Function and Expression in Functional Grammar 9783110872620, 9783110134070

Citation preview

Function and Expression in Functional Grammar

Functional Grammar Series 16

Editors

A. Machtelt Bolkestein Simon C. Dik Casper de Groot J. Lachlan Mackenzie

Mouton de Gruyter Berlin · New York

Function and Expression in Functional Grammar

edited by

Elisabeth Engberg-Pedersen Lisbeth Falster Jakobsen Lone Schack Rasmussen

Mouton de Gruyter Berlin · New York

1994

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter & Co., Berlin.

@ Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication

Data

Function and expression in functional grammar / edited by Elisabeth Engberg-Pedersen, Lisbeth Falster Jakobsen, Lone Schack Rasmussen. p. cm. — (Functional grammar series ; 16) Paper presented at the fourth International Conference on Functional Grammar which was held June 1990. University of Copenhagen. Includes bibliographical references and index. ISBN 3-11-013407-1 1. Functionalism (Linguistics) —Congresses. I. Engberg-Pedersen, Elisabeth, 1 9 5 2 II. Jakobsen, Lisbeth Falster, 1939— . III. Rasmussen, Lone Schack. IV. International Conference on Functional Grammar (4th : 1990 : University of Copenhagen) V. Series. P147.F8614 1994 415-dc20 94-14582 CIP

Die Deutsche Bibliothek — Cataloging-in-Publication

Data

Function and expression in functional grammar / ed. by Elisabeth Engberg-Pedersen ... — Berlin ; New York : Mouton de Gruyter, 1994 (Functional grammar series ; 16) ISBN 3-11-013407-1 NE: Engberg-Pedersen, Elisabeth [Hrsg.]; GT

© Copyright 1994 by Walter de Gruyter & Co., D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Typesetting and Printing: Arthur Collignon GmbH, Beriin. — Binding: Dieter Mikolai, Berlin. Printed in Germany.

Contents

Introduction

1

The end of the chain: Where does decomposition of lexical knowledge lead us eventually? Piek Vossen

11

Semantic fiinctions in perspective - reconsidering meaning definitions Lone Schack Rasmussen

41

Transitivity and the treatment of (non)prototypicality in Functional Grammar Louis Goossens

65

A valence based theory of grammatical relations Michael Herslund-Finn Serensen

81

The study of ergativity in Functional Grammar Vit Bubenik

97

Object assignment in a Functional Grammar of Croatian revisited Mario Brdar

109

Perspective, markedness, and paradigmatic relations between predicates. A case study of Danish Lisbeth Falster Jakobsen

127

Syntactic fiinctions, topic, and grammatical relations Hartmut Haber land-Ole Nedergaard Thomsen

153

Speech acts and information structure in Functional Grammar Ole Togeby

183

Ancient Greek warfare - A case study in constituent ordering HelmaDik

197

Contextualizing constituent as topic, non-sequential background and dramatic pause: Hebrew and Aramaic evidence Randall Buth 215 Dutch subordinators and PI in a Functional Grammar word order template Chris Braecke

233

vi

Contents

From CASE to FOCUS in the pronouns of some Wessex-based dialects of English Harold Paddock 255 Auxiliary verbs in Arabic Martine Cuvalay

265

Term-to-phrase mapping rules: A case study from Arabic Ahmed Moutaouakil

285

On the generation of English temporal satellite terms John H. Connolly

303

Not without you, I won't: Special utterance types in Functional Grammar YaelZiv 317 A ftmctional typology of speech reports Marijke De Roeck

331

Computational description of verbal complexes in English and Latin Simon C.Dik

353

Prolog for a functional grammar of Hungarian: A programmer's look at grammar writing Kwee Tjoe Liang 385 Expression rules using f-structures Hans Weigand

403

Some formal and computational aspects of the Functional Grammar machine model DikBakker 421 Subject index

443

Introduction

1. The Fourth International Conference on Functional Grammar The Fourth International Conference on Functional Grammar held at the University of Copenhagen in June 1990 has resulted in two volumes of papers witljin the Functional Grammar framework, one volume entitled "Layered Structure and Reference in a Functional Perspective" edited by Michael Fortescue, Peter Harder, and Lars Kristoflfersen and published by John Benjamins, and the present volume published in the Functional Grammar Series which in April 1991 was taken over by Mouton de Gruyter when Foris Publications was integrated with this publisher. The Fourth International Conference on Functional Grammar 1990 was arranged by a group of Danish linguists who have for a number of years been working within a functional approach to language description in order to establish a common theoretical framework. The group has been strongly influenced by Simon Dik and the school of Functional Grammar, but has also drawn inspiration from work by other linguists who see language not as a formal system, but seek to describe the linguistic sign as a combination of meaning and expression, a consequence of the view that language is a social phenomenon with a psychological correlate. The group has also renewed ideas from Louis Hjelmslev's Glossematics with its emphasis on the distinction between meaning and expression. The aspects of ftinctionality propounded by Simon Dik (1989) were reflected very strongly at the Fourth Conference; but the contributors, of course, chose various points of departure and also in some cases demonstrated a wider approach to the description of functionality. Among other things the productive mode of the theory, going from the semantics of grammatical categories and structures to their expression, inspired applications to the structures of languages hitherto not described in Functional Grammar terms and explorations nof the possibility of computer implementations of the model. Diachronic explanations of phenomena in particular languages, using functional notions, were also presented. Some contributors took their starting point in subjects or modes of description outside the well-defined scope of Dik (1989), such as the lexicon or prototypical descriptions of grammatical phenomena in particular languages.

2

Introduction

2. Functional Grammar 1978 and 1989 compared The Fourth International Conference was the first to be held after the publication of Simon Dik's Theory of Functional Grammar. Part I: The Structure of the Clause (1989). That volume is a considerable expansion of the first version of the theory in Functional Grammar (1978), and the expansion reflects the large number of contributions to Functional Grammar inspired by and elaborating the 1978 version. In accordance with its title, Functional Grammar (1978) was concerned above all with the Semantic, Syntactic, and Pragmatic Functions and how they are expressed within independent linguistic expressions. This approach restricted the focus of interest to a subset of the NPs of the clause, mainly the arguments which are related to predicates in predicate frames. Whereas the theoretical discussion concerning arguments was very detailed, very little was said about satellites, and practically nothing about other aspects of the clause, such as temporality, aspectuality, modality, and illocutionary force. The theory is enlarged in Dik 1989 with the Layered Model, in which a layering of these elements operators, is combined with an ordering of satellites which correlates semantically with the layers. Ideas on a hierarchical layering of operators relevant to the predicate were advanced in different forms by Foley and Van Valin (1984), Bybee (1985), and Hengeveld (1989). According to the version adopted in Dik (1989), the underlying structure of the clause can be represented as follows: (1)

Nucleus or Nuclear Predication = predicate frame as selected from the lexicon or produced by predicate formation rules, applied to appropriate terms = Stem (Arg"), where Stem is the form of the predicate and Arg" are its arguments.

(2)

Core Predication = PI [Nucleus] SI, where PI are predicate operators and SI are predicate satellites. Predicate operators are grammatically expressed modifiers of the nuclear State of Affairs (SoA). They include such "Inner Aspect" values as "progressive", "imperfective", "perfective".

(3)

Extended Predication = P2 [Core] S2, where P2 are predication operators and S2 predication satellites. Predication operators are grammatically expressed distinctions which serve to "localize" or "situate" the SoA as expressed by the Core Predication. They include values for the categories Tense, (Outer) Aspect (e.g. "perfect"), and Polarity.

Introduction

3

(4)

Proposition = P3 [Extended Fred] S3, where P3 and S3 are propositional operators and satellites, respectively. Prepositional operators are grammatically expressed distinctions which define the Speaker's attitude towards, or evaluation of, the content expressed in the Extended Predication. They include such (subjective) modal values as "possible" and "desired".

(5)

Clause = P4 [Proposition] S4, where P4 and S4 are illocutionary operators and satellites, respectively. Illocutionary operators are grammatically expressed distinctions which express or modify the illocutionary force of the linguistic expression. They include such values as "declarative" and "interrogative".

The overall structure of the clause can thus be represented as follows: (6)

[P4 [P3 [P2 [PI [Stem (Arg")] SI] S2] S3] S4]

At the same time the three types of Function are attributed to different stages in the model: the Semantic Functions are part of the predicate frame, which forms nuclear predications in combination with arguments at stage (1). The arguments may be assigned Syntactic Functions between stage (2) and (3); and finally the Pragmatic Functions are assigned between stage (4) and (5). The result is a model which integrates the ordering of the Functions with an ordering of the other grammatical phenomena which enter into the construction of the internal clause structure. The concentration on the internal structure of the clause has entailed cuts in the scope of the new framework. Most important for the present volume is the absence of the notions Theme and Tail among the Pragmatic functions. In return, the treatment of Topic and Focus is much more detailed in Dik (1989). The Layering with its concentration on other aspects of the clause has enriched the categorization of the States of Affairs, and consequently brought the relation of Nuclear Semantic Functions to the Semantic Function Hierarchy into focus; this results in a distinction between arguments as such, argi, arg2, arg3, and classes of arguments which carry the same number in relation to the ordering of the arguments in predicate frames: A', A^, A^. This again can be seen as a result of the direct application of the theory to different languages, highlighting the problem of balancing the universal functional approach to classification with what is actually the case in particular languages, where Functions are combined with expressions. Accordingly, the enrichment of the theory has brought out a much more facetted treatment of the nature of expression rules and how they operate.

4

Introduction

Finally, theorizing on methodology and basic linguistic concepts has been expanded and the notion of a 'Model Natural Language User' as a natural extension of a functional theory brought forward.

3. Overview of the papers The basic level of the grammatical model in Functional Grammar, "the Fund", contains all the predicates and terms from which predications can be construed. Their combinatorial potential is specified by means of predicate frames, which indicate the quantitative and qualitative valencies of the predicate, i. e. the number and type of arguments that the predicate takes to form nuclear predications. The qualitative valencies are specified by the Semantic Functions of the arguments and the selectional restrictions imposed on them. The model distinguishes nuclear and satellite Semantic Functions, and the nuclear Semantic Functions are furthermore divided into classes of first, second and third argument Semantic Functions, A', A^, A^. The first argument Semantic Functions partially reflect the typology of States of Affairs. Each basic frame in the lexicon is associated with a number of meaning postulates, through which the predicate is semantically related to other predicates of the language. If these meaning postulates add up to a fiill specification of the meaning of the predicate they are called meaning definitions. The first two papers stress the need within Functional Grammar to develop a systematic account of lexical knowledge. The paper by Vossen, "The end of the chain: Where does decomposition of lexical knowledge lead us eventually?", discusses the problem of how to posit a set of atomic predicates that can handle both classificational and fimctional definitions in the form of meaning postulates. In "Semantic functions in perspective — reconsidering meaning definitions", Schack Rasmussen advocates a bipartite model which establishes the semantic and the formal potential of the predicate as two comparable, equipollent dimensions. The paper proposes a typology of meaning definitions which defines the parameters contained in the basic set and demonstrates that a typology of meaning definitions may account for all the semantic information conveyed by the semantic functions except Agent and Instrument. Several contributors are concerned with the question of how the Semantic Functions, the valency of predicate fi-ames. Syntactic Functions and the expressions of arguments and satellites interrelate.

Introduction

5

In "Transitivity and the treatment of (non)prototypicality in Functional Grammar", Goossens proposes a more consistent use of the Goal category which integrates the notion transitivity as conceived in Cognitive Grammar into the underlying structure of Functional Grammar. He provides a procedure for Functional Grammar which does justice to the insight of Cognitive Grammar that transitivity is a graded category and advocates the idea that, in general, linguistic categorization is graded and displays prototype behaviour. In "A valence based theory of grammatical relations", Herslund and Serensen compare their own valency-based theory of grammatical relations with the Functional Grammar treatment of arguments and Syntactic Functions. They introduce a new grammatical relation, Adject, and present a unified semantic inteφretation of the relations between A' and A^ in 2-place predicates and between A^ and A^ in 3-place predicates. In "The study of ergativity in Functional Grammar", Bubenik demonstrates that the Functional Grammar model of the development of ergativity in terms of markedness shift does not specify an exact route. In particular, in Indo-Aryan languages intransitives have always been passivizable, and these languages have never had ergative constructions without passive ones. Moreover, a number of other changes contributed to the emergence of the ergative, and the stages in the development were not discrete, as suggested by the Functional Grammar model. Different assignments of Syntactic Functions, Subject and Object, capture oppositions such as that between active and passive constructions and thus define different perspectives over States of Affairs as designated by the predication. These fianctions are, unlike Semantic and Pragmatic Functions, not considered universal, but when made use of in particular languages they have universal properties. The quite specific conditions laid down by Dik for assuming Syntactic Functions as a feature of a particular language are debated by two authors. Brdar challenges in his paper "Object assignment in a Functional Grammar of Croatian revisited" the Functional Grammar view that Object assigment is irrelevant in a functional grammar of Croatian. Instead of a dichotomous distinction between Object assignment and nonassignment, he advocates a typological continuum along which languages can be ordered according to the degree of object relevance. Such a continuum is due to the general asymmetric duality of the linguistic sign. Falster Jakobsen proposes in "Perspective, markedness, and paradigmatic relations between predicates. A case study of Danish" another type of assignment: Object assignment to a first argument. She also discusses the relations between Semantic Functions, first arguments (argi), and the A' in Functional

6

Introduction

Grammar, and proposes a stricter distinction between Semantic Functions as conceptual entities and as entities syntagmatically linked with predicates. Finally, she suggests a level of sentence constituents as functional categories in their own right: they are carriers of degrees of markedness that provide the basis for Pragmatic Functions. By Pragmatic Functions, Dik means 'functions which specify the informational status of the constituents in relation to the wider communicative setting in which they are used' (1989: 264). The communicative setting is to be understood in terms of the Speaker's estimate of the Addressee's pragmatic information at the moment of speaking, where the Addressee's pragmatic information consists of his knowledge, beliefs, feelings, and preconceptions at that moment. Dik distinguishes extra-clausal and intra-clausal pragmatic functions. The extra-clausal functions which will be treated in volume 2 of The Theory of Functional Grammar, include among others Theme (As for the students, they won't be invited) and Clarification or Tail {He s a nice chap, your brother). In volume 1, the focus is on intra-clausal pragmatic functions, which are described in terms of topicality and focality. Topicality characterizes those entities "about" which information is provided or requested in the discourse, while the focal information is that information which is relatively the most important or salient in the given communicative setting and considered by the Speaker to be most essential for the Addressee to integrate into his pragmatic information. Dik further subdivides Topic and Focus into subcategories such as New Topic, Given Topic, New Focus, and Parallel Focus. In the chapters on constituent ordering, Dik introduces PI, a clause-initial position which in some languages is obligatorily occupied by specific categories of constituents such as question words or subordinators. The articles on pragmatic functions in this volume contribute to a clarification of the distinction between sentence topic and discourse topic. In "Syntactic functions, topic, and grammatical relations", Haberland and Nedergaard Thomsen discuss Syntactic Functions as defined in Functional Grammar theory and point out that the Functional Grammar Topic is really a discourse topic defined in terms of givenness. The syntactic functions subject and object can, therefore, be defined in terms of topicality understood as what the sentence is about, with the subject as the primary internal topic and the object as the secondary internal topic. In light of their pragmatic definitions of syntactic functions and topic the authors then describe a number of marked mapping constructions and cases of circumstantials as subject and (discourse) topic. In "Speech acts and information structure in Functional Grammar", Togeby discusses two aspects of Functional Grammar. First, he points out prob-

Introduction

7

lems that arise in a speech act analysis based on grammatical sentence types. Second, he sets up rules for identifying the topic and focus of a sentence. In "Ancient Greek warfare - A case study in constituent ordering", H. Dik demonstrates that Ancient Greek constituent order can neither be considered random nor governed by syntactic ordering rules. Instead, it can be accounted for by pragmatic factors such as discourse topic, sentence topic, and focus. It is suggested that these pragmatic functions should not be restricted to terms but be assigned directly to semantic functions. Buth suggests in "Contextualizing constituent as topic, nonsequential background and dramatic pause: Hebrew and Aramaic evidence" that Topic in Functional Grammar should be broadened to accomodate insights from a discourse analysis perspective. He draws attention to a special kind of PI topicalization which relates to the clause as a whole and not to a single marked constituent. The effect is one of backgrounding the clause and thereby producing textual discontinuity. In "Dutch subordinators and PI in a Functional Grammar word order template", Braecke analyses the finite verb in main clauses and the subordinator -in subordinate clauses in Dutch as marking the start of syntactic ordering of the clause. Both types of constituents are seen as appearing in the second position of the clause, after PI. This analysis makes the fillers of PI more homogeneously pragmatic than the analysis presented by Dik (1989). The last article to focus on functions also forms a transition to the articles on expression. It is Paddock's "From CASE to FOCUS in the pronouns of some Wessex-based dialects of English". In a diachronic shift in pronouns from a case system to a focus system in some Wessex-based dialects of English and their sister or half-sister dialects in parts of Newfoundland, the subject/object distinction has been replaced by a strong/weak distinction. The change in expression is interpreted as the result of an inteφlay of semantic, syntactic, and pragmatic functions. In the Functional Grammar model the abstract underlying clause structures are mapped onto actual linguistic expressions by means of Expression Rules, which determine the form, the order, and the prosodie contour of the constituents, given their structural and functional status within the underlying clause structure. The following three articles treat mapping rales from underlying stractures to different types of auxiliary verbs, terms, and temporal satellites. Cuvalay discusses in "Auxiliary verbs in Arabic" an aspect of verb complexes in Arabic which is relevant to the formulation of expression mies in Functional Grammar. She points out the need to distinguish different kinds of so-called defective verbs in Arabic on the basis of distribution and meaning and demonstrates how they can be accounted for by means of the different types of π-operators proposed by Dik (1989).

8

Introduction

In "Term-to-phrase mapping rules: A case study from Arabic", Moutaouakil formulates a subset of expression rules for Modem Standard Arabic which allows a prediction of the manner in which the logico-semantic structure of a given term is converted to the actual syntactic configuration of a phrase. Two sorts of phrase constituency rules can be distinguished: rales concerning a) the internal syntactic stmcture of phrases and b) the relationship between phrases and the predicate. In particular, the paper discusses the rales for quantifiers and numerators which have a different syntactic behaviour in Arabic than normally stated in Functional Grammar. In his paper "On the generation of English temporal satellite terms", Connolly suggests a notation of rales for the formal expression of temporal satellites and presents a set of rales for temporal satellites in English. Furthermore, he suggests a set of semantic subcategories of the function "temporal" and describes the factors that determine whether or not the expression of a temporal satellite should include an adposition, and if so, which one. Two papers discuss constractions which are outside the scope of Dik (1989). The paper by Ziv "Not without you, I won't: Special utterance types in Functional Grammar" analyses a special utterance type "Not without you, I won't". After having shown its peculiar syntactic, semantic, and pragmatic properties, the author discusses its sentential vs. discourse status, proposing that this type of sentence fragment requires a discourse sequence as a part of its frame of reference or even as a unit of analysis. On the basis of a sample of 40 languages. De Roeck discusses in "A functional typology of speech reports" two claims in Functional Grammar treatments of speech reports: I) speech reports are complements of verbs of saying; 2) there are basically two types of speech reports, direct and indirect speech. Distinguishing parataxis, hypotaxis, and embedding (i. e. the clause modifying or replacing a noun), she finds that all three types of constraction can be found with speech reports in the languages of the sample. With respect to the second claim, the author demonstrates that there is a scale between direct and indirect speech and that in some languages direct speech reports can be used for the expression of a proposition and even of a predication. The Layered Model, which presents a hierarchical ordering of grammatical phenomena, invites computational implementations. Dik's "Computational description of verbal complexes in English and Latin" presents a prolog program called VERBAL COMPLEXES which generates verbal expressions from operator combinations in five languages: English, Dutch, Latin, French, and Japanese. By means of a detailed presentation of the rales for English and a comparison of these with the Latin rales, the author demonstrates that much the same strategy of formal expression can be

Introduction

9

implemented for analytically and synthetically expressed verbal complexes. The author introduces the "conveyor belt" principle, i. e. the principle that the application of one operator feeds the application of another operator, and demonstrates that all formal expression rules can be formulated according to one general format. Finally, the paper provides new insight into the function of modal operators. Kwee in "Prolog for a functional grammar of Hungarian: A programmer's look at grammar writing" investigates the feasibilitiy of translating the existing computer model of Functional Grammar (Algol68) into a declarative language like Prolog and of adapting this model to a free word order language, exemplified by Hungarian. In "Expression rules using f-structures" Weigand proposes a new formalism for representing Functional Grammar structures in the form of feature structures and discusses the advantages of this formalism. He also shows how expression rules can be formulated with the help of feature structures. The formalism is illustrated by a fragment of a German grammar. Bakker in "Some formal and computational aspects of Functional Grammar machine model" presents the outline of a computational model for Functional Grammar which includes both a universal level and the grammars of individual languages. He further suggests that the formal framework of hierarchical feature-value structures may be extended in such a way that it can be used to formalize some nontrivial structures and rales in a Functional Grammar-like grammar within the proposed model. The publication of Function and Expression in Functional Grammar has been supported by the Danish Research Council for the Humanities and the University of Copenhagen. We would like to thank Soren Wichmann and Henrik Morell for preparing the manuscripts for publication and Michael Fortescue for his assistance with the English. Copenhagen, March 1993

Elisabeth Engberg-Pedersen Lisbeth Falster Jakobsen Lone Schack Rasmussen

The end of the chain: Where does decomposition of lexical knowledge lead us eventually? Piek Vossen 1. Introduction^ In Functional Grammar (henceforth FG) semantic knowledge on predicates, as far as it concerns conceptual knowledge, is stored in the form of Meaning Postulates (Dik 1978, 1987, 1989: 4.5). One of the basic assumptions about the storage of this knowledge in the lexicon is that the Meaning Postulates do not contain abstract features to express the meaning of words but predicates of the same language. As a result all predicates are semantically interrelated via the lexicon in a specific way, which has been described by Dik (1978) in terms of the principle of Stepwise Lexical Decomposition. According to this principle the meaning of a word is decomposed in terms of the more general words found in its meaning postulate, as in the following example of bachelor: (1)

bachelor —» an unmarried man

Since the meaning of man is also described in the same lexicon, its semantic content can in its tum be decomposed in even more abstract terms: (2)

man —• a male person

This process can be repeated until predicates are reached which cannot be further defined. In addition to these decompositional effects, expressing the meaning of predicates in terms of other predicates of the same language will lead to a certain degree of language dependency of meaning, since languages will not have and use the same set of equivalent words and expressions to define words. To quote Dik: '"defining meaning' is a language-internal affair" (Dik 1989: 86). Describing meaning in terms of predicates will probably reflect a cultural and linguistic bias which, however, is a typical part of the meaning of the words. So we can talk about sunrise and even make inferences on the rising-process, knowing that it is actually an earthtum. A dictionary, in this respect, can be seen as a more or less empirical collection of such knowledge on the meanings of words which is also ex-

12

Piek Vossen

pressed in terms of words of the same language, compare the following examples from Longman^:

(3)

Entry word

Meaning description

co-star actor man male person being thing object

a famous actor or actress who... a man who acts a part in a play a fully-grown human male a male person or animal human being a living thing any material object a thing

This practice of dictionaries (referred to as Ordinary Dictionary Practice by Dik 1978) will necessarily lead to circularity in those cases in which a language has no more abstract words left to describe other words, as is the case with object and thing. In the Links project^, the definitions of one such dictionary on computertape, the Longman Dictionary of Contemporary English (henceforth LDOCE), have been automatically analyzed to make its semantic content systematically accessible. This makes it possible to evaluate the specific semantic interrelatedness of words via their definitions on a massive scale. In this paper, I will describe some results of such inventory of semantic relations between nominal predicates. This inventory can be regarded from two points of view: (i) (ii)

the kinds of relations that hold between nominal predicates the distribution of predicates related to each other: which predicates are related to which other predicates?

In the next section, I will give a typology of the basic relations expressed in the definitions of nouns and compare this typology of relations with a typological classification of nominal predicates within FG by RijkhofF (1990). From this comparison I conclude that it is necessary to add a new dimension to this classification to provide for the various relations expressed in dictionary definitions. In the third section, the application of stepwise lexical decomposition to the data resulting fi-om the Links project is described, which results in a distributional overview of related nominal predicates exhibiting three major levels: a bottom-level of words not occurring as heads in definitions, a core-level of most frequent definition heads, and a core-level-defining top-level which consists of a rather small set of words circularly defined. The top-level which represents a basic semantic classification of the words in the lexicon is not only circular but also contains less reliable relations.

Decomposition of lexical knowledge

13

Assuming that the only solution to this circularity and unreliability would be to define a set of atomic predicates, the following questions will be raised: (i) (ii)

which elements would be candidates for atomic predicates? how are atomic predicates defmed?

To answer the first question I will discuss some notions from cognitive theories on abstraction levels (Berlin 1972) and categories (Rosch 1977). The second question can be only partially answered using the classificational model described in the second section because it only describes the constitutional meaning of predicates. From the data found in LDOCE it appears that there is also another mode of defining words, namely: a fiinctional perspective. To some extent the functional classification illustrates that the grammar is in the lexicon. Particular predicates are essentially classified in terms of typical "functions" such as "place", "time", "instrument", "manner" which correspond to semantic functions in Functional Grammar. More important, however, are the consequences for the principle of stepwise lexical decomposition which has to be widened to be able to infer the properties of predicates needed in the Grammar. Such a broad notion of decomposition is also needed for those predicates which have kernels which express a complex composition or components of complex entities.

2. The kinds of relations between nominal predicates The relation between the entry word and the words of the definition in a dictionary is basically determined by the kernel phrase of the definition. Other constituents such as pre- and postmodifiers and determiners only add information to the characterization of the type of thing designated by the kernel. Within the Links project different types of kernels have been distinguished based on structural properties and their semantic effect on the entry word (for a more extensive account, see Vossen 1989): i. Noncomplex kernels In the case of a noncomplex kernel, the syntactic kernel (the part which is sensitive to congruence) is also the semantic head:

(4) (5) (6)

Entry word

Meaning description

watchdog sheepdog abattoir

a fierce dog used to guard property a dog trained to drive sheep... slaughterhouse

14

Piek Vossen

In (4) and (5), the kernel designates the category of things to which the entry word refers. In this case, a hyponymy-relation is implicitly expressed. In the case of (6), however, we must speak of a synonymy-relation, since slaughterhouse is hardly more general than abattoir. The absence of modifiers is, in this respect, typical. ii. Complex kernels In the case of complex kernels, the pure syntactic kernel is followed by an o/complement which contains the most important semantic element:

(7) (8) (9) (10)

Entry word

Meaning description

candy stomach waste band

a type of sweet a front part of the body a stretch of land a group of musicians

The syntactic kernel (e.g. part, stretch in (8) and (9)) explicitly specifies the relation between the entry word {stomach, waste) and the main element (body, land). As can be seen in (7) hyponymy-relations can also be explicitly expressed in this way. iii. Derivations Two types of derivational structure have been found to express the meaning of higher order nominal predicates:

(11) (12) (13)

Entry word

Meaning description

ambiguity gasp advent

a condition of being ambiguous an act of gasping the coming of Christ

In (11) and (12), the syntactic kernel condition and act constitute very general classes of entities, and they are followed by verb phrases which fill in more specific properties. Only a small set of words fonctions as kernel of these constructions: act, activity, action, state, condition, way. In (13), the kernel itself is a derivation of a verb or adjective. The oy^complement of these structures refers to an argument of the underlying verb or adjective. Nominal predicates with such a definition always designate higher order entities. 2.1. Typology of relations in definitions of nominal predicates In case of a noncomplex kernel or a derivation, a hyponymy relation is always expressed (except for some synonyms). To explicitly express nonhyponymy

Decomposition of lexical knowledge

15

relations a complex construction has to be used. In the end, these relators can be described in terms of five basic relations: i.

type vs. class expressed by:

default relation·, type; sort; kind; form; class; race; family; breed; etc.

(14)

candy

a type of sweet

ii.

component vs. whole expressed by: part; piece; whole; body; system; etc.

(15) (16)

cockpit circulatory system

iii.

quantity vs. mass expressed by: part; piece; amount; quantity; band; bank; fold; beam; spot; mass; paste; etc.

(17) (18)

mat waste

iv.

member vs. group expressed by: member; piece; group; set; number; team; list; band; row; pattern; etc.

(19) (20)

papist band

V.

instance vs. concept expressed by: piece; work; example; case; idea; art; etc.

(21) (22)

classic demolition

the part of a plane or racing car the the system of blood, blood vessels, and heart, ...

α mass of hair α stretch of land

a member of the Roman Catholic Church a group of musicians

α piece of art a case of this''

Each relation can be expressed by a variety of elements which often also add other properties to the meaning of a word. For example beam not only expresses any quantity о / b u t also shape and amount relative to some norm. Some relations are bidirectional, i. e. part of a whole (15) but also a whole or system (consisting) of particular parts (16); a piece or quantity of a mass (18), but also a mass of pieces (17); a group of members (20) and a member of a group (19), so that they can be distinguished in terms of their direction. Finally, the same word can be used to express different relations depending on the type of complement it is applied to, part of a substance means a quantity, whereas part of an object means component.

16

Piek Vossen

2.2. Classifications of nominal predicates in FG The effect of many of these relators shows a strong resemblance to what has been called subcategorial conversion by Dik (1987). In this respect these words change the typological status of a term as a whole. The term a group of animals can be said to refer to an entity of the type collective whereas animal as such in its unmarked form and sense is a noun denoting an individual. Similarly, a piece of wood is a term referring to a countable entity whereas wood is a noncountable mass-denoting noun. Within FG four basic types of entities are described which correspond with four basic types of nominal predicates: individuals, collectives, masses and concept nouns. Following Rijkhoff (1990) these four types can be distinguished on the basis of two aspectual features: i. ii.

shape structure

having an implied boundary in the spatial dimension divisible into smaller parts without losing essential properties

Things with shape can be distinguished from their environment and, therefore, can be quantified or are countable. If something has structure, it can be divided into parts without losing its essential properties: a part of an amount of water is still water, and a subset of cattle is still cattle. The absence or presence of these features leads to a classification of nouns into four nominal aspects (see Table 1). Table I. Classification in terms of aspectual features structure shape

+ + collective cattle

+ mass water

+ individual man

concept birdiness

Examples of concept nouns in studies have only been given for nouns in classifier languages which have to be marked for shape when quantified, paraphrasable as three stick cigarette (Allan 1977). Perhaps derivations of verbs and adjectives in English in their generic sense (like demolition or birdiness) come closest to what could be called concept nouns. At least their definitions and grammatical coding (uncountable) suggest this. It is possible to say / don't like birdiness without making explicit that you are talking about instances (actual birds), the breeding of birds or any other aspect of it. Following this line one could suggest that the typological status of adjectives and concept nouns is similar. If we look at the frequency list of words^ below which have been analyzed in the Links project as relation-designators, then we see that the most frequent

Decomposition of lexical knowledge

17

words, type, piece, part, group, etc., express basic types of relations. Most of the other words either refer to quantities of substances or to sets of things. The less frequent a relator is, however, the more specific are the constitutional properties it expresses. So pair does not only refer to a set of things, but it also implies a cardinality of two. A row, on the other hand, does not add a cardinality value to the set, but shape. Similar things can be said of list, pattern, organization, network, ring, chain, team, etc. Also elements basically expressing a quantity of a mass can, in addition to that, refer to specific constitutional properties of the entity: mixture, band, stretch, sheet, ball, block, bar, drop. When we look at the least frequent elements, it is even doubtful whether the complement after of carries more semantic information than the relator, e.g. traffic of goods, tongue of metal, pint of beer, phrase of English, battle of shots. In a sense the selection restriction on the argument of the relation is so specific that it can almost be filled in by default, e.g. a loaf of bread (see Table 2). Table 2. List of relators distinguished in the Links project battle

1

curl

1

phrase 01 pint

1 1

tongue

1

cloud

2

loaf

2

traffic

2

beam division 04/02 stick 04 chain family 05/06 circle drop network rush stream

10 10 10 11 12 13 13 13 13 13

association heap line 02 organization 02 bar lump article 01 block bunch ball pile bit line 09 quantity class flow body 04 row sheet pattern side combination division 02 body 05 sort branch 03

14 14 14 14 15 16 17 17 18 19 19 20 21 24 25 26 27 28 28 32 38 39 42 47 47 49

stretch length band collection end period pair systemOl/02 list arrangement area mixture mass example form amount number member set kind group part piece type

50 56 59 60 60 60 65 70 75 80 84 104 114 135 136 162 177 208 246 257 437 679 766 1308

18

Piek Vossen

In general it is not possible to express these more specific properties in terms of the two aspectual features described by Rijkhoflf. This can also be said of one of the basic relations, component vs. whole, which does not refer to external shape and/or internal structure as defined by Rijkhoff (since a component of a whole is not similar to the whole). Without affecting the original classification into four subtypes which correspond to types of nominal predicates distinguished on grammatical grounds, it is possible to account for these compositional features by adding a new dimension (see Table 3). Table 3. Subdivision of nominal basic types

collective mass individual

+ composition

— composition

team of players mixture of foods body of parts

group of people mass of food piece of wood

Apparently composition can be applied to all three^ types of entities leading to new subtypes within. Components can then be seen as dependable but distinguishable entities located within other complex entities which, thus, have necessarily inner composition. The notion composition in its tum seems to be a scalar being determined by other features, such as number of components, heterogeneity of the components, and dependency to the whole. If we look at individuals, a highly complex entity as a person consists of more than one component of different types such as limbs, organs, blood, flesh, hair, skin which hardly occur independently of the whole and without which the whole can hardly exist. A stick, on the other hand, only consists of one thing: wood, and it can be broken into pieces resulting in more than one, although shorter, sticks (other examples of homogeneous entities are rope, river, stone). If we look at collectives, the components of a group of people are homogeneous and can easily occur independently. Dividing the group results in smaller groups, but a team of players cannot as easily be divided into smaller teams. The components of the latter complex collective {attackers, defenders, a single goal-keeper, etc.) also have a ftinction within the whole. A difference between individuals and collectives could be the possibility of components to stand alone. Also masses can have composition. Not only do we find all kinds of mixtures of masses, but also very complex masses as dish or chemical com-

Decomposition of lexical knowledge

19

pounds. The latter two, although being "distributive" in the sense that they can be divided into smaller portions of masses, still loose their status as a whole when being decomposed. Table 4. Classification in terms of aspectual and compositional features several components

heterogeneous

pendency to the whole

shape

structure

cattle chain

+

-

+

+

+

+ / -

+

-

organization food dish stick

+

+

+

-

list person

+ +

•

-

-

-

-

+

+

+ / -

+

-

+

-

-

-

+

-

-

+

+

+ +

-

-

Reconsidering the list of relators, we see that those expressing noncomponent relations often add shape, number and/or composition to an entity of another type. A pile of bars thus becomes a collective with little complexity which can easily be divided into smaller piles but a framework of bars should be carefully split into smaller pieces of frame because it has composition as well. Other relators such as system, arrangement, whole, body, network even represents more complex organizations of parts. In these definitions the complexity is often explicitly expressed in the kernel phrase in which heterogeneous entities are coordinated with and:

(23) (24) (25) (26)

Entry word

Meaning description

batter mortar block and tackle assembly line

a mixture of flour, eggs, and milk a mixture of lime, sand, and water an arrangement of wheels and ropes an arrangement of workers and machines

Somewhat less complex are entities with similar relators, but a more or less homogeneous set of entities as a complement (often in the plural):

(27) (28)

Entry word

Meaning description

plexus village

a network of nerves or blood vessels a collection of houses and other buildings

Looking at the distribution of the various relations in the lexicon, it appears that, despite the variety of relators used, hyponymy is still the most frequently expressed relation (see Table 5).

20

Piek Vossen

Table 5. Distribution of types of relations Noncomplex relations:

68%

c. 3 0 0 0 0

Complex relation: {type, kind'.

16%

c.

3% 16%

c. 1500) c. 7 000

Derivations: Total of entry-kernel relations Total of entry words

7000

c. 4 4 0 0 0 c. 23 000

From the 7000 complex relations, still in 1500 cases the explicitly expressed relation was also a hyponymy relation {type and kind). This leaves only about 13% of all senses of nominal predicates being related to other predicates via a nonhyponymy relation. These figures, however, still do not tell us anything about distribution of types of nominal predicates. Many predicates which refer to parts of the body are not directly described as such, but are described as, for instance, organs and limbs. Via organ and limb the relation to body is expressed only once for the whole class of words (see Figure 1). Something similar holds for collectives, which are described as kinds of a collective which is described only once as a group of individuals, e.g. navy, enterprise and church are organizations and an organization is a group of people. Finally, up to now we have only been discussing complex relations expressed by the kernel part of the definitions. Many part-whole relations are

organ—PART

animal or person

I ISA

I ISA I compound eye orb peeper

I eye nose heart conk

Figure 1. Parts indirectly related to their whole

| ISA | snout snitch hooter snoot

Decomposition of lexical knowledge

21

instead expressed in the differentiae of the definitions and not in the kernel part itself: Entry word

Meaning description

(29) (30)

window knob

(31)

spoon

(32)

head

a space in a wall, esp. in a house, ... a round lump on the surface or at the end of something a tool for mixing, serving, and eating food, consisting of a small bowl with a handle the part of the body which contains the eyes, nose and mouth, and the brain

In these examples a noncomplex kernel is used, whereas in case of (23) and (24) the component-whole relation is expressed, in a sense, by locating the part in another entity. In the same way, wholes, (31) and (32), are decomposed via predicates as consist of contain, have and the preposition with in the differentiae.

3. Stepwise lexical decomposition in LDOCE In order to get more insight into the distribution of the types of nominal predicates it is essential to look at the decompositional effect. For this purpose a program has been developed, called Word-DeviP (Word DEcomposition Via the Lexicon, Vossen — Serail 1990) which automatically generates decomposition chains like the one in Figure 2 based on the analysis of the definitions.

remembrarle e 00.01 memory 00.01 ability 00.00 skill 00.00 < Ы М К > power 01.01 < Ы М К > ability CIRC hemorrhoid 00.00 < Ы Ж > vessel 00.03 tube 00.03 organ 00.01 animal 01.01 creature 00.01 animal CIRC Figure 2. Decomposition chains of the Word-Devil program

22

Piek Vossen

In these chains words are related to their heads which are recursively looked up in the dictionary being related to other heads. The numbers after each word indicate the homograph (the number before the dot: "power 07.01") and the sense (the number after the dot: "power 01.07") of the word. In case of 00 there was only a single homograph and/or sense. The kind of relation that holds between two predicates is expressed between angled brackets, e.g. organ to animal by (PART). In case no relation was explicitly expressed the value LINK is filled in. 3.1. Levels of predicates in the lexicon By ordering these chains^ the other way round (from end to beginning) one gets an overview of all the fields, relations per field and circularities that occur, as in Figure 3. Within these chains it is possible to distinguish three levels of words on the basis of their position: 1. words which do not occur as heads of definitions of other words: so-called bottom-level words (words to the right in the chains in figure 3), 2. words which very frequently occur as kernels and which, therefore, play a major role in relating other words: so-called core-level words (words in the middle of the chains in Figure 3), 3. and finally there is a rather small set of circularly defined words in which all chains end: this is the so-called top-level (words to the left of the chains in Figure 3). Especially the top-level is important since it incoφoгates a kind of semantic classification of the nominal predicates in English. Furthermore what is stated of nouns at the highest level can be inherited by nouns on more specific levels. As such a lexicon may function as a very efficient device to store

CIRC CIRC CIRC CIRC CIRC CIRC CIRC CIRC CIRC

ability power ability m e m o r y . ability < Ы Ж > power ability patience ability < Ы М К > power < ы ш ; > ability sympathy ability power ability hmnour . animal creature animal breed . animal creature animal snake . animal creature animal dog . . animal creature animal organ . animal creature animal goat . .

Figure 3. Sorted chains resulting in a taxonomy

Decomposition of lexical knowledge

23

information only once which can then be inferred for whole classes of words. This also means, however, that when essential information at the highest level is not expressed in a lexicon derived from a dictionary, such a lexicon will fail to provide this information for a whole class of words. Figure exemplifies the major circularities at this top-level of which the object-thing cycle is most important. Not all predicates related to these topnodes are given, but only some of the predicates which constitute major fields of basically concrete entities. Below each predicate the frequency as a head is given. Going to the left: the head of each word as an entry word is given, and going to the right: entry words of which it is the head are found. To save space homograph-sense numbers and relation types have been omitted. Looking at these top-level relations we not only find circularity but also other kinds of phenomena. First of all at the highest levels totally distinct concepts like act and being, or fruit and instrument are attached to the same categories, respectively thing and object. Apparently, these two top-nodes are not so informative that they create homogeneous sets. In addition, very closely related concepts are sometimes not attached to the same category or even the same branch of categories, compare aircraft and aeroplane respectively, described as machine and vehicle, and skin versus other bodyparts like bone and organ. Furthermore, the cycles at the bottom like animal which is circularly defined by creature stand alone: i. e. they are in no way related to other beings like plant and person. Similar islands are formed by the clothinggarment cycle which is not related to cloth and substance, and the ship-boatvessel cycle which is not related to vehicle. Other typical properties of top-level words are a high degree of polysemy and rather abstract definitions which are only vaguely distinguished (for a more detailed discussion, see Vossen 1991). All these phenomena can be related to the fact that a lexicographer has to use words to define words whose meaning is either so basic that it is difficult to explain in words, or whose meaning is more perceptual than conceptual. In some extreme cases this can lead to very complex and long definitions as in: Entry word

Meaning description

(33)

colour

(34)

matter

the quality which allows the eyes to see the difference between (for example) a red flower and a blue flower when both are the same size and shape the material which makes up the world and everything in space which can be seen or touched, as opposed to thought or mind

24

Piek Vossen

object—instrument-machine 1 8 9

I

I

1 3 6

-rocket I

1 7 6

5

-aircraft--helicopter 3 7

-tool

9

1 0 5

— fruit

2

-spoon

nut

-groundnut-peanut 9 4 1 8 2 1 —article furniture- -bed airbed 4 6 3 0 20 1 thing something--building— -hotel 4 3 4

9 1 6

I

3

9 4

—covering-- -skin I

9 4

—food I

leather-calfskin

4 9

3 5

-meal

1 2 9

1

dinner

3 0

1

— v e h i c l e — -aeroplane 1 0 7 2 person -people force- -navy

—being3 7

2 5 5 8

I

3 6 3

5 6

6

-official-judge 7 0

1 1

-body 5 7

referee 1

bone

I

rib

3 7

2

organ

heart

5 4

—anything 8 7

I

container- -box 61

1 3 3

5

coffin 2

material-- -substance-flesh—meat—veal 2 4 5

I

3 0 0

2 5

-wood 1 4 3

4 8

-paper 1 3 3

4 0

2 1 5

3 2 3

3 3

1 4

1 3

toilet_paper mirror 2

-cloth -seed-

2

1

-glass

—plant-

6 7

board—shelf-cupboard

underwear 3

-grain 26

rice 2

animal--dog

garment--shirt

boat--motorboat

cord—cable

3 0 0

1 1 1

7 3

2 1

creature 60

4 9

clothing 22

6

4

ship

string

8 9

21

vessel 2

rope

Figure 4. Top-level of relations between concrete nouns

4 3

2

Decomposition of lexical knowledge

25

or to compositional definitions (involving instantiation) instead of decompositional definitions:

(35) (36)

Entry word

Meaning description

colour human

red, blue, green, black, brown, yellow, white, etc. a man, woman, or child, not an animal

Apparently, the information found in dictionaries at this top-level is less valuable. What do these top-level words represent? Is it possible to interpret them in terms of a classification of nominal predicates? Dik (1978, 1986, 1989) avoids this problem of circularity by assuming a set of atomic predicates which are not further defined in terms of predicates of the same language. This, however, raises two other questions: i. ii.

which elements will those atomic predicates be? how are the semantics of these atomic predicates described?

3.2. Atomic predicates To answer the first question it is essential to know what the different levels of words represent. Despite the continuously growing amount of literature on concepts, inheritance networks, categories, etc., only little is known of the relation between words and different levels of abstraction. Berlin (1972) describes five different levels of abstraction in animal and plant names based on his studies of ethnobotanic classifications in various languages: unique beginner: life form: generic name: specific name: varietal name:

plant, animal tree, bush, flower pine, oak, maple, elm Pondorosa pine, white pine northern Pondorosa pine

The most basic level, according to Berlin, is the generic level. It distinguishes itself from the other levels in a number of aspects. First of all most names for animals and plants are to be found at this level. There are considerably fewer life form and unique beginner terms. Berlin even claims that these latter levels were absent for a long time and developed out of the generic level relatively late in the terminological history of plant names. He compares this with a similar development which is taking place now with the category animacy in English on top of plants, animals and humans for which there is also no term.

26

Piek Vossen

Another aspect is that generic names are always unmarked whereas life form and varietal names are always marked. In most cases they are composed of the generic name and a specifier. Furthermore, Berlin claims that empirical studies show that generic names are learned first by children before extension to other levels takes place. A final claim for which, however, no empirical proof is given is that generic names are supposed to be more useful: that is, in most contexts life form names and unique beginner names will be too ambiguous and more specific names will be unnecessarily specific or even incorrect to refer to a particular plant or animal. This latter hypothesis corresponds with the notion of basic category in Rosch (1977a, 1977b). Good categories represent a level of abstraction that differs maximally from its neighbours in a taxonomy, and at the same time a level below the subtypes that are maximally common. This becomes clear if one looks at the following taxonomic example from Rosch: musical instruments / guitar

/

classical guitar

\ drums folk guitar

Guitar and drums are basic categories and can also be said to be perceptually very different. The differences between a classical and a folk guitar, however, are only minimal. If we extend Berlin's hypothesis that most names for concepts will be found at the generic level to dictionary definitions then the most frequent heads of these definitions will be either unique beginners or life form names. Since there will be fewer more specific terms than generic terms the frequency of the generic terms as heads has to be significantly smaller. Going back to the words related to object and thing in Figure 4, we can indeed observe a certain cut-off point in frequency at some chains as demonstrated by Figure 5. According to these distributions and given the fact that the average frequency as a head is 10 (4500 heads and 44 ООО relation tokens between heads and entry words) hotel, rocket, spoon, and aeroplane would be basic concepts at a generic level, whereas building, machine, tool, and vehicle would be unique beginners or life form names. The frequency-drop between container and box, and between place and school is less significant, and both box and school have more than average frequencies, which makes these cases less clear-cut. With regard to the distinctiveness at the basic level of concepts and

Decomposition of lexical knowledge machine vs. rocket 176 5 tool vs. spoon 105 9 building vs. hotel 94 3

27

vehicle vs. aeroplane 107 2 container vs. box vs. frame 133 61 1 place vs. school vs. college 531 37 8

Figure 5. Distributional cut-off points

the commonness below this level, a similar conclusion could be made. What we can refer to as places, tools, machines, vehicles and containers can vary considerably in perceptual features. However, types of hotels, rockets, spoons and aeroplanes do not vary much. It is still to be seen, however, whether these differences are statistically significant. The fact that LDOCE has used a controlled vocabulary of 2000 words will certainly have had some effect on these figures, although it still has to be shown that other dictionaries use more and/or other words. Until such comparative studies have been done the empirical scope of these data is restricted to this particular dictionary. Another point is that considering single steps is not enough; the shape and distribution of complete branches should be looked at. Frequency of heads should be compared to the total number of predicates related to it via the more specific predicates below (Amsler 1980) and the way in which these specific predicates are balanced. If we presume that these figures are significant, then the heads of these potential atoms {place, tool, machine, vehicle, etc.) have to be replaced by a classificational definition in terms of their aspectual feature. As far as the concrete nouns in Figure 4 are concerned, we would get quite a number of atomic predicates which are primarily distinguished as individuals and masses and are further distinguished by the differentiae of their dictionary definition. Within FG the information on the type of nominal predicate can either be stored in the predicate frame or in the Meaning Postulate. If, for the sake of the argument, we choose the latter solution, this information will be entailed by the first restrictor of the Postulate which corresponds with the head of the dictionary definition on the lower levels, e.g.:

(37) (38) (39) (40) (41)

Predicate

Meaning Postulate

rocket bed car air meat

[machine: driven by burning gases: etc.] [furniture: to sleep on] [vehicle: with 3 or usually 4 wheels: etc.] [gas: which surrounds the earth: etc.] [flesh: of animals: which is eaten]

28

Piek Vossen

In the case of atomic predicates, the heads will be replaced by indications of aspectual feature:

(42) (43) (44) (45) (46)

Predicate

Meaning Postulate

machine furniture vehicle flesh gas

[(shape): man-made: uses power...etc.] [(shape): large: quite movable: etc.] [(shape): in or on which people...etc.] [kcstructure): soft: including fat: etc.] [kcstructure): not solid: not liquid]

The differentiae found in definitions discriminate between the various atoms. Nominal predicates like object, something, anything and thing which are found on top of these atoms, will then be defined by completely void definitions which not only lack differentiae but in some cases also nominal aspect:

(47) (48) (49) (50) (51)

Predicate

Meaning Postulate

thing object substance something anything

[] [(shape)] [(structure)]

[] []

In the case of anything and something, this voidness represents their pronominal meaning. While object appears to be restricted to nonsubstances and substance to the complementary type of entities, thing is used for both types of entities and therefore reflects no nominal aspect. 3.3. Deviations in distributional properties One major problem with this approach is that these assumptions about concepts, categories and levels of abstraction from cognitive theories are based on studies that focus on concrete entities and perceptual properties. In addition, also the classification model set up in the previous section only describes constitutional properties. However, if we look at the semantic features which are used in the Longman tape version (not in the book) to label nominal entries, it appears that 50% of all noun-senses (22 ООО senses) are labelled as abstract, a class of nouns which is not fiirther divided into more specific fields. Figure 6 contains some major branches and cycles at the toplevel based on the heads of those abstract senses, filling in some finer distinctions.

Decomposition of lexical knowledge thing

something--fact

4 3 4

9 1 6

8 2

—disagreement

I

2 4

presence 5

--sound 2 2 9

word

name

2 3 9

6 4

sobriquet 1

--statement--description 116 I 26 --saying report--account—story I

9

1 9

3 3

2

1 5

—quality 4 5 1

strength I

muscle

2 3

0

colour

brown--beige

1 3 0

4

1

--activity 8 3

--result

infection

8 6

4

--action 4 2 0

--cause

reason

5 3

2 8

1 3

1 0 7

appearance

1 4 4 0 1

5 5

discovery 3

show 1 0

--permission I

argument 2 9

--happening--event —act

1 7

passage 1

—declaration--guarantee

I

1 5

1

- -movement I

refi ex

1 5 6

1

--punishment 2 4

hanging 1

condition 2 4 8

state

need

lack

disorder

5 3 9

1 2

6 1

1 8

I

freedom I

2 7

consciousness--feeling-admiration 1

1 4 2

5 6

--summary

--promise

7

fear 1 3

Figure 6. Top-level of relations between abstract nouns

disease 1 2 2

29

30

Piek Vossen

For these abstract nouns it is not immediately clear whether we can distinguish similar levels as for concrete nouns. If we look at the distribution of the heads (their frequency as a head), then we can see that condition, state and act constitute large fields which are hardly subdivided into proportionate specific fields. Instead they represent very shallow branches. Most words described as acts, states and conditions (also activity and action) are not heads of other words, and therefore belong to the bottom-level of the dictionary. These abstract nouns, which mainly refer to higher order entities as described by Dik (1989: 181), have often a derivational definition structure as described in section 2. Therefore it is possible to infer more specific classificational information on these nouns from the verbs and adjectives which are the main predicates in these definitions. In this respect it is also interesting that 30% of all these nouns labelled as abstract have derivational suffixes like -tion, -ment, -ism, -ity, etc. Other higher order nouns are represented by the major cycles in Figure 7. Another concept that constitutes an extraproportionate field is "person" (see Figure 4). Here we see the same phenomenon: it is not smoothly divided into proportionate subfields but constitutes a shallow branch with numerous bottom-level words. The branch person is also atypical in that it is not subdivided on the basis of constitutional properties (subtypes of person do not vary maximally in perceptual features, but on the contrary share most perceptual features, except for the few races of human beings), but to a large extent on the basis of social, cultural and cognitive properties like character, skills, behaviour, habits, activity, social rank and status, origin, etc. Finally, Figures 4 and 6 only contain the major fields related to object, something and thing, that is words which themselves have a high frequency as heads. The frequency of object, thing and something, however, is also rather high (189, 434 and 916, respectively). Apparently, there are still a number of rather specific words which are directly related to the highest nodes in the taxonomy, but do not occur as heads themselves. Replacing these circular top-nodes by a typological feature system automatically turns all these words into atomic predicates. However, if we look more closely at what kind of words they are, it appears that many of them are lexicalized derivations of the following kind:

(52) (53) (54) (55)

Entry word

Meaning description

buzzer stiffener acquisition adaptation

a thing that buzzes a thing which stiffens something or someone acquired something that has been adapted

Decomposition of lexical knowledge period occasion—meal—breakfast 172 I 45 30 3 I --season time I 9 266 --century 3 way 202 I I I manner 47 I method 79

means resource 32 1 behaviour—habit scepticism 105 24 1

part stage—degree—value 212 I 13 85 I 26 I I I I I —system I piece 129 -rank 56 52 ability green fingers 141 I 1 I —speech I 1 20 powerOl —recognition 62 I 2 I —memory skill 3 70

rate par 24 I 1 speed 21 position06-leadership 4 2

forceOl 71 I | I I power03 36

life 36 violence 5 lightOl 71

signOl line02—figure07-square 78 I 95 30 11. I --brace I

I

1

markOS --question mark 39 0

Figure 7. Other cycles representing major abstract nouns

31

32

Piek Vossen

Again, the meaning of these words is not formulated in terms of their constitutional properties (see person above) but in terms of the involvement of the entity in some kind of State of Affairs (SoA). Their constitutional vagueness is also indicated by the fact that in many cases (40% of all head-occurrences of thing) these heads are coordinated with completely different heads, e.g. person or thing: Entry word

Meaning description

(56)

arrival

(57) (58)

puzzler threat

a person who or thing that arrives or has arrived a person who or thing that puzzles one a person, thing, or idea regarded as a possible danger

It is of little importance whether the denotated entity is human or not, as long as it fulfills the role described. The consequence of removing the top nodes would be that all these nonconstitutional predicates, in a sense, will become atomic. Instead of a rather small set, we would end up with a set of about 1500 predicates which have to be primarily defined in terms of their nominal aspect and which are only distinguished in terms of their differentiae (here the controlled vocabulary might certainly be showing its effect, i. e. resulting in large fields at the top which have not been differentiated because the more specific words are not in the controlled vocabulary). 3.4. Different modes of defining predicates Some of the major fields in Figure 4 also seem to refer to a particular SoAinvolvement or functionality only, compare: place, food, vehicle, instrument, covering, container. The various subtypes of these categories can vary substantially in constitution but share their typical associated SoA-involvement. One level deeper the different kinds of, for instance, containers are divided into more specific subtypes which can differ in constitutio, e.g. bag, box, bowl, vessel, etc., or with regard to more specific containments, e.g. what is contained, how much, how long, etc. A top-node like material/substance (see Figure 8), on the other hand, typically represents a concept basically distinguished within the constitutional dimension. At the subtype level of substance, however, finer distinctions are also made in terms of different SoA-involvement. A result is that, for instance, not all liquids are stored below liquid. Things described as medicine, lime, chemical, or explosive can be liquid too, but are not primarily categor-

Decomposition of lexical knowledge

33

substance--fies meat—veal I-medicine—drug—tranquilizer I syrup I-chocolate I -gas air tornado I -lime quicklime I -metal gold I money I nail I -powder cement I dust I-chemical I-explosive-nitroglycerine -liquid tear petrol water dishwater Figure 8. Functional and constitutional predicates related to substance

ized as such. In principle all substances could be classified as either a solid, a gas or a liquid at a certain temperature, but it is perhaps typical for lexical knowledge that several dimensions are mixed. A functional classification in terms of food, medicine, explosive, etc. can be made completely independently from this constitutional classification. As such these predicates, which are in fact all related to the same predicate substance, cannot be seen as cohyponyms because they are in no sense complementary. Things which fall in one category {solid) can also be part of another (food). Types of substances are only complementary as far as they are distinguished within the same dimension. The notion of cohyponym, thus, appears to be relative to the particular dimension in which a category is subdivided. Also within other fields such as animal we can observe a similar shift in perpective. Among the different dog races, for example, we also find watchdog and sheepdog which can be of any dog-type (as far as constitutional properties are involved) as long as they are used or trained for some рифозе. Generalizing we can state that there are two basic modes of primary classifying concepts between which the lexicographers of Longman have made a choice: i. ii.

functional mode which describes an entity primary in terms of its involvement in a particular SoA; constitutional mode which describes an entity primary in terms of its constitutional properties.

34

Piek Vossen

In this respect it is interesting to see what the most frequent modifiers in noun-defínitions are: -

used and made as main predicates of attributive clauses and participials postmodifying the head; - with flowers, with leaves, with tail, for people, with legs as adpositional modifiers; - small, large, strong, short, long, thin, narrow as adjectival modifiers. These modifiers strongly stress the distinction between the two modes of defining. If a functional word at the highest level constitutes a large field, it is often necessary to distinguish the subtypes in terms of their constitution and vice versa. If the data from this dictionary can be generalized, we can state that entities can be viewed from both perspectives, but nominal predicates always entail a choice between them. Given some entity in some world, any set of features can be chosen to describe it, and given a particular set of features chosen to describe it, in principle, any of these features can be used to categorize it. However, when such an entity is referred to by a particular predicate, then the meaning of this predicate should be seen as an instruction to look at the entity as typically having at least some particular properties (which does not imply that it has no other properties) and primarily regard it as the kind of thing represented by one particular feature as its category. Thus, using a predicate such as food to name an entity should be interpreted as an instruction to conceptualize it primarily in terms of its involvement in a SoA, whereas using another predicate such as substance means that the same entity should be look at basically in terms of its constitution. This would imply that since skin is primarily described as a covering, all organs and bones are directly related to it. A similar variation is also reflected by the following two lists. The first one demonstrates coverings consisting of a material, the second one a substance of material functioning as a covering: Entry word

Meaning description

(59)

armour

(60)

awning

(61) (62) (63)

blanket boot chap

strong protective metal covering on fighting vehicles, ships, and aircraft a movable covering, esp. one made of canvas, used to protect... a thick especially woollen covering... a covering of leather or rubber for the foot... a fleshy covering of the jaw

Decomposition of lexical knowledge

35

Entry word

Meaning description

(64)

carpet

(65)

buckram

(66)

daub

(67)

drugget

heavy woven often woollen material for covering floors or stairs stiff cloth used for covering books, stiffening hats... (a) soft sticky material for covering surfaces like walls rough heavy woollen material used esp. as a floor covering

What makes a lexicographer decide to describe, for instance, soup as food first and in the second instance as liquid, and eatables first as substance or object and then in terms of fiinctional properties (e.g. meat and fruit) could be part of the meaning of soup in English (or the meaning of soup for this lexicographer). Provided that these differences have to be taken seriously, these mixed perspectives might, in a sense, reflect the language and cultural dependency of conceptual lexical knowledge stored in a dictionary. How are, for example, terms described in LDOCE as container, described in a Dutch dictionary in which such a derivational category is not available, and what does it mean when these Dutch predicates turn out to be defined in terms of their constitution? Do they have a different meaning in Dutch? Comparison with data from other dictionaries and dictionaries of other languages will probably yield interesting differences and correspondences.

4. Consequences for the principle of stepwise lexical decomposition To some extent the functional classification illustrates that the grammar is in the lexicon. Particular predicates are essentially classified in terms of typical fiinctions such as place, time, instrument, manner which correspond to semantic functions in Functional Grammar. The current development in FG to distinguish different types of satellites operating at different levels makes it interesting to look at these predicates which apparently, are very well suited to fulfil such functions of satellites. The ultimate functional approach in this respect is probably pointed out by Mackenzie (1992) who claims that particular nouns denoting things like places and times in fact are nonentities. In that case there is no constitution at all at the highest level and these nouns should be described in terms of functional properties only.

36

Piek Vossen

One of the main reasons for organising the definitions in the FG lexicon in the form of Meaning Postulates is that it makes it possible to infer this type of semantic information on predicates indirectly by following the decomposition chains. The major effect of building stepwise hierarchies is that the information is gradually distributed over all levels in the taxonomy so that it only has to be expressed once but can be inherited for all predicates related to it as subtypes. The fact that veal is a mass (see Figure 4) is not expressed directly at its definition but is inherited from material via meat, flesh, and substance. The findings described in this paper, however, illustrate that it is not enough to simply follow these decomposition chains. First of all complex kernels as we have seen in section 2 block the standard inheritance principles. If we look back at Figure 4, then we see that not all predicates below substance and material are also noncountable masses. Those predicates which were related to a substance via a complex kernel, e.g. board {piece of wood) and mirror (piece of glass) are basically as countable terms as any other individual predicate and thus cannot inherit the property (structure) from substance. Something similar holds for the predicates below people (force, navy) which cannot be related to person as such without taking care of the group-aspect. The semantic effect of subcategorial conversion, for instance, from mass to individual and from individual to collective via relators such as piece and group should therefore be described in such a way (perhaps by means of a rule) that the inheritance of particular properties is overridden or blocked. For the component-whole relations, however, the situation is more complex. As we have seen, entities can have a heterogeneous composition consisting of components of different types (possibly dependent on each other in constituting the whole), so that no predictions can be made on the basis of the kernel-information only. The body parts in Figure 4 below body most certainly cannot be seen as predicates inheriting all properties of person and higher predicates: how animate is a hand or a nail, how countable is hair! A property such as animacy only applies to the whole because of the functionality of the parts, but it does not apply to the parts as such. Some parts of objects can even be liquid, contrasting in aspectual features:

(68)

Entry word

Meaning description

juice

the liquid part of a fruit, vegetables, and meat

These components can in no way inherit such essential semantic properties as (structure) or (animacy) directly via their decomposition chains. From its definition kernel part of a fruit alone we will never end up in liquid and not in substance either.

Decomposition of lexical knowledge

37

Guiding inheritance along component-whole relations, therefore, can only be done for entities which have a homogeneous composition. The type of predicate (the category from which it can inherit essential semantic properties) can thus be predicted from the type of the complement and the relator that has been used. A homogeneous entity like wood automatically results in an homogeneous individual when portioned and shaped (such as stick), and a homogeneous group of animals such as cattle results in a collective. However, to make a similar typing of predicates for heterogeneous complex entities possible, the notion of decomposition has to be widened so that also the information expressed in the differentiae is used to get at other categories {juice will be related to the category liquid and substance because the necessary properties to do so are found in the modifiers). In case of entities such as batter which are explicitly described as a complex whole (a mixture offlour, eggs, and milk, beaten together,...), even looking at the differentiae will not directly make clear to what entity it belongs {mass), unless very complex inferences are made. The second conclusion is based on the different modes of classifying predicates encountered. The drawback of having these modes is that some of the subtypes of a category will be found below other categories. For example not all substances are classified as such. Those defined as covering (e.g. chap) or product (e.g. soap) will never inherit the properties of substance (such as (structure)) because their decomposition chains do not lead to it. The fact that we are dealing with noncountable matter can again only be inferred from the other properties expressed in the definition, although, apparently, linguists have to be chemists as well to do so:

(69)

Entry Word

Meaning Description

soap

a product made from fat and ALKALI, for use with water to clean the body or other things

The more general conclusion is that the classical idea that definitions are composed of a category which is specified by differentiae is too strict. What the category is and what properties are specifications can vary, making inheritance principles far more complex. One way of dealing with this phenomenon described by Briscoe and Copestake (forthcoming), consists of allowing multiple inheritance from more than one category, e.g. soap both from product and substance. The problem is then how to infer automatically to which other category it belongs on the basis of the differentiae. When making inferences like a product of two substances is a new compound substance, its possible inheritance via the differentiae is possible too.

38

Piek Vossen

Notes 1 This paper has been made with the help of Willem Meijs, Simon Dik and Jan Rijkhoff who gave me many suggestions. 2 All examples of meaning descriptions in this paper are based on the definitions of the Longman Dictionary of Contemporary English, 1978 edition. 3 The Links project was a project funded by the Dutch Organization for Academic Research (N.W.O.) under no. 300-169-007 for a period of three years from 1-2-1986 until 1-2-1989, carried out at the English Department of the University of Amsterdam. Participating Research Assistants: M. den Brooder, I. van den Hurk, R Vossen. Research coordinator: W. Meijs. 4 "This" refers to the previous sense of "demolition" in which the concept is defined: "the action of demolishing" as an uncountable concept. In case we want to speak of an individualization of this concept the first sense is converted to a countable entity via the relator "case". In LDOCE the individualized meaning is listed as a separate sense with a different grammatical category. 5 Some of the relators in this list are followed by numbers which indicate the sense or senses in which they are used according to the sense numbering of LDOCE. These sense numbers have been added by hand as part of the analysis of the definitions. 6 No examples of concepts nouns with a complex composition are given because the existence of concept nouns in English is doubtful. This does not mean that concept nouns in classifier languages cannot have an inner composition. 7 The program has been developed at the Computer Department of the University of Amsterdam using L-tree modules. L-tree is a very efficient and fast lexical database system written in Pascal (Skolnik 1980). 8 To save space the homograph and sense numbers have been left out. 9 Figure 4 is an abstraction of what has been found since many words are related to more than one category via coordination. For example article is related to thing or object but only the object link is represented here. Similarly body is also related to animal and not only to person. Within the current version of the Devil-system coordinated heads have been split so that in practice nodes like article and body will occur more than once in the network. A more adequate solution still has to be developed.

References Allan, K. 1977 Amsler, R. A. 1980 Berlin, Brent 1972

"Classifiers", Language 53: 285-311. The structure of the Merriam-Webster Pocket Dictionary. Ph.D. Thesis, Texas University.

"Speculations on the growth of ethnobotanical nomenclature", in: Language in Society, I: 5 1 - 8 6 . Briscoe, T. - A. Copestake forthcoming Paper presented at the LlKE-workshop, Tilburg, 1991. Dik, Simon C. 1978 Stepwise lexical decomposition. Lisse: Peter de Ridder Press.

Decomposition of lexical knowledge

39

Dik, Simon C. "Lexicon en Syntaxis in Functionele Grammatica", in: GLOT 9 no. 1/2: 15—27. 1986 Dik, Simon C. "Linguistically motivated knowledge representation", in: M. Nagao (ed.). Lan1987 guage and artificial intelligence. Amsterdam: North Holland, 145-170. Dik, Simon C. 1989 The theory of Functional Grammar. Part I: The structure of the clause. (Functional grammar series 9.) Dordrecht: Foris. Lyons, John 1977 Semantics, Vols. 1 - 2 . Cambridge: Cambridge University Press. MacKenzie, J. Lachlan 1992 Places and things, in: Michael Fortescue - Peter Harder — Lars Kristoffersen (eds.). Layered structure and reference in a functional perspective. Amsterdam - Philadelphia: John Benjamins, 253-276. Medin, D. L. - E. E. Smith 1984 "Concepts and concept formation" in: Annual Review of Psychology 35: 113-138. Procter, P. (ed.) 1978 Longman dictionary of contemporary English. London: Longman. Rijkhoff, Jan 1990 "Towards a unified analysis of terms and predications", in: Jan Nuyts — A. Machtelt Bolkestein — Co Vet (eds.). Layers and levels of representation in language theory. A functional view. Amsterdam - Philadelphia: John Benjamins. Rosch, Eleanor "Classification of real world objects: Origins and representation in cognition", 1977a in: P. N. Johnson-Laird - P. C. Wason (eds.). Thinking: Readings in cognitive science. Cambridge: Cambridge University Press, 212-222. "Linguistic relativity", in: P. N. Johnson-Laird - P. C. Wason (eds.). Thinking: 1977b Readings in cognitive science. Cambridge: Cambridge University Press, 501-521. Skolnik, J. 1980 "L-trees", Paper presented at the 6th symposium of the association for literary and linguistic computing, Cambridge. Vossen, Piek 1989 "The structure of lexical knowledge as envisaged in the Links project", in: J. Connolly - S. Dik (eds.). Functional grammar and the computer. Dordrecht: Foris, 177-199. 1991 "Polysemy and vagueness of meaning descriptions in the Longman dictionary of contemporary English", in: S. Johansson - A. Stenstrom (eds.), English Computer Corpora: Selected papers and research guide. Berlin - New York: Mouton de Gruyter. Vossen, P. - 1 Serail 1990 "Devil: A taxonomy browser for decomposition via the lexicon", Esprit BRA3030 Acquilex WP No. 009, University of Amsterdam, Amsterdam. Wilks, Y. - D. Fass - С. Guo - E. McDonald - Τ Plate - В. Slator 1988 "Machine Tractable Dictionaries as Tools and Resources for Natural Language Processing", in: Proceedings of the 12th conference on computational linguistics. Budapest, 750-755.

Semantic functions in perspective reconsidering meaning definitions Lone Schack Rasmussen

1. Introduction ^ This paper discusses the organization of the basic component in FG, the Fund, especially the way of presenting verbal predicates. In the current version of FG the information about a predicate is largely concentrated in its predicate frame, while its meaning definition plays no role in the grammatical analysis. The paper advocates the point of view that a bipartite model which establishes the semantic and the formal potential of the predicate as two parallel, equipollent dimensions will lead to grammatical descriptions with a higher degree of generality and descriptive capacity. Stating the semantic information at the most basic level of the model increases the possibility of relating the predicate structures more directly to the higher layers of the model. This avoids unnecessary complication of the layered sentence model and makes it possible to account for cases of lexical ambiguity, incoφoration of lexical elements, and other phenomena which may influence the formal structure of the sentence. In order to make the two dimensions comparable we need a further elaboration of the system of meaning definitions. A proposal for a typology of meaning definitions which defines the parameters contained in the basic set of meaning definitions will be suggested. The semantic functions are discussed as an example of how such a bipartite model may simplify both theoretical and concrete aspects of the grammatical model. In the current version of FG the semantic functions constitute a problematic category, both because of their internal characteristics and because of the hidden existence of not only one, but two sets of semantic fiinctions. The paper intends to show that a typology of meaning definitions may account for all the semantic information conveyed by the semantic functions except Agent and Instrument, these being the only semantic notions which need to be indicated separately.

2. The Fund: predicate frames, semantic functions and meaning definitions The notions here mentioned all belong to the basic level of the grammatical model in FG, the Fund, which contains all the predicates and terms from

42

Lone Schack Rasmussen

which predications can be constructed. Predicates are themselves considered as structures called predicate frames, which contain a kind of "blueprint" for the predications in which they can be used. Each basic frame in the lexicon is associated with a number of meaning postulates, through which the predicate is semantically related to other predicates of the language. If these meaning postulates add up to a full specification of the meaning of the predicate, they may be named meaning definitions (see Dik 1989: 54). The predicate frame indicates the quantitative and qualitative valencies which establish the actual expression potential of the predicates and reflect the semantic characteristics of the predicate by means of semantic functions (henceforth SFs) qualifying its arguments. The model distinguishes nuclear and satellite positions, and the nuclear SFs, to which I will confine myself in this paper, are fiirthermore divided into three types: first, second and third argument SFs, A', A^, A^. The SFs of A' are devised in such a way that they partially reflect the typology of SoAs, i. e. the states of affairs designated by the nuclear predication. A's can have the SFs: "Agent, Positioner, Force, Processed, and Zero". They represent complex meanings, dividing into a controlled type (Agent and Positioner) versus an uncontrolled one, the latter being further analyzable into an instigated and a noninstigated type (Processed and Force). The nuclear, nonfirst argument SFs are further specified as: Goal (which is A^), Recipient, Location, Direction, Source, Reference, together with the two partly overlapping "secondary" SFs "Goal(Exp)" and "Recipient Go > Ree > Ben > Instr > Loc > Temp" (Dik 1989: 226) The other more detailed one accounts for assigning subject and object at the language specific level. It is on this level that we are introduced to a hierarchal ordering of the SFs in the form of a division into A', A^, A^, which represent a syntagmatically defined set, derived from the typology of SoAs.

Semantic functions and meaning definitions

45

The main question that arises in this connection is what kind of relationship exists between the two hierarchies - and in particular, what the status is of SFs as Positioner, Processed, Force, and Zero that only appear in one of the hierarchies? As Falster Jakobsen (this volume) asks about the two hierarchies: Are they "mutually exclusive or do they combine?"

3. A proposal for a typology of meaning definitions The points mentioned in 2. constitute some of the reasons why it has not been possible to establish a definitive inventory of semantic functions in FG. They seem to indicate that the main problem is the concentration of the whole semantic load of the predicate frame on the category of SFs. This suggests that it would be preferable to establish a model where all the semantic information is introduced in the form of different independent parameters indicating type of SoA, argument values (freed from intersection with the SoA), and other relevant factors such as "semantic field" — a notion which is not taken into account by Dik. The problem is where in the FG framework it is most appropriate to introduce these notions. It cannot be done in the predicate frame because it is formulated as dependent on the formal coding of the predicate as a sentence constituent. What I propose then is to upgrade the proper semantic component of the Fund, the meaning definitions, formulating them in such a way that they become sensitive to these semantic factors. As Dik indicates, the terminal point of the meaning postulates and meaning definitions is a set of the semantically most basic predicates including notions like "cause" and "come about", "the meaning of which cannot be defined in terms of the meanings of other predicates of the same language." (Dik 1989: 85). By constructing a limited set of nonarbitrary formal predicates, modelled on the set of basic "meaning definitions" it would be possible to obtain a model better equipped to show in detail the organization of the semantic structure of the predicate. This approach may seem to be in contradiction with that of Dik, but only apparently so, because such predicates are not abstract items detached from language; on the contrary they are abstractions fi-om actual language structure, and thereby fulfill the claim that "defining meaning is a language internal affair, dealing with a network of implicational relations between the predicates of a language." (Dik 1989: 86). But, as Dik also observes: When a predicate is basic, this means that the speaker must leam and know it as such in order to be able to use it correctly. It does not mean that the predicate

46

Lone Schack Rasmussen has no internal structure. Indeed, the meaning of the most basic predicates is such that it can be analysed in terms of combinations of the meaning of semantically simpler predicates. Thus, the English predicates die, kill, and murder are basic predicates, since there is no rule of English by means of which they could be formed; they therefore belong to the lexicon of English. Their meanings, however, are structured in the sense that they can be analysed in terms of more elementary predicates. (Dik 1989: 68)

I totally agree with this point of view. I just want to call the attention to the fact that a model which can represent the last step of the meaning definitions in terms of more elementary predicates is still lacking in the FG theory. By elaborating the ultimate meaning definitions as a level which is structured in a way that makes it fully comparable with the predicate frame, it becomes possible to compare semantically similar terms that have different predicate frames, detecting what exactly is the difference that obtains between them, and to show where predicates which differ semantically may have similarities. Furthermore, separating the meaning definitions from the predicate frame level will also facilitate the comparison between the different layers of the sentence, as well as between languages. The advantage obtained by introducing such a set of elementary predicates is very similar to the one Dik obtains by introducing a more abstract "definiteness operator" (d), instead of "definite article": FG uses a more abstract "defmiteness operator" (d), which on the one hand captures the essential property common to all definite noun phrases, while at the same time allowing for different forms of expression. (Dik 1989: 16)

4. A model for a basic set of meaning definitions In the following I want to suggest how such a basic set of meaning definitions of verbal predicates could be shaped. The model was elaborated from the basic semantic structure of Spanish verbs (Schack Rasmussen 1987), so Spanish will be used for exemplification. The conclusion of my investigations was that the basic semantic pattern of Spanish verbs can be expressed by means of two dimensions, the action scheme and the semantic field. The classification of the action schemes appeared as a result of the analysis of, in the first place, a cognitively and linguistically focal group of verbs, the verbs of concrete location, i. e. verbs that designate position or movement in physical space. These verbs convey three types of activity patterns: Position (of an entity in a place), movement (of an entity in a place), and causation (of either position or movement).

Semantic functions and meaning definitions

47

A formal account of the verbs was given by means of the following basic predicates (where E stands for entity, L for location, К for instigator, ESTAR for 'be at', IR for 'move', and CAUSAR for 'cause', and s is a variable): Position·. Motion·. Causative:

ESTAR(E,L) indicates that E is located on L; IR(E,L) indicates that E moves on L; CAUSAR(K,s) ( = CAUSAR(K,ESTAR(E,L)) / CAUSAR(K,IR(E,L))) indicates that К causes either the location or movement of E.

However, these notions only partially capture the lexical structure of the locational verbs, so, in order to attain a more satisfying model, the apparatus was supplemented with a formula for directional movement (where S stands for the starting point and Τ for the terminal point): IR(E,S,T) indicates that E moves in relation to one or both of the two locative points of reference S and T, i. e. movement of an entity to or from a place. This structure has the crucial function of designating the crossing of the borderline from nonmovement to movement or vice versa. It is also important to indicate when a particular verb, apart from focusing on the nonlocal participiants, К and E, concentrates on special locational points of reference, either S, T, or L in its totality, making it figure as a profiled argument, i. e. as an obligatorily prominent element. This is signalled in the predicate frame by writing the argument with italics, thus leading to a further differentiation in formulas such as: ESTAR(£^:), lR(E,S,n IR(^,S,7), IR(E,S,T), CAOSAR(K,\R(E,S,T) CAOSAR{K,IR{E,S,T), etc. Already this framework accounts for a number of the SoA parameters in an explicit way. The position formula, ESTAR(£',L), represents the nondynamic SoAs expressed by verbs such as contener 'contain', estar 'be, stand' or tener 'have': (4)

Juan estaba en la cocina. 'Juan was/stood in the kitchen.'

(5)

Juan tiene el libra 'Juan has the book.'

We also find the nondynamic SoA as part of the lexical structure of verbs denoting an instigated SoA: CAUSAR(A:,ESTAR(£',L)): aguantar 'sustain', sostener 'support': (6)

Juan sostiene la tabla. 'Juan supports the board.'

48

Lone Schack Rasmussen

The movement formula, Ι Κ ί ^ , ^ , represents dynamic, nondirectional SoAs, such as denoted by andar 'walk', correr 'run': (7)

Juan corría por el prado. 'Juan ran in the meadow.'

It can also be instigated, CAUSAR(Ar,IR(£',L)): llevar 'carry', mover 'move': (8)

Juan lleva la maleta. 'Juan carries the suitcase.'

The movement formula, IR(£',S,T), represents dynamic, directional SoAs without prominent locational points such as those denoted by caer 'fall': (9)

Cae el avión. falls the plane 'the plane is falling.'

The formulas with profiled arguments indicate telic SoAs, both in a nondirectional and a directional action scheme: expressed by recorrer 'travel (through)': (10)

Juan recorrió todo el país. Juan travelled all the country 'Juan travelled all over the country.'

and IR(E,S,T)) denoted by cruzar 'cross', descender 'descend': (11)

Juan cruzó la calle. 'Juan crossed the street.'

They are also found instigated - CAUSAR(A:,IR(£,5,T)): quitar 'take away', sacar 'take out': (12)

Juan sacó un pollo de la nevera. Juan took out a chicken from the fridge 'Juan took a chicken out of the fridge.'

CAUSAR(/:,IR(£',S,7)): llenar 'fill', traer 'bring': (13)

Juan llenó la casa de amigos. 'Juan filled the house with friends.'

(14)

Juan trajo a casa a un amigo. Juan brought to home aOBJ friend 'Juan brought home a friend.'

Semantic functions and meaning definitions CA\JSAR(K,ЩЕ,3,Т)): (15)

49

trasladar 'move from one place to another':

Juan trasladó el sofá. Juan moved from one place to another the sofa 'Juan moved the sofa.'

The momentaneous SoAs, which I consider a special type of telle SoAs, are accounted for in a movement formula profiling only one locative argument, S o r Τ: a. IR(£',5',T): abandonar 'abandon, leave', salir 'go out': (16)

Juan salió. 'Juan went out.'

(17)

Juan abandonó la casa. 'Juan left the house.'

b. Ι^^,Β,Τ): alcanzar 'reach', llegar 'get to, arrive': (18)

Juan ha llegado. 'Juan has arrived'

(19)

Juan alcanzó la cima. 'Juan reached the summit'

The model turns out to be applicable to other verbs than those of concrete location, serving to demonstrate which type of SoA they signal lexically. Actually, in combination with the other parameter, the semantic field, the majority of Spanish verbs can be described by means of these formulas (together with a few other, rarer, variants). The notion of semantic field makes it possible to account also for the rest of the verbs that denote external SoAs, demonstrating how they integrate in a multidimensional semantic network. These verbs tum out to indicate four main types of SoAs. Apart from the already mentioned Locational type, we find verbs denoting Possession, or, rather, 'possessional' relations, Condition, i. e. SoAs expressing 'characteristics' or 'disposition', and Circumstance, i. e. SoAs with the character of 'instantiation'. The four types of SoAs are considered basic semantic fields. They are included in the model as a restriction on the predicate, giving way to more specified formulas as: ESTARcond(£',Z,), IRcirc(£',L), IRpos(£:,S,7), CAUSAR(Â',IRloc(£',S,7)), etc. The membership of a verb in one of these semantic fields depends on which selectional restrictions they ascribe to their arguments. The arguments are evaluated with respect to their characteristics as

50

Lone Schack Rasmussen

extensional or intensional, on the one side, and as denoting individuals or phases of individuals, on the other. The first paradigm is taken from logic. The notions are defined as follows: "By the extension of a term is meant the class of the things to which it is correctly applied ... The intension of a term as the set of essential properties which determines the apphcability of the term" (Lyons 1977: 158). The last paradigm, taken from Carlsson (1980), "relation between individuals" versus "relation between phases of individuals", is comparable with the contrastive concept of "essence / accidens" found in Aristotle (van Rijen 1989). All the semantic factors here enumerated are included in the predicate frame as restrictions on the predicate, thus providing more specialized formulas for individual verbs.^ As the four semantic fields constitute an obligatory choice for all verbs except operator verbs, they provide us with a set of criteria for determining the semantic type of the verb (see Table 1). Table 1. Criteria for determining the semantic type of the verb Extension Location Possession Condition Circumstance

Intension

+ +

Individuals

Phases +

+ +

+ +

"E E "E = = "-E =

+

The following examples of meaning definitions show how the two parameters in combination express the basic semantic structure of the verbs concerned: Location. ESTARloc( "Z,)': IRloc("£,"S,"T)': CAUSAR(/Í:,IR1OC("£,"L)')J:

permanecer 'stay' caer'ЫУ transportar 'transport'

Possession. ESTARpos(£'^.)': IRpos(£',S,7)': CAUSAR(A:,IRpos(£',S,7)')^:

pertenecer 'belong' heredar 'inherit' regalar 'give as a present'

Condition. ESTARcond(£','Z,)': IRcond(£','S,'7)': CAUSAR(/:,IRcond(£',"S,'r)')^: CA[JSAR{K,ÌRcond{E,'S,'RICOyy:

existir 'exist' enviudar 'become a widow' convertir 'make into' enriquecer 'make rich'

Semantic functions and meaning definitions Circumstance. IRcirc("£','L)·: IRcircC^/S,'?)': CAUSAR(/:,IRcirc("£,X)'y: CAUSARÍ/CJRcircIRC^.'S/T)')^:

51

ira%'ar'work' empezar 'begin' conducir 'drive' emborrachar 'make drunk'

To illustrate the distinction 'relation between individuals' versus 'relation between phases of individuals', let us take the verb tener, which means 'to have something', as a possession, or as a located object; cf. (5) Juan tiene el libra. 'Juan has the book'. Its ambiguity can be explained as due to a change in the type of relationship. In the first case it designates an essential relation between individuals, and in the second, an accidental relation between phases of individuals, in other words, a more situation-bound SoA in both cases between extensional arguments: tener: ESTARpos(£:,L)' / ESTARloc('Έ, "L)' The distinction between individuals and phases of individuals is found in the Spanish 'to be' verbs: ser and estar. The former differs fiirther from the latter in having an intensional locative argument. This set of criteria is sufficient to account for the basic meaning of these verbs, and clearly shows what their similarities and dissimilarities are: ser: ESTARcond(£',U)' (20)

Juan es ciego. 'Juan is blind.' (as condition)

(21)

Juan es rico. 'Juan is rich.' (as condition)

(22)

Juan es borracho. Juan is drunk (as condition) 'Juan is a drunkard.'

estar. ESTARloc("£,'X)' (4) / ESTARcirc("£,U)' (23)

Juan está ciego. Juan is blind (as circumstance) 'Juan is drugged.'

(24)

Juan está rico. Juan is rich (as circumstance) 'Juan is tasty.'

(25)

Juan está borracho. 'Juan is drunk.' (as circumstance)

52

Lone Schack Rasmussen

As can be seen, the locative theory does not exclude copula constructions from verbal predicates. On the contrary, the copulas serve to express the basic part of all four semantic fields. This means, for example, that by analyzing (20) - (25) as nondynamic predicates in the semantic fields Condition and Circumstance, respectively, we are able to give an exact description of the structural relationship to related predications such as (26), (27), and (28): (26)

Cegaron a Juan. They blinded JuanOBJ 'They blinded Juan.'

(27)

La guerra enriqueció a Juan. the vv'ar made rich JuanOBJ 'the war made Juan rich.'

(28)

Emborracharon a Juan. they made drunk JuanOBJ 'they made Juan drunk.'

(26) only implies (20), not (23), and (27) only (21), not (24), while (28) only implies (25), not (22). The implication pattern (with ser) shows that cegar and enriquecer belong to the semantic field Condition, and emborrachar (with estar) to that of Circumstance. The interpretation of ser and estar along this line suggests that instead of regarding the copula as an "empty" element, "a semantically innocent supportive element with no deep conceptual or philosophical significance" (Dik 1989: 173), it would be preferable to take it as a signal of a 'nondynamic SoA in some semantic field'. Taking this definition as a language-independent starting point, language-individual differences in copula choice could be explained as different solutions to the problem of how much of this meaning it is necessary to signal. In this way the theory can account for the Russian nonmanifestation as well as for the sixteen copulas of a language like Koyukon from the same perspective (see Portesene 1992). Because the copula constructions have the most simple structure, they may be taken as representations of the basic level of verbal predicates. Their exclusion from the set of verbal predicates would lead to a distortion of the whole system and prevent the lexical structure of all verbs fi-om being accounted for within the same implicational fi-amework. The analysis here proposed suggests that the basic system of meaning definitions for verbs should be conceived of as a compositional set of factors, as a semantic frame comprising the basic semantic principles for their lexical structuring. The basic scheme comprises two paradigms: the action scheme, containing the three basic types 'static, dynamic, and causative', with the

Semantic functions and meaning definitions

53

possibility of differentiation by profiling the arguments and indicating +/directionality, and the semantic field notion which differentiates four basic types by means of the criteria individuals or phases of individuals and extension/intension. The meaning definitions and predicate frames of the individual predicates can be seen as choices from the basic typology. The difference between Dik's approach and the present semantic model lies in that what we might call the Localist model is able to account directly not only for all the different types of SoAs but also, at the same time, the exact relationship among all the arguments involved in the SoA, making more explicit the exact location of the predicate in relation to the rest of the lexicon by means of the semantic field notion. Furthermore, it appears more typologically adequate. Because of its higher degree of formalisation it is not so easily bound to idiosyncratic features of one language. Spanish, for example, does not have a lexical item like English move expressing 'nondirectional movement' without further nuances. The notion is expressed by means of a reflexive construction moverse (similar to Danish bevœge sig). This type of construction appears as a product of 'predicate formation', and as such I consider them to be disqualified as basic meaning definitions. As for Spanish, this means that the most basic lexical item for expressing movement will be ir, which is directional. This confirms that there is a need for a basic set of meaning definitions figuring as an independent level.

5. The relationship between SFs and meaning definitions Assuming that a semantic model such as the one proposed here can determine the meaning definitions of verbal predicates, the next problem that arises is what the relationship between the meaning definitions and the SFs of the predicate frame is. The meaning definitions, which represent the ultimate definition of the semantic potential of the predicates, were formulated as predicates with a number of arguments; these can be variables or a fixed value filled by a semantic component (CIEGO 'blind', for instance, is Terminal Point in cegar 'to make blind'(26)). Assuming that there is a relation between the semantic and the expression potential of a predicate, there must exist a correspondence between the two types of arguments, the Semantic Arguments of the meaning definition and the semantic fimctions (SFs) of the predicate frame. An obvious analysis is that Semantic Arguments are notions that form part of the action schemes, while SFs reflect the type of SoA of the predicate frame. Therefore the SFs can be defined as 'manifest Semantic Arguments'.

54

Lone Schack Rasmussen

The Localist model differs from the FG model in presenting the three types of action schemes as the most important factor, with the other semantic factors dependent on or related to them. The argument values attached to the different types of action schemes were defined as Causator, Entity, Location, Starting Point, and Terminal Point, corresponding to the arguments K, E, L, S, and T. The question, then, is if this inventory of Semantic Arguments contains a sufficient amount of information to account adequately for the relational values of the predicate frame, the SFs. An analysis of the possibility of letting the inventory of Semantic Arguments represent the SFs has shown that in Spanish only one additional amplification is needed in order to account for the semantic values related to the sentence constituents. Apart from the argument values, the only further information required is an indication of whether the first argument, either К or E, is an Agent.

6. Agent, Instrument, and Force As the notion of Agent appears with lexemes with both dynamic and causative action schemes, it is defined as an extra value that can be attached to the first argument of these schemes: (29)

Juarix bajó la cuesta. 'Juan went down the hill'

(30)

Juarip, abrió la puerta. 'Juan opened the door'

Because Agency in itself implies an activity which is typically, but not necessarily, accompanied by the features Control and Intentionality, it is expressed by means of a causative predicate that modifies the first argument, К or E. This modified argument is furthermore coreferential with the modifying Kargument. The formal paraphrase is as follows: (31) a. IRx(£,L) [A] b. CAUSAR(^,5) [A]

(where E = A) (where К = A)

- where χ = a semantic field variable, and [A] = CAUSAR(K,IR(E,L)). As the Agent formula is rather unwieldy, the modification will normally be stated simply in the form of the [A], unless an Instrument is involved (see (34)).

Semantic functions and meaning definitions

55

By considering Agent a restriction on the meaning definition, the model obtains a higher descriptive capacity. It can state precisely the ambiguity of 'IR'schemes, which may denote intentional or unintentional movement, without having to assign two different action schemes ('movement' versus 'causation') to the verb, as is a frequent, but counterintuitive solution to this problem: (32)

(32')

Juan rodaba por la cuesta. Juan rolled on the hill 'Juan rolled down the hill.' \K\oc{"Juan,"cuestaf [+/-A]

Another advantage is that the unmodified 'CAUSAR' scheme can be reserved for the description of "pure" causal relations such as: (33)

La guerra arruinó a la familia. the war ruined OBJ the family 'the war ruined the family.'

When an Instrument is involved, either manifestly, as in example (34), or lexically incorporated, as in verbs like acuchillar, which can be paraphrased 'matar con cuchillo' 'kill with a knife', or manosear, paraphrasable as 'tocar con las manos', the Agent modification has to be spelled out in fiill because Instrument is part of it: (34)

Juan abrió la puerta con una Have. 'John opened the door with a key.'

(34')

CAUSAR("/Man,IRloc( "puertaAPERTURA,Tyy [A] - where [A] = CA\JSAR{"Juan,mioc{"Have,S,"puerta))

Instrument is defined as the E-argument in the modifying causative function of Agency, thereby demonstrating the exact relation between Agent and Instrument, namely that the Agent causes the Instrument to effect an SoA. This analysis has the advantage of clarifying the somewhat chameleonic status of Instrument, which has caused linguists so many problems. Furthermore it explains how in sentences such as (35), Instrument, normally realized as a satellite, may appear in a higher place in the semantic hierarchy than expected. The explanation is simply that an Instrument in a subject function does not break the hierarchy, because if the first argument slot is not filled.

56

Lone Schack Rasmussen

the E-argument in the modifying causative function is the first argument to be expressed: (35)

Esta llave abrió la puerta. 'This key opened the door.'

(35 ')

CAUSAR(/:,IRloc( "puerta APERTURA,!)^ [A] - where [A] = CAUSAR(i:,IRloc('7/ave,S, "/JMertó))

In FG, Agency is defined in terms of the SoA parameter Control. If we analyze Control as a feature that appears when the Causator is animated, thus being dependent on the restriction (anim) of the K-argument, we can avoid establishing an extra SF, Force, as is done in FG in order to account for the differences in Control observed between (30) and (36) (Dik 1989: 106). By defining the SF of "viento" in (36) as an inanimate Agent, its lack of Control is fully accounted for. (36)

El viento abrió la puerta. 'the wind opened the door.'

In addition, the representation of "viento" (36) and "Have" as K-and E-argument of the modifying causative predicate, respectively, adequately reflects the semantic difference between them, namely that only "Have" is controllable.

7. Reconsidering the inventory of SFs With this inventory of semantic argument values as our point of departure we can simplify the inventory of SFs; some of the SFs are variants that arise because of differences, not in the arguments as such, but somewhere else in the semantic composition of predicates. The description of SFs can be simplified in two ways. Taking into account the parameter of semantic fields. Reference, Possessor, and Recipient appear to be locational variants. The semantic analysis of predications where these SFs appear, shows that the first two both signal Location (L), Possessor in the semantic field of Possession and Reference in that of Condition: (37)

This house is John's. (Dik 1989: 175)

(37')

ESTAKposQiouseJohny (E) (L)

(38)

The boy resembles his father (Dik 1989: 101)

(38')

ESlhKconá{boy:fatherJ (L) (E)

Semantic functions and meaning definitions

57

Besides Location, Direction, and Source, Dik's A^ also comprises Recipient, which can be identified as an animate Terminal point (T) in nonlocational semantic fields (=dir): (39)

Juan le dio el libro a Maria. Juan her gave the book to María 'Juan gave the book to Maria.'

(39')

CAUSAR(ywa«,IRpos(//èro,Juan,Maria)i)^) [A] (E) (S) [an](T)

With respect to the SF Experiencer, Dik states that it must be treated as "a secondary semantic function, a "footnote" to the other first argument functions" (Dik 1989: 102), because "More generally, it often seems to be the case that languages model their (exp) constructions on the pattern of the corresponding ( - e x p ) constructions. This implies that the opposition (+exp) does not necessarily have a deep impact on the grammatical organization of natural languages" (Dik 1989: 99). The theory outlined here makes it possible to explain the reason why. Experience is a semantic factor, but not of the same type as Agent or telic. The similarities between ± Exp. constructions are due to the fact that they exhibit the same types of pattern of activity, i. e. types of functions, and the same basic semantic fields. And the difference can be ascribed to the fact that they belong to two different 'semantic universes', one designating 'mental' (epistemic) SoAs, the other 'nonmental' (external) So As. This way, for example poseer 'own' should be analyzed as as ESTARpos.ext(£'^,)', trabajar 'work' as IRcirc.ext(£,'L)', and saber as ESTARpos.epi.(E',L')', pensar 'think' as IRcirc.epi(£','L)^ etc.^ The signalling of another semantic Universe fully explains its secondary status in relation to the other SFs. The other parameter to take into account is that of 'action schemes'. By distinguishing the argument values from the action schemes they are part of we can identify several of the SFs as variants. Positioner can be specified as "the Agentive Causator in a nondynamic SoA": (40)

Juan guarda el dinero en un viejo calcetín. 'Juan keeps the money in an old sock.'

(40 ' )

CAUSAR( "yMa«,ESTARloc( "su dinero, "viejo [A]

calcetin)^

The combination of nondynamic structure with Agency is acceptable because it is part of a causative scheme. Since Positioner and Force react in the same

58

Lone Schack Rasmussen

way to the passive test, they appear as Agent variants, to judge from their syntactic behaviour. (41)

£•/. dinero era guardado por Juan en un viejo calcetín. 'the money was kept by Juan in an old sock.'

(42)

La puerta fue abierta por el viento. 'the door was opened by the wind.'

This amounts to saying that on a more basic level these SFs correspond to the same semantic notion. Thereby both Positioner and Force tum out to be superfluous as independent SFs (see 6.) Processed and Zero can be analyzed as the Entity in a dynamic and a nondynamic SoA, respectively. This solution avoids overlapping and offers, at the same time, a positive definition of the SF Zero: (43)

La roca cayó. 'the rock fell down.'

(43')

IRloc('Voca,S,T)' i

(44)

El plato estaba en la mesa. 'the plate was on the table.'

(44')

^SlAR\oc{"plato,"mesaf

Dik argues strongly in favour of separating Goal and Processed. But the separation is not necessary. In the model presented here they both represent the Entity argument — that is why the similarity arises - but in different types of SoA, namely in caused versus noncaused movement. Therefore Goal can be analyzed as the Entity in a dynamic SoA that fills the s slot in a causative function, as Dik does. But since the difference is derivable from the type of action scheme it figures in, we avoid redundancy by simply defining it as an Entity: (45)

Juan movió la roca. 'Juan moved the rock.'

(45')

CAUSAR( "7маи,1Шос( "госаХ)^У) [A]

The examination of the possibility of letting the inventory of Semantic Arguments represent the SFs shows that the only additional piece of information which is needed in order to account for the semantic values related to the sentence constituents is an indication of whether the first argument, either К

Semantic functions and meaning definitions

59

or E, is an Agent. Therefore Agent, and the Agent-dependent notion Instrument, must be included in the typology of meaning definitions, together with the values of the Semantic Arguments, Entity (E) together with a locative point of reference, Location (L), Source (S) or Terminal point (T), and Causator (K), as part of the semantic values that can be attached to the set of basic meaning definitions. In conclusion, we can say that the analysis seems to confirm that a division of labour prevents unnecessary overlapping. In a bipartite model the predicate frame will state the formal (quantitative) properties of a predicate, while the meaning definition accounts for the semantic (qualitative) properties. Apart fi-om reducing the number of semantic functions, the establishing of a semantic parallel to the predicate frame in the form of a set of basic meaning definitions has the advantage on a more general level of giving us exact criteria for deciding how many SFs there can be, and also offers an explanation of why a SF has the fiinction of A', A^ or A^. A comparison between the Semantic Arguments of the basic set of meaning definitions and the three argument types A', A^, A^ of the predicate frame demonstrates that A' corresponds to the first (nonlocative) argument of the predicate, i. e. К or E, depending on the predicate type +!— causative); A^ represents the second argument, i. e. E or L, depending on the predicate type { + ! - causative), and A^ the locative argument (in a three-placed relation +!— causative) (see Table 2). Table 2. Semantic Arguments in meaning definitions / semantic functions in predicate frames Meaning definition

A'

A^

A3

CAUSAR

К Agent Positioner Force

E Goal S Source

L Location

IR

E Agent Processed

L Location

ESTAR

E Zero

L Location Possessor Reference

Τ Direction Recipient Τ Direction

Some of the Semantic Arguments occur more than once in order to show which values A', A^, and A^ can take in the different predicate frames. The analysis of the correspondence shows that, even though there is no one to one relationship between the two systems — the values of the Semantic

60

Lone Schack Rasmussen

Arguments and the semantic functions of the arguments in the predicate frame - there does exist a relatively high correspondence between them. The division of SFs into A l ' , A^, and A^ mirrors the difference in the semantic argument status: the nonlocative ones correspond to A' and A^, and the locative ones to A^ and A^. The distinction between local and nonlocal Semantic Arguments can be substituted by the more vague notion of "central/peripheral" used by Dik to characterize arguments (Dik 1989: 226). And this fuzzy border can be fiilly explained as a product of the appearance of the same semantic argument values, but in different types of action schemes, which thus represent different SoAs — L (Location), for example, is A^ in noncausative schemes, and A^ in causative ones. This fact, though, can only be appreciated if the structure of the predicate frame is observed from the point of view of the meaning definitions (Table 2). This proves that both levels are pertinent, for analytical as well as for theoretical reasons. The existence of two parallel dimensions, meaning definitions and predicate frames, makes us able to define the borderline between A', A^, A^ not as fuzzy but as flexibleлvithin a definite system. This means that the model makes possible a reconciliation of the two sets of SFs by defining the ones belonging to the larger set as variants. Another result of elaborating the meaning definitions as a semantic parallel to the predicate frame is the possibility of giving a more stringent treatment of the prototypical assignment of A', A^, and A^ (see Table 2). The expression potential of the lexical items can be accounted for in a more systematic way by showing that a predicate can have various predicate frames attached to its meaning definition. An example illustrates that the difference between semantically closely related predicates such as salir 'go out' and abandonar 'leave' does not lie so much in the difference of action shemes as in their different array of predicate frames: abandonar Meaning definition: IRloc(£',5,T) Predicate fi-ame: АЦЕд), A2(S) a. Juan abandonó la ciudad. 'Juan left the town.' b. *Juan abandonó la ciudad al campo. 'Juan left the town into the countryside.' salir Meaning definition: IRloc(£',5',T) Predicate frame 1: A1(E±A), A2(S) predicate frame 2: АЦЕд), A2(T) predicate frame 3: A1(EA), A 2 ( S ) , A 3 ( T )

Semantic functions and meaning definitions

61

(47) a. Juan salió de la ciudad. 'Juan went out of the town.' b. Juan salió al campo. 'Juan went out into the countryside.' c. Juan salió de la ciudad al campo. 'Juan went out of the town into the countryside.' This approach also facilitates the description of predicates with nonprototypical assignment of A', A^, and A?. They can be described in terms of markedness. Typical examples are 'filling' verbs, 'instrumental' verbs, and verbs with incorporated lexical elements: (48) a. JohUi^ filled water^ into the bottkj. b. John^^ filled the bottle^ with water

(see Dik 1989: 106)

(49) a. Johny^ filled water^ into the bottlej with a fiinnel^. b. Johny^ used a fiinnel to fill the bottlej with water^. c. John^ used a funnel to fill water^ into the bottlej. (46b), (47b), and (47c) can be analyzed as nontypical assignments of the A2 position to the Terminal Point. In the case of (46) this amounts to saying that (a) and (b) represent two different predicate frames, which are considered as alternative representations of the same meaning definition. We account thereby at the same time for the lexical identity of the verbal predicate and for its different coding possibilities. Another important consequence of the improved SF inventory is that it makes it possible to reformulate the semantic function hierarchy proposed by Dik (1989: 226), in such a way that it accounts for the relationship between semantic and syntactic functions without leaving any of the semantic functions out of the hierarchy. A semantic function hierarchy based on the Localist model would be: (50)

Agent (as a restriction attached to the first argument) > Instrument (because of its dependence on an Agent) > Causator (K) > Entity > Locative (with Source and Terminal as variants) > and Modification (corresponding to satellite arguments) - or: A, I, К ,E, L, S, T, M

Taking into account my inteφretation of Instrument, this semantic function hierarchy is in accordance with that of Dik: Ag(=K,E) > Go(=E) > Rec(=T) > Ben(= outer T) > Ins > Loc(=L) > Temp

62

Lone Schack Rasmussen

I agree with Dik's view on Beneficiary as a satellite SF (Dik 1989: 195), and analyze it as an "outer" Terminal Point. Its appearence in the semantic function hierarchy in spite of its being a satellite (not commented on by Dik) is most likely to be explainable as due to the fact that it prototypically refers to an animate entity, a factor which has a strongly upgrading effect (Dik 1989: 34).

8. Conclusion The Localist model offers a framework with a high degree of generality and descriptive capacity, because it does not insist on concentrating the entire relational meaning in the SFs but, instead, presents this part of the sentence meaning as depending on a combination of different semantic factors: action scheme, type of argument, semantic field, semantic universe, and selectional restrictions. In this way the SFs appear as a set established by independent criteria, and not as a generalization over syntactic functions. By means of meaning definitions, lexical meaning is defined as a multidimensional network combining several semantic parameters. This approach avoids a conflated model on the lexical semantic level, in the same way as Dik avoids conflation in the description of the organization of the sentence structure by using the layered model, at the same time as it satisfies his claim that meaning should be defined as "a language internal affair, dealing with a network of implicational relations between the predicates of a language." (Dik 1989: 86). By using the Localist model we not only obtain a simpler inventory of SFs, but also state the necessary semantic information at the most basic level of the model. This increases the possibility of relating predicate structures to the higher layers of the model and accounting more directly for its constructional and derivational possibilities. Notes 1 I would like to thank Simon Dik, Elisabeth Engberg-Pedersen, Lisbeth Falster Jakobsen, Casper de Groot, and Peter Harder for helpful suggestions. Errors are all my own. 2 In the formal meaning definition intensionality is expressed by " and relation between Phases of Individuals by ". To make these formulas fit for representing the coded meaning of actual verbs in an adequate way it is furthermore necessary to add a time variable indicating 'what can be said on the basis of information available at some reference point (tj) about the occurrence of some SoA at some moment (t) or interval (I)'. In the formulas these specifications are expressed as t = moment; I, J = variables of interval. All dynamic verbs are related to Interval, because a movement implies necessarily reference to at least two moments, while nondynamic verbs only needs reference to one moment to have a truth value. Many

Semantic functions and meaning definitions

63

verbs also have idiosyncratic restrictions of individual verbs, such as the 'direction' or 'speed' of the movement, etc. My analysis of "the lexical structure of the Spanish verbs" indicates that the distinction ± experience is just the tip of the iceberg. To obtain a system that can account for all types of verbal lexemes, it is necessary to operate with more than two universes. For the moment it seems that a model, to be adequate, requires at least the following semantic universes: An External, an Epistemic, a Deontic, a Normative, and a Logical Universe. They are formally included in the model as a further modification on the semantic field restriction.

References Carlsson, Gregory N. 1980 Reference to kinds in English. New York: Garland Publishers. Dik, Simon C, 1978a Stepwise lexical decomposition. Lisse: The Peter de Ridder Press. Dik, Simon C. 1978b Functional Grammar. Amsterdam: North Holland. Dik, Simon C. 1989 The theory of Functional Grammar Part I: The structure of the clause. Dordrecht: Poris. Falster Jakobsen, Lisbeth this volume "Peφective, markedness, and paradigmatic relations between predicates. A case study of Danish". Fillmore, Charles 1968 "The case for case" in Emmond Bach - Robert T. Harms (eds.), Universals in linguistic theory. New York: Holt, Rinehart and Winston. Fortescue, Michael 1992 "Aspect and superaspect in Koyokon: An application of the Functional Grammar model to a polysynthetic language", in: Michael Fortescue — Peter Harder - Lars Kristoffersen (eds.). Layered structure and reference in a functional perspective. Amsterdam/Philadelphia: Benjamins, 99-142. Hengeveld, Kees 1986 "Copular verbs in a functional grammar of Spanish", Linguistics 24: 393 —420. Hoekstra, Teun - Harry van der Hulst - Michael Moortgat (eds.) 1981 Perspectives on Functional Grammar Dordrecht: Foris. Lyons, John 1977 Semantics I-IL Cambridge: Cambridge University Press. Mackenzie, J. Lachlan 1981 "Functions and 'cases'", in: Teun Hoekstra - Harry van der Hulst - Michael Moortgat (eds.), 299-318. Schack Rasmussen, Lone 1987 panske verbers semantiske grundstruktur En lokalistisk funktionsteori. [The basic semantic structure of Spanish verbs. A localistic functional theory. Ph.D. dissertation. University of Copenhagen.] forthcoming El sistema semántico de los verbos españoles. [The semantic system of the Spanish verbs.] van Rijen, J. 1989 Aspects of Aristotle's Logic of Modalities. (Synthese Historical Library, vol. 35.) Dordrecht: Kluwer.

Transitivity and the treatment of (non)prototypicality in Functional Grammar Louis Goossens

1. Introduction' It is somewhat suφгising that the term transitivity, which all of us have been educated with linguistically and which is a key notion in about all descriptive grammars as well as in a lot of more theoretically oriented linguistic work, is manifestly avoided within the tradition of Functional Grammar. Although it was, be it sparingly, present in Dik (1978), it is, for example, strikingly absent in the index of Dik (1989). The main reason for this, is, of course, that other concepts have taken over: Object and (marked) Subject assignment has been an important area of investigation in FG from the start. Yet it seems to me, there are reasons to rediscover the notion transitivity as it figures in recent work within the framework of Cognitive Grammar, notably Taylor (1989), Langacker (1991) and an unpublished PhD thesis by Sally Rice (1987). Since FG has, and (in my view) should have, pretensions of being a cognitively oriented theory of language, it is worth asking the question how the insights formulated within Cognitive Grammar can be worked into the underlying structures of FG. In what follows I first try to clarify the Cognitive Grammar view of transitivity, discussing material from Rice (1987) which consists in acceptable and nonacceptable English passives, and where transitivity as conceived by Cognitive Grammar is assigned an explanatory role. In sections 3 and 4 I explore how and to what extent FG can (could) incoφorate these insights. The final section draws a number of conclusions, both in relation to the transitivity issue and concerning the implications of my exploration for FG.

2. Transitivity in Cognitive Grammar 2.1. Some generalities As Taylor (1989: 198) points out, "Cognitive Grammar ... views a construction as the pairing of a specification of form with a specification of meaning." A first step to be taken therefore is to provide an initial characterization of the form of the transitive construction. Again we can tum to Taylor

66

Louis Goossens

(1989: 206), who provides us with a "formula" for this form. We quote it under (1). (1)

The syntactic properties of the (transitive) construction may be presented by the formula NPiVtransNP2, where NPi and NP2 stand for the subject and direct object, and Varans is a transitive verb. In its protoypical instantiations, both NPs have specific reference, while the verb is realis, i. e. affirmative and indicative, and in a reporting tense (either present or past). These latter characteristics fall out from a specification of the construction's meaning.

Obviously such a characterization is circular as long as we do not know what a transitive verb is. Here Taylor follows traditional wisdom, which identifies transitive verbs on the basis of a number of distributional properties, the most important of which is, of course, that transitive verbs allow passivization. As regards the meaning of a transitive construction, Taylor makes clear that it is difficult to characterize in a few words. Rather, prototypical instantiations correlate with a series of semantic properties which together are intended as the analytical presentation of "an experientially primitive gestalt, cognitively simpler than any of its component parts" (Taylor 1989: 206, where a list of eleven properties can be found). We do find a formulation of the transitive prototype, however, in Rice (1987: xii), which is reproduced here as (2). (2)

... transitivity is advanced as a grammatical category that has an underlying cognitive basis. It is shown to reflect certain prototype effects that Rosch demonstrated are characteristic of human categorization. The prototypical transitive event is one in which two asymmetrically related entities are involved in some unilateral activity. The activity requires forceful movement or some energized transfer instigated by one entity resulting in either contact with or some observable effect in the other. A view of transitivity as a graded category organized around a prototype can accomodate sentences coding events that depart from the canon, either because the events are construed as transpiring in nonphysical space or because they do not invoke enough facets of the prototype, (my emphasis)

Armed with this prototype account of the notion transitivity as conceived in Cognitive Grammar, we can now proceed to a discussion of how it is handled by Rice to explain nonprototypical passives in English.

Transitivity and (non)prototypicality

67

2.2. Passivization and nonprototypical transitivity Much of Rice's dissertation is concerned with an investigation of the extent to which sentences coding events that do not come under the transitive prototype can be construed as if they were transitive. Since passivization is the crucial test in English to find out about this, she surveys a series of sentence types where in some cases the passive is acceptable, but to be rejected in others. Without trying to do justice to all the details of Rice's exploration, we survey some of her test sentences together with the essence of her (cognitive) explanations. Note that we take over both Rice's (American English) spelling and her judgements about acceptability. It is clear that not all speakers of English will without any qualification agree about each of those judgements, but this is unavoidable when we are concerned with the nonprototypical realizations of a linguistic category, as is the case here. As far as I can see, however, this does not affect the general validity of Rice's account. A first group of instances is quoted under (3) to (8). (3) a. The narrow footbridge was trampled on by the kindergartners. b. *The narrow footbridge was gone on by the kindergartners. c. Their licentious behavior was frowned on by the neighbors. d. The new policy was reported on at great length by the droning bureaucrat. e. *The town is bordered on by the river and the lumbermill. (4) a. John was rushed to by Mary, who needed a favor. b. John was spoken to by Mary. c. *The drawing-room was retired to by the dinner guests. d. The poor and needy were attended to by the priest. e. *Members of the hospital staff were spread to by the virus. (5) a. The referee is being hit at by the boxer. b. The British soldier was aimed at by the IRA gunman. c. John was barked at by his shrewish sister. d. * London was started at by our European tour. e. *The sight of blood is sickened at by most men. (6) a. *The bed was fallen off by the child. b. *The family trust can be lived off by John, once he has reached 21. (7) a. *The treehouse was descended from by the child. b. *The meadow was strayed from by a few errant cows.

68

Louis Goossens

(8) a.

In the parking lot after the game, the referee was fought with by the losing coach. b. The doctor was conferred with by an old patient at the cocktail party. c. *John couldn't be lived with by Jim, so he moved out. d. ^Leaves were clogged with by the gutter.

In Rice's terminology, this first group comes together under the heading prepositional verbs. The term is not unproblematic, because for most of these sentences it would be difficult to claim that the preposition belongs with the verb (in other words, that the prepositional phrase would be a prepositional object). On the other hand, it is clear that they have a common denominator: in all of these instances the active would have a prepositional phrase, which, to put it in FG terms, represents the first level-one satellite qualifying a oneplace nuclear predication, or, in some cases, the second argument in the nuclear predication. Since we have prepositional phrases, we obviously have no canonical transitive constructions (as defined in (1)). From that point of view there is nothing extraordinary about the fact that some of them do not permit passivization. A number of others do, however. An explanation is required to account for this difference in behaviour with respect to the way in which they can be matched with the transitive prototype to the point that they permit passivization. Simplifying a little, the question is whether the event represented can be viewed as involving an energized impact from one entity onto another (in casu the entity in the prepositional phrase). With respect to (3) this is the case for (3 a), where trample may suggest this energized impact, but not for (3b), which contains a neutral verb of movement. In (3c) and (3d) the state of affairs is metaphorically viewed as involving an impact on their licentious behavior and on the new policy respectively (both are, as it were, "attacked"), but not in (3e), which represents a "configurational" state of affairs; the "event" is not construed as involving a transfer of energy. Similar accounts can be offered for (4) — (8). In general, it appears that to, at and on are highly transitivizing: as a rule there is an idea of directionality towards the entity in the prepositional phrase; if that directionality is also conceived of as energized, it can be matched with the prototype. The contrary is true for off and from, whose basic meaning indicates directionality away from; hence it is impossible to reinterpret the instances in (6) and (7) as expressing an idea of energized impact from A (the first argument) onto B. The reason why (4c) is impossible, is that, although there is directionality towards the drawing-room, there is no idea that the guests retiring to it have

Transitivity and (non)prototypicality

69

an impact on it. (4e) is different in this respect; the members of the hospital staff are indeed affected by the virus. On the other hand, the event is construed, not as a directed approach, but (metaphorically, of course) as outward, diversified extension which is too different from the canonical transitive event to be associated with it. As regards (5d) and (5e), at does not express direction towards, but (literal or metaphorical) location at, which explains the impossibility of the passive. Finally, there are the items where the preposition is with, which appears to go either way. Again the explanation amounts to a construal as directed transfer of energy in (8a) and (8b), but such a conceptualization is not possible for (8c) or (8d). Note, for that matter, that this difference in the use of with reflects the diachronic development from an original meaning which can be paraphrased as 'against' to the later "accompaniment/association" sense, from which in tum the "instrumental" sense has developed. A second group is termed imperfective verbs. Again somewhat unfelicitously, but also here the common feature is clear. All the instances have nondynamic verbs, as can be ascertained from the exemplification under (9) to (11). (9)

a. *The swimming pool is contained by the yard. b. The lake is contained by the dam. c. The house is encircled by (the) trees.

(10) a. *The couple next door is known by John. b. Mary is loved by John. c. The broadcast is seen by millions. (11) a. b. c. d.

Line Line Line Line

A В В A

intersects with line B. intersects with line A. is intersected by line A. is intersected by line B.

Given the nondynamic character of these states of affairs, the normal expectation would be that they do not match the (essentially dymamic) transitive prototype. It is therefore the nonstarred passives that require further explanation. (9b) can be construed as an energized event, because there is an idea that the dam as it were counteracts the pressure exerted on it by the water of the lake. For (9c), Rice assumes that the "configurational" state of affairs is mentally construed as a metaphorically dynamic process in which the trees encircle the house. A similar explanation is offered for the passives under (11), where a symmetrical static configuration is conceived as dynamic.

70

Louis Goossens

The items in (10) all contain Experiences (to use FG terminology). Whereas knowing is difficult to construe as an energized event, this is not so for loving. With respect to seeing, Langacker (1991: 304) observes the follovi^ing: "The extension is presumably grounded metaphorically, either through specific metaphors such as SEEING IS TOUCHING, or more generally, through the shared path-like nature attributed to such phenomena as energy flow, gaze, and directed attention." The final set of examples is of a miscellaneous nature. Rice categorizes them as involving "nonagentive participants" ((12) and (13)), "cognate objects" (14), and "metric expressions" (15). We leave it to the reader to construct the "active" counterparts on which passivization is attempted. (12) a. *Increased employment was seen by 1986. b. Ί*Αη ankle was fractured by John. c. *A pipe was burst by the cooling machine. (13) a. *A Nobel prize was won (for Garcia Marquez) by Cien Años de Soledad. b. *John was struck by the answer in a dream. (14) a. *The death of a saint was died by Susan. b. The song was sung by Caruso. (15) a. 7Five miles were/was run by the entire team. b. The five miles was run by the entire team.

The reason why most of these instances are rejected is probably obvious by now. The acceptability of (14b) is connected with the fact that the song had some sort of préexistence before Caruso sang it (Caruso "performed" it, and in that sense, metaphorically at least, did something to it). Similarly, (15b) is better than (15a), because there is an idea of a premeasured distance that is as it were "performed" by the team.

3. Towards a notion "prototypical transitivity" in FG 3.1. Preliminary observations The preceding section has sufficiently illustrated, I hope, that the Cognitive Grammar view of transitivity, which recognizes a conceptually defined transitive prototype and which, as it were, measures other "events" (states of af-

Transitivity and (non)prototypicality

71

fairs) involving two entities related by some predicate with respect to the way in which they match this prototype, has considerable explanatory power. The question at this point is how this account relates to the FG view of Subject and Object assignment, which is essentially a multi-factor matter (see Dik 1989: chapters 10 and 11, and especially 11.4). As a first point we should emphasize that Cognitive Grammar is aware of the "discourse function" of grammatical relations (like subject and object), voice, and case. Langacker, for example, stresses that the cognitive salience of subjects can be understood in terms of their "topicworthiness" or "topicality", which is "resolvable into several factors that pertain to different aspects of the conception of clausal participants" (Langacker 1991: 306). Secondly, although the Semantic Function Hierarchy, especially in its revised form (Dik 1989: 11.3), goes some way towards a "graded" account of Subject/Object assignment prior to a consideration of its discourse functions, it does not suffice to give a full account of the role played by transitivity. Saying that second arguments are "more central" than third arguments and that among second arguments Goal has a privileged position with respect to marked Subject assignment (and passivization) does not yet explain why this is so, nor under what conditions other nonfirst arguments than Goals can become marked subjects. Let us next see how we can work a more explicit account of transitivity into the present state of the art in FG.

3.2. Goal as the key to prototypical transitivity in FG The obvious starting point is to associate the transitive prototype with the presence of the semantic function Goal, as defined in Dik (1989: 103), repeated here as (16). (16)

Goal: the entity affected or effected by the operation of some controller (Agent/Positioner) or Force.

A state of affairs which contains a Goal can be viewed as coming close to matching the transitive prototype. Indeed, it involves two asymmetrically related entities, and the Goal is affected or effected by the Agent, Positioner or Force. On the other hand, by including So As involving an Effected Goal into the prototype, we have broadened this prototype to the extent that an Effected Goal denotes an entity which depends for its existence on the successful realization of the SoA to which it belongs. The SoA does not denote contact between two entities, nor observable impact of one entity on another, because the second entity only comes into being as a result of it. Moreover, if the

72

Louis Goossens

first argument is a Force, it is often less individuated; and if the first argument is a Positioner the SoA is nondynamic. Still, we would like to posit that the transitive prototype for English covers all SoAs involving a Goal, whether Affected or Effected, because, as far as I can see, that is what is indicated by the facts of passivization. The implication of this position is, of course, that the definition of a prototype is to some extent a language-particular matter. It is, therefore, from the point of view of English that the consequences of this view will be further explored. In (17) we give a number of instances that come under this extended prototype. Only (17a) fiilly matches the prototype as defined in (2), but the facts of passivization in English leave little doubt that also (17b — e) can be taken to express centrally transitive states of affairs. Note that for our purposes it is useful to add the specification Aff(ected) or Eff(ected), or, in a case like (I7e) (the active counteφart of (14b)) a combination of the two. (17) a. b. c. d. e.

JohriAg hit Billoo (Aft) JohnA,g built this summer-houseao (E«) Geoffpo kept the топеуоо (AFF) The windfo blew Johnnyoo (AA) the hedge^r Carusosang the songoo (AÍT ЕЙ)

It follows, however, that only those second arguments that come under this (extended) prototype can be given the semantic function Goal. In other words, the assignment of Goal will have to be somewhat more restricted than in average FG practice so far. We have a look at a number of second arguments from that point of view in the next section.

3.3. Second arguments/terms other than Goal The following instances contain second arguments (or just second terms) which are not Goals, though some, or even most of them, have been labelled as Goals in at least some FG publications. Since my point here is that these second terms are not Goals, I have offered other characterizations, but I restrict my justification for them to a minimum; these proposals do not all have to be regarded as definitive. (18) a. The blue boxo contains the red balls^^f b. Johno poss has the largest housej^^f c. Johno resembles Jac^Ref

Transitivity and (non)prototypicality

73

Ref is in line with the definition for Reference as "the second or third term of a relation with reference to which the relation is said to hold" (Dik 1989: 103). With the added characterization of the first argument in (18b) as Possessor I add some further subcategorization to the rather diffuse category of States, but this has, of course, no further implications for the problem that concerns us here. Sentences like those in (19) are somewhat more problematic. I take both of these to contain a satellite rather than a second argument, but especially the characterization of the so-called cognate object in (19a) is tentative. What matters most in this context is that there are no Goals in (19). (19) a. /о/гпргос Exp died the death of a saint^^n b. JohnA,g ran two mileSjyistance Next we consider Experiences as I defined them in Goossens (1990a: 4.3 and 4.4), i. e. as SoAs which have an Experiencer as their first argument. This makes Experiences incompatible with a Goal as their second argument, given the fact that Goals require a first argument that is Agent, Positioner or Force. Again these second arguments may be taken to be References. If we want to emphasize the specificity of Experiences, we can give them a label of their own, for example Phen(omenon), which is the name for the second argument of Mental Processes in Systemic Functional Grammar. I list Ref and Phen side by side here, without trying to settle the issue. Note in passing that in (20c) Mary is a metonym for what Mary said. (20) a. b. c. d.

JohriE^p saw Mary^^f/ph^ /о/гиЕхр likes Mary^^f/Phen JohtiE^p believed Mary^^f/p^^en /оАИехр knows Mai^Ref/Phen

A final group of instances contains linguistic action verbs. These do not, of course, form a monolithic group, and they certainly deserve a lot more attention than they have received so far in FG. What I want to do here is no more than illustrate that assigning the semantic function Goal to the second argument of two- or three-place linguistic action verbs is not obvious, and that a careful exploration of the conceptual field of linguistic action within the framework of FG would be of considerable interest (for a recent contribution in a universalistic perspective, see M. De Roeck, this volume). In (21) we have listed three patterns for the verb say. Only (21 a) can be passivized as such, (2lb) requires preparatory it, and even in that form is somewhat exceptional, and (21c) is impossible in the passive. For a more

74

Louis Goossens

refined and corpus-based discussion of passives with say, cf Goossens (1982: 108-110). (21) a. John said the same thing. (The same thing was said by John) b. John said that it was raining. (?It was said (by John) that it was raining) c. John said "It's raining". (*It was said (by John) "It's raining") From our point of view, therefore, it would be problematic to adopt the predicate frame in (22a), even if the first argument of say can be taken to be agentively involved, and if the second argument is in some sense effected. There is also, for that matter, an argument against treating the optional third argument as a Recipient, since it does not behave like other Recipients with respect to Object and Subject assignment (it only appears as a io-phrase). Given the specificity, as well as the vastness of the field of linguistic action, I would at this point tentatively propose separate semantic functions like Sp(eaker), Mess(age) and Addr(essee), as in (22b). (22) a. sayv (xl)Ag (x2)go (хЗ)кес (where x3 is optional) b. sayv (xl)sp (x2)Mess (x3)Addr (where x3 is optional) Another illustration can be provided with contextualizations for the verb tell (23). Again, there is no attempt to be exhaustive (for a full-scale investigation I refer the reader to Putseys 1982), but the instances show that it is the Addressee, rather than the Message, that becomes subject in passive sentences with tell. Only when it is used in its narrate-ssme does it permit the Message to be subjectivalized. (23) a. John told Mary that he was ready. (Mary was told (by John) that he was ready) (*It was told Mary that John was ready) (?It was told to Mary that John was ready) b. John told Mary a story. (The story was told to Mary) (Mary was told the same story) Also here the predicate frame that would perhaps first come to mind, viz. (24a), will have to be rejected. My preference, therefore, would go to (24b).

Transitivity and (non)prototypicality

75

But, as I have already indicated, this is open to ftirther investigation. Not only as far as tell is concerned (for which we will have to distinguish between its different senses: (24b) only represents the inform-sensé), but also because the lexicalization of linguistic action will have to be explored more globally. (24) a. tell, (xl)Ag (X2)go (хЗ)кес b. tellv (xl)sp (x2)Addr (X3)m ess To conclude this subsection, there are a considerable number of second arguments that do not come under our extended prototype. They have, therefore, to be specified with other semantic functions than Goal, as indicated.

4. Nonprototypical transitivity in FG 4,1. Affected as an additional (conceptual) construal Having adopted the position that (in English at least) the transitive prototype falls together with the presence of a Goal as defined in (16), the next question is how we have to account for nonprototypical instances of transitivity. The great majority of nonprototypical passives, we think, can be captured by adding an extra specification Affected to a second argument with another semantic functiori than Goal, or to a third argument or to a level-1 satellite, which through this addition becomes more central in the state of affairs, and which I therefore take to acquire argument status. Note that, although the addition is to the semantic fiinction of a single argument or level-1 satellite, the idea is that it affects the whole (nuclear or core) predication. Accordingly, the conceptual condition for this added construal relates to the SoA as a whole. It must be possible to view the entity to whose semantic function the specification is added as somehow (i. e. either literally or metaphorically) affected in the state of affairs expressed by the nuclear or the core predication. In line with the cognitive definition of transitivity in (2), construing an argument or satellite as Affected implies that the state of affairs is (often metaphorically) conceived as involving a transfer of energy from the first argument to some other asymmetrical entity. This is the case in instances like (25), (26) and (27). In the simplified FG representations we restrict ourselves to the core predication and to the specification of the semantic functions. The addition of the "feature" Affected makes an argument or satellite eligible for Subject marking (sometimes also for Object assignment, but we do not spell out the details there). Note also that we do not consider the question to what extent we have instances of predicate formation here. For the time being I

76

Louis Goossens

think that they do not involve predicate formation, but, obviously, there is some reinterpretation going on in the marked instances. I leave the matter for further investigation. (25)

Mary was given this necklace by Bill. givcv (xl:Bill(xl))Ag (х2:песк1асе(х2))оо (хЗ:Магу(хЗ))кес AÍT

(26)

Maty was told by John that he was ready. telW (xl:John(xl))sp (x2:Mary(x2))Addr Aff (x3:X(x3))Mess

(27)

The terrace was written on by John. writCv (xl:John(xl))Ag (x2:terrace(x2))LOC AÍT

This additional construal can be adopted for most of the nonprototypical passives discussed in section 2. We restrict ourselves to a few examples, again with considerable simplification. (28)

The footbridge was trampled on by the kindergartners. tramplCv (xl:kindergartners(xl))Ag (x2:footbridge(x2))Loc AÍT

(29)

Their behaviour was frowned on by the neighbours. FROWNV (xl:neighbours(xl))po (x2:behaviour(x2))Loc(M) АЩМ) (The addition of (M) underscores the metaphorical character of the specification)

(30)

John was rushed to by Mary. rushv (xl:Mary(xl))Ag (x2:John(x2))Dir A«·

(31)

The lake is contained by the dam. containv (xl:dam(xl))o (x2:lake(x2))Ref АЩМ) (Somewhat redundantly from our point of view, we might have given the first argument the specification О Fo(M))

(32)

The man had been seen by a neighbour. (xl:neighbour(xl))Exp (х2:тап(х2))КЕ№ЬЕП

SEEV

(33)

АЩМ)

Line В is intersected by line A. intersectv (xl:line A(xl))o (x2:line B(x2))Ref ащм) (The remark made for (31) also applies here)

4.2. Effected as an additional construal? So far we have adopted the position that the presence of a Goal as defined in FG provides us with a prototype for transitivity in English and that we can interpret most other SoAs which are assimilated to that prototype as involving an additional transitive construal, which can be captured adequately by ad-

Transitivity and (non)prototypicality

77

ding the feature Affected to a nonfirst argument or level-1 satellite. Since this accounts for most, but not for all nonprototypical passives, the next question is whether a similar procedure involving the addition of a feature Effected can be used for some, if not all, of the remaining cases. The condition is, of course, that this should mirror the conceptual construal that can be posited. As far as I can see there are only two contexts for which this line of thinking may be adopted. The first of these is that of linguistic action verbs, where we find that a nonfirst-argument Message, especially if it is presented in a sufficiently individuated way as in (21a), can become a passive subject. One might postulate here that assimilation to the (extended) prototype is possible because the feature Effected, which is as it were inherently relevant for Messages, is actualized. This may be captured in an FG representation like that in (34). (34)

(=21a) The same thing was said by John. Xsayv (xl:John(xl))sp (x2:same thing(x2))Mess Efr

The other case is represented by (35) (= 15b). A distance covered by running can be conceptualized as Effected. Note again, that individuation promotes transitive construal. (35)

The five miles was run by the entire team. шПу (χ 1: entire team(xl))Ag (d 1x2: five miles(x2))Distance Eff

Summing up, we find use for the procedure where assimilation to the transitive prototype involves the addition of a feature Effected, but only to a restricted extent. This is not really surprising, because the inclusion of Effected Goals into our transitive prototype is already one step removed irom the prototype as identified in Cognitive Grammar. 4.3 Nontransitive passives There still remain a number of passives which cannot be assimilated to the transitive prototype, and which for that reason I would like to label nontransitive. This means that, although a transitive prototype with its cognitive extensions as a rule underlies passivization in English, there are a few cases where this is not so. (36) and (37) exemplify this. (36)

John is believed to be dishonest.

(37) a. John is said to have lefi for Copenhagen. b. This flower is said to be a ranunculus.

78

Louis Goossens

What we get here is Subject assignment to an argument of an embedded predication, commonly referred to as raising-to-subject. A full-scale discussion is outside our scope, but we can refer the reader to Dik (1989: 11.5) and to Goossens (1990b) for some further suggestions. What is important here, is that there is no idea of an energized impact of the first argument of the matrix predicate on the "raised" argument, or any possibility to inteφret the raised argument as Effected. It is the discourse function of the passive that is sufficient to motivate it here, whereas as a rule transitivity, or assimilation to the transitive prototype, is a precondition. Obviously, passives of this sort provide us with a maximal deviation from the prototypical passive.

5. Conclusions and further perspective In this paper we have tried to integrate the notion transitivity as conceived in Cognitive Grammar into the underlying structures of FG. Basically, this amounts to setting up a transitive prototype, which we equated with the presence in a state of affairs of a second argument whose semantic function is Goal. In accordance with Dik (1989), Goal includes not only Affected, but also Effected, entities and correlates with the presence of a first argument which is Agent, Positioner or Force. This broadens the prototype as conceived in Cognitive Grammar, but appears to be adaquate for English. A consequence of this position is that Goal should be avoided for second arguments which do not live up to this broadened prototype; a number of (tentative) proposals were therefore put forward to arrive at a more consistent use of Goal in FG. The next step was to provide a procedure that does justice to the Cognitive Grammar insight that transitivity is a graded category, and that nonprototypical cases can be construed as assimilating to the prototype. Our proposal is to add a feature Affected, more rarely a feature Effected, to a nonfirst argument other than Goal. It was found that most nonprototypical passives can be accounted for in this way, with the exception of expressions involving raising-to-suject. For these, transitivity, or assimilation to the transitive prototype, does not appear to be a precondition for passivization. In these cases the discourse fiinction of passives, which normally builds on a transitive construal, completely overrides the transitivity condition. Obviously, this both testifies to the importance of the discourse/perspectivization function of the passive in English and to the nonprototypicality of the passive in instances of that kind.

Transitivity and (non)prototypicality

79

With respect to the theory of FG, I hope that I have been able to demonstrate that it is both possible and necessary to view linguistic categorizations as graded and as involving prototype effects. Also that further work is still ahead in the exploration of the typology of states of affairs, where especially states of affairs involving linguistic action were singled out as an understudied area. More technically, the relation between additional construals with the features Affected or Effected and predicate formation was left unexplored; again, this requires a broader discussion context. The only thing I want to emphasize here, probably a little redundantly, is that also with respect to predicate formation we have to think in terms of central and peripheral instances; in other words, not only the categories of natural languages, but also those of linguistic frameworks have to be thought of in terms of prototypical cases, nonprototypical instances and prototype effects. Which is only obvious, considering that linguists' categories have to be mapped onto the categories of natural language.

Notes 1 I have profited considerably from comments on two oral presentations of earlier versions of this paper, one at the "Werkgemeenschap FG" at the University of Amsterdam, the other at the FG Conference in Copenhagen. Among the discussants, I would like to thank in particular Simon Dik and Peter Kahrel.

References Dik, Simon C. 1978 Functional Grammar. Amsterdam: North-Holland. Dik, Simon C. 1989 The theory of Functional Grammar. Part I: The structure of the clause. (Functional Grammar Series 9.) Dordrecht - Providence: Foris. Dirven, René - Louis Goossens - Yvan Putseys - Emma Vorlat (eds.) 1982 The scene of linguistic action and its perspectivization by SPEAK, TALK, SAY and TELL. (Pragmatics & Beyond III: 6.) Amsterdam - Philadelphia: Benjamins. Goossens, Louis 1982 "SAY: focus on the message", in: Dirven et al., 85-131. 1990a "Mental processes and relational verbs and the typology of states of affairs in FG", in: Mike Hannay - Elseline Vester (eds.), 167-186. 1990b "FG reflections on tobacco is said to be harmful", in: Sylviane Granger (ed.), 61-69. Granger, Sylviane (ed.) 1990 Perspectives on the English lexicon. A tribute to Jacques Van Roey. (Cahiers de l'Institut de Linguistique de Louvain 1 6 . 3 ^ . ) Louvain-la-Neuve.

80

Louis Goossens

Hannay, Mike - Elseline Vester (eds.) 1990 Working with Functional Grammar: Descriptive and computational applications. (Functional Grammar Series 13.) Dordrecht — Providence: Foris. Langacker, Ronald W. 1991 Foundations of Cognitive Grammar. Vol. 2: Descriptive application. Stanford: Stanford University Press. Putseys, Yvan 1982 "Aspects of the linguistic action scene with TELL", in: Dirven et al., 133-163. Rice, Sally Α. 1987 Towards a cognitive model of transitivity. [Unpublished Ph.D. dissertation, University of California at San Diego.] Roeck, Marijke De this volume "A functional typology of speech reports". Taylor, John 1989 Linguistic categorization. Prototypes in linguistic theory. Oxford: Clarendon Press.

A valence based theory of grammatical relations Michael Herslund and Finn Serensen 1. Grammatical Relations in linguistic theories All linguistic theories seem to make use of some notion of Grammatical Relations (GRs) such as Subject and Object. And yet, such notions are one of the least clarified areas of contemporary linguistic theory. In GenerativeTransformational Grammar, GRs have no theoretical status, insofar as they are derived from structural configurations, but the importance of GRs for our thinking about language is eloquently illustrated by their reappearance as cover terms in e. g. "Subject to Object Raising", "The Specified Subject Constraint", etc. In contradistinction to generative grammar. Relational Grammar endows GRs with an axiomatic theoretical status: they are undefined primitives of the theory. In different brands of typologically oriented or universalist approaches (as, for instance, Givón 1984), GRs are simply there and it is assumed that they can be identified, or even defined, by different sets of coding and behavioural properties (as, for instance, Keenan 1976).

2. Grammatical Relations in FG. The status of GRs in FG is remarkable in at least two respects. First, GRs, of which only two, viz. Subject (S) and Object (O) are recognised, do not constitute a distinct syntactic level, insofar as they are assigned to and interspersed among semantically characterised terms, among which they are said to define a certain "perspective" upon the state of affairs denoted by the predication. In a clause like (1), S and О are assigned, but the other terms receive no GRs and remain as purely semantically marked clause members: (1)

Alberts gave Victoriao a kiss on her cheek.

The S and О functions are so to speak superposed upon the predication already constituted by the predicate and its arguments (and possible satellites), and they are said to impose a certain perspective, or point of view, on the predication. The second point is the notional characterisation of S and O. The following quotation is quite characteristic in this respect: "Subject is assigned to

82

Michael Herslund-Finn Sürensen

the argument that refers to the most saUent participant in the state of affairs described, whereas the constituent with object function gives the secondary vantage point" (Hoekstra 1981: 9). In general, S and О are seen as "pointers" (Dik 1989: 213), as defining the "perspective" from which the content of the predication is seen (cf e. g. Dik 1978: 71). Dik's presentation is thouroughly metaphoric: "point of view", "perspective", "pictures" (Dik 1978: 71), "pointer" (Dik 1989: 213), S and О are said to "trace a(n unambiguous) path through the predication" (Dik 1978: 73, 1980: 14), and so on. What this amounts to is a not very clear picture of the functions fulfilled by S and O, especially when it is repeatedly emphasised that this kind of perspective should not be confused with pragmatic fianctions such as 'topic' (e. g. Dik 1978: 87, 92). Yet, the explicitation of the functional content of S in Dik (1989: 214) is, with one possible exception, of a clearly pragmatic nature. To state it briefly: the main function of GRs in FG is the description of voice. S and О are, as mentioned above, said to constitute a path through the predication, the axis, so to speak, around which the active-passive distinction revolves. But we still do not really leave the realm of metaphor, because it remains unclear what a vantage point is, primary for S, secondary for O, and what kind of path these GRs trace. We agree, of course, with FG that S and О are essential for voice distinctions, but we also think that FG has not fully done justice to the GRs. The FG assignment of syntactic functions is always illustrated by ditransitive verbs such as give. Both assignments are said to introduce some kind of perspective: S assignment accounts for voice alternations, О assignment accounts for dative-shift types of alternations. The important thing to retain here is that apparently the same notional feature, i. e. "perspective", is used to characterise both functions. But from the fact that syntactic functions are only assigned when necessary, i. e. in order to account for some voice-like alternation, follow a number of consequences. The first consequence is that in ordinary transitive clauses, there is no need to assign the О function once the S function has been assigned (the other logical possibility, viz. О assignment but no S assignment is ruled out by fiat in Dik 1989: 210). The second consequence is that in intransitive clauses no syntactic functions are assigned. How FG then accounts for the common coding and behavioural properties of all subjects and of all objects in a given language, to say nothing of the quite widespread common properties of О and intransitive S to which we return below, is unclear. It is however obvious that one could not solve this problem in terms of semantic roles. It could probably be done in terms of the notions of first and second argument, introduced in Dik (1989: 232), but this solution

Valence and grammatical relations

83

really amounts to saying S and О in different words. The third consequence is of course that some languages simply do not have syntactic fiinctions: Serbo-Croatian is presented as such a language in Dik and Gvozdanovic (1981). Such languages thereby seem to regularise the somewhat bizarre situation of most European languages, which have some syntactic functions, but apparently not in all clause types, and where they exist they are interspersed among purely semantically characterised constituents. But then, such languages are also "perspectiveless" and it is still unclear to us what functions S and О fulfill in the English clause which are lacking in the Serbo-Croatian clause. It is still the case that the S in such a language must be in the nominative case, regardless of the semantic role of the argument in question, the О in the accusative, and so on.

3. An alternative view on GRs In this section we would like to suggest that the FG view on GRs, as sketched above, is inadequate and that natural languages have a kind of syntactic structuring which FG is unable to capture. According to the fundamental view of (different) valence theories, the verb is the organisational centre of the clause and dictates accordingly which other constituents can (or must) be present in the clause. It follows from this view that a valence theory not only should yield a classification of verbs but also predict a typology of clause types as a function of the established verb classes. We shall present such a view below.

3.1. Subject and Object Let us take as our point of departure the notion of predication found in FG. In this approach the predication is a combination of a predicate and a number of arguments (three at most, Dik 1989: 69) to which the predicate assigns semantic functions such as Agent, Goal, etc. So far, we have no quarrel with FG. But the crucial point, we maintain, before this kind of predication can qualify as a linguistic expression of a natural language, is the assignment of GRs, which are far more important than FG would have us believe, insofar as they impose a structure upon the "flat" predication, predicate plus semantically characterised arguments. In any predication there is one argument which is the fundamental argument in the sense that one cannot understand or even conceive the state of affairs denoted by the predication without this argument. This fundamental

84

Michael Herslund-Finn Serensen

argument is most easily illustrated by the class of symmetrical verbs such as English break, as in (2): (2) a. b.

The cigar broke. Ernest broke the cigar

The fundamental argument is thus the О of a transitive verb or the subject of an intransitive verb (Si). Let us say that this fundamental argument together with the predicate forms a syntactic predicate: it is the predicate constituting argument. In our view, a predication, in order to qualify as a linguistic expression, must contain a subject, which is the pivot of the predication, the argument on which hinges the assignment of a truth value to the proposition. We believe that this feature is crucial for the predications of natural languages. These are not just descriptions of states of affairs, they are also, and above all, propositions about states of affairs. And according to a certain philosophical tradition, the fundamental property of a predication is the creation of a truth or falsity yielding combination, a proposition, cf Strawson (1974: 21), Searle (1969: 124). In our proposal, then, one argument is chosen as subject, receives a special treatment, as it were. This choice is not identical to the FG choice of perspective, but, and we think this is important, it is not incompatible with FG's assignment of the S function. We can then characterise S as the predication constituting argument {the cigar of (2a), Ernest of (2b)). We can consequently represent the GRs assigned so far as in (3): (3)

predicate formation predication formation

0

Si

s.

+ —

-I-I-

-I-

We feel that FG lacks a level of analysis where the predication is validated with respect to one special argument which is crucial for the assignment of a truth value. We would accordingly like to make a case for the view that such a syntactic level is necessary in order for a linguistic theory to be sufficiently general. Our theory is based, as the predications of FG, upon a hypothesis about the relations obtaining between a verbal element and its arguments, its valence. Compared to FG, the level we have in mind resembles most closely the level where classes of arguments (Al, A2 and A3, cf. Dik (1989: 232)) are introduced, but our theory attempts to identify and characterise the relations between the arguments in the different slots of such a simple clause

Valence and grammatical relations

85

structure. We find it most expedient to consider subject, object and adject (see 3.2 below) as fiindamental and in fact universal relations of clause structure. The relations we postulate between the arguments, and their different functions in the building of the clause, are of a general nature and not dependent upon their form. The arguments are above all syntactic entities and not characterised only by their semantic functions (case or role). The possibility of changing the perspective, although important, should not be the crucial feature of subject and object assignments, as it is the case in FG. The role of the object (and intransitive subject) relation is crucial in our proposal: together with the verb this relation constitutes a complex predicate which is then predicated of the subject. From the preliminary characterisation of the two basic GRs, as in (3), follow a number of consequences: in transitive clauses, О is closer to or more integrated into the verb than St; the fundamental, predicate constituting arguments, О and Sj, should have a lot in common; and the Si, which is both a predicate and a predication constituting argument, has a kind of double function in the clause. Let us briefly explore and comment upon these three consequences (or predictions). As for the first point, in what sense can the О of a transitive clause be said to be closer to, or more integrated into the verb than the St? That it is indeed the case is easily seen in examples like those of (4): (4) a. b. c. d. e.

Peter stillede potteplanterne pâ bordet. 'Peter put the flowerpots upon the table.' Peter stillede et dumt sporgsmál. 'Peter asked a silly question.' Peter stillede en forkert diagnose. 'Peter made a wrong diagnosis.' Peter stillede alle sine ure. 'Peter set all his clocks.' Peter stillede trœskoene. 'Peter kicked the bucket.'

It is the О which, so to speak, cuts into the polysemy of the verb, here exempliñed by the Danish verb stille 'put', and determines the reading, which varies as a function of the lexical material of the О phrase (cf Marantz 1984). On the other hand, one can vary the lexical material of the St ad infinitum without obtaining the slightest variation in the interpretation of the verb. This integration of the О into the verb is also revealed by the facts that verbs in many languages dictate the case of the О phrase, which may vary as a function of the positive/negative distinction in the predicate, and/or the

86

Michael Herslund—Finn S0rensen

definiteness of the О phrase ((5) is from Comrie 1981:65, (6) from Groenke 1972: 72): (5) a.

b.

(6) a.

b.

Russian: Masa kupila sapkuAcc. MashaNom buyp ast hatAcc 'Masha bought a cap.' Masa ne kupila sapkiQ^nMashaNom not buyp^s, hatoen 'Masha did not buy a cap.' Finnish: Hän luki kirjarip^cc· HCNom readpast Ьоокдсс 'He read the book.' Hän luki kirjaapanHeNom readpast bookp^« 'He was reading a book.'

That such alternations should take place precisely in the О should by now come as no suφrise. What is at stake here is, inter alia, the distinction between a holistic and a partial reading, as this distinction has been described repeatedly and extensively for the English smear paint class of verbs. It seems indeed natural that this distinction should be located in the predicate forming argument, the O, and that, where alternations of the smear paint type occur, the holistic reading should be associated with the О function, the predicate constituting argument. Somewhat similar facts can be found in caseless languages like Danish. Nouns have no case inflection, but an undetermined О phrase attracts the main stress (cf Rischel 1980, Thomsen 1990); accordingly the verb loses its stress and is pronounced in the present tense without the "st0d" (glottal catch) which otherwise characterises a large class of verbs. Compare (7): (7) a.

b.

c.

Hun She 'She Hun She 'She Hun She 'She

láser readpres is reading.' láser begerne. readpres books-the is reading the books.' lœser boger readpres books reads books.'

Valence and grammatical relations

87

This phonological integration of the verb and its О is accompanied by a kind of semantic semi-incorporation (cf the related notion of incorporation as predicate formation in FG, Dik 1980: 39ss.), which again underlines the status of О as the fundamental, predicate forming argument. As for the second point, common properties of О and Sj, let us notice first of all that these are the two GRs involved in most voice and voice-like alternations, like active vs. passive and middle, or lexical alternations like (2) above. We know of no clear cases of an О alternating with an St. This constant relation we can simply formulate as (8): (8)

Si —

О

which subsumes the regular voice or voice-like alternations between О and Sj. This deep identity, or functional equivalence, between the two fundamental GRs is probably also what is found in ergative languages, where the same case marks the intransitive subject and (what corresponds to) the object of a transitive clause (in European type languages), as in (9) from Greenlandic: (9) a.

b.

QimmeqAhs aterpoq. DogAbs walk downpres 'The dogsj is walking down.' Arnap^ex qimmeqAbs takuvaa. WomanRei dogAbs seepres 'The womans; sees the dogo.'

Besides these two features, there are many indications even in the better known European type languages which point in the same direction. We shall only mention a few af them here (cf further Herslund 1988a: 12ss. and 1988b: 50ss.). Selectional restrictions requiring that a certain argument be plural affect О or Si alike, but to our knowledge never S,: (10) a. Han talte sine frimœrker. He countpret hisRefl stamps 'He counted his stamps.' b. *Han talte sit frimœrke. He countpret hisRefl stamp 'He counted his stamp.' (11) a. Myrerne myldrede rundt от syltetojet. Ants-the swarmpret round about jam-the 'The ants were swarming around the jam.' b. *Myren myldrede rundt от syltetejet. Ant-the swarmpret round about jam-the 'The ant was swarming around the jam.'

88

Michael Herslund-Finn Serensen

The Aktionsart of verbs is often modified by the context they appear in. Again О and S; are crucial is this respect, compare the following from French; the a. examples are perfective, the b. examples imperfective: (12) a. Elle She 'She b. Elle She 'She

a has has a has has

bu le cognac. drank the brandy drank the brandy.' bu du cognac. drank some brandy been drinking brandy.'

(13) a. Le vélo tombe. the bike fallpres 'The bike falls.' b. La neige tombe. The snow fallpres 'The snow is falling.' We know of no certain examples of an St modifying the basic perfective/ imperfective distinction. The last feature we shall mention which underlines the basic functional equivalence of О and S; is the fact that clitics in French can be extracted from О phrases and from certain Si phrases; the distinction between unaccusative and unergative predicates certainly intervenes here, but we cannot go into that question here. Under no circumstances, however, can a clitic be extracted from an St phrase. Compare the behaviour of en and lui in (14) and (15): (14) a. II ещ perd [le contrôle Cj]. 'Не loses control over it.' b. [Le contrôle Cj] lui ещ échappe. 'The control over it escapes him.' c. *[Le contrôle e;] 1'ещ ennuie. 'The control over it bores him.' (15) a. II luii pince [le nez e;]. 'He pinches her nose.' b. [La tête Ci] luii tourne. 'Her head turns.' (i. е. 'She is dizzy') c. *[Les yeux Ci] ne 1ш\ voient rien. 'Her eyes see nothing.'

Valence and grammatical relations

89

One could certainly find many other properties in different languages which point in the same direction: the deep-rooted functional equivalence between О and S¡, an equivalence which will prove crucial for the discussion of the next section. As for the third point, it is inherent in this description that the Si is a kind of double function, cf (3): it shares some (most?) properties with O, viz. the semantic integration with the verb, and another property with St, viz. that of predication formation. In this perspective, it seems quite natural that the basic formula (8) should be so often implemented by a reflexive construction which precisely expresses the fundamental argument twice. There are indeed languages where phenomena such as those illustrated in (2) must be expressed by a reflexive construction. Russian is such a language: (16) a. Okno slomalos'. 'The window brokcRefl.' b. Ivan slomal okno. 'Ivan broke the window.' Also reflexive passives are well known from a number of languages. 3.2. The Adject Having established the two central GRs, the predicate constituting О or Sj, and the predication constituting S; or St, there is so to speak room for no more around the verb (the syntactic predicate). Only in one case can the valence of the verb specify one fiirther GR. This GR is the one which encodes a special relation, not only between the predicate and one of its arguments, but also between this argument and some other argument. The argument which contracts such a relation is always, not suφrisingly, one of the fundamental GRs, О or Si. Let us illustrate this third GR, which we shall call Adject (A), with the class of verbs represented by English send. This verb represents a major class of trivalent verbs which, at least in European languages, occur in two environments, as in the English examples (17): (17) a. Victoria sent Albert to Berlin. b. Victoria sent Berlin a message. In (17a) we have a local complement, a Dir in FG terms, in (17b) we have an indirect object, a RecObj in FG terms. Our proposal is then simply that both illustrate the A relation, which consequently can be realised in different

90

Michael Herslund—Finn S0rensen

ways. One immediate argument for this position is of course the complementary distribution of local and dative complements after many verbs in many languages: they cannot co-occur, the verb must choose between two manifestations of the A relation. So what we have is, in FG terms, a regular and constant alternation between Dir and Ree functions with a substantial class of verbs (the empirical basis for this claim is exposed in some detail in Herslund 1986 for Danish, and in Herslund and Sorensen 1985, 1987 and Herslund 1988b for French). And we can go one step fiirther: it is invariably the case in predications of this type that a special relation is established between the Go argument and the Dir/Rec argument, between the О and the A in terms of the theory we are advocating here. In the "flat" predications of FG one might equally well expect such a relation to obtain between the Ag and the Dir/Rec argument in some cases, but that is never the case. It is in fact always the case that part of the meaning of these verbs is that the Go argument is located with respect to the Dir/Rec argument. The send, put, give class of verbs is of course typical in this respect, but even more abstract predicates like show, tell or promise are not very difficult to understand in terms of a location of (the entity denoted by) the О constituent with respect to (the entity denoted by) the A constituent. And this special relation, whose semantic basis is clearly of a locative nature, is encoded as a syntactic relation of a predicative nature: the relation between Sj/O and A constitutes a secondary predication in the clause: (18) a. S V О A b. S V A This concept of a secondary predication is useful in understanding the difference between locatives and datives, as in (17). What distinguishes them is the orientation of the secondary predication. One possible paraphrase of (17a) is (19a), of (17b) (19b): (19) a. Victoria CAUSE {Albert BE in Berlin) b. Victoria CAUSE {Berlin HAVE a message) So one way of analysing dative complements as a special subset of As, is to simply say that they are inverted locatives: it is the argument denoting the place that is subject in the secondary predication (for further details, see Herslund 1988b: lOOff.). We would like to point out that this analysis also is compatible with the FG analysis of dative shift, as in (17) b., insofar as the idea of the dative A as a kind of secondary subject can lend some content to the FG idea of the О function as the secondary vantage point, whereby dative

Valence and grammatical relations

91

shift is assimilated to the voice dictinctions primarily associated with the S function. We cannot go into all the details of the question here (see Herslund 1986, 1988a, 1988b), but it is not very difficult to find arguments which support the conclusion that datives are a kind of secondary subjects. One argument in favour of this view would of course be to show that datives share some ftindamental subject properties. One example of this, the only one we shall mention in this context, is the behaviour of reflexives in (some varieties of) Danish. Reflexives are, as in most languages, controlled by subjects, as in (20), where we have the reflexive possessive article in b.: (20) a. Han solgte hendes bil. He sellpret her car 'He sold her car.' b. Han solgte sin bil. 'He sold his-Refl car.' Now in general, only subjects can control such reflexives. But examples like the following are far from uncommon: (21 )

Filmen "Intermezzo " gav Ingrid Bergmann sit folkelige gennembrud i Sverige. 'The film "Intermezzo" gave Ingrid Bergmann her-Refl popular break-through in Sweden.'

One natural explanation of data like this is of course that reflexives are always controlled by subjects, primary ("real" Ss) or secondary (dative As). Having introduced the basic idea of an A, we would like to present some other types of As. We use RA as an abréviation for the A relation. Consider the sentences in (22) - (23): (22)

Peter bor i Paris. 'Peter lives in Paris.'

(23)

Peter opbevarer ollene i koleskabet. 'Peter keeps the beers in the refrigerator.'

In (22)—(23) it is the prepositional phrases which function as A, i. e. i Paris and i koleskabet. In both cases, the postulated RA implies that Si/0 is located in relation to the A. This is supported by the facts since the S of (22) and the О of (23) clearly are situated in relation to the places denoted by the As. Notice also that (23) cannot express a state of affairs in which the subject is located in relation to the A.

92

Michael Herslund-Finn Serensen

Approximately the same pattern can be seen to exist in the following sentences: (24)

Peter er íj'g. 'Peter is ill.'

(25)

Peter finder losningen uacceptabel. 'Peter finds the solution unacceptable.'

In the case of (24) and (25), the grammatical tradition has always recognised a particular relation between Peter and syg in (24), and between losningen and uacceptabel in (25). In the present framework it is claimed that this wellknown relation is RA, and thus also that the adjective phrases in (24) and (25) are As. In short, the present framework takes the well-known relation between a subject and its complement, and between an object and its complement, to be a particular instance of a more general pattern, i. e. the pattern introduced by Кд. The identification of the relations present in (22)—(23) and in (24)-(25) makes it natural to suppose that at least some verbs allow both the locative pattern of (22)—(23) and the predicative pattern of (24)—(25). This linking power of the verb is illustrated in (26) and (27) where the verbs used in (24) and (25) are used with prepositional phrases of the locative type: (26)

Peter er i Paris. 'Peter is in Paris.'

(27)

Peter fandt piben under bordet. 'Peter found the pipe under the table.'

This variation is quite natural if it is assumed that the underlined constituents of (24) - (25) and (26) - (27) are As related to Sj/O by RA. As expected, given the present framework's general conception of RA, and thus of As, it is natural also to find verbs allowing a predicative complement or a dative complement. This is the case with gore ('make/do') as illustrated below: (28)

Peter gjorde sin far gal. 'Peter made his father furious.'

(29)

Peter gjorde mig denne tjeneste. 'Peter did me this favour.'

This variation and the fact that the Os are located respectively in relation to gal og mig strongly support the idea of a particular relational pattern. This is what the present framework seeks to capture by the RA and by the introduction of the adject function.

Valence and grammatical relations

93

So far, we have illustrated the kinds of complements that are taken to be As in the present framework. From the examples given in (17)-(29), it should be clear that this label is applied to constituents which the grammatical tradition has labelled with a series of partly unrelated terms such as: subject complement, object complement, prepositional complement, dative object (or complement), indirect object, and place adverbial. The primary claim made in the present framework is thus that the RA and the adject fiinction allow a generalisation concerning the structure of all sentences containing these apparently unrelated constituents. This generalisation does not mean that there is no distinction to be made within the class of As. More or less explicitly, we have already distinguished between three types of As: the locative type, the predicative type, and the dative type. The generalisation implies, however, that these types of As cannot cooccur in the same sentence and that a particular verb may combine with one or more types of As in different sentences, and that the RA is always realised between S/O and A. The proposed analysis also explains the otherwise curious fact that whereas traditional grammar recognises subject and object complements, no one has ever, to our knowledge, come up with such a thing as an "indirect object complement". This lacuna in the inventory of traditional grammatical functions is of course readily explained in the present framework: since indirect object and subject/object complement realise the same GR, viz. A, they cannot of course co-occur; a verb followed by both could not exist. 3.3. A Clause Typology We are now able to propose a typology of verbs and, hence, clauses for European languages. There is a basic distinction between transitive and intransitive verbs and clauses. Upon this basic distinction is supeφosed, as it were, the presence versus absence of the A relation. This relation, which, as argued above, is of a predicative nature, covers a lot of traditional GRs such as indirect object, prepositional object, some local complements (especially two classes, to and from complements), subject and object complement, and possibly others as well. The proposed verb typology emerges as the following compact schema: (30) transitive symmetrical intransitive

+ S S S S

A V V V V

О A О A A A

-A S V О S V О SV SV

94

Michael Herslund—Finn S0rensen

This classification is of course only a statement of the overall types. It has much the same status as the basic vocalic triangle (i-a-u) in phonology, insofar as it states the limits within which clauses are constructed, in much the same way that the vocalic triangle defines the limits within which vowel articulations are possible. Languages of the common European type vary within these limits as to which classes of A they allow, some languages lack e. g. the symmetrical class of verbs, viz. Russian, and so on. But one interesting fact about (30) is that all the major voice and voice-like alternations, such as passive, middle and causative, can be described as variations within the limits defined by this schema.

4. Conclusion One major concern of FG seems to be descriptive economy: the reason for not recognising other GRs than S and О is that in this way one can achieve a more economical description. But in the proposals put forward here, another kind of economy is obtained: if we propose to introduce the GR A and analyse the dative complement as a special manifestation of this relation, it is at the same time possible to dispense with the semantic function Recipient; this fiinction will correspond exactly to the subclass of Dirs which hold the S relation in the secondary predication associated with A. So we do not think that one could use economy of description as an argument against the proposed analysis. And in general, semantic fiinctions in FG are as ill-defined as the GRs of any theory. On the other hand, one positive and ecumenical conclusion is that most of our proposals seem to be compatible with the basic principles of FG. References Comrie, Bernard 1981 Language universals and linguistic typology. Oxford: Blackwell. Dik, Simon C. 1978 Functional Grammar. Dordrecht: Foris. 1980 Studies in Functional Grammar. London: Academic Press. 1989 The theory of Functional Grammar. Part I: The structure of the clause. Dordrecht; Foris. Dik, Simon C. - Jadranka Gvozdanovic 1981 "Subject and object in Functional Grammar", in: Hoekstra et al. (eds.), 21-39. Givón, Talmy 1984 Syntax. A functional-typological introduction. Amsterdam: John Benjamins.

Valence and grammatical relations

95

Groenke, Ulrich 1972 Grundzüge der Struktur des Finnischen. Hamburg: Ropp. Herslund, Michael 1986 "The double object construction in Danish", in: Hellan - Koch Christensen (eds.), Topics in Scandinavian syntax. Dordrecht: Reidei 125-147. 1988a "On valence and grammatical relations", in: Finn Serensen (ed.). Valency. Three studies on the linking power of verbs. (Copenhagen Studies in Language 11.), 3-34. 1988b Le datif en français. Louvain-Paris: Peeters. Herslund, Michael - Finn S0rensen 1985 De franske verber. En valensgrammatisk fremstilling. I. Verbernes syntaks. Copenhagen: University of Copenhagen. 1987 De franske verber 2. En valensgrammatisk fremstilling. II. Klassifikation. Copenhagen: Copenhagen Business School. Hoekstra, Teun 1981 "An outline of Functional Grammar". In: Hoekstra et al. (eds.), 3-18. Hoekstra, Teun - Harry van der Hulst — Michael Moorgat (eds.) 1981 Perspectives on Functional Grammar. Dordrecht: Foris. Keenan, Edward L. 1976 "Towards a universal definition of "Subject"", in: Charles N. Li (ed.), Subject and Topic. New York-London: Academic Press, 303-333. Marantz, Alec P. 1984 On the nature of grammatical relations. Cambridge, MA: MIT Press. Rischel, Jorgen 1980 "Phrasenakzent als Signal des Objekts ohne 'determiner' im Dänischen", in: Festschrift für Gunnar Bech. (Kopenhagener Beiträge zur Germanistischen Linguistik, Sonderband 1.), 262-279. Searle, John R. 1969 Speech acts. Cambridge: Cambridge University Press. Strawson, P. F. 1974 Subject and predicate in logic and grammar. London: Methuen. Thomsen, Ole Nedergaard 1990 "Unit accentuation as an expression device for predicate formation in Danish", in: Lisbeth Falster Jakobsen (ed.), Functional Grammar in Denmark. Copenhagen: University of Copenhagen, 66-86.

The study of ergativity in Functional Grammar Vit Bubenik 1. Introduction In this paper, in section 1, I propose to survey briefly several theories of the origin of ergative organization in Modern Indo-Aryan (Modern Indo-Aryan) languages; in section 2 the position of Functional Grammar in this respect will be examined in some detail and certain questionable assumptions pinpointed; in section 3 Middle Indo-Aryan evidence for the existence of the passive construction and its coexistence with the incipient ergative construction will be presented; and finally, in section 4, the contribution of the study of Indo-Aryan ergativity to the theory of Functional Grammar will be outlined. The Indo-Aryan data deserve special attention because this group of languages supplies us with the only historically documented example of the full development from passive to ergative. To facilitate our investigation we may wish to examine the nature of ergativity in Modern Hindi-Urdu in terms of its morphology (case marking and verbal agreement). In Hindi-Urdu, the transitive verb in the perfective aspect requires the agent to be marked with a special agentive postposition ne. Contrast the following nonperfective and perfective sentences with intransitive and transitive verbs (in what follows + indicates the т о ф Ь е т е boundary, and = indicates the boundary between the clitic and its host): (1)

larkä cala (Hindi-Urdu) boy go+pp+masc 'The boy went.'

(2)

larkT call girl go+pp-bfem 'The girl went.'

(3)

larkä кит dekhtâ hai boy dog seeing be+3/Sg 'The boy sees a dog.'

(4)

larkä kitäb dekhtä hai boy book seeing be+3/Sg 'The boy sees a book.'

98

Vit Bubenik

(5)

larke=ne kuttâ dekhâ hai boy=erg dog see+PP+mase be+3/Sg 'The boy has seen a dog.'

(6)

larke=ne kitäb dekhâ hai boy=erg book see+pp+fem be+3/Sg 'The boy has seen a book.'

We may observe that intransitive subjects in (1) and (2), and transitive objects in (3) through (6) are not marked with any postposition. On the other hand, transitive subjects in (5) and (6) are marked with the ergative postposition ne. Contrasting (3) and (4) with (5) and (6) we may conclude that the determining factor is the aspect of the predicate: i. е., the transitive subjects are marked with the ergative postposition only if the verb is perfective; otherwise, as in (3) and (4) with the imperfective verb, the subjects are unmarked. We may further observe that in (1) and (2) the verb agrees with the intransitive subject, in (3) and (4) with the transitive subject; but in (5) and (6) the verb agrees with the transitive object. However, if the transitive object is marked by the defmitizing postposition ко the objective agreement is blocked and the verb appears in the unmarked form; contrast (7) and (8) with (5) and (6); the postposition ко is glossed DAT/ACC because it can also mark indirect objects: (7)

larke=ne kutte=ko dekhâ hai boy=erg dog=dat/acc see+pp+m/n be+3/Sg 'The boy has seen the dog.'

(8)

larke=ne larkî=ko dekhâ hai boy=ERG girl=dat/acc see+pp+m/n be+3/Sg 'The boy has seen the girl.'

Both (7) and (8) show the neutral agreement; this actually can be seen only if the object is feminine, as in (6) vs. (8), since in Hindi-Urdu the unmarked form of the verb is the masculine singular form. Marathi, which preserved the neuter gender, will show a different neuter form in all instances when the transitive object is definite (marked with the postposition /a); contrast (9) and (10): (9)

tyâ=ne vahâ pahilî (Marathi) he=erg exercise book see+pp+fem 'He saw an exercise book.'

The study of ergativity (10)

99

tyä=ne muT=la pahilT he=erg girl=dat/acc see+pp+neut 'He saw the girl.'

On the basis of data such as the above, Hindi-Urdu, Marathi and other Western Indo-Aryan languages have been described as being of ergative typology. At this point, I would like to emphasize that Western Indo-Aryan languages belong to those ergative languages which possess a fully developed system of the passive voice. Examples (11) and (12) are passive counterparts to (3) and (7) with passive indicated by the participle forms of the verb to 'go': (11)

kuttä larke=se dekhâ jata hai (passive counterpart to 3) dog boy=abl seen pass Ье-ЬЗ/Sg Ά dog is seen by the boy.'

(12)

kutte^ko larke=se dekha gayä hai (passive соип1ефай to 7) dog=dat/acc boy=abl seen pass be+3/Sg 'The dog has been seen by the boy.'

The verb in the passive sentence (in 11) has to agree with the unmarked NP, i. е., the agreement is with the Goal of the action which has been assigned Subj function. In (12), however, the neutral agreement of the ergative counterpart has been retained. There are essentially two theories about the origin of the ergative construction in Modem Indo-Aryan languages. The first one assumes that the way in which ergative developed in the Indo-Aryan family, in perfective aspect, was through the loss of the inflectional perfect, and replacement by the periphrastic construction based on a participle that was passive in form - in other words, the passive construction was recategorized as an ergative one (Anderson 1977; Comrie 1978). The other theory assumes that the passive construction of Old Indo-Aryan (Old Indo-Aryan) was already ergative (Klaiman 1978; Wallace 1982). Thus according to Wallace (1982: 148) "in Sanskrit, the to-participle is ergative" or it has "an ergative force". A compromise stance is taken by Hock (1986) who concludes that the modem ergative constructions reflect both older ergatives and older passives (i. е., that neither the passive origin hypothesis nor the ergative hypothesis are fully adequate).

100

Vit Bubenik

2. The Functional Theory of the Origins of Ergativity The position of Functional Grammar is that ergative systems can arise through markedness shift operating on the active-passive opposition of a nominative language (cf Dik 1980: 113-126 and 1989: 242-246). Markedness shift was defined by Dik (1978: 111) as a historical process in which a certain expression type which is marked at an earlier stage in the development of a language may become unmarked at a later stage (see Chart 1). Chart 1. Markedness shift (Dik 1980: 115) obsolete Stage 1 Stage 2 Stage 3

At Stage 1 there is an opposition between an unmarked form Ej and a marked form E2. At Stage 2 the marked form E2 has become unmarked, and has ousted Ej from being used. At Stage 3 a newly created marked form E3 has been introduced, so that the original opposition is restored. The operation of markedness shift through the active-passive opposition according to Dik (1989: 243-244) is summarized in Chart 2. In the chart, [ag] in square brackets under passive indicates the expression of the Agent phrase. If markedness shift applies to the accusative language with marked passive construction in Type I, this construction by being used more frequently becomes the dominant expression for transitive predications, and ends up by putting the original unmarked active construction out of work. Thus we arrive at Type II: Accusative language with unmarked passive. In this type the opposition between active and passive has disappeared which may lead to further changes: the passive moφhology is either lost or reinterpreted as a marker of transitivity, and the unmarked passive construction is reinteφreted as an active one, with a concomitant shift of Subj and Obj function. This will result in Type III: ergative language without passive. Here the case marking of the agent of the transitive predicate is ergative, but the syntactic properties of the agent can be described in terms of Subj which also covers the absolutive term of the intransitive predicate. Dik (1989: 244) observes that in Type III languages the case marking correlates neither with the semantic, nor the syntactic ftinctions, and expects (i) a further adaptation of case marking to correspond with Subj/Obj articulation and (ii) the intro-

The study of ergativity

101

duction of some new kind of passive construction. The conjunction of these two processes will bring us back full circle to Type I. Chart!. The rise of an ergative construction through a reinterpretation of an earlier passive construction (according to Dik 1989: 2 4 3 - 2 4 4 ) . Type I. Accusative language with marked passive

intr. trans.

Active [unmarked]

Passive [marked]

V AgSubj [nom] V AgSubj [nom] GoObj [acc]

Vpass GoSubj [nom] Ag [ag]

Type II. Accusative language with unmarked passive Active [unmarked] intr.

Passive [unmarked]

V AgSubj [nom]

trans.

Vpass GoSubj [nom] Ag [ag]

Type III. Ergative without passive

intr. trans.

Active V AgSubj [abs] V[tr] AgSubj [erg] GoObj [abs]

Passive

Fleshing the above scenario with Old, Middle and Modem Indo-Aryan data - presented in Chart 3 - it would indeed appear that the rise of ergative organization in Indo-Aryan languages has to do with markedness shift with respect to Subj assignment. However, several qualifications are necessary if we want to adopt the scenario in Chart 2 to Indo-Aryan data. (i)

(ii) (iii)

(iv)

The intransitive verbs were passivizable through the whole history of Indo-Aryan (i. е., there should be no empty slot for intransitive passive in Chart 2). There were two types of the active and the passive "perfective" moφhology in Old Indo-Aryan: the aorist and the perfect. Even more significantly, there were always two (even three) types of the passive voice available to the transitive verbs during the Middle Indo-Aryan period. In other words. Type III. Ergative without passive is illusory in the history of Indo-Aryan. This point will be taken in the next section. Given (iii) we cannot credit markedness shift with the introduction of a new passive construction; these two are not in a causal nexus. Put differently, we have to reconsider the relationship of the passive and the ergative constructions.

102

Vit Bubenik

Chart 3. Markedness shift in the history of Indo-Aryan languages

(obsolete) OIA

MIA (early) MIA (late)

Active (unmarked) akârsat (Aorist) ,cakära (Perfect) 'he (has) made'

Passive (marked) (tena) akâri (Aorist) (tena) krtám (Perfect) -'it was/has been made (by him)'

akasi tena kata" PROTO-ERGATIVE 'he (has) made it'

(teha) kata 'BE'-PASSIVE /it was made (by him)' (tena) kata gata 'GO'-PASSIVE 'it was made (by him)^

NIA

us=ne kiyâ ERGATIVE 'he (has) made it'

(us=se) kiyä gayâ-· 'GO'-PASSIVE 'it was made (by him)'

3. Typological Changes in Middle Indo-Aryan The systemic potential of Old Indo-Aryan included active and mediopassive forms in both aspects for the expression of past events (called traditionally Imperfect, Perfect, Pluperfect and Aorist). In addition there was the verbal adjective which was also used in passive constructions. As shown in Chart 3, in Old Indo-Aryan the past perfective event such as "he has made it" could be expressed in three ways: (i)

active construction using the asigmatic or sigmatic aorist: ákarat or 'he (has) made' or the perfect: сакйга 'he (has) made'. finite passive construction using the aorist: àkâri 'it was/has been made' or the perfect: cakre 'it was/has been made' (the latter form is not shown in Chart 3). nonfinite passive construction (using the verbal adjective in -to)·, tena krtám 'it was/has been made by him'. áicrsat

(ii)

(iii)

In the last mentioned construction we have the right moφhological characteristics for an ergative pattern; however, in Vedic Sanskrit the construction is a derived passive one, rather than an ergative one. It was only with the disappearance of the finite passive construction (äkäri) and the atrophy of the active personal forms {cakâra and ákürsat) that this derivational relationship became opaque.

The study of ergativity

103

The gradual disappearance of the active finite forms for the expression of past (perfective) events was accompanied by an incorporation into the verbal system of the above mentioned nominalized deverbal form with stative force (the ία-participle). After the loss of the sigmatic aorist in the late Middle Indo-Aryan the nonfmite passive construction tena kata 'it was made by him' became the only means of expressing the past actions. It is obvious that this construction continues both the finite and nonfmite passives of Old IndoAryan, i. е., tena akâri and tena krtám by-him made; however, its position in the verbal system of Old Indo-Aryan and Middle Indo-Aryan, respectively, was fundamentally different. In Middle Indo-Aryan this nonfinite passive construction filled in the gap in the system of temporal contrasts; otherwise only the present and future finite forms were available in Middle Indo-Aryan. On the other hand, in Old Indo-Aryan the system of temporal contrasts had no gaps in either voice. The speakers of Old Indo-Aryan could choose among the active imperfect, perfect and aorist for the narration of past imperfective and perfective actions. The nonfmite passive construction tena krtám (byhim made) was only an alternative way of saying 'he made' {cakära or äkärsat) in Old Indo-Aryan, whereas in Middle Indo-Aryan tena kata became the only way of saying it. In other words, with the demise of the active forms (sigmatic aorist) the motivation to treat the original passive construction tena krtám as the passive disappears; and in the absence of any active-passive contrast we are entitled to treat this construction as syntactically ambiguous between passive ( = the original meaning) 'it was (has been) made by him' and the incipient ergative ( = the new meaning) 'he (has) made it'. This complex construction for the active meaning was reanalyzed as an ergative construction during the Modem Indo-Aryan stage as a consequence of another cardinal change which took place at the end of the Middle Indo-Aryan period: starting in Apabhrarnsa (sixth century A.D.) the inherited distinction of nominative and accusative was wiped out by sound changes and the absolutive case made its appearance. The ambiguity between the protoergative tena kata and the 'be'-passive tena kata in late Middle Indo-Aryan is shown in Chart 3 by the tilde (~). To be sure, this ambiguity existed only in the 3rd person singular. In all other persons passive was the only interpretation available. The following is a late Middle Indo-Aryan example of the 'be'-passive in the first person singular: (13)

Suravariria pësiu=mhi king-Hinstr sent+masc=be+l/Sg Ί am/have been sent by the king' [12th c.. Sc. 724.4]

104

Vit Bubenik

If there is no agentive phrase, the constructions with the participle agreeing with the Goal are, of course, to be inteφreted passively; an example of a prototypical agentless passive is given in (14). (14)

taha kösalliu divvu ehu pësiu you+gen/dat gift+neut heavenly+neut this+neut sent+neut 'This divine gift was/has been sent to you' [12th c., Sc. 724.6.7]

If there is an agentive phrase we have to pay attention to the postulates of pragmatics in determining whether we are dealing with the ergative or the passive constructions. The word order of Middle Indo-Aryan inherited its freedom from Old Indo-Aryan (but at the end of the Middle Indo-Aryan period the trend towards the verb-fmal and agent-initial position statistically increases). That means that sequences with a sentence-initial agent can be interpreted ergatively whereas those with the Ag in marked post-verbal position Go PP Ag (or PP Go Ag or PP Ag Go) passively; examples for the ergative interpretation are given in (15), those for the passive interpretation in (16): (15)

AgSubj GoObj PP (ergative inteφretation) paT... manuattu pattu you+instr manness-bneut reach-l-pp-bneut 'You reached existence as man.' [12th c., KPP 56.1-2] AgSubj PP GoObj Siriena tak-khani khâdiu tâsu sîsu Srîyaka+instr that-moment+loc cut+pp+neut he+gen head+neut 'In that moment SiTyaka eut off his head.' [12th е., KPP 48.4-5]

Examples for the passive interpretation: (16)

GoSubj PP Ag (passive interpretation) chakkhada-vasumai... uvasâhiya... Sanatkumärina six-part-earth -I- fem subdue+caus+pass+fem S.+instr 'The six-part earth was subdued by Sanatkumâra.' [12th c., Sc.719.3-8] PP Ag GoSubj πιο vayâsihï Thûlabhaddu kösähi ghari led+mase friends+instr S.+mase K.+gen house+loe 'Sthülabhadra was led by [his] friends into Kösä's house.' [12th c., KPP 9.4]

The study of ergativity

105

In correlative sentences the passive inteφretation of the main clause may be preferable to avoid subject-switching ( = to preserve topic continuity to use Givón's terminology 1984: 137). This is shown in (17): (17)

aha ... ja là-ci sä muddha samullavai further when something that+fem charming+fem speak+3/Sg [12th c., Sc. 629.11-5] tava dittha ... tina ... purisina then seen+fem that+instr man+instr 'Further, while the charming [one] was speaking, she was seen by that man.'

Ergative interpretation of (17), i. е., 'while the charming [one] was speaking, that man saw her' would not preserve topic continuity; to opt for it, the second correlative clause would have to be structured differently, i. е., with the Ag in unmarked preverbal position. Another ambiguity of Middle Indo-Aryan concerns the relationship of the incipient ergative construction to its nominalization (or the nominalization of its passive counterpart). These two cannot be distinguished as shown in (18): (18)

guruna bhaniu pâyachittu teacher+instr said+neut penance+neut

which is ambiguous between 'The teacher ordered penance' and its nominalization 'Penance ordered by the teacher'. The way out of this ambiguity is to form an adjectival determinative compound shown in (19): (19)

guru-bhaniu pâyachittu teacher-said+neut penance+neut 'Penance ordered by the teacher.' [12th c., Sc. 683.2-3]

Modem Hindi-Urdu keeps the agentive postposition se of the passive voice but the past participle has to be reinforced by the past participle of the copula, as shown in (20); (20)

guru=se likhr huí kitäb (Hindi-Urdu) teacher=abl write+pp+fem b e + p p + f e m book Ά book written by the teacher. '

A significant "improvement" on this state of affairs came with the appearance of a new analytic passive construction with the auxiliary jânâ 'go'. This

106

Vit Bubenik

construction is shown in Chart 3 as coexisting with the 'be'- passive. Its importance is obvious: it supplies us with a new passive fonnation allowing thus for an unambiguous interpretation of the 'be'- passive as an ergative. Thus, historically speaking, in Modem Hindi-Urdu, the past participle in the ergative construction continues the old passive participle, whereas in the passive construction the past participle has to be reinforced by the present or the past participle of the verb to "go". Contrast the ergative us=ne kiyâ 'he (has) made it' with the passive us=se kiya gaya 'it was made by him' in the last line of Chart 3. The final step in the evolution towards ergative organization in Western Indo-Aryan languages was taken as late as the sixteenth to seventeenth centuries when distinct postpositions for the ergative subject vs. the agentive phrase in the passive construction appeared. As you may gather fi-om examples (5) - (8) and (11) — (12), in Hindi-Urdu the agent in the ergative construction is marked with ne whereas that in the passive construction with se. In previous stages of the history of Indo-Aryan languages both of these ftinctions were expressed uniformly by the instrumental suffix inherited from Old Indo-Aryan. The preceding discussion is summarized in Chart 4, which shows that in Middle Indo-Aryan the incipient ergative construction could be distinguished from its passive counterpart on syntactico-pragmatic grounds but not morphologically. At the end of the evolution, in Modem Indo-Aryan (Hindi-Urdu) this distinction is fiilly anchored in morphology. Chart 4. Evolution of ergative and passive from late MIA to modem lA languages

Ergative Passive

late MIA (Apabhramsa) "free" word order

Modem lA (Hindi-Urdu) S-final word order

N + INSTR V + P P V+PP N+INSTR

N=ne V+PP N=je V+PP'go'+PP

Ergative may be distinguished from Passive on syntactic/ pragmatic grounds (word order, topic-continuity) but not morphologically

Ergative differs from Passive moφhologically. (ERG = ne, ABL = se, passive AUX)

4. Conclusions The value of markedness shift with respect to Subj assignment in the trajectory from the accusative typology of Old Indo-Aryan to the ergative typology

The study of ergativity

107

of Hindi-Urdu was discussed in Section 2. It was argued that FG has to take into account a considerable amount of additional data to arrive at a more convincing presentation of the precise routes of markedness shifts through the entire spectrum. Most notably: the intransitive verbs were passivizable through the whole history of Indo-Aryan and there never was a stage which could be called Ergative without passive. Substantial Middle Indo-Aryan evidence for the existence of two types of passive constructions was presented in Section 3. The 'be'-passive (inherited from the Old Indo-Aryan verbal adjective in -to) was recategorized ultimately as an ergative construction during the Modem Indo-Aryan period as a consequence of the demise of the active forms of perfect and aorist. During the Middle Indo-Aryan period the 'be'-passive was ambiguous between ergative and passive and this ambiguity was alleviated by a new passive construction with an auxiliary to 'go' at the end of Middle Indo-Aryan period. Put differently, markedness shift did not create an empty slot in the system for a new passive construction; this slot was occupied by the 'be'-passive which was ousted from it by the innovative 'go'-passive supplying a new unambiguous passive construction. In addition, as usually in linguistic history, a number of other changes "conspired" with the 'go'-passive in the ergative reorganization of Indo-Aryan: the emergence of the absolutive case (since the sixth century A.D.), the emergence of the defmitizing postposition kahû > ko (since the fourteenth century) and finally the emergence of the ergative postposition ne (since the sixteenth to seventeenth centuries). On the whole, the four stages in markedness shift were far from being discrete; FG - when used in historical linguistics - has to put more emphasis on continuity of change than on discrete typologically motivated stages. Of equal importance, more attention has to be paid to moφhosyntactic phenomena - as we saw, the existence of two distinct analytic passives (the 'be'- and the 'go'-passive) was of fundamental importance for the reconstruction of the trajectory from Type II to Type IV - an aspectually split language with marked passive. References Anderson, Stephen R. 1977 "On mechanisms by which languages become ergative", in: Charles N. Li (ed.), Mechanisms of syntactic change. Austin: University of Texas Press, 317-363. Bubenik, Vit 1989

"On the origins and elimination of ergativity in Indo-Aryan languages", Canadian Journal of Linguistics 34: 377-398.

Comrie, Bernard 1978 "Ergativity", in: Winfred P. Lehmann (ed.), Syntactic Typology. Austin: University of Texas Press, 329-394.

108

Vit Bubenik

Dik, Simon C. 1978 Functional Grammar. Amsterdam: North Holland. 1980 Studies in Functional Grammar. London: Academic Press. 1989 The theory of Functional Grammar. Part I: The structure of the clause. Dordrecht: Foris. Givón, Talmy 1984 Syntax: A functional-typological introduction. Vol. 1. Amsterdam: Benjamins. Hock, H. H. 1986 "P-oriented constructions in Sanskrit", in: Bh. Krishnamurti at al. (eds.). South Asian languages: Structure, convergence and diglossia. Delhi: Motilal Banarsidass, 15-26. Klaiman, M. H. 1978 "Arguments against a passive origin of the ergative", Chicago Linguistic Society 14: 204-216. Schokker, G. H. 1969-70 "Theyäna-passive in the Modem Indo-Aryan languages", IIJ Ì2: 1-23. Stump, G. T. 1983 "The elimination of ergative patterns of case-marking and verbal agreement in Modem Indie languages", Ohio State University Working Papers in Linguistics 27: 140-164. Trask, R. L. 1979 "On the origins of ergativity", in: Frans Plank (ed.), Ergativity. London: Academic Press, 285^04. Wallace, W. D. 1982 "The evolution of ergative syntax in Nepali", Studies in the Linguistic Sciences 12.2: 147-211.

Object assignment in a Functional Grammar of Croatian revisited Mario Brdar 0. Introduction The aim of the present paper is twofold: on the one hand we intend to address the question of (ir)relevance of Object assignment in a FG of Croatian; and on the other hand, at the theoretical level, to speculate in the light of our findings on some possible consequences for the model of FG. The organization of the paper is as follows: in the first part we review in tum the criteria for the relevance of Object assignment as discussed in Dik (1978, 1980) and Dik (1989) in some detail, and then proceed to consider their output in Croatian. In the second part of the paper we will try to broaden the perspective and cursorily discuss some Hungarian and Russian data relevant for our topic. These typological considerations will enable us to formulate a series of questions concerning the gap between some basic principles underlying FG and their implementation in actual language description.

1.

Object assignment in FG of Croation?

1.1.

Criteria for the relevance of Object assignment in FG

1.1.1. Obj-assignment in Functional Grammar (Dik 1978, 1980) There are several practical diagnostic procedures in the model of FG that are used in Dik (1978: 99-105, 1980: 129-132) to determine whether a particular language has a distinctive operation of Obj-assignment. They are all based on the existence of distinctive oppositions where Obj function is alternatively assigned to Goal or to some other semantic function. The first type of opposition that is criterial for Obj-assignment is the socalled dative alternation illustrated in the following pair of English sentences: (1) a. b.

Petór(AgSubj) gave the money(о^оы) to /'eter(AgSubj) gave /oA«(Recobj) the money^Qo)

According to Dik (1980: 129) these sentences exhibit a series of crucial properties. The members of such pairs of sentences are "synonymous" to the extent that they designate the same set of SoAs. We also note a formal differ-

110

Mario Brdar

enee in the expression of the Recipient: in the second sentence it gets the same formal expression as the Goal in the first, i. e. its prepositional marker of the semantic function is suppressed. At the same time a significant difference in the ordering of constituents is evident: in the second sentence the Recipient occupies the same position as the Goal in the first. The second type of opposition relevant for Obj-assignment concerns socalled "Raising" constructions. In Dik (1979: 124) it is assumed that pairs such as (2) a. b.

John believed that Peter had killed the farmer. John believed Bill to have killed the farmer

can be accounted for if we regard them as the result of alternative syntactic ñmction assignment. In (2b) the Object function at the level of the main clause is exceptionally assigned to the Subject of the embedded clause. The formal differences exhibited in (2a) and (2b) are consequences of this special form of Obj-assignment. Dik (1980: 130) points out that oppositions of this type can be described in terms of differential Obj-assignment only if the "raised" constituent displays behaviour typical of an object of the matrix clause, both regarding its form and its position. Finally, the third type of opposition that is regarded in Dik (1978: 99-105, 1980: 130-131) as being criterial for the relevance of Object function in a given language is found in pairs of causative constructions such as (3) and (4) in Dutch and English respectively: (3) a. b. (4) a. b.

Marie Mary Marie Mary

liet let liet let

de boodschappen doen the shopping do Jan de boodschappen John the shopping

door Jan. by John doen. do

Mary had her health-book signed (by the secretary). Mary had the secretary sign her health-book.

We observe that the alternative expressions for the Causee in (3) and (4) differ with respect to their form and their position: in (3a) and (4a) the Causee is expressed by means of Agent phrases {door Jan and by the secretary), while in (3b) and (4b) it behaves like an object of the causative verb (i. e. Jan and the secretary immediately follow the verb and have no prepositional markers). 1.1.2. Obj-assignment in The theory of Functional Grammar (Dik 1989) In The theory of Functional Grammar the basic requirement that Obj-assignment is relevant in a language only if semantic roles other than just Goal are

Object assignment in Croatian

111

accessible to it is retained, but there are some new elements that are significantly different from the earlier proposal. Certain of these shifts can be shown to run contrary to the very core of the fimctional approach to linguistic phenomena and are, actually, in conflict with certain other innovations. One of the most striking developments in The theory of Functional Grammar that should be highlighted is the general tendency towards a more unified account of both syntactic functions. The operations of Subj- and Obj-assignment were recognized at the very outset of Functional Grammar as imposing alternative perspectives, i. e. as expressing primary and secondary perspective in relation to a single SoA, but in The theory of Functional Grammar they become more tightly linked. It is explicitly and implicitly stressed throughout Dik (1989) that the different examples of Subj- and Obj-assignment are far from randomly spread across languages and that there is a common pattern in the variation found (cf Dik 1989: 219, 236). Languages are claimed to vary from no Subj- and Obj-assignment at all to maximal exploitation of the two strategies. In between we find an orderly progression of a typological hierarchy. The operation of Obj-assignment is in fact argued to be dependent on Subj relevance in a given language. The possibilities could be summarized, following de Groot (1989: 97), in the following implicational scale: (5) language type 1 language type 2 language type 3

Subj + 1

-

Obj + -

A language for which Subj-assigment is relevant may but need not have a distinctive operation of Obj-assignment. However, if for a given language Subj-assignment is irrelevant, it is predicted that it will also lack a distinctive operation of Obj-assignment. If the syntactic function Obj is relevant to the description of a certain language, the syntactic function Subj will be relevant too (cf Dik 1989: 235). The practical procedure used in The theory of Functional Grammar to establish whether a language has a distinctive operation of Obj-assigment or not is illustrated by reference to a single criterion. There is no mention of causative constructions at all, and "Raising constructions", exemplified in (2a) and (2b), do not seem to be criterial either. They are rather seen as actually correlating with the relevance of Obj. Discussing the relevance of Subj-assigment in general Dik (1989: 219-220) postulates that it is relevant to a language only if that language has a regular opposition between active and corresponding passive constructions such that passive constructions are

112

Mario Brdar

alternative expressions of a predication which can also be expressed in the active, i. e. the valency of the predicate must be the same. Alternative expression of the same SoA is a necessary but not a sufficient condition for recognizing an active-passive opposition: in the passive construction, some nonfirst argument must have acquired the coding and behavioural properties which characterize the first argument in the active. The relevant coding properties in English are: occurence in positions reserved for subjects; no prepositional marking; nominative case (if pronominal), and agreement in person and number with the finite verb. The behavioural properties in the sense of Keenan (1976: 313-320) are related to grammatical processes and constructions typically sensitive to the Subject function, e. g. reflexivization, relativization, control phenomena in infinitival complements, participial constructions, and the existence of raising constructions (Dik 1989: 221-223). Obj-assignment is to be treated in a similar fashion: analogous criteria should apply, i. e. identity of SoA and sharing of coding and behavioural properties in common with the second argument of the active construction. This is exemplified on pairs of sentences with dative alternation in English and German repeated here: (6) a. b.

The boy gave the flowers to the girl. The boy gave the girl the flowers.

(7) a. b.

Der Junge gab die Blumen dem Mädchen. Der Junge gab dem Mädchen die Blumen.

German, unlike English, does not allow two ways of presenting the situation. The difference between (7a) and (7b) can be explained in terms of alternative placement rules. The act of giving is alternatively expressed according to the relative prominence of the participants. However, the whole issue of Obj-assignment is, in comparison with Subjassigment, dealt with in a brief and rather vague manner. It is, in particular, not clear which coding and behavioural properties are meant. We could, with a high degree of certainty, make guesses about some of them for English, such as occurrence in positions reserved for objects, no prepositional marking, accusative - i. e. objective - marking for pronominal objects, etc. Such a proposal, though, gives rise to a number of problems, as we shall demonstrate in the following sections. Another innovation worth stressing is that it is suggested (Dik 1989: 235) that accessibility to Subj/Obj cannot be one-dimensionally defined in terms of semantic fiinctions. A pluridimensional, "multifactor" approach is required, in which different parameter values may contribute to the relative accessibility of a term to Subj/Obj assignment.

Object assignment in Croatian

113

1.2. Approaches to Obj-assignment in a FG of Croatian Although we have seen that some of the criteria for Obj relevance are no longer considered to be decisive in The theory of Functional Grammar, we will, in order to facilitate a comparison of the two approaches and the problems they lead to, review in the following part of the paper the output of all the three criteria in Croatian. Let us begin with the two criteria that are considered redundant in the most recent version of the model. As for the type of oppositions found in pairs of English and Dutch causative constructions, there can really be no doubt that there are no corresponding structures in Croatian. Contrary to Dik's (1980: 131) claim that Croatian has no causative construction at all, we find analytic constructions like (8) and (9): (8)

Ivan si je dao sasiti odijelo. Ivan himself aux let sew suit 'Ivan had a suit made.'

(9)

Kralj ga je dao zatvoriti. king him aux let imprison 'The king had him imprisoned.'

This type of construction, however, cannot be explained by adding a Causee Agent. Regarding the so-called "Raising constructions", Dik and Gvozdanovic (1981: 23) claim on the basis of the following Croatian equivalents of EngHsh Raising structures in (10) that Croatian does not appear to display the prerequisite behaviour: (10) a. John believed that Peter played a sonata. b. John believed Peter to play a sonata. (11) a. Ivan mis И Ivan thinks 'Ivan thinks b. Ivan misli Ivan thinks 'Ivan thinks

da Petar svira sonatu. that Petarnos plays sonataacc that Petar plays a sonata.' Petar da svira sonatu Petarnom that plays sonataacc that Petar plays a sonata.'

It is obvious that, in the above Croatian examples, there is no Raising at all: Petar is in both cases in the nominative, in (lib) it is merely displaced from the embedded clause.

114

Mario Brdar

Dik and Gvozdanovic (1981: 24) also cite pairs where there is nominative case in the one and accusative case in the other construction: (12) a. Ivan vidi Ivan sees 'Ivan sees b. Ivan vidi Ivan sees 'Ivan sees

kako Petar svira sonatu. how Petarnom is playing asonataacc how Petar is playing a sonata.' Petra kako svira sonatu. Petaracc how plays sonataacc Petar, how he is playing a sonata.'

But again, it is reasonable to assume that Petar^^c in (12b) is a real matrix Goal argument rather than the argument of the embedded verb. Note that the embedded verb is still finite (it is inflected for the 3rd person singular). It should also be pointed out that there are accusative-with-infinitive constructions in Croatian with verbs such as vidjeti 'see', cuti, 'hear', slusati 'listen to' b: (13)

Mi smo ga culi pjevati tu pjesmu vise puta. we aux him heard sing that song several times 'We heard him sing that song several times.'

Such cases are not discussed in Gvozdanovic's (1981) paper on displacement phenomena, although the construction is possible in Croatian. It is not certain that it should be explained away in a fashion similar to de Groot's (1989) treatment of accusative-with-infinitive in Hungarian, discussed briefly in 2.1. below. However, concluding from this that there are no Raising constructions in Croatian for which alternative Object assignment would be required would mean overshooting the mark considerably. The existence of such constructions was pointed out and discussed en passant by Kucanda (1984: 108). He shows that in (13b) the Subject of the embedded predication has been "raised" to Object on the level of the main predication: (14) a. On/'subj smatraju da je /vansubj izdajnik. they think that cop Ivannom traitornom 'They think that Ivan is a traitor.' b. Oni'subj smatraju Ivanaobj izdajnikom. they think IvaUacc traitorinsw 'They think Ivan a traitor.' According to Mihailovic (1976: 57) there are a number of verbs in Serbian and Croatian that allow Subject-to-Object-raising (e. g. drzati 'take ioT\proglasiti 'declare', odrediti 'appoint', primati 'recognize').

Object assignment in Croatian

115

Concerning the dative alternation, which is in The theory of Functional Grammar rehed upon as the only criterion, Dik and Gvozdanovic (1981: 2223) maintain on the basis of sentences such as (15) a. Ivan je dao jabuku Ivanki. Ivan aux given applCacc Ivankajat 'Ivan has given the apple to Ivanka.' that there is no dative alternation in Croatian. In the above example the Goal is in the accusative case, and the Recipient in the dative. The relative order of the Agent, the Goal and the Recipient may vary as shown by (16b) (16e): (16) b. c. d. e.

Ivan je dao Ivanki jabuku. Jabuku je Ivan dao Ivanki. Jabuku je Ivanki dao Ivan. Ivanki je Ivan dao jabuku.

But in all these orderings we note that the formal expression of the Goal and the Recipient is constant, so that Object function would invariantly be assigned to the Goal. Kucanda (1984: 108-109) argues that sentences: (17) a. 0«i(AgSubj) su /va«a(RecObj) ponudili bijelim vinom. theynom aux IvaUacc offered white wincinstr 'They offered Ivan white wine.' b. 0«i(AgSubj) su Ivanu^R^^c) ponudili bijelo vino^ooobj) theynom aux Ivandat offered white winca ^'acc 'They offered white wine to Ivan.' falsify Dik and Gvozdanovic's claim: the Recipient term in (17a) is marked with the same case form as the Goal in (17b). Kucanda proposes a common semantic structure for both (17a) and (17b): (18)

ponuditiy (xl: animate (xl))Ag (x2)go (x3:animate(x3))Rec

In her rejoinder to Kucanda, Gvozdanovic (1986: 105) tries to show that (17a) and (17b) do not designate the same SoA. The semantic structure proposed by Kucanda in (18) only underlies (17b) in her analysis. She argues that Ivana^cc in (17a) is semantically not a Recipient but rather a Goal, as in (19)

Oni su Ivana ponudili Petru za dopisnika. theynom aux IvaUacc offered Petarjat for correspondentacc 'They offered Ivan to Petar as a correspondent.'

116

Mario Brdar

This parallel drawn by Gvozdanovic seems rather arbitrary: while Ivana in (19) is, in a manner of speaking, the transferred entity, the same cannot be maintained for (17a). Gvozdanovic stipulates that in (17a) we are in fact dealing with a 2-place predicate frame derived from (18), where the only arguments are the Agent and the Goal, the latter being the person that is in her analysis offered something. This predicate-frame can be extended as in (17a). This view is argued to be supported by the grammaticality of (19a), and ungrammaticality of (19b): (19) a. Oni su Ivana ponudili. theynom aux Ivanacc offered 'They offered (sth. to, i. e. treated) Ivan.' b. *Oni su Ivanu ponudili. they aux Ivanjat offered 'They offered (sth.) to Ivan.' The grammaticality of (19a) is also assumed to be related to the fact that this derived predicate-frame involves a semantic shift where ponuditi does not mean 'to offer' but 'to treat', without specifying by means of what. The grammaticality of (19c): (19) c. Oni

su

ponudili

vino.

theynom aux offered winCacc 'They offered wine.' is, according to Gvozdanovic, connected with the semantic shift by which ponuditi does not mean only 'to offer', but 'to offer for sale'. At first sight, this line of argumentation seems convincing. However, a number of serious problems arises as soon as one goes into details of the solution propounded by Gvozdanovic. We have already pointed out that most native speakers of Croatian would judge the parallel drawn by Gvozdanovic between the two occurrences of /vanflacc in (17a) and (19a) as counter-intuitive. Ivana^^ in (17a) is definitely felt to be more Recipient-like. It would be extremely interesting to ascertain how many different shades of meaning the verb ponuditi — or some similar verb — can have, and how many different predicate-frames would be required to cover all these possibilities. There is no doubt that the verb ponuditi, at least in the dialect of Croatian we tested, is considered to be highly polysemous, i. e. it need not necessarily mean only 'to offer for sale' in (19c) or 'to treat' in (19a). These senses

Object assignment in Croatian

117

are rather difficult to keep apart without an appropriate context. The Croatian verb does not seem to enforce this type of distinction. Compare: (20)

Oni su ponudili macke. theynom aux offered catSacc 'They offered cats (for sale).' 'They treated cats (with sth.).' 'They treated (someone) with cats.'

As can be seen from the above glosses, the verb ponuditi could be inteφreted in a number of ways, either as meaning 'to offer for sale' or 'to offer someone some food'. The sentence is, in fact, ambiguous if the verb refers to an act of offering food: cats may be getting something to eat, but at the same time they may be interpreted as something offered as food, due to the fact that one and the same expression may be interpreted either as a Goal or as a Recipient, ponuditi being inherently a 3-place predicate. But there are even worse cases of syntactic ambiguity: (21)

Oni su ponudili ^//û(RecObj)/Goaiobj) гпаскатпа^ао^х^ц^^) theynom aux offered Alfacc catSdat/instr 'They offerd cats to A l f or 'They offered Alf to cats.'

Either of the two NPs in the oblique case could be interpreted as the Recipient or as the Goal. (22) a. Oni su ponudili Alfu^^t machetee 'They offered cats to Alf.' b. Oni su ponudili machetee ^IfoniinstT 'They offered Alf to cats.' (22a) and (22b) clearly refer to two different SoAs, but for each of them there is an alternative rendering collapsed in the ambiguous sentence (21) above. This seems to show that any ad hoc attempts to relate (17a) and (17b) to various shifts in meaning could not be extended to cover all cases and that some such pairs could still be assumed to designate the same state of affairs. In other words, there are at least some cases of dative alternation with ponuditi. Even if the above arguments are rejected as far-fetched and hinging on extreme examples with case syncretism, there are some other verbs in Croa-

118

Mario Brdar

tian showing the same type of alternation for which semantic shifts can hardly be assumed: (23) a. On je darovao slugu s 5 diñara. he aux presented servantacc with 5 dinarSinstr 'He presented the servant with 5 dinars.' b. On je darovao sluzi 5 diñara. he aux presented servantjat 5 dinarSacc 'He presented 5 dinars to the servant.' Similar behavior characterizes verbs such as podariti, zaduziti, razduziti. We also note that our candidates for Obj-assignment in (17) and (23) exhibit one of the characteristic coding properties of objects in Croatian, i. e. the accusative marking, whereas the other expression, without syntactic function, is in the instrumental or dative case, or is prepositionally marked. In sum, contrary to Dik and Gvozdanovic, we have so far found at least some evidence that Croatian meets some of the requirements proposed and that its status as a prime example of a language for which Obj-assignment is irrelevant would be in need of careftil reexamination within the model presented in Dik (1979, 1980). It is also obvious that it cannot be unequivocally classified as a non-Obj-assigning language within the model developed in Dik (1989).

2. Assessment of the two approaches 2.1. Broadening the perspective: Object assignment in Hungarian and Russian Since Croatian is, of course, not the only language for which Object assignment has been maintained to be irrelevant, it would be illuminating to broaden our perspective and discuss briefly some Hungarian and Russian data pertinent to our topic. In de Groot (1981a: 43, 1981b: 77, 1989: 95) it is claimed that Object assignment is irrelevant in Hungarian. However, upon a closer look, we find an interesting situation: one of the three criteria used in FG seems to be satisfied. There appears to be no dative alternation in Hungarian, nor is the status of the Causee Agent the same as in English or Dutch, since it is not in the objective case. But we do find Raising constructions. They are not altogether alien to Hungarian, although they are, as Kenesei (1980: 181-184)

Object assignment in Croatian

119

points out, of more limited range than in English. According to Kenesei, Subject-to-Object-raising occurs in Hungarian with verbs such as lát 'see' or hall 'hear': (24) a. János János 'János b. János János 'János

látta, hogy Péter elment. see-past that Péter leaves saw that Péter had left.' látta Pétert elmeni. see-past Péteracc leave saw Péter leave.'

Kenesei points out that these verbs do not allow Raising unless the action of the embedded clause is simultaneous with that of the matrix clause. De Groot (1989: 111) argues against any raising analysis and assumes that an explanation of the co-occurrence of sentences such as (24a) and (24b) is to be found in the semantic difference between the two. Predication (24b) is true only if János actually saw Péter. On the other hand, there is no such truth condition for (24a), as de Groot illustrates. It may have been the case that János saw that Péter's car was not in the parking-lot, from which he concluded that Péter had left the place. However, even if we agree with this analysis, there seem to be other, relatively frequent constructions in Hungarian for which a raising analysis in the sense of Dik (1979) seems to be necessary: (25) a. A tanárok azt tartják, the teachers that consider hogy Péter rendkívül értelmes. that Péter extremely intelligent 'The teachers consider that Péter is extremely intelligent.' b. A tanárok rendkívül értelmesnek tartják Pétert. the teachers extremely intelligentdat consider Péteracc 'The teachers consider Péter extremely intelligent.' c. A fiúk azt mondják, hogy Mária csinos. the boys that claim that Mária nice 'The boys claim that Mária is nice.' d. A fiúk Máriát csinosnak mondják. The boys Máriaacc nicejat claim 'The boys claim Mária to be nice.'

120

Mario Brdar

There are no causative constructions in Russian comparable to (3) and (4) in Dutch and English respectively. As for the dative alternation, we do find a handful of cases, but the construction in (26b) is marked as obsolete: (26) a. Ja blagodaren vam za vnimanie, I thankful youjat for attention kotoroe vy mn'e darili. whichacc you mcdat lend-past 'I am thankfiil to you for the attention you showed to me.' b. Ja blagodaren vam za vnimanie, I thankful youjat for attention kotorym vy men'a darili. whichinstr you mCacc lend-past Ί am thankfiil to you for the attention you showed to me.' However, "Raising constructions" are fairly productive in Russian (cf Rùzicka 1976: 14-16): (27) a. Maria Maria 'Maria b. Maria Maria 'Maria

scitaet, considers considers scitaet considers considers

cto Boris intel'igentnyj. that Borisnom intelligent that Boris is intelligent.' Borisa intel'igentnim. BoriSacc intelligentinstr Boris intelligent.'

2.2. How functional is the FG approach to Object assignment? The two approaches to Object assignment clearly show some difficulties which FG is bound to run into if the apparent tendency towards its more precise (but at the same time more rigid) formalisation is to continue. Both appoaches could be criticised on general grounds because, as far as we can see, there is a case par excellence of isomoφhism implicit in both analyses. If the Object function were indeed invariably assigned to Goals in Croatian or some other language, we would have a nearly perfect one-to-one correspondence between a semantic function and its moφhological and/or syntactic encoding. There is, undoubtedly, a certain degree of isomoφhism and iconicity in all languages, but this cannot be expected to be perfectly systematic and regular. From all we know about languages, such a correspondence would be highly unlikely to obtain in a natural language and would go against one of the basic tenets of FG, namely that language is primarily an instrument of social interaction. Since communicative needs are in a constant

Object assignment in Croatian

121

process of change, natural languages must keep pace and be flexible enough to adapt to these needs. This flexibility may manifest itself in many respects. Generally speaking, we observe that natural languages have an in-built mechanism for this self-regulation: the asymmetric duality of the linguistic sign as described by Karcevskij (1929) enables changes to occur in the course of which a linguistic form may come to cover new contents — mainly due to the processes of analogy and metaphorical extension. This omnipresent duality and open-endedness gradually destroys ground for isomorphism in a language. We know that the use of morphological and syntactic devices can be extended to encode a number of semantic functions. This is true even in a language with quite a high degree of semantic transparency such as Hungarian, where a variety of semantic functions may be collapsed into a single morphological marker. So the postposition -t does not always mark Goals nor does -val always mark Instruments. We saw that the accusative case — the prototypical Object-marking case form in Croatian - can encode not only Goals but also Recipients, and that Goals could be moφhologically encoded not only by the accusative but by the instrumental or genitive as well. Falster Jakobsen (1989: 62 and correction sheet) gives a set of examples in German where a similar situation obtains: (28) a. jemandem^ú^t^^ etwas^ccOo rauben b. jemanden^ccRec ^iner SachCgenGo berauben c. Jemanden^ccRec berauben As for the treatment of Obj-assignment in FG, we note that the data cited by Dik and Gvozdanovic show quite a high degree of correlation between the criteria for Object relevance: unlike in English, where all the three criteria are satisfied, Croatian allegedly displays none of the required behavioural properties. Such a high degree of correlation would make any theory quite appealing and striking for its formal elegance, but it also kindles a spark of suspicion. We hope to have shown that the situation is not so clear-cut in Croatian, for at least some of the requirements discussed above can be met. Note that we cannot evade the problem by calling into question the productivity of certain constructions, a matter which we will address presently in connection with the analysis proposed in The theory of Functional Grammar. To put in a nutshell, the problem FG, and any similar analysis, is faced with is that we simply lack a principled answer to the following question: How are we to treat cases where only some, one or two from the whole set of criteria are satisfied? Can the assignment of a particular function still be considered irrelevant? Let us note in passing that there is possibly one more

122

Mario Brdar

problem involving the differences between languages such as Hungarian, Russian and Croatian. We could solve the problem by assuming that all the three criteria must be satisfied. However, prior to adopting this view, one should scrutinize the status of these criteria. How do we, after all, arrive at a set of valid criteria? Can we really be confident that the approach advocated by Dik and Gvozdanovic is not biased towards certain languages? It should be kept in mind that there may be other types of oppositions involving Object assignment in various languages. Since our criteria are primarily meant to check whether the Object function is invariably assigned to Goals or not, it seems that it would not be totally out of place to suppose that Object assignment is not irrelevant in a model working with a set of criteria if some of them are met. If it is not irrelevant, we might as well ask whether it is relevant in the same way as it is in English? The streamlining of the set of criteria for Object relevance we witness in The theory of Functional Grammar appears to be related to the fact that the two syntactic functions are now tightly linked to each other. However, the methodological parallel between Subject and Object assignment, although formally elegant, is suspect on empirical grounds. We have already noted that it is far from clear which coding and behavioural properties are meant in the case of Object assignment. In spite of much fruitñil and stimulating work on objects in a cross-linguistic perspective (e. g. Hopper and Thompson 1980, Hopper and Thompson 1982, Plank 1984) hardly anyone has dared yet to put forward such a list of object properties parallel to Keenan's (1976) list for subjects. Moreover, if the dative alternation is to be accepted as a universal and alldecisive criterion parallel to passivisation, there arises the problem of the difference in scope between the two phenomena. The former seems to apply in English, on the one hand, only to a semantically defined subset of 3-place predicates, i. e. mainly to those denoting some kind of transference of an entity between two animate participants. On the other hand, dative alternation is by no means as productive as passivisation, i. e. it does not apply freely even to all predicates denoting transference, e. g. donate. Admittedly, the number of verbs that exhibit dative alternation in Croatian is not large, but as long as they display the relevant behaviour they fall into the domain of the criterion, no matter whether they follow a productive pattern or not. Besides, there seems to be a gradual transition between unproductive and productive constructions, and it is obviously impossible to decide on the exact number of triggers that will suffice for a pattern to be considered productive. A similar objection is expressed in Falster Jakobsen (1989: 62), who also

Object assignment in Croatian

123

notes another problem with Object assignment in The theory of Functional Grammar, namely the problem of the sameness of the SoA in dative alternation: is the whole difference between alternative renderings just a matter of different perspective in the SoA? She shows that the Danish constructions (29) a. at to b. at to

give give give give

nogen somebody noget something

noget something til nogen to somebody

are not synonymous because, apart from the change of perspective, they do not necessarily have the same content, since (29b) has a more polysémie content: 'to present with' and/or 'to hand over to' as opposed to the more monosemic content of (29a) 'to present with', a phenomenon which caused many problems in the analysis of Croatian verbs such as ponuditi. Considering all these difficuhies, we ftilly agree with Falster Jakobsen (1989: 63) that Object assignment in its present form, as formulated in The theory of Functional Grammar, seems rather narrow in its grammatical scope in descriptive practice and is not applicable to all - or even a small range of - genetically closely related languages. In the first part of the present paper we hinted at the fact that there are some innovations present in The theory of Functional Grammar which actually go unheeded in the descriptive fragments there, but could contribute considerably towards a solution of present difficulties. Although we note that the whole notion of Object assignment practically hinges on the existence or non-existence of dative alternation, it is suggested elsewhere in Dik (e. g. 1989: 235) that accessibility to Subj/Obj cannot be one-dimensionally defined in terms of semantic fiinctions. A pluridimensional, "multifactor" approach is required, in which different parameter values may contribute to the relative accessibility of a term to Subj/Obj-assignment. Various contributing factors are, unfortunately, dealt with in a rather brief manner, some of them actually being phenomena argued elsewhere to be explicable in terms of Subj/Obj-assignment. It appears to be clear at this point that all the facts suggest a multifacet approach. A simple dichotomous distinction between Obj-assigning and nonObj-assigning languages does not appear to be fine enough to capture what seems to be scale or hierarchy of Object relevance. A possible way out of this stalemate between theoretical pleas and descriptive practice would be, bearing in mind that explanation and the description of language facts has primacy over methodological issues in a fiinctional model, to recognise that in order to do justice to Obj-assignment in various

124

Mario Brdar

languages and to its various degrees of relevance, we do not need fewer but more criteria. We should re-evaluate the set proposed in Functional Grammar and possibly complete it by adding some new tests rather than rely on a single one, check these tests against each other, and then try to arrive at a sort of hierarchy of criteria, and at the same time model a typological continuum along which languages could be ordered according to the degree of Object relevance. This seems to be in line with the view expressed en passant by Dik (1989: 215, footnote 8), following Kucanda (1984), that Subj-assignment may after all be marginally relevant in a FG of Croatian. This is significant because it opens up new perspectives in the treatment of syntactic functions in FG and transcends the rigid either/or approach. Another important recent contribution towards a better understanding of syntactic functions in FG is de Groot's (1989: 97) proposal to consider their relevance at various levels of analysis and thus establish holistic typologies. A syntactic function may tum out to be relevant in a language in all domains, or just at the level of finite predications, but not at other levels. Obviously, if we adopt the two proposals by Dik and de Groot mentioned above, we could recognize various degrees of relevance of syntactic functions and at the same time be able to work out a typological hierarchy for the exploitation of the two fonctions as suggested in Dik. Otherwise there is hardly any room for introducing an orderly progression in the implicational scale of their relevance. This proposal may be criticized for failing to provide empirical results for the determination of the relevance of the Object fonction as clear-cut as in previous analyses, but the notion of scalar, as opposed to binary distinctions has in general much to recommend it, particularly in the light of the omnipresent multiple vagueness of natural languages and the asymmetric duality of linguistic signs. References Danes, Frantisele 1966 "The relation of centre and periphery as a language universal". Travaux Linguistiques de Prague 2: 9-24. Dik, Simon C. 1978 Functional Grammar. Amsterdam: North-Holland 1979 "Raising in a Functional Grammar", Lingua: 47: 119-141. 1980 Studies in Functional Grammar. London and New York: Academic Press. 1989 The theory of Functional Grammar Part I: The structure of the clause. Dordrecht: Poris.

Object assignment in Croatian

125

Dik, Simon C. - Jadranka Gvozdanovic 1981 "Subject and object in Functional Grammar", in: T. Hoekstra — H. van der Hulst - M. Moortgat (eds.), Perspectives on Functional Grammar. Dordrecht: Foris, 21-39. Falster Jakobsen, Lisbeth 1989 "Der kom en soldat marcherende hen ad landevejen: En, To! On Subject and Object assignment in Danish", in: Lisbeth Falster Jakobsen (ed.), Functional Grammar in Denmark. A preprint on the Occasion of the Fourth International Conference on Functional Grammar. Copenhagen: University of Copenhagen, 46-65. Givón, Talmy 1983 Syntax: A functional-typological introduction. Vol. 1. Ann Arbor: Karoma Publishers. Groot, Casper de 1981a "Sentence-intertwining in Hungarian", in: A.Machtelt Bolkestein et al. (eds.), Predication and expression in Functional Grammar. London and New York: Academic Press, 41-62. 1981b "On Theme in Functional Grammar", in: Teun Hoekstra - Harry van der Hulst — Michael Moortgat (eds.), Perspectives on Functional Grammar. Dordrecht: Foris, 75-88. 1989 Predicate structure in a Functional Grammar of Hungarian. Dordrecht: Foris. Gvozdanovic, Jadranka 1981 "Word order and displacement in Serbo-Croatian", in: A. Machtelt Bolkestein et al. (eds.). Predication and expression in Functional Grammar. London and New York: Academic Press, 125-141. 1986 "Subject and object in Serbo-Croatian and evidence for linguistic theories". Filologija, 14: 97-108. Hopper, Paul J. — Sandra A. Thompson 1980 "Transitivity in grammar and discourse". Language 56: 251-299. Hopper, Paul J. - Sandra A. Thompson (eds.) 1982 Studies in transitivity. New York: Academic Press. Karcevskij, Serge 1929 "Du dualisme asymétrique du signe linguistique", Travaux de Cercle Linguistique de Prague 1: 33-38. Keenan, Edward L. 1976 "Towards a universal definition of "Subject", in: Charles N. Li (ed.). Subject and Topic. New York and London: Academic Press, 303-333. Kenesei, István 1980 "Sentence embedding in English and Hungarian", in: László Dezsö — William Nemser (eds.), Studies in English and Hungarian contrastive linguistics. Budapest: Akadémiai Kiadó, 177-205. Kucanda, Dubravko 1984 "On subject-assignment in Serbo-Croatian within the framework of Functional Grammar", Lingua 64: 99-114. Mihailovic, Ljiljana 1976 "Passive sentences in English and Serbo-Croatian. Part II", in: Rudolf Filipovic (ed.), Yugoslav Serbo-Croatian - English Contrastive Project Reports 10. Zagreb: Institute of Linguistics, 50-108.

126

Mario Brdar

Plank, Frans (ed.) 1984 Ruzicka, R. 1976

Objects. Towards a theory of grammatical relations. London: Academic Press. "Über vergleichende und konfrontierende Sprachwissenschaft", in: Ernst Eichler - Josef Filipec - Bohuslav Havránek - Rudolf Rüzicka (eds.), Beiträge zur konfrontierenden Sprachwissenschaft. Halle: VEB Max Niemeyer Verlag, 9-21.

Perspective, markedness, and paradigmatic relations between predicates. A case study of Danish Lisbeth Falster Jakobsen

Introduction ^ In languages which do not characterize pragmatic functions directly through morphological markers, such as Danish, the pragmatic functions are often expressed through a combination of features from different contentive and expressional levels of the language; cf. Dik (1989: 278) for ways in which Focus may be expressed, especially the two first points: (i) (ii)

prosodie prominence: emphatic accent; special constituent order: special positions for Focus constituents in the linear order of the clause.

These points imply that there is another way of realising the clause which is unmarked compared to the aforementioned marked ways. This means that in order to understand these constructions as marked they must be understood within a paradigm containing unmarked and marked coding possibilities. "Unmarked" is understood in the sense of having a higher token frequency (Dik 1989: 39) and thus a more neutral impact. I want to argue that the paradigmatic relation between unmarked and marked clausal constructions forms the basis of the different pragmatic functions of the clause in languages such as Danish. And that a third factor plays a part in the marking besides the above mentioned two features: the formal coding of all the sentence constituents triggered by the relational value (such as subject, object, adverbial, etc., in the traditional sense of the terms) given to them in the surface construction. This surface coding must be seen as a level in its own right, no matter which value the underlying argument or satellite might have. I want to argue further that in languages which have the possibility of reperspectivizing through a differentiated subject and object assignment, such as Danish, the syntactic functions are most adequately understood through a paradigmatic relation between the basic perspective of the predicate frame, which is the unmarked way of perspectivizing, and the "reperspectivization" which takes place through the different possibilities of coding the predication.

128

Lisbeth Falster Jakobsen

In this way the reperspectivizing construction is understood as marked in relation to the original, basic predicate frame. In this way too the perspective of the arguments can appear as marked in a clause where the sentence constituents are otherwise pragmatically totally unmarked. Finally, it is my claim that Object assigment, in the form described by Dik (1989) is not adequate for Danish; but that the reperspectivization of the Object assigment is more adequately understood through a paradigmatic relation between different basic predicates, consisting of the same or a similar verb root combined with different formal valencies. I want moreover to argue that as the Pragmatic ftmctions operate very late in the formation of the clause structure, i. e. when the functions have obtained such a functional "individuality" that they are almost ready to be mapped on to an expression, it might be worth considering the qualities of the ftmctions at a stage where they have reached a more "sign-like" state.^ This state is reached when the expression must be taken into consideration in describing the total ftmctionality of the constituent in the clause structure. Another sign-like quality of a constituent must be its syntagmatic potential, in terms of various selectional qualities such as government and agreement. The Predicate in FG must be understood as an individual lexical item with a syntagmatic potential, thus as a sign. At least there seems nowhere else to place the varying selectional qualities of semantically related predicates across languages, if they are not understood as signs. This means considering the possibility of a dimension in the linguistic description of FG where the effect of the interrelatedness of the two sides of a language, the content side and the expression side, is taken into account throughout the model. So starting with the semantic fimctions at the bottom of the layered model I propose four functional levels, each consisting of two parallel dimensions, to be considered for any term ending up as a constituent in the realised clause. These levels are: 1) 2) 3) 4)

I semantic ftinction perspectivization function reperspectivization ftinction pragmatic ftinction

II grammaticalized unit function argument function syntactic function^ sentence constituent ftinction

The left hand column represents the levels of FG-ftinctions which a single functional unit goes - or may go - through on its way to the top of the model. It is a purely semantic view, so to speak, a state where the unit goes through several levels of "speculative" semantic syntagmatization; spéculât-

Perspective, markedness, and paradigmatic relations

129

ive in so far as there is no combination with grammatical features. In this way the column is suited to descriptions across languages. The right hand column represents the corresponding grammaticalized view: the relational conditions needed for triggering the correct expression at the top of the model are taken into account too. The right-hand functions will thus be more bound to the actual linguistic signs and therefore more suited to the description of an individual language. The syntactic and the sentence constituent functions are two separate levels because the sentence constituents, as items in the surface structure, carry not only relational, but also prosodie and word ordering features. And because all arguments and satellites end up in sentence constituents, regardless of their having been subjected to syntactic assignment.

2. The relation between the semantic functions, A\ and argi It is quite clear from Dik version (1989) that there is a difference between the systematization and use of the "free" semantic ftinctions, e.g in the Semantic Function Hierarchy (SFH), and the semantic ftinctions related to predicates, at least for the first arguments of predicates. The Semantic Function Hierarchy is set up as follows (1989: 226):

Subj Obj

Ag > Go > Ree > Ben > Instr > Loc > Temp -Ь>-Ь>+ >-Ь >-Ь >+ >-Ь - > + > - Ь >-|- > + >+ > +

On the whole the semantic function hierarchy is understood as a crosslinguistic typology with respect to accessibility to Subj/Obj-assigment. This view is a development of a previous understanding of the hierarchy; in Dik (1989: 68-70) the predicate frame is discussed at some length with 'to give' as the example. Here it is said that: the numbering of the argument positions Xi, Хг, хз... does indeed define an ordering over these positions; ...it reflects a priority hierarchy defined over the semantic functions,...in the sense that the Agent arguments are more "central" to the predication than Goals, and these more central than Recipients. In terms of such differences, we shall have reason to distinguish "first arguments" (such as Agent), "second arguments" (such as Goal), and "third arguments" (such as Recipients).

This passage gives the reader reason to believe that there is such a thing as general ordering of arguments in a predicate frame according to a Semantic Function Hierarchy.

130

Lisbeth Falster Jakobsen

Consider now the semantic functions of the class of A', i. e. the class of all first arguments of basic predicates, (Dik 1989: 234): A^ = {Agent,Positioner,Force,Processed,Zero} Apart from Agent, no other semantic function from the Semantic Function Hierarchy is represented. This new set of semantic functions is a set which is derived from the typology of States of Affairs - the SoA type being a compositional function of the semantic properties of both predicate and terms (Dik 1989: 90). Thus from the start this is devised as a syntagmatically defined set. A' is the class of first arguments, whereas a 1st argument is a particular entity belonging to a particular predicate frame which selects its particular semantic fiinction out of the set in A^ So obviously, the semantic functions of the Semantic Function Hierarchy and those of A' are two different sets: the semantic functions of the Semantic Function Hierarchy set and their ordering are not related to the particular basic nuclear predicate frame as opposed to the set of A', which has no ordering at all. In Dik (1989: 234) the idea of ordering arg2 and args of a predicate according to the Semantic Function Hierarchy is also abandoned in favour of integrating the notion of an unordered A^ into the Semantic Function Hierarchy: A' > Ag Pos Fo Proc 0

Az > R e c > B e n . . . Go Ree Ben Instr etc.

This integration seems to me a confusion between the idea of the original Semantic Function Hierarchy: a "pure" semantic statement about the ordering of semantic functions in relation to their possible matches with different syntactic expressions across languages, and on the other hand about the actual relation between semantic functions and the numbering of arguments belonging to individual predicates in individual languages. In cases where the predicate frame contains more than one argument, how is the 1st argument (a member of the class A^) then found? In reality it is found by matching the argument with the final expression of the term in a clausal structure with an active predication, thus numbering the argument in question in accordance with the desired expression. This means that there are no independent properties or functions which characterize a 1st argument or the class A'; all properties belong to the syntagmatic-relational or to the purely expressional dimension of language.^ Therefore A' has to be considered as a class of entities each having a coding potential which later trig-

Perspective, markedness, and paradigmatic relations

131

gers the final expression, and a matching potential for the correct lexicosemantic selection of the argument. In the case of A^ of the revised Semantic Function Hierarchy (Dik 1989: 234), this class also seems to be founded on the actual realisations of second arguments. But as to their semantic fonctions, they don't acquire a new set when becoming a part of a basic nuclear predicate frame like the first argument; A^ as a class gets only an unordered selection of semantic fiinctions from the originally ordered Semantic Function Hierarchy. This gives rise to two questions: 1)

If first arguments are affected by the nuclear predicate fi-ame in such a way as to be able to adopt semantic functions fi-om outside the Semantic Function Hierarchy, why are arg2 and arg3 not subject to a similar inñuence from the nuclear predicate fi-ame? Arguments and satellites, when expressed, are normally both expressed as sentence constituents. The relation between the semantic fimction of a satellite and its final expression is in FG viewed as a match between the term and a "marker" indicating the semantic fiinctions in a more or less unequivocal way (Dik 1989: 309-315), whereas second and third arguments ending up as objects are masked, i. e. neutralized as to marking. This neutralization raises the question of the semantic fiinctions of arg2 and arg3. Are their semantic fiinctions really unaffected by the predicate fi-ame? If so, why should the arguments then be masked?

2)

What is the general relation between the concept of semantic functions as seen in the Semantic Function Hierarchy and scattered through The theory of Functional Grammar and the set of semantic functions of A'? Are they mutually exclusive or do they combine?

Following this train of thought, arg2 and arg3 might also be semantically affected by the syntagmatic relation to a nuclear predicate;® from this it follows that there might be a new aspect of semantic fiinctions to be studied: the relation between syntagmafically unbound and bound semantic functions.

3. The relation between an argument, a constituent under Subject/Object assignment and a sentence constituent Dik assumes (1989: 222) that Subject assignment is relevant in a language if that language has a "passive" construction with the following properties: (i)

the construction provides an alternative way of expressing the same SoA as the corresponding "active" one;

132

Lisbeth Falster Jakobsen

(ii)

it contains a nonfirst argument which has a number of coding properties in common with the first argument of the corresponding active construction; and

(iii)

has a number of behavioural properties in common with the first argument of the active construction.

Similar criteria apply in the case of Object assigment: identity of SoA and common coding/behavioural properties, this time with the second argument of the active construction. This means that the whole set of first arguments is realised in the same sentence constituent, a "subject" in an active clause, defined by one set of surface properties. But the reverse does not hold: the set of sentence constituents carrying this set of surface properties has four derivational subsets: (i) (ii) (iii) (iv)

first argument in predications in the active voice, where the predicate has a possibility of passive assignment; second argument (or for some languages an argument with a higher number) in predications in the passive voice; first argument in predicates without an alternative possibility of assignment; a dummy (a "formal subject") in the case of predicates without an argument.^

As it is assumed (Dik 1989: 233-235) that A ' is a cover-all name for the class of first arguments, there is no merging of identity between any given Semantic function in the set of A\ such as Agent, and the Subject function; all the different functions in the class of A ' are elements of the class of subjects. There is no identity between the class of constituents under Subject assignment and the class of sentence constituents realised as "subject" either. Thus there is no 1:1 relation between the entities on the level of arguments, the level of Subject/Object assignment and the level of sentence constituents. In this way we either need two sets of names: one for the level of Subject/ Object assignment and one for the level of sentence constituents - or, as all constituents realised with the coded and behavioural properties of the subject constitute one class at their particular level, they must all be called subject by default, irrespective of assignment and possible relation to the functions of A ' . However this may be, the level of sentence constituents must be recognized as a level in its own right with respect to both function and expression, including the positional potential attached to the different sentence constituent values. The differences among languages, however, may make a more detailed description impossible within a general linguistic framework.

Perspective, markedness, and paradigmatic relations

133

4. The autonomy of the levels of the right hand column. Perspective and markedness Although there is no 1:1 relation between A' as a class, the class of arguments with a potential for Subject assignment (which includes the subject of passives) and the class of subject sentence constituents which I have proposed, there is of course a linking factor between the levels in particular cases: the predicate. The predicate gives its arguments a basic perspective which must be considered the prototypical, unmarked perspective. This is the perspective that we find realised in dictionaries: do alicui aliquid jemandem etwas geben iemand iets geven at give nogen noget to give somebody something Here with an unexpressed argi which is always taken to be standing to the left of the expression. These expressions are so to speak realisations of the basic predicate open in Х]. This unmarked perspective can be changed into a marked one by Subject assignment to an argument with a higher number than 1 and maybe by Object assignment as well. Markedness must necessarily have a counterpart in unmarkedness: in this case the markedness of the perspective at the level of Syntactic functions is understood against the background of the unmarked perspective of the basic predicate frame, in other words: there is a paradigmatic relation between the predicate frames on the two levels. When an argument has undergone Subject or Object assigment, it is expressed by a sentence constituent of some special formal marking. Sentence constituents, however, normally have both unmarked and marked positions in the clause; these positions are primarily connected with the value of the sentence constituent (nominal, verbal, adverbial, etc.) with some formal exceptions according to LIPOC (Dik 1989: 351). The final position of a sentence constituent is regulated by the demands of the pragmatic level. This means that the two levels, Syntactic fimctions and Sentence constituent fimctions of column II (cf 1. Introduction), form a paradigm where the arguments with their assigned Syntactic fimctions will be the unmarked basis, to be realised in an unmarked or marked way according to the final position of the sentence constituent. Take an arg2 at the level of Argument function:

134

Lisbeth Falster Jakobsen

it is related paradigmatically to the level of Syntactic functions. As a nonsubject arg2 is unmarked in relation to its argument number; if it is assigned Subject function, it is marked. As a subject surface constituent of a clause on level 4, the argument constituent can again be unmarked or marked in relation to level 3, this time regulated by prosody, word order or emphasizing particles, etc. When an argument in a syntagmatic construction, such as the aforementioned arg2, is part of two different paradigms and is marked in one and unmarked in the other paradigm, this may account for the slight markedness carried by all subjects in passive clauses, even when they are unmarked subject constituents in the surface structure. The markedness value of the members of a construction in itself contribute only a very weak abstract meaning; but the constituents so marked are the receptacles of the different assigments from the pragmatic level. All sentence constituents, with or without Syntactic function assignment in the FG sense of the term, are carriers of some pragmatic or informational function according to the management of the message, whether these functions are attached to the single constituents or to the clause structure as a whole.

5. Method It is the right hand column, especially, that I want to explore by tracing some types of sentence constituents down the track to the bottom of the model. As this column is connected with the formal-syntactical dimension of language, the best method seems to be to start top-down from final realisations in a specific language, here Danish, in order to try to test the adequacy of column II. I shall start with exploring the relationship between the sentence constituents and the function of markedness at the level of pragmatic function.

6. Some linguistic data from Danish Danish happens to be a language with Syntactic ftinctions (in the FG sense), and it thus meets all three functional levels of the FG model (Dik 1989: 222). It is a language with a fairly bound word order. As for realised main clauses, it is a strong V2-language with the finite verb in the position right after PI. The sentence constituents are realised in an order which in the Danish grammatical tradition is usually represented in P. Diderichsens almost 50

Perspective, markedness, and paradigmatic relations

135

years old "sentence schema" (Diderichsen 1966: 184), in many ways a very good specification of the FG representation of word order (see Figure 1). (Dik) P2

PI

V

V

S

S

P3

(Diderichsen) Extrapos.

(1)

Content field

Fundament

Nexus field

Extra pos.

X

V

η

a

V

N

Deifor

har

ikke

overrakt

dronningen

rode roser

there-

have

not

fore

+Pres

borgmesteren mayor + Sing +Def

present +Active PaP with

queen + Sing +Def

red rose -Sing -Def

A i denne by. in town in +Sing +Def +Dem

'Therefore the mayor has not presented the queen with red roses in this town.' Figure 1. Dik's and Diderichsen's representation of word order.

For my puφose, the interesting parts of this schema are the positions X, n, N, A (n,N: nomináis; a,A: adverbials). All NPs and pronouns in η (or strictly speaking the personal pronouns as the only forms displaying a case distinction nominative/accusative — by paradigmatic extrapolation the same holds for nouns) are nominative and, because of this, are normally considered subject of the clause; all NPs/pronouns in N are accusative, and by the same line of reasoning they can all be considered objects.® It is possible to realise argument 1 of the predicate of an active predication in N, thus in the accusative. This fact renders it impossible to consider this NP/Pron the subject of the clause, if one wants to preserve a definition of the subject as distinct from that of the object, because the definition in Danish hinges on the case difference alone. There is no agreement in person and number with the finite verb (cf Dik 1989: 221), and there is no fixed positional ordering between subject and object, as they can both enter into X. This phenomenon, which in reality is the assignment of a Syntactic Function in the opposite direction from a FG assignment, a sort of demotion of the argi, has not yet received any description within FG. I venture to call argi realised in N an Object assignment, on the grounds that there is no formal distinction between this NP and other NPs which are realised arg2 or arg3 in N (ex. 2b and c),^ and that this Object assignment can be brought into

136

Lisbeth Falster Jakobsen

a paradigm with Subject assignment of arg, in relation to the same basic predicate frame (2a and b). Without an assignment it would be impossible in FG to account for the fact that argi gets another case assignment than the one normally triggered by members of A'. (2) a.

X Store ulykker de Pron+Nom 'Great misfortunes/they

V overgik + Prêt+Active befell

N manden, ham Pron+Acc the man/him.'

b.

Der overgik manden store ulykker dummy ham Pron+Acc dem Pron +Acc 'There befell the man/him great misfortunes/them.'

c.

Transaktionerne skaffede de Pron+Nom + Prêt+Act 'The transactions/ brought they

manden ham Pron+Acc the man/ him

store ulykker dem Pron+Acc great misfortunes/ them.'

The position A of the schema (see Figure 1) pertains to the adverbials of the clause, and is only interesting here because the first of the possible constituents in A will be an argument — if there is one in the predicate frame. Among the members of this set of arguments is the so-called Adpositional Object. The NP of this adpositional phrase will be accusative as well: (3)

X Derfor

V har

η a V A vi aldrig talt от deden. Nom Adpos Acc Therefore have we never talked about death+Def 'Therefore we have never talked about death.'

In n, N and A, for that matter, there is a fixed combination of position and case (accusative) for the involved NP or the anaphoric personal pronoun; the only position with a choice of case, and thus with a functional paradigm in this dimension, is the position X, which can hold a nominative constituent, the subject (in the traditional sense of the word), or an accusative constituent — the object, or an adverbial: (4) a.

X V a V Han har aldrig set Nom 'He has never seen

η mig. Acc me.'

Perspective, markedness, and paradigmatic relations

b.

c.

Χ Mig Acc Me 'Me

137

ν η a V har han aldrig set. Nom has he never seen he has never seen.'

X V η a V От mig har han aldrig talt. AdPosAcc Nom 'About me has he never spoken.'

7. Sentence constituents and the Pragmatic functions The sentence constituents must be considered the syntagmatic level w^hich, among other things, carries the Pragmatic functions; they are, as linguistic signs with an expression and a content, so to speak receptacles of features from different parameters, such as the grammatical value of the sentence constituent, GIVENNESS, position, prosodie features, etc. (These parameters are discussed by Siewierska 1988: 29-103). The parameters each form a paradigm of plus/minus or more differentiated values, with a single value from each parameter in question realised per sentence constituent. The individual values from the different parameters can combine in the sentence constituents in more than one way, of course. The different combinations form the different values of a paradigm of different markedness to be interpreted as various pragmatic functions, according to the wider context of the clause. Let us consider some unmarked realisations of clauses with a maximum of arguments, i. e. 3 (cf. Siewierska 1988: 12): (5) a.

-Marked X V a V N Borgmesteren har aldrig overrakt dronningen de rode roser Nom,+Giv Acc,+Giv Acc,+Giv'° mayor+Def has never presented with queen+Def the red roses 'The mayor has never presented the queen with red roses.'

b.

Borgmesteren har aldrig overrakt en dronning rode roser Nom,+Giv Acc,-Giv Acc,-Giv mayor+Def a queen red roses

c.

Borgmesteren har aldrig overrakt dronningen rode roser Nom,+Giv Acc,+Giv Acc,—Giv mayor+Def queen+Def red roses

138

Lisbeth Falster Jakobsen

The clauses are equally unmarked with the nominative constituent in the position η instead of X: (6) a.

-Marked X V η a V N Detfor har borgmesteren aldrig overrakt dronningen de rode roser. 'Therefore the mayor has never presented the queen with the red roses. '

b.

Derfor har borgmesteren aldrig overrakt en dronning rode roser

c.

Derfor har borgmesteren aldrig overrakt dronningen rode roser

This shows that there is a domain for the realisation of the unmarked subject: the positions X and n, and another for unmarked objects: N. If the positions of schema (1) in section 6 themselves are taken as the constants, the grammatical values (here "Nom" for subject and "Acc" for object) and the GIVENNESS values of the constituents are the variables. (For the sake of brevity I have reduced the parameters to 2 in this paper.) One way of altering the markedness of the clauses - sometimes to the point of ungrammaticality - is to switch the values of the GIVENNESS parameter, thus altering the unmarked flow of information: (7) a.

+ Marked X V a V En borgmester har aldrig overrakt Nom,-Giv a mayor

N dronningen rode roser Acc,+Giv Acc,-Giv queen+Def red roses

b.

Derfor har en borgmester aldrig overrakt dronningen rode roser Nom,—Giv Acc,+Giv Acc-Giv

c.

Borgmesteren har aldrig overrakt en dronning de rode roser Nom,+Giv Acc,-Giv Acc,+Giv mayor+Def a queen the red roses

Another way of producing + marked clauses is to switch the grammatical values of the constituents in the positions of the schema. The normal marked possibility is to let the object stand in X: (8) a.

+Marked X V Dronningen har Acc,+Giv queen+Def

η a V borgmesteren aldrig overrakt Nom,+Giv mayor+Def

N de rode roser. Acc,+Giv the red roses

Perspective, markedness, and paradigmatic relations

b.

De r0de roser har borgmesteren aldrig overrakt dronningen. Acc,+Giv Nom,+Giv Acc,+Giv the red roses mayor+Def queen+Def

c.

En dronning har borgmesteren aldrig overrakt de rede roser Acc,-Giv Nom,+Giv Acc,+Giv a queen mayor+Def the red roses

d.

Rede roser har borgmesteren aldrig overrakt dronningen. Acc,-Giv Nom,+Giv Acc,+Giv red roses mayor+Def queen+Def

139

where the realisation of two unusual features renders the constituent in X still more marked than one feature. The variation will be the same for predicates with only two arguments: (9) ä·

b.

-Marked X V V Borgmesteren har dyrket Nom,+Giv 'The mayor has grown

N róseme. Acc,+Giv the roses'

Borgmesteren har dyrket roser Nom,+Giv Acc,-Giv mayor+Def roses

+Marked X V η V ä · Róseme har borgmesteren dyrket. Acc,+Giv Nom,+Giv 'Roses the mayor has grown' b'. Roser har borgmesteren Acc,—Giv Nom,+Giv

dyrket.

A more general formulation of the rule will be: (10)

- M a r k O b j : right part of the clause +MarkObj: left part of the clause - M a r k Subj: left part of the clause

(5,6) (8) (5,6)

+Mark Subj: left part of the clause

(7)

The asymmetry of the rule stems from the fact that there are more than one way to mark a constituent: to put it inside or outside the neutral positional

140

Lisbeth Falster Jakobsen

domain (the examples in (10) above the line), or to mark it in other ways, i. e. by placing constituents with a particular GIVENNESS-value "against the grain" of the flow of information (example (10) below the line). The subject constituent, when keeping its relational value, cannot move out of its neutral positional domain." If the predicate has got only one argument, typically in presentative clauses, it is a bit more complicated. Conforming to the schema (10), this argument is realised either as a -Marked Subject in X or n, or a -Marked object in N (in this position always in combination with a dummy der in X or n): (11)

-Marked X ν V A a. Borgmesteren/han er kommet nu. Nom,+Giv mayor+Def/he is come,PaP now 'The mayor/he has come now.' X V η V b. Nu er borgmesteren/han kommet. Now is mayor+Def/he come,PaP Nom,+Giv 'The mayor/he has come now.' X c. Der

V N er kommet en borgmester/ham Acc,-Giv There is come a mayor/him Ά mayor/he has come now'. V

A nu. now.

or, with an alteration of the values of the GIVENNESS parameter, the Giv.argument is +Marked as a subject, and the +Giv.argument is +Marked as an object: (12)

+ Marked X a. En borgmester/han

V V A er kommet nu.

Nom,-GÌ ν

A

mayor/he

Χ ν η b. Derfor er en borgmester/han Nom,-Giv a mayor/he

V A kommet nu.

Perspective, markedness, and paradigmatic relations

141

Χ ν V Ν Α C. Der er kommet borgmesteren/ham nu. dummy Acc,+Giv mayor+Def or, conforming to the schema of (10), the argument can be + Marked as an object in X: (13)

X V η V a. En borgmester/ham er der kommet. Acc,-Giv dummy b. IBorgmesteren/ham er der kommet. Acc,+Giv dummy mayor+Def

Possible positional realisations of argi with different Markedness-values, according to variations of the values of the parameters GIVENNESS and grammatical relation are as follows: (14)

-Mark +Mark +Mark +Mark

-Giv +Giv -Giv +Giv

-Mark +Mark

+Giv Subj: - G i v Subj:

Obj: Obj: Obj: Obj:

right part of the clause rigth part of the clause left part of the clause ?left part of the clause

(11c) (12c) (13a) (13b)

left part of the clause left part of the clause

(11 a,b) (12a,b)

A case of extreme markedness in X is an adpositional object (or adverbial argument):'^ (15)

++Marked X V η a a. От doden taler vi aldrig. Acc,+Giv Nom About death talk we never 'About death we never talk.'

but the markedness can be reduced to that of a fronted +Giv.object by preposition stranding, as in (15b). (15)

+ Marked X ν b. Deden taler Acc,+Giv Death talk

η a A vi aldrig от. Nom we never about

142

Lisbeth Falster Jakobsen

Conclusion: the mere relational value of the sentence constituent is at the same time a semantic feature'"' of the constituent, acquired at this near-surface level: together with the positions of the constituent, the relational value forms a paradigm of marked and unmarked combinations. Therefore the sentence constituents, as items in their own grammatical right and not conditioned merely by underlying functional factors, have to be taken into consideration for their contribution to the desired pragmatic value of the clause. A demonstration of which combinations of values from the different parameters express which actual pragmatic values of the clause in Danish is outside the scope of this paper — but surely the same combinations of features from the parameters may give different pragmatic results in different languages.'^

8. Subject Assigmnent Subject assignment makes possible a shift of perspective between arguments as compared to the perspective which the basic predicate has given its arguments in the frame (Dik 1989: 215). In other words, the basic frame in the Fund already represents some - though not strictly linear - ordering of the arguments (1989: 68-70). The predicate frame with its ordering of arguments is then the prerequisite for a possible reshuffling of the arguments on the level of syntactic fiinctions, whereby another perspective on the presentation of the arguments in the SoA is introduced. This means that the reperspectivization of an SoA, brought about by assigning the subject function to another argument than an argi, can only be understood on the paradigmatic background of the basic order. There are several grammatical ways of making a reperspectivization in Danish: by passive constructions and by various extended verbal constructions. 8.1. Passive constructions Danish follows a rule that says that, in principle, arguments higher than argi that are Goal or Recipient and have an accusative in their expression potential may undergo Subject assignment in passive predications: X V a V (16) a. Dronningen er ikke blevet overrakt Nom,+Giv The queen is not been presented with 'The queen has not been presented with red

N rederoser. Acc,-Giv red roses roses.'

Perspective, markedness, and paradigmatic relations

143

b. De rede roser er ikke blevet overrakt dronningen. Nom,+Giv Acc,+Giv c. Rode roser er ikke blevet overrakt dronningen. Nom,-Giv Acc,+Giv d. IRode roser er ikke blevet overrakt en dronning. Nom,—Giv Acc,—Giv e. Der er ikke blevet overrakt dronningen rede roser dummy Acc,+Giv Acc,-Giv ...etc. The grades of markedness conform to the schema (10). It is even possible to have a passive construction of a 3-place predicate without any subject assignment, thus with two objects (still conforming to (10)), as in (16e). And it is possible to make the NP of the adpositional phrase subject in a construction with preposition stranding: (17)

X ν V A Han er blevet kostet med som barn. Nom, PersPron AdPos He is get,PaP order,PaP about as child 'He has been ordered about when a child.'

Thus the potentiality of an argument to undergo subject assignment in positions X or η cannot serve as a distinction between the two constituents in N - they both have this potentiality. In other words: passivization cannot be used as an argument for construing an abstract order of the arguments in relation to the predicate. According to FG, the order of the arguments will be: (18)

overrœkkey (xi:(anim)(xi))Ag (x2)go (хз:(ап1т)(хз))кес

Accordingly, all my examples with the Recipient constituent to the left of the Goal constituent, will be the result of assignment of Object to the Recipient constituent. Thus (19a) is a case of reperspectivization from the basic perspective of the predicate (19b): (19)

X ν V N A a. Borgmesteren har overrakt dronningen rede roser Nom,Ag Acc,Rec Acc,Go b. Borgmesteren har overrakt rede roser til dronningen. Non,Ag Acc,Go AdPosAcc,Rec

There are reasons why I don't find this interpretation valid for Danish - one is that the basic perspective should be the most unmarked as well, and in the

144

Lisbeth Falster Jakobsen

case of (19b) this is in fact a more marked realisation than the one given in (19a). But I have gone into it here because of the problem that (17) raises in this connection: although cases like (17) are somewhat rare as tokens, the type does exist. The markedness thus lies in the token frequency of the whole construction, and not in the subject constituent han. The han of the clause fulfils two of the stipulated coding properties for subjects in English in FG (Dik 1989: 221): -

occurrence in positions reserved for subjects

-

nominative, if pronominal

whereas it is questionable what to do with the third property: -

no prepositional marking.

If one considers the two examples in (20) (cf (17) for a translation): (20)

X ν V A a. Han er blevet kostet med som barn. Nom,PersPron X V η V A b. Ham er der blevet kostet med som barn. Acc,PersPron dummy

there can be no doubt that han in (20) is the unmarked subject of the clause. This again means that we can have Subject assignment of the NP part of an adpositional phrase, giving it a new word order domain with new pragmatic possibilities - but leaving the preposition in its old position, and thus leaving behind a trace of the construction of the basic ordering and the basic grammatical value connected with the position. This being possible, it is remarkable that this type of Subject assignment does not seem to apply to 3-place predicates such as at overrcekke ('to present with'): (21)

X ν V N A a. *Hun er blevet overrakt rede roser til. Nom,PersPron AdPos X V η V N A b. Hende er der blevet overrakt rode roser til. Acc,PersPron dummy AdPos X V V N c. Hun er blevet overrakt rode roser Nom,PersPron

Perspective, markedness, and paradigmatic relations

145

This raises the question of whether the claim that the model should be able to trigger an expression in one go is correct: the non-grammaticality of (21a) compared with (21c) might be an indication that either (22)

overrcekkey (xi:(anim)(xi))Ag (х2:

A2 Go Ree Ben Instr Loc

>

A3 Ree Ben Instr Loc Dir

>

...

(Siewierska 1991: 110)

Ag Pos Fo Proc Zero

Figure 1. Semantic function hierachies.

The semantic ftmction hierarchy could be viewed as a "topicworthiness" hierarchy (cf. Givón's "topicality" or "topic hierarchy", Givón 1984a: 151, 1984b: 137). This would imply that A':Ag is a default primary topic, A^:Go a default secondary topic, and possibly that A^:Rec a default tertiary topic. The assigned topic values in a given judgment are then actual topics. The unmarked mappings^ between semantic arguments and actual topics are thus the following: A':Ag

A2:GO

A3:Rec

Tl°

T2°

T3°

We suggest the following notation that distinguishes between default (or merely potential) and actual topicality: the default topic is marked as (T), and

Syntactic functions, topics, and grammatical relations

161

the actual topic as [Т]. The table of unmarked mappings could then - trivially — be recast as: (ΤΓ)

(T2°)

(T3°)

[Tl°]

[T2°]

[T3°]

If A'(T1°) in a given language always shows up as A'[T1°] in actual judgments, then we will maintain that this language has no grammatical relation subject. If, on the other hand, A'(T1°) - besides the occurrence as A'[T1°], i. e. as the unmarked mapping — also is actualized as A' [Ad] accompanied by an (A^(T2°) actualized as) A^[T1°], then we will say that a grammatical relation subject exists in this language. So far, our analysis resembles FG's: a [Tl°] corresponds to FG's Sb, as a [T2°] corresponds to FG's Ob. The difference between the two proposals viz. our recognition of grammatical relations as different from actual topics - will be dealt with in section 1.2. We have said that [Tl°] may become distinctive, may become a syntactic function, i.e. "pragmatic pivot" (cf Foley - Van Valin 1984: 305 ff.), if it can be assigned to alternative argument ftmctions. It is possible to analyse an A' as distinctive if its various instantiations (by semantic relations) behave in identical fashion in certain respects. For example, A* — irrespective of its specific semantic value - is coded by the preposition by in English if it is an adjunct in the passive voice; and it may trigger reflexivization, even as an adjunct in the passive (cf Siewierska 1991: 80). A' under such conditions may be conceived of as a "semantic pivot". Even though these two types of distinctiveness are important, we shall not go into any further detail in this paper. Before we close this section, we have to stress that the semantic function hierarchy referred to earlier does not appear to be a linguistic universal. Rather, the concept of a semantic ftmction hierarchy is a typological variable. Thus, A', A^, A^, and lower degrees of semantic roles have different unmarked role values in different languages (or language types). 1.2. Grammatical relations in FG - a proposal Our next suggestion is to call A'(T1°)[T1°] a prototypical subject or the subject prototype® - potential and actual topic coincide. Likewise, A^(T2°)[T2°] is the direct object prototype, and A2(T3°)[T3°] is the indirect object prototype.

162

Hartmut Haberland-Ole Nedergaard Thomsen

However, we shall only speak about a subject and direct, or indirect, object prototype if nonprototypical instances occur in a given language. That is, if a language has a passive (as a perspectival variant of a SoA), implying that it has A2[T1°], dative subjects (Shibatani 1988a: 117), or other instances of marked (nonprototypical) subjects, then it makes sense to talk about subject as a grammatical relation. Shibatani (1988a) has shown how it is possible to argue that the nonprototypical cases (the extensions of the prototype) are related to this prototype. For Philippine languages, he considers two extensions of the subject prototype, viz. cases in which potential and actual topic do not coincide. The first extension consists of actual topics [Tl°] which are not actualisations of default topics (Tl°). These nonprototypical subjects share ten properties with prototypical subjects. But there is another extension, too, viz. the case of A'(Tl°)[Ti°] for i > 1. Here we have a default Topic 1° which does not actually occur as Topic 1°, but which still is A'. These nonprototypical subjects still share five properties with prototypical ones. The extension A'(T1°), it should be noted, is not demoted to A'[Ad] in Philippine languages, but remains inside the pragmatic core as A ' [ T i ° ] , i > l , and behaves like a prototypical subject in some respects: these actors trigger reflexivization and sometimes also deletion in coordinated clauses. Philippine languages are an important touchstone for grammatical theory, and Siewierska (1991: 82-86, 91-93) has demonstrated that standard FG is not yet capable of giving an adequate (explanatory) analysis of their elementary syntactic structures. We believe that the introduction of prototype categories in FG would overcome (some of these) problems.

1.3. The thetic and the categorical judgment In the following, we want to pursue a bit further the distinction between thetic and categorical judgments introduced above. Take the following Danish question — answer sequences: (13) a. A: Hvad sker der i Litauen? what happens Δ in Lithuania 'What's happening in Lithuania?' B: Russerne Javer è/o'kade. russians.DEF do blockade 'The Russians have set up a blockade.'

Syntactic functions, topics, and grammatical relations

163

b. A: Hvad sker der i Litauen? В: Der bliver Javet è/o'kade. Δ is.being done blockade 'There is a blockade being set up.' The predicate frame underlying both (B)-sentences is: (14)

olave'blo'kade (A':Ag(Tl°))

which is a noun incorporation predicate (a kind of derived predicate, cf Nedergaard Thomsen 1992). It has a prirnary argument which is by default the (primary) topic. In the first pair this default topic is actualized, whereas in the second pair it is not. The judgment established in the (B)-sentence of (13a) is therefore categorical, that in the (B)-sentence of (13b), thetic. The thetic judgment is, to repeat, a marked judgment, lacking an actual primary internal topic, viz. a [Т1°]. Before proceeding, we have to go into more detail with the analysis of a judgment. A judgment consists of a comment applied to zero, one, or more internal topics, and possibly also adjuncts. The comment is the head of a judgment: when a comment is applied to a topic, the result is a projection of the comment. The maximal projection of a comment is the judgment. The zero projection of the comment may contain an incorporated constituent (In), as in the examples (13a, b) above. [In] has the lowest, zero, degree of actual topicality. Now, a thetic judgment, as stated earlier, consists of a maximal projection of a comment which does not "command" a [Tl°], i. e. the thetic judgment is not primarily about a topic (or first order referent), but is about the denoted state of affairs as a whole, i. e. a second-order entity. (Sasse (1987) speaks about "recognition" of a state of affairs.) In some languages, the thetic judgment is coded by a separate sentence type, e.g. VS order (as against categorical SV). In other (subject-prominent) languages, the sentence type coding the categorical judgment is taken as a structural prototype. Thus, the structural slot of [Tl°] is empty in thetic judgments and has to be filled by a "dummy subject" (B). This - strongly nonprototypical - subject does, of course, not denote the "posed" SoA. But it denotes the space in which the SoA occurs (or obtains) — since one of the main functions of a thetic judgment is to package an existential state of affairs.^ The dummy subject is nonprototypical, but it is an actual [Tl°], as evidenced by English raisings like I believe there to be unicorns in the garden. - raising in EngHsh is only possible of embedded [Tl°]s (cf. I believe Max to be innocent). The nonprototypicality of this dummy consists in the fact

164

Hartmut Haberland—Ole Nedergaard Thomsen

that it is not a primary argument: in a thetic judgment, the primary argument (if there is one) is a default (T2°). And, as we have seen in (13b), a thetic judgment may have a secondary argument which is a default (T2°), there occurring as [In] (i. e. incoφoгated constituent). So nothing is left but to call the dummy subject a zero argument (A°). This is default (Tl°).^° We may conclude this section by giving a tree structure of a judgment (see Figure 2). judgment / Tl°

\ comment!" / \ T2° commenta" / \ Adjunct commentN" / \ Incorporated constituent

commentN+l°

Figure 2.

2. Marked mappings This section deals with the pragmatically motivated divergences from the unmarked mappings, where unmarked refers to active, ergative, prepositional dative, nonraising, and nonclefting." First of all we would like to point out that there is a parallel phenomenon to the ergative construction as far as objects are concerned, viz. the dechticaetiative (this is a term from Blansitt 1984; the objects involved are called "primary" and "secondary objects" in Dryer 1986). In the dechticaetiative construction, recipient or benefactive, or even more peripheral terms, appear as unmarked A} and [T2°]'^ (or "primary object" in Dryer's terminology). And there is possibly a parallel to the antipassive, viz. the antidative (cf Dryer 1986): the marked mapping of A^:Go on [T2°] (i. e. a "secondary object"). English has both a dechticaetiative (15a) and (15b), and an antidative (15c); (15) a. That remark cost you your job/*cost your job to you. b. He presented her with a book. c. He presented a book to her.

Syntactic functions, topics, and grammatical relations

165

(15a) is actually a dative shifted form which has no unshifted counterpart." (15b) is a dechticaetiative with the goal as [Ad], They will be given the following structural descriptions: costv (15) a', (obligatory dative shift) presentv b'. (dechticaetiative) c'. (antidative)

(A':Ag) (A^iGo) Tl° T3°

(A^rBen) T2° (unmarked)

(A':Ag) (A^iRec) (A^rGo) Tl° T2° Ad Tl° Ad T2°

(unmarked) (marked)

2.1. Passive A prototypical monotransitive sentence in accusative languages has an underlying orientation as is shown in (16a) (cf. Dik 1989: 212): (16) a. A^:Ag^f¡^A2:Go An actual example would be (16b): (16) b. Fred—»kicky—•the mule A passive sentence is analyzed in Dik (1989: 213) by assigning a "pointer" to the secondary argument: (16) c. Fred—•kicky—»the mule [1] In our approach we would add notation for default topicality to (16a), which is reformulated as (16a'). (16) a'. A':Ag(Tl°)^fi^A2:Go(T2°) In an actual active clause, default topics are actualized as actual topics, which gives (17a): (17) a. A':Ag[Tl°]—fi—A2:GO[T2°] The corresponding passive shows a marked assignment of actual topics: (17) b. A':Ag[Ad]—fi—A2:GO[T1°] Still we think that the difference between (17a) and (17b) is not enough to bring out the difference between active and passive. What is changed is not only the status of the arguments involved, but also the perspective on the

166

Hartmut Haberland-Ole Nedergaard Thomsen

causal relation (of "telicity") inherent in the Accomplishment denoted by the fj. This can be captured by the following notations: a. (ACT fj! predv [Accomplishment](fi)) (unmarked) b. (PASS fji predv [Accomplishment](fi)) (marked) 2.2. Antipassive The cardinal (syntactic) ergative construction (cf. Dixon 1972 on Dyirbal; Shibatani 1988b) is analyzed by us in the following way: (18) a. A>:Go[Tl°Hfi—A2:Ag[T2°] assuming an underlying (18a'): (18) a'. Ai:Go(Tl°)—fi—A2:Ag(T2°) That is, in ergative languages the semantic function hierarchy is: A':Go < a^:Ag < ... The antipassive is accordingly: (18) b. A':Go[AdHfi-A2:Ag[Tl°] In syntactically ergative languages the unmarked perspective on the causal relation is the ergative (corresponding to the passive perspective in accusative languages), the marked perspective is the antipassive (corresponding to the active). Notationally, they appear as a. (ERG fi: predy [Accomplishment](fi)) b. (ANTIPASS fi: predv [Accomplishment](fi)) Thus, the antipassive is not always just a means of reducing the number of arguments. (Nedergaard Thomsen (forthcoming) gives a somewhat fiiller discussion of Dyirbal syntactic ergativity and embedding.) 2.3. Dative shift and antidative Whereas passive and antipassive concern marked mappings of the primary topic, dative and antidative are the marked mappings of the secondary topic on the recipient/beneficient, or the goal, respectively. Parallel to the analyses given above we shall propose the following verbal operators for the secondary actantial perspectives in dative construction types: a. accusative: (ACC fi: predv [Accomplishment](fi)) (unmarked) b. dative: (DAT fi: predv [Accomplishment](fi)) (marked)

Syntactic functions, topics, and grammatical relations

167

and in dechticaetiative languages: c. dechticaetiative: (DECH f^ predy [Accomplishment](fj)) (unmarked) d. antidative: (ANTID f;: predy [Accomplishment](f;)) (marked) In dative languages, the relevant subpart of the semantic function hierarchy is accordingly: A^:Go < a^:Rec, whereas it is A^:Rec < a^:Go in dechticaetiative languages. In the following we shall investigate the interaction between dative shift and active, passive, and thetic. 2.3.1. Dative shift in the active Danish is a language with dative shift. Note that A^:Go(T2°) can be syntactically incorporated in a dative shift construction, as in (19b) and (19d): (19) a. Han ^gav hende ^begerne. he gave her books.DEF 'He gave her the books.' b. Han ogav hende ^boger. (incorporation) he gave her books 'He gave her books.' c. Han 'gov ^begerne til hende. He gave books.DEF to her 'He gave the books to her.' d. Han ogav ^boger til hende. (incorporation) He gave books to her 'He gave books to her.' To these are assigned the following descriptions: (19) (active) givey (Ai:Ag(Tl°)) a', (dative shift) Tl° b'. (dative shift) Tl° c'. (accusative) Tl° d'. (accusative) Tl° (In = incoφoгated constituent)

(A2:Go(T2°)) T3° In T2° In

(A3:Rec(Ad)) T2° T2° Ad Ad

168

Hartmut Haberland-Ole Nedergaard Thomsen

2.3.2. Dative shift in the passive The above examples have passive counterparts: (20) a. Hun blev she

^givet ^bogerne.

PASS.PAST given books.DEF

'She was given the books.' b. Hun blev she

ogivet ^beger. (incoφoration)

PASS.PAST given books

'She was given books.' c. Bogeme

blev

Ogivet til hende.

books.DEF PASS.PAST given to her 'The books were given to her.' d. Bogerne

blev

Ogivet hende.

books.DEF PASS.PAST given her 'The books were given her.'

(20) (passive) givey

(A':Ag(Tl°))

(A2:Go(T2°))

(A3:Rec(Ad))

a',

(dative shift)

-

T2°

ΤΓ

b'.

(dative shift)

-

In

Tl°

c'.

(accusative)

-

Tl°

Ad

d'.

(dative shift)

-

Tl°

T2°

2.3.3. Dative shift in thetic judgments The thetic judgment excludes the presence of an actual primary internal topic. Compare: (21) a. Der blev В

Ogivet hende nogle

PASS.PAST given her

beger.

some books

'She was given some books.' b. Der blev В

ogivet hende boger. (incoφoration)

PASS.PAST given her

books

'She was given books.' c. Der blev В

Ogivet nogle beger

til hende.

PASS.PAST given some books to her

'Some books were given to her.' d. Der blev В

ogivet boger

til hende. (incoφoration)

PASS.PAST given books to her

'Books were given to her.'

Syntactic functions, topics, and grammatical relations

169

which are assigned the following descriptions:

(21) (thetic. passive) a'. b'. c'. d'.

(A3:Rec(Ad))

givev (dative shift) (dative shift) (accusative) (accusative)

T3° In T2° In

T2° T2° Ad Ad

In these sentences, the passive is a prerequisite of the thetic judgment. Only if A ' : A g ( T r ) is demoted to A':Ag[Ad], a dummy subject can be inserted in the preverbal position to make the judgment thetic. 2.3.4. Antidative Danish has also an antidative construction, i. e. a marked construction presupposing the ranking of, prototypically, A^:Rec over A^:Goal. We will not give a prototypical example here (parallel to the English present somebody with something), but introduce the much-discussed dechticaetiative paint-wall constructions: while the accusative verbs profile the moving entity or figure, the paint-wall type verbs profile the ground in the dechticaetiative construction. male 'to paint' clearly shows ground profiling: (22) a. Han malede vœggen med den rode maling. (decht.) he painted wall.DEF with the red.DEF paint 'He painted the wall with red paint.' b. Han malede den rode maling pâ vœggen. (antidat.) he painted the red.DEF paint on wall.DEF 'He painted red paint on the wall.' c. den malede vœg the painted wall 'the painted wall' d. *den malede rode maling the painted red paint *'the painted red paint' A similar case i s f y l d e 'to fill':*'' (23) a. Han fyldte vasen med vandet. (dechticaetiative) he filled vase.DEF with water 'He filled the vase with the water.'

170

Hartmut Haberland-Ole Nedergaard Thomsen b. Han fyldte vandet i vasen. (antidative) he filled water.DEF in vase.DEF 'He filled the water in the vase.' c. den fyldte vase the filled vase 'the filled vase' d. *det fyldte vand the filled water *'the filled water'

But other verbs like sende 'to send', overdrage 'to transfer' exhibit figure profiling: (24) a. Han sendte Maria bogen. (dative shift) he sent Maria book.DEF 'He sent Maria the book.' b. Han sendte bogen til Maria, (accusative) he sent book.DEF to Maria 'He sent the book to Maria.' c. den sendte bog the sent book 'the sent book.' d. *den sendte Maria the sent Maria *'the sent Maria.' The dechticaetiative and antidative constructions (22a) and (22b) are analyzed in the following way: (22) (active) a', b'.

malev

(A':Ag(Tl°))

(A2:Loc(T2°)) (A3:Go(Ad))

(dechticaetiative) (antidative)

ТГ ΤГ

T2° Ad

Ad Т2°

Both the dechticaetiative and the antidative constructions may occur in the passive. It should not go unmentioned, of course, that standard FG does not analyze the often discussed paint-wall construction as a case of reperspectivization.

Syntactic functions, topics, and grammatical relations

171

FG claims that the variants do not denote the same SoA. And as a result of that, the marked variant is conceived of as the output of a predicate formation rule. We believe that sameness of SoA is in general impossible as a criterion for recognizing perspectival variants. Even ordinary dative shifts have been analyzed as denoting different SoAs from their unmarked counterparts (cf Gropen - Pinker - Hollander 1989; Langacker 1990). This is a huge topic which deserves to be treated in a separate paper.

2.4. Raising We have mentioned raising as a clear case of a construction type where the separation of predication and judgment can be ascertained. Here we go into more detail with raising phenomena. First, consider the following nonraising examples: (25) a. John believed that Bill had kicked the mule. b. It was believed by John that Bill had kicked the mule. which are analyzed: believev (25) a', (active) b'. (passive)

(A'(T1°)) (A2(T2°) =kickv (Tl°,-) (T2°,-) (active) (Ad,-) (Tl°,-) (active)

(A'(T1°)) (-,T1°) (-,T1°)

(A2(T2°))) (-,T2°) (-,T2°)

(where the parentheses mean ([Tjnatrix],[Tsub])) Their raising counterparts are: (25) c. John believed Bill to have kicked the mule. d. Bill was believed by John to have kicked the mule. analyzed as: behevev (25) c'. (active) d'. (passive)

(А'(ТГ)) (A2(T2°) =kickv (Tl°,-) (active) (Ad,-) (active)

(Α·(Τ1°)) (T2°,T1°) (T1°,T1°)

(A2(T2°))) (-,T2°) (-,T2°)

Raising and passive may also be combined: (25) e. John believed the mule to have been kicked by Bill. f. The mule was believed by John to have been kicked by Bill.

172

Hartmut Haberland—Ole Nedergaard Thomsen

These examples are analyzed as: believev (25) e'. (active) f . (passive)

(A'(T1°)) (A2(T2°)= kicky (A*(T1°)) (Tl°,-) (passive) ( - , A d ) (Ad,-) (passive) ( - , A d )

(A2(T2°))) (T2°,T1°) (T1°,T1°)

Common for all the above examples is that only embedded [Tl°]'s may function as matrix [Tl°]'s. Also relevant in this connection is the phenomenon of "sentence-intertwining" (cf. de Groot 1981a,b). In Hungarian, even points of time satellites may be crossreferenced on the matrix verb as a [T2°], like in (26) (from de Groot 1981b: 57): (26)

Elmér holnap akarja hogy vezessek. Elmer tomorrow wants.DEF that drive.I [+def] [T2°] 'Elmer wants that I drive tomorrow.'

2.5. Clefting Cleft constructions are said to denote the same SoA as their noncleft counterparts (cf Foley - Van Valin 1985: 359). There are two identificational subtypes in English: (a)

WH-clefts

(b)

iT-clefts

The following exemplify those types (from Foley - Van Valin 1985: 358): (27) a. What Ron ate was a sandwich. b. It was a sandwich that Ron ate. They both consist of a focus and a "background" (the relative clause). The formal difference between them is that WH-clefts consist of a background followed by a focus, whereas iT-clefts have their focus precede their background. The focus is a replica of an argument or a satellite within the relative clause background. In the relative clause the term is a bound variable (Rxj). In the focus constituent the variable has a term head (ilxj: sandwichN(xj))Foc· In terms of the thetic/categorical distinction, (27a) would be categorical (with the relative clause as [ΤΓ]), whereas (27b) could be analyzed as thetic (with the focus and background constituting an NP) (see Figure 3).

Syntactic functions, topics, and grammatical relations Categorical

173

Thetic

commenti" T2°Foc

commenta"

Foc a sandwich

that Ron ate

was

A^: Id'ed

Id'er

Id'er

Id'ed

Figure 3.

3. Further perspectives: satellites and topicality It goes without saying that there still are many problems in the field of relational typology. We will mention some of them in this tentative section. First, there is the problem of topic and so-called topicalization. We don't consider topic a structural position of the clause. Although preverbal or sentence-initial positions often are topic markers, not all fi-onted constituents have to be topics. Thus, we do not analyze the secondary argument this door in (28b) ( = (25b) in Dik 1989: 220) as a primary internal topic: (28) a. John cannot open this door. ( = 25a) b. This door John cannot open. ( = 25b) c. This door cannot be opened by John. If this door in (28b) were a primary topic, it would be an A^:Go(T2°)[Tl°], something which we would expect in a passive like (28c). The fronting of this door means, of course, that it is pragmatically foregrounded or highlighted, but we analyse this prominence as a sign of focality (in FG focus is "the relatively most salient information within a given setting"): this door in (28b) is [T2°] and focus. Likewise, the tertiary (beneficiary) argument for Mary in (29a) is not topicalized: (29) a. For Mary, Fred bought a rose. but is a focus.

174

Hartmut Haberland-Ole Nedergaard Thomsen

In contrast, we have [Τ1°] in the passive sentence (29b). In a left-dislocated sentence like (29c), we get a [Theme] (external topic) {Mary) coreferential with an internal secondary topic (the beneficiary, her): (29) b. Mary was bought a rose by Fred. c. Mary, Fred bought her a rose. (29) may then be analyzed as: buyv (29) a', (active, accusative) b'. (passive, dative shift) c'. (active, dative shift)

(A':Ag(Tl°)) Tl° Ad Tl°

(A2:Go(T2°)) T2° T2° T3°

(A3:Ben(Ad)) AdFoc Tl° T2°

Thus when A' is [Tl°] and A^ is [T2°], there is an option whether the beneficiary should be a non-focal or a focal (29a) [Ad]. In Danish sentence-intertwinings a focus ([Foe]) may be assigned to an adjunct satellite: (30) a. Hvornâr regnede vi ud at de ките komme? when figured we out COMP they could come [Foe] 'When did we figure out they could come?' which is analyzed as: (30) a'. otegne-'ud (A'(T1°)) (active) ( T l ° , - )

(A2(T2°) = kommev ( А ^ Т Г ) ) (T2°,-) (active) ( - , T 1 ° )

(Sat:Temp(Ad)) (Foc,Ad)

This temporal satellite may be clefted (focalized) too: (30) b. Hvornâr var det at vi regnede ud when was it COMP we figured out [Foe] at de kunne komme? COMP they could come 'When was it we figured out that they could come?' Moreover, the borderline between topics and adjuncts is hard to draw, for although it may (seem to) be easy to find a cut-off point on the semantic ftinction hierarchy for different pragmatically motivated perspectival ftinction

Syntactic fimctions, topics, and grammatical relations

175

assignments in a given language (so that what is to the left of the cut-off point can figure as a default topic of some degree, that to the right an adjunct), still several semantic roles to the right of the cut-off point may become actual internal topics in a given sentence. There are two cases which need discussion. The first is the case where some satellite is promoted to a topical position which at the same time is, for instance, subject. The second case concerns languages where some means of formal marking of topics can apply both to arguments and satellites: in the positional scheme of the Danish main clause (a V/2 structure), there is a preverb position used for [Tl°], theme, and focus constituents, as well as given information; and if a satellite occurs in this position, it is difficult to determine which of the latter three pragmatic fimctions is involved. As an example of the first case, consider so-called "pseudopassives" with prepositional stranding, which have been discussed, among others, by Riddle - Scheintuch (1983): (31) a. This bed hasn ì been slept in by anyone, (local) b. This ball has never been played (*this game) with, (instram.) Such sentences (where satellite roles are integrated topically into the syntactic-pragmatic structure) might, within standard FG, be handled as instances of predicate formation (whereby a former satellite becomes primary argument). But we believe this is an ad hoc approach. A similar case of satellites which become internal topics [Tl°] are "topicmanner constructions" (cf Dixon 1976) and "agentivalizations" (cf. Schlesinger 1979, 1989) in English: (32) a. This pan fries well, (instrumental) b. This tent sleeps five persons, (local) c. This spray kills cockroaches instantly, (instrumental) These examples (where the promoted satellite actually becomes a subject) show that it is not possible to avoid the problem by claiming that satellites always become external topics (themes) when they become topics. The above subjects can hardly be themes: they are not predication-extemal. Another problem with these examples is that besides the satellite becoming actual primary topic, do they also become primary argument? Schlesinger's (1979, 1989) answer is in the affirmative: there is a semantic assimilation between the satellite and the semantic role agent. This is evidence of the existence of a grammatical relation subject, i. e. agent-cum-topic, assigned to the satellite function. This implies that not only topicality values could be assigned to terms, but also grammatical relations.

176

Hartmut Haberland-Ole Nedergaard Thomsen

The last problem to mention is the problem of demarcation between external topic (ET), and focal and nonfocal adjuncts. Consider these examples: (33) a. Igár, der var vejret godt. yesterday there was weather.DEF good [ET] 'As for yesterday, the weather was good.' Ά.'.*Da var vejret godt. then was weather.DEF good [AdFoc] 'Then the weather was good' b. Igár var vejret godt. yesterday was weather.DEF good [AdFoc] 'Yesterday the weather was good.' c. Vejret var godt igár weather.DEF was good yesterday [Ad] 'The weather was good yesterday.' The problem is whether igár in (33a) actually is a theme. (33a') shows that even though this constituent is coreferential with a demonstrative der (according to writing a local adverb but in the spoken language representing a neutralization of the distinction local vs. temporal {da)), this demonstrative does not have the possibility of functioning as a focal (fronted) adjunct of its own. This rounds off the catalogue of problems for the FG approach to grammatical relations, problems we hope will be solved in near future research.

4. Conclusion In this paper we hope to have shown that the (perspectival) syntactic fiinctions of FG should be viewed as functions of aboutness and lower degrees of topicality. We have introduced the distinction between default topicality and actual topicality, the former manifested in semantic function hierarchies (i. e. topicworthiness hierarchies) and in the predicate frames; the latter in actual judgment perspectives.

Syntactic functions, topics, and grammatical relations

177

Topicality functions may be syntacticized as pragmatic pivots. Semantic (transitivity) functions may be syntacticized as semantic pivots. The intersection between topicality and transitivity is Grammatical Relations. Thus, the subject prototype is primary argument-cum-primary topic, the direct object prototype secondary argument-cum-secondary topic, and the indirect object prototype tertiary argument-cum-tertiary topic. Grammatical relations comprise, in addition to their prototypes, also extensions, nonprototypical members, which behave like prototypes in some ways but differ from them in others. Grammatical structure has two core articulations, viz. a semantic articulation as predication; and a topicality articulation as judgment. These two articulations can vary independently of each other (as is seen, i. a., in raising phenomena). Another "pragmatic" articulation is in terms of information structure (Given/New and other illocutionarily dependent structures). Grammar is a multistrata! structure.

Notes 1 We have had two trial runs of some of the ideas presented her: one on May 2, 1990 for the Functional Grammar study group of Copenhagen University and one at the Fourth Functional Grammar Conference in Copenhagen on June 26, 1990. On both occasions, we received comments and criticism, from Peter Harder, Lisbeth Falster Jakobsen, J. Lachlan Mackenzie, Johan van der Auwera, MachteU Bolkestein, and Ole Togeby, which we gratefijlly acknowledge. All shortcomings are, as always, our own responsibility. 2 This standard example is not unproblematic, cf Borg and Comrie's warning (1984: 123) that verbs like to give in many languages are atypical ditransitive verbs. 3 Even discourse topics are not necessarily repeated. The following ехсеф1 from a Modem Greek novel (Maria lordanidou's Loxandra) illustrates this very well. We have marked what we analyze as discourse topics in the original text by (...). (We have tried to keep the English as close to the original as possible.) None of these topics is held over more than one clause, (i) (I konstandinúpoli) ekini tin epoM itan érta xarmáni αρά ôjaforés polities, proástia ke xorjá, skorpizména páno sta parália tiz mikrás asias ke tis evrópis. ke (i káOe politía, to káOe proàstio, to káOe xorjó) i\e to dopikó tu xaraktira, ta ίθί ke ta éOima tu plidizmü pu plionopsifúse. {Tin evwpaikí όχθί to vospóru) tin katikúsan perisótero élines ke jeniká evropéi - ifo méja rèma, to bujúk-deré, ta Oerapjá, óla ekina ta proástia) Bmizane evrópi. {i asiatiki óxffl) itane anatoli. (Constantinople), in those days, was a blend of different towns, suburbs and villages, spread upon the shores of Asia Minor and Europe. And (each town, each suburb, each village) had its local character, the ways and the customs of the population that was in the majority. (The European shore) was mostly inhabited by Greeks and generally Europeans - (Mega Rema, Büjük-Dere, Therapeia, all these suburbs) reminded of Europe. (The Asian shore) was the East.

178

Hartmut H a b e r l a n d - O l e Nedergaard Thomsen

4 In the following reanalyses of (7) and (8) we have analysed Dik's topic and focus constituents in terms of given and new. (7') A: Question: Interrogative E (trueX: [Past pos e: [(fj;: Ьарреп(4)) (what е')ме« (drnxj! demonstratorN(Xj))] (e)] (X)) B: Answer: Assertive E (trueX: [Past pos e: [(ί^: removev(fk))New (dlXi: policeN(xi))New (dix,: platformN(Xk))New (Axj)] (e)] (X)) (8') A: Question: Interrogative E

5 6 7 8 9 10

(trueX: [Past pos e: [(f^: rernovev(fk)) (who Xi)New (dmxj: demonstratorN(Xj))] (e)] (X)) B: Answer: Assertive E (trueX: [Past pos e: [(Afk,Xj) (dlxj: роИсеы(х;))ме»] (e)] (X)) All constituents not marked as new are given. We are talking about clauses here in a preliminary fashion; it will later on become clear that we would prefer to talk about judgments. Sasse's "thetic statement" corresponds to "sentence focus" in Lambrecht (1987) and Van Valin (1990); the corresponding term for "categorical judgment" is "predicate focus". There are reasons to assume that not even this scale of unmarked mappings is universal; at most it represents the most common, so to say unmarked variant of such a scale. The "Actor-Topic" in Philippine languages, cf. Shibatani (1988a). The dummy subject pronoun in thetic sentences, originally a distal (Danish „der 'there') or proximal (Danish Jier 'here') local deictic. FG analyzes perspectival variants by assigning perspectival functions to semantic roles. Now, it seems that the two different judgment types determine the valency of the verbal predicate in the predication they are mapped onto: if the thetic judgment is assigned to transitive predications in Danish, there is no primary argument, and the default topicality value of that argument is inherited by the dummy argument. In standard FO, thetic judments would be a result of predicate formation rules on account of the valency reduction and replacement. An alternative we would prefer is to have judgmental operators determine this variation. Thus, we introduce the following operators: CAT determines the unmarked, categorical perspective, THET the marked, thetic one. olave'blo'kadey would then be analysed as having two variants: (CAT f¡ = olave'blo'kadev (fi)) ( A ' = A g ( T r ) ) (THET f¡ = olave-blo'kadev (fi)) (A hin > Ίη > 'и. Since Wessex-type dialects have grammatical gender systems (Paddock 1988) in which most count nouns belong to the same "gender" as do sexual males (such as man, boy, bull and stallion), this nasal clitic is used as a proform for a wide range of nouns, as the following data indicate. (12)

Give-'n 'is book. 'Give him his book'

(13)

G/ve-'n to Tom. 'Give him/it to Tom'

(14)

Give-'n to-'n. 'Give him/it to him '

(15)

Kick- 'n fer me. 'Kick him/it for me'

(16)

Well, that's the endo'-'n.^ 'Well, that's the end of him/it'

Such facts as the above make it almost impossible to draw up neat paradigms for Wessex-type pronouns, even within a fairly restricted dialect area of southwestern England. When one attempts to display all the pronoun forms

258

Harold Paddock

for the Wessex area in one table, as does Rogers (1979: 34-5) it seems that Wessex-type dialects have almost complete free variation among pronoun forms. Nothing could be further from the truth.

3. Interlectal developments The above typical intralectal developments (within specific regional and/or social dialects) have been modified by interlectal developments. For example, in the social dimension no doubt all nonstandard Wessex-type dialects have now been modified by contact with more standard varieties of English. This is as true of the transplanted Wessex-type dialects in Newfoundland as it is of the Wessex-type dialects still surviving in southwestern England. Within southwestern England itself one can find significant regional differences resulting from the loss of the subject versus object case distinctions (such as I/me, thou/thee, she/her, and we/us). In general, the Peninsular Wessex or Devon-type regional dialects (found in Devon, West Somerset, and East Cornwall) tended to assign broader functions to former object forms (especially to us and her) whereas the Mainland Wessex or Dorset-type regional dialects (found in most of Dorset and some adjacent parts of East Somerset, Wiltshire, and Hampshire) tended to assign broader functions to former subject forms (especially to Г). This typical regional difference between Devon and Dorset is neatly summed up in the following Dorset stereotyping of Devon speakers: (17)

Hark a' thay Debm volk ussin' an' 'errin'. 'Listen to those Devon folk M5-ing and /гег-ing'

What such stereotyping Dorset speakers notice as peculiar in the speech of Devon speakers is of course not the mere presence of the forms us/es and her/'er but their higher frequencies and, above all, the extended functions of us/es and her/'er in Devon-type dialects. In transitional areas where extensive mixing of the two types has occurred we may be able to identify four-step (A, B, C, D) continua of the following type: (18) a. b. c. d.

A. B. C. D.

WE wouldn' do that, would we? WE wouldn ' do that, would QS? es wouldn ' do that, would es.^ US wouldn ' do that, would es.^*

(19) a. b. c. d.

A. B. С. D.

SHE wouldn' do that, would she.? SHE wouldn' do that, would 'ет? 'er wouldn ' do that, would 'er? HER wouldn' do that, would 'er.?

From CASE to FOCUS in pronouns

259

We may characterize the above four stages of the continuum as follows: Stage A:

Stage B: Stage C: Stage D:

Like Standard English. Preverbal we and she may be stressed (for focus) or unstressed. Postverbal we and she are not stressable because they do not occupy a possible focus position. Preverbal we and she may be stressed (for focus) or unstressed. Postverbal es and 'er are unstressable clitics. Both preverbal and postverbal es and 'er are unstressable clitics. Preverbal us and her may be stressed (for focus) or unstressed. Postverbal es and 'er are unstressable clitics.

Stage B, with the standard subject forms we and she in preverbal position and the non-standard subject forms es and 'er in postverbal position, is particularly interesting. It is theoretically interesting because the segmental form of the pronoun is not governed by its syntactic ftinction (i. е., subject versus object) but rather by the relative order of constituents (compare Dik 1989: 290-291, etc.).^ It is diachronically interesting because it may throw light on the stages by which the syntactic case distinction of subject/object was generally lost in Wessex-type dialects. If this is true, we may regard stage D as the most innovating variant and stage A as the most conservative one (Bailey 1973: 67-86). Stage В is also sociolinguistically (and psycholinguistically) interesting because speakers who normally use this stage seem quite unconscious of their own use of (unstressable) clitics es and 'er as subjects in postverbal position and sometimes "accuse" С and D stage speakers of using the us/es and her/'er forms incorrectly. We should note here that the above four stages are not merely geographical continua. They are also involved in social continua, both of the interpersonal type based on class, age, sex, etc. and the intrapersonal type based on stylistic ranges of individuals. However, it seems that Wessex speakers themselves distinguish most sharply between stages В and C, regardless of whether the continuum involved is geographical or social. In other words, Wessex speakers seem less conscious of the postwQrbdX forms in variants of type В but more conscious of the /preverbal forms in variants of types С and D. It seems, then, that it is not just the fimction of the forms us and her as syntactic subjects which is significant. What seems even more significant is their use in preverbal position in type С variants and their ability to accept focus marking (including full stress) in type D variants. In fact, it is my general experience that unstressable clitics seem to lie at a lower level of consciousness than do independent words. This seems reasonable in that unstressable clitics are in fact bound forms attached to free forms. They are therefore

260

Harold Paddock

closely related to affixes and are perhaps classified by speakers under morphology (i. е., word formation) rather than under syntax. However, another possibility is that speakers tend to be less conscious of any morph that is not capable of undergoing pragmatic "foregrounding" (Dik 1989: 218) or "highlighting" (Dik 1989: 343, etc.). The fates of Wessex-type pronoun forms and usages are particularly interesting in Newfoundland. In general, the Devon-type use of former objects (us and her) as subjects is by now extremely rare in Newfoundland, whereas the Dorset-type use of former subjects (/, he, she, we, they) as objects is still quite common in some parts of Newfoundland. At least four types of conditioning (linguistic, historical, geographical, and social) have led to this present day situation. The linguistic conditioning is evident in the fact that former objects used as subjects have survived best in Newfoundland when they were used as postvQvhdX unstressable clitics in the above type В variants; conversely, former subjects used as objects have survived best in Newfoundland when used as stressed focus forms as in (6) and (7) above. This means that the former syntactic case distinction (of subject versus object), wholly or partially lost in the Wessex mother dialects, was generally replaced by a pragmatic focus distinction in the Newfoundland daughter dialects. In other words, the contact between different Wessex-type dialects in Newfoundland led to the process which Trudgill (1986: 110) calls reallocation. New pragmatic functions (Dik 1978: 127-56) were reallocated to old pronoun forms which had formerly lost their syntactic functions (Dik 1978: 18) in one or more of the contributing mother dialects. The non-linguistic conditioning (historical, geographical and social) can be established with a fair degree of certainty. The Avalon Peninsula of Newfoundland was generally settled earlier, mostly by Devon-type speakers from Peninsular Wessex; the Mainland of Newfoundland was generally settled later (usually after 1800), mostly by Dorset-type speakers from Mainland Wessex.^ The Devon-type speakers on the Avalon Peninsula were subjected to two major standardizing influences. One of these was the strong influence of Anglo-Irish or Hiberno-English dialects from southeastern Ireland. The other was urbanization, especially in the larger communities of Conception Bay and in the capital city of St. John's. Both of these standardizing influences were considerably weaker for the Dorset-type speakers in Mainland Newfoundland, because fewer Irish settled there and less urbanization occurred there. The non-linguistic conditioning therefore tended to support the linguistic conditioning, with the general result in Newfoundland that Dorset-type uses of former subject pronouns as objects have survived better than Devontype uses of former object pronouns as subjects.

From CASE to FOCUS in pronouns

261

4. Conclusion The mixing of Wessex-type dialects with different pronoun forms and/or fiinctions has led to reformulations of what Dik (1989: 289) has called "expression rules, which determine the form, the order, and the prosodie contour of the constituents, given their structural and functional status within the underlying clause structure." Subtle interplays of semantic, syntactic, and pragmatic fonctions are involved in such expression rules. I will illustrate this claim with examples from third person singular pronouns. Since Wessex-type dialects have a covert system of grammatical gender (Paddock 1988) based on semantic distinctions such as mass/count, animate/ inanimate, and mobile/nonmobile; the expression rules for pronouns must refer to such semantic features in order to select the correct gender of the pronoun. For example, in my own native Newfoundland dialect a singular nonanimate count noun selects a masculine pronoun if it is nonmobile (like hat, coat, or bucket) but a feminine pronoun if it is mobile (like car, boat, or airplane). After the semantic rules have selected the gender of the pronoun the pragmatic rales must decide whether a focus form is required or not. Again, in my own native dialect, the choice of a third person singular focus form would restrict further choices since the focus form is always he for masculines, she for feminines, and that for neuters (since it has been reduced to the unstressable clitic V). However, if a nonfocus form were selected, then further choices would have to be made by the syntactic component of the expression rales. As was illustrated above, the syntactic rales would not only have to distinguish between syntactic fiinctions such as subject and object, but also between different word orders resulting from different basic illocutions (Dik 1989: 256) such as Declaratives versus Interrogatives. Finally, phonological rales might be needed to make adjustments to the phonetic output. If we apply all the above processes to a Wessex-type dialect with a full range of forms for masculine third person singialar pronouns, we would get the following results: (a) Semantic rales select masculine rather than feminine or neuter. (b) Pragmatic rales choose between the focus form he and the clitic forms uh/ur or en/'n. (c) If a clitic form is required, the syntactic rales choose between subject forms uh/ur and object forms en/'n. (d) Phonological rales (help) choose between proclitic uh and enclitic ur, or between syllabic en and nonsyllabic 'n.

262

Harold Paddock

Although the above is presented as an ordered sequence in what Dik (1989: 52) has called a "quasi-productive mode" I agree with Dik (1989: 52) when he says that "I do not wish to suggest that this order of presentation necessarily simulates the various steps that a speaker takes in producing linguistic expressions." I further agree with Dik (1989: 52) that "the order of actual production is not even organized in a sequential way" because "natural language users have rather strong capacities for the parallel processing of information." However, what the sequencing does indicate is that semantic, pragmatic, syntactic, and phonological rales are all necessary in order to account for the expression rales that convert abstract underlying clause stractures into concrete linguistic expressions. In a synthesizing monograph. Bailey (1973) tried to incorporate the methods and insights of the Transformational-Generative (TG) model of language into a "new framework" meant to solve problems of describing language variation (his "homogeneity paradox") and language change (his "temporal paradox"). In particular, Bailey hailed the TG model for its ability to "capture" the dynamic nature of language continua in space (both social and geographical space) and in time (both "real" historical time and "apparent" generational time), particularly through the use of ordered rales. The attempts of Bailey and others to apply the TG model to language variation and change enabled variationists and historical linguists to reach what Chomsky (1967: 28-59) would have called his second level of success for linguistic description. In other words, it enabled such linguists to move from the level of success Chomsky called "observational adequacy" to the level he called "descriptive adequacy." However, the TG model did little to help linguists move on to Chomsky's third level of success, which he called "explanatory adequacy." There are several reasons for this fact. One was an excessive attention to the formalism"' of rales and to the ordering of rales. Many linguists were attracted to and impressed by the pseudomathematical elegance of TG descriptions. But this often resulted in a masking of the building itself (i. е., the stracture of the language) by the TG scaffolding erected to examine that building (i. е., the formalism and notational conventions of the TG model). I feel that functional grammar (FG) provides us with new hope of moving towards "explanatory adequacy" in the fields of language variation and change. There are several reasons for this hope. One is that FG is not yet overcluttered by formalism. Another is that FG has broken out of the straitjacket of rale ordering in a radical way. Not only has it abandoned the ordering of rales (as an explanatory device) within a given component of a language (such as syntax or phonology), but, more importantly, it has aban-

From CASE to FOCUS in pronouns

263

doned what Dik (1989: 3) has called the "methodological priority" of syntax over semantics and pragmatics. This new equality among structural rules, which govern linguistic expressions, and pragmatic rules, which govern verbal interaction, should enable linguists to produce better explanations (and even better descriptions?) of language variation and change. To use Dik's terminology (1989: 2), the "functional paradigm" should enable us to overcome some of the limitations of the "formal paradigm." This hope seems reasonable because language variation and change arise in the functioning (i. е., learning and using) of languages and not in the describing of languages.

Notes 1 Such mutual cliticization of preposition followed by pronoun is typical of Wessex dialects of English. It occurs not only with of+en as o"n but also with of+thee as o"ee, of+her as о "r, of+it as о "t, of+us as о "s, and of+them as о "т. Compare the "prepositional pronouns" of Celtic languages. See, for example, Lockwood on Welsh (1975: 40), Cornish (1975: 59), and Irish (1975: 105). 2 Compare the situation in Modem Standard English where the prescriptive rule is based on syntactic case to yield It's /, It's he. It's she. It's we and It's they, but the descriptive rule is based on word order to yield It's me. It's him. It's her. It's us, and It's them. 3 For a detailed analysis of sources and periods of English settlement in Newfoundland see Handcock (1989). He shows that the source area which I have called Peninsular Wessex was mostly South Devon, and that the one which I have called Mainland Wessex was all of Dorset plus some adjacent parts of Hampshire, Somerset, and Wiltshire (Handcock 1989: 145-153). 4 TG linguists sometimes graciously confess to their excessive formalism. See, for example, the admission by Chomsky and Halle (1968: 400) concerning their own generative phonology: "The problem is that our approach to features, to rules and to evaluation has been overly formal ... In particular, we have not made use of the fact that the features have intrinsic content."

References Bailey, Charles-James N. 1973 Variation and linguistic theory. Arlington, Virginia: Center for Applied Linguistics. Chambers, J. K. - Peter Trudgill 1980. Dialectology. London: Cambridge University Press. Chomsky, Noam 1967 Current issues in linguistic theory. (Third printing.) The Hague: Mouton. Chomsky, Noam - Morris Halle 1968 The sound pattern of English. New York: Harper and Row. Dik, Simon C. 1978 Functional grammar. Amsterdam: North-Holland. 1989 The theory offunctional grammar Part I: The structure of the clause. Dordrecht: Foris

264

Harold Paddock

Handcock, W. G. 1977 "English migration to Newfoundland", in: John J. Mannion (ed.), The peopling of Newfoundland. St. John's: Institute of Social and Economic Research, Memorial University of Newfoundland, 15-48. 1989 Soe long as there comes noe women: Origins of English settlement in Newfoundland. St. John's: Breakwater Books. Lockwood, W. B. 1975 Languages of the British Isles past and present. London: André Deutsch. Paddock, Harold J. 1988 "The actuation problem for gender change in Wessex versus Newfoundland", in: J. Fisiak (ed.). Historical dialectology. Berlin: Mouton de Gruyter, 377-395. Rogers, Norman 1979 fVessex dialect. Bradford-on-Avon: Moonraker Press. Story, G. M. - W. J. Kirwin - J. D. A. Widdowson (eds.) 1982 Dictionary of Newfoundland English. Toronto: University of Toronto Press. Trudgill, Peter 1986 Dialects in contact. Oxford and New York: Basil Blackwell. Wakelin, Martyn F. 1984 "Rural dialects in England", in: P. Trudgill (ed.). Language in the British Isles. Cambridge: Cambridge University Press, 70-93. Zwicky, Arnold M. 1985 "Clitics and particles". Language 61: 283-305.

Auxiliary verbs in Arabic Martine Cuvalay 0. Introduction' This paper discusses the treatment in Functional Grammar of auxiliaries in Arabic. An earlier attempt to account for verbs with auxiliary and copular functions in Modem Standard Arabic by Moutaouakil (1986) will be reviewed in the light of later developments in FG theory. After an outline of the verbal system in Arabic, arguments will be advanced against Moutaouakil's proposal to describe all the verbs in question as being the direct expression of predicate or predication operators. The main point to be made concerns the special status of the verb капа 'to be',^ which should be distinguished from the other verbs on the basis of distribution and meaning. It will be argued that the occurrences of this verb can best be accounted for by a copula support rule that in the case of Arabic must be adapted to allow for the insertion of kâna in verbal sentences too. The operation of such an "auxiliary support rule" will be demonstrated with examples from Modem Standard Arabic and from some of the spoken dialects of Arabic. With respect to the other auxiliary verbs, it will be questioned whether they are all grammaticalized to the extent that their insertion should be captured by expression rules. A more differentiated approach will be advocated by positing altemative solutions.

1. The verbal system in Arabic 1.1. The Arabic language The Arabic language group contains many dialects without a standardized written form. There is a formal language, called Modem Standard Arabic, which is practically uniform throughout the Arab world and quite close to Classical Arabic, the old literary form which is associated with the early Koranic tradition. The contemporary spoken dialects as a group are called New Arabic. The New Arabic dialects are learned as native languages and are used in informal situations. Modem Standard Arabic is taught in school for written and formal communication.

266

Martine Cuvalay

1.2. The opposition between prefixed and suffîxed verb forms In all varieties of Arabic, there is a basic morphological opposition between two sets of verbal forms. In the first set, inflection for person, number and gender is achieved by adding suffixes to a stem. The forms in the other set consist of a stem with a prefix, or, for some person, gender and number combinations, a circumfíx. All verbs^ can be inflected in both ways. Most grammarians have described the main fiinction of this opposition in terms of being either aspectual or temporal,"* and the forms have been called "preterite and future", "perfective and imperfective", and "completed and uncompleted", to name but a few. In this article I will use the neutral terms "suffixed form" (SF) and "prefixed form" (PF) to avoid the confusion that arises from a functionally based terminology.^ In fact, the description of the functions that the two main verb forms may fulfil is not limited to the domain of tense and aspect alone. Beeston (1968: 48-49) recognizes three levels of differentiation: aspectual:

temporal: modal:

"the perfect [SF] points to a single action, regarded as instantaneous in its occurrence [perfective], the imperfect [PF] to habitual or repeated action, or to one visualized as covering a space of time [imperfective]" "the perfect points to past time, the imperfect to present or future time" "the perfect points to a fact, the imperfect to a conceptual idea not necessarily realized in fact"

Even this semantic classification does not account for all the ñmctions involved, like for instance the expression of wishes with the SF in Classical Arabic and Modem Standard Arabic, and the obligatory use of this form in the protasis of most types of conditional sentences.

1.3. The subdivisions of the PF In Classical Arabic and Modem Standard Arabic a further subdivision of the PF is achieved by different endings: и for indicative, a for subjunctive, and the absence of a short vowel for jussive forms.^ There is also an optional preverbal fiiture marker sawfa or sa-, used with the indicative PF. The New Arabic dialects have PFs without such different endings, but most of them do have a pair of preverbal markers to indicate tense, aspect or mood distinctions. The meanings and forms of these preverbal markers are not necessarily the same for the individual dialects.

Auxiliary verbs in Arabic

267

In Colloquial Egyptian Arabic, for instance, the preverbal marker bi- expresses progressive or habitual aspect. The preverbal marker ha- is used to indicate prospective aspect or future tense. In most Moroccan dialects the preverbal markers ka- (or ta-) and gadi- (ga-) cover more or less the same values. Although the way in which the PF is ftirther differentiated creates a rather sharp distinction between Classical Arabic and Modem Standard Arabic on the one hand, and the colloquial dialects on the other hand, the general setup of the verbal systems is essentially the same.

1.4. Simple verb forms For all varieties, there are only a few simple verb forms."' If they occur without auxiliaries or particles that emphasize the intended meaning, their interpretation is dependent on the context. In sentences (la) and (lb) from Classical Arabic, the intended time references are made explicit by the adverbials min qablu 'before' and yawma l-qiyamati 'the Day of Resurrection' (which is supposed to be located in the future).

(1) a.

b.

Classical Arabic (Fischer 1972; Koran 2: 91) lima taqtulma l-'anbiyä'a min qablu? why kill:PF.2mp the prophets before? 'Why did you use to kill the prophets before?' Classical Arabic (Comrie 1976; Koran 2: 113) llahu yahkumu bayna -hum yawma l-qiyämati ... God judge:PF.3ms between them day resurrection... 'God will judge between them on the Day of Resurrection...'

In both sentences we have prefixed forms, but their interpretation is different. In (la), the adverbial situates the predication in the past, so the PF (which refers to nonpast on the temporal level) is not used to indicate time reference. It has to be interpreted on the aspectual or modal level and is here taken to be the expression of imperfective aspect, pointing at habitual action. In example (lb), the intended time reference as indicated by the adverbial and the interpretation of the verb form on the temporal level are compatible. Interpretations on the modal (not presented as a fact) or aspectual (ongoing or habitual/repeated action) level seem to be less likely. It should be noted that the obligation to express a certain category is hard to define, owing to differences in style and discourse type. Within the tem-

268

Martine Cuvalay

poral category present time reference is not necessarily expressed, but past and future tense generally are, at least in Modem Standard Arabic and most New Arabic dialects.

1.5. The relation between meaning and form From the above description of the verbal system in Arabic it will be obvious that simple verb forms do not have a specific meaning that can be established without considering the context. Although all the different functions of one verbal form could be accounted for by postulating a rather vague unified semantic inteφretation, like, for instance, "closedness" for the SF, I will not try to do so, and take the position that the use of these forms can best be described in the way Dik (1989: 302) proposed for verbal categories which serve a variety of puφoses. Dik (1989: 300-303) distinguishes three types of morphosyntactic operators (μ-operators) that together determine the form of an expression by applying rules of the format: (2)

Operator[Operandum] = Value

Primary μ-operators are present in the underlying clause structure and have a direct semantic inteφretation. Auxiliary μ-operators do not occur as such in the underlying clause structure, but are introduced by expression rules in order to trigger later expression rules. These auxiliary μ-operators serve to express a variety of semantic relations, but have a unified formal effect on their operandum. Contextual μ-operators are primarily used to capture agreement relations. According to Dik, verbal categories such as "infinitive", "present participle" and "subjunctive", that appear to have no specific unified meaning, are best accounted for by auxiliary μ-operators. They first translate a semantic category, like for instance Progressive Aspect, into the verbal category "present participle", and then proceed to effectuate the corresponding formal inflection. In the case of Arabic, different predicate, predication, proposition and illocutionary operators (π-operators) in the underlying clause structure may be expressed through the same verb form. I will therefore consider these primary operators to get their expression through auxiliary μ-operators, that first assign a formal category (SF or PF) to the semantic value of a π-operator, and then apply the expression rules corresponding to this formal category.

Auxiliary verbs in Arabic

269

One of the theoretical implications of this approach is that I consider the SF and PF to be ambiguous rather than vague. So instead of presupposing one meaning with different "implicatures" according to the context, I assume that there are conceptually different meanings in the clause structure which eventually get the same expression. I take the same position with respect to New Arabic verb forms like the ¿г-PF in Colloquial Egyptian and the ka-?V in Moroccan Arabic, which also may have two or more different meanings. Some of the ambiguity inherent to the main verb forms in Arabic is avoided by using verbal complexes, in which two or even more verbs form one predication. This paper is dedicated to a fimctional description of the auxiliary verbs that may occur in these verbal complexes.

2. Verbal complexes in Arabic 2.1. Auxiliary verbs In Arabic, the information about person, gender and number of the first argument term is coded in each verbal form, whether this term is overtly expressed, as in sentences (lb) and (3), or not, as in (la), (4) and (5). In all varieties of the language it is possible to form complex predicates with more than one verb. The first verb (and in complexes with three verbs also the second) has to be taken from a restricted class of auxiliary verbs, and all forms are inflected for person, number and gender. The cross-reference relations between the verb forms in a complex construction suggest a certain degree of (moφhological) independence. In fact, a formal analysis in which the verbal complex is syntactically inteφreted as a matrix predicate with a complement or adverbial clause could be defended in all cases. For reasons to be discussed later, such an analysis will be favored for some combinations. Complex forms with the verb капа 'to be', however, are always considered to represent one syntactical unit, consisting of a predicate (the last verb) with an auxiliary.

2.2. Combinations with kâna The most frequently used verbal complexes involve a form of the verb капа 'to be'. Both the SFs and PFs of the verb капа, and in the New Arabic dialects also PFs of капа with a preverbal marker, may be combined with inflected forms of another verb. In sentences (3), (4) and (5) I have repro-

270

Martine Cuvalay

duced examples with an SF of капа in Modem Standard, Egyptian and Moroccan Arabic. (3)

(4)

(5)

Modem Standard Arabic (Beeston 1970) kanat al-mahkamatu tajlisu kulla yawm be:SF.3fsg .the-court sit:PF.3fsg every day 'The court used to sit daily.' Egyptian Arabic (Hinds-Badawi 1986) kän samai il-wägib lamma ruhti-l-u be:SF.3msg do:SF.3msg the-homework when go:SF.lsg-to-him 'He had done his homework when I went to see him.' Moroccan Arabic (Harrell 1966) elas ma-kanet-s ka-tehder msa-h? why NEG-be:SF.3fsg-NEG PMl-talk:PF.3fsg with-him 'Why wasn't she talking to him?' (PM = preverbal marker)

In these examples, the SF of капа puts the main proposition in the past, while the form of the second verb (the lexical predicate) specifies an aspectual value. Before further investigating the use of капа as an auxiliary, we now turn to a discussion of the functions of this verb as a copula and its treatment in FG. 2.3. The treatment of käna as a copular verb in FG In Arabic, sentences with a nonverbal predicate are not expressed with an equivalent of the English verb "to be" when they refer to the Present and are neutral with respect to the aspectual and modal distinctions relevant to the specific variety of the language. So a simple statement with an adjectival predicate like "intelligent" can be expressed as in (6):

(6)

Modem Standard Arabic (Moutaouakil 1986) hind-un ф4акТу-а1-ип Hind-NOM intelligent-f-NOM 'Hind is intelligent.'

If there are temporal, aspectual, or modal distinctions to be made, a copular verb will be added, resulting in a sentence like (7): (7)

Modem Standard Arabic (Moutaouakil 1986) kânat Hind-un dpkTy-at-an be:SF.3fs Hind-NOM intelligent-f-ACC 'Hind was intelligent.'

Auxiliary verbs in Arabic

271

the first effort to account for the use of käna as a copula in Modern Standard Arabic within the theoretical framework of FG resulted in the following rule:® (8)

Copula support in Arabic: input: π predicateß (xi) (x2)...(Xn) conditions: π = Past, Future, Generic (Temps-zéro) β = Nominal, Adjectival, Adpositional, Adverbial output: π kânay predicatep (xj) (x2)...(xn)

This rule was introduced and later abandoned by Moutaouakil (1986: 2), who felt it was insufficient for the following reasons: (i) (ii)

the insertion of the verb капа is not only sensitive to the temporal operator, but also to the aspectual operator; other copular verbs can appear in nonverbal constructions to express different combinations of temporal and aspectual operator values.

He subsequently formulated a new set of expression rales to account for the insertion of käna and the use of several other verbs with copular and auxiliary functions. In section 2.4. some characteristics of these other verbs will be discussed. 2.4. The defective verbs Traditionally, Arab grammarians classified verbs with copular and auxiliary functions in a group, which they referred to as käna wa akawätuhä 'käna and its sisters'. Apart from капа itself, this group contains the verbs laysa 'not to be',' sßra 'to become', and three subgroups of three or more phasal aspect modifying verbs each, with meanings like 'to become', 'to remain' and 'not to cease'. Although most of these verbs can also function as full (independent) verbs, their most common use is "incomplete", in the sense that they occur in complex constractions with another verb or with a nonverbal predicate. As such, their lexical meaning is reduced, or empty. If a nominal or adjectival predicate is associated with one of the verbs in this group, it has to be marked for the accusative case (as opposed to the nominative case, which would have been used in the absence of such a verb). Apart from this characteristic inflection there are no formal signs of complementation.

272

Martine Cuvalay

In sentence (9) we see an example with one of the sisters of капа and a verbal predicate:

(9)

Modem Standard Arabic (Moutaouakil 1986) 'asbaha kâlidun yaktubu al-qisas become:SF.3msg Kälid write:PF.3msg the-stories 'Kälid has started to write stories. '

As I already mentioned in 2.1., both verb forms are inflected to agree with the first argument term kâlidun. The occurrence in constructions with a verbal predicate is characteristic of two other groups of verbs (meaning 'to be about' and 'to begin'), that, together with kâna and its sisters, are called al-afsâlu an-näqisah 'the incomplete or defective verbs'. Except for kâna, which may be combined with any other verb form (see section 2.2 sentences (3), (4) and (5)), the defective verbs usually form constructions with a verb in PF. As stated in section 2.3. above, the similarities between the verb kâna and the other defective verbs led Moutaouakil to abandon the copula support rule as in (8) in order to integrate it with the expression rules he devised to account for the insertion of the other "copular" verbs.

2,5. The treatment of the defective verbs in FG According to Moutaouakil (1986: 5-7), the aspectual values relevant to Modem Standard Arabic are perfective and imperfective. The imperfective is further differentiated by the operator values inchoative (continuous or noncontinuous), durative (habitual or iterative), nonachieved (in the sense of still going on) and "approxitive" (immediate prospective). The temporal operator may assign the values present, past and future, the past being fiirther subdivided into absolute and relative past. As it is of little avail to the current discussion of the treatment of the defective verbs in FG, the adequacy of this classification will remain unquestioned. I will give two examples to show how this classification is used in the expression rules that Moutaouakil formulated to account for the interaction of temporal and aspectual values. Moutaouakil (1986: 9) states: "Le prédicat verbal auquel est associée la catégorie aspectuelle du "Perfectif prend la forme de l'Accompli lorsqu'il est au "Passé absolu" ... et la forme de l'Accompli à laquelle est adjoint le verbe auxiliaire kan ... lorsqu'il est au "Passé Relatif." (The verbal predicate which is associated with the aspectual category Perfective takes the SF when

Auxiliary verbs in Arabic

273

it is in the Absolute Past, and the construction with the SF and the auxiliary verb капа w^hen it is in the Relative Past.) On this basis he formulates the following rules: (10) a. input: output: b. input: output:

[Perf[AbsPast[predv(x')-..(x")]]] [SF-predv(x')...(x")] [Perf[RelPast[predv(x')...(x")]]] [käna-SF-predv(x') · · -(x")]

These rules are supposed to account for the different forms of the predicate in sentences (11a) and (lib). Modem Standard Arabic (Moutaouakil 1986) (11) a. najahat Zaynab-u succeed. SF.3fs Zaynab-NOM 'Zaynab has succeeded.' b. sa'usiru-ka l-kitaba gadan 'id FUT-lend:PF.ls-you the book tomorrow because sa 'акйпи 'anhaytu gira 'ata-hu FUT-be:PF.ls fmish:SF.ls reading-his Ί will lend you the book tomorrow as I will have finished reading it then.' When the predicate is characterized aspectually by the categories imperfective, inchoative and noncontinuous, it is realized with the PF and one of the auxiliary verbs of the group of sarasa 'to begin': sarasa, tafiqa, bada'a, and jasala. It is realized in this form, whether it is in the past (12a), present (12b), future (12c) or generic (universal) tense (12d). Modem Standard Arabic (Moutaouakil 1986) (12) a. sarasa kâlid-un yaktubu r-risälata begin:SF.3ms p l i d - N O M write:PF.3ms the-letter 'Kâlid began to write the letter.' b. yasrasu

kälid-un

yaktubu

r-risâlata

begin:PF.3ms Kalid-NOM write:PF.3ms the letter 'Kälid begins to write the letter.' c.

sa-yasrasu

kälid-un

yuharriru

maqâla-hu

FUT-begin:PF.3ms p l i d - N O M edit:PF.3ms article-his 'Kälid will begin to edit his article tomorrow.'

gadan tomorrow

274

Martine Cuvalay d.

yasraeu

t-t_alj-u

yadübu

fi

r-rabïei

begin:PF.3ms the-snow-NOM melt:PF.3ms in the-spring 'The snow begins to melt in the spring.' The rule which is responsible for the form of the predicate in this type of constructions is formulated by Moutaouakil (1986: 13) as in (13):

(13)

input:

[ImperftInch[NonCont[

output: [

saraea tafiqa bada'a jaeala

Past Present Future T0

[pred,(x')...(x")]]]]]

PF-predv(x')...(x")]

The other rules are built in the same way, and although there remains of course much more to explain, it will be clear from the examples that according to this later proposal the verb kâna as well as the other defective verbs are all inserted through the direct mapping of operator values to corresponding (complex) forms in the expression component. In my opinion, this introduction of a rather large number of expression rules does not do justice to the special characteristics of капа, which distinguish this verb from the other members of the defective group. In the following section I will enumerate some of the most important differences and discuss how they can be accounted for in the FG model.

3. Another approach to the Arabic auxiliaries in FG 3.1. The differences between käna and other defective verbs First of all, кйпа is the only verb which can be said to be lexically empty. It may be used freely, without considering the semantic features of the State of Affairs (SoA) designated by the predicate it supports. The other defective verbs do add certain aspects of meaning to the verbal complexes they form, and for each of them individual selection restrictions need to be formulated to guarantee semantic compatibility with the lexical predicate. The verbs jaeala 'to make', and bada'a 'to begin' from rule (13), for instance, can both be used in complex constructions to mark the beginning of an SoA that is not necessarily still going on at the moment of speech (or

Auxiliary verbs in Arabic

275

other reference point). The verb jaeala ('to make'), however, can as a defective verb only be combined with an SoA that is conceived of as [+ control], i. e. the first argument must have the power to determine whether this SoA will obtain or not. The verb bada 'a ('to begin') is not sensitive to this feature, and may form complexes with SoAs that are either [+control] or [-control]. A second significant difference is, that капа may be combined with verbs in all possible forms, whereas the inflection of verbal predicates in construction with the other defective verbs is limited to the PR The defective verbs may themselves occur in the SF, and can even be preceded by a form of кйпа, like in sentence (14):

(14)

Modem Standard Arabic (Messaoudi 1985) капа 'akada yastaseiru râhatan... be:SF.3ms start:SF.3ms feel:PF.3ms rest 'He started to feel at ease...'

As would be expected on the basis of these differences, complex forms with кйпа are much more frequent than combinations with any of the other defective verbs. In her analysis of a text in Modem Standard Arabic, Messaoudi (1985: 175) found 360 combinations with капа, whereas 15 other auxiliary verbs together were responsible for 315 complex forms. A final remark concems the uniqueness of капа, in the sense that this verb has no synonyms. When a form of капа is used, it can not be replaced by another auxiliary verb without altering the meaning of the sentence. The other defective verbs all belong to smaller subgroups of verbs with about the same meanings and functions, and are as such often interchangeable. This interchangeability is reflected in the output of rale (13), where a choice can be made between four verbs meaning 'to begin'. Moutaouakil (1986) opted for the same solution in the expression mies for the other defective verbs: the first position in the output rales is always occupied by two or more verbs between braces. I think that the differences between капа and the other defective verbs should have consequences for their representation in FG. According to Moutaouakil's proposal, all verbs with copular and auxiliary functions are considered to be the direct expression of an operator value. The verb капа, however, is only a mediator which makes the expression of an operator possible by its ability to be inflected. The most powerful way to account for a verb with such an intermediary function in FG is through a general copula support rale, which in the case of Arabic may be extended to include auxiliary support too.

276

Martine Cuvalay

3.2. A general auxiliary support rule The conditions that motivate 1he use of капа with nonverbal predicates may also occur in structures with verbal predicates. In sentences with nonverbal predicates, the expression of one single TMA value already involves the insertion of a copular verb. With verbal predicates, the necessity to insert a supportive verb only comes up when there are more TMA values that have to be expressed through verbal inflection. The following principles would account for "auxiliary support" in these cases: (i) (ii) (iii)

only one distinction of tense/mood/aspect can be expressed on a verb by inflection; therefore, if two distinctions are to be expressed at the same time a supportive auxiliary is needed; the distinctions that are closest to the stem in underlying clause structure have priority for being expressed on the lexical verb, thus forcing the "outer" operator value to be expressed on the auxiliary.

This last principle is a reflection of the "projective" nature of the expression of grammatical operators, within the FG model represented in a languageindependent hierarchical layering of TMA categories. For the relative positions of π-operators in the model I refer to Hengeveld (1989) and Dik (1989: 56-60). See also Dik's article on verbal complexes in this volume for a detailed description of the "projectivity principle". For an illustration of how these three principles together account for the formation of verbal complexes with the auxiliary капа we may look at the examples in section 2.2. In sentence (3), two operator values are expressed, namely Past Tense and Habitual Aspect. We assume the aspectual operator value Habitual to be closer to the verbal predicate jalasa 'to sit' in the underlying clause structure than the temporal operator Past. ' ' This means that this "inner" operator value will be handled by the auxiliary μ-operators (see section 1.5.) in the expression component first. The μ-operators "translate" the value Habitual Aspect to the inflectional category PF, and the as yet uninflected verb is coded to get the PF agreement affixes later, when the contextual μ-operators will do their work. Now that the inflectional code corresponding to the aspectual value Habitual Aspect has been assigned to the lexical predicate the temporal value Past Tense will be handled by the expression rules. As this value is also expressed through verbal inflection the auxiliary μ-operators search for an uncoded verb first. This operation fails, because the verb jalasa ('to sit') already received a PF code. The inability to assign the inflectional code corresponding to Past

Auxiliary verbs in Arabic

277

Tense triggers auxiliary support. After the insertion of капа, this second part of the procedure is repeated, and the supportive verb käna is coded for the inflectional category SF. Sentence (4) shows two SFs, one to mark Past Tense and another to indicate Perfective or Perfect Aspect. Although it is not relevant for the expression of this sentence, I assume that the expression rules operate on the underlying clause structure in the same order. They assign the SF code associated with the aspectual operator to the lexical verb first and use the intermediary function of the supportive verb käna to assign the second SF code corresponding to the temporal value. The same principles account for the occurrence of käna in sentence (5) from Moroccan Arabic. In the expression component the auxiliary μ-operators determine that Habitual Aspect is to be expressed by inflecting a verb according to the PF with a preverbal marker ka- (the ^a-PF). They succesfully assign this inflection to the verbal predicate and look for a next operator value. This turns out to be Past Tense, which also needs to be mapped on a verb. The auxiliary support rule is triggered and provides for the verb капа. This verb is then coded to be inflected with suffixes. One of the advantages of the principles discussed here is that we are now able to account for the different meanings of pairs of verb complexes like in (15) without formulating special rules for all possible combinations. Egyptian Arabic (15) a. SF + bi-PF / bi-PF-b SF kân biyiktib 'he was writing' or 'he used to write' biykün katab 'he usually has written' (at a certain recurring reference point) b.

SF + ha-PF / ha-PF + SF kan hayiktib 'he was going to write' haykûn katab — 'he will have written'

c.

bi-PF + ha-PF / ha-PF + bi-PF biykun hayiktib 'he usually is about to write' (at a certain recurring reference point) haykûn biyiktib — 'he will be (habitually) writing'

The inteφretation of these complexes is always in line with the projectivity principle. If we accept that the number of operator values that have to be expressed by verbal inflection and the order in which they are dealt with by the expression rules determine whether a certain value will be expressed on the lexical predicate or on the auxiliary капа, we don't need to make strong

278

Martine Cuvalay

statements like "tense is always expressed on the auxiliary and aspect on the predicate" (see for instance Jelinek (1981: 29)), that have to be mitigated later. In this section I showed that both the auxiliary and copular functions of the verb капа can be accounted for by a supportive device which is sensitive to the absence of an uninflected verb form and the presence of an operator value which has to be expressed through verbal inflection. In this way, it is not the operator value itself that is directly responsible for the insertion of капа, although it does determine its inflectional form.

3.3. The representation of the other defective verbs In section 3.1.1 argued that the differences between кйпа and the other defective verbs should be reflected in their treatment in FG. If we accept the general auxiliary support rule as described in 3.2. for the insertion of forms of the verb кйпа only, and conceive of the other defective verbs as the direct expression of operator values, such a different treatment is established. In this case, the verb кйпа would be accounted for by the support rale, triggered by auxiliary μ-operators, whereas the other defective verbs would be associated with primary μ-operators. The primary μ-operators assign the PF code to the verbal predicate (if present) and introduce the as yet uninflected and uncoded defective verb corresponding with the specific operator value. The form of the defective verb may be influenced by an "outer" operator. This is reflected in rale (13), where the form of the verb from the defective group is not fixed. It can be used to express one of the temporal or modal values indicated in the input phrase. In fact, the defective verb can be said to take over this fiinction from the lexical verb, which now obligatorily carries the PF code. With nonverbal predicates, the insertion of one of the defective verbs provides for an opportunity to express operator values that would not be available without this verb. As such, they have a copular function similar to the verb кйпа. It should be noted again that this is never their primary function. If there is no other operator value that needs to be expressed, the defective verbs are usually inflected with suffixes. Although this description seems to be convenient with respect to the desired differentiation between капа and the other defective verbs, there are some problems that remain to be solved. A first problem concerns the supposed interchangeability of some of the verbs.

Auxiliary verbs in Arabic

279

Except for the one that conditions the insertion of laysa 'not to be', Moutaouakil's rules leave a choice to be made between the verbs in a subgroup. Since these rules are supposed to apply autonomously in the expression component, this solution requires fiill synonymy betw^een the alternatives. I already mentioned a difference in the use of the verbs jasala and bada 'a. When followed by a verb in PF, they both mean 'to begin to', but the former with the strong association of 'to set out to', and the latter with the possible translation of 'to set in' (see Wehr 1961: 127, 44). In the other groups we also fmd different shades of meaning, for instance between 'istamarra and baqiya. Both verbs are used with the meaning of 'to continue to', but 'istamarra has a connotation of persistence that baqiya does not have. Although further investigation is needed, I assume that these differences can not be accounted for by predicate operators. A second problem is that many verbs that may be combined with a verbal predicate in the PF are not classified in the defective group. Some of them are quite common and may even be more frequently used in complex constructions than defective verbs are. The verb eada 'to return', for instance, may form a construction with a PF with the specific meaning of 'to resume (an activity)'. The verb rada 'to go' could be added to the group of verbs in (13), because it is also translated with 'to begin, set out to do' if combined with a following PF. Considering these two points, I think there are reasons to question the adequacy of a representation in which all the defective verbs are taken to be the expression of a predicate operator value. For some of them, it would be very hard to come up with special characteristics which set them apart from "normal" verbs, that also may have a predication as a complement. The verbs bada'a 'to begin', baqiya 'to remain, stay, continue to be', and käda 'to be on the point, be about', for instance, apparently have no reduced meaning in their functions as defective verbs in Modem Standard Arabic. Even the verbs that do have a less specific meaning in their defective use are, in this respect, not different from some of the verbs outside the defective group that may have a reduced meaning in combinations with another predicate too. For all these verbs, inside and outside the special defective group, we may consider formulating predicate formation rules that operate in the fund. If we do so, we avoid the problem associated with the choices to be made between several verbs conveying more or less the same meaning. In this part of the model there is no objection to different selection possibilities. Whether this is the best option, however, has to be investigated for each verb in all varieties of Arabic separately. In his work on the auxiliarization of the English modals, Goossens (1987: 139-140) discusses some criteria that may help to decide which representation is most adequate.

280

Martine Cuvalay

3.4. A grammaticalization scale On the basis of his detailed investigation of the diachronic development of EngUsh modal verbs, Goossens (1987: 118-119) proposed to relate the degree of grammaticalization'^ of a given construction to a differentiation in the way it is represented in the underlying clause structure. Within FG, he found that an increase in grammaticalization could be reflected by a scale as in (16). (16)

filli predicates < predicate formation < predicate operators

Although he explicitly states that this three-point grammaticalization scale is not refined enough, Goossens demonstrated its usefiilness for a first differentiation between complex constructions that are not grammaticalized at all (full predicates), constructions that are grammaticalized to a certain degree (predicate formation), and fiilly grammaticalized combinations (predicate operators). The following conclusions of Goossens' (1987: 139-140) may be relevant for an analysis of auxiliarization in Arabic: (i)

(ii)

(iii)

In deciding the choice between independent predicate status or predicate formation for a modal verb a crucial argument appeared to be whether the modal verb brings along its own argument structure (in which case we take it to be an independent predicate) or whether a combining predicate ... imposes its argument structure on the whole combination (this gives rise to treatment under predicate formation). To assign predicate operator status to a given modal it should, in addition to the fact that it does not have an argument structure of its own, be used in specific grammatical functions such as the expression of tense, the marking of certain types of subclauses and the like. ... it [the grammaticalization scale] appears to correlate with a decreasing specifity in the (semantic) combinatorial possibilities for the item that can be shown to develop from independent predicates to the other end of the scale.

The information on defective and other verbs in grammars and dictionaries of Arabic is not detailed enough to support anything more than the general remark that the defective verbs in Classical Arabic and Modem Standard Arabic seem to be partially grammaticalized, if at all. Whether a certain verbal complex is best classified as a ñill predicate with a complement, or

Auxiliary verbs in Arabic

281

should be seen as derived by a predicate formation rule, can only be determined after the detailed investigation of a софиз, in the way Goossens did with the English modals.'^ In the New Arabic dialects, we find much more variation with repect to the degree of grammaticalization of verbal complexes, and the need to differentiate between at least three representations in the underlying structure is felt more strongly than in Classical Arabic and Modern Standard Arabic. Again, further investigations of the subject will have to confirm the relevance of such a differentiated approach to verbal complexes in Arabic.

4. Conclusions In this paper I stated that the general supportive fimctions of the verb кйпа in all varieties of Arabic can best be represented in Functional Grammar by a copula support rule adapted to account for its use in verbal sentences too. The other verbs with auxiliary and copular functions should be analysed individually to determine their position on a "grammaticalization scale" and to decide how they can best be incorporated into the FG model.

Notes 1 This article has profited considerably from the detailed comments and suggestions of Simon Dik and the editors of this volume. I am furthermore grateful to Casper de Groot, Kees Hengeveld, and Hella Olbertz, who helped me shape my ideas on the subject. 2 In accordance with the generally accepted conventions the cited form of the verb is the 3rd person singular masculine of the suffixed verb form (SF). In this article, the Modem Standard Arabic form käna is meant to refer to the equivalents of this verb in the spoken dialects too, although the phonetic realizations may vary. 3 Except for the verb laysa 'not to be', see also note 9. 4 For a survey of the literature on the subject I refer to Aartun (1963), Kharma (1983) and Messaoudi (1985). 5 Neutral terms are also prefered by Cohen (1989): "cs" and "cp", Kharma (1983): "l.S.F." and "2.S.F." and Messaoudi (1985): "Fs" and "Fp". The 3rd person masculine forms of the verb 'to do', faeala and yafealu, are also commonly used to indicate the two forms. 6 The forms that are marked by a prefix and a suffix have other endings for the indicative, subjunctive and jussive inflections. 7 A full description of the verbal system in Arabic would have to include the imperative and active participle forms. As the present discussion of auxiliary verbs can do without these forms, they are left out to avoid unnecessary complexity. 8 I adapted the notation slightly to fit the general schema in Dik (1989). 9 The verb laysa can only be inflected with suffixes and is used for the negation of predications with present time reference.

282

Martine Cuvalay

10 This is not true for the verb laysa 'not to be', which has no synonyms, like käna. The verb laysa, however, has only one function (see also note 9), whereas käna has many. 11 This assumption is not based on Hengeveld's classification. He considers both Tense and Quantificational Aspect to be predication operators (π2-operators). Dik (this volume) states, that the projectivity principle should also hold for the different operator categories that may be ordered inside one of the layers. 12 Goossens measured the degree of grammaticalization by using six parameters, defined by Lehmann (1982). 13 See also Olbertz (1989) for a discussion of the criteria involved.

References Aartun, Kjell 1963 Zur Frage altarabischer Tempora. Oslo: Scandinavian University Books. Beeston, Alfred F.L. 1968 Written Arabic, an approach to the basic structures. Cambridge: Cambridge University Press. 1970 The Arabic language today. London: Hutchinson University Press. Cohen, David 1989 L'aspect verbal. Paris: Presses Universitaires de France. Comrie, Bernard 1976 Aspect. Cambridge: Cambridge University Press. Dik, Simon C. 1989 The theory of Functional Grammar. Part I: The structure of the clause. Dordrecht: Poris. Fischer, Wolfdietrich 1972 Grammatik des Klassischen Arabisch. Wiesbaden: Harrassowitz. Goossens, Louis 1987 "The auxiliarization of the English modals", in: Martin Harris - Paolo Ramat (eds.), 111-143. Harrell, Richard S. (ed.) 1966 A dictionary of Moroccan Arabic. Washington: Georgetown University Press. Harris, Martin - Paolo Ramat (eds.) 1987 Historical development of auxiliaries. Berlin: Mouton de Gruyter Hengeveld, Kees 1989 "Layers and operators in Functional Grammar", Journal of Linguistics 25: 127157. Hinds, Martin - El-Said Badawi 1986 A dictionary of Egyptian Arabic. Beirut: Librairie du Liban. Jelinek, Eloise 1981 On defining categories: Aux and predicate in colloquial Egyptian Arabie. [Dissertation, University of Arizona]. University Microfilms, Ann Arbor Kharma, Nayef 1983 A contrastive analysis of the use of verb forms in English and Arabic. Heidelberg: J. Groos. Lehmann, Christian 1982 Thoughts on grammaticalization. Köln: Arbeiten des Kölner UniversalienProjekts.

Auxiliary verbs in Arabic

283

Messaoudi, Leila 1985 Temps et aspect, approche de la phrase simple en Arabe écrit. Paris: Geuthner. Moutaouakil, Ahmed 1986 "Les constructions copulaires en Arabe". [Paper. University Mohammed V, Rabat.] Olbertz, Hella 1989 "Periphrastic Aspect in Spanish". Working Papers in Functional Grammar 32. Wehr, Hans 1961 A dictionary of modern written Arabic (Arabic-English). Wiesbaden: Harrasowitz.

Term-to-phrase mapping rules: A case study from Arabic Ahmed Moutaouakil 0. Introduction Within the Functional Grammar framework, entities referred to in a sentence receive two different kinds of representation. They are represented as logicosemantic structures (i. e. terms) at the predicative structure level and as morphosyntactic structures (i. e. phrases) at the constituent structure level. A considerable amount of research has been devoted, within this framework, to the description of the logicosemantic properties of terms. We also need a complete and precise formulation of a subset of expression rules which would enable us to predict (in a principled way) the manner in which a logicosemantic structure (i. e. a term) is converted to a syntactic configuration (i. e. a phrase). The aim of this paper is to try to propose an explicit formulation of this subset of mapping rules and to show how it accounts for the data of Modem Standard Arabic. I will be chiefly concerned with the rules that deal with the main processes involved in the conversion of terms into phrases, namely: (a) head selection, (b) morphological realization of term operators, (c) word ordering within phrases, and (d) case marking. Special attention will be paid to the morphosyntactic properties of nominalized terms.

1. The organization of Functional Grammar According to the general organization of grammar in the theory of Functional Grammar, a sentence is derived through the progressive building of three structures: Predicative structure. Functional structure, and Constituent structure.' These three structures represent the input of the successive application of three sets of rules which operate in the following way: the Fund (which subsumes a lexicon and predicate formation rules) delivers (basic and derived) nuclear predicate frames which are extended through the adding of satellite positions. A fully specified predicative structure is obtained by the application of two ñirther subsets of rules to a (nuclear or extended) predicate frame: Term insertion rules and Predicate operator specification rules. The

286

Ahmed Moutaouakil

former subset of rules inserts the appropriate lexical items in the argument (and satellite) positions. The latter subset determines the Mood, Aspect and Tense features associated with the predicate. The functions assignment rules specify the fiinctional status of the terms by assigning to them syntactic functions (Subject and Object) and pragmatic functions (Topic, Focus). We get a fiilly specified functional structure by assigning syntactic and pragmatic functions to the terms and specifying the predication operator, that is, by determining the (literal and implicated) illocutionary forces associated with the entire predication. The expression rules determine the way in which the functional structure is mapped onto a constituent structure. We can distinguish the following subsets of expression rules: (i) (ii) (iii)

(iv) (v)

The phrase constituency rules which determine the form in which terms are realized (i. e. which map terms onto phrases); the predicate constituency rules determining the form in which the predicate itself is realized; the relator insertion rules by virtue of which different sorts of relators (adpositions, subordinators, illocutionary force indicating devices) are inserted in the appropriate slots; the placement rules which determine the order of constituents, the stress and intonation assignment rules.

The output of the application of these five subsets of expression rules is a prephonological representation which serves as input to phonological rules. In the remainder of this paper, I will concentrate on the first subset of expression rules, i. e. phrase constituency rules.

2. Head selection As noted above, phrase constituency rales are those expression rales which determine the way in which terms (conceived of as logicosemantic stractures) are mapped onto phrases, that is, syntactic confígurational stractures. Two sorts of phrase constituency rales can be distinguished: a) the rales which determine the internal syntactic stracture of phrases, and b) the rales which specify the different syntactic relationships holding between phrases and the predicate. The first sort of rales must be formulated so as to fulfil the following three main tasks: determining the syntactic status of restrictors, specifying the form in which operators are realized, and specifying the positions in which these elements are placed within phrases.

Arabic term-to-phrase mapping rules

287

In the predicative structure, a term consists of a term operator and a restrictor or a sequence of restrictors. In order to convert a term to a phrase, one of the restrictors must be selected as a head whereas the other restrictors get the status of modifiers. For selecting the head, the following general procedure is used: when a term contains only one restrictor, this unique restrictor becomes the head as in (1): (1)

haraja r-rajulu went-out the-man-nom 'The man went out.'

In the case of terms containing more than one restrictor, the first restrictor is selected as the head of the phrase while the other restrictors become modifiers. Sentences (2a-b) illustrate the point: (2) a.

b.

zurtu madînatan faransiyyatan visited-I city-acc French-acc Ί visited a beautifial French city.'

jamttatan beautiful-acc

qabaltu zawja ?uhti Zaydin met-I husband-acc sister-gen Zayd-gen Ί met Zayd's sister's husband.'

The terms madînatan faransiyyatan jamtìtan and zawja ?uhti Zaydin in (2a) and (2b) have, at the predicative structure level, the structures (3a) and (3b) respectively: (3) a.

b.

(ilfxi: madTnatN(xi): faransiyyatA (xi); jamTlatA(xi))

(dlmxi: zawjN(xi) (dlfxj: ?uhtN(xj) (dlmxk: ZaydN(xk))))

If these generalizations are correct, then we can formulate the "Head selection rule" as follows: (4)

Head selection rule: "The restrictor selected as the head of the phrase is: (i) the unique restrictor of terms containing one restrictor; (ii) the first restrictor of terms containing more than one restrictor."

The general schema according to which elements constituting a term (i. e. operators and restrictors) get their syntactic status is schema (5) where the commas indicate that the phrase is at this step of its formation an unordered set: (5)

{Det, Head, (Modifiers)}

288

Ahmed Moutaouakil

3. From operators to determiners The category of term operators includes, among others, defmiteness, demonstratives, question-terms, quantifiers and numerators. The defmiteness operator is generally realized in the form of an article. In Arabic, this article is the prefix al when the term is definite and the suffix (un, an, in) when the term is indefinite.^ Here are some examples of these two kinds of articles: (6)

(7) a.

b.

c.

ra? ay tu l-walada saw-I the-boy-acc Ί saw the boy.' ja?a waladun came boy-nom Ά boy has come.' ra?aytu waladan saw-I boy-acc Ί saw a boy.' l-taqaytu bi waladin met-I with boy-gen Ί met a boy.'

The rules responsible for the realization of the defmiteness vs. indefmiteness operator in Arabic can be formulated in the following way: (8)

d^ al i — Г(и)п~ (a)n L(i)n J

The formal realization of demonstrativeness implies a combination of term operators: the demonstrative operator itself and the number and gender operators. One can formulate the rules responsible for the realization of the demonstrative term operator as follows: (9) a. b. c. d. e. f

Dem Dem Dem Dem Dem Dem

1 m—hâdâ, dâlikâ 2 m—» hädäni, hädayni η m — hä?ulä?i, ?ulâ?ika 1 f — hädihi, tilka 2 f —• hätäni, hätayni nf hä?ulä?i, ?ulä?ika

where Dem = demonstrative; 1 = singular; 2 = dual; η = plural; m = masculine; f = feminine.

Arabie term-to-phrase mapping rules

289

Rules ( 9 a - f ) account for the occurrence of demonstrative operators in sentences like (lOa-f): (10) a. ?ufaddilu {hâda/dalika} l-kitäba prefer-I this/that the-book-acc Ί prefer this (that) book.' b. Lam ?ara hädavni r-rajulavni not saw-I those-acc the-men-acc Ί haven't seen those two men.' c. ипфгг ?ilâ {hâ?ulâ?i/?ulâ?ika} look at these/those 'Look at these (those) men!'

r-rijâli men

d. '^asiqtu {hâdihi/tilka} l-fatâta loved-I this/that the-girl-acc Ί loved this (that) girl.' e. najahat hâtâni t-tälibatäni succeeded those-nom the-students-nom 'Those two students have succeeded.' f . la ?a'=rifu {hä?ulä?i/?ulä?ika} n-nisä?a not know-I these/those the-women-acc Ί don't know these (those) women.' In so-called wh-questions, the question terms are represented, at the fiinctional structure level, by means of term-positions containing the term operator Q. According to this analysis, the fimctional structure of (11) is (12): (11)

mäduä ?akala Zaydun ? what ate Zayd-nom 'What did Zayd eat?'

(12)

(Int (Perf (Pass ?ака1ау (dlmx·: ZaydN (x')) Ag subj Top (Qx^:(food)(x2))G„objFoc)))

By the application of the appropriate expression rules, the questioned term is realized as an interrogative pronoun. The questioned term in (12), for instance, is converted into the interrogative pronoun mâda, as we can see from

(11). In Arabic, the rales responsible for the morphological realization of questioned terms are sensitive to: a) the selection restriction (animate vs. nonanimate, human vs. nonhuman, ... etc.) imposed on the term-position, and b)

290

Ahmed Moutaouakil

the syntactic (or semantic) fiinction this term-position bears. These two sets of information are coded, as shown by (12), in the functional structure. The rule accounting for the realization of a questioned term as mâdâ, for example, can be formulated in the following way: (13)

(Q xi: non-human (xi)) subj/obj FOC — (xi: mâdâ (xi)) subj/obj FOC

Let us tum now to the problematic status of quantifiers and numerators. In English, quantifiers and numerators are, obviously, term operators, conforming to the general definition proposed in Functional Grammar for characterizing this category of elements. As expected, they are syntactically realized as determiners in the phrases within which they appear, as we can see from sentences (14a — b) and (15a — b): (14) a. / have read all these books. b. Some cities were destroyed. (15) a. Mary bought only one coat. b. John gave Mary three flowers. As for Arabic, quantifiers and numerators exhibit a quite different syntactic behaviour. They constitute the heads of the phrases within which they occur. Like ordinary nominal heads, they assign the genitive case to the first nominal modifier in so-called genitive constructions. Compare: (16) a. jâ?a ?аЬй Zaydin came father-nom Zayd-gen 'Zayd's father has come.' b. jâ?a kullu r-rijâli came all-nom the-men-gen 'All the men have come.' c. jä?a tßlätatu rijâlin came three-nom men-gen 'Three men have come.' More importantly, they bear the case assigned to the phrase containing them: (17) a. "ada kullu l-?asdiqü?i came back all-nom the-friends-gen 'All the friends have come back.'

Arabic term-to-phrase mapping rules

291

b. ra?aytu kulla l-?asdiqa?i saw-I all-acc the-friends-gen Ί saw all the friends.' c. l-taqaytu bi-kulli l-?asdiqâ?i met-I with-all-gen the-friends-gen Ί met all the friends.' (18) a. fawqa l-maktabi talätßtu kutubin on the-desk-gen three-nom books-gen 'There are three books on the desk.' b. staraytu talatata kutubin bought-I three-acc books-gen Ί bought three books.' c. wajadtu l-halla fl talätati kutubin found-I the-solution-acc in three-gen books-gen Ί found the solution in three books.' If we deduce from these observations that quantifiers and numerators are heads of the phrases within which they appear, then how can we represent them at the predicative (i. e. logicosemantic) structure level? Only the following two possibilities, as far as I can judge, can be considered: (i) One can regard these two kinds of words as restrictors (more specifically first restrictors) which become heads of phrases at the constituent structure level, as predicted by rule (25). According to this view, the terms kullu r-rijali and talätatu rijälin in sentences (16b — c) can be represented as (19) and (20) respectively: (19) (20)

(dmxi: kull (xi) : rijalw (xi)) (dmxi: talätat (xi): rijälN (xi))

(ii) One can also regard them as term operators quantifying or numerating the set of entities referred to by the whole term which contains them. If one were to adopt such an analysis, the structures of the terms occurring in the sentences (16b — c) would be something like (21) and (22): (21) (22)

(Vdm x': rijaWx')) (i3m χ: rijälN (x'))

In order to account for the syntactic behaviour of these term operators, one could formulate a phrase constituency rule which allows some operators in some languages (like Arabic) to become heads of the phrases which correspond to the terms containing them.

292

Ahmed Moutaouakil

I am convinced that both these analyses — which I propose to call restrictor analysis and operator analysis — are far from adequate. However, it seems to me that the operator analysis, although it overly complicates the term-phrase grammar, is less counterintuitive than the restrictor analysis where an element having the logicosemantic properties of a term operator is artificially given the status of a restrictor.^ In any case, having for the moment, no means to favour one of the two competing analyses, I will not go into the discussion of this problem in more detail.

4. Internal word ordering Term operators and restrictors form a structure which is logicosemantically ordered. By logicosemantically ordered I mean that these elements are ordered according to their restricting role with respect to the set of entities referred to by the term. More explicitly speaking, the less restricting element occupies the first restrictor position; the more restricting element occupies the last restrictor position. This logicosemantic ordering is not necessarily preserved when terms are mapped onto phrases, which means that we must formulate some principles and rules that enable us to account for the order of the elements occurring in a phrase, i. e. determiners, head, and modifiers. First of all, one must distinguish between two types of languagesprefield languages and postfield languages (cf Dik 1989: 345-355). In prefield languages modifiers precede the head, whereas in postfield languages they follow it. Thus, the internal ordered structure of a phrase conforms to one of the two following patterns: (23) a. {Det Mod Head} b. {Det Head Mod} It is well known that Arabic is a postfield language^ in which the elements occurring within phrases are ordered according to the pattern represented in (23b). The comparison between sentences (24a - b) and (25a - b) illustrates the point: (24) a. näma t-tiflu s-sagTru slept the-boy-nom the-little-nom 'The little boy has slept.' b. ntuqida maqalu Hindin was criticized article-nom Hind-gen 'Hind's article was critized.'

Arabie term-to-phrase mapping rules (25) a. *nama s-sagTru

293

t-tìflu

slept the-little-nom the-boy-nom b. *ntuqida Hindin maqâlu was criticized Hind-gen article-nom With these observations in mind, we can formulate the rule dealing with the patterning of the constituents occurring in phrases in Arabic (and probably in all postfield languages) as follows: (26)

{Det, Head, Mod} — {Det Head Mod}

In order to account in a satisfactory manner for the order of the constituents within phrases, we must also specify the parameters determining the order of determiners and modifiers of those phrases which contain a combination of determiners and more than one modifier. As far as Arabic is concerned, the examined data suggest that in a phrase containing as determiners both a demonstrative particle (i. e. hâda, hâdihi, ... etc.) and a definite article (i. e. аГ), the former determiner obligatorily occupies the first determiner position as is shown by the contrast between the following two sentences: (27) a. lam ?antaqid hada l-kitäba not criticized-I this the-book-acc Ί haven't criticized this book.' b. *lam ?antaqid al hâda l-kitäba not criticized-I the this the-book-acc Taking for granted that the term operators which surface in Arabic as determiners not suffixed to the nominal head are demonstrative particles and definite articles, we can formulate the rule describing the order of these two determiners as follows: (28)

{Dem, Def Art} — {Dem Def Art} where Dem = demonstrative; Def Art = definite article.

As for modifiers, they are ordered within the phrase according to two main parameters: a logicosemantic parameter and a syntactic (categorial) parameter. In general, the elements which become modifiers of a phrase keep the positions they had as restrictors in the underlying term. The fact that the semantically motivated order of restrictors is reflected by the syntactic con-

294

Ahmed Moutaouakil

figuration in which they appear as modifiers becomes clear from the comparison of terms (29a) and (30a) and their syntactic realizations in sentences (29b) and (30b): (29) a. (dnfx' ; mudunN (x'): magribiyyatN(x') : sahiliyyatN(x')) b. ?uhibbu l-muduna l-magribiyyata s-sahiliyyata like-I the-cities-acc the-morocan-acc coasting-acc 'I like the Moroccan coastal cities.' (30) a. (dlmx': kitâbN (x') (dlmxJ; ? u s t â ^ (xJ): (dlfx'': HindN(x'')))) b. staraytu kitäba ?ustadi Hindin bought-I book-acc teacher-gen Hind-gen Ί bought Hind's teacher's book.' This semantically motivated order is neutralized, however, when restrictors differ with respect to their categoria! complexity. In this case, the most complex restrictor occupies the final position in the phrase, as we can deduce from the comparison of (31a) and (31b): (31) a. ja?a

r-rajulu l-karTmu

Ι-Ιαφ: tahaddatnä '^anhu

came the-man the-generous-nom who

talked-we

about-him

'The generous man about whom we talked, came.' b. *jä?a r-rajulu Ι-Ιαφ: tahaddatnä '^anhu came the-man-nom who talked-we about-him l-karmu the-generous-nom This phenomenon is dealt with within the Functional Grammar framework by the general ordering principle referred to by LIPOC (i. e. Language Independent Preferred Order of Constituents). According to this principle the categorially complex constituents prefer to be placed after the categorially simple ones. A clause modifier, for instance, cannot precede an adjectival modifier as is clear from the contrast (31a) vs. (31b).

5. Case marking Let us tum now to the other subset of phrase constituency rules, those rales which are meant to account for the syntactic relationships between phrases and the predicate of the sentences within which they occur. As already men-

Arabie term-to-phrase mapping rules

295

tioned, this subset of rules deals with the (abstract)® cases which may be assigned to phrases. Cases are assigned, within the Functional Grammar framework, according to the following procedure: an abstract case (i. e. Nominative, Accusative or Genitive) is assigned to a phrase according to its functional status, i. e. according to the (semantic or syntactic) function it bears in the underlying functional structure. The fimctionally motivated case is assigned to the whole phrase as is clear from the case-specified structure (32b) of sentence (32a): (32) a. takassara ka?su Hindin broke cup-nom Hind-gen 'Hind's cup broke.' b. [Ass [Ind [Perf [Pass takassary {(ka?s) (Hind)} proc subj Top]]]] Nom However, the constituent on which the case is realized is the head of the phrase, as I have shown above. Therefore, we need a case rule which enables us to account for the percolation of the case bome by a phrase to its head. In Moutaouakil (1990) I proposed to formulate the case percolation rale in the following way: (33)

{Det Head Mod} — {Det

l·^

Mod}

Ω Ω where Ω = some abstract case The case of the phrase being assigned to the head according to rale (33), we must be able to explain the case marking of the constituents which function as modifiers. As far as Arabic is concerned, the cases which can be assigned to modifiers are of two sorts: agreement cases and functional cases. By agreement cases, I mean the case assigned to a constituent В by agreement with the case bome by a constituent A, A and В occurring'in a configuration where В syntactically depends on A, i. e. is a dependent of A. By functional case, I mean the case assigned to some constituent according to its functional status, i. e. to the (semantic or syntactic) function it bears in the underlying fiinctional stracture. These two kinds of cases are assigned to modifiers according to the following procedure: (i) A modifier is assigned an agreement case when it is an adjective as we can see from the case-specified stractures (34a — b), surfacing as sentences (35a) and (35b) respectively:

296

Ahmed Moutaouakil

(34) a. [Ass [Perf [Pass ^asiqv {dtu} Ag subj Top { ( 1 - f a t ^ (1-jamIlatA)} Go Obj Foc]]]

^ ^ Acc

'

\

Acc

'

Acc

b. [Ass [Perf [Past najahv {(l-fatâtN)(d-dakiyyatA)} Ag s^ Top]]] I » Nom Nom t ι (35) a. '^asiqtu l-fatäta l-jamTlata loved-I the-girl-acc the-beautifiil-acc Ί loved the beautiful girl.'

Nom

b. najahati l-fatâtu d-daìdyyatu succeeded the-girl-nom the-clever-nom 'The clever girl has succeeded.' In structures (34a) and (34b), the case assigned to the whole phrase (Object phrase and Subject phrase respectively) is copied on its head; this case is borne by the adjectival modifier, which agrees with the head. (ii) When the modifier is a noun, it is assigned a semantic (and possibly a syntactic) internal function with respect to the head. In so-called possessive constructions, the second restrictor (i. e. the restrictor which will become a modifier) receives in the underlying functional structure the semantic function Possessor as is clear from the ftinctional structure (36b) of sentence (36a): (36) a. labistu m f taf a Zaydin put-on-I coat-acc Zayd-gen Ί put on Zayd's coat.' (36) b. [Ass [Perf [Pass labisy (dix': tu (x')Ag subj Top (dimxj: mi'^tafN (xj) (dlmx·^: ZaydN(x'') Poss^O obj F O C ] ] ] In Arabic, the modifier which bears Possessor fianction is typically assigned the genitive case. The case-specified structure of sentence (36a) is thus (37): (37)

[Ass [Perf [Pass labisy {tu}Ag Subj Top {(mi'^taf) (Zayd) POSS}go obj foc ]]] Acc k

Gen

Acc I

Nominalized predications functioning as terms with respect to a superordinate predication differ from verbal predications in that they must model their ex-

Arabic term-to-phrase mapping rules

297

pression after ordinary nominal terms, i. e. in that they acquire typical nominal properties.^ In Arabic, the predicate of a nominalized verbal predication is a nominal predicate (i. e. Action noun. Agent noun, or Goal noun) which has lost some of its basic verbal features (i. e. Mood, Aspect and Tense features) and acquired the main features generally characterizing ordinary nominal terms. It also determines the (genetive) case of its nominal modifier as ordinary nominal heads do. The term qatla Halidin Hindan in sentence (38) illustrates the point: (38)

'^alimtu qatla Hâlidin Hindan knew-I killing-acc I^lid-gen Hind-acc Ί knew the killing of Hind by Halid.'

However, like basic verbal predicates, a nominalized predicate keeps the property of taking arguments which can be assigned semantic and syntactic functions. For instance, qatla in (38) is a two-place predicate taking as arguments an Agent-Subject and a Goal-Object. It thus behaves like its verbal counterpart qatal as is clear from the comparison between (38) and (39): (39)

qatala Halidun Hindan killed Hälid-nom Hind-acc 'Hahd killed Hind.'

On the basis of these observations, we can represent the underlying fimctional structure of sentence (38) as follows: (40)

[Ass [Perf [Pass "^alimy (dlx':tu (x^) Po Subj Top (dlmx^: {qatlN (dlmx': HalidN (x')) Ag Subj (dlfxj:HindN (xj)) Go Obj} (x^)) Go Obj Foe]]]

Bearing semantic (and possibly) syntactic functions, the arguments of a nominalized predicate become accessible to case assignment. In nominalized predications, cases are assigned according to the general procedure adopted in Functional Grammar, i. e. according to the functional status of the arguments. This procedure faces, however, a rather serious problem: nominalized predicates acquire among other nominal properties, the property of having one argument bearing the genitive case, that is a case which does not conform to its functional properties represented in the underlying functional structure. The first argument (x') of the nominalized predication in structure (40), for example, should be assigned a nominative case according to its functional status (i. e. to the syntactic function it bears). Contrary to one's expectation, this argument is formally expressed in the genitive, i. e. in a case which is

298

Ahmed Moutaouakil

generally associated with the Possessor semantic function. In order to deal with this phenomenon, one could argue that the argument formally expressed in the genitive case in this kind of constructions acquires the functional properties of the so-called Possessor argument. Such an analysis would be backed by the general hypothesis that nominalized verbal predications are under pressure to adjust to the pattern of nominal tems. It seems to me that as far as Arabic is concerned, it would be counterintuitive to regard the modifier expressed in the genitive in nominalized predications as a Possessor. The nominalized predicate, unlike the nominal head of so-called Possessive constructions keeps the verbal property of designating a state of affairs (i. e. an Action, a Process, a Position, or a State). Accordingly, its arguments automatically bear semantic (and syntactic) functions with respect to the state of affairs it designates. More specifically, the argument expressed in the genitive can be an Agent-Subject, a Recipient-Object or a Goal. In sentences (41a), (41b) and (41c), for example, the arguments Halidin, "Amrin and lmäla get, with respect to the nominalized predicates duhalu and ?(^ia?u, the fimctions Agent-Subject, Recipient-Object and Goal respectively: (41) a. sarram duhûlu HälTdin pleased-me going-in-nom Halid-gen 'Halid's entering has pleased me.' b. balagam ?fta?u '^Amrin l-mäla reached-me giving-nom '=Amr-gen the-money-acc Ί was informed about giving the money to ''Amr.' c. balagam ?ftâ?u l-mali '^Amran reached-me giving-nom the-money-gen '^Amr-acc Ί was informed about giving money to '^Amr.' With these observations in mind, I will assume that the genitive case in which one of the arguments of nominalized predicates is formally expressed is not a functional but a structural case determined by the (syntactic) configuration in which the argument in question appears. This means that the first argument of nominalized predicates is assigned a fimctional case which is masked by the genitive, this case being the typical formal expression of the first modifier of nominal terms. If we relate this claim to the adjustment hypothesis mentioned above, then it would mean that the arguments of nominalized predicates keep their fimctional and case properties, except for the first argument which, on surface, adjusts^ to the case model of basically nominal terms in that it gets the case typically borne by the nominal modifier of this kind of terms.^

Arabic term-to-phrase mapping rules

299

6. Conclusion Like all logicosemantically based grammars, Functional Grammar must be provided with a set of rules functioning as a mapping device which permits to convert logicosemantic underlying structures to actual linguistic expressions. In this study, I have tried to contribute to the formulation of a limited subset of these rules, the rules which map terms onto phrases. On the basis of the analyses previously proposed within the Functional Grammar framework, I have outlined the form (and the conditions of the application) of the Head selection rule, the rales specifying the realization of term-operators, the rales responsible for the internal ordering of phrase constituents and the mies accounting for the assignment of cases to and within phrases. The conclusions which can be drawn from this study are the following: (i) (ii)

(iii)

(iv)

(v)

Term-operators are realized as determiners; numerators and quantifiers, however, exceptionally function as heads. Among the restrictors forming a term, the first restrictor is the best candidate for constituting the head of the corresponding phrase; all the other restrictors become modifiers. In Arabic - which is a verb-initial language, thus a postfield language — all modifiers follow the head whatever their syntactic category is: the intemal order of phrases in this language (and presumably in all "real" postfield languages) conforms to the pattern: Determiner(s)-Head(s)-Modifier(s). As regards modifiers themselves, they keep their underlying logicosemantic order, except when they differ from each other with respect to their categorial complexity, in which case the more categorially complex modifier occupies the last modifier-position within the phrase. This property is captured, within the Functional Grammar framework, by the general principle of Language Independent Preferred Order of Constituents. According to their functional status with respect to the predicate of the sentence, phrases are assigned (abstract) cases which generally surface as morphological case markings. The case assigned to the entire phrase percolates to the nominal head. As for modifiers, they get agreement cases when they are adjectival modifiers and functional cases when they are nomináis. Adjectival modifiers inherit the case assigned to the head; nominal modifiers are assigned the cases required by their own semantic (or syntactic) functions. In basically nominal terms, the nominal modi-

300

Ahmed Moutaouakil fier is assigned the genitive case according to the semantic function it bears, i. e. Possessor function. In nominalized predications functioning as terms, the arguments keep their basic functional properties and, accordingly, are assigned cases corresponding to their semantic (or syntactic) functions. The first argument, however, tends to adjust to the case model of basically nominal terms: its functionally determined case is generally masked by the genitive case, which is typically assigned to the nominal modifier bearing the Possessor function in so-called possessive constructions. In nominalized predications, only one argument can be formally expressed in the genitive. Except in some highly marked constructions, it is the first argument (i. e. the Agent-Subject argument) that occupies the genitive position.

(vi)

(vii)

Notes 1 Predicative structure, Functional structure, and Constituent structure are the labels I have adopted so far in my writings (Moutaouakil 1984, 1985, 1989) to refer to the representations indicated in the FG literature (cf Dik 1978, 1980a, 1980b) by the standard terminologies predication, fully specified predication, and syntactic form. The motivations behind the use of these labels are given in Moutaouakil (1984), where it is stressed that this terminological difference has no theoretical import. Notice that in the recent version of FG (Dik 1989) the two structures labelled predication and fully specified predication {Predicative structure and Functional structure) are collapsed in one structure (called underlying clause structure) in which is coded all the information on which Expression rules operate in order to build a linguistic expression (i. e. what 1 call Constituent structure). 2 In fact, the suffix meant here is η (i. е. 'nounation' as it is commonly called in the Arabicist literature). I assume that this suffix is an article by means of which the indefiniteness operator is formally realized. This claim is confirmed by the well-known fact that η and the definite article cannot co-occur, as is shown by the ungrammaticality of sentence (ib): (i) a.

b.

saribtu l-qahwata drank-I the-coffee-acc Ί drank the coffee.' *saribtu l-qahwatan drank-I the-coffee-acc

3 It follows from the definition of restrictors that they must be referring expressions, i. e. expressions which refer to entities in some world. The restricting process can take place only if this requirement is satisfied. Now quantifiers and numerators obviously are not referring expressions for they do not refer to any entity in any world. Therefore, it would be difficult to regard them as restrictors. 4 The properties of these two types of languages are described in detail in Dik (1983) and Dik (1989). 5 Verb initial languages are generally postfield languages where the head (i. e. Verb or Noun) occupies the initial position in its domain.

Arabic term-to-phrase mapping rules

301

6 In my work on Functional Grammar of Arabic, two kinds of cases are distinguished: abstract cases and concrete cases. Abstract cases (i. e. nominative, accusative, genitive) are those cases assigned to phrases according to the ftmctional information coded in the underlying structure. Concrete cases are morphological (i. e. suffixed marked) realizations of abstract cases. The relevance of this distinction lies in the fact that, in Arabic, abstract cases may not be moφhologically realized at all. Furthermore, an abstract case may be realized by means of a suffix which does not correspond to it. On the basis of this distinction, two subsets of case rules must be formulated in Functional Grammar (in a Functional Grammar of Arabic at least): (a) rales which assign abstract cases to phrases according to their functional status in the underlying stracture, and (b) rales which specify the form in which abstract cases are realized. Rules belonging to the latter subset may be roughly formulated as ii: (ii)

NOM > u ACC > a GEN > i

7 The expression model of nonderived nominal terms is stated in (Dik 1985: 3) as follows: (iii)

Prototypical Expression Model (РЕМ) of Terms: "The РЕМ is the expression model for a term with a non-derived nominal as its head, possibly modified by attributive adjectives and possessor phrases and determined by one or more term operators."

8 The same analysis was adopted by Mackenzie (1985) in order to account for the case properties of the т о ф Ь е т е 's in English. 9 According to Ancient Arab grammarians, the first argument of nominalized predicates may in Classical Arabic be expressed in the case corresponding to its syntactic function. Thus, sentence (iv), where the first argument (Hindun) of the embedded nominalized predication is expressed in the nominative case, is grammatical in those grammarians's view, although it sounds highly marked: (iv)

?ahzanani qatlun Hindun Hälidan saddened-me killing-nom Hind-nom Hälid-acc 'It has saddened me that H. killed H.'

10 In Arabic, only one argument can be expressed in the genitive. This restriction leads to a certain competition between the arguments of n-place-nominalized predicates. However, it is the first argument (i. e. the Agent-Subject argument) that generally gets this case, as is clear from the contrast between the following two sentences: (v) a.

b.

?ahzanani qatlu Hindin Hälidan saddened-me killing-nom Hind-gen Halid-acc 'It has saddened me that Hind killed Halid.' ? ?ahzanani qatlu Halidin Hindun saddened-me killing-nom Halid-gen Hind-nom

Although they are grammatical, sentences like (vb) where the genitive position is filled by the second argument (i. e. the Goal-Object argument) do not occur, as far as I can judge, in contemporary texts.

302

Ahmed Moutaouakil

References Bolkestein, Α. Machtelt - Casper de Groot - J. Lachlan Mackenzie (eds.) 1985a Syntax and pragmatics in Functional Grammar. (Functional Grammar Series 1.) Dordrecht: Foris. 1985b Predicates and terms in Functional Grammar. (Functional Grammar Series 2.) Dordrecht: Foris. Dik, Simon C. 1978 Functional Grammar. (North-Holland Linguistic Series 37.) Amsterdam: NorthHolland. 1980a "Seventeen sentences: Basic principles and application of Functional Grammar", in: Edith Moravcsik — Jessica Wirth (eds.), 45-75. 1980b Studies in Functional Grammar. London: Academic Press. 1985 "Formal ^nd semantic adjustment of derived constructions", in: A. Machtelt Bolkestein - Casper de Groot - J. Lachlan Mackenzie (eds.) (1985b), 1-28. 1989 The theory of Functional Grammar. Part 1: The structure of the clause. Dordrecht: Foris. Hannay, Mike - Elseline Vester (eds.) 1990 Working with Functional Grammar: Descriptive and computational applications. Dordrecht: Foris. Mackenzie, J. Lachlan 1985 "Nominalization and valency reduction", in: A. Machtelt Bolkestein - Casper de Groot - J. Lachlan Mackenzie (eds.) (1985b), 29-^7. Moravcsik, Edith — Jessica R. Wirth (eds.) 1980 Syntax and semantics 13: Current approaches to syntax. New York: Academic Press. Moutaouakil, Ahmed 1984 "Le focus en arabe: Vers une analyse fonctionelle", Lingua 64: 115-176. 1985 "Topic in Arabic: Towards a functional analysis", in: A. Machtelt Bolkestein Casper de Groot - J. Lachlan Mackenzie (eds.) (1985a), 75-89. 1989 Pragmatic functions in a Functional Grammar of Arabic. Dordrecht: Foris. 1990 "Restrictive relatives in Arabic: A functional approach", in: Mike Hannay Elseline Vester (eds.), 51-72.

On the generation of English temporal satellite terms John H. Connolly 1. Introduction Time-related concepts and categories can appear in any of the main parts of the sentence recognised in FG. They manifest themselves within the verbal predicate via the systems of tense and aspect, and they often also find expression within terms. Temporal terms may be either of nuclear status, as in (1), or of satellite status, as in (2). (1) a. b.

Saturday was enjoyable. (Temporal term as subject) We enjoyed Saturday. (Temporal term as object)

(2) a. b. c. d. e. {.

They quite recently visited Wales. Last week they visited Wales. They stayed in Wales for a week. They visited Wales a year ago. When the clock struck they sprang into action. While filling in the form I noticed a misprint.

It may be noted that temporal satellites are of the "level two" type in Dik's hierarchy (1989b: 205-206). In English, satellite temporal terms may be realised by adverbial phrases, as in (2a), noun phrases, as in (2b), adpositional phrases (usually prepositional, as in (2c), though ago is postpositional, as in (2d)), or subordinate clauses (which may be finite, as in (2e), or nonfinite, as in (2f)). Furthermore, as these examples show, such terms may be placed in a variety of positions within the sentence. As well as being subject to the structural variety just outlined, temporal terms also show diversity at the semantic level. For example, the clause when the clock struck in (2e) refers to a point in time, while the expression for a week in (2c) indicates the duration of an event. It is, therefore, apparent that the generation of temporal terms by means of a FG is not a simple matter. Among the most important problems raised are the following: (3) a.

What factors determine whether a temporal term assumes nuclear or satellite status?

304

John H. Connolly b.

c. d. e.

What semantic subcategories of the function temporal need to be distinguished in order to allow the formulation of adequate expression rules? What constraints exist between the choice of temporal terms and the tense/aspect of the verbal predicate? What factors determine whether or not an adposition occurs within a particular temporal term, and if so, which adposition is chosen? How is the placement of temporal terms to be handled?

Some of these questions, namely (3a - c), pertain essentially to the predication component, while the others relate to the expression component. As space does not permit the discussion of all of them, this paper will confine itself to those which are relevant to the formal expression of temporal satellite terms, i. e. (3b) and (3d).

2. Semantic categories Employing the terminology used by Quirk et al. (1985: chs. 8-9), temporal terms may be classified into three main categories: (4) a. b. c.

Position in time. Duration. Frequency.

Terms representing position in time may refer either to just an instant, as in (5a), or to a longer period regarded as a single unit of time, as in (5b): (5) a. b.

The clock struck at one o'clock. The competition took place in 1980.

Terms of this kind generally answer the question 'when'. Durational terms, on the other hand, generally answer the question 'how long' and almost always refer to a span of time. Examples are found in (6): (6) a. b. c.

They stayed for an hour. They have been here since yesterday. They will stay until tomorrow.

In some cases, the period of time referred to is related to a fixed reference point. For example, in (6b) the reference point is the present moment and corresponds to the end-point of the period referred to by the temporal ex-

English temporal satellite terms

305

pression, which therefore represents a "backward span" (ibid). Again, in (6c) the reference point is the present moment, but in this case it constitutes the starting point of the period referred to by the temporal expression, which thus represents a "forward span" (ibid). On the other hand, (6a) furnishes an example of a durational expression with no definite reference point indicated. As for frequency terms, these generally answer the question 'how often'. Examples are given in (7): (7) a. b.

We go to FG conferences every two years, FG conferences are always interesting.

The three categories listed in (4) are essentially those recognised in Dik (1978: 50). The major question that now arises, is whether they are sufficient for the purpose of formulating adequate expression rules. Consider, by way of example, the temporal satellites in (8): (8) a. b. c. d.

The police The police The police The police

called called called called

at noon. on Monday. in the morning. by night.

All the temporal terms in (8) belong to the category of position in time, but each one is (and has to be) introduced by a different preposition. How can we devise expression rules which select the correct preposition if the expressions in question are all given the same semantic category label in the underlying predication? Similarly, consider again the examples in (6), all of which (as stated above) are assigned the category of duration. Once again, how can we formulate the correct expression rules, so as to choose the appropriate prepositions, if the underlying terms are not distinguished in respect of their semantic function? Faced with problems such as these, the obvious temptation is to propose a more delicate semantic subclassification of temporal satellites, so that the choice of preposition is automatically determined by the precise semantic function of the term concerned. For example, duration might be subclassified as "forward-span" as in (6c), "backward-span" as in (6b), and "nonspecificspan" as in (6a), thus making it much more straightforward to write expression rules which select the right preposition. The drawback of such a strategy, however, is that it could well lead to the postulation of so many subcategories that the end result would be a semantic description of temporal expressions that constituted little more than a rather unenlightening form of

306

John H. Connolly

"featurese". Consequently, it would be preferable to attempt to work within the bounds of the existing classification system if possible. Let us consider how we might proceed. Clearly, it is not going to be possible to formulate expression rules which stand a chance of working unless sufficient relevant information can be made available to those rules on the basis of the predications which act as their input, combined with the lexical information contained in the fund (see Dik 1980: 4-7). Furthermore, the information concerned will have to be either explicitly represented or formally deducible, at least if we are interested in developing the FG*C*M*NLU (Functional Grammar Computational Model of the Natural Language User); see Dik (1989a). It is, therefore, necessary to investigate how the requisite information may, indeed, be applied to the expression rales. Naturally, the contents of the predication will always play a significant part in determining the outcome of applying the expression rales. For example, if the predication refers to a time of day (e.g. noon, as in (8a)), then prepositions such as at or before may legitimately be inserted by those rales, whereas prepositions like in or on may not. However, the information represented in conventional FG predications is not guaranteed to be sufficient to ensure the precise choice of preposition required. Firstly, it may need to be supplemented by further lexical information, e.g. the fact that noon is a time of day (a fact which may be intuitively obvious but which a formalised grammar must state explicitly; in FG this will be done using a meaning postulate). Secondly, it seems unavoidable that some additional information will have to be included within the predication itself, over and above that which has hitherto been incorporated therein. Let us now investigate this issue more closely.

3. Temporal detail within predications Consider the problem of generating the following two sentences: (9) a. b.

The parcel arrived days ago. The parcel arrived after three days.

As a first approximation, let us propose (10) as the common element in the predications underlying both (9a) and (9b). The representation in (10) has been formulated with the help of the notation employed by Vet (1986) and

English temporal satellite terms

307

Hengeveld (1988), in which the variable Ci represents the state-of-affairs described in a particular sentence as an event situated in time and space: (10)

(PAST ei:[arrive(dlxi:parcel(x,))proc](ei): [(i3x2:day(x2))Time-pos](ei))

The question now is what information to add to (10) in order to cause it to be particular to (9a) or to (9b), rather than to serve for both of these sentences as it does at present. In the case of (9a), we need to specify that the term Хг represents a position in time prior to the present moment, i. е.: (11)

X2 < El

where Ej designates the deictic centre (cf. Comrie 1985: 36; Hengeveld 1988: 11; Dik 1989b: 37) and hence the present instant. (Let us call constructs like (11) "relational formulae". Note that (11) does not merely indicate a PAST state of affairs, but shows that an explicit comparison between X2 and Ei is being made). However, in the case of (9b), the important fact is that Хг is a position in time subsequent to some other (though in this case unstated) reference point Хз, i. е.: (12)

Хз < Х2

Accordingly, the predications underlying (9a) and (9b) may be formulated as (13a) and (13b) respectively, by adding the relevant relational formulae to each of them: (13) a. (PAST ei:[arrive(dlxi:parcel(xi))proc](ei): [(Í3x2:day(x2):x2 < EOxime-posKei)) b. (PAST e,:[...](e,):[(i3x2:day(x2):x3 < X2)Time-pos](ei)) The same approach can be applied to the development of appropriate, distinguishable underlying predications for other temporal satellite expressions. Some examples are given in the predication-expression pairs (14-17): (14) a. b. (15) a. b. (16) a. b. (17) a. b.

(FUT ei:[...](e,):[(i3x2:day(x2):E, < X2)Time-pos](ei)) The parcel will arrive in three days. (FUT e,:[...](ei):[(dlx2:(dlx3:noon(x3):x2 < X3))Time-pos](ei)) The parcel will arrive before noon. (FUT ei:[...](ei):[(dlx2:(dlx3:noon(x3):x2 < X3))Time-pos](ei)) The parcel will arrive by noon. (FUT ei:[...](ei):[(dIx2:noon(x2))Time-pos](ei)) The parcel will arrive at noon.

Note that in this last example, no relational formula is necessary.

308

John H. Connolly

The examples (14-17) all relate to expressions of position in time. However, expressions of frequency are amenable to the same kind of treatment, as illustrated in (18): (18) a. (PRES ei:[bring(dlxi:postman(xi))Ag (iNx2:magazine(x2))Go](e i ): [(dNxj : Saturday(x3)Freq](e i)) b. The postman brings magazines on Saturdays. Durational satellites, on the other hand, are slightly more complicated to deal with. Consider first of all the following sentence: (19)

The amnesty lasted until Sunday.

In (19) the period of time referred to by the temporal satellite is presented as stretching from one point in time (in this instance, an unspecified point in the past) up to a subsequent point (last Sunday). Thus, the expression introduced by the preposition until refers to the end point of a period of time. In order to capture such information within the corresponding predication, we may use the notation shown in (20) to signify the end-point of a period of time. (20)

(xOI

We may now formulate the predication underlying (19) with the help of this notation. (21)

(PAST ei:[last(dlxi:amnesty(x,))0](ei): [(dlx2:(dlx3:Sunday(x3):x3 = (х2)|))оиг](е,))

Similarly, we may employ the notation given in (22) to indicate the startingpoint of a period of time. (22)

|(Xi)

An example of the use of this notation can be seen in the following predication-expression pair: (23) a. (PERFei:[...](ei): [(dlx2:(dlx3:Sunday(x3):x3 = |(x2)))Dur](ei)) b. The amnesty has lasted since Sunday. Some durational expressions, for instance those introduced by for, simply refer to a timespan coextensive with that of the event concerned. The terms underlying such expressions can be formulated without any special notation: (24) a. (FUT ei:[last(dlx,:amnesty(xi))0](ei): [(i2x2:week(x2))Dur](ei)) b. The amnesty will last for two weeks.

English temporal satellite terms

309

4. Expression rules for the form of temporal satellites In the light of the proposals so far, we may now give an indication of some of the expression rules needed to ensure that temporal satellite terms are realised by surface-level expressions of the appropriate form. The rules to be presented are far from a complete and exhaustively detailed set, but they will show the lines along which a fuller treatment might be developed. They draw on the description in Quirk et al. (1985), but have been drawn up in such a way as to suggest how they might be implemented in a computing system. Owing to constraints of space, only the rules which apply to phrases, as opposed to clauses, will be considered. Let us begin with time-position terms with a noun as their underlying head. To deal with these requires an ordered set of rules after the fashion of (25), in which the TERM that appears in any given rule is assumed to be associated with the variable Xi: (25) a. Time-pos[TERM] = by TERM (if TERM is generic day or generic night, and does not contain a relational formula) b. Time-pos[TERM] = at TERM (if TERM is generic night, and does not contain a relational formula) c. Time-pos[TERM] = at TERM (if TERM is a point in time or a holiday period, and does not contain an item expressed by last, next, this, that or some, and does not contain a relational formula) d. Time-pos[TERM] = TERM (if TERM is a point in time or a holiday period, and contains an item expressed by last, next, this, that or some, and does not contain a relational formula) e. Time-pos[TERM] = on TERM (if TERM is a day or a day-division, and does not contain an item expressed by last, next, this, yesterday, today or tomorrow, and does not contain a relational formula, and if in the case of a day-division: TERM RESTRICTORS are not null) f Time-pos[TERM] = TERM (if TERM is a day or a day-division, and contains an item expressed by last, next, this, that, some, yesterday, today or tomorrow, and does not contain a relational formula, and if in the case of a day-division: TERM RESTRICTORS are not null)

310

John H. Connolly g. Time-pos[TERM] = in TERM (otherwise, provided that: TERM does not contain an item expressed by last, next, this, that, some, yesterday, today or tomorrow, and does not contain a relational formula) h. Time-pos[TERM] = TERM (otherwise, provided that: TERM does not contain a relational formula) i. Time-pos[TERM] = during TERM (if TERM is not a point in time, and does not contain a relational formula) j.

Time-pos[TERM] = TERM ago (if PREDICATION is PAST and not PERF, and X; < Ei) k. Time-pos[TERM] = in ТЕЬШ (ifEi t" entry and a "type => human" entry. It may contain more information, such as a nominal predicate, and operators etc., and the type may be more specific, for example, [human, female]. It is easily seen that the f-structure in (7) contains all the information present in the predicate frame. In fact, it contains more, such as the "type" and "cat" slots. This extra information is not superfluous; in traditional FG, this should be accounted for somewhere else. The f-structure, because of its flexible format, can easily incoφorate more information where the predicate frame would become clumsy. A typical example is morphological or semimorphological information. In German, as well as in many other languages, verbal predicates can consist of a root plus a separable prefix. The separable prefix belongs to the predicate, but is often expressed separately, so it has to be distinguished inside the predicate. With f-structures, this is done easily. For example, the verb einholen ('to overtake') becomes: (8)

(pred (stem "hoi-"; sepref "ein"; cat cat => p; type => [action]; sem ^ (ag (type [mobile]); go => (type => [mobile])))

v);

408

Hans Weigand

By splitting the predicate up in two parts, expression rules can address them separately.* A second example where the predicate frame can be enriched with more information when it is written in f-structure format, concerns things like gender and irregular forms. For example, the plural form of das Auge ('the eye') is Augen, which is an exception with respect to the plural form of neutral nouns. Instead of shifting this information to the expression rules, we can encapsulate it in the predicate frame as follows: (9)

(pred => (stem "aug-"; cat n; gdr => neut; forms => (plural "augen")); type [organ]; cat t)

In the expression rules (for plural operator) we then only need a rule that tests whether pred.forms.plural is empty or not. If it is not empty, the form listed should be taken, otherwise the general rule can be applied. This use of the lexical priority rule can also be incoφorated into the specification method so that it need not be repeated for each particular rule. For simplicity, we stick here to the explicit specification method, but uhimately a method incoφorating some kind of default inheritance is, although formally equivalent, much more userfriendly. In FG, the Fund is built up as a lexicon of basic predicate frames, such as the ones listed above, plus a number of predicate formation rales. A predicate formation mie is a lexical synchronically productive rale by which new predicate frames can be derived. In our notation, these rales are functions from fstractures to f-stractures. An example is nominalization, as in German (one simplified rale): ( 10)

NOM-1 ((pred => (stem => X; sepref =)• Y; cat => v); cat => p; sem => (F => A); type ^ T)) = (pred => (stem Y + + X + +"en"; cat => n; gdr => neut); cat ^ t; num => sing; sem (F => A); type => T)

Expression rules using f-stmctures

409

For example, the verb einholen is mapped to the term das Einholen. The noun is made from the prefix, the stem, and the suffix "en", and takes neutral gender. The semantic type of the term is inherited from the semantic type of the verb. The term frame is marked for singular since the nominalization does not have a plural form. The nominalization can also inherit an argument (Agent, Goal, Instrument,...) or restrictor from the predicate frame. Using the f-structure format for predicate frames and the subsumption relationship defined on them, we come to the following definition of the Fund. We assume a given Thesaurus as a finite set of predicate frames (see Weigand 1990: 85-86). Then the Fund is the smallest set of predicate frames such that Thesaurus IN Fund and if predicate formation rule R applies to some frame / then R(/) is contained in Fund as well. A complex predicate frame can be defined then as a frame/such that (i) (ii)

/ i s subsumed by some predicate frame from the Fund; and all subframesf\ w, for some address w i n / w i t h tail a semantic ñinction, are complex predicate frames.

Ignoring the disdnction we made in Weigand (1990: 114, 115) between predicate frame and predicate schema, we can now define a core predication (Dik 1989: 103) as a complex predicate frame. Furthermore, an embedded (instantiated) predication can be defined as a frame E such that E is the unification of some core predication and some instantiation frame. A possible term instantiation frame is: (var eO; cat t; num => sing; def frame (tmp =!• tO; loc => 10))

d;

(We refrain here from defining the possible instantiation frames - this can be done by using subsumption as well.) Higher levels of the underlying clause can be defined now on top of these structures. For example, we might define a proposition as a frame of category "prop" where the predicative part is an instantiated predication and the argument structure may contain an entry for the pragmatic function Topic (that must co-refer with one of the arguments of the predication, that is, the Topic must be a term with some semantic function assigned). The illocutionary frame can be defined as a frame with certain slots, one of which is the propositional content of type "proposition". As in Weigand (1990: 160), a distinction can be made between the level of the instantiated predication and the extended predication; the categories are pi and p2 respectively. In the same way, the term can be split up over two levels tl and t2 (cf Rijkhoffl988: 15).

410

Hans Weigand

3. Formal expression rules When the underlying clause is in f-structure format, formal expression rules can be expressed as subsumption rules. An example is the case marking of articles. In FG, formal expression rules are defined as functions, where the input is (part of) a predication, and the output some expression. It is also possible that the output is arrived at in stages, by means of auxiliary operators. Case operators such as Nom, Gen, Dat, Acc, are given in Dik (1989: 301, 302) as an example of such an auxiliary (or secondary) operator (in Latin). In the f-structure format, the formal operator rules are defined slightly different. In the first place, we let both input and output of the rule be f-structures. In this way, we avoid the problem of mixing up strings and structures. In the second place, we let the rule not replace the predication (part), but only augment it. This ensures that the rule can never be destructive, which makes the grammar more transparent and can be crucial for parser generation (Janssen 1989). To do this, we elaborate on the above-mentioned "auxiliary operators". In the third place, our notation makes it easier to formulate conditions on the rules. Subsumption was defined in section 2. We now first give a rule for case marking of the article without using auxiliary operators. (11)

(poss => (num => sing; pred.gdr => neut; cat => t; def ^ d)) < (poss => (article => "des"; pred => (head X + -b"es"; stem

X)))

This rule can be informally described as: suppose you have a subframe of type t (term), being definite, neutral and singular, in the context of a semantic function "possessive". Then it can be inferred that this subframe also contains a slot "article", filled in by the string "des", and that the head of the pred slot is composed of the stem of the pred slot and the suffix "es". Slots like "article" and "head" are called secondary slots. These are slots introduced by the grammar rules and corresponding to certain linguistic categories, but whose content is completely determined by more basic "deep-level" information. Using auxiliary operators, the same rule can be described as: (12) a. (poss => (cat => t)) < (poss => GEN) b. GEN (num => sing; def => d; pred.gdr neut) < (article "des"; pred => (head => X + + " e s " ; stem nm X))

Expression rules using f-structures

411

The first rule tells you that the term addressed by "poss" gets a GEN marking. The second rale tells you what slots a (neutral, singular, definite) GEN marked subfirame contains. Auxiliary operators are meant for capturing generalizations, if, for example, the genitive case occurs also with other functions than "possessive". Note that the rale does not tell you yet in what order the article and predicate head are to be expressed. It doesn't even say whether they will be expressed as separate words. We stress that subsumption rales are completely monotonie. They do not destroy or replace information in the predication stracture, but just augment it with some additional slots. The subsumption rales can be applied in any order and as many times as you want. If we assume that the subsumption rales do not introduce cycles in the frame (a reasonable restriction, since it also applies to the predicate stractures themselves), the process of subsumptive inferencing will always stop with some finite welldefined f-stracture. If the rales are inconsistent, the end result will be the bottom element. In the resulting stracture, all the elements of the surface expression are prepared and stored. What remains to be done is expressing these "prefabs" in the right order.

4. Linear precedence rules In the ID/LP grammar of Uszkoreit (1987) the grammar consists basically of three parts: a set of basic grammar rales (Immediate Dominance rales), a metagrammar component that generates derived grammar rales, and a couple of Linear Precedence (LP) rales that function as fibers on the grammar rales produced by the metagrammar. Only those rales are acceptable that satisfy all LP rales. Examples of LP rales are: (13) a. X < V -MC b. +PRONOUN

which we conveniently write as: (16)

(t) article < X (p) X < pred \ NOT main_clause

The first rale we already saw above, the second one says that the pred subframe (the verbal predicate) is expressed after all other arguments if the clause is not a main clause.

414

Hans Weigand

The definition of LP rales can be made even more powerful if we follow Uszkoreit's suggestion of complex LP rules. A complex LP rule consists of a set of simple LP rales. As an example, he gives: (17)

(+NOM +NOM +DAT -FOCUS +PRO

< < < <
"setz"; sepref => "ber"; cat cat => p; type [action, move, telic]; sem => (ag (type => [human]); go (type => [thing]); path => OVER(type => [thing]))) 1.2 (pred => (stem

" h e l f ; cat => v; forms

v);

(past ^ " h a l f ; pp =» "geholfen"));

cat => p; type => [action]; sem => (ag => (type => [animate]); go => (type => [animate]; case => DAT))) 1.3 (pred => (stem => "glaub"; cat => v); cat => p; type => [position, mental]; sem => (pos => [human]; comp => (type => [proposition]; clause => sub))) 1.4 (pred => (stem => "firau"; cat => n; gdr => fem); cat => t; type => @ mensch) 1.5 (pred => (stem => "mann"; cat => n; gdr => mase); cat =!· t; type @ mensch) 1.6 NOM-1 ((pred => (stem X; sepref => Y; cat => v); cat => p; sem => (F => A); type ^ T)) (pred => (stem => Y + + X + + " e n " ; cat => n; gdr => neut); cat =î> t; num => sing; sem => (F => A); type^T)

Expression rules using f-structures

2. Subsumption rules 2.1 (subj.cat = > t ) < (subj => NOM(cat

t))

2.2 NOM(num plur; < (pred => (stem pred.gdr fem)

X; head => X + + "en"))

2.3 (voice => active; < (sem.go => ACC) sem.go (cat t; case -)) 2.4 (voice => active; < (sem.go => Z:(case => Z)) sem.go (cat => t; case => +)) 2.5

(cat => p2) < (num => X; person => Y subj => (num ^ X; person Y))

2.6 (subj pred

X; < (pred.voice => passive) (go => X))

2.7 (num => sing; def => def; < (det =;· "der") pred.gdr => mase) 3. LP rules 3.1 (prop) topic < pred \ focus focus < X 3.2 (p2) vfín < X \ clause

-

main

3.3 (pi) X < pred 3.4 (tl) res < pred 3.5 (t2) pred < res 3.6 (t) det < X 3.7 (pi) (.NOM < .DAT; .NOM < .ACC; .DAT < .ACC; .(focus -) < .(focus +); .(pred.cat => pro) < .(pred.cat ^ n))

417

418

Hans Weigand

Comments In the lexicion the predicate frames are specified containing subcategorization information, irregular forms and semantic information. 1.6 contains one nominalization rule. The first subsumption rule, 2.1, specifies that subjects get nominative case. Rule 2.2. specifies the suffix for plural nominative feminine nouns; similar rules must be made for all cases and numbers. Rule 2.3 and 2.4 specify the case marking of goal arguments in active sentences. The case marking is either derived from the lexical entry (2.4.) or, by default, the accusative. Rule 2.5 is an agreement rule; it ensures that the verbal predication agrees in number and person with the subject (category p2 stands for the extended predication). Rule 2.6 ensures that w^hen the goal co-refers with the subject, the predication gets passive voice. This rule can also be considered as an integrity constraint. Rule 2.7 is one of the group of rules that specify the form of the noun phrase determiner. LP rule 3.1 says that the focus of a proposition is expressed first (in other words, comes at PI position). If there is no focus function, the topic is expressed first. Rule 3.2 says that in main clauses the finite verb must be expressed first (at extended predication level), whereas 3.3 specifies that the verbal predicate is expressed as the last one. The other rules concern the term (noun phrase). Rule 3.4 says that restrictors in the embedded term predication are expressed before the nominal predicate, whereas restrictors attached to the extended term predication, such as relative clauses, are expressed after (3.5). The determiner is expressed first (3.6). Rule 3.7 is the complex LP rule that governs the ordering of the arguments. Nominatives tend to be expressed before datives, and accusatives; constituents with focus are normally expressed more to the end; pronouns are expressed before nonpronouns. Any of these tendencies can be violated by force of another tendency. Notes 1 A more elaborate grammar of German should take into account that lexical material from the predicate frame can be incoφorated in the nominalization as well. For example: (i)

das Schnellerfahren 'driving faster'

(ii)

das Busfahren 'going by bus'

In these cases, we assume that the lexemes schneller and bus are not included in the predicate of the frame that serves as input for the nominalization, as with the separable prefixes.

Expression rules using f-stractures

419

but are part of the argument structure (cf. Wunderlich 1983; Uzkoreit 1987). Thus we have the following levels of attachment for verbal predicate frames: (1)

the verbal predicate is one unit Example: (pred => (stem "übersetz-"; cat => v); cat => p; sem (ag => [human]; go => [communication])) übersetzen in the sense of 'to translate'

(2)

the predicate consists of a stem plus a separable prefix Example: (pred => (stem => "setz-"; cat => v; sepref => "über"); cat => p; sem => (ag [human]; go => [material])); übersetzen in the sense of 'to transpose'

(3)

the predicate frame consists of a predicate plus other lexical material Example: (pred => (stem => "fahr-"; cat => v); cat => p; sem => (ag => [human]; go => (head => "bus"))

A difference between (1) and (2) is that the prefix part of the predicate can be expressed separately from the stem in finite forms. Another difference is that the infinitive form puts the particle zu ('to') in front of the predicate in (1) and between prefix and stem in (2) and (3). A difference between (2) and (3) is that the sepref in (2) cannot get an emphatic focus, whereas the lexical material in (3) can: (iii) (iv)

*Uber kann ich ihn nicht setzen. Ί can not transfer him' Bus fährt er selten. 'He seldom goes by bus'

To produce the nominalization Busfahren, the NOM-1 should be adapted such that pred.stem of the output becomes Z + + Y+ + X + +"en", where X is the stem of the verb, Y a possible sepref, and Ζ a possible lexical head of one of the arguments. The statement that in (1) the predicate is just a stem, does not rule out moφhological structure; this is not taken into account. A filli discussion of ordering principles of German is outside the scope of this paper. Uzkoreit (1987: 24) summarizes them (informally) as follows: focus follows nonfocus * unmarked order is SUBJ, lOBJ, DOBJ (but any order of arguments can be acceptable) * personal pronouns precede other NPs * definite NPs precede nondefinite NPs * light constituents precede heavy constituents * if a focused constituent precedes a nonfocused it will carry the focus accent the first position of the clause is special and can be filled by any consituent (marked as topic or focus For FGians, all these principles will sound very familiar as being instantiations of universal ordering principles spelled out already in Dik (1978). It should be noted that the introduction of an "article" slot is one way of writing the grammar. Another possibility is to change the subframe " d e f , which now just refers to a value definite/ indefinite, into a composite frame that not only specifies the definiteness, but also has a

420

Hans Weigand lexical head that can be filled with, for example, the genitive pronoun des. Then the LP rule could be formulated in a more semantic fashion as def < X

stating that (the expression of) the definiteness operator precedes everything (locally). Deciding between such alternatives, which for our fragment are fully equivalent, is beyond the scope of this paper. 4 It has been remarked that LP rules defined as such might act as a filter, which is to be avoided in FG. However, the same problem arises with the use of patterns and placement rules. What is important about filters is that such a device is not introduced to repair the erroneous overproduction of other modules. This is not at stake here, since the input for the LP rules is the product of a grammar process completely adhering to the principles of FG.

References Ait-Kaci, Hassan 1986 "Type subsumption as a model of computation", in: L. Kerschberg, (ed.), Expert database systems. Menlo-Park: Benjamin/Cummings, 115-140. Connolly, John 1986 "Testing Functional Grammar placement rales using PROLOG", International Journal of Man-Machine Studies 24: 623-632. Dignum, Frank 1989 "Parsing an English text using Functional Grammar", in: J. Connolly — S. C. Dik (eds.). Functional Grammar and the computer. (FG Series 10.) Dordrecht: Foris, 109-134. Dik, Simon C. 1989 The Theory of Functional Grammar, Part 1: The structure of the clause. Dordrecht: Foris. Janssen, Theo 1989 "Towards a universal parsing algorithm for Functional Grammar", in: J. Connolly - S.C. Dik (eds.). Functional Grammar and the computer, (FG Series 10.) Dordrecht: Foris, 65-75. Pollard, Cari - Ivan Sag 1987 Information-based syntax and semantics I. CSLI Lecture Notes. Stanford CA. Rijkhoff, Jan 1988 "A typology of operators", Working Papers in Functional Grammar 29. University of Amsterdam. Uszkoreit, Hans 1987 Word order and constituent structure in German. CSLI Lecture Notes. Stanford CA. Weigand, Hans 1990 Linguistically motivated principles of knowledge base systems. (FG Series 12.) Dordrecht: Foris. Wunderiich, Dietee 1983 On the compositionality of german prefix verbs, in: R. Bäuerle - С. Schwarze - Α. von Stechow (eds.). Meaning, use and interpretation of language. Beriin: De Grayter, 452^65.

Some formal and computational aspects of the Functional Grammar machine model Dik Bakker

0. Introduction In Bakker (1990) a computational model of FG was presented, called Functional Grammar Machine, in which a strict separation was made between theoretical concepts, the formalism to express specific rules for languages, the algorithms used to make a grammar "work", and implementation matters such as parsing strategies and the programming language and representation system used. In this paper I will discuss in more detail the formalism and its procedural aspects. This will be done in sections 2 and 3, respectively. But first I will consider some problems around modelling in linguistics in general, and within FG in particular.

1. Models and FG theory If we assume linguistics to be an empirical science, then a linguistic theory should provide us with a consistent and complete set of concepts and assumptions about language. It is only in a model of a theory, however, that the relation is established between those concepts and assumptions on the one hand and, on the other hand, empirical facts, i. e. sentences of some language. In all prevailing linguistic fi^ameworks, both in the formal and the functional paradigm, this is traditionally done by specifying the path that relates a sentence to its underlying syntactic or semantic representation(s), stated in the terminology of the theory concerned. The main function of a model is then to make the theoretical concepts operational, and ready for verification or falsification, thus providing the necessary link in the empirical cycle in which the theory is (deductively) tested and (inductively) improved. Therefore, models introduce a procedural aspect not, or not necessarily, present in the corresponding theory. In the case of a computational model, a second aspect is added: in order for a computer to simulate the procedures implicated by the model, all internal versions of the representations of underlying sentences must be fully formalized. Thus, models are in some respect more complete and more specific than theories normally are. On the other hand, however, in

422

Dik Bakker

all known models abstractions and restrictions are introduced, that make them less comprehensive than the theory they were derived from. I briefly mention the restrictions and abstractions most often made:' a) One language. Models almost never represent knowledge on more than one specific language, although their level of abstraction implies that they be interchangeable between languages. However, the relation between languages is excluded, while there is no (explicit) representation of typological and universal linguistic knowledge. b) No dialect or idiolect. Abstraction is made to the "ideal" speaker/hearer of some standard language. c) Grammar. Only information is represented that is traditionally referred to as grammatical knowledge. Pragmatic knowledge, often labelled "knowledge of the world", is generally excluded. d) Synchrony. Current models normally exclude knowledge of language change. e) Competence. With the exception of psycholinguistically oriented models (e.g. de Smedt 1990) knowledge of language use is not represented. f) Sentences. The sentence is taken to be the highest level of description/ representation: discourse aspects are not normally taken into consideration. g) Mode. Most models just establish the relationship between grammatical rules and (sets of) sentences, and do not distinguish between production and reception, and spoken or written language. These simplifications in relation to the theory deprive models of certain types of adequacy: with restriction (d) we may lose important parts of the model's explanatory power; with (a) we lose typological adequacy; (e) and (g) reduce the model's psychological adequacy; and (c) and (f) diminish pragmatic adequacy. In the FG literature, two models have been introduced, one noncomputational and one computational. These will be discussed briefly in section 1.1. In section 1.2 a third model will be considered, which meets some of the problems that the existing models exhibit.

1.1. Models in FG The noncomputational model was introduced in Dik (1978). Apart from the additions implied by the layered version of the underlying clause, it has not been changed fundamentally since. The version given in Figure I is fi-om Dik (1989b).

The Functional Grammar machine model

rfRAGMAlic^ Vfunctioi^

Figure 1. Outline of the FG model (Dik 1989b: 53)

423

424

Dik Bakker

Although it is based on a number of typological considerations, it essentially gives the modules and the internal organization of the grammar of one language L; it does not reflect explicitly typological and universal rules. In fact, all the abstractions mentioned above are made. As to its procedural aspects, some problems may be observed. Two concern the underlying clause. First of all, the proposed inside out construction of the underlying clause proposed in it runs counter to the scope relations and the corresponding constraints on values that exist between the operators of the respective layers. Secondly, pragmatic functions are assigned at a very late stage, i. e. when all material that they may be assigned to is present in the underlying clause. However, when psycholinguistic adequacy is taken into account, topical (and focal) information may be assumed to be about the first to be available.^ A third point bears upon the expression rales. These are thought to operate in a mn-of-the-mill fashion, i. e. without any interpretation or filtering device interfering. However, this causes problems in the case of gaps in the paradigms of predicates present in underlying clauses that have no semantic explanation. Many examples of this phenomenon are to be found in the Russian lexicon (cf Zaliznjak 1977). E.g. the verb exat 'go' has no imperative, and pisat 'write', zc/at 'wait', and paxat 'plow' lack gerand forms. A special case are inflected forms that are allowed, but have a distribution that deviates from that of the root. An example of this phenomenon in Russian is nosok ('nose' + dimin). The use of its oblique cases is restricted to 'nose of a shoe'. The noninflected form has no such restriction. All these will cause the expression rales to block, and thus take the form of a filtering device. The relations causing such nonexisting forms to be generated may be so complicated (some may be triggered by secondary operators that are introduced only during the execution of other expression rales) that excluding them from underlying clauses may call for a preran of the expression rales. A solution would be to admit a limited amount of filtering in the model, precisely for this type of cases. Another way out, which avoids filtering, would be to make predicates - or larger parts of underlying clauses - abstract to a certain extent, in order for the expression rales to have more options for expressing the contents of underlying clauses, without real interpretation. However, this would also ran counter to an important principle of the theory. A computational model of FG is presented in Dik (1989c: 4; see Figure 2). In contrast to the model just discussed, this is a model of the language user. As such it makes only (a), (b) and (d) of the above abstractions characteristic of theoretical models. Hesp (1990: 21, 39) signals several problems involved in this model, that will not be discussed here. These problems mainly concern procedural aspects of the model mentioned above in connection with the grammar model and psychological adequacy.

The Functional Grammar machine model

425

и

С

kíumladgm Ьашт predlemtionM 6·

λ· pbonetic raprmamtatiott

5»

8·

3 3» 4

rapreaaatatim of urittan form

prodicition

pradícatioa и

i—в»·

a*

Figure 2. Rough outline of C*M*NLU

1.2. The Functional Grammar Machine model Neither a grammar type model nor a model of the language user constitutes in itself a complete model of FG theory: the former abstracts from the language user, and fails to meet most of the adequacy requirements. The latter never caters for diachronic and typological rules that have no empirical соип1ефаП in the speakers of any language, which makes it fall short with respect to typological and explanatory adequacy.^ An integrated model is conceivable, but rather unattractive. It would be hybrid between two different theoretical concepts: language (or grammar) and its user, and would therefore be less transparent for empirical testing as, in fact, we would not know what we were modelling in the first place. Apart from that, the complexity of such a framework would render any computer implementation of it virtually infeasible. In modelling it seems best, therefore, to separate these two approaches, the grammatical and the psychological, and aim at two models, that will, of course, have a great deal in common. A model of grammar should then ideally capture aspects (a), (b) (dialect), (d), (f), and (g) above, while a user model should ideally take aspects (b) (idiolect), (c), (e), (f), and (g) into consideration. The Functional Grammar Machine model is of the first type. Its major extension relative to the grammar model of section 1.1 is the integration of grammatical knowledge on the typological and universal level. It takes the global form given in Figure 3. At the top level all grammatical entities, i. e. rules and structures, are represented that have a universal character. The bottom level contains all language specific entities. At the intermediate levels we find entities that.

426

Dik Bakker

и

L1

L2

L3

L4

L5

Figure 3. The FGM model

going from bottom to top, pertain to still larger subsets of languages, and which are of an increasingly abstract nature."* In a dynamic view, the framework embodies the empirical cycle referred to above. On the one hand, abstract, more theoretical entities may be inferred from the lower levels; this may lead to the instantiation of typologies and (absolute and statistical) universals. On the other hand, concrete, more specific entities may be deduced from higher levels, leading to the instantiation of grammars for specific languages. In section 3 some examples of both processes will be provided. But first we will have a look at the formal apparatus necessary for the representation of grammatical entities, and for the respective operations to be performed on them.

2. Formalization of FG entities Requirements one would like to impose on a formalism in cases like the present are that it be as simple as possible, and that it should stay close to the formalisms used in the literature. Hierarchic feature-value structures (henceforth hfv; these are related to f-structures, see Kay 1979) come close to this requirement, and may be instrumental in providing solutions for the shortcomings signalled above. For FG, they were proposed in Bakker (1989) and Weigand (1989). The basic form of hfv's is a label, followed by the equivalence operator = and a value. The latter may be either atomic, or a list of hfv's, or a variable. We may have: (1) a. b.

tense=present argument = [term=T, sem_function=ag, synt_ftinc=subject]

The Functional Grammar machine model

427

(la) is atomic, and (lb) is complex, the first element of its value having a variable as a value in its tum (I will follow the convention of using upper case for variables). In complex hfv's, the outermost feature will be called the label of the hfv. In order to accomodate FG-like representations, some extensions and additions must be made to this basic scheme, of which the most important will be mentioned below. 2.1. Disjunction At some stage in the development of underlying representations features may not yet be fully specified, but the range of values they may take may be restricted to a subset of the ones possible for that feature. An example is provided by the (provisional) definition of a well-formed term, where the semicolon stands for the logical (inclusive) OR: (2)

term=[def=D, num=N, head = [cat=noun, arg=[var=V]], restrictor=[[cat=adjective] ; [term=T] ; [relative=R]]]

This definition constrains the type of filler for the restrictor in a term to one of the three mentioned. There may be an undetermined number of restrictors of each type (see 2.3). 2.2. Operators In feature-value representations, apart from the equality operator (=), three other operators may be used. The greater than ( > ) and less than ( < ) operators, to be read as 'right from' and 'left from' respectively, may be used in the following situation. If, in a grammar, we have this definition: (3)

sem_fiinc=(ag; go; ree; ben; loc; temp)

then, in an underlying representation, (4a) is short for (4b): (4) a. sem_func > ag b. sem_ftinc=(go; ree; ben; loc; temp) It is required that (3) be defined in the grammar as an ordered set (see section 2.7). The # is used to signal unequality. Opposite to > and < it may be employed in the case of unordered and infinite sets, cf (5) a. b. c.

tense # present cardinality # 1 tense # Τ

428

Dik Bakker

In (5c) it is stated that tense should be different from the value or set of values that might be bound to variable T, at whatever stage binding might take place. It is a way of forcing different values upon two features, whatever those values may be (see also section 2.4). 2.3. Repetition In FG-like representations we have two ways of repeating entities. The first is the use of recursion: in (2) a term was introduced in the definition of a term. The second is by iteration, as in an extended version of (2): (2')

term=[def=D, num=N, head=[cat=noun, arg=[var=V] ], restrictor=[([cat=adjective] / [0, infinite]); ([term=T] / [0, infinite]); ([relative=R] / [0, infinite])]]

(2') states that, in a term, there may be any number of restrictors, of the three types mentioned. 2.4. Scope In underlying clauses there may exist interdependencies between features on different levels, to be expressed by scope relations between them. I assume that, by default, operators and satellites of some layer L have scope over all entities of layers embedded in L. Within L, operators have scope over satellites. In order to distinguish between embedded layers and features with complex values, the former have to be defined in the grammar: embedding (proposition —• extended_predication) The scope of some entity of layer L is never extended beyond a cycle. A cycle is started within L by any (recursive) feature that itself (in)directly embeds L in the grammar. For example, a relative sentence, itself an extended predication or proposition, starts a cycle within some term, that is in its tum part of some predication. This restriction ensures that features have a high degree of uniqueness within a cycle: we may speak of "the" tense, "the" main predicate, etc. Scope relations may be restricted by providing an explicit definition. To give some examples: (6) a. b.

scope (clause —• proposition) scope (satellite=[sem_fijnc=subj_modal] —• core_pred)

The Functional Grammar machine model

429

(6a) restricts the scope of clausal entities to the prepositional level; (6b) restricts the scope of some type of prepositional satellite. The scope of variables is restricted to the (complex) expression that introduces them. The following definition ensures that adjectival term restrictors will have the same argument as the nominal head of the term: (2")

term=[def=D, num=N, head=[cat=noun, arg=[var=V]], restrictor=[([cat=adjective, arg=[var=V]] / [0, infinite])]]

2.5. Hierarchy For some rules of grammar it makes sense to generalize over subsets of feature values. E.g. in FG-like grammars we find the concept of "first argument". This includes all semantic functions that may be assigned to the first argument of some predicate. It may be introduced by way of the following definition: (7)

sem_func=(first=(ag; posit; force; proc; zero); go; ree; ben; ...)

The character of the value "first" is an abbreviation for the subset of values that it defines. Value hierarchies may be of any depth. 2.6. Rules The combinations of feature values that may theoretically occur in a complex structure may be restricted by setting constraints on them. These may take the form of complex structures, as in (8a), or logical implications, as in (8b): (8) a. b.

constraint(predicate=[cat=(noun; adjective; verb), arg=(A/[l,4])]) constraint(illocution=imperative —• (volitional # (wish; hope), tense=present))

(8a) sees to it that any predicate in the lexicon (eventually produced by a predicate formation rule) will have one of the categories mentioned, and will have between one and four arguments. (8b) provides some constraints on operator values for imperative sentences. The (8a) type of constraint must be true for any entity with the label specified; the implication type must be true for all entities that obey the condition. Cycles introduce a lower bound for features with unrestricted scope. However, feature values may have influence on values within embedded cycles. In order to represent such relations, constraints should be used. But, in contrast to intracyclical constraints, we have to specify the path between the features concerned.

430

Dik Bakker

Rule (9) specifies that, in a relative sentence, there is at least one argument or satellite the term variable of which is the same as that of the term in which it is embedded: (9)

constraint((term=[term_var=Vl, restrictor= [proposition= [ext_pred= [core_pred=[nucl_pred= [argument=[term=[term_var=V2]]]]]]]]; term=[term_var=V1, restrictor=[ext_pred= [satellite=[term=[term_var=V2]]]]]) VI = V2)

Note that the second disjunct of the condition only refers to all satellites of the extended predicational level and lower. 2,7. Ordering There may be several reasons to assume that the values of some features in a grammar are ordered. There may be rules working on specific subsets of values, such as is the case for the Semantic Function Hierarchy (SFH), that gives a language-specific cut-off point in the universally ordered set of values for semantic functions, to the right of which no syntactic function assignment is possible. This gives rise to the following constraints, among others: (10)

SFH: Dutch: constraint(synt_ftinc=subject —» sem_func < ree) English: constraint(synt_func=subject —• sem_fiinc < ben) Tagalog: constraint(synt_func=subject —• sem_func < temp)

The fact that some feature has ordered values, allowing the use of the < and > operators, has to be acknowledged explicitly in the grammar. So we must define: ordered(sem_func). A special type of ordering is typological ordering, which represents the relative relevance of values in a grammar. An example is syntactic function, which may or may not be realized in a given language. A full blown definition for its values would give us: (11)

synt_func=(subject; object) typol_ordered(synt_ftmc)

This implies that, for any language, there is a specific cut-off point to the right of which values are not relevant. Typological ordering has a number of implications for the frequency of use and relative markedness of the values involved (cf Dik 1989b: 30 on typological irequency).

The Functional Grammar machine model

431

2.8. Other grammar entities Above only (parts of) underlying clauses were given as examples of the use of hfv's. There are, however, other entities of an FG-like grammar which may be also formalized by means of hfv's, such as predicate frames, predicate formation rales and syntactic templates. For the latter, see Bakker (1989), For reasons of space, we will give no examples of those here.

3. Procedural aspects Hfv's as an instrument for the representation of linguistic entities have two appealing characteristics: they look more or less like the usual FG representations; and their formal properties are rather well-known (cf Shieber 1986, and references there). Because of the latter characteristic, certain well-defined operations may be performed on them that establish relations between tuples of entities in hfv form. A procedure that rather naturally applies to hfv's is unification. In general, two hfv's unify if and only if none of their substructures contradict. Thus, the tuples of (14a) unify, while those of (14b) do not: (14) a. pred=[lex=unify, cat=verb] pred=[lex=L, cat=verb] b. pred=[lex=unify] pred=[lex=general, cat=noun]

U pred=[lex=unify] U pred=[lex=unify] U pred=[lex=generalize] U pred=[lex=general, cat=adj ]

When two stractures unify, there is precisely one most general unifier, i. e. a stracture that unites all material present in them. In the most general unifier, free variables will be bound with corresponding values. For both tuples of (14a), the most general unifier is pred=[lex=unify, cat=verb]. In general, a most general unifier is more specific than both of the input stmctures. From a static perspective, unification may be seen as a method of checking whether structures conform to constraints given in the grammar. This may be illustrated by the following example. In (15a) we find the representation of an underlying clause. (15b) is a constraint that is meant to block the combination of imperative illocution and several operator values on the propositional level. (15) a. clause=[illocution=imp, proposition=[volitional=wish, ext_pred=[...]]] b. constraint(clause=[illocution=imp] —• proposition=[volitional # (wish; hope)])

432

Dik Bakker

Since the premise of the constraint and the underlying clause unify, and the conclusion does not, (15a) is considered to be not well-formed. From a dynamic perspective, the repeated and ordered application of unification to the most general unifier resulting from some previous unification may lead to the construction of yet more specified underlying structures, such as fully specified underlying clauses and rule templates, and to the generation of sentences. For examples see section 3.1 below; 3.2 gives another application of unification. See Knight (1989) for a comprehensive introduction to unification. Another operation on hfv's that is in many respects the inverse of unification is generalization. When two structures generalize, there is exactly one most specific generalizer (MSG), i. e. a structure that is a more general version of both of them. The tuples in (16) generalize, with the corresponding most specific generalizer: (16) a. pred=[lex=generalize, cat=verb] pred=[lex=unify, cat=verb] MSG: pred=[lex=L, cat=verb] b. pred=[lex=general, cat=adj] pred=[lex=general, cat=noun] MSG: pred=[lex=general, cat=C] Section 3.2 gives an example of the application of generalization within the Functional Grammar Machine framework. 3.1. Sentence generation We may see the process of sentence production in an FG-like grammar as the stepwise specification of a fully unspecified underlying clause, resulting in a fully specified underlying clause, followed by the recursive specification of rule templates for yet more specific, linearly ordered constituents, finally leading to an ordered, prephonological tree.^ As was indicated above, unification may qualify for precisely this task: input to each application of unification are a partially unspecified underlying clause in the first phase and a ditto rule template in the second, on the one hand, and some specified entity that may fill a hitherto unspecified substructure of that underlying clause or rule template, on the other. As a side effect of each (successful) application, the most general unifier serves as the output, i. e. a more specific version of the input underlying clause or rule template. This, in its tum, is the input underlying clause or rule template for the following step. For standard unification to fit such a process, it has to be adapted in several ways.^ Standard unification applies to two structures that have the

The Functional Grammar machine model

433

same label. So if we would like to add, say, a term to some underlying clause with the label 'clause', then, in order for unification to apply at all, we should provide the unifier with that underlying clause and a second structure with 'clause' for its label, and the complete feature path to the term concerned. This would overburden the grammar to a large extent. However, if its label is not unique already, we may make the substructure to be inserted identifiable by providing it with the extra information necessary to disambiguate it sufficiently within the current cycle of the underlying clause. The following specifications for a (simplified) term render it yet more specific, and thus fit for a decreasing number of still unspecified term positions in some underspecified underlying clause: (17) a. term=[def=d, num=sg, lex=horse, var=V] b. arg=[term=[def=d, num=sg, lex=horse, var=V] ] c. arg=[term=[def=d, num=sg, lex=horse, var=V], sem_fimc=ag] (17a) may fit any (unspecified) term position, for both arguments and satellites, provided that it is in the scope of the clause. (17b) applies to arguments only; (17c) only to an argument specified for the agent fiinction. In case of ambiguity — which may transpire in the case of (17a) and (17b) - all versions will be generated successively. The respective entities must be specified such that precisely the sentence or set of sentences aimed at are produced. The generation process starts with a ftilly (or partially) unspecified initial underlying structure, typically an underlying clause or a part of it, such as an extended predication or an empty term, and any set of fully, or partially, specified substructures of it, such as terms and main predicates. The latter material is put on a stack, and will be used to further specify the initial structure. I assume at least the core predication, and noncore entities marked for topic or focus, to be present on the stack before any further processing takes place. The specification of the initial structure then proceeds in an outside-in fashion: the respective operators are assigned values, and the relevant constraints are applied. The specification of extra terms and embedded predications or propositions may be called during the construction process, if empty slots are encountered for which there is no material on the stack. In that case, the specification process is invoked recursively. As was menticmed above, on applying unification to a complex structure, the order of the substructures will not be changed; in case of conflicting orders, that of the first entity is taken. The resulting most general unifier will serve as the (more specified) underlying clause for the following application of unification. When all stack material has been used, and the underlying clause is fully specified, its material will be inserted into corresponding rule templates. Again, this is a series of recursive applications of unification. In this case.

434

Dik Bakker

the input structures are (the next empty slot of) a rule template and some substructure of the fully specified underlying clause that is about to fill it. The unifier is employed for the procedural tasks involved, i. e. the localization, comparison, extraction and insertion of hfv's. The procedural kernel of the unifier algorithm is therefore extended with these four basic procedures.

3.2. Grammar derivation and the inference of universale Within the framework of Figure 2, we may see the relation of a grammar G(L) for language L vis-à-vis the universal level as basically defined in terms of selection, restriction and (further) specification. Provided that the hfv formalism is used throughout for representations, unification then comes in as the natural operation for deriving G(L) from the U level, and generalization comes in as the operation that does the opposite, i. e. infers (more) general rules from the specific ones in the set of all G(L), thereby establishing typologically or universally relevant abstractions. Together these operations embody the empirical cycle mentioned above. In order to support these procedures, the и level should have the appropriate organization. Within the Functional Grammar Machine model, it takes the form given in Figure 4 below.

UNIVERSAL META RULES

\

/

SELECTION-RESTRICTION

Figure 4. O r g a n i z a t i o n o f t h e F G M u n i v e r s a l level

The Functional Grammar machine model

435

The 'elements' and 'structures' sets constitute the inventory of basic and complex structures that any grammar may contain. So in the former we find an exhaustive enumeration of all operators and functions together with their values as they may be found in any language. Cf for semantic fiinctions: (18)

sem_func=(semf_soa= (semf_nuc_pred=first=(ag; pos; fo; proc; zero); go; ree) ; semf_core_pred=(addit_partic=(ben; comp); means= (instr; man; speed; qual); spatial=(so; pa; dir))); semf_ext_pred=(temp; loc; ...); semf_propos=(volition; ...); semf_clause=(mitigation; ... ))

The complex structure set gives maximal definitions of underlying clauses, rule templates, predicate formation rules and lexical entries, specifying all elements that might be found as substructures for them in any language. An example is definition (2") for terms. Next to these inventories, which may be considered the building blocks of the theory, the two rule sets contain the (universal) constraints on what may actually be found in a specific language. Paradigmatic rules define what may, and may not, be found in a grammar; they determine what is a well-formed grammar. The majority of the universale in Greenberg (1963) are of this type, cf Universal 36. If a language has the category of gender, it always has the category of number.

In hfv format: (19)

constraint(gender=G —• number=N)

Syntagmatic rules determine what may, and may not, be found in an actual fully specified underlying clause or rule template; together they determine what is a well-formed sentence. Examples are the constraint of (9) and those below: (20) a. constraint(synt_fimc=X —• sem_fiinc = semf_soa) b. constraint(core_pred=[term=[synt_fiinc=subject, sem_fimc = X], term=[synt_func=object, sem_fiinc = Y]] ^ X < Y) (20a) is a very general rule that specifies that any syntactic function is only assigned to substructures with a semantic function that pertains to the level below the extended predication (see (18)). (20b) checks whether, if there is

436

Dik Bakker

both a subject and an object, the former is assigned to a term with a semantic ftmction "left" of that of the object term. Starting from these definitions, I will now show how the grammar rules for the SFH for a set of languages might be derived. Given the room available here, it will only be a very partial account of the phenomena concerned. I will only consider the subject, as it seems to be relatively straightforward in comparison to the object in FG. For English, it seems that we could do with a further restriction on (20a): (20') a. SFH - English (Indo-European): constraint( synt_iunc=subject —• sem_fiinc < ben) For some Amerindian languages such a simple scheme does not suffice. In Jacaltec, for example, there is the further restriction that transitive verbs may only have animate direct agents as subject in the active and in the (four versions of) passive (cf Craig 1976: 111). This could be formalized in the following way: (21) a. SFH — Jacaltec (Amerindian/Northem/Almosan^): constraint (nucl_pred=[arg=[synt_func=S, (term=[head=[arg= [anim#animate]]] ; sem_fimc > proc)]], arg=B], — S # subject) A similar case is that of Nitinaht, where person plays a crucial role in subject assignment. In this language, if a third person is to act upon a first person, the use of the passive is obligatory in order to let the first person term end up as the subject (cf Siewierska 1988: 50). This is expressed by: (21) b. SFH - Nitinaht (Amerindian/Northem/Penutian): constraint (nucl_pred= [arg=[term= [person=[speaker=yes, addressee=no]]], arg=[term=[person=[speaker=no, addressee=no], synt_func=S]]], —• S # subject) In Huichol, recipient and beneficiary, not goal, are promoted to subject in passive constructions with bitransitives (cf. Comrie 1982: 107). This infringe-

The Functional Grammar machine model

437

ment on the continuity within the set of semantic functions open for subject assignment may be represented as: (21) c. SFH - Huichol (Amerindian/Central): constraint (nucl_pred= arg=[sem_ftmc=goal, synt_ftinc=S], arg=[sem_fiinc=(rec; ben)] — S # subject) Given the typical nature of the terms that normally fill recipient and beneficiary slots, we may very well inteφret this as an example of animacy being required of a subject term; at least a certain amount of 'animacy' is associated with them. If these were all the Amerindian languages in our framework, a first hypothesis may now be that in this family, animacy and person are factors that codetermine subject assignment. This may lead to the following tentative typological rule: (22)

SFH - Amerindian: constraint (nucl_pred= [arg=[term=[animacy=A; person=P]], arg=[term=[animacy>A; person>P], synt_func=S]], — S # subject)

That the case of English is also more complicated than (20a') may be gathered from the following sentences: (23) a. That outlook / A mad dog frightened me. b. I was frightened by that outlook / a mad dog. The sentences of (23a) are rather marked whereas those of (23b) seem to be the unmarked way to express the corresponding clause in most contexts. So a better alternative to (20a') may be: (21) d. SFH - English (Indo-European): constraint (nucl_pred=[cat=verb, arg=[term= [def=D, anim=A]], arg=[term= [def > D, anim > A ,synt_func=S]]] — S # subject) This may be overruled if topic or focus is assigned to the respective terms.^ In addition to animacy and person, definiteness and pragmatic functions seem to play a role in subject assignment, too. All these factors cluster around such

438

Dik Bakker

notions as "empathy" and "familiarity" (cf. Kuno - Kaburaki 1977 and Allan 1987, respectively). In terms of FG we could say that these features express different types of centrality to the Speaker/Addressee that are specified on the clausal level. In all, a definition of SFH in terms of a number of such elements, including semantic functions, will most probably tum out to have more universal validity than semantic function alone will have. The semantic functions could, at least partially, be decomposed into some of the others. For that matter, we might henceforth read SFH rather as Syntactic Function assignment Hierarchy. We have started with the traditional version of SFH at the U level as embodied by (20a). On instantiating versions for the Amerindian languages we have introduced the more elaborate rules (21a-c), that led to the generalization of (23). Now with (2Id) added to them, we may go yet a step further. Generalization over (21a-d) leads to the following universal rule (assuming that these four languages are the only ones represented in the framework): (24)

SFH - Universal: constraint (nucl_pred= [arg=[term=[centrality=N] ], arg=[term=[centrality > N], synt_func=S]], S # subject) constraint(synt_func=subject —• sem_func=semf_soa)®

The 'centrality' feature is a generalization over the set of features involved in this process. In order for the generalizer to infer (24) from (21a-d) we therefore need the following definition: (25)

centrality=(pragm_func; anim; person; def; sem_func)

Any language would have its own subset of values instantiated for it.'" Now let us assume that (24) will do for the assignment of subject in the great majority of languages. These all conform to the definition of syntactic function as highlighting one of the terms that take part in the state of affairs expressed by some utterance. However, even though examples are hard to find, and of restrictive use, there are cases where subject may be assigned to terms outside the domain of the core predication. In Cebuano and Kalagan, both Philippine languages, subject may be assigned to temporal satellites, thus to a term of the extended predicate level, i. e. outside the domain of the state of affairs descriptors, and therefore in conflict with the FG definition of subject." This phenomenon, however, may well be explained by the working of another rule. In these languages, relativization may only be applied to

The Functional Grammar machine model

439

subject terms, i. e. relative pronouns obligatorily are assigned the subject function in the relative sentence. Indeed in the corresponding languages, assignment of subject to temporals only takes place in case these are relativized; in all other cases, the 'normal' constraints hold (cf. Dik 1989b: 230). Apparently, the relativization force is stronger than the other factors determining subject assignment. In order to leave the universal (24), which states an important aspect of FG theory, as it is, I introduce the following, language specific type of rale: (26)

SFH - Cebuano (Austrie/Centrai Philippines): extension (term=[var=V, restrictor = [relative=[ proposition=[extended_predication=[ satellite=[ term=[var=V, syntf= subject]]]]]]])

This rale assigns subject to any relativized satellite on the extended predication level and below. Extensions are ordered before the corresponding constraints, but are never generalized to a higher level of the framework: they are strictly local. It may, however, be among such rales that we find important changes in the stracture of language or grammar.

4. Conclusion In the foregoing, I have presented the outlines of a computational model of Functional Grammar, that includes both a universal level and that of the grammars of individual languages. I hope to have shown that a well-known formal framework like hierarchic feature-value stractures may be extended such that it might be used to formalize some nontrivial stractures and rales in an FG-like grammar in that model. Notes 1 Recent versions of models for Government and Binding, and its offspring, Lexical-Functional Grammar and Generalized Phrase Structure Grammar, are to be found in Sells (1985). Models for FG are discussed below. 2 In Dik (1989b) the model is presented as an outline of the theory that does not necessarily provide full procedural adequacy. Given the requirements attributed above to a fully-fledged model of a grammar, it must be seen as a stylized, not completely worked out model. 3 If we would assume the innate character of some (highly abstract) linguistic knowledge (or the existence of a Language Acquisition Device), at least some universals might be rep-

440

4 5 6 7 8

9 10 11

Dik Ваккег resented in a user model. In FG, they would be of a pragmatic rather than a grammatical nature. However, no less abstract universale, such as those generally presented in the typological literature, could be accomodated in such a model. In Dik (1989a) a more or less similar picture is sketched of the relation between language grammars and typological levels of grammar. The process of phonematization is a further refinement that is left outside the Functional Grammar Machine model, as is done in greater part of the FG literature. In order to avoid too many technicalities, a number of aspects of the unifier and generalizer, such as the marking of obligatory and optional substructures will not be discussed here. I follow the classification in Ruhlen (1987). In general, in a fully developed framework, the ordering of rule application is a major issue that I will not go into here. For now, I assume that there are universal ordering principles, such as: "pragmatics prevails over semantics; semantics prevails over syntax" etc. There may be language specific exceptions to them. The second constraint is still necessary for intransitives, among others. All features occurring in it have to be defined as ordered. Centrality is probably an ordered set itself (cf Allan 1987). In a broader definition satellites of the extended predicational level should be included for (potential) subject assignment. Cf for English: (i)

These beds have been slept in.

Since these cases are rather exceptional and restricted, there seem to be arguments for having extension type rules in a grammar.

References Allan, K. 1987 Bakker, Dik 1989

"Hierarchies and the choice of left conjuncts (with particular attention to English)", Journal of Linguistics 23: 51-77.

"A formalism for FG expression rules", in: John H. Connolly - Simon C. Dik (eds.), 45-64. 1990 "A Functional Grammar Machine", in: Mike Hannay - Elseline Vester (eds.), 229-250. Comrie, Bernard 1982 "Grammatical relations in Huichol", in: P. Hopper - S. A. Thompson (eds.). Syntax and semantics ¡5. Studies in transitivity. New York: Academic Press, 95-115. Connolly, John H. - Simon C. Dik (eds.) 1989 Functional Grammar and the computer. Dordrecht: Foris. Craig, e.G. 1976 "Properties of basic and derived subjects in Jacaltec", in: C. Li (ed.) Subject and topic. New York: Academic Press, 99-123. Dik, Simon C. 1978 Functional Grammar. Dordrecht: Foris. 1989a "Functional Grammar and its relevance to grammar writing", in; G. Graustein — G. Leitner (eds.), Reference grammars and modem linguistic theory. Tübingen: Max Niemeyer Verlag, 33-55.

The Functional Grammar machine model 1989b

441

The theory of Functional Grammar. Part 1: The structure of the clause. Dordrecht: Foris. 1989c "FG*C*M*NLU: Functional Grammar computational model of the natural language user" in: John H. Connolly - Simon C. Dik (eds.), 1-28. Greenberg, Joseph H. 1963 "Some universale of grammar with particular reference to the order of meaningful elements", in: Joseph H. Greenberg (ed.), Universals of language. Cambridge: MIT Press, 58-90. Hannay, Mike - Elseline Vester (eds.) 1990 Working with Functional Grammar. Dordrecht: Foris. Hesp, C. 1990 "A critique of FG-CMNLU". Working Papers in Functional Grammar 35. Amsterdam: University of Amsterdam. Kay, M. 1979 "Functional Grammar", in: Proceedings of the 5th annual meeting of the Berkeley Linguistic Society. Berkeley. Knight, K. 1989 "Unification: a multidiscipinary survey", in: ACM Computing Surveys Vol. 211, 93-124. Kuno, Susumo - Etsuko Kaburaki 1977 "Empathy and syntax". Linguistic Inquiry 8: 627-672. Ruhlen, Merritt 1987 A guide to the World's languages. Vol. 1: Classification. London: Edward Arnold. Sells, Peter 1985 Lectures on contemporary syntactic theories. Stanford: CSLI. Shieber, S. M. 1986 An introduction to unification-based approaches to grammar. Chicago: University of Chicago Press. Siewierska, Anna 1988 Word order rules. London: Croom Helm. Smedt, K. J. M. J. de 1990 Incremental sentence generation. A computer model of grammatical encoding. [PhD dissertation. University of Nijmegen.] Weigand, H. 1989 Linguistically motivated principles of knowledge based systems. Dordrecht: Foris. Zaliznjak, A. A. 1977 Grammaticeskij slovar' russkogo jazyka. Moscow.

Subject índex

addressee (semantic/pragmatic function) 5, 74-75, 221,335-337, 348, 438 affected (semantic feature) 75-76, 78-79, 334-335 agent 4, 41, 42, 44, 54-56, 57-58, 59, 61, 73, 78, 90, 97, 100-101, 104-105, 106, 110, 115, 116, 118, 129-130, 154, 155-156, 158, 175, 199, 202-205, 210-211, 297, 298, 325, 362, 436 agreement 97, 98-99, 104, 112, 128, 237, 268, 295, 299, 359, 397 Aktionsart 88, 192-193 analytic expression 105, 107, 113, 354, 359, 370,371-372, 373, 381-382 animacy 25, 56, 122, 261, 289, 362, 436, 437 argument 4, 49-50, 53-62, 71, 81, 83, 89, 127-147 passim, 157, 175, 195, 286, 297, 298, 300, 338, 341, 409, 433 - A' 3, 4, 5, 42, 43, 44, 59-61, 72, 78, 82-83, 84, 129-133, 160, 161, 165-172, 174, 360, 389, 396,419, 429 - A^ 3, 4, 5, 42, 43, 44, 59-61, 71-75, 78, 82-83, 84, 131-132, 160, 161, 165-172, 174, 337, 348 - A3 3, 4, 5, 42, 44, 57, 59-61, 75, 84, 160, 161, 165-172 - argi 3, 5, 129-131, 133, 135, 136, 141, 142, 148 - arg2 3, 130, 131, 133-134, 135 - argjS, 130, 131, 135, 145, 167 aspect 2, 266-281 passim, 286, 297, 303, 304, 356-382 passim atomic predicate, see predicate, atomic attitude 183-188 auxiliary 7, 105, 265-281, 321, 322, 354, 357, 361, 366, 369, 371, 372 auxiliary operator, see operator, auxiliary auxiliary μ-operator, see μ-operator, auxiliary beneficiary 164, 174, 436 case 86, 97, 100, 112, 113-114, 115, 117, 118, 121, 135, 136, 142, 145, 148, 285, 292,

294-298, 299, 301, 391, 393, 396, 410-411 causative 47-53, 54-59, 90, 94, 110, 111, 113, 120 clause 3, 333-334, 338, 344, 354, 380, 433 cleft sentence 164, 172-173, 189, 191, 194 clitic 88, 97, 199, 236, 244, 255-261, 404 cognate object 70, 73 Cognitive Grammar 4, 65, 70-71, 78 complement 92-93, 189, 199, 204-211, 331-332, 334-342, 346, 395, 407 conjunction 339, 341, 343 conjunctive 376, 382-383 context 155-156, 186, 188, 190, 217-22 passim contextual μ-operator, see μ-operator, contextual conveyor belt 8, 361-362, 364, 369, 370, 371-374, 377, 378, 380, 381, 382 coordination 338, 339 copula 51-52, 105, 265-281 core predication, see predication, core dative 90-93, 94 dative alternation/shift 82, 90-91, 109-110, 112, 115-118, 120, 122, 123, 154-155, 159-160, 165, 166-171, 193-194 dative subject 162 declarative 3, 183, 186, 195, 261, 369 defective verb 272-279 definite operator, see operator, definite defmiteness 46, 86, 98, 107, 148, 188, 189, 191, 195, 288, 293, 300, 419-420 deictic centre 332-333, 342-344 demotion of argument 146 determiner 287, 288-292, 299 diachrony 5, 7, 99-107, 224-225, 280 direct speech, see reported speech and quotation direction (semantic function) 42, 89—90, 94 discourse 6, 7, 8, 156, 197, 216-217, 226, 318-327, 345-346 discourse principle 318, 321

444

Subject index

displacement, see sentence intertwining do-support 366-367 dummy 132, 147, 163-164, 169, 178, 234, 252, 392, 393 dummy-preposing 392, 393, 395-396 effected (semantic feature) 76-77, 78-79. 334-335 ellipsis 320, 321, 322, 324, 325 embedding 8, 157-158, 331-332, 338-342, 346, 348, 390, 391-394, 395, 396, 397, 409, 428, 433 ergativity 5, 87, 97-107, 164, 166 experience (SoA) 70, 73 expression 128-147 passim, 188 expression, triggering the expression in one go 145, 367 expression rule 7, 8, 9, 255, 261-263, 2 6 5 281 passim, 285-300 passim, 304-315 passim, 353-382 passim, 393-400 passim, 403-418 extended predication, see predication, extended extra clausal constituent 209, 210 feature-value structure 9 FG*C«M*NLU 306, 385, 425 filtering 411 focus 3, 6, 127, 148, 156, 172, 173, 174, 175, 176, 178, 188-195, 197-198, 203, 204-212, 222-223, 228, 234, 248, 250, 251, 255-261, 318, 320, 391, 392, 395-396, 397, 404, 437 force (semantic function) 42, 54, 57-58, 73, 78, 130 free ride 368, 382 fronting 319-320, 327 function - pragmatic 2, 3, 5 - 6 , 82, 127, 128-147 passim, 153-177, 188-195, 197-212 passim, 215-230 passim, 233-252 passim, 255, 259-260, 261, 286, 325-326, 343, 391-394, 395-396, 397, 403, 424 - semantic 2, 3, 4, 5, 4 1 - 6 2 , 71-79, 82-83, 85, 89-94, 100, 109-111, 112, 120-121, 123, 128, 129-131, 153-154, 160, 175-177, 178, 195, 198, 201, 255, 295, 297, 298, 299, 325-326, 334, 388,

391-393, 396, 397, 403, 409, 429, 435, 436-439 - syntactic 2, 3, 4 - 5 , 6, 81-94, 100-104, 109-124, 128-147 passim, 153-177, 195, 198, 201, 215, 255, 259-260, 261, 263, 286, 295, 297, 298, 299, 325-326, 389-390, 391,403, 438 functional sentence perspective 217, 228 fund 4, 41, 142, 146-147, 279, 285, 306, 408-409 gender 255-256, 261, 266, 269, 359-361, 372, 373, 383, 408 givenness 137-141, 148, 153, 156, 177, 188-191, 192, 197-198, 327 Glossematics 1 goal 4, 42, 43-44, 58, 71-78, 90, 99, 109-111, 114-117, 120, 121, 122, 143, 155, 165-172, 297, 298, 334-337, 436 grammaticalization 154, 159, 161-162, 177, 265, 280-281 Gricean principle 326

288, 178,

104, 142, 332, 175,

hypotaxis 338-339, 346 illocution (see also speech act) 3, 177, 183-188, 261, 286, 320, 324, 359, 366, 369, 380,381,387, 409,413, 431 illocutionary operator, see operator, illocutionary illocutionary satellite, see satellite, illocutionary indirect object 161-162, 193 indirect speech, see reported speech and quotation information unit 319, 322-324, 327, 328 instrument (semantic function) 4, 13, 35, 41, 44, 55-56, 59, 61, 106, 121 interrogative 3, 183, 186, 187, 195, 261, 289, 318, 366, 369, 381 intonation 158, 185, 286, 319, 322, 327, 332, 339, 342, 403 intransitive 84, 85, 87, 93, 97-98, 101, 107 islands constraints (subjacency) 323—324 isomorphism 120-121 layered model 2, 8, 128, 147, 187, 276, 331, 333, 338, 346, 348, 353, 354-359, 404, 424, 428

Subject index lexical decomposition 4, 11-37 linear ordering, see word order linguistic action verb 73, 77, 79, 331-332, 334-337, 342, 346, 348 LIPOC 133, 244, 294, 299, 404, 412 location (semantic fonction) 42, 56, 89-93 μ-operator 268, 276, 277 - auxiliary 268, 278 - contextual 268 - primary 268, 278 markedness 5, 127-147, 158-159, 161, 163, 164-172, 198, 216-228 passim, 247-248, 414 markedness shift 5, 100-102, 106-107 meaning definition 41-62 meaning postulate 11, 27-28, 36, 42, 45 metaphor 69, 70, 75, 82, 121, 147 metonymy 73 middle voice 87, 94, 102 modal operator, see operator, modal modal verb 186-187, 279 mood 266-281 passim, 286, 297, 356-382 passim тофЬо8уп1ас11с operator, see operator, morphosyntactic nonsentential utterance 317-328 nonverbal predicate, see predicate, nonverbal nuclear predication, see predication, nuclear nuclear semantic fonction 3, 4, 42, 303 number 266, 269, 288, 355, 359-361, 372, 373, 408 numerator 8, 288, 290-291, 299, 300 object 6, 66, 81-94, 98, 100-101, 104-105, 109-124, 127, 135, 136, 154-177 passim, 193, 257-260 passim, 296, 298, 331, 336, 342 - adpositional 141, 143 - assignment 5, 44, 65, 71, 75, 81, 85, 109-124, 127, 128, 129, 131-142 passim, 145-146, 155, 332, 337, 389-390 operational moφhosyntax 382 operator 8, 178, 187, 274, 275, 276, 278, 286, 287, 321, 326, 353-382 passim, 388, 424, 428, 431,433 - auxiliary 382, 403, 404, 410-411

445

-

definite 46 illucutionary 3, 183, 184, 187, 268, 320, 321 - modal 186-187, 279, 356-360, 367, 369, 376, 382 - morphosyntactic 353, 355, 403 - predicate 2, 268, 279, 280, 285 - predication 2, 187, 265, 268, 286, 403 - proposition 2 - 3 , 183, 184, 187, 268, 333 - term 192, 285, 287, 288-293, 299, 301 ordering, linear, see vk-ord order π-operator 7, 18, 187, 191, 192, 268, 276, 334, 354 PI 6, 7, 135, 200, 215-230 passim, 233-252, 391-392, 393, 404,412,415 P4 212 paradigm 137, 146-147, 362-363, 371, 378 paradigmatic relation 128, 133-134, 142, 185, 187 parataxis 338-339, 346 parsing 415 passive (see also voice) 58, 66, 67-79, 82, 87, 89,94, 97-107, 111-112, 122, 131-132, 134, 142-145, 154-161, 165-169, 174, 336, 389, 390 - nontransitive 77-78, 101, 107 - perspectivization fonction {see also perspective and perspectivization) 78 patient 325, 362 person 266, 269, 359-361, 436-437 person-number ending 266, 374, 377-380 perspective 84, 85, 127-147, 155-160, 165-166, 174-175, 176, 178, 332 perspectivization 81-83, 111, 157, 391 polarity 2, 85-86, 359-360, 366, 369, 380 polysemy 85, 116, 123 positioner (semantic fonction) 42, 57-58, 73, 78, 130 possessor (semantic fonction) 296, 298, 300 postfield language 292, 299, 300 pragmatic module 325 predicate 2,4, 11-37,41-62,81,83, 89, 128, 133, 185, 192, 195, 199, 202-210, 265, 269, 280, 286, 294, 297, 303, 353, 388, 391,394, 429 - atomic 13, 25-28 - nonverbal 276, 279-298, 301

446

Subject index

predicate constituting argument 8 3 - 8 9 predicate formation 2, 53, 146-147, 163, 175, 178, 279, 280, 285, 408-409, 431 predicate frame 2, 4, 27, 41-62, 116, 127-147 passim, 155, 176, 285, 362-363, 371, 381, 389, 392, 393, 403, 4 0 7 - 4 0 9 , 4 1 8 - 4 1 9 , 431 predicate operator, see operator, predicate predication 4, 5, 8, 56, 81, 82, 83, 90, 112, 127, 146, 157, 175, 177, 192, 235, 269, 279, 300, 304, 306, 338, 346, 386-387, 388, 389, 391, 392, 396, 403, 405, 407, 410 predication, core 2, 75, 354, 409, 433, 438 - extended 2, 354, 409, 428, 430, 439 - nuclear 2, 42, 43, 75, 199, 362 - secondary 90, 94 predication constituting argument 84, 85, 89, 100 predication operator, see operator, predication prefield language 292 preposition 68-69, 112, 141, 143, 144, 156, 160, 175, 190, 191 prepositional phrase 68, 9 1 - 9 2 prepositional verb 6 8 - 6 9 primary μ-operator, see μ-operator, primary processed (semantic function) 43-44, 58, 407 projective mapping 353, 354, 362, 370 projectivity principle 277, 282, 354, 369 pronoun 7, 135, 136, 148, 185, 193, 238-239, 244, 255-261, 332, 336, 342, 380, 404 proposition 2, 8, 84, 183, 185, 186, 235, 333, 338, 344, 345, 346, 348, 354, 409, 428, 431 proposition operator, see operator, proposition proposition satellite, see satellite, proposition prototypicality 66-79, 147, 161-162, 163, 177, 334, 337 quantifier 8, 288, 290-291, 299, 300, 312 quotation 185, 191-192, 331-347 raising 78, 81, 110, 111, 112, 113, 114, 118-120, 157-158, 163, 164, 171-172, 177, 389-390, 391, 392, 396, 397 recipient (semantic function) 42, 57, 74, 89-90, 94, 110, 115-117, 121, 142, 143, 145, 155, 160, 164, 166-172, 221, 298, 436

recursion 395-396, 398, 428 reference (semantic function) 42, 56, 73 reflexive 89, 91, 112, 161, 162 relative clause 172, 235, 238-240, 388, 397-398, 438-439 relator 286 reperspectivization (see also perspective and perspectivization) 127-128, 142-145, 146, 155, 170 reported speech 8, 73-74, 331-347, 376 restrictor 286, 287-294, 296, 299, 300, 309, 311,409, 427,428, 429 rule ordering {see also conveyor belt) 361, 367,413-414 satellite 2, 4, 7, 42, 43, 55, 62, 73, 75, 81, 127, 129, 131, 172, 173-176, 174, 175, 183-185, 191, 194, 195, 200, 201, 209, 234, 235, 246, 247, 251, 285, 286, 303-315, 326, 333-334, 391, 428, 430, 438 - illocutionary 3, 184, 321, 333 - proposition 2 - 3 , 183 scalar distinction 123-124, 146, 178, 331-347 scope 156, 184-188, 192, 193, 356, 369, 424, 428-429 secondary predication, see predication, secondary selection restriction 87, 274, 289, 388 semantic function hierarchy 44, 61, 71, 129-131, 160, 161, 167, 174, 176, 430, 436-439 sentence intertwining 172, 174, 392, 393, 395-396 sentential status 319-321, 323, 342, 345 side-tracking 372, 374, 381-382 source (semantic function) 42 speaker (semantic/pragmatic function) 5, 74-75, 183-186, 189, 191, 198, 331-346 passim, 356, 358, 438 speech act {see also illocution) 6, 183-188, 195, 320, 321, 333 state of affairs (SoA) 2, 3, 4, 5, 32-34, 42, 44, 45, 47-53, 57-60, 68-71, 75, 79, 83, 84, 91, 109, 111, 112, 115, 117, 123, 130, 131-132, 142, 156, 159, 163, 171, 235, 274, 275, 298, 307, 345, 358

Subject index stress 86, 255-261, 286, 369 subject 6, 7, 66, 71, 75, 8 1 - 9 4 , 98, 100-101, 104-105, 110, 112, 114, 127, 135, 154-177 passim, 198, 202, 204, 212, 216, 221, 228, 236, 259, 260, 296, 298, 324, 331, 342, 436-439 - secondary 91 subject assignment 5, 44, 65, 71, 78, 8 1 - 8 2 , 85, 106, 111-112, 122, 123, 124, 127, 129, 131-145, 155, 332, 336, 360, 389-390, 4 3 6 - 4 3 9 subordinator 7, 233-252, 286 syntagmatic relation 128, 130, 131, 153 synthetic expression 354, 359, 370, 371 - 3 7 2 , 377,381-382 Systemic Functional Grammar 73 tag 319-320, 321,327 tail 3, 6, 228 temporal satellite 7, 8, 303-315, 4 3 8 - 4 3 9 tense 2, 266-281 passim, 286, 297, 303, 304, 332, 343, 344, 3 5 6 - 3 8 2 passim term 4, 6, 7, 41, 73, 81, 128, 156, 195, 235, 246, 285-300, 303-315, 3 8 8 - 3 9 8 passim, 407, 409, 413, 427, 428, 430 theme 3, 6, 174, 175, 176, 200, 209, 217, 218, 228, 320, 324, 326, 327

447

topic 3, 6, 7, 71, 82, 153-178, 188-195, 197-198, 202-212, 215-230, 234, 235, 248, 250-251, 320, 326, 327, 391-398 passim, 404, 409, 437 transitive 6 6 - 7 7 , 82, 84, 87, 93, 9 7 - 9 8 , 100, 158, 165, 332, 334-337, 342, 346 - ditransitive 82, 8 9 - 9 3 , 177 transitivity 4, 6 5 - 7 9 , 85, 100, 177, 335-337 underlying clause structure 165, 187, 319-321, 354, 360, 393, 403, 413, 421, 424, 4 3 1 - 4 3 3 unification 405, 4 3 1 - 4 3 4 valency 4, 42, 8 3 - 9 4 , 116-117, 122, 128, 178 variation, linguistic 256-263, 265-267, 281 verbal complex 7, 8, 353-382, 4 0 3 - 4 0 4 voice {see also passive) 82, 87, 94, 160, 359-360, 364-365, 372, 374, 377-379, 390, 436 word order 112, 115, 127, 129, 133, 134-144, 153, 163, 173-174, 193, 197-212, 259, 261, 263, 285, 286, 292-294, 304, 390-394, 395, 397, 403, 411-412, 419 word order template 13, 175, 233-252, 404, 431