New Perspectives on Argument Structure in Functional Grammar [Reprint 2015 ed.] 9783110895872, 9783110173932

Citation preview

New Perspectives on Argument Structure in Functional Grammar

W G DE

Functional Grammar Series 25

Editors

A. Machtelt Bolkestein | Casper de Groot J. Lachlan Mackenzie

Mouton de Gruyter Berlin · New York

New Perspectives on Argument Structure in Functional Grammar

edited by

Ricardo Mairal Usón María Jesús Pérez Quintero

Mouton de Gruyter Berlin · New York

2002

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, 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 New perspectives on argument structure in functional grammar / edited by Ricardo Mairal Usón, Maria Jesús Pérez Quintero, p. cm. - (Functional grammar series ; 25) Includes bibliographical references and index. ISBN 3 11 017393 X (he. : alk. paper) 1. Functionalism (Linguistics) 2. Grammar, Comparative and general — Verb phrase. 3. Semantics. I. Mairal Usón, Ricardo, 1965- II. Pérez Quintero, María Jesús, 1968III. Series. P147.N48 2002 410'.l'8-dc21 2002026499

Die Deutsche Bibliothek — Cataloging-in-Publication Data New perspectives on argument structure in functional grammar / ed. by Ricardo Mairal Usón ; María Jesús Pérez Quintero. — Berlín ; New York : Mouton de Gruyter, 2002 (Functional grammar series ; 25) ISBN 3-11-017393-X

© Copyright 2002 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted 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. Printing: WB-Druck, Rieden/Allgäu. Binding: Lüderitz & Bauer, Berlin. Printed in Germany.

Contents

Introduction

vii

The predicate in Functional Grammar J. Lachlan Mackenzie

1

Functional Grammar and lexical templates Ricardo Mairal Usón and Pamela Faber

39

Do we need predicate frames? Daniel Garcia Velasco and Kees Hengeveld

95

Adpositions, the lexicon and expression rules Dik Bakker and Anna Siewierska

125

Ontological Semantics and lexical templates: Mowing the grass from the other side of the fence Antonio Moreno Ortiz and Chantal Pérez Hernández

179

On the syntax-semantics interface in word formation: The case of English -er nominalizations Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero 213 Catching a glimpse of linguistic reality: Modelling the complexity of CATCH in the Functional Lexematic Model Chris Butler

247

Lexical templates and syntactic variation: The syntax-semantics interface of the Old English speech verb secgan Marta María González Orta 281

vi

Contents

Lexical-semantic explorations in English verbs of physical contact: Iconicity and linguistic representations Margarita Goded Rambaud and Rocío Jiménez Briones

303

Another look at second and third arguments Javier Martin Arista

331

Index of subjects

349

Introduction

Since Dik (1978), Functional Grammar (FG) has undergone a number of substantial modifications. Many of the refinements of the theory have been oriented towards enriching the structure of the clause by giving it a multi-layered dimension (Dik [1989] 1997a, 1997b; Hengeveld 1989), as well as providing an in-depth account of pragmatic functions (Connolly et. al. 1997). A great effort has also been made to elaborate higher units of description at the discourse level (Hengeveld 1997, to appear; Kroon 1997). However, other aspects of the theory such as expression rules, the internal structure of the lexicon, and the syntax-semantics interface have not received the attention they evidently deserve, notable exceptions being Bakker (1999, 2001) and Olbertz et. al. (1998) for the expression rule component and the internal structure of the lexicon, respectively. The FG lexical component - the Fund - consists of a lexicon, a set of predicate formation rules, which account for derivational processes, and a set of term formation rules. In the lexicon, each basic predicate is represented in the form of a predicate frame, which also contains some specification of the meaning of the predicate. The predicate frame is conceived of as a construct that plays a vital role in the construction of underlying predications and in the instruction of the expression rule component to produce the appropriate output. J. Lachlan Mackenzie, starting from a discourse perspective, offers a complete account of the role of predicate frames in the existing theory. The author also evaluates other types of research that constitute a radical reformulation of the very notion of predicate frame. This results in a complete picture of the state of the art in argument structure representation in FG. He also touches upon other controversial issues within the theory, such as the status of Predicate Formation Rules. Against this general background, this volume investigates a number of possible innovations within the lexical component of FG.

viii

Introduction

These innovations have to do with the interaction of meaning definitions with their corresponding predicates and with the possibility of constructing meaning definitions on the basis of an abstract metalanguage. These two theoretical topics are the leading issues of Ricardo Mairal Usón and Pamela Faber's key article in this volume, and in fact they also constitute the background against which the rest of the papers have been written. A reconsideration of the predicate frame in these two areas eventually leads to the formulation of a new design for the syntax-semantics interface in FG, which is much more in consonance with the theoretical underpinnings of a functional theory. In connection with this, it is important to note that the key paper as well as most papers in the volume - heavily bears upon insights from Role and Reference Grammar (RRG - Van Valin and LaPolla 1997). This theoretical move seems to be in consonance with the fact that both models are regarded as moderate functional approaches, and share many theoretical premises in a number of specific points, viz. the adoption of a layered structure model for clause analysis. More particularly, one of the leading thesis is that the metalanguage posited in RRG logical structures can enrich the format of the predicate frame. Furthermore, logical structures unify into one compact, single representation both the semantic and the syntactic distinctive parameters of each predicate. However, what RRG is almost silent about concerns the development of a further semantic decompositional system of the primitives that form part of the lexical representation. In this sense, the inclusion of the semantic analysis done in the framework of the Functional Lexematic Model (which is a further development of the standard FG view) in the inventory of RRG logical structures constitutes the central point of departure of the key paper in this volume. Ricardo Mairal Usón and Pamela Faber argue in their paper for a new system of representation, the lexical template, which conflates both syntactic and semantic information into one unified representation; hence the predicate frame and its corresponding meaning definition are no longer conceived of as two separate modules but as one unified representation. This theoretical move is a radical departure from the orthodox FG view. The authors, drawing on previous work

Introduction

ix

on the lexicon and its division into articulated semantic classes (Faber and Mairal 1999), argue that each lexical class is characterized by a lexical template, a theoretical construct which contains the basic conceptual parameters shared by all the verbs belonging to a lexical class. The twofold division into internal and external variables allows the representation of the full set of regularities within a lexical class into one unified representation. The authors proceed to discuss how to account for syntactic variation within the same lexical class without having to postulate a different entry for each complementation structure. They claim that syntactic variation can be explained by modeling processes which operate upon the canonical lexical template through the application of lexical rules which relate lexical entries to their corresponding complement configurations. A lexical template is posited for the lexical class of manner-of-cutting verbs, and the alternations typical of these predicates are analyzed with respect to the maximal canonical template. However, as the authors conclude, an issue for further research concerns the description of the inventory of lexical rules, as well as the actual format of the fusion processes involved. From a more syntactically-driven perspective, Daniel Garcia Velasco and Kees Hengeveld introduce the notion of predication frame, which replaces the notion of predicate frame with a view to making the theory more adequate psychologically, pragmatically, and typologically. The authors claim that predicate frames should be replaced by predicates and frames, the former corresponding to lexemes and the latter to the aforementioned predication frames. They suggest that a mechanism should be posited to link the two in the grammar. Predication frames represent the full range of underlying configurations in which a lexeme can be inserted. Lexemes, which are represented in the Fund, are provided with an abstract meaning definition which, in turn, serves to anchor each lexeme to its appropriate predication frame. Thus, a speaker selects a particular predication frame and then he/she chooses an appropriate lexeme, a process which is guided by the conceptual information encoded in lexemes. Like Mairal Usón and Faber, these authors agree that syntactic alter-

χ

Introduction

nations are better not accounted for by predicate formation rules, but rather by lexeme derivation rules. Without questioning the validity of the notion of predicate frame and the epistemological nature of meaning definitions, Dik Bakker and Anna Siewierska, drawing on their dynamic version of the expression rule component, address the issue of the syntax-semantics interface in FG. They reconsider the type of material which should be stored in the lexicon and provide a detailed discussion of the status of adpositions in an FG grammar from both a diachronic and a typological perspective. They provide evidence that some types of adpositions should be treated as predicates. A new conception of the predicate is posited within the larger framework of the lexicon, underlying representation and expression such that the inventory of expression rules can produce the appropriate output. This paper thus sheds new light on the status of expression rules, which are vital in the design of a powerful linking algorithm. In Mairal Usón and Faber's paper, a strong claim is made towards the use of an ontology as a device for further semantic decomposition of the set of constants which permeate a lexical template. Antonio Moreno Ortiz and Chantal Pérez Hernández, drawing on work in Ontological Semantics and more particularly on the Mikrokosmos ontology, explore the kind of impact that the incorporation of a language-independent ontology of concepts would have on an FG lexicon organized in terms of a hierarchy of lexical templates, and proceed to provide a detailed account of the way both internal and external variables can be further decomposed semantically by anchoring these to an ontology of concepts. Moreover, as the authors claim, this type of endeavor turns out to be very helpful in the study of ontology and of acquisition of the lexicon. Within the realms of morphology, Francisco J. Cortés Rodriguez and María Jesús Pérez Quintero present an alternative account of agent nominalizations in terms of the notion of lexical template as formulated in the key article. These authors argue against the use of predicate formation rules for lexical creation since it is highly questionable whether a single predicate formation rule can account for the full gamut of possible input predicates. They claim instead that af-

Introduction

xi

fixes are lexical units in their own right, and therefore, should be represented in the lexicon. The type of representation they suggest is an expansion of the notion of lexical template. They then proceed to discuss the way affixal representations fuse with their corresponding basis. Of central importance is the notion of lexical template as an alternative to the predicate frame. In order to test the explanatory power of this construct, the rest of the papers in this volume concentrate on investigating the methodological scope of this proposal by applying it to several key issues in lexical semantics and the semantics of syntax. Chris Butler examines in detail the notion of lexical template, exemplifying it by one of the most polysemous predicates in English, i.e. catch. The author provides empirical evidence from informant testing and corpus data extracted from dictionaries and the British National Corpus and proceeds to assess to which extent the full complexity of the behavior of catch can be captured within a lexical template. The author, though admitting the shortcomings of both the notion of predicate frame and the use of natural language phrases for meaning representation, concludes that it is impossible to posit a lexical template for high-complement taking predicates like catch such that all examples can be accounted for. Finally, he opens a new line of thought on the status of primitives and tentatively considers the use of Wierzbicka's semantic primitives for the construction of an ontology. Marta María González Orta, using a representative sample of data in Old English, proposes a typology of lexical templates for the lexical domain of speech, applying them to the lexeme SECGAN as the major representative of this lexical class. The author proposes a maximal lexical template and shows how the different alternations of this predicate can be accounted for by means of the application of the Lexical Template Modeling Process. Similarly, though centering on the analysis of Modern English, Margarita Goded Rambaud and Rocío Jiménez Briones examine the iconic nature of a lexical subclass of the domain of feeling. After extracting the parameters en-

xii

Introduction

coded in this lexical subdomain, they move on to propose a lexical template for this group of verbs. The last paper addresses the status of arguments in FG against the background of the key article. Javier Martín Arista, drawing on the notion of lexical template, proposes a refinement of the classification of second and third arguments as represented in the Semantic Function Hierarchy. The author claims that such a theoretical move is motivated semantically and argues that Result and the telic version of Source, Path, Direction and Location are nuclear semantic functions with argument status. Finally, his Principle of Lexical Template Instantiation, a slightly modified version of Mairal Usón and Faber's Lexical Modeling Process, accounts for the well-formedness conditions on semantic and syntactic realization. In all, this volume contains a number of ground-breaking proposals which challenge the orthodox view of argument structure in FG as configured in terms of a predicate frame and a meaning definition. Furthermore, this volume describes new paths of research regarding the syntax-semantic interface in that it highlights the interaction between predicate frames and meaning definitions, a methodological issue which has as yet not been addressed in Functional Grammar. We are confident that the contributions set forth in the present volume represent a step forward in the complex linguistic enterprise of accounting for the mapping of syntax to semantics. February 2002 Ricardo Mairal Usón Maria Jesús Pérez Quintero * We are indebted to Kees Hengeveld and Casper de Groot for their comments on an earlier version of this text.

Introduction

xiii

References Bakker, Dik 1999

Functional Grammar expression rules: From templates to constituent structure. Working Papers in Functional Grammar 67. 2001 The FG expression rules: A dynamic model. In: María Jesús Pérez Quintero (ed.), Challenges and Developments in Functional Grammar, 15-53. (Revista Canaria de Estudios Ingleses 42.) La Laguna: Servicio de publicaciones. Connolly, John, Roel M. Vismans, Christopher Butler and Richard Gatward (eds.) 1997 Discourse and Pragmatics in Functional Grammar. Berlin/New York: Mouton de Gruyter. Dik, Simon C. 1978 Functional Grammar. Amsterdam: North Holland. 1997a The Theory of Functional Grammar. Part 1: The Structure of the Clause. Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter. First published Dordrecht: Foris [1989]. 1997b The Theory of Functional Grammar. Part 2: Complex and Derived Constructions. Edited by Kees Hengeveld. Berlin/Mouton: Mouton de Gruyter. Faber, Pamela and Ricardo Mairai Usón 1999 Constructing a Lexicon of English Verbs. Berlin/New York: Mouton de Gruyter. Hengeveld, Kees 1989 Layers and operators in Functional Grammar. Journal of Linguistics 25.1: 127-157. 1997 Cohesion in Functional Grammar. In: John Connolly, Roel M. Vismans, Christopher Butler and Richard Gatward (eds.), Discourse and Pragmatics in Functional Grammar, 1-16. Berlin/New York: Mouton de Gruyter. to appear The architecture of a Functional Discourse Grammar. In: J. Lachlan Mackenzie and Maria de los Ángeles Gómez González (eds.), A New Architecture for Functional Grammar. Berlin/New York: Mouton de Gruyter. Kroon, Caroline 1997 Discourse markers, discourse structure and Functional Grammar. In: John Connolly, Roel M. Vismans, Christopher Butler and Richard Gatward (eds.), Discourse and Pragmatics in Functional Grammar, 17-32. Berlin/New York: Mouton de Gruyter. Olbertz, Hella, Kees Hengeveld and Jesús Sánchez (eds.) 1998 The Structure of the Lexicon in Functional Grammar. Amsterdam/Philadelphia: John Benjamins.

xiv

Introduction

Van Valin, Robert D. and Randy J. LaPolla 1997 Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

The predicate in Functional Grammar J. Lachlan Mackenzie

1. The predicate as an ascriptive item In his seminal proposal for a new architecture of Functional Grammar (FG), Hengeveld (to appear) sees communication as consisting of a succession of moves. A dialogue is thus in essence an alternation of moves, much as in a chess game. Each move consists of one or more utterances, i.e. acts of utterance. Some utterances are purely interactional, as with the first utterance recorded in (1), which serves merely to indicate assent (Yes). Other utterances combine interaction with propositional material, as with the second utterance recorded in (1), the train has arrived. (1)

Yes, the train has arrived.

The act of utterance the train has arrived is analysed as containing two basic subacts and a number of operators and functions that further specify those subacts and their interplay. On the one hand, the words has arrived reflect a subact of ascription, whereby the speaker ascribes arrival to an entity, and a number of operators (Present, Perfect) that modify that subact of ascription. On the other hand, the words the train reflect a subact of reference in which the speaker invokes the entity to which arrival is ascribed, as well as a number of operators (Definite, Specific, Singular) and functions (Processed, Subject, Topic). In this particular case, the speaker has judged that the subact of reference is best supported by a further (sub)subact of ascription. If she had said It has arrived, then it would have been purely referential, unsupported by ascription. In (1), however, the subact of reference is further restricted by the introduction of an ascriptive subact invoking the lexical item TRAIN. This complex inter-

2

J. Lachlan Mackenzie

play of ascription and reference is what underlies the evocation of States of Affairs and is organized by the structures that underlie the construction of a propositional utterance. Thus the layered structure imposes a strict separation of operators, functions, ascriptive and referential material: the predicate frame ensures the primacy of the ascriptive subact with its dependent subacts of reference (reflected in arguments), just as the structure of the term constrains ascriptive subacts within the term to appear as restrictors upon the primary subact of reference. In (1), the two ascriptive subacts are strongly associated with the items TRAIN and ARRIVE. These are drawn from the Fund, which contains the totality of items, basic and derived, available for subacts of ascription in the language under consideration. It is these items that are known in FG as predicates. The Fund also includes other items, notably proper names such as Malibu or Geraldine, which are not predicates but "basic items", items ready-made for subacts of reference; other occupants of the Fund include items that are called upon by the expression rules to indicate the presence of functions (e.g. English by to mark an Agent that lacks the syntactic function Subject) or the presence of operators (e.g. Spanish estar to mark the presence of the operator Progressive, cf. Olbertz 1998:479); and then there are items such as English .yes, well, etc. that are directly indicative of an interactional intention. The examples given hitherto all correspond to words. But the Fund also contains items that are both less than and more than words: thus we will find there affixes, both inflectional and derivational, as well as affixes that have ascriptive value and which therefore need to be regarded as predicates (Fortescue 1992 gives some examples for Koyukon; see also Mithun 1999: 49 on Yup'ik "affixes with verb-like meanings"); but the items in the Fund may also be multi-word, extending from phrasal verbs like English PUT OFF 'postpone' through idioms like ALL HELL BREAK LOOSE, to entire texts, e.g. the Lord's Prayer. FG distinguishes a central component of the Fund, the Lexicon, which contains all those items that are not derived by some procedure (generally that of predicate formation, see section 4 below) from a simpler item. The Lexicon can thus be defined as the totality of basic

The predicate in Functional Grammar

3

predicates, whether these take the form of a morpheme, a word, or a fixed (i.e. synchronically non-derived) combination of words. It will be clear that the Lexicon is central to the Fund in a number of respects: synchronically, it provides the building blocks from which derived predicates, e.g. compounds and other multi-word items, are formed; historically, many proper names (e.g. The White House) originate in items from the lexicon; and many function words (i.e. items introduced by the expression rules) derive by grammaticalisation from lexical predicates (see section 6 below). The defining property of the predicate will be its availability for the communicative subact of ascription. Ascription will be understood here in its normal sense of the assignment of property to a (single) participant, as in the discussion of (1) above, but will also be extended to cover the naming of a relation between two or more participants in that relation. Thus in (2): (2)

Keith gave Mary his lecture notes.

the predicate GIVE ascribes a relation that holds among the three participants in the State of Affairs. Ascription will also be further extended to cover instances where no participant, i.e. no dependent subact of reference, is present, as in the case of meteorological predicates like English RAIN; in (3), it does not reflect an act of reference: (3)

It was raining.

These distinctions are reflected in the quantitative valency of the predicate, to be further discussed in section 3 below. The predicate is indissolubly linked to the notion of a State of Affairs (SoA). Whereas each subact of reference invokes an entity of some kind, each subact of ascription invokes an SoA. An SoA is emphatically not an objective situation in the external world, but a speaker's subjective (and possibly even illusory) conceptualisation (cf. Dik 1997a: 51). SoAs can be embedded within one another. Thus in the second utterance in (1) above, the SoA of being a train is embedded within the SoA of arriving: in FG terms, the propositional

4

J. Lachlan Mackenzie

material being imparted is essentially as follows: "I ascribe arrival to an entity to which I ascribe the property of being a train". SoAs are classified according to their aspectual characteristics. Prominent among these are control (whether or not there is a participant who/which determines the occurrence of the SoA, as opposed to merely undergoing it) and dynamism (whether or not the SoA involves a change). In an Action SoA, both control and dynamism are present; in a State SoA, both are absent. FG also recognises two other kinds of SoA: Positions (+control, -dynamism) and Processes (control, +dynamism). These distinctions do not apply in FG to predicates, but to the SoA that is invoked in the subact of ascription. Thus in the second utterance in (1) above, the SoA is a Process, since the train does not control its own arrival. The aspectuality of the SoA determines the semantic functions of the arguments associated with the predicate. Thus, if the SoA is an Action, the first argument of the corresponding argument will be Agent; if the SoA is a Process, its first argument will be Processed; see Dik (1997a: 117-124) for further details. There has been a tendency in FG to ascribe particular arrays of semantic functions to specific predicates. For example, GIVE, as exemplified in (2) above, has been used time and again as a classical example of a predicate almost inherently associated with Agent, Goal (= Patient), and Recipient. Butler (2001) has demonstrated the folly of this approach, showing that in actual usage, which deserves priority in a functionalist approach, GIVE has a much broader privilege of occurrence. The conclusion must be that the assignment of semantic functions is dependent upon the conceptualisation that is reflected in the particular SoA. In English, not only with the predicates just discussed but quite generally, the distinctions between different semantic functions of first arguments are not expressed morphologically. The evidence for their presence must be drawn from their co-occurrence potential with adverbs (Agent with carefiilly, Agent, Force or Processed with rhythmically, etc.). The opposition between 'falling on purpose' and 'falling accidentally' can in English only be expressed by adding some such satellite as inadvertently. Van Valin and LaPolla (1997: 119) report that in Tsova-Tush, this very distinction is reflected in the

The predicate in Functional Grammar

5

distinction between ergative and absolutive case (for a more general treatment of the morphological consequence of the [±control] parameter, see Mithun 1999: 128-131): (4)

a. Vuiz-n-as fall-TNS-lsgERG Ί fell down on purpose.' b. Voz-en-sO fall-TNS-1 sgABS

Having established the status of the predicate as an occupant of the Fund with an ascriptive function, we will now progress, in section 2, to a consideration of how predicates are represented in FG, expanding the discussion in section 3 to the predicate frame. Section 4 will deal with predicate formation rules (PFRs), showing how each element of the frame is subject to their operation. Section 5 treats a particular kind of PFR, term predicate formation, and section 6 looks at the potential for predicates to undergo grammaticalisation.

2. The representation of predicates in the Fund Predicates are listed in the Fund in the following form: (5)

(f¡: predicate (f¡))

The variable (f¡) was introduced in Dik (1989: 50), but was developed above all by Keizer (1992a) and Hengeveld (1992: 51-55). The formalism given in (5) has the effect of assimilating predicates to terms (i.e. referring expressions), which since the inception of FG have had the general format (Variable: Predicate (Variable)), a format which has since been generalised to each of the layers that have continued to characterise the model in its various manifestations (cf. Hengeveld to appear for the most recent of these).

6

J. Lachlan Mackenzie

There are various justifications for the variable (f¡). Firstly, there are predicates that take other predicates as arguments, for example the predicate LOOK, as exemplified in (6): (6)

You look ill.

Here the predicate ILL occurs as the head of the second argument of the predicate LOOK (Hengeveld 1992: 42-43): if arguments are defined as terms, then it is appropriate that the argument should take the form associated with argumenthood. Secondly, on the assumption that anaphora is a matter of co-indexing (but see Cornish 2000 for a rejection of that assumption), a predicate variable allows anaphora to predicates to be handled in an analogous way to anaphora to terms (see Keizer 1992a: 4; Hengeveld 1992: 53): (7)

You look ill, and so does he. ... (f¡: ill (fO)... (Ai)... where A is an anaphoric operator

Similarly, predicates can be the antecedent for relativisation, as in (8) (Keizer 1992a: 4): (8)

I look ill, which I am not. ... (f¡: ill (fi))... (Rf¡) ... where R is a relativisation operator

Thirdly (and this was already implicit in the representation given in (7)), predicates can be subject to operators; and operators apply in FG to variables. An example of such an operator is Progressive. We have already mentioned the application of the operator Progressive to verbal predicates in Spanish, yielding the auxiliary estar (Olbertz 1998: 479); Hengeveld (1992: 37) advocates an analogous analysis for the occurrence of copular estar with adjectival predicates:1

The predicate in Functional Grammar

(9)

Antonio está Antonio COP.PRES.3SG 'Antonio is being crazy.'

7

loco crazy

... (Progr fi: loco (fi)) (dlxi: Antonio (xi))zero ··· Analogously, the Ingressive operator applied to an adjectival predicate induces the copula become in English (cf. Hengeveld 1992: 38). According to Rijkhoff (1991), nominal predicates, too, can display aspectual distinctions: thus noun classifiers such as tel ('plant') in Jacaltec will be analysed as operators on the predicate variable (f¡): (10)

a.

kaj tel tahnaj ixpij (Grinevald 2000:65) red PLANT ripe tomato 'The ripe tomato is red.' b. tel tahnaj ixpij (dx,: (PLANTfi: ixpij (fO) (xi): (f2: tahnaj (f2)) (x,)) 'the ripe tomato'

In much current work in FG (for example in most of Dik 1997a, 1997b), the second occurrence of the (f¡) variable, where it appears as an argument of the predicate, is omitted. This applies to all other variables, too, with the aim of increasing the legibility of representations; cf. Dik (1997a: 63). The representation in (10b) would then appear as (11): (11)

(dxi: (PLANTfi: ixpij): (f2: tahnaj))

This emendation was first proposed in Mackenzie (1987), but not merely with the aim of simplifying representations. The intention was to make the representation less like a logical structure put together a posteriori, and to emphasise the fact that the speaker is with each predicate ascribing a property on-line, as suggested in section 1 above. Thus (11) would be read as follows: "there is a definite entity such that I ascribe the plant-property IXPIJ such that I ascribe the property TAHNAJ".

8

J. Lachlan Mackenzie

This approach to representation has the added advantage of presenting prototypical (i.e. "non-relational") nominal predicates as having no arguments. In orthodox FG, non-relational nouns such as English TOMATO appear in the Fund with an obligatory Zero-argument: (12)

(fi: tomato) (xi)zero

However, this type of representation predisposes nouns towards predicative use, i.e. towards occurrence in utterances in which the application of that noun to an argument is being asserted. The predicative use of nouns is of course not unknown (although it is rather restricted in English, cf. *This is tomato: we will see in section 5 below how we arrive at This is a tomato). Nevertheless, nouns are defined in FG in terms of their prototypical use as heads of terms (cf. Hengeveld 1992: 58), where the property expressed by the noun is, as we have seen, being ascribed rather than asserted as such. For such prototypical cases, the avalent representation suffices: (13)

(fi: tomato)

(13) can be expanded, in languages that permit "bare nominal predicates" into (12) through valency extension (see section 4 below) where the speaker wishes to assert the application of a nominal predicate to an argument. In this approach, relational nouns, too, will lack a Zero argument, but they will be distinguished by having an obligatory argument with a different semantic function; thus the relational noun FRIEND will appear in the Fund of English as (14a), with the Reference term being available for the befriendee, as shown in (14b): (14)

a. b.

(fi: friend) (xi)Ref Bill's friend (dlxi: (fi: friend) (dlx 2 : Bill)Ref)

The predicate in Functional Grammar

9

The above discussion has assumed, with traditional FG, that predicates occur in the Fund with a number of arguments that ranges upwards from zero. These arguments form part of the predicate frame to be described in section 3. Before progressing, however, we should take note of Garcia Velasco and Hengeveld's (this volume) radical departure from this orthodox FG position: they suggest that predicates, as they are represented in the Fund, lack any frame whatsoever. In their approach the frame is assigned on-line, both by the speaker in production and by the hearer in comprehension. They propose a universal set of frames of which each language makes a subset available to its users. Certain Papuan languages, for example, make no use of trivalent frames, expressing three-place relations through serialization. In formulating an utterance, the speaker selects a particular predicate and then chooses an appropriate frame from the array permitted by her language. In Garcia Velasco and Hengeveld's view, this selection process is guided by the conceptual representation of the predicate in question, but they do not exclude the possibility of creative deviations from the standard choices. The hearer similarly constructs the frame from the coded utterance against the background of the conceptual representations and contextual information available to him. A clear advantage of this proposal is that the various frames associated with one and the same predicate no longer have to be formed by predicate formation rules from some basic frame - for who is to say which frame is basic? A disadvantage, which may be overcome in time, is that the proposal is dependent upon an as yet rather sketchily described conceptual representation. To the extent that the development of that conceptual representation is based upon linguistic evidence, there is a danger of circularity. Without taking a position on this recent and untested proposal, let us turn in the following section to the nature of the predicate frame. What we will find there will apply either within the Fund as envisioned in orthodox FG or to the result of the on-line creation of frames proposed by Garcia Velasco and Hengeveld.

10

J. Lachlan Mackenzie

3. The predicate frame As has been implicit in the preceding discussion, each predicate is associated with a frame, a structured bundle of information that is vital in creating the underlying representation of utterances and in instructing the expression rules to produce the appropriate output. Dik (1997a: 77-104) requires five kinds of information to be included in every predicate frame. These will now be dealt with in turn.

3.1. The form of the predicate

Firstly, the Fund must contain all unpredictable information about the phonological form of the predicate: this is generally given in the form of a string of phonemes, possibly with an indication of tones; where relevant, there will also be an indication of the characteristic accent position (CAP, see Dik 1997a: 448). The extent to which supplementary information is required will vary enormously from one language type to another. For an English noun, the formation of the plural need not be specified in the Fund, since it is fully predictable from the phonological form of the predicate: the familiar allomorphy rules will be located among the expression rules. Only exceptional predicates like ox, CHERUB, KNIFE, etc. will have their plural form specified in the Fund. The gender of the noun can also be left unspecified, since the choice of third-person singular anaphoric pronouns (the only forms affected by gender) is determined by the perceived animacy/non-animacy and, in the case of animacy, the perceived sex of the referent. In English, gender is thus a referential category, and is not associated with the predicate, the ascriptive component of a term. For a German noun, however, the frame of each noun will have to contain information about the morphological paradigm to which that noun belongs, from which the plural form as well as various case forms will become predictable for the expression rules, but also information about which of the three gender classes it belongs to, since that determines the expression of anaphoric reference; neither paradigm class nor gender are predictable from the form of the predicate,

The predicate in Functional Grammar

11

although the two matters are partially interlinked. Even greater complications may be foreseen for the analysis of Athabaskan languages, in which each verbal predicate has an imperfective, a perfective and a future/progressive form, and where "[e] specially for verbs of position ... motion ... or handling, different verb stems are chosen according to the shape, texture, and number of objects involved" (Mithun 1999: 363-364). The purpose of providing all this morphophonological information with each predicate is to reduce the work done by the expression rules to automatic and fully regular formal processes; after all, the underlying representation is designed to contain all the information that the expression rules need to do their work. An alternative architecture is thinkable, in which the expression rules respond not only to the underlying representation, but also, in parallel, consult the Fund for morphophonological information. This would not only be more efficient, but may be psychologically more adequate, reflecting parallel rather than linear production of morphophonological and semantico-syntactic expression.

3.2. The category of the predicate The predicate frame, secondly, will contain a statement of the category, the "part of speech" (Hengeveld 1992), to which the predicate belongs. This is conventionally indicated in square brackets immediately after the predicate in question. The emic distinctions to be made here are those that are relevant for the operation of the expression rules of the language in question. Hengeveld (1992) has shown that languages draw differently from the etic categories {Verb, Noun, Adjective, Adverb} such that a language that has a particular category will also have the categories listed to the left: thus no language will have Adjective that lacks Noun. It follows that there will be languages with only Verbs, and indeed Hengeveld has identified Tuscarora (and some others) as just such a language: a language like that could be said to have no syntactic categories at all. At the same time, languages can be typologised as either rigid or flexible, with an in-

12

J. Lachlan Mackenzie

termediate category of specialised languages, represented inter alia by English: in a specialised language, all four categories are present; in a rigid language, one or more parts of speech is lacking; in a flexible language, a single part of speech is used in different functions. In work since 1992, Hengeveld has shown that his systématisation of parts of speech has considerable predictive power for understanding the properties of languages. Other traditionally recognised parts of speech (Article, Conjunction, etc.) are never used for ascription and thus do not apply to predicates. These will all be introduced by expression rules: thus articles will express various term operators, and subordinating conjunctions will express the semantic functions attaching to embedded or dependent clauses. In orthodox FG, adpositions have similarly been regarded as expressing semantic functions, and therefore (in contrast with generative grammar, cf. Jackendoff 1977) Adposition has not been included in the list of parts of speech, so that adpositions are not regarded as predicates. This seems correct for those that have a purely functional status, such as English by asa marker of the semantic function Agent or Japanese vcaasa marker of the pragmatic function Topic. Mackenzie (1992,2001), however, has argued that certain adpositions in English, namely those that indicate refinements of general semantic functions such as Loc(ative) and Temp(oral), do have ascriptive status and should be included in the Fund. This would entail the recognition of an etic category Adposition. Mackenzie (2001), however, argues for a hypercategory Ad which would cover both Adverb and Adposition - note that he takes spatial and temporal adverbs to represent the basic class of adverbs rather than, as Hengeveld (1992) does, manner adverbs, since the latter, in those languages that possess them, typically appear to be adjectives that require morphological adjustment to fulfil their special function as modifiers of non-nominal predicates.

The predicate in Functional Grammar

13

3.3. Quantitative valency The third type of information contained by the predicate frame is the quantitative valency of the predicate, i.e. a statement of the number of arguments it requires. An SoA may contain an unlimited number of participants. In consequence, a predicate may be accompanied, at least in certain language types, by an unlimited quantity of associated terms. But FG claims that only a restricted number of these, the maximum being three, have argument status, the others all being satellites, i.e. terms that are added to the predicate frame stored in the Fund. The distinction between arguments and satellites has plagued every theory that has proposed such a contrast, and FG is no exception. Textual analysis has shown that many predicates assumed to have a certain number of arguments frequently occur with one or more of those arguments unspecified (cf. Garcia Velasco and Portero Muñoz to appear). Consider the following example, drawn from a television show in which A is asking Β his impression of her work on his house: (15)

A: Do you like? B: Ilove!

Under the standard analysis, the second arguments of the verbs LIKE and LOVE are lacking here; the conceptualisation of the SoAs strongly suggests that the missing terms are not satellites. Since FG spurns deletion rules (Dik 1997a: 19), and even a notion such as zero realisation seems suspiciously close to deletion, it is forced to take recourse to a process of valency reduction to explain such examples: such a move is descriptively adequate, but explains little. A radical alternative suggests itself, one which the proposal by Garcia Velasco and Hengeveld (this volume) mentioned above could presumably accommodate with ease: assume that all predicates are fundamentally avalent but that speakers are typically, but not always, constrained by communicative circumstances, and sometimes by lexical requirements, to specify terms that are dependent upon those

14 J. Lachlan Mackenzie

predicates ("arguments"); they are, however, also free to leave terms unspecified. This would be an application of the Gricean Maxim of Quality to quantitative valency. It is to be expected that verbs, with their prototypically predicative use, will typically be accompanied by arguments, whereas nouns, as prototypical heads of terms, will have less need for such specifications (cf. in this regard Mackenzie's 1994 argument that even relational nouns are avalent in the Fund). To return to (15), we may assume that speakers A and Β have been so committed to expressing their feelings that they place the relevant predicates in utterance-final (focal) position. A further case in point is the analysis of action nominalisations in FG. Dik (1997b: 165) argues that both DESTROY and its nominalisation DESTRUCTION are bivalent, with Agent/Force and Goal arguments. This proposal is based upon the observation that parallelisms of the following kind can be perceived: (16)

a. b.

The enemy destroyed the city. the enemy's destruction of the city

In generative grammar, this kind of observation led to the lexicalist hypothesis (Chomsky 1970) and X-bar theory (Jackendoff 1977). In FG, it has engendered Dik's proposal to locate the observed parallelism in the respective Fund entries of DESTROY and DESTRUCTION and all other such pairs, with the expression rules being primed (1997b: 168) to realise Agent/Force as enclitic 's and Goal as of where these are arguments of a nominal predicate. Mackenzie (1996: 336) has argued that this approach is stipulative rather than explanatory, and above does no justice to the patent fact, which emerges clearly from textual analysis, that the arguments of nominalisations occur much less frequently than those of the corresponding verbs. The alternative he proposes is to regard such nominalisations as DESTRUCTION as avalent, and to permit the speaker to add satellites as the communicative situation demands; in languages such as English, these satellites take the form of possessors, metaphorical owners of the quasi-thing that arises as the result of nominalising a verb.

The predicate in Functional Grammar

15

To abolish quantitative valency altogether may be going too far: whereas English EAT can regularly occur without mention of its Goal, that seems more difficult for DEVOUR; if the second argument of PUT (in the sense of 'place, position') is specified, the third (Locative) argument is generally required, although that is not true of either PLACE or POSITION:

(17)

a. He ate all night. b. ?He devoured all night.

(18)

a. I then positioned the lamp. b. ?/ then put the lamp.

Perhaps the answer is to retain the hypothesis of the avalency of predicates but to recognise, on the basis of empirical analysis of actual language use, a scale of explicitness. Languages also seem to differ in this regard. Some are more explicit in requiring the specification of participants in SoAs than others: thus Mithun finds that predicate-argument structures "are relatively rare in many North American languages" (1999: 187), showing how in Tuscarora "[sentences often consist simply of a verb serving as a predicate".

3.4. Qualitative valency Fourthly, the frame indicates the qualitative valency of the predicate, i.e. the semantic properties of its associated arguments - let us continue to grant that predicates can indeed take arguments. Qualitative valency covers the semantic functions of the arguments and the selection restrictions that apply to them. Arguments are classified as first, second or third (Al, A2, A3) on the basis of the combinations permitted by the language: thus the semantic functions of arguments associated with monovalent predicates will be assigned to Al; with bivalent predicates, the Al will be that argument that shares the same semantic function as the sole argument of monovalent predicates, the other argument being A2; and

16

J. Lachlan Mackenzie

analogously for A3. The semantic functions assigned to Al, regardless of the quantitative valency of the predicate, are determined by the control and dynamism features of the SoA: (19)

SoA

Al

[+control, +dynamism] [-^control, -dynamism] [-control, +dynamism] [-control, -dynamism]

Agent Positioner Processed or Force Zero

The semantic functions assigned to A2 are much more loosely connected to the type of SoA: A2 is typically Goal (= Patient), but in certain non-telic SoAs A2 may have the semantic function Reference; where a spatial meaning attaches to A2, such semantic functions as Locative, Direction and Source may be applied. The sets of semantic functions applying to Al and A2 respectively are non-overlapping: thus FG insists that in (20), the semantic function of the Al of (20a) is distinct from that of the A2 in (20b): (20)

a. b.

The rac^Processed moved. The students Agent moved the

rocfe^.

Whereas languages generally conflate the distinctions between Processed and Goal (in ergative systems) or between Processed and Agent (through Subject assignment), the FG position is supported by such languages as Nez Perce, which Mithun (1999: 229) cites as displaying all three distinctions: (21)

Nez Perce Semantic function Agent Al Processed Al Goal A2

Marking -nim (ergative suffix) -Zero (intransitive subject) -ne (accusative suffix)

The predicate in Functional Grammar

17

But, as Mithun (1999: 229) also observes, "[^ripartite systems such as this ... are rare, probably because they are not maximally efficient". The semantic functions of A3 arguments, in those languages that can be claimed to have these, are typically spatial; with verbs of giving, teaching, etc., where the beneficiary is high on the animacy scale, A3 may be a Recipient. Whereas FG permits some overlap between A2 and A3 (both may bear a spatial semantic function), no overlap between Al and A3 is possible. Across many languages, however, Recipient marking is often found on a solitary Al in an experiential predication, cf. German (22) and Koasati (23; Mithun 1999: 238): (22)

(23)

a. Er gab mir 3SG.NOM give.PRET 1SG.DATIVE Buch book.ACC 'He gave me a book.' b. Mir graust vor der ÎSG.DATIVE shudder.PRES before DEF Ί dread the thought of the operation.'

ein INDEF

Operation. operation

a.

st-am-il ENSTR-1 SG.DAT-arrive-here 'Bring it to me!' b. am-ayóhki-s 1 SG.DAT-feel.acrophobia-PAST Ί have a fear of heights.'

In FG, experiential predicates are analysed as parasitic upon nonexperiential ones, with the sole difference that Exp is appended to the semantic function to trigger the appropriate expression rules. Thus Dative, in both German and Koasati, will be occasioned by either Recipient or ProcessedExp; FG in its present formulation cannot offer any synchronic explanation for this recurrent syncretism.

18

J. Lachlan Mackenzie

Another facet of qualitative valency is the selection restriction. Dik (1997a: 91-97) recognises that selection restrictions have been controversial since their introduction into linguistics by Katz and Postal (1964: 13); cf. McCawley (1971: 219), who refers the utterer of My toothbrush is alive and trying to kill me to a psychiatric clinic rather than a remedial English course. Yet Dik insists that selection restrictions must form part of the predicate frame: (a) there are predicates like blond the argument of which must refer to hair (or by metonymy to the person from whom the hair grows); (b) hearers are able to infer relevant aspects of the meaning of an unfamiliar word from the selection restrictions of the predicates applied to that word; (c) grammaticalisation often involves the loosening of selection restrictions, a fact which could not be captured if they were banned from the grammar; and (d) certain predicate formation rules (see section 4 below) affect only selection restrictions. In the spirit of McCawley's bon mot, however, Dik recognises that selection restrictions are violable, in which case a metaphorical interpretation can arise. Thus (24) transgresses both selection restrictions on GUZZLE, i.e. that the Agent be and the Goal : (24)

My car guzzles gasoline.

Dik assumes that the hearer of (24) and similar utterances observes the clash between the predication that is presented and the selection restrictions of GUZZLE and applies special interpretation strategies to resolve this clash, which involve a quest for a metaphorical understanding of GUZZLE that is comparable with its basic meaning. This view, however, based as it is on the assumption that literal interpretations are the first to occur to the hearer, underestimates the extent to which much of our cognition is metaphorical; top-down processing may well cause the hearer to go directly to the metaphorical understanding without following the garden path that leads to a textually irrelevant literal interpretation (for a careful discussion of this issue, see Steen 1994: 90-94).

The predicate in Functional Grammar

19

3.5. The meaning definition The final component of the predicate frame is the meaning definition. The meanings of predicates are, in distinction to what is proposed in RRG (Van Valin and LaPolla 1997), defined exclusively in terms of other predicates of the same language. This is conducted in a stepwise manner (cf. Dik 1978), such that the meaning of a complex predicate is defined as a specification of a predicate that has one degree fewer of complexity; this defining predicate is in turn also subject to definition, until, at the end of the line, a predicate is reached which cannot be further defined. The embryonic theory of meaning definitions proposed by Dik (1997a: 97-103) has been thoroughly elaborated by Martin Mingorance (1998) and Faber and Mairal Usón (1999) into the Functional Lexematic Model. This model develops the hierarchical view of the Fund, in which the paradigmatic dimension (the relations among predicates) is shown to have as much importance as the syntagmatic dimension that has otherwise been most prominent in FG. In particular, Faber and Mairal Usón demonstrate that the paradigmatic dimension is highly relevant for syntax: the possibilities for the syntactic complementation of English verbs are predictable from the position of those verbs in the paradigmatic hierarchy. Semantics is in this way directly linked to syntactic form, so that the meaning definition becomes much more than an appendage to the predicate frame. Indeed, Faber and Mairal Usón propose replacing the predominantly syntagmatic predicate frame with the "predicate schema", in which both axes receive equal attention. Mairal Usón and Faber (this volume) take this line of argument several steps further, rejecting stepwise lexical decomposition in favour of meaning definitions that quite explicitly pertain to RRG as presented in Van Valin and LaPolla (1997). They are particularly critical of the FG stance on such alternations as those identified by Levin (1993). They point out that FG simply assumes different predicate frames for the various patterns associated with any one predicate, without showing how these are generated; at best, productive relations between predicate frames are handled by predicate formation rules, as we shall see in section 4 below. In a detailed study of

20

J. Lachlan Mackenzie

verbs of cutting (hew, hack, whittle, ...), they devote attention to the actual generation of predicate frames, their proposal being that all the frames within one semantic domain are derived from a single canonical "lexical template". Each instantiated predicate frame (conative, middle, unspecified object, ...) results from a constrained reduction of this maximally specified template. The undoubted gain in explanatory force achieved by this proposal is bought at the expense of introducing a semantic metalanguage, which represents a radical departure from the original principles of FG. However, if it can be shown that this metalanguage can be properly constrained and that its use contributes to enhancing the cognitive adequacy of the model, Mairal Usón and Faber's proposals may well lead to a revaluation of the meaning definition, which they rightly say "need to play a more active role" (this volume: 41). In any case, it will be worth exploring the relation between their position and that of Hengeveld and Garcia Velasco (2001), summarised in section 2 above, who claim that one of the factors determining the speaker's selection of predicate frame is a conceptual representation. Future work will have to determine whether this representation can be equated with the lexical template.

4. Predicate formation rules Basic predicates (listed in the Lexicon) and derived predicates are linked through predicate formation rules (PFRs), which take the former as their input and deliver the latter. PFRs may also form more complex derived predicates from simpler derived predicates. Each PFR affects at least one of the five properties of predicates laid out in section 3; the various kinds of effect will now be studied in the same order.

The predicate in Functional Grammar 21

4.1. Affecting the form of the input predicate PFRs typically, but not always, affect the form of the predicate to which they apply. PFRs typically add affixes (prefixes, suffixes, or infixes) to the input predicate, or extend the input through reduplication, although cases of subtractive morphology have also been observed (cf. Dressler's 1987 observation that Russian nouns such as BIOLOG 'biologist' are derived from BIOLOGIJA 'biology' by subtraction of a portion of the stem). Another possibility is apophony (stem mutation; cf. Junger 1987: 62-96 for examples of relevant binyanimaffecting PFRs in Hebrew). The PFR may, however, have no morphological effect on the input: in English, for example, the Instance Nominalisation is formed from the verb stem without any change of form (RUN [V] > RUN [N]; JUMP [V] > JUMP [N); with polysyllabic input, however, a shift of CAP is generally observed: reMAKE [V] > REmake [N]. The effect need not be limited to the morphology of the input predicate. Thus Dik (1980: chapter 3) argues that causativisation in Dutch involves the introduction of an auxiliary verb laten, and Baron and Herslund (1998) see English support verb constructions such as make a survey, make threats (against), which involve the cooccurrence of a verb of general meaning and an instance nominalisation, as the result of predicate formation.

4.2. Affecting the category of the input predicate The syntactic category of the input predicate is of course changed by PFRs of nominalisation, verbalisation and adjectivalisation. In the derivation of English régionalisation, the nominal input region is expanded into an adjective regional, which in turn becomes a verb régionalisé, which then turns into the noun régionalisation. Other PFRs, however, do not alter the syntactic category of the input: thus in Chukchee (Muravyova 1998: 536), the diminutive suffix -qej applies to nouns to yield other nouns; similarly for the Chukchee collective suffixes -giniw and -mk. Causativisation rules typically apply to verbs to give other verbs, although an adjectival or nominal input is also possible, cf. English broad > broaden, length >

22

J. Lachlan Mackenzie

is also possible, cf. English broad > broaden, length > lengthen, etc.; as will be clear from the example of régionalisé, nouns may also undergo causativisation via an intermediary stage of adjectivalisation.

4.3. Affecting the quantitative valency of the input predicate As to quantitative valency, PFRs may be valency-extending, valencyreducing or valency-preserving. Valency extension has been associated in FG above all with causativising PFRs, which are seen as adding an argument with the semantic function Agent (or Causer) to the array of arguments in the input frame. Dik (1985) points out that the result of this operation is frequently subject to a Prototypical Expression Model (PEM). In other words, the expression rules tend to normalise the output according to the Principle of Formal Adjustment, so that, for example, the demoted Agent in French causative constructions is expressed in (25) as if it were a Recipient, the AgentGoal-Recipient combination being an example of a PEM: (25)

Elle fait laver la 3SG.FEM cause.3SG wash the son mari. 3SG.POSS husband 'She has her husband do the dishes.'

vaisselle dishes

à REC

Valency reduction involves either the suppression of one argument from the array in the input predicate frame or its incorporation into the predicate. In the case of a bivalent predicate, either Al or A2 may be suppressed. Al-suppression in an Action is associated with the speaker's decision not to mention the Agent. Consider as an example the German verb gehen 'go', normally associated with the arguments Agent and Direction. Suppression of the Agent does not affect the verb form of the verb, but calls for an impersonal (i.e. nonreferential) subject es:

The predicate in Functional Grammar 23

(26)

a.

Dann gingen wir ins Gebirge. Then go.PRET 1PL.NOM in.the mountain.COLL 'We went into the mountains.' b. Dann ging es ins Then go.PRET 3 SG.NEUT.NOM in.the Gebirge. mountain.COLL ' We/they/... were off into the mountains.'

A2-suppression, it is emphasised in FG, offers a puzzle to the interpreter: a normally η-place predicate is presented as having (n-1) places. The solution, Dik (1997b: 13-14) claims, will be found in one or more of three possibilities: (a) the A2 should be interpreted as non-specific (and therefore unspecified); (b) the A2 is identical to Al (the reflexive interpretation); (c) if Al is plural, the relation between Al and A2 is reciprocal. I have not been able to find a language in which all three possibilities are present with the same form. English (see (27)) allows the same form for the (a) and (b) interpretations, but requires each other for (c); German (see (28)) must distinguish between (a) and (b), but allows (b) and (c) to be expressed identically, although the adverb gegenseitig may be added to clarify that a reciprocal meaning is intended. Note that German marks valency reduction with interpretations (b) and (c) by means of a reduction marker (R), the "reflexive" pronoun sich. (27)

a. Men do not like to wash (valency reduction; interpretations (a) and (b)) b. Men do not like to wash each other (no valency reduction; interpretation (c))

24

(28)

J. Lachlan Mackenzie

a. Männer waschen nicht gern. man.PL wash.PRES NEG with.pleasure (valency reduction; interpretation (a)) b. Männer waschen sich nicht gern. man.PL wash.PRES R NEG with.pleasure (valency reduction; interpretation (b) or (c)) c. Männer waschen sich gegenseitig nicht man.PL wash.PRES R reciprocally NEG gern. with.pleasure (valency reduction; interpretation (c))

Valency reduction is also at play in incorporation: here the reduced argument survives in the expression, but now loses referentiality. Van Valin and LaPolla (1997: 123) give the following example from Lakhota: (29)

a.

Wiéhása ki chq ki kaksá-he. Man the wood the chop-CONT 'The man is chopping the wood.' b. Wichása ki chqkáksa-he. Man the wood-chop-CONT 'The man is chopping wood.'

Dik (1997a: 12) assumes that a similar rule applies in English, but that the output may not occur as such; it is necessarily input to a further PFR of Agent Noun Formation: cf. work wood > *wood-work > woodworker. Mackenzie (1990: 138-139) considers this possibility for English, but concludes that it is costly to require of the incorporation rule that its output should never be the basis for forming a predication. He prefers to analyse such forms as woodworker as deriving from regular nominal compounding of Ni wood and N2 worker, with no fixed semantic relation between Ni and N2. Ni will be most commonly interpreted as the Goal of the SoA of N2, but other possibilities are equally valid: headworker (Instr), field worker (Loc), star worker (Qual), etc.

The predicate in Functional Grammar 25

Finally, there are PFRs that do not affect the valency of the input. A familiar example of these is Completive Predicate Formation, the rule that links such pairs as the following: (30)

a. b.

Nora stuffed dresses into her suitcase. Nora stuffed her suitcase with dresses.

In both cases, the predicate STUFF is trivalent. What differs is the qualitative valency, Agent-Goal-Direction in (30a) and Agent-GoalInstrument in (30b), with the further complication that the Goal of (30a) corresponds to the Instrument of (30b) and the Direction of (30a) to the Goal of (30b). The rule is so-called because in (30b) A2 is understood as being completely affected: there is no room left in the suitcase, whereas (30a) leaves that possibility open.

4.4. Affecting the qualitative valency of the input predicate (30) is an example of a PFR that affects the semantic functions of the arguments, whereas the valency-extending and valency-reducing rules discussed above do not do so. Nevertheless, we cannot simply assume that valency-affecting PFRs preserve the argument roles and the valency-preserving rules change the roles. De Groot (1989: 147) has shown that Denominal Verb Formation in Hungarian involves both valency-extension and a change in role of the Al; and Diminutive Noun Formation across languages affects only the predicate, while preserving both the quantitative valency and the semantic function of any argument the Noun may have (cf. De Groot 1989:145).

4.5. Affecting the meaning of the input predicate Finally, the question arises whether every PFR affects the meaning of the input predicate. Whereas this generally is the case (for why else have the rule?), Bolkestein and Risselada (1985) have pointed to a PFR in Latin that appears to be sensitive to textual rather than seman-

26

J. Lachlan Mackenzie

tic considerations, and De Groot (1989:144) argues that the apparent Hungarian equivalent to Completive Verb Formation in fact has no comparable semantic effect, the sense of completeness being achieved, if desired, through application of the perfective aspect to the verbal predicate. If there is a semantic difference, De Groot continues, then that lies in the area of telicity.

4.6. Formalizing the Predicate Formation Rule

The conclusion from the preceding overview of the impact of PFRs on input predicates is that none of their effects is obligatory, but that every PFR has some identifiable effect, be it on the form, the category, the valency (qualitative and/or quantitative), or the meaning of the predicate - as far as we know, there are no necessary connections between any of these, although this matter remains to be researched. The predicate formation rule thus has the following general form: (31 )

Input: (f¡: predicate [a]) (xi)sfy... (x„)Sß where (a) α = syntactic category (b) (xi ... x„) are arguments of whatever layer (c) sfy and sfô are semantic functions (d) i > 0 (e) η < 4 Output: (f¡: predicate{-M} [β]) {(xi)sft}...{(xm)sfç} where (a) {-M} is a possible trigger for an expression rule (b) α may or may not be identical to β (c) (γ ...δ) may nor may not be identical to (ε ...ζ) (d) i > 0 (e) m < 5 Meaning: Specification of the effect, if any, on the meaning of the input predicate Conditions: ...

The predicate in Functional Grammar

27

FG also permits a variant on (31), in which there are two inputs, and the output serves to combine the two inputs into one. This applies, most obviously, to instances of compounding: thus English library book derives by PFR from two nominal input predicates LIBRARY and BOOK. An analogous analysis suggests itself for the linking of verbal predicates in serial verb constructions in those cases where one predicate operator applies to multiple verbs in the series. PFRs are productive in the sense of applying only where the relationship between input and output is fully regular and systematic. Nevertheless, FG makes provision for the possibility that a predicate with the appropriate meaning and use is already available in the Fund: then the PFR is blocked for that predicate. This explains the absence of such potential Agentive Nouns as English *cycler, * studier, * inhabiter (cf. Mackenzie 1990: 134). Conversely, not every form that satisfies the output requirements of a PFR is derived by that rule: through lexicalisation, a derived predicate can move into the Lexicon, cf. the many senses of sleeper other than 'one who sleeps'. Many PFRs are subject to conditions which are added as a fourth clause to the format shown in (31) above: for example, as was mentioned above, Bolkestein and Risselada (1985) found that a certain valency-preserving rule of Latin was subject to discourse conditions. An example of a rule that is subject to a range of conditions, phonological, morphological, valency-related, semantic and pragmatic, is Dutch Durative Nominalisation, studied by Mackenzie (1985). This rule, which attaches the prefix ge- to the stem of a verb to yield an uncountable neuter noun, is formulated as follows: (32)

Input: (f¡: predicate [V]) (xi)Ag/Fo/Proc Output: (-e f¡: predicate-DUR [N]) (yi)p0ss where -c = non-countable -DUR is a trigger for the prefix ge(y¡) is a satellite Meaning: The output designates an SoA with protracted or habitual occurrence of the SoA designated by the input

28

J. Lachlan Mackenzie

Not all verbal predicates lend themselves equally willingly to this rule: in the corpus analysed by Mackenzie, none of the input predicates began with a consonant; there were also no examples of input predicates with an inseparable prefix and only one with a separable prefix; as is suggested by the formulation of the input in rule (32), the great majority of the input predicates in the corpus were monovalent; all of the input predicates designated dynamic SoAs; the great majority designated a perceptible SoA, with a strong preference for aural perception; many, but certainly not all, input predicates had a negative connotation; and there is, finally, a statistically significant greater frequency of this nominalisation in novels and novelettes than in, say, popular-science articles. None of the preceding can be formulated as an absolute condition upon the application of the rule: thus it is possible to find it (albeit rarely) applying to bivalent predicates, and it is certainly not excluded from writings on popular science. Mackenzie suggests that there is a prototypical input to this rule, and example of which would be Dutch JENGEL 'whine', which satisfies all the criteria. More generally, where a PFR is less than fully productive, it may be worth examining whether its domain of application may be circumscribed in terms of prototype theory. The PFRs discussed in this section all derive predicates from predicates. Let us now turn to a special class of PFRs, which creates predicates from terms.

5. Term predicate formation FG allows for predicates to be derived from terms by means of a special subset of PFRs known as Term Predicate Formation Rules (TPFRs). The output of a TPFR is identical to the output in (31) above, with the difference that the derived predicate has no syntactic category. TPFRs are valency-extending, as can be seen from the following general schema of such rules:

The predicate in Functioned Grammar 29

(33)

Input: (t)|sfal where (a) (t) = term (b) |sfa| is an optional semantic function Output: {(t)|sfa)> (Xi)zero where {...} is a derived predicate

Note that the argument of the derived predicate has the semantic function Zero: the resultant predication designates a State. The formation of predicates from terms is designed to account for the observation that expressions that otherwise function as arguments and satellites can also appear with predicative function in stative predications. The clear parallelism in meaning and function between the non-italicised segments of (34a) and (34b) justifies their being given a fundamentally identical analysis: (34)

a. She teaches. b. She is a teacher.

The option of assigning a semantic function to the input arises in such cases as (35b): (35)

a. b.

They are working, They are at work.

The predication of (35b) will correspondingly be represented as in (36): (36)

Pres e,: (fi: {(e2: (f2: work)

[N])LOC})

(dmxi: [-S, - A ] ) z e r o

with the preposition ai realising the semantic function Loc(ative). Given the close connection between possession and location, possessive predicates are given analogous treatment: (37)

a. b.

That book is mine. Pres ei: (ft: {(dlxi: [+S,-A]) p 0 s s}) (dldistx,: (f2: book [N]))zero

30

J. Lachlan Mackenzie

And existentials, too, are analysed in the same vein, with the interesting difference that the predicate is here formed from a zero-location; the argument is necessarily presentative, i.e. NewFocus (Hannay 1985): (38)

a. There are difficulties. b. Pres ei: {(0)LOC} (dmei: (f2: difficulty [N]))zen>NewFoc

Controversy has arisen concerning the analysis of identifying predications such as (39), which Dik (1997a: 205) treats in the same way as classifying predications such as (37b): (39)

George W. Bush is the President of the United States.

Whereas a term that enters into a TPFR loses its ability to refer, the term the President of the United States in (39) is fully referential. This issue was first taken up within FG by Keizer (1990). She points out that the distinction between classifying and identifying constructions cannot be correlated, as Dik does, with an indefinite and definite term predicate, respectively: thus (40), as an answer to Who is John?, is indeed identifying, but as an answer to What is John? (in an appropriate context) is classifying: (40)

John is the author.

Keizer suggests that term predicate formation can be retained for classifying uses, but that identifying uses call for an underlying predication with the predicate BE and two Zero arguments, one for each of the two referential expressions in an utterance such as (39). Hengeveld (1992: 80) abolishes term predicate formation in favour of permitting terms to be used directly as predicates under the (fi) variable; he retains, however, Dik's equation of identification with definiteness and classification with indefiniteness:

The predicate in Functional Grammar 31

(41)

Hengeveld (1992: 91; innocuously adapted for consistency, JLM) Identification (fi: (dxi)) (x2)z«o Classification (fi: (ixi)) (x2)zero

Keizer (1992b), who retains her critique of Dik, also rejects term predicate formation but permits the definite and indefinite operators to apply to the (f¡) variable. This allows her to propose the following analysis of the two types of construction: (42)

Keizer (1992b: 347-349) Identification (a¡)Sfl (oj)Sfi where α is a variable over entity-types Classification (f¡) (a¡)Sf

In Keizer's approach the predication in (34b) would be analysed as in (43a), while Hengeveld's analysis would be as in (43b): (43)

a. Pres ei: (ifi: teacher [N]) (dlxi: [-S, -A, - M ] ) z e r o b. Pres eu (fi: (ilxi: teacher [Ν])) (dlxu [-S, -A, -M])zero

The dispute may be finally resolved in the framework of the new architecture for FG proposed by Hengeveld (to appear). Here, in the interactional level, a fundamental distinction is drawn between two kinds of subact, ascription (represented by T) and reference (represented by R). At the representational level, both classifying and identifying expressions may be shown in the same format; but they will differ at the interactional level in that the former involves the Tvariable and the latter the R-variable.

32

J. Lachlan Mackenzie

6. The grammaticalisation of predicates Predicates belong to the Fund, but some are potentially subject to grammaticalisation: as a result, they may lose, partially or fully, their ascriptive function, now appearing in the output of predicate formation or expression rules. Olbertz (1998) offers a comprehensive study of verbal periphrases in Spanish, observing how various verbal predicates have developed grammatical meanings in the areas of aspect and modality; to a greater or lesser extent, each of these predicates has adopted a non-ascriptive function. Predicates expressing completion, location, direction or manner of movement have in Spanish come to assume the mantle of aspect markers, while predicates expressing obligation and permission have taken on epistemic-modal senses, and a thinking verb (pensar) has evolved into an expression of "participant-oriented volitional modality", i.e. "intend, be determined to, feel like". Olbertz shows that in each case of partial periphrasis a conceptual expansion has occurred, and that with a fully periphrastic construction there has been a total conceptual shift. Some predicates still display residues of their lexical base, in which case she proposes analysis by means of a PFR; where such a predicate is fully periphrastic, the predicate may already be better analysed as an operator, she argues. Where there is no residue of the lexical base, the predicate is in every case treated as expressing an operator, i.e. is fully integrated (in its relevant uses) into the grammar. A similar picture, I would claim, applies to adpositions. Whereas some adpositions, those that are ascriptive in function, belong in the Fund (see section 3.2 above), others belong to the grammar, expressing the semantic, syntactic or pragmatic function of the term to which it attaches. The latter type seem generally to have arisen historically through grammaticalisation of a predicate. For example, adpositions used to express the semantic function Agent (English by, Dutch door 'through', German von 'from', etc.) have their origins in various spatial adpositional predicates. Other grammatical adpositions appear to have derived from the grammaticalisation of a verbal predicate that has lost some of its ascriptive status. In languages like English, the verbal predicate in question is non-finite, and thus already less than

The predicate in Functional Grammar

33

fully verbal: given your objections, concerning the proposal, come Christmas. In other languages verbs that enter into serial constructions yield new prepositions: thus in Mandarin Chinese the marker of the Goal function bä derives from an ancient verb of that form, used in a serial construction, and meaning 'take'. Nominal predicates, often those designating parts of the human body, also regularly become adpositions. This operation appears to take place first in the Fund as recategorisation. The scenario typically involves the introduction of a noun as head of a term; this noun gives further specificity to a spatial expression. The noun, being functionally subservient to the locative relation then gradually loses its categoriality, just as the locative marker also loses its rationale and may be reduced or even disappear. As a result, what was the nominal predicate, which retains its ascriptive value, now takes on the full functional load of expressing the locative relation and is recategorised as a spatial adposition. This adposition may then go on to be grammaticalised as a marker of a function. We may, not entirely speculatively, schematise the development as follows: (44) Stage 1 on the mountain (dip,: (fi: mountain [N]))u>c further specification of position at the summit desired Stage 2 on the top of the mountain (dipi: (fi: top [N] (dlp2: mountain [N]W))LOC specification through introduction of new head noun top Stage 3 on top of the mountain (pi: (f,: top [?] (dlpa: mountain [N]>Rrf)Xx>c partial decategorisation of noun top, which loses its determiner2 Stage 4 on top of the mountain (pi: (f^ top [Ad] (dlp2: mountain [ N ] ) ^ ) ^ recategorisation of noun as adpositional predicate Stage S atop the mountain (pi: (fi: top [Ad] (dlp2: mountain [ N ] ) ^ ) ) ^ phonological attrition (univerbation)

On top of (but not the regionally restricted atop) is now available for more abstract grammatical uses, e.g. to express the semantic function Addition: (45)

On top of being a nice guy, he's a brilliant linguist.

34

J. Lachlan Mackenzie

7. Conclusion The purpose of this article has been to give an overview of the FG view of the predicate as a linguistic item with ascriptive function. We have seen that each predicate is in correspondence with a State of Affairs and that the aspectual characteristics of the predicate flow from the conceptualisation of that SoA. We have also mentioned investigations of the possibility that many other facets of the conceptualisation also impact the predicate. This research has led to a radical rethink of the predicate frame, ranging from a thoroughgoing expansion of the meaning representation to a proposal to do away with the frame altogether within the Fund. We have also revisited some of the debates surrounding predicate formation, and have seen that PFRs retain their potential to generate worthwhile research. And we have concluded by introducing the diachronic dimension, showing how certain predicates may come to lose their ascriptive function and take on a grammatical role.

Notes 1. The correlation is, however, not so straightforward as Hengeveld suggests here. In a webpage about Salvador Dalí . we read A los que ya entonces le acusaban de ser un loco les decía: «La única différencia entre yo y un loco, es que yo no estoy loco». ('To those who then accused him of playing the fool, he said "The sole difference between me and a fool is that I am not foolish'"), where ser is used (with a nominal predicate) for the progressive SoA of playing the fool and estar (with an adjectival predicate) for the non-progressive SoA of (not) being foolish. I am grateful to Pilar Guerrero Medina for discussion of this example. 2. It also loses its possibility for pluralisation: cf. on the tops of the mountains, *on tops of the mountains.

The predicate in Functional Grammar

35

References Baron, Irène and Michael Herslund 1998 Support verb constructions as predicate formation. In: Hella 01bertz, Kees Hengeveld and Jesús Sánchez García (eds.), 99-116. The Structure of the Lexicon in Functional Grammar. Amsterdam/Philadelphia: John Benjamins. Bolkestein, A. Machtelt and Rodie Risselada 1985 De tekstuele funktie van valentie: Latijnse drieplaatsige predikaten in kontekst [The textual function of valency: Latin threeplace predicates in their context]. TTT: Tijdschrift voor Taal- en Tekstwetenschap 5: 161-176. Butler, Christopher 2001 A matter of GIVE and TAKE: Corpus linguistics and the predicate frame. In: Ma Jesús Pérez Quintero (ed.), Challenges and Developments in Functional Grammar, 55-78. (Revista Canaria de Estudios Ingleses 42.) La Laguna: Servicio de publicaciones. Chomsky, Noam 1970 Remarks on nominalization. In: Roderick A. Jacobs and Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar, 184-221. Waltham, M.A.: Ginn. Cornish, Francis 2000 The Functional Grammar conception of discourse anaphora. Working Papers in Functional Grammar 73. Dik, Simon C. 1978 Stepwise Lexical Decomposition. Lisse: Peter de Ridder Press. 1980 Studies in Functional Grammar. London/New York: Academic Press. 1985 Formal and semantic adjustment of derived constructions. In: A. Machtelt Bolkestein, Casper de Groot, and J. Lachlan Mackenzie (eds.), 1-28. Predicates and Terms in Functional Grammar. Dordrecht: Foris. 1989 The Theory of Functional Grammar. Part 1: The Structure of the Clause. Dordrecht: Foris. 1997a The Theory of Functional Grammar. Part 1: The Structure of the Clause. Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter. 1997b Hie Theory of Functional Grammar. Part 2: Complex and Derived Constructions. Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

36

J. Lachlan Mackenzie

Dressier, Wolfgang U. 1987 Subtraction in a polycentristic theory of Natural Morphology. In: Edmund Gussmann (ed.), Rules and the Lexicon, 67-78. Lublin: Redakcja Wydawnictw Katolickiego Uniwersytetu Lubelskiego. Faber, Pamela B. and Ricardo Mairal Usón 1999 Constructing a Lexicon of English verbs. Berlin/New York: Mouton de Gruyter. Fortescue, Michael 1992 Aspect and superaspect in Koyukon: An application of the Functional Grammar model to a polysynthetic language. In: Michael Fortescue, Peter Harder and Lars Kristoffersen (eds.), Layered Structure and Reference in a Functional Perspective, 99-141. Amsterdam/Philadelphia: John Benjamins. Garcia Velasco, Daniel and Kees Hengeveld this volume Do we need predicate frames? Garcia Velasco, Daniel and Carmen Portero Muñoz to appear Understood objects in Functional Grammar. Working Papers in Functional Grammar. Grinevald, Colette 2000 A morphosyntactic typology of classifiers. In: Gunter Senft (ed.), Systems of Nominal Classification, 50-92. Cambridge: Cambridge University Press. Groot, Casper de 1989 Predicate Structure in a Functional Grammar of Hungarian. Dordrecht: Foris. Hannay, Michael 1985 English Existentials in Functional Grammar. Dordrecht: Foris. Hengeveld, Kees 1992 Non-verbal Predication: Theory, Typology, Diachrorty. Berlin/New York: Mouton de Gruyter. to appear The architecture of a Functional Discourse Grammar. In: J. Lachlan Mackenzie and María de los Angeles Gómez González (eds.), A New Architecture for Functional Grammar. Berlin/New York: Mouton de Gruyter. Jackendoff, Ray 1977 X' Syntax: A Study of Phrase Structure. Cambridge, M.A.: MIT Press. Junger, Judith 1987 Predicate Formation in the Verbal System of Modem Hebrew. Dordrecht: Foris.

The predicate in Functional Grammar

37

Katz, Jerry J. and Paul Postal 1964 An Integrated Theory of Linguistic Description. Cambridge, M.A.: MIT Press. Keizer, M. Evelien 1990 Referentiality and the representation of predicate nomináis. In: Mike Hannay and Elseline Vester (eds.), Working with Functional Grammar: Descriptive and Computational Applicatons, 145-166. Dordrecht: Foris. 1992a Predicates as referring expressions. In: Michael Fortescue, Peter Harder and Lars Kristoffersen (eds.), Layered Structure and Reference in a Functional Perspective, 1-27. Amsterdam/Philadelphia: John Benjamins. 1992b Reference, predication and (in)definiteness in Functional Grammar: A functional approach to English copular sentences. Ph.D. dissertation, Vrije Universiteit Amsterdam. Levin, Beth 1993 English Verb Classes and Alternations: A Preliminary Investigation. Chicago: University of Chicago Press. McCawley, James D. 1971 Where do noun phrases come from?. In: Danny D. Steinberg and Leon A. Jakobovits (eds.), Semantics: An Interdisciplinary Reader in Philosophy, Linguistics and Psychology, 217-231. London: Cambridge University Press. Mackenzie, J. Lachlan 1985 Genominaliseer [Nominalizing]. TTT: Tijdschrift voor Taal- en Tekstwetenschap 5: 177-198. 1987 The representation of nominal predicates in the fund. Working Papers in Functional Grammar 25. 1990 First-argument nominalization in a Functional Grammar of English. Linguistica Antverpiensia 24: 119-147. 1992 Places and things. In: Michael Fortescue, Peter Harder and Lars Kristoffersen (eds.), Layered Structure and Reference in a Functional Perspective, 253-276. Amsterdam/Philadelphia: John Benjamins. 1994 Relationality and nouns: where grammar hands over to pragmatics. In: Ronny Boogaart and Jan Noordegraaf (eds.), 155-160. Nauwe betrekkingen [Close Relationships]. Münster: Nodus. 1996 English nominalizations in the layered model of the sentence. In: Betty Devriendt, Louis Goossens and Johan van der Auwera (eds.), Complex Structures: A Functionalist Perspective, 325355. Berlin/ New York: Mouton de Gruyter.

38

J. Lachlan Mackenzie

2001

Adverbs and adpositions: Hie Cinderella categories of Functional Grammar. In: Maria Jesús Pérez Quintero (ed.), Challenges and Developments in Functional Grammar, 119-135. (Revista Canaria de Estudios Ingleses 42.) La Laguna: Servicio de publicaciones. Mairal Usón, Ricardo and Pamela Faber this volume Functional Grammar and lexical templates. Martin Mingorance, Leocadio 1998 El modelo lexemático-fimcional. Edited by Amalia Marín Rubiales. Granada: Universidad de Granada. Mithun, Marianne 1999 The Languages ofNative North America. Cambridge: Cambridge University Press. Muravyova, Irena A. 1998 Chukchee (Paleo-Siberian). In: Andrew Spencer and Arnold M. Zwicky (eds.), The Handbook of Morphology, 521-538. Oxford: Blackwell. Olbertz, Hella 1998 Verbal Periphrases in a Functional Grammar of Spanish. Berlin/ New York: Mouton de Gruyter. Rijkhoff, Jan 1991 Nominal aspect. Journal of Semantics 8:291-309. Steen, Gerard 1994 Understanding Metaphor in Literature. London: Longman. Van Valin Jr., Robert D. and Randy J. LaPolla 1997 Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Functional Grammar and lexical templates Ricardo Mairal Usón Pamela Faber

0. Introduction1 The question of argument structure has become one of the key issues in lexical semantics. Research in this area is mainly concerned with the explanatory power of lexical representations, and the predictive power of the lexicon, or more specifically, the way arguments project onto syntax. Each linguistic theory addresses this question from a different perspective, and in this respect, Simon Dik's Functional Grammar (FG) is no exception. Despite the fact that much has been written on this issue, it is far from being resolved. We would like to add impetus to the original FG working program on argument structure by bringing the interaction of semantic and syntactic information to the foreground. In this article, we place the FG perspective on argument structure within the wider framework of other proposals in order to see how it can be made more powerful and robust. As first formulated in Dik (1978a), the lexicon has a prominent position in FG theoiy. Lexical representations, encoded in terms of structured predicate frames, constitute the initial step in the analysis of the underlying structure of the clause. This means that predicate frames can be regarded as "blue prints" for predications since the predicate structure is built around them. According to Dik (1997: 79), a predicate frame contains the following information: - Predicate variable: Each predicate is characterized by a variable fi, which symbolizes the property or relation designated by the predicate. This notational feature signals anaphorical reference to properties or relations within a text.

40

Ricardo Mairal Usón and Pamela Faber

- Predicate form: The form of the predicate contains information concerning the phonological representation of the predicate together with specifications of tone and characteristic accent position, if applicable. - Sub-category : The (sub-) category of the predicate is marked by a label (Ν, V, A, etc.). - Quantitative valency: The number of arguments that a given predicate takes. - Qualitative valency: The semantic roles of the arguments in the SoAs designated by the predication. - Selection restrictions: Selection restrictions specify the semantic or pragmatic nature of the arguments and are specified in terms of a range of a feature model which contains fillers like "animate", "human", etc. Another type of information that is introduced into the notion of predicateframeis that of a meaning definition. FG presents a system of lexical representation which Dik (1978b) terms Stepwise Lexical Decomposition. One of its principal corollaries is that meaning definitions are encoded in terms of meaning postulates, which are defined by natural language phrases. The idea behind this procedure is that meaning definitions constitute a web in which more specific predicates are defined in terms of more general ones. The explanatoiy power of Dik's proposal can best be judged within the context of the following general types of lexical representation that have been postulated: - Lexical representations with only syntactically relevant information (cf. RRG's logical structures or Rappaport and Levin's (1998) lexical templates). This format excludes other primarily semantic factors, which play a fundamental role in the meaning definition of a lexical item. However, by focusing only on grammatically relevant factors, the resulting format of the syntaxsemantics algorithm, though precise and elegant in its formulation, is incomplete in its structure because of the lack of a deeper semantic representation.

Functional Grammar and lexical templates

41

- Lexical representations with an enriched semantic component together with a rich inventory of syntactic constructions which form the core grammar of a language. Such constructions follow a decompositional system and only capture those elements which have a syntactic impact. What is interesting is the way the meaning of predicates and constructions fuse. The basic claim behind this approach is that lexicon and grammar form a continuum, and consequently, there is no need to specify linking rules. Instead, the representation of constructions themselves determine which predicates can fuse with a certain construction. As a result, there is no need to postulate a new lexical entry for each syntactic configuration since the syntactic structures a predicate subcategorizes for cannot be predicted from its meaning definition. In this approach, constructions are the essence of grammar, while lexical representations accommodate themselves to constructional meaning. The problem with this approach is that the linking algorithm is largely unspecified. Some of the major exponents of this paradigm come from Cognitive Linguistics, more specifically, the work done within Construction Grammar (cf. Goldberg 1995; Langacker 1987, 1991; Lakofif 1987, etc.). - Ontologically-driven lexical semantics: Following Nirenburg and Levin (1992), the task of ontology-driven lexical semantics is to determine and specify meanings of lexical units and suggest how they must be represented (in the lexicon) and how they contribute to the overall representation of text meaning. FG representations belong to the first group, although with the exceptional premise that no metalanguage is used. The striking thing about the approaches in this group is that they only focus on those aspects which are grammatically relevant, thus reducing to a bare minimum the multifaceted nature of word meaning. In contrast to other theories, FG incorporates a meaning definition into the system, but for FG argument structure to become more semantically powerful, an articulated semantic theory must be developed for their lexical representations. This signifies that meaning definitions, resulting from stepwise lexical decomposition, need to play a more active role

42

Ricardo Mairal Usón and Pamela Faber

than they actually do. Consider the format of the predicates hew and prune from the lexical class of CUT verbs formalized in terms of predicate frames: (1)

hew [V] (xi: animate^ (X2: hard material^ df = cut [V] (xi)Ag (X2)GO (σι: rough & difficult)Manner

(2)

prune [V] (xi: animate) Ag fa'· branches)oo (X3: trees/bushes)source df = cut [V] (xi)Ag (X2)GO (X3: trees/bushes)source [Fut ej: [grow [V](x2)(x4:good[A]) Manner ]Purp

Evidently, these verbs are related semantically, but the only evidence of this in the preceding representations is the fact that both are defined in terms of cut. No mention is made of other types of semantic relations evident in the differentiating information, or the fact that semantic relatedness is directly connected with shared syntactic behavior. In the last few years there have been a number of important contributions in FG lexical semantics (Vossen 1994,1995; Meijs 1990; Meijs and Vossen 1992; Schack Rasmussen 1994; Olbertz et al. 1998) and the work done within the Functional Lexematic Model (henceforth FLM) (Martin Mingorance 1995; Faber and Mairal 1998, 1999). Although these proposals all coincide in regarding the lexicon as the storage place for syntax, they also share the view that there should also be a mechanism to explain how argument structure, as formalized in predicate frames, interacts with the information in meaning definitions. This is one of the cornerstones of the FLM (Faber and Mairal 1999), as well as the approach described by Schack Rasmussen (1994), who draws on Jackendoff s Conceptual Structures as a mechanism to conflate both types of information. Although our formulation of an alternative proposal to the predicate frame is basically motivated by the preceding considerations, it also coincides to some degree with Hengeveld's (1992: 80, 90-94) proposal for "general predicate frames".

Functional Grammar and lexical templates 43

In principle, this may sound contradictory to FG in that Dik (1997: 101-102) claims that meaning definitions do not have a direct role in the construction of the underlying clause structure. Nevertheless, he admits that under certain circumstances meaning definitions may be "unpacked" when necessary. It is our belief that this is one of the most promising lines of research in FG today. In the original FG lexicon, lexical entries are represented by a predicate frame and a meaning definition, but no explanation is given of how predicate frames are generated or to what extent a predicate's meaning may be constrained by its complementation patterns. Nor is any mention made of the role of stepwise lexical decomposition in the design of lexical representations. The FLM has gone a step forward in proving the adequacy of postulating a lexicon organized into coherent lexical classes, a methodological position that has been endorsed in much of the recent work done within lexical semantics. In fact, in the following example, Dik (1997: 85) also acknowledges the importance of this type of information in predicting syntactic behavior. Since only predicates of movement can be combined with directional satellites, he suggests the addition of a feature [move] to the predicate frame: (3)

walk [V, move, atelic] (xi: )Ag

This is important since the organization of predicates into lexical classes shows how syntactic configurations can be largely predicted from the semantics of predicates. Our proposal takes Dik's assumption further by integrating the information in predicate frames into more syntactically-driven RRG lexical representations, and enhancing the semantic component by basing it on the FLM system of articulated semantic classes. Our proposal thus redesigns predicate frames, and, at the same time, enhances RRG lexical representations. The first section of this article presents the methodological corollaries of the new proposal with the aim of showing why argument structure in FG should be further elaborated so that the resulting lexicon is in greater harmony with the functional orientation of the theory. Section 2 focuses on the description of the lexical template, the

44

Ricardo Mairal Usón and Pamela Faber

theoretical construct which represents the salient conceptual information within a given lexical class. Section 3 shows how lexical templates can be applied to the lexical class of CUT verbs. Section 4 describes the structural configurations of this class of predicates in terms of lexical templates. Finally, Section 5 gives a brief account of how lexical templates can be related to their different syntactic configurations.

1. Towards a new conception of argument structure in Functional Grammar Initially, the primary focus of the FLM was the development of an enriched version of the lexicon in Simon Dik's FG (Faber and Mairal 1999). However, after structuring the lexicon in articulated semantic classes, we found that this type of organization was highly informative about the relation between semantics and syntax. As a result, we are now in the process of designing a syntax-semantics interface based on an inventory of lexical templates which operate within lexical classes. This entails important changes in Functional Grammar and also brings a number of fundamental issues to the foreground: (i) The design of a new system of lexical representation based on a metalanguage·. The metalanguage proposed is part of a definitional system which conflates both the semantic and the syntactic information into one unified format, i.e. lexical templates. Lexical templates grow out of the need to account for both the semantic and syntactic behavior of lexical units, thus dispensing with the need to keep the semantic definition and the predicate frame of lexemes in separate compartments. This means that lexical templates are conceived as a reformulation of the orthodox FG argument structure proposal into one unified, compact representation. This is an evident departure from one of the key axioms of stepwise lexical decomposition, which in its final version, would ultimately lead to a set of predicates that cannot be defined any further. The question of whether these undefinables would have the status of concealed primitives, which can be used in the

Functional Grammar and lexical templates

45

construction of a more cognitive-conceptual structure, still remains unanswered. In order to design a lexicon which captures both syntactic and semantic properties as well as generalizations, we need a more powerful explanatory device than that provided by natural language phrases. As Butler (this volume) shows, stepwise lexical decomposition functions fairly well for the textbook examples, but shows a rather limited potential in terms of establishing a definitional system with sufficient explanatory power. In order to formulate a system powerful enough to capture the full range of significant linguistic generalizations within a given lexical class, we adopt a more abstract mechanism than that provided by language-dependent phrases. Our lexical templates are in accordance with the type of argumentation employed for grammatical operators where a metalanguage is used, and are represented by means of a modified version of the logical structures of Role and Reference Grammar (RRG) (Van Valin 1993a, 1993b), complemented with the FLM semantic component (ii) An inventory of lexical templates which operate within lexical classes (Faber and Mairal 1998): We start from a lexicon organized into coherent semantic classes, and then explore how a lexical class can be represented under one unified structure which captures the full range of syntactic and semantic factors characteristic of that semantic area. Consequently, information in lexical entries can be reduced to a minimum and important generalizations can be posited across different types of predicates. This theoretical move does not entail a radical departure from FG, but rather complements in many respects Hengeveld's (1992: 80, 92-94) "general predicate frames". The resulting lexicon would thus consist of a set of lexical templates which would interact in non-trivial way with their respective predicational representations. (iii) The format of a lexical template, consisting of a set of internal and external variables, each capturing the conceptual substance of a lexical class: The format of a lexical template in terms of internal and external variables sheds light on selection restrictions, one of the

46

Ricardo Mairal Usón and Pamela Faber

most pervasive issues in FG. We show that the internal properties of arguments are anchored to an ontology of concepts. This signifies a division between the conceptual and the lexical level as word meanings are represented partly in the lexicon and partly in the ontology. The fact that the format of a lexical template consists of both internal and external variables coincides with Mackenzie's (1986: 18) proposal of "arguments (x)", "implied satellites (y)", and "non-implied satellites (z)". (iv) The development of a set of lexical rules, which establish the systematic relations between the lexical template and the different syntactic configurations relevant to each predicate·. Since we postulate a maximal representation for those predicates within a lexical class, it is also necessary to account for the actual syntactic structures of those predicates. We thus establish a lexical modeling principle, which explains how the lexical template fuses with its different syntactic configurations (cf. Sections 4 and 5). Although we only outline the modeling process here, this principle generates a corresponding set of lexical rules, which have a wider scope than predicate formation rules in the sense that the scope of the former actually covers lexical classes. Accordingly, each lexical class thus becomes a grammar in itself, a lexical grammar. Given that a full account of these methodological principles is beyond the scope of this article, we will focus on the notion of a lexical template as applied to the lexical class of CUT verbs. A more explicit account of the actual form of lexical rules will appear in future publications (cf. Mairal Usón 2002). In order to exemplify our approach, we have taken the lexical class of CUT verbs, which has a characteristic set of syntactic and semantic features. Firstly, as claimed above, the fact that FG postulates a separate predicate frame for each syntactic configuration signifies that it makes no allowance for information shared by sets of predicates, and as a result, is unable to account for regularities such as the distinctive sets of syntactic alternations that characterize cer-

Functional Grammar and lexical templates

47

tain classes of predicates. The predicate frames of certain CUT verbs are shown in the following examples: (4)

slash [V] (xi: animate^ (X2: object^ df = cut [V] (xi)Ag (x2>Go (X3: sharp-edged tooOinsmunent (σι: movement [Ν]: quick [A] & strong [A] & swinging [A])Manner

(5)

lop [V] (xi: animate)Ag (X2: o b j e c t ) ^ df = cut [V] (xi)Ag (X2>G0 (X3: object)source

(σι: stroke [N]:

quick [A] & Strong [A])Manner (6)

slice [V] (xi: animate)Ag (X2: object)G 0

df = cut [V] (xi)Ag (x2)go (X3: pieces [N]: thin[A]: flat [A])Result?

(7)

chop [V] (xi: animate)Ag (X2: object)Go df = cut [V] (xi)Ag (X2)GO (X3: pieces) Result? [Iter ei: [hit [V] (Xl)Ag (X2>Go (σι: sharp-edged tOOl)instrument]Means

(8)

hack [V] (xi: animate)Ag (X2: object^ df = cut [V] (xi)Ag (X2)GO (X3: pieces: uneven)Rcsuit? (σι: way [N]: rough [A]: violent [A])Manner

(9)

whittle [V] (xi: animate)Ag (X2: wood)Qo df = cut [V] (xi)Ag (X2)GO (X3: size [N]: small [A]) R esuit? [ei: [remove [V] (xi)Ag (X2:pieces [N]: small [A]: thin [A])GO ]Manner

These lexical entries, represented as predicate frames, show the problem of using natural language phrases for lexical representation. For example, FG's inventory of semantic functions cannot satisfactorily account for the Result component codified in CUT verbs. This explains why we have used "?" to mark the impossibility of encoding this feature in FG terms. It is also difficult to understand why there is a repetition of the argument structure of the predicate and the meaning definition. There are two separate components in these represen-

48

Ricardo Mairal Usón and Pamela Faber

tations without any indication of the way they interact. Another important drawback is the impossibility of accounting for the different alternations these predicates show (cf. example 10). Dik's predicate formation rules do not say in which area(s) of the lexicon such alternations occur. In contrast, the type of lexical rules in our proposal operate within semantic classes to account for the syntactic variations in each lexical template. In this way, lexical templates give lexical productivity a new dimension in the sense that expression rules and lexical rules interact in non-trivial ways.3 A case in point can be found in the CUT verbs, which, according to Levin (1993), show the following alternations: (10)

CUT

ALTERNATION

EXAMPLE

Causative/inchoative alternation

Margaret cut the bread. *The bread cut. Margaret cut the bread. Margaret cut at the bread. I cut the twig offthe branch. I cut the twig and the branch apart. Margaret cut herself on the thumb. Margaret cut her thumb. The butcher cuts the meat. The meat cuts easily. Margaret cut the meat. *Margaret cut. Margaret cut her finger. *Margaret cut. Margaret cut herself. *Margaret cut. I cut the bread with this knife. The knife cut the bread I cut the bread with this knife. This knife cuts the bread The knife doesn 't cut. Carol cut herself.

Conative alternation Apart reciprocal alternation (transitive) Body-part possessor ascension alternation Middle alternation Unspecified object alternation Understood body-part object alternation Understood reflexive object alternation Instrument subject alternation Characteristic property of instrument alternation Unintentional interpretation with reflexive object

Functional Grammar and lexical templates Unintentional interpretation with body-part object Inalienably possessed body-part object Path phrase

49

Carol cut her finger. Carol cut her/*his [body-part]. Carol cut the paper from one end to the other. His throat was cut from ear to ear

[BNC]. Resultative phrase Benefactive alternation Material/Product alternation (transitive)

Total transformation alternation (transitive) [Non-alternating into only]

Carol cut the bread to pieces. Carol cut the envelope open. James cut a rose for Emily. James cut Emily a rose. James cut a mask out of the cardboard James cut the cardboard into a mask James cut the cardboard into a mask *James cut the cardboardfrom a box into a mask

According to Levin (1993), this inventory of alternations differentiates CUT verbs from other semantic classes of verbs. Nevertheless, she does not say how to distinguish between verbs such as slash, prune, chop, and hack, which belong to the same class and are characterized by many of the same alternations as cut. Given that diversity of meaning does not correlate with syntactic variation on a one-to-one basis, syntax by itself is insufficient as a basis for the determination of semantic classes because it cannot provide a satisfactory explanation of semantic constraints. As Pustejovsky (1995:10) points out: [...] participation in one grammatical alternation does not sufficiently determine the semantic class of the verb. In fact, even once a complete cataloguing of participation in alternation classes is achieved, we must ask ourselves just what we have accomplished

The FLM operates on the premise that lexical representations can be derived from a lexicon organized in areas of meaning. Furthermore, from the fìlli set of areas of meaning, which constitute the semantic

50

Ricardo Mairal Usón and Pamela Faber

architecture of the lexicon, complementation is derived. As anyone who has worked in lexical semantics knows, predicates in the same semantic space show striking similarities in their respective syntactic patterns. The distribution of the different syntactic configurations is not accidental, but is in accordance with their location within the hierarchical semantic organization of semantic domains in the lexicon (cf. Principle of Lexical Iconicity in Faber and Mairal 1999: 186-187). In this regard, CUT verbs are located within the domain of ACTION and show the following internal organization: (11)

CUT verbs

CUT to make an opening/wound/mark in sth/sb with a sharp-edged tool. hew to CUT a large piece out of a rock, stone or another hard material in a rough way, usu. with difficulty, prune to CUT off branches from trees/bushes/plants that they will grow better in the future, shave to CUT off hairfrom the face/body-part very close to the skin with a razor/shaver, gash to make a long, deep CUT in the skin. nick to make a small, shallow CUT in sth, usu. accidentally, slash to CUT sth with a sharp-edged tool with a quick, strong, swinging movement (a long, deep, cut), lop to CUT sth from what it was attached to with quick, strong stroke, slice to CUT sth into thin,flatpieces, chop to CUT sth into pieces by repeatedly hitting it with a sharp-edged tool (axe, knife, etc.). hack to CUT sth into uneven pieces in a rough, violent way. whittle to CUT sth (wood) to a smaller size by removing small, thin pieces. chisel to CUT sth (stone/wood/metal) into a special shape as if with a chisel, carve to CUT wood/stone into a special shape, saw to CUT sth (branch/tree) with a saw.

Functional Grammar and lexical templates

51

clip

to CUT a piece/s from sth with scissors or another sharp-edged tool to make them shorter/neater, snip to CUT sth as if with scissors with short, quick movements. shear to CUT sth (wool/hair) off (as i f ) with shears, mow to CUT sth (grass, plants with long stems, what grows in a field, etc.), using a machine/ scythe. We have observed that this type of hierarchical organization runs in tandem with the syntactic behavior of predicates. Not only do these predicates share the same area of meaning but also the same syntactic territory. This signifies that their degree of predictability can be enhanced if we postulate a meta-entry or lexical template. As a result, in the FLM lexicon, each domain and subdomain is characterized by a template, which represents the core meaning of all predicates within the same area of semantic space (Faber and Mairal 1999). On the basis of this lexical template it is possible to predict the inventory of syntactic configurations, relevant within a lexical class.

2. Lexical templates Lexical representations generally contain both structural and idiosyncratic elements. Structural parameters determine the semantic classes of verbs that are grammatically relevant, whereas idiosyncratic information distinguishes a verb from others in the same class (Rappaport and Levin 1998: 106). If the sole purpose of a system of lexical representation is to capture grammatically relevant aspects of meaning and not to provide a fine-grained lexical semantic representation, the semantic metalanguage need not be as detailed. However, this is not the idea behind a lexical template, which we conceive as a semantic enhancement of an initial syntactically-driven structure. In this respect, predicate frames, which only capture grammatically relevant features, need to be enhanced by introducing the semantic features in their meaning definitions. When this is done, lexical representations can be postulated for entire verb classes.

52

Ricardo Mairal Usón and Pamela Faber

In order to obtain that degree of generalization, a metalanguage is necessary. In this respect, we have used RRG logical structures, given the compatibility of both FG and RRG. Both theories are regarded as moderate functional approaches,4 and share certain similarities, since the FG layered structure of the clause is reminiscent of RRG. It is thus not surprising that we adopt the metalanguage proposed in this paradigm to enrich the predicate frame. RRG uses a decompositional representation for representing the semantic structure and argument structure of verbs and other predicates (their Logical Structure, LS). It is based on the Aktionsart distinctions proposed in Vendler (1967), and the decompositional system is a variant of the one proposed in Dowty (1979). Verb classes are divided into states, activities, achievements, and accomplishments together with their corresponding causatives.5 States and activities are primitives, whereas accomplishments and achievements consist of either a state or activity predicate plus a BECOME and an INGR operator:6 The inventory of RRG logical structures is shown in (12). There are a number of tests which determine which class the verb in a clause is assigned to (Van Valin and LaPolla 1997:90-102). The LS of the verb forms the core of the semantic representation of the clause, and it is this semantic representation that is related to the syntactic representation by means of the linking algorithm. The linking algorithm is bi-directional since it maps the semantic representation onto the syntactic representation, and vice versa. In the semantics-tosyntax linking, there are principles which project the appropriate syntactic representation from the semantic representation of the sentence.

Functional Grammar and lexical templates

53

(12) VERB CLASS

LOGICAL STRUCTURE

State Activity Achievement

predicate' (x) or (x,y)

Accomplishment

do' (x, [predicate' (x) or (x,y)] INGR predicate' (x) or (x,y), or INGR do' (x, [predicate' (x) or (x,y)] BECOME predicate' (x) or (x,y), or BECOME do' (x, [predicate' (x) or (x,y)]

Active accomplishment

do' (x, [predicate,' (x, (y))] & BECOME predicate2' (z,x) or (y)

Causative

α CAUSES β where α, β are LS of any type

(Van Valin and LaPolla 1997: 109)7 Nevertheless, these structures still need further semantic decomposition, and this entails the inclusion of an enhanced semantic component, which is presently not contemplated in RRG logical structures or the FG predicate frame. For example, ACTION is made up of the causative domains of more basic categories such as MOVEMENT and CHANGE. Since CUT verbs belong to the domain of ACTION, they lexicalize both manner (how the movement/activity is carried out) and result (the change brought about by the movement/activity), CUT verbs can thus be said to consist of two subevents: an activity and a resulting state: (13)


Resulting state

54

Ricardo Mairal Usón and Pamela Faber

The definitional structure of these verbs reveals that the basic differentiating parameter is the manner in which the activity is carried out. For example, hew is differentfromcut in that the cutting activity is done roughly and with difficulty. Possible reasons for this are the nature of the affected entity (its degree of "cuttability") or specific physical/mental properties of the effector (physical strength/emotional state). However, in the case of hew, it is a question of the nature of the affected entity, which is prototypically wood, stone or some other hard material. Such semantic parameters must necessarily be included in any representation of the activity expressed by this verb, which in turn should also be derivable from the generic lexical representation for CUT verbs. Each lexical class is characterized by a lexical template that encodes regularities and maximizes information in the lexicon with a minimum cost of representation. Lexical templates conflate both syntactic information (those aspects of the meaning of a word which are grammatically relevant) and semantic information (those aspects which act as distinctive parameters within a whole lexical class) into one unified representation. Van Valin and LaPolla (1997: 91)8 write: [...] arriving at a decompositional system is a compromise between the demands of semantics (make all the necessary distinctions relevant to meaning) and those of syntax (make syntactically relevant distinctions that permit the expression of significant generalizations).

The objective of an FG lexicon would be to formulate a more complete representation in which the set of semantic and syntactic factors that hold within this lexical class are present in one unified structure. Lexical templates thus would condense meaning definitions and their syntactic configurations into one representation for all the predicates within a given lexical class.

Functional Grammar and lexical templates

55

3. A lexical template for CUT verbs The lexical template for CUT verbs can be used to derive the meaning definitions for all of the other verbs in the same lexical class. Their distinguishing characteristics are encoded in metalanguage, and then added on to the basic template. The semantic-syntactic scenario for CUT verbs is encoded in the following way: (14)

[[do' (w, [use.sharp-edged.tool(a)in(ß)manner' (w, x)])] & [BECOME be-af (y, x)]] CAUSE [[do' (x, [make.cut.on' (x, y)])] CAUSE [BECOME pred* (y, (z))]], a = x.

The representation in (14) contains an effector (w) who carries out the cutting activity upon a patient (y) by means of a sharp-edged tool (x). As a result, the affected entity acquires a new state, that of being cut. This structure can be more specifically interpreted as follows: an effector (w) uses a sharp-edged tool (x) in such a way that the tool becomes in contact with a patient (y), causing an event such that χ makes a cut on y, and this, in turn, causes y to become cut. Furthermore, a new variable (z) is introduced to account for those cases where the final result is further specified (into pieces, in strips, open etc.). Suffice the following chart to illustrate the set of external variables involved and their corresponding referents: (15) Type of external variable w y X ζ

Referent Effector Patient Instrument Result State

When this structure is compared with any of the predicate frames postulated in (4)-(9), it becomes evident that the lexical template conflates both the argument structure properties of these predicates together with their idiosyncratic features. Thus, the result component

56

Ricardo Mairal Usón and Pamela Faber

codified in a predicate like slice, which was problematic in FG terms, is easily captured by means of the final causal chain [BECOME pred' (y, (z))]. The fact that the instrument of a predicate like chop can become the subject in an Instrument Subject Construction is easily explained by the implement use of χ introduced in the intermediate causal chain CAUSE [do' (x, [make.cut.on' (x, y)])]. Furthermore, the internal variables postulated in the first part of the lexical template find their corresponding lexical expression for each of these predicates. Hence chop instantiates the manner variable (iterative) and the instrument. (16)

slice [V] (xi: animate)/^ (x2: object)a0 df = cut [V] (xi)Ag (X2)GO(X3:pieces [N]: thin[A]: flat

[A]) Re suit

(17)

chop [V] (xi: animate)/^ (x2· object^ df = cut [V] (xi)Ag (x2)GO (X3: pieces) Re suit? [Iter ei: [hit [V] (Xl) A g (X2)GO (σι: sharp-edged t00l)instrument]Meani;

(18)

He chopped through the ropes with the cleaver. [BNC] The cleaver chopped through the tortured ropes. [BNC]

A lexical template has two types of variables: external variables and internal variables. External variables are those aspects of the meaning of a word that are realized syntactically. In contrast, internal variables are those semantic parameters which characterize an entire lexical class. Their internal structure is ontologically-driven. External variables are marked by the letters, w, x, y, and z, which appear in this template: w encodes the effector, χ encodes the instrument; y, the goal-affected entity; and z, a new state. The internal variables are INSTRUMENT, MANNER, AFFECTED OBJECT, and RESULT. They are marked by Greek letters (α, β), which act as place-holders for the ontological constants, which will define the lexical class in question. A further question that arises is whether a new intermediate variable should be introduced, given the fact that there are variables which can have an optional grammatical impact, e.g. the instrument variable as exemplified above. In contrast, other variables are clearly internal in

Functional Grammar and lexical templates

57

the sense that they never have a grammatical impact. This would be the case of MANNER. This is very much in line with Mackenzie's (1986: 18) "Intimacy Scale", where he distinguishes three variables in terms of their proximity to the predicate: The resultant picture is one of scale of closeness to the predicate, with the most intimate relationship being enjoyed by arguments (with their own internal hierarchy of first, second, and third argument), then y-satellites (always there, but not always mentioned), and ultimately z-satellites (optionally added as modifiers).

We believe that Mackenzie's proposal should be integrated into our approach and consequently three types of variables should be posited. In fact in Faber and Mairal (1999: 145-186) we distinguish the following three parameters in virtue of their potential syntactic encoding: (i)

Lexically-realized grammatical parameters, which are obligatory and always explicit in a verb's complementation structure. (ii) Lexically-realized optional parameters, which are semantically implicit, but not always activated in the actual linguistic expression. (iii) Lexically-realized contextual parameters, which are not syntactically projected, but serve as clues for contextual setting. This classification does not conflict with RRG theoretical premises. In RRG terms, our lexical template might appear to violate the Completeness Constraint (Van Valin and La Polla 1997: 325) because it introduces new elements which are not syntactically realized. However, this is not so because the Completeness Constraint applies only to external argument positions, i.e. pred' (x, y), and not to internal arguments. Since the instantiations that we propose are either internal argument variables or part of the semantic decomposition, the Completeness Constraint is irrelevant.9 With regard to the structure of the lexical template in (14), the first part of the template (before the first CAUSE) is the basic activity predicate of cutting, which underlies all of the verbs in the group. The

58

Ricardo Mairal Usón and Pamela Faber

second part of the template marks the causative accomplishment reading of these predicates. At the end of this second segment (after the second CAUSE), is the RESULT part of the template: [CAUSE [BECOME pred' (y, (z))]], which can take either one or two arguments, as the brackets show. In the minimal expression, the representation is a single argument predicate, like BECOME cut' (y). However, for sentences like "cut y into pieces" or "sliced y open", the representation would be BECOME be* (y, [pieces']) or BECOME be' (y, [open']). The result is thus treated as a secondary predicate. In the lexical template, the elements in boldface are primitives, and those in capitals (BECOME, CAUSE) are operators. Although RRG representations capture syntactically relevant aspects of meaning, they cannot account for semantic decomposition. For example, RRG does not say if CUT verbs are defined in terms of cut' (19) or in terms of BECOME plus the predicate (20): (19)

[do' (x, 0)] CAUSE [BECOME cut' (y)]

(20)

[do' (x, 0)] CAUSE [BECOME pruned' (y)]

Based on our research, the generic term (cut) should be regarded as a primitive in the lexical representations of its subordinate terms (e.g. hew, hack, chop, slice, slash, etc.). This avoids definitional circularity and establishes a solid foundation for the use of metalanguage. The greater the decomposition, the more delicate the analysis will be. The bracketing in the lexical template is not accidental. In fact, it has important consequences for the prediction of syntactic behavior. For example, one motivation for bracketing the second two sub-LSs together is that whenever there is an instrumental subject with these verbs, the reading must be telic (causative accomplishment). Moreover, the basic conative alternation in (1) (Margaret cut at the bread) can be derived from this template since y (bread), but not χ (instrument) is specified. Finally, each lexical template also has a corresponding set of lexical rules to encode syntactic productivity, and show how the template

Functional Grammar and lexical templates

59

can be modeled to account for the syntactic alternations shown by these verbs. With regard to the relations between the canonical template and the different configurations, the full lexical template is the one that underlies the class as a whole. Individual verbs, including cut, can be represented by all or part of it. From this perspective, all of the alternations involve different instantiations from the maximal lexical template underlying the class. We shall now take a closer look at the internal variables. Although all CUT verbs have the same event structure, their meaning definitions reveal semantic differences of MANNER, INSTRUMENT, AFFECTED OBJECT and RESULT. In this sense, we claim that these are part of the meaning definition of each predicate as encoded in, say, a dictionary entry. A further query that might arise is whether there are any preestablished diagnostic test frames to uncover these patterns (de Groot p.c.).

3.1. Internal variables 3.1.1. Manner as its name indicates, encodes the way in which the agent carries out the activity lexicalized by the verb. The following predicates can be differentiated from each other by their respective manner component (See (21)). This component can be subdivided into MANNER-effector (information dependent on specific physical/mental properties of the person carrying out the action, such as his/her physical strength/emotional state) and MANNER-movement (description of the movement involved in the cutting activity). MANNER,

60

Ricardo Mairal Usón and Pamela Faber

(21)

Manner in CUT verbs

Rred. hew

hack

to cut

to cut

Affected Resulting object state a large rock, stone or other hard material sth into uneven pieces

w a y [MANNER-

Instrument

MANNER

in a rough way, with difficulty [MANNEREFFECTOR]

in a rough, violent EFFECTOR]

lop

to cut

sth

with a quick, strong

slash

to cut

sth

with a quick, strong, swinging movement

s t r o k e [MOVEMENT]

with a sharpedged tool

[MOVEMENT]

chop

to cut

sth

snip

to cut

sth

into pieces

by repeatedly hitting with a sharpit [ITERATIVE edged tool MOVEMENT] (axe, knife, etc.) with short, quick (as if) with m o v e m e n t s [ITERAscissors TIVE MOVEMENT]

(22)

MANNER-EFFECTOR

hew

a.

b. hack

a. b.

[rough'

Λ

difficult'

[do'

(w,

0)]]

CAUSE

[BECOME cut'(y)] [do' (Tom, 0)] CAUSE [BECOME cut' (stone)] b1. Tom hewed the stone, [rough' A violent' [do' (w, 0)]] CAUSE [BECOME cut' (y)] & [BECOME exist' (pieces)] [do' (men, 0)] CAUSE [BECOME cut' (vegetation)] b'. Men hacked the vegetation.

Functional Grammar and lexical templates

(23)

61

MANNER-MOVEMENT

slash a.

[quick' Λ strong' Λ swing' [do' (w, [move' (w)])]] CAUSE [BECOME cut' (y)] & [BECOME have' (y, deep cut)] b. [do' (Ron, 0)] CAUSE [BECOME cut' (y)] b'. Ron slashed her throat, chop a. [repeated' [do' (w, [hit' (w, y)])]] CAUSE [BECOME cut' (y)] & [BECOME haveVexist' (y, pieces)] b. [do' (cook, 0)] CAUSE [BECOME cut' (y)] b'. The cook chopped the turnips. The meaning definitions of CUT verbs show that all of these predicates have the same underlying template. For example, in the case of MANNER, the distinction between syntactic and semantic elements can be represented by an ontological constant,10 which is replaceable by a property-entity indicating the manner in which the activity is performed. These constants have no grammatical realization, but appear in the following lexical template: (24)

[ [(do' (w, 0)]] CAUSE [BECOME cut' (y)]

The instantiation of this variable naturally depends on the characteristics of the predicate involved. Further decomposition is the result o: the application of canonical realization rules,1 whose internal struc ture would be derived from a conceptual ontology (cf. Section 3.2).

3.1.2. Instrument Another parameter that differentiates verbs in this lexical class is that of INSTRUMENT, as can be seen in their respective definitions:

62

Ricardo Mairal Usón and Pamela Faber

(25)

Instrument in CUT verbs

slash

to cut

Affected object Resulting state hair from the face/body part very close to the skin sth

chop

to cut

sth

into pieces

clip

to cut

a piece of sth

making it shorter/neater

snip

to cut

sth

Pred. shave to cut

chisel to cut

saw

to cut

shear

to cut

mow

to cut

into a special sth (stone/wood/ shape metal) sth (branch/tree) sth (wood/hair) sth (grass/plants with long stems in a field)

Manner

INSTRUMENT

with a razor/shaver

with a quick, strong, swinging movement by repeatedly hitting it

with short, quick movements

with a sharpedged tool

with a sharpedged tool (axe, knife, etc.) with a sharpedged tool (scissors, clippers) (as if) with scissors (as if) with a chisel with a saw (as if) with shears (as if) with a machine/ scythe

This semantic parameter is encoded in the lexical template by adding another subevent. The agent brings a sharp-edged instrument into contact with the affected entity, which results in the acquisition of a new state. Two examples of how this event structure can be represented can be seen in saw and shear :

Functional Grammar and lexical templates

(26)

63

INSTRUMENT

saw

a.

[do' (w [use' (w, x)])] CAUSE [do' (x, [BECOME be-af (y, x)])] CAUSE [BECOME cut' (y)] b. [do' (carpenter, [use' (carpenter, saw)])] CAUSE [do' (saw, [BECOME be-at' (wood, saw)])] CAUSE [BECOME cut' (wood)] b\ The carpenter sawed the wood with a power saw. shear a. [do' (w [use' (w, x)])] CAUSE [do' (x, [BECOME be-at' (y, x)])] CAUSE [BECOME cut' (y)] b. [do' (John, [use' (John, electric shears)])] CAUSE [do' (electric shears, [BECOME be-at' (sheep, electric shears)])] CAUSE [BECOME cut' (sheep)] b\ John sheared the sheep with electric shears. These new structures can be conflated by postulating an ontological constant of the type INSTRUMENT. The resulting lexical template has the following format: (27)

[ [(do' (w, 0)]] CAUSE [do' ] CAUSE [BECOME cut' (y)]

The selection restrictions for the instantiation of instrument variables in the lexical template should account for the potential range of entities that can appear in the definition of these verbs. As in the case of manner, the instantiations of instrument are also derived from an ontology. This lexical template inherits the properties of the one in (24) since there are various CUT predicates which lexicalize both MANNER and INSTRUMENT. Moreover, both parameters modify the activity predicate. MANNER acts as a modifier of the whole activity, whereas INSTRUMENT adds a new causal chain to the activity structure: (28)

[ [(do' (w, 0)]] [CAUSE [do' ]]

64

Ricardo Mairal Usón and Pamela Faber

3.1.3. Result The result of the cutting activity specifies the change of state experienced by the affected entity. Certain verbs not only lexicalize the final state of being cut but also the way in which the cutting produces a new form (cf. 4.9). Thus, that explains why we have introduced a new variable (z) which accounts for this syntactic constituent: (29)

[do' (w, 0)] CAUSE [BECOME cut' (y)] & BECOME have' (y,(z))l

(30)

[do· (w, 0)] CAUSE [BECOME cut' (y)] & fBECOME have' (y. new shape(sY)]

The resulting state appears in the second part of the lexical template as a second causal chain: (31)

[ [(do' (w,0)]] CAUSE [do' (x) ] CAUSE [BECOME cut' (y)] & [BECOME have* (z) ]

Thus far, an analysis of the major semantic parameters reveals that the structure proposed for the whole lexical class condenses the major semantic properties of this class of verbs.

3.1.4. Affected object CUT verbs not only differ in the way the cutting activity is performed, but also in the type of affected entity encoded. The specification of the referential scope of the second argument indicates a change in Aktionsart as well as an unspecified object alternation. In other words, these predicates can either lexicalize the result of the activity or just the activity itself since the affected object expresses an intrinsic facet of the meaning of the verb.

Functional Grammar and lexical templates

(32)

65

CUT verbs with an inherent second argument

Pred.

AFFECTED OBJECT

shave

to cut

hew

to cut

whittle to cut

chisel

to cut

carve

to cut

saw

to cut

prune

to cut

shear

to cut

mow

to cut

hair from the fece/body part very close to the skin a large rock, stone or other hard material sth (wood)

sth (stone/wood/ metal) sth (wood/stone) sth (wood/branch/t ree) branches from trees/bushes/pl ants sth (wood/hair) sth (grass/plants with long stems in a field)

Resulting state

Manner

Instrument with a razor/ shaver

in a rough way, with difficulty to a smaller size into a special shape

by removing small, thin pieces (as if) with a chisel

into a special shape (as if) with a saw to make them neater/grow better (as if) with shears (as if) with a machine/ scythe

The majority of CUT verbs lexicalize both manner and result as shown in the following examples:

66

Ricardo Mairal Usón and Pamela Faber

(33) 1ST ARG

CUTVERB

2ND ARG

John

cuts

She

slices

He

clips

wood a piece ofwood bread the bread hedges the hedges

i

[ACTIVITY]

"J [ACCOMPLISHMENT]

\ [ACTIVITY] { [ACCOMPLISHMENT] ·( [ACTIVITY]

i

[ACCOMPLISHMENT]

This phenomenon is easily justified in those verbs whose affected entity is intrinsic to the meaning of the verb since they generate an activity reading when the object is omitted: (34)

The slaves provided the unskilled labor, hewing and hauling where they had hoped to plunder.

(35)

Let's prune so that we lose neither the old trunk nor the new branches.

(36)

The officers waited while he shaved.

However, this shift in the internal aspectual and temporal properties of the predicates also holds for verbs whose intrinsic object is referential. These predicates can be regarded as both activities and accomplishments since they lexicalize manner and result respectively. These verbs thus would have one lexical template for the activity reading and another for the accomplishment reading. The activity template is an instantiation of the accomplishment template. (37)

[MANNER [do* (w, [cut' (w, y) ])]] [activity]

(38)

[MANNER [do' (w, 0)] ] CAUSE [BECOME cut' (y)] [accomplishment]

Functional Grammar and lexical templates

67

The distinction made between internal and external variables is very important Internal variables form part of the semantic representation of the verb, and furthermore, are intrinsic to the meaning of the predicate. They are not expressed because they can be recovered from context. The lexical template is shown in (39) as well as examples of its application to specific predicates: (39)

[[do' (w, [use.sharp-edged.tool(a)in(ß)manner' (w, x)])] & [BECOME be-at'(y, x)]] carve

[do' (w, 0)] CAUSE [BECOME cut' (wood/ stone)] & [BECOME have' (wood/stone, special shape)] He was learning how to carve.

shave

[do* (w, [use' (w, x)])] CAUSE [do' (z, [BECOME be-at· (hair, x)])] CAUSE [BECOME cut' (hairYl & [BECOME NOT be-at' (face/body, hair)] The officers waited while he shaved

mow

[do' (w, [use' (w, x)])] CAUSE [do' (z, [BECOME be-at' (grass/plants, x)])] CAUSE [BECOME cut' (grass/plants)! Each tenant had to mow three days in the autumn.

saw

[do' (w, [use' (w, x)])] CAUSE [do' (x, [BECOME be-at' (branch/tree, x)])] CAUSE [BECOME cut' (branch/tree)] He did nothing but saw all day.

68

Ricardo Mairal Usón and Pamela Faber

3.2. Linking curto an ontology Selection restrictions have become one of the most pervasive issues in lexical semantics. FG adopts a featured-based model which contains fillers like , , , , etc. Nevertheless, these features either cannot account for the semantic scope of the arguments, or are so general that they do not really impose any type of constraint, e.g. . Another alternative is that of using Pustejovsky's qualia structures, as suggested by Van Valin and La Polla (1997: 184-189). However, qualia (at least in their present form) have been criticized as constituting a simplification, or even impoverishment, of the semantic zone of the lexical entry. Nirenburg and Raskin (to appear) write that although the generative lexicon approach appears to be a step forward in semantics, it is still very much attached to syntax, and committed to minimizing its semantic exposure because semantics that cannot be directly tested on syntax is left untouched. In this respect, a very promising alternative is that which stems out of the work done within ontological semantics. Our analysis shows that for a more refined representation, further decomposition of the second argument is needed. However, as we have seen with CUT verbs, a more lexical-conceptual representation is necessary in order to distinguish between verbs within the same lexical class. A first step to specifying such a representation would be anchoring each lexical template to a well-designed conceptual ontology by means of which word senses would be related to each other on the basis of an underlying model of the world. As Nirenburg and Raskin (to appear) so accurately point out, only then can one justify the postulation of a certain number of theoretical concepts, a certain set of roles and features, and a prescribed range of values. As is well-known, any representation of the knowledge base of a natural language user should be in the form of a semantic network reflecting associative structure. (Semantic networks are conceptual representations of what we know.) Any type of understanding naturally implies the generation, extension, and linkage of a structure in which new knowledge is integrated into previous knowledge in such a way as

Functional Grammar and lexical templates

69

to make explicit the set of relations between concepts. This type of interrelated knowledge enables us to combine ideas, make inferences, extrapolate, and otherwise use this information in a productive and creative way. The fact that such structure is essentially hierarchical is to be expected, given that hierarchies are central to human cognition. JackendofF (1997:16) writes: From the point of view of psychology and neuroscience, of course, redundancy is expected. Moreover, so are multiple sources of infinite variability, each with hierarchical structure. One can understand an unlimited number of hierarchically organized visual scenes and conjure up an unlimited number of visual images; one can plan and carry out an action in an unlimited number of hierarchically organized ways; one can appreciate an unlimited number of hierarchically organized tunes.

When this type of organization is applied to the ordering of a semantic domain, the resulting structure is an ontology (in the artificial intelligence sense rather than the philosophical sense), defined by Gruber (1993), as an explicit specification of a conceptualization. One crucial property of a conceptual system is that no concept can be described without an account of its relationships to various other concepts (Lamb 1998: 147). For example, in the Mikrokosmos approach, an ontology is conceived as a language-neutral body of knowledge about the world. It constitutes a repository of primitive symbols used in meaning representation, which are interconnected by means of a rich system of semantic and discourse-pragmatic relations defined among the concepts (Mahesh and Nirenburg 1995:1). In relation to CUT verbs, each verb with its corresponding argument structure means the systematic activation of different areas of the ontology where the entitiesreferredto are located. For example, in the Mikrokosmos ontology (Mahesh and Nirenburg 1995; Viegas et al. 1999), the instantiation of the INSTRUMENT parameter would have the location shown in (40).

Ricardo Mairal Usón and Panela Faber

O •

ALL O

L

υ • •

EVENT un PROPERTY OBJECT • INTANGIBLE-OBJECT • MENTAL-OBJECT PHYSICAL-OBJECT • MATERIAL • PERCEPTUAL • PLACE • SEPARABLE-ENTITY I • ANIMATE • • INANIMATE • INGES TIB LE ARTIFACT • NATURAL-OBJECT • AIRPORT-ARTIFACT • ANIMAL-RELATED-ARTIFACT • BUILDING-ARTIFACT • DECORATIVE-ARTIFACT • DEVICE

Li Li

L!

• a

EVERYDAY-ARTIFACT a BED-LINEN CLEANING-ARTIFACT a CLOTHING-ARTIFACT COOKING-EQUIPMENT a [...] IMPLEMENT a BRUSH • MECHANICAL-IMPLEMENT • WOODWORKING-IMPLEMENT

•

•

•

[...]

•

CUTTING-IMPLEMENT AX a CHISEL RAZOR a SAW a SCISSORS a SWORD a CUTLERY • GRATER • KNIFE

• •

Functional Grammar and lexical templates

71

The other parameters for CUT verbs would have the following paths within the ontology: (41)

AFFECTED ENTITY OBJECT —• PHYSICAL OBJECT —• SEPARABLE ENTITY —• ANIMATE —• ANIMATE PART —» ANIMAL PART —• EXTERNAL ANIMAL PART —> HAIR OBJECT —• PHYSICAL OBJECT —• SEPARABLE ENTITY —• ANIMATE —• ANIMATE PART —• PLANT PART —• EXTERNAL PLANT PART —> TREE-BRANCH, PLANT-STEM, FLOWER, LEAF OBJECT —• PHYSICAL OBJECT —• MATERIAL —> PLANT-DERIVED SUBSTANCE —• WOOD OBJECT —• PHYSICAL OBJECT —• MATERIAL —>• SOLED —• EARTH MATERIAL —• ROCK OBJECT - » PHYSICAL OBJECT —• MATERIAL —• PLANT-DERIVED SUBSTANCE —• WOOD OBJECT —• PHYSICAL OBJECT —• MATERIAL —> SOLID —• EARTH MATERIAL —> MINERAL —>· LIMESTONE —• MARBLE OBJECT —»• PHYSICAL OBJECT —• SEPARABLE ENTITY —»• ANIMATE - > PLANT —• BUSH. GRASS. TREE. CROP PLANT

It goes without saying that this ontology is still incomplete. The integration of new concepts and the reorganization of old ones would be achieved by corpus analysis. In other words, the constraints and parameters for such a system of concepts would be derived from texts. For example, the following concordances from the British National Corpus reflect important aspects of the conceptual profile of the entity MARBLE, one of the affected entities typical of CUT verbs. The conceptual relations derived are both horizontal as well as vertical. Vertical relations are those such as ISA and HASA, which are the basic structuring relations of a conceptual hierarchy.

72

Ricardo Mairal Usón and Pamela Faber

(42)

MARBLE ISA Χ

4

ly been re-modelled inside in glass and MARBLE and with all the latest computer has progressedfrombronze to stone and MARBLE in the course of a career marked to carve a variety of materials such as MARBLE, plastic and wood. Duplito vaporise the dirtfromlimestone and MARBLE sculpture. John Larson, Head of I

29 43 ce

(43)

MARBLE HAS-TYPEX

10

his carvedfrom hard dolomitic Thasian MARBLE, but this could again point to a a new standing block of Calacatta Siena MARBLE, delicately decorated with incisi hose name derives from the pink Georgia MARBLE of the original building former a different structure from the Carrara MARBLE of Italy with which we are more f ingress of acid rain. The white Makrana MARBLE of which the Taj is built is of a fatal chill while soaking in an Italian MARBLE tub in 1875. Bird Vincent,

16 36 37 38 61

In (42), we can see that marble is a kind of material (concordance 44). Other possible candidates for coordinate concepts, or concepts at the same level as MARBLE, are also mentioned, such as LIMESTONE, GLASS, BRONZE, STONE, PLASTIC, and WOOD, which will have to be taken into account when elaborating an ontology. The concordances in (43) above all appear in the BNC as types of marble. These are examples of concepts which would appear at more specific levels of the ontology. They show that marble is categorized in terms of geographical precedence.

Functional Grammar and lexical templates

Ti

Apart from basic structuring relations, there are also horizontal relations which indicate links to concepts in other areas of the ontology. Examples of such relations are the following: (44)

MARBLE MATERIAL-OF X X MADE-OF MARBLE

9 11 12 13 17 18 19 20 21 22 26 28

atron, of all people, had turned into a MARBLE bust which yet incomprehensibly a pean sculpture and works of art. A pink MARBLE carved figure of Christ on the co ace and hands of the figure, set in its MARBLE chair, in that circle offantasti I include two George II Siena and white MARBLE chimneypieces, both from Northumb ρ and legs, on a bronze throne. A black MARBLE eagle crouches at his feet. The t ros and stole eighteen exhibits, mostly MARBLE figurines, dated to the Cycladic r. Alcove shelves at either side of the MARBLE fireplace are crammed with books k is currently underway on the coloured MARBLE floor and the crypt, with the aid e nineteenth-century Italian Travertine MARBLE fountain with sea horses, putti a hese under the chandeliers in the grand MARBLE foyer of one of the seven Staliner had any links with Lenin. The unique MARBLE hall, which gave its name to the the form of a rectangle, and place the MARBLE hearth on it.

In (44) we can see that the conceptual relations that can be derived from the concordances are MATERIAL-OF and MADE-OF. MARBLE is a material of which various artifacts are made. In the case of marble, these artifacts can be divided basically in three categories: BUILDINGPARTS, FURNITURE and STATUES. Accordingly each conceptual relation

74

Ricardo Mairal Usón and Pamela Faber

has a corresponding inverse relation. In the same way that marble is the MATERIAL-OF building-parts, furniture and statues, building-parts, furniture and statues are MADE-OF marble. (45)

MARBLE HAS PROPERTY X X PROPERTY-OF MARBLE

7 11 17 24 31 34 35 48 52 53 62

the mystical department, deploys white MARBLE balls at Mary Boone (until the pean sculpture and works of art. A pink MARBLE carved figure of Christ on the co ρ and legs, on a brortze throne. A black MARBLE eagle crouches at his feet. Thet th its remarkable colours - it was pure MARBLE glistening and sparkling with mie uter shell, the skin smooth and pale as MARBLE. Inside he was Pan. Lazy, lusty Ρ . The throne is decorated with coloured MARBLE, jewelled mosaics, ivory angels, s out into the open and sets as hard as MARBLE. Not only is "thank you " superflu t's eyelids, cool in the water, cool as MARBLE Sitting in the dentist's waiting ngs. My mother used to lament the heavy MARBLE table blown to powder by the V2. subtle interplay of stucco and ancient MARBLE The restoration has also reveale Wordsworth I can still feel the cold MARBLE under my bare feet. 1 was standin

The concordances in (45) are indicative of properties of marble such as HAS-COLOR (black, white, pink), HAS-TEMPERATURE (cool, cold), HASWEIGHT (heavy), REFLECTS-LIGHT (glisten, sparkle), and HAS-AGE (ancient). Although these are not necessarily all of the characteristics of marble, they are certainly the most salient and defining.

Functional Grammar and lexical templates

75

This process would be carried out for all of the entities that typically appear with the verb in question. The resulting system of concepts would be organized in a tangled subsumption hierarchy that are the basis for one type of lexical rules. In this way, it would be possible to elaborate a conceptualization of the act of cutting that would serve as a basis for the creation of lexical templates characteristic of the meaning area in question. 4. Syntactic configurations and lexical templates Following Levin (1993), we have elaborated a typology of syntactic configurations, based on information extracted from the BNC corpus. The inventory of syntactic alternations that Levin (1993) posits for CUT verbs has been contrasted with corpus data, and a template formulated for each one.

4.1. Transitive (46)

[do' (x, 0)] CAUSE [BECOME cut' (y)]

The lexical template in (46) encodes an activity that leads to a resulting state. This is in fact the minimal expression of the lexical template, something which coincides with what Talmy (1996: 235) calls "windowing of attention" or "gapping of attention". The reason why the causative/inchoative alternation is not possible lies in the syntactic selection principle, which stipulates that all arguments specified in the lexical template must be realized syntactically (Van Valin and LaPolla 1997: 173):12 Syntactic Selection Principle The number of syntactic slots for arguments and for argument-adjuncts within the core is equal to the number of distinct specified argument positions in the semantic representation of the core.

76

Ricardo Mairal Usón and Pamela Faber

The transitive representation is the canonical representation of CUT verbs, all of which show the syntactic pattern in (47): (47)

The man

cut hewed pruned clipped shaved slashed lopped sliced chopped hacked chiseled whittled nicked sawed snipped sheared mowed carved

the bread the stone the apple trees the hedges his beard her throat the tree branch the ham the wood the vegetation the rock the block of wood the furniture the planks the ribbon the sheep the grass the marble

4.2. Conative alternation (48)

[[do' (w, [use.sharp-edged.tool(a)in(ß)manner' (w, x)] & [BECOME be-af (y, x)]]

The template for the conative alternation is less productive than the previous one. The activity part is modified by another chain, which introduces the idea of attempted action. The representation shows an intransitive verb and a second argument, obliquely marked by at, which does not have the role of undergoer. This predicts that the conative alternation is impossible with an instrumental subject (ac-

Functional Grammar and lexical templates

77

tor), because, as Van Valin and LaPolla (1997: 121-122) argue, only implement-effectors which are part of a causal chain can function as actor, and there is no causal chain in the top part of the template. In the top part, χ is a simple instrument and not an implement-effector. Thus, we observe that there is an accommodation from the canonical lexical template in (14). The following CUT verbs illustrate this particular use: (49)

The man

cut hewed clipped slashed chopped hacked chiseled whittled nicked sawed snipped

at at at at at at at at at at at

the bread the stone the hedges her throat the wood the vegetation the rock the block of wood bits of grass the planks the chain link fence

4.3. Middle alternation (50)

be' ([[do'(0,0)] CAUSE [BECOME cut' (y)]], [predicate»])

In the middle alternation template, the predicate13 is an obligatory element codifying an attribute of the subject (e.g. cutting the meat is easy, nicking the furniture is easy, etc.). These adverbs, unlike manner adverbs, assign a property to the whole predication. As Van Valin and LaPolla (1997: 417) point out, such a representation captures the fact that middle constructions always occur with stative predicates, thus blocking expressions with dynamic verbs. Although Levin (1993) affirms that this alternation is typical of all CUT verbs, the BNC corpus only produced results for the following predicates:

78

Ricardo Mairal Usón and Pamela Faber

(51) The meat The furniture The bread The wood The wires The ribbon This grass

cuts nicks slices sawed clipped snipped mows

y

easily

As becomes clear, this alternation is also derived from the canonical lexical template in (14) with different patterns of specified and unspecified arguments. In this construction only >> is specified. 4.4. Unspecified object alternation (52)

[MANNER [do' (x, [cut' (χ, y) ])]] [activity]

The template of the unspecified object alternation focuses on the activity part of CUT verbs. Since manner is brought to the foreground as one of the principal differentiation parameters of these verbs, a variable is included which instantiates the different predicates of the lexical class. This change in the Aktionsart is possibly due to the nature of the object. If the object is non-referential, then it can freely be omitted, thus producing an activity interpretation. In such cases, the second argument is underlined. In terms of macrorole assignment, these verbs are intransitive since one of the prerequisites for assigning the role of undergoer is that the object must be referential. The verbs showing this alternation in the BNC corpus are the following:

Functional Grammar and lexical templates

(53)

79

f

The man continued to

hew prune shave chop chisel \ whittle saw snip shear mow carve ν

4.5. Instrument subject alternation (54)

[[do' (0, [use.sharp-€dged.tool(a)in(ß)manner' (0, χ)])] & [BECOME be-at' (y, x)]] CAUSE [[do' (x, [make.cuton' (x, y)])] CAUSE [BECOME pred' (y)]]

The representation of instrument-subject alternation reflects the implement use of these verbs. An activity is performed by an effector, who uses an instrument to cut the affected object. The fact that the instrument appears three times in the causal chain, explains why it can take the role of actor, and be the first argument of an activity predicate in the absence of an effector. Consequently, in terms of the actor-undergoer hierarchy, this argument is the default option for actor. The verbs susceptible to this alternation are the following:

80

Ricardo Mairal Usón and Pamela Faber

(55) INSTRUMENT

The knife The power axe The secateurs The hedge-clippers The new razor The razor The cleaver The knife The cleaver Their machetes The razor The power saw The secatuers The electric shears The lawn mower

CUT VERB

cut hewed pruned clipped shaved slashed lopped (off) sliced chopped hacked nicked sawed snipped sheared mowed

AFFECTED ENTITY

the bread the stone the apple trees the shrubs his beard her throat his leg. the chicken breast the meat the vegetation his chin the wood the flowers the sheep the grass

4.6. Characteristic property of instrument alternation (56)

[predicate' [do* (0, [use' (0, x)]) & [BECOME be-af (0, x)]]

CAUSE [[do' (x, [make.cut.on' (x,0)])] CAUSE [BECOME

pred' (0)]]

This template for the characteristic-property-of-instrument alternation involves the medial activity cut' section of the lexical template plus a state predicate, which is instantiated as a manner adverb and functions as a one-place predicate taking a subpart of the lexical representation as its argument (Van Valin and La Polla 1997: 162-171). This is not surprising since manner adverbs modify activities. In this representation, all arguments are unspecified except for the instrument, as seen in the following examples:

Functional Grammar and lexical templates

81

(57) INSTRUMENT

The knife The new razor The secateurs The hedge-clippers The power saw The scissors The electric shears The lawn mower

CUT VERB + MANNER ADVERB

MANNER ADVERB

cuts shaves prune clip saws don't snip don't shear mows

well well badly well well well well well

If we compare this structure with (14), there has evidently been a modeling process in the sense that all arguments but χ are left unspecified.

4.7. Unintentional interpretation The notion of agent plays an important role in RRG. Predicates which are always unmarked for agent (e.g. murder) are represented in the logical structure by an operator DO, which signals that the first argument always has this function. However, the first argument of CUT verbs is not necessarily an agent since this action can be carried out inadvertently or unintentionally, and for this reason, the operator DO is not included in the corresponding lexical template. Nevertheless, an effector may assume the role of agent under special circumstances (cf. Holisky 1987: 8; Van Valin and Wilkins 1996: 309-310), but in such a case, the interpretation is a question of pragmatic inference or implicature (cf. Holisky's Pragmatic Principle).

82

Ricardo Mairal Usón and Pamela Faber

(58)

Unintentional interpretation with reflexive object [do' (w, 0)] CAUSE [BECOME cut' (y)]

(59)

Carol

cut

herself

Steven

nicked

himself with a safety razor

Unintentional interpretation with body-part object [do* (w, 0)] CAUSE [BECOME cut' ([haye.as.part' (x, y)])] Carol cut her thumb Steven nicked his wrist with the plastic shards He slashed his wrist on the broken glass He gashed his knee last Saturday

4.8. Path phrase (60)

[[do' (w, [use.sharp-edged.tool(a)in(ß)manner' (w, x)])] & [BECOME be-at' (y, x)]] CAUSE [[do' (x, [make.cut.on' (x, y)])] CAUSE [BECOME pred' (y, (z))]]

The notion of path signifies movement from one point to another. Following Jolly (1993), we adopt the primitive be-via' to specify path. Unlike be-at' or NOT be-at' which encode the beginning and end of motion respectively, be-via' specifies both meanings. As Dowty (1979:142) writes: Travellingfromplace A to place Β is not merely changingfrombeing at A to not being at A, nor is it changing from not being at Β to being at B, but is apparently the conjunction of these two changes.

In CUT verbs, path specifies the conjunction of both the start and end points because the cutting activity implies movementfrompoint A to a

Functional Grammar and lexical templates

83

point B. This does not mean that the cutting is no longer at A, or that the cutting is only at B. This representation requires four arguments: the effector (x), the affected entity (y), and the initial and terminal points of the movement encoded in the activity (z, w). (61) 1ST ARGUMENT

CUT VERB 2ND

Carol

cut

the paper

They He

hewed pruned

their way the garden

He He He The knife The man

shaved slashed gashed sliced chopped

his head her throat her throat his face his way

He

hacked

his way

He He

chiseled sawed

his way the board

He

snipped

He He

mowed carved

ARGUMENT

a path

PATH

from one end to the other through the jungle from one end to the other from ear to ear from ear to ear from ear to ear across the cheek through the rhododendrons out of the wilderness into the cavern from one side to the other through her hearing aid through the facts through the traffic

These constructions can possibly be regarded as elaborations of the BECOME be-at' (y, x) part of the top lexical template. However, a better formulation for this part of the template would be BECOME beLOC' (y, x).

84

Ricardo Mairal Usón and Pamela Faber

4.9. Resultatile phrase (62)

[[do' (w, [use.sharp-edged.tool(a)in(ß)manner' (w, χ)])] & [BECOME be-at' (y, χ)]] CAUSE [[do' (χ, [make.cut.on' (χ, y)])] CAUSE [BECOME pred' (y, (ζ))]]

The template for the resultative phrase indicates the existence of two predicates (cut, and the result predicate). These verbs act as separate nuclei and function as a complex predicate, thus resulting in a nuclear juncture (cf. Van Valin and LaPolla 1997: 442-444). A representative sample of this construction appears in (63): (63) 1ST ARGU-

CUT VERB

2ND ARGU-

RESULT

MENT

MENT

Carol They He He He They

cut hewed pruned shaved slashed clipped

envelope her the hedges his head her throat the evergreens

This He He Thieves The magician He

sliced chopped hacked chiseled sawed snipped

the building meat the body the safe the woman the apricots

open /topieces into pieces bare clean open into formal shapes into segments into small cubes to pieces open in two into small pieces

The result part of the lexical template, CAUSE [BECOME pred' (y (z))], can have either one or two arguments depending on the verb and the construction, in the minimal expression, it would be a single argument predicate, like BECOME cut' (y), but for sentences like cut y to pieces, sliced y open etc., it could be represented as BECOME be' (y, [pieces'])

Functional Grammar and lexical templates

85

or BECOME be* (y, [open']), where the result is treated as a secondary predicate. Predicate frames, at least in their present form, are unable to capture this type of component.

4.10. Creation and transformation alternations (64)

[[do' (w, [use.sharp-edged.tool(a)in(ß)manner' (w, x)])] & [BECOME be-at' (y, x)]] CAUSE [[do' (x, [make.cut.on' (x, y)])] CAUSE [BECOME NOT exist' (y) Λ BECOME exist'

m

The template encoding creation and transformation alternations specifies simultaneous changes of state symbolized by A. Because of the cutting activity, the affected entity loses its original form, and becomes a new entity. Again, change (and thus movement) occurs as well as a result. The following CUT verbs are sensitive to this alternation: (65) 1 s t ARG.

CUT VERB

EFFECTED OBJ. / AFFECTED OBJ.

AFFECTED OBJ. / EFFECTED OBJ.

James

cut

He

hewed

He

whittled

He

chiseled

He

carved

the cardboard into a mask the block of stone into a statue the faggot into a cudgel the marble block into a statue the marble block into a likeness of his father

He

sawed

a mask out of the cardboard a statue out of the block of stone a cudgel from the faggot a statue from the marble block a likeness of his father out of the marble block writing and drawing sticks from the graphite

the graphite into drawing and writing sticks

86

Ricardo Mairal Usón and Pamela Faber

5. From lexical templates to syntactic structure It is now necessary to ask ourselves if there is a systematic relationship between the canonical structure of CUT verbs and its different instantiations. If we posit that the full lexical template underlies the class as a whole, then individual verbs, including cut, have some or all of it. In much the same way as predicates share a semantic area, predicates highlight one part, or parts of the lexical template. By semantically relating predicates to each other, we postulate a semantic content which groups a number of lexemes together. These predicates focus on a particular area within that vast semantic content. The choice of one predicate over the rest entails a further degree of lexical delicacy. The specific syntactic configurations of each predicate either involve the licensing of the whole structure or part of the structure as encoded in the lexical template, or else, these involve further degrees of syntactic delicacy. From this perspective, all of the alternations involve a licensing process from the maximal lexical template underlying the class. It would consequently be an empirical question as to which assumption, basic activity or basic causative accomplishment, best accounts for the alternations that cut enters into. The fact that all of the alternations, both in terms of the different verbs in the class, and the alternations that a single verb enters into, can be derived from the canonical, maximal lexical template has been shown for most of the alternations presented.. This does not clash with the methodological principle of FG "avoid deletion of specified arguments" (cf. Dik 1997: 19-21) since this principle seems to apply in the case of syntactic derivation. However, in the lexicon predicates form an intricate web in which one choice determines a further one, and so on, until we ultimately arrive at the most delicate one. The format of a lexical template is a maximal expression but a different issue is the way this template interacts with specific syntactic configurations. In this regard, we claim that the accommodation of a lexical template to a specific construction can take a maximized or reduced format. That is why we prefer to use the term modeling process. This avoids the issue of whether lexical templates involve

Functional Grammar and lexical templates

87

reductions or expansions, a theoretical issue which is highly controversial. Broadly speaking, the modeling process can operate at various levels, each of which would count as a distinct lexical rule: - Lexical template accommodation (e.g. leaving out " c a u s e [...]" for the activities). - Internal variable instantiation (e.g. beta would be realized in all of the manner alternations, but not necessarily in the others). - External variable instantiation (e.g. leaving χ and y unspecified in the activity use of cut). - Lexical templates (e.g. inserting the Lexical Template into the attributive template to derive both the middle and the characteristic property of instrument alternation). In line with this assertion, the mapping between the lexical template of a lexical class and the full set of different syntactic structures is governed by the following process: Lexical Template Modeling Process Lexical templates can be modeled by accommodating external variables, instantiating internal variables and operators (e.g. CAUSE), or else, by introducing elements resulting from the fusion with other templates iff there is a compatibility between the features in the lexical template and the syntactic construction under scrutiny.

For example, a middle or a resultative construction could not be fused with an atelic lexical template since there would be a clash between the information of the lexical template and that required by the construction. Resultative constructions entail the final stage of an activity and thus their incompatibility with an atelic predicate. Typological evidence shows that languages admit both processes, either reductions or expansions. We believe that in the case of the lexicon, it is not so much a question of reduction vs. expansion process, as is the case in certain approaches like Rappaport and Levin (1998), who adopt an expansion mechanism (cf. the Template Augmentation Principle). It is rather a question of a modeling or accommodation

88

Ricardo Mairal Usón and Pamela Faber

process with a set of accompanying lexical rules, which accounts for the various ways a predicate licenses a part or the whole of the lexical template. As a result, we are in favor of lexical rules, but not as an unconstrained part of the system. This is an issue which is extensively developed in Mairal Usón (2002). Finally, the format of each lexical entry would consist of its lexical template, which would include its unique selectional properties along with the alternations it undergoes. Such alternations would be constrained to a certain degree by the lexical template itself. Initially it would be necessary to specify which rules apply, but with further study of lexical organization and the information encoded within, principles might well emerge which would allow one to predict the rules that each lexical template would be sensitive to.

6. Conclusions We claim that FG argument structure could be further enhanced if meaning definitions, as postulated in terms of meaning postulates and argument structure properties, could interact. In this regard, we suggest the lexical template as an alternative to the predicate frame. Lexical templates conflate predicate frames into one unified representation in that meaning definitions and predicate fìames are now presented in one compact representation. We also advocate the use of metalanguage in order to capture significant linguistic generalizations, given that the scope of predicate frames is insufficient to represent such information. The format of lexical templates consists of internal and external variables. Internal variables are those features which are implicitly encoded in the FG meaning definition, whereas external variables capture those aspects of the meaning of a predicate which have a syntactic impact. In this way, lexical templates unify meaning postulates with their corresponding predicate ñame. An FG lexicon should be organized into articulated semantic classes, each of which is characterized by a lexical template. The resulting FG lexicon consists of a set of hierarchically related lexical

Functional Grammar and lexical templates

89

templates, which are compatible with Hengeveld's (1992) general predicate fiâmes. Furthermore, we demonstrate the validity of this proposal by applying it to the lexical class of CXJT verbs. The decomposition of cut permits the elaboration of a meta-entry, which encodes shared information in the form of constants. However, as has been shown, cut is sensitive to at least ten different complementation patterns, something that is also true, though to a lesser extent, for its hyponyms. Thus, the problem is how to account for syntactic variations within the same lexical class without postulating a lexical entry for each complementation structure. In our opinion, syntactic variation can be explained by modeling processes which operate upon the canonical lexical template through the application of lexical rules that which relate lexical entries to their corresponding complement configurations. A lexical template is posited for the lexical class of CUT verbs, and the alternations typical of these predicates are analyzed with respect to the maximal canonical template. Different types of lexical templates (e.g. result, middle, transitive, conative, etc.) are thus revealed as concealed constructions which are in direct correspondence to what we have called the semantic architecture of the lexicon. A further point of research is the elaboration of the inventory of lexical rules which operate in an FG lexicon as well as the actual linking mechanisms involved (Mairal Usón 2002). In conclusion, our proposal enriches the predicate frame by integrating the information encoded in meaning definitions into the same format of the predicate frame itself. This entails reducing the two levels of representation in FG into a single level of representation.

Notes 1. This research is part of research project PB98-1342, funded by the Spanish Ministry of Education. We should like to thank Robert Van Valin, Christopher Butler, Casper de Groot and Kees Hengeveld for their comments on an earlier version of this paper. Needless to say, the responsibility for all errors, inaccuracies and misconceptions remains entirely ours.

90

Ricardo Mairal Usón and Pamela Faber

2. The term lexical template is similar in many respects to that of predicate schema (Faber and Mairal 1998). However, for die sake of clarity, we will use lexical template to describe this construct. 3. The impact of lexical rules upon the expression rule component is beyond the scope of this article (cf. Mairal Usón forthcoming). 4. For a comparison of both theories and the way they can mutually benefit we refer the reader to Mairal and Van Valin (2001). 5. FG recognizes the impossibility of organizing verb classes in terms of the notion of Aktionsart since the SoA type is compositionally derived from the properties of the arguments and satellites. Dik (1997: 106) states: "Setting up a typology of SoAs then raises the question of what properties of predicates and terms enter into the definition SoA types." Nevertheless, in our opinion, the full range of syntactic and semantic parameters that converge within a given lexical class can best be accounted for in terms of SoA types plus semantic information typical of the lexical domain that the predicates belong to. Van Valin and LaPolla (1997: chapter 3) argue that the nature of the state of affairs determines the participants therein, and likewise the properties of the predicate, especially in terms of Aktionsart, determine its argument structure. In this regard, FG is not so distant from this theoretical issue in the sense that Dik (1997) also acknowledges the inclusion of state of affairs type to account for certain syntactic configurations. 6. For a more complete version of the system of lexical representation in RRG, we refer the reader to Van Valin and LaPolla (1997: chapter 3,4 and 7). 7. Rappaport and Levin (1998) also propose an inventory of templates, which are very similar in form to those of RRG. Both consist of a set of structural elements or primitives and a set of constants. 8. Paradoxically enough, RRG primarily concentrates on syntax and only proposes a very tentative semantic analysis, i.e. Van Valin and Wilkins's (1993) analysis of the verb remember and Van Valin and LaPolla's (1997: 116-118) analysis of speech act verbs. 9. Van Valin and LaPolla (1997: 325) describe The Completeness Constraint as follows: All of the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in the sentence, and all of the referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic representation of a sentence. 10. The set of constants proposed in this paper resemble to some extent those proposed in Jackendoff (1990) and also Rappaport and Levin (1998). 11. This term is taken from Rappaport and Levin (1998: 108), although our proposal differs from theirs in various respects. Firstly, our rules are semantically grounded in that they serve to further decompose a variable into the possible

Functional Grammar and lexical templates 91 realizations which hold within a given particular class, in this case, CUT verbs. However, Rappaport and Levin's rules establish the associations between a variable and its corresponding event structure template, but no mention is made as to how this variable is semantically instantiated. 12. Both the Completeness Constraint and the Syntactic Selection Principle underlie an isomorphism between the structure of the semantic and the syntactic representation. Similarly, Rappaport and Levin (1998: 112-113) formulate the "Subevent Identification Condition" and the "Argument Realization Condition", which point to the same direction. 13. The instantiation of this predicate would be an adverb. In this regard, Van Valin and LaPolla (1997: 163) claim that manner adverbs can be interpreted as oneplace predicates.

References Butler, Christopher S. this volume Catching a glimpse of linguistic reality: Modelling the complexities of verb meanings in the Functional Lexematic Model. Dik, Simon C. 1978a Functional Grammar. Dordrecht: Foris. 1978b Stepwise Lexical Decomposition. Lisse: de Ridder. 1997 The Theory of Functional Grammar. Part I: The Structure of the Clause. Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter. Dowty, David 1979 Word Meaning and Montague Grammar. Dordrecht: Reidel. Faber, Pamela and Ricardo Mairal 1998 Towards a typology of predicate schemata in a Functional Lexematic Model. In: Gerd Wotjak (ed.), Towards a Functional Lexicology, 11-37. Frankfurt: Peter Lang. 1999 Constructing a Lexicon of English verbs. Berlin/New York: Mouton de Gruyter. Goldberg, Adele A Construction Grammar Approach to Argument Structure. 1995 Chicago: University of Chicago Press. Gruber, Thomas R. 1993 A translation approach to portable ontology specifications. Knowledge Acquisitions·. 199-220.

92

Ricardo Mairal Usón and Pamela Faber

Hengeveld, Kees 1992 Non-Verbal Predication: Theory, Typology, Diachrony. Berlin/New York: Mouton de Gruyter. Holisky, Dee Α. 1987 The case of the intransitive subject in Tsova-Tush (Batsbi). Lingua 71:103-132. Jackendoff, Ray 1990 Semantic Structures. Cambridge, M.A.: MIT Press. 1997 The Architecture of the Language Faculty. Cambridge, M.A.: ΜΓΓ Press. Jolly, Julia 1993 Preposition Assignment in English. In: Robert Van Valin (ed.), Advances in Role and Reference Grammar, 275-310. Amsterdam/Philadelphia: John Benjamins. LakoflÇ George 1987 Women, Fire and Dangerous Things. Chicago: Chicago University Press. Lamb, Sidney M. 1998 Pathways of the Brain: The Neurocognitive Basis of Language. Amsterdam/Philadelphia: John Benjamins. Langacker, Ronald 1987 Foundations of Cognitive Grammar. Stanford: Sanford University Press. 1991 Concept, Image and Symbol: The Cognitive Basis of Grammar. Berlin/New York: Mouton de Gruyter. Levin, Beth 1993 English Verb Classes and Alternations: A Preliminary Investigation. Chicago: University of Chicago Press. Mackenzie, Lachlan 1986 Aspects of English nominalization in English and Dutch. Working Papers in Functional Grammar 15. Mahesh, Kavi and Sergei Nirenburg 1995 A situated ontology for practical NLP. A Paper Presented at the IJCAI '95 Workshop on Basic Ontological Issues in Knowledge Sharing, August 19-21, Montreal, Canada. Mairal Usón, Ricardo 2002 Lexical representations and syntactic structures: Towards the design of a Lexical Grammar Model. Unpublished research monograph. forthcoming Lexical rules and expression rules: A preliminary discussion.

Functional Grammar and lexical templates 93 Mairal Usón, Ricardo and Robert J. Van Valin 2001 What Role and Reference Grammar can do for Functional Grammar. In: María Jesús Pérez Quintero (ed.), Challenges and Developments in Functional Grammar, 137-166. (Revista Canaria de Estudios Ingleses 42.) La Laguna: Servicio de publicaciones. Martin Mingorance, Leocadio 1995 Lexical logic and structural semantics: Methodological underpinnings in the structuring of a lexical database for natural language processing. In: Ulrich Hoinkes (ed.), Panorama der Lexikalischen Semantik, 461-474. Tübingen: Gunter Narr. Meijs, Willem 1990 Spreading the word: knowledge-activation in a functional perspective. In: John Connolly and Simon C. Dik (eds.), Functional Grammar and the Computer, 201-217. Dordrecht: Foris. Meijs, Willen and Piek Vossen 1992 In so many words: Knowledge as a lexical phenomenon. In: James Pustejovsky and Sabine Beigier (eds.), Lexical Semantics and Knowledge Representation, 137-153. Berlin: Springer. Nirenburg, Sergei and Lori Levin 1992 Syntax-driven and ontology-driven lexical semantics. In: James Pustejovsky and Sabine Beigier (eds.), Lexical Semantics and Knowledge Representation, 5-20. Berlin: Springer. Nirenburg, Setgei and Victor Raskin to appear Ten choices for lexical semantics. Computational Linguistics. Olbeitz, Hella, Kees Hengeveld and Jesús Sánchez (eds. ) 1998 The Structure of the Lexicon in Functional Grammar. Amsterdam/Philadelphia: John Benjamins. Pustejovsky, James 1995 The Generative Lexicon. Cambridge, M.A.: MIT Press. Rappaport, Malka and Beth Levin 1998 Building verb meanings. In: Miriam Butt and William Geuder (eds.), The Projection of Arguments: Lexical and Compositional Factors, 97-134. Stanford: CSLI. Schack Rasmussen, Lone 1994 Semantic functions in perspective - reconsidering meaning definitions. In: Elisabeth Engberg-Pedersen, Lisbeth Falster Jakobsen and Lone Schack Rasmussen (eds.), Function and Expression in Functional Grammar, 41-63. Berlin/New York: Mouton de Gruyter.

94

Ricardo Mairal Usón and Pamela Faber

Talmy, Leonard 1996 The windowing of attention in language. In: Masayoshi Shibatani and Sandra A. Thompson (eds.), Grammatical Constructions. Their Form and Meaning, 235-287. Oxford: Clarendon Press. Van Valin, Robert D. (ed.) 1993a Advances in Role and Reference Grammar. Amsterdam/ Philadelphia: John Benjamins. 1993b Synopsis of Role and Reference Grammar. In: Robert D. Van Valin (ed.), Advances in Role and Reference Grammar, 1-164. Amsterdam/Philadelphia: John Benjamins. Van Valin, Robert D. and Randy LaPolla 1997 Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Van Valin, Robert D. and D. Wilkins 1993 Predicting syntactic structure from semantic representations: Remember in English and its equivalents in Mpamtwe Arrernte. In: Robert D. Van Valin (ed.), Advances in Role and Reference Grammar, 499-534. Amsterdam/Philadelphia: John Benjamins. 1996 The case for 'Effector': Case roles, agents, and agencyrevisited.In: Masayoshi Shibatani and Sandra A. Thompson (eds.), Grammatical Constructions. Their Form and Meaning, 289-322. Oxford: Clarendon Press. Vendler, Zeno 1967 Linguistics in Philosophy. Cornell: Cornell University Press. Viegas, Evelyne, Kavi Mahesh, Sergei Nirenburg and S. Beale 1999 Semantics in action. In: Patrick Saint-Dizier (ed.), Predicative Form in Natural Language and Lexical Knowledge Bases, 171-204. Dordrecht: Kluwer. Vossen, Piek 1994 The end of the chain: Where does decomposition of lexical knowledge lead us eventually?. In: Elisabeth Engberg-Pedersen, Lisbeth Falster Jakobsen and Lone Schack Rasmussen (eds.), Function and Expression in Functional Grammar, 11-39. Berlin/New York: Mouton de Gruyter. 1995 Grammatical and Conceptual Individuation of the Lexicon. Amsterdam: IFOTT.

Do we need predicate frames? Daniel Garcia Velasco Kees Hengeveld

1. Introduction1 In this article we discuss the role of predicate frames within the theory of Functional Grammar (FG) and present the alternative possibility of replacing them by general predication frames into which lexemes, listed in the lexicon without frames, are inserted through the application of linking rules. In doing so we combine two ideas we have presented in other places. The idea of introducing predication frames2 into FG was originally proposed in Hengeveld (1992a: 80, 92-94). There it was meant to offer an alternative to Dik's (1980) Term-Predicate Formation Rule. This rule covers the predicative use of adpositional phrases in constructions like Sheila is in the garden and involves, in Dik's approach, the addition of a semantic function to a term outside of a predication. As noted by Mackenzie and Hannay (1982), such a proposal cannot be maintained in a model in which semantic functions always obtain within a predication. Hengeveld (1992a) proposes to solve this problem by creating predication frames which define the relation between the argument term and the term used predicatively. In this paper we generalize the idea of using predication frames to all kinds of predicates. Following ideas presented in Nuyts (1992), Garcia Velasco (1998) discusses the possibility of enriching the FG model with a prelinguistic conceptual level. One of the consequences of that move is that a great deal of the information present in predicate frames in current FG may be derived from Lexical-Conceptual Structures, as repre-

96

Daniel Garcia Velasco and Kees Hengeveld

sented by abstract meaning definitions, if a proper linking system is developed. In this article we bring these proposals together and suggest that the notion of predicate frame has to be replaced by a combination of predication frames on the one hand, lexemes provided with abstract meaning definitions on the other, and a linking mechanism joining these two together. We will argue that such a step is not only desirable, but also helps the theory to attain a higher degree of typological, psychological and pragmatic adequacy. The feasibility of our proposal from the perspective of these three standards of adequacy is discussed in section 2 of this article. Section 3 then introduces the notion of predication frame and shows the need for a linking mechanism. This linking mechanism itself is the subject of section 4, which goes deeper into the question of how lexemes become associated with frames on the basis of their abstract meaning definitions. In the concluding section 5 we look at the consequences of our proposal for the FG model.

2. Predicate frames and the FG standards of adequacy 2.1. Predicate frames and psychological adequacy

The standard of psychological adequacy requires that FG should "relate as closely as possible to psychological models of linguistic competence and linguistic behaviour" (Dik 1997: 13). Predicate frames play a central role in the organization of standard FG in that they constitute the input for clausal generation and contain a great deal of the syntactic and semantic information relevant to the interpretive process. It is to be expected, therefore, that psycholinguistic evidence confirm their privileged status both in sentence production and comprehension. An interesting case to evaluate the psychological adequacy of predicate frames concerns the behavior of those verbs which allow alternative argument structures. Among others, these include verbs

Do we need predicate frames? 97

participating in the so-called locative and causative alternations as illustrated in (1) and (2) respectively: (1)

a. Bill loaded bricks onto the truck. b. Bill loaded the truck with bricks.

(2)

a. Bill opened the door. b. The door opened.

Standard FG treats these alternations by means of Predicate Formation Rules; for example, the locative alternation in (1) is taken care of by means of rule (3) (Dik 1980: 37): (3)

COMPLETIVE VERB FORMATION IN ENGLISH INPUT: predv (xi)Ag (x2>go (x3>loc OUTPUT: predv (xi)Ag (x3)go feW CONDITION: the input predicate-frame must indicate an Action by which something is applied to some surface in such a way that the surface gets covered with the something as a result of the Action.

and the causative alternation in (2) receives the following treatment (Siewierska 1991: 28): (4)

INCHOATIVE FORMATION INPUT: predv (xi)Ag (x2)go OUTPUT: predv feW

From a psycholinguistic perspective, the FG approach suggests that the interpretation and production of the output frames from both (3) and (4) will have to refer to the relevant predicate formation rule to establish the relation with the input frame and arrive at the appropriate semantic interpretation of the derived predicate. As Kahrel (1989: 136) points out "the implicit assumption in FG is that during the processing of language, predicate formation rules apply to derive

98

Daniel García Velasco and Kees Hengeveld

productively derivable predicates". Consequently, derived predicate frames should take longer to process than basic ones since these are already available in the lexicon. Carlson and Tanenhaus (1988) have tested this hypothesis by contrasting the interpretation of these predicates with ambiguous verbs such as set in the following examples: (5)

a. Bill set the alarm clockfor six in the morning. b. Bill set the alarm clock onto the shelf.

The verb set has two possible senses, "adjust" as in (5a) and "place" as in (5b). In the interpretation of these examples, disambiguation does not take place until the presentation of the final prepositional phrase. There seems to be agreement in the psycholinguistic literature that multiple senses of ambiguous verbs are normally accessed in parallel until one of them is selected on the basis of the context and general knowledge (Boland 1993; Boland et al. 1995; Carlson and Tanenhaus 1988; Shapiro et al. 1989). In the case of the set examples above, lexical access will make available all senses of the verb, but only the contextually more appropriate will remain active. Upon selecting the wrong sense, reinterpretation forces the natural language user to retrieve the right sense by unfolding the lexical entry again. This process should obviously cost processing effort. Unlike set, verbs participating in alternations such as the ones mentioned above do not present different senses, but rather, different argument structures. When encountering one of these verbs, the speaker/hearer will have to "decide" which predicate frame the verb is attached to. Consider the following examples (Carlson and Tanenhaus 1988: 274): (6)

a. Bill loaded the truck onto the ship. b. Bill loaded the truck with bricks.

These sentences show a case of "thematic ambiguity", in the sense that the semantic function which the sequence the truck should receive is not evident until the next constituent arrives. According to Carlson and Tanenhaus, the truck in (6b) is a Location, whereas in

Do we need predicate

frames?

99

(6a) it is a Theme. Dik's rule above assigns Goal function to the object of load in both alternatives, but this does not undermine the reasoning, as the role of the truck in (6b) derives from a Locative reading of the same argument within the input predicate frame, as illustrated in (la). The prediction stemming from the FG organization of the lexicon is, firstly, that sentence (6b) should take longer to process than (6a), since it requires to unfold the relevant predicate formation rule to arrive at the right interpretation for the predicate. Secondly, thematic ambiguities should take as much time or even longer to process than sense ambiguities since they need to interact with the predicate formation component, whereas sense ambiguities require to "look up" the lexicon again for another sense of the relevant lexical item.3 In Carlson and Tanenhaus' (1988) experiment both sentences with sense ambiguities and sentences with thematic ambiguities were paired with unambiguous controls. The subjects' task was to decide as quickly as possible whether the sentence "made sense". The authors proved that sense ambiguities take longer to process than unambiguous control sentences and are less often judged to make sense. In contrast, thematic ambiguities did not show significant differences with respect to the processing of their controls. This goes against FG expectations. Under the predicate formation rule approach advocated by FG, these facts are difficult to explain. However, if we assume that the different thematic variants of the load class of verbs are obtained from the same common core meaning, associated with two differentframes,there is no need for the processor to reopen the lexical entry since the meaning of the verb remains constant. In such a case, the results of the experiment make perfect sense. Similar arguments can be put forward on the basis of the causative alternation. Consider the following example (attributed in Carlson and Tanenhaus 1988 to Stowe 1987): (7)

Even before the police stopped the driver was getting nervous.

100

Daniel García Velasco andKees Hengeveld

In the absence of punctuation, the causative/ergative verb stop is interpreted as a transitive predicate with the term phrase the driver as its object. This results in a garden-path effect in the example. The authors observe that if the subject of stop is replaced by an inanimate entity, readers are more likely to analyze the verb as intransitive: (8)

Even before the truck stopped the driver was getting nervous.

It seems then that the selection restriction affects the interpretation of the transitivity of the predicate. In the experiment described, when the first argument of the verb stop was animate, subjects' reading times were longer than the relevant controls, but when it was inanimate no corresponding effect was found. This is unexpected if the intransitive variant were to be derived from the transitive one as FG claims. Again, it seems that this evidence might be better accounted for if we assume that alternating verbs present a common core meaning to which a syntactic frame is assigned on-line on the basis of the contextual information available during sentence comprehension. This move would also reduce the power of predicate formation rules, which, in our view, should be limited as much as possible.4

2.2. Predicate frames and pragmatic adequacy The concept of predicate frame has been examined by Butler (1998, 2001) in the light of the standard of pragmatic adequacy. He argues that, given FG's commitment to this standard, "the model of predicate-argument structure encapsulated in the predicate frame should be able to account for the ways in which predicates and their arguments are used in actual communicative discourse" (Butler 2001: 56). Butler (2001) examines the use of the verbal predicate give in the Cobuild Bank of English. Dik's (1997) representation of this predicate is as follows:

Do we need predicate frames? 101

(9)

(fi: give) [V]

(XI:

(xi))Ag

(X2)GO

fa'·

(X3))Rec

From his carefixl study of the nature of the participants in the givepredications and the type of transfers involved, Butler concludes that the structure in (9) is too simple to account for the actual use of the verb. One finding which illustrates this general observation is that in almost 20% of the examples the Agent is not animate, as the frame requires. Consider the following example (Butler 2001: 58): (10)

His years of composing Greek verse gave him confidence.

The high percentage of non-animate agents mentioned by Butler suggests that this example is not just a violation of the selection restrictions on the first argument, but has more serious implications. FG defines the semantic function Agent as the entity controlling an Action, that is, an entity having the power to determine whether an SoA will obtain (Dik 1997: 112). But under Dik's own tests for [+control] SoAs, the first argument in Butler's example emerges as a non-Agent: (11)

Predication complement of order, persuade, request *I ordered/persuaded/requested his years of composing Greek verse to give him confidence. I ordered/persuaded/requested Mary to give him confidence.

(12)

Predication combined with Beneficiary satellites *His years of composing Greek verse gave him confidence for my sake. Mary gave him confidence for my sake.

This entails that the 20% of inanimate first arguments in Butler's study should be assigned a different semantic function. Given FG's inventory of semantic functions, the most likely candidate seems to be that of Force.

102

Daniel García Velasco and Kees Hengeveld

This alternation between Forces and Agents in first argument position is not restricted to the verbal predicate give. In fact, it has long been noted that instrumental satellites may take subject position, a process which in FG entails the conversion of Instrument to Force (e.g. Mairal Usón and Faber this volume): (13)

John cut the meat with a knife. The knife cut the meat.

Given the productivity of this alternation, one could try and formulate a Predicate Formation Rule with the following general format: (14)

INSTRUMENT-FORCE FORMATION INPUT: (fi: predv (fi): {(f2: (x 3 W (fc))} (fi)) (xi)ab (X2)go OUTPUT: (fi: predv (fi)) (χ3)ρογ Λ (R2)) e.g. burn in Peter burned his fingers 1.2.4. trans, +con, +dyn (T,: (f,: • (fi)) (T,)) (Ri: (τ , ) ^ (Ri)) (R2: (x2)Pat (R2)) e.g. Λ/Μ in Peter kissed Sheila 2. Head of term phrase Any entity type: 2.1. property/relation

(Ri : (τ ι : • (τ 0) (Ri)... (RN))

( R , : ( f , : • ( f , ) ) ( R 1 ) . . . (RN))

e.g. color in the color of her coat 2.2. individual (RI: (xi: • (x,)) (R T )... (RN))

e.g. coat in her new coat 2.3. state of affairs (RI: (ei: • (e,)) ( R , ) . . . (RN))

e.g. wedding in her third wedding 2.4. propositional content (Ri:(pi: • (PI))(RI) ··. (RN)) e.g. hope in her last hope

112

Daniel García Velasco and Kees Hengeveld

3. Modifier of term phrase (Ri: (τ i: - (τ ifc: (fi: • (fi)) (τ ,)β (Ri)) e.g. new in her new coat 4. Modifier of predicate phrase (Ti: (f,: - (fO: (f 2 : • (f2)) (fO) (T,)) e.g. beautifully in She dances beautifully 5. Modifier of predication (e,: Η (ei): • (ei)0) e.g. yesterday in I met her yesterday 6. Modifier of proposition (PI: H (Pi): • (Pi)o) e.g. probably in She's probably gone by now Note that we are here considering frames that capture the syntactic environments in which lexemes may occur. But frames can also be used within syntax itself to define possible syntactic configurations. Thus, the general frame for predications would be as in (30), which is an expanded version of (29.5): (30)

(e,: [(T,: ( f , ) ( T , ) ) (R,: (τ I) S F ( R I ) ) . . . ( R N : (TN)SF (RN))] (ei):

(Φι) (ei)) Frame (29.3) may now be turned into a more general frame, as in (31): (31)

(RI:

(τ Ι: - (τ fo · ) ( R 0 ... (RN))

where the " · " indicates the position where a term modifier has to be inserted. This modifier may either be a lexical frame, or a syntactic

Do we need predicate frames? 113

frame, such as (30). The latter case then accounts for relative clauses. Since we are dealing with lexemes in this paper, we will not go into this issue any further. 4. Linking rules 4.1. Meaning definitions as mediators

When lexemes are separated from frames, the question of course arises how lexemes are linked to the predication frames in which they are allowed to occur. In current FG the predicate frames define the quantitative and qualitative valency of lexemes. When separating lexemes from their frames an alternative solution to the problem of linking lexicon to syntax has to be developed. The solution we adopt here follows up on Garcia Velasco (1998), itself inspired by, particularly, Jackendoff (1990) and Van Valin and LaPolla (1997). The solution necessarily involves the introduction of abstract meaning definitions. The abstract meaning components serve as triggers in linking the lexeme involved to a specific predication frame. If linking were to be based on concrete meaning definitions, every link would have to be specified separately. Thus, we arrive at the same conclusion as Mairal Usón and Faber (this volume), even though we arrive at this conclusion on the basis of our study of syntactic configurations, whereas their study take sit point of departure in the lexicon. We will assume that lexemes are listed in the Fund together with a specification of their category (which generally restricts the number of frames in which they may occur) and of their idiosyncratic morphological properties. Lexemes can furthermore give access to a lexical paradigm (Dik 1997: 84) in which irregular forms, collocational patterns, etc. are included. Abstract meaning definitions, in turn, should capture the syntactically relevant information which accounts for the lexeme's distributional properties. In the approach advocated here, abstract meaning definitions should include the information necessary to link the lexeme to an appropriate predication frame.

114

Daniel García Velasco and Kees Hengeveld

In the construction of our definitions we shall be making use of standard primitive relations, which, in a spirit similar to that of Jackendoff (1990), will be assumed to define basic ontological categories of conceptual nature. Lexemes should thus specify the ontological category they designate and the number of participants required in the lexical relation. The linking procedure is thus mediated by the number of entities present in the abstract meaning definition which will have to be projected onto syntax. This restriction captures the basic intuition that syntactic constructions must be tied to a semantic interpretation. It furthermore involves recognizing the relevance of FG's ontological categories at the conceptual level (Nuyts 1992).8 By way of illustration, let us consider a simple case of linking. The following could be the abstract meaning definition of the lexeme open: (32)

open [V] [fi: [CAUSE (xi) [BECOME open' (x2)]]]

This definition expresses that the lexeme open designates a relation (as represented by the "f ' variable) between two entities (as represented by the "x" variables). The presence of these variables, together with the restriction introduced above, will guide the linking process towards the selection of a transitive predication frame. The lexeme will thus become the head of a predicate phrase. If we further accept that semantic functions can be defined on the basis of the position which a variable takes in a given abstract meaning definition (Jackendoff 1990; Schack-Rasmussen 1994; Van Valin and LaPolla 1997), the role of each participant in the predication can also be obtained from the lexemes's abstract meaning definition. Thus, we may assume that the argument of a CAUSE function will be assigned the role of Agent whereas the argument of the BECOME function will receive the function Patient. On the basis of this information, the lexeme open may be linked to frame 1.2.4 in (29), with the result presented in (33):

Do we need predicate frames? 115

(33)

(T,: (fi: open [V] (f,)) (T,)) (R,: (X ,)Ag (R,)) (R2: (Χ2)Ρ* (Ra))

which would be used to account for sentences such as: (34)

Sheila opened the door.

However, the evidence shown in section 2 suggests that the linking of lexemes to predication frames cannot be developed on a one-toone basis. Quite on the contrary, we showed in that section that lexemes may take different frames if they have more than one reading. Thus, the lexeme open has another meaning definition, given in (35): (35)

open [V] [fi: [BECOME open' (xi)]]

In this case, the information available in the abstract meaning definition links the lexeme open to the intransitive frame 1.1.3 in (29), with the result presented in (36): (36)

(T,: (f,: open [V] (f,)) (T,)) (R i: (τ ,)Pal (R,))

which would be used to account for sentences such as: (37)

The door opened.

Thus, one lexeme may be linked to more than one predication frame if it has more than one meaning definition.

4.2. Predicate formation The approach presented in the previous section seems a reasonable solution for those cases in which the existence of more than one meaning definition for a single lexeme is relatively arbitrary and unpredictable. The fact, however, that open may be used in the two ways illustrated above is not an arbitrary fact of English, but is an

116

Daniel García Velasco and Kees Hengeveld

example of the causative alternation that we discussed in 2.1. As shown there, FG accounts for this alternation by means of a Predicate Formation Rule. In our approach, this option is no longer available. Predicate Formation Rules operate on predicate frames: they take a predicate frame as input and produce a predicate frame as output. In our proposal predicate frames have disappeared, so another procedure has to be developed to account for productive alternations. Let us once more contrast the two meaning definitions for the verbal lexeme open·. (38)

open [V] 1. [fi: [CAUSE (xi) [BECOME open' (x2)]]] 2. [fi: [BECOME open' (xi)]]

The relation between the two meaning definitions of open is evident: the second one consists of a subpart of the first one. This means that it is relatively easy to formulate a rule which derives one abstract meaning definition of open from the other, and then link each of the definitions to a different predication frame. Predicate Formation may thus be replaced by Lexeme Derivation. It remains an open question whether in (38) meaning definition 2 is derived from 1 (as in Mairal Usón and Faber this volume) or the other way around (as in Rappaport and Levin 1998), although the latter solution seems most consonant with FG's constraints on deletion (Dik's 1997: 18-24). We should admit that this solution does not solve any of the problems involved in dealing with alternations through Predicate Formation, but it does not add new problems either. Whatever the nature of the rules, they can only be defined in terms of some open-ended class of input predicates. In most cases alternations do not seem to be fully productive, and derived lexemes may acquire specific meanings not predicted by the rule. In this respect Cuvalay-Haak (1997: 120) makes a very useful distinction between "incidental predicate formation" and "recurrent predicate formation". In her approach, incidental predicate formation rules "account for the ability to form new predicates on the basis of perceived regularities in the lexicon", while recurrent predicate formation rules "specify formal and semantic modi-

Do we needpredicate frames? 117

fications which are completely predictable in terms of their effect on the input predicate". Only in the latter case would we make use of Lexeme Derivation Rules. In the former case, there is a creative process of lexeme formation which is not part of the grammar. Just like a speaker may manipulate selection restrictions to create new meanings for existing lexemes, so he may manipulate perceived regularities to create new meanings for new lexemes. In doing so the speaker combines his knowledge of the world with his knowledge of the language to improve his instrument of communication. This process should be described in terms of the wider cognitive abilities of speakers, rather than in terms of their linguistic capacities. We explore this issue a bit further in the next section.

5. Conclusion: Grammar and cognition Jackendoff (1983:19) has rightly observed that one could understand lexical meaning as "those conceptual structures that happen to be verbally expressible". As such, the lexicon is a mediator between our cognitive-pragmatic abilities and our linguistic ones. This means that it is difficult to draw a sharp line between grammar and cognition, and between abstract meaning definitions and what some authors have called Lexical-Conceptual Structures (Jackendoff 1990; Hale and Keyser 1986). As soon as a speaker wants to transmit a "conceptual structure" that is not "verbally expressible" he may want to coin a new expression. The possibilities this speaker has are restricted by the language system, so, as stated earlier, he will have to invoke his knowledge of the language next to his knowledge of the world. The linking mechanism that we proposed in the previous section is a static one and is part of the grammar. A lexeme has one or more meaning definitions, and every meaning definition links up to a specific predication frame. The creative use of language, however, often involves the use of lexemes in configurations in which they normally would not occur, and requires more than the consistent application of a grammatical system. Yet, the creative use of language may lead to the successful introduction of a new meaning definition of a lexeme

118

Daniel García Velasco and Kees Hengeveld

into the language (cf. Cuvalay-Haak 1997: 121). This can be illustrated by returning to an example from one of the flexible languages introduced in 2.3, Mundari. In this language lexemes are not tied to a particular syntactic slot. As a result, a single lexeme may be used with many different, albeit related, meanings. Thus, a lexeme like buru in (22) may mean, depending on the syntactic slot in which it appears: (i) mountain; (ii) to heap up; (iii) to keep up certain feasts or to hold a fair (because they take place on mountains); (iv) to call something a mountain. Meaning (iii) is not as evident as the other ones, probably because it is culturally determined. This particular meaning of the lexeme must have arisen at some point. Let us imagine a speaker who has no lexeme available to refer to a feast, or is unsatisfied with the one he has. In conceptualizing a feast, the image of a mountain dominates in the general picture of our speaker, since in the Munda community feasts take place on mountains. Taking advantage of the syntactic possibilities of Mundari, our speaker then coins the word buru in the head position of the predication frame of an intransitive predicate phrase. Our speaker is successful, and the set of meaning definitions of buru now includes this new meaning. This creative process is revealing as regards the interface between cognition and language, or between conceptualization and semantics. In our linguistic analysis we can only deal with the conventionalized uses of lexemes. But these conventionalized uses are pointers to a body of knowledge associated with the properties or relations these lexemes designate.9 Yet this wider body of knowledge falls outside the scope of our linguistic analysis unless systematic relations between (meaning definitions of) lexemes can be established. This doesn't mean that conceptualization shouldn't bother us. In fact, in Functional Discourse Grammar, the most recent variant of FG (Hengeveld to appear a), the grammatical module operates within the setting of a more-encompassing cognitive one. Clearly, one aspect of this new approach that needs further elaboration concerns precisely the internal structure of the cognitive component and its interaction with the grammatical component. This paper confirms that the lexicon is the right place to start investigating this problem.

Do we need predicate frames? 119

Notes 1. The authors wish to thank an anonymous referee for his/her comments on an earlier version of this paper. 2. In Hengeveld (1992a) these were called "general predicate frames". 3. It might well be possible that, through frequent usage, some derived predicates are available in the lexicon and related to their basic counterparts through redundancy rules (see De Groot (1989: 133) and Kahrel (1989) for some discussion). Still, unfolding the predicate formation rule seems to be necessary to interpret the output frame correctly, since some of its semantic properties (including its meaning definition) derive from the input predicate frame. 4. Compare also Kahrel (1989: 137) who calls for a more "liberal" lexicon in which many derived predicates are listed, thus reducing the heavy burden placed on predicate formation rules in language processing. 5. See, however, Kristoffersen (1992), who assumes that predicate formation rules can take predications and propositions as their input in his account of West Greenlandic morphology. 6. The labels "Actor" and "Undergoer" are taken from Role and Reference Grammar (Foley and Van Valin 1984; Van Valin and LaPolla 1997). The macrorole "Benefactive" is added here since it is needed to account for the differences between languages with and without applicative constructions. 7. The abbreviations used for semantic functions are the following: SF = any semantic function, 0 = Zero, Po = Positioner, Pat = Patient, Ag = Agent, Ref= Reference. Note that we do not distinguish between Goal and Processed, as does Dik (1997), but use the semantic function Patient for both. 8. This intuition has been formalized in different forms recently; see for example Van Valin and LaPolla's (1997) Completeness Constraint and Rappaport and Levin's (1998) Subevent Identification Condition. 9. Part of this body of knowledge extends over what Leech (1974) names Connotative Meaning, defined as the real world experience one associates with an expression.

120

Daniel García Velasco and Kees Hengeveld

References Bhat, D.N.S. 1997

Noun-verb distinction in Munda languages. In: Anvita Abbi (ed.), Languages of Tribal and Indigenous Peoples of India, 227-251. Delhi: Motilal Banarsidass Publishers. Boland, Julie E., Michael K. Tanenhaus, Susan M. Garsney and Greg N. Carlson 1995 Verb argument structure in parsing and interpretation: Evidence from wh-questions. Journal of Memory and Language 34: 774806. Boland, Julie E. 1993 The role of argument structure in sentence processing: Distinguishing between syntactic and semantic effects. Journal of Psycholinguistic Research 22: 133-152. Butler, Christopher S. 1998 Enriching the FG lexicon. In: Hella Olbertz, Kees Hengeveld and Jesús Sánchez (eds.), The Structure of the Lexicon in Functional Grammar, 171-194. Amsterdam/Philadelphia: John Benjamins. 2001 A matter of GIVE and TAKE: Corpus linguistics and the predicate frame. In: Maria Jesús Pérez Quintero (ed.), Challenges and Developments in Functional Grammar, 55-78. (Revista Canaria de Estudios Ingleses 42.) La Laguna: Servicio de publicaciones. Carlson, Greg Ν. and Michael Κ. Tanenhaus 1988 Thematic roles and language comprehension. In: Wendy Wilkins (ed.), Syntax and Semantics. Thematic Relations, 263-288. New York: Academic Press. Comrie, Bernard 1981 Language Universals and Linguistic Typology. Oxford: Blackwell. Cuvalay-Haak, Martine 1997 The Verb in Literary and Colloquial Arabic. Berlin/New York: Mouton de Gruyter. Dik, Simon C. 1980 Studies in Functional Grammar. London: Academic Press. 1997 The Theory of Functional Grammar. Part 1: The Structure of the Clause. Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter. Fillmore, Charles J. 1986 Pragmatically controlled zero anaphora. BLS XII: 95-107.

Do we need predicate frames? 121 Foley, William A. and Robert D. Van Valin, Jr. 1984 Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Garcia Velasco, Daniel 1998 La predicación nuclear en la GF de S.C. Dik: una perspectiva conceptual. Ph.D. dissertation, Universidad de Oviedo. García Velasco, Daniel and Carmen Portero Muñoz to appear Understood objects in Functional Grammar. Working Papers in Functional Grammar. Groot, Casper de 1989 Predicate Structure in a Functional Grammar of Hungarian. Dordrecht: Foris. Hale, Kenneth and Samuel J. Keyser 1986 Some transitivity alternations in English. Lexicon Project Working Papers 7. Hengeveld, Kees 1992a Non-verbal Predication: Theory, Typology, Diachrony. Berlin/New York: Mouton de Gruyter. 1992b Parts of speech. In: Michael Fortescue, Peter Harder and Lars Kristoifersen (eds.), Layered Structure and Reference in a Functional Perspective, 29-55. Amsterdam/Philadelphia: John Benjamins. to appear a The architecture of a Functional Discourse Grammar. In: J. Lachlan Mackenzie and María de los Angeles Gómez González (eds.), A New Architecture for Functional Grammar. Berlin/New York: Mouton de Gruyter. to appear b State-of-Affairs concepts. In: Geert Booij, Christian Lehmann and Joachim Mugdan (eds.), Morphology: A Handbook on Inflection and Word Formation. Berlin/New York: Walter de Gruyter. Hoffman, John 1903 Mundari Grammar. Calcutta: Bengal Secretariat Press. Jackendoff, Ray 1983 Semantics and Cognition. Cambridge: MIT Press. 1990 Semantic Structures. Cambridge: MIT Press. Junger, Judith 1987 Predicate Formation in the Verbal System of Modern Hebrew. Dordrecht: Foris. Kahrel, Peter 1989 On the representation of the lexicon in FG. In: John H. Connolly and Simon C. Dik (eds.), Functional Grammar and the Computer, 135-150. Dordrecht: Foris.

122

Daniel García Velasco andKees Hengeveld

Kristoffersen, Lars 1992 Derivation and inflection in a Functional Grammar of West Greenlandic. In: Michael Fortescue, Peter Harder and Lars Kristoffersen (eds.), Layered Structure and Reference in a Functional Perspective, 143-171. Amsterdam/Philadelphia: John Benjamins. Kuipers, Aert H. 1968 The categories verb-noun and transitive-intransitive in English and Squamish. Linguali: 610-626. Leech, Geoffrey 1974 Semantics. Harmondsworth: Penguin. Mackenzie, J.Lachlan and Mike Hannay 1982 Prepositional predicates and focus constructions in a Functional Grammar of English. Lingua 56:43-57. Mairal Usón, Ricardo and Pamela Faber this volume Functional Grammar and lexical templates. Mosel, Ulrike and Einar Hovdhaugen 1992 Samoan Reference Grammar. (Instituttet for sammenlignende kulturforskning, Series B: Skrifter, LXXXV.) Oslo: Scandinavian University Press. Nuyts, Jan 1992 Aspects of a Cognitive-Pragmatic Theory of Language. Amsterdam/Philadelphia: John Benjamins. Rappaport, Malka and Beth Levin 1998 Building verb meanings. In: Miriam Butt and William Geuder (eds.), The Projection of Arguments: Lexical and Compositional Factors, 97-134. Stanford: CSLI. Schack-Rasmussen, Lone 1994 Semantic functions in perspective - reconsidering meaning definitions. In: Elisabeth Engberg-Pedersen, Lisbeth Falster Jakobsen and Lone Schack Rasmus sen (eds.), Function and Expression in Functional Grammar, 41-63. Berlin/New York: Mouton de Gruyter. Shapiro, Lewis P., Edgar B. Zurif and Jane Grimshaw 1989 Verb processing during sentence comprehension: Contextual impenetrability. Journal of Psycholinguistic Research 18: 223243. Siewierska, Anna 1991 Functional Grammar. London: Routledge. Spruit, Arie 1986 Abkhaz Studies. Ph.D. dissertation, University of Leyden.

Do we needpredicate frames? 123 Stowe, Laurie 1987

Thematic structure and sentence comprehension. In: Greg N. Carlson and Michael K. Tanenhaus (eds.), Linguistic Structure in Language Processing, 319-357. Dordrecht: Reidel. Van Valin, Robert D. and Randy J. LaPolla 1997 Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Adpositions, the lexicon and expression rules Dik Bakker Anna Siewierska

0. Introduction In Functional Grammar, there is a strict demarcation line between the lexicon, underlying representations and expression rules. The lexicon is the depository of predicates, their form, category and meaning. Four types of predicates are distinguished, which are the major ("open") word classes: Verbs, Nouns, Adjectives and Adverbs. They appear as such in underlying representations of which they are the semantic backbone. Other lexical elements, belonging to minor ("closed") word classes such as auxiliaries, adpositions and articles, are not located in the lexicon. They are considered to have no inherent meaning but only a grammatical meaning. They are coded in underlying representations via operators and functions and introduced in the utterance by expression rules, just like purely morphological and phonological material. It is not clear where all this grammatical material is stored if not in the lexicon; it is heavily suggested that it forms part of and comes with the expression rules themselves. This may be feasible for a restricted set of morphological markers, which have few or no phonologically independent allomorphes. It is much more complicated for auxiliaries, which in many respects behave like verbal predicates, and for adpositions, which in many languages share morphosyntactic properties with nouns and verbs. In particular, however, this black-and-white approach to the different types of lexical elements creates a problem for the treatment of grammaticalization processes. In such cases, an auxiliary emerges from what was once a full verb, or an adposition develops from what was once a full noun. However, these processes take a long time,

126

Dik Bakker and Anna Siewierska

typically hundreds of years. Between the two extremes of "pure" verbal predicate and "pure" auxiliary verb lies a whole trajectory of minimal, often hardly noticeable changes in form and function and of variation between speakers at any one time. There is no single moment somewhere in the middle of this process, neither from the perspective of the individual speakers nor from that of the language as a system, that could be pinpointed as The Moment of the Shift, when the element under consideration jumps from the lexicon into the grammar. In this contribution, we outline a possible solution to this problem. Our proposal centers around two points. The first is a slight modification of the contents of the lexicon; i.e. the inclusion of elements with a (partial or a full) grammatical meaning. The second is the use of our dynamic version of the expression rules, discussed in full in Bakker (2001), which allows for a more shaded way of introducing lexical material. The text is organized as follows. In section 1, in order to support the later discussion we present a brief version of our expression rule proposal. In section 2 we unfold our ideas about the organization of the lexicon in FG. Section 3 provides information about the domain we would like to use for our exercise: adpositions. In sections 4, 5, 6 and 7 we discuss four different diachronic stages in the development of prepositions, in terms of our ideas on the lexicon and on expression. Several examples will be given of concrete cases of expression in these terms. Finally, section 8, mentions some aspects of adpositions that had to remain outside the discussion, but which should be considered for a proper evaluation of our proposals.

1. Overview of the dynamic model of expression In their standard form, as presented in Dik (1997), the expression rules (ERs) come in three phases. Starting out from a fully specified underlying representation (UR), which contains the predicates and the relevant functions and operators, first the grammatical elements are generated. This gives us both the independent grammatical mor-

Adpositions, the lexicon and expression rules

127

phemes such as determiners, auxiliaries and prepositions, and the respective kinds of bound morphemes, such as tense markers and agreement suffixes. In a second step, all lexical and morphological material is ordered into a linear string. Thirdly, this string is given its final phonological form, leading to the expression as such. In Bakker (1994, 1999, 2001) it is shown that this canonical organization of the ER in three sequential steps has two major shortcomings. Firstly it over generates, i.e. it may produce all kinds of forms that do not occur in any existing language. This precludes one from gaining insight into the constraints on (some components of) grammar. But overgeneration is not only a theoretical issue. It puts into question the leamability of a FG grammar, and as such endangers its cognitive adequacy. And secondly, the ER component as it stands also under generates, i.e. it cannot produce certain forms that actually occur in languages. Linguistically, this is a more directly obvious shortcoming, since it endangers the descriptive adequacy of the theory. A clear, and ubiquitous example of the latter are constructions in which form and order interact. (1) from Breton (Borsley and Stephens 1989) and (2) from Dutch are a case in point. (1)

(2)

a.

Ar vagale ne lenn-ont ket levrioù the children PCL read-3PL not books b. Ne lenn ket ar vugale levrioù PCL read not the children books 'The children do not read books.' a. Jij speel-t goed vandaagl you play-2SG well today 'You play well today! ' b. Vandaag speel je goedl Today play-0 you well 'Today you play well! ' c. Ik zie dat jij vandaag goed speel-t\ I see that you today well play-2SG Ί see that you play well today!'

128

Dik Bakker and Anna Siewierska

In the (a) examples, with SV order, we have a subject agreement marker on the verb. In the (b) examples, with VS order, there is no such marker; the verb appears as a bare stem. The complement clause of (2c), with its more or less fixed SOV order, the 2 person singular agreement marker is always expressed. If the expression rules first generate the grammatical forms and only then establish linear order, there is no way to determine whether the suffix should be generated or not. The only solution would be to look at the factors that determine the final order - here: pragmatic aspects of the subject constituent and the availability of another potential PI filler and the level of embedding - but this would imply "prerunning" the ordering rules in some way.1 A possible escape from this apparent vicious circle is the integration of Stage I and II into one process, which interleaves form and order aspects. This is what has been proposed in Bakker (1999). As in the standard model of expression, the point of departure is a UR, with all the predicates present. Also the templates and the placement rules are retained, be it in another guise. The first addition to the standard approach is that the templates necessary for the expression of the UR concerned are combined into a tree-like structure as in Figure 1. Although such a hierarchical organization of templates has always been possible - and maybe even been implied - in the theory, it has never been explicitly introduced as such, nor has any other form of constituent structure been posited. Although Figure 1 may be seen as the traditional (static) representation of the constituent structure underlying the corresponding expression, it codes in fact the history of the (dynamic) process leading to that expression. In other words, it shows what templates have been invoked for the respective slots of higher ordered templates, and on the basis of which UR material they have been selected. Thus, it is precisely the concrete functional linguistic information in a template slot which has been selected from the information in the slot above it on the basis of slotspecific placement rules which in its turn determines the way in which it will be expressed by selecting the right template. The recursive process of slot filling and template selection can be represented

Adpositions, the lexicon and expression rules

129

statically as in Figure 1 for the (simplified) underlying representation in (3). MAIN CLAUSE f=UR)

VINFIN

TERMfSUBJECTi

VERB [IMPRF.GERND]

growing DET NOUN NOMRESTR [DEF,PL] [PL]

'the'

'trees'

TERM

PREP [POSS]

[REM,SG] NOUN

'of

'that'

'theory'

Figure 1. Constituent structure tree of nested templates

(3)

[DECL E,: [X,: [PRES en [PROG [grow [V] (DEF PL x,: tree [N]: (REM SG x2: theory [N])POSS)ZERO, SUBJECT]]]]] 'The trees of that theory are growing.'

130

Dik Bakker and Anna Siewierska

In Figure 1, templates are found as the row of categories at the respective levels. E.g. PI VFIN VINFIN on the second row is the appropriate template for the clause level. The appropriate template for the actual filler of the PI position, the subject term in this case, turns out to be DET NOUN NOMRESTR. This process of repeated template selection and slot filling provides us with a tree in which functional categories (the ones in bold print) and formal categories alternate. The process bottoms out in a row of nodes with lexical and grammatical categories, such as NOUN and ADPOS, or a cluster of grammatical elements, such as [REM, SG]. By organizing the distribution of information in this way, we provide a dynamic character to the process of tree construction, in which the sequence of slot filling, and therefore time plays a crucial role. This makes it possible to postpone the determination of the final form of both predicates and grammatical elements, maximally till the moment at which they are expressed and minimally till the moment at which their own choice and shape determine certain aspects of the expression. We will organize this process of dynamic tree construction by applying the following principles.2 Principle 1: Constituent structures are developed top down from the material in the UR. Thus, first the higher syntactic constituents, such as noun phrases, are developed, then the individual free morphemes, next the bound morphemes and finally their phonological form. In this way, and in contrast to the standard model, the explanatory hierarchy of FG, as discussed in Dik (1986) is respected, which runs from syntax to morphology to phonology. Principle 2: Development takes place from left to right. This is the "natural" order in which linguistic forms are uttered in the first place. Left to right ordering may be expected to shape and have shaped language over time, at least to some extent, and more than any other order. Principle 3: Development works depth first. This means that of any two contiguous elements to be expressed, at whatever level, the left-

Adpositions, the lexicon and expression rules

131

most one will be completely expanded up to its terminal forms before the rightmost one will be considered, and this recursively. An implication of this is that of the complete information which is necessary for the production of the whole utterance, only a fraction will be available at any one time. This reduces considerably the burden for short term memory, inherent to breadth first development. In combination with Principle 2, this may solve a number of problems of the type illustrated in example (1) above. Principle 4: For any node Ν in the tree, all features found on a direct path from that node to the top node are in principle available to be inherited by N. Trivially, this implies that overt primary operators would be available for every node in the tree. However, a distinction will be made between "raw" UR material (functions, π and ω operators, inherent features of predicates such as Animacy or Gender) on the one hand, and μ operators on the other hand. In order to be accessible for inheritance, it will be assumed that any UR feature has to be transformed into a μ operator, either in a one-to-one or a more-to-one fashion, as in the case of portemanteau operators. An implication of this is that only those features of URs qualify as such if they are turned into a μ operator at some stage during expression. Apart from this there may be universal or language (type) specific downward barriers that make features inaccessible to lower nodes. Barriers may be both of a functional and a formal nature, i.e. determined by a layer or substructure in the UR or by some syntactic or morphological boundary, i.e. a specific node in the tree. Principle 5: Operators may percolate, i.e. move upwards to higher nodes. For percolation there may be universal and language (type) specific upward barriers, that make them inaccessible to higher nodes in the tree. Barriers to percolation are only of a formal nature. These principles should contribute to the (cognitive and typological) adequacy of the dynamic model. On the one hand, this organization means that the linguistic forms are produced precisely in the order in which they are uttered by the speaker in a live setting, thus giving the

132

Dik Bakker and Anna Siewierska

model a real time procedural flavor. On the other hand, several constraints follow from these principles which restrict the formal power of the expression component, by determining what information has to be available at what stage in the tree expansion process. In this way, it may be determined more precisely what information necessarily has to come in from the UR as opposed to during the course of expression and what comes during expression, and thus, what has predicate status in the language under consideration and what has not. Furthermore, such constraints provide a means of determining the order in which grammatical material becomes available. E.g. a case marking postposition should be available on some node in the tree before the elements on which that case is expressed are created. Since grammatical forms may be created at several stages in the expansion process, we also get an instrument to distinguish between more and less grammaticalized elements; this may reflect the length of their diachronic path through the language, and their relative role in syntax and morphology. The constraints implied by principles 1-5 above determine in fact the information flow through the constituent structure tree that is dynamically constructed by them. Crucial is now what functional and formal information may be available at which node during tree expansion. In order to accommodate this information, we think that nodes should have the following internal structure, depicted in Figure 2.

The label (abbreviation: Lab) represents the category of a slot in a template, such as PI, Subject or Vinfin. The configuration (Config) is a specification of some part of the underlying representation that will be expressed by this node. It can not be just any part of the UR; it should be part of the configuration of the mother node of the node under consideration. Examples are the specification of the subject term, a temporal satellite, the head of a term, and the main verbal predicate. Typically, such descriptions will contain one or more predicates; in those cases we will call them functional configurations. A configuration may also be the description of a grammatical element, such as an auxiliary, a demonstrative, or a tense suffix. In that case we will call them formal.

A ¿¡positions, the lexicon and expression rules

133

The functional features (FncFtrs) are the primary μ operators relevant for this slot. They may be derived from operators, functions and lexical elements of the Config, or they are directly inherited from the FncFtrs field of the mother node. Examples of functional features are Number, Tense, Animacy. Formal features (FrmFtrs) are the auxiliary μ operators relevant for this slot. They may be derived from the Config and the FncFtrs, be inherited from the mother node, or percolate upwards from a daughter node. Examples of formal features are Case and Finiteness. Finally, the Subcategorization (SubCat) is a template consisting of nodes for the categories into which Config is to be expressed. The appropriate template is selected from the set of templates of the language on the basis of the concrete value of the functional part of the node, via more or less general rules. A template may also be inserted on the basis of lexical information, typically from the head of the structure in Config, for cases where the default node specifications would lead to the selection of the wrong syntactic environment. In that case we have an instance of lexical priority. All these aspects together represent what in the standard model of the expression rules are Stage I and Stage II. We have seen that the templates and placement rules of Stage II are replaced by the subcategorizations and the functional configurations of our model, respectively. Furthermore, formal configurations and the feature sets replace the grammatical formation rules of Stage I. All these aspects together represent what in the standard model of the expression rules are Stage I and Stage II. We have seen that the templates and placement rules of Stage II are replaced by the subcategorizations and the functional configurations of our model, respectively. Furthermore, formal configurations and the feature sets replace the grammatical formation rules of Stage I.

134

Dik Bakker and Anna Siewierska

LABEL functional aspects CONFIGURATION

FUNCTIONAL FEATURES formal aspects FORMAL FEATURES

SUBCATEGORIZATION

Figure 2. Structure of a node

To round off this short sketch of the dynamic model of the expression component, we give the complete derivation of a very simple sentence. It is the expression of the UR given in (4). (4)

[DECL Ei: [Xi: [PRES ei: ['smart' [A] (REM PL x,: 'girl' [N])ZERO, SUBJECT]]]]

'Those girls are smart.' The first step is the selection of a node that will express the UR of (3). For its Config field, this node will have a very general representation of a UR, with variables for all relevant functions, operators and predicates. After its selection, the Config field will be unified with the UR, and the respective variables will get the corresponding values. (5) gives a simplified version of the top node before and (6) after unification. Upper case labels indicate free variables; lower case represent the values for these variables after binding with the corresponding values from the UR.3

Adpositions, the lexicon and expression rules

135

(5)

NODE 1 (uninstantiated) Lab: sentence Config: PLLOC [[TENSE [PRED CAT ARG1]]]] FncFtrs: ILLOC, TENSE, CAT, NUMBER.SUBJECT, PERSON.SUBJECT FimFtrs: SubCat:

(6)

NODE 1 (partially instantiated) Lab: sentence Config: [DECL Ei: [Xi: [PRES en ['smart' [A] (REM PL xi: 'girl' [N])zero, subject]]]] FncFtrs: DECL, PRES, [A], PL, [-sp,-hr] FrmFtrs: SubCat:

The last step for this node is the selection of the right template for its SubCat field. The information in the Config field is instrumental in this. The result is given in (7). (7)

NODE 1 (fully instantiated) Lab: sentence Config: [DECL Ei: [Xi: [PRES e,: ['smart' [A] (REM PL x i : 'girl' [N])zero,subject]]]] FncFtrs: DECL, PRES, [A], PL, [-sp,-hr] FrmFtrs: SubCat: pi, subject, vfin, mainpred

This node is fully instantiated because there are no further variables to be bound. If this were to be the case, then they should get their values via percolation from lower nodes. The next step will be the development of the pi position. Typically, there will be several alternative versions for this node in the grammar, e.g. with a Focus, Topic or Subject term for its configuration, but also with adverbials. In our case, the Subject will be the UR element chosen. This will trigger a term template for the filler of the

136

Dik Bakker and Anna Siewierska

SubCat of Node 2. The next step will then be the expansion of the Determiner node of this template, assuming that this is the first slot. Figure 3 gives the complete flow of the expansion of Node 1, up till the expression of the "mainpred" slot. This should serve for a brief overview. It should set the stage for the next sections, in which we will discuss some more complicated cases of expression. In particular, we will be involved with the consequences of the model for the way in which predicates and nonpredicational material are represented in the lexicon.

1: SENTENCE

7: VFIN

2: PI

/•w 'girl' Figure 3. Full tree expansion

9: MAINPRED

ii 12

's'

iL 13

16

19

'smart'

Adpositions, the lexicon and expression rules

137

2. The lexicon in Functional Grammar As reflected in the contributions in Olbertz et al. (1998) there is a general agreement within the FG community that the lexicon consists at least of the complete set of lexical entries - also called predicate frames - of the language concerned. A lexical entry typically contains the following fields of information. Fl. The form of the predicate. This is a phonological representation rather than a "citation" form as found in dictionaries. In fact it is the stem or the root, i.e. it provides precisely the information necessary to derive regular inflected forms. F2. The predicate variable, f. Through this variable, reference may be made to the predicate when it appears in a text, where it will be uniquely indexed. Although in the FG literature it is often left out of concrete representations in the literature, and in fact we will be omitting it too, we will assume that it has been accepted as part of the theory. F3. The word class. This is not restricted to a plain predicate type V, N, A or Adv. The type may be extended with a subcategory necessary not only for its proper use in underlying representations, but also in expression. Thus, for nouns we may need gender information for the right choice of article or anaphoric pronoun (e.g. [N, neuter]). For verbs the state of affairs that the verb codes may be represented in this way, with all its implications for the semantic context in which it may appear. E.g. subcategorization [V, action] indicates that the verb in question may be used with a direction satellite, as opposed to verbs of category [V, state]. Such subcategories may also be necessary for the right choice of paradigm in case Fl is not sufficient for this. F4. The quantitative valency, i.e. the number of arguments the predicate takes.

138

Dik Bakker and Arma Siewierska

F5. The qualitative valency, i.e. the nature of the arguments of the predicate as reflected for instance in their semantic functions. Directly related to the latter are: F6. The selection restrictions on the fillers of the argument positions. Typically, these are stated in terms of grammatical notions such as , but they may also be more specific, and represented by predicates of the language, such as "piece of clothing". F4-F6 is often considered to be the predicate frame in a more restricted sense. We will use the notion "predicate frame" in this sense below; we will use "lexical entry" for the complete set of information available for any entry in the lexicon. F7. The meaning definition. This is a more or less complete representation of the meaning of the predicate in terms of other predicates of the language. They are constructed according to the same principles as underlying representations of clauses. The Theory of Stepwise Decomposition (Dik 1978) determines which are the atomic predicates of the language, i.e. those predicates of which the meaning is not further decomposable. So, instead of a meaning definition, based on some type of formal logic, Functional Grammar, following Carnap (1956: 222f), prefers in fact to provide the lexicon with meaning postulates, which are based on a non-logical vocabulary, e.g. that of the (natural) language under consideration. We will therefore use this notion - in short MP - rather than "meaning definition" below. A second point is whether MP's are encyclopedic, i.e. contain all a speaker knows about a certain predicate, or linguistic, i.e. restricted to what is necessary to decide on the wellformedness of a sentence. Although we think that a functional theory eventually should make explicit how we get from lexical elements to their use in actual discourse, including all their connotations and implications, we will follow Allan (2001: 84f) here in assuming a link between the lexicon and the encyclopedic knowl-

Adpositions, the lexicon and expression rules

139

edge of the speaker which we will leave out of the discussion for our purposes here.4 According to the theory of FG, part of the information discussed under F1-F7 above may be redundant. E.g. may be implied by the function Agent; this function itself may be predicted from the fact that a verb has category + subcategory [V, action]. Redundancy rules will take care of these generalizable aspects. Furthermore, lexical elements whose form cannot be derived from existing predicates via productive rules of derivation or inflection should be represented separately in the predicate frame they are semantically related to. E.g. the frame for the English verbal predicate give should contain the irregular forms gave and given (cf. Dik 1997: 84). The same goes for morphologically complex elements whose meaning may not be derived from the composing elements in a straightforward way, as is the case with most compound nouns, and some terms. They will receive their own lexical entry, and become part of the extended lexicon, or Fund. The above could be viewed as the most economical representation of the lexicon, seen from the perspective of a purely linguistic model of the grammar. It is, however, not necessarily the right model of the lexicon to fit a model of the language user, i.e. the speaker or the hearer. The latter is envisaged in the Model of the Natural Language User (M.NLU) presented in Dik (1997: If) as the ultimate goal of linguistic theory. If we assume that, from a monolingual perspective, there is just one mental lexicon to serve both modes of language use, then "burying" irregular forms of a nominal, verbal or adjectival paradigm in the entry of the corresponding stem or root seems to be counterintuitive. They are hard to retrieve by the hearer, since they have no lexical entry of their own, and may differ completely from their stem or root in a formal sense. It is also often the case that such irregular forms are highly frequent in actual language use. Some authors, notably Weigand and Hoppenbrouwers (1998), basing their view on psycholinguistic research, suggest that even regular inflected forms of more than average frequency should get their own entry in the (mental) lexicon, to account for the high speed with which they are retrieved. We will follow these authors in assuming that, apart

140

Dik Bakker and Anna Siewierska

from ail irregular forms, all more frequently used derived and inflected forms will be readily available as entries in the mental lexicon.5 It will be assumed that such entries are underspecified in the sense that they mainly contain the extra grammatical information that distinguishes them from the root or stem, such as 3SG or PAST. Furthermore, there is a bidirectional link to the corresponding stem or root where the rest of the information of type F1-F7 may be found. For these reasons, we add another field to the lexical entry. F8. Semantic link. In the case of an irregular element of a paradigm this link leads to the main entry (root or stem). In the case of the main entry, there is a set of links leading to all related forms in the lexicon. These links may be seen as a concrete implementation of the idea of semantic fields (or its dynamic counterpart "priming") which we find e.g. in the Functional Lexematic Model in Faber and Mairal (1999). They are simply the most closely related forms in the lexical network. Somewhat less direct links relate predicates which share crucial parts of their meaning postulates, and form a semantic field.6 A second element that is not systematically represented in the standard FG lexicon is the relationship between the predicate frame proper and the meaning postulate on the one hand and the way these are formalized on the other hand. Siewierska (1993), discussing Jackendoff (1990) among others, addresses this issue for the first time within an FG context. In terms of the FG model this would mean that there may exist systematic relationships between meaning postulates, predicate frames and the morphosyntactic ways in which the predicate receives its expression. Apart from the predicate as such, this concerns those elements embedded in its meaning postulate which are actually expressed, obligatorily as arguments, or optionally as satellites. Jackendoff s meaning representations contain abstract predicates, such as TRANS, MOVE and AGENT. Dik and Hengeveld (1991) give a typological approach to the way perception verb complements are expressed. More specifically, they show that there is an iconic relationship within languages between the layer on

Adpositions, the lexicon and expression rules

141

which the second argument of the different types of perception verbs - often of a clausal nature - should be located semantically and the syntactic complexity of its expression. In general, the higher the layer the more complex the expression in that language, running from a term or a participial construction on layer 1 to a full blown finite clause on layer 4. However, the authors do not propose a way in which these facts should be represented in the lexicon of a specific language, nor how this information could be linked to the relevant expression rules. Faber and Mairal Usón (1999) and Mairal Usón and Faber (this volume), following Jackendoff (1990), Pustejovsky (1995), Van Valin and LaPolla (1997) and Rappaport and Levin (1998), postulate meta-entries (or: lexical templates) for the meaning representation of a set of semantically related predicates. One of their examples are English verbs that express diverse ways of cutting, such as cut, hew, prune and shave. The corresponding meta-entry gives a maximally generalized meaning representation for these predicates in all their aspects. The individual predicates have the defining part of the meta-entry in common - in this case the postulate of the basic predicate cut - plus that aspect of the meta-entry which is specific for the predicate concerned. As opposed to traditional FG meaning postulates, which contain formal expressions over predicates of the object language only, lexical templates also contain abstract predicates such as CAUSE and BECOME. Furthermore, they employ two types of variables: external and internal ones. The former remain implicit in the meaning postulate; only the latter are realized syntactically. In this sense, lexical templates conflate the traditional meaning postulate and the predicate frame. Postulating a Principle of Lexical Iconicity, Mairal Usón and Faber predict that there will be a direct relationship between lexical templates and their morphosyntactic expression, thereby bypassing the predicate frame as an intermediate representation platform. Unfortunately, they do not show in detail how their templates interact with the FG expression rules. Instead, they refer to the Role and Reference Grammar linking rules, which connect meaning representations with the corresponding syntactic tree structures in a bidirectional fashion (cf. Van Valin and LaPolla 1997: 317f). It is not clear to the current authors to what extent RRG link-

142

Dik Bakker and Anna Siewierska

ing rules translate into FG terms, even after the introduction of lexical templates.7 Just like Faber and Mairal (1999), Bakker (1994) reasons for abstract predicates in meaning postulates. Arguments are derived from the treatment of selection restrictions and the formulation of (generalized) predicate formation rules, where elements from meaning postulates must be matched with grammatical notions such as ANIMATE, HUMAN, DEFINITE and PERCEPTION. A third argument is derived from lexical decomposition. Apart from a restricted set of basic predicates of the language, meaning postulates might well "bottom out" in a number of abstract, cognitive notions of a more or less language independent, crosscultural nature. For the rest of this article, we will assume that there may indeed be abstract predicates in meaning postulates. In general, for their grounding, they are preferably to be found in other cognitive domains as well, although this is not necessarily always the case. As for having separate predicate frames (fields F4 to F6), this seems to be an empirical matter. Predicate frames would indeed be redundant if it were to be shown that the (optimal) meaning postulate contains all the semantic and pragmatic information necessary and sufficient to drive predicate formation, clause formation and expression rules. Thus, it should be clear which of the elements internal to a meaning postulate are actually expressed, and to what extent the selection restrictions and the semantic functions play a role in expression. The latter are necessary for the assignment of syntactic functions, for the right type of relativization (cf. Bakker and Hengeveld 1999), for case assignment and other fimction-to-form hierarchies and projections within FG theory. The external variables of Mairal Usón and Faber may be instrumental here. But since there often are multiple representations of these variables in lexical templates (and as a consequence, in meaning postulates), it is not altogether clear whether this may always be done in a straightforward way. Therefore, in the absence of optimal meaning postulates and the corresponding set of linking or projection rules, we think that predicate frames may serve as the interface between semantics and syntax, at

Adpositions, the lexicon and expression rules

143

least for the time being. As a side effect, they make explicit which lexical information is necessary for expression and which is not.8 Even if we assume that in the majority of cases expression is handled by general rules and default mechanisms, there may be cases where specific lexical idiosyncrasies determine the form of an expression. Typically, these are synchronically arbitrary aspects of subcategorization. The following are potential examples. For instance, there may be an adposition or case coupled to a verbal argument for which there seems to be no semantic basis in the synchronic sense, as in (8) for English and (9) for German. (8)

I look at you, but you don't listen to me

(9)

Ich danke dir herzlich ÍSG.NOM thank 2SG.DAT cordially Ί thank you very much.'

Another example may be the choice of an auxiliary for a verbal predicate which expresses some type of aspect, as in (10) for Dutch, or a specific type of copula for the predicative use of some adjective, as in (11) for Spanish. (10)

a. Ik heb in het bos gewandeld 1SG have in DEF wood walk.PPART Ί have walked in the wood.' b. Ik ben naar de bioscoop gegaan ÌSG be to the cinema go.PPART Ί have gone to the cinema.'

(11)

a. Está muerto COP.TEMP.3SG dead.M b. Es muerto COP.PERM.3 SG dead.M 'He is dead.'

144

Dik Bakker and Anna Siewierska

For such cases of lexical priority, just as for irregular forms in a paradigm, the lexical entry should provide the necessary instructions. Therefore, there should be yet another field in the lexical entry. F9. Unpredictable formal aspects of the subcategorization of the predicate. This includes elements such as obligatory adpositions or cases that mark a verbal argument. Also negative information may be included here, such as gaps in the paradigm for which there is no semantic basis.9 A third point that concerns us here is the location of nonpredicational lexical elements such as auxiliaries, adpositions, articles, pronouns, clitics, and of morphological elements such as derivational and inflectional affixes. The standard view in FG is that they are all located in the grammar, i.e. that they come with the expression rules that introduce them, and which in their turn are triggered by the relevant primary or auxiliary operators. Mutatis mutandis, from the perspective of the hearer grammatical elements operate as markers which trigger the corresponding expression rules in reverse order, giving rise to the respective operators and functions in the inferred underlying representation. In section 3 a more shaded treatment of the several types of grammatical elements will be suggested. It will be argued that instead of a clear cut dichotomy between full predicates and grammatical elements there is in fact a continuum. Full predicates like (English) give, tree and beautiful may be examples of the one extreme; at the other end we find suffixes like the (English) plural marker -s and tense marker -ed. Elements such as auxiliaries and adpositions, which often show clear formal and semantic traces of a verbal, nominal or adjectival origin will be found somewhere in the middle of the continuum. The dynamic model of the expression rules as sketched in section 1 provides an environment that may give support to such a position. In the next sections we will discuss the expression of adpositions from this perspective.

Adpositions, the lexicon and expression rules

145

3. From predicate to grammatical marker: The case of adpositions The dynamicity of our ER model allows us to provide two extra dimensions to the way non-predicational material is expressed. These are the order of expression (the "horizontal" dimension) and the relative position of elements in the hierarchy of the tree in which they come into existence (the "vertical" one; absent in the standard model). Here we will be concerned especially with the latter. If we follow current FG practice, then among all lexical items there is a subset of "real" predicates, which are located in the lexicon with all fields F1-F9 discussed above provided with lexical information. Before expression all relevant predicates are inserted into the UR, and therefore are available by definition at the top level of the expression tree. As such, their lexical information is maximally accessible in terms of the role it may play in the eventual form of the expression with the exception of the meaning postulates, which are supposed to be non-transparent for the expression rules. The availability of this lexical information is necessary for all kinds of side effects, such as the introduction of classifiers, the choice of the corresponding articles, gender agreement coded on other constituents, the selection of the right subcategorization pattern, etcetera.10 At the bottom end of the tree, i.e. at the terminal nodes, we find bound grammatical elements, such as affixes, which have no lexical meaning of their own, and which mark e.g. plurality on a noun or past tense on a verb. They are inserted right before their expression, which gives them minimum accessibility for other stages of expression. Often their form is completely determined by the context. If they have any influence at all on the formal aspects of the expression it is typically restricted to their immediate phonological surroundings. We will follow the standard model here, and assume that such elements are not located in the lexicon, but that they "come with the rule". In our terms: they are to be found in the corresponding template slots.11 This leaves us with a certain amount of space for traditionally non-predicational lexical elements to exert influence on the shape of

146

Dik Bakker and Anna Siewierska

the overall expression. Indeed, we will assume that there is a direct correspondence between the level on which lexical material is introduced via the expression rules and the role it may play in morphosyntax. As to the location of this material, our position will be that such elements are to be found in the lexicon as well, however with a reduced set of fields F1-F9. By definition, the presence or absence of an accomplished meaning postulate will determine for us whether a lexical element is accessible for the process of UR formation, and may appear directly in URs or only indirectly via operators or functions. Arguably, the loss of certain aspects of its meaning might be the first step of some lexical element on the grammaticalization path. However, such an element does not lose its semantics altogether in one single step. Crucial semantic information from the meaning postulate of the original predicate may be transformed into subcategorial information. E.g. the Future Aspect implied by the meaning of the verb go may be transformed into the subcategory FUT in the F3 field of the English future tense auxiliary go. Further reduction of the meaning postulate will mark further grammaticalization of the corresponding element. However, an element will loose its place in the lexicon only when it has no independent syntactic status anymore and has either a completely fixed position in the sentence or has become part of the morphology of the language. It will end up being an integrated part of the grammar then, i.e. introduced via a specific node. In this way, the place and role of a non-predicational element in the expression process is a more or less direct reflection of its diachronic history. For the same reason, it is the expression process that will shape and determine the future position of any lexical element of the language. We think that adpositions are a good example of such a diachronic process. Their status in the grammar is probably the most disputed of the parts of speech. Traditionally, adpositions are grouped with the closed classes of "minor" elements, such as subordinators and particles. Standard FG theory follows this tradition, under the assumption that adpositions have no semantics of their own. Therefore they are denied predicational status, and are exclusively located in the expression component. Within more syntactically oriented frameworks,

Adpositions, the lexicon and expression rules

147

which concentrate on the formal aspects of elements rather than their semantic ones, adpositions are often assumed to be a lexical category, or indeed a predicate, alongside V, Ν and A. The typical (often implicit) argument for this position is that adpositions function as syntactic heads of adpositional phrases and that they may assign case (cf. Bresnan (ed.) 1982: 302 for Lexical Functional Grammar; Gazdar et al. 1985: 13 If for Generalized Phrase Structure Grammar; van Riemsdijk and Williams 1986: 43 for Generative Grammar; Pollard and Sag 1994: 22 for Head-driven Phrase Structure Grammar). Interestingly, Van Valin and La Polla (1997: 52f; 376f) follow Bresnan (1982) when they develop a syntactic theory that matches the underlying semantic representations of Role and Reference Grammar. They distinguish between two types of adpositions. On the one hand, there are predicative adpositions, which have an argument position of their own, and therefore license a noun phrase. Ulis is the case for the English preposition in in example (12) below. On the other hand, there are non-predicative adpositions, which mark an argument of some predicate, and which do not themselves license the noun phrase that fills that position, such as to in (12). (12)

Kim gave the book to Sandy in the library.

Note that many prepositions of English may be found in both functions; the two uses of at in (13) are a case in point:12 (13)

I got mad at Max at the station.

However, the category of adpositions is not undisputed within FG either. Mackenzie (1992) argues that the great majority of the adpositions in English should in fact be given predicate status, since they have a clear meaning. On these grounds they should be located in the lexicon, and therefore be present in underlying representations, rather than be introduced via expression rules. Only a restricted subset, viz. at, from, via, to and toward(s) have lost their predicate status and have grammaticalized. They are the direct expression of a semantic function, Location, Source, Path, Allative and Approach, respec-

148

Dik Bakker and Arnia Siewierska

tively. Therefore they are part of the expression rules much in the way that the English suffix -s expresses the plural term operator PL on the nominal head of a term. On the other hand, prepositions such as outside should be seen as two-place relational predicates, much as the noun father, and should be interpreted as the (semantic) head of the (locative) term that they form part of.13 Samuelsdorff (1998) argues against a "grammatikon", i.e. a separate set of elements independent from the lexicon. He suggests that at least part of the grammatical elements, notably pronouns, adpositions and adverbs, should be located in the lexicon itself. Extending Mackenzie's proposal for English prepositions, he suggests a grammaticalization cline, where some elements of these three word classes are in the lexicon and others are in the grammar. For Samuelsdorff, in order to be in the lexicon a full meaning postulate does not seem to be a requirement. Elements may appear in underlying representations as second restrictors only and never as first restrictors; this is assumed to be the case for the preposition outside. They may even appear in an abstract form. E.g. under is represented as a term operator SUBES SIVE but is still seen as part of the lexicon. It is only for the subset of fully grammaticalized prepositions, corresponding to semantic functions such as the five proposed by Mackenzie for English that a purely expression rule status is claimed. Interestingly, not many authors, within and outside FG mention the fact that adpositions are not at all a universal category; around 20% of the worlds languages do not possess them, mainly from Australia, the Americas and the Indo-Pacific area.14 Such languages often have elements or constructions with the same functionality as adpositions, such as verbal particles, relational nouns and serial verb constructions. See examples (14), (15) and (16), respectively: (14)

Wardaman (Non-Pama Nyungan; Merlan 1994) wu-dab-(b)a werr-ma 0-bu-ndi PFX-cave-ABL out-SFX 3SG-AUX-PAST 'He came out of the cave.'

Adpositions, the lexicon and expression rules

(15)

Maricopa (Hokan; Gordon 1986) h 'a-sh

ha

kw 'ur-ii

ν 'aw-m

tree-SBJ water side-at 'The tree is by the water.' (16)

149

stand-REAL

Yoruba (Benue-Kwa; Lord 1993) mo

mú

ìwé

wá

ilé

1SG take book come home Ί brought a book home.' Potentially but not necessarily such languages may be in a preadpositional state. Eventually, relational nouns, serial verbs and particles may develop into adpositions. At the other end of the cline we find languages lacking adpositions but displaying many case markers which arguably have developed from adpositions. These languages may be seen as postadpositional. See example (17). (17)

Kayardild (Australian; Evans 1995) dathin-ki

mijil-I

mutha-wuru

malji-wuru

that-LOC net-LOC many-PROP hole-PROP 'Those nets have many holes.' Interestingly, it seems to be mainly postpositions that develop into case suffixes. Prepositions, if they grammaticalize further at all, generally develop into verbal suffixes (cf. Reh 1986, and section 7 below). Lehmann (1982: 86) mentions Mangarayi as a rare case of a language with case prefixes. Although there may be a functional division of labor between categories such as relational nouns, serial verbs, adpositions and case affixes, this is by no means a straightforward, balanced affair. The overall picture is definitely a complex one, as the languages in Table 1 show.

150

Dik Bakker and Arma Siewierska

Table 1. Numerical distribution of different functional elements Language

Adpositions

Cases

Lezgian (Dagestanian) Pipil (Aztecan) Tzutujil (Mayan) Kayardild (Australian) Cantonese (Chinese) Kanurí (Saharan) Tarascan (Chibchan)

18

18

8 (+3 Spanish) 2 -

-

Serial Verbs (converbs)

>36

2

> 12

-

>25

-

>20

-

>22

-

Yes

>20 >20

13 (+7 verbal)

Relational Nouns

TOTAL

-

-

2

(many converbs)

8

-

-

-

8

2

6

-

-

8

There are two things which are striking about Table 1. Firstly, there is a big difference between the languages in the total number of elements belonging to the four relational categories. This strongly suggests that, as far as this type of relationship is concerned our list is not exhaustive. And secondly, the majority of the languages prefer to concentrate on one category rather than distribute the functionality over several categories. We might tentatively try and order the above languages on a pre- to post-adpositional scale. This is done in (18) below. Note that the diachronic process is circular, and that we could have put Kayardild at the left hand side as well. (18)

Cantonese - Tzutujil - Pipil - Kanurí - Lezgian - Tarascan Kayardild

An extreme case of adpositions versus relational nouns, brought to our attention by Peter van Baarle is presented by Arawak, a Carib language from the Guyanas. This language has around 60 postpositions, and 5 suffixes with adpositional function. Moreover there is a

Adpositions, the lexicon and expression rules

151

large number of relational nouns, some clearly related to one or other of the postpositions (Peter van Baarle p.c.). Taking all the above into consideration, we will now see how adpositions might be treated by our expression rule model. We will first look, in section 4, at the birth of a preposition. Sections 5 and 6 discuss two stages of prepositions, a "young" and an "old" one. Finally, in section 7, we will look at a bound case marker that has developed out of a preposition. We do not claim that ours is a complete survey in any way. Given the perspective we chose of adpositions as a more or less distinctive stage on a diachronic cline from independent predicate to grammatical marker, there will undoubtedly be all kinds of variants and intermediate versions in the languages of the world. We think, however, that our approach gives a dynamic dimension to adpositions, and therefore to the notion of "predicate", both from a synchronic and a diachronic perspective.

4. The birth of a preposition Often, adpositions originate from nouns. Sometimes we may observe this process taking place in a language; we witness an adposition in statu ηascendi. A possible example is the Dutch noun richting 'direction'. Compare the following examples. (19)

a. b.

Gaat deze tram in de richting van het station? go-3SG PROX tram in DEF direction of DEF station Gaat deze tram richting (het) station? go-3SG PROX tram direction (DEF) station 'Does this tram go into the direction of the station?'

152

(20)

Dik Bakker and Anna Siewierska

De Consumentenbond start een campagne DEF consumer's organization start.3SG INDEF campaign richting de politiek. direction DEF politics 'The Consumer's organization starts a campaign directed to the political institutions.'

In (19b) we find a contraction of the double PP construction used in (19a) which is often made by speakers in the context of a literal destination, such as a geographic name or a well-known building. The motivation for this partial "holophrase" may well be the fact that we are dealing with a frequent combination which is hardly ever modified in any way. The normally obligatory (definite) article of the head noun may be dropped. In (20) we see a more recent development, frequently found in the context of a political discussion. The use of the contracted construction of (19b) is extended to abstract and sometimes also to indefinite referents. The original construction in (19a) remains fully acceptable, and is used in literal contexts, but less so in contexts such as (20). In the latter case, the contracted construction seems to have become the preferred one, with or without the expression of the article of the dependent noun phrase. Arguably, what we are witnessing here the introduction of a new preposition into Dutch. Quite probably the contraction in (19b) was the precursor of the construction in (20). Apart from the fact that there are some constraints still on the use of "richting" as in (20), it is hard to distinguish from the more or less equi-íunctional circumposition in (21a) and the preposition in (21b). (21)

a. De Consumentenbond start een campagne naar de politiek toe, b. De Consumentenbond start een campagne iegens de politiek

In eight steps we will sketch a (very tentative) scenario for the grammaticalization of richting. These steps are not necessarily minimal steps in the sense of Bakker (1998). All kinds of intermedi-

Adpositions, the lexicon and expression rules

153

ate stages are possible, and even highly likely, such as phonological reduction, subtle shifts in the semantics of predicates, and different contextual restrictions. Also diverse, shifting levels of optionality are possible for the respective alternative constructions. We start from the situation in which only constructions like the one in (19a) are possible. There is a more or less standard lexical entry for the nominal predicate richting, with the corresponding meaning postulate, abbreviated as MP]. A directional phrase built around the predicate richting such as the one in (22) below has the preposition in 'in' as an obligatory element. (22)

in de richting van Haarlem 'in the direction of Haarlem'

Synchronically, there is no particular (semantic) reason why this should be the preferred preposition. Thus, this fact has to be coded in the F9 field of the lexical entry as a specific subcategorial aspect of the predicate concerned. In the representations below we will use labeled bracket notation rather than (partial) trees as in Figure 1 to represent specific details of subcategorization. Furthermore, the underlying representation of the phrase in (22) has a Possessive term for the restrictor of the head noun. By default, this Possessive is expressed by way of the preposition van. This gives us the following representation for the initial stage. The F9 field is in curly brackets. We follow Mackenzie (1987) in not having a (zero) argument position traditionally claimed for nouns. Finally, we only specify the elements relevant to the current discussion; various details have been omitted.

154

Dik Bakker and Anna Siewierska

STAGEI EXPRESSION 'in de richting van Haarlem' UR= (DEF SG χι : 'richting' [N]: (DEF SG x2: 'Haarlem')poss)DIR LEX.ENTRYi= 'richting', [Ν, non-neuter],..., MP h {Lab: term Config: (VAR:'richting')DIR SubCat: [TERM [PREP 'in' [P]]]} So, whenever during expression richting is found to be the head of a term with the semantic function Direction, then a subcategorial frame will be inserted which contains the preposition in as its syntactic head. The frame as specified in SubCat is incomplete. It will be unified with the standard term frame that will be provided by the expression rules. A first change takes place when the preposition van looses its semantic aspects in this construction and is reanalyzed as the mere expression of the relationship between richting and its object. This leads to a slightly altered lexical entry, with van as a marker of the object of the head noun. The relevant changes are printed in bold. Note that there is a constraint now on the nature of the object in Config, since this change in the nature of the preposition takes place only in the case of objects that represent a location.16 STAGE II 'in de richting van Haarlem' UR= (DEF SG xi: 'richting' [N]: (DEF SG x2: 'Haarlem')^ )DIR LEX.ENTRYI= 'richting', [ Ν , non-neuter],..., M P I , {Lab: term Config: (DEF SG VARI: 'richting': (VAR2: PRED [])REF )DIR SubCat: [TERM [PREP 'in' [P]] [TERM [PREP 'van* [P]] VAR2]} EXPRESSION=

Adpositions, the lexicon and expression rules

155

Note that this is intended to code an alternative expression for the nominal predicate richting. For cases that do not fit into the Config specified above the normal expression rules will apply. In a following stage, the preposition van, which expresses a more or less "empty" semantic relation may be left out of the expression. This is only allowed when the object of richting is a proper name, in which case the preposition and the head are the only elements of the term phrase. A second restriction is that the head term is definite and singular. Since the semantic function REF has no standard expression in Dutch, we assume that it will be expressed by zero in case there is no explicit form mentioned. STAGE III E X P R E S S I O N ^ 'in de richting Haarlem' UR= (DEF SG xi: 'richting' [N]: (DEF SG x2: 'Haarlem')^ )DIR LEX.ENTRYI= 'richting', [Ν , non-neuter],..., M P I , {Lab: term Config: (DEF SG VARI: 'richting': (def VAR2: PRED [ P N < L O C > ] ) R E F )DIR SubCat: [TERM [PREP 'in' [P]] [TERM [PN PRED]]]}

The structure of Stage II remains viable for all other configurations of richting. By this time the sequence of preposition and definite article in de, which is almost completely predictable in this context, has started to weaken phonologically, and is then optionally left out. The underlying representation is not affected, but the subcategorization schema is. Note that the major difference between Stage III and Stage IV is the presence of the logical NOT operator. This operator prevents the expression rules from inserting a (default) preposition to express the DIR function and the otherwise obligatory definite article.

156

Dik Bakker and Anna Siewierska

S T A G E

IV

EXPRESSION 'richting Haarlem' UR= (DEF SG XI: 'richting' [N]: (DEF SG x2: 'Haarlem')^ )DIR LEX.ENTRYI= 'richting', [Ν, non-neuter],..., MP,, {Lab: term Config: (DEF SG VARI: 'richting': (DEF VAR2: PRED [PN < L O C > ] ) R E F )DIR SubCat: [TERM N O T [ ] N O T [ A R T ] [TERM [PN PRED]]]} P R E P

This suppression of grammatical material is by no means a general rule of the synchronic grammar of Dutch. It might, however, turn out to be a diachronic rule. The use of this prepositionless construction now spreads to other types of directional objects, which are not rendered by a proper noun but do indicate a unique location. See example (23). (23)

richting het station, het museum, de dierentuin, het vliegveld direction the station, the museum, the zoo, the airport 'in the direction of the station, the museum, the zoo, the airport'

The constraint on the Configuration is relaxed correspondingly. In the expression below, *OR* is the logical disjunction operator. S T A G E

V

EXPRESSION= 'richting het station' UR= (DEF SG xi : 'richting' [N]: (DEF SG x2: 'station')«^ )DIR LEX.ENTRYi= 'richting', [Ν, non-neuter],..., MPi, {Lab: term Config: (DEF SG VARI: 'richting': (def VAR2: PRED [ P N * O R * Ν ] ) R E F )DIR

SubCat:

[TERM NOT PRED]]]}

[PREP]

NOT

[ A R T ] [TERM [PN/N

Arguably, the new more abstract meaning may make the modified selection restriction superfluous. Furthermore, the gender information may be left out of the description. It is no longer relevant since there is no article. Also, other modifiers which are sensitive to gender information, such as adjectives and relative clauses, are forbidden at

158 Dik Bakker and Anna Siewierska this stage on semantic grounds. The complex subcategorial expression, which in Stage VI codes an obligatory rather than an optional dependent category, may best be represented now by an argument position. Furthermore, we follow Mackenzie (1992) in adding a Zero first argument to express the notion of relationship. The possible selection restrictions derive from the modified meaning postulate. What we are left with then is in fact a sort of relational noun. It could be argued that this nominal subcategory codes precisely all the modifications that the original entry underwent during the earlier stages in the process. It may in fact be claimed that a relational noun distinguishes itself from 'regular' nouns above all by its specific (morpho)syntactic behavior. Stage VI may therefore make way for Stage VII. STAGE VII EXPRESSION 'richting de politiek' UR= (xi: 'richting' [Ν, Rei]: (DEF SG x2: 'politiek')REF )DIR LEX.ENTRY 2 = 'richting', [N, Rei], ((xjWo(XU)REF ),..., MP 2 ,... Note that explicit specification of the F9 field is no longer necessary since expression rules will take care of terms with a relational noun as their head. This seems to be the current situation in Dutch. However, the stage is set for the next move: the reanalysis of the relational noun richting as a preposition by a new generation of speakers. There are some factors that may indeed make this happen in the near future. First there is the applicability of the construction to a growing set of more general contexts, coupled to a process of "desemanticization" of the predicate concerned (cf. Lehmann 1982). In our terms that would mean that the meaning postulate will contain more abstract predicates as discussed in section 2. Furthermore, the obligatory Referent argument, the nature of the constraints on its fillers and the set of corresponding expression rules make its position in the grammar very much like - and indeed indistinguishable from - the one for established prepositions in the language. In short, we can say that the

Adpositions, the lexicon and expression rules

159

noun richting in the relevant context has grammaticalized to a large extent towards a preposition, with its relational nounship as an intermediate stage. A factor that could keep the predicate in Stage VII for some time to come would be the ongoing unrestricted use of the original noun, especially as long as the relational version keeps its full form and would not undergo phonological reduction. Maybe only when native speakers' intuitions about the relation between the "full" noun richting and its relational version disappear, because of formal differences or because of the fact that the former is no longer part of the language, may the relational noun enter the ranks of the prepositions. This would lead to Stage VIII. Note that, like the previous stages, there is no reason not to locate the entry in the lexicon, since it has several defining aspects of a predicate, such as a meaning postulate and a predicate frame, and selection restrictions. STAGE VIII EXPRESSION 'richting de politiek' UR= (xi: 'richting' [P]: (DEF SG x2: 'politick')^ )DIR LEX.ENTRY2= 'richting', [P], ((xk)ZERO(xiOref),..., MP2 , ... A final point that needs to be made in this section concerns the status of constructions based on this predicate. The input construction at Stage I had the nominal predicate as the head of a satellite or a term restrictor. It seems to be counterintuitive that, as a result of this long grammaticalization process, it would be promoted to the head of the relational construction, as in example (26a) below. Rather, we assume a dependent position for the construction, with a variable for the filler of the first argument position, and the construction in the role of a nominal (or predicational) restrictor, as in (26b). Only then do we preserve iconicity with the way the construction is expressed, as in (25), i.e. with the referent of the zero argument variable as the head, not the relational noun/preposition.

160

Dik Bakker and Anna Siewierska

(25)

een campagne richting de politiek INDEF campaign direction DEF politics 'a campaign directed to the political institutions'

(26)

a.

(xi: 'richting' [P]: (INDEF SG x2: 'campagne')ZERO (DEF SGX 3 : 'politiek')REF)

b. (INDEF SG xi: campagne [Ν]: (χι: 'richting' [P] (x2: X O Z E R O (DEF SG x3 : 'politiek')REF )ZERO)

5. Prepositions: The early years In this section we will have a look at the lexical representation and the expression of a preposition that has featured in the English language for some time, viz. above, as in example (27). (27)

Harpo is flying above the circus.

Again following Mackenzie (1992: 15), and the classification of English prepositions in Meijer (1995), we assume that above has a complex set of semantic features, procuring it a full blown entry in the lexicon. The semantics of the meaning postulate may be rendered by a formal expression over several abstract locational predicates. Below, an attempt is made at providing such a meaning postulate. The abstract predicates are represented in capital letters. LEX.ENTRY=

'above', [P], ( (XOZERO (XOZERO

(xj)ooAL:

),···, [SUPERIOR (VERTICAL)DIR & NOT (XJ)REF

(CONTACT ( X I ) z E R O ( X j ) G O A L ) ] UR= [[[PRES ei: [IMPF 'fly' [V] (xj: 'Harpo' [PN])A0]: (pi: 'above' [P] (ei)ZERo (DEF SG x2: 'circus' [N])REF)LOC ] ] ]

The expression of the prepositional constituent will take place in a more or less straightforward way, since all important lexical material

Adpositions, the lexicon and expression rules

161

is present in the underlying representation. The most relevant part of it is given below. Node k specifies a configuration with a prepositional predicate as the head. Node k+1 deals with the expression of this predicate. NODE k (uninstantiated) Lab: PP Config: [VAR FORM [P]] SubCat: NODE k (partially instantiated) Lab: PP Config: [(pi: 'above' [P] (ei) ZE Ro (DEF SG x2: 'circus' SubCat: NODE k (fully instantiated) Lab: PP Config: [(pi: 'above' [P] (EOZERO (DEF SG x2: 'circus' SubCat: PREP, TERM

[N])REF)LOC]

[N])RE F )LOC]

NODE k+1 (uninstantiated) Lab: PREP Config: [FORM [P]] SubCat: FORM NODE k+1 (fully instantiated) Lab: PREP Config: ['above' [P]] SubCat: 'above' With their complex semantics, these adpositions should be considered as a predicate. Therefore they can head a copula construction without the application of any predicate formation rule, as in (28).

162

(28)

Dik Bakker and Anna Siewierska

The sun is above the horizon. U R = [ [ [ P R E S ei: ['above' [P] ( D E F (DEF SGx 2 : 'horizon' [N])ref)]]]]

SG

xi: 'sun' [N])zero

6. Prepositions: The later years Over time, prepositions may grammaticalize further. This is noticeable in two ways. Their use may spread to more contexts while their meaning postulate may become more and more abstract, thus stretching the selection restrictions. Typically, this is accompanied by a simplification of the form. "Old" prepositions like English at, from and to are monosyllabic, and have a wide variety of application, far beyond the original locational domain. Their meaning may be rendered by a simple abstract predicate rather than a complex expression as the one for above. For the three prepositions mentioned this may be LOCATION, SOURCE and ALLATIVE. These correspond in a more or less one to one fashion to the three semantic functions Loc, So and All, which we find expressed by one marker in all kinds of genetically and areally unrelated languages. That an adposition grammaticalizes to become the expressor of a semantic function in itself is not strange at all. Adpositions start out as two-place predicates, relating their two arguments, typically two terms or a term and a predication. A semantic function does precisely the same. The grammaticalization in question is on the dependent (the term to which it originally belongs, and which is expressed as a PP) rather than on the head (the other term or the predication). We will assume that a preposition has fully grammaticalized, and has lost its predicate status when its meaning postulate consists of precisely one abstract predicate which corresponds to a typologically established function or operator in the grammar. We may then say that it expresses precisely that function or that operator. In URs we will find it in the form of the corresponding function or operator, not as a predicate like the "younger" prepositions. The expression rules will select it from the lexicon, where we would still prefer to locate it

Adpositions, the lexicon and expression rules

163

as long as it plays a role in the syntax of the language. An argument supporting the above is that in many languages prepositions assign case to other elements of the phrase, or are themselves the target of agreement phenomena. And from the perspective of semantics, once we have allowed abstract elements to appear in the meaning postulate of "real" predicates, as proposed above, we see no principled reasons why prepositions of the category of at, from and to would not also be located in the lexicon. Only when linguistic elements are exclusively dealt with by the morphology, as in the case of affixes, should they become part of the word formation system and "disappear" from the lexicon. Example (29) below contains the "old" preposition at. In the corresponding UR it is represented as the LOC function on the second term. (29)

Groucho works at the circus UR= [[[PRES ei: [work [V] (DEF SG xi: 'Groucho' [PN])ag]: (DEF S G x 2 : 'circus' [N])loc ]]]

For the preposition at we propose the following lexical entry. Note that the only element of the meaning postulate is the abstract predicate LOC (for LOCATION), corresponding to the semantic function LOC in the grammar. The fact that the meaning postulate is a single abstract predicate prevents the lexical element from appearing as such in an UR.17 LEX.ENTRY=

' a t ' , [P], ((xOzero (Χ])κερ), ... , [LOC (xí)zero (xj)ref]

When expressing the UR of (28) we will get at the stage where the locative term should be expressed. This may go as follows. Below, Node η is an alternative to Node k in section 5 for the formation of a Prepositional Phrase in English. In its uninstantiated form in the grammar it specifies a nominal head instead of a prepositional one. On top of this, it contains an exhaustive list of semantic functions that are expressed via a preposition.

164

Dik Bakker and Anna Siewierska

NODE η (uninstantiated) Lab: PP Config: [VARFORM [N] SEMF={LOC, SO, ALL,...}] FncFtrs: SEMF SubCat: NODE η (partially instantiated) Lab: PP Config: [(DEF SG x2 : 'circus' [N])LOC] FncFtrs: LOC SubCat: NODE η (fully instantiated) Lab: PP Config: [(pi: 'above' [P] (EOZERO (DEF SG x2: 'circus' FncFtrs: LOC SubCat: PREP, TERM

[N])REF)LOC]

NODE n+1 (uninstantiated) Lab: PREP Config: [FORM [P] FUNC] FncFtrs: FUNC SubCat: FORM NODE n+1 (partially instantiated) Lab: PREP Config: [FORM [P] LOC] FncFtrs: LOC SubCat: FORM This constellation - a partially instantiated node with a lexical entry in its Configuration field - will be interpreted by the expression rules as an instruction to search the lexicon for the corresponding form. The Config unifies with the entry for at as specified above. This will lead to the following fully instantiated node.

A (¡positions, the lexicon and expression rules

165

NODE n+1 (fully instantiated) Lab: PREP Config: ['at' [P] LOC] FncFtrs: LOC SubCat: 'at' The rest of the expression takes place according to the general rules outlined in section 1. 7. From preposition to affix: The end of the affair Finally we will have a look further down the scale: a preposition disappearing into the morphology. The example we use is from Reh (1986). She presents data from three related Nilo-Saharan languages on the use of the Dative marker m. In Alur (Ringe 1948 as cited by Reh) this marker has the form of a preposition, and marks Recipients, provided that they follow the Patient (this language seems to have Dative Shift): (30)

Tornasi o-miyo bongo m Olum T. PRF-give clothes to O. 'Tom gave clothes to Olu.'

In Dholuo (Stafford 1967 as cited by Reh) we find a clear cognate to the Alur marker. However, in this language it has spread to all contexts. Furthermore, when used with pronouns, it is suffixed to the preceding verb. Cf. (31)

Otieno o-kelo

m

Odhiambo

kitabu

O. PRF-bring DAT O. book 'Otieno has brought a book to Odhiambo.'

166

Dik Bakker and Anna Siewierska

But: (32)

O-kelo-n-a kitabu 3 SG-bring-DAT-1SG book 'S/he brings me a book.'

Finally, in Lango (Noonan 1981 as cited by Reh) the corresponding marker has integrated completely into the verbal morphology. It has become a Benefactive marker which obligatorily appears as a suffix on the verb in the presence of a Beneficiary or Recipient. Cf. (33)

a. o-kelo 3SG.PRF-bring 'S/he brought it.' b.

o-kel-li

dako

3SG.PRF-bring-BEN woman 'S/he brought it for the woman.' The cases of Alur (30) and Dholuo (31) should probably be handled like English at in section 6. The complex case of (32) will not be discussed here since more should be known about the verbal morphology and the pronominal system of Dholuo, which is outside the scope of the present article. The Lango case of (33) may be handled in two ways. A first option is that the clause level top node is provided with a functional feature BEN marking the presence/absence of a Benificiary or Recipient term. This feature codes the grammaticalization of the Benefactory function at the clause level. It is inherited by the node that expresses the finite verb. During this expression the marker -m is suffixed to the verb in case the BEN feature has value present. In contrast to Alur and Dholuo, the marker is not retrieved from the lexicon but is an integral part of an expression rule at the morphological level. Cf.

Adpositions, the lexicon and expression rules

167

NODE ρ (uninstantiated) Lab: VF Config: [FORM [V]] FncFtrs: BEN SubCat: NODE ρ (partially instantiated) Lab: VF Config: ['kel' [V]] FncFtrs: yes SubCat: NODE ρ (fully instantiated) Lab: VF Config: ['kel' [V]] FncFtrs: yes SubCat: V, BENSFX NODE p+i (fully instantiated) Lab: BENSFX FncFtrs: yes SubCat: -m In this case we can say that the original lexical marker m has grammaticalized in the form of a morphological rule. Finally, after the introduction of the suffix phonological rules assimilate the initial nasal of the marker to the final consonant of the verb stem, leading to the form in (33b).18 A second option for the Lango case is lexical and operates via a predicate formation rule. This rule is applicable to a restricted set of verbs and extends the predicate frame of these verbs with a Ree or Ben argument. The final verb form is created by the same type of expression rules as those necessary for the first solution. This time the rules work on the lexicon, possibly in the form of a redundancy rule. A more thorough study of the Lango verbal system, and typological evidence for a clausal BEN marker are necessary ingredients for a right choice in this case. It seems, however, that our expression

168

Dik Bakker and Arma Siewierska

rules can handle both cases, whatever option seems to be the best one. 8. Conclusions; remaining points In our contribution we have discussed the status of adpositions in an FG grammar. We have suggested that there may be several different types of adpositions, ranging from elements that are just about to lose their status as relational nouns via full predicational adpositions and long established and almost fully grammaticalized adpositions that are inserted by the expression rules to erstwhile prepositions now turned into case suffixes. All adpositional elements, we suggested, have their own entry in the lexicon. Their predicational status - in other words: whether they may appear as such in underlying representations - depends on the form of their meaning postulate. If the latter consists of precisely one abstract predicate then it is assumed that the lexical element does not have predicate status while remaining part of the lexicon. In all other cases adpositions should be considered as predicates. We may only hope that the above discussion has shed some new light on the notion of predicate in FG in its relation to the lexicon, underlying representations and expression, and more specifically on the position of adpositions. However, a number of points concerning the latter have not yet been touched upon. We mention just a few. A first point is the assignment of case by non-predicational adpositions. It is obvious that the necessary case information, which is stored in the lexical entry of the adposition concerned has to be available once the elements within the PP which are sensitive to case are to be expressed. In terms of our model of the expression rules this means that the adposition in question has to be retrieved in time. For prepositions that may be the moment in which they are expressed, since the case information which becomes available after their retrieval may percolate upwards to the PP node and from there to all material that follows.19 However, for postpositions that would be too late. Our expectation is that there are no postpositions of the non-

Adpositions, the lexicon and expression rules

169

predicational type that assign case, or at least not unpredictable case, e.g. other than the accusative. If this turns out to be incorrect then the expression rules should be adapted to enable the relevant adposition to be retrieved at an earlier stage, probably at the level of the PP node itself. All types of adposition could then be seen as the (syntactic) head of the PP. A second point that should be mentioned here is the fact that certain adpositions may be modified, by adverbial expressions, adverbs and other adpositions, the latter sometimes fused into one new adposition. The examples in (34) below from English are a case in point.20 (34)

a. The truck stopped three centimeters before the edge. b. It happened rieht at the corner. c. I walked into the garden.

For predicational adpositions such as before in (34a) the underlying representation does not seem to be problematic; like any other predicate they may be restricted via the : operator. However, for nonpredicational adpositions such as at in (34b) there seems to be a representational problem. Related to this, especially from the perspective of the diachrony of adpositions and the development of case markers out of them are examples of multiple case marking as in the example below from Huallaga Quechua in (35), taken from Weber (1989: 227), where an accusative and a commitative are suffixed to the same noun: (35)

Hwan-ta-wan Pablu-ta rika-shka-:. Juan-ACC-COM Pablo-ACC see-PRF-1 Ί saw Paul with John.'

A third point that deserves attention is the case of languages which do not have adpositions at all but are about to introduce them. This may be on the basis of serial verb constructions or relational nouns. Or it may be via language contact, as in the case of the Amerindian language Otomi from Mexico which lacks adpositions as a category

170

Dik Bakker and Anna Siewierska

but nevertheless imported quite a few prepositions from Spanish (Bakker and Hekking 1999). Finally we mention the case of adpositions that are coupled to an argument position of a verb, already introduced briefly in example (8), repeated here for convenience. (8)

I look at you, but you don't listen to me.

Our type of expression rules seem quite capable of correctly positioning such unpredictable elements in the expression, via the representations in the lexicon. However, some problems as to their status remain, one of them being passivization. In English these prepositions seem indeed to be mere markers of the Goal function, given that passivization is possible, as is shown in (36). However, this does not hold for Dutch as shown in (37). (36)

a. I look at you. b. You are looked at (by me).

(37)

a. Ik kijk naar jullie. b. Jullie worden naar gekeken (door mij).

However, these problems have to wait for a later discussion. Abbreviations 1 2 3 A ABL ACC BEN COM COP DAT DECL

1st person 2nd person 3"1 person Adjective (predicate) Ablative (case) Accusative (case) Benefactive (case) Commitative (case) Copula Dative (case) Declarative

Adpositions, the lexicon and expression rules DEF INDEF LOC M Ν NOM Ρ PAST PERM PFX PL PPART PRES PRF PROG PROP PROX REAL REM SBJ SFX SG TEMP V

171

Definite Indefinite Locative (case) Masculine Noun (predicate) Nominative (case) Preposition (predicate) Past Permanent Prefix Plural Past Participle Present Perfective Progressive Proprietive (case) Proximate Realis Remote Subject Suffix Singular Temporal Verb (predicate)

Notes 1. Of course, we may take an extreme functional position here in the sense of Croft (1995), where any space for autonomous operations is denied to syntax altogether. It might turn out to be possible then in the grammar of individual languages to determine the combination of all underlying factors which eventually lead to certain order patterns. These could be made available as μ operators to the Phase I rule that determines the form of the finite verb. Unless we see the form of the agreement marker as the expression of this conglomerate of factors, i.e. not only of person/number aspects of the Subject, but also as information about its pragmatic status, this would indeed boil down to prerunning the expression rules in a more or less disguised and sophisticated way. 2. Technically speaking, this is by far not the only way the strategy envisaged here may be implemented. However, in this form we follow models in the psy-

172

3.

4.

5.

6.

7.

Dik Bakker and Arma Siewierska

cholinguistic literature, such as Kempen and Hoenkamp (1987) and Levelt (1989), thereby adding to the cognitive adequacy of the FG model. For a fully worked out implementation of the dynamic model a feature-value notation will be assumed including a set of logical operators and a unification mechanism. Such a formal system is necessary to make the representations explicit and unambiguous, and to define the operations on them in a straightforward way. A proposal for such a system may be found in Bakker (1994: 263296). We will stick to a more informal version here. When given in the form of an implication as in (i) below rather than an equivalence as in (ii) meaning postulates are typically semantically incomplete as well (example from Dik 1997: 97f). (i) bachelor (x) -> not (married (x)) (ii) bachelor (x) ·

Structure)

ni,[[Φν] Serlsufln

Deverbal nouns ι[[Φη] ^erJsufln Denominai nouns m niι[[Φλ4)1 #er]sudn Deadjectival nouns

OUTPUT CONDinONS Allomorphies

1) Full development of epenthetic pre-suffix [r]. Iff: base word ends in -ure [(u)a(r>]. E.g. manufacturer, insurer,

[L_(u)a (r) ]B«e θ(Γ)]]η

treasurer.

L_(u)9ra (r) ]n

222

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero Truncations 1) Apocope of base final [i]. Iff: base word ends in -ology, -onomy, -ography, -osophy. E.g. astrologer, astronomer, cartographer, philosopher.

0l9d39(r)]]n

LotedaiW^l,

[

[[ [[

L_os3fe(r)]]n

[[

onSmijBast os8ft]Base9(r)]]n (r)

0grafi]Ba«9 ]]o

Suprasegmental

->

[

ografe(r)]]n

Vacuous

The first block of information codified in the affixal entry refers to the graphemic and morphophonological behavior of the affix. Thus, under the heading Spelling alternants there is a description of the graphic alternation of -er with other two forms -or and -ar that have no phonological impact. The Phonological description of the affix is done following the distinction - already established in the generative tradition (Chomsky and Halle 1968; Siegel 1979; Aronoff 1979; Booij 1977) - between at least two types of affixes, with different phonological boundaries and behaviors: Class I (with boundary +) and Class 2 (boundary #). This distinction has explanatory power since it is postulated that Class I is pre-phonological, that is, it takes place before some processes like stress assignment, syllable distribution and truncations. With regard to -er its Affixal type (#x##) is the typical of Class II affixes, that is, the attachment of -er to the base word does not affect the original syllabic structure of the base, nor does it cause any stress displacement. This is represented under the format of these rules: Stress Location

[ 2 a( r )] ! f' s o] Base [3a(r)]/[2So'Solease

where superscripts stand for stress distribution and "S" means "syllable". The suffix -er lacks primary stress irrespective of whether the base has one syllable or more than one.

On the syntax-semantics interface in wordformation

223

The morphophonological behavior of the affix is described in the sections Morpho-lexical operations (where a description of the internal morphological structure of the derived words is provided) and Output conditions where all morphophonological readjustments are described by means of two set of rules (Aronoff 1979): allomorphy and truncation rules. In the case of -er there is an allomorphic variation that occurs with bases ending in -ure: in these occasions it is necessary to fully develop an epenthetic [r] base final sound. Although it is not expected that there should be any case of productive truncation with Class II affixes, there are cases of non-productive variations of this type, such as the one codified in the entry for -er with denominal bases ending in -ology, -onomy, -osophy, -ography, where final [i] is lost and there is a syllabic redistribution. Such variations are codified since these rules can be considered, for non-productive cases, as mere redundancy rules.10 The second basic block of information in the entry concerns the syntactico-semantic classification of the types of base words to which the affix can be attached. This type of information is fundamental to predict the syntactico-semantic behavior of the derived word and to find the relevant distinctions among competing affixes (such as -er vs. -ant or -ist, in the case of Agent nominalizations), since it enables us to establish the lexico-semantic domains where each affix operates. For the description of verbal bases we have adopted the system proposed in Van Valin and LaPolla (1997: chapter 3): verbs are classified according to a system of lexical decomposition which groups them in terms of different Aktionsarts (Vendler 1967).11 The typology of lexical classes is arrived at by considering the inherent temporal properties of verbs, thus rendering the following classification: States Activities Accomplishments Achievements

[+static], [-telic], [-punctual] [-static], [-telic], [-punctual] [-static], [+telic], [-punctual] [-static], [+telic], [+punctual]

224

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero

These four types have a corresponding causative class; each of these eight classes of Aktionsart is represented according to the following logical structures: Verb class

Logical structure

State Activity Achievement

predicate' (x) or (x, y) do' (x, [predicate* (x) or (x, y)])

Accomplishment

BECOME predicate' (x) or (x, y), or

Causative

INGR predicate' (x) or (x, y), or INGR do' (x, [predicate' (x) or (x, y)]) BECOME do' (x, [predicate' (x) or (x, y)]) α CAUSE β, where α, β are LSs of any type

Van Valin and LaPolla (1997: 105-108) add another class type Active accomplishment - with its corresponding causative, whose logical structure is: Active accomplishment

do' (x, [predicate!' (x (y))]) & BECOME predicate2' (z, x) or (y)

With regard to nominal bases, they are described according to Pustejovsky's (1991: 426-427) classification in terms of the following Qualia properties:12 a. Constitutive role: the relation between an object and its constituents, or proper parts 1. material 2. weight 3. parts and component elements b. Formal role: that which distinguishes the object within a larger domain 1. orientation 2. magnitude 3. shape

On the syntax-semantics interface in wordformation

225

4. dimensionality 5. color 6. position c. Telle role: purpose and function of the object 1. purpose that an agent has in performing an act 2. built-in function or aim that specifies certain activities d. Agentive role: factors involved in the origin or "bringing about" of an object 1. creator 2. artifact 3. natural kind 4. causal chain The two first Qualia are useful for the semantic characterization of the nominal domains where the affix in question shows productivity, whereas the telic and agentive Qualia would be especially useful to account for the underlying structure of the derived words: as shall be seen later, these features provide the verbal nexus that binds the derived word to its semantic characterization in the Modeling Process. In the description of adjectival bases, they are treated as a lexical class parallel to nouns and verbs. RRG does not seem consistent in their representations: they are considered as NP operators when used attributively, and are part of LSs only when they have a predicative function (supported by a copular verb). From a lexical perspective - as the one defended in the FLM - it is more coherent to treat them as lexemes with their own valency and semantic features (cf. Martín Mingorance 1984b; Cortés 1986, 1988). In fact, all adjectives in our analysis are described as monovalent statai events, whose logical structure is predicate' (x).13 We now turn our attention to the last section of affixal entries: the underlying representation of the syntactico-semantic features of affixes. These formulas - described formerly as affix schémas (Cortés 1997a, 1997d) - provide a grammatical and semantic description of the words derived by means of the corresponding affix. As can be inferred, the FLM - in opposition to the valency-centered FG approach - proposes as the point of departure for the generation of

226

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero

word formation products a rich decompositional approach whose aim is to reconstruct the full gamut of operators and participants in the underlying syntactico-semantic structure of derived words. In this sense, the FLM lies closer to the RRG approach for semantic representation. The basic underlying structure for -er nouns is as follows:14 (1)

Φ'Ν : PRED (x',... [LS (χ',...)]) where LS = logical structure of any type and PRED= any type of predicate

The Φ symbol is a lexical variable that represents any constant ("word") whose lexical representation corresponds with the logical structure that follows the ":" symbol. This variable is co-indexed with the (x) argument of the logical structure to indicate that all nominalizations are "oriented" towards that argument (cf. Laca 1993); that is, the creation of -er formations involves a process of profiling one entity - the trajectory - in the semantic scenario. As Langacker (1987: 312) points out: [CLIMB] designates a process in which a trajector, through time, assumes a succession of positions on the surface of a landmark. ( . . . ) [ER] is a nominalizing predicate, profiling a thing identifiable solely by its role as trajector in a schematically characterized process; the process constitutes the base of the predicate. This schematic base process functions as the e-site (elaboration site) in the valence relation: it corresponds to the profile of [CLIMB], and is elaborated by that predicate. [ER] is the profile determinant, so the composite [CLIMB-ER] designates a thing characterized solely as the trajector of the process [CLIMB],

Within this general representation, therefore, the variable PRED would be substituted by the constants corresponding to the logical structures (LSs, henceforth) of the various possible formations. Thus, in the case of deverbal nominalizations the following formulas are different possible instantiations of the basic maximal structure in (1):

On the syntax-semantics interface in wordformation

(2) (3) (4)

(5) (6) (7) (8) (9)

227

Φ'Ν : pred' (x') or (x', y) e.g. believer Φ'Ν : do' (x\ [pred' (x') or (x', y)]) e.g. hunter, shaker Φ*Ν : DO (x\ [do' (χ1, [pred' (x¡) or (x¡, y)])])

exploiter Φ'Ν : BECOME pred' (χ1) or (x¡, y) e.g. learner Φ'Ν : INGR pred» (x¡) or (x\ y) e.g. blower Φ'Ν : INGR do' (x\ [pred* (y) or (x',y)]) e.g. beginner Φ'Ν : [do' (x\ 0)] CAUSE [BECOME pred' (y) or (x¡, y)] e.g. dryer, comforter Φ'Ν : [do' (x\ 0)] CAUSE [INGR pred* (x¡, y)] e.g. stapler

As can be inferred from the indexation with superscript1 these LSs are the representation of formations oriented towards the first argument. It must be stressed that this argument does not need to be an instigator of the state of affairs expressed by the base. In FG terms, the range of semantic functions of this argument would go from the more typical Agent trough Processed and Force to Zero. Maybe we could describe these formations as Al (First Argument) nominalizations; but still we would not have solved the problem of Instrumental or Locative formations such as the ones mentioned above. Again, the system for semantic decomposition that we have adopted turns out to be more satisfactory. The alternance between Agents or Forces (RRG's Effectors) and Instruments is a well known factor in the behavior of verbal predicates (e.g. Tom cut the bread with the knife vs. The knife cut the bread). Van Valin and LaPolla's (1997: 121-122) proposal to account for this alternation becomes valid also for the representation of instrumental -er formations. The LSs are generalized descriptions susceptible to different Modeling processes by means of which it is possible to account for a whole range of semantic alternations. Thus, starting from a logical structure such as the one

228

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero

proposed in Mairal Usón and Faber (this volume: 55) for manner-ofcutting verbs: (10)

[[do' (w, [use.sharp-edged.tool(a)in(ß)manner' (w, x)])] A [BECOME be-at* (y, x)]] CAUSE [[do* (x, [make.cut.on' (x, y)])] CAUSE [BECOME pred' (y, (z))]]

it is possible to obtain the following logical structure for instrument alternations by means of the Modeling Process: (11)

[do' (w, [use' (w, x)])] CAUSE [[do' (x, [BECOME be-at' (y, x)])] CAUSE [BECOME cut' (y)]]

Therefore, the structure of Tom cut the bread with the knife would be: (12)

[do' (Tom, [use· (Tom, knife)])] CAUSE [[do* (knife, [BECOME be-at' (knife, bread)])] CAUSE [BECOME cut' (bread)]]

Van Valin and LaPolla (1997: 121) explain this structure in the following way: The causing event in [this structure] is complex, and the INSTRUMENT argument appears three times in the logical structure: as the IMPLEMENT of use' and as the EFFECTOR of do' (x, [cut' (x, y)]). It is possible, if the first argument of the highest do' were left unspecified, to say The knife cut the bread, with the INSTRUMENT knife as actor.

In a parallel fashion, if we start from the formula: (13)

Φ'Ν ·' do' (x, [cut']) [do' (w, [use' (w, x)])] CAUSE [[do' (x, [BECOME be-at' (y, x)])] CAUSE [BECOME cut' (y)]]

On the syntax-semantics interface in wordformation

229

We can model it along the following format to account for the instrumental formation cutter. (14)

cutter^: [do' (x, [use' (x, y')])] CAUSE [[do' (y\ [BECOME be-at' (y\ z)])] CAUSE [BECOME cut' (z)]]

The general structure, then, for instruments is: (15)

Φ'Ν: [do* (x, [...])] CAUSE [[...do* (y\ [...])] CAUSE [BECOME/ESfGRpred· (...)]]

The formula in (15) can, consequently, be considered as a constructional template for instrumental formations (see Figure 1). As regards locative formations, modeling would involve the addition of a locative predicate: (16)

Φ'Ν: [be-loc' (x\ [PRED (y,... [LS (y,...)])])]

The LS of the event is treated as an entity being located in a spatial relation. The representations of, for instance, the noun diner would be: (17)

dinerV: [be-in' (x1, [do' (y, [dine* (y)])])]

There is a further structure that can be expanded from the basic LS; in this case a semantic parameter of manner15 is added: (18)

Φ'Ν: [PRED

(x\ ... [LS (x ¡ ,...)])]

Such a variable is necessary in order to establish some semantic differentiations among several derived nouns; at the same time it motivates semantically some variations in the morpho-syntactic behavior of these new lexemes, as shall be seen below. As Mairal Usón and Faber (this volume: 61) claim, this parameter "is replaceable by a property-entity indicating the manner in which

230

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero

the activity is performed". Thus manner can be further decomposed by means of the following canonical realization rule: Manner —• repeatedly, professionally The rule permits to rewrite the structure (18) as (19)

Φ'Ν: [ do' (x1, [pred' (x") or (x*, y)] )] (B2)

FUSION OF LEXICAL MATERIAL

(B2.1)

BASE WORD SELECTION

pred' = interpret' (B2.2)

AFFIX SELECTION

x' = -ist:

discard. Non-congruence with base morphology (and others), -ician: discard. Non-congruence with base morphology (and others), -ant/ent: discard. Non-congruence with template restrictions, -er: selected.

On the syntax-semantics interface in wordformation

235

CYCLE I-OUTPUT:

Φ'Ν : [ do' (-er", [interpret' (-er", y)])] CYCLE II - LINKING RULES II: ASSIGNMENT OF MORPHOSYNTAGMATIC

FUNC-

TIONS: Φ'Ν (C)

· -er1 dnennimtum ïnterpretdetcnmnant

MORPHOSYNTAGMA(MACROROLE ASSIGNMENT):

-er" determinatus (Profiled Actor)

ÍBterpretdjtenniMiit (0 MR)

CYCLE III: EXPRESSION RULES (D)

REDUCTION TO LEXICAL UNIT: (Dl)

LINEARIZATION:

Φ' Ν -> [[interpret] tterfo (D2) (E)

ADJUSTMENT RULES: NONE

OUTPUT:

[interrita]

There are occasions where formations show a more complex structure, since derived nomináis can appear modified by either a Prepositional Phrase or by a nominal Premodifier which are the expression of the Undergoer macrorole. In this respect, Actor nominalizations behave in a similar fashion to action nomináis (cf. Nunes 1993; Levi 1978) showing an ergative pattern; that is, Actor nominalizations are Macrorole Atransitive in

236

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero

most occasions since the Actor is a profiled constituent of the derived word and U-assignment does not take place. Only in the cases of bivalent bases there is the possibility of having M-intransitive (Uassignment) constructions. Van Valin and LaPolla (1997: 188) describe this phenomenon in the following terms: When transitive verbs are nominalized by this rule, there are two possible realizations of the undergoer argument. It may appear as a direct coren argument marked by of, as in a drinker of beer, a painter of houses, a hunter of ducks or a killer of cops, or it may be incorporated into the derived nominal, creating beerdrinker, house-painter, duckhunter or copkiller.

There are, however, some points to comment on this statement: firstly, there is no description of the conditions that trigger or block the apparition of the Undergoer19 in the derived constructions;20 secondly, noun incorporation and some cases of postmodification with an of phrase involve the presence of inherent arguments which, consequently, cannot be assigned macrorole status. In the remainder of this paper we turn to this issue. The most outstanding cases where it is necessary to make overt reference to the behavior of the Undergoer are connected with the activity/active accomplishment alternation of some verbs (e.g. Robert ate a dish of rabbit Active Accomplishment vs. Robert eats energy barsActivity). One of the most important features of many activity verbs is the non-referential character of the second argument: Thus the second argument with an activity verb like eat will be called an INHERENT ARGUMENT, an argument which expresses an intrinsic facet of the meaning of the verb and does not refer specifically to any participant in an event denoted by the verb; it serves to characterize the nature of the action rather than to refer to any of the participants. It is not fixed, in that it can be used to characterize a number of different types of actions expressible by a particular verb, e.g. drinking beer, drinking coffee, drinking tea, drinking milk etc. (Van Valin and LaPolla 1997: 123)

On the syntax-semantics interface in wordformation

237

Activity verbs with inherent second arguments are often the base for noun incorporation processes (e.g. cab/ taxi/ tractor/ engine/ bus/ screw driver; game/ goal/ book/ lock/ door/ shop keeper), where there is no U-assignment - given the obligatorily referential nature of this macrorole (cf. Van Valin and LaPolla 1997:149-159). When the second argument of activity verbs is non-referential the construction permits modification with the of phrase (e.g. a writer of books). It must be stressed that in this case, there is no U-assignment either and therefore the appearance of a constituent with this role is optional. The only restriction is the necessity for the complement of the preposition to be expressed in plural form in the case of count nouns. In the cases where the second argument of activity verbs is referential, U-assignment takes place and, therefore, of postmodification is compulsory. This explains, for instance, the fact that leader (78 occurrences) never appears alone in the LOB corpus. It is interesting to consider also the case of lover formations that appear in the LOB corpus. Only 8 of its 24 occurrences appear with a modifier structure; this seems to run counter to our argumentation, but in all other 16 occurrences the Undergoer is an entity already accessible from the preceding context. Again it seems that pragmatic factors can cancel the linking of the Undergoer argument.21 With regard to the active accomplishment use of these verbs the possibility of noun incorporation is blocked; thus, expressions like Robert is an energy bar eater can never be understood as *Robert ate an energy bar. Undergoer assignment would necessarily require the ¿»/•construction: the eater of an energy bar. There are occasions where postmodification is constructed with other prepositional phrases. In these cases the derived nominal inherits the same non-predicative preposition of the corresponding argument from the base's LS. E.g. the emerger from the dark, a believer in sincerity / simplicity / best-sellers / the moderates ' way; the contender for Joe's title /for the challenge.

238

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero

4. Conclusions and prospects The treatment of word formation is still a matter of controversy for several grammatical models, among which FG should be included. In this article some of the inconsistencies of Predicate Formation Rules have been used as a background to propose a different approach. The Word Formation Component of the Functional Lexematic Model starts from a joint view of word formation as both a lexical and a morphosyntagmatic phenomenon. The analysis of -er formations from this perspective has proved the explanatory power of this approach: it accounts for the wide range of semantic variability of these formations and also for the interrelation between their semantic structure and the syntactic configuration of deverbal NPs. This has been made possible by the integration of a rich decompositional system and linking devices, as described in Van Valin and LaPolla's Role and Reference Grammar, within the formal apparatus of the WFC. Notes 1. This paper is part of the research project Diccionario nuclear sintáctico de base semántica del léxico en inglés antiguofinancedby the Gobierno Autónomo de Canarias, code number PI 1999/136. We are grateful to Drs. Butler, Mairal and Van Valin for their insightful comments on earlier versions of this paper. 2. Our analysis is based on the compilation of -er derived nouns in the Collins COBUILD dictionary and in the search for these formations in the LOB Corpus. 3. Note that other factors that are left unanswered as regards all possible agent formations in English are: (i) how to predict which agent affix {-er, -ist, -ant, ...) is to be selected; (ii) in what way all agent formations, from different bases and with different affixes, are part of a common derivational process; and (iii) under what conditions the resulting words display valency variations (e.g. writer vs. novel writer or the writer of this poem). 4. Note that the more we advance in this argumentation the less appropriate it seems to use the term "agent" nouns; in fact, in the remainder of this paper it will be used as mnemonic for all formations with the -er suffix; later a new terminological proposal will be made.

On the syntax-semantics interface in wordformation

239

5. This seems to be a satisfactory solution for instrumental -er formations in general. 6. This concept of schema later will be reformulated as "template", since its major function will be to account for the morphosyntactic structure of derived words. 7. Note that this involves a harmonious view of morphology, but at the expense of sacrificing the lexical aspects of WF. It is interesting to recall that both Construction Grammar (Golberg 1995) and RRG (Van Valin and LaPolla 1997) support a view similar to ours. 8. By way of example, let's recall AronofFs (1979: 53-54) description of a morphological restriction imposed by -al deadjectival suffix: it will not be attached to nouns of the form Xy-ment (i.e. the class of nouns of the form Xment where X is an independently occurring verb: *employmental < employment < employ). Note that this restriction is strictly morphological, and not phonological, since al can be attached to nouns ending in the sequence -ment when this is part of its root; e.g.: ornamental. 9. The development of this theory will come about in the model of Lexical Phonology (cf. Kiparsky 1982a, 1982b, 1983; Mohanan 1982, 1986; Strauss 1982; Harris 1983). The problems of a strictly (morpho)phonological approach to word-formation (basically the famous bracketing paradoxes that it caused) led to the resurgence of a syntacticist approach to derivational morphology; within the Government and Binding model we may cite the works of Marantz (1984, 1988), Fabb (1984), Sproat (1985) or Baker (1988), among others, in which this point of view is adopted. Cf. Cortés (1994: chapters 2 and 4) for a critical reassessment of these approaches. 10. Note that FG disallows the existence of non-productive rules of the kind proposed here. Dik (1997a: 342-345) considers this type of lexically-governed generalizations as "regularities" that would appear stored in the lexicon. The theoretical distance between FG's regularities and our redundancy rules is so short that the latter may be taken as a mere formalization of the firs. From this perspective, redundancy rules are stored in the word formation componenet and, despite their non-productive character, may be occassionally "activated" to produce incidental innovations, in much the same way as FG's regularities. 11. Note that within FG, verbs are classified in terms of a typology of States of Affairs which does not correspond with Vendler's proposal. As a matter of fact, in previous works (Cortés 1994) bases were grouped following the FG proposal; however, the analysis of verbal bases in terms of Aktionsart types (especially the opposition between telic and non-telic classes, involved in the distinction between activities and accomplishments) has proven more useful to predict the morpho-syntagmatic behavior of derived formations, as will be described later. 12. Again, the system of description is the same as the one in RRG. The description provided in the lexical entry above is, however, not exhaustive. A full classiti-

240

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero

cation would need to link the different Qualia to an ontological classification of concepts, such as the one proposed by Mahesh and Nirenburg (1995). 13. This analysis lies closer to FG's treatment of adjectival lexemes as predicates. 14. This pattern is, in fact, the basic structure for all derived agent formations in English. Therefore, we can posit that this formula would be located in the second module of the WFC. All derivative nouns with -er, -ant or -ist would be expansions from this logical structure. Their differences are codified in the phonological, morphological and lexical conditions of their respective entries. In fact, these conditions affect the range of possible slot-fillers of the underlying structure. This skeletal representation justifies the existence of the second module of the WFC, the set of basic templates for WF processes that can be instantiated by several derivational affixes. For a detailed description of these suffixes cf. Cortés (1994). 15. The introduction of this kind of ontological constant is explained in Rappaport and Levin (1998) and in Mairal Usón and Faber (this volume). Note that the Modeling Process is subject to a number of matching conditions. One of these conditions would affect the possibility of adding this kind of constants, since it must be compatible with the logical structure of the bases. Thus, among the best candidates for expansion with a manner constant would be formations with an activity predicate given that activities are typically done in a certain way, whereas locational stative predicates are difficult to be expanded in this way. 16. We do not describe deadjectival formations, since they follow the same structure as the other - deverbal- statai -er derivatives. 17. Γη RRG, the term semantic macroroles is introduced in order to account for "generalizations across argument-types found with particular verbs which have significant grammatical consequences; it is they, rather than specific arguments in logical structure, that grammatical rules refer to primarily." (Van Valin and LaPolla 1997: 139). Two macroroles are distinguished: Actor, the generalized Agent-type role, and Undergoer, the generalized Patient-type role. 18. This feature is especially revealing to differentiate activities from active accomplishments which, in opposition to the durative meaning of the former, are resultative by nature. 19. We are not considering the cases in which discourse conditions enable the speaker to introduce this kind of construction, since on these occasions the appearance of the incorporated nominal or the o/"-phrase is optional form a syntactico-semantic perspective; that is, factors such as pragmatic accessibility (Lambrecht 1994) can activate this kind of structure while at the same time canceling some of the rules that are posited in this paper. The topic of the interaction of pragmatics and word-formation should be the matter for further research. 20. One interesting restriction of deverbal nomináis in general is that they only have one direct core argument (cf. Nunes 1993: 389-390), as is shown by the ungrammaticality of the following examples:

On the syntax-semantics interface in wordformation

241

(a) *the teacher of linguistics of the students (b) "the linguistics teacher of the students (c) *the student teacher of linguistics (d) *the student linguistics teacher The difference between action and -er nomináis is that, in the case of the second ones, the core argument can appear as a postmodifier of phrase (the only possible construction for action nomináis) or as premodifier noun (noun incorporation). Notice that it is possible to say Luisa's teacher of linguistics, but as argued in Nunes (1993) and Van Valin and LaPolla (1997) the genitive NP is not a core argument of the deverbal NP; they are to be interpreted as optional possessors that appear in a pre-core position (NP initial position, Van Valin and LaPolla (1997: 56), or Left-Detached Position, Nunes (1993: 386-387). 21. In the case of non-referential arguments there is no possibility of recovering them from discourse; that is, the optionality of the of phrases in these situations cannot be accounted for by pragmatic accessibility.

References Aronoff, Mark 1979 Baker, Mark 1988 Booij, Geert 1977

Word Formation in Generative Grammar, 2nd edition. Cambridge, M.A.: MIT Press. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press.

Dutch Morphology. A Study of Word-Formation in Generative Grammar. Lisse: The Peter de Ridder Press. Chomsky, Noam and Morris Halle 1968 The Sound Pattern ofEnglish. New York: Harper and Row. Collins COBUILD English Language Dictionary 1987 Collins COBUILD English Language Dictionary. London: Collins. Cortés Rodriguez, Francisco José 1986 Análisis lexemático-funcional de los adjetivos del campo de percepciones gustativas y olfativas en inglés y español. M.A. Thesis, Universidad de Granada. 1988 'Que huele mal': un ejemplo de análisis adjetival contrastivo. Revista Canaria de Estudios Ingleses 16: 225-238.

242

Francisco J. Cortés Rodriguez and María Jesús Pérez Quintero

1994 1997a 1997b 1997c

1997d

Lexicón onomasiològico de afijos nominalizadores españoles e ingleses. Ph.D. dissertation, Universidad de La Laguna. La creación léxica. Una aproximaciónfimcional.Universidad de La Laguna: Servicio de Publicaciones. The Functional Lexematic Model of word formation. Atlantis 19: 79-98. La morfología derivativa en la Gramática Funcional de Dik: ¿formación de palabras o formación de predicados?. Alfinge 9: 117-134. Sobre la relación entre léxico-génesis, diacronia y categorización. In: Gerd Wotjak (ed.), Towards a Functional Lexicology, 227243. Frankfurt: Peter Lang.

Coseriu, Eugenio 1978 Gramática, semántica, universales. Madrid: Gredos. 1981 Principios de semántica estructural. Madrid: Gredos. Dik, Simon C. 1997 The Theory of Functional Grammar. Part 2: Complex and Derived Constructions. Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter. Fabb, Nigel 1984 Syntactic Affixation. Cambridge, M.A.: MIT Press. Faber, Pamela and Ricardo Mairal Usón 1999 Constructing a Lexicon of English Verbs. Berlin/New York: Mouton de Gruyter. Foley, William A. and Robert D. Van Valin, Jr. 1984 Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Goldberg, Adele E. 1995 A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Groot, Casper de 1989 Predicate Structure in a Functional Grammar of Hungarian. (Functional Grammar Series 11.) Dordrecht: Foris. Harris, James W. 1983 Syllable Structure and Stress in Spanish: A Nonlinear Analysis. Cambridge, M.A.: MIT Press. Kiparsky, Paul 1982a Explanation in Phonology. Dordrecht: Foris. 1982b From cyclic phonology to lexical phonology. In: Harry van der Hulst and Norval Smith (eds.), The Structure of Phonological Representations, part 1,131-175. Dordrecht: Foris.

On the syntax-semantics interface in wordformation 1983

243

Word-formation and the lexicon. Proceedings of the 1982 MidAmerica Linguistics Conference, 3-22. Lawrence: University of Kansas. Kristoffersen, Lars 1991 Verbal derivation and inflection in a Functional Grammar of West Groenlandie. Ph.D. dissertation, University of Copenhagen. Laca, Brenda 1993 Las nominalizaciones orientadas y los derivados españoles en dor y -nte. In: Soledad Varela Ortega (ed.) Fundamentos de morfología, 180-204. Madrid: Taurus Universitaria. Lambrecht, Knud 1994 Information Structure and Sentence Form. Cambridge: Cambridge University Press. Langacker, Ronald W. 1987 Foundations of Cognitive Grammar, Volume 1. Stanford: Stanford University Press. 1991 Foundations of Cognitive Grammar, Volume 2. Stanford: Stanford University Press. Levi, Judith 1978 The Syntax and Semantics of Complex Nomináis. New York: Academic Press. Mackenzie, J. Lachlan 1986 Aspects of nominalization in English and Dutch. Working Papers in Functional Grammar 15. 1987 The representation of nominal predicates in the fund: A new proposal. Working Papers in Functional Grammar 25. Mahesh, Kavi and Sergei Nirenburg 1995 A situated ontology for practical NLP. A Paper Presented at the IJCAI '95 Workshop on Basic Ontological Issues in Knowledge Sharing, August 19-21, Montreal, Canada. Mairal Usón, Ricardo and Pamela Faber this volume Functional Grammar and lexical templates. Marantz, Alec 1984 On the Nature of Grammatical Relations. Cambridge, M.A.: MIT Press. 1988 Clitics, morphological merger, and the mapping to phonological structure. In: M. Hammond and Michael Noonan (eds.), Theoretical Morphology. Approaches in Modern Linguistics, 253-271. San Diego: Academic Press, Inc.

244

Francisco J. Cortés Rodríguez and María Jesús Pérez Quintero

Marchand, Hans 1969 The Categories and Types of Present-Day English WordFormation. A Synchronic-Diachronic Approach, 2nd edition. München: C.H. Beck'sehe Verlagsbuchhandlung. Martin Mingorance, Leocadio 1984a Lexical fields and Stepwise Lexical Decomposition in a contrastive English-Spanish verb valency dictionary. In: Reinhard Κ. Hartmann (ed.), Lexeter '83 Proceedings, 226-236. Tübingen: Max Niemeyer Verlag. 1984b La semántica sintagmática del adjetivo: Parámetros para la organización de un lexicón inglés/español de valencias adjetivales. Actas del 2o congreso de A.E.S.L.A., Universidad de Granada: 329-340. Bases metodológicas para un estudio contrastivo del léxico 1985 derivado. Revista Española de Lingüística Aplicada Año I: 3754. Classematics in a Functional-Lexematic grammar of English. 1987 Actas delXCongreso Nacional de AEDEAN, Zaragoza: 377-388. Mohanan, K.P. 1982 Lexical Phonology. Bloomington: Indiana University Linguistics Club. The Theory of Lexical Phonology. Dordrecht: Reidel. 1986 Nunes, Mary L. Argument linking in English derived nomináis. In: Robert D. Van 1993 Valin (ed.), Advances in Role and Reference Grammar, 375-432. Amsterdam/Philadelphia: John Benjamins. Pustejovsky, James 1991 The generative lexicon. Computational Linguistics 17:409-441. Rappaport, Malka and Beth Levin 1998 Building verb meaning. In: Miriam Butt and William Geuder (ed.), The Projection of Arguments: Lexical and Compositional Factors, 97-134. Stanford: CSLI. Scalise, Sergio 1987 Morfología Generativa. Madrid: Alianza. Shibatani, Masayoshi and T. Kageyama 1988 Word formation in a modular theory of grammar: Postsyntactic compounds in Japanese. Language 64 (3): 451-482. Siegel, Dorothy L. 1979 Topics in English Morphology. New York: Garland Publishing, Inc. Sproat, Richard W. 1985 On Deriving the Lexicon. Cambridge, M.A.: MIT Press.

On the syntax-semantics interface in wordformation

245

Strauss, Steven L. 1982 On the theory of word-formation and its role in phonological analysis. Linguistic Analysis 9: 253-276. Van Valin, Robert D. (ed.) 1993 Advances in Role and Reference Grammar. Amsterdam / Philadelphia: John Benjamins. Van Valin, Robert D and Randy J. LaPolla 1997 Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Vendler, Zeno 1967 Linguistics in Philosophy. Ithaca: Cornell University Press.

Catching a glimpse of linguistic reality: Modelling the complexity of CATCH in the Functional Lexematic Model Chris Butler

1. Introduction1 Considerable advances in accounting for the richness of lexicogrammatical patterning have been made in recent years within the Functional Lexematic Model (see Martin Mingorance 1998, Faber and Mai ral Usón 1999). There are, however, some aspects of the FLM, as developed up to now, which, from the point of view of someone outside the model, could be seen as limitations: - So far the FLM has tended to deal with a restricted range of meanings regarded as central for lexemes within particular lexical fields (see e.g. Mairal Usón and Faber, this volume), rather than with the full range of meanings listed in dictionaries for particular lexical items.2 This orientation is a natural one, given the overall emphasis of the FLM on building up a picture of the lexicon in broad terms, rather than concentrating in great detail on particular, small areas. - Although aligning itself with the central tenet of functionalism, namely that language should be studied in relation to its centrally communicative role, the proponents of the FLM have only just started making use of the rich insights into actual usage provided by corpus analysis.3 - Although the model combines the paradigmatic and syntagmatic dimensions in insightful ways, it has not yet integrated systematic information on the syntagmatic lexical patterns we refer to as

248

Chris Butler

collocation (see, however, Butler 1998,1999 for suggestions as to how this might be done). - Despite the explicitly cognitive orientation of the FLM, little attention has so far been given to testing the idea that different senses of a lexeme differ in their psychological saliency. In this paper, which has been written against the background of both Faber and Mairal Usón's (1999) detailed account of the FLM and the ground-breaking proposals made in their contribution to the present volume, I will bring evidence from dictionaries, corpora and informant testing to bear on one particular issue: the tendency of many high and middle frequency verbs in English to have a wide range of uses, many of which involve collocation and colligation with items of varying degrees of specificity. The meanings involved are systematically related to one another, but also to other quite different lexemes. The challenge is to capture the semantic, syntactic and pragmatic aspects of these complex relationships within an integrated model which does full justice to the complexity of the evidence. The discussion will centre on the verb CATCH,4 which occurs with considerable frequency in English.5 Having first looked at how this verb has been handled so far within the FLM framework, I will present a schematic network for the meanings of CATCH, derived from the definitions given in the Longman Dictionary of the English Language, the corpus-based Collins COBUILD English Dictionary and the Collins English Dictionary, and also from informant testing on the relative salience of the meanings. I will then examine data from the British National Corpus which reveals the quantitative importance of the co-occurrence of CATCH with nominal expressions, including an important group referring to perceptual phenomena (e.g. CATCH a glimpse/whiff, CATCH + /Äe/possessive eye). An attempt will be made to assess the extent to which the data from informant testing and corpus analysis can be captured within the FLM, especially in the version articulated in Mairal Usón and Faber (this volume). We shall see that although the proposals made by Mairal Usón and Faber in the present volume fare rather better than those in their 1999 book,

Catching a glimpse of linguistic reality

249

in accounting for the data to be presented here, there is still some way to go before the full complexity of the behaviour of CATCH, as revealed by dictionary, corpus and informant data, can be adequately modelled.

2. The verb CATCH in the FLM In work done within the FLM framework, the verb CATCH has been treated as part of the lexical domain of POSSESSION, for which the basic lexeme ("genus") is HAVE. Faber, Márquez Linares and Fernández Sánchez (1998), Faber and Mairal Usón (1999: 101-109, 301-04) and Mairal Usón (1999: 86-87) present the subdomain το COME το HAVE, for which the most general lexeme is GET. Within this subdomain, one of the sub-subdomains is GETTING SOMETHING FROM SOMEONE/ SOMETHING ELSE, the general lexeme here being TAKE. Taking by force is represented by SEIZE, WREST and CAPTURE, these being distinguished by seizing having a component of officialdom, wresting being marked as archaic, while capturing is said to be for the purpose of control, SEIZE has the hyponym ANNEX, indicating the seizure of an area of land, CATCH comes in as a hyponym of CAPTURE, defined as capturing after a chase. The goal can be a person, in which case the appropriate hyponyms of CATCH are APPREHEND (formal) or NAB (informal), while if the goal is an animal, we have lexemes such as FISH, BAG, HOOK and TRAP (the last three specifying the instrument). Faber, Márquez Linares and Fernández Sánchez (1998) also add ARREST as a hyponym of CATCH, and COLLAR ('catch sb holding them so that they cannot get away') at the same level as NAB. For catching animals, they also add POACH at the same level as the lexemes specified above, and SNARE as a hyponym of TRAP.

250

Chris Butler

3. The verb CATCH in three dictionaries of English The meanings of CATCH were first explored by searching three dictionaries of English: the second edition of the Collins Cobuild English Language Dictionary, the Longman Dictionary of the English Language and the Collins English Dictionary.6 These three dictionaries have rather different criteria for the ordering of their entries: Collins places any "core" meaning first, but otherwise places emphasis on frequency of use and arranging senses to give a coherent account of the meanings; Longman places senses with unrestricted geographical usage first, in their historical order; Cobuild gives, as the first sense, a common one which is clear and quite easy to understand, and in the second edition many items are assigned to "superheadwords" relating to groupings of senses for a particular word. The entries in the Longman dictionary cover the main senses, and are summarized in Figure 1, by way of exemplification of the range of meanings involved. Examples are given here only when necessary to make the sense intended clear. vt 1 a) to capture or seize, esp after pursuit b) to take or entangle in or as if in a snare c) to deceive d) to discover unexpectedly, find e) to become suddenly aware of (/ caught myself swearing) f) to get (something) entangled 2 a) 1 to take hold of, snatch, esp to receive and hold (a moving object) in the hands 2 to dismiss (a batsman in cricket) by taking and holding the ball... b) to affect suddenly (the idea caught his imagination immediately) c) to intercept (they caught itjust in time to prevent a major outbreak) d) to avail oneself of, take (/ try and catch what sleep I can) e) to obtain through effort 3 to become affected by (the grease caught fire), e.g.: a) to contract (e.g. a cold) b) to respond sympathetically to the point of becoming imbued with (e.g. the spirit of an occasion)

Catching a glimpse of linguistic reality

251

c) to be struck by a) to take in and retain (a barrel... rainwater) b) to fasten 5 to take or get momentarily or quickly (a glimpse of a friend, sight of her) 6 to be in time for (catch the bus) 7 a) to grasp by the senses or the mind, apprehend (/ didn't quite catch it) b) to capture a likeness of

4

vi 1 to become caught, stick 2 a) of a crop, to sprout and become established b) of a fire, to catch alight and become established 3 to play the position of catcher on baseball team catch it (informal) to incur blame, reprimand, or punishment Figure 1. Entries for CATCH in the Longman Dictionary of the English Language

From the list of senses in the three dictionaries, a number of interrelated dimensions of meaning emerge: - The element of possession foregrounded in the current FLM account is clearly important, in the sense that catching someone or something, whether a criminal, a rabbit, a cold, a bus, or a glimpse, results in "having" whatever has been caught, in a very general sense of that verb. At a very basic and abstract level, then, the FLM is correct in regarding CATCH as a verb of possession. - A second component which links all the various senses of CATCH is that of contact. In sense la of the Longman list, which is the one represented by the FLM treatment of the verb, the contact is clearly physical, as it is also for (the literal part oí) sense lb, If, 2a, 4a and 4b. The intransitive uses 1, 2 and 3 are clearly related to transitive senses lb/f, 3 and 2a respectively. For other uses, however, the contact is perceptual or cognitive, lb, le, 2b, 3b, 5 and 7a perhaps being the clearest cases.

252

Chris Butler

- A further dimension of meaning which is common to one subgroup of senses is the notion that the contact is being used to control what is caught by preventing its escape, or further movement, and that this prevention may involve an element of difficulty: senses la, lb, 2a, 4a, 4b involve physical capture, while Id, le, 2c, 2d, 5, 6 and 7b can be subsumed under this heading if we extend the meaning to include control over non-physical entities. In Id, we have a sense which is often used when we have come across someone doing something they should not be doing (iCaught you! / You wouldn't catch me doing that), in which case they are prevented from escaping from the situation unnoticed. In le there is the implication that becoming aware of what one is doing leads to stopping oneself doing it. The example given for 2c clearly indicates the controlling, preventative nature of the situation, while that for 2d describes a situation in which sleep is fugitive and needs to be "captured" before it "escapes". If we catch a glimpse of someone or catch sight of them, our sight "captures" them before they "escape" from our view, while catching a bus involved managing to be at the bus stop in time to prevent the bus leaving without us. Finally, capturing a likeness of someone (7b) again involves "fixing" their essential characteristics in some artistic medium, so preventing their "escape". - The "immobility" referred to above is just one way, though perhaps the most frequent, in which the process of catching can affect the entity which is caught. The affectedness can, however, be of other kinds, specified in nomináis following the verb: this is clearest in the group of senses in 3: for instance, the grease is affected by contact with heat or fire so that it goes up in flames; the person's body is affected by contact with germs so that these cause illness.7

Catching a glimpse of linguistic reality

253

4. An informant test on the meanings of the verb CATCH In order to assess, in a rather simplistic way, the psychological salience of various senses of the verb CATCH, 160 native speakers of English, who were students in the Education facility of Cardiff University, UK, were asked to write down three sentences containing a form of the verb CATCH.8 The examples given were then categorized into the various senses represented, with the results shown in Table 1. Table 1. The most psychologically salient senses of CATCH, as revealed by an informant test SENSE

IN ALL SENTENCES GIVEN

catch moving object catch disease catch means of transport catch in trap, net, etc. catch after chase TOTAL of these categories

157 (32.7%) 76 (15.8%) 76 (15.8%) 47 (9.8%) 28 (5.8%) 384 (80.0%)

AS FIRST SENTENCE GIVEN

88 (55.0%) 17 (10.6%) 14 (8.8%) 19(11.9%) 4 (2.5%) 142 (88.8%)

It can be seen that the group of senses concerned with catching some object to arrest its mobility (i.e. the first, fourth and fifth categories in Table 1) accounts for 48.3% of the total of 480 sentences provided, while in the first responses this proportion rises to 69.4%. This gives clear support for the treatment of this type of meaning as the central one. This is in line with other studies showing that the most salient sense (i.e. the one most people come up with first when asked about a word) is usually a concrete rather than an abstract one.9 Note that the sense analysed as basic in the FLM account ('catch after chase'), though it does belong to this overall group, is actually not very frequent in the responses (5.8% of total responses, 2.5% of first responses). Note also that neither CATCH a glimpse not CATCH sight of appears even once in the informant data, and CATCH . . . eye occurs only 3 times.

254

Chris Butler

5. A tentative schematic network for the meanings of the verb CATCH

Figure 2 is an attempt to capture the main meanings of the verb CATCH as evidenced in the dictionaries consulted, building in the data on psychological salience derived from our informant test. It takes the basic sense of the verb to be 'possession through contact', and within this, treats as central the groups of meanings concerned with arresting the motion of a person, object or substance. Two major groups of extensions radiate from this central core. One is related through the concept of perception, giving rise to subgroups of meanings concerned with visual and mental perception. One of the meanings within the mental subgroup has the component of "affectedness" in common with the meanings concerned with catching a disease and catching fire. A second major extension of the central group of senses is through abstraction, to yield the idea of availing oneself of something which is difficult to get, often implying the exercising of some effort. One aspect which emerges very clearly from this exercise is the importance of the other lexical items with which forms of CATCH collocate. In order to investigate this in more detail, we must turn to corpus evidence.

6.

CATCH

in the British National Corpus

Table 2 shows the frequencies for the forms catch, catches, caught, catching in the British National Corpus (100 million words in total, 90% written, 10% spoken), according to the part of speech tags assigned by the CLAWS tagging system.10 From these data, we can produce the simpler Table 3, showing the frequencies of items tagged as clearly verbal, nominal or adjectival, as well as those where the tagging program was unable to decide between two classifications. It is clear that the use of forms of CATCH in the corpus is overwhelmingly verbal, and that of the clearly verbal uses, the order of frequency of forms is caught > catch > catching > catches.

Catching a glimpse of linguistic reality

255

X ëυ •e SÌ

£ 1st arg. of do' (x,... > 1st arg. of pred'(x, y) > 2nd arg. of pred'(x, y) > arg. of pred'(x)

In accordance with this hierarchy, the criteria to select the PSA depending on the type of construction are the following: (3)

Privileged Syntactic Argument Selection Principles a. Syntactically accusative constructions: highest-ranking macrorole is default choice. b. Syntactically ergative constructions: lowest-ranking macrorole is default choice.

Finally, case assignment rules are in a direct relation to the assignment of macroroles and the PSA (Van Valin and LaPolla 1997: 359): (4)

Case assignment rules a. Assign nominative case to the highest-ranking macrorole argument (in terms of the Privileged syntactic argument selection hierarchy). b. Assign accusative case to the other macrorole argument. c. Assign dative case to non-macrorole arguments (default).

Therefore, taking into account the Default Macrorole Assignment Principles, the Actor - Undergoer Hierarchy and the interaction existing between macroroles and grammatical relations, the information to be included in the lexical representations of verbs will be very much reduced. Firstly, syntactic information will be derived from the semantic notion of transitivity being based on the Default Macrorole Assignment Principles. Secondly, it is not necessary to specify the macroroles assigned to the arguments of the verbs since the Actor - Undergoer Hierarchy will provide this information. Thirdly, grammatical relations will be inferred from the Default Macrorole Assignment

288

Marta María González Orta

Principles and from the Privileged Syntactic Argument Selection Principles of the different languages. 3. Hypothesis The speech lexical domain in Old English has been constructed turning the information from the Thesaurus of Old English (Roberts and Kay 1995) about Old English speech verbs into the structure of the speech domain in Present-day English, following Faber and Mairal Usón's (1999: 288-290) proposal. According to Cortés Rodríguez and Mairal Usón (forthcoming): The basic assumption behind this method is that semantic domains are kept constant diachronically for its more part, needless to say that the internal microstructure of these domains has been substantially altered throughout the history of the English language.

In order to provide the lexicon of a historical language with an organization in lexical hierarchies, semantics alone fails to do it since there is no access to specific meaning definitions. As Cortés Rodríguez and Mairal Usón (forthcoming) state: By combining the information from different lexicographical sources we will be able to group lexical units in terms of their genus; it is the level of differentiae speciflcae that seems impossible to determine: a definite ascertainmnet of sense-relations among lexemes is implausible unless further sources of information are used.

In that sense, applying the Lexical Iconicity Principle - Beta Reading, syntactic information may contribute to reconstruct these lexical hierarchies in Old English determining the semantic space and the location of lexemes within a subdomain. This hypothesis states that departing from the Lexical Iconicity Principle that Faber and Mairal Usón (1999: 187) propose:

Lexical templates and syntactic variation

(5)

289

Lexical Iconicity Principle The greater the semantic scope of a lexeme, the greater its syntactic variation.5

Cortés Rodríguez and Mairal Usón (forthcoming) will apply this principle although in a reverse direction, what they call the Lexical Iconicity Principle - Beta Reading: (6)

Lexical Iconicity Principle - Beta Reading The greater the syntactic coverage of a lexical unit, the higher its position in the semantic hierarchy within a given subdomain.

Since this research is based on The Dictionary of Old English Corpus,6 speech verbs will be located in this corpus in order to obtain the contexts in which these lexemes appear.7 That is what we have done with the speech verb SECGAN and the results will be shown in the next section. Accordingly, this lexeme seems to be one of the most generic lexemes of the Old English speech lexical domain, since it presents a great variety of complementation patterns which according to the Lexical Iconicity Principle - Beta Reading will prove its higher position in the semantic hierarchy of this domain.

4. A proposal of lexical templates for the lexical class of old English speech verbs Van Valin and LaPolla (1997: 116-118) state that speech verbs or verbs of saying are considered originally activity verbs and the basic logical structure which will correspond to them is the one that is presented below, from which the logical structure of more specific speech verbs can be derived: (7)

do* (x, [express.(a).to.(ß).in.language.(Y)' (χ, y)])

290

Marta María González Orta

This logical structure shows two external variables χ, y, or external argument positions, which will have a syntactic representation, and three internal variables α, β, γ (marked by Greek letters), which make reference to the content of the expression, to the addressee and to the language used, respectively. These internal variables are part of the semantic representation of these verbs; that is, they function as ontological constants of this verbal class. Their introduction will permit to add a semantic decomposition to the logical structure of activity verbs, giving place to the lexical template for the lexical class of speech verbs.8 Firstly, activity verbs are considered intransitive verbs, since their second argument cannot be assigned the macrorole Undergoer due to its non-referential, inherent nature.9 According to the Default Macrorole Assignment Principles and the Case assignment rules, these verbs take only one macrorole, Actor, which corresponds to the variable χ (it is also the PSA) and which will be assigned Nominative case, and if there is a non-macrorole direct core argument (y), this will be assigned Dative case. Secondly, these verbs can take a second argument realized by a prepositional construction, whose logical structure will be extracted from the lexicon and added to the original logical structure of the verb. The following logical structure corresponding to the lexeme SECGAN illustrates this idea: (8)

do' (x, [express.(a).to.(ß).in.language.(y)' (χ, 0)]) λ [about' (y)], where α = [about' (y)] pa ongan se Hœlynd secgan be Iohanne (Mt (WSCp) B8.4.3.1)

'Then God began to speak about John.' Thirdly, within this verbal class the alternation activity-active accomplishment (or intransitivity-transitivity) can take place, depending on the context in which these verbs appear.10 As we already saw in the description of the different verbal classes in terms of their Aktionsart, the semantic feature which differentiates activities form active accomplishments is the telicity of the latter. This semantic

Lexical templates and syntactic variation

291

difference will be reflected in their logical structures. Therefore, the logical structure of an active accomplishment will have the same internal variables that are characteristic of the semantics of speech verbs (α, β, γ), but instead of the two external variables of activities logical structures, this one will present three external variables, x, z, y, where χ will make reference to the speaker, ζ to α or the content of the expression, and .y to β or the hearer. Applying again the Default Macrorole Assignment Principles and the Case assignment rules, the variable χ will take the macrorole Actor (it will be also the PSA) and it will be assigned Nominative case, the variable ζ will take the macrorole Undergoer and it will be assigned Accusative case if it is not realized by a 7%a/-clause, Whclause or Direct speech, and finally the variable^, which corresponds to a non-macrorole direct core argument, will be assigned Dative case. The corresponding logical structure will be presented below: (9)

[do' (x, [express.(α).to.(ß).m.language.(γ, (x, y)] )] [BECOME aware.of (y, z)], where y = β, ζ = α

&

The two logical structures above will conform two lexical templates of this verbal class, since both of them contain syntactic (external argument positions x, y, z) and semantic information (internal variables α, β, γ) within the same lexical representation.11 Following Mairal Usón and Faber (this volume: 87), the process that governs the mapping between a lexical template and the different syntactic structures is called Lexical Template Modeling Process, which says: Lexical templates can be modeled by accommodating external variables, instantiating internal variables and operators (e.g. CAUSE), or else, by introducing elements resulting from the fusion with other templates iff there is a compatibility between the features in the lexical template and the syntactic construction under scrutiny.

The last point to be dealt with is our proposal of incorporating the typology of modality operators signaling the speaker's personal commitment to the truth of the proposition developed by Faber and Mairal Usón (1999: 132-133) into the analysis of clausal subordina-

292

Marta María González Orta

tion in RRG, since they seem to be complementary. In that sense, when the content of the expression of a speech verb being represented by the external variable ζ is realized by a 77utf-clause or Whclause, the following scale of operators will be applied: -

-

-

-

The speaker, who accepts the proposition as true, presents its content as being the case with no alternatives available. Such predicates act as lexical triggers because they presuppose the truth of their complement clause. PROBABLE: The speaker, slightly less committed to the truth of the proposition, presents the propositional content as expected, but offers alternatives. POSSIBLE: The speaker presents the information as possibly true with a preference for one alternative. This operator also explains the non-occurrence of satellites which negate the expectation of the SoA being the case. INDETERMINATE: The speaker presents the propositional content as an open question with two equal alternatives. CERTAIN:

The following examples corresponding to SECGAN illustrate the operator CERTAIN: (10)

ne mihte he for sceame him openlice secgan pœt his eawfœste wif on dcere wodnysse lœg (MC Horn II, 10 Bl.2.11) 'He might not say to him openly for shame that his wife fell in madness.' S VOI + O-That clause: χ (Nom), ζ [Certain indirect dicourse], y (Dat)

(11)

ac we wyllad eow secgan nu cerest hu he com to ôœre byrig hierusalem OECHomI, 14.1 Bl.1.15) 'But we will say to you now how he came to Jerusalem.' S VOI + O-Wh clause: χ (Nom), ζ [Certain indirect discourse], y (Dat)

Lexical templates and syntactic variation

293

Turning now to the Old English speech lexical domain, the lexeme SECGAN, due to its syntactic variability and, consequently, its prototypical nature within this lexical domain (.Lexical Iconicity Principle-Beta Reading), will be analyzed as representative of it. This predicate is characterized by two main lexical templates. The first is conformed by the logical structure corresponding to an active accomplishment and the second by that of an activity verb. Applying the Lexical Template Modeling Process, the second template will be derived by a reduction process from the first one, which in this case is the maximal lexical template, and likewise the different syntactic constructions that characterize the lexeme SECGAN. Firstly, from the maximal lexical template another template can be obtained corresponding to an active accomplishment where only two external variables (x and z) are syntactically realized: (12)

do' (x, [express.(a).to.(β).in.language.(y)' (x, 0)]) & [BECOME aware.of (0, z)], where 0 = β, ζ = α

Secondly, from the lexical template conformed by an activity logical structure the following templates can be derived (13-16): (13)

do' (x, [express.(a).to.(ß).in.language.(y)' (x, y)]) A [about' (ζ)], where α = [about* (ζ)]

(14)

do' (χ, [express.(a).to.(ß).in.language.(Y)' (χ, 0)]) a [about' (y)], where α = [about' (y)]

where one of the arguments is realized by a prepositional construction introduced by Old English be or embe and represented in the LS by [about' (y)] or [about' (z)], depending on whether there are two (or and y) or three (x, y, and z) external variables which are syntactically realized. (15)

do' (x, [express.(a).to.(ß).in.language.(y)' (χ, 0)]) a [of (y)], where α = [of (y)]

294

Marta María González Orta

where the second argument is realized by a prepositional construction introduced by Old English fram and represented as [of (y)] in the LS, since only two external variables (x and y) have a syntactic expression. (16)

do' (x, [express.(a).to.(ß).in.language.(Y)' (χ, o)])

where only the external variable χ has a syntactic realization. Table 2 contains the maximal lexical template corresponding to SECGAN along with its syntactic alternations and examples that illustrate them. 5. Concluding remarks Following the Lexical Iconicity Principle - Beta Reading, which says that the syntactic variability of a lexeme is connected with its higher position within the semantic hierarchy of a given (sub-)domain, the results obtained from the syntagmatic analysis of the Old English lexeme SECGAN have proved its prototypical nature within the Old English speech lexical domain. In the syntagmatic analysis of the FLM the notion of lexical template has been integrated as a way of representing the interaction between syntax and semantics, since predicate frames have been proved inadequate. Accordingly, the FLM will provide the logical structures developed by Van Valin and LaPolla (1997) with a semantic decomposition which will permit to capture generalizations within verbal classes, reducing the information to be included in the lexical entries. Moreover, the Lexical Template Modeling Process will govern the mapping between lexical templates and the different syntactic structures of a lexeme. Finally, what we have done in this paper is only a starting point towards the hierarchical configuration of the Old English speech lexical domain. Therefore, in the same way as SECGAN is represented by a maximal lexical template from which its syntactic alternations will be derived, within the speech lexical domain each sub-

Lexical templates and syntactic variation

295

domain is characterized by a maximal lexical template to account for the different syntactic configurations corresponding to the predicates that integrate the subdomains. Table 2. Syntactic alternations of SECGAN SECGAN do' (x, [express.(a).to.(ß).in.language.(y)' (χ, (y))]) & [BECOME aware-of* ((y), ζ)], where y = β, ζ = α do' (χ, [express.(a).to.(ß).in.language.(y)' (x,y)]) & [BECOME aware.of* (y,z)], where y = β, ζ = α SV01+ 02: χ (Nom), ζ (Acc), y (Dat) e.g. Se mœssepreost sceal secgan sunnandagum and mœssedagum pœs godspelles angyton engliscpam folce OEletl (Wulfsige Xa) Bl.8.1) 'The masspriest will say to the people on Sundays and massdays the meaning of the gospel in English.' SVOl + O-That clause: χ (Nom), ζ [Certain indirect dicourse], y (Dat) e.g. ne mihte he for sceame him openlice secgan poet his eawfceste wif on âœre wodnysse lœg (MC Horn II, 10 Β1.2.11 ) 'He might not say to him openly for shame that his wife fell in madness.' SVOl + O-Wh clause: χ (Nom), ζ [Certain indirect discourse], y (Dat) e.g. ac we wyllaô eow secgan nu cerest hu he com to ôœre byrig hierusalem OECHomI, 14.1 Bl.1.15) 'But we will say to you now how he came to Jerusalem.' SVOl + O-Direct speech: χ (Nom), ζ [Direct discourse], y (Dat) e.g. and hu miht pu secgan to Ôinum breâer pus: geôafa, min broôor, pœt ic pat mot ateo of ôinum eagan nu (JE Hom 14 Bl.4.14) 'And how you might say to your brother thus: "Admit, my brother, that I remove now that mote of this eye".'

296

María María González Oria

do' (x, [express.(a).to.(ß).in.language.(y)' (χ, β)]) & [BECOME aware.of (0,z)], where 0 = β, ζ = α SVO: χ (Nom), ζ (Acc) e.g. Hyrde ic secgan... pcet word (Part A3.18) Ί heard to say that word.' SVO-That clause: χ (Nom), ζ [Certain indirect discourse] e.g. swa we cer gehyrdon secgan pœt se eadiga Simeon wœs soôfcest & clœne & godfyrht on his life (LS19 (PurifMaryVerc 17) B3.3.19) 'So we heard to say that the blessed Simeon was honest and clean and godfearing in his life.' SVO-Wh clause: χ (Nom), ζ [Certain indirect discourse] e.g. Ic wylle secgan, hwœt hi ealle habbaâ (JE Gram Β1.9.1) Ί will say what they all have.' SVO-Direct speech: χ (Nom), ζ [Direct discourse] e.g. Apollonius sœde: Forôam pe ic bœd his dohtor me to gemœccan B4.1)

(ApT

'Apollonius said: "Therefore I commanded his daughter to cohabit with me . do' (x, [express.(a).to.(ß).in.language.(y)' (χ, y)]) λ [about* (ζ)], where α = [about' (ζ)] V + 01+ be + Dat: χ (Nom), y (Dat), [be + Dat PP] e.g. Gyt we wyllad eow secgan be sumon gesœligon cyninge Β 1.4.27)

(jEHom 27

'Yet we will talk to you about some prosperous king.' V + OI+ embe + Acc: χ (Nom), y (Dat), [embe + Acc PP] e.g. Git ic pe wille secgan embe da twelf ymbrendagas (Comp 10.2 (Henel) B20.10.2)

'Yet I will talk to you about the twelve Emberdays.'

Lexical templates and syntactic variation

do' (x, [express.(ot).to.(ß).in.language.(y)' (χ, 0)]) λ [about' (y)], where α = [about* (y)] V + be + Dat: χ (Nom), [be + Dai PP] e.g. pa ongan se Hœfyndsecgan be Iohanne (Mt (WSCp) B8.4.3.1) 'Then God began to speak about John.' V + embe + Acc: χ (Nom), [embe + Acc PP] e.g. and nan man ne mœg on ôysum middanearde fullfremedlice secgan embe pone sodan God (JE Hex Bl.5.13) 'And no man in this world may completely speak about the true God.' do' (X, [express.(a).to.(ß).in.language.(Y)f (Χ, 0)]) Λ [of (y)], where a = [of (y)] V + fram + Dat: x (Nom), [fram + Dat PP] e.g. ¡xet heframSigemundes secgan hyrde ellendcedum (Beo A4.1 ) 'That he heard to speak of the heroic deeds of Sigemunde.'

do' (x, [express.(a).to.(ß).in.language.(y)' (x, y)]) SVO: x (Nom), y (Dat) e.g. ac we wyllaâ eow secgan (iEHom 19 B1.4.19) 'But we will talk to you.' do' (x, [express.(a).to.(ß).in.language.(y)' (χ, o)]) SV: x (Nom) e.g. swa se Hœlend sœde œr his prowunge (iElet4 (SigeweardZ) Β1.8.4.4) 'Such as God said before his passion.'

297

298

Marta Maria González Orto

Appendix12 1. men sceolon...secgan drihtnepane (Sat A 1.4) Ic eow secgan mag sod orgete (And A2.1 ) We moton eow secgan eowre sawle pearfe (jECHom 1,17 (App) Β1.1.19.4) Gif se sacerd ne mœg deem lœwedum marmum larspel secgan (¿ECHom II, 41

Β 1.2.45) Nu wylle we eow secgan sume Petres wundra (JELS (Peter's Chair) Β1.3.11) pa began se preost...secgan done soôan geleafan pam arwurpan Albane

(Alban) Β 1.3.20)

OELS

him snytro on sefan secgan mihte, nales òy pe he pœt moste oôôe genuman woldepœthe paragifenagodepancode (Dan Al.3) ic ne mœg nanne manflndan,pe me seege hweet hit behealde (Gen B8.1.4.1 ) magum to secgan...hu him on side gelomp. (Jul A3.5) And he pa ongan him secgan eal hu he pœne eniht gehœlde purh cristes naman

(LS 30 (Pantaleon) B3.3.30) Drihten pe het secgan pas ping Forlœt min fole, pœt hit qffrie me.

(Exod

B8.1.4.2) porrne mœg ic pe for seod secgan: gelyf me gyf pu wille, ôœt rihte on pœre ylcan tyde pu wast œall pœt du nu wilnast to witanne, and hœfst œall pœt pu wilnast to hœbenne. (Solil 1 B9.4.2) 2. an wundorlic dincg, pe we willad secgan nu. (¿ELS (Maccabees) Β1.3.25) peh ic hit nu scortlice secgan scyle (Or 5 B9.2.6) pane ic wolde secgan (Solil 1 B9.4.2) Sculon we ane cydnesse his mœgenes secgan (Bede 2 B9.6.4) on hwylce time man sceall secgan ALLELUIA. (Ben RW B10.3.4) Swa we oft on bocum gehyrdon secgan pœt we mid pœre halegan prynnysse rode and mid his halegra reliquium (HomS 44 (Baz-Cr) B3.2.44) ic mœg to sode secgan pœt heo his agen gemœcca wœre. (ApT B4.1) ic gehyrde secgan pœt pu wœre gleaw pœron. (Gen B8.1.4.1 ) Men ne cunnon secgan to sode...hwapœm hlœste onfeng (Beo A4.1) mœg secgan se de wyle sod specan (Beo A4.1 ) Ac ic secgan mœg...hupa sacerdas sace niwiad, (Seasons A31) swa swa we her secgad: Sum halig bisceop wœs gehaten Gregorius (iEHom 8

Β 1.4.8) 3. us lyst nu eow secgan be pam halgan godspelle (jEHom 2 Β1.4.2)

4. Ac we wyllad nu secgan be his fordside. (¿ELS (Martin) Β1.3.30) Da pohte ic pœt pœt wœre seo helle pe ic oft on life embe secgan gehyrde

OECHom II, 23 Β 1.2.26) ne embe secgan ne gehyrde. (jEHom 22 Β1.4.22)

Lexical templates and syntactic variation

299

6. me soôlice secgan cunnon (El A2.6) ponne ic pe secgan mœge. (Met A6) we wyllad gedqfenlice eow secgan on engliscum gereorde (¿EHom 19 Β 1.4.19) 7. swa we secgan ne magon. (¿EHom 19 Β1.4.19) swa swa we lange œr seedon on disse ilcan bec (Bo B9.3.2) swa ge secgan on andetnysse. (MonCa 1 (Korhammer) C12.1 )

Notes 1. The research carried out for the writing of this paper has been supported by the Gobierno de Canarias research project PI 1999/136, "Diccionario Nuclear Sintáctico de Base Semántica del Léxico en Inglés Antiguo". This grant is hereby gratefully acknowledged. 2. In Dik (1978), the semantic postulates of Stepwise Lexical Decomposition provide the paradigmatic structuring of the lexicon. 3. Moreover, as we will see later on, in the case of a historical language like Old English it seems impossible to establish meaning definitions in terms of Stepwise Lexical Decomposition, since only the genus of the definition can be determined. 4. Activity verbs are considered intransitive verbs, independently of the number of syntactic arguments they present (Van Valin and LaPolla 1997: 122-125). 5. That is, the general tendency of the most prototypical lexemes in a subdomain is to present a greater number of complementation patterns than the less prototypical ones. 6. The Dictionary of Old English was conceived by Angus Cameron as an historical dictionary in the tradition established by the Oxford English Dictionary and is based on records written in English between 600 and 1150 A.D. The Dictionary of Old English electronic corpus is a complete record of surviving Old English. There are 3037 texts in the corpus including poetry, prose, interlinear glosses, glossaries, runic inscriptions and inscriptions in the Latin alphabet (http://ets.umdl.umich.edU/o/oec/moreoec.html). 7. The examples from the corpus can be found in Table 2 and in the Appendix. 8. For a deeper discussion about external variables and internal variables, see Mairal Usón and Faber (this volume). They also indicate that the Completeness Constraint, which says "All of the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in the sentence, and all of the referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic

300

Marta María González Orta

representation of the sentence" (Van Valin and LaPolla 1997: 325), will apply only to external arguments. 9. See Van Valin and LaPolla (1997: 122-125,147-154). 10. Van Valin and LaPolla (1997:111-113), when dealing with this type of alternation represented as do' (x, [pred' (x, y)]) o do' (x, [pred' (x, y)]) & BECOME pred' (y), argue that the basic lexical meaning of these verbs as activity verbs will be the one represented in their lexical entries in the lexicon, whereas the active accomplishment alternation will be related to their meaning in a particular context. On the contrary, Mairal Usón and Faber (this volume: 59) state that, applying the Lexical Template Modeling Process that we will present below, "all of the alternations involve different instantiations from the maximal lexical template underlying the class". In the case of Old English SECGAN, its corresponding maximal lexical template will be an active accomplishment and from this one, following a reduction process, the alternation activity will be derived. This will be shown in Table 2. 11. Within this verbal class, there can be other lexical templates. For instance, the lexical template of the subdomain in which the predicate tell is included will present the logical structure of a causative active accomplishment. [do* (x, [express.(a).to.(ß).in.language.(Y)' (x, y)] )] CAUSE [BECOME aware.of* (y, z)], where y = β, ζ = α 12. The numbers that precede the extra examples below correspond to each of the seven alternations derived from the two main lexical templates associated to SECGAN. In the case of number 5 the only example found is that appearing in Table 2.

References Cortés Rodríguez, Francisco J. and Ricardo Mairal Usón forthcoming A preliminary design for a syntactic dictionary of Old English on semantic principles. Cortés Rodríguez, Francisco J. and María Jesús Pérez Quintero 2001 Finding relief for FG lexical representations: A syntacticsemantic description of Old English verbs of 'Healing'. In: Maria Jesús Pérez Quintero (ed.), Challenges and Developments in Functional Grammar, 79-101. (Revista Canaria de Estudios Ingleses 42.) La Laguna: Servicio de publicaciones. Dik, Simon C. 1978 Stepwise Lexical Decomposition. Lisse: de Ridder.

Lexical templates and syntactic variation

301

Faber, Pamela and Ricardo Mairal Usón 1994 Methodological underpinnings for the construction of a functional lexicological model. Misceláneo: A Journal of English and American Studies 15: 193-217. 1997a The paradigmatic and syntagmatic structure of the lexical field of EXISTENCE in the elaboration of a semantic macronet. Studies in Language 21 (1): 129-167. 1997b The paradigmatic and syntagmatic structure of the lexical field of feeling. Cuadernos de Investigación Filológica XXIII-XXIV: 3560.

1998a

Towards a typology of predicate schemata in a Functional Lexematic Model. In: Gerd Wotjak (ed.), Towards a Functional Lexicology, 11-37. Frankfurt: Peter Lang. 1998b Methodological criteria for the elaboration of a functional lexicon-based grammar of the semantic domain of cognitive verbs. In: Hella Olbertz, Kees Hengeveld and Jesús Sánchez García (eds.), The Structure of the Lexicon in Functional Grammar, 324. Amsterdam/Philadelphia: John Benjamins Publishing Company. 1999 Constructing a Lexicon of English Verbs. Berlin/New York: Mouton de Gruyter. Healey, Antonette di P. (ed.) The Dictionary of Old English Corpus. Web site ISBN 0-47200277-5. Lakofíj George 1987 Women, Fire and Dangerous Things. Chicago: University of Chicago Press. Langacker, Ronald W. 1987 Foundations of Cognitive Grammar, Volume 1. Stanford: Stanford University Press. 1991a Foundations of Cognitive Grammar, Volume 2. Stanford: Stanford University Press. 1991b Concept, Image and Symbol: The Cognitive Basis of Grammar. Berlin/New York: Mouton de Gruyter. Mairal Usón, Ricardo 1993 Complementation patterns of cognitive, physical perception and speech act verbs in the English language. A Functional-Cognitive approach. Ph.D. dissertation, Universidad de Zaragoza. 1997 Fundamentos teóricos para la integración del eje cognitivo en el Modelo Lexemático-Funcional. In: Juan de Dios Luque Durán and A. Pamies Bertrán (eds.), 31-54. Problemas de lexicología y lexicografta. Granada: Método Ediciones.

302

Marta María González Oria

1998

El Modelo Lexemático Funcional. El legado lingüístico de Leocadio Martín Mingorance. Edited by Amalia Marín Rubiales. Granada: Universidad de Granada. Mairal Usón, Ricardo and Pamela Faber this volume Functional Grammar and lexical templates. Roberts, Jane and Christian Kay 1995 A Thesaurus of Old English. London: King's College London Medieval Studies. Van Valin, Robert D. and Randy LaPolla 1997 Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Lexical-semantic explorations in English verbs of physical contact: Iconicity and linguistic representations Margarita Goded Rambaud Rocío Jiménez Briones

1. Introduction The purpose of this paper is to explore the possibilities of lexical templates as representational tools accounting for iconicity as a possible conceptualization and organization strategy. It is, precisely, because what is being transmitted in the communication process is meaning (abstractions of one kind or another not directly accessible to the speaker and addressee) that we need to operate with the only available empirical data. Evidently, meaning cannot be considered empirical data as we do not have direct access to it. However, what we do have direct access to is the lexical vehicle which conveys a certain meaning. Such a lexical vehicle and its morphosyntactic and pragmatic organization is our only available source of purely linguistic data if we rule out all extralinguistic contextual information which can be processed through other perception and sensory systems. That is why we need a representational tool that could account for the syntax-semantics interface in the lexicon. The predicate frames proposed by Functional Grammar (FG) encapsulate most of the information relevant for the description of all elements contributing to the codification of linguistic items ranging from an isolated lexical entry to a complete expression. In addition, the formalism proposed in Dik's (1997) predicate frames makes it possible to structure major meaning components at a very high level of influence and it is, precisely, this positioning that constitutes a

304

Margarita Goded Rambaud and Rocío Jiménez Briones

major advance in the pursuit of meaning representation of lexical items in this model. However, in order to describe an area of meaning that we postulate to be iconic with certain physical experiences, a model of representation which organizes its entries in a way that links a hierarchy in the organization of meaning with its semantic and syntactic codification is expected to be much more transparent. FG predicate frames fail to offer an explicit connection between syntax and semantics in a way we feel to be essential in connection with meaning representation. Exploring the lexicon of restricted areas of human experience suggests that meaning is not only highly hierarchical and structured, but it also presents some degree of iconicity, as has been extensively discussed in the literature (Haiman 1985: 3,21-159; Langacker 1987: 12; Wierzbicka 1985: 334, among many others). For this reason, iconicity is suggested in this paper as a possible conceptualization strategy that could help explain the lexical organization of the two selected verbal subdomains under study. Following Wierzbicka (1985), we postulate that "the relationship is iconic in the sense that the system of formal distinctions and the system of conceptual distinctions are mutually isomorphic." Iconicity, then, is explored here as a possible driving force to be identified in the structure of inheritance mechanisms within the templates proposed by Mairal Usón and Faber (this volume) in the Functional Lexematic Model (FLM). As a result, the parameters of intensity and iteration are suggested as ways of codifying iconicity in the subdomains analyzed. Furthermore, these semantic parameters reflect the syntax-semantic interface showing how certain grammatical features can be derived from an accurate lexical-semantic representation. In order to codify this, a number of English verbs belonging to the lexical domain of physical contact have been selected as data to be analyzed in relation with a pattern of certain human physical experiences. However, the study of their relative position in a preestablished construct may be more clearly shown if linguistic codification follows certain patterns.

Lexical-semantic explorations in English verbs of physical contact

305

The methodological approach taken to achieve this purpose is an application of Faber and Mahal's (1999) expansion of the FLM, where a selection of English verbs belonging to the subdomain chosen are analyzed and represented in a continuum or scale. This coding, including both semantic and syntactic patterns, can be more transparently represented if we use lexical templates as representational tools applied to the architecture of the two subdomains defined as "physical contact: to touch someone to show or give pleasure" and "physical contact: to hit someone in order to hurt". This paper is organized in three broad sections. Firstly, the architecture of the two subdomains is presented, and the alternations characterizing these verbs are explained. Secondly, the parameters of iteration and intensity are identified and discussed as the elements capturing iconicity in the subdomains under study. Finally, the linguistic representation of this isomorphic relation is discussed, comparing FG predicate frames with FLM lexical templates to propose the latter framework as a representational tool linking meaning and the experiences represented.

2. The FLM applied to the architecture of two subdomains of the verbs of physical contact In order to analyze the intensity parameter in depth, that is, the different degrees by means of which a type of contact is effected (from lightly to forcefully), we, firstly, have to extract the syntacticsemantic information of each of the verbs that, in principle, codify such a parameter in English. In other words, it is necessary to provide the two subdomains under consideration with a syntacticsemantic organization. Since one of the most accurate studies on the architecture of the lexicon has been carried out by the FLM, its main tenets and methodology will be applied to such subdomains. The FLM was developed as an attempt to enrich the lexicon component developed of Dik's Functional Grammar. It was elaborated by Martin Mingorance (1984, 1985ab, 1990), who wisely integrated Dik's FG and Coseriu's Lexematic Theory. Martin Mingorance

306

Margarita Goded Rambaud and Rocío Jiménez Briones

claimed that meaning should be approached from both a microstructural and macrostructural stance. Microstructurally, the meaning definition of a lexeme plays a role in the syntax-semantics interface, which, as will be seen later, is not present in the FG meaning definitions. Macrostructurally, lexemes have a place within the larger context of their lexical domain, as well as with lexemes from other domains. As a result of this twofold analysis, lexemes are conceived as dynamic representations of our own conceptual structure, instead of a frozen list of words (Faber and Mairal 1999: 3). The microstructural analysis is currently being developed by the FLM research group (see Mairal Usón and Faber this volume), whereas the macrostructural study has already been accomplished, resulting in the architecture of the English lexicon into highlyinformative lexically-articulated semantic subdomains, such as these below: (1)

physical contact: to touch someone to show or give pleasure 1. TOUCH: to lightly come into contact with a body part using

your hands or fingers in order to show affection or give pleasure. 1.1. TICKLE: to touch with the fingers a sensitive part of the body to make someone laugh. 1.2. PAT: to touch gently and repeatedly with the flat hand to show friendliness. 1.3. STROKE: to touch someone with your hands gently

and lovingly for pleasure. 1.3.1. CARESS: to stroke someone gently to show affection or love. 1.3.2. FONDLE: to stroke someone gently in a loving or sexual way.

Lexical-semantic explorations in English verbs of physical contact

(2)

307

physical contact: to hit someone in order to hurt 1. HIT: to hurt someone forcefully by using the hand or something held in the hand. 1.1. STRIKE: fml/legal. to hit someone. 1.2. BASH: infini, to hit someone hard, esp. in a fight. 1.3. PUNCH: TO hit someone hard with your clenched fist. 1.4. SLAP: To hit someone with the flat part of your hand. 1.5. BEAT: to hit someone severely again and again. 1.5.1. BATTER: to beat women/children with something heavy. 1.6. BRUISE: to hit part of the body without breaking the skin, usu. producing a purple mark on it. 2. SCRATCH: to hurt your skin slightly with your fingernails or a sharp object.

To obtain the organization of (1) and (2) above, the FLM bases its analysis on the complementation of syntagmatic and paradigmatic relations. Whereas the latter codify how lexemes are arranged on the axis of selection through an onomasiological organization, the former account for the grammatical information vital for the correct usage of each lexeme. Since, as will be stated below, it is the place of a lexeme in the paradigmatic structure of its domain that determines its syntagmatic information, the paradigmatic arrangement of (1) and (2) will be better handled now, postponing the syntactic phenomena to the end of this section. Unlike many thesauri that start from a preexisting concept and then move downwards to "fill it in" with lexical items, we have worked upward from the meaning definitions of each lexeme. From these meaning definitions, which have been obtained by looking up each lexical item in some of the most thorough dictionaries of the English language, those verbs that share the same genus are arranged in the same subdomain. These have been restricted to those predicates that express physical contact by using hands and unspecified

308

Margarita Goded Rambaud and Rocío Jiménez Briones

objects, which explains why verbs such as kiss, kick, whip, club, etc., are not present in the subdomains under consideration. This idea of the genus as the prototypical element of a subdomain and the one with the wider syntactic spectrum finds its parallel in the way we handle central basic human experiences. As Langacker (1991: 295) puts it: Certain recurrent and sharply differentiated aspects of our experience emerge as archetypes, which we normally use to structure our conceptions insofar as possible. Since language is a means by which we describe our experience, it is natural that such archetypes should be seized upon as the prototypical values of basic linguistic constructs.

For instance, the genus touch arranges those lexemes of (1), whereas hit does so for (2). Those distinctive features particular to each lexeme allow us to distinguish and organize them hierarchically as dependent on one or other hypernym term or archilexeme. Such is the case, for example, with stroke and caress, where the latter is distinguished from its hypernym through the distinctive feature to show affection or love. As a result of this methodical process of analysis, our lexical domains are hardly related to those lexical fields proposed by either traditional thesauri or lexicons. Since intensity is one of those distinctive features (along with instrument and purpose) that help to distinguish certain lexical items and subdomains from each other, it is this type of organization that allows us to check the way this parameter is semantically present in the two subdomains under study. Thus, adverbs such as lightly, gently and lovingly, codify the intensity feature for touch or stroke in (1) above, while forcefully, hard or severely do the same for hit, bash and beat. The hierarchical organization of the subdomains in (1) and (2) sees its correlation in the syntagmatic axis; that is, the syntactic complementation of each of these lexical items depends on its position in the semantic hierarchy. This close interrelation of syntax and semantics is what the FLM puts forward as the Principle of Lexical Iconicity:

Lexical-semantic explorations in English verbs ofphysical contact

309

The greater the semantic scope of a lexeme, the greater its syntactic variation. (Faber and Mairal 1999:187)

If applied to our subdomains, we would expect that touch, hit and scratch, since they are the most prototypical terms in their respective subdomains, would show a greater syntactic coverage than their hyponyms. Let us now approach each of the subdomains in turn. Adapting the terminology used in Levin (1993),1 Tables 1 and 2 show the syntactic alternations in which the hypernym touch and its several hyponyms and lexicalized parameters appear:2 Table 1. Alternations of touch ALTERNATION

EXAMPLE

Transitive

Mrs Travers' fingers touched the little girl's cheek (CC)

Body-part possessor ascension alternation

He touched the girl on the arm to get her attention (CIDE) He touched the girl's arm (our example) He touched his sister with the tips of his fingers (our example) a. Reflexive object: Carrie touched herself (Levin) b. Body-part object: Carrie touched her hair (Levin)

Instrument "with" Intentional interpretation only

Table 2 shows how the hyponyms of touch are more restricted in their syntactic alternations, since the intentional interpretation only pattern, if not completely ungrammatical, is, at least, marginal for some of them: (3)

*/?Carrie tickled/patted/stroked/caressed/fondled herself. */? Carrie tickled/patted/stroked/caressed/fondled her hair.

310

Margarita Goded Rambaud and Rocío Jiménez Briones

Table 2. Alternations of "to touch someone to show or give pleasure"

Touch Tickle [inst] fingers [purp] to make s.o. laugh Pat [inten] gently [iter] repeatedly [inst] palms [purp] to show friendliness Stroke [inten] gently & lovingly [inst] hands Caress [inten] gently [purp] to show affection or love Fondle [inten] gently [man] lovingly, sexually

Transitive

Body-part ascension

Instrument "with"

Intentional interpretation

+ +

+ +

+ +

+

+

+

—

—

+

+

+

+

+

+

+

+

+

In relation to the second subdomain ('Ho hit someone in order to hurt"), the Principle of Lexical Iconicity is also validated. The more general terms hit and scratch display the greater number of syntactic alternations:

Lexical-semantic explorations in English verbs of physical contact

311

Table 3. Alternations of hit and scratch ALTERNATION

EXAMPLE

Transitive

His father used to hit him (Garrudo) She scratched his face with her fingernails (LLCE) Instrument "with" He hit him with a stick (our example) She scratched his face with her fingernails (LLCE) Body-part possessor ascension The vicar was so angry, he hit the burglar on the alternation head with a candlestick (CC) She scratched him on the face (our example) He hit the burglar's head with a candlestick (our example) She scratched his face (our example) Unintentional interpretation a. Reflexive object: Paula hit herself on the doorknob (Levin) Be careful not to scratch yourself on the roses (CIDE) b. Body-part object: I hit my knee on/against the chair (LDCE) He's scratched his hand on a nail (OALD) Conative Alternation He hit at his opponent but missed (Garrudo) He scratched at Paul/at her knee (our example) Resultative Phrase He hit Paul to death (our example) *An American businessman was scratched unconscious (our example) Instrument Subject Alterna- He hit Paul with a stick (our example) The stick hit Paul (our example) tion She scratched his face with her fingernails (LLCE) Herfingernailsscratched his face (our example)

Table 4 shows the wider syntactic coverage of hit and scratch over punch, slap, beat, batter, and bruise. The only two verbs that can also take the same seven possible syntactic patterns are strike and bash, although their pragmatic features of formality and informality hierarchically restrict them to being hyponyms of the most neutral hit.

312

Margarita Goded Rambaud and Rocío Jiménez Briones

Table 4. Alternations of "to hit someone in order to hurt"

Hit Strike [prag] formality Bash [prag] informal Punch [inten] hard [inst] fist Slap [inst] palm Beat [inten] severely [iter] again & again Batter [goal] women & children [inst] sth heavy Bruise [goal] part of the body [man] no breaking

Trans.

Instr. Body-part Unintent Conative Result. Instr. "with" possessor Interpret, alternat. Phrase Subj. ascension

+ +

+ +

+

+

+

+

+

+

+

+ +

+ +

+ +

+

+

+

+

+

—

+

-

+

+

—

+

—

+

-

+

+

—

+

+

+

+

+

—

+

+

+

+

+

—

+

—

+

+

+

+

—

+

+

+

+

—

+

+ Scratch [inten] slightly [inst] fingernail

To sum up, the premises laid down by the FLM provide us with a useful tool to arrange these two subdomains into highly organized lexical-semantic-syntactic sets. Through this type of organization, the fact that predicates such as punch, slap, beat and batter do not take part in the unintentional interpretation and the instrument subject syntactic alternations, is associated with the semantics of these predi-

Lexical-semantic explorations in English verbs ofphysical contact

313

cates: they are agentive verbs that always imply the use of an instrument on the part of a volitional subject. This explains the ungrammaticality of (4): (4)

*Paula punched/slapped/beat/battered herself on the doorknob. *He punched/slapped/beat/battered his arm against a shelf.

As will be shown below, one can still go further and claim the existence of one lexical-semantic-syntactic construction for the whole subdomain, that is, lexical templates à la Role and Reference Grammar (RRG). But, for a complete representation of both subdomains, it is necessary to first deal with the interpretation and codification of the parameters of intensity and iteration.

3. Iconicity and its interpretation: the parameters of intensity and iteration in the subdomains analyzed After the lexical-semantic-syntactic study of the predicates corresponding to two vital areas of human experience, such as physical contact that results in pleasure or pain, some reflection about the degree of iconicity between our experience and the linguistic elements available to express is needed. When analyzing the subdomain of verbs of contact, Faber and Mairal (1999: 159-162) identified an iterative parameter that they related to the type of blow. We suggest here an extension of their description and, instead of establishing the iconicity on the basis of the type of blow, we argue for the type of contact as the key feature. In other words, we put forward the intensity of contact as the element that seems to draw the line that separates pleasure from pain. Whether such a parameter is linguistically codified, and how, will also be of concern in this section. The intensity parameter is a semantic parameter that encodes a value which is a function of the meaning area where it operates (Faber and Mairal 1999: 107-108). This is clearly shown in the two sub-

314

Margarita Goded Rambaud and Rocío Jiménez Briones

domains under study, since each of them would represent a point of a possible scale or continuum of intensity. At one end, we would find touch ("to lightly come into contact with a body part using your hands or fingers in order to show affection or give pleasure") and its hyponyms, whereas, at the other end, we would find hit ("to hurt someone forcefully by using the hand or something held in the hand") and scratch ("to hurt your skin slightly with your fingernails or a sharp object"), as well as their hyponyms: (5)

_>

(-) < touch tickle pat stroke caress fondle

scratch

(+)

hit strike bash punch slap beat batter bruise

Thus, the codification of the physical experience of pleasure or pain through contact is firmly grounded on the intensity of such contact. Moreover, this scale of intensity in (5) is iconic to the lexicalsemantic-syntactic organization of the predicates that speakers and addressees use when talking about different types of contact (Tables 2 and 4). To put it differently, the iconic relationship embodied in this semantic parameter of intensity supports the syntax-semantics interface, "the aspects of grammar where these two interact" (Van Valin and LaPolla 1997: 139), or the increasingly-assumed thesis that linguistic structures are not independent from their lexicalsemantic meaning. This interface between syntax and semantics is made explicit by the existence of what Faber and Mairal (1998a: 38) define as "the synsem parameters". These syntactic-semantic parameters operate throughout the lexicon and play an important part in the generation of the different syntactic clause structures. Since not all of them find

Lexical-semantic explorations in English verbs ofphysical contact

315

a direct correlation in syntax, the synsem parameters have accordingly been divided into three types: (a) (b) (c)

lexically-realized grammatical parameters, which are obligatory and directly affecting verb complementation. lexically-realized optional parameters, which are semantically implicit, though not always present in the linguistic expression. lexically-realized contextual parameters, which are not syntactically projected, but provide us with clues for a correct contextual setting.

The intensity parameter would belong to those lexically-realized contextual parameters which, though not syntactically projected, are useful to hierarchically organize the lexemes of each subdomain in contrasting ways. As shown in Tables 2 and 4, this is the case in our examples: the parameter of intensity, along with instrument and purpose (which are also lexically-realized contextual parameters), allows us to contrast predicates inside the same subdomain and across subdomains. Thus, punch differsfromscratch because of the features [hard] & [fist] vs. [slightly] & [fingernails], and, in turn, from stroke, due to features such as [gently & lovingly] and [hands]. In other words, this parameter is lexicalized in both subdomains, although it does not show any effect on the syntactic alternations of each of them. This is what Portero (1997: 138) regards as intensificación cualitativa (qualitative intensification), that is, the type of intensity associated with different degrees of a verb quality. As shown above, the predicates belonging to our first subdomain are associated with the lowest degree of intensity, unlike those belonging to the second one, in which the degree increases towards the other end of the continuum of intensity. However, if intensity finds no correlation in the syntactic alternations of these subdomains, where, and how, is our distinct physical experience of pain and pleasure through contact reflected linguistically? The answer partly lies in the study of another type of parameter: the lexically-realized grammatical parameter of iteration.

316

Margarita Goded Rambaud and Rocío Jiménez Briones

All the verbs included in the second subdomain could be classified as iterative verbs, i.e. they encode hitting someone repeatedly. That is why their co-occurrence with the adverbial once is not allowed, especially in the case of beat and batter where iteration is lexically codified as well: (6)

*His father used to hit/strike/bash/punch/slap/beat/batter him once. *She scratched his face with herfingernailsonce.

It seems that, apart from its intensity, it is this continuous, repetitive contact with a body part or an object that turns it into pain. Furthermore, "the repetition of the action results in the attainment of a final state" (Faber and Mairal 1998a: 48); the agent may affect the patient in such a way that the latter achieves a state of unconsciousness, senselessness, or even death. This final state finds a syntactic codification in this subdomain by means of the addition of a resultative phrase (an adjectival complement or a prepositional phrase), which proves that this pattern belongs to the lexically-realized grammatical ones, since, as can be observed below, it affects verb complementation: (7)

An American businessman was hit/stricken/bashed/punchedV slapped/beaten/bruised unconscious/to death.

However, there must be something else, apart from iteration, that distinguishes when a type of contact is felt as pain or pleasure, since the repetitive nature of the contact is also present in the predicates of the first subdomain, although they do not allow the resultative phrase alternation: (8)

*Babies want to be tickled/patted/stroked/fondled/caressed once.

(9)

*He ticlded/patted/stroked/fondled/caressed her sister unconscious/to tears.

Lexical-semantic explorations in English verbs of physical contact

317

Again one needs to identify where the difference between these two subdomains lies or why iteration is syntactically expressed through the resultative phrase in the second subdomain, but not in the first one. The explanation may be found in the co-occurrence of both parameters. We argue, then, that it is not only iteration that brings about such different syntactic alternations, but rather iteration AND the highest degree of intensity. In this way, one can only feel pain if the contact is repeated and/or the intensity with which it is performed is high on the scale of intensity, i.e. forcefully, hard, severely, etc. This fact would also explain the syntactic realization of the resultative phrase alternation when the intensity is "negative" (the second subdomain), unlike its "positive" counterpart (the first subdomain). As is recurrent in any language, what is conceptually salient and deviates from the norm, from the unmarked type, receives syntactic codification (Siewierska 1991: 5). It seems, then, that happiness is the unmarked value when talking about feelings: (...) another evidence that the default value in FEELING is not a neutral emotional state but a positive one. In other words, our normal state is happiness and move away from that state necessarily means a change for the worse (Faber and Mairal 1998b: 53)

In the case under consideration, "happiness" could be interpreted as lack of physical pain. Thus, this change for the worse is iconically mirrored through the adjectival complement or the prepositional phrase of the resultative alternation. For example, a predicate such as scratch, which codifies iteration, but neither light nor extreme intensity, cannot appear with the resultative phrase alternation: (10)

*An American businessman was scratched unconscious/to death.

On the other hand, if the predicates belonging to the first domain (the "happy" one) are used in such a way that the iterative action and the intensity reach the negative pole of the cline, their syntactic cooccurrence with the resultative phrase is possible:

318

Margarita Goded Rambaud and Rocío Jiménez Briones

(11)

She tickled his feet until he couldn 't stop laughing. (CIDE)

(12)

She tickled her boyfriend to death, (metaphorical interpretation)

4. Iconicity and its codification: the linguistic formalization of the parameters of intensity and iteration in the subdomains analyzed So far we have seen that, following the premises of the FLM, the structuring of our two subdomains in the articulated semanticsyntactic sets of (1), (2) and Tables 2 and 4 brings to the foreground the close relation between syntax and semantics. Moreover, it has been explained how this syntax-semantics interface is deeply rooted in our natural experience of the intensity and repetitiveness of a physical contact, which is iconically codified by means of two lexically-realized synsem parameters: intensity, as a contextual parameter, and iteration, as a grammatical one. The issue to be dealt with next concerns the highly controversial linguistic formalization of such interface. In the next subsection, FG's unsuccessful account of this question will be discussed, leaving for the last subsection Mairal Usón and Faber's reconsideration on this topic (see this volume) as applied to our analysis. 4.1. The FG approach: the limitations of its argument structures In FG all lexical items are handled as predicates, appearing in the form of structures called predicate frames (Dik 1997: 59). These are a type of metalanguage-free formulae that specify the syntactic category of the predicate, the number of arguments, their semantic functions and their selection restrictions. Furthermore, predicate frames are "the most basic building blocks at the morphosemantic level of linguistic organization" (Dik 1997: 58) from which the underlying predication is derived.

Lexical-semantic explorations in English verbs of physical contact

319

As for the meaning of lexical items, each predicate frame is associated with a meaning definition, which also takes the format of a predicate frame, as (13) and (14) make explicit: (13)

hit [V]

(XI:

< prototyp.3

human>)AGENT ( * 2 :

)c¡oAL D f = hurt [V] (xi)AGENT (X2: )GOAL ( σ ι : )INSTRUMENT(CTi: )MANNER

(14)

punch [V] (xi: < prototyp. human>)AGENT (*2'· )INSTRUMENT (