mamplan

The structures posited by PG for these sentences can be represented as follows: (9)

dron

b.

a.

boh-mamplan

dron

boh-mamplan

RELATIONAL GRAMMAR AND FG

325

The crucial difference is not whether or not semantic roles are represented in the diagrams, but rather whether grammatical relations are represented at more than one level. The different structures posited in FG and RG make different notions available for the statement of Achenese Verb Agreement. RG can use the notion of 'initial 1', stating the rule informally as follows.-^ (10)

Verb Agreement in Achenese The verb of a clause b agrees with the initial 1 of b.

(10) correctly accounts for the agreement in (6-7) and in the other Achenese sentences cited by Lawler (1977)^. In the FG representation of (7b) in (9b), however, the second-person nominal dron is not a 1 at any level. FG therefore cannot state Achenese Verb Agreement in terms of lhood. Dik's (1978:117) proposal is to state it in terms of the semantic function 'Agent':

(11)

Agreement in Achenese according to FG . The verb agrees with the Agent.

(11) correctly accounts for the agreement in (6-7). Dik claims that Achenese Verb Agreement provides an argument for FG over RG. This conclusion is echoed by Watters (1980:165-66) . But Dik does not cite data from Achenese to show that the semantic notion of Agent is the correct one for stating agreement; (7a-b) are the only sentences he cites. I will now show that (11) makes incorrect predictions about Achenese agreement, even if attention is restricted to the data cited by Lawler (1977). When additional data is considered, the situation is even worse for the FG account. The RG account in (10), however, correctly accounts for the data. In FG, sentences are viewed as predications about "states of affairs", which can be of any of four types [Dik (1978-33)]:

(12)

State = uncontrolled situation = uncontrolled non-dynamic state of affairs Process = uncontrolled event = uncontrolled dynamic state of affairs Position - controlled situation = controlled non-dynamic state of affairs Action = controlled event = controlled dynamic state of affairs

326

PERLMUTTER

The semantic function 'Agent' is defined in terms of the notion of Action (p. 37) : "If an Action predication has only one argument, this argument will necessarily designate the entity controlling the Action. To arguments having this property we shall assign the semantic function of Agent (Ag). If an Action predication has a second argument, this will typically designate the entity to which the Action is applied by the Agent. To this function we will assign the semantic function Goal (Go)." About State predications, Dik (p. 38) says: "If a State predication has only one argument, I shall say that that argument has zero semantic function 0. I see no good reason to assign any particular semantic function to such arguments. Two-place State predications will most often have two zero function arguments, and there may even be three-place State predications with three zero function arguments. A second argument in State predications may, however, also have Time or Location function." Since (11) states Achenese agreement in terms of "Agent", the predication is that there will be agreement only in Action predications. State, Process, and Position predications should lack agreement. Dik (p. 34) explicitly cites the sentence (13)

John saw a beautiful bird

as an example of a State predication. As the quotations above make clear, both John and a beautiful bird in (13) are "zero function arguments" and not Agents. Thus, the FG account of Achenese agreement claims that Achenese sentences like (13) will not have agreement. But this is incorrect. Lawler cites the pair: (14)

a.

Gopnyan gi-nging ani7 agam nyan. 3h 3h-see child male that 'He sees the boy.'

b.

Ani? agam nyan g-±-nging le-gopnyan 'The boy is seen by him.'

7 The FG analysis of (14a-b) is given in (15), and the RG analysis in (16-17): (15)

a.

gopnyan ant' agam nyan

b.

7 gopnyan ant agam nyan

327

RELATIONAL GRAMMAR AND FG (16)

(17)

Under the FG account, (11) and (15) together predict no agreement in (14), but there is agreement with gopnyan. This is exactly what is predicted under the RG account in (10) and (16). (14) is not the only example cited by Lawler which would be a State under an FG analysis, but which nonetheless manifests verb agreement: (18)

Gopnyan gi-dawa limo nyan ji-cu le-jih. 3h 3h-consider cow that 31-steal by-31 'He. considers the cow to be stolen by him^.

Gopnyan would be a zero-function argument here rather than an Agent under an FG analysis, but an initial 1 under the RG account. The latter correctly accounts for the agreement. Thus, the FG account proposed by Dik does not even account for Lawler's Achenese data. If we extend the data base beyond what Lawler cites, we find a large number of examples with agreement in what FG characterizes as States: (19)

(20)

a.

Adowa Ion ji-pi-tsaya cirita nyan. ygr sibling I 31-CAUS-believe story that 'My younger sibling believes that story'.

b.

Cirita nyan ji-pi-tsaya le adowa Ion. 'That story is believed by my younger sibling'.

a.

Gopnyan gi-musulet gr±-rasa le mahkamah. 3h 3h-lie 3h-feel by court-of-justice 'That she lied was felt by the court.'

b.

Gopnyan g±-musulet ji-rasa le anil gopnyan. 'That she lied was felt by her child.'

(21)

Gopnyan ji-banci le adowa-gi. 3h 31-hate by younger-sibling-her 'She is hated by her younger sibling.'

(22)

Gopnyan ji-sanka banyai le adowe gopnyan. 3h 31-regard dumb by younger-sibling 3h 'She is regarded as dumb by her younger sibling.'

g

328

PERLMUTTER (23)

Timpat gopnyan ji-pi-rusa' le pikara nyan. position 3h 31-undermine by matter that 'His position is undermined by that matter.'

(24)

Bubong nyan j'i-tumpang le tameh. roof that 31-support by columns 'The roof is supported by columns.'

(25)

G°pnyan gi-ingat. 3h 3h-remember 'He remembers.'

Since these sentences are predications of States according to Dik's typology, their arguments are zero-function arguments, and should not determine agreement in Achenese. But in each case there is agreement, and in each case it is determined by the initial 1 (a zero-function argument) and not by the initial 2 (also a zero-function argument, where present). This follows from the RG rule in (10). It is not clear how the FG rule in (11) could be patched up to account for this data. Now consider an example Dik would classify as a Process (uncontrolled event): (26)

Lon ji-pi-ingat gadoh gopnyan le haba. I 31-CAUS-remember disappearance 3h by message 'I was reminded of his disappearance by a message.'

About Process predications, Dik says (p. 37): "Two-place Process predications may have an argument which, though it does not control the Process, is nevertheless presented as an autonomous cause or instigator of it, as in The wind opened the door. To such an argument (here: the wind) we will assign the semantic function Force (Fo), with Huddleston 1970. The other argument will be given the Goal function." Haba 'message' in (26) would thus be a Force. Since it determines agreement on the verb, the FG agreement rule in (11) would have to be complicated in some way to account for cases where agreement is determined by a Force rather than an Agent. Since haba in (26) is the initial 1 in RG, (10) accounts for such sentences without any further complication. Now consider some further examples where agreement can be accounted for in terms of initial grammatical relations but not in terms of semantic functions. (27b) and (28b) are the stratal diagrams for (27a) and (28a), respectively.

RELATIONAL GRAMMAR AND FG (27)

a.

329

Gopnyan gi-tirimon surat nyan. 3h

3h-receive letter that

'He received the letter1.

Surat nyan gi-tirimon

le-gopnyan.

'The letter was received by him'. (28)

a.

Gopnyan ka gi-bre buku nyan ki-kano. 3h

PERF 3h-give book that to-us

' He gave the book to us'.

/H\

bre c.

gopnyan buku nyan

kamo

Buku nyan ka gi-bre ki-kamo 1 e-gopnyan. 'The book was given to us by him'.

The lhood of gopnyan in the (a)-sentences is confirmed by its appearance as a Passive chomeur in the (b)-sentences. Thus, the RG formulation of Achenese Verb Agreement in (10) accounts for the fact that agreement is determined by the initial 1 gopnyan

in these examples. The fact that gopnyan

is seman-

tically a Recipient in (27) is irrelevant. For the FG account of Achenese Verb Agreement in (11), however, (27) poses a serious problem. The structure of (27a) in FG terms would be:

330

PERLMUTTER

Since (11) specifies that the verb agrees with the Agent, and since neither gopnyan nor surat nyan is an Agent here, there should be no agreement in (27). But the verb agrees with gopnyan. To account for this in FG, it would be necessary to modify (11), stating that Recipients as well as Agents determine agreement. But then (28) would pose a problem. Here the verb does not agree with the Recipient kamo. Thus, (27) and (28) together pose a problem for the FG account of Achenese Verb Agreement. In brief, the FG account of Achenese verb agreement in (11), in conjunction with the FG analysis of predications in terms of Actions, States, Processes, and Positions, fails to account for the data. At best, the rule would have to be a complicated disjunction of the different semantic functions defined in the theory, since there are cases where Forces, Recipients, and zero-function arguments (at least) determine agreement. But even such a disjunction would not work, as has been pointed out, since some zero-function arguments determine agreement while others do not, and some Recipients do while others do not. In all cases, a formulation in terms of 'initial 1' accounts for the data. FG's failure to account for Achenese Verb Agreement is not due solely to the typology of the four types of predications and the definitions of their associated semantic functions. Even if one were to adopt a different class of semantic functions, including Experiencer, Cognizer, etc. [cf. Fillmore (1968)], a formulation in terms of those semantic notions would still fail to capture the generalization. The correct generalization requires reference to the notion 'initial 1', and that requires a theory 9 that recognizes grammatical relations at more than one level. This is what RG provides. FG's failure to account for Achenese Verb Agreement is a consequence of its failure to recognize grammatical relations at more than one level. Achenese Verb Agreement is not the only phenomenon stated in terms of initial grammatical relations in RG. The notion of initial 1 is used by Bell (1974, 1976, to appear), to state the conditions on Equi and reflexives in Cebuano, a Phillippine language. Harris (1976, 1981) uses it to state the condition on tav-reflexives in Georgian, as well as in a constraint on Inversion. Harris (to appear) argues that the notion of initial 1 is needed to state Verb Agreement in Udi, a Caucasian language. Aside from rules using the notion of initial 1, various other rules have been argued to involve reference to the initial level. Harris (1976, 1981) argues that reference to this level is necessary in the rules governing verb stem suppletion and

RELATIONAL GRAMMAR AND FG

331

the distribution of the preverbs da- and ga- in Georgian. OzkaragSz (1980) argues that the constraint governing -ArAk adverbial clauses in Turkish must refer to the initial level. Allen and Frantz (1978) and Allen (to appear) argue that the verb agreement rule in Southern Tiwa, a Tanoan language of New Mexico, refers to the initial level. Based on these results and the findings of Allen and Gardiner (1977) , Allen, Frantz, Gardiner, and Perlmutter (to appear) argue that initial grammatical relations figure crucially in the rules for verb agreement, the choice of agreement morphemes, noun incorporation, and the suppletion of certain verb stems in Southern Tiwa. Hubbard (1980) argues that the condition on antecedents of reflexives that are final terms in Albanian refers to initial grammatical relations. Perlmutter (to appear e) argues that reference to the initial level is also needed in the grammar of Luiseno, a Uto-Aztecan language of southern California. Since FG does not recognize grammatical relations at non-final levels, FG would not be able to state the relevant rules and generalizations in terms of initial grammatical relations, as in RG. It would therefore have to resort to statements in terms of semantic functions, as with Achenese Verb Agreement. The discussion of Achenese Verb Agreement in this paper has shown some of the differences between formulations in terms of 'initial 1' and 'Agent'i Similar differences exist between formulations in terms of other initial grammatical relations and semantic functions as defined in FG. Thus, the two frameworks' statements of the relevant rules in various languages will not be equivalent. A defender of FG must show that statement in terms of FG's semantic functions yields a better account of the data than formulation in terms of initial grammatical relations in RG. In sum, I have argued here that the FG account of Achenese Verb Agreement is flawed because FG does not recognize grammatical relations at non-final levels. The FG treatment of Achenese Agreement in terms of semantic functions fails to account for the data. The RG approach in terms of initial lhood, on the other hand, succeeds where the FG approach fails. Finally, I have indicated a number of other cases where RG uses initial grammatical relations. Investigation of these cases will reveal further empirical differences between the two frameworks. 3.2. Generalizations in terms of Grammatical Relations at More than One Level There are several types of generalizations that cannot be stated in FG because of its failure to recognize grammatical relations at more than one level.

PERLMUTTER

332

The first type of generalization at issue are those in terns of the notion 'nominal heading an n-arc', where n is a variable over R-signs (names of grammatical relations)• In such cases, a rule refers to nominals bearing a certain grammatical relation, but is does not matter in which stratum or strata it does so. I illustrate this type of generalization with the condition governing possible antecedents of reflexives in Russian. The form of argument is in some respects similar to that in ) 3.1. First I show how RG captures the generalization in terms of lhood because it posits more than one syntactic level. Next I show that FG cannot capture it in terms of its notion of Subject because it posits subjecthood at only one level. Then I argue that FG cannot capture the generalization in terms of semantic functions either. As argued in Perlmutter (to appear c), three rules of Russian are sensitive to final lhood. First, only final Is can be Equi victims. Second, verb agreement is determined by the final 1 of the clause. Third, a nominal that is the final 1 of the highest clause in which it heads a central GR arc is in the nominative case. Thus, nominals which cannot be Equi victims, do not determine verb agreement, and are not in the nominative case cannot be final Is. Although only Is can antecede reflexives in Russian, in some cases the antecedent is not a final 1. This is crucial to the argument. The argument is based on a necessary (but not sufficient] condition for a nominal to be the antecedent of a reflexive: (30)

Necessary Condition on Antecedents of Reflexives in Russian Only a nominal heading a 1-arc can serve as antecedent of a reflexive.

This condition governs both antecedents of the reflexive pronoun sebja (and its case variants sebe and soboj), and the reflexive possessive svoj where svoj requires an antecedent. The reflexive pronouns are inflected for case but not for gender or number. The reflexive possessive svoj is adjectival in form, agreeing with the head in gender, number, and case. The condition in (30) is formulated in Perlmutter (to appear c), where some qualifications are discussed.10 It builds on the work of Peskovskij (1956), Klenin (1974), Chvany (1975), and Timberlake (1979). Some of the examples cited here are taken from these sources.

RELATIONAL GRAMMAR AND FG

333

(30) accounts for the non-ambiguity of:** (31)

Ja rasskazal Borisu anekdot o sebe. joke about REFL I told DAT 'I told Boris a joke about myself.'

Ja can antecede the reflexive because it heads a 1-arc in the associated HN; Borisu cannot because it heads a 3-arc but not a 1-arc. A nominal heading only a 2-arc cannot antecede a reflexive: (32)

Anna otpravila rebenka k svoim roditeljam. NOM sent child/ACC to REEL'S parents

Anna qualifies as antecedent because it is a 1; rebenka does not because it is a 2. The condition cannot be stated in terms of lineair order, as the following examples shew: (33)

Anekdoty o sebe nam rasskazyvali tol'ko inostrancy. jokes about REFL us/DAT told only foreigners 'Only foreigners told us jokes about themselves.'

(34)

Borisu rasskazyval anekdot o sebe tol'ko Sergej. DAT told joke about REFL only NOM 'Only Sergej^ told Boris^ a joke about himself^'.

In (33), both nam and inostrancy follow the reflexive. Only inostrancy can be the antecedent, for it heads a 1-arc; nam heads only a 3-arc. In (34), Borisu cannot be the antecedent of sebe although it precedes it because it heads a 3-arc but not a 1-arc in the associated RN. Sergej is the antecedent, although it follows the reflexive, because it heads a 1-arc. (35) contrasts with (32): (35)

Rebenok byl otpravlen k svoim roditeljam. child was sent to REFL's parents 'The child was sent to his parent.'

In (32) rebenka is not a possible antecedent because it is a 2 but not a 1. In (35), however, rebenok is a 1 in the second stratum:

PERLMÜTTER

334 (36)

Contrasts like that between (32) and (35) show that the condition on antecedents of reflexives cannot be stated in terms of semantic functions; this will be crucial in our discussion of the FG approach to this condition. In all

the examples cited so far, the antecedent of the reflexive is a

final 1. However, there are two types of cases where the antecedent of a reflexive heads a 1-arc, but not in the final stratum. First, as Klenin (1974) shows, there are cases where what in relational terms is a Passive chomeur is antecedent of a reflexive: (37)

a.

Eta kniga byla kuplena Borisom dlja sebja. this book was bought INSTR for REFL 'This book was bought by Boris for himself.'

In (37), the antecedent is an initial 1 but not a final 1. The second type of example comes from Inversion clauses, in which the initial 1 is a final 3. The evidence for Inversion in Russian is given in Perlmutter (to appear c), and will not be repeated here. The following example is relevant:

(38)

a. Mne nuzna bolee udobnaja kvartira me/DAT needs more comfortable apartment ne dlja sebja, a djla svoej sem'i. NEG for REFL but for REFL's family 'I need a more comfortable apartment not for myself, but for my family.'

335

RELATIONAL GRAMMAR AND FG b. V/l / 2 I

/ f

p nu&ia

^ mne

2 J

X

Ben^X.

»

sebja

bolee udobnaja kvartira

The RN for (38a) includes the subnetwork in (38b). The evidence is that the antecedent of the reflexive in (38a), like that in (37a), is not a final 1 comes from three things: it is not in the nominative case, it does not determine verb agreement, and it cannot be an Equi victim. It qualifies as antecedent of a reflexive because it heads an initial-stratum 1-arc, even though it does not head a final 1-arc. Additional evidence for its initial lhood comes from its ability to control Equi in gerund constructions. In many Inversion clauses, the initial 2 advances to 1 and heads a final 1-arc. Advancees to 1 in Inversion clauses can also antecede reflexives: (39)

a.

Boris ne nuzen svoim detjam. NOM NEG needs REFL's children/DAT 'Boris's children don't need him'.

In (39) Boris, the initial 2 and final 1, is the antecedent of the reflexive svoim.

Reviewing the data on Russian reflexives presented here, we see that there is no single stratum in which a nominal must head a 1-arc in order to qualify as antecedent of a reflexive; it qualifies as long as it heads a 1-arc in some stratum. This is captured by the formulation in (30) . The necessary condition on antecedents of reflexives in Russian is thus a condition stated in RG in terms of the notion 'nominal heading a 1-arc', with no further specification. This condition captures the generalization uniting the various nominals that can be antecedents of reflexives.

336

PERLMUTTER

Now let us attempt to state the generalization in FG. We will consider two ways of stating it: in terms of grammatical relations and in terms of semantic functions. Since FG does not recognize grammatical relations at more than one level, it cannot state the generalization in terms of grammatical relations. While most antecedents of Russian reflexives are subjects in FG terms, not all of them are. The crucial examples are here (37a) and (38a). The antecedents of reflexives in these examples are in oblique cases, they do not determine verb agreement, and they cannot be Equi victims [Perlmutter (to appear c)J. In RG, this means that they are not final Since FG in effect recognizes only final

Is.

subjecthood, and does not posit

grammatical relations at non-final levels, this means that they cannot be subjects at all in FG. As a result, the condition on antecedents of reflexives cannot be stated in terms of subjecthood in FG. The other possibility in FG is to state the condition in terms of semantic functions. At best, such a statement would have to refer to a clumsy disjunction of semantic functions. Some antecedents are Agents, as in (31-34) and (37) . Sane are Positioners: (40)

Boris derzal rebenka v svoix rukax. NOM held child/ACC in REFL's arms 1 Boris held the child in his arms'.

And, as will be seen below, some antecedents are Goals, some are Recipients, and some are zero-function arguments. But the need to resort to a clumsy disjunction is not the worst problem confronting an FG attempt to state the generalization in terms of semantic functions. Such a statement simply will not work. First consider what can antecede a reflexive in State predications. Recall that "two-place State predications will most often have two zerofunction arguments, and there may even be three-place State predications with three zero-function arguments. A second argument in State predications may, however, also have Time or Location

function."

[Dik (1978:38)]

No other semantic functions for State predications are given. We must therefore assume that where a State predication has two arguments that are not Time or Location, they are both zero-function arguments. If the condition on antecedents of Russian reflexives is to be stated in terms of semantic functions, then the two zero-function arguments in State predications should behave alike. But they do not:

RELATIONAL GRAMMAR AND FG (41)

a.

Boris o£en' ljubit svoix detej. NOM very loves REFL's children/ACC 'Boris loves his children very much'.

b.

*Svoi deti ocen" ljubjat Borisa. REFL's children/NOM very love ACC 'REFL's children love Boris very much'.

337

This data is accounted for by the RG condition in (30); Boris in (41a) is a 1, but Borisa in *(41b) is not, so only the former can antecede a reflexive. Linear order, already shown to be irrelevant in (33-34), is irrelevant here as well. If (41a) is rearranged so that the reflexive precedes the antecedent, the result is still grammatical:

(42)

Svoix detej Boris oien" ljubit. 'His own children Boris loves very much'•

Rearranging *(41b) so that the putative antecedent precedes the reflexive only yields ungrammatical sentences: (43)

a.

*Borisa svoi deti oien' ljubjat.

b.

*Borisa ocen' ljubjat svoi deti.

In sentences with two zero-function arguments, subjecthood is crucial for anteceding reflexives. This can also be seen in: (44)

Tanja uvazaet Marinu nesmotrja na svoe obsiee NOM respects ACC despite REFL's general prezrenie k zen§£inam. contempt to women 'Tanja. respects Marina^ despite her. general contempt for 3 women'?

The possessive reflexive svoe in (44) can refer only to Tanja, not to Marinu. Since this is a State predication, both are zero-function arguments. But only Tanja heads a 1-arc in the associated RN; Marinu heads only a 2-arc. The examples cited above show that a formulation of the condition on antecedents of reflexives in terms of the notion 'zero-function argument' will not work. The same holds for Goals, which Dik (p. 38) defines as follows: "If an Action predication has a second argument, this will typically designate the entity to which the Action is applied by the Agent. To this

PERLMUTTER

338

entity we will assign the semantic function Goal (Go) ." Goals antecede reflexives in (45a) and (46a), but they cannot do so in *(45b) and *(46b): (45)

a.

b.

(46)

a.

b.

Boris byl ubit svoim mladsim bratom. NOM was killed REFL's younger brother/INSTR 'Boris was killed by his younger brother.' *Svoj mladsij brat ubil Borisa. REFL's younger brother/NOM killed ACC 'REFL's younger brother killed Boris.' Rebenok byl otpravlen domoj svoej mater'ju. child/NOM was sent home REFL's mother/INSTR 'The child was sent home by his mother.1 *Svoja mat' otpravila rebenka domoj. REFL's mother sent child/ACC home 'REFL's mother sent the child home.'

The difference, of course, is that (45a) and (46a) are passive, so that in RG the Goal heads a 1-arc in the second stratum. *(45b) and *(46b) are active, so that, in RG, the Goal heads only a 2-arc. Thus, (45a) and (46a) are grammatical, while *(45b) and *(46b) are not. But in terms of semantic functions, there is no difference between (45a/46a) and *(45b/46b). An FG account of the condition on antecedents of Russian reflexives in terms of semantic functions would fail to account for this data. A similar argument is possible for cases where the antecedent of a reflexive has the semantic function of Recipient: (47)

a.

V to vremja ja polucal mnogo pisem ot at that time I received many letters from svoix druzej. REFL's friends 'At that time I received many letters from my friends.'

b.

*V to vremja svoi druz'ja mne posylali at that time REFL's friends me/DAT sent mnogo pisem. many letters 'At that time REFL's friends sent me many letters."

In both sentences, the first person nominal has the semantic function of Recipient. But in (47a) it heads a 1-arc, while in *(47b) it heads only a 3-arc. And only in the former can it antecede a reflexive. Similarly:

339

RELATIONAL GRAMMAR AND FG (48)

Ja peredam eto pis'mo Borisu dlja svoix druzej. I give this letter DAT for REFL's friends 'I will give this letter to Boris for my friends.'

The reflexive in (48) can refer only to ja,

because only ja heads a

1-arc. It cannot refer to the Recipient Borisu,

which heads only a 3-arc.

An FG formulation of the condition on antecedents of reflexives in terms of semantic functions will not distinguish such cases. In brief, an FG attempt to state the condition on antecedents of Russian reflexives in terms of semantic functions fails. FG cannot use its notion of subjecthood to state the condition because it recognizes subjects at only one level, which corresponds to the final level in RG. Thus, an FG statement in terms of subjecthood would not account for examples like (37) and (38), where the antecedent is an initial 1 but not final 1 in RG. FG's inability to state the generalization is a direct consequence of its failure to posit grammatical relations at more than one level. The condition on antecedents of Russian reflexives is not an isolated example. A number of other generalizations in grammars refer to the notion 'nominal heading an

n-arc', with no further specification. The condition

on controllers of gerund constructions in Russian, discussed in Perlmutter (to appear c), has exactly the same form. So do the generalizations governing nominative case marking and antecedents of reflexives in Choctaw, a Muskogean language of Missisippi and Oklahoma [Davies (this volume)]. Klokeid (to appear) argues that three generalizations in Nitinaht (a Wakashan language of British Columbia) have the same form: those governing pronominal encliticization, nominative preposition marking, and preposition 12

suppression in questions.

There are also generalizations stated as con-

junctions of two conditions of this form. Hubbard (1979, 1980) argues that the condition governing non-active verbal morphology in Albanian references the notion 'nominal heading a 1-arc and a 2-arc', and Perlmutter (in preparation b) argues that the condition governing the perfect auxiliary essere in Italian references the notion 'nominal heading a 1-arc and an object arc'.^ All of these generalizations have the same property that makes (30) unstatable in FG: they refer to grammatical relations at more than one level. FG's failure to recognize more than one syntactic level makes it unable to capture such generalizations.

340

PERLMUTTER Generalizations stated in terms of the notion 'working 11 defined and

exemplified in Perlmutter (1979) are also unstatable in FG. A nominal is a 'working 1' if it heads a 1-arc and a final term (1, 2, or 3) arc. Thus, in Passive clauses the initial 2 is a working 1 by virtue of the fact that it is a final 1. In Inversion clauses, where a 1 (usually the initial 1) is a final 3, this nominal also qualifies as a working 1. Such Inversion nominals prevent FG from capturing the relevant generalizations. For example, as argued in Perlmutter (1979, to appear d), only a working 1 can control the consecutive da + infinitive construction in Italian. Inversion nominals cannot be treated as subjects in FG because they behave differently from subjects with respect to case, verb and adjective agreement, Equi, and Quantifier Float; this is why they are treated as final 3s in RG. But they can control the consecutive da + infinitive construction by virtue of the fact that they are initial Is. But since FG does not recognize non-final subjecthood, it cannot state the generalization in terms of subjecthood. It is then left with the semantic functions. But attempts to state the generalization in terms of semantic functions result in the same kinds of difficulties that arose in connection with the condition on antecedents of Russian reflexives; some Goals are possible controllers while others are not, some zero-function arguments are while others are not, and so on. An argument of exactly the same form against the FG approach can be constructed for generalizations stated in terms of the notion 'working 1'. Further , FG cannot state generalizations in terms of the notions 'acting 1', 'acting 2', etc. This class of notions is defined in Perlmutter and Postal (to appear c). These notions are used where a nominal behaves like a final chomeur rather than a final term, but behaves like a term with respect to certain phenomena. The behavior as a final non-term would rule out an FG-analysis in which the nominal in question is a subject or object. Then there is no notion of non-final grammatical relation to capture the fact that such nominals behave like terms with respect to certain phenomena. These notions of acting termhood are used in Perlmutter and Postal (to appear c), to account for pronominal clitics in Welsh, and for case assignment and object agreement in Maasai, a Nilotic language of Kenya. Jackson (1980) uses the notion 'acting 1' to account for the reflexive-reciprocal morpheme -koL- in Tamil. Marlett (to appear) and Tohsaku (in preparation) use the notion of 'acting 2' to account for data in Seri (a Hokan language of northern Mexico) and Ainu, respectively.

RELATIONAL GRAMMAR AND FG

341

There are also other generalizations for whose statement more than one syntactic level is needed. One of these is the 1-Advancement Exclusiveness Law of Perlmutter and Postal (to appear b). Another is the condition on sereflexives in Clause Union constructions in Romance discussed by Raposo (to appear). In ^ 3.2 I have indicated a number of generalizations in

terms of gram-

matical relations at more than one level. Because it recognizes grammatical relations at only one level, FG cannot capture such generalizations.^

I

have used the condition on antecedents of Russian reflexives to illustrate what happens if FG attempts a statement in terms of semantic functions. These are not the right notions for the data. Parallel problems would arise for the other generalizations mentioned here which require grammatical relations at more than one level. 4. Some Further Inadequacies of FG 4.1. Dummies Dummy nominals such as those italicized below have figured prominently in the linguistic literature: (49)

(50)

English a.

There is a daffodil under the pillow.

b.

It was obvious that he was guilty.

c.

I figured it out that he was lying.

d.

It's amazing the things he's getting away with.

e.

It1s raining.

French a.

II est arrive trois inspecteurs de produits laitiers.

b.

Il est évident qu'il est coupable.

c.

C'est étonnant le nombre de gens qui croient cela.

'Three dairy product inspectors arrived.' 'It is obvious that he's guilty.' 'It's amazing the number of people who believe that.' (51)

Dutch a.

Br spelen twee kinderen in de tuin. 'There are two children playing in the garden.1

PERLMUTTER

342 b.

Er wordt door de kinderen op het ijs geschaatst. "It is skated by the children on the ice." Het is niet zeker dat het al te laat is. 'It is not certain that it is already too late.'

(52)

Welsh a.

Yr oedd hi yn bwrw glaw ddoe. was

she

throw rain yesterday

'It was raining yesterday.' (53)

Hungarian Janos azt mondta, hogy visszajon. it

said

that return

1

Janos said that he would return.'

In certain cases, such as that of there in English sentences like (49a), evidence was amassed in transformational frameworks showing that this there behaves like an NP, and specifically like a subject; for the main arguments, see Perlmutter and Soames (1979:42-45, 412-14, and 464-65). For Dutch, Perlmutter and Zaenen (to appear) give arguments based on linear position in both main and subordinate clauses that er in (51) is a subject. They base further arguments on case and agreement in sentences like (51a) in Dutch and the corresponding sentences of German. Dummy nominals behave syntactically like subjects or objects, but unlike most other nominals, they do not correspond to arguments in semantic representations.*^

In RG this is captured by means of RNs in which dummies

do not head arcs in initial strata. For example, the RN corresponding to (51a) can be abbreviated as the stratal diagram: (54)

spelen twee kinderen de tuin

er

RELATIONAL GRAMMAR AND FG

343

In a theory in which the correspondences between nomináis' semantic roles and grammatical relations are stated in terms of initial-stratum grammatical relations, the fact that dummies systematically lack the former is reflected in the fact that they systematically lack the latter as well. FG ignores the possibility that dummies can be subjects or objects. It attempts a definition of the notion of subject in terms that make dummy subjects impossible: "Our definition of the subject function is basically a semantic one: we defined the Subject as that constituent which refers to the entity which is taken as a point of departure for the presentation of the state of affairs in which it participates."

Dik (1978:87)

This defi-

nition cannot be reconciled with the existence of dummy subjects, as in most of the examples in (49-52). The existence of dummies (including dummy subjects) points up a basic inadequacy of FG. Like the need for multiple levels in syntax, dummies show that a theory of syntax cannot limit itself to semantic functions and the assignment of a single grammatical relation to nomináis with specific semantic functions. 4.2. The 3-Relatiori FG recognizes only two grammatical relations: subject and object. Dik (1978:18) makes this explicit: "As far as syntactic functions are concerned, Subject and Object are sufficient to take care of the relevant differences between linguistic expressions which express the same predication." FG thus assumes that the traditional notions of direct and indirect object can be subsumed under a single relation. However, evidence from numerous languages can be cited to show: (55)

a.

a 3-relation must be set up distinct from the 1- and 2-relations because 3s behave differently from Is and 2s with respect to certain phenomena

b.

the 3-relation must be regarded as a member of the class of term relations, alongside the 1- and 2-relations; these form a class apart from the various non-term relations.

Although evidence of the kind required could be cited from diverse languages, I limit myself here to Harris's (1976, 1981) evidence from Georgian. First, there is ample evidence of type (55b), since a number of generalizations

PERLMUTTER

344

refer to the class of terms (Is, 2s, and 3s); (56)

a.

Final terms determine Person Agreement on the verb.

b.

Final first and second person terms determine Number Agreement on the verb.

c.

Final terms can be antecedents of tavisi-reflexives.

d.

A nominal must be a final term to qualify as antecedent of a tav-reflexive (a necessary but not sufficient condition).

e.

Unemphatic pronouns corresponding to final terms do not appear overtly.

f.

There are case marking rules referring to terms bearing one of the "retirement relations" chomeur and Emeritus.

Next, 3s behave differently from both Is and 2s in a number of ways: (57)

a.

They are marked with different cases.

b.

They determine different person agreement markers on the verb.

c.

The marking of retired 3s is different from that of retired Is and 2s.

Finally, 3s behave differently from 2s in a number of ways: (58)

a.

2s but not 3s can advance to 1 in Passive clauses.

b.

Object Raising is possible for 2s but not 3s.

c.

Some verbs have suppletion for number or animacy of the initial 2, but not for the initial 3.

d.

The Object Camouflage construction occurs where the 2 is first or second person; it is oblivious to 3s.

e.

3s but not 2s can be put en chomage in the Inversion construction. 3s but not 2s can be put en chomage in Version constructions.

f.

It might be thought that FG can account for differences in behavior between 2s and 3s by using the semantic function of Recipient where RG uses the 3-relation. However, this is not the case. As Harris (1976, 1981) makes clear, there are three classes of 3s in Georgian that are not semantically Recipients: 3s in Clause Union causatives, 3s in the Inversion construction, and advancees to 3 in Version constructions. These are all noninitial 3s. They preclude the possibility of capturing 3hood in terms of the semantic function of Recipient.

345

RELATIONAL GRAMMAR AND FG

In brief, a theory like FG that does not distinguish between 2s and 3s cannot even account for the data internal to Georgian. Additional languages would provide further evidence for distinguishing the two object relations and for treating the three term relations as a class. 4.3. The Chômeur

Relation

The chômeur relation posited in RG has no real counterpart in other frameworks. The Motivated Chômage Law of Perlmutter

and Postal (1977, to appear

a) limits the chômeur relation to nominals bearing a term relation in an earlier s t r a t u m . T h u s , a notion equivalent to that of chômeur can only be posited in a framework that recognizes more than one syntactic level. Crucial for FG are cases where a rule must refer to the chômeur relation. I will cite three such cases here. First, Chung (1976) shows that in Indonesian, quantifiers can appear displaced from their nominals. Quantifier Shift is possible for a wide variety of nominals (both final terms and non-terms), but not for chômeurs, in either the Passive or 3-2 Advancement constructions. The restriction can be stated as follows in RG:

(59)

Quantifier Shift is not possible for a nominal heading a Cho arc.

In FG, on the other hand, there is no notion of chômeur. FG cannot use its notions of subject and object to state the restriction, since the nominals in question are not final terms and therefore would not be subjects or objects in FG. One cannot have the restriction refer to nominals that are not subjects or objects, because Obliques (which are not subjects or objects in FG terms) allow Quantifier Shift. FG cannot use semantic functions to characterize the class of nominals disallowing Quantifier Shift, because there are pairs of sentences (active and passive, on the one hand, and with and without 3-2 Advancement, on the other) whose nominals have the same

seman-

tic functions, but which differ with respect to whether or not they allow Quantifier Shift. This is captured by (59) in RG, but cannot be in FG. Second, Seiler and Frantz (to appear) show that the liupiat dialect of Eskimo has several advancements to 2, and that in each case the 2 put en chômage by the advancement appears in the instrumental case. The generalization is easily stated in RG in terms of the notion 12-chomeur' (a nominal heading a 2-arc in one stratum and a Cho arc in the next):

346

PERLMUTTER (60)

A 2-chomeur is marked with instrumental case.

The notion '2-chomeur', which is crucial to (60), inherently involves reference to more than one syntactic level. FG is therefore unable to state the generalization. Third, Harris (1976, 1979, 1981) shows that Georgian has case marking rules for "retired terms", that is, nominals bearing both a term relation and a "retirement relation" (chomeur or Emeritus). In RG, the nominals in question bear the two relations at distinct levels. As in the case of Indonesian Quantifier Shift pnd Inupiat Instrumental Marking, FG cannot state the generalization. Since the nominals in question do not behave like final terms, they cannot be assigned the subject or object relations in FG. And the semantic functions provide no basis for characterizing the relevant class of nominals. Since no nominal can head a Cho arc in the initial stratum,^ the chomeur relation figures only in structures involving more than one syntactic level. The generalizations captured by rules referring to the chomeur relation cannot be captured in FG or other frameworks positing only one syntac tic level.

5. Conclusions FG and RG have enough in common to make comparison possible. The major issue dividing them concerns multiple syntactic levels: RG posits grammatical relations at levels other than final, whereas FG does not. The issue is whether there are generalizations in natural languages that make it necessary to posit non-final syntactic levels. In ^ 3 . 1 it is argued that Achenese Verb Agreement refers to subjecthood in the initial stratum, and in ^ 3.2 it is argued that the condition on antecedents of reflexives in Russian specifies the notion 'nominal heading a 1-arc', with no specification of stratum. Neither the syntactic nor semantic functions posited in FG make it possible to state the generalizations. Achenese Verb Agreement and Russian reflexives are representative of two classes of phenomena requiring multiple syntactic levels; other cases presently known are indicated. ^ 4 briefly mentions three further inadequacies of FG: its failure to recognize that dummies can be subjects or objects, its failure to distinguish between direct and indirect objects, and its failure to posit the chomeur relation.

RELATIONAL GRAMMAR AND FG

347

NOTES *I am indebted to Valentina Forskaja and T. Iskandar for intuitions on sentences of Russian and Achenese, respectively. Teun Hoekstra also helped with the Achenese data. I am also indebted to Sarah Bell, Simon Dik, Donna Gerdts, and Geoffrey Pullum for comments on the first draft of this paper. Errors and shortcomings are my own. This work was supported in part by the National Science Foundation through grant No. BNS78-17498 to the University of California, San Diego and an I.W. Killam Senior Fellowship at the University of British Columbia. *The possibility is not excluded in either framework that these rules might, in particular cases, be sensitive to other things as well. Dik (1978) makes explicit claims concerning the sensitivity of syntactic rules to other types of information. i Explanations of the basic notions of RG, terminology, and relational networks (RNs) and stratal diagrams are given in Perlmutter and Postal (1977, to appear a) and Perlmutter (1980, to appear d). See also Davies (this volume). 3 The status of the watch in (2b-c) in FG is not clear. There is apparently no principled reason either for assigning or for not assigning this nominal the 2-relation in these sentences. «The glosses '3h' and '3b' indicate third person nominals designating individuals higher and lower in status than the speaker, respectively. The nominal determining agreement is italicized in each example. 5 Dik (p. 117) formulates the RG rule differently. Dik's account of RG contains a number of misunderstandings and misconceptions. Some of these are due to changes in the framework since the earliest presentations in e.g. Perlmutter and Postal (1974) and Botha (1974), while others are misunderstandings even of the earliest work. The reader can get a more accurate picture of RG from the papers in Perlmutter (to appear a, b). 6

A number of problems concerning Achenese Verb Agreement remain. See Perlmutter (to appear d) for discussion. "^Dik's decision to have two or three zero-function arguments in State predications leads to serious problems for FG. In an Action predication with an Agent and Goal, it is possible to give straightforward rules for active and passive morphology: if the Agent is Subject, the clause has active morphology, and if the Goal is Subject, it has passive morphology. But in State predications with two (or more) zero-function arguments, how does the grammar account for the morphological contrast between an active and the corresponding passive? For example, consider the English glosses of (14a-b), whose structure in FG terms would be like (15). What determines which sentence is active and which is passive? This is obviously related to the question of accounting for the semantics of such sentences. If both nominals have the same semantic function in FG, how does the grammar express the distinction between the one that sees and the one that is seen? This semantic problem will arise for all languages. The problem of accounting for the contrast between active and passive morphology arises for all languages that have such a morphological distinction. g The transcription used here differs from that in Lawler (1977) in a few details.

348

PERLMUTTER

For this reason, the Achenese agreement data also provides an argument against Fillmore's (1968) theory of passivization, as well as against other frameworks that fail to recognize more than one syntactic level. This matter is discussed in Perlmutter (1978b, in preparation a). ^This condition is also discussed in Perlmutter (1978b, to appear d, in preparation a). ^ I n the examples both reflexives and their antecedents are italicized. 12

As Klokeid notes, the relevant rules of Nitinaht could be stated in terms of either of two notions: "nominal heading a 1-arc' or 'acting l' (discussed briefly below). 1 ^'2' and '3' are collectively called object R-signs. For further discussion of the Italian data, see Perlmutter (in preparation b) and Rosen (1981).

14 A number of other current syntactic frameworks posit only one syntactic level and consequently have essentially the same difficulties as FG. This is discussed in Perlmutter (1978b, to appear b, in preparation a). is also true of so-called "pronominal copies." 16 Keenan (1975) and Comrie (1977) have proposed analyses of impersonal passives that violate the Motivated Chomage Law. For arguments supporting an alternative analysis and hence the Motivated Chomage Law itself, see the papers in Perlmutter (to appear b).

^This is a consequence of the Motivated Chomage Law.

REFERENCES ALLEN, Barbara J. (to appear) "Goal Advancement in Southern Tiwa." ALLEN, Barbara J. and Donald G. FRANTZ 1978 "Verbal Agreement in Southern Tiwa," Proceedings of the Fourth Annual Meeting of the Berkeley Linguistic Society (University of California, Berkeley). Revised version in Perlmutter (to appear a). ALLEN, Barbara J., Donald G. FRANTZ, Donna B. GARDINER, and David M. PERLMUTTER (to appear) "Possessor Ascension and Syntactic Levels in Southern Tiwa," in Perlmutter (to appear a). ALLEN, Barbara J. and Donna B. GARDINER 1977 "Noun Incorporation in Isleta," Work papers of the Summer Institute of Linguistics, University of North Dakota, Vol. 21. BELL, Sarah J. 1974 "Two Consequences of Advancement Rules in Cebuano," Papers from the Fifth Annual Meeting of the North Eastern Linguistic Society (Harvard University, Cambridge).

RELATIONAL GRAMMAR AND FG

349

BELL, Sarah J. 1976 Cebuano Subjects in Two Frameworks, Unpublished doctoral dissertation, MIT. Distributed by Indiana University Linguistics Club, Bloomington, Indiana. BELL, Sarah J. (to appear) "Advancements and Ascensions in Cebuano," in Perlmutter (to appear a). BOTHA, R.P. 1974 Kursus oor Perlmutter en Postal se Relasionele Grammatika (Studiegroep vir taalwetenskap, Universiteit van Pretoria). CHUNG, Sandra 1976 "An Object-Creating Rule in Bahasa Indonesia", Linguistic Inquiry 7, 41-87. CHVANY, Catherine 1975 On the Syntax of BE-Sentences in Russian, Slavica Publishers, Cambridge. COLE, Peter and Jerrold SADOCK (eds.) 1977 Syntax and Semantics 8: Grammatical Relations, Academic Press, New York. COMRIE, Bernard 1977 "In Defense of Spontaneous Demotion: The Impersonal Passive," in Cole and Sadock (1977). DAVIES, William (this volume) "Choctaw Subjects and Multiple Levels of Syntax." DIK, Simon 1978 Functional Grammar, North-Holland Publishing Co., Amsterdam, New York, and Oxford. FILLMORE, Charles J. 1968 "The Case for Case," in Emmon Bach and Robert Harms (eds.) Universals in Linguistic Theory, Holt, Rinehart, and Winston, New York. HARRIS, Alice 1976 Grammatical Relations in Modern Georgian, Unpublished doctoral dissertation, Harvard University. HARRIS, Alice 1979 "Retired Term Marking in Old Georgian," in Paul R. Clyne, William F. Hanks, and Carol L. Hofbauer (eds.), The Elements: A Parasession on Linguistic Units and Levels (Chicago Linguistic Society, University of Chicago) . HARRIS, Alice 1981 Georgian Syntax : a Study in Relational Grammar, Cambridge University Press, Cambridge. HARRIS, Alice (to appear) "Case Marking, Verb Agreement, and Inversion in Udi," in Perlmutter (to appear a).

350

Perlmutter

HOEKSTRA, Teun 1978 "Funktionele grammatika," Forum der Letteien 19, 293-312. HUBBARD, Philip L. 1979 "Albanian Neapolitan Morphology: Passive, Multi-Attachment, and the Onaocusative Hypothesis," Linguistic Notes from La Jolla, No. 6. HUBBARD, Philip L. 1980 The Syntax of the Albanian Verb Complex, Unpublished doctoral dissertation, University of California, San Diego. HUDDLESTON, Rodney 1970 "Remarks on Case-Grammar," Linguistic Inquiry 1, 501-511. JACKSON, Charles H. 1980 "Relational Grammar versus Lexicalist Syntax: The Issue of Levels of Syntax in Tamil" (Unpublished paper. University of California, San Diego). KEENAN, Edward L. 1975 "Some Universals of Passive in Relational Grammar," Papers from the Eleventh Regional Meeting of the Chicago Linguistic Society (University of Chicago). KLENIN, Emily R. 1974 Russian Reflexive Pronouns and the Semantic Roles of Noun Phrases in Sentences, Unpublished doctoral dissertation, Princeton University. KLOKEID, Terry J. (to appear) "Levels in Syntax: Evidence from Nitinaht." LAWLER, John M. 1977 "A Agrees with B in Achenese: A Problem for Relational Grammar," in Cole and Sadock (1977). MARLETT, Stephen A. (to appear) "Personal and Impersonal Passives in Seri," in Perlmutter (to appear b). OZKARAGOZ, Inci 1980 "Evidence from Turkish for the Unaccusative Hypothesis," Proceedings of the Sixth Annual Meeting of the Berkeley Linguistic Society (University of California, Berkeley). PERLMUTTER, David M. 1978a "Impersonal Passives and the Unaccusative Hypothesis," Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society (University of California, Berkeley). PERLMUTTER, David M. 1978b "Empirical Evidence Distinguishing Some Current Approaches to Syntax," Colloquium lecture presented at the annual meeting of the Linguistic Society of America, Boston. PERLMUTTER, David M. 1979 "Working Is and Inversion in Italian, Japanese, and Quechua," Proceedings of the Fifth Annual Meeting of the Berkeley Linguistics Society (University of California, Berkeley).

RELATIONAL GRAMMAR AND FG

351

PERLMUTTER, David M. 1980 "Relational Grammar," in Edith Moravcsik and Jessica Wirth (eds.) Syntax and Semantics 13: Current Approaches to Syntax, Academic Press, New York. PERLMUTTER, David M. (ed.) (to appear a) Studies in Relational Grammar. PERLMUTTER, David M. (ed.) (to appear b) Passivization and Related Issues. PERLMUTTER, David M. (to appear c) "Evidence for Inversion in Russian and Kannada," in Perlmutter (to appear a). PERLMUTTER, David M. (to appear d) "Syntactic Representation, Syntactic Levels, and the Notion of Subject," in Pauline Jacobson and Geoffrey Pullum (eds.) The Nature of Syntactic Representation, D. Reidel Publishing Co., Dordrecht and Boston. PERLMUTTER, David M. (to appear e) "Possessor Ascension and Some Relational Laws," in Perlmutter (to appear a). PERLMUTTER, David M. (in preparation a) "Empirical Evidence Distinguishing Some Current Approaches to Syntax: The Subjecthood of Passive Chomeurs." PERLMUTTER, David M. (in preparation b) "Multiattachment and the Unaccusative Hypothesis: The Perfect Auxiliary in Italian." PERLMUTTER, David M. and Paul M. POSTAL 1974 Lectures on Relational Grammar, Summer Linguistic Institute of the Linguistic Society of America, University of Massachusetts at Amherst. PERLMUTTER, David M. and Paul M. POSTAL 1977 "Toward a Universal Characterization of Passivization," Proceedings of the Third Annual Meeting of the Berkeley Linguistics Society (University of California, Berkeley). PERLMUTTER, David M. en Paul M. POSTAL (to appear a) "Some Proposed Laws of Basic Clause Structure," in Perlmutter (to appear a). PERLMUTTER, David M. and Paul M. POSTAL (to appear b) "The 1-Advancement Exclusiveness Law," in Perlmutter (to appear b) . PERLMUTTER, David M. and Paul M. POSTAL (to appear c) "Impersonal Passives and Some Relational Laws," in Perlmutter (to appear b). PERLMUTTER, David M. and Scott SOAMES 1979 Syntactic Argumentation and the Structure of English, University of California Press, Berkeley and Los Angeles.

352

PERLMUTTER

PERLMUTTER, David M. and Annie ZAENEN (to appear) "The Indefinite Extraposition Construction in Dutch and German," in Perlmutter (to appear b). PESKOVSKIJ, Aleksandr M. ? t 1956 Russkij sintaksis v naucnom osvescenii , Gosudarstvennoe ucebnopedagogiceskoe izdatel'stvo Ministerstva ProsvelSenija RSFSR, Moskva. RAPOSO, Eduardo P. (to appear) "The Interaction of Clause Union and SE-Reflexivization in Romance." ROSEN, Carol 1981 The Relational Structure of Reflexive Clauses: Evidence from Doctoral dissertation. Harvard University.

Italian,

SEILER, Wolf and Donald G. FRANTZ (to appear) "The Instrumental Case in Inupiat." TIMBERLAKE, Alan 1979 "Reflexivization and the Cycle in Russian," Linguistic Inquiry 109-141. WATTERS, John R. 1980 Review of Dik (1978), Lingua 50, 155-189.

10,

ERRATA

Owing to an unfortunate error at the printer, the pages 21 and 41 are wrong. Please correct the following errors: p. 21

SUBJECT AND OBJECT IN FUNCTIONAL GRAMMAR, read: SUBJECT AND OBJECT IN SERBO-CROATIAN Simon Dik & Gvozdanovic Jadranka, read: Simon Dik & Jadranka Gvozdanovic

p. 41

Simon Dik et al., read: Simon Dik, Maria E. Hoffmann, Jan R. de Jong, Sie Ing Djiang, Harry Stroomer, and Lourens de Vries

Please note that the Table of Contents should be changed accordingly.

FORIS PUBLICATIONS